OWASP corerulet warning "invalid http request line" triggered by CONNECT method - owasp

Question

Summary:
I have setup a basic WAF with mod-security and the OWASP coreruleset 3.3.2. When using the WAF I see lots of warnings in modsec_audit.log regarding the CONNECT method, which trigger crs rule 920100:
Message: Warning. Match of "rx
^(?i:(?:[a-z]{3,10}\s+(?:\w{3,7}?://[\w\-\./](?::\d+)?)?/[^?#](?:\?[^#\s])?(?:#[\S])?|connect
(?:\d{1,3}\.){3}\d{1,3}\.?(?::\d+)?|options
\)\s+[\w\./]+|get /[^?#](?:\?[^#\s])?(?:#[\S])?)$" against
"REQUEST_LINE" required. [file
"/etc/httpd/modsecurity.d/owasp-modsecurity-crs/rules/REQUEST-920-PROTOCOL-ENFORCEMENT.conf"]
[line "63"] [id "920100"] [msg "Invalid HTTP Request Line"] [data
"CONNECT oneofmy.longer.hostname.here.abcde.com:443 HTTP/1.1"]
[severity "WARNING"] [ver "OWASP_CRS/3.3.2"] [tag "application-multi"]
[tag "language-multi"] [tag "platform-multi"] [tag "attack-protocol"]
[tag "paranoia-level/1"] [tag "OWASP_CRS"] [tag "capec/1000/210/272"]
Details regarding setup:
I set up the WAF with mod_security 2.9.2 and httpd. I used the crs-setup.conf.example provided by crs and only modified these two settings regarding threshhold and allowing the CONNECT http method. (CONNECT method is used in our setup for proxy purposes).
SecAction \
"id:900110,\
phase:1,\
nolog,\
pass,\
t:none,\
setvar:tx.inbound_anomaly_score_threshold=5,\
setvar:tx.outbound_anomaly_score_threshold=4"
SecAction \
"id:900200,\
phase:1,\
nolog,\
pass,\
t:none,\
setvar:'tx.allowed_methods=GET HEAD POST OPTIONS CONNECT'"
What I have tried so far:
Google a lot about crs and mod-security rules. As per my current, admittedly limited mod-security/crs understanding, basically the regex of the rule 920100 does not match against the request line CONNECT oneofmy.longer.hostname.here.abcde.com:443 HTTP/1.1
Tried to find more details how the regex for 920100 is constructed.
looked up how the request line for a CONNECT string is defined in RFC. To me, the request line looks fine as per the RFC.
tried to debug the regexp with a site like regex101 and my request line. I did not succeed in debugging.

Answer 1

Looking at the part of the regular expression for rule 920100 that deals with the CONNECT method:
connect (?:\d{1,3}\.){3}\d{1,3}\.?(?::\d+)?
we can see that it's looking for an IPv4 address. For example, it would match on (and hence not raise an alert for) a request line like the following:
CONNECT 1.2.3.4:80 HTTP/1.1
As you say, the RFC seems to allow for domain names to be used with the CONNECT method, however, rule 920100 is not currently written to allow for this. This seems to be the cause of your issue.
With further investigation and testing, the rule can probably be expanded in a future CRS release to allow for domain names in CONNECT requests.
What can you do here and now?
To address this issue right now, you can add a rule exclusion to your Apache configuration. There are two ways you can do this:
Remove the offending rule completely by excluding it at configure time, like so:
#
# -- CRS Rule Exclusion: 920100 - Invalid HTTP Request Line
#
# Turn off rule due to issues with CONNECT requests
#
SecRuleRemoveById 920100
As a configure time rule exclusion, this should be placed after the inclusion of the Core Rule Set in your Apache configuration (because the rule that's being removed must be added/included in the first place before it can then be removed!).
Remove the offending rule conditionally, only for CONNECT requests, by excluding it at runtime, like so:
#
# -- CRS Rule Exclusion: 920100 - Invalid HTTP Request Line
#
# Turn off rule for CONNECT requests, which cause issues
#
SecRule REQUEST_METHOD "#streq CONNECT" \
"id:1000,\
phase:1,\
pass,\
nolog,\
ctl:ruleRemoveById=920100"
As a runtime time rule exclusion, this should be placed before the inclusion of the CRS in your Apache configuration (because the rule that's being modified must be modified before it fires off during execution).
A small word of warning
I can't find any mention of anyone else using ModSecurity with the CRS in this way, i.e. fielding CONNECT requests. The CRS rules aren't written to accommodate having a domain name in the request line, so you may find that you run into further false positives as a result. You would need to deal with those in a similar way to that described above.
(You can find an excellent tutorial on the subject of writing rule exclusions here: https://www.netnea.com/cms/apache-tutorial-8_handling-false-positives-modsecurity-core-rule-set/).
Hopefully this helps.