[NeilBrown <neilb@ownmail.net>]

This new version
 - includes some reviewed-bys
 - adds an explanation to second patch of why ext4_fc_eligible() was
   placed where it was in one case
 - modifies the last patch (but keeps the Reviews - hope that is OK)
   to keep the changes to ->i_nlink in __ext4_link
    https://sashiko.dev/#/patchset/20260317224638.3809014-1-neilb%40ownmail.net
   points out that removing it for the replay path might not be correct,
   and I cannot now see any reason to move those changes (ext4_inc_count() 
   and drop_nlink())

From previous email:

My particular interest in changing ext4 is to remove the use of
d_alloc().  I will want to deprecate d_alloc() as it doesn't fit the new
model well.

The use of d_alloc() in ext4 is incidental to the actual task at hand.
The code really wants to pass around a parent directory and a name, and
only uses the dentry because that seems convenient.  As these patches
show the code actually becomes simpler when we avoid the dentry.

NeilBrown

 [PATCH v2 1/3] ext4: split __ext4_add_entry() out of ext4_add_entry()
 [PATCH v2 2/3] ext4: add ext4_fc_eligible()
 [PATCH v2 3/3] ext4: move dcache manipulation out of __ext4_link()