[Snort-devel] BUG: corner case involving http_cookie

Will Metcalf william.metcalf at ...2499...
Wed Mar 10 17:56:57 EST 2010


How about this one, using the same pcap. I have both client and server
flow_depth set to 0.

#test 72 http_cookie + byte_test.
#hmm interesting using http_cookie in combination with byte_test seems
to fail always. Removing either the byte_test check or the http_cookie
modifier will cause this sig to fire.  Notice that the byte_test
#check isn't even relative to the content match.
#
#file oisfsearchnums.pcap
alert tcp any any -> any any (msg:"e6504ae48c99f09df7f58996aacbb6b0 +
http_cookie + byte_test"; content:"e6504ae48c99f09df7f58996aacbb6b0";
http_cookie; byte_test:2,=,12,596,string,dec; classtype:bad-unknown;
sid:72; rev:1;)


Regards,

Will

On Wed, Mar 10, 2010 at 1:54 PM, Will Metcalf <william.metcalf at ...2499...> wrote:
> I can see the need for the cut-off.  You have to make compromises
> between accuracy and performance this tends to be the nature of the
> beast with IDS's,  and sure there will always be evasions abut the
> dynamic flow_depth thing still isn't making much sense to me and/or at
> least this behavior should be documented somewhere.  Did I just miss
> something in the readme?  Btw there is a typo both in the manual and
> the README regarding client_flow_depth.  "It primarily eliminates
> Snort fro inspecting larger HTTP Cookies that appear at the end of
> many client request Headers."
>
>
> Regards,
>
> Will
>
> On Wed, Mar 10, 2010 at 1:18 PM, Steven Sturges
> <steve.sturges at ...402...> wrote:
>> There are different settings for the requests & responses.
>> See client_flow_depth and server_flow_depth in the readme/manual.
>>
>> Setting either of the flow depth values to 0 would inspect
>> all of the respective request or response, but incurs a
>> pretty big performance hit.
>>
>> And yes, the default (in code) is 300 for both.
>>
>> Will Metcalf wrote:
>>> Ahhh ok so I want to make sure I understand correctly in 2.8.5 the
>>> default 300 byte cutoff is not only applied to http_responses but also
>>> http_requests (README.http_inspect) if there are http normalizations.
>>> Is there not still a potential evasion here based on the tests below
>>> as packet being normalized or potentially matching on one or more
>>> rules can be fairly unpredictable?  Would your recommended
>>> configuration then be to set client_flow_depth to 0 in the
>>> http_inspect preproc for 2.8.5 users if they wanted to avoid the
>>> potential evasion?
>>>
>>> Regards,
>>>
>>> Will
>>>
>>> On Wed, Mar 10, 2010 at 12:03 PM, Steven Sturges
>>> <steve.sturges at ...402...> wrote:
>>>> Seems that what is really coming into play is the flow depth.
>>>> That is going to limit how much of the raw data is searched
>>>> with the pattern matcher.
>>>>
>>>> To summarize how it works with 2.8.6:
>>>>
>>>> 1) If HTTP normalizes headers and cookies, the amount of raw data
>>>> searched by fast pattern matcher is determined by client (or server)
>>>> flow depth settings.
>>>>
>>>> 2) If there are no HTTP normalizations, flow depth is not applied, so
>>>> all of the raw data is searched.
>>>>
>>>> With 2.8.5, in 1) above, the flow depth was applied only if there were
>>>> rules searched for one of those HTTP normalized buffers.
>>>>
>>>> By default, client flow depth is 0 -- which is what Will is probably
>>>> hitting in his tests.
>>>>
>>>> Alerts w/ combinations of rules & different flow depths in 2.8.6:
>>>>
>>>> Default flow depth (0)
>>>> SID 59, SID 68 enabled
>>>> 1 alert (sid 59)
>>>>
>>>> Default flow depth (0)
>>>> SID 68 enabled
>>>> 0 alerts
>>>>
>>>> [Same as above for client_flow_depth 300]
>>>>
>>>> Client flow depth (1460)
>>>> SID 59, SID 68 enabled
>>>> 3 alerts (sid 59, sid 68x2)
>>>>
>>>> Client flow depth (1460)
>>>> SID 68 enabled
>>>> 2 alerts (sid 68x2)
>>>>
>>>> -steve
>>>>
>>>> Matt Jonkman wrote:
>>>>> Appreciate the clarification Steve. But I'm concerned, this will make
>>>>> hundreds if not a few thousand rules not work correctly in our set and
>>>>> in vrt/snort gpl. If someone has http_inspect on in a recent snort, but
>>>>> does NOT have ALL of their http related rules converted to the new form
>>>>> using the http_* modifiers (which we have none converted) then they're
