btrfs: move and comment read-only check in btrfs_cancel_balance

Balance cannot be started on a read-only filesystem and will have to
finish/exit before eg. going to read-only via remount.

In case the filesystem is forcibly set to read-only after an error,
balance will finish anyway and if the cancel call is too fast it will
just wait for that to happen.

The last case is when the balance is paused after mount but it's
read-only and cancelling would want to delete the item. The test is
moved after the check if balance is running at all, as it looks more
logical to report "no balance running" instead of "read-only
filesystem".

Reviewed-by: Anand Jain <anand.jain@oracle.com>
Signed-off-by: David Sterba <dsterba@suse.com>

diff --git a/fs/btrfs/volumes.c b/fs/btrfs/volumes.c
index ed230247ae5d5f577cec2719bb17c1d28015a05e..ff97f4749028fef82da952e505e104d55fc8299b 100644 (file)
--- a/fs/btrfs/volumes.c
+++ b/fs/btrfs/volumes.c
@@ -4083,15 +4083,22 @@ int btrfs_pause_balance(struct btrfs_fs_info *fs_info)
 
 int btrfs_cancel_balance(struct btrfs_fs_info *fs_info)
 {
-       if (sb_rdonly(fs_info->sb))
-               return -EROFS;
-
        mutex_lock(&fs_info->balance_mutex);
        if (!fs_info->balance_ctl) {
                mutex_unlock(&fs_info->balance_mutex);
                return -ENOTCONN;
        }
 
+       /*
+        * A paused balance with the item stored on disk can be resumed at
+        * mount time if the mount is read-write. Otherwise it's still paused
+        * and we must not allow cancelling as it deletes the item.
+        */
+       if (sb_rdonly(fs_info->sb)) {
+               mutex_unlock(&fs_info->balance_mutex);
+               return -EROFS;
+       }
+
        atomic_inc(&fs_info->balance_cancel_req);
        /*
         * if we are running just wait and return, balance item is