Merge commit '0072d11f' into tmp-learn

4 files changed, 139 insertions, 31 deletions

diff --git a/proto/bgp/attrs.c b/proto/bgp/attrs.c
index 084c9b63..a7b1a7ed 100644
--- a/proto/bgp/attrs.c
+++ b/proto/bgp/attrs.c
@@ -1901,7 +1901,6 @@ bgp_out_table_export_start(struct rt_exporter *re, struct rt_export_request *req
   pool *p = rp_new(c->c.proto->pool, "Export hook");
   struct rt_export_hook *hook = mb_allocz(p, sizeof(struct rt_export_hook));
   hook->pool = p;
-  hook->lp = lp_new_default(p);
   hook->event = ev_new_init(p, bgp_out_table_feed, hook);
   hook->feed_type = TFT_HASH;
 
@@ -1919,6 +1918,7 @@ bgp_setup_out_table(struct bgp_channel *c)
   };
 
   init_list(&c->prefix_exporter.hooks);
+  init_list(&c->prefix_exporter.pending);
 
   c->c.out_table = &c->prefix_exporter;
 }
@@ -1929,10 +1929,10 @@ bgp_setup_out_table(struct bgp_channel *c)
  */
 
 int
-bgp_preexport(struct channel *c, rte *e)
+bgp_preexport(struct channel *C, rte *e)
 {
   struct proto *SRC = e->src->proto;
-  struct bgp_proto *p = (struct bgp_proto *) (c->proto);
+  struct bgp_proto *p = (struct bgp_proto *) C->proto;
   struct bgp_proto *src = (SRC->proto == &proto_bgp) ? (struct bgp_proto *) SRC : NULL;
 
   /* Reject our routes */
@@ -2546,27 +2546,57 @@ bgp_rte_recalculate(rtable *table, net *net, rte *new, rte *old, rte *old_best)
     return !old_suppressed;
 }
 
-rte *
-bgp_rte_modify_stale(struct rte *r, struct linpool *pool)
+void
+bgp_rte_modify_stale(struct rt_export_request *req, const net_addr *n, struct rt_pending_export *rpe UNUSED, rte **feed, uint count)
 {
-  eattr *ea = ea_find(r->attrs, BGP_EA_ID(BA_COMMUNITY));
-  const struct adata *ad = ea ? ea->u.ptr : NULL;
-  uint flags = ea ? ea->flags : BAF_PARTIAL;
+  struct bgp_channel *c = SKIP_BACK(struct bgp_channel, stale_feed, req);
+  struct rt_import_hook *irh = c->c.in_req.hook;
 
-  if (ad && int_set_contains(ad, BGP_COMM_NO_LLGR))
-    return NULL;
+  /* Find our routes among others */
+  for (uint i=0; i<count; i++)
+  {
+    rte *r = feed[i];
+
+    if (
+	!rte_is_valid(r) ||			/* Not a valid route */
+	(r->sender != irh) ||			/* Not our route */
+	(r->stale_cycle == irh->stale_set))	/* A new route, do not mark as stale */
+      continue;
+
+    eattr *ea = ea_find(r->attrs, BGP_EA_ID(BA_COMMUNITY));
+    const struct adata *ad = ea ? ea->u.ptr : NULL;
+    uint flags = ea ? ea->flags : BAF_PARTIAL;
 
-  if (ad && int_set_contains(ad, BGP_COMM_LLGR_STALE))
-    return r;
+    /* LLGR not allowed, withdraw the route */
+    if (ad && int_set_contains(ad, BGP_COMM_NO_LLGR))
+    {
+      rte_import(&c->c.in_req, n, NULL, r->src);
+      continue;
+    }
+
+    /* Route already marked as LLGR, do nothing */
+    if (ad && int_set_contains(ad, BGP_COMM_LLGR_STALE))
+      continue;
 
-  _Thread_local static rte e0;
-  e0 = *r;
+    /* Store the tmp_linpool state to aggresively save memory */
+    struct lp_state tmpp;
+    lp_save(tmp_linpool, &tmpp);
 
