You cannot select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.
suricata/src/stream-tcp-reassemble.c

3684 lines
121 KiB
C

tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* Copyright (C) 2007-2016 Open Information Security Foundation
*
* You can copy, redistribute or modify this Program under the terms of
* the GNU General Public License version 2 as published by the Free
* Software Foundation.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* version 2 along with this program; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA
* 02110-1301, USA.
*/
/**
* \file
*
* \author Gurvinder Singh <gurvindersinghdahiya@gmail.com>
* \author Victor Julien <victor@inliniac.net>
*
* Reference:
* Judy Novak, Steve Sturges: Target-Based TCP Stream Reassembly August, 2007
*
*/
#include "suricata-common.h"
#include "suricata.h"
#include "debug.h"
#include "detect.h"
#include "flow.h"
#include "threads.h"
#include "conf.h"
#include "flow-util.h"
#include "threadvars.h"
Add per packet profiling. Per packet profiling uses tick based accounting. It has 2 outputs, a summary and a csv file that contains per packet stats. Stats per packet include: 1) total ticks spent 2) ticks spent per individual thread module 3) "threading overhead" which is simply calculated by subtracting (2) of (1). A number of changes were made to integrate the new code in a clean way: a number of generic enums are now placed in tm-threads-common.h so we can include them from any part of the engine. Code depends on --enable-profiling just like the rule profiling code. New yaml parameters: profiling: # packet profiling packets: # Profiling can be disabled here, but it will still have a # performance impact if compiled in. enabled: yes filename: packet_stats.log append: yes # per packet csv output csv: # Output can be disabled here, but it will still have a # performance impact if compiled in. enabled: no filename: packet_stats.csv Example output of summary stats: IP ver Proto cnt min max avg ------ ----- ------ ------ ---------- ------- IPv4 6 19436 11448 5404365 32993 IPv4 256 4 11511 49968 30575 Per Thread module stats: Thread Module IP ver Proto cnt min max avg ------------------------ ------ ----- ------ ------ ---------- ------- TMM_DECODEPCAPFILE IPv4 6 19434 1242 47889 1770 TMM_DETECT IPv4 6 19436 1107 137241 1504 TMM_ALERTFASTLOG IPv4 6 19436 90 1323 155 TMM_ALERTUNIFIED2ALERT IPv4 6 19436 108 1359 138 TMM_ALERTDEBUGLOG IPv4 6 19436 90 1134 154 TMM_LOGHTTPLOG IPv4 6 19436 414 5392089 7944 TMM_STREAMTCP IPv4 6 19434 828 1299159 19438 The proto 256 is a counter for handling of pseudo/tunnel packets. Example output of csv: pcap_cnt,ipver,ipproto,total,TMM_DECODENFQ,TMM_VERDICTNFQ,TMM_RECEIVENFQ,TMM_RECEIVEPCAP,TMM_RECEIVEPCAPFILE,TMM_DECODEPCAP,TMM_DECODEPCAPFILE,TMM_RECEIVEPFRING,TMM_DECODEPFRING,TMM_DETECT,TMM_ALERTFASTLOG,TMM_ALERTFASTLOG4,TMM_ALERTFASTLOG6,TMM_ALERTUNIFIEDLOG,TMM_ALERTUNIFIEDALERT,TMM_ALERTUNIFIED2ALERT,TMM_ALERTPRELUDE,TMM_ALERTDEBUGLOG,TMM_ALERTSYSLOG,TMM_LOGDROPLOG,TMM_ALERTSYSLOG4,TMM_ALERTSYSLOG6,TMM_RESPONDREJECT,TMM_LOGHTTPLOG,TMM_LOGHTTPLOG4,TMM_LOGHTTPLOG6,TMM_PCAPLOG,TMM_STREAMTCP,TMM_DECODEIPFW,TMM_VERDICTIPFW,TMM_RECEIVEIPFW,TMM_RECEIVEERFFILE,TMM_DECODEERFFILE,TMM_RECEIVEERFDAG,TMM_DECODEERFDAG,threading 1,4,6,172008,0,0,0,0,0,0,47889,0,0,48582,1323,0,0,0,0,1359,0,1134,0,0,0,0,0,8028,0,0,0,49356,0,0,0,0,0,0,0,14337 First line of the file contains labels. 2 example gnuplot scripts added to plot the data.
14 years ago
#include "tm-threads.h"
#include "util-pool.h"
#include "util-unittest.h"
#include "util-print.h"
#include "util-host-os-info.h"
#include "util-unittest-helper.h"
#include "util-byte.h"
#include "util-device.h"
#include "stream-tcp.h"
#include "stream-tcp-private.h"
#include "stream-tcp-reassemble.h"
#include "stream-tcp-inline.h"
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
#include "stream-tcp-list.h"
#include "stream-tcp-util.h"
#include "stream.h"
#include "util-debug.h"
#include "app-layer-protos.h"
#include "app-layer.h"
#include "app-layer-events.h"
#include "detect-engine-state.h"
#include "util-profiling.h"
#include "util-validate.h"
#ifdef DEBUG
static SCMutex segment_pool_memuse_mutex;
static uint64_t segment_pool_memuse = 0;
static uint64_t segment_pool_memcnt = 0;
#endif
static PoolThread *segment_thread_pool = NULL;
/* init only, protect initializing and growing pool */
static SCMutex segment_thread_pool_mutex = SCMUTEX_INITIALIZER;
/* Memory use counter */
SC_ATOMIC_DECLARE(uint64_t, ra_memuse);
/* prototypes */
TcpSegment *StreamTcpGetSegment(ThreadVars *tv, TcpReassemblyThreadCtx *);
void StreamTcpCreateTestPacket(uint8_t *, uint8_t, uint8_t, uint8_t);
void StreamTcpReassemblePseudoPacketCreate(TcpStream *, Packet *, PacketQueue *);
void StreamTcpReassembleInitMemuse(void)
{
SC_ATOMIC_INIT(ra_memuse);
}
/**
* \brief Function to Increment the memory usage counter for the TCP reassembly
* segments
*
* \param size Size of the TCP segment and its payload length memory allocated
*/
void StreamTcpReassembleIncrMemuse(uint64_t size)
{
(void) SC_ATOMIC_ADD(ra_memuse, size);
SCLogDebug("REASSEMBLY %"PRIu64", incr %"PRIu64, StreamTcpReassembleMemuseGlobalCounter(), size);
return;
}
/**
* \brief Function to Decrease the memory usage counter for the TCP reassembly
* segments
*
* \param size Size of the TCP segment and its payload length memory allocated
*/
void StreamTcpReassembleDecrMemuse(uint64_t size)
{
#ifdef UNITTESTS
uint64_t presize = SC_ATOMIC_GET(ra_memuse);
if (RunmodeIsUnittests()) {
BUG_ON(presize > UINT_MAX);
}
#endif
(void) SC_ATOMIC_SUB(ra_memuse, size);
#ifdef UNITTESTS
if (RunmodeIsUnittests()) {
uint64_t postsize = SC_ATOMIC_GET(ra_memuse);
BUG_ON(postsize > presize);
}
#endif
SCLogDebug("REASSEMBLY %"PRIu64", decr %"PRIu64, StreamTcpReassembleMemuseGlobalCounter(), size);
return;
}
uint64_t StreamTcpReassembleMemuseGlobalCounter(void)
{
uint64_t smemuse = SC_ATOMIC_GET(ra_memuse);
return smemuse;
}
/**
* \brief Function to Check the reassembly memory usage counter against the
* allowed max memory usgae for TCP segments.
*
* \param size Size of the TCP segment and its payload length memory allocated
* \retval 1 if in bounds
* \retval 0 if not in bounds
*/
int StreamTcpReassembleCheckMemcap(uint32_t size)
{
if (stream_config.reassembly_memcap == 0 ||
(uint64_t)((uint64_t)size + SC_ATOMIC_GET(ra_memuse)) <= stream_config.reassembly_memcap)
return 1;
return 0;
}
/* memory functions for the streaming buffer API */
/*
void *(*Malloc)(size_t size);
*/
static void *ReassembleMalloc(size_t size)
{
if (StreamTcpReassembleCheckMemcap(size) == 0)
return NULL;
void *ptr = SCMalloc(size);
if (ptr == NULL)
return NULL;
StreamTcpReassembleIncrMemuse(size);
return ptr;
}
/*
void *(*Calloc)(size_t n, size_t size);
*/
static void *ReassembleCalloc(size_t n, size_t size)
{
if (StreamTcpReassembleCheckMemcap(n * size) == 0)
return NULL;
void *ptr = SCCalloc(n, size);
if (ptr == NULL)
return NULL;
StreamTcpReassembleIncrMemuse(n * size);
return ptr;
}
/*
void *(*Realloc)(void *ptr, size_t orig_size, size_t size);
*/
static void *ReassembleRealloc(void *optr, size_t orig_size, size_t size)
{
if (size > orig_size) {
if (StreamTcpReassembleCheckMemcap(size - orig_size) == 0)
return NULL;
}
void *nptr = SCRealloc(optr, size);
if (nptr == NULL)
return NULL;
if (size > orig_size) {
StreamTcpReassembleIncrMemuse(size - orig_size);
} else {
StreamTcpReassembleDecrMemuse(orig_size - size);
}
return nptr;
}
/*
void (*Free)(void *ptr, size_t size);
*/
static void ReassembleFree(void *ptr, size_t size)
{
SCFree(ptr);
StreamTcpReassembleDecrMemuse(size);
}
/** \brief alloc a tcp segment pool entry */
void *TcpSegmentPoolAlloc()
{
if (StreamTcpReassembleCheckMemcap((uint32_t)sizeof(TcpSegment)) == 0) {
return NULL;
}
TcpSegment *seg = NULL;
seg = SCMalloc(sizeof (TcpSegment));
if (unlikely(seg == NULL))
return NULL;
return seg;
}
int TcpSegmentPoolInit(void *data, void *initdata)
{
TcpSegment *seg = (TcpSegment *) data;
/* do this before the can bail, so TcpSegmentPoolCleanup
* won't have uninitialized memory to consider. */
memset(seg, 0, sizeof (TcpSegment));
if (StreamTcpReassembleCheckMemcap((uint32_t)sizeof(TcpSegment)) == 0) {
return 0;
}
#ifdef DEBUG
SCMutexLock(&segment_pool_memuse_mutex);
segment_pool_memuse += sizeof(TcpSegment);
segment_pool_memcnt++;
SCLogDebug("segment_pool_memcnt %"PRIu64"", segment_pool_memcnt);
SCMutexUnlock(&segment_pool_memuse_mutex);
#endif
StreamTcpReassembleIncrMemuse((uint32_t)sizeof(TcpSegment));
return 1;
}
/** \brief clean up a tcp segment pool entry */
void TcpSegmentPoolCleanup(void *ptr)
{
if (ptr == NULL)
return;
StreamTcpReassembleDecrMemuse((uint32_t)sizeof(TcpSegment));
#ifdef DEBUG
SCMutexLock(&segment_pool_memuse_mutex);
segment_pool_memuse -= sizeof(TcpSegment);
segment_pool_memcnt--;
SCLogDebug("segment_pool_memcnt %"PRIu64"", segment_pool_memcnt);
SCMutexUnlock(&segment_pool_memuse_mutex);
#endif
}
/**
* \brief Function to return the segment back to the pool.
*
* \param seg Segment which will be returned back to the pool.
*/
void StreamTcpSegmentReturntoPool(TcpSegment *seg)
{
if (seg == NULL)
return;
seg->next = NULL;
seg->prev = NULL;
PoolThreadReturn(segment_thread_pool, seg);
}
/**
* \brief return all segments in this stream into the pool(s)
*
* \param stream the stream to cleanup
*/
void StreamTcpReturnStreamSegments (TcpStream *stream)
{
TcpSegment *seg = stream->seg_list;
TcpSegment *next_seg;
if (seg == NULL)
return;
while (seg != NULL) {
next_seg = seg->next;
StreamTcpSegmentReturntoPool(seg);
seg = next_seg;
}
stream->seg_list = NULL;
stream->seg_list_tail = NULL;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/** \internal
* \brief check if segments falls before stream 'offset' */
static inline int SEGMENT_BEFORE_OFFSET(TcpStream *stream, TcpSegment *seg, uint64_t offset)
{
if (seg->sbseg.stream_offset + seg->sbseg.segment_len <= offset)
return 1;
return 0;
}
/** \param f locked flow */
void StreamTcpDisableAppLayer(Flow *f)
{
if (f->protoctx == NULL)
return;
TcpSession *ssn = (TcpSession *)f->protoctx;
StreamTcpSetStreamFlagAppProtoDetectionCompleted(&ssn->client);
StreamTcpSetStreamFlagAppProtoDetectionCompleted(&ssn->server);
StreamTcpDisableAppLayerReassembly(ssn);
}
/** \param f locked flow */
int StreamTcpAppLayerIsDisabled(Flow *f)
{
if (f->protoctx == NULL || f->proto != IPPROTO_TCP)
return 0;
TcpSession *ssn = (TcpSession *)f->protoctx;
return (ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED);
}
int StreamTcpReassemblyConfig(char quiet)
{
uint32_t segment_prealloc = 2048;
ConfNode *seg = ConfGetNode("stream.reassembly.segment-prealloc");
if (seg) {
uint32_t prealloc = 0;
if (ByteExtractStringUint32(&prealloc, 10, strlen(seg->val), seg->val) == -1)
{
SCLogError(SC_ERR_INVALID_ARGUMENT, "segment-prealloc of "
"%s is invalid", seg->val);
return -1;
}
segment_prealloc = prealloc;
}
if (!quiet)
SCLogInfo("stream.reassembly \"segment-prealloc\": %u", segment_prealloc);
stream_config.prealloc_segments = segment_prealloc;
int overlap_diff_data = 0;
ConfGetBool("stream.reassembly.check-overlap-different-data", &overlap_diff_data);
if (overlap_diff_data) {
StreamTcpReassembleConfigEnableOverlapCheck();
}
if (StreamTcpInlineMode() == TRUE) {
StreamTcpReassembleConfigEnableOverlapCheck();
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
stream_config.sbcnf.flags = STREAMING_BUFFER_NOFLAGS;
stream_config.sbcnf.buf_size = 2048;
stream_config.sbcnf.Malloc = ReassembleMalloc;
stream_config.sbcnf.Calloc = ReassembleCalloc;
stream_config.sbcnf.Realloc = ReassembleRealloc;
stream_config.sbcnf.Free = ReassembleFree;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
return 0;
}
int StreamTcpReassembleInit(char quiet)
{
/* init the memcap/use tracker */
StreamTcpReassembleInitMemuse();
if (StreamTcpReassemblyConfig(quiet) < 0)
return -1;
#ifdef DEBUG
SCMutexInit(&segment_pool_memuse_mutex, NULL);
#endif
StatsRegisterGlobalCounter("tcp.reassembly_memuse",
StreamTcpReassembleMemuseGlobalCounter);
return 0;
}
void StreamTcpReassembleFree(char quiet)
{
SCMutexLock(&segment_thread_pool_mutex);
if (segment_thread_pool != NULL) {
PoolThreadFree(segment_thread_pool);
segment_thread_pool = NULL;
}
SCMutexUnlock(&segment_thread_pool_mutex);
SCMutexDestroy(&segment_thread_pool_mutex);
#ifdef DEBUG
SCLogInfo("segment_pool_memuse %"PRIu64"", segment_pool_memuse);
SCLogInfo("segment_pool_memcnt %"PRIu64"", segment_pool_memcnt);
SCMutexDestroy(&segment_pool_memuse_mutex);
#endif
}
TcpReassemblyThreadCtx *StreamTcpReassembleInitThreadCtx(ThreadVars *tv)
{
16 years ago
SCEnter();
TcpReassemblyThreadCtx *ra_ctx = SCMalloc(sizeof(TcpReassemblyThreadCtx));
if (unlikely(ra_ctx == NULL))
return NULL;
memset(ra_ctx, 0x00, sizeof(TcpReassemblyThreadCtx));
ra_ctx->app_tctx = AppLayerGetCtxThread(tv);
App layer API rewritten. The main files in question are: app-layer.[ch], app-layer-detect-proto.[ch] and app-layer-parser.[ch]. Things addressed in this commit: - Brings out a proper separation between protocol detection phase and the parser phase. - The dns app layer now is registered such that we don't use "dnstcp" and "dnsudp" in the rules. A user who previously wrote a rule like this - "alert dnstcp....." or "alert dnsudp....." would now have to use, alert dns (ipproto:tcp;) or alert udp (app-layer-protocol:dns;) or alert ip (ipproto:udp; app-layer-protocol:dns;) The same rules extend to other another such protocol, dcerpc. - The app layer parser api now takes in the ipproto while registering callbacks. - The app inspection/detection engine also takes an ipproto. - All app layer parser functions now take direction as STREAM_TOSERVER or STREAM_TOCLIENT, as opposed to 0 or 1, which was taken by some of the functions. - FlowInitialize() and FlowRecycle() now resets proto to 0. This is needed by unittests, which would try to clean the flow, and that would call the api, AppLayerParserCleanupParserState(), which would try to clean the app state, but the app layer now needs an ipproto to figure out which api to internally call to clean the state, and if the ipproto is 0, it would return without trying to clean the state. - A lot of unittests are now updated where if they are using a flow and they need to use the app layer, we would set a flow ipproto. - The "app-layer" section in the yaml conf has also been updated as well.
12 years ago
SCMutexLock(&segment_thread_pool_mutex);
if (segment_thread_pool == NULL) {
segment_thread_pool = PoolThreadInit(1, /* thread */
0, /* unlimited */
stream_config.prealloc_segments,
sizeof(TcpSegment),
TcpSegmentPoolAlloc,
TcpSegmentPoolInit, NULL,
TcpSegmentPoolCleanup, NULL);
ra_ctx->segment_thread_pool_id = 0;
SCLogDebug("pool size %d, thread segment_thread_pool_id %d",
PoolThreadSize(segment_thread_pool),
ra_ctx->segment_thread_pool_id);
} else {
/* grow segment_thread_pool until we have a element for our thread id */
ra_ctx->segment_thread_pool_id = PoolThreadGrow(segment_thread_pool,
0, /* unlimited */
stream_config.prealloc_segments,
sizeof(TcpSegment),
TcpSegmentPoolAlloc,
TcpSegmentPoolInit, NULL,
TcpSegmentPoolCleanup, NULL);
SCLogDebug("pool size %d, thread segment_thread_pool_id %d",
PoolThreadSize(segment_thread_pool),
ra_ctx->segment_thread_pool_id);
}
SCMutexUnlock(&segment_thread_pool_mutex);
if (ra_ctx->segment_thread_pool_id < 0 || segment_thread_pool == NULL)
abort();
16 years ago
SCReturnPtr(ra_ctx, "TcpReassemblyThreadCtx");
}
void StreamTcpReassembleFreeThreadCtx(TcpReassemblyThreadCtx *ra_ctx)
{
16 years ago
SCEnter();
App layer API rewritten. The main files in question are: app-layer.[ch], app-layer-detect-proto.[ch] and app-layer-parser.[ch]. Things addressed in this commit: - Brings out a proper separation between protocol detection phase and the parser phase. - The dns app layer now is registered such that we don't use "dnstcp" and "dnsudp" in the rules. A user who previously wrote a rule like this - "alert dnstcp....." or "alert dnsudp....." would now have to use, alert dns (ipproto:tcp;) or alert udp (app-layer-protocol:dns;) or alert ip (ipproto:udp; app-layer-protocol:dns;) The same rules extend to other another such protocol, dcerpc. - The app layer parser api now takes in the ipproto while registering callbacks. - The app inspection/detection engine also takes an ipproto. - All app layer parser functions now take direction as STREAM_TOSERVER or STREAM_TOCLIENT, as opposed to 0 or 1, which was taken by some of the functions. - FlowInitialize() and FlowRecycle() now resets proto to 0. This is needed by unittests, which would try to clean the flow, and that would call the api, AppLayerParserCleanupParserState(), which would try to clean the app state, but the app layer now needs an ipproto to figure out which api to internally call to clean the state, and if the ipproto is 0, it would return without trying to clean the state. - A lot of unittests are now updated where if they are using a flow and they need to use the app layer, we would set a flow ipproto. - The "app-layer" section in the yaml conf has also been updated as well.
