Re: [PATCH] ext4: attempt to shrink directory on dentry removal

["Theodore Y. Ts'o" <tytso@mit.edu>]

On Sun, Aug 25, 2019 at 11:07:46PM -0600, Andreas Dilger wrote:
> This may not always detect empty directory blocks properly, because
> ext4_generic_delete_entry() will only merge deleted entries with the
> previous entry.  It at least appears possible that if entries are not
> deleted in the proper order (e.g. in reverse of the order they are
> listed in the directory) there may be multiple empty entries in a block,
> and the above check will fail.

I don't think that's a problem.  We always merge with the previous
entry, whether it's an empty/deleted entry or an in-use entry.  So
long as all implementations do this, it works just fine.  If there is
an ext2/3/4 implementation which deletes the entry by simply clearing
the inode number *without* merging with the previous one, it's
possible that we might get confused.

But that's easily fixed, too.  In ext4_generic_delete_entry(), we just
need to add a check so that we check to see if the subsequent entry
(if it exists) has a zero de->inode value.  If so, then we absorb the
current directory entry to include the deleted subsequent entry and
repeat the check.

						- Ted