>>>>> going to have massive problems, no? I think a lot of people are missing
>>>>> a lot of things right now.
>>>>>
>>>>> Is there a way to make http_inspect not do this? We have a lot of sigs
>>>>> that won't work on normalized data, for instance the recent trojan sigs
>>>>> that look at the order of the parameters in an http post for uniqieness.
>>>>> With the normalized only we can't see that can we?
>>>>>
>>>>> Thanks!
>>>>>
>>>>> Matt
>>>>>
>>>>> On 3/10/10 10:26 AM, Will Metcalf wrote:
>>>>>>> 1) Use http_cookie in the rule as you note in the case that works.
>>>>>> Right so the reason that I cc'd the emerging list is that they are not
>>>>>> using the http_* modifiers to maintain compatibility with older
>>>>>> versions of snort.  Shouldn't this buffer at least be available to
>>>>>> match on via rawbytes  which is what would be consistent with what you
>>>>>> have done with telnet and dcerpc.  Does this also not add a potential
>>>>>> evasion method if this is the intended behavior, one that perhaps
>>>>>> VRT/ET should be made aware of.  I haven't gone through very many of
>>>>>> the sigs but since the normalized buffer begins with the "Cookie:"
>>>>>> instead of the value isn't there a now a potential evasion if I can
>>>>>> get another sig to trip on the same packet as say sid 2136.
>>>>>>
>>>>>> Regards,
>>>>>>
>>>>>> Will
>>>>>>
>>>>>> On Wed, Mar 10, 2010 at 8:15 AM, Steven Sturges
>>>>>> <steve.sturges at ...402...> wrote:
>>>>>>> Will--
>>>>>>>
>>>>>>> In the 2nd rule the only content is an HTTP cookie.  Without using
>>>>>>> http_cookie, it would try to match the raw data.
>>>>>>>
>>>>>>> With HTTP Inspect enabled, it is separating the headers, cookie,
>>>>>>> method, etc from the raw data, hence all of the modifier keywords
>>>>>>> that you can use with content -- can use more than one together.
>>>>>>>
>>>>>>> Without specifying http_cookie in the rule and when HTTP Inspect
>>>>>>> enabled (and cookie inspection enabled in 2.8.6), if the pattern
>>>>>>> matcher searches any of the HTTP buffers, it doesn't search the raw
>>>>>>> data.  If no rules use the specific HTTP buffers or there are no HTTP
>>>>>>> buffers, the pattern matcher will search the raw data, which is why
>>>>>>> the rule byte itself (without http_cookie) works.
>>>>>>>
>>>>>>> This was done to avoid going over the same data twice in the pattern
>>>>>>> matcher, and it is working as it is designed.
>>>>>>>
>>>>>>> Two options:
>>>>>>>
>>>>>>> 1) Use http_cookie in the rule as you note in the case that works.
>>>>>>>
>>>>>>> 2) Turn off HTTP Inspect (not realistic)
>>>>>>>
>>>>>>> Cheers
>>>>>>> -steve
>>>>>>>
>>>>>>> Will Metcalf wrote:
>>>>>>>> hmmm I don't think so.  Look at first test.  both rules fire.
>>>>>>>>
>>>>>>>> Regards,
>>>>>>>>
>>>>>>>> Will
>>>>>>>>
>>>>>>>> On Tue, Mar 9, 2010 at 10:31 PM, beenph <beenph at ...2499...> wrote:
>>>>>>>>> I will try a wild guess, what is your event_queue size like?
>>>>>>>>>
>>>>>>>>> Its probably a bug or something  that need clarification regarding
>>>>>>>>> http_cookie and http_inspect, but mabey http_cookie enable a modifier
>>>>>>>>> in http_inspect that alter alerting behavior when event_queue is at 1
>>>>>>>>> (since i guess both "alerts" are part of the same normalized http
>>>>>>>>> stream)
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> -elz
>>>>>>>>> ps: didin't run the pcap and rules test.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> On Tue, Mar 9, 2010 at 11:15 PM, Will Metcalf <william.metcalf at ...1066....2499...> wrote:
>>>>>>>>>> failing to use the http_cookie modifier on a rule where there is
>>>>>>>>>> another rule that matches the same packet makes a rule that should
>>>>>>>>>> fire fail.
>>>>>>>>>>
>>>>>>>>>> src/snort -V
>>>>>>>>>>
>>>>>>>>>>   ,,_     -*> Snort! <*-
>>>>>>>>>>  o"  )~   Version 2.8.5.3 (Build 124)
>>>>>>>>>>   ''''    By Martin Roesch & The Snort Team:
>>>>>>>>>> http://www.snort.org/snort/snort-team
>>>>>>>>>>           Copyright (C) 1998-2009 Sourcefire, Inc., et al.
>>>>>>>>>>           Using PCRE version: 7.8 2008-09-05
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> src/snort -k none -q -A console -c etc/snort.conf -l ./ -r oisfsearchnums.pcap
>>>>>>>>>>
>>>>>>>>>> #this combo works
