Re: [PATCH 2/2] simplify xfs_lock_for_rename

[Christoph Hellwig <hch@xxxxxx>]

> 
> It seems that do_rename() in namei.c calls lookup_hash/__lookup_hash
> which does an inode->i_op->lookup() to get the dentry. This will do
> the xfs_lookup anyway, taking a refernence if the target exists.
> 
>       error = do_path_lookup(newdfd, newname, LOOKUP_PARENT, &newnd);
>       ...
>       new_dentry = lookup_hash(&newnd);
>       ...
>       error = vfs_rename(old_dir->d_inode, old_dentry,
>                                  new_dir->d_inode, new_dentry);
> 
> Therefore, target_ip/ipp2 is already known then when xfs_rename/
> xfs_lock_for_rename is called.

You're right.  Back to the drawing board to make rename even simpler.
None of the other patches I sent yesterday depends on this one so they
can go in independently.  And I'd especially like to see the lookup one
before lookup gets more complicated again for CI support.