[PATCH] xfs: clarify lock ordering comment

[PATCH] xfs: clarify lock ordering comment

[Ross Zwisler]

Replace "i_mmap_lock" with "mmap_lock" in the lock ordering comment above
xfs_filemap_page_mkwrite().  The lock in question is actually the
XFS_MMAPLOCK_SHARED rw_semaphore (no leading "i"), and this comment is
easily confused with the "i_mmap_lock_[read|write]" functions that operate
on struct address_space->i_mmap_rwsem.  This clarification is especially
important because address_space->i_mmap_rwsem is taken down in the DAX
code as part of this fault path.

Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
---
 fs/xfs/xfs_file.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
index f429662..b190033 100644
--- a/fs/xfs/xfs_file.c
+++ b/fs/xfs/xfs_file.c
@@ -1477,7 +1477,7 @@ xfs_file_llseek(
  *
  * mmap_sem (MM)
  *   sb_start_pagefault(vfs, freeze)
- *     i_mmap_lock (XFS - truncate serialisation)
+ *     mmap_lock (XFS - truncate serialisation)
  *       page_lock (MM)
  *         i_lock (XFS - extent map serialisation)
  */
-- 
2.1.0

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: [PATCH] xfs: clarify lock ordering comment

[Dave Chinner]

On Thu, Oct 08, 2015 at 03:58:01PM -0600, Ross Zwisler wrote:
> Replace "i_mmap_lock" with "mmap_lock" in the lock ordering comment above
> xfs_filemap_page_mkwrite().  The lock in question is actually the
> XFS_MMAPLOCK_SHARED rw_semaphore (no leading "i"), and this comment is

struct xfs_inode {
....
	        mrlock_t                i_mmaplock;     /* inode mmap IO lock */
....

> easily confused with the "i_mmap_lock_[read|write]" functions that operate
> on struct address_space->i_mmap_rwsem.  This clarification is especially
> important because address_space->i_mmap_rwsem is taken down in the DAX
> code as part of this fault path.
> 
> Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
> ---
>  fs/xfs/xfs_file.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
> index f429662..b190033 100644
> --- a/fs/xfs/xfs_file.c
> +++ b/fs/xfs/xfs_file.c
> @@ -1477,7 +1477,7 @@ xfs_file_llseek(
>   *
>   * mmap_sem (MM)
>   *   sb_start_pagefault(vfs, freeze)
> - *     i_mmap_lock (XFS - truncate serialisation)
> + *     mmap_lock (XFS - truncate serialisation)

As per above, the XFS lock is "i_mmaplock"...

The lock names are annotated with the subsystem the lock belongs to
to avoid this confusion. Along with the lock ordering (inside
sb_start_pagefault) this should indicate that it's not the
"i_mmap_lock (MM - vma serialisation)" lock... ;)

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs

Re: [PATCH] xfs: clarify lock ordering comment

[Ross Zwisler]

On Fri, Oct 09, 2015 at 09:24:50AM +1100, Dave Chinner wrote:
> On Thu, Oct 08, 2015 at 03:58:01PM -0600, Ross Zwisler wrote:
> > Replace "i_mmap_lock" with "mmap_lock" in the lock ordering comment above
> > xfs_filemap_page_mkwrite().  The lock in question is actually the
> > XFS_MMAPLOCK_SHARED rw_semaphore (no leading "i"), and this comment is
> 
> struct xfs_inode {
> ....
> 	        mrlock_t                i_mmaplock;     /* inode mmap IO lock */
> ....
> 
> > easily confused with the "i_mmap_lock_[read|write]" functions that operate
> > on struct address_space->i_mmap_rwsem.  This clarification is especially
> > important because address_space->i_mmap_rwsem is taken down in the DAX
> > code as part of this fault path.
> > 
> > Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
> > ---
> >  fs/xfs/xfs_file.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/fs/xfs/xfs_file.c b/fs/xfs/xfs_file.c
> > index f429662..b190033 100644
> > --- a/fs/xfs/xfs_file.c
> > +++ b/fs/xfs/xfs_file.c
> > @@ -1477,7 +1477,7 @@ xfs_file_llseek(
> >   *
> >   * mmap_sem (MM)
> >   *   sb_start_pagefault(vfs, freeze)
> > - *     i_mmap_lock (XFS - truncate serialisation)
> > + *     mmap_lock (XFS - truncate serialisation)
> 
> As per above, the XFS lock is "i_mmaplock"...
> 
> The lock names are annotated with the subsystem the lock belongs to
> to avoid this confusion. Along with the lock ordering (inside
> sb_start_pagefault) this should indicate that it's not the
> "i_mmap_lock (MM - vma serialisation)" lock... ;)

Ah, that makes sense, thanks. :)

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs