Bug #5780
closedHTTP/2 - FN when matching on multiple http2.header contents
Added by Brandon Murphy almost 2 years ago. Updated over 1 year ago.
Description
It appears that when attempting to combine two different HTTP2 headers into a single rule, no alert is produced.
Consider the following rules and the attached pcap, which contains a single tcp session with a single HTTP2 stream.
alert http2 $HOME_NET any -> any any (msg:"HTTP2 - Single Header - Authority"; flow:established,to_server; http2.header; content:"authority: bugertor"; sid:1;) alert http2 $HOME_NET any -> any any (msg:"HTTP2 - Single Header - Method"; flow:established,to_server; http2.header; content:"method: GET"; sid:2;) alert http2 $HOME_NET any -> any any (msg:"HTTP2 - Two Headers - Authority/Method"; flow:established,to_server; http2.header; content:"method: GET"; content:"authority: bugertor.com"; sid:3;)
Current Behavior¶
Only sid:1 and sid:2 fire
Expected Behavior¶
All three signatures should fire.
HTTP Keyword Overloading¶
Once the correct http2 configuration option is enabled (http1-rules), the using the standard http1 keywords (http.method, http.host) the below signature works as expected.
alert http $HOME_NET any -> any any (msg:"HTTP2 - Overload Test"; flow:established,to_server; http.method; content:"GET"; http.host; content:"bugertor.com"; sid:4;)
Files
http2_multiple_headers.pcap (1.7 KB) http2_multiple_headers.pcap | Brandon Murphy, 01/10/2023 03:53 PM |
Updated by Brandon Murphy almost 2 years ago
- File http2_multiple_headers.pcap http2_multiple_headers.pcap added
- Description updated (diff)
- Affected Versions 6.0.9, git master added
Updated by Victor Julien almost 2 years ago
- Status changed from New to Assigned
- Assignee changed from OISF Dev to Philippe Antoine
Updated by Philippe Antoine almost 2 years ago
Workaround use http.header
instead of http2.header
cf alert http2 any any -> any any (msg:"HTTP2 - Two Headers - Authority/Method"; http.header; content:"method: GET"; content:"authority: bugertor.com"; sid:4;)
Updated by Philippe Antoine almost 2 years ago
Not HTTP2 specific, multi-buffer thing
Another reproducer is
alert mqtt any any -> any any (msg:"HTTP2 - Two Headers - Authority/Method"; mqtt.subscribe.topic; content:"topicX"; mqtt.subscribe.topic; content:"topicY"; sid:5;) alert mqtt any any -> any any (msg:"HTTP2 - Two Headers - Authority/Method"; mqtt.subscribe.topic; content:"topicY"; sid:6;) alert mqtt any any -> any any (msg:"HTTP2 - Two Headers - Authority/Method"; mqtt.subscribe.topic; content:"topicX"; sid:7;)
on suricata-verify/tests/mqtt-sub-rules/mqtt5_sub_userpass.pcap
Updated by Brandon Murphy almost 2 years ago
Philippe Antoine wrote in #note-4:
Workaround use
http.header
instead ofhttp2.header
cfalert http2 any any -> any any (msg:"HTTP2 - Two Headers - Authority/Method"; http.header; content:"method: GET"; content:"authority: bugertor.com"; sid:4;)
Oh, that's interesting! Do you happen to know if this only works when overloading is enabled (in 6.0.x)?
Updated by Philippe Antoine almost 2 years ago
Do you happen to know if this only works when overloading is enabled (in 6.0.x)?
I tested on master.
And I am not sure what you mean by overloading... What is it ?
By the way, thanks Brandon for this interesting report ;-)
Also, I would not expect http2.header; content:"method: GET"; content:"authority: bugertor.com";
. to match because you want both contents in a single header (like http2.header; content:"method"; content:"GET";
does match)
But I would expect http2.header; content:"method: GET"; http2.header; content:"authority: bugertor.com";
to match and it does not...
What do you think about it as a rule writer ?
Do you understand "http header" as a single key value pair, or as all the key value pairs in a HTTP request/response ?
Updated by Brandon Murphy almost 2 years ago
Philippe Antoine wrote in #note-7:
Do you happen to know if this only works when overloading is enabled (in 6.0.x)?
I tested on master.
And I am not sure what you mean by overloading... What is it ?
Enabling the http1-rules option within the http2 app layer configuration item. I stole the term from https://redmine.openinfosecfoundation.org/issues/4067
By the way, thanks Brandon for this interesting report ;-)
Also, I would not expect
http2.header; content:"method: GET"; content:"authority: bugertor.com";
. to match because you want both contents in a single header (likehttp2.header; content:"method"; content:"GET";
does match)
But I would expecthttp2.header; content:"method: GET"; http2.header; content:"authority: bugertor.com";
to match and it does not...