12 years ago
AppLayerDestroyCtxThread(ra_ctx->app_tctx);
SCFree(ra_ctx);
16 years ago
SCReturn;
}
/**
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
* \brief check if stream in pkt direction has depth reached
*
* \param p packet with *LOCKED* flow
*
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
* \retval 1 stream has depth reached
* \retval 0 stream does not have depth reached
*/
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
int StreamTcpReassembleDepthReached(Packet *p)
{
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (p->flow != NULL && p->flow->protoctx != NULL) {
TcpSession *ssn = p->flow->protoctx;
TcpStream *stream;
if (p->flowflags & FLOW_PKT_TOSERVER) {
stream = &ssn->client;
} else {
stream = &ssn->server;
}
return (stream->flags & STREAMTCP_STREAM_FLAG_DEPTH_REACHED) ? 1 : 0;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
return 0;
}
/**
* \internal
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
* \brief Function to Check the reassembly depth valuer against the
* allowed max depth of the stream reassmbly for TCP streams.
*
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
* \param stream stream direction
* \param seq sequence number where "size" starts
* \param size size of the segment that is added
*
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
* \retval size Part of the size that fits in the depth, 0 if none
*/
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
static uint32_t StreamTcpReassembleCheckDepth(TcpSession *ssn, TcpStream *stream,
uint32_t seq, uint32_t size)
{
SCEnter();
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* if the configured depth value is 0, it means there is no limit on
reassembly depth. Otherwise carry on my boy ;) */
if (ssn->reassembly_depth == 0) {
SCReturnUInt(size);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* if the final flag is set, we're not accepting anymore */
if (stream->flags & STREAMTCP_STREAM_FLAG_DEPTH_REACHED) {
SCReturnUInt(0);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
uint64_t seg_depth;
if (SEQ_GT(stream->base_seq, seq)) {
if (SEQ_LEQ(seq+size, stream->base_seq)) {
SCLogDebug("segment entirely before base_seq, weird: base %u, seq %u, re %u",
stream->base_seq, seq, seq+size);
SCReturnUInt(0);
}
seg_depth = STREAM_BASE_OFFSET(stream) + size - (stream->base_seq - seq);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
} else {
seg_depth = STREAM_BASE_OFFSET(stream) + ((seq + size) - stream->base_seq);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* if the base_seq has moved passed the depth window we stop
* checking and just reject the rest of the packets including
* retransmissions. Saves us the hassle of dealing with sequence
* wraps as well */
SCLogDebug("seq + size %u, base %u, seg_depth %"PRIu64" limit %u", (seq + size),
stream->base_seq, seg_depth,
stream_config.reassembly_depth);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (seg_depth > (uint64_t)stream_config.reassembly_depth) {
SCLogDebug("STREAMTCP_STREAM_FLAG_DEPTH_REACHED");
stream->flags |= STREAMTCP_STREAM_FLAG_DEPTH_REACHED;
SCReturnUInt(0);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("NOT STREAMTCP_STREAM_FLAG_DEPTH_REACHED");
SCLogDebug("%"PRIu64" <= %u", seg_depth, stream_config.reassembly_depth);
#if 0
SCLogDebug("full depth not yet reached: %"PRIu64" <= %"PRIu32,
(stream->base_seq_offset + stream->base_seq + size),
(stream->isn + stream_config.reassembly_depth));
#endif
if (SEQ_GEQ(seq, stream->isn) && SEQ_LT(seq, (stream->isn + stream_config.reassembly_depth))) {
/* packet (partly?) fits the depth window */
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (SEQ_LEQ((seq + size),(stream->isn + 1 + ssn->reassembly_depth))) {
/* complete fit */
SCReturnUInt(size);
} else {
stream->flags |= STREAMTCP_STREAM_FLAG_DEPTH_REACHED;
/* partial fit, return only what fits */
uint32_t part = (stream->isn + 1 + ssn->reassembly_depth) - seq;
DEBUG_VALIDATE_BUG_ON(part > size);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (part > size)
part = size;
SCReturnUInt(part);
}
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCReturnUInt(0);
}
/**
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
* \brief Insert a packets TCP data into the stream reassembly engine.
*
* \retval 0 good segment, as far as we checked.
* \retval -1 badness, reason to drop in inline mode
*
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
* If the retval is 0 the segment is inserted correctly, or overlap is handled,
* or it wasn't added because of reassembly depth.
*
*/
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
int StreamTcpReassembleHandleSegmentHandleData(ThreadVars *tv, TcpReassemblyThreadCtx *ra_ctx,
TcpSession *ssn, TcpStream *stream, Packet *p)
{
SCEnter();
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (ssn->data_first_seen_dir == 0) {
if (PKT_IS_TOSERVER(p)) {
ssn->data_first_seen_dir = STREAM_TOSERVER;
} else {
ssn->data_first_seen_dir = STREAM_TOCLIENT;
}
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* If the OS policy is not set then set the OS policy for this stream */
if (stream->os_policy == 0) {
StreamTcpSetOSPolicy(stream, p);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if ((ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED) &&
(stream->flags & STREAMTCP_STREAM_FLAG_NEW_RAW_DISABLED)) {
SCLogDebug("ssn %p: both app and raw reassembly disabled, not reassembling", ssn);
SCReturnInt(0);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* If we have reached the defined depth for either of the stream, then stop
reassembling the TCP session */
uint32_t size = StreamTcpReassembleCheckDepth(ssn, stream, TCP_GET_SEQ(p), p->payload_len);
SCLogDebug("ssn %p: check depth returned %"PRIu32, ssn, size);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (stream->flags & STREAMTCP_STREAM_FLAG_DEPTH_REACHED) {
/* increment stream depth counter */
StatsIncr(tv, ra_ctx->counter_tcp_stream_depth);
}
if (size == 0) {
SCLogDebug("ssn %p: depth reached, not reassembling", ssn);
SCReturnInt(0);
}
DEBUG_VALIDATE_BUG_ON(size > p->payload_len);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (size > p->payload_len)
size = p->payload_len;
TcpSegment *seg = StreamTcpGetSegment(tv, ra_ctx);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (seg == NULL) {
SCLogDebug("segment_pool is empty");
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
StreamTcpSetEvent(p, STREAM_REASSEMBLY_NO_SEGMENT);
SCReturnInt(-1);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
TCP_SEG_LEN(seg) = size;
seg->seq = TCP_GET_SEQ(p);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* proto detection skipped, but now we do get data. Set event. */
if (stream->seg_list == NULL &&
stream->flags & STREAMTCP_STREAM_FLAG_APPPROTO_DETECTION_SKIPPED) {
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
AppLayerDecoderEventsSetEventRaw(&p->app_layer_events,
APPLAYER_PROTO_DETECTION_SKIPPED);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (StreamTcpReassembleInsertSegment(tv, ra_ctx, stream, seg, p, TCP_GET_SEQ(p), p->payload, p->payload_len) != 0) {
SCLogDebug("StreamTcpReassembleInsertSegment failed");
SCReturnInt(-1);
}
SCReturnInt(0);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
static uint8_t StreamGetAppLayerFlags(TcpSession *ssn, TcpStream *stream,
Packet *p, enum StreamUpdateDir dir)
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
{
uint8_t flag = 0;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (!(stream->flags & STREAMTCP_STREAM_FLAG_APPPROTO_DETECTION_COMPLETED)) {
flag |= STREAM_START;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (ssn->state == TCP_CLOSED) {
flag |= STREAM_EOF;
}
if (p->flags & PKT_PSEUDO_STREAM_END) {
flag |= STREAM_EOF;
}
if (dir == UPDATE_DIR_OPPOSING) {
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (p->flowflags & FLOW_PKT_TOSERVER) {
flag |= STREAM_TOCLIENT;
} else {
flag |= STREAM_TOSERVER;
}
} else {
if (p->flowflags & FLOW_PKT_TOSERVER) {
flag |= STREAM_TOSERVER;
} else {
flag |= STREAM_TOCLIENT;
}
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (stream->flags & STREAMTCP_STREAM_FLAG_DEPTH_REACHED) {
flag |= STREAM_DEPTH;
}
return flag;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/**
* \brief Check the minimum size limits for reassembly.
*
* \retval 0 don't reassemble yet
* \retval 1 do reassemble
*/
static int StreamTcpReassembleRawCheckLimit(const TcpSession *ssn,
const TcpStream *stream, const Packet *p)
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
{
SCEnter();
/* if any of these flags is set we always inspect immediately */
#define STREAMTCP_STREAM_FLAG_FLUSH_FLAGS \
( STREAMTCP_STREAM_FLAG_DEPTH_REACHED \
| STREAMTCP_STREAM_FLAG_TRIGGER_RAW \
| STREAMTCP_STREAM_FLAG_NEW_RAW_DISABLED)
if (stream->flags & STREAMTCP_STREAM_FLAG_FLUSH_FLAGS) {
if (stream->flags & STREAMTCP_STREAM_FLAG_DEPTH_REACHED) {
SCLogDebug("reassembling now as STREAMTCP_STREAM_FLAG_DEPTH_REACHED "
"is set, so not expecting any new data segments");
}
if (stream->flags & STREAMTCP_STREAM_FLAG_TRIGGER_RAW) {
SCLogDebug("reassembling now as STREAMTCP_STREAM_FLAG_TRIGGER_RAW is set");
}
if (stream->flags & STREAMTCP_STREAM_FLAG_NEW_RAW_DISABLED) {
SCLogDebug("reassembling now as STREAMTCP_STREAM_FLAG_NEW_RAW_DISABLED is set, "
"so no new segments will be considered");
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCReturnInt(1);
}
#undef STREAMTCP_STREAM_FLAG_FLUSH_FLAGS
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* some states mean we reassemble no matter how much data we have */
if (ssn->state > TCP_TIME_WAIT)
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCReturnInt(1);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (p->flags & PKT_PSEUDO_STREAM_END)
SCReturnInt(1);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* check if we have enough data to do raw reassembly */
if (PKT_IS_TOSERVER(p)) {
if (STREAM_LASTACK_GT_BASESEQ(stream)) {
uint32_t delta = stream->last_ack - stream->base_seq;
/* get max absolute offset */
uint64_t max_offset = STREAM_BASE_OFFSET(stream) + delta;
int64_t diff = max_offset - STREAM_RAW_PROGRESS(stream);
if ((int64_t)stream_config.reassembly_toserver_chunk_size <= diff) {
SCReturnInt(1);
} else {
SCLogDebug("toserver min chunk len not yet reached: "
"last_ack %"PRIu32", ra_raw_base_seq %"PRIu32", %"PRIu32" < "
"%"PRIu32"", stream->last_ack, stream->base_seq,
(stream->last_ack - stream->base_seq),
stream_config.reassembly_toserver_chunk_size);
SCReturnInt(0);
}
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
} else {
if (STREAM_LASTACK_GT_BASESEQ(stream)) {
uint32_t delta = stream->last_ack - stream->base_seq;
/* get max absolute offset */
uint64_t max_offset = STREAM_BASE_OFFSET(stream) + delta;
int64_t diff = max_offset - STREAM_RAW_PROGRESS(stream);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if ((int64_t)stream_config.reassembly_toclient_chunk_size <= diff) {
SCReturnInt(1);
} else {
SCLogDebug("toclient min chunk len not yet reached: "
"last_ack %"PRIu32", base_seq %"PRIu32", %"PRIu32" < "
"%"PRIu32"", stream->last_ack, stream->base_seq,
(stream->last_ack - stream->base_seq),
stream_config.reassembly_toclient_chunk_size);
SCReturnInt(0);
}
}
}
SCReturnInt(0);
}
/**
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
* \brief see what if any work the TCP session still needs
*/
int StreamNeedsReassembly(const TcpSession *ssn, uint8_t direction)
{
const TcpStream *stream = NULL;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
#ifdef DEBUG
char *dirstr = NULL;
#endif
if (direction == STREAM_TOSERVER) {
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
stream = &ssn->client;
#ifdef DEBUG
dirstr = "client";
#endif
} else {
stream = &ssn->server;
#ifdef DEBUG
dirstr = "server";
#endif
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
int use_app = 1;
int use_raw = 1;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if ((ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED) ||
(stream->flags & STREAMTCP_STREAM_FLAG_GAP))
{
// app is dead
use_app = 0;
}
if (stream->flags & STREAMTCP_STREAM_FLAG_DISABLE_RAW) {
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
// raw is dead
use_raw = 0;
}
uint64_t right_edge = STREAM_BASE_OFFSET(stream) + stream->sb.buf_offset;
SCLogDebug("%s: list %p app %"PRIu64" (use: %s), raw %"PRIu64" (use: %s). Stream right edge: %"PRIu64,
dirstr,
stream->seg_list,
STREAM_APP_PROGRESS(stream), use_app ? "yes" : "no",
STREAM_RAW_PROGRESS(stream), use_raw ? "yes" : "no",
right_edge);
if (use_raw) {
if (right_edge > STREAM_RAW_PROGRESS(stream)) {
SCLogDebug("%s: STREAM_HAS_UNPROCESSED_SEGMENTS_NEED_ONLY_DETECTION", dirstr);
return STREAM_HAS_UNPROCESSED_SEGMENTS_NEED_ONLY_DETECTION;
}
}
if (use_app) {
if (right_edge > STREAM_APP_PROGRESS(stream)) {
SCLogDebug("%s: STREAM_HAS_UNPROCESSED_SEGMENTS_NEED_ONLY_DETECTION", dirstr);
return STREAM_HAS_UNPROCESSED_SEGMENTS_NEED_ONLY_DETECTION;
}
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("%s: STREAM_HAS_UNPROCESSED_SEGMENTS_NONE", dirstr);
return STREAM_HAS_UNPROCESSED_SEGMENTS_NONE;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
#ifdef DEBUG
static uint64_t GetStreamSize(TcpStream *stream)
{
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (stream) {
uint64_t size = 0;
uint32_t cnt = 0;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
TcpSegment *seg = stream->seg_list;
while (seg) {
cnt++;
size += (uint64_t)TCP_SEG_LEN(seg);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
seg = seg->next;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("size %"PRIu64", cnt %"PRIu32, size, cnt);
return size;
}
return (uint64_t)0;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
static void GetSessionSize(TcpSession *ssn, Packet *p)
{
uint64_t size = 0;
if (ssn) {
size = GetStreamSize(&ssn->client);
size += GetStreamSize(&ssn->server);
//if (size > 900000)
// SCLogInfo("size %"PRIu64", packet %"PRIu64, size, p->pcap_cnt);
SCLogDebug("size %"PRIu64", packet %"PRIu64, size, p->pcap_cnt);
}
}
#endif
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/** \internal
*
* Get buffer, or first part of the buffer if data gaps exist.
*
* \brief get stream data from offset
* \param offset stream offset */
static void GetAppBuffer(TcpStream *stream, const uint8_t **data, uint32_t *data_len, uint64_t offset)
{
const uint8_t *mydata;
uint32_t mydata_len;
if (stream->sb.block_list == NULL) {
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("getting one blob");
StreamingBufferGetDataAtOffset(&stream->sb, &mydata, &mydata_len, offset);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
*data = mydata;
*data_len = mydata_len;
} else {
StreamingBufferBlock *blk = stream->sb.block_list;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (blk->offset > offset) {
SCLogDebug("gap, want data at offset %"PRIu64", got data at %"PRIu64,
offset, blk->offset);
*data = NULL;
*data_len = 0;
} else if (offset > blk->offset && offset <= (blk->offset + blk->len)) {
SCLogDebug("get data from offset %"PRIu64". SBB %"PRIu64"/%u",
offset, blk->offset, blk->len);
StreamingBufferSBBGetDataAtOffset(&stream->sb, blk, data, data_len, offset);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("data %p, data_len %u", *data, *data_len);
} else {
StreamingBufferSBBGetData(&stream->sb, blk, data, data_len);
}
}
}
static inline bool CheckGap(TcpStream *stream, Packet *p)
{
const uint64_t app_progress = STREAM_APP_PROGRESS(stream);
uint64_t last_ack_abs = STREAM_BASE_OFFSET(stream);
if (STREAM_LASTACK_GT_BASESEQ(stream)) {
/* get window of data that is acked */
uint32_t delta = stream->last_ack - stream->base_seq;
DEBUG_VALIDATE_BUG_ON(delta > 10000000ULL && delta > stream->window);
/* get max absolute offset */
last_ack_abs += delta;
// last_ack > app_progress, but not data. We won't get it either because of last_ack.
if (last_ack_abs > app_progress+1) {
// account for our too liberal ack acceptance & pseudo packet last_ack hackery
if (SEQ_LT(stream->next_seq, stream->last_ack)) {
if (SEQ_GT(stream->next_seq, stream->base_seq)) {
uint64_t next_seq_abs = STREAM_BASE_OFFSET(stream) + (stream->next_seq - stream->base_seq);
if (next_seq_abs > app_progress+1) {
/* fall through */
} else {
return false;
}
}
}
SCLogDebug("packet %u GAP! last_ack_abs %u > app_progress %u, at no data.", (uint)p->pcap_cnt, (uint)last_ack_abs, (uint)app_progress);
return true;
}
}
return false;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/** \internal
* \brief get stream buffer and update the app-layer
* \retval 0 success
*/
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
static int ReassembleUpdateAppLayer (ThreadVars *tv,
TcpReassemblyThreadCtx *ra_ctx,
TcpSession *ssn, TcpStream *stream,
Packet *p, enum StreamUpdateDir dir)
{
const uint64_t app_progress = STREAM_APP_PROGRESS(stream);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("app progress %"PRIu64, app_progress);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("last_ack %u, base_seq %u", stream->last_ack, stream->base_seq);
const uint8_t *mydata;
uint32_t mydata_len;
GetAppBuffer(stream, &mydata, &mydata_len, app_progress);
if (mydata == NULL || mydata_len == 0) {
if (CheckGap(stream, p)) {
/* send gap signal */
SCLogDebug("sending GAP to app-layer");
AppLayerHandleTCPData(tv, ra_ctx, p, p->flow, ssn, stream,
NULL, 0,
StreamGetAppLayerFlags(ssn, stream, p, dir)|STREAM_GAP);
AppLayerProfilingStore(ra_ctx->app_tctx, p);
/* set a GAP flag and make sure not bothering this stream anymore */
SCLogDebug("STREAMTCP_STREAM_FLAG_GAP set");
stream->flags |= STREAMTCP_STREAM_FLAG_GAP;
StreamTcpSetEvent(p, STREAM_REASSEMBLY_SEQ_GAP);
StatsIncr(tv, ra_ctx->counter_tcp_reass_gap);
}
return 0;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
//PrintRawDataFp(stdout, mydata, mydata_len);
SCLogDebug("stream %p data in buffer %p of len %u and offset %"PRIu64,
stream, &stream->sb, mydata_len, app_progress);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* get window of data that is acked */
if (StreamTcpInlineMode() == FALSE) {
if (p->flags & PKT_PSEUDO_STREAM_END) {
// fall through, we use all available data
} else {
uint64_t last_ack_abs = app_progress; /* absolute right edge of ack'd data */
if (STREAM_LASTACK_GT_BASESEQ(stream)) {
/* get window of data that is acked */
uint32_t delta = stream->last_ack - stream->base_seq;
DEBUG_VALIDATE_BUG_ON(delta > 10000000ULL && delta > stream->window);
/* get max absolute offset */
last_ack_abs += delta;
}
/* see if the buffer contains unack'd data as well */
if (app_progress + mydata_len > last_ack_abs) {
uint32_t check = mydata_len;
mydata_len = last_ack_abs - app_progress;
BUG_ON(mydata_len > check);
SCLogDebug("data len adjusted to %u to make sure only ACK'd "
"data is considered", mydata_len);
}
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* update the app-layer */
int r = AppLayerHandleTCPData(tv, ra_ctx, p, p->flow, ssn, stream,
(uint8_t *)mydata, mydata_len,
StreamGetAppLayerFlags(ssn, stream, p, dir));
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* see if we can update the progress */
if (r == 0 && mydata_len > 0 &&
StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(stream))
{
SCLogDebug("app progress %"PRIu64" increasing with data len %u to %"PRIu64,
app_progress, mydata_len, app_progress + mydata_len);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
stream->app_progress_rel += mydata_len;
SCLogDebug("app progress now %"PRIu64, STREAM_APP_PROGRESS(stream));
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
} else {
SCLogDebug("NOT UPDATED app progress still %"PRIu64, app_progress);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCReturnInt(0);
}
/**
* \brief Update the stream reassembly upon receiving a packet.
