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-list.c

987 lines
34 KiB
C

/* Copyright (C) 2007-2022 Open Information Security Foundation
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.
11 years ago
*
* 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
*
* Segment list functions for insertions, overlap handling, removal and
* more.
*/
#include "suricata-common.h"
#include "rust.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.
11 years ago
#include "stream-tcp-private.h"
#include "stream-tcp.h"
#include "stream-tcp-reassemble.h"
#include "stream-tcp-inline.h"
#include "stream-tcp-list.h"
#include "util-streaming-buffer.h"
#include "util-print.h"
#include "util-validate.h"
#include "app-layer-frames.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.
11 years ago
static void StreamTcpRemoveSegmentFromStream(TcpStream *stream, TcpSegment *seg);
static int check_overlap_different_data = 0;
void StreamTcpReassembleConfigEnableOverlapCheck(void)
{
check_overlap_different_data = 1;
}
/*
* Inserts and overlap handling
*/
RB_GENERATE(TCPSEG, TcpSegment, rb, TcpSegmentCompare);
int TcpSegmentCompare(struct TcpSegment *a, struct TcpSegment *b)
{
if (SEQ_GT(a->seq, b->seq))
return 1;
else if (SEQ_LT(a->seq, b->seq))
return -1;
else {
if (a->payload_len == b->payload_len)
return 0;
else if (a->payload_len > b->payload_len)
return 1;
else
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.
11 years ago
/** \internal
* \brief insert segment data into the streaming buffer
* \param seg segment to store stream offset in
* \param data segment data after overlap handling (if any)
* \param data_len data length
*
* \return 0 on success
* \return -1 on error (memory allocation error)
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.
11 years ago
*/
static inline int InsertSegmentDataCustom(TcpStream *stream, TcpSegment *seg, uint8_t *data, uint16_t data_len)
{
uint64_t stream_offset;
uint32_t data_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.
11 years ago
if (likely(SEQ_GEQ(seg->seq, stream->base_seq))) {
stream_offset = STREAM_BASE_OFFSET(stream) + (seg->seq - 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.
11 years ago
data_offset = 0;
} else {
/* segment is partly before base_seq */
data_offset = stream->base_seq - seg->seq;
stream_offset = STREAM_BASE_OFFSET(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.
11 years ago
}
SCLogDebug("stream %p buffer %p, stream_offset %"PRIu64", "
"data_offset %"PRIu16", SEQ %u BASE %u, data_len %u",
stream, &stream->sb, stream_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.
11 years ago
data_offset, seg->seq, stream->base_seq, data_len);
DEBUG_VALIDATE_BUG_ON(data_offset > data_len);
if (data_len <= data_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.
11 years ago
SCReturnInt(0);
}
int ret = StreamingBufferInsertAt(
&stream->sb, &seg->sbseg, data + data_offset, data_len - data_offset, stream_offset);
if (ret != 0) {
/* StreamingBufferInsertAt can return -2 only if the offset is wrong, which should be
* impossible in this path. */
DEBUG_VALIDATE_BUG_ON(ret != -1);
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.
11 years ago
}
#ifdef DEBUG
{
const uint8_t *mydata;
uint32_t mydata_len;
uint64_t mydata_offset;
StreamingBufferGetData(&stream->sb, &mydata, &mydata_len, &mydata_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.
11 years ago
SCLogDebug("stream %p seg %p data in buffer %p of len %u and offset %"PRIu64,
stream, seg, &stream->sb, mydata_len, mydata_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.
11 years ago
//PrintRawDataFp(stdout, mydata, mydata_len);
}
#endif
SCReturnInt(0);
}
/** \internal
* \brief check if this segments overlaps with an in-tree seg.
* \retval true
* \retval false
*/
static inline bool CheckOverlap(struct TCPSEG *tree, TcpSegment *seg)
{
const uint32_t re = SEG_SEQ_RIGHT_EDGE(seg);
SCLogDebug("start. SEQ %u payload_len %u. Right edge: %u. Seg %p",
seg->seq, seg->payload_len, re, seg);
/* check forward */
TcpSegment *next = TCPSEG_RB_NEXT(seg);
if (next) {
// next has same seq, so data must overlap
if (SEQ_EQ(next->seq, seg->seq))
return true;
// our right edge is beyond next seq, overlap
if (SEQ_GT(re, next->seq))
return true;
}
/* check backwards */
TcpSegment *prev = TCPSEG_RB_PREV(seg);
if (prev) {
// prev has same seq, so data must overlap
if (SEQ_EQ(prev->seq, seg->seq))
return true;
// prev's right edge is beyond our seq, overlap
const uint32_t prev_re = SEG_SEQ_RIGHT_EDGE(prev);
if (SEQ_GT(prev_re, seg->seq))
return true;
}
SCLogDebug("no overlap");
return false;
}
/** \internal
* \brief insert the segment into the proper place in the tree
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.
