[PATCH] jbd2: Fix forever sleeping process in do_get_write_access()

[PATCH] jbd2: Fix forever sleeping process in do_get_write_access()

[Jan Kara]

In do_get_write_access() we wait on BH_Unshadow bit for buffer to get
from shadow state. The waking code in journal_commit_transaction() has
a bug because it does not issue a memory barrier after the buffer is moved
from the shadow state and before wake_up_bit() is called. Thus a waitqueue
check can happen before the buffer is actually moved from the shadow state
and waiting process may never be woken. Fix the problem by issuing proper
barrier.

Reported-by: Tao Ma <boyu.mt@taobao.com>
Signed-off-by: Jan Kara <jack@suse.cz>
---
 fs/jbd2/commit.c |    9 +++++++--
 1 files changed, 7 insertions(+), 2 deletions(-)

 Analogous JBD fix has been queued in my tree...

diff --git a/fs/jbd2/commit.c b/fs/jbd2/commit.c
index 2e5d370..3a958c7 100644
--- a/fs/jbd2/commit.c
+++ b/fs/jbd2/commit.c
@@ -768,8 +768,13 @@ wait_for_iobuf:
                    required. */
 		JBUFFER_TRACE(jh, "file as BJ_Forget");
 		jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
-		/* Wake up any transactions which were waiting for this
-		   IO to complete */
+		/*
+		 * Wake up any transactions which were waiting for this IO to
+		 * complete. The barrier must be here so that changes by
+		 * jbd2_journal_file_buffer() take effect before wake_up_bit()
+		 * does the waitqueue check.
+		 */
+		smp_mb();
 		wake_up_bit(&bh->b_state, BH_Unshadow);
 		JBUFFER_TRACE(jh, "brelse shadowed buffer");
 		__brelse(bh);
-- 
1.7.1

Re: [PATCH] jbd2: Fix forever sleeping process in do_get_write_access()

[Eric Sandeen]

On 5/5/11 7:10 AM, Jan Kara wrote:
> In do_get_write_access() we wait on BH_Unshadow bit for buffer to get
> from shadow state. The waking code in journal_commit_transaction() has
> a bug because it does not issue a memory barrier after the buffer is moved
> from the shadow state and before wake_up_bit() is called. Thus a waitqueue
> check can happen before the buffer is actually moved from the shadow state
> and waiting process may never be woken. Fix the problem by issuing proper
> barrier.

needed for jbd/commit.c as well, I guess?

-Eric

> Reported-by: Tao Ma <boyu.mt@taobao.com>
> Signed-off-by: Jan Kara <jack@suse.cz>
> ---
>  fs/jbd2/commit.c |    9 +++++++--
>  1 files changed, 7 insertions(+), 2 deletions(-)
> 
>  Analogous JBD fix has been queued in my tree...
> 
> diff --git a/fs/jbd2/commit.c b/fs/jbd2/commit.c
> index 2e5d370..3a958c7 100644
> --- a/fs/jbd2/commit.c
> +++ b/fs/jbd2/commit.c
> @@ -768,8 +768,13 @@ wait_for_iobuf:
>                     required. */
>  		JBUFFER_TRACE(jh, "file as BJ_Forget");
>  		jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
> -		/* Wake up any transactions which were waiting for this
> -		   IO to complete */
> +		/*
> +		 * Wake up any transactions which were waiting for this IO to
> +		 * complete. The barrier must be here so that changes by
> +		 * jbd2_journal_file_buffer() take effect before wake_up_bit()
> +		 * does the waitqueue check.
> +		 */
> +		smp_mb();
>  		wake_up_bit(&bh->b_state, BH_Unshadow);
>  		JBUFFER_TRACE(jh, "brelse shadowed buffer");
>  		__brelse(bh);

Re: [PATCH] jbd2: Fix forever sleeping process in do_get_write_access()

[Jan Kara]

On Thu 05-05-11 08:49:14, Eric Sandeen wrote:
> On 5/5/11 7:10 AM, Jan Kara wrote:
> > In do_get_write_access() we wait on BH_Unshadow bit for buffer to get
> > from shadow state. The waking code in journal_commit_transaction() has
> > a bug because it does not issue a memory barrier after the buffer is moved
> > from the shadow state and before wake_up_bit() is called. Thus a waitqueue
> > check can happen before the buffer is actually moved from the shadow state
> > and waiting process may never be woken. Fix the problem by issuing proper
> > barrier.
> 
> needed for jbd/commit.c as well, I guess?
  Yes, I was already queued in my tree. I just sent it to the list as well.

								Honza
> > diff --git a/fs/jbd2/commit.c b/fs/jbd2/commit.c
> > index 2e5d370..3a958c7 100644
> > --- a/fs/jbd2/commit.c
> > +++ b/fs/jbd2/commit.c
> > @@ -768,8 +768,13 @@ wait_for_iobuf:
> >                     required. */
> >  		JBUFFER_TRACE(jh, "file as BJ_Forget");
> >  		jbd2_journal_file_buffer(jh, commit_transaction, BJ_Forget);
> > -		/* Wake up any transactions which were waiting for this
> > -		   IO to complete */
> > +		/*
> > +		 * Wake up any transactions which were waiting for this IO to
> > +		 * complete. The barrier must be here so that changes by
> > +		 * jbd2_journal_file_buffer() take effect before wake_up_bit()
> > +		 * does the waitqueue check.
> > +		 */
> > +		smp_mb();
> >  		wake_up_bit(&bh->b_state, BH_Unshadow);
> >  		JBUFFER_TRACE(jh, "brelse shadowed buffer");
> >  		__brelse(bh);
> 
-- 
Jan Kara <jack@suse.cz>
SUSE Labs, CR

Re: [PATCH] jbd2: Fix forever sleeping process in do_get_write_access()

[Eric Sandeen]

On 5/5/11 9:11 AM, Jan Kara wrote:
> On Thu 05-05-11 08:49:14, Eric Sandeen wrote:
>> On 5/5/11 7:10 AM, Jan Kara wrote:
>>> In do_get_write_access() we wait on BH_Unshadow bit for buffer to get
>>> from shadow state. The waking code in journal_commit_transaction() has
>>> a bug because it does not issue a memory barrier after the buffer is moved
>>> from the shadow state and before wake_up_bit() is called. Thus a waitqueue
>>> check can happen before the buffer is actually moved from the shadow state
>>> and waiting process may never be woken. Fix the problem by issuing proper
>>> barrier.
>>
>> needed for jbd/commit.c as well, I guess?
>   Yes, I was already queued in my tree. I just sent it to the list as well.

sorry, sometimes "we" forget :)

Thanks!
-Eric

Re: [PATCH] jbd2: Fix forever sleeping process in do_get_write_access()

[Ted Ts'o]

On Thu, May 05, 2011 at 02:10:39PM +0200, Jan Kara wrote:
> In do_get_write_access() we wait on BH_Unshadow bit for buffer to get
> from shadow state. The waking code in journal_commit_transaction() has
> a bug because it does not issue a memory barrier after the buffer is moved
> from the shadow state and before wake_up_bit() is called. Thus a waitqueue
> check can happen before the buffer is actually moved from the shadow state
> and waiting process may never be woken. Fix the problem by issuing proper
> barrier.
> 
> Reported-by: Tao Ma <boyu.mt@taobao.com>
> Signed-off-by: Jan Kara <jack@suse.cz>

Thanks, I've added this to the ext4 tree.  (Currently in the dev
branch, pending testing.)

					- Ted