Implemented qname minimisation

git-svn-id: file:///svn/unbound/trunk@3554 be551aaa-1e26-0410-a405-d3ace91eadb9

doc/Changelog

@@ -1,3 +1,6 @@
+30 November 2015: Ralph
+	- Implemented qname minimisation
+
 30 November 2015: Wouter
 	- Fix for #724: conf syntax to read files from run dir (on Windows).
 

doc/unbound.conf.5.in

@@ -587,6 +587,12 @@ queries.  For domains that do not support 0x20 and also fail with fallback
 because they keep sending different answers, like some load balancers.
 Can be given multiple times, for different domains.
 .TP
+.B qname\-minimisation: \fI<yes or no>
+Send minimum amount of information to upstream servers to enhance privacy.
+Only sent minimum required labels of the QNAME and set QTYPE to NS when 
+possible. Best effort approach, full QNAME and original QTYPE will be sent when
+upstream replies with a RCODE other than NOERROR. Default is off.
+.TP
 .B private\-address: \fI<IP address or subnet>
 Give IPv4 of IPv6 addresses or classless subnets. These are addresses
 on your private network, and are not allowed to be returned for

iterator/iterator.c

@@ -81,6 +81,21 @@ iter_init(struct module_env* env, int id)
 		log_err("iterator: could not apply configuration settings.");
 		return 0;
 	}
+	if(env->cfg->qname_minimisation) {
+		uint8_t dname[LDNS_MAX_DOMAINLEN+1];
+		size_t len = sizeof(dname);
+		if(sldns_str2wire_dname_buf("ip6.arpa.", dname, &len) != 0) {
+			log_err("ip6.arpa. parse error");
+			return 0;
+		}
+		iter_env->ip6arpa_dname = (uint8_t*)malloc(len);
+		if(!iter_env->ip6arpa_dname) {
+			log_err("malloc failure");
+			return 0;
+		}
+		memcpy(iter_env->ip6arpa_dname, dname, len);
+	}
+
 	return 1;
 }
 
@@ -101,6 +116,8 @@ iter_deinit(struct module_env* env, int id)
 	if(!env || !env->modinfo[id])
 		return;
 	iter_env = (struct iter_env*)env->modinfo[id];
+	if(env->cfg->qname_minimisation)
+		free(iter_env->ip6arpa_dname);
 	free(iter_env->target_fetch_policy);
 	priv_delete(iter_env->priv);
 	donotq_delete(iter_env->donotq);
@@ -145,6 +162,12 @@ iter_new(struct module_qstate* qstate, int id)
 	/* Start with the (current) qname. */
 	iq->qchase = qstate->qinfo;
 	outbound_list_init(&iq->outlist);
+	if (qstate->env->cfg->qname_minimisation)
+		iq->minimisation_state = INIT_MINIMISE_STATE;
+	else
+		iq->minimisation_state = DONOT_MINIMISE_STATE;
+	
+	memset(&iq->qinfo_out, 0, sizeof(struct query_info));
 	return 1;
 }
 
@@ -590,6 +613,11 @@ generate_sub_request(uint8_t* qname, size_t qnamelen, uint16_t qtype,
 		subiq->qchase = subq->qinfo;
 		subiq->chase_flags = subq->query_flags;
 		subiq->refetch_glue = 0;
+		if(qstate->env->cfg->qname_minimisation)
+			subiq->minimisation_state = INIT_MINIMISE_STATE;
+		else
+			subiq->minimisation_state = DONOT_MINIMISE_STATE;
+		memset(&subiq->qinfo_out, 0, sizeof(struct query_info));
 	}
 	return 1;
 }
@@ -1042,6 +1070,8 @@ processInitRequest(struct module_qstate* qstate, struct iter_qstate* iq,
 			iq->query_restart_count++;
 			iq->sent_count = 0;
 			sock_list_insert(&qstate->reply_origin, NULL, 0, qstate->region);
+			if(qstate->env->cfg->qname_minimisation)
+				iq->minimisation_state = INIT_MINIMISE_STATE;
 			return next_state(iq, INIT_REQUEST_STATE);
 		}
 