11 years ago
* don't worry about the data or overlaps
*
* \retval 2 not inserted, data overlap
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.
11 years ago
* \retval 1 inserted with overlap detected
* \retval 0 inserted, no overlap
* \retval -ENOMEM memcap reached
* \retval -EINVAL seg out of seq range
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.
11 years ago
*/
static int DoInsertSegment (TcpStream *stream, TcpSegment *seg, TcpSegment **dup_seg, 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.
11 years ago
{
/* in lossy traffic, we can get here with the wrong sequence numbers */
if (SEQ_LEQ(SEG_SEQ_RIGHT_EDGE(seg), stream->base_seq)) {
return -EINVAL;
}
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.
11 years ago
/* fast track */
if (RB_EMPTY(&stream->seg_tree)) {
SCLogDebug("empty tree, inserting seg %p seq %" PRIu32 ", "
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.
11 years ago
"len %" PRIu32 "", seg, seg->seq, TCP_SEG_LEN(seg));
TCPSEG_RB_INSERT(&stream->seg_tree, seg);
stream->segs_right_edge = SEG_SEQ_RIGHT_EDGE(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.
11 years ago
return 0;
}
/* insert and then check if there was any overlap with other segments */
TcpSegment *res = TCPSEG_RB_INSERT(&stream->seg_tree, seg);
if (res) {
SCLogDebug("seg has a duplicate in the tree seq %u/%u",
res->seq, res->payload_len);
/* exact duplicate SEQ + payload_len */
*dup_seg = res;
return 2; // duplicate has overlap by definition.
} else {
if (SEQ_GT(SEG_SEQ_RIGHT_EDGE(seg), stream->segs_right_edge))
stream->segs_right_edge = SEG_SEQ_RIGHT_EDGE(seg);
/* insert succeeded, now check if we overlap with someone */
if (CheckOverlap(&stream->seg_tree, seg) == true) {
SCLogDebug("seg %u has overlap in the tree", seg->seq);
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.
11 years ago
}
}
SCLogDebug("seg %u: no overlap", seg->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.
11 years ago
return 0;
}
/** \internal
* \brief handle overlap per list segment
*
* For a list segment handle the overlap according to the policy.
*
* The 'buf' parameter points to the memory that will be inserted into
* the stream after the overlap checks are complete. As it will
* unconditionally overwrite whats in the stream now, the overlap
* policies are applied to this buffer. It starts with the 'new' data,
* so when the policy states 'old' data has to be used, 'buf' is
* updated to contain the 'old' data here.
*
* \param buf stack allocated buffer sized p->payload_len that will be
* inserted into the stream buffer
*
* \retval 1 if data was different
* \retval 0 data was the same or we didn't check for differences
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.
11 years ago
*/
static int DoHandleDataOverlap(TcpStream *stream, const TcpSegment *list,
const TcpSegment *seg, uint8_t *buf, 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.
11 years ago
{
SCLogDebug("handle overlap for segment %p seq %u len %u re %u, "
"list segment %p seq %u len %u re %u", seg, seg->seq,
p->payload_len, SEG_SEQ_RIGHT_EDGE(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.
11 years ago
list, list->seq, TCP_SEG_LEN(list), SEG_SEQ_RIGHT_EDGE(list));
int data_is_different = 0;
int use_new_data = 0;
if (StreamTcpInlineMode()) {
SCLogDebug("inline mode");
if (StreamTcpInlineSegmentCompare(stream, p, list) != 0) {
SCLogDebug("already accepted data not the same as packet data, rewrite packet");
StreamTcpInlineSegmentReplacePacket(stream, p, list);
data_is_different = 1;
/* in inline mode we check for different data unconditionally,
* but setting events still depends on config */
if (check_overlap_different_data) {
StreamTcpSetEvent(p, STREAM_REASSEMBLY_OVERLAP_DIFFERENT_DATA);
}
}
/* IDS mode */
} else {
if (check_overlap_different_data) {
if (StreamTcpInlineSegmentCompare(stream, p, list) != 0) {
SCLogDebug("data is different from what is in the list");
data_is_different = 1;
}
} else {
/* if we're not checking, assume it's different */
data_is_different = 1;
}
/* apply overlap policies */
if (stream->os_policy == OS_POLICY_LAST) {
/* buf will start with LAST data (from the segment),
* so if policy is LAST we're now done here. */
return (check_overlap_different_data && data_is_different);
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.
