single-thread API: add usrsctp_get_timeout()

[ibc]

PR #339 makes usrsctp work in single thread environments by forcing the application call usrsctp_handle_timers() periodically. While it's good enough (it does work), the optimized way of doing it would be by having an API like DTLSv1_get_timeout() and DTLSv1_handle_timeout in OpenSSL and BoringSSL.

usrsctp already has usrsctp_handle_timers() so the only missing API would be usrsctp_get_timeout().

This is, the application provides an input to the lib (via usrsctp_socket(), usrsctp_close(), usrsctp_connect(), usrsctp_conninput(), etc) and retrieves from the lib (via usrsctp_get_timeout()) the time that must elapse before the lib (usrsctp) would need to do something (like sending a retransmission, a chunk or whatever). And then the app would set its own timer with that expiration value and call usrsctp_handle_timers() when it expires. This would be much more efficient that the current approach in which the app must call usrsctp_handle_timers() every N ms just in case the lib has something to do at that time.

As @tuexen said:

usrsctp_get_timeout() should give the app the number of milliseconds until the next timer runs off. It should walk the (unordered) list of timers to compute that value. It would even be more efficient if the timer list would be ordered, but then starting a timer would be more complicated.

IMPORTANT: It should be very clear and well documented when the app should call usrsctp_get_timeout(), this is: the app should call usrsctp_get_timeout() and prepare a new timer after calling which usrsctp functions and after which usrsctp events.

[milahu]

why was this closed?

fixed by #591

[ibc]

What do you mean? This issue was closed because it was done here: #339

[milahu]

#339 adds

static unsigned get_tick_count(void)
{
#ifdef _WIN32
	return GetTickCount();
#else
	struct timeval te;
	gettimeofday(&te, NULL); // get current time
	unsigned milliseconds = te.tv_sec*1000LL + te.tv_usec/1000; // calculate milliseconds
	return milliseconds;
#endif
}

but that looks very different from #591

int
usrsctp_get_timeout(void)
{
	return sctp_get_next_tick();
}

int
sctp_get_next_tick(void)
{
	int ret;
	sctp_os_timer_t *c;

	SCTP_TIMERQ_LOCK();
	c = TAILQ_FIRST(&SCTP_BASE_INFO(callqueue));
	if (c) {
		uint32_t min_delta = UINT32_MAX;
		while (c) {
			uint32_t delta = c->c_time - ticks;
			min_delta = (delta < min_delta) ? delta : min_delta;
			c = TAILQ_NEXT(c, tqe);
		}
		ret = min_delta;
	} else {
		ret = -1;
	}
	SCTP_TIMERQ_UNLOCK();

	return ret;
}

usrsctp_get_timeout is used in frida-core

[ibc]

@milahu do you mean that PR #591 (not merged and unfortunately obviously ignored) is not a replacement but an addition and improvement to make single thread behavior way better than before?