*
* For IDS mode, the stream is in the opposite direction of the packet,
* as the ACK-packet is ACK'ing the stream.
*
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
* One of the utilities call by this function AppLayerHandleTCPData(),
* has a feature where it will call this very same function for the
* stream opposing the stream it is called with. This shouldn't cause
* any issues, since processing of each stream is independent of the
* other stream.
*/
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
int StreamTcpReassembleAppLayer (ThreadVars *tv, TcpReassemblyThreadCtx *ra_ctx,
TcpSession *ssn, TcpStream *stream,
Packet *p, enum StreamUpdateDir dir)
{
SCEnter();
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* this function can be directly called by app layer protocol
* detection. */
if ((ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED) ||
(stream->flags & STREAMTCP_STREAM_FLAG_NOREASSEMBLY)) {
SCLogDebug("stream no reassembly flag set or app-layer disabled.");
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCReturnInt(0);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("stream->seg_list %p", stream->seg_list);
#ifdef DEBUG
PrintList(stream->seg_list);
GetSessionSize(ssn, p);
#endif
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* Check if we have a gap at the start of the stream. 2 conditions:
* 1. no segments, but last_ack moved fwd
* 2. segments, but clearly some missing: if last_ack is
* bigger than the list start and the list start is bigger than
* next_seq, we know we are missing data that has been ack'd. That
* won't get retransmitted, so it's a data gap.
* 3. check if next_seq is smaller than last_ack, indicating next_seq
* has fallen behind the data that is already acked.
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
*/
{
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
int ackadd = (ssn->state >= TCP_FIN_WAIT2) ? 2 : 1;
if ((stream->seg_list == NULL && /*1*/
stream->base_seq == stream->isn+1 &&
SEQ_GT(stream->last_ack, stream->isn + ackadd))
||
(stream->seg_list != NULL &&
( /*2*/
(SEQ_GT(stream->seg_list->seq, stream->base_seq) &&
SEQ_LT(stream->seg_list->seq, stream->last_ack))
||
(SEQ_GT(stream->seg_list->seq, stream->base_seq) &&
SEQ_LT(stream->next_seq, stream->last_ack))
)
)
)
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
{
if (stream->seg_list == NULL) {
SCLogDebug("no segs, last_ack moved fwd so GAP "
"(base %u, isn %u, last_ack %u => diff %u) p %"PRIu64,
stream->base_seq, stream->isn, stream->last_ack,
stream->last_ack - (stream->isn + ackadd), p->pcap_cnt);
}
/* send gap signal */
SCLogDebug("sending GAP to app-layer");
AppLayerHandleTCPData(tv, ra_ctx, p, p->flow, ssn, stream,
NULL, 0,
StreamGetAppLayerFlags(ssn, stream, p, dir)|STREAM_GAP);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
AppLayerProfilingStore(ra_ctx->app_tctx, p);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* set a GAP flag and make sure not bothering this stream anymore */
SCLogDebug("STREAMTCP_STREAM_FLAG_GAP set");
stream->flags |= STREAMTCP_STREAM_FLAG_GAP;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
StreamTcpSetEvent(p, STREAM_REASSEMBLY_SEQ_GAP);
StatsIncr(tv, ra_ctx->counter_tcp_reass_gap);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCReturnInt(0);
}
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* if no segments are in the list or all are already processed,
* and state is beyond established, we send an empty msg */
TcpSegment *seg_tail = stream->seg_list_tail;
if (seg_tail == NULL ||
SEGMENT_BEFORE_OFFSET(stream, seg_tail, STREAM_APP_PROGRESS(stream)))
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
{
/* send an empty EOF msg if we have no segments but TCP state
* is beyond ESTABLISHED */
if (ssn->state >= TCP_CLOSING || (p->flags & PKT_PSEUDO_STREAM_END)) {
SCLogDebug("sending empty eof message");
/* send EOF to app layer */
AppLayerHandleTCPData(tv, ra_ctx, p, p->flow, ssn, stream,
NULL, 0,
StreamGetAppLayerFlags(ssn, stream, p, dir));
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
AppLayerProfilingStore(ra_ctx->app_tctx, p);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCReturnInt(0);
}
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* no segments, nothing to do */
if (stream->seg_list == NULL) {
SCLogDebug("no segments in the list to reassemble");
SCReturnInt(0);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (stream->flags & STREAMTCP_STREAM_FLAG_GAP) {
SCReturnInt(0);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* with all that out of the way, lets update the app-layer */
return ReassembleUpdateAppLayer(tv, ra_ctx, ssn, stream, p, dir);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
}
/** \internal
* \brief get stream data from offset
* \param offset stream offset */
static int GetRawBuffer(TcpStream *stream, const uint8_t **data, uint32_t *data_len,
StreamingBufferBlock **iter, uint64_t offset, uint64_t *data_offset)
{
const uint8_t *mydata;
uint32_t mydata_len;
if (stream->sb.block_list == NULL) {
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("getting one blob");
uint64_t roffset = offset;
if (offset)
StreamingBufferGetDataAtOffset(&stream->sb, &mydata, &mydata_len, offset);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
else {
StreamingBufferGetData(&stream->sb, &mydata, &mydata_len, &roffset);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
*data = mydata;
*data_len = mydata_len;
*data_offset = roffset;
} else {
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (*iter == NULL)
*iter = stream->sb.block_list;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (*iter == NULL) {
*data = NULL;
*data_len = 0;
*data_offset = 0;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
return 0;
}
if (offset) {
while (*iter && ((*iter)->offset + (*iter)->len < offset))
*iter = (*iter)->next;
if (*iter == NULL) {
*data = NULL;
*data_len = 0;
*data_offset = 0;
return 0;
}
}
SCLogDebug("getting multiple blobs. Iter %p, %"PRIu64"/%u (next? %s)", *iter, (*iter)->offset, (*iter)->len, (*iter)->next ? "yes":"no");
StreamingBufferSBBGetData(&stream->sb, (*iter), &mydata, &mydata_len);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if ((*iter)->offset < offset) {
uint64_t delta = offset - (*iter)->offset;
if (delta < mydata_len) {
*data = mydata + delta;
*data_len = mydata_len - delta;
*data_offset = offset;
} else {
*data = NULL;
*data_len = 0;
*data_offset = 0;
}
} else {
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
*data = mydata;
*data_len = mydata_len;
*data_offset = (*iter)->offset;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
*iter = (*iter)->next;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
return 0;
}
/** \brief does the stream engine have data to inspect?
*
* Returns true if there is data to inspect. In IDS case this is
* about ACK'd data in the packet's direction.
*
* In the IPS case this is about the packet itself.
*/
bool StreamReassembleRawHasDataReady(TcpSession *ssn, Packet *p)
{
TcpStream *stream;
if (PKT_IS_TOSERVER(p)) {
stream = &ssn->client;
} else {
stream = &ssn->server;
}
if (stream->seg_list == NULL) {
return false;
}
if (stream->flags & (STREAMTCP_STREAM_FLAG_NOREASSEMBLY|
STREAMTCP_STREAM_FLAG_DISABLE_RAW))
return false;
if (StreamTcpInlineMode() == FALSE) {
if ((STREAM_RAW_PROGRESS(stream) == STREAM_BASE_OFFSET(stream) + stream->sb.buf_offset)) {
return false;
}
if (StreamTcpReassembleRawCheckLimit(ssn, stream, p) == 1) {
return true;
}
} else {
if (p->payload_len > 0 && (p->flags & PKT_STREAM_ADD)) {
return true;
}
}
return false;
}
/** \brief update stream engine after detection
*
* Tasked with progressing the 'progress' for Raw reassembly.
* 2 main scenario's:
* 1. progress is != 0, so we use this
* 2. progress is 0, meaning the detect engine didn't touch
* raw at all. In this case we need to look into progressing
* raw anyway.
*
* Additionally, this function is tasked with disabling raw
* reassembly if the app-layer requested to disable it.
*/
void StreamReassembleRawUpdateProgress(TcpSession *ssn, Packet *p, uint64_t progress)
{
TcpStream *stream;
if (PKT_IS_TOSERVER(p)) {
stream = &ssn->client;
} else {
stream = &ssn->server;
}
if (progress > STREAM_RAW_PROGRESS(stream)) {
uint32_t slide = progress - STREAM_RAW_PROGRESS(stream);
stream->raw_progress_rel += slide;
stream->flags &= ~STREAMTCP_STREAM_FLAG_TRIGGER_RAW;
/* if app is active and beyond raw, sync raw to app */
} else if (progress == 0 &&
STREAM_APP_PROGRESS(stream) > STREAM_RAW_PROGRESS(stream) &&
!(ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED) &&
!(stream->flags & STREAMTCP_STREAM_FLAG_GAP))
{
/* if trigger raw is set we sync the 2 trackers */
if (stream->flags & STREAMTCP_STREAM_FLAG_TRIGGER_RAW)
{
uint32_t slide = STREAM_APP_PROGRESS(stream) - STREAM_RAW_PROGRESS(stream);
stream->raw_progress_rel += slide;
stream->flags &= ~STREAMTCP_STREAM_FLAG_TRIGGER_RAW;
/* otherwise mix in the tcp window */
} else {
uint64_t tcp_window = stream->window;
if (tcp_window > 0 && STREAM_APP_PROGRESS(stream) > tcp_window) {
uint64_t new_raw = STREAM_APP_PROGRESS(stream) - tcp_window;
if (new_raw > STREAM_RAW_PROGRESS(stream)) {
uint32_t slide = new_raw - STREAM_RAW_PROGRESS(stream);
stream->raw_progress_rel += slide;
}
}
}
/* app is dead */
} else if (progress == 0) {
uint64_t tcp_window = stream->window;
uint64_t stream_right_edge = STREAM_BASE_OFFSET(stream) + stream->sb.buf_offset;
if (tcp_window < stream_right_edge) {
uint64_t new_raw = stream_right_edge - tcp_window;
if (new_raw > STREAM_RAW_PROGRESS(stream)) {
uint32_t slide = new_raw - STREAM_RAW_PROGRESS(stream);
stream->raw_progress_rel += slide;
}
}
stream->flags &= ~STREAMTCP_STREAM_FLAG_TRIGGER_RAW;
} else {
SCLogDebug("p->pcap_cnt %u: progress %u app %u raw %u tcp win %u",
(uint)p->pcap_cnt, (uint)progress, (uint)STREAM_APP_PROGRESS(stream),
(uint)STREAM_RAW_PROGRESS(stream), (uint)stream->window);
}
/* if we were told to accept no more raw data, we can mark raw as
* disabled now. */
if (stream->flags & STREAMTCP_STREAM_FLAG_NEW_RAW_DISABLED) {
stream->flags |= STREAMTCP_STREAM_FLAG_DISABLE_RAW;
SCLogDebug("ssn %p: STREAMTCP_STREAM_FLAG_NEW_RAW_DISABLED set, "
"now that detect ran also set STREAMTCP_STREAM_FLAG_DISABLE_RAW", ssn);
}
SCLogDebug("stream raw progress now %"PRIu64, STREAM_RAW_PROGRESS(stream));
}
/** \internal
* \brief get a buffer around the current packet and run the callback on it
*
* The inline/IPS scanning method takes the current payload and wraps it in
* data from other segments.
*
* How much data is inspected is controlled by the available data, chunk_size
* and the payload size of the packet.
*
* Large packets: if payload size is close to the chunk_size, where close is
* defined as more than 67% of the chunk_size, a larger chunk_size will be
* used: payload_len + 33% of the chunk_size.
* If the payload size if equal to or bigger than the chunk_size, we use
* payload len + 33% of the chunk size.
*/
static int StreamReassembleRawInline(TcpSession *ssn, const Packet *p,
StreamReassembleRawFunc Callback, void *cb_data, uint64_t *progress_out)
{
SCEnter();
int r = 0;
TcpStream *stream;
if (PKT_IS_TOSERVER(p)) {
stream = &ssn->client;
} else {
stream = &ssn->server;
}
if (p->payload_len == 0 || (p->flags & PKT_STREAM_ADD) == 0 ||
(stream->flags & STREAMTCP_STREAM_FLAG_NOREASSEMBLY))
{
*progress_out = STREAM_RAW_PROGRESS(stream);
return 0;
}
uint32_t chunk_size = PKT_IS_TOSERVER(p) ?
stream_config.reassembly_toserver_chunk_size :
stream_config.reassembly_toclient_chunk_size;
if (chunk_size <= p->payload_len) {
chunk_size = p->payload_len + (chunk_size / 3);
SCLogDebug("packet payload len %u, so chunk_size adjusted to %u",
p->payload_len, chunk_size);
} else if (((chunk_size / 3 ) * 2) < p->payload_len) {
chunk_size = p->payload_len + ((chunk_size / 3));
SCLogDebug("packet payload len %u, so chunk_size adjusted to %u",
p->payload_len, chunk_size);
}
uint64_t packet_leftedge_abs = STREAM_BASE_OFFSET(stream) + (TCP_GET_SEQ(p) - stream->base_seq);
uint64_t packet_rightedge_abs = packet_leftedge_abs + p->payload_len;
SCLogDebug("packet_leftedge_abs %"PRIu64", rightedge %"PRIu64,
packet_leftedge_abs, packet_rightedge_abs);
const uint8_t *mydata = NULL;
uint32_t mydata_len = 0;
uint64_t mydata_offset = 0;
/* simply return progress from the block we inspected. */
bool return_progress = false;
if (stream->sb.block_list == NULL) {
/* continues block */
StreamingBufferGetData(&stream->sb, &mydata, &mydata_len, &mydata_offset);
return_progress = true;
} else {
/* find our block */
StreamingBufferBlock *iter = stream->sb.block_list;
for ( ; iter != NULL; iter = iter->next) {
uint64_t iter_re_abs = iter->offset + iter->len;
DEBUG_VALIDATE_BUG_ON(packet_leftedge_abs < iter->offset &&
packet_rightedge_abs > iter->offset &&
packet_rightedge_abs < iter_re_abs);
if (iter->offset <= packet_leftedge_abs && iter_re_abs >= packet_rightedge_abs) {
StreamingBufferSBBGetData(&stream->sb, iter, &mydata, &mydata_len);
mydata_offset = iter->offset;
break;
}
}
}
/* this can only happen if the segment insert of our current 'p' failed */
if (mydata == NULL || mydata_len == 0) {
SCLogDebug("no data: %p/%u", mydata, mydata_len);
*progress_out = STREAM_RAW_PROGRESS(stream);
return 0;
}
if (mydata_offset >= packet_rightedge_abs ||
(packet_leftedge_abs >= (mydata_offset + mydata_len))) {
SCLogDebug("no data overlap: %p/%u", mydata, mydata_len);
*progress_out = STREAM_RAW_PROGRESS(stream);
return 0;
}
/* adjust buffer to match chunk_size */
uint64_t mydata_rightedge_abs = mydata_offset + mydata_len;
SCLogDebug("chunk_size %u mydata_len %u", chunk_size, mydata_len);
if (mydata_len > chunk_size) {
uint32_t excess = mydata_len - chunk_size;
SCLogDebug("chunk_size %u mydata_len %u excess %u", chunk_size, mydata_len, excess);
if (mydata_rightedge_abs == packet_rightedge_abs) {
mydata += excess;
mydata_len -= excess;
mydata_offset += excess;
SCLogDebug("cutting front of the buffer with %u", excess);
} else if (mydata_offset == packet_leftedge_abs) {
mydata_len -= excess;
SCLogDebug("cutting tail of the buffer with %u", excess);
} else {
uint32_t before = (uint32_t)(packet_leftedge_abs - mydata_offset);
uint32_t after = (uint32_t)(mydata_rightedge_abs - packet_rightedge_abs);
SCLogDebug("before %u after %u", before, after);
if (after >= (chunk_size - p->payload_len) / 2) {
// more trailing data than we need
if (before >= (chunk_size - p->payload_len) / 2) {
// also more heading data, devide evenly
before = after = (chunk_size - p->payload_len) / 2;
} else {
// heading data is less than requested, give the
// rest to the trailing data
after = (chunk_size - p->payload_len) - before;
}
} else {
// less trailing data than requested
if (before >= (chunk_size - p->payload_len) / 2) {
before = (chunk_size - p->payload_len) - after;
} else {
// both smaller than their requested size
}
}
/* adjust the buffer */
uint32_t skip = (uint32_t)(packet_leftedge_abs - mydata_offset) - before;
uint32_t cut = (uint32_t)(mydata_rightedge_abs - packet_rightedge_abs) - after;
DEBUG_VALIDATE_BUG_ON(skip > mydata_len);
DEBUG_VALIDATE_BUG_ON(cut > mydata_len);
DEBUG_VALIDATE_BUG_ON(skip + cut > mydata_len);
mydata += skip;
mydata_len -= (skip + cut);
mydata_offset += skip;
}
}
/* run the callback */
r = Callback(cb_data, mydata, mydata_len);
BUG_ON(r < 0);
if (return_progress) {
*progress_out = (mydata_offset + mydata_len);
} else {
/* several blocks of data, so we need to be a bit more careful:
* - if last_ack is beyond last progress, move progress forward to last_ack
* - if our block matches or starts before last ack, return right edge of
* our block.
*/
uint64_t last_ack_abs = STREAM_BASE_OFFSET(stream);
if (STREAM_LASTACK_GT_BASESEQ(stream)) {
uint32_t delta = stream->last_ack - stream->base_seq;
DEBUG_VALIDATE_BUG_ON(delta > 10000000ULL && delta > stream->window);
/* get max absolute offset */
last_ack_abs += delta;
}
SCLogDebug("last_ack_abs %"PRIu64, last_ack_abs);
if (STREAM_RAW_PROGRESS(stream) < last_ack_abs) {
if (mydata_offset > last_ack_abs) {
/* gap between us and last ack, set progress to last ack */
*progress_out = last_ack_abs;
} else {
*progress_out = (mydata_offset + mydata_len);
}
} else {
*progress_out = STREAM_RAW_PROGRESS(stream);
}
}
return r;
}
/** \brief access 'raw' reassembly data.
