block: Try to handle busy underlying device on discard

Commit 384d87ef2c95 ("block: Do not discard buffers under a mounted
filesystem") made paths issuing discard or zeroout requests to the
underlying device try to grab block device in exclusive mode. If that
failed we returned EBUSY to userspace. This however caused unexpected
fallout in userspace where e.g. FUSE filesystems issue discard requests
from userspace daemons although the device is open exclusively by the
kernel. Also shrinking of logical volume by LVM issues discard requests
to a device which may be claimed exclusively because there's another LV
on the same PV. So to avoid these userspace regressions, fall back to
invalidate_inode_pages2_range() instead of returning EBUSY to userspace
and return EBUSY only of that call fails as well (meaning that there's
indeed someone using the particular device range we are trying to
discard).

Link: https://bugzilla.kernel.org/show_bug.cgi?id=211167
Fixes: 384d87ef2c95 ("block: Do not discard buffers under a mounted filesystem")
CC: stable@vger.kernel.org
Signed-off-by: Jan Kara <jack@suse.cz>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

diff --git a/fs/block_dev.c b/fs/block_dev.c
index 4aa1f88d5bf8b390db16fab9d8999f0c9200b588..03166b3dea4db2208e6786f6c8a7f991f655c409 100644 (file)
--- a/fs/block_dev.c
+++ b/fs/block_dev.c
@@ -118,13 +118,22 @@ int truncate_bdev_range(struct block_device *bdev, fmode_t mode,
        if (!(mode & FMODE_EXCL)) {
                int err = bd_prepare_to_claim(bdev, truncate_bdev_range);
                if (err)
-                       return err;
+                       goto invalidate;
        }
 
        truncate_inode_pages_range(bdev->bd_inode->i_mapping, lstart, lend);
        if (!(mode & FMODE_EXCL))
                bd_abort_claiming(bdev, truncate_bdev_range);
        return 0;
+
+invalidate:
+       /*
+        * Someone else has handle exclusively open. Try invalidating instead.
+        * The 'end' argument is inclusive so the rounding is safe.
+        */
+       return invalidate_inode_pages2_range(bdev->bd_inode->i_mapping,
+                                            lstart >> PAGE_SHIFT,
+                                            lend >> PAGE_SHIFT);
 }
 
 static void set_init_blocksize(struct block_device *bdev)