btrfs: get ->kill_sb() of its own
authorAl Viro <viro@zeniv.linux.org.uk>
Thu, 17 Nov 2011 06:22:46 +0000 (01:22 -0500)
committerAl Viro <viro@zeniv.linux.org.uk>
Mon, 9 Jan 2012 00:33:23 +0000 (19:33 -0500)
... and move free_fs_info() to that, out of ->put_super().  Do NOT
set ->s_fs_info to NULL in the latter; we need it for sget() to
be able to see and wait for fs in the middle of umount if we get a
mount/umount race.

Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
fs/btrfs/super.c

index a3f435e..eca4862 100644 (file)
@@ -147,14 +147,13 @@ void __btrfs_std_error(struct btrfs_fs_info *fs_info, const char *function,
 
 static void btrfs_put_super(struct super_block *sb)
 {
-       struct btrfs_root *root = btrfs_sb(sb);
-       int ret;
-
-       ret = close_ctree(root);
-       free_fs_info(root->fs_info);
-       sb->s_fs_info = NULL;
-
-       (void)ret; /* FIXME: need to fix VFS to return error? */
+       (void)close_ctree(btrfs_sb(sb));
+       /* FIXME: need to fix VFS to return error? */
+       /* AV: return it _where_?  ->put_super() can be triggered by any number
+        * of async events, up to and including delivery of SIGKILL to the
+        * last process that kept it busy.  Or segfault in the aforementioned
+        * process...  Whom would you report that to?
+        */
 }
 
 enum {
@@ -1223,11 +1222,21 @@ static int btrfs_statfs(struct dentry *dentry, struct kstatfs *buf)
        return 0;
 }
 
+static void btrfs_kill_super(struct super_block *sb)
+{
+       struct btrfs_fs_info *fs_info = NULL;
+       if (sb->s_root)
+               fs_info = btrfs_sb(sb)->fs_info;
+       kill_anon_super(sb);
+       if (fs_info)
+               free_fs_info(fs_info);
+}
+
 static struct file_system_type btrfs_fs_type = {
        .owner          = THIS_MODULE,
        .name           = "btrfs",
        .mount          = btrfs_mount,
-       .kill_sb        = kill_anon_super,
+       .kill_sb        = btrfs_kill_super,
        .fs_flags       = FS_REQUIRES_DEV,
 };