Introduce a per prefix weight. The weight is used to tip prefixes with equal

long AS pathes in one or the other direction.  It weights a prefix at a very
late stage in the decision process. This is a nice bgpd feature to traffic
engineer networks where most AS pathes are equally long.
OK henning@

8 files changed, 101 insertions, 14 deletions

diff --git a/usr.sbin/bgpd/bgpd.conf.5 b/usr.sbin/bgpd/bgpd.conf.5
index 0db4f1bdafa..eb3ab76300a 100644
--- a/usr.sbin/bgpd/bgpd.conf.5
+++ b/usr.sbin/bgpd/bgpd.conf.5
@@ -1,4 +1,4 @@
-.\" $OpenBSD: bgpd.conf.5,v 1.47 2005/04/12 14:22:46 claudio Exp $
+.\" $OpenBSD: bgpd.conf.5,v 1.48 2005/04/12 14:32:00 claudio Exp $
 .\"
 .\" Copyright (c) 2004 Claudio Jeker <claudio@openbsd.org>
 .\" Copyright (c) 2003, 2004 Henning Brauer <henning@openbsd.org>
@@ -892,6 +892,23 @@ Prepend the local AS
 .Ar number
 times to the
 .Em AS path .
+.Pp
+.It Ic weight Ar number
+The
+.Em weight
+is used to tip prefixes with equal long AS pathes in one or
+the other direction.
+It weights a prefix at a very late stage in the decision process.
+If
+.Ar number
+starts with a plus or minus sign, the
+.Em weight
+will be adjusted by adding or subtracting
+.Ar number ;
+otherwise it will be set to
+.Ar number .
+.Em Weight
+is a local non-transitive attribute and a bgpd specific extension.
 .El
 .Sh FILES
 .Bl -tag -width "/etc/bgpd.conf" -compact
diff --git a/usr.sbin/bgpd/bgpd.h b/usr.sbin/bgpd/bgpd.h
index a1fd5298bc4..909f7fe5c0b 100644
--- a/usr.sbin/bgpd/bgpd.h
+++ b/usr.sbin/bgpd/bgpd.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: bgpd.h,v 1.162 2005/03/28 15:16:46 henning Exp $ */
+/*	$OpenBSD: bgpd.h,v 1.163 2005/04/12 14:32:00 claudio Exp $ */
 
 /*
  * Copyright (c) 2003, 2004 Henning Brauer <henning@openbsd.org>
@@ -544,6 +544,8 @@ enum action_types {
 	ACTION_SET_RELATIVE_LOCALPREF,
 	ACTION_SET_MED,
 	ACTION_SET_RELATIVE_MED,
+	ACTION_SET_WEIGHT,
+	ACTION_SET_RELATIVE_WEIGHT,
 	ACTION_SET_PREPEND_SELF,
 	ACTION_SET_PREPEND_PEER,
 	ACTION_SET_NEXTHOP,
diff --git a/usr.sbin/bgpd/parse.y b/usr.sbin/bgpd/parse.y
index 0eddbd03158..a2d47a30084 100644
--- a/usr.sbin/bgpd/parse.y
+++ b/usr.sbin/bgpd/parse.y
@@ -1,4 +1,4 @@
-/*	$OpenBSD: parse.y,v 1.157 2005/04/12 14:26:58 claudio Exp $ */
+/*	$OpenBSD: parse.y,v 1.158 2005/04/12 14:32:00 claudio Exp $ */
 
