- Applied patch from Daisuke HIGASHI for rrset-roundrobin and

minimal-responses features.


git-svn-id: file:///svn/unbound/trunk@2658 be551aaa-1e26-0410-a405-d3ace91eadb9

doc/CREDITS

@@ -18,3 +18,4 @@ cz.nic - sponsoring 'summer of code' development by Zdenek and Marek.
 Brett Carr - windows beta testing.
 Luca Bruno - patch for windows support in libunbound hosts and resolvconf().
 Tom Hendrikx - contributed split-itar.sh a useful script to 5011-track ITAR.
+Daisuke HIGASHI - patch for rrset-roundrobin and minimal-responses.

doc/Changelog

@@ -1,3 +1,7 @@
+10 April 2012: Wouter
+	- Applied patch from Daisuke HIGASHI for rrset-roundrobin and
+	  minimal-responses features.
+
 5 April 2012: Wouter
 	- fix bug #443: --with-chroot-dir not honoured by configure.
 	- fix bug #444: setusercontext was called too late (thanks Bjorn

doc/example.conf.in

@@ -309,6 +309,13 @@ server:
 	# if yes, perform key lookups adjacent to normal lookups.
 	# prefetch-key: no
 
+	# if yes, Unbound rotates RRSet order in response.
+	# rrset-roundrobin: no
+
+	# if yes, Unbound doesn't insert authority/additional sections
+	# into response messages when those sections are not required.
+	# minimal-responses: no
+
 	# module configuration of the server. A string with identifiers
 	# separated by spaces. "iterator" or "validator iterator"
 	# module-config: "validator iterator"

doc/unbound.conf.5.in

@@ -570,6 +570,18 @@ If yes, fetch the DNSKEYs earlier in the validation process, when a DS
 record is encountered.  This lowers the latency of requests.  It does use
 a little more CPU.  Also if the cache is set to 0, it is no use. Default is no.
 .TP
+.B rrset-roundrobin: \fI<yes or no>
+If yes, Unbound rotates RRSet order in response (the random number is taken
+from the query ID, for speed and thread safety).  Default is no.
+.TP
+.B minimal-responses: \fI<yes or no>
+If yes, Unbound doesn't insert authority/additional sections into response
+messages when those sections are not required.  This reduces response
+size significantly, and may avoid TCP fallback for some responses.
+This may cause a slight speedup.  The default is no, because the DNS
+protocol RFCs mandate these sections, and the additional content could
+be of use and save roundtrips for clients.
+.TP
 .B module\-config: \fI<"module names">
 Module configuration, a list of module names separated by spaces, surround
 the string with quotes (""). The modules can be validator, iterator.

util/config_file.c

@@ -195,6 +195,8 @@ config_create(void)
 	cfg->remote_control_enable = 0;
 	cfg->control_ifs = NULL;
 	cfg->control_port = UNBOUND_CONTROL_PORT;
+	cfg->minimal_responses = 0;
+	cfg->rrset_roundrobin = 0;
 	if(!(cfg->server_key_file = strdup(RUN_DIR"/unbound_server.key"))) 
 		goto error_exit;
 	if(!(cfg->server_cert_file = strdup(RUN_DIR"/unbound_server.pem"))) 
@@ -399,6 +401,8 @@ int config_set_option(struct config_file* cfg, const char* opt,
 	else S_MEMSIZE("key-cache-size:", key_cache_size)
 	else S_POW2("key-cache-slabs:", key_cache_slabs)
 	else S_MEMSIZE("neg-cache-size:", neg_cache_size)
+	else S_YNO("minimal-responses:", minimal_responses)
+	else S_YNO("rrset-roundrobin:", rrset_roundrobin)
 	else S_STRLIST("local-data:", local_data)
 	else S_YNO("control-enable:", remote_control_enable)
 	else S_STRLIST("control-interface:", control_ifs)
@@ -651,6 +655,8 @@ config_get_option(struct config_file* cfg, const char* opt,
 	else O_LST(opt, "control-interface", control_ifs)
 	else O_LST(opt, "domain-insecure", domain_insecure)
 	else O_UNS(opt, "val-override-date", val_date_override)
+	else O_YNO(opt, "minimal-responses", minimal_responses)
+	else O_YNO(opt, "rrset-roundrobin", rrset_roundrobin)
 	/* not here:
 	 * outgoing-permit, outgoing-avoid - have list of ports
 	 * local-zone - zones and nodefault variables
@@ -1079,6 +1085,8 @@ config_apply(struct config_file* config)
 	MAX_TTL = (uint32_t)config->max_ttl;
 	MIN_TTL = (uint32_t)config->min_ttl;
 	EDNS_ADVERTISED_SIZE = (uint16_t)config->edns_buffer_size;
+	MINIMAL_RESPONSES = config->minimal_responses;
+	RRSET_ROUNDROBIN = config->rrset_roundrobin;
 	log_set_time_asc(config->log_time_ascii);
 }
 