@@ -1599,6 +1629,8 @@ processLastResort(struct module_qstate* qstate, struct iter_qstate* iq,
 			iq->refetch_glue = 1;
 			iq->query_restart_count++;
 			iq->sent_count = 0;
+			if(qstate->env->cfg->qname_minimisation)
+				iq->minimisation_state = INIT_MINIMISE_STATE;
 			return next_state(iq, INIT_REQUEST_STATE);
 		}
 	}
@@ -1975,9 +2007,76 @@ processQueryTargets(struct module_qstate* qstate, struct iter_qstate* iq,
 		}
 	}
 
+	if(iq->minimisation_state == INIT_MINIMISE_STATE) {
+		/* (Re)set qinfo_out to (new) delegation point, except
+ 		 * when qinfo_out is already a subdomain op dp. This happens
+		 * when resolving ip6.arpa dnames. */
+		if(!(iq->qinfo_out.qname_len 
+			&& dname_subdomain_c(iq->qchase.qname, 
+				iq->qinfo_out.qname)
+			&& dname_subdomain_c(iq->qinfo_out.qname, 
+				iq->dp->name))) {
+			iq->qinfo_out.qname = iq->dp->name;
+			iq->qinfo_out.qname_len = iq->dp->namelen;
+			iq->qinfo_out.qtype = LDNS_RR_TYPE_NS;
+			iq->qinfo_out.qclass = iq->qchase.qclass;
+		}
+
+		iq->minimisation_state = MINIMISE_STATE;
+	}
+	if(iq->minimisation_state == MINIMISE_STATE) {
+		int labdiff = dname_count_labels(iq->qchase.qname) -
+			dname_count_labels(iq->qinfo_out.qname);
+
+		iq->qinfo_out.qname = iq->qchase.qname;
+		iq->qinfo_out.qname_len = iq->qchase.qname_len;
+
+		/* Special treatment for ip6.arpa lookups.
+		 * Reverse IPv6 dname has 34 labels, increment the IP part 
+		 * (usually first 32 labels) by 8 labels (7 more than the 
+		 * default 1 label increment). */
+		if(labdiff <= 32 &&
+			dname_subdomain_c(iq->qchase.qname, ie->ip6arpa_dname)) {
+			labdiff -= 7;
+			/* Small chance of zone cut after first label. Stop
+			 * minimising */
+			if(labdiff <= 1)
+				labdiff = 0;
+		}
+
+		if(labdiff > 1) {
+			verbose(VERB_QUERY, "removing %d labels", labdiff-1);
+			dname_remove_labels(&iq->qinfo_out.qname, 
+				&iq->qinfo_out.qname_len, 
+				labdiff-1);
+		}
+		if(labdiff < 1 || 
+			(labdiff < 2 && iq->qchase.qtype == LDNS_RR_TYPE_DS))
+			/* Stop minimising this query, resolve "as usual" */
+			iq->minimisation_state = DONOT_MINIMISE_STATE;
+		else {
+			struct dns_msg* msg = dns_cache_lookup(qstate->env, 
+				iq->qinfo_out.qname, iq->qinfo_out.qname_len, 
+				iq->qinfo_out.qtype, iq->qinfo_out.qclass, 
+				qstate->query_flags, qstate->region, 
+				qstate->env->scratch);
+			if(msg && msg->rep->an_numrrsets == 0)
+				/* no need to send query if it is already 
+				 * cached as NOERROR/NODATA */
+				return 1;
+		}
+		
+	}
+	if(iq->minimisation_state == SKIP_MINIMISE_STATE)
+		/* Do not increment qname, continue incrementing next 
+		 * iteration */
+		iq->minimisation_state = MINIMISE_STATE;
+	if(iq->minimisation_state == DONOT_MINIMISE_STATE)
+		iq->qinfo_out = iq->qchase;
+
 	/* We have a valid target. */
 	if(verbosity >= VERB_QUERY) {
-		log_query_info(VERB_QUERY, "sending query:", &iq->qchase);
+		log_query_info(VERB_QUERY, "sending query:", &iq->qinfo_out);
 		log_name_addr(VERB_QUERY, "sending to target:", iq->dp->name, 
 			&target->addr, target->addrlen);
 		verbose(VERB_ALGO, "dnssec status: %s%s",
@@ -1986,8 +2085,8 @@ processQueryTargets(struct module_qstate* qstate, struct iter_qstate* iq,
 	}
 	fptr_ok(fptr_whitelist_modenv_send_query(qstate->env->send_query));
 	outq = (*qstate->env->send_query)(
-		iq->qchase.qname, iq->qchase.qname_len, 
+		iq->qinfo_out.qname, iq->qinfo_out.qname_len, 
-		iq->qchase.qtype, iq->qchase.qclass, 
+		iq->qinfo_out.qtype, iq->qinfo_out.qclass, 
 		iq->chase_flags | (iq->chase_to_rd?BIT_RD:0), EDNS_DO|BIT_CD, 
 		iq->dnssec_expected, iq->caps_fallback || is_caps_whitelisted(
 		ie, iq), &target->addr, target->addrlen, iq->dp->name,
@@ -2042,6 +2141,9 @@ processQueryResponse(struct module_qstate* qstate, struct iter_qstate* iq,
 	enum response_type type;
 	iq->num_current_queries--;
 	if(iq->response == NULL) {
+		/* Don't increment qname when QNAME minimisation is enabled */
+		if (qstate->env->cfg->qname_minimisation)
+			iq->minimisation_state = SKIP_MINIMISE_STATE;
 		iq->chase_to_rd = 0;
 		iq->dnssec_lame_query = 0;
 		verbose(VERB_ALGO, "query response was timeout");
@@ -2142,6 +2244,15 @@ processQueryResponse(struct module_qstate* qstate, struct iter_qstate* iq,
 			sock_list_insert(&qstate->reply_origin, 
 				&qstate->reply->addr, qstate->reply->addrlen, 
 				qstate->region);
+		if(iq->minimisation_state != DONOT_MINIMISE_STATE) {
+			/* Best effort qname-minimisation. 
+			 * Stop minimising and send full query when RCODE
+			 * is not NOERROR */
+			if(FLAGS_GET_RCODE(iq->response->rep->flags) != 
+				LDNS_RCODE_NOERROR)
+				iq->minimisation_state = DONOT_MINIMISE_STATE;
+			return next_state(iq, QUERYTARGETS_STATE);
+		}
 		return final_state(iq);
 	} else if(type == RESPONSE_TYPE_REFERRAL) {
 		/* REFERRAL type responses get a reset of the 
@@ -2201,6 +2312,8 @@ processQueryResponse(struct module_qstate* qstate, struct iter_qstate* iq,
 		 * point to the referral. */
 		iq->deleg_msg = iq->response;
 		iq->dp = delegpt_from_message(iq->response, qstate->region);
+		if (qstate->env->cfg->qname_minimisation)
+			iq->minimisation_state = INIT_MINIMISE_STATE;
 		if(!iq->dp)
 			return error_response(qstate, id, LDNS_RCODE_SERVFAIL);
 		if(!cache_fill_missing(qstate->env, iq->qchase.qclass, 
@@ -2280,6 +2393,8 @@ processQueryResponse(struct module_qstate* qstate, struct iter_qstate* iq,
 		/* set the current request's qname to the new value. */
 		iq->qchase.qname = sname;
 		iq->qchase.qname_len = snamelen;
+		if (qstate->env->cfg->qname_minimisation)
+			iq->minimisation_state = INIT_MINIMISE_STATE;
 		/* Clear the query state, since this is a query restart. */
 		iq->deleg_msg = NULL;
 		iq->dp = NULL;
@@ -2353,6 +2468,8 @@ processQueryResponse(struct module_qstate* qstate, struct iter_qstate* iq,
 	/* LAME, THROWAWAY and "unknown" all end up here.
 	 * Recycle to the QUERYTARGETS state to hopefully try a 
 	 * different target. */
+	if (qstate->env->cfg->qname_minimisation)
+		iq->minimisation_state = SKIP_MINIMISE_STATE;
 	return next_state(iq, QUERYTARGETS_STATE);
 }
 
@@ -2968,7 +3085,7 @@ process_response(struct module_qstate* qstate, struct iter_qstate* iq,
 	prs->flags &= ~BIT_CD;
 
