Merge commit '98bb80a243b58c43453e9be69d19d0350286549c' into int-new

4 files changed, 207 insertions, 70 deletions

diff --git a/proto/radv/config.Y b/proto/radv/config.Y
index b5f4b5f2..0e43c237 100644
--- a/proto/radv/config.Y
+++ b/proto/radv/config.Y
@@ -27,7 +27,7 @@ static u8 radv_mult_val;	/* Used by radv_mult for second return value */
 CF_DECLS
 
 CF_KEYWORDS(RADV, PREFIX, INTERFACE, MIN, MAX, RA, DELAY, INTERVAL,
-	MANAGED, OTHER, CONFIG, LINK, MTU, REACHABLE, TIME, RETRANS,
+	MANAGED, OTHER, CONFIG, LINGER, LINK, MTU, REACHABLE, TIME, RETRANS,
 	TIMER, CURRENT, HOP, LIMIT, DEFAULT, VALID, PREFERRED, MULT,
 	LIFETIME, SKIP, ONLINK, AUTONOMOUS, RDNSS, DNSSL, NS, DOMAIN,
 	LOCAL, TRIGGER, SENSITIVE, PREFERENCE, LOW, MEDIUM, HIGH)
@@ -80,6 +80,7 @@ radv_iface_start:
   RADV_IFACE->max_ra_int = DEFAULT_MAX_RA_INT;
   RADV_IFACE->min_delay = DEFAULT_MIN_DELAY;
   RADV_IFACE->current_hop_limit = DEFAULT_CURRENT_HOP_LIMIT;
+  RADV_IFACE->linger_time = DEFAULT_LINGER_TIME;
   RADV_IFACE->default_lifetime = -1;
   RADV_IFACE->default_lifetime_sensitive = 1;
   RADV_IFACE->default_preference = RA_PREF_MEDIUM;
@@ -94,6 +95,7 @@ radv_iface_item:
  | LINK MTU expr { RADV_IFACE->link_mtu = $3; }
  | REACHABLE TIME expr { RADV_IFACE->reachable_time = $3; if ($3 > 3600000) cf_error("Reachable time must be in range 0-3600000"); }
  | RETRANS TIMER expr { RADV_IFACE->retrans_timer = $3; }
+ | LINGER TIME expr { RADV_IFACE->linger_time = $3; if ($3 > 3600) cf_error("Linger time must be in range 0-3600"); }
  | CURRENT HOP LIMIT expr { RADV_IFACE->current_hop_limit = $4; if ($4 > 255) cf_error("Current hop limit must be in range 0-255"); }
  | DEFAULT LIFETIME expr radv_sensitive {
      RADV_IFACE->default_lifetime = $3;
diff --git a/proto/radv/packets.c b/proto/radv/packets.c
index 9ea8feee..e07296e1 100644
--- a/proto/radv/packets.c
+++ b/proto/radv/packets.c
@@ -70,36 +70,6 @@ struct radv_opt_dnssl
   char domain[];
 };
 
-
-static struct radv_prefix_config default_prefix = {
-  .onlink = 1,
-  .autonomous = 1,
-  .valid_lifetime = DEFAULT_VALID_LIFETIME,
-  .preferred_lifetime = DEFAULT_PREFERRED_LIFETIME
-};
-
-
-static struct radv_prefix_config *
-radv_prefix_match(struct radv_iface *ifa, struct ifa *a)
-{
-  struct radv_proto *p = ifa->ra;
-  struct radv_config *cf = (struct radv_config *) (p->p.cf);
-  struct radv_prefix_config *pc;
-
-  if (a->scope <= SCOPE_LINK)
-    return NULL;
-
-  WALK_LIST(pc, ifa->cf->pref_list)
-    if (net_in_netX(&a->prefix, (net_addr *) &pc->prefix))
-      return pc;
-
-  WALK_LIST(pc, cf->pref_list)
-    if (net_in_netX(&a->prefix, (net_addr *) &pc->prefix))
-      return pc;
-
-  return &default_prefix;
-}
-
 static int
 radv_prepare_rdnss(struct radv_iface *ifa, list *rdnss_list, char **buf, char *bufend)
 {
@@ -234,6 +204,36 @@ radv_prepare_dnssl(struct radv_iface *ifa, list *dnssl_list, char **buf, char *b
   return -1;
 }
 