util/config_file.h

@@ -290,6 +290,12 @@ struct config_file {
 
 	/** daemonize, i.e. fork into the background. */
 	int do_daemonize;
+
+	/* minimal response when positive answer */
+	int minimal_responses;
+
+	/* RRSet roundrobin */
+	int rrset_roundrobin;
 };
 
 /**

util/configlexer.lex

@@ -241,6 +241,8 @@ control-cert-file{COLON}	{ YDVAR(1, VAR_CONTROL_CERT_FILE) }
 python-script{COLON}		{ YDVAR(1, VAR_PYTHON_SCRIPT) }
 python{COLON}			{ YDVAR(0, VAR_PYTHON) }
 domain-insecure{COLON}		{ YDVAR(1, VAR_DOMAIN_INSECURE) }
+minimal-responses{COLON}	{ YDVAR(1, VAR_MINIMAL_RESPONSES) }
+rrset-roundrobin{COLON}		{ YDVAR(1, VAR_RRSET_ROUNDROBIN) }
 <INITIAL,val>{NEWLINE}		{ LEXOUT(("NL\n")); cfg_parser->line++; }
 
 	/* Quoted strings. Strip leading and ending quotes */

util/configparser.h

@@ -161,7 +161,9 @@
      VAR_SSL_SERVICE_PEM = 379,
      VAR_SSL_PORT = 380,
      VAR_FORWARD_FIRST = 381,
-     VAR_STUB_FIRST = 382
+     VAR_STUB_FIRST = 382,
+     VAR_MINIMAL_RESPONSES = 383,
+     VAR_RRSET_ROUNDROBIN = 384
    };
 #endif
 /* Tokens.  */
@@ -290,6 +292,8 @@
 #define VAR_SSL_PORT 380
 #define VAR_FORWARD_FIRST 381
 #define VAR_STUB_FIRST 382
+#define VAR_MINIMAL_RESPONSES 383
+#define VAR_RRSET_ROUNDROBIN 384
 
 
 
@@ -306,7 +310,7 @@ typedef union YYSTYPE
 
 
 /* Line 2068 of yacc.c  */
-#line 310 "util/configparser.h"
+#line 314 "util/configparser.h"
 } YYSTYPE;
 # define YYSTYPE_IS_TRIVIAL 1
 # define yystype YYSTYPE /* obsolescent; will be withdrawn */

util/configparser.y

@@ -104,7 +104,7 @@ extern struct config_parser_state* cfg_parser;
 %token VAR_PREFETCH_KEY VAR_SO_SNDBUF VAR_HARDEN_BELOW_NXDOMAIN
 %token VAR_IGNORE_CD_FLAG VAR_LOG_QUERIES VAR_TCP_UPSTREAM VAR_SSL_UPSTREAM
 %token VAR_SSL_SERVICE_KEY VAR_SSL_SERVICE_PEM VAR_SSL_PORT VAR_FORWARD_FIRST
-%token VAR_STUB_FIRST
+%token VAR_STUB_FIRST VAR_MINIMAL_RESPONSES VAR_RRSET_ROUNDROBIN
 
