Jim, many thanks for your pointers. I think I left out too
much of the context earlier in this thread.
The problem is communicating with a mailserver behind a
Symantec Raptor Firewall. This firewall has a syn-flood
protection which works like this:

your server sends a SYN
the firewall answers with ACK (without SYN). The ack number
  is that from your SYN + 1000000
your server _should_ send a RST
The firewall takes this RST as a sign that the SYN sent
  earlier is no syn flood and responds with a SYN ACK to
  your original SYN, thus establishing the connection

The problem now is getting ipfilter to send this RST. This
yields the problems described in my earlier posting. The
term 'broken server' is used here for servers behind a
Raptor Firewall.

At 11:34 AM 6/10/2003 -0400, Jim Sandoz wrote:

Sensille wrote:
The problem with that is when a delayed ack is suddenly popping up,
your connection is dropped. If you do this, only enable it for the
host that has this anti-syn-flooding behaviour.
It would be great to find a generic ruleset to allow communication
with these 'broken' servers. Adding all these servers with one rule
per server is impractical, because I see several dozens of them.
On the other hand, risking to reset a good connection is even worse.

the problem is not "broken servers". any time a packet is dropped enroute, packets arrive out of sequence at the receiver. you can't control this. the generic ruleset you need is to reply with reset *only* to SYN packets. otherwise you're just going to be killing off valid connections via the return-rst, as you have found out.

read the problem description and example here:
http://marc.theaimsgroup.com/?l=ipfilter&m=97234715121908&w=2

and check the FAQ, here:
http://www.phildev.net/ipf/IPFprob.html#9

jim




Reply via email to