11 years ago
}
/* start at the same seq */
if (SEQ_EQ(seg->seq, list->seq)) {
SCLogDebug("seg starts at list segment");
if (SEQ_LT(SEG_SEQ_RIGHT_EDGE(seg), SEG_SEQ_RIGHT_EDGE(list))) {
SCLogDebug("seg ends before list end, end overlapped by list");
} else {
if (SEQ_GT(SEG_SEQ_RIGHT_EDGE(seg), SEG_SEQ_RIGHT_EDGE(list))) {
SCLogDebug("seg ends beyond list end, list overlapped and more");
switch (stream->os_policy) {
case OS_POLICY_LINUX:
if (data_is_different) {
use_new_data = 1;
}
break;
}
} else {
SCLogDebug("full overlap");
}
switch (stream->os_policy) {
case OS_POLICY_OLD_LINUX:
case OS_POLICY_SOLARIS:
case OS_POLICY_HPUX11:
if (data_is_different) {
use_new_data = 1;
}
break;
}
}
/* new seg starts before list segment */
} else if (SEQ_LT(seg->seq, list->seq)) {
SCLogDebug("seg starts before list segment");
if (SEQ_LT(SEG_SEQ_RIGHT_EDGE(seg), SEG_SEQ_RIGHT_EDGE(list))) {
SCLogDebug("seg ends before list end, end overlapped by list");
} else {
if (SEQ_GT(SEG_SEQ_RIGHT_EDGE(seg), SEG_SEQ_RIGHT_EDGE(list))) {
SCLogDebug("seg starts before and fully overlaps list and beyond");
} else {
SCLogDebug("seg starts before and fully overlaps list");
}
switch (stream->os_policy) {
case OS_POLICY_SOLARIS:
case OS_POLICY_HPUX11:
if (data_is_different) {
use_new_data = 1;
}
break;
}
}
switch (stream->os_policy) {
case OS_POLICY_BSD:
case OS_POLICY_HPUX10:
case OS_POLICY_IRIX:
case OS_POLICY_WINDOWS:
case OS_POLICY_WINDOWS2K3:
case OS_POLICY_OLD_LINUX:
case OS_POLICY_LINUX:
case OS_POLICY_MACOS:
if (data_is_different) {
use_new_data = 1;
}
break;
}
/* new seg starts after list segment */
} else { //if (SEQ_GT(seg->seq, list->seq)) {
SCLogDebug("seg starts after list segment");
if (SEQ_EQ(SEG_SEQ_RIGHT_EDGE(seg), SEG_SEQ_RIGHT_EDGE(list))) {
SCLogDebug("seg after and is fully overlapped by list");
} else if (SEQ_GT(SEG_SEQ_RIGHT_EDGE(seg), SEG_SEQ_RIGHT_EDGE(list))) {
SCLogDebug("seg starts after list and ends after list");
switch (stream->os_policy) {
case OS_POLICY_SOLARIS:
case OS_POLICY_HPUX11:
if (data_is_different) {
use_new_data = 1;
}
break;
}
} else {
SCLogDebug("seg starts after list and ends before list end");
}
}
}
SCLogDebug("data_is_different %s, use_new_data %s",
data_is_different ? "yes" : "no",
use_new_data ? "yes" : "no");
/* if the data is different and we don't want to use the new (seg)
* data, we have to update buf with the list data */
if (data_is_different && !use_new_data) {
/* we need to copy list into seg */
uint32_t list_offset = 0;
uint32_t seg_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.
11 years ago
uint32_t list_len;
uint16_t seg_len = p->payload_len;
uint32_t list_seq = list->seq;
const uint8_t *list_data;
StreamingBufferSegmentGetData(&stream->sb, &list->sbseg, &list_data, &list_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.
11 years ago
if (list_data == NULL || list_len == 0)
return 0;
BUG_ON(list_len > USHRT_MAX);
/* if list seg is partially before base_seq, list_len (from stream) and
* TCP_SEG_LEN(list) will not be the same */
if (SEQ_GEQ(list->seq, stream->base_seq)) {
;
} else {
list_seq = stream->base_seq;
list_len = SEG_SEQ_RIGHT_EDGE(list) - stream->base_seq;
}
if (SEQ_LT(seg->seq, list_seq)) {
seg_offset = list_seq - seg->seq;
seg_len -= seg_offset;
} else if (SEQ_GT(seg->seq, list_seq)) {
list_offset = seg->seq - list_seq;
list_len -= list_offset;
}
if (SEQ_LT(seg->seq + seg_offset + seg_len, list_seq + list_offset + list_len)) {
list_len -= (list_seq + list_offset + list_len) - (seg->seq + seg_offset + seg_len);
}
SCLogDebug("here goes nothing: list %u %u, seg %u %u", list_offset, list_len, seg_offset, seg_len);
//PrintRawDataFp(stdout, list_data + list_offset, list_len);
//PrintRawDataFp(stdout, buf + seg_offset, seg_len);
memcpy(buf + seg_offset, list_data + list_offset, list_len);
//PrintRawDataFp(stdout, buf, p->payload_len);
}
return (check_overlap_different_data && data_is_different);
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.
