Current-Users archive

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index][Old Index]

Re: troubles with ipnat TCP entries



2008/5/7 Erik Bertelsen <bertelsen.erik%gmail.com@localhost>:
> 2008/5/4 Manuel Bouyer <bouyer%antioche.eu.org@localhost>:
>> Hi,
>>  I upgraded my home router to yesterday's current, and since then I have
>>  troubles with ipnat: it seems to keep states for a lot of connections
>>  which have been closed by either the application or the server,
>>  while it closes TCP connection which are still active (e.g. an imaps
>>  session initiated from mutt). FWIW, this is on a sparc (so big-endian)
>>  machine.
>>
>>  Attached is the output of ipnat -lv on this box.
>>  Notice that there's a lot of TCP map to remote host port 80 which have been
>>  closed from the host or server side (a netstat on the nated host confirmes
>>  this). These have a long TTL.
>>  On the other hand, my connection to 132.227.86.2 port 993 (the first entry 
>> in
>>  the output below) has a ttl of only 465. This is the connection which is
>>  dropped by the NAT box quite fast, while mutt had the connection to the
>>  server still open.
>>
>>  Does anyone else have noticed this problem, or have an idea about it ?
>>
>>  --
>>  Manuel Bouyer <bouyer%antioche.eu.org@localhost>
>>      NetBSD: 26 ans d'experience feront toujours la difference
>>  --
>>
>
> Yes, for some time (at least a month) I have seen a similar behaviour:
> I have a i386-box running current functioning as my NAT gateway and
> after some time it collects a lot of NAT entries. As a user I
> typically observe this when I use gmail from a machine on my local LAN
> and the browser after a while cannot connect to the server.
>
> It took me some time to suspect my gateway machine and then reboot
> this machine would solve the problem for a while. Later I observed
> that when it is stuck it typically has 1500-2000+ entries in the NAT
> table and that running ipnat -F would clear it up without a reboot.
>
> - Erik
>

This seems to have been fixed even if I did not notice any commit message
targeted against this problem.
My gateway machine now runs a -current kernel 4.99.63 of 29 May and
without having to flush the NAT tables (current no. of entries < 30) it
seems to be able to keep running.


- Erik


Home | Main Index | Thread Index | Old Index