0x20 fallback code.

git-svn-id: file:///svn/unbound/trunk@1285 be551aaa-1e26-0410-a405-d3ace91eadb9

daemon/worker.c

@@ -211,7 +211,7 @@ worker_handle_reply(struct comm_point* c, void* arg, int error,
 	e.qsent = NULL;
 
 	if(error != 0) {
-		mesh_report_reply(worker->env.mesh, &e, 0, reply_info);
+		mesh_report_reply(worker->env.mesh, &e, reply_info, error);
 		worker_mem_report(worker, NULL);
 		return 0;
 	}
@@ -222,11 +222,12 @@ worker_handle_reply(struct comm_point* c, void* arg, int error,
 		|| LDNS_QDCOUNT(ldns_buffer_begin(c->buffer)) > 1) {
 		/* error becomes timeout for the module as if this reply
 		 * never arrived. */
-		mesh_report_reply(worker->env.mesh, &e, 0, reply_info);
+		mesh_report_reply(worker->env.mesh, &e, reply_info, 
+			NETEVENT_TIMEOUT);
 		worker_mem_report(worker, NULL);
 		return 0;
 	}
-	mesh_report_reply(worker->env.mesh, &e, 1, reply_info);
+	mesh_report_reply(worker->env.mesh, &e, reply_info, NETEVENT_NOERROR);
 	worker_mem_report(worker, NULL);
 	return 0;
 }
@@ -241,7 +242,7 @@ worker_handle_service_reply(struct comm_point* c, void* arg, int error,
 
 	verbose(VERB_ALGO, "worker svcd callback for qstate %p", e->qstate);
 	if(error != 0) {
-		mesh_report_reply(worker->env.mesh, e, 0, reply_info);
+		mesh_report_reply(worker->env.mesh, e, reply_info, error);
 		worker_mem_report(worker, sq);
 		return 0;
 	}
@@ -253,11 +254,12 @@ worker_handle_service_reply(struct comm_point* c, void* arg, int error,
 		/* error becomes timeout for the module as if this reply
 		 * never arrived. */
 		verbose(VERB_ALGO, "worker: bad reply handled as timeout");
-		mesh_report_reply(worker->env.mesh, e, 0, reply_info);
+		mesh_report_reply(worker->env.mesh, e, reply_info, 
+			NETEVENT_TIMEOUT);
 		worker_mem_report(worker, sq);
 		return 0;
 	}
-	mesh_report_reply(worker->env.mesh, e, 1, reply_info);
+	mesh_report_reply(worker->env.mesh, e, reply_info, NETEVENT_NOERROR);
 	worker_mem_report(worker, sq);
 	return 0;
 }

doc/Changelog

@@ -1,5 +1,7 @@
 6 October 2008: Wouter
 	- jostle-timeout option, so you can config for slow links.
+	- 0x20 fallback code.  Tries 3xnumber of nameserver addresses
+	  queries that must all be the same.  Sent to random nameservers.
 
 2 October 2008: Wouter
 	- fixup unlink of pidfile.

iterator/iter_utils.c

@@ -484,3 +484,60 @@ int iter_msg_from_zone(struct dns_msg* msg, struct delegpt* dp,
 		return 1;
 	return 0;
 }
+
+/**
+ * check equality of two rrsets 
+ * @param k1: rrset
+ * @param k2: rrset
+ * @return true if equal
+ */
+static int
+rrset_equal(struct ub_packed_rrset_key* k1, struct ub_packed_rrset_key* k2)
+{
+	struct packed_rrset_data* d1 = (struct packed_rrset_data*)
+		k1->entry.data;
+	struct packed_rrset_data* d2 = (struct packed_rrset_data*)
+		k2->entry.data;
+	size_t i, t;
+	if(k1->rk.dname_len != k2->rk.dname_len ||
+		k1->rk.flags != k2->rk.flags ||
+		k1->rk.type != k2->rk.type ||
+		k1->rk.rrset_class != k2->rk.rrset_class ||
+		query_dname_compare(k1->rk.dname, k2->rk.dname) != 0)
+		return 0;
+	if(d1->ttl != d2->ttl ||
+		d1->count != d2->count ||
+		d1->rrsig_count != d2->rrsig_count ||
+		d1->trust != d2->trust ||
+		d1->security != d2->security)
+		return 0;
+	t = d1->count + d1->rrsig_count;
+	for(i=0; i<t; i++) {
+		if(d1->rr_len[i] != d2->rr_len[i] ||
+			d1->rr_ttl[i] != d2->rr_ttl[i] ||
+			memcmp(d1->rr_data[i], d2->rr_data[i], 
+				d1->rr_len[i]) != 0)
+			return 0;
+	}
+	return 1;
+}
+
+int 
+reply_equal(struct reply_info* p, struct reply_info* q)
+{
+	size_t i;
+	if(p->flags != q->flags ||
+		p->qdcount != q->qdcount ||
+		p->ttl != q->ttl ||
+		p->security != q->security ||
+		p->an_numrrsets != q->an_numrrsets ||
+		p->ns_numrrsets != q->ns_numrrsets ||
+		p->ar_numrrsets != q->ar_numrrsets ||
+		p->rrset_count != q->rrset_count)
+		return 0;
+	for(i=0; i<p->rrset_count; i++) {
+		if(!rrset_equal(p->rrsets[i], q->rrsets[i]))
+			return 0;
+	}
+	return 1;
+}

