Nest: Maintain separate IPv4, IPv6 and LLv6 preferred addresses

Also redesign preferred address selection and update protocols to use
appropriate preferred address.

Based on a previous work by Jan Maria Matejka.

7 files changed, 93 insertions, 99 deletions

diff --git a/sysdep/bsd/krt-sock.c b/sysdep/bsd/krt-sock.c
index df639816..be8b50d6 100644
--- a/sysdep/bsd/krt-sock.c
+++ b/sysdep/bsd/krt-sock.c
@@ -287,18 +287,21 @@ krt_send_route(struct krt_proto *p, int cmd, rte *e)
 #endif
   {
     /* Fallback for all other valid cases */
-    if (!i->addr)
-    {
-      log(L_ERR "KRT: interface %s has no IP addess", i->name);
-      return -1;
-    }
 
 #ifdef RTF_CLONING
     if (cmd == RTM_ADD && (i->flags & IF_MULTIACCESS) != IF_MULTIACCESS)	/* PTP */
       msg.rtm.rtm_flags |= RTF_CLONING;
 #endif
 
-    sockaddr_fill(&gate, ipa_is_ip4(i->addr->ip) ? AF_INET : AF_INET6, i->addr->ip, NULL, 0);
+    struct ifa *addr = (net->n.addr->type == NET_IP4) ? i->addr4 : (i->addr6 ?: i->llv6);
+
+    if (!addr)
+    {
+      log(L_ERR "KRT: interface %s has no IP addess", i->name);
+      return -1;
+    }
+
+    sockaddr_fill(&gate, af, addr->ip, i, 0);
     msg.rtm.rtm_addrs |= RTA_GATEWAY;
     break;
   }
@@ -1124,13 +1127,11 @@ kif_sys_shutdown(struct kif_proto *p)
   krt_buffer_release(&p->p);
 }
 
-
-struct ifa *
-kif_get_primary_ip(struct iface *i UNUSED)
+int
+kif_update_sysdep_addr(struct iface *i)
 {
-#if 0
   static int fd = -1;
-  
+
   if (fd < 0)
     fd = socket(AF_INET, SOCK_DGRAM, 0);
 
@@ -1140,20 +1141,10 @@ kif_get_primary_ip(struct iface *i UNUSED)
 
   int rv = ioctl(fd, SIOCGIFADDR, (char *) &ifr);
   if (rv < 0)
-    return NULL;
-
-  ip_addr addr;
-  struct sockaddr_in *sin = (struct sockaddr_in *) &ifr.ifr_addr;
-  memcpy(&addr, &sin->sin_addr.s_addr, sizeof(ip_addr));
-  ipa_ntoh(addr);
+    return 0;
 
-  struct ifa *a;
-  WALK_LIST(a, i->addrs)
-  {
-    if (ipa_equal(a->ip, addr))
-      return a;
-  }
-#endif
+  ip4_addr old = i->sysdep;
+  i->sysdep = ip4_from_ipa(ipa_from_sa4(&ifr.ifr_addr);
 
-  return NULL;
+  return !ip4_equal(i->sysdep, addr);
 }
diff --git a/sysdep/bsd/sysio.h b/sysdep/bsd/sysio.h
index 0e895e20..68296e65 100644
--- a/sysdep/bsd/sysio.h
+++ b/sysdep/bsd/sysio.h
@@ -38,12 +38,12 @@
  */
 
 #define INIT_MREQ4(maddr,ifa) \
-  { .imr_multiaddr = ipa_to_in4(maddr), .imr_interface = ipa_to_in4(ifa->addr->ip) }
+  { .imr_multiaddr = ipa_to_in4(maddr), .imr_interface = ip4_to_in4(ifa->sysdep) }
 
 static inline int
 sk_setup_multicast4(sock *s)
 {
-  struct in_addr ifa = ipa_to_in4(s->iface->addr->ip);
+  struct in_addr ifa = ip4_to_in4(s->iface->sysdep);
   u8 ttl = s->ttl;
   u8 n = 0;
 
diff --git a/sysdep/linux/netlink.c b/sysdep/linux/netlink.c
index 2b7b13fb..c9d5cdec 100644
--- a/sysdep/linux/netlink.c
+++ b/sysdep/linux/netlink.c
@@ -2030,3 +2030,9 @@ void
 kif_sys_shutdown(struct kif_proto *p UNUSED)
 {
 }
+
+int
+kif_update_sysdep_addr(struct iface *i UNUSED)
+{
+  return 0;
+}
diff --git a/sysdep/unix/krt.Y b/sysdep/unix/krt.Y
index b261c91e..3bf7da65 100644
--- a/sysdep/unix/krt.Y
+++ b/sysdep/unix/krt.Y
@@ -14,6 +14,7 @@ CF_DEFINES
 
