Re: [PATCH 3/3] Enable support for tmpfs quotas

[Jan Kara <jack@suse.cz>]

On Wed 24-01-24 12:42:39, Carlos Maiolino wrote:
> On Wed, Jan 17, 2024 at 06:59:54PM +0100, Jan Kara wrote:
> > On Tue 09-01-24 14:46:05, cem@kernel.org wrote:
> > > diff --git a/quotasys.c b/quotasys.c
> > > index 903816b..1f66302 100644
> > > --- a/quotasys.c
> > > +++ b/quotasys.c
> > > @@ -1384,7 +1390,11 @@ alloc:
> > >  			continue;
> > >  		}
> > >
> > > -		if (!nfs_fstype(mnt->mnt_type)) {
> > > +		/*
> > > +		 * If devname and mnt->mnt_fsname matches, there is no real
> > > +		 * underlyin device, so skip these checks
> > > +		 */
> > > +		if (!nfs_fstype(mnt->mnt_type) && strcmp(devname, mnt->mnt_fsname)) {
> > >  			if (stat(devname, &st) < 0) {	/* Can't stat mounted device? */
> > >  				errstr(_("Cannot stat() mounted device %s: %s\n"), devname, strerror(errno));
> > >  				free((char *)devname);
> > 
> > I'm a bit uneasy about the added check because using device name the same
> > as filesystem name is just a common agreement but not enforced in any way.
> > So perhaps just add an explicit check for tmpfs? Later we can generalize
> > this if there are more filesystems like this...
> 
> What about adding a new tmpfs_fstype() helper, to mimic nfs_fstype, and use it
> here? like:
> 
> if (!nfs_fstype(mnt->mnt_type) && tmpfs_fstype(mnt->mnt_type))) {
> 	/* skipe S_ISBLK && S_ISCHR checks */
> }
> 
> We could open code !strcmp(mnt->mnt_type, MNTTYPE_TMPFS), but it seems to me
> adding a new tmpfs_fstype() helper is easier on the eyes, and also OCFS2 does
> something similar.
> 
> Perhaps that's exactly what you meant as having an explicit check for tmpfs, but
> I'm not really sure?!

Yes, this would be a nice way of doing it.

								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR