Re: [PATCH v4 10/17] vfs: make vfs_create break delegations on parent directory

[Christian Brauner]

On Mon, Nov 03, 2025 at 07:30:57PM -0500, Jeff Layton wrote:
> On Tue, 2025-11-04 at 11:14 +1100, NeilBrown wrote:
> > On Mon, 03 Nov 2025, Jeff Layton wrote:
> > > In order to add directory delegation support, we need to break
> > > delegations on the parent whenever there is going to be a change in the
> > > directory.
> > > 
> > > Add a delegated_inode parameter to struct createdata. Most callers just
> > > leave that as a NULL pointer, but do_mknodat() is changed to wait for a
> > > delegation break if there is one.
> > > 
> > > Signed-off-by: Jeff Layton <jlayton@xxxxxxxxxx>
> > > ---
> > >  fs/namei.c         | 26 +++++++++++++++++---------
> > >  include/linux/fs.h |  2 +-
> > >  2 files changed, 18 insertions(+), 10 deletions(-)
> > > 
> > > diff --git a/fs/namei.c b/fs/namei.c
> > > index fdf4e78cd041de8c564b7d1d89a46ba2aaf79d53..e8973000a312fb05ebb63a0d9bd83b9a5f8f805d 100644
> > > --- a/fs/namei.c
> > > +++ b/fs/namei.c
> > > @@ -3487,6 +3487,9 @@ int vfs_create(struct createdata *args)
> > >  
> > >  	mode = vfs_prepare_mode(idmap, dir, mode, S_IALLUGO, S_IFREG);
> > >  	error = security_inode_create(dir, dentry, mode);
> > > +	if (error)
> > > +		return error;
> > > +	error = try_break_deleg(dir, args->delegated_inode);
> > >  	if (error)
> > >  		return error;
> > >  	error = dir->i_op->create(idmap, dir, dentry, mode, args->excl);
> > > @@ -4359,6 +4362,8 @@ static int may_mknod(umode_t mode)
> > >  static int do_mknodat(int dfd, struct filename *name, umode_t mode,
> > >  		unsigned int dev)
> > >  {
> > > +	struct delegated_inode delegated_inode = { };
> > > +	struct createdata cargs = { };
> > 
> > If we must have 'createdata', can it have a 'struct delegated_inode'
> > rather than a pointer to it?
> > 
> 
> If we do that, then we'd need some way to signal that the caller
> doesn't want to wait on the delegation break. Currently that's
> indicated by setting cargs.delegated_inode to NULL. I suppose we could
> add a bool for this or something.
> 
> I confess that I too am lukewarm on struct createdata. I can live with
> it, but it's not clearly a win to me either.
> 
> Christian, thoughts?

If two stable voices of the community seem to have consensus that this
isn't worth it then it's obviously fine to not do it.

> 
> > 
> > >  	struct mnt_idmap *idmap;
> > >  	struct dentry *dentry;
> > >  	struct path path;
> > > @@ -4383,18 +4388,16 @@ static int do_mknodat(int dfd, struct filename *name, umode_t mode,
> > >  	switch (mode & S_IFMT) {
> > >  		case 0:
> > >  		case S_IFREG:
> > > -		{
> > > -			struct createdata args = { .idmap = idmap,
> > > -						   .dir = path.dentry->d_inode,
> > > -						   .dentry = dentry,
> > > -						   .mode = mode,
> > > -						   .excl = true };
> > > -
> > > -			error = vfs_create(&args);
> > > +			cargs.idmap = idmap,
> > > +			cargs.dir = path.dentry->d_inode,
> > > +			cargs.dentry = dentry,
> > > +			cargs.delegated_inode = &delegated_inode;
> > > +			cargs.mode = mode,
> > > +			cargs.excl = true,
> > > +			error = vfs_create(&cargs);
> > >  			if (!error)
> > >  				security_path_post_mknod(idmap, dentry);
> > >  			break;
> > > -		}
> > >  		case S_IFCHR: case S_IFBLK:
> > >  			error = vfs_mknod(idmap, path.dentry->d_inode,
> > >  					  dentry, mode, new_decode_dev(dev));
> > > @@ -4406,6 +4409,11 @@ static int do_mknodat(int dfd, struct filename *name, umode_t mode,
> > >  	}
> > >  out2:
> > >  	end_creating_path(&path, dentry);
> > > +	if (is_delegated(&delegated_inode)) {
> > > +		error = break_deleg_wait(&delegated_inode);
> > > +		if (!error)
> > > +			goto retry;
> > > +	}
> > >  	if (retry_estale(error, lookup_flags)) {
> > >  		lookup_flags |= LOOKUP_REVAL;
> > >  		goto retry;
> > > diff --git a/include/linux/fs.h b/include/linux/fs.h
> > > index b61873767b37591aecadd147623d7dfc866bef82..cfcb20a7c4ce4b6dcec98b3eccbdb5ec8bab6fa9 100644
> > > --- a/include/linux/fs.h
> > > +++ b/include/linux/fs.h
> > > @@ -2116,12 +2116,12 @@ struct createdata {
> > >  	struct mnt_idmap *idmap;	// idmap of the mount the inode was found from
> > >  	struct inode *dir;		// inode of parent directory
> > >  	struct dentry *dentry;		// dentry of the child file
> > > +	struct delegated_inode *delegated_inode; // returns parent inode, if delegated
> > >  	umode_t mode;			// mode of the child file
> > >  	bool excl;			// whether the file must not yet exist
> > >  };
> > >  
> > >  int vfs_create(struct createdata *);
> > > -
> > >  struct dentry *vfs_mkdir(struct mnt_idmap *, struct inode *,
> > >  			 struct dentry *, umode_t, struct delegated_inode *);
> > >  int vfs_mknod(struct mnt_idmap *, struct inode *, struct dentry *,
> > > 
> > > -- 
> > > 2.51.1
> > > 
> > > 
> 
> -- 
> Jeff Layton <jlayton@xxxxxxxxxx>