When Suricata picks up a flow it assumes the first packet is
toserver. In a perfect world without packet loss and where all
sessions neatly start after Suricata itself started, this would be
true. However, in reality we have to account for packet loss and
Suricata starting to get packets for flows already active be for
Suricata is (re)started.
The protocol records on the wire would often be able to tell us more
though. For example in SMB1 and SMB2 records there is a flag that
indicates whether the record is a request or a response. This patch
is enabling the procotol detection engine to utilize this information
to 'reverse' the flow.
There are three ways in which this is supported in this patch:
1. patterns for detection are registered per direction. If the proto
was not recognized in the traffic direction, and midstream is
enabled, the pattern set for the opposing direction is also
evaluated. If that matches, the flow is considered to be in the
wrong direction and is reversed.
2. probing parsers now have a way to feed back their understanding
of the flow direction. They are now passed the direction as
Suricata sees the traffic when calling the probing parsers. The
parser can then see if its own observation matches that, and
pass back it's own view to the caller.
3. a new pattern + probing parser set up: probing parsers can now
be registered with a pattern, so that when the pattern matches
the probing parser is called as well. The probing parser can
then provide the protocol detection engine with the direction
of the traffic.
The process of reversing takes a multi step approach as well:
a. reverse the current packets direction
b. reverse most of the flows direction sensitive flags
c. tag the flow as 'reversed'. This is because the 5 tuple is
*not* reversed, since it is immutable after the flows creation.
Most of the currently registered parsers benefit already:
- HTTP/SMTP/FTP/TLS patterns are registered per direction already
so they will benefit from the pattern midstream logic in (1)
above.
- the Rust based SMB parser uses a mix of pattern + probing parser
as described in (3) above.
- the NFS detection is purely done by probing parser and is updated
to consider the direction in that parser.
Other protocols, such as DNS, are still to do.
Ticket: #2572
This patch provides a working expectation system. This will allow
suricata to have a way to identify parallel connections opened by
a protocol such as FTP.
Expectation are a chained list and there is a cleaning by timeout
of the entries.
This patch also defined a counter of expectations that is also
used to check if we need to query IPPairs. This way we only query
the IPPairs store if we have an expectation.
Check counter id before updating a counter. In case of a disabled
parser with the protocol detection enable, the id can be 0. In
debug mode this would lead to a BUG_ON.
Add API calls to upgrade to TLS or to request a protocol change
without a specific protocol expectation.
If the HTTP CONNECT session includes a port on the url, use that to
look up the probing parser during protocol detection. Solves a
missed detection of a SSLv2 session that upgrades to TLSv1. SSLv2
relies on the probing parser which is limited to certain ports.
In case of STARTTLS in SMTP and FTP, the port is hardcoded to 443.
A new event APPLAYER_UNEXPECTED_PROTOCOL is set if there was a
mismatch.
Support changing the application level protocol for a flow. This is
needed by STARTTLS and HTTP CONNECT to switch from the original
alproto to tls.
This commit allows a flag to be set 'FLOW_CHANGE_PROTO', which
triggers a new protocol detection on the next packet for a flow.
Set flags by default:
-Wmissing-prototypes
-Wmissing-declarations
-Wstrict-prototypes
-Wwrite-strings
-Wcast-align
-Wbad-function-cast
-Wformat-security
-Wno-format-nonliteral
-Wmissing-format-attribute
-funsigned-char
Fix minor compiler warnings for these new flags on gcc and clang.
Implement the inline mode for raw content inspection. Packets
are leading, and when a packet's payload has been added to the
stream, the packet is inspected in the context of the stream.
Reassembly will return a buffer with the packet data with older
data in front of it and after it, if available.
At flow timeout, we no longer need to first run reassembly in
one dir, then inspection in the other. We can do both in single
packet now.
Disable pseudo packets when receiving stream end packets. Instead
call the app-layer parser in the packet direction for stream end
packets and flow end packets.
These changes in handling of those stream end packets make the
pseudo packets unnecessary.
