Bug #7252
closedstream/reassemble: GetBlock implies gap without searching the entire tree for block
Description
GetBlock
fn has this logic:
for ( ; blk != NULL; blk = SBB_RB_NEXT(blk)) {
if (blk->offset >= offset) {
return blk;
} else if ((blk->offset + blk->len) > offset) {
return blk;
}
}
return NULL;
This means that the moment a block with an offset greater than the asked offset was found, it was returned.
In the caller, therefore, the following is done:
/* block past out offset */
else if (blk->offset > offset) {
SCLogDebug("gap, want data at offset %"PRIu64", "
"got data at %"PRIu64". GAP of size %"PRIu64,
offset, blk->offset, blk->offset - offset);
*data = NULL;
*data_len = blk->offset - offset;
}
and then, the data offset is adjusted as per some gap handling logic.
This is incorrect because the point of GetBlock
fn is to get the block containing a given offset. Entire tree should have been searched for the given offset instead of returning the first block greater than equal to the given offset.
Note that if a block has offset equal to the given offset, it is perfect. It is incorrect in the other case i.e. the block offset is greater than the given offset.
Updated by Shivani Bhardwaj about 1 month ago
- Status changed from New to Rejected
- Assignee changed from Community Ticket to Shivani Bhardwaj
This bug was based on an incorrect assumption: that sb->head
points to the root of the tree. Turns out it points to the minimum value in the SBB tree so it really just makes up for a linear search starting at the lowest value node.
Rejecting ticket as invalid.
Tested and verified after creating a pcap with appropriate gaps matching the criteria where I thought this would be buggy.