Re: 2.6.31-rc8: CIFS with 5 seconds hiccups

[Jeff Layton <jlayton@redhat.com>]

On Mon, 14 Sep 2009 16:10:47 -0400 (EDT)
Christoph Lameter <cl@linux-foundation.org> wrote:

> On Thu, 10 Sep 2009, Jeff Layton wrote:
> 
> > In any case, I think we need to look closely at what's happening at
> > mount time. First, I'll need some other info:
> >
> > 1) output of "/sbin/mount.cifs -V" from both machines
> 
> The 32 bit machine
> 
> #/sbin/mount.cifs -V
> mount.cifs version: 1.5
> //chiprodfs2/company /mnt -ouser=clameter,domain=xxx
> mount -t cifs //chiprodfs2/company /mnt -ouser=clameter,domain=xxx
> 

Yow, that version of mount.cifs is really old. I wonder if it may be
passing bad mount options to the kernel? Might be interesting to strace
that. Something like:

# strace -f -s 256 -e mount mount -t cifs //chiprodfs2/company /mnt -ouser=clameter,domain=xxx

...it'll probably have a cleartext password in it so you might want to
doctor the options a bit before sending along if you do.

Alternately, you might just want to try a newer version of mount.cifs
and see whether that fixes this.

> 64 bit machine
> 
> $ /sbin/mount.cifs -V
> mount.cifs version: 1.12-3.4.0
> 
> mount -t cifs //chiprodfs2/company /mnt -ouser=clameter,domain=w2k
> 
> > 3) wire captures from mount attempts on both machines. Try to mount the
> > "clameter" dir on both boxes and do captures of each attempt. Maybe
> > this time use -s 0 with tcpdump so we get all of the traffic.
> 
> I cannot mount the clameter dir on the 32 bit box. Hangs. So I will mount
> /company.
> 

Actually, the trace of a hanging mount would probably be interesting.

Does the 32-bit capture that you sent represent a mount attempt that
hung? Or was it successful?

> > There may be crackable password hashes in the captures, so you may want
> > to send them to me privately and not cc the list.
> 
> Ok will follow.
> 

Thanks for the info, I had a look at the captures. They both look
fairly similar. The main difference is that the 32-bit box doesn't seem
to have sent any more calls after sending a QPathInfo call to the
server for the root inode of the mount.

What's the "devname" that you're giving to the mount command for the
"clameter" dir? If there's more than 1 path component after the
hostname, then the problem may be in the old version of mount.cifs.
Some of them had broken handling for path prefixes.

-- 
Jeff Layton <jlayton@redhat.com>