i386/103664: kmem_map_too_small panic after about 7d uptime on 6.1-release

[Igor Soumenkov]

The following reply was made to PR i386/103664; it has been noted by GNATS.

From: "Igor Soumenkov" <2igosha at gmail.com>
To: "Ed Maste" <emaste at phaedrus.sandvine.ca>
Cc: freebsd-gnats-submit at freebsd.org
Subject: Re: i386/103664: kmem_map_too_small panic after about 7d uptime on 6.1-release
Date: Wed, 27 Sep 2006 07:42:46 +0400

 I found out that it was ipfw that caused the memory leak and lead to the panic.
 And I found the fix in version 1.141 of file ip_fw2.c
 
 3944,3945c3944
 < rule = layer3_chain.reap;
 < layer3_chain.reap = NULL;
 ---
 > rule = layer3_chain.reap, layer3_chain.reap = NULL;
 3947c3946
 < if (rule != NULL)
 ---
 > if (layer3_chain.reap != NULL)
 
 Seems that there are no leaks now.
 
 +Warning: memory type IpFw/IpAcct leaked memory on destroy (108606
 allocations, 6993216 bytes leaked).
 
 On 9/27/06, Ed Maste <emaste at phaedrus.sandvine.ca> wrote:
 > > I am running 6.1-RELEASE on a two-xeon machine with 2GB ram, acpi and HT disabled. The system is running the SMP kernel and each week it panics with "kmem_map too small" error.
 >
 > You can have a look at vmstat -m output over time, to see if one
 > particular allocation type is constantly growing.  If that's the
 > case it might suggest the location of a leak.  You can also run
 > vmstat -m on a core file to examine the malloc statistics.
 >
 
 
 -- 
 Igor Soumenkov.