 #define THIS_KRT ((struct krt_config *) this_proto)
 #define THIS_KIF ((struct kif_config *) this_proto)
+#define KIF_IFACE ((struct kif_iface_config *) this_ipatt)
 
 static void
 krt_set_merge_paths(struct channel_config *cc, uint merge, uint limit)
@@ -25,6 +26,17 @@ krt_set_merge_paths(struct channel_config *cc, uint merge, uint limit)
   cc->merge_limit = limit;
 }
 
+static void
+kif_set_preferred(ip_addr ip)
+{
+  if (ipa_is_ip4(ip))
+    KIF_IFACE->pref_v4 = ip;
+  else if (!ipa_is_link_local(ip))
+    KIF_IFACE->pref_v6 = ip;
+  else
+    KIF_IFACE->pref_ll = ip;
+}
+
 CF_DECLS
 
 CF_KEYWORDS(KERNEL, PERSIST, SCAN, TIME, LEARN, DEVICE, ROUTES, GRACEFUL, RESTART, KRT_SOURCE, KRT_METRIC, MERGE, PATHS)
@@ -88,18 +100,38 @@ CF_ADDTO(kif_proto, kif_proto kif_item ';')
 
 kif_item:
    proto_item
+ | INTERFACE kif_iface
  | SCAN TIME expr {
       /* Scan time of 0 means scan on startup only */
       THIS_KIF->scan_time = $3;
    }
- | PRIMARY opttext net_or_ipa {
-     struct kif_primary_item *kpi = cfg_alloc(sizeof (struct kif_primary_item));
-     kpi->pattern = $2;
-     kpi->addr = $3;
-     add_tail(&THIS_KIF->primary, &kpi->n);
-   }
  ;
 
+kif_iface_start:
+{
+  this_ipatt = cfg_allocz(sizeof(struct kif_iface_config));
+  add_tail(&THIS_KIF->iface_list, NODE this_ipatt);
+  init_list(&this_ipatt->ipn_list);
+}
+
+kif_iface_item:
+   PREFERRED ipa { kif_set_preferred($2); }
+ ;
+
+kif_iface_opts:
+   /* empty */
+ | kif_iface_opts kif_iface_item ';'
+ ;
+
+kif_iface_opt_list:
+   /* empty */
+ | '{' kif_iface_opts '}'
+ ;
+
+kif_iface:
+  kif_iface_start iface_patt_list_nopx kif_iface_opt_list;
+
+
 CF_ADDTO(dynamic_attr, KRT_SOURCE { $$ = f_new_dynamic_attr(EAF_TYPE_INT | EAF_TEMP, T_INT, EA_KRT_SOURCE); })
 CF_ADDTO(dynamic_attr, KRT_METRIC { $$ = f_new_dynamic_attr(EAF_TYPE_INT | EAF_TEMP, T_INT, EA_KRT_METRIC); })
 
diff --git a/sysdep/unix/krt.c b/sysdep/unix/krt.c
index f0241777..0349a09f 100644
--- a/sysdep/unix/krt.c
+++ b/sysdep/unix/krt.c
@@ -89,6 +89,16 @@ static struct kif_config *kif_cf;
 static timer *kif_scan_timer;
 static bird_clock_t kif_last_shot;
 
+static struct kif_iface_config kif_default_iface = {};
+
+struct kif_iface_config *
+kif_get_iface_config(struct iface *iface)
+{
+  struct kif_config *cf = (void *) (kif_proto->p.cf);
+  struct kif_iface_config *ic = (void *) iface_patt_find(&cf->iface_list, iface, NULL);
+  return ic ?: &kif_default_iface;
+}
+
 static void
 kif_scan(timer *t)
 {
@@ -116,57 +126,6 @@ kif_request_scan(void)
     tm_start(kif_scan_timer, 1);
 }
 
-static inline int
-prefer_addr(struct ifa *a, struct ifa *b)
-{
-  int sa = a->scope > SCOPE_LINK;
-  int sb = b->scope > SCOPE_LINK;
-
-  if (sa < sb)
-    return 0;
-  else if (sa > sb)
-    return 1;
-  else
-    return ipa_compare(a->ip, b->ip) < 0;
-}
-
-static inline struct ifa *
-find_preferred_ifa(struct iface *i, const net_addr *n)
-{
-  struct ifa *a, *b = NULL;
-
-  WALK_LIST(a, i->addrs)
-    {
-      if (!(a->flags & IA_SECONDARY) &&
-	  (!n || ipa_in_netX(a->ip, n)) &&
-	  (!b || prefer_addr(a, b)))
-	b = a;
-    }
-
-  return b;
-}
-
-struct ifa *
-kif_choose_primary(struct iface *i)
-{
-  struct kif_config *cf = (struct kif_config *) (kif_proto->p.cf);
-  struct kif_primary_item *it;
-  struct ifa *a;
-
-  WALK_LIST(it, cf->primary)
-    {
-      if (!it->pattern || patmatch(it->pattern, i->name))
-	if (a = find_preferred_ifa(i, &it->addr))
-	  return a;
-    }
-
-  if (a = kif_get_primary_ip(i))
-    return a;
-
-  return find_preferred_ifa(i, NULL);
-}
-
-
 static struct proto *
 kif_init(struct proto_config *c)
 {
@@ -224,15 +183,15 @@ kif_reconfigure(struct proto *p, struct proto_config *new)
       tm_start(kif_scan_timer, n->scan_time);
     }
 