iterator/iter_utils.h

@@ -180,4 +180,14 @@ int iter_msg_has_dnssec(struct dns_msg* msg);
 int iter_msg_from_zone(struct dns_msg* msg, struct delegpt* dp, 
 	enum response_type type, uint16_t dclass);
 
+/**
+ * Check if two replies are equal
+ * For fallback procedures
+ * @param p: reply one. The reply has rrset data pointers in region.
+ * 	Does not check rrset-IDs
+ * @param q: reply two
+ * @return if one and two are equal.
+ */
+int reply_equal(struct reply_info* p, struct reply_info* q);
+
 #endif /* ITERATOR_ITER_UTILS_H */

iterator/iterator.c

@@ -1153,11 +1153,41 @@ processQueryTargets(struct module_qstate* qstate, struct iter_qstate* iq,
 		tf_policy = ie->target_fetch_policy[iq->depth];
 	}
 
+	/* if in 0x20 fallback get as many targets as possible */
+	if(iq->caps_fallback) {
+		int extra = 0;
+		size_t naddr, nres, navail;
+		if(!query_for_targets(qstate, iq, ie, id, -1, &extra)) {
+			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
+		}
+		iq->num_target_queries += extra;
+		if(iq->num_target_queries > 0) {
+			/* wait to get all targets, we want to try em */
+			verbose(VERB_ALGO, "wait for all targets for fallback");
+			return 0;
+		}
+		/* did we do enough fallback queries already? */
+		delegpt_count_addr(iq->dp, &naddr, &nres, &navail);
+		/* the current caps_server is the number of fallbacks sent.
+		 * the original query is one that matched too, so we have
+		 * caps_server+1 number of matching queries now */
+		if(iq->caps_server+1 >= naddr*3) {
+			/* we're done, process the response */
+			verbose(VERB_ALGO, "0x20 fallback had %d responses "
+				"match for %d wanted, done.", 
+				(int)iq->caps_server+1, (int)naddr*3);
+			iq->caps_fallback = 0;
+			iq->state = QUERY_RESP_STATE;
+			return 1;
+		}
+		verbose(VERB_ALGO, "0x20 fallback number %d", 
+			(int)iq->caps_server);
+
 	/* if there is a policy to fetch missing targets 
 	 * opportunistically, do it. we rely on the fact that once a 
 	 * query (or queries) for a missing name have been issued, 
 	 * they will not be show up again. */
-	if(tf_policy != 0) {
+	} else if(tf_policy != 0) {
 		int extra = 0;
 		verbose(VERB_ALGO, "attempt to get extra %d targets", 
 			tf_policy);
@@ -1760,7 +1790,8 @@ process_response(struct module_qstate* qstate, struct iter_qstate* iq,
 	if(event == module_event_noreply || event == module_event_error) {
 		goto handle_it;
 	}
-	if(event != module_event_reply || !qstate->reply) {
+	if( (event != module_event_reply && event != module_event_capsfail)
+		|| !qstate->reply) {
 		log_err("Bad event combined with response");
 		outbound_list_remove(&iq->outlist, outbound);
 		(void)error_response(qstate, id, LDNS_RCODE_SERVFAIL);
@@ -1805,6 +1836,37 @@ process_response(struct module_qstate* qstate, struct iter_qstate* iq,
 		log_dns_msg("incoming scrubbed packet:", &iq->response->qinfo, 
 			iq->response->rep);
 	
