Routing tables now have their own loops.

This basically means that:
* there are some more levels of indirection and asynchronicity, mostly
  in cleanup procedures, requiring correct lock ordering
* all the internal table operations (prune, next hop update) are done
  without blocking the other parts of BIRD
* the protocols may get their own loops very soon

8 files changed, 75 insertions, 27 deletions

diff --git a/proto/bgp/attrs.c b/proto/bgp/attrs.c
index 9b9013f9..1080db77 100644
--- a/proto/bgp/attrs.c
+++ b/proto/bgp/attrs.c
@@ -2140,7 +2140,7 @@ use_deterministic_med(struct rte_storage *r)
 }
 
 int
-bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best)
+bgp_rte_recalculate(rtable_private *table, net *net, rte *new, rte *old, rte *old_best)
 {
   rte *key = new ? new : old;
   u32 lpref = key->attrs->pref;
diff --git a/proto/bgp/bgp.c b/proto/bgp/bgp.c
index dc845550..aac1f45c 100644
--- a/proto/bgp/bgp.c
+++ b/proto/bgp/bgp.c
@@ -1800,10 +1800,12 @@ bgp_channel_start(struct channel *C)
   ip_addr src = p->local_ip;
 
   if (c->igp_table_ip4)
-    rt_lock_table(c->igp_table_ip4);
+    RT_LOCKED(c->igp_table_ip4, t)
+      rt_lock_table(t);
 
   if (c->igp_table_ip6)
-    rt_lock_table(c->igp_table_ip6);
+    RT_LOCKED(c->igp_table_ip6, t)
+      rt_lock_table(t);
 
   c->pool = p->p.pool; // XXXX
   bgp_init_bucket_table(c);
@@ -1884,10 +1886,12 @@ bgp_channel_cleanup(struct channel *C)
   struct bgp_channel *c = (void *) C;
 
   if (c->igp_table_ip4)
-    rt_unlock_table(c->igp_table_ip4);
+    RT_LOCKED(c->igp_table_ip4, t)
+      rt_unlock_table(t);
 
   if (c->igp_table_ip6)
-    rt_unlock_table(c->igp_table_ip6);
+    RT_LOCKED(c->igp_table_ip6, t)
+      rt_unlock_table(t);
 
   c->index = 0;
 
diff --git a/proto/bgp/bgp.h b/proto/bgp/bgp.h
index 7cb4df1f..60f93bce 100644
--- a/proto/bgp/bgp.h
+++ b/proto/bgp/bgp.h
@@ -586,7 +586,7 @@ void bgp_free_prefix(struct bgp_channel *c, struct bgp_prefix *bp);
 
 int bgp_rte_better(struct rte *, struct rte *);
 int bgp_rte_mergable(rte *pri, rte *sec);
-int bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best);
+int bgp_rte_recalculate(rtable_private *table, net *net, rte *new, rte *old, rte *old_best);
 void bgp_rte_modify_stale(struct rt_export_request *, const net_addr *, struct rt_pending_export *, rte **, uint);
 u32 bgp_rte_igp_metric(struct rte *);
 void bgp_rt_notify(struct proto *P, struct channel *C, const net_addr *n, rte *new, const rte *old);
diff --git a/proto/mrt/mrt.c b/proto/mrt/mrt.c
index 9d78438d..b40592d2 100644
--- a/proto/mrt/mrt.c
+++ b/proto/mrt/mrt.c
@@ -228,7 +228,7 @@ mrt_next_table_(rtable *tab, rtable *tab_ptr, const char *pattern)
        NODE_VALID(tn);
        tn = tn->next)
   {
-    tab = SKIP_BACK(struct rtable, n, tn);
+    tab = SKIP_BACK(rtable, n, tn);
     if (patmatch(pattern, tab->name) &&
 	((tab->addr_type == NET_IP4) || (tab->addr_type == NET_IP6)))
       return tab;
@@ -243,13 +243,21 @@ mrt_next_table(struct mrt_table_dump_state *s)
   rtable *tab = mrt_next_table_(s->table, s->table_ptr, s->table_expr);
 
