LDAP Injection Rule

[dnkolegov]

The following LDAP injection vectors from Alonso-Parada research are not detected by current LDAP Injection Rule:

• foo)(sn=100
• printer)(uid=*)
• printer)(department=fa*)
• printer)(department=*fa*)

Burp Suite uses the following vectors to test an LDAP Injection and they are also not detected:

• eb9adbd87d)(sn=*
• eb9adbd87d)!(sn=*
• *)(sn=*
• *)!(sn=*

Also it is not obvious the purpose of the top and middle parts of regular expression that check values beginning with (:

(?:\((?:\W*?(?:objectc(?:ategory|lass)|homedirectory|[gu]idnumber|cn)\b\W*?=|[^\w\x80-\xFF]*?[\!\&\|][^\w\x80-\xFF]*?\()) 

Which LDAP injection context is supposed here?

[dune73]

@dnkolegov : Let's try and get this sorted out.

Do you have the capacity to test any new rules proposed here? My ldap knowledge is fairly limited and I do not want to introduce false positives here.

I think together we can get it done.

[dnkolegov]

I propose to consider the following regex:

^[^:\(\)\&\|\!\<\>\~]*\)(?:\((?:[^,\(\)\=\&\|\!\<\>\~]+[><~]?=|[&!|](?:\)|\()?)|\)\([\&\|\!]|[&!|]\([^\(\)\=\&\|\!\<\>\~]+[><~]?=[^:\(\)\&\|\!\<\>\~]*)

Tests:

foo)(sn=100
foo)(&)
documents)(security_level=*))(&(directory=documents
printer))(&
printer)(&
printer)(uid=*)
printer)(department=fa*)
printer)(groupType:1.2.840.113556.1.4.803:=2147483648)
printer)(department=*fa*)
*pri*nter*)(department=*fa*)
*)(objectClass=*))(&(objectClass=void
*)(objectClass=users))(&(objectClass=foo
void)(objectClass=users))(&(objectClass=void)
eb9adbd87d)(sn=*
eb9adbd87d)!(sn=*
*)(sn=*
*)!(sn=*
ka0x)(|(homedirectory=*)
5faa0382d747b754)(sn=*
5faa0382d747b754)!(sn=*
*)(uid=*))(|(uid=*
aaa*aaa)(groupType:1.2.840.113556.1.4.803:=2147483648)
aaa*aaa)(memberof:1.2.840.113556.1.4.1941:=cn=Group1,OU=groupsOU,DC=x)
aaa*aaa)(cn>=bob)
cn=Group1,OU=groupsOU,DC=x)(memberof:1.2.840.113556.1.4.1941:=cn=Group1,OU=groupsOU,DC=x)
cn=Group1,OU=groupsOU,DC=x)(memberof:1.2.840.113556.1.4.1941:=cn=Group1,OU=groupsOU,DC=x))(&(memberof:1.2.840.113556.1.4.1941:=cn=Group1,OU=groupsOU,DC=x)
eb9adbd87d)(userAccountControl:1.2.840.113556.1.4.803:=65536)

[dune73]

Thank you. That looks remarkable.

So this is a list of attack strings and a regex, which detects them?

What about false positives?

[dnkolegov]

I tested this regex only for some valid LDAP filters:

(&(objectCategory=computer)  (userAccountControl:1.2.840.113556.1.4.803:=8192))
(objectSID=S-1-5-21-73586283-152049171-839522115-1111)
(userAccountControl:1.2.840.113556.1.4.803:=67108864)(&(objectCategory=group)(groupType:1.2.840.113556.1.4.803:=2147483648))

[csanders-git]

@dune73 i guess this never got added - it is great work though I think we should put it on the 3.1 list.

[dune73]

I like this as well. But I think we need somebody with real ldap experience telling us, if FPs are likely.
LDAP traffic is rare so I totally see us ship this regex and kill some site with it.

[dnkolegov]

Updated regex (fixed whitespaces):

^[^:\(\)\&\|\!\<\>\~]*\)\s*(?:\((?:[^,\(\)\=\&\|\!\<\>\~]+[><~]?=|\s*[&!|]\s*(?:\)|\()?\s*)|\)\s*\(\s*[\&\|\!]\s*|[&!|]\s*\([^\(\)\=\&\|\!\<\>\~]+[><~]?=[^:\(\)\&\|\!\<\>\~]*)

[lifeforms]

A new exploit: https://blog.ripstech.com/2017/joomla-takeover-in-20-seconds-with-ldap-injection-cve-2017-14596/

[dune73]

This is still not added and given our lack of knowledge for everything LDAP we would rely on @dnkolegov's word to go and include this into our rule set.

So @dnkolegov, are you actively using this rule on your server(s)? Any feedback with regards to false positives?

[dnkolegov]

This rule was developed and used within a proprietary Positive Technologies' Application Firewall. I do not remember customer requests regarding this rule. But I have not been working there for 2 years, so I do not have actual information.

[dune73]

Thank you for writing in again, @dnkolegov!

So what do we do guys: I see this as a new paranoia level 3 rule, or we drop the endeavour despite the good contribution by @dnkolegov because we simply lack the knowledge to support this.

[lifeforms]

I would argue to bring back the rule in the default set (paranoia level 1). Vulnerability to LDAP injection may be somewhat rare, but nevertheless it's a well published attack class, I could see it being important to enterprise users, and even OWASP and Webappsec wikis have extensive pages on it.

The LDAP injection rule was present in CRS 2.x but removed, I have no data why, but I can't recall ever having operations trouble with the rule myself.

Our old rule was apparently not having enough coverage, but with the improvements by @dnkolegov
it's a good start, he has supplied material for adding unit tests, and if users turn up more payloads we can handle them incrementally.

A risk could be false positives, but we'll have our regular development cycle with the release candidates to flesh them out. We can be a little more bold at the start of a cycle, and if we run into trouble we can always put the rule in a higher paranoia level.

[dune73]

OK. Let's do that. I'll prepare a PR.