On Thu, 4 Feb 2021 09:45:37 +0000
Stefan Hajnoczi <stefanha@redhat.com> wrote:

> On Thu, Feb 04, 2021 at 09:25:28AM +0100, Greg Kurz wrote:
> > On Wed, 3 Feb 2021 17:00:06 +0000
> > Stefan Hajnoczi <stefanha@redhat.com> wrote:
> > 
> > > On Wed, Feb 03, 2021 at 03:20:14PM +0100, Greg Kurz wrote:
> > > > On Wed,  3 Feb 2021 11:37:18 +0000
> > > > Stefan Hajnoczi <stefanha@redhat.com> wrote:
> > > > 
> > > > > lo_do_lookup() finds an existing inode or allocates a new one. It
> > > > > increments nlookup so that the inode stays alive until the client
> > > > > releases it.
> > > > > 
> > > > > Existing callers don't need the struct lo_inode so the function doesn't
> > > > > return it. Extend the function to optionally return the inode. The next
> > > > > commit will need it.
> > > > > 
> > > > > Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
> > > > > ---
> > > > >  tools/virtiofsd/passthrough_ll.c | 29 +++++++++++++++++++++--------
> > > > >  1 file changed, 21 insertions(+), 8 deletions(-)
> > > > > 
> > > > > diff --git a/tools/virtiofsd/passthrough_ll.c b/tools/virtiofsd/passthrough_ll.c
> > > > > index e63cbd3fb7..c87a1f3d72 100644
> > > > > --- a/tools/virtiofsd/passthrough_ll.c
> > > > > +++ b/tools/virtiofsd/passthrough_ll.c
> > > > > @@ -831,11 +831,13 @@ static int do_statx(struct lo_data *lo, int dirfd, const char *pathname,
> > > > >  }
> > > > >  
> > > > >  /*
> > > > > - * Increments nlookup and caller must release refcount using
> > > > > - * lo_inode_put(&parent).
> > > > > + * Increments nlookup on the inode on success. unref_inode_lolocked() must be
> > > > > + * called eventually to decrement nlookup again. If inodep is non-NULL, the
> > > > > + * inode pointer is stored and the caller must call lo_inode_put().
> > > > >   */
> > > > >  static int lo_do_lookup(fuse_req_t req, fuse_ino_t parent, const char *name,
> > > > > -                        struct fuse_entry_param *e)
> > > > > +                        struct fuse_entry_param *e,
> > > > > +                        struct lo_inode **inodep)
> > > > >  {
> > > > >      int newfd;
> > > > >      int res;
> > > > > @@ -845,6 +847,10 @@ static int lo_do_lookup(fuse_req_t req, fuse_ino_t parent, const char *name,
> > > > >      struct lo_inode *inode = NULL;
> > > > >      struct lo_inode *dir = lo_inode(req, parent);
> > > > >  
> > > > > +    if (inodep) {
> > > > > +        *inodep = NULL;
> > > > > +    }
> > > > > +
> > > > 
> > > > Is this side-effect needed ? If lo_do_lookup() returns an error, it
> > > > rather seems that the caller shouldn't expect anything to be written
> > > > here, i.e. the content of *inodep still belongs to the caller and
> > > > whatever value it previously put in there (as patch 3/3 does) should
> > > > be preserved IMHO.
> > > > 
> > > > Apart from that LGTM.
> > > 
> > > I like this approach because it prevents accessing uninitialized memory
> > > in the caller:
> > > 
> > >   struct lo_inode *inode;
> > > 
> > >   if (lo_do_lookup(..., &inodep) != 0) {
> > >     goto err;
> > >   }
> > >   ...
> > > 
> > >   err:
> > >   lo_inode_put(&inode); <-- uninitialized in the error case!
> > 
> > My point is that it is the caller's business to ensure that inode
> > doesn't contain garbage if it is to be used irrespective of the
> > outcome of lo_do_lookup(). This is precisely what patch 3/3 does,
> > so I don't understand the ultimate purpose of nullifying the
> > inode pointer _again_ in lo_do_lookup()...
> 
> APIs should be designed to eliminate classes of errors where possible
> IMO. Taking care regarding the uninitialized pointer in the error case
> could be the caller's responsibility, but what's the advantage?
> 

Because it is more explicit. FWIW caller is still responsible since it
calls lo_inode_put() in the end : initializing inode to NULL like patch
3/3 does warrants that no matter what happens in lo_do_lookup() or even
if it isn't called at all, inode can be safely passed to lo_inode_put().

But this change doesn't hurt, especially with the benefits of the rest
of this series, so:

Reviewed-by: Greg Kurz <groug@kaod.org>

> (There's a related thing with lo_inode_put(&inode) where it sets *inode
> = NULL to eliminate use-after-free bugs in callers. It would have been
> possible to use the same approach as free(3) where it's the caller's
> responsiblity, but that API design decision in free(3) has caused
> many bugs in applications.)
> 
> Stefan