   if (s->table)
-    rt_unlock_table(s->table);
+  {
+    RT_LOCK(s->table);
+    rt_unlock_table(RT_PRIV(s->table));
+    RT_UNLOCK(s->table);
+  }
 
   s->table = tab;
   s->ipv4 = tab ? (tab->addr_type == NET_IP4) : 0;
 
   if (s->table)
-    rt_lock_table(s->table);
+  {
+    RT_LOCK(s->table);
+    rt_lock_table(RT_PRIV(s->table));
+    RT_UNLOCK(s->table);
+  }
 
   return s->table;
 }
@@ -573,14 +581,23 @@ mrt_table_dump_init(pool *pp)
 static void
 mrt_table_dump_free(struct mrt_table_dump_state *s)
 {
-  if (s->table_open)
-    FIB_ITERATE_UNLINK(&s->fit, &s->table->fib);
-
   if (s->table)
-    rt_unlock_table(s->table);
+  {
+    RT_LOCK(s->table);
+
+    if (s->table_open)
+      FIB_ITERATE_UNLINK(&s->fit, &RT_PRIV(s->table)->fib);
+
+    rt_unlock_table(RT_PRIV(s->table));
+    RT_UNLOCK(s->table);
+  }
 
   if (s->table_ptr)
-    rt_unlock_table(s->table_ptr);
+  {
+    RT_LOCK(s->table_ptr);
+    rt_unlock_table(RT_PRIV(s->table_ptr));
+    RT_UNLOCK(s->table_ptr);
+  }
 
   config_del_obstacle(s->config);
 
@@ -596,8 +613,14 @@ mrt_table_dump_step(struct mrt_table_dump_state *s)
   s->max = 2048;
   s->bws = &bws;
 
+  rtable_private *tab;
+
   if (s->table_open)
+  {
+    RT_LOCK(s->table);
+    tab = RT_PRIV(s->table);
     goto step;
+  }
 
   while (mrt_next_table(s))
   {
@@ -606,15 +629,18 @@ mrt_table_dump_step(struct mrt_table_dump_state *s)
 
     mrt_peer_table_dump(s);
 
-    FIB_ITERATE_INIT(&s->fit, &s->table->fib);
+    RT_LOCK(s->table);
+    tab = RT_PRIV(s->table);
+    FIB_ITERATE_INIT(&s->fit, &tab->fib);
     s->table_open = 1;
 
   step:
-    FIB_ITERATE_START(&s->table->fib, &s->fit, net, n)
+    FIB_ITERATE_START(&tab->fib, &s->fit, net, n)
     {
       if (s->max < 0)
       {
 	FIB_ITERATE_PUT(&s->fit);
+	RT_UNLOCK(s->table);
 	return 0;
       }
 
@@ -634,6 +660,7 @@ mrt_table_dump_step(struct mrt_table_dump_state *s)
     mrt_peer_table_flush(s);
   }
 
+  RT_UNLOCK(s->table);
   return 1;
 }
 
@@ -661,7 +688,11 @@ mrt_timer(timer *t)
   s->always_add_path = cf->always_add_path;
 
   if (s->table_ptr)
-    rt_lock_table(s->table_ptr);
+  {
+    RT_LOCK(s->table_ptr);
+    rt_lock_table(RT_PRIV(s->table_ptr));
+    RT_UNLOCK(s->table_ptr);
+  }
 
   p->table_dump = s;
   ev_schedule(p->event);
@@ -734,7 +765,11 @@ mrt_dump_cmd(struct mrt_dump_data *d)
   s->filename = d->filename;
 
   if (s->table_ptr)
-    rt_lock_table(s->table_ptr);
+  {
+    RT_LOCK(s->table_ptr);
+    rt_lock_table(RT_PRIV(s->table_ptr));
+    RT_UNLOCK(s->table_ptr);
+  }
 
   this_cli->cont = mrt_dump_cont;
   this_cli->cleanup = mrt_dump_cleanup;
diff --git a/proto/mrt/mrt.h b/proto/mrt/mrt.h
index 4ff94c12..04865089 100644
--- a/proto/mrt/mrt.h
+++ b/proto/mrt/mrt.h
@@ -40,7 +40,7 @@ struct mrt_proto {
 
 struct mrt_dump_data {
   const char *table_expr;
-  struct rtable *table_ptr;
+  rtable *table_ptr;
   const struct filter *filter;
   const char *filename;
 };
@@ -60,7 +60,7 @@ struct mrt_table_dump_state {
 
