block: fix locking for struct block_device size updates

Two different callers use two different mutexes for updating the
block device size, which obviously doesn't help to actually protect
against concurrent updates from the different callers.  In addition
one of the locks, bd_mutex is rather prone to deadlocks with other
parts of the block stack that use it for high level synchronization.

Switch to using a new spinlock protecting just the size updates, as
that is all we need, and make sure everyone does the update through
the proper helper.

This fixes a bug reported with the nvme revalidating disks during a
hot removal operation, which can currently deadlock on bd_mutex.
Reported-by: Xianting Tian <xianting_tian@126.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

block/partitions/core.c

@@ -592,8 +592,8 @@ int bdev_resize_partition(struct block_device *bdev, int partno,
 	if (partition_overlaps(bdev->bd_disk, start, length, partno))
 		goto out_unlock;
 
-	part_nr_sects_write(part, (sector_t)length);
-	i_size_write(bdevp->bd_inode, length << SECTOR_SHIFT);
+	part_nr_sects_write(part, length);
+	bd_set_nr_sectors(bdevp, length);
 
 	ret = 0;
 out_unlock:

drivers/block/aoe/aoecmd.c

@@ -900,9 +900,7 @@ aoecmd_sleepwork(struct work_struct *work)
 		ssize = get_capacity(d->gd);
 		bd = bdget_disk(d->gd, 0);
 		if (bd) {
-			inode_lock(bd->bd_inode);
-			i_size_write(bd->bd_inode, (loff_t)ssize<<9);
-			inode_unlock(bd->bd_inode);
+			bd_set_nr_sectors(bd, ssize);
 			bdput(bd);
 		}
 		spin_lock_irq(&d->lock);

drivers/md/dm.c

@@ -2097,18 +2097,6 @@ static void event_callback(void *context)
 	dm_issue_global_event();
 }
 
-/*
- * Protected by md->suspend_lock obtained by dm_swap_table().
- */
-static void __set_size(struct mapped_device *md, sector_t size)
-{
-	lockdep_assert_held(&md->suspend_lock);
-
-	set_capacity(md->disk, size);
-
-	i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT);
-}
-
 /*
  * Returns old map, which caller must destroy.
  */
@@ -2131,7 +2119,8 @@ static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t,
 	if (size != dm_get_size(md))
 		memset(&md->geometry, 0, sizeof(md->geometry));
 
-	__set_size(md, size);
+	set_capacity(md->disk, size);
+	bd_set_nr_sectors(md->bdev, size);
 
 	dm_table_event_callback(t, event_callback, md);
 

drivers/s390/block/dasd_ioctl.c

@@ -55,10 +55,7 @@ dasd_ioctl_enable(struct block_device *bdev)
 
 	dasd_enable_device(base);
 	/* Formatting the dasd device can change the capacity. */
-	mutex_lock(&bdev->bd_mutex);
-	i_size_write(bdev->bd_inode,
-		     (loff_t)get_capacity(base->block->gdp) << 9);
-	mutex_unlock(&bdev->bd_mutex);
+	bd_set_nr_sectors(bdev, get_capacity(base->block->gdp));
 	dasd_put_device(base);
 	return 0;
 }
@@ -91,9 +88,7 @@ dasd_ioctl_disable(struct block_device *bdev)
 	 * Set i_size to zero, since read, write, etc. check against this
 	 * value.
 	 */
-	mutex_lock(&bdev->bd_mutex);
-	i_size_write(bdev->bd_inode, 0);
-	mutex_unlock(&bdev->bd_mutex);
+	bd_set_nr_sectors(bdev, 0);
 	dasd_put_device(base);
 	return 0;
 }

fs/block_dev.c

@@ -876,6 +876,7 @@ struct block_device *bdget(dev_t dev)
 	bdev = &BDEV_I(inode)->bdev;
 
 	if (inode->i_state & I_NEW) {
+		spin_lock_init(&bdev->bd_size_lock);
 		bdev->bd_contains = NULL;
 		bdev->bd_super = NULL;
 		bdev->bd_inode = inode;
@@ -1290,6 +1291,7 @@ static void check_disk_size_change(struct gendisk *disk,
 {
 	loff_t disk_size, bdev_size;
 
+	spin_lock(&bdev->bd_size_lock);
 	disk_size = (loff_t)get_capacity(disk) << 9;
 	bdev_size = i_size_read(bdev->bd_inode);
 	if (disk_size != bdev_size) {
@@ -1299,11 +1301,15 @@ static void check_disk_size_change(struct gendisk *disk,
 			       disk->disk_name, bdev_size, disk_size);
 		}
 		i_size_write(bdev->bd_inode, disk_size);
-		if (bdev_size > disk_size && __invalidate_device(bdev, false))
+	}
+	bdev->bd_invalidated = 0;
+	spin_unlock(&bdev->bd_size_lock);
+
+	if (bdev_size > disk_size) {
+		if (__invalidate_device(bdev, false))
 			pr_warn("VFS: busy inodes on resized disk %s\n",
 				disk->disk_name);
 	}
-	bdev->bd_invalidated = 0;
 }
 
 /**
@@ -1328,14 +1334,11 @@ int revalidate_disk(struct gendisk *disk)
 	if (!(disk->flags & GENHD_FL_HIDDEN)) {
 		struct block_device *bdev = bdget_disk(disk, 0);
 
-		if (!bdev)
-			return ret;
-
-		mutex_lock(&bdev->bd_mutex);
+		if (bdev) {
 			check_disk_size_change(disk, bdev, ret == 0);
-		mutex_unlock(&bdev->bd_mutex);
 			bdput(bdev);
 		}
+	}
 	return ret;
 }
 EXPORT_SYMBOL(revalidate_disk);
@@ -1373,9 +1376,9 @@ EXPORT_SYMBOL(check_disk_change);
 
 void bd_set_nr_sectors(struct block_device *bdev, sector_t sectors)
 {
-	inode_lock(bdev->bd_inode);
+	spin_lock(&bdev->bd_size_lock);
 	i_size_write(bdev->bd_inode, (loff_t)sectors << SECTOR_SHIFT);
-	inode_unlock(bdev->bd_inode);
+	spin_unlock(&bdev->bd_size_lock);
 }
 EXPORT_SYMBOL(bd_set_nr_sectors);
 

include/linux/blk_types.h

@@ -38,6 +38,7 @@ struct block_device {
 	/* number of times partitions within this device have been opened. */
 	unsigned		bd_part_count;
 	int			bd_invalidated;
+	spinlock_t		bd_size_lock; /* for bd_inode->i_size updates */
 	struct gendisk *	bd_disk;
 	struct backing_dev_info *bd_bdi;