 %%
 toplevelvars: /* empty */ | toplevelvars toplevelvar ;
@@ -160,7 +160,8 @@ content_server: server_num_threads | server_verbosity | server_port |
 	server_edns_buffer_size | server_prefetch | server_prefetch_key |
 	server_so_sndbuf | server_harden_below_nxdomain | server_ignore_cd_flag |
 	server_log_queries | server_tcp_upstream | server_ssl_upstream |
-	server_ssl_service_key | server_ssl_service_pem | server_ssl_port
+	server_ssl_service_key | server_ssl_service_pem | server_ssl_port |
+	server_minimal_responses | server_rrset_roundrobin
 	;
 stubstart: VAR_STUB_ZONE
 	{
@@ -1096,6 +1097,26 @@ server_local_data_ptr: VAR_LOCAL_DATA_PTR STRING_ARG
 		}
 	}
 	;
+server_minimal_responses: VAR_MINIMAL_RESPONSES STRING_ARG
+	{
+		OUTYY(("P(server_minimal_responses:%s)\n", $2));
+		if(strcmp($2, "yes") != 0 && strcmp($2, "no") != 0)
+			yyerror("expected yes or no.");
+		else cfg_parser->cfg->minimal_responses =
+			(strcmp($2, "yes")==0);
+		free($2);
+	}
+	;
+server_rrset_roundrobin: VAR_RRSET_ROUNDROBIN STRING_ARG
+	{
+		OUTYY(("P(server_rrset_roundrobin:%s)\n", $2));
+		if(strcmp($2, "yes") != 0 && strcmp($2, "no") != 0)
+			yyerror("expected yes or no.");
+		else cfg_parser->cfg->rrset_roundrobin =
+			(strcmp($2, "yes")==0);
+		free($2);
+	}
+	;
 stub_name: VAR_NAME STRING_ARG
 	{
 		OUTYY(("P(name:%s)\n", $2));

util/data/msgencode.c

@@ -443,9 +443,9 @@ static int
 packed_rrset_encode(struct ub_packed_rrset_key* key, ldns_buffer* pkt, 
 	uint16_t* num_rrs, uint32_t timenow, struct regional* region,
 	int do_data, int do_sig, struct compress_tree_node** tree,
-	ldns_pkt_section s, uint16_t qtype, int dnssec)
+	ldns_pkt_section s, uint16_t qtype, int dnssec, size_t rr_offset)
 {
-	size_t i, owner_pos;
+	size_t i, j, owner_pos;
 	int r, owner_labs;
 	uint16_t owner_ptr = 0;
 	struct packed_rrset_data* data = (struct packed_rrset_data*)
@@ -461,26 +461,28 @@ packed_rrset_encode(struct ub_packed_rrset_key* key, ldns_buffer* pkt,
 	if(do_data) {
 		const ldns_rr_descriptor* c = type_rdata_compressable(key);
 		for(i=0; i<data->count; i++) {
+			/* rrset roundrobin */
+			j = (i + rr_offset) % data->count;
 			if((r=compress_owner(key, pkt, region, tree, 
 				owner_pos, &owner_ptr, owner_labs))
 				!= RETVAL_OK)
 				return r;
 			ldns_buffer_write(pkt, &key->rk.type, 2);
 			ldns_buffer_write(pkt, &key->rk.rrset_class, 2);
-			if(data->rr_ttl[i] < timenow)
+			if(data->rr_ttl[j] < timenow)
 				ldns_buffer_write_u32(pkt, 0);
 			else 	ldns_buffer_write_u32(pkt, 
-					data->rr_ttl[i]-timenow);
+					data->rr_ttl[j]-timenow);
 			if(c) {
-				if((r=compress_rdata(pkt, data->rr_data[i],
+				if((r=compress_rdata(pkt, data->rr_data[j],
-					data->rr_len[i], region, tree, c))
+					data->rr_len[j], region, tree, c))
 					!= RETVAL_OK)
 					return r;
 			} else {
-				if(ldns_buffer_remaining(pkt) < data->rr_len[i])
+				if(ldns_buffer_remaining(pkt) < data->rr_len[j])
 					return RETVAL_TRUNC;
-				ldns_buffer_write(pkt, data->rr_data[i], 
+				ldns_buffer_write(pkt, data->rr_data[j],
-					data->rr_len[i]);
+					data->rr_len[j]);
 			}
 		}
 	}
@@ -528,7 +530,7 @@ static int
 insert_section(struct reply_info* rep, size_t num_rrsets, uint16_t* num_rrs,
 	ldns_buffer* pkt, size_t rrsets_before, uint32_t timenow, 
 	struct regional* region, struct compress_tree_node** tree,
-	ldns_pkt_section s, uint16_t qtype, int dnssec)
+	ldns_pkt_section s, uint16_t qtype, int dnssec, size_t rr_offset)
 {
 	int r;
 	size_t i, setstart;
@@ -540,7 +542,7 @@ insert_section(struct reply_info* rep, size_t num_rrsets, uint16_t* num_rrs,
 			setstart = ldns_buffer_position(pkt);
 			if((r=packed_rrset_encode(rep->rrsets[rrsets_before+i], 
 				pkt, num_rrs, timenow, region, 1, 1, tree,
-				s, qtype, dnssec))
+				s, qtype, dnssec, rr_offset))
 				!= RETVAL_OK) {
 				/* Bad, but if due to size must set TC bit */
 				/* trim off the rrset neatly. */
@@ -553,7 +555,7 @@ insert_section(struct reply_info* rep, size_t num_rrsets, uint16_t* num_rrs,
 			setstart = ldns_buffer_position(pkt);
 			if((r=packed_rrset_encode(rep->rrsets[rrsets_before+i], 
 				pkt, num_rrs, timenow, region, 1, 0, tree,
-				s, qtype, dnssec))
+				s, qtype, dnssec, rr_offset))
 				!= RETVAL_OK) {
 				ldns_buffer_set_position(pkt, setstart);
 				return r;
@@ -564,7 +566,7 @@ insert_section(struct reply_info* rep, size_t num_rrsets, uint16_t* num_rrs,
 			setstart = ldns_buffer_position(pkt);
 			if((r=packed_rrset_encode(rep->rrsets[rrsets_before+i], 
 				pkt, num_rrs, timenow, region, 0, 1, tree,
-				s, qtype, dnssec))
+				s, qtype, dnssec, rr_offset))
 				!= RETVAL_OK) {
 				ldns_buffer_set_position(pkt, setstart);
 				return r;
@@ -595,6 +597,31 @@ insert_query(struct query_info* qinfo, struct compress_tree_node** tree,
 	return RETVAL_OK;
 }
 
