[PATCH 4/6] fuse: return correct dentry for ->mkdir

[NeilBrown]

On Sat, 22 Feb 2025, Al Viro wrote:
> On Fri, Feb 21, 2025 at 10:36:33AM +1100, NeilBrown wrote:
> 
> > @@ -871,7 +870,12 @@ static int fuse_mknod(struct mnt_idmap *idmap, struct inode *dir,
> >  	args.in_args[0].value = &inarg;
> >  	args.in_args[1].size = entry->d_name.len + 1;
> >  	args.in_args[1].value = entry->d_name.name;
> > -	return create_new_entry(idmap, fm, &args, dir, entry, mode);
> > +	de = create_new_entry(idmap, fm, &args, dir, entry, mode);
> > +	if (IS_ERR(de))
> > +		return PTR_ERR(de);
> > +	if (de)
> > +		dput(de);
> > +	return 0;
> 
> Can that really happen?

Probably now.  It would require S_IFDIR to be passed in the mode to
vfs_mknod().  I don't think any current callers do that, but I don't see
any code in vfs_mknod() to prevent it.

> 
> > @@ -934,7 +939,12 @@ static int fuse_symlink(struct mnt_idmap *idmap, struct inode *dir,
> >  	args.in_args[1].value = entry->d_name.name;
> >  	args.in_args[2].size = len;
> >  	args.in_args[2].value = link;
> > -	return create_new_entry(idmap, fm, &args, dir, entry, S_IFLNK);
> > +	de = create_new_entry(idmap, fm, &args, dir, entry, S_IFLNK);
> > +	if (IS_ERR(de))
> > +		return PTR_ERR(de);
> > +	if (de)
> > +		dput(de);
> > +	return 0;
> 
> Same question.

That definitely cannot happen. - because we *know* that d_splice_alias()
never returns a dentry for any but an S_IFDIR inode (how might we
explain that to the rust type system I wonder :-).

I was going for "obviously correct" without try to optimise, but you are
correct that testing for a non-NULL non-ERR dentry should be optimsed
away as impossible in all cases except mkdir.

Thanks,
NeilBrown


> 
> > +	de = create_new_entry(&invalid_mnt_idmap, fm, &args, newdir, newent, inode->i_mode);
> > +	if (!IS_ERR(de)) {
> > +		if (de)
> > +			dput(de);
> > +		de = NULL;
> 
> Whoa...  Details, please.  What's going on here?
>