 	/* normalize and sanitize: easy to delete items from linked lists */
-	if(!scrub_message(pkt, prs, &iq->qchase, iq->dp->name, 
+	if(!scrub_message(pkt, prs, &iq->qinfo_out, iq->dp->name, 
 		qstate->env->scratch, qstate->env, ie)) {
 		/* if 0x20 enabled, start fallback, but we have no message */
 		if(event == module_event_capsfail && !iq->caps_fallback) {

iterator/iterator.h

@@ -112,6 +112,32 @@ struct iter_env {
 	 * array of max_dependency_depth+1 size.
 	 */
 	int* target_fetch_policy;
+
+	/** ip6.arpa dname in wireformat, used for qname-minimisation */
+	uint8_t* ip6arpa_dname;
+};
+
+/**
+ * QNAME minimisation state
+ */
+enum minimisation_state {
+	/**
+	 * (Re)start minimisation. Outgoing QNAME should be set to dp->name.
+	 * State entered on new query or after following refferal or CNAME.
+	 */
+	INIT_MINIMISE_STATE = 0,
+	/**
+	 * QNAME minimisataion ongoing. Increase QNAME on every iteration.
+	 */
+	MINIMISE_STATE,
+	/**
+	 * Don't increment QNAME this iteration
+	 */
+	SKIP_MINIMISE_STATE,
+	/**
+	 * Send out full QNAME + original QTYPE
+	 */
+	DONOT_MINIMISE_STATE,
 };
 
 /**
@@ -322,6 +348,15 @@ struct iter_qstate {
 
 	/** list of pending queries to authoritative servers. */
 	struct outbound_list outlist;
+
+	/** QNAME minimisation state */
+	enum minimisation_state minimisation_state;
+
+	/**
+	 * The query info that is sent upstream. Will be a subset of qchase
+	 * when qname minimisation is enabled.
+	 */
+	struct query_info qinfo_out;
 };
 
 /**

util/config_file.c

@@ -240,6 +240,7 @@ config_create(void)
 	cfg->ratelimit_for_domain = NULL;
 	cfg->ratelimit_below_domain = NULL;
 	cfg->ratelimit_factor = 10;
+	cfg->qname_minimisation = 0;
 	return cfg;
 error_exit:
 	config_delete(cfg); 
@@ -473,6 +474,7 @@ int config_set_option(struct config_file* cfg, const char* opt,
 	else S_MEMSIZE("ratelimit-size:", ratelimit_size)
 	else S_POW2("ratelimit-slabs:", ratelimit_slabs)
 	else S_NUMBER_OR_ZERO("ratelimit-factor:", ratelimit_factor)
+	else S_YNO("qname-minimisation:", qname_minimisation)
 	/* val_sig_skew_min and max are copied into val_env during init,
 	 * so this does not update val_env with set_option */
 	else if(strcmp(opt, "val-sig-skew-min:") == 0)
@@ -747,6 +749,7 @@ config_get_option(struct config_file* cfg, const char* opt,
 	else O_DEC(opt, "ratelimit-factor", ratelimit_factor)
 	else O_DEC(opt, "val-sig-skew-min", val_sig_skew_min)
 	else O_DEC(opt, "val-sig-skew-max", val_sig_skew_max)
+	else O_YNO(opt, "qname-minimisation", qname_minimisation)
 	/* not here:
 	 * outgoing-permit, outgoing-avoid - have list of ports
 	 * local-zone - zones and nodefault variables

util/config_file.h

@@ -364,6 +364,8 @@ struct config_file {
 	struct config_str2list* ratelimit_below_domain;
 	/** ratelimit factor, 0 blocks all, 10 allows 1/10 of traffic */
 	int ratelimit_factor;
+
+	int qname_minimisation;
 };
 
