Has anyone else seen hangs trying to umount ext3 volumes? I am seeing
this repeatedly after running tiobench on an ext3 volume. This was only
showing up in the mm tree, but as of 2.5.72-bk2 I am now seeing it in
the mainline.
Steve
Steven Pratt <[email protected]> wrote:
>
> Has anyone else seen hangs trying to umount ext3 volumes?
Nope.
> I am seeing
> this repeatedly after running tiobench on an ext3 volume. This was only
> showing up in the mm tree, but as of 2.5.72-bk2 I am now seeing it in
> the mainline.
It would have been nice to have heard about it before this...
Could you please debug it a bit? sysrq-T, etc?
Does the system hang, or does the umount hang?
Andrew Morton wrote:
>Steven Pratt <[email protected]> wrote:
>
>
>>Has anyone else seen hangs trying to umount ext3 volumes?
>>
>>
>
>Nope.
>
>> I am seeing
>>this repeatedly after running tiobench on an ext3 volume. This was only
>>showing up in the mm tree, but as of 2.5.72-bk2 I am now seeing it in
>>the mainline.
>>
>>
>
>It would have been nice to have heard about it before this...
>
>
Sorry, lots of thing on my plate.
>Could you please debug it a bit? sysrq-T, etc?
>
Yes, I will try to get this as soon as possible.
>Does the system hang, or does the umount hang?
>
Just the umount hangs, but it is severe enough to not be able to
shutdown the system. Have to power off.
Steve
>
>
Andrew Morton wrote:
>Steven Pratt <[email protected]> wrote:
>
>
>>Has anyone else seen hangs trying to umount ext3 volumes?
>>
>>
>Could you please debug it a bit? sysrq-T, etc?
>
Here is the trace of the hung process:
umount D 00000001 290213268 18747 18746 (NOTLB)
Call Trace:
[<c01a1ae8>] journal_kill_thread+0xa8/0xe0
[<c011c780>] default_wake_function+0x0/0x30
[<c016d8c5>] destroy_inode+0x35/0x60
[<c01a2b80>] journal_destroy+0x10/0x1c0
[<c019a97e>] ext3_xattr_put_super+0x1e/0x30
[<c0196209>] ext3_put_super+0x29/0x1a0
[<c015b64b>] generic_shutdown_super+0x16b/0x180
[<c015c03d>] kill_block_super+0x1d/0x50
[<c015b3fd>] deactivate_super+0x5d/0x90
[<c0170fef>] sys_umount+0x3f/0xa0
[<c0171067>] sys_oldumount+0x17/0x20
[<c010afaf>] syscall_call+0x7/0xb
Here is a link to the console log with full sysrq trace:
http://www-124.ibm.com/developerworks/opensource/linuxperf/logs/ext3umount
Steve
Steven Pratt <[email protected]> wrote:
>
> Here is the trace of the hung process:
>
> umount D 00000001 290213268 18747 18746 (NOTLB)
> Call Trace:
> [<c01a1ae8>] journal_kill_thread+0xa8/0xe0
whoops. I bet you're seeing this when using some script which does the
unmount.
Might this help?
diff -puN fs/jbd/journal.c~kjournald-shutdown-fix fs/jbd/journal.c
--- 25/fs/jbd/journal.c~kjournald-shutdown-fix 2003-06-19 12:58:43.000000000 -0700
+++ 25-akpm/fs/jbd/journal.c 2003-06-19 12:58:43.000000000 -0700
@@ -161,7 +161,7 @@ loop:
del_timer_sync(journal->j_commit_timer);
journal_commit_transaction(journal);
spin_lock(&journal->j_state_lock);
- goto loop;
+ goto end_loop;
}
wake_up(&journal->j_wait_done_commit);
@@ -210,7 +210,7 @@ loop:
journal->j_commit_request = transaction->t_tid;
jbd_debug(1, "woke because of timeout\n");
}
-
+end_loop:
if (!(journal->j_flags & JFS_UNMOUNT))
goto loop;
@@ -230,12 +230,16 @@ static void journal_start_thread(journal
static void journal_kill_thread(journal_t *journal)
{
+ spin_lock(&journal->j_state_lock);
journal->j_flags |= JFS_UNMOUNT;
while (journal->j_task) {
wake_up(&journal->j_wait_commit);
+ spin_unlock(&journal->j_state_lock);
wait_event(journal->j_wait_done_commit, journal->j_task == 0);
+ spin_lock(&journal->j_state_lock);
}
+ spin_unlock(&journal->j_state_lock);
}
/*
_
Andrew Morton wrote:
>Steven Pratt <[email protected]> wrote:
>
>
>>Here is the trace of the hung process:
>>
>> umount D 00000001 290213268 18747 18746 (NOTLB)
>> Call Trace:
>> [<c01a1ae8>] journal_kill_thread+0xa8/0xe0
>>
>>
>
>whoops. I bet you're seeing this when using some script which does the
>unmount.
>
>
Yes, I was. :-)
>Might this help?
>
>
Will try this tonight. Should have an answer tomorrow.
Thanks for the quick response.
Steve
Andrew Morton wrote:
>Steven Pratt <[email protected]> wrote:
>
>
>>Here is the trace of the hung process:
>>
>> umount D 00000001 290213268 18747 18746 (NOTLB)
>> Call Trace:
>> [<c01a1ae8>] journal_kill_thread+0xa8/0xe0
>>
>>
>
>whoops. I bet you're seeing this when using some script which does the
>unmount.
>
>Might this help?
>
>
Yes, this fixed the problem. Thanks.
Steve