Date: Fri, 7 Mar 2014 14:01:36 -0500
From: "J. Bruce Fields" <bfields@fieldses.org>
To: Trond Myklebust <trond.myklebust@primarydata.com>
Cc: "Yan, Zheng" <zheng.z.yan@intel.com>, David Howells <dhowells@redhat.com>,
        linux-fsdevel@vger.kernel.org,
        Viro Alexander <viro@zeniv.linux.org.uk>, linux-nfs@vger.kernel.org,
        ceph-devel@vger.kernel.org, sage@inktank.com
Subject: Re: [PATCH][RESEND] vfs: Make __d_materialise_dentry() set the
 materialised dentry name correctly
Message-ID: <20140307190136.GB30256@fieldses.org>
References: <1394115376-17109-1-git-send-email-zheng.z.yan@intel.com>
 <6487FCC2-FB5C-4A32-94EF-5D3352A39781@primarydata.com>
 <20140307185552.GA30256@fieldses.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
In-Reply-To: <20140307185552.GA30256@fieldses.org>
Sender: linux-nfs-owner@vger.kernel.org

On Fri, Mar 07, 2014 at 01:55:52PM -0500, J. Bruce Fields wrote:
> On Thu, Mar 06, 2014 at 09:35:01AM -0500, Trond Myklebust wrote:
> > 
> > On Mar 6, 2014, at 9:16, Yan, Zheng <zheng.z.yan@intel.com> wrote:
> > 
> > > From: David Howells <dhowells@redhat.com>
> > > 
> > > Make __d_materialise_dentry() set the materialised dentry name correctly by
> > > flipping the arguments to switch_names().
> > > 
> > > switch_names() is lazy: if both names are internal to their dentries, it'll
> > > overwrite that of the first dentry with that of the second, and won't update
> > > that of the second.
> > > 
> > > In the case of __d_materialise_dentry(), the second is an already extant
> > > anonymous dentry that we want to insert into the tree in place of the dentry we
> > > just looked up[*].  However, the dentry we just looked up carries the name we
> > > actually want to use.
> > > 
> > > [*] This is used by NFS to join a mount of a subtree into a mount of a tree
> > > nearer the root when the two meet, where both mounts share a superblock and
> > > thus a set of dentries.
> > > 
> > > Signed-off-by: David Howells <dhowells@redhat.com>
> > > ---
> > > fs/dcache.c | 2 +-
> > > 1 file changed, 1 insertion(+), 1 deletion(-)
> > > 
> > > diff --git a/fs/dcache.c b/fs/dcache.c
> > > index 265e0ce..ff779d4 100644
> > > --- a/fs/dcache.c
> > > +++ b/fs/dcache.c
> > > @@ -2698,7 +2698,7 @@ static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
> > > 
> > > 	dparent = dentry->d_parent;
> > > 
> > > -	switch_names(dentry, anon);
> > > +	switch_names(anon, dentry);
> > > 	swap(dentry->d_name.hash, anon->d_name.hash);
> > > 
> > > 	dentry->d_parent = dentry;
> > 
> > Well spotted...
> > 
> > We ought to better document the fact that ’switch_names’ is asymmetrical. Perhaps change it to ‘update_target_name’, and then switch the argument names so that the ’target’ really is the thing that gets updated?
> 
> Probably not worth it if Miklos's RENAME_EXCHANGE operation is making it
> symmetrical soon anyway?:
> 
> 	http://mid.gmane.org/<1380643239-16060-4-git-send-email-miklos@szeredi.hu>

To clarify: I mean, the suggested cleanup may not be worth it.  The
__d_materialise_dentry bugfix of course is.

(What I wonder about is the duplication between this and __d_move.  Is
there any way to make d_materialise_unique a special case of d_move?)

--b.