 /*
  * Copyright (c) 2002, 2003, 2004 Henning Brauer <henning@openbsd.org>
@@ -152,7 +152,7 @@ typedef struct {
 %token	FROM TO ANY
 %token	PREFIX PREFIXLEN SOURCEAS TRANSITAS COMMUNITY
 %token	SET LOCALPREF MED METRIC NEXTHOP REJECT BLACKHOLE NOMODIFY
-%token	PREPEND_SELF PREPEND_PEER PFTABLE
+%token	PREPEND_SELF PREPEND_PEER PFTABLE WEIGHT
 %token	ERROR
 %token	IPSEC ESP AH SPI IKE
 %token	<v.string>		STRING
@@ -1206,6 +1206,32 @@ filter_set_opt	: LOCALPREF number		{
 			$$->type = ACTION_SET_RELATIVE_MED;
 			$$->action.relative = -$3;
 		}
+		| WEIGHT number				{
+			if (($$ = calloc(1, sizeof(struct filter_set))) == NULL)
+				fatal(NULL);
+			$$->type = ACTION_SET_WEIGHT;
+			$$->action.metric = $2;
+		}
+		| WEIGHT '+' number			{
+			if ($3 > INT_MAX) {
+				yyerror("weight to big: max %u", INT_MAX);
+				YYERROR;
+			}
+			if (($$ = calloc(1, sizeof(struct filter_set))) == NULL)
+				fatal(NULL);
+			$$->type = ACTION_SET_RELATIVE_WEIGHT;
+			$$->action.metric = $3;
+		}
+		| WEIGHT '-' number			{
+			if ($3 > INT_MAX) {
+				yyerror("weight to small: min -%u", INT_MAX);
+				YYERROR;
+			}
+			if (($$ = calloc(1, sizeof(struct filter_set))) == NULL)
+				fatal(NULL);
+			$$->type = ACTION_SET_RELATIVE_WEIGHT;
+			$$->action.relative = -$3;
+		}
 		| NEXTHOP address		{
 			if (($$ = calloc(1, sizeof(struct filter_set))) == NULL)
 				fatal(NULL);
@@ -1417,7 +1443,8 @@ lookup(char *s)
 		{ "tcp",		TCP},
 		{ "to",			TO},
 		{ "transit-as",		TRANSITAS},
-		{ "transparent-as",	TRANSPARENT}
+		{ "transparent-as",	TRANSPARENT},
+		{ "weight",		WEIGHT}
 	};
 	const struct keywords	*p;
 
diff --git a/usr.sbin/bgpd/printconf.c b/usr.sbin/bgpd/printconf.c
index a5b90a16273..2464a9faa54 100644
--- a/usr.sbin/bgpd/printconf.c
+++ b/usr.sbin/bgpd/printconf.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: printconf.c,v 1.38 2005/04/07 23:45:21 henning Exp $	*/
+/*	$OpenBSD: printconf.c,v 1.39 2005/04/12 14:32:00 claudio Exp $	*/
 
 /*
  * Copyright (c) 2003, 2004 Henning Brauer <henning@openbsd.org>
@@ -95,6 +95,12 @@ print_set(struct filter_set_head *set)
 		case ACTION_SET_RELATIVE_MED:
 			printf("metric %+d ", s->action.relative);
 			break;
+		case ACTION_SET_WEIGHT:
+			printf("weight %u ", s->action.metric);
+			break;
+		case ACTION_SET_RELATIVE_WEIGHT:
+			printf("weight %+d ", s->action.relative);
+			break;
 		case ACTION_SET_NEXTHOP:
 			printf("nexthop %s ", log_addr(&s->action.nexthop));
 			break;
diff --git a/usr.sbin/bgpd/rde.h b/usr.sbin/bgpd/rde.h
index c61a6523c7b..dc09c897489 100644
--- a/usr.sbin/bgpd/rde.h
+++ b/usr.sbin/bgpd/rde.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rde.h,v 1.63 2005/03/11 12:54:20 claudio Exp $ */
+/*	$OpenBSD: rde.h,v 1.64 2005/04/12 14:32:00 claudio Exp $ */
 
 /*
  * Copyright (c) 2003, 2004 Claudio Jeker <claudio@openbsd.org> and
@@ -161,6 +161,7 @@ struct rde_aspath {
 	char				 pftable[PFTABLE_LEN];
 	u_int32_t			 med;		/* multi exit disc */
 	u_int32_t			 lpref;		/* local pref */
+	u_int32_t			 weight;	/* low prio lpref */
 	u_int16_t			 flags;	/* internally used */
 	u_int16_t			 prefix_cnt; /* # of prefixes */
 	u_int16_t			 active_cnt; /* # of active prefixes */
diff --git a/usr.sbin/bgpd/rde_decide.c b/usr.sbin/bgpd/rde_decide.c
index 6f3a7116df0..98117ae4ae0 100644
--- a/usr.sbin/bgpd/rde_decide.c
+++ b/usr.sbin/bgpd/rde_decide.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rde_decide.c,v 1.40 2004/11/11 10:35:15 claudio Exp $ */
+/*	$OpenBSD: rde_decide.c,v 1.41 2005/04/12 14:32:01 claudio Exp $ */
 
 /*
  * Copyright (c) 2003, 2004 Claudio Jeker <claudio@openbsd.org>
@@ -154,21 +154,31 @@ prefix_cmp(struct prefix *p1, struct prefix *p2)
 			return -1;
 	}
 
