- Fix timeouts to keep track of query type, A, AAAA and other, if

another has caused timeout blacklist, different type can still probe.


git-svn-id: file:///svn/unbound/trunk@2613 be551aaa-1e26-0410-a405-d3ace91eadb9

doc/Changelog

@@ -1,5 +1,7 @@
 10 February 2012: Wouter
 	- Slightly smaller critical region in one case in infra cache.
+	- Fix timeouts to keep track of query type, A, AAAA and other, if
+	  another has caused timeout blacklist, different type can still probe.
 
 9 February 2012: Wouter
 	- Fix AHX_BROKEN_MEMCMP for autoheader mess up of #undef in config.h.

services/cache/infra.c

@@ -52,6 +52,11 @@
 /** Timeout when only a single probe query per IP is allowed. */
 #define PROBE_MAXRTO 12000 /* in msec */
 
+/** number of timeouts for a type when the domain can be blocked ;
+ * even if another type has completely rtt maxed it, the different type
+ * can do this number of packets (until those all timeout too) */
+#define TIMEOUT_COUNT_MAX 3
+
 size_t 
 infra_sizefunc(void* k, void* ATTR_UNUSED(d))
 {
@@ -196,6 +201,9 @@ data_entry_init(struct infra_cache* infra, struct lruhash_entry* e,
 	data->rec_lame = 0;
 	data->lame_type_A = 0;
 	data->lame_other = 0;
+	data->timeout_A = 0;
+	data->timeout_AAAA = 0;
+	data->timeout_other = 0;
 }
 
 /** 
@@ -250,6 +258,9 @@ 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;
+		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;
 		lock_rw_unlock(&e->lock);
 		e = infra_lookup_nottl(infra, addr, addrlen, nm, nmlen, 1);
 		if(e) {
@@ -259,9 +270,13 @@ infra_host(struct infra_cache* infra, struct sockaddr_storage* addr,
 			data_entry_init(infra, e, timenow);
 			wr = 1;
 			/* TOP_TIMEOUT remains on reuse */
-			if(old >= USEFUL_SERVER_TOP_TIMEOUT)
+			if(old >= USEFUL_SERVER_TOP_TIMEOUT) {
 				((struct infra_data*)e->data)->rtt.rto
 					= USEFUL_SERVER_TOP_TIMEOUT;
+				((struct infra_data*)e->data)->timeout_A = tA;
+				((struct infra_data*)e->data)->timeout_AAAA = tAAAA;
+				((struct infra_data*)e->data)->timeout_other = tother;
+			}
 		}
 	}
 	if(!e) {
@@ -358,8 +373,8 @@ infra_update_tcp_works(struct infra_cache* infra,
 
