Re: [PATCH v2] USB: EHCI: fix dereference of ERR_PTR

[Alan Stern]

On Fri, 18 Sep 2015, Sudip Mukherjee wrote:

> On Wed, Sep 16, 2015 at 12:54:03PM -0400, Alan Stern wrote:
> > On Wed, 16 Sep 2015, Sudip Mukherjee wrote:
> > 
> > > On error find_tt() returns either a NULL pointer or the error value in
> > > ERR_PTR. But we were dereferencing it directly without even checking if
> > > find_tt() returned a valid pointer or not.
> > > 
> > > Signed-off-by: Sudip Mukherjee <sudip@xxxxxxxxxxxxxxx>
> > > ---
> <snip>
> > > @@ -1373,6 +1375,8 @@ static void reserve_release_iso_bandwidth(struct ehci_hcd *ehci,
> > >  		}
> > >  
> > >  		tt = find_tt(stream->ps.udev);
> > > +		if (IS_ERR_OR_NULL(tt))
> > > +			return;
> > >  		if (sign > 0)
> > >  			list_add_tail(&stream->ps.ps_list, &tt->ps_list);
> > >  		else
> > 
> > This patch isn't needed.  In both reserve_release_intr_bandwidth() and 
> > reserve_release_iso_bandwidth() it is known that find_tt() will return 
> > a valid pointer.
> > 
> > This is because each of those functions is called from only one place.  
> > For example, reserve_release_intr_bandwidth() is called only at the end
> > of qh_schedule().  But near the start of qh_schedule() there is earlier
> > call to tt_find(), and there we do test for error pointers.  If the
> > first call doesn't return an error then the second call won't either.
> > 
> > The same sort of thing happens in reserve_release_iso_bandwidth().
> Yes, I should have looked more before sending. Sorry for the noise.
> But in those checkes for find_tt() only IS_ERR is checked, shouldn't we
> check for IS_ERR_OR_NULL as find_tt() can return NULL also?

I knew someone would ask about that!  :-)

The NULL case is similar to the ERR_PTR case, but more complicated.  
Basically, find_tt() returns NULL only when the device doesn't lie 
below a TT -- all the other invalid returns are ERR_PTRs.

In reserve_release_intr_bandwidth(), for instance, the call to 
find_tt() occurs only if tt_usecs = qh->ps.tt_usecs is nonzero.  This 
value is guaranteed to be 0 if the device doesn't run at low speed or 
full speed -- see qh_make() in ehci-q.c, where qh->ps.tt_usecs is 
initialized only when urb->dev->speed != USB_SPEED_HIGH.

To see that udev->tt is non-NULL whenever the speed isn't 
USB_SPEED_HIGH, you have to look through the hub driver code.  The 
relevant routine is hub_port_init() in core/hub.c, the section headed 
by the comment:

	/* Set up TT records, if needed  */

Similar reasoning applies to reserve_release_iso_bandwidth(); here the 
condition is that stream->splits is nonzero, which is true only if the 
device is full speed (see iso_stream_init()).

Alan Stern

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/