-  if (!EMPTY_LIST(o->primary) || !EMPTY_LIST(n->primary))
+  if (!EMPTY_LIST(o->iface_list) || !EMPTY_LIST(n->iface_list))
     {
       /* This is hack, we have to update a configuration
        * to the new value just now, because it is used
-       * for recalculation of primary addresses.
+       * for recalculation of preferred addresses.
        */
       p->cf = new;
 
-      ifa_recalc_all_primary_addresses();
+      if_recalc_all_preferred_addresses();
     }
 
   return 1;
@@ -254,7 +213,7 @@ kif_init_config(int class)
 
   kif_cf = (struct kif_config *) proto_config_new(&proto_unix_iface, class);
   kif_cf->scan_time = 60;
-  init_list(&kif_cf->primary);
+  init_list(&kif_cf->iface_list);
 
   kif_sys_init_config(kif_cf);
   return (struct proto_config *) kif_cf;
@@ -266,14 +225,13 @@ kif_copy_config(struct proto_config *dest, struct proto_config *src)
   struct kif_config *d = (struct kif_config *) dest;
   struct kif_config *s = (struct kif_config *) src;
 
-  /* Copy primary addr list */
-  cfg_copy_list(&d->primary, &s->primary, sizeof(struct kif_primary_item));
+  /* Copy interface config list */
+  cfg_copy_list(&d->iface_list, &s->iface_list, sizeof(struct kif_iface_config));
 
   /* Fix sysdep parts */
   kif_sys_copy_config(d, s);
 }
 
-
 struct protocol proto_unix_iface = {
   .name = 		"Device",
   .template = 		"device%d",
diff --git a/sysdep/unix/krt.h b/sysdep/unix/krt.h
index cb404de3..089c97ad 100644
--- a/sysdep/unix/krt.h
+++ b/sysdep/unix/krt.h
@@ -94,17 +94,20 @@ void krt_got_route_async(struct krt_proto *p, struct rte *e, int new);
 
 extern struct protocol proto_unix_iface;
 
-struct kif_primary_item {
-  node n;
-  byte *pattern;
-  net_addr addr;
-};
-
 struct kif_config {
   struct proto_config c;
   struct kif_params sys;	/* Sysdep params */
+
+  list iface_list;		/* List of iface configs (struct kif_iface_config) */
   int scan_time;		/* How often we re-scan interfaces */
-  list primary;			/* Preferences for primary addresses (struct kif_primary_item) */
+};
+
+struct kif_iface_config {
+  struct iface_patt i;
+
+  ip_addr pref_v4;
+  ip_addr pref_v6;
+  ip_addr pref_ll;
 };
 
 struct kif_proto {
@@ -116,6 +119,7 @@ extern struct kif_proto *kif_proto;
 
 #define KIF_CF ((struct kif_config *)p->p.cf)
 
+struct kif_iface_config * kif_get_iface_config(struct iface *iface);
 struct proto_config * krt_init_config(int class);
 
 
@@ -150,6 +154,6 @@ void kif_sys_copy_config(struct kif_config *, struct kif_config *);
 
 void kif_do_scan(struct kif_proto *);
 
-struct ifa *kif_get_primary_ip(struct iface *i);
+int kif_update_sysdep_addr(struct iface *i);
 
 #endif
diff --git a/sysdep/unix/unix.h b/sysdep/unix/unix.h
index dcaab729..4b0fb005 100644
--- a/sysdep/unix/unix.h
+++ b/sysdep/unix/unix.h
@@ -80,6 +80,9 @@ static inline ip_addr ipa_from_sa(sockaddr *sa)
 static inline struct in_addr ipa_to_in4(ip_addr a)
 { return (struct in_addr) { htonl(ipa_to_u32(a)) }; }
 
+static inline struct in_addr ip4_to_in4(ip4_addr a)
+{ return (struct in_addr) { htonl(ip4_to_u32(a)) }; }
+
 static inline struct in6_addr ipa_to_in6(ip_addr a)
 { return (struct in6_addr) { .s6_addr32 = { htonl(_I0(a)), htonl(_I1(a)), htonl(_I2(a)), htonl(_I3(a)) } }; }