[PATCH] xfs: improve xfs_isilocked

[PATCH] xfs: improve xfs_isilocked

[Christoph Hellwig]

Use rwsem_is_locked to make the assertations for shared locks work.

Signed-off-by: Christoph Hellwig <hch@lst.de>

Index: xfs/fs/xfs/xfs_iget.c
===================================================================
--- xfs.orig/fs/xfs/xfs_iget.c	2010-05-25 11:40:59.216005587 +0200
+++ xfs/fs/xfs/xfs_iget.c	2010-05-27 20:59:09.244004330 +0200
@@ -740,30 +738,24 @@ xfs_ilock_demote(
 }
 
 #ifdef DEBUG
-/*
- * Debug-only routine, without additional rw_semaphore APIs, we can
- * now only answer requests regarding whether we hold the lock for write
- * (reader state is outside our visibility, we only track writer state).
- *
- * Note: this means !xfs_isilocked would give false positives, so don't do that.
- */
 int
 xfs_isilocked(
 	xfs_inode_t		*ip,
 	uint			lock_flags)
 {
-	if ((lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) ==
-			XFS_ILOCK_EXCL) {
-		if (!ip->i_lock.mr_writer)
-			return 0;
+	if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
+		if (!(lock_flags & XFS_ILOCK_SHARED))
+			return !!ip->i_lock.mr_writer;
+		return rwsem_is_locked(&ip->i_lock.mr_lock);
 	}
 
-	if ((lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) ==
-			XFS_IOLOCK_EXCL) {
-		if (!ip->i_iolock.mr_writer)
-			return 0;
+	if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
+		if (!(lock_flags & XFS_IOLOCK_SHARED))
+			return !!ip->i_iolock.mr_writer;
+		return rwsem_is_locked(&ip->i_iolock.mr_lock);
 	}
 
-	return 1;
+	ASSERT(0);
+	return 0;
 }
 #endif

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

Re: [PATCH] xfs: improve xfs_isilocked

[Alex Elder]

On Thu, 2010-05-27 at 15:05 -0400, Christoph Hellwig wrote:
> Use rwsem_is_locked to make the assertations for shared locks work.

So you're changing it so it answers "yes it's locked"
even it it's only a read lock now, right?

Previously it was basically (once each for ilock and
iolock):  "If the exclusive flag is set, but there is no
writer, then it is not locked; otherwise it is."

Now it's "If the exclusive flag is set, but no writer,
it's not locked.  Otherwise if the shared flag is
set it's locked if rwsem_is_locked() says we are.
Otherwise (ASSERT(0) and) it is not locked."

That last part is wrong I think.  It should be OK to
call xfs_isilocked() with neither flag set, in which
case the result should be 0.  And if the exclusive
flag is set, and there *is* a writer, it *is* locked,
so it should return 1.

					-Alex

> Signed-off-by: Christoph Hellwig <hch@lst.de>
> 
> Index: xfs/fs/xfs/xfs_iget.c
> ===================================================================
> --- xfs.orig/fs/xfs/xfs_iget.c	2010-05-25 11:40:59.216005587 +0200
> +++ xfs/fs/xfs/xfs_iget.c	2010-05-27 20:59:09.244004330 +0200
> @@ -740,30 +738,24 @@ xfs_ilock_demote(
>  }
>  
>  #ifdef DEBUG
> -/*
> - * Debug-only routine, without additional rw_semaphore APIs, we can
> - * now only answer requests regarding whether we hold the lock for write
> - * (reader state is outside our visibility, we only track writer state).
> - *
> - * Note: this means !xfs_isilocked would give false positives, so don't do that.
> - */
>  int
>  xfs_isilocked(
>  	xfs_inode_t		*ip,
>  	uint			lock_flags)
>  {
> -	if ((lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) ==
> -			XFS_ILOCK_EXCL) {
> -		if (!ip->i_lock.mr_writer)
> -			return 0;
> +	if (lock_flags & (XFS_ILOCK_EXCL|XFS_ILOCK_SHARED)) {
> +		if (!(lock_flags & XFS_ILOCK_SHARED))
> +			return !!ip->i_lock.mr_writer;
> +		return rwsem_is_locked(&ip->i_lock.mr_lock);
>  	}
>  
> -	if ((lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) ==
> -			XFS_IOLOCK_EXCL) {
> -		if (!ip->i_iolock.mr_writer)
> -			return 0;
> +	if (lock_flags & (XFS_IOLOCK_EXCL|XFS_IOLOCK_SHARED)) {
> +		if (!(lock_flags & XFS_IOLOCK_SHARED))
> +			return !!ip->i_iolock.mr_writer;
> +		return rwsem_is_locked(&ip->i_iolock.mr_lock);
>  	}
>  
> -	return 1;
> +	ASSERT(0);
> +	return 0;
>  }
>  #endif
> 
> _______________________________________________
> xfs mailing list
> xfs@oss.sgi.com
> http://oss.sgi.com/mailman/listinfo/xfs



