Re: UBSAN: Undefined behaviour in linux-4.7.2/drivers/usb/core/devio.c:1713:25

From: Alan Stern
Date: Mon Aug 22 2016 - 17:06:30 EST


On Mon, 22 Aug 2016, BjÃrn Mork wrote:

> Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> writes:
>
> > On Mon, 22 Aug 2016, BjÃrn Mork wrote:
> >
> >> Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> writes:
> >>
> >> > On Sun, 21 Aug 2016, Jiri Slaby wrote:
> >> >
> >> >> Cc: proper lists.
> >> >>
> >> >> ep->desc.bInterval seems to be 0 here.
> >
> >> > As far as I can see, this isn't possible. The usb_parse_endpoint()
> >> > routine in drivers/usb/core/config.c is supposed to guarantee that
> >> > ep->desc.bInterval is never 0.
> >>
> >> That is if it is an ISO endpoint, right?
> >
> > I can't tell; the bug report doesn't say. However, ep->desc.bInterval
> > is ignored for bulk and control endpoints, so it must be either
> > isochronous or interrupt.
>
> So what if the endpoint is not isochronous or interrupt here?

Oops. You're right; for some reason I thought that code was executed
only if the endpoint type was known to be isochronous or interrupt.

> > USB uses two different encodings for endpoint intervals. The second
> > encoding above just gives the interval in frames; this is used for low-
> > and full-speed interrupt endpoints. The first encoding above is
> > exponential (it gives n where the actual interval is 2^(n-1) frames or
> > microframes); this is used for all isochronous endpoints and for
> > high-speed (or SuperSpeed etc.) interrupt endpoints.
>
> OK, I am still puzzled: Won't the code I quoted above do the shift for
> *any* uurb->type and endpoint type? There doesn't seem to be any test
> for isochronous or interrupt endpoint around it?

Yes, that's the reason for the UBSAN report. I'll write a patch to fix
it.

Alan Stern