Re: [PATCH 5/7] btrfs: clear BTRFS_ROOT_DEAD_RELOC_TREE before dropping the reloc root

[David Sterba <dsterba@suse.cz>]

On Mon, Mar 02, 2020 at 02:51:11PM -0500, Josef Bacik wrote:
> On 3/2/20 2:31 PM, David Sterba wrote:
> > On Mon, Mar 02, 2020 at 01:47:55PM -0500, Josef Bacik wrote:
> >> We were doing the clear dance for the reloc root after doing the drop of
> >> the reloc root, which means we have a giant window where we could miss
> >> having BTRFS_ROOT_DEAD_RELOC_TREE unset and the reloc_root == NULL.
> >>
> >> Signed-off-by: Josef Bacik <josef@toxicpanda.com>
> >> ---
> >>   fs/btrfs/relocation.c | 13 +++++++------
> >>   1 file changed, 7 insertions(+), 6 deletions(-)
> >>
> >> diff --git a/fs/btrfs/relocation.c b/fs/btrfs/relocation.c
> >> index e60450c44406..acd21c156378 100644
> >> --- a/fs/btrfs/relocation.c
> >> +++ b/fs/btrfs/relocation.c
> >> @@ -2291,18 +2291,19 @@ static int clean_dirty_subvols(struct reloc_control *rc)
> >>   
> >>   			list_del_init(&root->reloc_dirty_list);
> >>   			root->reloc_root = NULL;
> >> -			if (reloc_root) {
> >> -
> >> -				ret2 = btrfs_drop_snapshot(reloc_root, NULL, 0, 1);
> >> -				if (ret2 < 0 && !ret)
> >> -					ret = ret2;
> >> -			}
> >>   			/*
> >>   			 * Need barrier to ensure clear_bit() only happens after
> >>   			 * root->reloc_root = NULL. Pairs with have_reloc_root.
> >>   			 */
> >>   			smp_wmb();
> >>   			clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state);
> >> +
> >> +			if (reloc_root) {
> >> +
> >> +				ret2 = btrfs_drop_snapshot(reloc_root, NULL, 0, 1);
> >> +				if (ret2 < 0 && !ret)
> >> +					ret = ret2;
> >> +			}
> > 
> > This reverts fix 1fac4a54374f7ef385938f3c6cf7649c0fe4f6cd that moved if
> > (reloc_root) before the clear_bit.
> > 
> 
> Hmm we should probably keep this and move the
> 
> if (root->reloc_root)
> 
> thing after the
> 
>          if (!rc || !rc->create_reloc_tree ||
>              root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
>                  return 0;
> 
> to properly fix this.  I'll add this and send an updated series.  Thanks,

Also please update the changelog, it's too vague for a code that had
several bugs regarding the reloc_root lifetime.