From 9d38459d08577245f170188ebde3219513f8b7a7 Mon Sep 17 00:00:00 2001
From: Avery Pennarun <apenwarr@gmail.com>
Date: Tue, 11 Jun 2013 00:40:26 -0400
Subject: settime() sooner after connecting to the network.

If we called priv_settime(0) because none of the NTP servers were valid,
that would prevent future calls to settime/adjtime until 8 successful
packets were received, which is a long time when your clock is totally
wrong.  (It's intended to make sure adjtime() isn't called too often or with
inaccurate values, when your clock is *mostly* right.)

Change it so this delay only kicks in after the first *real* settime() call.
Also add some log messages to make it obvious when this occurs.

The symptom of this bug is that write_synced() wouldn't be called for
several minutes after booting *if* no clock warping was required (eg. if
you had already run ntpd an then quit it, and are now starting it again).
---
 ntp.c |    7 ++++++-
 1 files changed, 6 insertions(+), 1 deletions(-)

diff --git a/ntp.c b/ntp.c
index 9f83f1a..4c33f7e 100644
--- a/ntp.c
+++ b/ntp.c
@@ -462,6 +462,8 @@ priv_adjtime(void)
 			    &peers[offset_cnt / 2]->addr->ss)->sin_addr.s_addr;
 		else
 			conf->status.refid = conf->status.refid4;
+	} else {
+		log_debug("priv_adjtime: offset_cnt == 0");
 	}
 
 	free(peers);
@@ -493,7 +495,10 @@ priv_settime(double offset)
 	struct ntp_peer *p;
 
 	imsg_compose(ibuf_main, IMSG_SETTIME, 0, 0, &offset, sizeof(offset));
-	conf->settime = 0;
+	if (offset && conf->settime) {
+		log_debug("disabling settime after offset=%fs", offset);
+		conf->settime = 0;
+	}
 
 	TAILQ_FOREACH(p, &conf->ntp_peers, entry) {
 		if (p->next)
-- 
1.7.9.dirty