Re: zfs: Fatal trap 12: page fault while in kernel mode

on 27/07/2009 22:58 O. Hartmann said the following:
> Juergen Unger wrote:
[snip]
>>> _sx_xlock(3c,0,874aa28d,70f,8ae9a9f8,...) at _sx_xlock+0x43
>>> dmu_buf_update_user(0,8ae9a9f8,0,0,0,...) at dmu_buf_update_user+0x35
>>> zfs_znode_dmu_fini(8ae9a9f8,874b312d,1114,110b,879ab000,...) at zfs_znode_dmu_f3
>>> zfs_freebsd_reclaim(fcd29c3c,1,0,8ec63754,fcd29c60,...) at zfs_freebsd_reclaim+0
>>> VOP_RECLAIM_APV(874b65a0,fcd29c3c,0,0,8ec637c8,...) at VOP_RECLAIM_APV+0xa5
>>> vgonel(8ec637c8,0,80c77037,386,0,...) at vgonel+0x1a4
>>> vnlru_free(80f2a0f0,0,80c77037,300,3e8,...) at vnlru_free+0x2d5
>>> vnlru_proc(0,fcd29d38,80c652bc,33e,871932a8,...) at vnlru_proc+0x80
>>> fork_exit(8090d960,0,fcd29d38) at fork_exit+0xb8
>>> fork_trampoline() at fork_trampoline+0x8
[snip]
> 
> I see a similar problem on two SMP boxes (is your SMP?), but in my case,
> it seems not to be ZFS related although I also use ZFS as /home filesystem

In this case this does seem to be caused by ZFS.
>From the backtrace we see that _sx_xlock() is called on bogus struct sx pointer
(0x3c) and this is caused by dmu_buf_update_user() called with NULL first argument
(dmu_buf_t). Which means that znode_t z_dbuf was NULL - this could have been
caught by ASSERT in zfs_znode_dmu_fini if it were enabled.

If you have the crash dump, then it would be interesting to examine znode_t
structure ('zp' argument) in zfs_znode_dmu_fini.

P.S. I see that zfs_inactive checks for z_dbuf being NULL and there is the
following comment:
/*
 * The fs has been unmounted, or we did a
 * suspend/resume and this file no longer exists.
 */
Maybe zfs_freebsd_reclaim should do the same?

P.P.S. I am not a VFS or ZFS expert.

-- 
Andriy Gapon