*
* Access data as tracked by 'raw' tracker. Data is made available to
* callback that is passed to this function.
*
* In the case of IDS the callback may be run multiple times if data
* contains gaps. It will then be run for each block of data that is
* continuous.
*
* The callback should give on of 2 return values:
* - 0 ok
* - 1 done
* The value 1 will break the loop if there is a block list that is
* inspected.
*
* This function will return the 'progress' value that has been
* consumed until now.
*
* \param ssn tcp session
* \param p packet
* \param Callback the function pointer to the callback function
* \param cb_data callback data
* \param[out] progress_out absolute progress value of the data this
* call handled.
*/
int StreamReassembleRaw(TcpSession *ssn, const Packet *p,
StreamReassembleRawFunc Callback, void *cb_data,
uint64_t *progress_out)
{
SCEnter();
int r = 0;
/* handle inline seperately as the logic is very different */
if (StreamTcpInlineMode() == TRUE) {
return StreamReassembleRawInline(ssn, p, Callback, cb_data, progress_out);
}
TcpStream *stream;
if (PKT_IS_TOSERVER(p)) {
stream = &ssn->client;
} else {
stream = &ssn->server;
}
if ((stream->flags & (STREAMTCP_STREAM_FLAG_NOREASSEMBLY|STREAMTCP_STREAM_FLAG_DISABLE_RAW)) ||
StreamTcpReassembleRawCheckLimit(ssn, stream, p) == 0)
{
*progress_out = STREAM_RAW_PROGRESS(stream);
return 0;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
StreamingBufferBlock *iter = NULL;
uint64_t progress = STREAM_RAW_PROGRESS(stream);
uint64_t last_ack_abs = STREAM_BASE_OFFSET(stream); /* absolute right edge of ack'd data */
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* get window of data that is acked */
if (STREAM_LASTACK_GT_BASESEQ(stream)) {
SCLogDebug("last_ack %u, base_seq %u", stream->last_ack, stream->base_seq);
uint32_t delta = stream->last_ack - stream->base_seq;
DEBUG_VALIDATE_BUG_ON(delta > 10000000ULL && delta > stream->window);
/* get max absolute offset */
last_ack_abs += delta;
SCLogDebug("last_ack_abs %"PRIu64, last_ack_abs);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* loop through available buffers. On no packet loss we'll have a single
* iteration. On missing data we'll walk the blocks */
while (1) {
const uint8_t *mydata;
uint32_t mydata_len;
uint64_t mydata_offset = 0;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
GetRawBuffer(stream, &mydata, &mydata_len, &iter, progress, &mydata_offset);
if (mydata_len == 0)
break;
//PrintRawDataFp(stdout, mydata, mydata_len);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("raw progress %"PRIu64, progress);
SCLogDebug("stream %p data in buffer %p of len %u and offset %u",
stream, &stream->sb, mydata_len, (uint)progress);
if (p->flags & PKT_PSEUDO_STREAM_END) {
// inspect all remaining data, ack'd or not
} else {
if (last_ack_abs < progress) {
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("nothing to do");
goto end;
}
SCLogDebug("last_ack_abs %"PRIu64", raw_progress %"PRIu64, last_ack_abs, progress);
SCLogDebug("raw_progress + mydata_len %"PRIu64", last_ack_abs %"PRIu64, progress + mydata_len, last_ack_abs);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* see if the buffer contains unack'd data as well */
if (progress + mydata_len > last_ack_abs) {
uint32_t check = mydata_len;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
mydata_len = last_ack_abs - progress;
BUG_ON(check < mydata_len);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("data len adjusted to %u to make sure only ACK'd "
"data is considered", mydata_len);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
}
if (mydata_len == 0)
break;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("data %p len %u", mydata, mydata_len);
/* we have data. */
r = Callback(cb_data, mydata, mydata_len);
BUG_ON(r < 0);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (mydata_offset == progress) {
SCLogDebug("raw progress %"PRIu64" increasing with data len %u to %"PRIu64,
progress, mydata_len, STREAM_RAW_PROGRESS(stream) + mydata_len);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
progress += mydata_len;
SCLogDebug("raw progress now %"PRIu64, progress);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* data is beyond the progress we'd like, and before last ack. Gap. */
} else if (mydata_offset > progress && mydata_offset < last_ack_abs) {
SCLogDebug("GAP: data is missing from %"PRIu64" (%u bytes), setting to first data we have: %"PRIu64, progress, (uint32_t)(mydata_offset - progress), mydata_offset);
SCLogDebug("last_ack_abs %"PRIu64, last_ack_abs);
progress = mydata_offset;
SCLogDebug("raw progress now %"PRIu64, progress);
} else {
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("not increasing progress, data gap => mydata_offset "
"%"PRIu64" != progress %"PRIu64, mydata_offset, progress);
}
if (iter == NULL || r == 1)
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
break;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
end:
*progress_out = progress;
return r;
}
/** \internal
* \brief update app layer based on received ACK
*
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
* \retval r 0 on success, -1 on error
*/
static int StreamTcpReassembleHandleSegmentUpdateACK (ThreadVars *tv,
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
TcpReassemblyThreadCtx *ra_ctx, TcpSession *ssn, TcpStream *stream, Packet *p)
{
SCEnter();
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("stream->seg_list %p", stream->seg_list);
int r = 0;
if (StreamTcpReassembleAppLayer(tv, ra_ctx, ssn, stream, p, UPDATE_DIR_OPPOSING) < 0)
r = -1;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("stream->seg_list %p", stream->seg_list);
SCReturnInt(r);
}
int StreamTcpReassembleHandleSegment(ThreadVars *tv, TcpReassemblyThreadCtx *ra_ctx,
TcpSession *ssn, TcpStream *stream,
Packet *p, PacketQueue *pq)
{
SCEnter();
SCLogDebug("ssn %p, stream %p, p %p, p->payload_len %"PRIu16"",
ssn, stream, p, p->payload_len);
/* we need to update the opposing stream in
* StreamTcpReassembleHandleSegmentUpdateACK */
TcpStream *opposing_stream = NULL;
if (stream == &ssn->client) {
opposing_stream = &ssn->server;
} else {
opposing_stream = &ssn->client;
}
/* default IDS: update opposing side (triggered by ACK) */
enum StreamUpdateDir dir = UPDATE_DIR_OPPOSING;
/* inline and stream end and flow timeout packets trigger same dir handling */
if (StreamTcpInlineMode()) {
dir = UPDATE_DIR_PACKET;
} else if (p->flags & PKT_PSEUDO_STREAM_END) {
dir = UPDATE_DIR_PACKET;
} else if (p->tcph && (p->tcph->th_flags & TH_RST)) { // accepted rst
dir = UPDATE_DIR_PACKET;
} else if (p->tcph && (p->tcph->th_flags & TH_FIN) && ssn->state > TCP_TIME_WAIT) {
dir = UPDATE_DIR_PACKET;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* handle ack received */
if (dir == UPDATE_DIR_OPPOSING &&
StreamTcpReassembleHandleSegmentUpdateACK(tv, ra_ctx, ssn, opposing_stream, p) != 0)
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
{
SCLogDebug("StreamTcpReassembleHandleSegmentUpdateACK error");
SCReturnInt(-1);
}
/* if this segment contains data, insert it */
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (p->payload_len > 0 && !(stream->flags & STREAMTCP_STREAM_FLAG_NOREASSEMBLY)) {
SCLogDebug("calling StreamTcpReassembleHandleSegmentHandleData");
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (StreamTcpReassembleHandleSegmentHandleData(tv, ra_ctx, ssn, stream, p) != 0) {
SCLogDebug("StreamTcpReassembleHandleSegmentHandleData error");
SCReturnInt(-1);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
p->flags |= PKT_STREAM_ADD;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* in stream inline mode even if we have no data we call the reassembly
* functions to handle EOF */
if (dir == UPDATE_DIR_PACKET) {
SCLogDebug("inline (%s) or PKT_PSEUDO_STREAM_END (%s)",
StreamTcpInlineMode()?"true":"false",
(p->flags & PKT_PSEUDO_STREAM_END) ?"true":"false");
if (StreamTcpReassembleAppLayer(tv, ra_ctx, ssn, stream, p, dir) < 0) {
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCReturnInt(-1);
}
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCReturnInt(0);
}
/**
* \brief get a segment from the pool
*
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
* \retval seg Segment from the pool or NULL
*/
TcpSegment *StreamTcpGetSegment(ThreadVars *tv, TcpReassemblyThreadCtx *ra_ctx)
{
TcpSegment *seg = (TcpSegment *) PoolThreadGetById(segment_thread_pool, ra_ctx->segment_thread_pool_id);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("seg we return is %p", seg);
if (seg == NULL) {
/* Increment the counter to show that we are not able to serve the
segment request due to memcap limit */
StatsIncr(tv, ra_ctx->counter_tcp_segment_memcap);
} else {
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
seg->next = NULL;
seg->prev = NULL;
memset(&seg->sbseg, 0, sizeof(seg->sbseg));
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
return seg;
}
/**
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
* \brief Trigger RAW stream reassembly
*
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
* Used by AppLayerTriggerRawStreamReassembly to trigger RAW stream
* reassembly from the applayer, for example upon completion of a
* HTTP request.
*
* It sets a flag in the stream so that the next Raw call will return
* the data.
*
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
* \param ssn TcpSession
*/
void StreamTcpReassembleTriggerRawReassembly(TcpSession *ssn, int direction)
{
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
#ifdef DEBUG
BUG_ON(ssn == NULL);
#endif
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (ssn != NULL) {
if (direction == STREAM_TOSERVER) {
ssn->client.flags |= STREAMTCP_STREAM_FLAG_TRIGGER_RAW;
} else {
ssn->server.flags |= STREAMTCP_STREAM_FLAG_TRIGGER_RAW;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("flagged ssn %p for immediate raw reassembly", ssn);
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
#ifdef UNITTESTS
/** unit tests and it's support functions below */
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
#define SET_ISN(stream, setseq) \
(stream)->isn = (setseq); \
(stream)->base_seq = (setseq) + 1
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/** \brief The Function to create the packet with given payload, which is used
* to test the reassembly of the engine.
*
* \param payload The variable used to store the payload contents of the
* current packet.
* \param value The value which current payload will have for this packet
* \param payload_len The length of the filed payload for current packet.
* \param len Length of the payload array
*/
void StreamTcpCreateTestPacket(uint8_t *payload, uint8_t value,
uint8_t payload_len, uint8_t len)
{
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
uint8_t i;
for (i = 0; i < payload_len; i++)
payload[i] = value;
for (; i < len; i++)
payload = NULL;
}
/** \brief The Function Checks the reassembled stream contents against predefined
* stream contents according to OS policy used.
*
* \param stream_policy Predefined value of stream for different OS policies
* \param stream Reassembled stream returned from the reassembly functions
*/
int StreamTcpCheckStreamContents(uint8_t *stream_policy, uint16_t sp_size, TcpStream *stream)
{
if (StreamingBufferCompareRawData(&stream->sb, stream_policy,(uint32_t)sp_size) == 0)
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
{
//PrintRawDataFp(stdout, stream_policy, sp_size);
return 0;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
}
return 1;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
static int VALIDATE(TcpStream *stream, uint8_t *data, uint32_t data_len)
{
if (StreamingBufferCompareRawData(&stream->sb,
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
data, data_len) == 0)
{
SCReturnInt(0);
}
SCLogInfo("OK");
PrintRawDataFp(stdout, data, data_len);
return 1;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
#define MISSED_START(isn) \
TcpReassemblyThreadCtx *ra_ctx = NULL; \
TcpSession ssn; \
ThreadVars tv; \
memset(&tv, 0, sizeof(tv)); \
\
StreamTcpUTInit(&ra_ctx); \
\
StreamTcpUTSetupSession(&ssn); \
StreamTcpUTSetupStream(&ssn.server, (isn)); \
StreamTcpUTSetupStream(&ssn.client, (isn)); \
\
TcpStream *stream = &ssn.client;
#define MISSED_END \
StreamTcpUTClearSession(&ssn); \
StreamTcpUTDeinit(ra_ctx); \
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
PASS
#define MISSED_STEP(seq, seg, seglen, buf, buflen) \
StreamTcpUTAddPayload(&tv, ra_ctx, &ssn, stream, (seq), (uint8_t *)(seg), (seglen)); \
FAIL_IF(!(VALIDATE(stream, (uint8_t *)(buf), (buflen))));
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/**
* \test Test the handling of packets missed by both IDS and the end host.
* The packet is missed in the starting of the stream.
*
* \retval On success it returns 1 and on failure 0.
*/
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
static int StreamTcpReassembleTest25 (void)
{
MISSED_START(6);
MISSED_STEP(10, "BB", 2, "\0\0\0BB", 5);
MISSED_STEP(12, "CC", 2, "\0\0\0BBCC", 7);
MISSED_STEP(7, "AAA", 3, "AAABBCC", 7);
MISSED_END;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/**
* \test Test the handling of packets missed by both IDS and the end host.
* The packet is missed in the middle of the stream.
*
* \retval On success it returns 1 and on failure 0.
*/
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
static int StreamTcpReassembleTest26 (void)
{
MISSED_START(9);
MISSED_STEP(10, "AAA", 3, "AAA", 3);
MISSED_STEP(15, "CC", 2, "AAA\0\0CC", 7);
MISSED_STEP(13, "BB", 2, "AAABBCC", 7);
MISSED_END;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/**
* \test Test the handling of packets missed by both IDS and the end host.
* The packet is missed in the end of the stream.
*
* \retval On success it returns 1 and on failure 0.
*/
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
static int StreamTcpReassembleTest27 (void)
{
MISSED_START(9);
MISSED_STEP(10, "AAA", 3, "AAA", 3);
MISSED_STEP(13, "BB", 2, "AAABB", 5);
MISSED_STEP(15, "CC", 2, "AAABBCC", 7);
MISSED_END;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/**
* \test Test the handling of packets missed by IDS, but the end host has
* received it and send the acknowledgment of it. The packet is missed
* in the starting of the stream.
*
* \retval On success it returns 1 and on failure 0.
*/
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
static int StreamTcpReassembleTest28 (void)
{
MISSED_START(6);
MISSED_STEP(10, "AAA", 3, "\0\0\0AAA", 6);
MISSED_STEP(13, "BB", 2, "\0\0\0AAABB", 8);
ssn.state = TCP_TIME_WAIT;
MISSED_STEP(15, "CC", 2, "\0\0\0AAABBCC", 10);
MISSED_END;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/**
* \test Test the handling of packets missed by IDS, but the end host has
* received it and send the acknowledgment of it. The packet is missed
* in the middle of the stream.
*
* \retval On success it returns 1 and on failure 0.
*/
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
static int StreamTcpReassembleTest29 (void)
{
MISSED_START(9);
MISSED_STEP(10, "AAA", 3, "AAA", 3);
ssn.state = TCP_TIME_WAIT;
MISSED_STEP(15, "CC", 2, "AAA\0\0CC", 7);
MISSED_END;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
static int StreamTcpReassembleTest33(void)
{
TcpSession ssn;
Packet *p = PacketGetFromAlloc();
FAIL_IF(unlikely(p == NULL));
Flow f;
TCPHdr tcph;
TcpReassemblyThreadCtx *ra_ctx = NULL;
ssn.client.os_policy = OS_POLICY_BSD;
uint8_t packet[1460] = "";
StreamTcpUTInit(&ra_ctx);
StreamTcpUTSetupSession(&ssn);
PacketQueue pq;
memset(&pq,0,sizeof(PacketQueue));
memset(&f, 0, sizeof (Flow));
memset(&tcph, 0, sizeof (TCPHdr));
ThreadVars tv;
memset(&tv, 0, sizeof (ThreadVars));
FLOW_INITIALIZE(&f);
f.protoctx = &ssn;
App layer API rewritten. The main files in question are: app-layer.[ch], app-layer-detect-proto.[ch] and app-layer-parser.[ch]. Things addressed in this commit: - Brings out a proper separation between protocol detection phase and the parser phase. - The dns app layer now is registered such that we don't use "dnstcp" and "dnsudp" in the rules. A user who previously wrote a rule like this - "alert dnstcp....." or "alert dnsudp....." would now have to use, alert dns (ipproto:tcp;) or alert udp (app-layer-protocol:dns;) or alert ip (ipproto:udp; app-layer-protocol:dns;) The same rules extend to other another such protocol, dcerpc. - The app layer parser api now takes in the ipproto while registering callbacks. - The app inspection/detection engine also takes an ipproto. - All app layer parser functions now take direction as STREAM_TOSERVER or STREAM_TOCLIENT, as opposed to 0 or 1, which was taken by some of the functions. - FlowInitialize() and FlowRecycle() now resets proto to 0. This is needed by unittests, which would try to clean the flow, and that would call the api, AppLayerParserCleanupParserState(), which would try to clean the app state, but the app layer now needs an ipproto to figure out which api to internally call to clean the state, and if the ipproto is 0, it would return without trying to clean the state. - A lot of unittests are now updated where if they are using a flow and they need to use the app layer, we would set a flow ipproto. - The "app-layer" section in the yaml conf has also been updated as well.
12 years ago
f.proto = IPPROTO_TCP;
p->src.family = AF_INET;
p->dst.family = AF_INET;
p->proto = IPPROTO_TCP;
p->flow = &f;
tcph.th_win = 5480;
tcph.th_flags = TH_PUSH | TH_ACK;
p->tcph = &tcph;
p->flowflags = FLOW_PKT_TOSERVER;
p->payload = packet;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
p->tcph->th_seq = htonl(10);
p->tcph->th_ack = htonl(31);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
p->payload_len = 10;
FAIL_IF(StreamTcpReassembleHandleSegment(&tv, ra_ctx,&ssn, &ssn.client, p, &pq) == -1);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
p->tcph->th_seq = htonl(20);
p->tcph->th_ack = htonl(31);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
p->payload_len = 10;
FAIL_IF(StreamTcpReassembleHandleSegment(&tv, ra_ctx,&ssn, &ssn.client, p, &pq) == -1);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
p->tcph->th_seq = htonl(40);
p->tcph->th_ack = htonl(31);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
p->payload_len = 10;
FAIL_IF(StreamTcpReassembleHandleSegment(&tv, ra_ctx,&ssn, &ssn.client, p, &pq) == -1);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
p->tcph->th_seq = htonl(5);
p->tcph->th_ack = htonl(31);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
p->payload_len = 30;
FAIL_IF(StreamTcpReassembleHandleSegment(&tv, ra_ctx,&ssn, &ssn.client, p, &pq) == -1);
StreamTcpUTClearSession(&ssn);
StreamTcpUTDeinit(ra_ctx);
SCFree(p);
PASS;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
static int StreamTcpReassembleTest34(void)
{
TcpSession ssn;
Packet *p = PacketGetFromAlloc();
FAIL_IF(unlikely(p == NULL));
Flow f;
TCPHdr tcph;
TcpReassemblyThreadCtx *ra_ctx = NULL;
ssn.client.os_policy = OS_POLICY_BSD;
uint8_t packet[1460] = "";
StreamTcpUTInit(&ra_ctx);
StreamTcpUTSetupSession(&ssn);
PacketQueue pq;
memset(&pq,0,sizeof(PacketQueue));
memset(&f, 0, sizeof (Flow));
memset(&tcph, 0, sizeof (TCPHdr));
ThreadVars tv;
memset(&tv, 0, sizeof (ThreadVars));
FLOW_INITIALIZE(&f);
f.protoctx = &ssn;
App layer API rewritten. The main files in question are: app-layer.[ch], app-layer-detect-proto.[ch] and app-layer-parser.[ch]. Things addressed in this commit: - Brings out a proper separation between protocol detection phase and the parser phase. - The dns app layer now is registered such that we don't use "dnstcp" and "dnsudp" in the rules. A user who previously wrote a rule like this - "alert dnstcp....." or "alert dnsudp....." would now have to use, alert dns (ipproto:tcp;) or alert udp (app-layer-protocol:dns;) or alert ip (ipproto:udp; app-layer-protocol:dns;) The same rules extend to other another such protocol, dcerpc. - The app layer parser api now takes in the ipproto while registering callbacks. - The app inspection/detection engine also takes an ipproto. - All app layer parser functions now take direction as STREAM_TOSERVER or STREAM_TOCLIENT, as opposed to 0 or 1, which was taken by some of the functions. - FlowInitialize() and FlowRecycle() now resets proto to 0. This is needed by unittests, which would try to clean the flow, and that would call the api, AppLayerParserCleanupParserState(), which would try to clean the app state, but the app layer now needs an ipproto to figure out which api to internally call to clean the state, and if the ipproto is 0, it would return without trying to clean the state. - A lot of unittests are now updated where if they are using a flow and they need to use the app layer, we would set a flow ipproto. - The "app-layer" section in the yaml conf has also been updated as well.