 /** from cfg username, after daemonise setup performed */

util/configlexer.lex

@@ -205,6 +205,7 @@ SQANY     [^\'\n\r\\]|\\.
 	/* note that flex makes the longest match and '.' is any but not nl */
 	LEXOUT(("comment(%s) ", yytext)); /* ignore */ }
 server{COLON}			{ YDVAR(0, VAR_SERVER) }
+qname-minimisation{COLON}	{ YDVAR(1, VAR_QNAME_MINIMISATION) }
 num-threads{COLON}		{ YDVAR(1, VAR_NUM_THREADS) }
 verbosity{COLON}		{ YDVAR(1, VAR_VERBOSITY) }
 port{COLON}			{ YDVAR(1, VAR_PORT) }

util/configparser.h

@@ -1,8 +1,8 @@
-/* A Bison parser, made by GNU Bison 2.7.  */
+/* A Bison parser, made by GNU Bison 2.7.12-4996.  */
 
 /* Bison interface for Yacc-like parsers in C
    
-      Copyright (C) 1984, 1989-1990, 2000-2012 Free Software Foundation, Inc.
+      Copyright (C) 1984, 1989-1990, 2000-2013 Free Software Foundation, Inc.
    
    This program is free software: you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
@@ -204,7 +204,8 @@ extern int yydebug;
      VAR_RATELIMIT_FACTOR = 413,
      VAR_CAPS_WHITELIST = 414,
      VAR_CACHE_MAX_NEGATIVE_TTL = 415,
-     VAR_PERMIT_SMALL_HOLDDOWN = 416
+     VAR_PERMIT_SMALL_HOLDDOWN = 416,
+     VAR_QNAME_MINIMISATION = 417
    };
 #endif
 /* Tokens.  */
@@ -367,20 +368,21 @@ extern int yydebug;
 #define VAR_CAPS_WHITELIST 414
 #define VAR_CACHE_MAX_NEGATIVE_TTL 415
 #define VAR_PERMIT_SMALL_HOLDDOWN 416
+#define VAR_QNAME_MINIMISATION 417
 
 
 
 #if ! defined YYSTYPE && ! defined YYSTYPE_IS_DECLARED
 typedef union YYSTYPE
 {
-/* Line 2058 of yacc.c  */
+/* Line 2053 of yacc.c  */
 #line 64 "./util/configparser.y"
 
 	char*	str;
 
 
-/* Line 2058 of yacc.c  */
+/* Line 2053 of yacc.c  */
-#line 384 "util/configparser.h"
+#line 386 "util/configparser.h"
 } YYSTYPE;
 # define YYSTYPE_IS_TRIVIAL 1
 # define yystype YYSTYPE /* obsolescent; will be withdrawn */

util/configparser.y

@@ -122,6 +122,7 @@ extern struct config_parser_state* cfg_parser;
 %token VAR_RATELIMIT VAR_RATELIMIT_SLABS VAR_RATELIMIT_SIZE
 %token VAR_RATELIMIT_FOR_DOMAIN VAR_RATELIMIT_BELOW_DOMAIN VAR_RATELIMIT_FACTOR
 %token VAR_CAPS_WHITELIST VAR_CACHE_MAX_NEGATIVE_TTL VAR_PERMIT_SMALL_HOLDDOWN
+%token VAR_QNAME_MINIMISATION
 
 %%
 toplevelvars: /* empty */ | toplevelvars toplevelvar ;
@@ -186,7 +187,7 @@ content_server: server_num_threads | server_verbosity | server_port |
 	server_ratelimit_size | server_ratelimit_for_domain |
 	server_ratelimit_below_domain | server_ratelimit_factor |
 	server_caps_whitelist | server_cache_max_negative_ttl |
-	server_permit_small_holddown
+	server_permit_small_holddown | server_qname_minimisation
 	;
 stubstart: VAR_STUB_ZONE
 	{
@@ -1318,6 +1319,16 @@ server_ratelimit_factor: VAR_RATELIMIT_FACTOR STRING_ARG
 		free($2);
 	}
 	;
+server_qname_minimisation: VAR_QNAME_MINIMISATION STRING_ARG
+	{
+		OUTYY(("P(server_qname_minimisation:%s)\n", $2));
+		if(strcmp($2, "yes") != 0 && strcmp($2, "no") != 0)
+			yyerror("expected yes or no.");
+		else cfg_parser->cfg->qname_minimisation = 
+			(strcmp($2, "yes")==0);
+		free($2);
+	}
+	;
 stub_name: VAR_NAME STRING_ARG
 	{
 		OUTYY(("P(name:%s)\n", $2));