 int 
 infra_rtt_update(struct infra_cache* infra, struct sockaddr_storage* addr,
-	socklen_t addrlen, uint8_t* nm, size_t nmlen, int roundtrip,
+	socklen_t addrlen, uint8_t* nm, size_t nmlen, int qtype,
-	int orig_rtt, uint32_t timenow)
+	int roundtrip, int orig_rtt, uint32_t timenow)
 {
 	struct lruhash_entry* e = infra_lookup_nottl(infra, addr, addrlen,
 		nm, nmlen, 1);
@@ -377,9 +392,24 @@ infra_rtt_update(struct infra_cache* infra, struct sockaddr_storage* addr,
 	data = (struct infra_data*)e->data;
 	if(roundtrip == -1) {
 		rtt_lost(&data->rtt, orig_rtt);
+		if(qtype == LDNS_RR_TYPE_A) {
+			if(data->timeout_A < TIMEOUT_COUNT_MAX)
+				data->timeout_A++;
+		} else if(qtype == LDNS_RR_TYPE_AAAA) {
+			if(data->timeout_AAAA < TIMEOUT_COUNT_MAX)
+				data->timeout_AAAA++;
+		} else {
+			if(data->timeout_other < TIMEOUT_COUNT_MAX)
+				data->timeout_other++;
+		}
 	} else {
 		rtt_update(&data->rtt, roundtrip);
 		data->probedelay = 0;
+		if(qtype == LDNS_RR_TYPE_A)
+			data->timeout_A = 0;
+		else if(qtype == LDNS_RR_TYPE_AAAA)
+			data->timeout_AAAA = 0;
+		else	data->timeout_other = 0;
 	}
 	if(data->rtt.rto > 0)
 		rto = data->rtt.rto;
@@ -456,9 +486,23 @@ infra_get_lame_rtt(struct infra_cache* infra,
 	host = (struct infra_data*)e->data;
 	*rtt = rtt_unclamped(&host->rtt);
 	if(host->rtt.rto >= PROBE_MAXRTO && timenow < host->probedelay
-		&& rtt_notimeout(&host->rtt)*4 <= host->rtt.rto)
+		&& rtt_notimeout(&host->rtt)*4 <= host->rtt.rto) {
 		/* single probe for this domain, and we are not probing */
-		*rtt = USEFUL_SERVER_TOP_TIMEOUT;
+		/* unless the query type allows a probe to happen */
+		if(qtype == LDNS_RR_TYPE_A) {
+			if(host->timeout_A >= TIMEOUT_COUNT_MAX)
+				*rtt = USEFUL_SERVER_TOP_TIMEOUT;
+			else	*rtt = USEFUL_SERVER_TOP_TIMEOUT-1000;
+		} else if(qtype == LDNS_RR_TYPE_AAAA) {
+			if(host->timeout_AAAA >= TIMEOUT_COUNT_MAX)
+				*rtt = USEFUL_SERVER_TOP_TIMEOUT;
+			else	*rtt = USEFUL_SERVER_TOP_TIMEOUT-1000;
+		} else {
+			if(host->timeout_other >= TIMEOUT_COUNT_MAX)
+				*rtt = USEFUL_SERVER_TOP_TIMEOUT;
+			else	*rtt = USEFUL_SERVER_TOP_TIMEOUT-1000;
+		}
+	}
 	if(timenow > host->ttl) {
 		/* expired entry */
 		/* see if this can be a re-probe of an unresponsive server */

services/cache/infra.h

@@ -91,6 +91,13 @@ struct infra_data {
 	uint8_t lame_type_A;
 	/** the host is lame (not authoritative) for other query types */
 	uint8_t lame_other;
+
+	/** timeouts counter for type A */
+	uint8_t timeout_A;
+	/** timeouts counter for type AAAA */
+	uint8_t timeout_AAAA;
+	/** timeouts counter for others */
+	uint8_t timeout_other;
 };
 
 /**
@@ -195,6 +202,7 @@ int infra_set_lame(struct infra_cache* infra,
  * @param addrlen: length of addr.
  * @param name: zone name
  * @param namelen: zone name length
+ * @param qtype: query type.
  * @param roundtrip: estimate of roundtrip time in milliseconds or -1 for 
  * 	timeout.
  * @param orig_rtt: original rtt for the query that timed out (roundtrip==-1).
@@ -203,7 +211,7 @@ int infra_set_lame(struct infra_cache* infra,
  * @return: 0 on error. new rto otherwise.
  */
 int infra_rtt_update(struct infra_cache* infra, struct sockaddr_storage* addr,
-	socklen_t addrlen, uint8_t* name, size_t namelen,
+	socklen_t addrlen, uint8_t* name, size_t namelen, int qtype,
 	int roundtrip, int orig_rtt, uint32_t timenow);
 