12 years ago
f.proto = IPPROTO_TCP;
p->src.family = AF_INET;
p->dst.family = AF_INET;
p->proto = IPPROTO_TCP;
p->flow = &f;
tcph.th_win = 5480;
tcph.th_flags = TH_PUSH | TH_ACK;
p->tcph = &tcph;
p->flowflags = FLOW_PKT_TOSERVER;
p->payload = packet;
SET_ISN(&ssn.client, 857961230);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
p->tcph->th_seq = htonl(857961230);
p->tcph->th_ack = htonl(31);
p->payload_len = 304;
FAIL_IF(StreamTcpReassembleHandleSegment(&tv, ra_ctx,&ssn, &ssn.client, p, &pq) == -1);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
p->tcph->th_seq = htonl(857961534);
p->tcph->th_ack = htonl(31);
p->payload_len = 1460;
FAIL_IF(StreamTcpReassembleHandleSegment(&tv, ra_ctx,&ssn, &ssn.client, p, &pq) == -1);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
p->tcph->th_seq = htonl(857963582);
p->tcph->th_ack = htonl(31);
p->payload_len = 1460;
FAIL_IF(StreamTcpReassembleHandleSegment(&tv, ra_ctx,&ssn, &ssn.client, p, &pq) == -1);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
p->tcph->th_seq = htonl(857960946);
p->tcph->th_ack = htonl(31);
p->payload_len = 1460;
FAIL_IF(StreamTcpReassembleHandleSegment(&tv, ra_ctx,&ssn, &ssn.client, p, &pq) == -1);
StreamTcpUTClearSession(&ssn);
StreamTcpUTDeinit(ra_ctx);
SCFree(p);
PASS;
}
/** \test Test the bug 76 condition */
static int StreamTcpReassembleTest37(void)
{
TcpSession ssn;
Flow f;
TCPHdr tcph;
TcpReassemblyThreadCtx *ra_ctx = NULL;
uint8_t packet[1460] = "";
PacketQueue pq;
ThreadVars tv;
Packet *p = PacketGetFromAlloc();
FAIL_IF(unlikely(p == NULL));
StreamTcpUTInit(&ra_ctx);
StreamTcpUTSetupSession(&ssn);
memset(&pq,0,sizeof(PacketQueue));
memset(&f, 0, sizeof (Flow));
memset(&tcph, 0, sizeof (TCPHdr));
memset(&tv, 0, sizeof (ThreadVars));
FLOW_INITIALIZE(&f);
f.protoctx = &ssn;
App layer API rewritten. The main files in question are: app-layer.[ch], app-layer-detect-proto.[ch] and app-layer-parser.[ch]. Things addressed in this commit: - Brings out a proper separation between protocol detection phase and the parser phase. - The dns app layer now is registered such that we don't use "dnstcp" and "dnsudp" in the rules. A user who previously wrote a rule like this - "alert dnstcp....." or "alert dnsudp....." would now have to use, alert dns (ipproto:tcp;) or alert udp (app-layer-protocol:dns;) or alert ip (ipproto:udp; app-layer-protocol:dns;) The same rules extend to other another such protocol, dcerpc. - The app layer parser api now takes in the ipproto while registering callbacks. - The app inspection/detection engine also takes an ipproto. - All app layer parser functions now take direction as STREAM_TOSERVER or STREAM_TOCLIENT, as opposed to 0 or 1, which was taken by some of the functions. - FlowInitialize() and FlowRecycle() now resets proto to 0. This is needed by unittests, which would try to clean the flow, and that would call the api, AppLayerParserCleanupParserState(), which would try to clean the app state, but the app layer now needs an ipproto to figure out which api to internally call to clean the state, and if the ipproto is 0, it would return without trying to clean the state. - A lot of unittests are now updated where if they are using a flow and they need to use the app layer, we would set a flow ipproto. - The "app-layer" section in the yaml conf has also been updated as well.
12 years ago
f.proto = IPPROTO_TCP;
p->src.family = AF_INET;
p->dst.family = AF_INET;
p->proto = IPPROTO_TCP;
p->flow = &f;
tcph.th_win = 5480;
tcph.th_flags = TH_PUSH | TH_ACK;
p->tcph = &tcph;
p->flowflags = FLOW_PKT_TOSERVER;
p->payload = packet;
ssn.client.os_policy = OS_POLICY_BSD;
p->tcph->th_seq = htonl(3061088537UL);
p->tcph->th_ack = htonl(1729548549UL);
p->payload_len = 1391;
ssn.client.last_ack = 3061091137UL;
SET_ISN(&ssn.client, 3061091309UL);
/* pre base_seq, so should be rejected */
FAIL_IF (StreamTcpReassembleHandleSegment(&tv, ra_ctx,&ssn, &ssn.client, p, &pq) != -1);
p->tcph->th_seq = htonl(3061089928UL);
p->tcph->th_ack = htonl(1729548549UL);
p->payload_len = 1391;
ssn.client.last_ack = 3061091137UL;
FAIL_IF(StreamTcpReassembleHandleSegment(&tv, ra_ctx,&ssn, &ssn.client, p, &pq) == -1);
p->tcph->th_seq = htonl(3061091319UL);
p->tcph->th_ack = htonl(1729548549UL);
p->payload_len = 1391;
ssn.client.last_ack = 3061091137UL;
FAIL_IF(StreamTcpReassembleHandleSegment(&tv, ra_ctx,&ssn, &ssn.client, p, &pq) == -1);
StreamTcpUTClearSession(&ssn);
StreamTcpUTDeinit(ra_ctx);
SCFree(p);
PASS;
}
/**
* \test Test to make sure that we don't return the segments until the app
* layer proto has been detected and after that remove the processed
* segments.
*
* \retval On success it returns 1 and on failure 0.
*/
static int StreamTcpReassembleTest39 (void)
{
Packet *p = PacketGetFromAlloc();
FAIL_IF(unlikely(p == NULL));
Flow f;
ThreadVars tv;
StreamTcpThread stt;
TCPHdr tcph;
PacketQueue pq;
memset(&pq,0,sizeof(PacketQueue));
memset (&f, 0, sizeof(Flow));
memset(&tv, 0, sizeof (ThreadVars));
memset(&stt, 0, sizeof (stt));
memset(&tcph, 0, sizeof (TCPHdr));
TcpSession *ssn = NULL;
FLOW_INITIALIZE(&f);
f.flags = FLOW_IPV4;
App layer API rewritten. The main files in question are: app-layer.[ch], app-layer-detect-proto.[ch] and app-layer-parser.[ch]. Things addressed in this commit: - Brings out a proper separation between protocol detection phase and the parser phase. - The dns app layer now is registered such that we don't use "dnstcp" and "dnsudp" in the rules. A user who previously wrote a rule like this - "alert dnstcp....." or "alert dnsudp....." would now have to use, alert dns (ipproto:tcp;) or alert udp (app-layer-protocol:dns;) or alert ip (ipproto:udp; app-layer-protocol:dns;) The same rules extend to other another such protocol, dcerpc. - The app layer parser api now takes in the ipproto while registering callbacks. - The app inspection/detection engine also takes an ipproto. - All app layer parser functions now take direction as STREAM_TOSERVER or STREAM_TOCLIENT, as opposed to 0 or 1, which was taken by some of the functions. - FlowInitialize() and FlowRecycle() now resets proto to 0. This is needed by unittests, which would try to clean the flow, and that would call the api, AppLayerParserCleanupParserState(), which would try to clean the app state, but the app layer now needs an ipproto to figure out which api to internally call to clean the state, and if the ipproto is 0, it would return without trying to clean the state. - A lot of unittests are now updated where if they are using a flow and they need to use the app layer, we would set a flow ipproto. - The "app-layer" section in the yaml conf has also been updated as well.
12 years ago
f.proto = IPPROTO_TCP;
p->flow = &f;
p->tcph = &tcph;
FLOWLOCK_WRLOCK(&f);
int ret = 0;
StreamTcpUTInit(&stt.ra_ctx);
/* handshake */
tcph.th_win = htons(5480);
tcph.th_flags = TH_SYN;
p->flowflags = FLOW_PKT_TOSERVER;
p->payload_len = 0;
p->payload = NULL;
if (StreamTcpPacket(&tv, p, &stt, &pq) == -1)
goto end;
ssn = (TcpSession *)f.protoctx;
if (StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ||
StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ||
f.alproto != ALPROTO_UNKNOWN ||
f.alproto_ts != ALPROTO_UNKNOWN ||
f.alproto_tc != ALPROTO_UNKNOWN ||
ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED ||
FLOW_IS_PM_DONE(&f, STREAM_TOSERVER) || FLOW_IS_PP_DONE(&f, STREAM_TOSERVER) ||
FLOW_IS_PM_DONE(&f, STREAM_TOCLIENT) || FLOW_IS_PP_DONE(&f, STREAM_TOCLIENT) ||
ssn->client.seg_list != NULL ||
ssn->server.seg_list != NULL ||
ssn->data_first_seen_dir != 0) {
printf("failure 1\n");
goto end;
}
16 years ago
/* handshake */
p->tcph->th_ack = htonl(1);
p->tcph->th_flags = TH_SYN | TH_ACK;
p->flowflags = FLOW_PKT_TOCLIENT;
p->payload_len = 0;
p->payload = NULL;
if (StreamTcpPacket(&tv, p, &stt, &pq) == -1)
goto end;
if (StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ||
StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ||
f.alproto != ALPROTO_UNKNOWN ||
f.alproto_ts != ALPROTO_UNKNOWN ||
f.alproto_tc != ALPROTO_UNKNOWN ||
ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED ||
FLOW_IS_PM_DONE(&f, STREAM_TOSERVER) || FLOW_IS_PP_DONE(&f, STREAM_TOSERVER) ||
FLOW_IS_PM_DONE(&f, STREAM_TOCLIENT) || FLOW_IS_PP_DONE(&f, STREAM_TOCLIENT) ||
ssn->client.seg_list != NULL ||
ssn->server.seg_list != NULL ||
ssn->data_first_seen_dir != 0) {
printf("failure 2\n");
goto end;
}
/* handshake */
p->tcph->th_ack = htonl(1);
p->tcph->th_seq = htonl(1);
p->tcph->th_flags = TH_ACK;
p->flowflags = FLOW_PKT_TOSERVER;
p->payload_len = 0;
p->payload = NULL;
if (StreamTcpPacket(&tv, p, &stt, &pq) == -1)
goto end;
if (StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ||
StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ||
f.alproto != ALPROTO_UNKNOWN ||
f.alproto_ts != ALPROTO_UNKNOWN ||
f.alproto_tc != ALPROTO_UNKNOWN ||
ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED ||
FLOW_IS_PM_DONE(&f, STREAM_TOSERVER) || FLOW_IS_PP_DONE(&f, STREAM_TOSERVER) ||
FLOW_IS_PM_DONE(&f, STREAM_TOCLIENT) || FLOW_IS_PP_DONE(&f, STREAM_TOCLIENT) ||
ssn->client.seg_list != NULL ||
ssn->server.seg_list != NULL ||
ssn->data_first_seen_dir != 0) {
printf("failure 3\n");
goto end;
}
/* partial request */
uint8_t request1[] = { 0x47, 0x45, };
p->tcph->th_ack = htonl(1);
p->tcph->th_seq = htonl(1);
p->tcph->th_flags = TH_PUSH | TH_ACK;
p->flowflags = FLOW_PKT_TOSERVER;
p->payload_len = sizeof(request1);
p->payload = request1;
if (StreamTcpPacket(&tv, p, &stt, &pq) == -1)
goto end;
if (StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ||
StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ||
f.alproto != ALPROTO_UNKNOWN ||
f.alproto_ts != ALPROTO_UNKNOWN ||
f.alproto_tc != ALPROTO_UNKNOWN ||
ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED ||
FLOW_IS_PM_DONE(&f, STREAM_TOSERVER) || FLOW_IS_PP_DONE(&f, STREAM_TOSERVER) ||
FLOW_IS_PM_DONE(&f, STREAM_TOCLIENT) || FLOW_IS_PP_DONE(&f, STREAM_TOCLIENT) ||
ssn->client.seg_list == NULL ||
ssn->client.seg_list->next != NULL ||
ssn->server.seg_list != NULL ||
ssn->data_first_seen_dir != STREAM_TOSERVER) {
printf("failure 4\n");
goto end;
}
/* response ack against partial request */
p->tcph->th_ack = htonl(3);
p->tcph->th_seq = htonl(1);
p->tcph->th_flags = TH_ACK;
p->flowflags = FLOW_PKT_TOCLIENT;
p->payload_len = 0;
p->payload = NULL;
if (StreamTcpPacket(&tv, p, &stt, &pq) == -1)
goto end;
if (StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ||
StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ||
f.alproto != ALPROTO_UNKNOWN ||
f.alproto_ts != ALPROTO_UNKNOWN ||
f.alproto_tc != ALPROTO_UNKNOWN ||
ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED ||
FLOW_IS_PM_DONE(&f, STREAM_TOSERVER) || !FLOW_IS_PP_DONE(&f, STREAM_TOSERVER) ||
FLOW_IS_PM_DONE(&f, STREAM_TOCLIENT) || FLOW_IS_PP_DONE(&f, STREAM_TOCLIENT) ||
ssn->client.seg_list == NULL ||
ssn->client.seg_list->next != NULL ||
ssn->server.seg_list != NULL ||
ssn->data_first_seen_dir != STREAM_TOSERVER) {
printf("failure 5\n");
goto end;
}
/* complete partial request */
uint8_t request2[] = {
0x54, 0x20, 0x2f, 0x69, 0x6e, 0x64,
0x65, 0x78, 0x2e, 0x68, 0x74, 0x6d, 0x6c, 0x20,
0x48, 0x54, 0x54, 0x50, 0x2f, 0x31, 0x2e, 0x30,
0x0d, 0x0a, 0x48, 0x6f, 0x73, 0x74, 0x3a, 0x20,
0x6c, 0x6f, 0x63, 0x61, 0x6c, 0x68, 0x6f, 0x73,
0x74, 0x0d, 0x0a, 0x55, 0x73, 0x65, 0x72, 0x2d,
0x41, 0x67, 0x65, 0x6e, 0x74, 0x3a, 0x20, 0x41,
0x70, 0x61, 0x63, 0x68, 0x65, 0x42, 0x65, 0x6e,
0x63, 0x68, 0x2f, 0x32, 0x2e, 0x33, 0x0d, 0x0a,
0x41, 0x63, 0x63, 0x65, 0x70, 0x74, 0x3a, 0x20,
0x2a, 0x2f, 0x2a, 0x0d, 0x0a, 0x0d, 0x0a };
p->tcph->th_ack = htonl(1);
p->tcph->th_seq = htonl(3);
p->tcph->th_flags = TH_PUSH | TH_ACK;
p->flowflags = FLOW_PKT_TOSERVER;
p->payload_len = sizeof(request2);
p->payload = request2;
if (StreamTcpPacket(&tv, p, &stt, &pq) == -1)
goto end;
if (StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ||
StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ||
f.alproto != ALPROTO_UNKNOWN ||
f.alproto_ts != ALPROTO_UNKNOWN ||
f.alproto_tc != ALPROTO_UNKNOWN ||
ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED ||
FLOW_IS_PM_DONE(&f, STREAM_TOSERVER) || !FLOW_IS_PP_DONE(&f, STREAM_TOSERVER) ||
FLOW_IS_PM_DONE(&f, STREAM_TOCLIENT) || FLOW_IS_PP_DONE(&f, STREAM_TOCLIENT) ||
ssn->client.seg_list == NULL ||
ssn->client.seg_list->next == NULL ||
ssn->client.seg_list->next->next != NULL ||
ssn->server.seg_list != NULL ||
ssn->data_first_seen_dir != STREAM_TOSERVER) {
printf("failure 6\n");
goto end;
}
/* response - request ack */
uint8_t response[] = {
0x48, 0x54, 0x54, 0x50, 0x2f, 0x31, 0x2e, 0x31,
0x20, 0x32, 0x30, 0x30, 0x20, 0x4f, 0x4b, 0x0d,
0x0a, 0x44, 0x61, 0x74, 0x65, 0x3a, 0x20, 0x46,
0x72, 0x69, 0x2c, 0x20, 0x32, 0x33, 0x20, 0x53,
0x65, 0x70, 0x20, 0x32, 0x30, 0x31, 0x31, 0x20,
0x30, 0x36, 0x3a, 0x32, 0x39, 0x3a, 0x33, 0x39,
0x20, 0x47, 0x4d, 0x54, 0x0d, 0x0a, 0x53, 0x65,
0x72, 0x76, 0x65, 0x72, 0x3a, 0x20, 0x41, 0x70,
0x61, 0x63, 0x68, 0x65, 0x2f, 0x32, 0x2e, 0x32,
0x2e, 0x31, 0x35, 0x20, 0x28, 0x55, 0x6e, 0x69,
0x78, 0x29, 0x20, 0x44, 0x41, 0x56, 0x2f, 0x32,
0x0d, 0x0a, 0x4c, 0x61, 0x73, 0x74, 0x2d, 0x4d,
0x6f, 0x64, 0x69, 0x66, 0x69, 0x65, 0x64, 0x3a,
0x20, 0x54, 0x68, 0x75, 0x2c, 0x20, 0x30, 0x34,
0x20, 0x4e, 0x6f, 0x76, 0x20, 0x32, 0x30, 0x31,
0x30, 0x20, 0x31, 0x35, 0x3a, 0x30, 0x34, 0x3a,
0x34, 0x36, 0x20, 0x47, 0x4d, 0x54, 0x0d, 0x0a,
0x45, 0x54, 0x61, 0x67, 0x3a, 0x20, 0x22, 0x61,
0x62, 0x38, 0x39, 0x36, 0x35, 0x2d, 0x32, 0x63,
0x2d, 0x34, 0x39, 0x34, 0x33, 0x62, 0x37, 0x61,
0x37, 0x66, 0x37, 0x66, 0x38, 0x30, 0x22, 0x0d,
0x0a, 0x41, 0x63, 0x63, 0x65, 0x70, 0x74, 0x2d,
0x52, 0x61, 0x6e, 0x67, 0x65, 0x73, 0x3a, 0x20,
0x62, 0x79, 0x74, 0x65, 0x73, 0x0d, 0x0a, 0x43,
0x6f, 0x6e, 0x74, 0x65, 0x6e, 0x74, 0x2d, 0x4c,
0x65, 0x6e, 0x67, 0x74, 0x68, 0x3a, 0x20, 0x34,
0x34, 0x0d, 0x0a, 0x43, 0x6f, 0x6e, 0x6e, 0x65,
0x63, 0x74, 0x69, 0x6f, 0x6e, 0x3a, 0x20, 0x63,
0x6c, 0x6f, 0x73, 0x65, 0x0d, 0x0a, 0x43, 0x6f,
0x6e, 0x74, 0x65, 0x6e, 0x74, 0x2d, 0x54, 0x79,
0x70, 0x65, 0x3a, 0x20, 0x74, 0x65, 0x78, 0x74,
0x2f, 0x68, 0x74, 0x6d, 0x6c, 0x0d, 0x0a, 0x58,
0x2d, 0x50, 0x61, 0x64, 0x3a, 0x20, 0x61, 0x76,
0x6f, 0x69, 0x64, 0x20, 0x62, 0x72, 0x6f, 0x77,
0x73, 0x65, 0x72, 0x20, 0x62, 0x75, 0x67, 0x0d,
0x0a, 0x0d, 0x0a, 0x3c, 0x68, 0x74, 0x6d, 0x6c,
0x3e, 0x3c, 0x62, 0x6f, 0x64, 0x79, 0x3e, 0x3c,
0x68, 0x31, 0x3e, 0x49, 0x74, 0x20, 0x77, 0x6f,
0x72, 0x6b, 0x73, 0x21, 0x3c, 0x2f, 0x68, 0x31,
0x3e, 0x3c, 0x2f, 0x62, 0x6f, 0x64, 0x79, 0x3e,
0x3c, 0x2f, 0x68, 0x74, 0x6d, 0x6c, 0x3e };
p->tcph->th_ack = htonl(88);
