[PATCH v2][BZ 16852] Do not clobber recvmmsg argument.

[Rich Felker]

On Wed, Apr 30, 2014 at 03:35:40PM +0200, Michael Kerrisk (man-pages) wrote:
> On Wed, Apr 30, 2014 at 1:17 PM, Michael Kerrisk (man-pages)
> <mtk.manpages@gmail.com> wrote:
> > Hello Ondřej,
> >
> > On 04/30/2014 11:07 AM, Ondřej Bílka wrote:
> >> On Wed, Apr 30, 2014 at 10:28:53AM +0200, Michael Kerrisk (man-pages) wrote:
> >>> On 04/28/2014 07:01 PM, David Miller wrote:
> >>>> From: Ondřej Bílka <neleai@seznam.cz>
> >>>> Date: Mon, 28 Apr 2014 18:04:20 +0200
> >>>>
> >>>>>    * sysdeps/unix/sysv/linux/recvmmsg.c (recvmmsg): Do not clobber
> >>>>>    timeout argument.
> >>>>
> >>>> It is extremely unfortunate if we've defined this argument as const,
> >>>> now you are making it so that the user has no mechanism to get the
> >>>> updated timeval other than to define their own syscall stubs.
> >>>>
> >>>> This is doubly unfortunately since there is absolutely no reason for
> >>>> us to have defined the interface different from what the kernel
> >>>> actually provides.
> >>>
> >>> Agreed. Surely, the right fix here is simply to remove the "const"
> >>> qualifier from the glibc API?
> >>>
> >> Now I am also for removing const. Initially I looked to documentation
> >> and this was undocumented feature so I was not sure about it.
> >
> > Yes, my mistake.
> 
> Sigh! Now I remember why that timeout behavior didn't get documented:
> 
> http://thread.gmane.org/gmane.linux.man/3477/focus=3500
> 
> The timeout argument is completely broken, AFAICT. I never did get a
> reply from Arnaldo (but, as you can see in the thread, others agreed
> tha there's a problem), and then I got distracted :-{.

Indeed, this looks like a big problem and I don't see any viable
solution. I think the documented behavior, for now, should be that the
behavior is unspecified unless the timeout is a null pointer.

Rich