[RFC] jbd2: use GFP_NOFS for blkdev_issue_flush

flush request is issued in transaction commit code path, so looks using
GFP_KERNEL to allocate memory for flush request bio falls into the classic
deadlock issue.  I saw btrfs and dm get it right, but ext4, xfs and md are
using GFP_KERNEL, which makes me confused. Hoping filesystem developers
clarify.

Signed-off-by: Shaohua Li <shli@fusionio.com>
---
  fs/jbd2/commit.c |    4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)

--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

On Fri, Apr 13, 2012 at 10:27:35AM +0800, Shaohua Li wrote:
> 
> flush request is issued in transaction commit code path, so looks using
> GFP_KERNEL to allocate memory for flush request bio falls into the classic
> deadlock issue.  I saw btrfs and dm get it right, but ext4, xfs and md are
> using GFP_KERNEL, which makes me confused. Hoping filesystem developers
> clarify.

Only one of the 5 calls XFS makes to blkdev_issue_flush() is
questionable. Probably should use GFP_NOFS by default, though, so we
don't trip over this problem...

Cheers,

Dave.

On Fri 13-04-12 10:27:35, Shaohua Li wrote:
> 
> flush request is issued in transaction commit code path, so looks using
> GFP_KERNEL to allocate memory for flush request bio falls into the classic
> deadlock issue.  I saw btrfs and dm get it right, but ext4, xfs and md are
> using GFP_KERNEL, which makes me confused. Hoping filesystem developers
> clarify.
> 
> Signed-off-by: Shaohua Li <shli@fusionio.com>
  Yeah, ext4 (jbd2) should use GFP_NOFS since we are in a kjournald and
reclaim could possibly need to start a transaction thus blocking on
kjournald. So you have my:
  Reviewed-by: Jan Kara <jack@suse.cz>

								Honza
> ---
>  fs/jbd2/commit.c |    4 ++--
>  1 file changed, 2 insertions(+), 2 deletions(-)
> 
> Index: linux/fs/jbd2/commit.c
> ===================================================================
> --- linux.orig/fs/jbd2/commit.c	2012-04-13 09:57:51.785506925 +0800
> +++ linux/fs/jbd2/commit.c	2012-04-13 09:58:29.585498773 +0800
> @@ -723,7 +723,7 @@ start_journal_io:
>  	if (commit_transaction->t_need_data_flush &&
>  	    (journal->j_fs_dev != journal->j_dev) &&
>  	    (journal->j_flags & JBD2_BARRIER))
> -		blkdev_issue_flush(journal->j_fs_dev, GFP_KERNEL, NULL);
> +		blkdev_issue_flush(journal->j_fs_dev, GFP_NOFS, NULL);
> 
>  	/* Done it all: now write the commit record asynchronously. */
>  	if (JBD2_HAS_INCOMPAT_FEATURE(journal,
> @@ -859,7 +859,7 @@ wait_for_iobuf:
>  	if (JBD2_HAS_INCOMPAT_FEATURE(journal,
>  				      JBD2_FEATURE_INCOMPAT_ASYNC_COMMIT) &&
>  	    journal->j_flags & JBD2_BARRIER) {
> -		blkdev_issue_flush(journal->j_dev, GFP_KERNEL, NULL);
> +		blkdev_issue_flush(journal->j_dev, GFP_NOFS, NULL);
>  	}
> 
>  	if (err)
> --
> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

On Fri, Apr 13, 2012 at 10:27:35AM +0800, Shaohua Li wrote:
> 
> flush request is issued in transaction commit code path, so looks using
> GFP_KERNEL to allocate memory for flush request bio falls into the classic
> deadlock issue.  I saw btrfs and dm get it right, but ext4, xfs and md are
> using GFP_KERNEL, which makes me confused. Hoping filesystem developers
> clarify.
> 
> Signed-off-by: Shaohua Li <shli@fusionio.com>

Thanks, applied.

					- Ted
--
To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html