Re: [PATCH] tmpfs: security xattr setting on inode creation

[Hugh Dickins <hughd@google.com>]

On Tue, 28 Feb 2012, Ware, Ryan R wrote:
> On Tue, Feb 28, 2012 at 7:46 AM, Andrew Morton <akpm@linux-foundation.org>wrote:
> > On Fri, 24 Feb 2012 19:19:22 -0800 (PST)
> > Hugh Dickins <hughd@google.com> wrote:
> >...
> > > +             if (!new_xattr->name) {
> > > +                     kfree(new_xattr);
> > > +                     return -ENOMEM;
> > > +             }
> > > +
> > > +             memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
> > > +                    XATTR_SECURITY_PREFIX_LEN);
> > > +             memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
> > > +                    xattr->name, len);
> > > +
> > > +             spin_lock(&info->lock);
> > > +             list_add(&new_xattr->list, &info->xattr_list);
> > > +             spin_unlock(&info->lock);
> > > +     }
> > > +
> > > +     return 0;
> > > +}
> >
> > So if there's a kmalloc failure partway through the array, we leave a
> > partially xattrified inode in place.
> >
> > Are we sure this is OK?
> >
> 
> I'm guessing Jarkko can clean that up a bit.  It wouldn't be a good idea to
> leave inaccurate data structures laying around during failure cases.

Andrew raises a good concern, but Jarkko got it just right and no
change is needed: any xattrs already allocated are properly linked
on info->xattr_list, then when security_inode_init_security() fails
(with an error other than EOPNOTSUPP) the failing inode is iput(),
which ends up in shmem_evict_inode(), which kfree()s those xattrs
(and their names) on info->xattr_list.

Hugh