Zombie NFS writing from FreeBSD clients to FreeBSD 8.0 server with ZFS

[Rick Macklem]

On Tue, 2 Feb 2010, alan bryan wrote:

> --- On Wed, 1/6/10, Rick Macklem <rmacklem at uoguelph.ca> wrote:

> From: Rick Macklem <rmacklem at uoguelph.ca>
> Subject: Re: Zombie NFS writing from FreeBSD clients to FreeBSD 8.0 server with ZFS
> To: "alan bryan" <alan.bryan at yahoo.com>
> Cc: freebsd-stable at freebsd.org
> Date: Wednesday, January 6, 2010, 12:19 PM
> 
> 
> On Wed, 6 Jan 2010, alan bryan wrote:
> 
> > I have a AMD64 FreeBSD 8.0 server with ZFS filesystem
> being shared via NFS.  These are being accessed by the
> clients.  The clients are a mix of FreeBSD 6.2 32bit
> and FreeBSD 7.0 64bit.  I have seen similar behavior
> from both versions of FreeBSD as clients.
> >
> [stuff related to lotsa writes being done snipped]
> >
> > Any ideas on what to try, where to look for more
> insight, etc...??
> >
> - Are they the same write being retried over and over?


yes - it appears so


> - Is the server replying to them with an error?

yes

> 
> The above might give some insight into what's happening,

Didn't help much. The server is replying with NFS3ERR_IO (EIO) for
some reason. Unfortunately EIO is the catch all for just about
everything. (There is only one case in the NFS server code that
generates EIO and that would be caused by a bogus data mbuf list.)
As such, I suspect the EIO is coming from ZFS for some reason, but
can't be sure.

Btw, a binary capture using tcpdump can be read into wireshark, which
is much better at decoding NFS etc than tcpdump.

rick