What do you think about it as a rule writer ?
I do not like it. I think this is a large difference from the behavior of http.header
which "contains all of the extracted headers in a single buffer".
Do you understand "http header" as a single key value pair, or as all the key value pairs in a HTTP request/response ?
"as all the key value pairs in a HTTP request/response" only because it is this way with http.header
Very often we need the ability to write rules on the unique combination of headers and their values. The inability to combine multiple headers names and their values severally limits the rules that can be created, in this case, specific to HTTP/2, though apparently other protocols as well.
The completion of https://redmine.openinfosecfoundation.org/issues/4067 and the new default configuration in suri 7 which enables http1-rules helps a lot to workaround the current limitations of http2.header, though I see it as a problematic limitation non the less.
Updated by Philippe Antoine almost 2 years ago
So, trying to sump up, there are 2 issues :
- http2.header should match http.header behavior even if I do not like this name, I can rename the current http2.header keyword to http2.single_header or such...
- multi buffer should have a mechanism to match different content on different buffer
How does that sound ?
alert http any any -> any any (msg:"HTTP2 - Two Headers - Authority/Method"; http.header; content:"method: GET"; http.header; content:"authority: bugertor.com"; sid:4;)
. works on master-6.0.x with http1-rules
Updated by Brandon Murphy almost 2 years ago
Philippe Antoine wrote in #note-9:
So, trying to sump up, there are 2 issues :
- http2.header should match http.header behavior even if I do not like this name, I can rename the current http2.header keyword to http2.single_header or such...
Ok, this seems great. I agree, I've never been a fan of the name and I've always wondered why it wasn't http.headers
(plural) but I assume the answer is legacy compatibility.
- multi buffer should have a mechanism to match different content on different buffer
Sorry, I'm not sure I fully understand/follow this, can you provide more detail?
How does that sound ?
alert http any any -> any any (msg:"HTTP2 - Two Headers - Authority/Method"; http.header; content:"method: GET"; http.header; content:"authority: bugertor.com"; sid:4;)
. works on master-6.0.x withhttp1-rules
Awesome, I learned something new there!
Updated by Philippe Antoine almost 2 years ago
- Status changed from Assigned to In Review
Brandon, what do you think about https://github.com/OISF/suricata/pull/8371 ?
I assume the answer is legacy compatibility.
I think it is the answer...
Updated by Philippe Antoine almost 2 years ago
- Related to Feature #5784: detect: allow cross buffer inspection on multi-buffer matches added
Updated by Brandon Murphy almost 2 years ago
Brandon, what do you think about https://github.com/OISF/suricata/pull/8371 ?
I'll be honest, I am not familiar enough with Suri's code base (or code in general) to make any judgment about how this looks.
I have to rely on the PR/Code Review process to flush this out!
Updated by Philippe Antoine almost 2 years ago
- Target version changed from TBD to 7.0.0-rc1
Updated by Philippe Antoine almost 2 years ago
- Target version changed from 7.0.0-rc1 to 7.0.0-rc2
Updated by Victor Julien over 1 year ago
Thinking about this I agree we need a solution but also agree that the http_header
/ http.header
keyword is poorly named. How about we do something like
http.request_header
- matches each individual header (the single_header suggestion)http.request_headers
- http.header behavior limited to requestshttp.request_headers.raw
- http.header.raw for requestshttp.response_header
http.response_headers
http.response_headers.raw
These would all apply to http1 and http2, where they can be limited to a http version through alert http1
or I suppose http.protocol; content:"HTTP/1"
Updated by Philippe Antoine over 1 year ago
But we keep old http.header
for compatibility, right ?
And do we change http2.header
behavior ? to match the one from http.header
(while creating http.request_header
to have the single header feature)
Updated by Victor Julien over 1 year ago
http_header
and http.header
need to stay as is.
Wrt http2.header
, we should probably remove it. It's not part of a stable release as a non-experimental release yet, so we still have freedom there.
Updated by Philippe Antoine over 1 year ago
Updated by Brandon Murphy over 1 year ago
Sorry, i'm a bit late getting caught up here, but regarding a solution to support buffers of individual headers, I'd rather see this feature request implemented
Updated by Philippe Antoine over 1 year ago
- Status changed from In Review to Resolved
Updated by Philippe Antoine over 1 year ago
https://github.com/OISF/suricata/pull/9013 for adding a warning in master6
Updated by Philippe Antoine over 1 year ago
- Status changed from Resolved to Closed
Master6 got a warning cf https://github.com/OISF/suricata/pull/9074