NetFilter
[Top] [All Lists]

NOTRACK not working as expected

To: <netfilter@lists.netfilter.org>
Subject: NOTRACK not working as expected
From: "Wilson, Richard E" <richard.wilson@eds.com>
Date: Mon, 27 Nov 2006 15:06:34 -0600
Delivered-to: sp-com-lists@consult.net
Delivered-to: netfilter-list1@securepoint.com
List-archive: </pipermail/netfilter>
List-help: <mailto:netfilter-request@lists.netfilter.org?subject=help>
List-id: General discussion and user questions <netfilter.lists.netfilter.org>
List-post: <mailto:netfilter@lists.netfilter.org>
List-subscribe: <https://lists.netfilter.org/mailman/listinfo/netfilter>, <mailto:netfilter-request@lists.netfilter.org?subject=subscribe>
List-unsubscribe: <https://lists.netfilter.org/mailman/listinfo/netfilter>, <mailto:netfilter-request@lists.netfilter.org?subject=unsubscribe>
Sender: netfilter-bounces@lists.netfilter.org
Thread-index: AccSZ+nW8AEBr51nRnO5J/SNJ/U+uw==
Thread-topic: NOTRACK not working as expected
All,

Running RH EL 4 Kernel 2.6.9-42.0.3.ELsmp, iptables v1.2.11

I added the following rule to prevent the ip_conntrack table from
tracking localhost connections:

iptables -t raw -A PREROUTING -i lo -j NOTRACK

The monitor script in place to allow us to reboot this system gracefully
when the ip_conntrack table runs out of space reports when it reaches
90% of capacity (determined by ip_conntrack | wc -l / ip_conntrack_max).
I now have a couple copies of the ip_conntrack table a short time before
it ran out of space and I find a LOT (5k+) of connections with
"src=127.0.0.1 dst=127.0.0.0" in the table.  Is there a better way to
create a rule to not track localhost connections?

This server has 4GB of RAM and uses the default value of 65536 for the
ip_conntrack table size.  I have already changed the
ip_conntrack_tcp_timeout_established value from its default of 432000 (5
days) to 172800 (2 days) and it still ran out of table slots.  This
change was made earlier today, a couple hours before the table filled
up.  The localhost traffic cited above is mostly udp traffic (complex
DNS stuff going on).  The server handles mail to/from the Internet, so
has a lot of short term connections.

>>From what I've read, increasing the size of the ip_conntrack table is
best done by powers of 2 -- is this still the case?  I calculated that
the current table size is roughly 40MB of non-swappable RAM, so I have
room to increase it if all else fails.

Thanks in advance, 

Richard Wilson
EDS
richard dot wilson at eds dot com




<Prev in Thread] Current Thread [Next in Thread>