Subject: Re: nfe, kvakmem wait, and arp_drain: locked; punting
To: Bernhard Moellemann <firstname.lastname@example.org>
From: David Maxwell <email@example.com>
Date: 03/15/2006 13:43:59
> David Maxwell wrote:
> > As a datapoint, for me, doing large file transfers from a server using
> > the recently imported nfe(4) driver, leads to a hang.
> > [...]
> > I'd hazzard a guess that there's an mbuf leak in nfe(4), though I don't
> > have KMEMSTATS enabled in this kernel to check yet.
> Ten seconds before the hang, netstat -m reported:
> # 136 mbufs in use:
> # 128 mbufs allocated to data
> # 6 mbufs allocated to packet headers
> # 2 mbufs allocated to socket names and addresses
> # 0 calls to protocol drain routines
If you have (or can compile) a kernel with options KMEMSTATS, run
before the transfer, and during it a few times, and retain the results.
Something should stand out as consuming much more memory after the
nfe seeing significant traffic.
I can't do so right now (and each crash is a 3h parity check on the
main raid volume, if mounted, on the server in question).