Remove the 'StreamMsg' approach from the engine. In this approach the
stream engine would create a list of chunks for inspection by the
detection engine. There were several issues:
1. the messages had a fixed size, so blocks of data bigger than ~4k
would be cut into multiple messages
2. it lead to lots of data copying and unnecessary memory use
3. the StreamMsgs used a central pool
The Stream engine switched over to the streaming buffer API, which
means that the reassembled data is always available. This made the
StreamMsg approach even clunkier.
The new approach exposes the streaming buffer data to the detection
engine. It has to pay attention to an important issue though: packet
loss. The data may have gaps. The streaming buffer API tracks the
blocks of continuous data.
To access the data for inspection a callback approach is used. The
'StreamReassembleRaw' function is called with a callback and data.
This way it runs the MPM and individual rule inspection code. At
the end of each detection run the stream engine is notified that it
can move forward it's 'progress'.
A GAP during protocol detection would lead to all reassembly
getting disabled, so also the raw reassembly. In addition, it
could prevent the opposing side from doing protocol detection.
This patch remove the 'disable reassembly' logic. Stream engine
will take the stream with GAP and app-layer will make the proto
detection as complete.
Add negated matches to match list instead of amatch.
Allow matching on 'failed'.
Introduce per packet flags for proto detection. Flags are used to
only inspect once per direction. Flag packet on PD-failure too.
The Flow::data_al_so_far was used for tracking data already
parsed when protocol for the current direction wasn't known yet. As
this behaviour has changed the tracking can be removed.
When the current direction doesn't get a protocol detection, but the
opposing direction did, previously we would send the current data to
the parser. Then when we'd be invoked again (until the protocol
detection finally failed) we'd get the same data + the new data. To
make sure we'd not send the same data to the parser again, the flow
kept track of how much was already sent to the app-layer using
data_al_so_far.
This patch changes the behaviour. Instead of sending the data for
the current direction right away, we only do this when protocol
detection is complete. This way we won't have to track anything.
This patch adds a transaction counter for application layers
supporting it. Analysis is done after the parsing by the
different application layers.
This result in new data in the stats output, that looks like:
```
"app-layer": {
"tx": {
"dns_udp": 21433,
"http": 12766,
"smtp": 0,
"dns_tcp": 0
}
},
```
To be able to add a transaction counter we will need a ThreadVars
in the AppLayerParserParse function.
This function is massively used in unittests
and this result in an long commit.
Instead of handling the packet update during flow lookup, handle
it in the stream/detect threads. This lowers the load of the
capture thread(s) in autofp mode.
The decoders now set a flag in the packet if the packet needs a
flow lookup. Then the workers will take care of this. The decoders
also already calculate the raw flow hash value. This is so that
this value can be used in flow balancing in autofp.
Because the flow lookup/creation is now done in the worker threads,
the flow balancing can no longer use the flow. It's not yet
available. Autofp load balancing uses raw hash values instead.
In the same line, move UDP AppLayer out of the DecodeUDP module,
and also into the stream/detect threads.
Handle TCP session reuse inside the flow engine itself. If a looked up
flow matches the packet, but is a TCP stream starter, check if the
ssn needs to be reused. If that is the case handle it within the
lookup function. Simplies the locking and removes potential race
conditions.
More exceptional cases for protocol detection. In very unbalanced flows,
where just a few bytes are sent toserver and many toclient, proto detect
might not complete in time on the toserver direction. This can lead to
queuing up many segments in the toclient direction.
Another case is that in come cases the stream is flagged as proto detect
done, but the flows proto detect flags are not set. This is now handled
by the ProtoDetectDone() check.
If the protocol required TOSERVER data first, but the SSN started with
a GAP, then the TOCLIENT side would get stuck in an expensive path:
1. it would run detection on TOCLIENT
2. it would try to force reassembly for TOSERVER
3. it would reset the detected protocol as TOSERVER failed
4. it would not evict any segment
This had 2 consequences:
1. on long running sessions this could lead to using lots of memory
on segments, denying other sessions resources
2. wasted cycles on protocol detection and segment list management
This patch introduces a fix. It checks in the (2) stage above, whether
the opposing stream (that we depend on) it is a NOREASSEMBLY state. If
so, it gives up on this side of the session as well.
