[patch] statfs does not detect -t nullfs -o union as a union mount

[Konstantin Belousov]

On Sat, Mar 02, 2013 at 01:00:33AM +0100, Jilles Tjoelker wrote:
> On Thu, Feb 28, 2013 at 12:21:44AM +0200, Konstantin Belousov wrote:
> > On Wed, Feb 27, 2013 at 10:31:42PM +0100, Jilles Tjoelker wrote:
> > > While testing recent changes to opendir(), I noticed that fstatfs() does
> > > not return the MNT_UNION flag for a -t nullfs -o union mount. As a
> > > result, opendir()/readdir() return files that exist in both top and
> > > bottom directories twice (at least . and ..). Other -o union mounts and
> > > -t unionfs mounts work correctly in this regard.
> 
> > > The below patch passes through just the MNT_UNION flag of the nullfs
> > > mount itself. Perhaps more flags should be passed through.
> 
> > > commit fce32a779af4eb977c9b96feb6e4f811d89f2881
> > > Author: Jilles Tjoelker <jilles at stack.nl>
> > > Date:   Sat Feb 23 22:22:39 2013 +0100
> > > 
> > >     nullfs: Preserve the MNT_UNION flag of the nullfs mount itself.
> > >     
> > >     This is needed so that opendir() can properly detect a union mount like
> > >     mount -t nullfs -o union dir1 dir2.
> > > 
> > > diff --git a/sys/fs/nullfs/null_vfsops.c b/sys/fs/nullfs/null_vfsops.c
> > > index 3724e0a..ff06f57 100644
> > > --- a/sys/fs/nullfs/null_vfsops.c
> > > +++ b/sys/fs/nullfs/null_vfsops.c
> > > @@ -313,7 +313,7 @@ nullfs_statfs(mp, sbp)
> > >  
> > >  	/* now copy across the "interesting" information and fake the rest */
> > >  	sbp->f_type = mstat.f_type;
> > > -	sbp->f_flags = mstat.f_flags;
> > > +	sbp->f_flags = (sbp->f_flags & MNT_UNION) | mstat.f_flags;
> > >  	sbp->f_bsize = mstat.f_bsize;
> > >  	sbp->f_iosize = mstat.f_iosize;
> > >  	sbp->f_blocks = mstat.f_blocks;
> 
> > Would it make sense to preserve more flags from the upper mount ?
> > I see a use for MNT_NOEXEC as well, at least.
> 
> Yes, preserving MNT_NOEXEC will make -t nullfs -o noexec work better, in
> particular rtld's check for libraries loaded via environment variables.
> 
> In the same way MNT_RDONLY, MNT_NOSUID and MNT_NOSYMFOLLOW could be
> preserved.
> 
> On the other hand, MNT_ROOTFS should probably not be passed through from
> the underlying filesystem.
> 
> This would give
> sbp->f_flags = (sbp->f_flags & (MNT_RDONLY | MNT_NOEXEC | MNT_NOSUID |
>     MNT_UNION | MNT_NOSYMFOLLOW) | (mstat.f_flags & ~MNT_ROOTFS);
I think this is fine.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 834 bytes
Desc: not available
URL: <http://lists.freebsd.org/pipermail/freebsd-hackers/attachments/20130302/a342302c/attachment.sig>