Subject: Re: panic: kmem_malloc: mb_map too small?
To: Chris G Demetriou <Chris_G_Demetriou@ux2.sp.cs.cmu.edu>
From: Pete Bentley <email@example.com>
Date: 06/06/1996 15:41:10
At Wed, 05 Jun 1996 10:36:03 EDT, Chris G Demetriou writes:
>> Is there anybody out there currently suffering from the
>> kmem_malloc: mb_map too small
>> problem who can test a fix for me.
>> People who've increased NMBCLUSTERS to avoid that panic, and are
>> willing to decrease it for a while to test my changes are also
>> encouraged to reply... 8-)
This reminds me, the machines we had to increase NMBCLUSTERS on
because of mb_map overflowing were things like web proxies which
suffer from mbufs tied up by sockets in CLOSE_WAIT state for clients
which have gone away (as described in Stevens Vol 3, pp196--200). This
problem is much alleviated by incorporating the 4.4BSD-lite 2 patches
which ensure persist timers are running for such sockets, as per my PR
kern/2335 (still marked as open), so eventually the mbufs get released.
I might be able to test your patch here if I take that patch out :)