 /**

services/outside_network.c

@@ -1166,7 +1166,7 @@ static struct serviced_query*
 serviced_create(struct outside_network* outnet, ldns_buffer* buff, int dnssec,
 	int want_dnssec, int tcp_upstream, int ssl_upstream,
 	struct sockaddr_storage* addr, socklen_t addrlen, uint8_t* zone,
-	size_t zonelen)
+	size_t zonelen, int qtype)
 {
 	struct serviced_query* sq = (struct serviced_query*)malloc(sizeof(*sq));
 #ifdef UNBOUND_DEBUG
@@ -1188,6 +1188,7 @@ serviced_create(struct outside_network* outnet, ldns_buffer* buff, int dnssec,
 		return NULL;
 	}
 	sq->zonelen = zonelen;
+	sq->qtype = qtype;
 	sq->dnssec = dnssec;
 	sq->want_dnssec = want_dnssec;
 	sq->tcp_upstream = tcp_upstream;
@@ -1566,8 +1567,8 @@ serviced_tcp_callback(struct comm_point* c, void* arg, int error,
 		 * huge due to system-hibernated and we woke up */
 		if(roundtime < TCP_AUTH_QUERY_TIMEOUT*1000) {
 		    if(!infra_rtt_update(sq->outnet->infra, &sq->addr,
-			sq->addrlen, sq->zone, sq->zonelen, roundtime,
+			sq->addrlen, sq->zone, sq->zonelen, sq->qtype,
-			sq->last_rtt, (uint32_t)now.tv_sec))
+			roundtime, sq->last_rtt, (uint32_t)now.tv_sec))
 			log_err("out of memory noting rtt.");
 		}
 	    }
@@ -1658,7 +1659,7 @@ serviced_udp_callback(struct comm_point* c, void* arg, int error,
 		}
 		sq->retry++;
 		if(!(rto=infra_rtt_update(outnet->infra, &sq->addr, sq->addrlen,
-			sq->zone, sq->zonelen, -1, sq->last_rtt,
+			sq->zone, sq->zonelen, sq->qtype, -1, sq->last_rtt,
 			(uint32_t)now.tv_sec)))
 			log_err("out of memory in UDP exponential backoff");
 		if(sq->retry < OUTBOUND_UDP_RETRY) {
@@ -1752,8 +1753,8 @@ serviced_udp_callback(struct comm_point* c, void* arg, int error,
 		 * above this value gives trouble with server selection */
 		if(roundtime < 60000) {
 		    if(!infra_rtt_update(outnet->infra, &sq->addr, sq->addrlen, 
-			sq->zone, sq->zonelen, roundtime, sq->last_rtt,
+			sq->zone, sq->zonelen, sq->qtype, roundtime,
-			(uint32_t)now.tv_sec))
+			sq->last_rtt, (uint32_t)now.tv_sec))
 			log_err("out of memory noting rtt.");
 		}
 	    }
@@ -1814,7 +1815,7 @@ outnet_serviced_query(struct outside_network* outnet,
 		/* make new serviced query entry */
 		sq = serviced_create(outnet, buff, dnssec, want_dnssec,
 			tcp_upstream, ssl_upstream, addr, addrlen, zone,
-			zonelen);
+			zonelen, qtype);
 		if(!sq) {
 			free(cb);
 			return NULL;

services/outside_network.h

@@ -312,6 +312,8 @@ struct serviced_query {
 	uint8_t* zone;
 	/** length of zone name */
 	size_t zonelen;
+	/** qtype */
+	int qtype;
 	/** current status */
 	enum serviced_query_status {
 		/** initial status */

testcode/fake_event.c

@@ -147,6 +147,7 @@ delete_fake_pending(struct fake_pending* pend)
 {
 	if(!pend)
 		return;
+	free(pend->zone);
 	ldns_buffer_free(pend->buffer);
 	ldns_pkt_free(pend->pkt);
 	free(pend);
@@ -554,7 +555,7 @@ do_infra_rtt(struct replay_runtime* runtime)
 	if(!dp) fatal_exit("cannot parse %s", now->variable);
 	rto = infra_rtt_update(runtime->infra, &now->addr,
 		now->addrlen, ldns_rdf_data(dp), ldns_rdf_size(dp),
-		atoi(now->string), -1, runtime->now_secs);
+		LDNS_RR_TYPE_A, atoi(now->string), -1, runtime->now_secs);
 	log_addr(0, "INFRA_RTT for", &now->addr, now->addrlen);
 	log_info("INFRA_RTT(%s roundtrip %d): rto of %d", now->variable,
 		atoi(now->string), rto);
@@ -562,6 +563,24 @@ do_infra_rtt(struct replay_runtime* runtime)
 	ldns_rdf_deep_free(dp);
 }
 