11 years ago
}
/** \internal
* \brief walk segment tree backwards to see if there are overlaps
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.
11 years ago
*
* Walk back from the current segment which is already in the tree.
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.
11 years ago
* We walk until we can't possibly overlap anymore.
*/
static int DoHandleDataCheckBackwards(TcpStream *stream,
TcpSegment *seg, uint8_t *buf, 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.
11 years ago
{
int retval = 0;
SCLogDebug("check tree backwards: insert data for segment %p seq %u len %u re %u",
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.
11 years ago
seg, seg->seq, TCP_SEG_LEN(seg), SEG_SEQ_RIGHT_EDGE(seg));
/* check backwards */
TcpSegment *tree_seg = NULL, *s = seg;
RB_FOREACH_REVERSE_FROM(tree_seg, TCPSEG, s) {
if (tree_seg == seg)
continue;
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.
11 years ago
int overlap = 0;
if (SEQ_LEQ(SEG_SEQ_RIGHT_EDGE(tree_seg), 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.
11 years ago
// segment entirely before base_seq
;
} else if (SEQ_LEQ(tree_seg->seq + tree_seg->payload_len, seg->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.
11 years ago
SCLogDebug("list segment too far to the left, no more overlap will be found");
break;
} else if (SEQ_GT(SEG_SEQ_RIGHT_EDGE(tree_seg), seg->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.
11 years ago
overlap = 1;
}
SCLogDebug("(back) tree seg %u len %u re %u overlap? %s",
tree_seg->seq, TCP_SEG_LEN(tree_seg),
SEG_SEQ_RIGHT_EDGE(tree_seg), overlap ? "yes" : "no");
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.
11 years ago
if (overlap) {
retval |= DoHandleDataOverlap(stream, tree_seg, seg, buf, 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.
11 years ago
}
}
return retval;
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.
11 years ago
}
/** \internal
* \brief walk segment tree in forward direction to see if there are overlaps
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.
11 years ago
*
* Walk forward from the current segment which is already in the tree.
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.
11 years ago
* We walk until the next segs start with a SEQ beyond our right edge.
*
* \retval 1 data was different
* \retval 0 data was the same
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.
11 years ago
*/
static int DoHandleDataCheckForward(TcpStream *stream,
TcpSegment *seg, uint8_t *buf, 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.
11 years ago
{
int retval = 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.
11 years ago
uint32_t seg_re = SEG_SEQ_RIGHT_EDGE(seg);
SCLogDebug("check list forward: insert data for segment %p seq %u len %u re %u",
seg, seg->seq, TCP_SEG_LEN(seg), seg_re);
TcpSegment *tree_seg = NULL, *s = seg;
RB_FOREACH_FROM(tree_seg, TCPSEG, s) {
if (tree_seg == seg)
continue;
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.
11 years ago
int overlap = 0;
if (SEQ_GT(seg_re, tree_seg->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.
11 years ago
overlap = 1;
else if (SEQ_LEQ(seg_re, tree_seg->seq)) {
SCLogDebug("tree segment %u too far ahead, "
"no more overlaps can happen", tree_seg->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.
11 years ago
break;
}
SCLogDebug("(fwd) in-tree seg %u len %u re %u overlap? %s",
tree_seg->seq, TCP_SEG_LEN(tree_seg),
SEG_SEQ_RIGHT_EDGE(tree_seg), overlap ? "yes" : "no");
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.
11 years ago
if (overlap) {
retval |= DoHandleDataOverlap(stream, tree_seg, seg, buf, 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.
11 years ago
}
}
return retval;
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.
11 years ago
}
/**
* \param tree_seg in-tree duplicate of `seg`
* \retval res 0 ok, -1 insertion error due to memcap
*/
static int DoHandleData(ThreadVars *tv, TcpReassemblyThreadCtx *ra_ctx,
TcpStream *stream, TcpSegment *seg, TcpSegment *tree_seg, 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.
11 years ago
{
int result = 0;
TcpSegment *handle = 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.