-	/* 7. nexthop costs. NOT YET -> IGNORE */
+	/*
+	 * 7. local tie-breaker, this weight is here to tip equal long AS
+	 * pathes in one or the other direction. It happens more and more
+	 * that AS pathes are equally long and so traffic engineering needs
+	 * a metric that weights a prefix at a very late stage in the
+	 * decision process.
+	 */
+	if ((asp1->weight - asp2->weight) != 0)
+		return (asp1->weight - asp2->weight);
+
+	/* 8. nexthop costs. NOT YET -> IGNORE */
 
 	/*
-	 * 8. older route (more stable) wins but only if route-age
+	 * 9. older route (more stable) wins but only if route-age
 	 * evaluation is enabled.
 	 */
 	if (rde_decisionflags() & BGPD_FLAG_DECISION_ROUTEAGE)
 		if ((p2->lastchange - p1->lastchange) != 0)
 			return (p2->lastchange - p1->lastchange);
 
-	/* 9. lowest BGP Id wins */
+	/* 10. lowest BGP Id wins */
 	if ((p2->peer->remote_bgpid - p1->peer->remote_bgpid) != 0)
 		return (p2->peer->remote_bgpid - p1->peer->remote_bgpid);
 
-	/* 10. lowest peer address wins (IPv4 is better than IPv6) */
+	/* 11. lowest peer address wins (IPv4 is better than IPv6) */
 	if (memcmp(&p1->peer->remote_addr, &p2->peer->remote_addr,
 	    sizeof(p1->peer->remote_addr)) != 0)
 		return (-memcmp(&p1->peer->remote_addr, &p2->peer->remote_addr,
diff --git a/usr.sbin/bgpd/rde_filter.c b/usr.sbin/bgpd/rde_filter.c
index 65b48ff379d..756afc2afdd 100644
--- a/usr.sbin/bgpd/rde_filter.c
+++ b/usr.sbin/bgpd/rde_filter.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rde_filter.c,v 1.26 2005/03/14 17:32:04 claudio Exp $ */
+/*	$OpenBSD: rde_filter.c,v 1.27 2005/04/12 14:32:01 claudio Exp $ */
 
 /*
  * Copyright (c) 2004 Claudio Jeker <claudio@openbsd.org>
@@ -131,6 +131,24 @@ rde_apply_set(struct rde_aspath *asp, struct filter_set_head *sh,
 					asp->med += set->action.relative;
 			}
 			break;
+		case ACTION_SET_WEIGHT:
+			asp->weight = set->action.metric;
+			break;
+		case ACTION_SET_RELATIVE_WEIGHT:
+			if (set->action.relative > 0) {
+				if (set->action.relative + asp->weight <
+				    asp->weight)
+					asp->weight = UINT_MAX;
+				else
+					asp->weight += set->action.relative;
+			} else {
+				if ((u_int32_t)-set->action.relative >
+				    asp->weight)
+					asp->weight = 0;
+				else
+					asp->weight += set->action.relative;
+			}
+			break;
 		case ACTION_SET_PREPEND_SELF:
 			/* don't apply if this is a incoming default override */
 			if (dir == DIR_DEFAULT_IN)
diff --git a/usr.sbin/bgpd/rde_rib.c b/usr.sbin/bgpd/rde_rib.c
index b45f4e4aa5e..6a47462580e 100644
--- a/usr.sbin/bgpd/rde_rib.c
+++ b/usr.sbin/bgpd/rde_rib.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rde_rib.c,v 1.65 2005/03/26 12:46:52 claudio Exp $ */
+/*	$OpenBSD: rde_rib.c,v 1.66 2005/04/12 14:32:01 claudio Exp $ */
 
 /*
  * Copyright (c) 2003, 2004 Claudio Jeker <claudio@openbsd.org>
@@ -128,6 +128,10 @@ path_compare(struct rde_aspath *a, struct rde_aspath *b)
 		return (1);
 	if (a->lpref < b->lpref)
 		return (-1);
+	if (a->weight > b->weight)
+		return (1);
+	if (a->weight < b->weight)
+		return (-1);
 
 	r = strcmp(a->pftable, b->pftable);
 	if (r == 0)
@@ -266,6 +270,7 @@ path_copy(struct rde_aspath *asp)
 	nasp->nexthop = asp->nexthop;
 	nasp->med = asp->med;
 	nasp->lpref = asp->lpref;
+	nasp->weight = asp->weight;
 	nasp->origin = asp->origin;
 
 	nasp->flags = asp->flags & ~F_ATTR_LINKED;
@@ -290,6 +295,7 @@ path_get(void)
 	asp->origin = ORIGIN_INCOMPLETE;
 	asp->lpref = DEFAULT_LPREF;
 	/* med = 0 */
+	/* weight = 0 */
 
 	return (asp);
 }