Instead, intruduce StreamTcpDisableAppLayer to disable app layer
tracking and reassembly. StreamTcpAppLayerIsDisabled can be used
to check it.
Replace all uses of FlowSetSessionNoApplayerInspectionFlag and
the FLOW_NO_APPLAYER_INSPECTION.
AppLayerTest09, AppLayerTest10 and AppLayerTest11 depended on a max
protocol detection pattern size of < 17. Update the tests to pass one
extra byte to the app layer. This makes the protocol detection code
flag the session as 'proto detection completed' again.
The entire body of these functions are protected by ifdef PROFILING.
If the functions are inlined, then this check removes the need for the
function entirely.
Previously, the empty function was still called, even when not built
for profiling. The functions showed as being 0.25% of total CPU time
without being built for profiling.
Some traffic is very unbalanced. For example a 4 bytes request
followed by 12mb of response data. If the 4 bytes don't lead to
the protocol being detected, then app layer processing won't
start, but it will not give up either. It will just wait for more
data. This leads to piling up data on the opposite side.
Observed:
TS: 4 bytes. PP failed (bit set), PM has not given up (bit not set).
This makes sense as max_depth is not yet reached.
TC: 12mb. PP and PM failed (bits set).
As ts-PM never gives up, we never consider proto detect complete,
so all segments in either direction are still kept in the session.
This patch adds a cutoff for this case:
- IF for TS we have PP bit set, PM not set, AND
- we have TC both bits set, AND
- proto is unknown, AND
- TC data is 100k already, THEN
- give up on protocol detection.
The same for the opposite direction.
If we're getting a lot of data in one direction and the proto for this
direction is unknown, proto detect will hold up segments in the segment
list in the stream. They are held so that if we detect the protocol on
the opposing stream, we can still parse this side of the stream as well.
However, some sessions are very unbalanced. FTP data channels, large
PUT/POST request and many others, can lead to cases where we would have
to store many megabytes worth of segments before we see the opposing
stream. This leads to risks of resource starvation.
In this patch, a cutoff point is enforced. If we've stored 100k in one
direction and we've seen no data in the other direction, we give up.
If we've given up, the applayer_proto_detection_skipped event is set.
app-layer-event: applayer_proto_detection_skipped;
If a TCP session is midstream, we may end up with a case where the
start of an HTTP request is missing. We won't detect HTTP based on
the request. However, the reply is fine, so we detect HTTP anyway.
This leads to passing the incomplete request to the htp parser.
This has been observed, where the http parser then saw many bogus
requests in the incomplete data. This is not limited to HTTP.
To counter this case, a midstream session MUST find it's protocol
in the toserver direction. If not, we assume the start of the
request/toserver is incomplete and no reliable detection and parsing
is possible. So we give up.
Move app layer event handling into app-layer-event.[ch].
Convert 'Set' macro's to functions.
Get rid of duplication in Set and SetRaw. Set now calls SetRaw.
Fix potentential int overflow condition in the event storage.
Update callers.
This patch updates the DNS counters from the main AppLayer entry
functions. Due to the limited scope of AppLayerThreadCtx some of
the logic had to be implemented in app-layer.c, where it doesn't
belong.
To be able to register counters from AppLayerGetCtxThread, the
ThreadVars pointer needs to be available in it and thus in it's
callers:
- AppLayerGetCtxThread
- DecodeThreadVarsAlloc
- StreamTcpReassembleInitThreadCtx
StreamMsgs would be stored in a per thread queue before being
attached to the tcp ssn. This is unnecessary, so this patch
removes this queue and puts the smsgs into the ssn directly.
Large patch as it affects a lot of tests.
