NLnetLabs
diff --git a/‎doc/example.conf.in
Copy file name to clipboard
+3Lines changed: 3 additions & 0 deletions b/‎doc/example.conf.in
Copy file name to clipboard
+3Lines changed: 3 additions & 0 deletions
diff --git a/‎doc/unbound.conf.5.in
Copy file name to clipboard
+6Lines changed: 6 additions & 0 deletions b/‎doc/unbound.conf.5.in
Copy file name to clipboard
+6Lines changed: 6 additions & 0 deletions
diff --git a/‎services/cache/infra.c
Copy file name to clipboard
+24-4Lines changed: 24 additions & 4 deletions b/‎services/cache/infra.c
Copy file name to clipboard
+24-4Lines changed: 24 additions & 4 deletions
diff --git a/‎services/cache/infra.h
Copy file name to clipboard
+2Lines changed: 2 additions & 0 deletions b/‎services/cache/infra.h
Copy file name to clipboard
+2Lines changed: 2 additions & 0 deletions
diff --git a/‎util/config_file.c
Copy file name to clipboard
+3Lines changed: 3 additions & 0 deletions b/‎util/config_file.c
Copy file name to clipboard
+3Lines changed: 3 additions & 0 deletions
diff --git a/‎util/config_file.h
Copy file name to clipboard
+2Lines changed: 2 additions & 0 deletions b/‎util/config_file.h
Copy file name to clipboard
+2Lines changed: 2 additions & 0 deletions
@@ -192,6 +192,9 @@ server:
 	# minimum wait time for responses, increase if uplink is long. In msec.
 	# infra-cache-min-rtt: 50
 
+	# enable to make server probe down hosts more frequently.
+	# infra-keep-probing: no
+
 	# the number of slabs to use for the Infrastructure cache.
 	# the number of slabs must be a power of 2.
 	# more slabs reduce lock contention, but fragment memory usage.
 
@@ -382,6 +382,12 @@ Lower limit for dynamic retransmit timeout calculation in infrastructure
 cache. Default is 50 milliseconds. Increase this value if using forwarders
 needing more time to do recursive name resolution.
 .TP
+.B infra\-keep\-probing: \fI<yes or no>
+If enabled the server keeps probing hosts that are down, in the one probe
+at a time regime.  Default is no.  Hosts that are down, eg. they did
+not respond during the one probe at a time period, are marked as down and
+it may take \fBinfra\-host\-ttl\fR time to get probed again.
+.TP
 .B define\-tag: \fI<"list of tags">
 Define the tags that can be used with local\-zone and access\-control.
 Enclose the list between quotes ("") and put spaces between tags.
 