+static int
+radv_prepare_prefix(struct radv_iface *ifa, struct radv_prefix *px,
+		    char **buf, char *bufend)
+{
+  struct radv_prefix_config *pc = px->cf;
+
+  if (*buf + sizeof(struct radv_opt_prefix) > bufend)
+  {
+    log(L_WARN "%s: Too many prefixes on interface %s",
+	ifa->ra->p.name, ifa->iface->name);
+    return -1;
+  }
+
+  struct radv_opt_prefix *op = (void *) *buf;
+  op->type = OPT_PREFIX;
+  op->length = 4;
+  op->pxlen = px->prefix.pxlen;
+  op->flags = (pc->onlink ? OPT_PX_ONLINK : 0) |
+    (pc->autonomous ? OPT_PX_AUTONOMOUS : 0);
+  op->valid_lifetime = (ifa->ra->active || !pc->valid_lifetime_sensitive) ?
+    htonl(pc->valid_lifetime) : 0;
+  op->preferred_lifetime = (ifa->ra->active || !pc->preferred_lifetime_sensitive) ?
+    htonl(pc->preferred_lifetime) : 0;
+  op->reserved = 0;
+  op->prefix = ip6_hton(px->prefix.prefix);
+  *buf += sizeof(*op);
+
+  return 0;
+}
+
 static void
 radv_prepare_ra(struct radv_iface *ifa)
 {
@@ -269,37 +269,11 @@ radv_prepare_ra(struct radv_iface *ifa)
     buf += sizeof (*om);
   }
 
-  struct ifa *addr;
-  WALK_LIST(addr, ifa->iface->addrs)
+  struct radv_prefix *prefix;
+  WALK_LIST(prefix, ifa->prefixes)
   {
-    if (addr->prefix.type != NET_IP6)
-      continue;
-
-    struct radv_prefix_config *pc;
-    pc = radv_prefix_match(ifa, addr);
-
-    if (!pc || pc->skip)
-      continue;
-
-    if (buf + sizeof(struct radv_opt_prefix) > bufend)
-    {
-      log(L_WARN "%s: Too many prefixes on interface %s", p->p.name, ifa->iface->name);
+    if (radv_prepare_prefix(ifa, prefix, &buf, bufend) < 0)
       goto done;
-    }
-
-    struct radv_opt_prefix *op = (void *) buf;
-    op->type = OPT_PREFIX;
-    op->length = 4;
-    op->pxlen = net6_pxlen(&addr->prefix);
-    op->flags = (pc->onlink ? OPT_PX_ONLINK : 0) |
-      (pc->autonomous ? OPT_PX_AUTONOMOUS : 0);
-    op->valid_lifetime = (p->active || !pc->valid_lifetime_sensitive) ?
-      htonl(pc->valid_lifetime) : 0;
-    op->preferred_lifetime = (p->active || !pc->preferred_lifetime_sensitive) ?
-      htonl(pc->preferred_lifetime) : 0;
-    op->reserved = 0;
-    op->prefix = ip6_hton(net6_prefix(&addr->prefix));
-    buf += sizeof(*op);
   }
 
   if (! ic->rdnss_local)
@@ -408,7 +382,7 @@ radv_err_hook(sock *sk, int err)
 int
 radv_sk_open(struct radv_iface *ifa)
 {
-  sock *sk = sk_new(ifa->ra->p.pool);
+  sock *sk = sk_new(ifa->pool);
   sk->type = SK_IP;
   sk->subtype = SK_IPV6;
   sk->dport = ICMPV6_PROTO;
diff --git a/proto/radv/radv.c b/proto/radv/radv.c
index fe371ab4..c96d7724 100644
--- a/proto/radv/radv.c
+++ b/proto/radv/radv.c
@@ -51,6 +51,16 @@ radv_timer(timer *tm)
 
   RADV_TRACE(D_EVENTS, "Timer fired on %s", ifa->iface->name);
 
+  /*
+   * If some dead prefixes expired, regenerate the prefix list and the packet.
+   * We do so by pretending there was a change on the interface.
+   *
+   * This sets the timer, but we replace it just at the end of this function
+   * (replacing a timer is fine).
+   */
+  if (ifa->prefix_expires && (ifa->prefix_expires <= current_time()))
+    radv_iface_notify(ifa, RA_EV_GC);
+
   radv_send_ra(ifa, 0);
 
   /* Update timer */
@@ -68,7 +78,136 @@ radv_timer(timer *tm)
   tm_start(ifa->timer, t);
 }
 
