Re: [PATCH] Short write in nfsd becomes a full write to the client

["J. Bruce Fields" <bfields@fieldses.org>]

On Fri, Jun 26, 2009 at 09:24:49PM -0700, Dale Stimson wrote:
> On Thu, 5 Mar 2009 20:16:14 -0500, David Shaw <dshaw-wh+mT2OhP0WF0gnf/s2wvA@public.gmane.org> wrote:
> > If a filesystem being written to via NFS returns a short write count
> > (as opposed to an error) to nfsd, nfsd treats that as a success for
> > the entire write, rather than the short count that actually succeeded.
> > 
> > For example, given a 8192 byte write, if the underlying filesystem
> > only writes 4096 bytes, nfsd will ack back to the nfs client that all
> > 8192 bytes were written.  The nfs client does have retry logic for
> > short writes, but this is never called as the client is told the
> > complete write succeeded.
> ...
> > Here is a patch to properly return the short write count to the
> > client.
> [patch elided]
> 
> I bring this to your attention so you may, if you choose, look into
> this further:
> 
> Problem synopsis:
> An old client (running RHL 9 with kernel "2.4.20-43.9.legacy")
> attempts to seek on a file mounted over nfs.  The operation fails
> with "Illegal seek" or "Input/Output error".  The server is running
> Fedora 11 kernel-PAE-2.6.29.5-191.fc11.i686, which includes the
> short write patch.  When this kernel is re-built without the short
> write patch, everything works as before.

Does that server kernel have some version of
a0d24b295aed7a9daf4ca36bd4784e4d40f82303 "nfsd: fix hung up of nfs
client while sync write data to nfs server" applied?

If not, would it be possible to get a network trace?  (Run "tcpdump -s0
-wTMP", then run the test case, then kill tcpdump and mail me a copy of
TMP.)

--b.

> 
> Detais are at
> https://bugzilla.redhat.com/show_bug.cgi?id=508174
> 
> Caveats: I am specifically referring to the patch file
> "linux-2.6-nfsd-report-short-writes.patch" as newly included in
> Fedora's file kernel-2.6.29.5-191.fc11.src.rpm .  I can't vouch
> that that patch file is identical to what was posted to this list
> or merged for 2.6.30-rc1.
> 
> (As an aside, in this case, the client was attempting a simple gcc
> compile and link.  The failing programs (invoked by fcc) were the
> assember ("as)" and "ld".)