[PATCH] xfs: buffer ->bi_end_io function requires irq-safe lock

[Brian Foster <bfoster@xxxxxxxxxx>]

Reports have surfaced of a lockdep splat complaining about an irq-safe
-> irq-unsafe locking order in the xfs_buf_bio_end_io() bio completion
handler. This only occurs when I/O errors are present because bp->b_lock
is only acquired in this context to protect setting an error on the
buffer. The problem is that this lock can be acquired with the
(request_queue) q->queue_lock held. See scsi_end_request() or
ata_qc_schedule_eh(), for example.

Replace bp->b_lock with a new irq-safe lock for use in bio completion.
This avoids the bad lock ordering and should be otherwise low impact as
the lock is only acquired on I/O error.

Signed-off-by: Brian Foster <bfoster@xxxxxxxxxx>
---

Hi all,

I'm not totally sure this is a real issue, but the fix seemed simple
enough...

Brian

 fs/xfs/xfs_buf.c | 10 ++++++----
 fs/xfs/xfs_buf.h |  1 +
 2 files changed, 7 insertions(+), 4 deletions(-)

diff --git a/fs/xfs/xfs_buf.c b/fs/xfs/xfs_buf.c
index 9a2191b..aa0ce4a 100644
--- a/fs/xfs/xfs_buf.c
+++ b/fs/xfs/xfs_buf.c
@@ -174,6 +174,7 @@ _xfs_buf_alloc(
        RB_CLEAR_NODE(&bp->b_rbnode);
        sema_init(&bp->b_sema, 0); /* held, no waiters */
        spin_lock_init(&bp->b_lock);
+       spin_lock_init(&bp->b_error_lock);
        XB_SET_OWNER(bp);
        bp->b_target = target;
        bp->b_flags = flags;
@@ -1100,21 +1101,22 @@ xfs_bwrite(
        return error;
 }
 
-STATIC void
+static void
 xfs_buf_bio_end_io(
        struct bio              *bio)
 {
-       xfs_buf_t               *bp = (xfs_buf_t *)bio->bi_private;
+       struct xfs_buf          *bp = (struct xfs_buf *)bio->bi_private;
+       unsigned long           flags;
 
        /*
         * don't overwrite existing errors - otherwise we can lose errors on
         * buffers that require multiple bios to complete.
         */
        if (bio->bi_error) {
-               spin_lock(&bp->b_lock);
+               spin_lock_irqsave(&bp->b_error_lock, flags);
                if (!bp->b_io_error)
                        bp->b_io_error = bio->bi_error;
-               spin_unlock(&bp->b_lock);
+               spin_unlock_irqrestore(&bp->b_error_lock, flags);
        }
 
        if (!bp->b_error && xfs_buf_is_vmapped(bp) && (bp->b_flags & XBF_READ))
diff --git a/fs/xfs/xfs_buf.h b/fs/xfs/xfs_buf.h
index 4eb89bd..c84529c 100644
--- a/fs/xfs/xfs_buf.h
+++ b/fs/xfs/xfs_buf.h
@@ -161,6 +161,7 @@ typedef struct xfs_buf {
        spinlock_t              b_lock;         /* internal state lock */
        unsigned int            b_state;        /* internal state flags */
        int                     b_io_error;     /* internal IO error state */
+       spinlock_t              b_error_lock;   /* irq-safe IO error lock */
        wait_queue_head_t       b_waiters;      /* unpin waiters */
        struct list_head        b_list;
        struct xfs_perag        *b_pag;         /* contains rbtree root */
-- 
2.4.11