 					/* Configuration information */
   const char *table_expr;		/* Wildcard for table name (or NULL) */
-  struct rtable *table_ptr;		/* Explicit table (or NULL) */
+  rtable *table_ptr;			/* Explicit table (or NULL) */
   const struct filter *filter;		/* Optional filter */
   const char *filename;			/* Filename pattern */
   int always_add_path;			/* Always use *_ADDPATH message subtypes */
@@ -73,7 +73,7 @@ struct mrt_table_dump_state {
 
   HASH(struct mrt_peer_entry) peer_hash; /* Hash for peers to find the index */
 
-  struct rtable *table;			/* Processed table, NULL initially */
+  rtable *table;			/* Processed table, NULL initially */
   struct fib_iterator fit;		/* Iterator in processed table */
   int table_open;			/* Whether iterator is linked */
 
diff --git a/proto/perf/perf.c b/proto/perf/perf.c
index 8b2cb69f..aa688d88 100644
--- a/proto/perf/perf.c
+++ b/proto/perf/perf.c
@@ -198,7 +198,9 @@ perf_loop(void *data)
     p->exp++;
   }
 
-  rt_schedule_prune(P->main_channel->table);
+  RT_LOCK(P->main_channel->table);
+  rt_schedule_prune(RT_PRIV(P->main_channel->table));
+  RT_UNLOCK(P->main_channel->table);
   ev_schedule(p->loop);
 }
 
diff --git a/proto/radv/radv.c b/proto/radv/radv.c
index fa228c69..d572c1b7 100644
--- a/proto/radv/radv.c
+++ b/proto/radv/radv.c
@@ -555,7 +555,10 @@ radv_check_active(struct radv_proto *p)
     return 1;
 
   struct channel *c = p->p.main_channel;
-  return rt_examine(c->table, &cf->trigger, c, c->out_filter);
+  RT_LOCK(c->table);
+  int active = rt_examine(RT_PRIV(c->table), &cf->trigger, c, c->out_filter);
+  RT_UNLOCK(c->table);
+  return active;
 }
 
 static void
diff --git a/proto/static/static.c b/proto/static/static.c
index 45791e8e..bd7f3f5b 100644
--- a/proto/static/static.c
+++ b/proto/static/static.c
@@ -491,10 +491,12 @@ static_start(struct proto *P)
     static_lp = lp_new(&root_pool, LP_GOOD_SIZE(1024));
 
   if (p->igp_table_ip4)
-    rt_lock_table(p->igp_table_ip4);
+    RT_LOCKED(p->igp_table_ip4, t)
+      rt_lock_table(t);
 
   if (p->igp_table_ip6)
-    rt_lock_table(p->igp_table_ip6);
+    RT_LOCKED(p->igp_table_ip6, t)
+      rt_lock_table(t);
 
   p->event = ev_new_init(p->p.pool, static_announce_marked, p);
 
@@ -521,10 +523,12 @@ static_shutdown(struct proto *P)
     static_reset_rte(p, r);
 
   if (p->igp_table_ip4)
-    rt_unlock_table(p->igp_table_ip4);
+    RT_LOCKED(p->igp_table_ip4, t)
+      rt_unlock_table(t);
 
   if (p->igp_table_ip6)
-    rt_unlock_table(p->igp_table_ip6);
+    RT_LOCKED(p->igp_table_ip6, t)
+      rt_unlock_table(t);
 
   return PS_DOWN;
 }