On Fri, Jun 11, 2021 at 12:17:02PM -0500, Scott Cheloha wrote:
> Hi,
>
> setitimer(2) has a one hundred million second upper bound for timers.
> Any timer interval larger than this is considered invalid and we set
> EINVAL.
>
> There is no longer any reason to use this particular limit. Kclock
> timeouts support the full range of a timespec, so we can trivially
> increase the upper bound without any practical risk of overflow.
>
> This patch increases the upper bound to UINT_MAX seconds.
>
> Why UINT_MAX? UINT_MAX is the largest possible input to alarm(3). We
> could then simplify the alarm(3) manpage and the libc alarm.c code in
> a subsequent patch. POSIX says alarm(3) "is always successful". Our
> implementation can fail. It would be nicer/simpler if ours were free
> of failure modes.
>
> ok?
1 week bump.
Updated patch: make the maximum value ("max") static and const.
Index: kern_time.c
===================================================================
RCS file: /cvs/src/sys/kern/kern_time.c,v
retrieving revision 1.153
diff -u -p -r1.153 kern_time.c
--- kern_time.c 11 Jun 2021 16:36:34 -0000 1.153
+++ kern_time.c 18 Jun 2021 01:40:42 -0000
@@ -709,15 +709,16 @@ out:
int
itimerfix(struct itimerval *itv)
{
+ static const struct timeval max = { .tv_sec = UINT_MAX, .tv_usec = 0 };
struct timeval min_interval = { .tv_sec = 0, .tv_usec = tick };
if (itv->it_value.tv_sec < 0 || !timerisvalid(&itv->it_value))
return EINVAL;
- if (itv->it_value.tv_sec > 100000000)
+ if (timercmp(&itv->it_value, &max, >))
return EINVAL;
if (itv->it_interval.tv_sec < 0 || !timerisvalid(&itv->it_interval))
return EINVAL;
- if (itv->it_interval.tv_sec > 100000000)
+ if (timercmp(&itv->it_interval, &max, >))
return EINVAL;
if (!timerisset(&itv->it_value))
