[Discussion] Suricata rule not giving alerts

[Anoop Saldanha]

On Sun, Jul 20, 2014 at 5:50 PM, Jelte <masterjel5000 at hotmail.com> wrote:
> Jelte schreef op 7/15/2014 12:08 AM:
>> Victor Julien schreef op 7/14/2014 9:27 AM:
>>> On 07/14/2014 12:21 AM, Jelte O. wrote:
>>>> Hello all,
>>>>
>>>> I have a rule from the ET rule-set to alert against an attack that is
>>>> used to exploit a vulnerability in nginx 1.3.9-1.4.0. In order to
>>>> trigger this rule I loaded an exploit module in Metasploit and fired it
>>>> on my server.
>>>>
>>>> The vulnerability:
>>>> https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2013-2028
>>>> The rule: http://doc.emergingthreats.net/bin/view/Main/2016918
>>>> The Metasploit module:
>>>> https://github.com/rapid7/metasploit-framework/blob/master/modules/exploits/linux/http/nginx_chunked_size.rb
>>>>
>>>> I'll repeat the rule here:
>>>>
>>>>
>>>>     alert tcp $EXTERNAL_NET any -> $HTTP_SERVERS $HTTP_PORTS (msg:"ET
>>>>     WEB_SERVER Possible NGINX Overflow CVE-2013-2028 Exploit Specific";
>>>>     flow:established,to_server; content:"chunked"; http_header; nocase;
>>>>     fast_pattern:only; pcre:"/Transfer-Encoding\x3a[^\r\n]*?chunked/Hi";
>>>>     pcre:"/^[\r\n\s]*?[^\r\n]+HTTP\/1\.\d[^\r\n]*?\r?\n((?!(\r?\n\r?\n)).)*?Transfer-Encoding\x3a[^\r\n]*?Chunked((?!(\r?\n\r?\n)).)*?\r?\n\r?\n[\r\n\s]*?(f{6}[8-9a-f][0-9a-f]|[a-f0-9]{9})/si";
>>>>     reference:url,www.vnsecurity.net/2013/05/analysis-of-nginx-cve-2013-2028/;
>>>>     reference:url,github.com/rapid7/metasploit-framework/blob/master/modules/exploits/linux/http/nginx_chunked_size.rb;
>>>>     classtype:attempted-admin; sid:2016918; rev:6;)
>>>>
>>>> My attack did not generate any alerts. However, as soon as I removed the
>>>> "http_header;" and changed "/Hi" to "/i" (in the first pcre) the rule
>>>> started generating alerts. From this it seems like the HTTP header is
>>>> not complete/not recognized by Suricata. However, when I do an extended
>>>> logging on the HTTP traffic, I do see entries like:
>>>>
>>>> 07/13/14-23:45:27.830342 - - Chunked HTTP/1.1 GET mifpudtilvpjqsjl / - 0
>>>> x.x.x.x:40590 -> y.y.y.y:80
>>>>
>>>> My "customformat" for the http-log contains "%{Transfer-Encoding}i",
>>>> which would actually be the "contents of the defined HTTP Request Header
>>>> name" according to the documentation (refer to
>>>> https://redmine.openinfosecfoundation.org/projects/suricata/wiki/Custom_http_logging).
>>>> <https://redmine.openinfosecfoundation.org/projects/suricata/wiki/Custom_http_logging%29.>
>>>>
>>>> I also created packet dumps of both legitimate web traffic and this
>>>> attack and analyzed the streams in Wireshark. In both dumps there are
>>>> TCP PDU's which are re-assembled, but in the valid web traffic Wireshark
>>>> labels the protocol for some of the fully assembled client-to-server
>>>> packets as HTTP while for the attack there are only TCP packets from the
>>>> client to the server.
>>>>
>>>> I am wondering why the HTTP header is not available. I am not sure if
>>>> this is caused by Suricata, my OS/network interface or the rule itself.
>>>> I hope someone can help me out!
>>> As a first step, I'd suggest walking through this page
>>> https://redmine.openinfosecfoundation.org/projects/suricata/wiki/Self_Help_Diagrams
>>>
>> Thanks Will and Victor for your replies! I didn't know of the Self Help
>> Diagrams yet, unfortunately, because it could have saved me some time on
>> an earlier issue I had with bad TCP checksums. Anyway, I have traffic in
>> the stats log and the http log, I already disabled checksum validation
>> and the startup log says the rule has successfully been loaded (for
>> testing purposes I have only included this single rule). Also, my server
>> is not part of a VLAN.
>>
>> Offloading is enabled for my NIC. If I issue the "ethtool --show-offload
>> eth0" command I see that the rx-checksumming, the tx-checksumming,
>> tcp-segmentation-offload and the generic-segmentation-offload are on.
>> The NIC does not seem to support turning off the rx-checksumming (I get
>> "Operation not supported") but I was able to turn off the others. This
>> didn't have any effect, though. I also created another packet dump of
>> the attack after disabling these settings and compared this one to the
>> one I created before turning off the tx-checksumming and segmentation
>> offloading and they both matched. I already disabled the checksum
>> validation in Suricata to get rid off the invalid checksum errors I had
>> in the beginning.
>>
>> I went to all the self help diagrams but I still couldn't find the
>> cause. I did notice that I forgot to mention two important things:
>>
>> - Snort (on the same system) does generate an alert for that particular
>> rule.
>> - Suricata does generate alerts for other rules in which I filter on
>> content from the http header. For instance, a HTTP request inside
>> Firefox generates alerts for a rule that includes 'content:"Mozilla";
>> http_header; nocase;'.
>>
>> It seems like it has to do with the specific exploit that is used in
>> Metasploit. Refer to
>> https://github.com/rapid7/metasploit-framework/blob/master/modules/exploits/linux/http/nginx_chunked_size.rb
>>
>> Do you know anything else I can do for debugging? I have included the
>> pcap of the attack in the attachment.
>>

Does changing the "http_header" to "http_raw_header" and 'H' to 'D'
make any difference?

-- 
-------------------------------
Anoop Saldanha
http://www.poona.me
-------------------------------