[Snort-users] Understanding Snort Internals

Matthew Watchinski mwatchinski at ...1935...
Tue Jun 12 10:50:25 EDT 2007


Giorgio Moscardi wrote:
> Hi,
> I'm new to this mailing list, so first of all hello everybody :).
> 
> I'm trying to get a deeper understanding of how Snort works internally, for a 
> university project. I have posted this message to the Snort forums, already, 
> but it seems nobody even read it, so I'm posting it here again, hoping to get 
> some replies and not to upset anybody.
>  
> My main interest is in how string matching (the "content"/"uri-content" rule 
> options,  basically) is exactly performed. I've taken a look at various 
> documents on the Net, but most are too vague or talk about earlier (and 
> slower) Snort versions.
> 
> I know that a multiple-string pattern matching algorithm is used (a 
> modification of Aho Corasick, right? Or is it Wu-Manber? I know that 
> different algorithms can also be chosen at runtime). I also know that rules 
> are being grouped by protocol first, and by some protocol parameters then (i. 
> e. ports for tcp/udp, type for icmp), so that for every packet only a 
>  subset of the rules is tested, while the other rules cannot surely be 
> matched. I have also understood how the grouping treats "any" rules, so this 
> is clear enough for me.

A heavily modified version of Aho Corasick is the default pattern
matcher.  Wu-Manber has been deprecated.

Rules are grouped essentially by port.

> 
> The basic Snort operation is as follows, as far as I can understand: 

> 1. A packet arrives. Let's assume it contains IP+TCP

bpf goes right here.

Stream4/5 and frag3 go here.

> 2. The patterns belonging to the proper subset of the TCP rules are tested 
> against the packet (all at a single time). 
> 3. For every matched pattern the rule it belongs to is identified. This set of 
> rules contains all possible match candidates, while all other rules are 
> surely not matched at this point.

Rules without a content match and contain the correct port grouping are
evaluated here.

> 4. For every candidate rule the rest of the options are tested, to see whether 
> the rule is fully matched or not.
> 5. Fully matched rules are added to the event queue.
> 6. The event queue is processed.
>  
> So, am I right this far? I'd like to get all the possible corrections and/or 
> details! 

Pretty close, and what I've added above isn't 100% correct, but it adds
a few additional important steps.

>  
> Another thing that is quite unclear to me is how rules with multiple "content" 
> options are dealt with. The code seems to suggest that only the longest 
> pattern is added to the algorithm. So the remaining patterns are tested one 
> at a time at point 4? But this would not benefit from a multiple-pattern 
> string matching algorithm, so I'm not sure.

Longest content per rule is selected for loading Aho, additional
contents are evaluated in order in the OTN tree's

> 
> OK, it's all for now. Thanks for the time to read and answer my questions! 
> There will surely be more ;). 

Hopefully that helps a bit.

Cheers,
-matt




More information about the Snort-users mailing list