failover to next server.

git-svn-id: file:///svn/unbound/trunk@691 be551aaa-1e26-0410-a405-d3ace91eadb9

doc/Changelog

@@ -17,6 +17,7 @@
 	- fixup removal of nonsecure items from the additional.
 	- reduced timeout values to more realistic, 376 msec (262 msec has
 	  90% of roundtrip times, 512 msec has 99% of roundtrip times.)
+	- server selection failover to next server after timeout (376 msec).
 
 16 October 2007: Wouter
 	- no malloc in log_hex.

iterator/iter_delegpt.c

@@ -152,6 +152,7 @@ delegpt_add_addr(struct delegpt* dp, struct region* region,
 	dp->usable_list = a;
 	memcpy(&a->addr, addr, addrlen);
 	a->addrlen = addrlen;
+	a->attempts = 0;
 	return 1;
 }
 

iterator/iter_delegpt.h

@@ -104,6 +104,8 @@ struct delegpt_addr {
 	struct sockaddr_storage addr;
 	/** length of addr */
 	socklen_t addrlen;
+	/** number of attempts for this addr */
+	int attempts;
 };
 
 /**

iterator/iter_utils.c

@@ -214,6 +214,8 @@ iter_server_selection(struct iter_env* iter_env,
 		return NULL;
 	if(num == 1) {
 		a = dp->result_list;
+		if(++a->attempts < OUTBOUND_MSG_RETRY)
+			return a;
 		dp->result_list = a->next_result;
 		return a;
 	}
@@ -231,6 +233,8 @@ iter_server_selection(struct iter_env* iter_env,
 	}
 	if(!a)  /* robustness */
 		return NULL;
+	if(++a->attempts < OUTBOUND_MSG_RETRY)
+		return a;
 	/* remove it from the delegation point result list */
 	if(prev)
 		prev->next_result = a->next_result;

iterator/iterator.h

@@ -64,6 +64,8 @@ struct iter_prep_list;
  * Equals RTT_MAX_TIMEOUT
  */
 #define USEFUL_SERVER_TOP_TIMEOUT	120000
+/** number of retries on outgoing queries */
+#define OUTBOUND_MSG_RETRY 4
 
 /**
  * Global state for the iterator. 

services/cache/infra.c

@@ -401,6 +401,23 @@ infra_set_lame(struct infra_cache* infra,
 	return 1;
 }
 
+void 
+infra_update_tcp_works(struct infra_cache* infra,
+        struct sockaddr_storage* addr, socklen_t addrlen)
+{
+	struct lruhash_entry* e = infra_lookup_host_nottl(infra, addr, 
+		addrlen, 1);
+	struct infra_host_data* data;
+	if(!e)
+		return; /* doesn't exist */
+	data = (struct infra_host_data*)e->data;
+	if(data->rtt.rto >= RTT_MAX_TIMEOUT)
+		/* do not disqualify this server altogether, it is better
+		 * than nothing */
+		data->rtt.rto = RTT_MAX_TIMEOUT-1;
+	lock_rw_unlock(&e->lock);
+}
+
 int 
 infra_rtt_update(struct infra_cache* infra,
         struct sockaddr_storage* addr, socklen_t addrlen,
@@ -410,6 +427,7 @@ infra_rtt_update(struct infra_cache* infra,
 		addrlen, 1);
 	struct infra_host_data* data;
 	int needtoinsert = 0;
+	int rto = 1;
 	if(!e) {
 		if(!(e = new_host_entry(infra, addr, addrlen, timenow)))
 			return 0;
@@ -421,11 +439,13 @@ infra_rtt_update(struct infra_cache* infra,
 	if(roundtrip == -1)
 		rtt_lost(&data->rtt);
 	else	rtt_update(&data->rtt, roundtrip);
+	if(data->rtt.rto > 0)
+		rto = data->rtt.rto;
 
 	if(needtoinsert)
 		slabhash_insert(infra->hosts, e->hash, e, e->data, NULL);
 	else 	{ lock_rw_unlock(&e->lock); }
-	return 1;
+	return rto;
 }
 
 int 

services/cache/infra.h

@@ -199,12 +199,21 @@ int infra_set_lame(struct infra_cache* infra,
  * @param roundtrip: estimate of roundtrip time in milliseconds or -1 for 
  * 	timeout.
  * @param timenow: what time it is now.
- * @return: 0 on error.
+ * @return: 0 on error. new rto otherwise.
  */
 int infra_rtt_update(struct infra_cache* infra,
         struct sockaddr_storage* addr, socklen_t addrlen,
 	int roundtrip, time_t timenow);
 