p->tcph->th_seq = htonl(1);
p->tcph->th_flags = TH_PUSH | TH_ACK;
p->flowflags = FLOW_PKT_TOCLIENT;
p->payload_len = sizeof(response);
p->payload = response;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (StreamTcpPacket(&tv, p, &stt, &pq) == -1)
goto end;
if (StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ||
!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ||
f.alproto != ALPROTO_HTTP ||
f.alproto_ts != ALPROTO_HTTP ||
f.alproto_tc != ALPROTO_UNKNOWN ||
ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED ||
!FLOW_IS_PM_DONE(&f, STREAM_TOSERVER) || !FLOW_IS_PP_DONE(&f, STREAM_TOSERVER) ||
FLOW_IS_PM_DONE(&f, STREAM_TOCLIENT) || FLOW_IS_PP_DONE(&f, STREAM_TOCLIENT) ||
ssn->client.seg_list == NULL ||
ssn->client.seg_list->next == NULL ||
ssn->client.seg_list->next->next != NULL ||
ssn->server.seg_list == NULL ||
ssn->server.seg_list->next != NULL ||
ssn->data_first_seen_dir != APP_LAYER_DATA_ALREADY_SENT_TO_APP_LAYER) {
printf("failure 7\n");
goto end;
}
/* response ack from request */
p->tcph->th_ack = htonl(328);
p->tcph->th_seq = htonl(88);
p->tcph->th_flags = TH_ACK;
p->flowflags = FLOW_PKT_TOSERVER;
p->payload_len = 0;
p->payload = NULL;
if (StreamTcpPacket(&tv, p, &stt, &pq) == -1)
goto end;
if (!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ||
!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ||
f.alproto != ALPROTO_HTTP ||
f.alproto_ts != ALPROTO_HTTP ||
f.alproto_tc != ALPROTO_HTTP ||
ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED ||
!FLOW_IS_PM_DONE(&f, STREAM_TOSERVER) || !FLOW_IS_PP_DONE(&f, STREAM_TOSERVER) ||
!FLOW_IS_PM_DONE(&f, STREAM_TOCLIENT) || FLOW_IS_PP_DONE(&f, STREAM_TOCLIENT) ||
ssn->client.seg_list == NULL ||
ssn->client.seg_list->next == NULL ||
ssn->client.seg_list->next->next != NULL ||
ssn->server.seg_list == NULL ||
ssn->server.seg_list->next != NULL ||
ssn->data_first_seen_dir != APP_LAYER_DATA_ALREADY_SENT_TO_APP_LAYER) {
printf("failure 8\n");
goto end;
}
/* response - acking */
p->tcph->th_ack = htonl(88);
p->tcph->th_seq = htonl(328);
p->tcph->th_flags = TH_PUSH | TH_ACK;
p->flowflags = FLOW_PKT_TOCLIENT;
p->payload_len = 0;
p->payload = NULL;
if (StreamTcpPacket(&tv, p, &stt, &pq) == -1)
goto end;
if (!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ||
!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ||
f.alproto != ALPROTO_HTTP ||
f.alproto_ts != ALPROTO_HTTP ||
f.alproto_tc != ALPROTO_HTTP ||
ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED ||
!FLOW_IS_PM_DONE(&f, STREAM_TOSERVER) || !FLOW_IS_PP_DONE(&f, STREAM_TOSERVER) ||
!FLOW_IS_PM_DONE(&f, STREAM_TOCLIENT) || FLOW_IS_PP_DONE(&f, STREAM_TOCLIENT) ||
ssn->client.seg_list == NULL ||
ssn->client.seg_list->next == NULL ||
ssn->server.seg_list == NULL ||
ssn->server.seg_list->next != NULL ||
ssn->data_first_seen_dir != APP_LAYER_DATA_ALREADY_SENT_TO_APP_LAYER) {
printf("failure 9\n");
goto end;
}
/* response ack from request */
p->tcph->th_ack = htonl(328);
p->tcph->th_seq = htonl(88);
p->tcph->th_flags = TH_ACK;
p->flowflags = FLOW_PKT_TOSERVER;
p->payload_len = 0;
p->payload = NULL;
if (StreamTcpPacket(&tv, p, &stt, &pq) == -1)
goto end;
if (!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ||
!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ||
f.alproto != ALPROTO_HTTP ||
f.alproto_ts != ALPROTO_HTTP ||
f.alproto_tc != ALPROTO_HTTP ||
ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED ||
!FLOW_IS_PM_DONE(&f, STREAM_TOSERVER) || !FLOW_IS_PP_DONE(&f, STREAM_TOSERVER) ||
!FLOW_IS_PM_DONE(&f, STREAM_TOCLIENT) || FLOW_IS_PP_DONE(&f, STREAM_TOCLIENT) ||
ssn->client.seg_list == NULL ||
ssn->client.seg_list->next == NULL ||
ssn->server.seg_list == NULL ||
ssn->server.seg_list->next != NULL ||
ssn->data_first_seen_dir != APP_LAYER_DATA_ALREADY_SENT_TO_APP_LAYER) {
printf("failure 10\n");
goto end;
}
/* response - acking the request again*/
p->tcph->th_ack = htonl(88);
p->tcph->th_seq = htonl(328);
p->tcph->th_flags = TH_PUSH | TH_ACK;
p->flowflags = FLOW_PKT_TOCLIENT;
p->payload_len = 0;
p->payload = NULL;
if (StreamTcpPacket(&tv, p, &stt, &pq) == -1)
goto end;
if (!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ||
!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ||
f.alproto != ALPROTO_HTTP ||
f.alproto_ts != ALPROTO_HTTP ||
f.alproto_tc != ALPROTO_HTTP ||
ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED ||
!FLOW_IS_PM_DONE(&f, STREAM_TOSERVER) || !FLOW_IS_PP_DONE(&f, STREAM_TOSERVER) ||
!FLOW_IS_PM_DONE(&f, STREAM_TOCLIENT) || FLOW_IS_PP_DONE(&f, STREAM_TOCLIENT) ||
ssn->client.seg_list == NULL ||
ssn->client.seg_list->next == NULL ||
ssn->server.seg_list == NULL ||
ssn->server.seg_list->next != NULL ||
ssn->data_first_seen_dir != APP_LAYER_DATA_ALREADY_SENT_TO_APP_LAYER) {
printf("failure 11\n");
goto end;
}
/*** New Request ***/
/* partial request */
p->tcph->th_ack = htonl(328);
p->tcph->th_seq = htonl(88);
p->tcph->th_flags = TH_PUSH | TH_ACK;
p->flowflags = FLOW_PKT_TOSERVER;
p->payload_len = sizeof(request1);
p->payload = request1;
if (StreamTcpPacket(&tv, p, &stt, &pq) == -1)
goto end;
if (!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ||
!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ||
f.alproto != ALPROTO_HTTP ||
f.alproto_ts != ALPROTO_HTTP ||
f.alproto_tc != ALPROTO_HTTP ||
ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED ||
!FLOW_IS_PM_DONE(&f, STREAM_TOSERVER) || !FLOW_IS_PP_DONE(&f, STREAM_TOSERVER) ||
!FLOW_IS_PM_DONE(&f, STREAM_TOCLIENT) || FLOW_IS_PP_DONE(&f, STREAM_TOCLIENT) ||
ssn->client.seg_list == NULL ||
ssn->client.seg_list->next == NULL ||
ssn->client.seg_list->next->next == NULL ||
ssn->server.seg_list == NULL ||
ssn->server.seg_list->next != NULL ||
ssn->data_first_seen_dir != APP_LAYER_DATA_ALREADY_SENT_TO_APP_LAYER) {
printf("failure 12\n");
goto end;
}
/* response ack against partial request */
p->tcph->th_ack = htonl(90);
p->tcph->th_seq = htonl(328);
p->tcph->th_flags = TH_ACK;
p->flowflags = FLOW_PKT_TOCLIENT;
p->payload_len = 0;
p->payload = NULL;
if (StreamTcpPacket(&tv, p, &stt, &pq) == -1)
goto end;
if (!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ||
!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ||
f.alproto != ALPROTO_HTTP ||
f.alproto_ts != ALPROTO_HTTP ||
f.alproto_tc != ALPROTO_HTTP ||
ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED ||
!FLOW_IS_PM_DONE(&f, STREAM_TOSERVER) || !FLOW_IS_PP_DONE(&f, STREAM_TOSERVER) ||
!FLOW_IS_PM_DONE(&f, STREAM_TOCLIENT) || FLOW_IS_PP_DONE(&f, STREAM_TOCLIENT) ||
ssn->client.seg_list == NULL ||
ssn->client.seg_list->next == NULL ||
ssn->client.seg_list->next->next == NULL ||
ssn->server.seg_list == NULL ||
ssn->server.seg_list->next != NULL ||
ssn->data_first_seen_dir != APP_LAYER_DATA_ALREADY_SENT_TO_APP_LAYER) {
printf("failure 13\n");
goto end;
}
/* complete request */
p->tcph->th_ack = htonl(328);
p->tcph->th_seq = htonl(90);
p->tcph->th_flags = TH_PUSH | TH_ACK;
p->flowflags = FLOW_PKT_TOSERVER;
p->payload_len = sizeof(request2);
p->payload = request2;
if (StreamTcpPacket(&tv, p, &stt, &pq) == -1)
goto end;
if (!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ||
!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ||
f.alproto != ALPROTO_HTTP ||
f.alproto_ts != ALPROTO_HTTP ||
f.alproto_tc != ALPROTO_HTTP ||
ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED ||
!FLOW_IS_PM_DONE(&f, STREAM_TOSERVER) || !FLOW_IS_PP_DONE(&f, STREAM_TOSERVER) ||
!FLOW_IS_PM_DONE(&f, STREAM_TOCLIENT) || FLOW_IS_PP_DONE(&f, STREAM_TOCLIENT) ||
ssn->client.seg_list == NULL ||
ssn->client.seg_list->next == NULL ||
ssn->client.seg_list->next->next == NULL ||
ssn->client.seg_list->next->next->next == NULL ||
ssn->server.seg_list == NULL ||
ssn->server.seg_list->next != NULL ||
ssn->data_first_seen_dir != APP_LAYER_DATA_ALREADY_SENT_TO_APP_LAYER) {
printf("failure 14\n");
goto end;
}
/* response ack against second partial request */
p->tcph->th_ack = htonl(175);
p->tcph->th_seq = htonl(328);
p->tcph->th_flags = TH_ACK;
p->flowflags = FLOW_PKT_TOCLIENT;
p->payload_len = 0;
p->payload = NULL;
if (StreamTcpPacket(&tv, p, &stt, &pq) == -1)
goto end;
if (!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ||
!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ||
f.alproto != ALPROTO_HTTP ||
f.alproto_ts != ALPROTO_HTTP ||
f.alproto_tc != ALPROTO_HTTP ||
ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED ||
!FLOW_IS_PM_DONE(&f, STREAM_TOSERVER) || !FLOW_IS_PP_DONE(&f, STREAM_TOSERVER) ||
!FLOW_IS_PM_DONE(&f, STREAM_TOCLIENT) || FLOW_IS_PP_DONE(&f, STREAM_TOCLIENT) ||
ssn->client.seg_list->next == NULL ||
ssn->client.seg_list->next->next == NULL ||
ssn->client.seg_list->next->next->next == NULL ||
ssn->server.seg_list == NULL ||
ssn->server.seg_list->next != NULL ||
ssn->data_first_seen_dir != APP_LAYER_DATA_ALREADY_SENT_TO_APP_LAYER) {
printf("failure 15\n");
goto end;
}
/* response acking a request */
p->tcph->th_ack = htonl(175);
p->tcph->th_seq = htonl(328);
p->tcph->th_flags = TH_ACK;
p->flowflags = FLOW_PKT_TOCLIENT;
p->payload_len = 0;
p->payload = NULL;
if (StreamTcpPacket(&tv, p, &stt, &pq) == -1)
goto end;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SCLogDebug("StreamTcpIsSetStreamFlagAppProtoDetectionCompleted %s, "
"StreamTcpIsSetStreamFlagAppProtoDetectionCompleted %s, "
"f.alproto %u f.alproto_ts %u f.alproto_tc %u",
StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ? "true" : "false",
StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ? "true" : "false",
f.alproto, f.alproto_ts, f.alproto_tc);
if (!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ||
!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ||
f.alproto != ALPROTO_HTTP ||
f.alproto_ts != ALPROTO_HTTP ||
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
f.alproto_tc != ALPROTO_HTTP)// ||
//ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED)// ||
//!FLOW_IS_PM_DONE(&f, STREAM_TOSERVER) || !FLOW_IS_PP_DONE(&f, STREAM_TOSERVER) ||
//!FLOW_IS_PM_DONE(&f, STREAM_TOCLIENT) || FLOW_IS_PP_DONE(&f, STREAM_TOCLIENT) ||
//ssn->client.seg_list != NULL ||
//ssn->server.seg_list == NULL ||
//ssn->server.seg_list->next != NULL ||
//ssn->data_first_seen_dir != APP_LAYER_DATA_ALREADY_SENT_TO_APP_LAYER)
{
printf("failure 15\n");
goto end;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
StreamTcpPruneSession(&f, STREAM_TOSERVER);
StreamTcpPruneSession(&f, STREAM_TOCLIENT);
/* request acking a response */
p->tcph->th_ack = htonl(328);
p->tcph->th_seq = htonl(175);
p->tcph->th_flags = TH_ACK;
p->flowflags = FLOW_PKT_TOSERVER;
p->payload_len = 0;
p->payload = NULL;
if (StreamTcpPacket(&tv, p, &stt, &pq) == -1)
goto end;
#if 0
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (//!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->server) ||
//!StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(&ssn->client) ||
//f.alproto != ALPROTO_HTTP ||
//f.alproto_ts != ALPROTO_HTTP ||
//f.alproto_tc != ALPROTO_HTTP ||
//ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED ||
!FLOW_IS_PM_DONE(&f, STREAM_TOSERVER) || !FLOW_IS_PP_DONE(&f, STREAM_TOSERVER) ||
!FLOW_IS_PM_DONE(&f, STREAM_TOCLIENT) || FLOW_IS_PP_DONE(&f, STREAM_TOCLIENT) ||
ssn->client.seg_list != NULL ||
ssn->server.seg_list != NULL ||
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
ssn->data_first_seen_dir != APP_LAYER_DATA_ALREADY_SENT_TO_APP_LAYER
) {
printf("failure 16\n");
abort();
goto end;
}
#endif
16 years ago
ret = 1;
end:
StreamTcpSessionClear(ssn);
StreamTcpUTDeinit(stt.ra_ctx);
SCFree(p);
FLOWLOCK_UNLOCK(&f);
return ret;
16 years ago
}
/**
* \test Test to make sure that we sent all the segments from the initial
* segments to app layer until we have detected the app layer proto.
*
* \retval On success it returns 1 and on failure 0.
*/
static int StreamTcpReassembleTest40 (void)
{
int ret = 0;
Packet *p = PacketGetFromAlloc();
if (unlikely(p == NULL))
return 0;
Flow *f = NULL;
16 years ago
TCPHdr tcph;
TcpSession ssn;
PacketQueue pq;
memset(&pq,0,sizeof(PacketQueue));
16 years ago
memset(&tcph, 0, sizeof (TCPHdr));
ThreadVars tv;
memset(&tv, 0, sizeof (ThreadVars));
StreamTcpInitConfig(TRUE);
StreamTcpUTSetupSession(&ssn);
TcpReassemblyThreadCtx *ra_ctx = StreamTcpReassembleInitThreadCtx(NULL);
uint8_t httpbuf1[] = "P";
uint32_t httplen1 = sizeof(httpbuf1) - 1; /* minus the \0 */
uint8_t httpbuf3[] = "O";
uint32_t httplen3 = sizeof(httpbuf3) - 1; /* minus the \0 */
uint8_t httpbuf4[] = "S";
uint32_t httplen4 = sizeof(httpbuf4) - 1; /* minus the \0 */
uint8_t httpbuf5[] = "T \r\n";
uint32_t httplen5 = sizeof(httpbuf5) - 1; /* minus the \0 */
uint8_t httpbuf2[] = "HTTP/1.0 200 OK\r\nServer: VictorServer/1.0\r\n\r\n";
uint32_t httplen2 = sizeof(httpbuf2) - 1; /* minus the \0 */
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SET_ISN(&ssn.server, 9);
ssn.server.last_ack = 10;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SET_ISN(&ssn.client, 9);
ssn.client.isn = 9;
f = UTHBuildFlow(AF_INET, "1.2.3.4", "1.2.3.5", 200, 220);
if (f == NULL)
goto end;
f->protoctx = &ssn;
App layer API rewritten. The main files in question are: app-layer.[ch], app-layer-detect-proto.[ch] and app-layer-parser.[ch]. Things addressed in this commit: - Brings out a proper separation between protocol detection phase and the parser phase. - The dns app layer now is registered such that we don't use "dnstcp" and "dnsudp" in the rules. A user who previously wrote a rule like this - "alert dnstcp....." or "alert dnsudp....." would now have to use, alert dns (ipproto:tcp;) or alert udp (app-layer-protocol:dns;) or alert ip (ipproto:udp; app-layer-protocol:dns;) The same rules extend to other another such protocol, dcerpc. - The app layer parser api now takes in the ipproto while registering callbacks. - The app inspection/detection engine also takes an ipproto. - All app layer parser functions now take direction as STREAM_TOSERVER or STREAM_TOCLIENT, as opposed to 0 or 1, which was taken by some of the functions. - FlowInitialize() and FlowRecycle() now resets proto to 0. This is needed by unittests, which would try to clean the flow, and that would call the api, AppLayerParserCleanupParserState(), which would try to clean the app state, but the app layer now needs an ipproto to figure out which api to internally call to clean the state, and if the ipproto is 0, it would return without trying to clean the state. - A lot of unittests are now updated where if they are using a flow and they need to use the app layer, we would set a flow ipproto. - The "app-layer" section in the yaml conf has also been updated as well.