11 years ago
SCLogDebug("insert data for segment %p seq %u len %u re %u",
seg, seg->seq, TCP_SEG_LEN(seg), SEG_SEQ_RIGHT_EDGE(seg));
/* create temporary buffer to contain the data we will insert. Overlap
* handling may update it. By using this we don't have to track whether
* parts of the data are already inserted or not. */
uint8_t buf[p->payload_len];
memcpy(buf, p->payload, p->payload_len);
/* if tree_seg is set, we have an exact duplicate that we need to check */
if (tree_seg) {
DoHandleDataOverlap(stream, tree_seg, seg, buf, p);
handle = tree_seg;
}
const bool is_head = !(TCPSEG_RB_PREV(handle));
const bool is_tail = !(TCPSEG_RB_NEXT(handle));
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.
11 years ago
/* new list head */
if (is_head && !is_tail) {
result = DoHandleDataCheckForward(stream, handle, buf, 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.
11 years ago
/* new list tail */
} else if (!is_head && is_tail) {
result = DoHandleDataCheckBackwards(stream, handle, buf, 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.
11 years ago
/* middle of the list */
} else if (!is_head && !is_tail) {
result = DoHandleDataCheckBackwards(stream, handle, buf, p);
result |= DoHandleDataCheckForward(stream, handle, buf, p);
}
/* we had an overlap with different data */
if (result) {
StreamTcpSetEvent(p, STREAM_REASSEMBLY_OVERLAP_DIFFERENT_DATA);
StatsIncr(tv, ra_ctx->counter_tcp_reass_overlap_diff_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.
11 years ago
}
/* insert the temp buffer now that we've (possibly) updated
* it to account for the overlap policies */
int res = InsertSegmentDataCustom(stream, handle, buf, p->payload_len);
if (res < 0) {
if (res == -1) {
StatsIncr(tv, ra_ctx->counter_tcp_segment_memcap);
}
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.
11 years ago
return -1;
}
return 0;
}
/** \internal
* \brief Add the header data to the segment
* \param rp packet to take the headers from. Might differ from `pp` in tunnels.
* \param pp packet to take the payload size from.
*/
static void StreamTcpSegmentAddPacketDataDo(TcpSegment *seg, const Packet *rp, const Packet *pp)
{
if (GET_PKT_DATA(rp) != NULL && GET_PKT_LEN(rp) > pp->payload_len) {
seg->pcap_hdr_storage->ts.tv_sec = rp->ts.tv_sec;
seg->pcap_hdr_storage->ts.tv_usec = rp->ts.tv_usec;
seg->pcap_hdr_storage->pktlen = GET_PKT_LEN(rp) - pp->payload_len;
/*
* pkt_hdr members are initially allocated 64 bytes of memory. Thus,
* need to check that this is sufficient and allocate more memory if
* not.
*/
if (seg->pcap_hdr_storage->pktlen > seg->pcap_hdr_storage->alloclen) {
uint8_t *tmp_pkt_hdr = StreamTcpReassembleRealloc(seg->pcap_hdr_storage->pkt_hdr,
seg->pcap_hdr_storage->alloclen, seg->pcap_hdr_storage->pktlen);
if (tmp_pkt_hdr == NULL) {
SCLogDebug("Failed to realloc");
seg->pcap_hdr_storage->ts.tv_sec = 0;
seg->pcap_hdr_storage->ts.tv_usec = 0;
seg->pcap_hdr_storage->pktlen = 0;
return;
} else {
seg->pcap_hdr_storage->pkt_hdr = tmp_pkt_hdr;
seg->pcap_hdr_storage->alloclen = GET_PKT_LEN(rp) - pp->payload_len;
}
}
memcpy(seg->pcap_hdr_storage->pkt_hdr, GET_PKT_DATA(rp),
(size_t)GET_PKT_LEN(rp) - pp->payload_len);
} else {
seg->pcap_hdr_storage->ts.tv_sec = 0;
seg->pcap_hdr_storage->ts.tv_usec = 0;
seg->pcap_hdr_storage->pktlen = 0;
}
}
/**
* \brief Adds the following information to the TcpSegment from the current
* packet being processed: time values, packet length, and the
* header data of the packet. This information is added to the TcpSegment so
* that it can be used in pcap capturing (log-pcap-stream) to dump the tcp
* session at the beginning of the pcap capture.
* \param seg TcpSegment where information is being stored.
* \param p Packet being processed.
* \param tv Thread-specific variables.
* \param ra_ctx TcpReassembly thread-specific variables
*/
static void StreamTcpSegmentAddPacketData(
TcpSegment *seg, Packet *p, ThreadVars *tv, TcpReassemblyThreadCtx *ra_ctx)
{
if (seg->pcap_hdr_storage == NULL || seg->pcap_hdr_storage->pkt_hdr == NULL) {
return;
}
if (IS_TUNNEL_PKT(p) && !IS_TUNNEL_ROOT_PKT(p)) {
Packet *rp = p->root;
StreamTcpSegmentAddPacketDataDo(seg, rp, p);
} else {
StreamTcpSegmentAddPacketDataDo(seg, p, 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.
