Re: [PATCH 2/2] btrfs: clear spurious free-space entries

[David Sterba <dsterba@suse.cz>]

On Wed, Sep 24, 2025 at 06:35:43AM +0930, Qu Wenruo wrote:
> 
> 
> 在 2025/9/24 01:24, Mark Harmstone 写道:
> > Version 6.16.1 of btrfs-progs fixes a broken btrfs check test for
> > spurious entries in the free-space tree, those that don't belong to any
> > block group. Unfortunately mkfs.btrfs had been generating these, meaning
> > that these filesystems will now fail btrfs check.
> > 
> > Add a compat flag BTRFS_FEATURE_COMPAT_NO_SPURIOUS_FREE_SPACE, and if on
> > mount we find this isn't set, clean any spurious entries from the
> > beginning of the free-space tree.
> 
> I found this compat flag a little overkilled.
> 
> Are we really going to introduce a new compat flag every time there is 
> something wrong with the free space tree?

Agreed, the compat flag does not make sense to me in this case.

> > +	ret = btrfs_search_slot(trans, fst, &key, path, 0, 0);
> > +	if (ret < 0)
> > +		goto end_trans;
> > +
> > +	while (true) {
> > +		leaf = path->nodes[0];
> > +		if (path->slots[0] >= btrfs_header_nritems(leaf)) {
> > +			ret = btrfs_next_leaf(fst, path);
> > +			if (ret < 0)
> > +				goto end_trans;
> > +			if (ret > 0)
> > +				break;
> > +			leaf = path->nodes[0];
> > +		}
> > +
> > +		btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
> > +
> > +		bg = btrfs_lookup_first_block_group(fs_info, key.objectid);
> 
> We can do the lookup out of the loop. With a parameter 0 for @bytenr.
> 
> As we just need to delete any entry before the first bg, there is no 
> need to lookup the bg every time.
> 
> Furthermore, since the stale entries are just from the temporary chunks, 
> they should be pretty small, thus we are able to afford the check at 
> every mount.

A one time check per mount sounds like the best solution, it has
negligible cost, compared to compat bits.