+/** perform exponential backoff on the timout */
+static void
+expon_timeout_backoff(struct replay_runtime* runtime)
+{
+	struct fake_pending* p = runtime->pending_list;
+	int rtt, vs;
+	uint8_t edns_lame_known;
+	int last_rtt, rto;
+	if(!p) return; /* no pending packet to backoff */
+	if(!infra_host(runtime->infra, &p->addr, p->addrlen, p->zone,
+		p->zonelen, runtime->now_secs, &vs, &edns_lame_known, &rtt))
+		return;
+	last_rtt = rtt;
+	rto = infra_rtt_update(runtime->infra, &p->addr, p->addrlen, p->zone,
+		p->zonelen, p->qtype, -1, last_rtt, runtime->now_secs);
+	log_info("infra_rtt_update returned rto %d", rto);
+}
+
 /**
  * Advance to the next moment.
  */
@@ -608,6 +627,7 @@ do_moment_and_advance(struct replay_runtime* runtime)
 	case repevt_timeout:
 		mom = runtime->now;
 		advance_moment(runtime);
+		expon_timeout_backoff(runtime);
 		fake_pending_callback(runtime, mom, NETEVENT_TIMEOUT);
 		break;
 	case repevt_back_reply:
@@ -929,6 +949,7 @@ pending_udp_query(struct outside_network* outnet, ldns_buffer* packet,
 	pend->timeout = timeout/1000;
 	pend->transport = transport_udp;
 	pend->pkt = NULL;
+	pend->zone = NULL;
 	pend->serviced = 0;
 	pend->runtime = runtime;
 	status = ldns_buffer2pkt_wire(&pend->pkt, packet);
@@ -982,6 +1003,7 @@ pending_tcp_query(struct outside_network* outnet, ldns_buffer* packet,
 	pend->timeout = timeout;
 	pend->transport = transport_tcp;
 	pend->pkt = NULL;
+	pend->zone = NULL;
 	pend->runtime = runtime;
 	pend->serviced = 0;
 	status = ldns_buffer2pkt_wire(&pend->pkt, packet);
@@ -1017,9 +1039,8 @@ struct serviced_query* outnet_serviced_query(struct outside_network* outnet,
 	uint16_t flags, int dnssec, int ATTR_UNUSED(want_dnssec),
 	int ATTR_UNUSED(tcp_upstream), int ATTR_UNUSED(ssl_upstream),
 	struct sockaddr_storage* addr, socklen_t addrlen, uint8_t* zone,
-	size_t ATTR_UNUSED(zonelen), comm_point_callback_t* callback,
+	size_t zonelen, comm_point_callback_t* callback, void* callback_arg,
-	void* callback_arg, ldns_buffer* ATTR_UNUSED(buff),
+	ldns_buffer* ATTR_UNUSED(buff), int (*arg_compare)(void*,void*))
-	int (*arg_compare)(void*,void*))
 {
 	struct replay_runtime* runtime = (struct replay_runtime*)outnet->base;
 	struct fake_pending* pend = (struct fake_pending*)calloc(1,
@@ -1062,6 +1083,10 @@ struct serviced_query* outnet_serviced_query(struct outside_network* outnet,
 	}
 	memcpy(&pend->addr, addr, addrlen);
 	pend->addrlen = addrlen;
+	pend->zone = memdup(zone, zonelen);
+	pend->zonelen = zonelen;
+	pend->qtype = qtype;
+	log_assert(pend->zone);
 	pend->callback = callback;
 	pend->cb_arg = callback_arg;
 	pend->timeout = UDP_AUTH_QUERY_TIMEOUT;

testcode/replay.h

@@ -323,6 +323,12 @@ struct fake_pending {
 	struct sockaddr_storage addr;
 	/** len of addr */
 	socklen_t addrlen;
+	/** zone name, uncompressed wire format (as used when sent) */
+	uint8_t* zone;
+	/** length of zone name */
+	size_t zonelen;
+	/** qtype */
+	int qtype;
 	/** The callback function to call when answer arrives (or timeout) */
 	comm_point_callback_t* callback;
 	/** callback user argument */

testcode/unitmain.c

@@ -445,7 +445,7 @@ infra_test(void)
 		&vs, &edns_lame, &to) );
 	unit_assert( vs == 0 && to == init && edns_lame == 0 );
 
