[PATCH] avoid lock oder violation between xattr.tdb and g_lock.tdb

[Volker Lendecke]

On Wed, Jul 20, 2016 at 12:06:16PM -0500, Steve French wrote:
> Added Yogesh to cc: since he helped discover and triage it
> 
> On Wed, Jul 20, 2016 at 12:03 PM, Steve French <smfrench at gmail.com> wrote:
> > On Wed, Jul 20, 2016 at 12:51 AM, Volker Lendecke <vl at samba.org> wrote:
> >> On Tue, Jul 19, 2016 at 08:47:47PM -0500, Steve French wrote:
> >>> Running this series of Volker's patches on 4.4-test (adding the one earlier
> >>> patch needed for the missing #define SERVER_ID_BUF_LENGTH
> >>> ) we got a different oops (in get_share_mode_write_time due to corrupt lock
> >>> contents).
> >>>
> >>> Are there other coreq patches that would need to backport this to 4.4-test
> >>> (or is it safer to use Michael's original small patch ... or ...)?
> >>
> >> Can you reproduce this somehow and get me more info like a stack trace
> >> or the corrupted contents? And, your exact patchset? I will need to
> >> backport this to 4.4 and 4.3 for a customer too, so any information
> >> will be very interesting for me.
> >>
> >> Thanks,
> >>
> >> Volker
> >
> > Created large directory tree and did ls -lR (from cifs mount to Samba
> > 4.4-test branch equivalent to about two weeks ago + your patch
> > series).   (By the way Michael's small original patch did work without
> > problems in the same scenario as an alternative).

Just spent an hour trying to reproduce this. smbclient or mount.cifs
(debian jessie) did not trigger this problem.

I would *really* like to fix this. Can you send me a link to your
exact Samba source, and more detailed steps to reproduce, like your
smb.conf, your directory tree etc? Or give me access to the system
under test?

Thanks,

Volker