If a bind() spuriously fails (as appears to happen once in a blue moon due

to some stupid ipv6 bug in particular), remove that 'listen' from the list
and continue operation.  issue spotted by naddy
ok henning

1 files changed, 17 insertions, 6 deletions

diff --git a/usr.sbin/ntpd/server.c b/usr.sbin/ntpd/server.c
index 6156d522914..56fe4d5522b 100644
--- a/usr.sbin/ntpd/server.c
+++ b/usr.sbin/ntpd/server.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: server.c,v 1.32 2008/06/10 03:46:09 naddy Exp $ */
+/*	$OpenBSD: server.c,v 1.33 2008/11/10 16:59:33 deraadt Exp $ */
 
 /*
  * Copyright (c) 2003, 2004 Henning Brauer <henning@openbsd.org>
@@ -75,9 +75,7 @@ setup_listeners(struct servent *se, struct ntpd_conf *lconf, u_int *cnt)
 		freeifaddrs(ifa);
 	}
 
-	TAILQ_FOREACH(la, &lconf->listen_addrs, entry) {
-		new_cnt++;
-
+	for (la = TAILQ_FIRST(&lconf->listen_addrs); la; ) {
 		switch (la->sa.ss_family) {
 		case AF_INET:
 			if (((struct sockaddr_in *)&la->sa)->sin_port == 0)
@@ -104,8 +102,21 @@ setup_listeners(struct servent *se, struct ntpd_conf *lconf, u_int *cnt)
 			log_warn("setsockopt IPTOS_LOWDELAY");
 
 		if (bind(la->fd, (struct sockaddr *)&la->sa,
-		    SA_LEN((struct sockaddr *)&la->sa)) == -1)
-			fatal("bind");
+		    SA_LEN((struct sockaddr *)&la->sa)) == -1) {
+			struct listen_addr	*nla;
+
+			log_warn("bind on %s failed, skipping",
+			    log_sockaddr((struct sockaddr *)&la->sa));
+			close(la->fd);
+			la->fd = -1;
+			nla = TAILQ_NEXT(la, entry);
+			TAILQ_REMOVE(&lconf->listen_addrs, la, entry);
+			free(la);
+			la = nla;
+			continue;
+		}
+		new_cnt++;
+		la = TAILQ_NEXT(la, entry);
 	}
 
 	*cnt = new_cnt;