Don't import/export MED and LOCAL_PREF on external links.

Added real comparison of BGP routes (inspired by the Cisco one).
Default local preference and default MED are now settable.
Defined filter keywords for all BGP attributes we know.

3 files changed, 88 insertions, 50 deletions

diff --git a/proto/bgp/attrs.c b/proto/bgp/attrs.c
index f330483..22c73ca 100644
--- a/proto/bgp/attrs.c
+++ b/proto/bgp/attrs.c
@@ -29,6 +29,7 @@ struct attr_desc {
   int expected_length;
   int expected_flags;
   int type;
+  int allow_in_ebgp;
   int (*validate)(struct bgp_proto *p, byte *attr, int len);
   void (*format)(eattr *ea, byte *buf);
 };
@@ -78,32 +79,24 @@ bgp_check_next_hop(struct bgp_proto *p, byte *a, int len)
     return 8;
 }
 
-static int
-bgp_check_local_pref(struct bgp_proto *p, byte *a, int len)
-{
-  if (!p->is_internal)			/* Ignore local preference from EBGP connections */
-    return -1;
-  return 0;
-}
-
 static struct attr_desc bgp_attr_table[] = {
-  { NULL, -1, 0, 0,						/* Undefined */
+  { NULL, -1, 0, 0, 0,						/* Undefined */
     NULL, NULL },
-  { "origin", 1, BAF_TRANSITIVE, EAF_TYPE_INT,			/* BA_ORIGIN */
+  { "origin", 1, BAF_TRANSITIVE, EAF_TYPE_INT, 1,		/* BA_ORIGIN */
     bgp_check_origin, bgp_format_origin },
-  { "as_path", -1, BAF_TRANSITIVE, EAF_TYPE_AS_PATH,		/* BA_AS_PATH */
+  { "as_path", -1, BAF_TRANSITIVE, EAF_TYPE_AS_PATH, 1,		/* BA_AS_PATH */
     bgp_check_path, NULL },
-  { "next_hop", 4, BAF_TRANSITIVE, EAF_TYPE_IP_ADDRESS,		/* BA_NEXT_HOP */
+  { "next_hop", 4, BAF_TRANSITIVE, EAF_TYPE_IP_ADDRESS, 1,	/* BA_NEXT_HOP */
     bgp_check_next_hop, NULL },
-  { "MED", 4, BAF_OPTIONAL, EAF_TYPE_INT,			/* BA_MULTI_EXIT_DISC */
+  { "MED", 4, BAF_OPTIONAL, EAF_TYPE_INT, 0,			/* BA_MULTI_EXIT_DISC */
     NULL, NULL },
-  { "local_pref", 4, BAF_OPTIONAL, EAF_TYPE_INT,		/* BA_LOCAL_PREF */
-    bgp_check_local_pref, NULL },
-  { "atomic_aggr", 0, BAF_OPTIONAL, EAF_TYPE_OPAQUE,		/* BA_ATOMIC_AGGR */
+  { "local_pref", 4, BAF_OPTIONAL, EAF_TYPE_INT, 0,		/* BA_LOCAL_PREF */
     NULL, NULL },
-  { "aggregator", 6, BAF_OPTIONAL, EAF_TYPE_OPAQUE,		/* BA_AGGREGATOR */
+  { "atomic_aggr", 0, BAF_OPTIONAL, EAF_TYPE_OPAQUE, 1,		/* BA_ATOMIC_AGGR */
     NULL, NULL },
-  { "community", -1, BAF_OPTIONAL | BAF_TRANSITIVE, EAF_TYPE_INT_SET,	/* BA_COMMUNITY */
+  { "aggregator", 6, BAF_OPTIONAL, EAF_TYPE_OPAQUE, 1,		/* BA_AGGREGATOR */
+    NULL, NULL },
+  { "community", -1, BAF_OPTIONAL | BAF_TRANSITIVE, EAF_TYPE_INT_SET, 1,  /* BA_COMMUNITY */
     NULL, NULL },
 #if 0
   { 0, 0 },									/* BA_ORIGINATOR_ID */
@@ -308,10 +301,9 @@ static struct bgp_bucket *
 bgp_get_bucket(struct bgp_proto *p, ea_list *old, ea_list *tmp, int originate)
 {
   ea_list *t, *new;
-  unsigned i, cnt;
+  unsigned i, cnt, hash, code;
   eattr *a, *d;
   u32 seen = 0;
-  unsigned hash;
   struct bgp_bucket *b;
 