@@ -244,6 +244,7 @@ infra_create(struct config_file* cfg)
 		return NULL;
 	}
 	infra->host_ttl = cfg->host_ttl;
+	infra->infra_keep_probing = cfg->infra_keep_probing;
 	infra_dp_ratelimit = cfg->ratelimit;
 	infra->domain_rates = slabhash_create(cfg->ratelimit_slabs,
 		INFRA_HOST_STARTSIZE, cfg->ratelimit_size,
@@ -297,6 +298,7 @@ infra_adjust(struct infra_cache* infra, struct config_file* cfg)
 	if(!infra)
 		return infra_create(cfg);
 	infra->host_ttl = cfg->host_ttl;
+	infra->infra_keep_probing = cfg->infra_keep_probing;
 	infra_dp_ratelimit = cfg->ratelimit;
 	infra_ip_ratelimit = cfg->ip_ratelimit;
 	maxmem = cfg->infra_cache_numhosts * (sizeof(struct infra_key)+
@@ -445,6 +447,7 @@ infra_host(struct infra_cache* infra, struct sockaddr_storage* addr,
 	if(e && ((struct infra_data*)e->data)->ttl < timenow) {
 		/* it expired, try to reuse existing entry */
 		int old = ((struct infra_data*)e->data)->rtt.rto;
+		time_t tprobe = ((struct infra_data*)e->data)->probedelay;
 		uint8_t tA = ((struct infra_data*)e->data)->timeout_A;
 		uint8_t tAAAA = ((struct infra_data*)e->data)->timeout_AAAA;
 		uint8_t tother = ((struct infra_data*)e->data)->timeout_other;
@@ -460,6 +463,7 @@ infra_host(struct infra_cache* infra, struct sockaddr_storage* addr,
 			if(old >= USEFUL_SERVER_TOP_TIMEOUT) {
 				((struct infra_data*)e->data)->rtt.rto
 					= USEFUL_SERVER_TOP_TIMEOUT;
+				((struct infra_data*)e->data)->probedelay = tprobe;
 				((struct infra_data*)e->data)->timeout_A = tA;
 				((struct infra_data*)e->data)->timeout_AAAA = tAAAA;
 				((struct infra_data*)e->data)->timeout_other = tother;
@@ -482,7 +486,8 @@ infra_host(struct infra_cache* infra, struct sockaddr_storage* addr,
 	*edns_vs = data->edns_version;
 	*edns_lame_known = data->edns_lame_known;
 	*to = rtt_timeout(&data->rtt);
-	if(*to >= PROBE_MAXRTO && rtt_notimeout(&data->rtt)*4 <= *to) {
+	if(*to >= PROBE_MAXRTO && (infra->infra_keep_probing ||
+		rtt_notimeout(&data->rtt)*4 <= *to)) {
 		/* delay other queries, this is the probe query */
 		if(!wr) {
 			lock_rw_unlock(&e->lock);
@@ -566,18 +571,27 @@ infra_rtt_update(struct infra_cache* infra, struct sockaddr_storage* addr,
 	struct lruhash_entry* e = infra_lookup_nottl(infra, addr, addrlen,
 		nm, nmlen, 1);
 	struct infra_data* data;
-	int needtoinsert = 0;
+	int needtoinsert = 0, expired = 0;
 	int rto = 1;
+	time_t oldprobedelay = 0;
 	if(!e) {
 		if(!(e = new_entry(infra, addr, addrlen, nm, nmlen, timenow)))
 			return 0;
 		needtoinsert = 1;
 	} else if(((struct infra_data*)e->data)->ttl < timenow) {
+		oldprobedelay = ((struct infra_data*)e->data)->probedelay;
 		data_entry_init(infra, e, timenow);
+		expired = 1;
 	}
 	/* have an entry, update the rtt */
 	data = (struct infra_data*)e->data;
 	if(roundtrip == -1) {
+		if(needtoinsert || expired) {
+			/* timeout on entry that has expired before the timer
+			 * keep old timeout from the function caller */
+			data->rtt.rto = orig_rtt;
+			data->probedelay = oldprobedelay;
+		}
 		rtt_lost(&data->rtt, orig_rtt);
 		if(qtype == LDNS_RR_TYPE_A) {
 			if(data->timeout_A < TIMEOUT_COUNT_MAX)
@@ -681,7 +695,12 @@ infra_get_lame_rtt(struct infra_cache* infra,
 		return 0;
 	host = (struct infra_data*)e->data;
 	*rtt = rtt_unclamped(&host->rtt);
-	if(host->rtt.rto >= PROBE_MAXRTO && timenow < host->probedelay
+	if(host->rtt.rto >= PROBE_MAXRTO && timenow >= host->probedelay
+		&& infra->infra_keep_probing) {
+		/* single probe, keep probing */
+		if(*rtt >= USEFUL_SERVER_TOP_TIMEOUT)
+			*rtt = USEFUL_SERVER_TOP_TIMEOUT-1000;
+	} else if(host->rtt.rto >= PROBE_MAXRTO && timenow < host->probedelay
 		&& rtt_notimeout(&host->rtt)*4 <= host->rtt.rto) {
 		/* single probe for this domain, and we are not probing */
 		/* unless the query type allows a probe to happen */
@@ -704,7 +723,8 @@ infra_get_lame_rtt(struct infra_cache* infra,
 		/* see if this can be a re-probe of an unresponsive server */
 		/* minus 1000 because that is outside of the RTTBAND, so
 		 * blacklisted servers stay blacklisted if this is chosen */
-		if(host->rtt.rto >= USEFUL_SERVER_TOP_TIMEOUT) {
+		if(host->rtt.rto >= USEFUL_SERVER_TOP_TIMEOUT ||
+			infra->infra_keep_probing) {
 			lock_rw_unlock(&e->lock);
 			*rtt = USEFUL_SERVER_TOP_TIMEOUT-1000;
 			*lame = 0;
 
@@ -114,6 +114,8 @@ struct infra_cache {
 	struct slabhash* hosts;
 	/** TTL value for host information, in seconds */
 	int host_ttl;
+	/** the hosts that are down are kept probed for recovery */
+	int infra_keep_probing;
 	/** hash table with query rates per name: rate_key, rate_data */
 	struct slabhash* domain_rates;
 	/** ratelimit settings for domains, struct domain_limit_data */
 
@@ -170,6 +170,7 @@ config_create(void)
 	cfg->infra_cache_slabs = 4;
 	cfg->infra_cache_numhosts = 10000;
 	cfg->infra_cache_min_rtt = 50;
+	cfg->infra_keep_probing = 0;
 	cfg->delay_close = 0;
 	if(!(cfg->outgoing_avail_ports = (int*)calloc(65536, sizeof(int))))
 		goto error_exit;
@@ -563,6 +564,7 @@ int config_set_option(struct config_file* cfg, const char* opt,
 	    IS_NUMBER_OR_ZERO; cfg->infra_cache_min_rtt = atoi(val);
 	    RTT_MIN_TIMEOUT=cfg->infra_cache_min_rtt;
 	}
+	else S_YNO("infra-keep-probing:", infra_keep_probing)
 	else S_NUMBER_OR_ZERO("infra-host-ttl:", host_ttl)
 	else S_POW2("infra-cache-slabs:", infra_cache_slabs)
 	else S_SIZET_NONZERO("infra-cache-numhosts:", infra_cache_numhosts)
@@ -959,6 +961,7 @@ config_get_option(struct config_file* cfg, const char* opt,
 	else O_DEC(opt, "infra-host-ttl", host_ttl)
 	else O_DEC(opt, "infra-cache-slabs", infra_cache_slabs)
 	else O_DEC(opt, "infra-cache-min-rtt", infra_cache_min_rtt)
+	else O_YNO(opt, "infra-keep-probing", infra_keep_probing)
 	else O_MEM(opt, "infra-cache-numhosts", infra_cache_numhosts)
 	else O_UNS(opt, "delay-close", delay_close)
 	else O_YNO(opt, "do-ip4", do_ip4)
 
@@ -181,6 +181,8 @@ struct config_file {
 	size_t infra_cache_numhosts;
 	/** min value for infra cache rtt */
 	int infra_cache_min_rtt;
+	/** keep probing hosts that are down */
+	int infra_keep_probing;
 	/** delay close of udp-timeouted ports, if 0 no delayclose. in msec */
 	int delay_close;