Skip site navigation (1)Skip section navigation (2)
Date:      Mon, 20 Aug 2007 18:59:42 +0700
From:      "Vadim Goncharov" <vadimnuclight@tpu.ru>
To:        "Eduardo Meyer" <dudu.meyer@gmail.com>
Cc:        ipfw@freebsd.org
Subject:   Re: All I have is one packet!
Message-ID:  <optxcwxs1c4fjv08@nuclight.avtf.net>
In-Reply-To: <d3ea75b30708190751i84ef2d2jeda35ff6d1a11c8c@mail.gmail.com>
References:  <d3ea75b30708060905p25019480i90fd1d71dc9120a2@mail.gmail.com> <optwri2uts4fjv08@nuclight.avtf.net> <d3ea75b30708190751i84ef2d2jeda35ff6d1a11c8c@mail.gmail.com>

Next in thread | Previous in thread | Raw E-Mail | Index | Archive | Help
19.08.07 @ 21:51 Eduardo Meyer wrote:

>> Yes, dynamic rules in ipfw are not intended for supporting state created
>> in the middle of the session, wuth the default sysctl settings it will  
>> be
>> kept for 1 second (which, however, is enough for shaping of fast
>> transfers). I think, precise controlling of dynamic rules from both
>> userland and kernel should be added to ipfw, to modify existing rules on
>> the fly (or even more features, like pfsync). As a hackish dirty
>> workaround, may be it should be only one keyword, something like
>> "keep-state-middle", to create normal dynamic rule without initial SYNs.
>>
>> But you've said about even more complex behaviour, like init on TCP,
>> continue with UDP. That's difficult to implement in kernel, and may be
>> even not suitable for ipfw. Currently (I think), you can try to emulate
>> this behaviour by divert'ing tagged by ng_tag packet to userland  
>> program,
>> like snort_inline (from ports collection) with needed scripting, which
>> will trigger adding proper rules to firewall (you should also care about
>> expiring that connection on FINs and RSTs, though).
>
> That's exactly the point. However, from a simplistic and probably
> ignorant point of view on this matter, like mine, I believed it to be
> in fact a much more simple "state", which would only compare IP
> addresses (src<->dst) for the match, so I could just
>
> ipfw add X allow { tcp or udp } from any to any keep-iponly-state tagged  
> Y
>
> It would be helpfull with many protocols which in fact use a transport
> proto (like TCP) to do actual session initialization while using
> another transport proto (UDP, DDP, whatever) for the real traffic;
> many things do this nowadays;
>
> Would such a feature be possible?

Yes, in theory. But I'm not sure that such a patch will be merged into  
official tree. Also, one can think about link negotiation with another IP  
(src<->dst) pair, not this hosts: imagine direct FTP transfer between two  
servers or P2P application where clients negotiate connection parameters  
via server, and then only actual data in undetectable connection flows  
between them. So in general case you again need specialized protocol  
analyzer. For example, in your case with only IPs - can you say when  
dynamic rule will expire?

-- 
WBR, Vadim Goncharov



Want to link to this message? Use this URL: <http://docs.FreeBSD.org/cgi/mid.cgi?optxcwxs1c4fjv08>