[PATCH 00/37] xfs: current 3.4 patch queue

[Dave Chinner]

On Mon, Apr 30, 2012 at 09:34:34AM -0500, Ben Myers wrote:
> While testing this patchset without Jan's freeze work, I hit this assert.
> 
> This one rings bells for me, but I can't find where it's been reported.
> 
> -Ben
> 
> 
> [56571.411824] BUG: unable to handle kernel NULL pointer dereference at 0000000000000028
> [56571.412798] IP: [<ffffffffa00bf477>] xfs_log_space_wake+0x17/0x140 [xfs]

Which is failing on this:

(gdb) l *(xfs_log_space_wake+0x17)
0xffffffff81482cb7 is in xfs_log_space_wake (fs/xfs/xfs_log.c:844).
839             struct xfs_mount        *mp)
840     {
841             struct log              *log = mp->m_log;
842             int                     free_bytes;
843
844 >>>>>>>     if (XLOG_FORCED_SHUTDOWN(log))
845                     return;

On this offset:

$ pahole fs/xfs/xfs_log.o |grep -A 6 "^struct log"
struct log {
        struct xfs_mount *         l_mp;                 /*     0     8 */
        struct xfs_ail *           l_ailp;               /*     8     8 */
        struct xfs_cil *           l_cilp;               /*    16     8 */
        struct xfs_buf *           l_xbuf;               /*    24     8 */
        struct xfs_buftarg *       l_targ;               /*    32     8 */
        uint                       l_flags;              /*    40     4 */
                                                         ^^^^^^^^^^^^^^^^^

So the log pointer passed is NULL. That's the first definite sign
that the problems you are seeing are all a result of log buffer IO
completing after the log has been torn down.

Cheers,

Dave.
-- 
Dave Chinner
david at fromorbit.com