-  bgp_set_attr_ptr(&e0.attrs, BA_COMMUNITY, flags,
-		   int_set_add(pool, ad, BGP_COMM_LLGR_STALE));
-  e0.pflags |= BGP_REF_STALE;
+    /* Mark the route as LLGR */
+    rte e0 = *r;
+    bgp_set_attr_ptr(&e0.attrs, BA_COMMUNITY, flags, int_set_add(tmp_linpool, ad, BGP_COMM_LLGR_STALE));
+    e0.pflags &= ~BGP_REF_NOT_STALE;
+    e0.pflags |= BGP_REF_STALE;
 
-  return &e0;
+    /* We need to update the route but keep it stale. */
+    ASSERT_DIE(irh->stale_set == irh->stale_valid + 1);
+    irh->stale_set--;
+    rte_import(&c->c.in_req, n, &e0, r->src);
+    irh->stale_set++;
+
+    /* Restore the memory state */
+    lp_restore(tmp_linpool, &tmpp);
+  }
 }
 
 
diff --git a/proto/bgp/bgp.c b/proto/bgp/bgp.c
index 6ffe8824..33849b0b 100644
--- a/proto/bgp/bgp.c
+++ b/proto/bgp/bgp.c
@@ -139,6 +139,9 @@ static void bgp_update_bfd(struct bgp_proto *p, const struct bfd_options *bfd);
 static int bgp_incoming_connection(sock *sk, uint dummy UNUSED);
 static void bgp_listen_sock_err(sock *sk UNUSED, int err);
 
+static void bgp_graceful_restart_feed(struct bgp_channel *c);
+
+
 /**
  * bgp_open - open a BGP instance
  * @p: BGP instance
@@ -373,6 +376,7 @@ bgp_close_conn(struct bgp_conn *conn)
   conn->keepalive_timer = NULL;
   rfree(conn->hold_timer);
   conn->hold_timer = NULL;
+
   rfree(conn->tx_ev);
   conn->tx_ev = NULL;
   rfree(conn->sk);
@@ -511,6 +515,7 @@ void
 bgp_stop(struct bgp_proto *p, int subcode, byte *data, uint len)
 {
   proto_notify_state(&p->p, PS_STOP);
+  p->uncork_ev->data = NULL;
   bgp_graceful_close_conn(&p->outgoing_conn, subcode, data, len);
   bgp_graceful_close_conn(&p->incoming_conn, subcode, data, len);
   ev_schedule(p->event);
@@ -760,25 +765,25 @@ bgp_handle_graceful_restart(struct bgp_proto *p)
       {
       case BGP_GRS_NONE:
 	c->gr_active = BGP_GRS_ACTIVE;
-	rt_refresh_begin(c->c.table, &c->c.in_req);
+	rt_refresh_begin(&c->c.in_req);
 	break;
 
       case BGP_GRS_ACTIVE:
-	rt_refresh_end(c->c.table, &c->c.in_req);
-	rt_refresh_begin(c->c.table, &c->c.in_req);
+	rt_refresh_end(&c->c.in_req);
+	rt_refresh_begin(&c->c.in_req);
 	break;
 
       case BGP_GRS_LLGR:
-	rt_refresh_begin(c->c.table, &c->c.in_req);
-	rt_modify_stale(c->c.table, &c->c.in_req);
+	rt_refresh_begin(&c->c.in_req);
+	bgp_graceful_restart_feed(c);
 	break;
       }
     }
     else
     {
       /* Just flush the routes */
-      rt_refresh_begin(c->c.table, &c->c.in_req);
-      rt_refresh_end(c->c.table, &c->c.in_req);
+      rt_refresh_begin(&c->c.in_req);
+      rt_refresh_end(&c->c.in_req);
     }
 
     /* Reset bucket and prefix tables */
@@ -796,6 +801,53 @@ bgp_handle_graceful_restart(struct bgp_proto *p)
   tm_start(p->gr_timer, p->conn->remote_caps->gr_time S);
 }
 