11 years ago
/**
* \return 0 ok
* \return -1 segment not inserted due to memcap issue
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.
11 years ago
*
* \param seg segment, this function takes total ownership
*
* In case of error, this function returns the segment to the pool
*/
int StreamTcpReassembleInsertSegment(ThreadVars *tv, TcpReassemblyThreadCtx *ra_ctx,
TcpStream *stream, TcpSegment *seg, Packet *p,
uint32_t pkt_seq, uint8_t *pkt_data, uint16_t pkt_datalen)
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.
11 years ago
{
SCEnter();
TcpSegment *dup_seg = 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.
11 years ago
/* insert segment into list. Note: doesn't handle the data */
int r = DoInsertSegment (stream, seg, &dup_seg, p);
if (IsTcpSessionDumpingEnabled()) {
StreamTcpSegmentAddPacketData(seg, p, 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.
11 years ago
if (likely(r == 0)) {
/* no overlap, straight data insert */
int res = InsertSegmentDataCustom(stream, seg, pkt_data, pkt_datalen);
if (res < 0) {
if (res == -1) {
StatsIncr(tv, ra_ctx->counter_tcp_segment_memcap);
}
StatsIncr(tv, ra_ctx->counter_tcp_reass_data_normal_fail);
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.
11 years ago
StreamTcpRemoveSegmentFromStream(stream, seg);
StreamTcpSegmentReturntoPool(seg);
if (res == -1) {
SCReturnInt(-ENOMEM);
}
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.
11 years ago
SCReturnInt(-1);
}
} else if (r == 1 || r == 2) {
SCLogDebug("overlap (%s%s)", r == 1 ? "normal" : "", r == 2 ? "duplicate" : "");
if (r == 2) {
SCLogDebug("dup_seg %p", dup_seg);
}
/* XXX should we exclude 'retransmissions' here? */
StatsIncr(tv, ra_ctx->counter_tcp_reass_overlap);
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.
11 years ago
/* now let's consider the data in the overlap case */
int res = DoHandleData(tv, ra_ctx, stream, seg, dup_seg, 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.
11 years ago
if (res < 0) {
StatsIncr(tv, ra_ctx->counter_tcp_reass_data_overlap_fail);
if (r == 1) // r == 2 mean seg wasn't added to stream
StreamTcpRemoveSegmentFromStream(stream, 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.
11 years ago
StreamTcpSegmentReturntoPool(seg);
SCReturnInt(-1);
}
if (r == 2) {
SCLogDebug("duplicate segment %u/%u, discard it",
seg->seq, seg->payload_len);
StreamTcpSegmentReturntoPool(seg);
#ifdef DEBUG
if (SCLogDebugEnabled()) {
TcpSegment *s = NULL, *safe = NULL;
RB_FOREACH_SAFE(s, TCPSEG, &stream->seg_tree, safe)
{
SCLogDebug("tree: seg %p, SEQ %"PRIu32", LEN %"PRIu16", SUM %"PRIu32"%s%s%s",
s, s->seq, TCP_SEG_LEN(s),
(uint32_t)(s->seq + TCP_SEG_LEN(s)),
s->seq == seg->seq ? " DUPLICATE" : "",
TCPSEG_RB_PREV(s) == NULL ? " HEAD" : "",
TCPSEG_RB_NEXT(s) == NULL ? " TAIL" : "");
}
}
#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.
11 years ago
}
SCReturnInt(0);
}
/*
* Pruning & removal
*/
static inline bool SegmentInUse(const TcpStream *stream, const TcpSegment *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.
11 years ago
{
/* if proto detect isn't done, we're not returning */
if (!(stream->flags & STREAMTCP_STREAM_FLAG_NOREASSEMBLY)) {
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.
11 years ago
if (!(StreamTcpIsSetStreamFlagAppProtoDetectionCompleted(stream))) {
SCReturnInt(true);
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.
11 years ago
}
}
SCReturnInt(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.
11 years ago
}
/** \internal
* \brief check if we can remove a segment from our segment list
*
* \retval true
* \retval 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.
11 years ago
*/
static inline bool StreamTcpReturnSegmentCheck(const TcpStream *stream, const TcpSegment *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.
11 years ago
{
if (SegmentInUse(stream, seg)) {
SCReturnInt(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.
11 years ago
}
if (!(StreamingBufferSegmentIsBeforeWindow(&stream->sb, &seg->sbseg))) {
SCReturnInt(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.
11 years ago
}
SCReturnInt(true);
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.