+	if(event == module_event_capsfail) {
+		if(!iq->caps_fallback) {
+			/* start fallback */
+			iq->caps_fallback = 1;
+			iq->caps_server = 0;
+			iq->caps_reply = iq->response->rep;
+			iq->state = QUERYTARGETS_STATE;
+			iq->num_current_queries--;
+			verbose(VERB_DETAIL, "Capsforid: starting fallback");
+			goto handle_it;
+		} else {
+			/* check if reply is the same, otherwise, fail */
+			if(!reply_equal(iq->response->rep, iq->caps_reply)) {
+				verbose(VERB_DETAIL, "Capsforid fallback: "
+					"getting different replies, failed");
+				outbound_list_remove(&iq->outlist, outbound);
+				(void)error_response(qstate, id, 
+					LDNS_RCODE_SERVFAIL);
+				return;
+			}
+			/* continue the fallback procedure at next server */
+			iq->caps_server++;
+			iq->state = QUERYTARGETS_STATE;
+			iq->num_current_queries--;
+			verbose(VERB_DETAIL, "Capsforid: reply is equal. "
+				"go to next fallback");
+			goto handle_it;
+		}
+	}
+	iq->caps_fallback = 0; /* if we were in fallback, 0x20 is OK now */
+
 handle_it:
 	outbound_list_remove(&iq->outlist, outbound);
 	iter_handle(qstate, iq, ie, id);

iterator/iterator.h

@@ -218,6 +218,13 @@ struct iter_qstate {
 	 */
 	struct delegpt* dp;
 
+	/** state for 0x20 fallback when capsfail happens, 0 not a fallback */
+	int caps_fallback;
+	/** state for capsfail: current server number to try */
+	size_t caps_server;
+	/** state for capsfail: stored query for comparisons */
+	struct reply_info* caps_reply;
+
 	/** Current delegation message - returned for non-RD queries */
 	struct dns_msg* deleg_msg;
 

libunbound/libworker.c

@@ -695,7 +695,7 @@ libworker_handle_reply(struct comm_point* c, void* arg, int error,
 	e.qsent = NULL;
 
 	if(error != 0) {
-		mesh_report_reply(lw->env->mesh, &e, 0, reply_info);
+		mesh_report_reply(lw->env->mesh, &e, reply_info, error);
 		return 0;
 	}
 	/* sanity check. */
@@ -705,10 +705,11 @@ libworker_handle_reply(struct comm_point* c, void* arg, int error,
 		|| LDNS_QDCOUNT(ldns_buffer_begin(c->buffer)) > 1) {
 		/* error becomes timeout for the module as if this reply
 		 * never arrived. */
-		mesh_report_reply(lw->env->mesh, &e, 0, reply_info);
+		mesh_report_reply(lw->env->mesh, &e, reply_info, 
+			NETEVENT_TIMEOUT);
 		return 0;
 	}
-	mesh_report_reply(lw->env->mesh, &e, 1, reply_info);
+	mesh_report_reply(lw->env->mesh, &e, reply_info, NETEVENT_NOERROR);
 	return 0;
 }
 
@@ -720,7 +721,7 @@ libworker_handle_service_reply(struct comm_point* c, void* arg, int error,
 	struct libworker* lw = (struct libworker*)e->qstate->env->worker;
 
 	if(error != 0) {
-		mesh_report_reply(lw->env->mesh, e, 0, reply_info);
+		mesh_report_reply(lw->env->mesh, e, reply_info, error);
 		return 0;
 	}
 	/* sanity check. */
@@ -730,10 +731,11 @@ libworker_handle_service_reply(struct comm_point* c, void* arg, int error,
 		|| LDNS_QDCOUNT(ldns_buffer_begin(c->buffer)) > 1) {
 		/* error becomes timeout for the module as if this reply
 		 * never arrived. */
-		mesh_report_reply(lw->env->mesh, e, 0, reply_info);
+		mesh_report_reply(lw->env->mesh, e, reply_info, 
+			NETEVENT_TIMEOUT);
 		return 0;
 	}
-	mesh_report_reply(lw->env->mesh,  e, 1, reply_info);
+	mesh_report_reply(lw->env->mesh,  e, reply_info, NETEVENT_NOERROR);
 	return 0;
 }
 

services/mesh.c

@@ -363,11 +363,16 @@ mesh_new_callback(struct mesh_area* mesh, struct query_info* qinfo,
 }
 
