[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <Pine.LNX.4.64.0803061449220.23263@bizon.gios.gov.pl>
Date: Thu, 6 Mar 2008 14:51:11 +0100 (CET)
From: Krzysztof Oledzki <olel@....pl>
To: Denys Fedoryshchenko <denys@...p.net.lb>
cc: netdev@...r.kernel.org
Subject: Re: DoS by cat /proc/net/ip_conntrack ?
On Thu, 6 Mar 2008, Denys Fedoryshchenko wrote:
> Hi again
Hi,
> On loaded router
> net.netfilter.nf_conntrack_count = 415633
> passing about 100-150 Mbps
> network cards 3xe100, 1xe1000e
>
> i tried to issue command cat /proc/net/ip_conntrack |grep 'something'
>
> Router went dead for about 2 minutes, even i disconnect ssh session.
> Ping was looks like this:
> 64 bytes from dotfib (10.184.184.1): icmp_seq=15 ttl=61 time=4321 ms
> 64 bytes from dotfib (10.184.184.1): icmp_seq=50 ttl=61 time=398 ms
> 64 bytes from dotfib (10.184.184.1): icmp_seq=122 ttl=61 time=15.3 ms
> 64 bytes from dotfib (10.184.184.1): icmp_seq=142 ttl=61 time=4452 ms
> 64 bytes from dotfib (10.184.184.1): icmp_seq=180 ttl=61 time=850 ms
> (system recovered)
> 64 bytes from dotfib (10.184.184.1): icmp_seq=182 ttl=61 time=0.681 ms
> 64 bytes from dotfib (10.184.184.1): icmp_seq=183 ttl=61 time=0.936 ms
> 64 bytes from dotfib (10.184.184.1): icmp_seq=184 ttl=61 time=2.94 ms
>
> I dont think it is normal, and such command taking a lot of system resources
> and cause whole system to hang.
>
> Kernel 2.6.24.2
The answer is quite simple here: don't do this. Instead use "conntrack -L"
as netlink is much more effective and better designed.
Best regards,
Krzysztof Olędzki
Powered by blists - more mailing lists