   /* Merge the attribute lists */
@@ -339,8 +331,14 @@ bgp_get_bucket(struct bgp_proto *p, ea_list *old, ea_list *tmp, int originate)
 #endif
       if (EA_PROTO(a->id) != EAP_BGP)
 	continue;
-      if (EA_ID(a->id) < 32)
-	seen |= 1 << EA_ID(a->id);
+      code = EA_ID(a->id);
+      if (code < ARRAY_SIZE(bgp_attr_table))
+	{
+	  if (!bgp_attr_table[code].allow_in_ebgp && !p->is_internal)
+	    continue;
+	}
+      if (code < 32)
+	seen |= 1 << code;
       *d = *a;
       if ((d->type & EAF_ORIGINATED) && !originate && (d->flags & BAF_TRANSITIVE) && (d->flags & BAF_OPTIONAL))
 	d->flags |= BAF_PARTIAL;
@@ -562,20 +560,53 @@ bgp_rte_better(rte *new, rte *old)
 {
   struct bgp_proto *new_bgp = (struct bgp_proto *) new->attrs->proto;
   struct bgp_proto *old_bgp = (struct bgp_proto *) old->attrs->proto;
-  eattr *new_lpref = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_LOCAL_PREF));
-  eattr *old_lpref = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_LOCAL_PREF));
+  eattr *x, *y;
+  u32 n, o;
 
   /* Start with local preferences */
-  if (new_lpref && old_lpref)		/* Somebody might have undefined them */
+  x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_LOCAL_PREF));
+  y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_LOCAL_PREF));
+  n = x ? x->u.data : new_bgp->cf->default_local_pref;
+  o = y ? y->u.data : old_bgp->cf->default_local_pref;
+  if (n > o)
+    return 1;
+  if (n < o)
+    return 0;
+
+  /* Use AS path lengths */
+  if (new_bgp->cf->compare_path_lengths || old_bgp->cf->compare_path_lengths)
     {
-      if (new_lpref->u.data > old_lpref->u.data)
+      x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
+      y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
+      n = x ? as_path_getlen(x->u.ptr) : 100000;
+      o = y ? as_path_getlen(y->u.ptr) : 100000;
+      if (n < o)
 	return 1;
-      if (new_lpref->u.data < old_lpref->u.data)
+      if (n > o)
 	return 0;
     }
 
+  /* Use origins */
+  x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGIN));
+  y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_ORIGIN));
+  n = x ? x->u.data : 2;
+  o = y ? y->u.data : 2;
+  if (n < o)
+    return 1;
+  if (n > o)
+    return 0;
+
+  /* Compare MED's */
+  x = ea_find(new->attrs->eattrs, EA_CODE(EAP_BGP, BA_MULTI_EXIT_DISC));
+  y = ea_find(old->attrs->eattrs, EA_CODE(EAP_BGP, BA_MULTI_EXIT_DISC));
+  n = x ? x->u.data : new_bgp->cf->default_med;
+  o = y ? y->u.data : old_bgp->cf->default_med;
+  if (n < o)
+    return 1;
+  if (n > o)
+    return 0;
+
   /* A tie breaking procedure according to RFC 1771, section 9.1.2.1 */
-  /* FIXME: Look at MULTI_EXIT_DISC, take the lowest */
   /* We don't have interior distances */
   /* We prefer external peers */
   if (new_bgp->is_internal > old_bgp->is_internal)
@@ -587,12 +618,6 @@ bgp_rte_better(rte *new, rte *old)
 }
 
 static int
-bgp_local_pref(struct bgp_proto *p, rta *a)
-{
-  return 0;				/* FIXME (should be compatible with Cisco defaults?) */
-}
-
-static int
 bgp_path_loopy(struct bgp_proto *p, eattr *a)
 {
   byte *path = a->u.ptr->data;
@@ -680,6 +705,8 @@ bgp_decode_attrs(struct bgp_conn *conn, byte *attr, unsigned int len, struct lin
 	    { errcode = 5; goto err; }
 	  if ((desc->expected_flags ^ flags) & (BAF_OPTIONAL | BAF_TRANSITIVE))
 	    { errcode = 4; goto err; }
+	  if (!desc->allow_in_ebgp && !bgp->is_internal)
+	    continue;
 	  if (desc->validate)
 	    {
 	      errcode = desc->validate(bgp, z, l);
@@ -747,20 +774,6 @@ bgp_decode_attrs(struct bgp_conn *conn, byte *attr, unsigned int len, struct lin
 	}
     }
 
-  /* Assign local preference if none defined */
-  if (!(seen[BA_LOCAL_PREF/8] & (1 << (BA_LOCAL_PREF%8))))
-    {
-      ea = lp_alloc(pool, sizeof(ea_list) + sizeof(eattr));
-      ea->next = a->eattrs;
-      a->eattrs = ea;
-      ea->flags = 0;
-      ea->count = 1;
-      ea->attrs[0].id = EA_CODE(EAP_BGP, BA_LOCAL_PREF);
-      ea->attrs[0].flags = BAF_OPTIONAL;
-      ea->attrs[0].type = EAF_TYPE_INT;
-      ea->attrs[0].u.data = bgp_local_pref(bgp, a);
-    }
-
   /* If the AS path attribute contains our AS, reject the routes */
   e = ea_find(a->eattrs, EA_CODE(EAP_BGP, BA_AS_PATH));
   ASSERT(e);
diff --git a/proto/bgp/bgp.h b/proto/bgp/bgp.h
index 722e12e..ba4f2d4 100644
--- a/proto/bgp/bgp.h
+++ b/proto/bgp/bgp.h
@@ -21,6 +21,9 @@ struct bgp_config {
   int multihop;				/* Number of hops if multihop */
   ip_addr multihop_via;			/* Multihop: address to route to */
   int next_hop_self;			/* Always set next hop to local IP address */
+  int compare_path_lengths;		/* Use path lengths when selecting best route */
+  u32 default_local_pref;		/* Default value for LOCAL_PREF attribute */
+  u32 default_med;			/* Default value for MULTI_EXIT_DISC attribute */
   unsigned connect_retry_time;
   unsigned hold_time, initial_hold_time;
   unsigned keepalive_time;
diff --git a/proto/bgp/config.Y b/proto/bgp/config.Y
index 635c4f6..548f5d0 100644
--- a/proto/bgp/config.Y
+++ b/proto/bgp/config.Y
@@ -15,7 +15,9 @@ CF_HDR
 CF_DECLS
 
