[linux-cifs-client] Fwd: [PATCH 0/4] cifs: fix oopses and mem corruption with concurrent mount/umount (try #4)

[Steve French]

---------- Forwarded message ----------
From: Steve French <smfrench at gmail.com>
Date: Thu, Oct 30, 2008 at 12:25 PM
Subject: Re: [PATCH 0/4] cifs: fix oopses and mem corruption with concurrent
mount/umount (try #4)
To: Jeff Layton <jlayton at redhat.com>
Cc: linux-fsdevel <linux-fsdevel at vger.kernel.org>, LKML <
linux-kernel at vger.kernel.org>


This is much better since it doesn't regress the tcon sharing (which we
still need to extend for the shared superblock case) - I particularly like
the new routines to put (free) the tcon and put the smb session, but think
the locking gets more complicated (with little performance gain) moving from
one global spinlock covering tcon/smb-session/cifs-tcp-socket to one
embedded within each structure, and also could be confusing since the cifs
tcp socket is already protected by a semaphore and now would have a spinlock
too.   I think we greatly increase the chance of deadlock having to nest
spinlocks.




On Thu, Oct 30, 2008 at 10:16 AM, Jeff Layton <jlayton at redhat.com> wrote:

> This patchset is intended to fix the oopses, memory corruption and mount
> failures when using the reproducer detailed here:
>
> https://bugzilla.samba.org/show_bug.cgi?id=5720
>
> This is the fourth attempt at this. Since the third attempt, Steve
> French has committed the patch to handle the server->tsk pointer more
> atomically so that patch is ommitted here. While that patch helps ensure
> that the shutdown of the demux thread happens cleanly, there are still
> other races.
>
> Andrew Morton has also taken the patch to clean up the server protocol
> handling for -mm. I've included this patch in the set since Steve has
> not yet taken it, and I've had to modify it slightly for changes that
> have gone into Steve's tree.
>
> This patchset is based on Steve French's cifs-2.6 git tree and should
> apply cleanly to its current state.
>
> The main differences in this patchset are that it fixes some bugs in
> list handling in the earlier patchsets. It also reenables the sharing
> of tree connects. With this, any structures that were previously
> shared should remain so (no loss of functionality).
>
> There's still some remaining cleanup work that can be done here. The
> cifs_mount code could stand to be broken up into smaller functions.
> cifs_debug_data_proc_show could also stand to be reorganized to better
> reflect the heirarchy of server->session->tcon. Those changes are
> probably more suitable in follow-on patches. I'd like to know whether
> these are acceptible before I spend time working on them.
>
> I've been able to run the reproducer in the above BZ overnight on this
> patchset. Without it, it usually crashes within a few minutes.
>
> Jeff Layton (4):
>  cifs: clean up server protocol handling for TCP_Server_Info
>  cifs: disable sharing session and tcon and add new TCP sharing code
>  cifs: reinstate sharing of SMB sessions
>  cifs: reinstate sharing of tree connections
>
>  fs/cifs/cifs_debug.c |  286 +++++++++++++++++++---------------
>  fs/cifs/cifsfs.c     |   33 +++--
>  fs/cifs/cifsglob.h   |   28 ++--
>  fs/cifs/cifssmb.c    |   54 +------
>  fs/cifs/connect.c    |  426
> +++++++++++++++++++++++++-------------------------
>  fs/cifs/misc.c       |   93 +++++------
>  6 files changed, 456 insertions(+), 464 deletions(-)
>
>


-- 
Thanks,

Steve



-- 
Thanks,

Steve
-------------- next part --------------
HTML attachment scrubbed and removed