It looks to me like you're only allowing *new* connections to port 80.  Which 
means if the state in conntrack times out, the conntrack entry will be removed, 
and a once existing connection now will be denied.

Check your timeouts in /proc/sys/net/ipv4/netfilter, in the 
ip_conntrack_tcp_timeout_* files.

Joe Gooch
K12 Systems, Inc.


> -----Original Message-----
> From: David Clark [mailto:[email protected]]
> Sent: Friday, November 13, 2009 1:23 PM
> To: [email protected]
> Subject: [Pound Mailing List] Intermittent iptables filter
> errors from pound requests
>
> Hi,
>
> Recently I've noticed some issues in the pound logs where a connection
> to the back-end server timed out, e.g.:
>
> Nov  4 03:13:11 balance1 pound: backend 166.70.134.194:80 connect:
> Connection timed out
> Nov  4 03:13:16 balance1 pound: BackEnd 166.70.134.194:80 resurrect
>
> After looking into it a bit, it seems that once in a while (say ~200
> times per day out of 2,000,000+ requests) iptables on the back-end
> server blocks a packet from the pound server. I say this because I can
> see entries like the following in the logs on the back-end
> web servers:
>
> Nov  4 03:10:11 wwwut3 [1280662.309643] RULE 5 -- DENY IN=eth0 OUT> 
> SRC6.70.134.196 DST6.70.134.194 LEN` TOS=0x00 PREC=0x00 TTLd
> IDe011 DF PROTO=TCP SPT3942 DPT€ WINDOWX40 RES=0x00 SYN URGP=0
>
> Nov  4 03:10:14 wwwut3 [1280665.307411] RULE 5 -- DENY IN=eth0 OUT> 
> SRC6.70.134.196 DST6.70.134.194 LEN` TOS=0x00 PREC=0x00 TTLd
> IDe012 DF PROTO=TCP SPT3942 DPT€ WINDOWX40 RES=0x00 SYN URGP=0
>
> Nov  4 03:10:20 wwwut3 [1280671.307415] RULE 5 -- DENY IN=eth0 OUT> 
> SRC6.70.134.196 DST6.70.134.194 LEN` TOS=0x00 PREC=0x00 TTLd
> IDe013 DF PROTO=TCP SPT3942 DPT€ WINDOWX40 RES=0x00 SYN URGP=0
>
> Nov  4 03:10:32 wwwut3 [1280683.307406] RULE 5 -- DENY IN=eth0 OUT> 
> SRC6.70.134.196 DST6.70.134.194 LEN` TOS=0x00 PREC=0x00 TTLd
> IDe014 DF PROTO=TCP SPT3942 DPT€ WINDOWX40 RES=0x00 SYN URGP=0
>
> Nov  4 03:10:56 wwwut3 [1280707.307406] RULE 5 -- DENY IN=eth0 OUT> 
> SRC6.70.134.196 DST6.70.134.194 LEN` TOS=0x00 PREC=0x00 TTLd
> IDe015 DF PROTO=TCP SPT3942 DPT€ WINDOWX40 RES=0x00 SYN URGP=0
>
> Nov  4 03:11:44 wwwut3 [1280755.307410] RULE 5 -- DENY IN=eth0 OUT> 
> SRC6.70.134.196 DST6.70.134.194 LEN` TOS=0x00 PREC=0x00 TTLd
> IDe016 DF PROTO=TCP SPT3942 DPT€ WINDOWX40 RES=0x00 SYN URGP=0
>
> which, to my reading, is blocking traffic that would normally
> be allowed
> by the firewall:
>
> iptables --list --numeric
> Chain INPUT (policy DROP)
> target     prot opt source               destination
> ACCEPT     all  --  0.0.0.0/0            0.0.0.0/0           state
> RELATED,ESTABLISHED
> ...
> Cid43519296.0  tcp  --  0.0.0.0/0        0.0.0.0/0
> tcp dpt:80
> state NEW
> ...
> RULE_5     all  --  0.0.0.0/0            0.0.0.0/0
> ...
> Chain Cid43519296.0 (1 references)
> target     prot opt source               destination
> ACCEPT     all  --  166.70.134.196       0.0.0.0/0
> ACCEPT     all  --  166.70.134.197       0.0.0.0/0
> ...
> Chain RULE_5 (3 references)
> target     prot opt source               destination
> LOG        all  --  0.0.0.0/0            0.0.0.0/0
> LOG flags 0
> level 6 prefix `RULE 5 -- DENY '
> DROP       all  --  0.0.0.0/0            0.0.0.0/0
>
> There's always a sequence like that, too - an initial packet gets
> blocked, followed by another 3 seconds later, then 6 seconds, etc.  I
> believe the sequence ends where it does because I have a 180 second
> timeout in pound to the back-end servers.
>
> In an effort to troubleshoot this situation, I switched to a different
> machine running pound, with no effect (both machines use a 2.6.24
> kernel).  I was previously using pound 2.3.2 but earlier this week
> upgraded to 2.4.5, again with no effect.  The back-end web servers are
> running kernel versions 2.6.29 to 2.6.30 and all seem affected more or
> less equally.  None of the servers involved are anywhere near capacity
> (CPU usage below 10%, memory usage below 33%).
>
> I would <love> to know what I could do to identify the reason behind
> these errors, and even better, eliminate them.  Any help or
> investigative steps to take would be <much> appreciated.
>
> -David Clark
>
> --
> To unsubscribe send an email with subject unsubscribe to
> [email protected].
> Please contact [email protected] for questions.
>
--
To unsubscribe send an email with subject unsubscribe to [email protected].
Please contact [email protected] for questions.

Reply via email to