Re: [oops] in text matching

[Patrick McHardy <kaber@trash.net>]

Pablo Neira Ayuso wrote:
> Patrick McHardy wrote:
>> Pablo Neira Ayuso wrote:
>>>> Pablo? I would prefer to get this fixed before 2.6.24, unless
>>>> there's a better fix I'm going to include my patch to reject
>>>> pattern lengths of 0. Thanks.
>>> Sorry for the late reply. Why should we accept zero length patterns?
>>> Would you consider this patch? It keeps consistent the return value of
>>> all textsearch approaches when a zero length pattern is passed.
>> The other ones actually seem to handle it fine, and I think
>> it should actually behave similar to memcmp or strcmp, so I'd
>> prefer a patch to handle it properly by always matching.
> 
> Indeed, KMP supports zero length patterns but I don't know how :).
> kmp_find() accesses kmp->pattern[0] -which actually points 4 bytes out
> of the ts_config structure- to compare it with text[i], then if those 4
> bytes doesn't match text[i], then returns a matching at position 1.


Indeed, it doesn't seem to handle it properly too.

> I think that a zero length pattern is a unspecified entry that we should
> reject. Also, returning some error to tell the user that is passing a
> zero length pattern -something that is probably what he didn't really
> want as it happens to Andrei.
> 
> I can cook a patch for iptables to reject zero length pattern with a
> nice error message, thus avoiding the EINVAL that will probably go nuts
> users.


Well, in that case I'll take your last patch, its textsearch which
is broken, not iptables :)