12 years ago
f->proto = IPPROTO_TCP;
p->flow = f;
tcph.th_win = htons(5480);
tcph.th_seq = htonl(10);
tcph.th_ack = htonl(10);
tcph.th_flags = TH_ACK|TH_PUSH;
p->tcph = &tcph;
p->flowflags = FLOW_PKT_TOSERVER;
16 years ago
p->payload = httpbuf1;
p->payload_len = httplen1;
ssn.state = TCP_ESTABLISHED;
16 years ago
TcpStream *s = NULL;
s = &ssn.client;
16 years ago
FLOWLOCK_WRLOCK(f);
if (StreamTcpReassembleHandleSegment(&tv, ra_ctx, &ssn, s, p, &pq) == -1) {
printf("failed in segments reassembly, while processing toserver packet (1): ");
goto end;
}
p->flowflags = FLOW_PKT_TOCLIENT;
p->payload = httpbuf2;
p->payload_len = httplen2;
tcph.th_seq = htonl(10);
tcph.th_ack = htonl(11);
s = &ssn.server;
ssn.server.last_ack = 11;
16 years ago
if (StreamTcpReassembleHandleSegment(&tv, ra_ctx, &ssn, s, p, &pq) == -1) {
printf("failed in segments reassembly, while processing toserver packet (3): ");
goto end;
}
p->flowflags = FLOW_PKT_TOSERVER;
p->payload = httpbuf3;
p->payload_len = httplen3;
tcph.th_seq = htonl(11);
tcph.th_ack = htonl(55);
s = &ssn.client;
ssn.client.last_ack = 55;
16 years ago
if (StreamTcpReassembleHandleSegment(&tv, ra_ctx, &ssn, s, p, &pq) == -1) {
printf("failed in segments reassembly, while processing toserver packet (5): ");
goto end;
}
16 years ago
p->flowflags = FLOW_PKT_TOCLIENT;
p->payload = httpbuf2;
p->payload_len = httplen2;
tcph.th_seq = htonl(55);
tcph.th_ack = htonl(12);
s = &ssn.server;
ssn.server.last_ack = 12;
if (StreamTcpReassembleHandleSegment(&tv, ra_ctx, &ssn, s, p, &pq) == -1) {
printf("failed in segments reassembly, while processing toserver packet (6): ");
goto end;
}
/* check is have the segment in the list and flagged or not */
if (ssn.client.seg_list == NULL ||
SEGMENT_BEFORE_OFFSET(&ssn.client, ssn.client.seg_list, STREAM_APP_PROGRESS(&ssn.client)))
{
printf("the list is NULL or the processed segment has not been flaged (7): ");
goto end;
}
p->flowflags = FLOW_PKT_TOSERVER;
p->payload = httpbuf4;
p->payload_len = httplen4;
tcph.th_seq = htonl(12);
tcph.th_ack = htonl(100);
s = &ssn.client;
ssn.client.last_ack = 100;
16 years ago
if (StreamTcpReassembleHandleSegment(&tv, ra_ctx, &ssn, s, p, &pq) == -1) {
printf("failed in segments reassembly, while processing toserver packet (10): ");
goto end;
}
p->flowflags = FLOW_PKT_TOCLIENT;
p->payload = httpbuf2;
p->payload_len = httplen2;
tcph.th_seq = htonl(100);
tcph.th_ack = htonl(13);
s = &ssn.server;
ssn.server.last_ack = 13;
if (StreamTcpReassembleHandleSegment(&tv, ra_ctx, &ssn, s, p, &pq) == -1) {
printf("failed in segments reassembly, while processing toserver packet (11): ");
goto end;
}
p->flowflags = FLOW_PKT_TOSERVER;
p->payload = httpbuf5;
p->payload_len = httplen5;
tcph.th_seq = htonl(13);
tcph.th_ack = htonl(145);
s = &ssn.client;
ssn.client.last_ack = 145;
16 years ago
if (StreamTcpReassembleHandleSegment(&tv, ra_ctx, &ssn, s, p, &pq) == -1) {
printf("failed in segments reassembly, while processing toserver packet (14): ");
goto end;
}
16 years ago
p->flowflags = FLOW_PKT_TOCLIENT;
p->payload = httpbuf2;
p->payload_len = httplen2;
tcph.th_seq = htonl(145);
tcph.th_ack = htonl(16);
s = &ssn.server;
ssn.server.last_ack = 16;
16 years ago
if (StreamTcpReassembleHandleSegment(&tv, ra_ctx, &ssn, s, p, &pq) == -1) {
printf("failed in segments reassembly, while processing toserver packet (15): ");
goto end;
}
if (f->alproto != ALPROTO_HTTP) {
printf("app layer proto has not been detected (18): ");
goto end;
}
16 years ago
ret = 1;
end:
StreamTcpUTClearSession(&ssn);
StreamTcpReassembleFreeThreadCtx(ra_ctx);
StreamTcpFreeConfig(TRUE);
SCFree(p);
FLOWLOCK_UNLOCK(f);
UTHFreeFlow(f);
return ret;
16 years ago
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/** \test Test the memcap incrementing/decrementing and memcap check */
static int StreamTcpReassembleTest44(void)
{
StreamTcpInitConfig(TRUE);
uint32_t memuse = SC_ATOMIC_GET(ra_memuse);
StreamTcpReassembleIncrMemuse(500);
FAIL_IF(SC_ATOMIC_GET(ra_memuse) != (memuse+500));
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
StreamTcpReassembleDecrMemuse(500);
FAIL_IF(SC_ATOMIC_GET(ra_memuse) != memuse);
FAIL_IF(StreamTcpReassembleCheckMemcap(500) != 1);
FAIL_IF(StreamTcpReassembleCheckMemcap((1 + memuse + stream_config.reassembly_memcap)) != 0);
StreamTcpFreeConfig(TRUE);
FAIL_IF(SC_ATOMIC_GET(ra_memuse) != 0);
PASS;
}
/**
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
* \test Test to make sure that reassembly_depth is enforced.
*
* \retval On success it returns 1 and on failure 0.
*/
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
static int StreamTcpReassembleTest45 (void)
{
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
TcpReassemblyThreadCtx *ra_ctx = NULL;
TcpSession ssn;
ThreadVars tv;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
memset(&tv, 0, sizeof(tv));
uint8_t payload[100] = {0};
uint16_t payload_size = 100;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
StreamTcpUTInit(&ra_ctx);
stream_config.reassembly_depth = 100;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
StreamTcpUTSetupSession(&ssn);
ssn.reassembly_depth = 100;
StreamTcpUTSetupStream(&ssn.server, 100);
StreamTcpUTSetupStream(&ssn.client, 100);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
int r = StreamTcpUTAddPayload(&tv, ra_ctx, &ssn, &ssn.client, 101, payload, payload_size);
FAIL_IF(r != 0);
FAIL_IF(ssn.client.flags & STREAMTCP_STREAM_FLAG_DEPTH_REACHED);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
r = StreamTcpUTAddPayload(&tv, ra_ctx, &ssn, &ssn.client, 201, payload, payload_size);
FAIL_IF(r != 0);
FAIL_IF(!(ssn.client.flags & STREAMTCP_STREAM_FLAG_DEPTH_REACHED));
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
StreamTcpUTClearStream(&ssn.server);
StreamTcpUTClearStream(&ssn.client);
StreamTcpUTClearSession(&ssn);
StreamTcpUTDeinit(ra_ctx);
PASS;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/**
* \test Test the unlimited config value of reassembly depth.
*
* \retval On success it returns 1 and on failure 0.
*/
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
static int StreamTcpReassembleTest46 (void)
{
int result = 0;
TcpReassemblyThreadCtx *ra_ctx = NULL;
TcpSession ssn;
ThreadVars tv;
memset(&tv, 0, sizeof(tv));
uint8_t payload[100] = {0};
uint16_t payload_size = 100;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
StreamTcpUTInit(&ra_ctx);
stream_config.reassembly_depth = 0;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
StreamTcpUTSetupSession(&ssn);
StreamTcpUTSetupStream(&ssn.server, 100);
StreamTcpUTSetupStream(&ssn.client, 100);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
int r = StreamTcpUTAddPayload(&tv, ra_ctx, &ssn, &ssn.client, 101, payload, payload_size);
if (r != 0)
goto end;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (ssn.client.flags & STREAMTCP_STREAM_FLAG_NOREASSEMBLY) {
printf("STREAMTCP_STREAM_FLAG_NOREASSEMBLY set: ");
goto end;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
r = StreamTcpUTAddPayload(&tv, ra_ctx, &ssn, &ssn.client, 201, payload, payload_size);
if (r != 0)
goto end;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
if (ssn.client.flags & STREAMTCP_STREAM_FLAG_NOREASSEMBLY) {
printf("STREAMTCP_STREAM_FLAG_NOREASSEMBLY set: ");
goto end;
}
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
result = 1;
end:
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
StreamTcpUTClearStream(&ssn.server);
StreamTcpUTClearStream(&ssn.client);
StreamTcpUTClearSession(&ssn);
StreamTcpUTDeinit(ra_ctx);
return result;
}
/**
* \test Test to make sure we detect the sequence wrap around and continue
* stream reassembly properly.
*
* \retval On success it returns 1 and on failure 0.
*/
static int StreamTcpReassembleTest47 (void)
{
Packet *p = PacketGetFromAlloc();
FAIL_IF(unlikely(p == NULL));
Flow *f = NULL;
TCPHdr tcph;
TcpSession ssn;
ThreadVars tv;
PacketQueue pq;
memset(&pq,0,sizeof(PacketQueue));
memset(&tcph, 0, sizeof (TCPHdr));
memset(&tv, 0, sizeof (ThreadVars));
StreamTcpInitConfig(TRUE);
StreamTcpUTSetupSession(&ssn);
TcpReassemblyThreadCtx *ra_ctx = StreamTcpReassembleInitThreadCtx(NULL);
uint8_t httpbuf1[] = "GET /EVILSUFF HTTP/1.1\r\n\r\n";
uint32_t httplen1 = sizeof(httpbuf1) - 1; /* minus the \0 */
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SET_ISN(&ssn.server, 572799781UL);
ssn.server.last_ack = 572799782UL;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
SET_ISN(&ssn.client, 4294967289UL);
ssn.client.last_ack = 21;
f = UTHBuildFlow(AF_INET, "1.2.3.4", "1.2.3.5", 200, 220);
FAIL_IF(f == NULL);
f->protoctx = &ssn;
App layer API rewritten. The main files in question are: app-layer.[ch], app-layer-detect-proto.[ch] and app-layer-parser.[ch]. Things addressed in this commit: - Brings out a proper separation between protocol detection phase and the parser phase. - The dns app layer now is registered such that we don't use "dnstcp" and "dnsudp" in the rules. A user who previously wrote a rule like this - "alert dnstcp....." or "alert dnsudp....." would now have to use, alert dns (ipproto:tcp;) or alert udp (app-layer-protocol:dns;) or alert ip (ipproto:udp; app-layer-protocol:dns;) The same rules extend to other another such protocol, dcerpc. - The app layer parser api now takes in the ipproto while registering callbacks. - The app inspection/detection engine also takes an ipproto. - All app layer parser functions now take direction as STREAM_TOSERVER or STREAM_TOCLIENT, as opposed to 0 or 1, which was taken by some of the functions. - FlowInitialize() and FlowRecycle() now resets proto to 0. This is needed by unittests, which would try to clean the flow, and that would call the api, AppLayerParserCleanupParserState(), which would try to clean the app state, but the app layer now needs an ipproto to figure out which api to internally call to clean the state, and if the ipproto is 0, it would return without trying to clean the state. - A lot of unittests are now updated where if they are using a flow and they need to use the app layer, we would set a flow ipproto. - The "app-layer" section in the yaml conf has also been updated as well.
12 years ago
f->proto = IPPROTO_TCP;
p->flow = f;
tcph.th_win = htons(5480);
ssn.state = TCP_ESTABLISHED;
TcpStream *s = NULL;
uint8_t cnt = 0;
for (cnt=0; cnt < httplen1; cnt++) {
tcph.th_seq = htonl(ssn.client.isn + 1 + cnt);
tcph.th_ack = htonl(572799782UL);
tcph.th_flags = TH_ACK|TH_PUSH;
p->tcph = &tcph;
p->flowflags = FLOW_PKT_TOSERVER;
p->payload = &httpbuf1[cnt];
p->payload_len = 1;
s = &ssn.client;
FAIL_IF(StreamTcpReassembleHandleSegment(&tv, ra_ctx, &ssn, s, p, &pq) == -1);
p->flowflags = FLOW_PKT_TOCLIENT;
p->payload = NULL;
p->payload_len = 0;
tcph.th_seq = htonl(572799782UL);
tcph.th_ack = htonl(ssn.client.isn + 1 + cnt);
tcph.th_flags = TH_ACK;
p->tcph = &tcph;
s = &ssn.server;
FAIL_IF(StreamTcpReassembleHandleSegment(&tv, ra_ctx, &ssn, s, p, &pq) == -1);
}
FAIL_IF(f->alproto != ALPROTO_HTTP);
StreamTcpUTClearSession(&ssn);
StreamTcpReassembleFreeThreadCtx(ra_ctx);
StreamTcpFreeConfig(TRUE);
SCFree(p);
UTHFreeFlow(f);
PASS;
}
/** \test 3 in order segments in inline reassembly */
static int StreamTcpReassembleInlineTest01(void)
{
int ret = 0;
TcpReassemblyThreadCtx *ra_ctx = NULL;
ThreadVars tv;
TcpSession ssn;
Flow f;
memset(&tv, 0x00, sizeof(tv));
StreamTcpUTInit(&ra_ctx);
StreamTcpUTInitInline();
StreamTcpUTSetupSession(&ssn);
StreamTcpUTSetupStream(&ssn.client, 1);
FLOW_INITIALIZE(&f);
uint8_t payload[] = { 'C', 'C', 'C', 'C', 'C' };
Packet *p = UTHBuildPacketReal(payload, 5, IPPROTO_TCP, "1.1.1.1", "2.2.2.2", 1024, 80);
if (p == NULL) {
printf("couldn't get a packet: ");
goto end;
}
p->tcph->th_seq = htonl(12);
p->flow = &f;
FLOWLOCK_WRLOCK(&f);
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 2, 'A', 5) == -1) {
printf("failed to add segment 1: ");
goto end;
}
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 7, 'B', 5) == -1) {
printf("failed to add segment 2: ");
goto end;
}
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 12, 'C', 5) == -1) {
printf("failed to add segment 3: ");
goto end;
}
ssn.client.next_seq = 17;
ret = 1;
end:
FLOWLOCK_UNLOCK(&f);
FLOW_DESTROY(&f);
UTHFreePacket(p);
StreamTcpUTClearSession(&ssn);
StreamTcpUTDeinit(ra_ctx);
return ret;
}
/** \test 3 in order segments, then reassemble, add one more and reassemble again.
* test the sliding window reassembly.
*/
static int StreamTcpReassembleInlineTest02(void)
{
int ret = 0;
TcpReassemblyThreadCtx *ra_ctx = NULL;
ThreadVars tv;
TcpSession ssn;
Flow f;
memset(&tv, 0x00, sizeof(tv));
StreamTcpUTInit(&ra_ctx);
StreamTcpUTInitInline();
StreamTcpUTSetupSession(&ssn);
StreamTcpUTSetupStream(&ssn.client, 1);
FLOW_INITIALIZE(&f);
uint8_t payload[] = { 'C', 'C', 'C', 'C', 'C' };
Packet *p = UTHBuildPacketReal(payload, 5, IPPROTO_TCP, "1.1.1.1", "2.2.2.2", 1024, 80);
if (p == NULL) {
printf("couldn't get a packet: ");
goto end;
}
p->tcph->th_seq = htonl(12);
p->flow = &f;
FLOWLOCK_WRLOCK(&f);
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 2, 'A', 5) == -1) {
printf("failed to add segment 1: ");
goto end;
}
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 7, 'B', 5) == -1) {
printf("failed to add segment 2: ");
goto end;
}
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 12, 'C', 5) == -1) {
printf("failed to add segment 3: ");
goto end;
}
ssn.client.next_seq = 17;
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 17, 'D', 5) == -1) {
printf("failed to add segment 4: ");
goto end;
}
ssn.client.next_seq = 22;
ret = 1;
end:
FLOWLOCK_UNLOCK(&f);
FLOW_DESTROY(&f);
UTHFreePacket(p);
StreamTcpUTClearSession(&ssn);
StreamTcpUTDeinit(ra_ctx);
return ret;
}
/** \test 3 in order segments, then reassemble, add one more and reassemble again.
* test the sliding window reassembly with a small window size so that we
* cutting off at the start (left edge)
*/
static int StreamTcpReassembleInlineTest03(void)
{
int ret = 0;
TcpReassemblyThreadCtx *ra_ctx = NULL;
ThreadVars tv;
TcpSession ssn;
Flow f;
memset(&tv, 0x00, sizeof(tv));
StreamTcpUTInit(&ra_ctx);
StreamTcpUTInitInline();
StreamTcpUTSetupSession(&ssn);
StreamTcpUTSetupStream(&ssn.client, 1);
FLOW_INITIALIZE(&f);
stream_config.reassembly_toserver_chunk_size = 15;
uint8_t payload[] = { 'C', 'C', 'C', 'C', 'C' };
Packet *p = UTHBuildPacketReal(payload, 5, IPPROTO_TCP, "1.1.1.1", "2.2.2.2", 1024, 80);
if (p == NULL) {
printf("couldn't get a packet: ");
goto end;
}
p->tcph->th_seq = htonl(12);
p->flow = &f;
p->flowflags |= FLOW_PKT_TOSERVER;
FLOWLOCK_WRLOCK(&f);
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 2, 'A', 5) == -1) {
printf("failed to add segment 1: ");
goto end;
}
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 7, 'B', 5) == -1) {
printf("failed to add segment 2: ");
goto end;
}
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 12, 'C', 5) == -1) {
printf("failed to add segment 3: ");
goto end;
}
ssn.client.next_seq = 17;
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 17, 'D', 5) == -1) {
printf("failed to add segment 4: ");
goto end;
}
ssn.client.next_seq = 22;
p->tcph->th_seq = htonl(17);
ret = 1;
end:
FLOWLOCK_UNLOCK(&f);
FLOW_DESTROY(&f);
UTHFreePacket(p);
StreamTcpUTClearSession(&ssn);
StreamTcpUTDeinit(ra_ctx);
return ret;
}
/** \test 3 in order segments, then reassemble, add one more and reassemble again.
* test the sliding window reassembly with a small window size so that we
* cutting off at the start (left edge) with small packet overlap.