-	unit_assert( infra_rtt_update(slab, &one, onelen, zone, zonelen, -1, init, now) );
+	unit_assert( infra_rtt_update(slab, &one, onelen, zone, zonelen, LDNS_RR_TYPE_A, -1, init, now) );
 	unit_assert( infra_host(slab, &one, onelen, zone, zonelen, 
 			now, &vs, &edns_lame, &to) );
 	unit_assert( vs == 0 && to == init*2 && edns_lame == 0 );

testdata/iter_timeout_ra_aaaa.rpl

@@ -0,0 +1,244 @@
+; config options
+server:
+	target-fetch-policy: "0 0 0 0 0"
+
+stub-zone:
+	name: "."
+	stub-addr: 193.0.14.129 	# K.ROOT-SERVERS.NET.
+CONFIG_END
+
+SCENARIO_BEGIN Test iterator with timeouts on reclame AAAA dropping server
+
+; K.ROOT-SERVERS.NET.
+RANGE_BEGIN 0 100
+	ADDRESS 193.0.14.129 
+ENTRY_BEGIN
+MATCH opcode qtype qname
+ADJUST copy_id
+REPLY QR NOERROR
+SECTION QUESTION
+. IN NS
+SECTION ANSWER
+. IN NS	K.ROOT-SERVERS.NET.
+SECTION ADDITIONAL
+K.ROOT-SERVERS.NET.	IN	A	193.0.14.129
+ENTRY_END
+
+ENTRY_BEGIN
+MATCH opcode subdomain
+ADJUST copy_id copy_query
+REPLY QR NOERROR
+SECTION QUESTION
+com. IN A
+SECTION AUTHORITY
+com.	IN NS	a.gtld-servers.net.
+SECTION ADDITIONAL
+a.gtld-servers.net.	IN 	A	192.5.6.30
+ENTRY_END
+
+ENTRY_BEGIN
+MATCH opcode qtype qname
+ADJUST copy_id
+REPLY QR AA NOERROR
+SECTION QUESTION
+ns.example.net. IN A
+SECTION ANSWER
+ns.example.net. IN A 1.2.3.4
+ENTRY_END
+
+ENTRY_BEGIN
+MATCH opcode qtype qname
+ADJUST copy_id
+REPLY QR AA NOERROR
+SECTION QUESTION
+ns.example.net. IN AAAA
+SECTION ANSWER
+SECTION AUTHORITY
+. IN SOA a. b. 1 2 3 4 5
+ENTRY_END
+
+ENTRY_BEGIN
+MATCH opcode qtype qname
+ADJUST copy_id
+REPLY QR NOERROR
+SECTION QUESTION
+a.gtld-servers.net. IN AAAA
+SECTION ANSWER
+SECTION AUTHORITY
+. IN SOA a. b. 1 2 3 4 5
+ENTRY_END
+
+RANGE_END
+
+; a.gtld-servers.net.
+RANGE_BEGIN 0 100
+	ADDRESS 192.5.6.30
+ENTRY_BEGIN
+MATCH opcode qtype qname
+ADJUST copy_id
+REPLY QR NOERROR
+SECTION QUESTION
+com. IN NS
+SECTION ANSWER
+com.	IN NS	a.gtld-servers.net.
+SECTION ADDITIONAL
+a.gtld-servers.net.	IN 	A	192.5.6.30
+ENTRY_END
+
+ENTRY_BEGIN
+MATCH opcode subdomain
+ADJUST copy_id copy_query
+REPLY QR NOERROR
+SECTION QUESTION
+example.com. IN A
+SECTION AUTHORITY
+example.com.	280 IN NS	ns.example.net.
+SECTION ADDITIONAL
+ns.example.net.		IN 	A	1.2.3.4
+ENTRY_END
+RANGE_END
+
+; ns.example.com.
+; This server is REC_LAME
+RANGE_BEGIN 0 100
+	ADDRESS 1.2.3.4
+ENTRY_BEGIN
+MATCH opcode qtype qname
+ADJUST copy_id
+REPLY QR RA NOERROR
+SECTION QUESTION