-static char* ev_name[] = { NULL, "Init", "Change", "RS" };
+static struct radv_prefix_config default_prefix = {
+  .onlink = 1,
+  .autonomous = 1,
+  .valid_lifetime = DEFAULT_VALID_LIFETIME,
+  .preferred_lifetime = DEFAULT_PREFERRED_LIFETIME
+};
+
+static struct radv_prefix_config dead_prefix = {
+};
+
+/* Find a corresponding config for the given prefix */
+static struct radv_prefix_config *
+radv_prefix_match(struct radv_iface *ifa, net_addr_ip6 *px)
+{
+  struct radv_proto *p = ifa->ra;
+  struct radv_config *cf = (struct radv_config *) (p->p.cf);
+  struct radv_prefix_config *pc;
+
+  WALK_LIST(pc, ifa->cf->pref_list)
+    if (net_in_net_ip6(px, &pc->prefix))
+      return pc;
+
+  WALK_LIST(pc, cf->pref_list)
+    if (net_in_net_ip6(px, &pc->prefix))
+      return pc;
+
+  return &default_prefix;
+}
+
+/*
+ * Go through the list of prefixes, compare them with configs and decide if we
+ * want them or not.
+ */
+static void
+radv_prepare_prefixes(struct radv_iface *ifa)
+{
+  struct radv_proto *p = ifa->ra;
+  struct radv_iface_config *cf = ifa->cf;
+  struct radv_prefix *pfx;
+
+  /* First mark all the prefixes as unused */
+  WALK_LIST(pfx, ifa->prefixes)
+    pfx->mark = 0;
+
+  /* Find all the prefixes we want to use and make sure they are in the list. */
+  struct ifa *addr;
+  WALK_LIST(addr, ifa->iface->addrs)
+  {
+    if ((addr->prefix.type != NET_IP6) ||
+	(addr->scope <= SCOPE_LINK))
+      continue;
+
+    net_addr_ip6 *prefix = (void *) &addr->prefix;
+    struct radv_prefix_config *pc = radv_prefix_match(ifa, prefix);
+
+    if (!pc || pc->skip)
+      continue;
+
+    /* Do we have it already? */
+    struct radv_prefix *existing = NULL;
+    WALK_LIST(pfx, ifa->prefixes)
+      if (net_equal_ip6(&pfx->prefix, prefix))
+      {
+	existing = pfx;
+	break;
+      }
+
+    if (!existing)
+    {
+      RADV_TRACE(D_EVENTS, "Adding new prefix %N on %s",
+		 prefix, ifa->iface->name);
+
+      existing = mb_allocz(ifa->pool, sizeof *existing);
+      net_copy_ip6(&existing->prefix, prefix);
+      add_tail(&ifa->prefixes, NODE existing);
+    }
+
+    /*
+     * Update the information (it may have changed, or even bring a prefix back
+     * to life).
+     */
+    existing->alive = 1;
+    existing->mark = 1;
+    existing->cf = pc;
+  }
+
+  /*
+   * Garbage-collect the prefixes. If something isn't used, it dies (but isn't
+   * dropped just yet). If something is dead and rots there for long enough,
+   * clean it up.
+   */
+  btime now_ = current_time();
+  btime expires = now_ + cf->linger_time S;
+  btime expires_min = 0;
+  struct radv_prefix *next;
+  WALK_LIST_DELSAFE(pfx, next, ifa->prefixes)
+  {
+    if (pfx->alive && !pfx->mark)
+    {
+      RADV_TRACE(D_EVENTS, "Marking prefix %N on %s as dead",
+		 pfx->prefix, ifa->iface->name);
+
+      pfx->alive = 0;
+      pfx->expires = expires;
+      pfx->cf = &dead_prefix;
+    }
+
+    if (!pfx->alive)
+    {
+      if (pfx->expires <= now_)
+      {
+	RADV_TRACE(D_EVENTS, "Removing prefix %N on %s",
+		   pfx->prefix, ifa->iface->name);
+
+	rem_node(NODE pfx);
+	mb_free(pfx);
+      }
+      else
+      {
+	/* Find minimum expiration time */
+	if (!expires_min || (pfx->expires < expires_min))
+	  expires_min = pfx->expires;
+      }
+    }
+  }
+
+  ifa->prefix_expires = expires_min;
+}
+
+static char* ev_name[] = { NULL, "Init", "Change", "RS", "Garbage collect" };
 
 void
 radv_iface_notify(struct radv_iface *ifa, int event)
@@ -83,6 +222,7 @@ radv_iface_notify(struct radv_iface *ifa, int event)
   switch (event)
   {
   case RA_EV_CHANGE:
+  case RA_EV_GC:
     ifa->plen = 0;
   case RA_EV_INIT:
     ifa->initial = MAX_INITIAL_RTR_ADVERTISEMENTS;
@@ -92,6 +232,8 @@ radv_iface_notify(struct radv_iface *ifa, int event)
     break;
   }
 