11 years ago
}
static inline uint64_t GetLeftEdgeForApp(Flow *f, TcpSession *ssn, TcpStream *stream)
{
const FramesContainer *frames_container = AppLayerFramesGetContainer(f);
if (frames_container == NULL)
return STREAM_APP_PROGRESS(stream);
const Frames *frames =
stream == &ssn->client ? &frames_container->toserver : &frames_container->toclient;
// const uint64_t x = FramesLeftEdge(stream, frames);
// BUG_ON(x != (frames->left_edge_rel + STREAM_BASE_OFFSET(stream)));
// return x;
const uint64_t o = (uint64_t)frames->left_edge_rel + STREAM_BASE_OFFSET(stream);
SCLogDebug(
"%s: frames left edge: %" PRIu64, &ssn->client == stream ? "toserver" : "toclient", o);
return o;
}
static inline uint64_t GetLeftEdge(Flow *f, TcpSession *ssn, 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.
11 years ago
{
uint64_t left_edge = 0;
const bool use_app = !(ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED);
const bool use_raw = !(stream->flags & STREAMTCP_STREAM_FLAG_DISABLE_RAW);
const bool use_log = stream_config.streaming_log_api;
SCLogDebug("use_app %d use_raw %d use_log %d", use_app, use_raw, use_log);
if (use_raw) {
uint64_t raw_progress = STREAM_RAW_PROGRESS(stream);
if (StreamTcpInlineMode() == TRUE) {
uint32_t chunk_size = (stream == &ssn->client) ?
stream_config.reassembly_toserver_chunk_size :
stream_config.reassembly_toclient_chunk_size;
if (raw_progress < (uint64_t)chunk_size) {
raw_progress = 0;
} else {
raw_progress -= (uint64_t)chunk_size;
}
}
/* apply min inspect depth: if it is set we need to keep data
* before the raw progress. */
if (use_app && stream->min_inspect_depth && ssn->state < TCP_CLOSED) {
if (raw_progress < stream->min_inspect_depth)
raw_progress = 0;
else
raw_progress -= stream->min_inspect_depth;
SCLogDebug("stream->min_inspect_depth %u, raw_progress %"PRIu64,
stream->min_inspect_depth, raw_progress);
}
if (use_app) {
const uint64_t app_le = GetLeftEdgeForApp(f, ssn, stream);
left_edge = MIN(app_le, raw_progress);
SCLogDebug("left_edge %" PRIu64 ", using both app:%" PRIu64 ", raw:%" PRIu64, left_edge,
app_le, raw_progress);
} else {
left_edge = raw_progress;
SCLogDebug("left_edge %"PRIu64", using only raw:%"PRIu64,
left_edge, raw_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.
11 years ago
} else if (use_app) {
const uint64_t app_le = GetLeftEdgeForApp(f, ssn, stream);
left_edge = app_le;
SCLogDebug("left_edge %" PRIu64 ", using only app:%" PRIu64, left_edge, app_le);
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.
11 years ago
} else {
left_edge = STREAM_BASE_OFFSET(stream) + stream->sb.buf_offset;
SCLogDebug("no app & raw: left_edge %"PRIu64" (full stream)", left_edge);
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.
11 years ago
}
if (use_log) {
if (use_app || use_raw) {
left_edge = MIN(left_edge, STREAM_LOG_PROGRESS(stream));
} else {
left_edge = STREAM_LOG_PROGRESS(stream);
}
}
uint64_t last_ack_abs = STREAM_BASE_OFFSET(stream);
if (STREAM_LASTACK_GT_BASESEQ(stream)) {
last_ack_abs += (stream->last_ack - stream->base_seq);
}
/* in IDS mode we shouldn't see the base_seq pass last_ack */
DEBUG_VALIDATE_BUG_ON(last_ack_abs < left_edge && StreamTcpInlineMode() == FALSE && !f->ffr &&
ssn->state < TCP_CLOSED);
left_edge = MIN(left_edge, last_ack_abs);
/* if we're told to look for overlaps with different data we should
* consider data that is ack'd as well. Injected packets may have
* been ack'd or injected packet may be too late. */
if (StreamTcpInlineMode() == FALSE && check_overlap_different_data) {
const uint32_t window = stream->window ? stream->window : 4096;
if (window < left_edge)
left_edge -= window;
else
left_edge = 0;
SCLogDebug("stream:%p left_edge %"PRIu64, stream, left_edge);
}
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.
11 years ago
if (left_edge > 0) {
/* we know left edge based on the progress values now,
* lets adjust it to make sure in-use segments still have
* data */
TcpSegment *seg = NULL;
RB_FOREACH(seg, TCPSEG, &stream->seg_tree) {
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.
