svn commit: r316393 - head/sys/compat/linux

[Konstantin Belousov]

On Sun, Apr 02, 2017 at 09:23:41PM +0300, Chagin Dmitry wrote:
> On Mon, Apr 03, 2017 at 03:57:29AM +1000, Bruce Evans wrote:
> > On Sun, 2 Apr 2017, Chagin Dmitry wrote:
> > 
> > > On Sun, Apr 02, 2017 at 04:36:51PM +0300, Konstantin Belousov wrote:
> > >> On Sun, Apr 02, 2017 at 07:46:13AM +0000, Dmitry Chagin wrote:
> > >>> Author: dchagin
> > >>> Date: Sun Apr  2 07:46:13 2017
> > >>> New Revision: 316393
> > >>> URL: https://svnweb.freebsd.org/changeset/base/316393
> > >>>
> > >>> Log:
> > >>>   As noted by bde@ negative tv_sec values are not checked for overflow,
> > >>>   so overflow can still occur. Fix that. Also remove the extra check for
> > >>>   tv_sec size as under COMPAT_LINUX32 it is always true.
> > >>>
> > >>>   Pointed out by:	bde@
> > >>>
> > >>>   MFC after:	1 week
> > >>>
> > >>> Modified:
> > >>>   head/sys/compat/linux/linux_time.c
> > >>>
> > >>> Modified: head/sys/compat/linux/linux_time.c
> > >>> ==============================================================================
> > >>> --- head/sys/compat/linux/linux_time.c	Sun Apr  2 07:11:15 2017	(r316392)
> > >>> +++ head/sys/compat/linux/linux_time.c	Sun Apr  2 07:46:13 2017	(r316393)
> > >>> @@ -125,8 +125,7 @@ native_to_linux_timespec(struct l_timesp
> > >>>
> > >>>  	LIN_SDT_PROBE2(time, native_to_linux_timespec, entry, ltp, ntp);
> > >>>  #ifdef COMPAT_LINUX32
> > >>> -	if (ntp->tv_sec > INT_MAX &&
> > >>> -	    sizeof(ltp->tv_sec) != sizeof(ntp->tv_sec))
> > >>> +	if (ntp->tv_sec > INT_MAX || ntp->tv_sec < INT_MIN)
> > >> This line reads as only tv_sec == INT_MAX case results in non-EOVERFLOW
> > >> condition.
> > >>
> > >
> > > should I rewrite it like:
> > >
> > > 	if (ntp->tv_sec < INT_MIN || ntp->tv_sec > INT_MAX)
> > > ?
> > 
> > I don't see the problem.   Do you mean that the compiler might remove
> > this code because the check is tautologically false on 32-bit natives,
> > but warn too?
> 
> I mean kib@ sentence that line 'reads like', as I'm sure that the code
> is correct.
> This code does not compile on 32-bit as it under COMPAT_LINUX32.
I indeed mis-read the context where the checks are done, I thought that
the members checked are ints.

Please ignore me.
> 
> 
> > 
> > The sizeof() comparison is even easier to evaluate at compile time.
> > Perhaps it acted a hint to the compiler to not warn.
> > 
> > Bruce
> 
> --