*/
static int StreamTcpReassembleInlineTest04(void)
{
int ret = 0;
TcpReassemblyThreadCtx *ra_ctx = NULL;
ThreadVars tv;
TcpSession ssn;
Flow f;
memset(&tv, 0x00, sizeof(tv));
StreamTcpUTInit(&ra_ctx);
StreamTcpUTInitInline();
StreamTcpUTSetupSession(&ssn);
StreamTcpUTSetupStream(&ssn.client, 1);
FLOW_INITIALIZE(&f);
stream_config.reassembly_toserver_chunk_size = 16;
uint8_t payload[] = { 'C', 'C', 'C', 'C', 'C' };
Packet *p = UTHBuildPacketReal(payload, 5, IPPROTO_TCP, "1.1.1.1", "2.2.2.2", 1024, 80);
if (p == NULL) {
printf("couldn't get a packet: ");
goto end;
}
p->tcph->th_seq = htonl(12);
p->flow = &f;
p->flowflags |= FLOW_PKT_TOSERVER;
FLOWLOCK_WRLOCK(&f);
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 2, 'A', 5) == -1) {
printf("failed to add segment 1: ");
goto end;
}
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 7, 'B', 5) == -1) {
printf("failed to add segment 2: ");
goto end;
}
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 12, 'C', 5) == -1) {
printf("failed to add segment 3: ");
goto end;
}
ssn.client.next_seq = 17;
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 17, 'D', 5) == -1) {
printf("failed to add segment 4: ");
goto end;
}
ssn.client.next_seq = 22;
p->tcph->th_seq = htonl(17);
ret = 1;
end:
FLOWLOCK_UNLOCK(&f);
FLOW_DESTROY(&f);
UTHFreePacket(p);
StreamTcpUTClearSession(&ssn);
StreamTcpUTDeinit(ra_ctx);
return ret;
}
/** \test 3 in order segments, then reassemble, add one more and reassemble again.
* test the sliding window reassembly with a small window size so that we
* cutting off at the start (left edge). Test if the first segment is
* removed from the list.
*/
static int StreamTcpReassembleInlineTest08(void)
{
TcpReassemblyThreadCtx *ra_ctx = NULL;
ThreadVars tv;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
memset(&tv, 0x00, sizeof(tv));
TcpSession ssn;
Flow f;
StreamTcpUTInit(&ra_ctx);
StreamTcpUTInitInline();
StreamTcpUTSetupSession(&ssn);
StreamTcpUTSetupStream(&ssn.client, 1);
FLOW_INITIALIZE(&f);
stream_config.reassembly_toserver_chunk_size = 15;
ssn.client.flags |= STREAMTCP_STREAM_FLAG_GAP;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
f.protoctx = &ssn;
uint8_t payload[] = { 'C', 'C', 'C', 'C', 'C' };
Packet *p = UTHBuildPacketReal(payload, 5, IPPROTO_TCP, "1.1.1.1", "2.2.2.2", 1024, 80);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
FAIL_IF(p == NULL);
p->tcph->th_seq = htonl(12);
p->flow = &f;
p->flowflags |= FLOW_PKT_TOSERVER;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
FAIL_IF(StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 2, 'A', 5) == -1);
FAIL_IF(StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 7, 'B', 5) == -1);
FAIL_IF(StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 12, 'C', 5) == -1);
ssn.client.next_seq = 17;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
FAIL_IF(StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 17, 'D', 5) == -1);
ssn.client.next_seq = 22;
p->tcph->th_seq = htonl(17);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
StreamTcpPruneSession(&f, STREAM_TOSERVER);
FAIL_IF (ssn.client.seg_list->seq != 2);
FLOW_DESTROY(&f);
UTHFreePacket(p);
StreamTcpUTClearSession(&ssn);
StreamTcpUTDeinit(ra_ctx);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
PASS;
}
/** \test 3 in order segments, then reassemble, add one more and reassemble again.
* test the sliding window reassembly with a small window size so that we
* cutting off at the start (left edge). Test if the first segment is
* removed from the list.
*/
static int StreamTcpReassembleInlineTest09(void)
{
int ret = 0;
TcpReassemblyThreadCtx *ra_ctx = NULL;
ThreadVars tv;
TcpSession ssn;
Flow f;
memset(&tv, 0x00, sizeof(tv));
StreamTcpUTInit(&ra_ctx);
StreamTcpUTInitInline();
StreamTcpUTSetupSession(&ssn);
StreamTcpUTSetupStream(&ssn.client, 1);
FLOW_INITIALIZE(&f);
stream_config.reassembly_toserver_chunk_size = 20;
ssn.client.flags |= STREAMTCP_STREAM_FLAG_GAP;
uint8_t payload[] = { 'C', 'C', 'C', 'C', 'C' };
Packet *p = UTHBuildPacketReal(payload, 5, IPPROTO_TCP, "1.1.1.1", "2.2.2.2", 1024, 80);
if (p == NULL) {
printf("couldn't get a packet: ");
goto end;
}
p->tcph->th_seq = htonl(17);
p->flow = &f;
p->flowflags |= FLOW_PKT_TOSERVER;
FLOWLOCK_WRLOCK(&f);
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 2, 'A', 5) == -1) {
printf("failed to add segment 1: ");
goto end;
}
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 7, 'B', 5) == -1) {
printf("failed to add segment 2: ");
goto end;
}
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 17, 'D', 5) == -1) {
printf("failed to add segment 3: ");
goto end;
}
ssn.client.next_seq = 12;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
ssn.client.last_ack = 10;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
/* close the GAP and see if we properly reassemble and update base_seq */
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 12, 'C', 5) == -1) {
printf("failed to add segment 4: ");
goto end;
}
ssn.client.next_seq = 22;
p->tcph->th_seq = htonl(12);
if (ssn.client.seg_list->seq != 2) {
printf("expected segment 1 (seq 2) to be first in the list, got seq %"PRIu32": ", ssn.client.seg_list->seq);
goto end;
}
ret = 1;
end:
FLOWLOCK_UNLOCK(&f);
FLOW_DESTROY(&f);
UTHFreePacket(p);
StreamTcpUTClearSession(&ssn);
StreamTcpUTDeinit(ra_ctx);
return ret;
}
/** \test App Layer reassembly.
*/
static int StreamTcpReassembleInlineTest10(void)
{
int ret = 0;
TcpReassemblyThreadCtx *ra_ctx = NULL;
ThreadVars tv;
TcpSession ssn;
Flow *f = NULL;
Packet *p = NULL;
memset(&tv, 0x00, sizeof(tv));
StreamTcpUTInit(&ra_ctx);
StreamTcpUTInitInline();
StreamTcpUTSetupSession(&ssn);
StreamTcpUTSetupStream(&ssn.server, 1);
ssn.server.last_ack = 2;
StreamTcpUTSetupStream(&ssn.client, 1);
ssn.client.last_ack = 2;
ssn.data_first_seen_dir = STREAM_TOSERVER;
f = UTHBuildFlow(AF_INET, "1.1.1.1", "2.2.2.2", 1024, 80);
if (f == NULL)
goto end;
f->protoctx = &ssn;
App layer API rewritten. The main files in question are: app-layer.[ch], app-layer-detect-proto.[ch] and app-layer-parser.[ch]. Things addressed in this commit: - Brings out a proper separation between protocol detection phase and the parser phase. - The dns app layer now is registered such that we don't use "dnstcp" and "dnsudp" in the rules. A user who previously wrote a rule like this - "alert dnstcp....." or "alert dnsudp....." would now have to use, alert dns (ipproto:tcp;) or alert udp (app-layer-protocol:dns;) or alert ip (ipproto:udp; app-layer-protocol:dns;) The same rules extend to other another such protocol, dcerpc. - The app layer parser api now takes in the ipproto while registering callbacks. - The app inspection/detection engine also takes an ipproto. - All app layer parser functions now take direction as STREAM_TOSERVER or STREAM_TOCLIENT, as opposed to 0 or 1, which was taken by some of the functions. - FlowInitialize() and FlowRecycle() now resets proto to 0. This is needed by unittests, which would try to clean the flow, and that would call the api, AppLayerParserCleanupParserState(), which would try to clean the app state, but the app layer now needs an ipproto to figure out which api to internally call to clean the state, and if the ipproto is 0, it would return without trying to clean the state. - A lot of unittests are now updated where if they are using a flow and they need to use the app layer, we would set a flow ipproto. - The "app-layer" section in the yaml conf has also been updated as well.
12 years ago
f->proto = IPPROTO_TCP;
uint8_t stream_payload1[] = "GE";
uint8_t stream_payload2[] = "T /";
uint8_t stream_payload3[] = "HTTP/1.0\r\n\r\n";
p = UTHBuildPacketReal(stream_payload3, 12, IPPROTO_TCP, "1.1.1.1", "2.2.2.2", 1024, 80);
if (p == NULL) {
printf("couldn't get a packet: ");
goto end;
}
p->tcph->th_seq = htonl(7);
p->flow = f;
p->flowflags = FLOW_PKT_TOSERVER;
FLOWLOCK_WRLOCK(f);
if (StreamTcpUTAddSegmentWithPayload(&tv, ra_ctx, &ssn.client, 2, stream_payload1, 2) == -1) {
printf("failed to add segment 1: ");
goto end;
}
ssn.client.next_seq = 4;
int r = StreamTcpReassembleAppLayer(&tv, ra_ctx, &ssn, &ssn.client, p, UPDATE_DIR_PACKET);
if (r < 0) {
printf("StreamTcpReassembleAppLayer failed: ");
goto end;
}
/* ssn.server.ra_app_base_seq should be isn here. */
if (ssn.client.base_seq != 2 || ssn.client.base_seq != ssn.client.isn+1) {
printf("expected ra_app_base_seq 1, got %u: ", ssn.client.base_seq);
goto end;
}
if (StreamTcpUTAddSegmentWithPayload(&tv, ra_ctx, &ssn.client, 4, stream_payload2, 3) == -1) {
printf("failed to add segment 2: ");
goto end;
}
if (StreamTcpUTAddSegmentWithPayload(&tv, ra_ctx, &ssn.client, 7, stream_payload3, 12) == -1) {
printf("failed to add segment 3: ");
goto end;
}
ssn.client.next_seq = 19;
r = StreamTcpReassembleAppLayer(&tv, ra_ctx, &ssn, &ssn.client, p, UPDATE_DIR_PACKET);
if (r < 0) {
printf("StreamTcpReassembleAppLayer failed: ");
goto end;
}
FAIL_IF_NOT(STREAM_APP_PROGRESS(&ssn.client) == 17);
ret = 1;
end:
UTHFreePacket(p);
StreamTcpUTClearSession(&ssn);
StreamTcpUTDeinit(ra_ctx);
FLOWLOCK_UNLOCK(f);
UTHFreeFlow(f);
return ret;
}
/** \test test insert with overlap
*/
static int StreamTcpReassembleInsertTest01(void)
{
TcpReassemblyThreadCtx *ra_ctx = NULL;
ThreadVars tv;
TcpSession ssn;
Flow f;
memset(&tv, 0x00, sizeof(tv));
StreamTcpUTInit(&ra_ctx);
StreamTcpUTSetupSession(&ssn);
StreamTcpUTSetupStream(&ssn.client, 1);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
ssn.client.os_policy = OS_POLICY_LAST;
FLOW_INITIALIZE(&f);
uint8_t payload[] = { 'C', 'C', 'C', 'C', 'C' };
Packet *p = UTHBuildPacketReal(payload, 5, IPPROTO_TCP, "1.1.1.1", "2.2.2.2", 1024, 80);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
FAIL_IF(p == NULL);
p->tcph->th_seq = htonl(12);
p->flow = &f;
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
FAIL_IF(StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 2, 'A', 5) == -1);
FAIL_IF(StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 7, 'B', 5) == -1);
FAIL_IF(StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 14, 'D', 2) == -1);
FAIL_IF(StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 16, 'D', 6) == -1);
FAIL_IF(StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 12, 'C', 5) == -1);
ssn.client.next_seq = 21;
FLOW_DESTROY(&f);
UTHFreePacket(p);
StreamTcpUTClearSession(&ssn);
StreamTcpUTDeinit(ra_ctx);
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
PASS;
}
/** \test test insert with overlaps
*/
static int StreamTcpReassembleInsertTest02(void)
{
int ret = 0;
TcpReassemblyThreadCtx *ra_ctx = NULL;
ThreadVars tv;
TcpSession ssn;
memset(&tv, 0x00, sizeof(tv));
StreamTcpUTInit(&ra_ctx);
StreamTcpUTSetupSession(&ssn);
StreamTcpUTSetupStream(&ssn.client, 1);
int i;
for (i = 2; i < 10; i++) {
int len;
len = i % 2;
if (len == 0)
len = 1;
int seq;
seq = i * 10;
if (seq < 2)
seq = 2;
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, seq, 'A', len) == -1) {
printf("failed to add segment 1: ");
goto end;
}
}
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 2, 'B', 1024) == -1) {
printf("failed to add segment 2: ");
goto end;
}
ret = 1;
end:
StreamTcpUTClearSession(&ssn);
StreamTcpUTDeinit(ra_ctx);
return ret;
}
/** \test test insert with overlaps
*/
static int StreamTcpReassembleInsertTest03(void)
{
int ret = 0;
TcpReassemblyThreadCtx *ra_ctx = NULL;
ThreadVars tv;
TcpSession ssn;
memset(&tv, 0x00, sizeof(tv));
StreamTcpUTInit(&ra_ctx);
StreamTcpUTSetupSession(&ssn);
StreamTcpUTSetupStream(&ssn.client, 1);
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, 2, 'A', 1024) == -1) {
printf("failed to add segment 2: ");
goto end;
}
int i;
for (i = 2; i < 10; i++) {
int len;
len = i % 2;
if (len == 0)
len = 1;
int seq;
seq = i * 10;
if (seq < 2)
seq = 2;
if (StreamTcpUTAddSegmentWithByte(&tv, ra_ctx, &ssn.client, seq, 'B', len) == -1) {
printf("failed to add segment 2: ");
goto end;
}
}
ret = 1;
end:
StreamTcpUTClearSession(&ssn);
StreamTcpUTDeinit(ra_ctx);
return ret;
}
#include "tests/stream-tcp-reassemble.c"
#endif /* UNITTESTS */
/** \brief The Function Register the Unit tests to test the reassembly engine
* for various OS policies.
*/
void StreamTcpReassembleRegisterTests(void)
{
#ifdef UNITTESTS
UtRegisterTest("StreamTcpReassembleTest25 -- Gap at Start Reassembly Test",
StreamTcpReassembleTest25);
UtRegisterTest("StreamTcpReassembleTest26 -- Gap at middle Reassembly Test",
StreamTcpReassembleTest26);
UtRegisterTest("StreamTcpReassembleTest27 -- Gap at after Reassembly Test",
StreamTcpReassembleTest27);
UtRegisterTest("StreamTcpReassembleTest28 -- Gap at Start IDS missed packet Reassembly Test",
StreamTcpReassembleTest28);
UtRegisterTest("StreamTcpReassembleTest29 -- Gap at Middle IDS missed packet Reassembly Test",
StreamTcpReassembleTest29);
UtRegisterTest("StreamTcpReassembleTest33 -- Bug test",
StreamTcpReassembleTest33);
UtRegisterTest("StreamTcpReassembleTest34 -- Bug test",
StreamTcpReassembleTest34);
UtRegisterTest("StreamTcpReassembleTest37 -- Bug76 test",
StreamTcpReassembleTest37);
UtRegisterTest("StreamTcpReassembleTest39 -- app proto test",
StreamTcpReassembleTest39);
UtRegisterTest("StreamTcpReassembleTest40 -- app proto test",
StreamTcpReassembleTest40);
UtRegisterTest("StreamTcpReassembleTest44 -- Memcap Test",
StreamTcpReassembleTest44);
UtRegisterTest("StreamTcpReassembleTest45 -- Depth Test",
StreamTcpReassembleTest45);
UtRegisterTest("StreamTcpReassembleTest46 -- Depth Test",
StreamTcpReassembleTest46);
UtRegisterTest("StreamTcpReassembleTest47 -- TCP Sequence Wraparound Test",
StreamTcpReassembleTest47);
UtRegisterTest("StreamTcpReassembleInlineTest01 -- inline RAW ra",
StreamTcpReassembleInlineTest01);
UtRegisterTest("StreamTcpReassembleInlineTest02 -- inline RAW ra 2",
StreamTcpReassembleInlineTest02);
UtRegisterTest("StreamTcpReassembleInlineTest03 -- inline RAW ra 3",
StreamTcpReassembleInlineTest03);
UtRegisterTest("StreamTcpReassembleInlineTest04 -- inline RAW ra 4",
StreamTcpReassembleInlineTest04);
UtRegisterTest("StreamTcpReassembleInlineTest08 -- inline RAW ra 8 cleanup",
StreamTcpReassembleInlineTest08);
UtRegisterTest("StreamTcpReassembleInlineTest09 -- inline RAW ra 9 GAP cleanup",
StreamTcpReassembleInlineTest09);
UtRegisterTest("StreamTcpReassembleInlineTest10 -- inline APP ra 10",
StreamTcpReassembleInlineTest10);
UtRegisterTest("StreamTcpReassembleInsertTest01 -- insert with overlap",
StreamTcpReassembleInsertTest01);
UtRegisterTest("StreamTcpReassembleInsertTest02 -- insert with overlap",
StreamTcpReassembleInsertTest02);
UtRegisterTest("StreamTcpReassembleInsertTest03 -- insert with overlap",
StreamTcpReassembleInsertTest03);
StreamTcpInlineRegisterTests();
StreamTcpUtilRegisterTests();
tcp: streaming implementation Make stream engine use the streaming buffer API for it's data storage. This means that the data is stored in a single reassembled sliding buffer. The subleties of the reassembly, e.g. overlap handling, are taken care of at segment insertion. The TcpSegments now have a StreamingBufferSegment that contains an offset and a length. Using this the segment data can be retrieved per segment. Redo segment insertion. The insertion code is moved to it's own file and is simplified a lot. A major difference with the previous implementation is that the segment list now contains overlapping segments if the traffic is that way. Previously there could be more and smaller segments in the memory list than what was seen on the wire. Due to the matching of in memory segments and on the wire segments, the overlap with different data detection (potential mots attacks) is much more accurate. Raw and App reassembly progress is no longer tracked per segment using flags, but there is now a progress tracker in the TcpStream for each. When pruning we make sure we don't slide beyond in-use segments. When both app-layer and raw inspection are beyond the start of the segment list, the segments might not be freed even though the data in the streaming buffer is already gone. This is caused by the 'in-use' status that the segments can implicitly have. This patch accounts for that when calculating the 'left_edge' of the streaming window. Raw reassembly still sets up 'StreamMsg' objects for content inspection. They are set up based on either the full StreamingBuffer, or based on the StreamingBufferBlocks if there are gaps in the data. Reworked 'stream needs work' logic. When a flow times out the flow engine checks whether a TCP flow still needs work. The StreamNeedsReassembly function is used to test if a stream still has unreassembled segments or uninspected stream chunks. This patch updates the function to consider the app and/or raw progress. It also cleans the function up and adds more meaningful debug messages. Finally it makes it non-inline. Unittests have been overhauled, and partly moved into their own files. Remove lots of dead code.
10 years ago
StreamTcpListRegisterTests();
StreamTcpReassembleRawRegisterTests();
#endif /* UNITTESTS */
}