Skip to contentSkip to conversation

fix spinlock recursion on xa_lock in xfs_buf_item_push

PATCH In review Brian Foster 3 participants 6 messages Feb 11, 2013 – Feb 14, 2013 XFS filesystem

lore
Share & tools
lore
#1 lore raw

PATCH v4 1/2 xfs: recheck buffer pinned status after push trylock failure

In reply to Brian Foster

The buffer pinned check and trylock sequence in xfs_buf_item_push() can race with an active transaction on marking the buffer pinned. This can result in the buffer becoming pinned and stale after the initial check and the trylock failure, but before the check in xfs_buf_trylock() that issues a log force. If the log force is issued from this context, a spinlock recursion occurs on xa_lock.

Prepare xfs_buf_item_push() to handle the race by detecting a pinned buffer after the trylock failure so xfsaild issues a log force from a safe context. This, along with various previous fixes, renders the log force in xfs_buf_trylock() redundant.

---
 fs/xfs/xfs_buf_item.c |   12 +++++++++++-
 1 files changed, 11 insertions(+), 1 deletions(-)
fs/xfs/xfs_buf_item.c +11 -1
index 9c4c050..cf26347 100644
@@ -469,8 +469,18 @@ xfs_buf_item_push(
 
 	if (xfs_buf_ispinned(bp))
 		return XFS_ITEM_PINNED;
-	if (!xfs_buf_trylock(bp))
+	if (!xfs_buf_trylock(bp)) {
+		/*
+		 * If we have just raced with a buffer being pinned and it has
+		 * been marked stale, we could end up stalling until someone else
+		 * issues a log force to unpin the stale buffer. Check for the
+		 * race condition here so xfsaild recognizes the buffer is pinned
+		 * and queues a log force to move it along.
+		 */
+		if (xfs_buf_ispinned(bp))
+			return XFS_ITEM_PINNED;
 		return XFS_ITEM_LOCKED;
+	}
 
 	ASSERT(!(bip->bli_flags & XFS_BLI_STALE));
 
Signature · 6 lines
1.7.7.6

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
#3 lore raw

PATCH v4 2/2 xfs: remove log force from xfs_buf_trylock()

In reply to Brian Foster

The trylock log force invoked via xfs_buf_item_push() can attempt to acquire xa_lock, thus leading to a recursion bug when called with xa_lock held.

This log force was originally added to xfs_buf_trylock() to address xfsaild stalls due to pinned and stale buffers. Since the addition of this behavior, the log item pushing code had been reworked to detect and track pinned items to inform xfsaild to issue a log force itself when necessary. As such, the log force on trylock failure is redundant and safe to remove.

---
 fs/xfs/xfs_buf.c |    2 --
 1 files changed, 0 insertions(+), 2 deletions(-)
fs/xfs/xfs_buf.c +0 -2
index fbbb9eb..4e8f0df 100644
@@ -951,8 +951,6 @@ xfs_buf_trylock(
 	locked = down_trylock(&bp->b_sema) == 0;
 	if (locked)
 		XB_SET_OWNER(bp);
-	else if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
-		xfs_log_force(bp->b_target->bt_mount, 0);
 
 	trace_xfs_buf_trylock(bp, _RET_IP_);
 	return locked;
Signature · 6 lines
1.7.7.6

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
#0 lore raw

v4 cover fix spinlock recursion on xa_lock in xfs_buf_item_push

Hi all,

Here is v4 of the spinlock recursion fix. The only update is to the comment in patch 1. These changes have been run through the reproducer over the weekend without a failure.

I've also run another xfstests run while including the following patch from Dave:

xfs: xfs_bmap_add_attrfork_local is too generic

... due to seeing the same issue Lukas reproduced in test 013 (I saw it in 070 as well). That problem no longer occurs with this patch included, but I have hit an error in test 178 that didn't fire in my baseline run. It persists if I revert back to TOT, so I'm posting this set as is and I'll see if I can trace where that one came from...

Brian

v4:
- Fix comment in patch 1.
v3:
- Patches reordered and revised to:
1.) Add the race detection in xfs_buf_item_push().
2.) Remove the log force from xfs_buf_trylock() entirely to fix the recursion
and rely on xfsaild to issue the log force.
v2:
- Patch 2 is reworked to detect the potential race and defer the log force to xfsaild by returning XFS_ITEM_PINNED.

Brian Foster (2):
xfs: recheck buffer pinned status after push trylock failure
xfs: remove log force from xfs_buf_trylock()

 fs/xfs/xfs_buf.c      |    2 --
 fs/xfs/xfs_buf_item.c |   12 +++++++++++-
 2 files changed, 11 insertions(+), 3 deletions(-)
Signature · 6 lines
1.7.7.6

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Reviewed-by on v4 1/2
#2 lore raw

Re: [PATCH v4 1/2] xfs: recheck buffer pinned status after push trylock failure

In reply to Brian Foster
Quoted 7 lines · Brian Foster

The buffer pinned check and trylock sequence in xfs_buf_item_push() can race with an active transaction on marking the buffer pinned. This can result in the buffer becoming pinned and stale after the initial check and the trylock failure, but before the check in xfs_buf_trylock() that issues a log force. If the log force is issued from this context, a spinlock recursion occurs on xa_lock.

Quoted 6 lines

Prepare xfs_buf_item_push() to handle the race by detecting a pinned buffer after the trylock failure so xfsaild issues a log force from a safe context. This, along with various previous fixes, renders the log force in xfs_buf_trylock() redundant.

Looks good, doesn't seem to cause any problems I can see here.

Signature · 7 lines
Dave Chinner
[email protected]

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
Reviewed-by on v4 2/2
#4 lore raw

Re: [PATCH v4 2/2] xfs: remove log force from xfs_buf_trylock()

In reply to Brian Foster
Quoted 4 lines · Brian Foster

The trylock log force invoked via xfs_buf_item_push() can attempt to acquire xa_lock, thus leading to a recursion bug when called with xa_lock held.

Quoted 8 lines

This log force was originally added to xfs_buf_trylock() to address xfsaild stalls due to pinned and stale buffers. Since the addition of this behavior, the log item pushing code had been reworked to detect and track pinned items to inform xfsaild to issue a log force itself when necessary. As such, the log force on trylock failure is redundant and safe to remove.

Looks good.

Signature · 7 lines
Dave Chinner
[email protected]

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs
#5 lore raw
In reply to Brian Foster
Quoted 8 lines · Brian Foster

Hi all,

Here is v4 of the spinlock recursion fix. The only update is to the comment in patch 1. These changes have been run through the reproducer over the weekend without a failure.

I've also run another xfstests run while including the following patch from Dave:

Quoted 7 lines

xfs: xfs_bmap_add_attrfork_local is too generic

... due to seeing the same issue Lukas reproduced in test 013 (I saw it in 070 as well). That problem no longer occurs with this patch included, but I have hit an error in test 178 that didn't fire in my baseline run. It persists if I revert back to TOT, so I'm posting this set as is and I'll see if I can trace where that one came from...

Applied these two.

Regards,
Ben

_______________________________________________
xfs mailing list
[email protected]
http://oss.sgi.com/mailman/listinfo/xfs