+static int
+positive_answer(struct reply_info* rep, uint16_t qtype) {
+	size_t i;
+	if (FLAGS_GET_RCODE(rep->flags) != LDNS_RCODE_NOERROR)
+		return 0;
+
+	for(i=0;i<rep->an_numrrsets; i++) {
+		if(ntohs(rep->rrsets[i]->rk.type) == qtype) {
+			/* in case it is a wildcard with DNSSEC, there will
+			 * be NSEC/NSEC3 records in the authority section
+			 * that we cannot remove */
+			for(i=rep->an_numrrsets; i<rep->an_numrrsets+
+				rep->ns_numrrsets; i++) {
+				if(ntohs(rep->rrsets[i]->rk.type) ==
+					LDNS_RR_TYPE_NSEC ||
+				   ntohs(rep->rrsets[i]->rk.type) ==
+				   	LDNS_RR_TYPE_NSEC3)
+					return 0;
+			}
+			return 1;
+		}
+	}
+	return 0;
+}
+
 int 
 reply_info_encode(struct query_info* qinfo, struct reply_info* rep, 
 	uint16_t id, uint16_t flags, ldns_buffer* buffer, uint32_t timenow, 
@@ -603,6 +630,7 @@ reply_info_encode(struct query_info* qinfo, struct reply_info* rep,
 	uint16_t ancount=0, nscount=0, arcount=0;
 	struct compress_tree_node* tree = 0;
 	int r;
+	size_t rr_offset; 
 
 	ldns_buffer_clear(buffer);
 	if(udpsize < ldns_buffer_limit(buffer))
@@ -630,11 +658,13 @@ reply_info_encode(struct query_info* qinfo, struct reply_info* rep,
 			return 0;
 		}
 	}
+	/* roundrobin offset. using query id for random number */
+	rr_offset = RRSET_ROUNDROBIN?id:0;
 
 	/* insert answer section */
 	if((r=insert_section(rep, rep->an_numrrsets, &ancount, buffer, 
 		0, timenow, region, &tree, LDNS_SECTION_ANSWER, qinfo->qtype, 
-		dnssec)) != RETVAL_OK) {
+		dnssec, rr_offset)) != RETVAL_OK) {
 		if(r == RETVAL_TRUNC) {
 			/* create truncated message */
 			ldns_buffer_write_u16_at(buffer, 6, ancount);
@@ -646,35 +676,42 @@ reply_info_encode(struct query_info* qinfo, struct reply_info* rep,
 	}
 	ldns_buffer_write_u16_at(buffer, 6, ancount);
 