+  radv_prepare_prefixes(ifa);
+
   /* Update timer */
   btime t = ifa->last + ifa->cf->min_delay S - current_time();
   tm_start(ifa->timer, t);
@@ -137,16 +279,18 @@ radv_iface_add(struct object_lock *lock)
 static void
 radv_iface_new(struct radv_proto *p, struct iface *iface, struct radv_iface_config *cf)
 {
-  pool *pool = p->p.pool;
   struct radv_iface *ifa;
 
   RADV_TRACE(D_EVENTS, "Adding interface %s", iface->name);
 
+  pool *pool = rp_new(p->p.pool, iface->name);
   ifa = mb_allocz(pool, sizeof(struct radv_iface));
+  ifa->pool = pool;
   ifa->ra = p;
   ifa->cf = cf;
   ifa->iface = iface;
   ifa->addr = iface->llv6;
+  init_list(&ifa->prefixes);
 
   add_tail(&p->iface_list, NODE ifa);
 
@@ -172,11 +316,7 @@ radv_iface_remove(struct radv_iface *ifa)
 
   rem_node(NODE ifa);
 
-  rfree(ifa->sk);
-  rfree(ifa->timer);
-  rfree(ifa->lock);
-
-  mb_free(ifa);
+  rfree(ifa->pool);
 }
 
 static void
diff --git a/proto/radv/radv.h b/proto/radv/radv.h
index 8324bb67..4672e3b2 100644
--- a/proto/radv/radv.h
+++ b/proto/radv/radv.h
@@ -35,6 +35,7 @@
 #define DEFAULT_MAX_RA_INT 600
 #define DEFAULT_MIN_DELAY 3
 #define DEFAULT_CURRENT_HOP_LIMIT 64
+#define DEFAULT_LINGER_TIME 300
 
 #define DEFAULT_VALID_LIFETIME 86400
 #define DEFAULT_PREFERRED_LIFETIME 14400
@@ -64,6 +65,9 @@ struct radv_iface_config
   u32 max_ra_int;
   u32 min_delay;
 
+  u32 linger_time;		/* How long a dead prefix should still be advertised with 0
+				   lifetime */
+
   u8 rdnss_local;		/* Global list is not used for RDNSS */
   u8 dnssl_local;		/* Global list is not used for DNSSL */
 
@@ -75,7 +79,7 @@ struct radv_iface_config
   u32 current_hop_limit;
   u32 default_lifetime;
   u8 default_lifetime_sensitive; /* Whether default_lifetime depends on trigger */
-  u8 default_preference;	/* Default Router Preference (RFC 4191) */
+  u8 default_preference;	 /* Default Router Preference (RFC 4191) */
 };
 
 struct radv_prefix_config
@@ -118,6 +122,19 @@ struct radv_proto
   u8 active;			/* Whether radv is active w.r.t. triggers */
 };
 
+struct radv_prefix		/* One prefix we advertise */
+{
+  node n;
+  net_addr_ip6 prefix;
+
+  u8 alive;			/* Is the prefix alive? If not, we advertise it
+				   with 0 lifetime, so clients stop using it */
+  u8 mark;			/* A temporary mark for processing */
+  btime expires;		/* The time when we drop this prefix from
+				   advertising. It is valid only if !alive. */
+  struct radv_prefix_config *cf; /* The config tied to this prefix */
+};
+
 struct radv_iface
 {
   node n;
@@ -125,6 +142,9 @@ struct radv_iface
   struct radv_iface_config *cf;	/* Related config, must be updated in reconfigure */
   struct iface *iface;
   struct ifa *addr;		/* Link-local address of iface */
+  struct pool *pool;		/* A pool for interface-specific things */
+  list prefixes;		/* The prefixes we advertise (struct radv_prefix) */
+  btime prefix_expires;		/* When the soonest prefix expires (0 = none dead) */
 
   timer *timer;
   struct object_lock *lock;
@@ -132,12 +152,13 @@ struct radv_iface
 
   btime last;			/* Time of last sending of RA */
   u16 plen;			/* Length of prepared RA in tbuf, or 0 if not valid */
-  byte initial;			/* List of active ifaces */
+  byte initial;			/* How many RAs are still to be sent as initial */
 };
 
 #define RA_EV_INIT 1		/* Switch to initial mode */
 #define RA_EV_CHANGE 2		/* Change of options or prefixes */
 #define RA_EV_RS 3		/* Received RS */
+#define RA_EV_GC 4		/* Internal garbage collection of prefixes */
 
 /* Default Router Preferences (RFC 4191) */
 #define RA_PREF_LOW	0x18