On Tue, Mar 16, 2021 at 6:49 AM robbieko [off-list ref] wrote:

Hi All,

The patch delayed find orphan roots.
Move to after orphan cleanup with tree_root.
I think this will cause all orphan items to be deleted
when orphan cleanup with tree_root.
Afterwards, find orphan roots cannot find
the subvolume being deleted.

Not entirely able to parse what you are trying to say.

I suppose your concern is that the call to:

btrfs_orphan_cleanup(fs_info->tree_root)

which now happens before calling btrfs_find_orphan_roots(), results in
the orphans for roots being accidentally deleted and therefore cause
no root deletions to happen later?
If that's your concern, than it does not happen because
btrfs_orphan_cleanup() skips deletion of orphan items for deleted
roots.

I've just created a test case to verify it's correct, for RW mounts,
RO mounts and remounts from RO to RW:

https://pastebin.com/raw/zSZjgn48

I couldn't find any regression.

Thanks.

quoted

  out:
      return ret;
  }

@@ -3383,10 +3384,6 @@ int __cold open_ctree(struct super_block *sb, struct btrfs_fs_devices *fs_device
              }
      }

-     ret = btrfs_find_orphan_roots(fs_info);
-     if (ret)
-             goto fail_qgroup;
-
      fs_info->fs_root = btrfs_get_fs_root(fs_info, BTRFS_FS_TREE_OBJECTID, true);
      if (IS_ERR(fs_info->fs_root)) {
              err = PTR_ERR(fs_info->fs_root);