RE: [PATCH] mei: bus: type promotion bug in mei_nfc_if_version()

[Julia Lawall <julia.lawall@lip6.fr>]

On Wed, 4 Jul 2018, Winkler, Tomas wrote:

>
> > On Wed, Jul 04, 2018 at 01:57:44PM +0000, Winkler, Tomas wrote:
> > > >
> > > > On Wed, Jul 04, 2018 at 01:59:14PM +0200, Julia Lawall wrote:
> > > > >
> > > > >
> > > > > On Wed, 4 Jul 2018, Dan Carpenter wrote:
> > > > >
> > > > > > We accidentally removed the check for negative returns without
> > > > > > considering the issue of type promotion.  The "if_version_length"
> > > > > > variable is type size_t so if __mei_cl_recv() returns a negative
> > > > > > then "bytes_recv" is type promoted to a high positive value and
> > > > > > treated as success.
> > > > > >
> > > > > > Fixes: 582ab27a063a ("mei: bus: fix received data size check in
> > > > > > NFC
> > > > > > fixup")
> > > > > > Signed-off-by: Dan Carpenter <dan.carpenter@oracle.com>
> > > > > >
> > > > > > diff --git a/drivers/misc/mei/bus-fixup.c
> > > > > > b/drivers/misc/mei/bus-fixup.c index 0208c4b027c5..fa0236a5e59a
> > > > > > 100644
> > > > > > --- a/drivers/misc/mei/bus-fixup.c
> > > > > > +++ b/drivers/misc/mei/bus-fixup.c
> > > > > > @@ -267,7 +267,7 @@ static int mei_nfc_if_version(struct mei_cl
> > > > > > *cl,
> > > > > >
> > > > > >  	ret = 0;
> > > > > >  	bytes_recv = __mei_cl_recv(cl, (u8 *)reply, if_version_length,
> > 0);
> > > > > > -	if (bytes_recv < if_version_length) {
> > > > > > +	if (bytes_recv < 0 || bytes_recv < if_version_length) {
> > > > >
> > > > > Is this preferred to adding an int cast?
> > > >
> > > > I don't think it matters.  I kind of like explicitly testing for
> > > > negative but maybe later people will just remove the check like we
> > > > did here?  You could do it a bunch of different ways:
> > > >
> > > > 1: if (ret < 0 || ret < ARRAY_SIZE(xxx))
> > > > 2: if (ret < (int)ARRAY_SIZE(xxx))
> > > > 3: if (ret != ARRAY_SIZE(xxx))
> > > >
> > > > They're all equivalent.  I guess I don't like casting too much.  My
> > > > first approach to fixing this was just to declare if_version_length
> > > > as an int, but then I saw that originally there was a "bytes_recv < 0"
> > > > check and decided to go that way instead.
> > >
> > > Actually bytes_recv should be probably of ssize_t type,  so could be the
> > if_version_length.
> > >
> > > How did you find this, I haven't seen it in reported by sparse, smatch and I
> > believe -Wsign-compare is suppressed in compilation warnings.
> >
> > It's a new thing.  Julia noticed this kind of bug first

Actually, I got it from Joe Perches, for the sizeof case :)

julia

> > and I have been mucking
> > around with it in Smatch as well.  My Smatch check has too many false
> > positives to publish right now because it thinks a some common functions
> > like ffs() return negative error codes.
>
> I guess this is why it is suppressed in the compilation warnings  in the first place.
> Maybe need to disable it selectively, like for fss, just not sure how bad is that with false positive reports.
>
>
> Thanks
> Tomas
>
>