+static void
+bgp_graceful_restart_feed_done(struct rt_export_request *req)
+{
+  req->hook = NULL;
+}
+
+static void
+bgp_graceful_restart_feed_dump_req(struct rt_export_request *req)
+{
+  struct bgp_channel *c = SKIP_BACK(struct bgp_channel, stale_feed, req);
+  debug("  BGP-GR %s.%s export request %p\n", c->c.proto->name, c->c.name, req);
+}
+
+static void
+bgp_graceful_restart_feed_log_state_change(struct rt_export_request *req, u8 state)
+{
+  struct bgp_channel *c = SKIP_BACK(struct bgp_channel, stale_feed, req);
+  struct bgp_proto *p = (void *) c->c.proto;
+  BGP_TRACE(D_EVENTS, "Long-lived graceful restart export state changed to %s", rt_export_state_name(state));
+
+  if (state == TES_READY)
+    rt_stop_export(req, bgp_graceful_restart_feed_done);
+}
+
+static void
+bgp_graceful_restart_drop_export(struct rt_export_request *req UNUSED, const net_addr *n UNUSED, struct rt_pending_export *rpe UNUSED)
+{ /* Nothing to do */ }
+
+static void
+bgp_graceful_restart_feed(struct bgp_channel *c)
+{
+  c->stale_feed = (struct rt_export_request) {
+    .name = "BGP-GR",
+    .list = &global_work_list,
+    .trace_routes = c->c.debug | c->c.proto->debug,
+    .dump_req = bgp_graceful_restart_feed_dump_req,
+    .log_state_change = bgp_graceful_restart_feed_log_state_change,
+    .export_bulk = bgp_rte_modify_stale,
+    .export_one = bgp_graceful_restart_drop_export,
+  };
+
+  rt_request_export(&c->c.table->exporter, &c->stale_feed);
+}
+
+
+
+
 /**
  * bgp_graceful_restart_done - finish active BGP graceful restart
  * @c: BGP channel
@@ -819,7 +871,7 @@ bgp_graceful_restart_done(struct bgp_channel *c)
     BGP_TRACE(D_EVENTS, "Neighbor graceful restart done");
 
   tm_stop(c->stale_timer);
-  rt_refresh_end(c->c.table, &c->c.in_req);
+  rt_refresh_end(&c->c.in_req);
 }
 
 /**
@@ -861,7 +913,7 @@ bgp_graceful_restart_timeout(timer *t)
       /* Channel is in GR, and supports LLGR -> start LLGR */
       c->gr_active = BGP_GRS_LLGR;
       tm_start(c->stale_timer, c->stale_time S);
-      rt_modify_stale(c->c.table, &c->c.in_req);
+      bgp_graceful_restart_feed(c);
     }
   }
   else
@@ -899,7 +951,7 @@ bgp_refresh_begin(struct bgp_channel *c)
   { log(L_WARN "%s: BEGIN-OF-RR received before END-OF-RIB, ignoring", p->p.name); return; }
 
   c->load_state = BFS_REFRESHING;
-  rt_refresh_begin(c->c.table, &c->c.in_req);
+  rt_refresh_begin(&c->c.in_req);
 }
 
 /**
@@ -920,7 +972,7 @@ bgp_refresh_end(struct bgp_channel *c)
   { log(L_WARN "%s: END-OF-RR received without prior BEGIN-OF-RR, ignoring", p->p.name); return; }
 
   c->load_state = BFS_NONE;
-  rt_refresh_end(c->c.table, &c->c.in_req);
+  rt_refresh_end(&c->c.in_req);
 }
 
 
@@ -1526,6 +1578,8 @@ bgp_start(struct proto *P)
   p->last_rx_update = 0;
 
   p->event = ev_new_init(p->p.pool, bgp_decision, p);
+  p->uncork_ev = ev_new_init(p->p.pool, bgp_uncork, p);
+
   p->startup_timer = tm_new_init(p->p.pool, bgp_startup_timeout, p, 0, 0);
   p->gr_timer = tm_new_init(p->p.pool, bgp_graceful_restart_timeout, p, 0, 0);
 