StreamMsg' flow reference was used mostly to make sure a flow would
not get removed from the hash before inspection. For this it needed
to reference the flow use_cnt reference counter. Nowadays we have
more advanced flow timeout handling. This will make sure that if
there still are pending smsgs' in a flow, these will still be
processed.
Preparation for removing flow pointer from StreamMsg. Instead of
getting the ssn indirectly through StreamMsg->flow, we pass it
directly as all callers have it already.
This patch adds a memory counter for HTP memory usage. As
there is no thread variables available in application layer
the counter has been added to the TCP reassembly thread.
For AppLayerThreadCtx, AppLayerParserState, AppLayerParserThreadCtx
and AppLayerProtoDetectThreadCtx, use opaque pointers instead of
void pointers.
AppLayerParserState is declared in flow.h as it's part of the Flow
structure.
AppLayerThreadCtx is declared in decode.h, as it's part of the
DecodeThreadVars structure.
app-layer.[ch], app-layer-detect-proto.[ch] and app-layer-parser.[ch].
Things addressed in this commit:
- Brings out a proper separation between protocol detection phase and the
parser phase.
- The dns app layer now is registered such that we don't use "dnstcp" and
"dnsudp" in the rules. A user who previously wrote a rule like this -
"alert dnstcp....." or
"alert dnsudp....."
would now have to use,
alert dns (ipproto:tcp;) or
alert udp (app-layer-protocol:dns;) or
alert ip (ipproto:udp; app-layer-protocol:dns;)
The same rules extend to other another such protocol, dcerpc.
- The app layer parser api now takes in the ipproto while registering
callbacks.
- The app inspection/detection engine also takes an ipproto.
- All app layer parser functions now take direction as STREAM_TOSERVER or
STREAM_TOCLIENT, as opposed to 0 or 1, which was taken by some of the
functions.
- FlowInitialize() and FlowRecycle() now resets proto to 0. This is
needed by unittests, which would try to clean the flow, and that would
call the api, AppLayerParserCleanupParserState(), which would try to
clean the app state, but the app layer now needs an ipproto to figure
out which api to internally call to clean the state, and if the ipproto
is 0, it would return without trying to clean the state.
- A lot of unittests are now updated where if they are using a flow and
they need to use the app layer, we would set a flow ipproto.
- The "app-layer" section in the yaml conf has also been updated as well.
The uint8_t *pkt in the Packet structure always points to the memory
immediately following the Packet structure. It is better to simply
calculate that value every time than store the 8 byte pointer.
If a session only has data in one direction, like ftp data sessions,
protocol detection will only run in one direction. This led to a
situation where reassembly would hold all the segments as proto
detection was never flagged as complete.
This patch introduces a limit for protocol detection in this case.
If the limit is reached, detection will give up.
In case of recursive call to protocol detection from within protocol
detection, and the recursively invoked stream still hasn't been ack'ed
yet, protocol detection doesn't take place. In such cases we will end up
still calling the app layer with the wrong direction data. Introduce a
check to not call app layer with wrong direction data.
When sockets are re-used reset all relevant vars correctly.
This commit fixes a bug where we were not reseting app proto detection
vars.
While fixing #989, we discovered some other bugs which have also been
fixed, or rather some features which are now updated. One of the feature
update being if we recieve wrong direction data first, we don't reset the
protocol values for the flow. We let the flow retain the detected
values.
Unittests have been modified to accomodate the above change.
Removed a flag parameter introuced earlier to indicate the data
that is first acceptable by the parser. We now use a differently
named parameter to carry out the same activity.
The logic we use currently is if we have already sent some data to
a parser before we figure out we have a proto mismatch, we use the
proto from the first direction from which we have already sent the
data to the parser, else we stick to the the to client direction.
PM and PP phase.
Replace the areas of the code that would otherwise rely on setting/reading
these flags with these macros.
Other minor tweaks to some api calls.
Victor Julien
Eric Leblond
Jason Ish
Mats Klepsland
Giuseppe Longo
maxtors
Ken Steele
jeka
Anoop Saldanha
Last G