 void mesh_report_reply(struct mesh_area* mesh, struct outbound_entry* e,
-        int is_ok, struct comm_reply* reply)
+        struct comm_reply* reply, int what)
 {
+	enum module_ev event = module_event_reply;
 	e->qstate->reply = reply;
-	mesh_run(mesh, e->qstate->mesh_info,
-		is_ok?module_event_reply:module_event_noreply, e);
+	if(what != NETEVENT_NOERROR) {
+		event = module_event_noreply;
+		if(what == NETEVENT_CAPSFAIL)
+			event = module_event_capsfail;
+	}
+	mesh_run(mesh, e->qstate->mesh_info, event, e);
 }
 
 struct mesh_state* 

services/mesh.h

@@ -289,11 +289,11 @@ int mesh_new_callback(struct mesh_area* mesh, struct query_info* qinfo,
  *
  * @param mesh: the query mesh.
  * @param e: outbound entry, with query state to run and reply pointer.
- * @param is_ok: if true, reply is OK, otherwise a timeout happened.
  * @param reply: the comm point reply info.
+ * @param what: NETEVENT_* error code (if not 0, what is wrong, TIMEOUT).
  */
 void mesh_report_reply(struct mesh_area* mesh, struct outbound_entry* e,
-	int is_ok, struct comm_reply* reply);
+	struct comm_reply* reply, int what);
 
 /* ------------------- Functions for module environment --------------- */
 

services/outside_network.c

@@ -1284,22 +1284,23 @@ serviced_callbacks(struct serviced_query* sq, int error, struct comm_point* c,
 				== LDNS_RCODE_NOERROR || 
 			 LDNS_RCODE_WIRE(ldns_buffer_begin(c->buffer))
 				== LDNS_RCODE_NXDOMAIN)) {
-			verbose(VERB_OPS, "no qname in reply to check 0x20ID");
-			log_addr(VERB_OPS, "from server", 
+			verbose(VERB_DETAIL, "no qname in reply to check 0x20ID");
+			log_addr(VERB_DETAIL, "from server", 
 				&sq->addr, sq->addrlen);
-			log_buf(VERB_OPS, "for packet", c->buffer);
+			log_buf(VERB_DETAIL, "for packet", c->buffer);
 			error = NETEVENT_CLOSED;
 			c = NULL;
 		} else if(ldns_buffer_read_u16_at(c->buffer, 4) > 0 &&
 			!serviced_check_qname(c->buffer, sq->qbuf, 
 			sq->qbuflen)) {
-			verbose(VERB_OPS, "wrong 0x20-ID in reply qname, "
-				"answer dropped");
-			log_addr(VERB_OPS, "from server", 
+			verbose(VERB_DETAIL, "wrong 0x20-ID in reply qname");
+			log_addr(VERB_DETAIL, "from server", 
 				&sq->addr, sq->addrlen);
-			log_buf(VERB_OPS, "for packet", c->buffer);
-			error = NETEVENT_CLOSED;
-			c = NULL;
+			log_buf(VERB_DETAIL, "for packet", c->buffer);
+			error = NETEVENT_CAPSFAIL;
+			/* and cleanup too */
+			pkt_dname_tolower(c->buffer, 
+				ldns_buffer_at(c->buffer, 12));
 		} else {
 			verbose(VERB_ALGO, "good 0x20-ID in reply qname");
 			/* cleanup caps, prettier cache contents. */

util/module.c

@@ -62,6 +62,7 @@ strmodulevent(enum module_ev e)
 	case module_event_pass: return "module_event_pass";
 	case module_event_reply: return "module_event_reply";
 	case module_event_noreply: return "module_event_noreply";
+	case module_event_capsfail: return "module_event_capsfail";
 	case module_event_moddone: return "module_event_moddone";
 	case module_event_error: return "module_event_error";
 	}

util/module.h

@@ -239,6 +239,8 @@ enum module_ev {
 	module_event_reply,
 	/** no reply, timeout or other error */
 	module_event_noreply,
+	/** reply is there, but capitalisation check failed */
+	module_event_capsfail,
 	/** next module is done, and its reply is awaiting you */
 	module_event_moddone,
 	/** error */

util/netevent.h

@@ -80,6 +80,8 @@ typedef int comm_point_callback_t(struct comm_point*, void*, int,
 #define NETEVENT_CLOSED -1
 /** to pass timeout happened to callback function */
 #define NETEVENT_TIMEOUT -2 
+/** to pass fallback from capsforID to callback function; 0x20 failed */
+#define NETEVENT_CAPSFAIL -3
 
 /**
  * A communication point dispatcher. Thread specific.