-	/* insert auth section */
+	/* if response is positive answer, auth/add sections are not required */
-	if((r=insert_section(rep, rep->ns_numrrsets, &nscount, buffer, 
+	if( ! (MINIMAL_RESPONSES && positive_answer(rep, qinfo->qtype)) ) {
-		rep->an_numrrsets, timenow, region, &tree,
+		/* insert auth section */
-		LDNS_SECTION_AUTHORITY, qinfo->qtype, dnssec)) != RETVAL_OK) {
+		if((r=insert_section(rep, rep->ns_numrrsets, &nscount, buffer, 
-		if(r == RETVAL_TRUNC) {
+			rep->an_numrrsets, timenow, region, &tree,
-			/* create truncated message */
+			LDNS_SECTION_AUTHORITY, qinfo->qtype,
-			ldns_buffer_write_u16_at(buffer, 8, nscount);
+			dnssec, rr_offset)) != RETVAL_OK) {
-			LDNS_TC_SET(ldns_buffer_begin(buffer));
+			if(r == RETVAL_TRUNC) {
-			ldns_buffer_flip(buffer);
+				/* create truncated message */
-			return 1;
+				ldns_buffer_write_u16_at(buffer, 8, nscount);
+				LDNS_TC_SET(ldns_buffer_begin(buffer));
+				ldns_buffer_flip(buffer);
+				return 1;
+			}
+			return 0;
 		}
-		return 0;
+		ldns_buffer_write_u16_at(buffer, 8, nscount);
-	}
-	ldns_buffer_write_u16_at(buffer, 8, nscount);
 
-	/* insert add section */
+		/* insert add section */
-	if((r=insert_section(rep, rep->ar_numrrsets, &arcount, buffer, 
+		if((r=insert_section(rep, rep->ar_numrrsets, &arcount, buffer, 
-		rep->an_numrrsets + rep->ns_numrrsets, timenow, region, 
+			rep->an_numrrsets + rep->ns_numrrsets, timenow, region, 
-		&tree, LDNS_SECTION_ADDITIONAL, qinfo->qtype, 
+			&tree, LDNS_SECTION_ADDITIONAL, qinfo->qtype, 
-		dnssec)) != RETVAL_OK) {
+			dnssec, rr_offset)) != RETVAL_OK) {
-		if(r == RETVAL_TRUNC) {
+			if(r == RETVAL_TRUNC) {
-			/* no need to set TC bit, this is the additional */
+				/* no need to set TC bit, this is the additional */
-			ldns_buffer_write_u16_at(buffer, 10, arcount);
+				ldns_buffer_write_u16_at(buffer, 10, arcount);
-			ldns_buffer_flip(buffer);
+				ldns_buffer_flip(buffer);
-			return 1;
+				return 1;
+			}
+			return 0;
 		}
-		return 0;
+		ldns_buffer_write_u16_at(buffer, 10, arcount);
+	} else {
+		ldns_buffer_write_u16_at(buffer, 8, nscount);
+		ldns_buffer_write_u16_at(buffer, 10, arcount);
 	}
-	ldns_buffer_write_u16_at(buffer, 10, arcount);
 	ldns_buffer_flip(buffer);
 	return 1;
 }

util/net_help.c

@@ -53,6 +53,12 @@
 /** default value for EDNS ADVERTISED size */
 uint16_t EDNS_ADVERTISED_SIZE = 4096;
 
+/** minimal responses when positive answer: default is no */
+int MINIMAL_RESPONSES = 0;
+
+/** rrset order roundrobin: default is no */
+int RRSET_ROUNDROBIN = 0;
+
 /* returns true is string addr is an ip6 specced address */
 int
 str_is_ip6(const char* str)

util/net_help.h

@@ -93,6 +93,12 @@ extern uint16_t EDNS_ADVERTISED_SIZE;
 /** DNSKEY secure entry point, KSK flag */
 #define DNSKEY_BIT_SEP 0x0001
 
+/** minimal responses when positive answer */
+extern int MINIMAL_RESPONSES;
+
+/** rrset order roundrobin */
+extern int RRSET_ROUNDROBIN;
+
 /**
  * See if string is ip4 or ip6.
  * @param str: IP specification.