On Tue, Apr 30 2019, Jeremie Courreges-Anglas <j...@wxcvbn.org> wrote:
> libevent contains a fallback path in case clock_gettime(CLOCK_MONOTONIC)
> fails. The fallback path tries to cope with time going backwards and
> reaches into the timeheap internals, as noticed by Tobias in
>
> https://marc.info/?l=openbsd-tech&m=155595247719664&w=2
>
> I doubt that we care about this kind of seldom-tested compat code, which
> was probably useful to get portable libevent to run on various
> (ancient?) systems.
>
> Our clock_gettime can only fail for two reasons:
> - invalid clock id. Lots of our userland uses CLOCK_MONOTONIC now,
> I think it can be considered a requirement.
> - invalid timespec pointer. Here, since we pass the address of
> a timespec structure on the stack something must be really wrong if we
> get EFAULT.
>
> So the diff below removes the fallback path and all associated code and
> variables.
> I have left out some minor cleanups for now to ease reviews.
Here's a diff that amends the signature of gettime() and makes use of
TIMESPEC_TO_TIMEVAL().
Index: event.c
===================================================================
--- event.c.orig
+++ event.c
@@ -77,23 +77,20 @@ static void event_process_active(struct
static int timeout_next(struct event_base *, struct timeval **);
static void timeout_process(struct event_base *);
-static int
+static void
gettime(struct event_base *base, struct timeval *tp)
{
struct timespec ts;
if (base->tv_cache.tv_sec) {
*tp = base->tv_cache;
- return (0);
+ return;
}
if (clock_gettime(CLOCK_MONOTONIC, &ts) == -1)
event_err(1, "libevent: clock_gettime failed");
- tp->tv_sec = ts.tv_sec;
- tp->tv_usec = ts.tv_nsec / 1000;
-
- return (0);
+ TIMESPEC_TO_TIMEVAL(tp, &ts);
}
struct event_base *
@@ -803,8 +800,7 @@ timeout_next(struct event_base *base, st
return (0);
}
- if (gettime(base, &now) == -1)
- return (-1);
+ gettime(base, &now);
if (timercmp(&ev->ev_timeout, &now, <=)) {
timerclear(tv);
--
jca | PGP : 0x1524E7EE / 5135 92C1 AD36 5293 2BDF DDCC 0DFA 74AE 1524 E7EE