 CF_KEYWORDS(BGP, LOCAL, NEIGHBOR, AS, HOLD, TIME, CONNECT, RETRY, KEEPALIVE,
-	MULTIHOP, STARTUP, VIA, NEXT, HOP, SELF, BGP_PATH)
+	MULTIHOP, STARTUP, VIA, NEXT, HOP, SELF, DEFAULT, PATH, METRIC,
+	BGP_PATH, BGP_LOCAL_PREF, BGP_MED, BGP_ORIGIN, BGP_NEXT_HOP,
+	BGP_ATOMIC_AGGR, BGP_AGGREGATOR, BGP_COMMUNITY)
 
 CF_GRAMMAR
 
@@ -27,6 +29,8 @@ bgp_proto_start: proto_start BGP {
      BGP_CFG->hold_time = 240;
      BGP_CFG->connect_retry_time = 120;
      BGP_CFG->initial_hold_time = 240;
+     BGP_CFG->default_med = ~0;
+     BGP_CFG->compare_path_lengths = 1;
   }
  ;
 
@@ -48,9 +52,27 @@ bgp_proto:
  | bgp_proto KEEPALIVE TIME NUM ';' { BGP_CFG->connect_retry_time = $4; }
  | bgp_proto MULTIHOP NUM VIA IPA ';' { BGP_CFG->multihop = $3; BGP_CFG->multihop_via = $5; }
  | bgp_proto NEXT HOP SELF ';' { BGP_CFG->next_hop_self = 1; }
+ | bgp_proto PATH METRIC NUM ';' { BGP_CFG->compare_path_lengths = $4; }
+ | bgp_proto DEFAULT BGP_MED NUM ';' { BGP_CFG->default_med = $4; }
+ | bgp_proto DEFAULT BGP_LOCAL_PREF NUM ';' { BGP_CFG->default_local_pref = $4; }
  ;
 
-CF_ADDTO(dynamic_attr, BGP_PATH { $$ = f_new_dynamic_attr(EAF_TYPE_AS_PATH, T_PATH, EA_CODE(EAP_BGP, BA_AS_PATH)); })
+CF_ADDTO(dynamic_attr, BGP_PATH
+	{ $$ = f_new_dynamic_attr(EAF_TYPE_AS_PATH, T_PATH, EA_CODE(EAP_BGP, BA_AS_PATH)); })
+CF_ADDTO(dynamic_attr, BGP_LOCAL_PREF
+	{ $$ = f_new_dynamic_attr(EAF_TYPE_INT, T_INT, EA_CODE(EAP_BGP, BA_LOCAL_PREF)); })
+CF_ADDTO(dynamic_attr, BGP_MED
+	{ $$ = f_new_dynamic_attr(EAF_TYPE_INT, T_INT, EA_CODE(EAP_BGP, BA_MULTI_EXIT_DISC)); })
+CF_ADDTO(dynamic_attr, BGP_ORIGIN
+	{ $$ = f_new_dynamic_attr(EAF_TYPE_INT, T_INT, EA_CODE(EAP_BGP, BA_ORIGIN)); })
+CF_ADDTO(dynamic_attr, BGP_NEXT_HOP
+	{ $$ = f_new_dynamic_attr(EAF_TYPE_IP_ADDRESS, T_IP, EA_CODE(EAP_BGP, BA_NEXT_HOP)); })
+CF_ADDTO(dynamic_attr, BGP_ATOMIC_AGGR
+	{ $$ = f_new_dynamic_attr(EAF_TYPE_OPAQUE, T_ENUM_EMPTY, EA_CODE(EAP_BGP, BA_ATOMIC_AGGR)); })
+CF_ADDTO(dynamic_attr, BGP_AGGREGATOR
+	{ $$ = f_new_dynamic_attr(EAF_TYPE_INT, T_INT, EA_CODE(EAP_BGP, BA_AGGREGATOR)); })
+CF_ADDTO(dynamic_attr, BGP_COMMUNITY
+	{ $$ = f_new_dynamic_attr(EAF_TYPE_INT_SET, T_CLIST, EA_CODE(EAP_BGP, BA_COMMUNITY)); })
 
 CF_CODE