Bug #849
closedNot alerting on invalid http request Content-Length
Description
In the pcaps atatched
1) - http-events-abnormal-No-Content-Length.pcap
2) - http-events-abnormal-Invalid-Content-Length.pcap
3) - InvalidContentLengthApacheResponse.pcap
4) - ValidContentLengthApacheResponse.pcap
- 2) is with invalid content lenght - "Content-Length: 2040\r\n" added to the http request, packet 4.
- 1) is with the valid content length
In situation 2) wireshark does not recognize the http request - just recognizes it as valid TCP segment , which would be correct I think, since the content length is invalid.
Suricata recognizes 2) as a http request.
It is the exact same situation if the same is mirror but for the http response.
However the rules :
alert http any any -> any any (msg:"SURICATA HTTP invalid content length field in request"; flow:established,to_server; app-layer-event:http.invalid_content_length_field_in_request; flowint:http.anomaly.count,+,1; classtype:protocol-command-decode; sid:2221007; rev:1;) alert http any any -> any any (msg:"SURICATA HTTP invalid content length field in response"; flow:established,to_client; app-layer-event:http.invalid_content_length_field_in_response; flowint:http.anomaly.count,+,1; classtype:protocol-command-decode; sid:2221008; rev:1;)
From http-events.rules do not generate an alert (or any of the rules in http-events.rules for that matter) in either respective case (request or response).
3) and 4) are the same case but real cases against Apache web server
Does not alert with 2.0dev (rev 5157ce1) or 1.4.3 or 2.0dev (rev cd7b4fa - latest git master )
Running Suri with:
suricata -c /etc/suricata/suricata.yaml -S /http-events.rules -r /root/Work/suricata/BUG/InvalidContentLength/http-events-abnormal-Invalid-Content-Length.pcap --runmode=single
Thanks
Files
Updated by Peter Manev over 11 years ago
Another such pacp added.
This is from the actual tool/website testing site added - http://noxxi.de/research/dubious-content-length.html
as reported by Will Metcalf.
Packet number 6 is one such case where the "Content-length" is 4891 but the actual data is 1448.
thanks
Updated by Peter Manev over 11 years ago
Interesting addition -
with the pcap file Content-length-bug-noxxi-de.pcap - from the actuall test web page, 1.4.3 alerts 25 times, like so
07/13/2013-11:46:03.824560 [**] [1:2221008:1] SURICATA HTTP invalid content length field in response [**] [Classification: Generic Protocol Command Decode] [Priority: 3] {TCP} 37.221.197.243:8001 -> 192.168.1.71:4153
07/13/2013-11:46:03.691918 [**] [1:2221008:1] SURICATA HTTP invalid content length field in response [**] [Classification: Generic Protocol Command Decode] [Priority: 3] {TCP} 37.221.197.243:8001 -> 192.168.1.71:4144
07/13/2013-11:46:03.835038 [**] [1:2221008:1] SURICATA HTTP invalid content length field in response [**] [Classification: Generic Protocol Command Decode] [Priority: 3] {TCP} 37.221.197.243:8001 -> 192.168.1.71:4154
07/13/2013-11:46:03.794177 [**] [1:2221010:1] SURICATA HTTP unable to match response to request [**] [Classification: Generic Protocol Command Decode] [Priority: 3] {TCP} 37.221.197.243:8001 -> 192.168.1.71:4148
but git master - 2.0dev (rev 73e27c1) - does not alert.
Thanks
Updated by Philippe Antoine over 5 years ago
To me, the content length should now be handled fine after https://github.com/OISF/libhtp/commit/c0c87b4c560aae3850c7d90de1bdcb4fe966c9f0
That means :
- duplicate content-length headers with same value
- duplicate content-length headers with different values (first one is then used as http evader showed us that was the case for browsers)
- less or more data than content-length (but these do not generate an alert about the content length)
From your examples, it looks like you are expecting an alert when a connection is interrupted before the whole content length could have been sent.
Is this the case ?
Updated by Victor Julien over 5 years ago
Philippe can you turn the pcaps into suricata-verify tests?
Updated by Philippe Antoine over 5 years ago
Victor, to me, these are duplicates of the http evader cases (by the same person) https://noxxi.de/research/http-evader.html
Updated by Victor Julien over 5 years ago
So can this be closed @Philippe Antoine ?