block: turn bdev_lock into a mutex
authorChristoph Hellwig <hch@lst.de>
Thu, 1 Jun 2023 09:44:46 +0000 (11:44 +0200)
committerJens Axboe <axboe@kernel.dk>
Mon, 5 Jun 2023 16:53:03 +0000 (10:53 -0600)
There is no reason for this lock to spin, and being able to sleep under
it will come in handy soon.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Acked-by: Christian Brauner <brauner@kernel.org>
Acked-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Link: https://lore.kernel.org/r/20230601094459.1350643-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
block/bdev.c

index 080b5c8..f5ffcac 100644 (file)
@@ -308,7 +308,7 @@ EXPORT_SYMBOL(thaw_bdev);
  * pseudo-fs
  */
 
-static  __cacheline_aligned_in_smp DEFINE_SPINLOCK(bdev_lock);
+static  __cacheline_aligned_in_smp DEFINE_MUTEX(bdev_lock);
 static struct kmem_cache * bdev_cachep __read_mostly;
 
 static struct inode *bdev_alloc_inode(struct super_block *sb)
@@ -467,9 +467,6 @@ long nr_blockdev_pages(void)
  *
  * Test whether @bdev can be claimed by @holder.
  *
- * CONTEXT:
- * spin_lock(&bdev_lock).
- *
  * RETURNS:
  * %true if @bdev can be claimed, %false otherwise.
  */
@@ -477,6 +474,8 @@ static bool bd_may_claim(struct block_device *bdev, void *holder)
 {
        struct block_device *whole = bdev_whole(bdev);
 
+       lockdep_assert_held(&bdev_lock);
+
        if (bdev->bd_holder) {
                /*
                 * The same holder can always re-claim.
@@ -515,10 +514,10 @@ int bd_prepare_to_claim(struct block_device *bdev, void *holder)
        if (WARN_ON_ONCE(!holder))
                return -EINVAL;
 retry:
-       spin_lock(&bdev_lock);
+       mutex_lock(&bdev_lock);
        /* if someone else claimed, fail */
        if (!bd_may_claim(bdev, holder)) {
-               spin_unlock(&bdev_lock);
+               mutex_unlock(&bdev_lock);
                return -EBUSY;
        }
 
@@ -528,7 +527,7 @@ retry:
                DEFINE_WAIT(wait);
 
                prepare_to_wait(wq, &wait, TASK_UNINTERRUPTIBLE);
-               spin_unlock(&bdev_lock);
+               mutex_unlock(&bdev_lock);
                schedule();
                finish_wait(wq, &wait);
                goto retry;
@@ -536,7 +535,7 @@ retry:
 
        /* yay, all mine */
        whole->bd_claiming = holder;
-       spin_unlock(&bdev_lock);
+       mutex_unlock(&bdev_lock);
        return 0;
 }
 EXPORT_SYMBOL_GPL(bd_prepare_to_claim); /* only for the loop driver */
@@ -562,7 +561,7 @@ static void bd_finish_claiming(struct block_device *bdev, void *holder)
 {
        struct block_device *whole = bdev_whole(bdev);
 
-       spin_lock(&bdev_lock);
+       mutex_lock(&bdev_lock);
        BUG_ON(!bd_may_claim(bdev, holder));
        /*
         * Note that for a whole device bd_holders will be incremented twice,
@@ -573,7 +572,7 @@ static void bd_finish_claiming(struct block_device *bdev, void *holder)
        bdev->bd_holders++;
        bdev->bd_holder = holder;
        bd_clear_claiming(whole, holder);
-       spin_unlock(&bdev_lock);
+       mutex_unlock(&bdev_lock);
 }
 
 /**
@@ -587,9 +586,9 @@ static void bd_finish_claiming(struct block_device *bdev, void *holder)
  */
 void bd_abort_claiming(struct block_device *bdev, void *holder)
 {
-       spin_lock(&bdev_lock);
+       mutex_lock(&bdev_lock);
        bd_clear_claiming(bdev_whole(bdev), holder);
-       spin_unlock(&bdev_lock);
+       mutex_unlock(&bdev_lock);
 }
 EXPORT_SYMBOL(bd_abort_claiming);
 
@@ -602,7 +601,7 @@ static void bd_end_claim(struct block_device *bdev)
         * Release a claim on the device.  The holder fields are protected with
         * bdev_lock.  open_mutex is used to synchronize disk_holder unlinking.
         */
-       spin_lock(&bdev_lock);
+       mutex_lock(&bdev_lock);
        WARN_ON_ONCE(--bdev->bd_holders < 0);
        WARN_ON_ONCE(--whole->bd_holders < 0);
        if (!bdev->bd_holders) {
@@ -612,7 +611,7 @@ static void bd_end_claim(struct block_device *bdev)
        }
        if (!whole->bd_holders)
                whole->bd_holder = NULL;
-       spin_unlock(&bdev_lock);
+       mutex_unlock(&bdev_lock);
 
        /*
         * If this was the last claim, remove holder link and unblock evpoll if