>>>>>>>>>> #alert tcp any any -> any any (msg:"http_client_body";
>>>>>>>>>> content:"searchword="; uricontent:"/index.php"; nocase;
>>>>>>>>>> classtype:bad-unknown; sid:59; rev:1;)
>>>>>>>>>> #alert tcp any any -> any any (msg:"http_cookie match ";
>>>>>>>>>> content:"e6504ae48c99f09df7f58996aacbb6b0=120e494ce857d6ceeef89f9678d4d703";
>>>>>>>>>> http_cookie; classtype:bad-unknown; sid:68; rev:1;)
>>>>>>>>>> #
>>>>>>>>>> #03/07-21:19:54.242506  [**] [1:59:1] http_client_body [**]
>>>>>>>>>> [Classification: Potentially Bad Traffic] [Priority: 2] {TCP}
>>>>>>>>>> 192.168.100.17:38111 -> 96.43.130.5:80
>>>>>>>>>> #03/07-21:19:54.242506  [**] [1:68:1] http_cookie match  [**]
>>>>>>>>>> [Classification: Potentially Bad Traffic] [Priority: 2] {TCP}
>>>>>>>>>> 192.168.100.17:38111 -> 96.43.130.5:80
>>>>>>>>>> #03/07-21:19:54.364173  [**] [1:68:1] http_cookie match  [**]
>>>>>>>>>> [Classification: Potentially Bad Traffic] [Priority: 2] {TCP}
>>>>>>>>>> 192.168.100.17:38111 -> 96.43.130.5:80
>>>>>>>>>>
>>>>>>>>>> #the second rule does not fire
>>>>>>>>>> #alert tcp any any -> any any (msg:"http_client_body + depth";
>>>>>>>>>> content:"searchword="; uricontent:"/index.php"; nocase;
>>>>>>>>>> classtype:bad-unknown; sid:59; rev:1;)
>>>>>>>>>> #alert tcp any any -> any any (msg:"http_cookie match";
>>>>>>>>>> content:"e6504ae48c99f09df7f58996aacbb6b0=120e494ce857d6ceeef89f9678d4d703";
>>>>>>>>>> classtype:bad-unknown; sid:68; rev:1;)
>>>>>>>>>> #
>>>>>>>>>> #03/07-21:19:54.242506  [**] [1:59:1] http_client_body + depth [**]
>>>>>>>>>> [Classification: Potentially Bad Traffic] [Priority: 2] {TCP}
>>>>>>>>>> 192.168.100.17:38111 -> 96.43.130.5:80
>>>>>>>>>>
>>>>>>>>>> #this rule fires when used on it's own.
>>>>>>>>>> #alert tcp any any -> any any (msg:"http_cookie match";
>>>>>>>>>> content:"e6504ae48c99f09df7f58996aacbb6b0=120e494ce857d6ceeef89f9678d4d703";
>>>>>>>>>> classtype:bad-unknown; sid:68; rev:1;)
>>>>>>>>>> #
>>>>>>>>>> #03/07-21:19:54.242506  [**] [1:68:1] http_cookie match [**]
>>>>>>>>>> [Classification: Potentially Bad Traffic] [Priority: 2] {TCP}
>>>>>>>>>> 192.168.100.17:38111 -> 96.43.130.5:80
>>>>>>>>>> #03/07-21:19:54.364173  [**] [1:68:1] http_cookie match [**]
>>>>>>>>>> [Classification: Potentially Bad Traffic] [Priority: 2] {TCP}
>>>>>>>>>> 192.168.100.17:38111 -> 96.43.130.5:80
>>>>>>>>>>
>>>>>>>>>> ------------------------------------------------------------------------------
>>>>>>>>>> Download Intel® Parallel Studio Eval
>>>>>>>>>> Try the new software tools for yourself. Speed compiling, find bugs
>>>>>>>>>> proactively, and fine-tune applications for parallel performance.
>>>>>>>>>> See why Intel Parallel Studio got high marks during beta.
>>>>>>>>>> http://p.sf.net/sfu/intel-sw-dev
>>>>>>>>>> _______________________________________________
>>>>>>>>>> Snort-devel mailing list
>>>>>>>>>> Snort-devel at lists.sourceforge.net
>>>>>>>>>> https://lists.sourceforge.net/lists/listinfo/snort-devel
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>> ------------------------------------------------------------------------------
>>>>>>>> Download Intel® Parallel Studio Eval
>>>>>>>> Try the new software tools for yourself. Speed compiling, find bugs
>>>>>>>> proactively, and fine-tune applications for parallel performance.
>>>>>>>> See why Intel Parallel Studio got high marks during beta.
>>>>>>>> http://p.sf.net/sfu/intel-sw-dev
>>>>>>>> _______________________________________________
>>>>>>>> Snort-devel mailing list
>>>>>>>> Snort-devel at lists.sourceforge.net
>>>>>>>> https://lists.sourceforge.net/lists/listinfo/snort-devel
>>>>>>>>
>>>>>> ------------------------------------------------------------------------------
>>>>>> Download Intel® Parallel Studio Eval
>>>>>> Try the new software tools for yourself. Speed compiling, find bugs
>>>>>> proactively, and fine-tune applications for parallel performance.
>>>>>> See why Intel Parallel Studio got high marks during beta.
>>>>>> http://p.sf.net/sfu/intel-sw-dev
>>>>>> _______________________________________________
>>>>>> Snort-devel mailing list
>>>>>> Snort-devel at lists.sourceforge.net
>>>>>> https://lists.sourceforge.net/lists/listinfo/snort-devel
>>>
>>
>




More information about the Snort-devel mailing list