+example.com. IN NS
+SECTION ANSWER
+example.com.	280 IN NS	ns.example.net.
+ENTRY_END
+
+ENTRY_BEGIN
+MATCH opcode qtype qname
+ADJUST copy_id
+REPLY QR RA NOERROR
+SECTION QUESTION
+www.example.com. IN A
+SECTION ANSWER
+www.example.com. 10 IN A	10.20.30.40
+SECTION AUTHORITY
+example.com.	280 IN NS	ns.example.net.
+ENTRY_END
+RANGE_END
+
+STEP 1 QUERY
+ENTRY_BEGIN
+REPLY RD
+SECTION QUESTION
+www.example.com. IN A
+ENTRY_END
+
+; recursion happens here.
+STEP 10 CHECK_ANSWER
+ENTRY_BEGIN
+MATCH all
+REPLY QR RD RA NOERROR
+SECTION QUESTION
+www.example.com. IN A
+SECTION ANSWER
+www.example.com. IN A	10.20.30.40
+SECTION AUTHORITY
+example.com.	IN NS	ns.example.net.
+ENTRY_END
+
+; query for (dropped) AAAA record.
+STEP 20 QUERY
+ENTRY_BEGIN
+REPLY RD
+SECTION QUESTION
+www.example.com. IN AAAA
+ENTRY_END
+
+; the AAAA query times out.
+STEP 21 TIMEOUT
+STEP 22 TIMEOUT
+STEP 23 TIMEOUT
+STEP 24 TIMEOUT
+STEP 25 TIMEOUT
+
+; we get servfail, but the AAA arrives again (after the servfail times 
+; out of the cache)
+STEP 30 CHECK_ANSWER
+ENTRY_BEGIN
+MATCH all
+REPLY QR RD RA SERVFAIL
+SECTION QUESTION
+www.example.com. IN AAAA
+ENTRY_END
+STEP 31 TIME_PASSES ELAPSE 6
+STEP 40 QUERY
+ENTRY_BEGIN
+REPLY RD
+SECTION QUESTION
+www.example.com. IN AAAA
+ENTRY_END
+
+; timeouts for AAAA keep happening.
+STEP 41 TIMEOUT
+STEP 42 CHECK_ANSWER
+ENTRY_BEGIN
+MATCH all
+REPLY QR RD RA SERVFAIL
+SECTION QUESTION
+www.example.com. IN AAAA
+ENTRY_END
+STEP 43 TIME_PASSES ELAPSE 12
+
+STEP 50 QUERY
+ENTRY_BEGIN
+REPLY RD
+SECTION QUESTION
+www.example.com. IN AAAA
+ENTRY_END
+; fallback queries
+STEP 51 TRAFFIC
+; and it fails, no parentside entries and so on.
+STEP 52 CHECK_ANSWER
+ENTRY_BEGIN
+MATCH all
+REPLY QR RD RA SERVFAIL
+SECTION QUESTION
+www.example.com. IN AAAA
+ENTRY_END
+STEP 53 TIME_PASSES ELAPSE 12
+
+STEP 60 QUERY
+ENTRY_BEGIN
+REPLY RD
+SECTION QUESTION
+www.example.com. IN AAAA
+ENTRY_END
+STEP 61 TIMEOUT
+STEP 62 TRAFFIC
+STEP 63 CHECK_ANSWER
+ENTRY_BEGIN
+MATCH all
+REPLY QR RD RA SERVFAIL
+SECTION QUESTION
+www.example.com. IN AAAA
+ENTRY_END
+
+STEP 70 QUERY
+ENTRY_BEGIN
+REPLY RD
+SECTION QUESTION
+www.example.com. IN A
+ENTRY_END
+STEP 72 TRAFFIC
+STEP 73 CHECK_ANSWER
+ENTRY_BEGIN
+MATCH all
+REPLY QR RD RA NOERROR
+SECTION QUESTION
+www.example.com. IN A
+SECTION ANSWER
+www.example.com. IN A	10.20.30.40
+SECTION AUTHORITY
+example.com.	IN NS	ns.example.net.
+ENTRY_END
+
+SCENARIO_END