+/**
+ * Update information for the host, store that a TCP transaction works.
+ * @param infra: infrastructure cache.
+ * @param addr: host address.
+ * @param addrlen: length of addr.
+ */
+void infra_update_tcp_works(struct infra_cache* infra,
+        struct sockaddr_storage* addr, socklen_t addrlen);
+
 /**
  * Update edns information for the host.
  * @param infra: infrastructure cache.

services/outside_network.c

@@ -61,7 +61,7 @@
 /** number of times to retry making a random ID that is unique. */
 #define MAX_ID_RETRY 1000
 /** number of retries on outgoing UDP queries */
-#define OUTBOUND_UDP_RETRY 4
+#define OUTBOUND_UDP_RETRY 1
 
 /** initiate TCP transaction for serviced query */
 static void serviced_tcp_initiate(struct outside_network* outnet, 
@@ -1021,6 +1021,9 @@ serviced_tcp_callback(struct comm_point* c, void* arg, int error,
 	sq->pending = NULL; /* removed after this callback */
 	if(error != NETEVENT_NOERROR && verbosity >= VERB_DETAIL)
 		log_addr("tcp error for address", &sq->addr, sq->addrlen);
+	if(error==NETEVENT_NOERROR)
+		infra_update_tcp_works(sq->outnet->infra, &sq->addr,
+			sq->addrlen);
 	if(error==NETEVENT_NOERROR && LDNS_RCODE_WIRE(ldns_buffer_begin(
 		c->buffer)) == LDNS_RCODE_FORMERR && 
 		sq->status == serviced_query_TCP_EDNS) {
@@ -1066,6 +1069,7 @@ serviced_udp_callback(struct comm_point* c, void* arg, int error,
 	struct serviced_query* sq = (struct serviced_query*)arg;
 	struct outside_network* outnet = sq->outnet;
 	struct timeval now;
+	int fallback_tcp = 0;
 	if(gettimeofday(&now, NULL) < 0) {
 		log_err("gettimeofday: %s", strerror(errno));
 		/* this option does not need current time */
@@ -1073,9 +1077,10 @@ serviced_udp_callback(struct comm_point* c, void* arg, int error,
 	}
 	sq->pending = NULL; /* removed after callback */
 	if(error == NETEVENT_TIMEOUT) {
+		int rto = 0;
 		sq->retry++;
-		if(!infra_rtt_update(outnet->infra, &sq->addr, sq->addrlen,
+		if(!(rto=infra_rtt_update(outnet->infra, &sq->addr, sq->addrlen,
-			-1, (time_t)now.tv_sec))
+			-1, (time_t)now.tv_sec)))
 			log_err("out of memory in UDP exponential backoff");
 		if(sq->retry < OUTBOUND_UDP_RETRY) {
 			log_name_addr(VERB_ALGO, "retry query", sq->qbuf+10,
@@ -1085,8 +1090,13 @@ serviced_udp_callback(struct comm_point* c, void* arg, int error,
 			}
 			return 0;
 		}
-		error = NETEVENT_TIMEOUT;
+		if(rto >= RTT_MAX_TIMEOUT) {
-		/* UDP does not work, fallback to TCP below */
+			fallback_tcp = 1;
+			/* UDP does not work, fallback to TCP below */
+		} else {
+			serviced_callbacks(sq, NETEVENT_TIMEOUT, c, rep);
+			return 0;
+		}
 	}
 	if(error == NETEVENT_NOERROR && sq->status == serviced_query_UDP_EDNS 
 		&& LDNS_RCODE_WIRE(ldns_buffer_begin(c->buffer)) 
@@ -1103,8 +1113,8 @@ serviced_udp_callback(struct comm_point* c, void* arg, int error,
 		}
 		return 0;
 	}
-	if(error != NETEVENT_NOERROR || 
+	if(LDNS_TC_WIRE(ldns_buffer_begin(c->buffer)) ||
-		LDNS_TC_WIRE(ldns_buffer_begin(c->buffer))) {
+		(error != NETEVENT_NOERROR && fallback_tcp)  ) {
 		/* fallback to TCP */
 		/* this discards partial UDP contents */
 		if(sq->status == serviced_query_UDP_EDNS)