OSDN Git Service

xfs: remove leftover CoW reservations when remounting ro
authorDarrick J. Wong <darrick.wong@oracle.com>
Thu, 14 Dec 2017 23:46:05 +0000 (15:46 -0800)
committerDarrick J. Wong <darrick.wong@oracle.com>
Thu, 21 Dec 2017 16:47:32 +0000 (08:47 -0800)
When we're remounting the filesystem readonly, remove all CoW
preallocations prior to going ro.  If the fs goes down after the ro
remount, we never clean up the staging extents, which means xfs_check
will trip over them on a subsequent run.  Practically speaking, the next
mount will clean them up too, so this is unlikely to be seen.  Since we
shut down the cowblocks cleaner on remount-ro, we also have to make sure
we start it back up if/when we remount-rw.

Found by adding clonerange to fsstress and running xfs/017.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
fs/xfs/xfs_icache.c
fs/xfs/xfs_icache.h
fs/xfs/xfs_super.c

index 58d2d42..3861d61 100644 (file)
@@ -870,7 +870,7 @@ xfs_eofblocks_worker(
  * based on the 'speculative_cow_prealloc_lifetime' tunable (5m by default).
  * (We'll just piggyback on the post-EOF prealloc space workqueue.)
  */
-STATIC void
+void
 xfs_queue_cowblocks(
        struct xfs_mount *mp)
 {
index bff4d85..d4a7758 100644 (file)
@@ -81,6 +81,7 @@ void xfs_inode_clear_cowblocks_tag(struct xfs_inode *ip);
 int xfs_icache_free_cowblocks(struct xfs_mount *, struct xfs_eofblocks *);
 int xfs_inode_free_quota_cowblocks(struct xfs_inode *ip);
 void xfs_cowblocks_worker(struct work_struct *);
+void xfs_queue_cowblocks(struct xfs_mount *);
 
 int xfs_inode_ag_iterator(struct xfs_mount *mp,
        int (*execute)(struct xfs_inode *ip, int flags, void *args),
index f663022..2db6a40 100644 (file)
@@ -1360,6 +1360,7 @@ xfs_fs_remount(
                        xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
                        return error;
                }
+               xfs_queue_cowblocks(mp);
 
                /* Create the per-AG metadata reservation pool .*/
                error = xfs_fs_reserve_ag_blocks(mp);
@@ -1369,6 +1370,14 @@ xfs_fs_remount(
 
        /* rw -> ro */
        if (!(mp->m_flags & XFS_MOUNT_RDONLY) && (*flags & MS_RDONLY)) {
+               /* Get rid of any leftover CoW reservations... */
+               cancel_delayed_work_sync(&mp->m_cowblocks_work);
+               error = xfs_icache_free_cowblocks(mp, NULL);
+               if (error) {
+                       xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
+                       return error;
+               }
+
                /* Free the per-AG metadata reservation pool. */
                error = xfs_fs_unreserve_ag_blocks(mp);
                if (error) {