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

Re: [PATCH] xfs: improve xfs_isilocked

[Christoph Hellwig]

On Fri, May 28, 2010 at 02:40:53PM -0500, Alex Elder wrote:
> On Thu, 2010-05-27 at 15:05 -0400, Christoph Hellwig wrote:
> > Use rwsem_is_locked to make the assertations for shared locks work.
> 
> So you're changing it so it answers "yes it's locked"
> even it it's only a read lock now, right?

If XFS_ILOCK_SHARED/XFS_IOLOCK_SHARED are in the flags we'll answer yes
it's locked for a read lock now, indeed.

> Previously it was basically (once each for ilock and
> iolock):  "If the exclusive flag is set, but there is no
> writer, then it is not locked; otherwise it is."

Yes.

> Now it's "If the exclusive flag is set, but no writer,
> it's not locked.  Otherwise if the shared flag is
> set it's locked if rwsem_is_locked() says we are.
> Otherwise (ASSERT(0) and) it is not locked."

Not exactly.  Now it's:

 - if excl is set but shared isn't return true if mr_writer is
   set, else false
 - if shared is set either alone or together with excl return
   if it is locked in any way (rwsem_is_locked).

Note that xfs_isilocked can be called like:

	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));

which means that either excl or shared is fine.

 - if either one or both of excl and shared are set and it's 

> That last part is wrong I think.  It should be OK to
> call xfs_isilocked() with neither flag set, in which
> case the result should be 0.

We can argue about removing the assert, but we currently don't
and should't call xfs_isilocked wit ha 0 argument - it's rather
pointless to do so.

> And if the exclusive
> flag is set, and there *is* a writer, it *is* locked,
> so it should return 1.

We do that right now.

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

Re: [PATCH] xfs: improve xfs_isilocked

[Alex Elder]

On Sat, 2010-05-29 at 05:50 -0400, Christoph Hellwig wrote:
> On Fri, May 28, 2010 at 02:40:53PM -0500, Alex Elder wrote:
> > On Thu, 2010-05-27 at 15:05 -0400, Christoph Hellwig wrote:
> > > Use rwsem_is_locked to make the assertations for shared locks work.
> > 
> > So you're changing it so it answers "yes it's locked"
> > even it it's only a read lock now, right?
. . .

> > Now it's "If the exclusive flag is set, but no writer,
> > it's not locked.  Otherwise if the shared flag is
> > set it's locked if rwsem_is_locked() says we are.
> > Otherwise (ASSERT(0) and) it is not locked."
> 
> Not exactly.  Now it's:
> 
>  - if excl is set but shared isn't return true if mr_writer is
>    set, else false
>  - if shared is set either alone or together with excl return
>    if it is locked in any way (rwsem_is_locked).

OK, that makes sense, I get it now.

> Note that xfs_isilocked can be called like:
> 
> 	ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL|XFS_ILOCK_SHARED));
> 
> which means that either excl or shared is fine.
> 
>  - if either one or both of excl and shared are set and it's 
> 
> > That last part is wrong I think.  It should be OK to
> > call xfs_isilocked() with neither flag set, in which
> > case the result should be 0.
> 
> We can argue about removing the assert, but we currently don't
> and should't call xfs_isilocked wit ha 0 argument - it's rather
> pointless to do so.

Yes, you're right.  I'd still say the function should
return the right answer even if given an unreasonable
request.  But that's being pedantic.

> 
> > And if the exclusive
> > flag is set, and there *is* a writer, it *is* locked,
> > so it should return 1.
> 
> We do that right now.

Yup.  Thanks for setting me straight.

Reviewed-by: Alex Elder <aelder@sgi.com>




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