@@ -1672,7 +1726,6 @@ bgp_init(struct proto_config *CF)
   P->rte_better = bgp_rte_better;
   P->rte_mergable = bgp_rte_mergable;
   P->rte_recalculate = cf->deterministic_med ? bgp_rte_recalculate : NULL;
-  P->rte_modify = bgp_rte_modify_stale;
   P->rte_igp_metric = bgp_rte_igp_metric;
 
   p->cf = cf;
diff --git a/proto/bgp/bgp.h b/proto/bgp/bgp.h
index 003893e0..469f0cb9 100644
--- a/proto/bgp/bgp.h
+++ b/proto/bgp/bgp.h
@@ -319,6 +319,7 @@ struct bgp_proto {
   struct bgp_socket *sock;		/* Shared listening socket */
   struct bfd_request *bfd_req;		/* BFD request, if BFD is used */
   struct birdsock *postponed_sk;	/* Postponed incoming socket for dynamic BGP */
+  event *uncork_ev;			/* Uncork event in case of congestion */
   struct bgp_stats stats;		/* BGP statistics */
   btime last_established;		/* Last time of enter/leave of established state */
   btime last_rx_update;			/* Last time of RX update */
@@ -371,6 +372,7 @@ struct bgp_channel {
 
   timer *stale_timer;			/* Long-lived stale timer for LLGR */
   u32 stale_time;			/* Stored LLGR stale time from last session */
+  struct rt_export_request stale_feed;	/* Feeder request for stale route modification */
 
   u8 add_path_rx;			/* Session expects receive of ADD-PATH extended NLRI */
   u8 add_path_tx;			/* Session expects transmit of ADD-PATH extended NLRI */
@@ -576,7 +578,7 @@ void bgp_done_prefix(struct bgp_channel *c, struct bgp_prefix *px, struct bgp_bu
 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);
-struct rte *bgp_rte_modify_stale(struct rte *r, struct linpool *pool);
+void bgp_rte_modify_stale(struct rt_export_request *req, const net_addr *n, struct rt_pending_export *rpe UNUSED, rte **feed, uint count);
 u32 bgp_rte_igp_metric(const rte *);
 void bgp_rt_notify(struct proto *P, struct channel *C, const net_addr *n, rte *new, const rte *old);
 int bgp_preexport(struct channel *, struct rte *);
@@ -609,6 +611,7 @@ void bgp_schedule_packet(struct bgp_conn *conn, struct bgp_channel *c, int type)
 void bgp_kick_tx(void *vconn);
 void bgp_tx(struct birdsock *sk);
 int bgp_rx(struct birdsock *sk, uint size);
+void bgp_uncork(void *vp);
 const char * bgp_error_dsc(unsigned code, unsigned subcode);
 void bgp_log_error(struct bgp_proto *p, u8 class, char *msg, unsigned code, unsigned subcode, byte *data, unsigned len);
 
diff --git a/proto/bgp/packets.c b/proto/bgp/packets.c
index 4d4ae3eb..de976588 100644
--- a/proto/bgp/packets.c
+++ b/proto/bgp/packets.c
@@ -3175,6 +3175,21 @@ bgp_rx_packet(struct bgp_conn *conn, byte *pkt, uint len)
   }
 }
 
+void
+bgp_uncork(void *vp)
+{
+  struct bgp_proto *p = vp;
+
+  if (p && p->conn && (p->conn->state == BS_ESTABLISHED) && !p->conn->sk->rx_hook)
+  {
+    struct birdsock *sk = p->conn->sk;
+    ASSERT_DIE(sk->rpos > sk->rbuf);
+    sk->rx_hook = bgp_rx;
+    bgp_rx(sk, sk->rpos - sk->rbuf);
+    BGP_TRACE(D_PACKETS, "Uncorked");
+  }
+}
+
 /**
  * bgp_rx - handle received data
  * @sk: socket
@@ -3189,6 +3204,7 @@ int
 bgp_rx(sock *sk, uint size)
 {
   struct bgp_conn *conn = sk->data;
+  struct bgp_proto *p = conn->bgp;
   byte *pkt_start = sk->rbuf;
   byte *end = pkt_start + size;
   uint i, len;
@@ -3198,6 +3214,12 @@ bgp_rx(sock *sk, uint size)
     {
       if ((conn->state == BS_CLOSE) || (conn->sk != sk))
 	return 0;
+      if ((conn->state == BS_ESTABLISHED) && rt_cork_check(conn->bgp->uncork_ev))
+      {
+	sk->rx_hook = NULL;
+	BGP_TRACE(D_PACKETS, "Corked");
+	return 0;
+      }
       for(i=0; i<16; i++)
 	if (pkt_start[i] != 0xff)
 	  {