Re: dead umount - call trace

Peter Osterlund <[email protected]> Mon, 24 Jan 2005 21:24:48 +0100 (CET)
Newsgroups gmane.linux.suse.packet-writing
Message-ID <[email protected]>
On Mon, 24 Jan 2005, Attila Body wrote:

> It seems that the dead umount process keeps freezing on the same place
> according to its call trace:
>
> umount        D C03F93E0     0 15738  15706                     (NOTLB)
> de3bbd68 00000082 d5c53540 c03f93e0 00000040 00000000 cbbba954 00000012
>        0003640e a7439200 000f5fe6 d5c53540 d5c53690 de3ba000 d8c7ba4c
> 00000282
>        d5c53540 c02ea6e0 d8c7ba54 00000001 d5c53540 c01170c0 d8c7ba54
> d8c7ba54
> Call Trace:
>  [<c02ea6e0>] __down+0x90/0x110
>  [<c01170c0>] default_wake_function+0x0/0x20
>  [<c017c511>] __mark_inode_dirty+0xd1/0x1c0
>  [<c02ea8ab>] __down_failed+0x7/0xc
>  [<e0d1fda0>] .text.lock.balloc+0x8/0x128 [udf]
>  [<e0d25767>] udf_write_aext+0xf7/0x170 [udf]
>  [<e0d1fbac>] udf_free_blocks+0xcc/0x100 [udf]
>  [<e0d2cda4>] extent_trunc+0x124/0x180 [udf]
>  [<e0d2d025>] udf_discard_prealloc+0x225/0x2e0 [udf]
>  [<e0d21301>] udf_clear_inode+0x41/0x50 [udf]
>  [<c017285e>] clear_inode+0xde/0x120
>  [<c01728df>] dispose_list+0x3f/0xb0
>  [<c0172a92>] invalidate_inodes+0x62/0x90
>  [<c015e9ad>] generic_shutdown_super+0x5d/0x140
>  [<c015f69d>] kill_block_super+0x2d/0x50
>  [<c015e84d>] deactivate_super+0x6d/0x90
>  [<c0175e6f>] sys_umount+0x3f/0xa0
>  [<c014bc8d>] do_munmap+0x13d/0x180
>  [<c014bd14>] sys_munmap+0x44/0x70
>  [<c0175ee7>] sys_oldumount+0x17/0x20
>  [<c0103263>] syscall_call+0x7/0xb
>
> If my memory serves me well, I've seen something very similar on alt
> +sysreq+T in previous freezes Hope it helps finding out what the hell
> could happen...

This shows that the umount command is waiting for file system to flush
dirty data to the disc, but doesn't show if the drive is frozen or just
writing very slowly. Do you see any activity if you run:

	watch cat /proc/driver/pktcdvd/pktcdvd0

Writing lots of small files using packet writing is unfortunately very
slow. See

	http://lists.suse.com/archives/packet-writing/2004-Oct/0004.html

for more details.

-- 
Peter Osterlund - [email protected]
http://web.telia.com/~u89404340