11 years ago
if (TCP_SEG_OFFSET(seg) > left_edge) {
SCLogDebug("seg beyond left_edge, we're done");
break;
}
if (SegmentInUse(stream, 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.
11 years ago
left_edge = TCP_SEG_OFFSET(seg);
SCLogDebug("in-use seg before left_edge, adjust to %"PRIu64" and bail", left_edge);
break;
}
}
}
return left_edge;
}
static void StreamTcpRemoveSegmentFromStream(TcpStream *stream, TcpSegment *seg)
{
RB_REMOVE(TCPSEG, &stream->seg_tree, 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.
11 years ago
}
/** \brief Remove idle TcpSegments from TcpSession
*
* Checks app progress and raw progress and progresses them
* if needed, slides the streaming buffer, then gets rid of
* excess segments.
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.
11 years ago
*
* \param f flow
* \param flags direction flags
*/
void StreamTcpPruneSession(Flow *f, uint8_t flags)
{
SCEnter();
if (f == NULL || f->protoctx == NULL) {
SCReturn;
}
TcpSession *ssn = f->protoctx;
TcpStream *stream = NULL;
if (flags & STREAM_TOSERVER) {
stream = &ssn->client;
} else if (flags & STREAM_TOCLIENT) {
stream = &ssn->server;
} else {
SCReturn;
}
if (stream->flags & STREAMTCP_STREAM_FLAG_NOREASSEMBLY) {
return;
}
if (stream->flags & STREAMTCP_STREAM_FLAG_DEPTH_REACHED) {
stream->flags |= STREAMTCP_STREAM_FLAG_NOREASSEMBLY;
SCLogDebug("ssn %p / stream %p: reassembly depth reached, "
"STREAMTCP_STREAM_FLAG_NOREASSEMBLY set", ssn, stream);
StreamTcpReturnStreamSegments(stream);
StreamingBufferClear(&stream->sb);
return;
} else if ((ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED) &&
(stream->flags & STREAMTCP_STREAM_FLAG_DISABLE_RAW)) {
SCLogDebug("ssn %p / stream %p: both app and raw are done, "
"STREAMTCP_STREAM_FLAG_NOREASSEMBLY set", ssn, stream);
stream->flags |= STREAMTCP_STREAM_FLAG_NOREASSEMBLY;
StreamTcpReturnStreamSegments(stream);
StreamingBufferClear(&stream->sb);
return;
}
const uint64_t left_edge = GetLeftEdge(f, ssn, stream);
if (left_edge && left_edge > STREAM_BASE_OFFSET(stream)) {
uint32_t slide = left_edge - STREAM_BASE_OFFSET(stream);
SCLogDebug("buffer sliding %u to offset %"PRIu64, slide, left_edge);
if (!(ssn->flags & STREAMTCP_FLAG_APP_LAYER_DISABLED)) {
AppLayerFramesSlide(f, slide, flags & (STREAM_TOSERVER | STREAM_TOCLIENT));
}
StreamingBufferSlideToOffset(&stream->sb, left_edge);
stream->base_seq += slide;
if (slide <= stream->app_progress_rel) {
stream->app_progress_rel -= slide;
} else {
stream->app_progress_rel = 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.
11 years ago
}
if (slide <= stream->raw_progress_rel) {
stream->raw_progress_rel -= slide;
} else {
stream->raw_progress_rel = 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.
11 years ago
}
if (slide <= stream->log_progress_rel) {
stream->log_progress_rel -= slide;
} else {
stream->log_progress_rel = 0;
}
SCLogDebug("stream base_seq %u at stream offset %"PRIu64,
stream->base_seq, STREAM_BASE_OFFSET(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.
11 years ago
}
/* loop through the segments and remove all not in use */
TcpSegment *seg = NULL, *safe = NULL;
RB_FOREACH_SAFE(seg, TCPSEG, &stream->seg_tree, safe)
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.
11 years ago
{
SCLogDebug("seg %p, SEQ %"PRIu32", LEN %"PRIu16", SUM %"PRIu32,
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.
11 years ago
seg, seg->seq, TCP_SEG_LEN(seg),
(uint32_t)(seg->seq + 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.
11 years ago
if (StreamTcpReturnSegmentCheck(stream, seg) == 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.
11 years ago
SCLogDebug("not removing segment");
break;
}
StreamTcpRemoveSegmentFromStream(stream, seg);
StreamTcpSegmentReturntoPool(seg);
SCLogDebug("removed segment");
continue;
}
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.
11 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.
11 years ago
/*
* unittests
*/
#ifdef UNITTESTS
#include "tests/stream-tcp-list.c"
#endif