[PATCH] xfs: annotate lockless b_flags read in xfs_buf_lock

[Cen Zhang]

xfs_buf_lock() reads bp->b_flags before acquiring the buffer semaphore
to check whether a stale, pinned buffer needs a log force:

    if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))

This races with xfs_trans_dirty_buf(), which modifies b_flags while
the buffer is locked by a transaction on another CPU.

The pre-semaphore check is a performance hint: if a stale pinned
buffer is detected, forcing the log avoids a long wait on the
semaphore.  Either outcome of the race is benign -- a false positive
triggers a harmless log force, and a false negative simply means the
caller blocks on the semaphore and the log force happens later.

Annotate the lockless read with READ_ONCE().

Fixes: ed3b4d6cdc81 ("xfs: Improve scalability of busy extent tracking")
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Cen Zhang <zzzccc427@xxxxxxxxx>
---
 fs/xfs/xfs_buf.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/xfs/xfs_buf.c b/fs/xfs/xfs_buf.c
index d2f3c50d80e7..6819477307bd 100644
--- a/fs/xfs/xfs_buf.c
+++ b/fs/xfs/xfs_buf.c
@@ -988,7 +988,7 @@ xfs_buf_lock(
 {
 	trace_xfs_buf_lock(bp, _RET_IP_);
 
-	if (atomic_read(&bp->b_pin_count) && (bp->b_flags & XBF_STALE))
+	if (atomic_read(&bp->b_pin_count) && (READ_ONCE(bp->b_flags) & XBF_STALE))
 		xfs_log_force(bp->b_mount, 0);
 	down(&bp->b_sema);
 
-- 
2.34.1