Merge branch 'master' into int-new-channels

doc/bird.sgml

@@ -318,8 +318,9 @@ protocol rip {
 <p><descrip>
 	<tag>include "<m/filename/"</tag>
 	This statement causes inclusion of a new file. <m/Filename/ could also
-	be a wildcard. The maximal depth is 8. Note that this statement could be
+	be a wildcard, in that case matching files are included in alphabetic
-	used anywhere in the config file, not just as a top-level option.
+	order. The maximal depth is 8. Note that this statement could be used
+	anywhere in the config file, not just as a top-level option.
 
 	<tag><label id="dsc-log">log "<m/filename/"|syslog [name <m/name/]|stderr all|{ <m/list of classes/ }</tag>
 	Set logging of messages having the given class (either <cf/all/ or
@@ -1119,9 +1120,12 @@ foot).
 
 	<cf><m/P/.last</cf> returns the last ASN (the source ASN) in path <m/P/.
 
+	<cf><m/P/.last_nonaggregated</cf> returns the last ASN in the non-aggregated part of the path <m/P/.
+
 	Both <cf/first/ and <cf/last/ return zero if there is no appropriate
 	ASN, for example if the path contains an AS set element as the first (or
-	the last) part.
+	the last) part. If the path ends with an AS set, <cf/last_nonaggregated/
+	may be used to get last ASN before any AS set.
 
 	<cf><m/P/.len</cf> returns the length of path <m/P/.
 
@@ -1859,6 +1863,11 @@ using the following configuration parameters:
 	in neighbor's implementation of 4B AS extension. Even when disabled
 	(off), BIRD behaves internally as AS4-aware BGP router. Default: on.
 
+	<tag>enable extended messages <m/switch/</tag>
+	The BGP protocol uses maximum message length of 4096 bytes. This option
+	provides an extension to allow extended messages with length up
+	to 65535 bytes. Default: off.
+
 	<tag>capabilities <m/switch/</tag>
 	Use capability advertisement to advertise optional capabilities. This is
 	standard behavior for newer BGP implementations, but there might be some
@@ -2608,8 +2617,8 @@ protocol ospf &lt;name&gt; {
 	updates. Default value is 5.
 
 	<tag>priority <M>num</M></tag>
-	On every multiple access network (e.g., the Ethernet) Designed Router
+	On every multiple access network (e.g., the Ethernet) Designated Router
-	and Backup Designed router are elected. These routers have some special
+	and Backup Designated router are elected. These routers have some special
 	functions in the flooding process. Higher priority increases preferences
 	in this election. Routers with priority 0 are not eligible. Default
 	value is 1.

filter/config.Y

@@ -282,7 +282,7 @@ CF_KEYWORDS(FUNCTION, PRINT, PRINTN, UNSET, RETURN,
 	LEN,
 	DEFINED,
 	ADD, DELETE, CONTAINS, RESET,
-	PREPEND, FIRST, LAST, MATCH,
+	PREPEND, FIRST, LAST, LAST_NONAGGREGATED, MATCH,
 	EMPTY,
 	FILTER, WHERE, EVAL)
 
@@ -743,6 +743,7 @@ term:
  | term '.' MASK '(' term ')' { $$ = f_new_inst(); $$->code = P('i','M'); $$->a1.p = $1; $$->a2.p = $5; }
  | term '.' FIRST { $$ = f_new_inst(); $$->code = P('a','f'); $$->a1.p = $1; }
  | term '.' LAST  { $$ = f_new_inst(); $$->code = P('a','l'); $$->a1.p = $1; }
+ | term '.' LAST_NONAGGREGATED  { $$ = f_new_inst(); $$->code = P('a','L'); $$->a1.p = $1; }
 
 /* Communities */
 /* This causes one shift/reduce conflict

filter/filter.c

@@ -1056,6 +1056,14 @@ interpret(struct f_inst *what)
     res.type = T_INT;
     res.val.i = as;
     break;
+  case P('a','L'):	/* Get last ASN from non-aggregated part of AS PATH */
+    ONEARG;
+    if (v1.type != T_PATH)
+      runtime( "AS path expected" );
+
+    res.type = T_INT;
+    res.val.i = as_path_get_last_nonaggregated(v1.val.ad);
+    break;
   case 'r':
     ONEARG;
     res = v1;

lib/lists.c

@@ -41,7 +41,7 @@ add_tail(list *l, node *n)
 {
   node *z = l->tail;
 
-  n->next = (node *) &l->null;
+  n->next = &l->tail_node;
   n->prev = z;
   z->next = n;
   l->tail = n;
@@ -60,7 +60,7 @@ add_head(list *l, node *n)
   node *z = l->head;
 
   n->next = z;
-  n->prev = (node *) &l->head;
+  n->prev = &l->head_node;
   z->prev = n;
   l->head = n;
 }
@@ -88,7 +88,7 @@ insert_node(node *n, node *after)
  * rem_node - remove a node from a list
  * @n: node to be removed
  *
- * Removes a node @n from the list it's linked in.
+ * Removes a node @n from the list it's linked in. Afterwards, node @n is cleared.
  */
 LIST_INLINE void
 rem_node(node *n)
@@ -96,23 +96,6 @@ rem_node(node *n)
   node *z = n->prev;
   node *x = n->next;
 
-  z->next = x;
-  x->prev = z;
-}
-
-/**
- * rem2_node - remove a node from a list, with cleanup
- * @n: node to be removed
- *
- * Removes a node @n from the list it's linked in and resets its pointers to NULL.
- * Useful if you want to distinguish between linked and unlinked nodes.
- */
-LIST_INLINE void
-rem2_node(node *n)
-{
-  node *z = n->prev;
-  node *x = n->next;
-
   z->next = x;
   x->prev = z;
   n->next = NULL;
@@ -150,9 +133,9 @@ replace_node(node *old, node *new)
 LIST_INLINE void
 init_list(list *l)
 {
-  l->head = (node *) &l->null;
+  l->head = &l->tail_node;
   l->null = NULL;
-  l->tail = (node *) &l->head;
+  l->tail = &l->head_node;
 }
 
 /**
@@ -172,6 +155,6 @@ add_tail_list(list *to, list *l)
   p->next = q;
   q->prev = p;
   q = l->tail;
-  q->next = (node *) &to->null;
+  q->next = &to->tail_node;
   to->tail = q;
 }

lib/lists.h

@@ -26,10 +26,23 @@ typedef struct node {
   struct node *next, *prev;
 } node;
 
-typedef struct list {			/* In fact two overlayed nodes */
+typedef union list {			/* In fact two overlayed nodes */
-  struct node *head, *null, *tail;
+  struct {				/* Head node */
+    struct node head_node;
+    void *head_padding;
+  };
+  struct {				/* Tail node */
+    void *tail_padding;
+    struct node tail_node;
+  };
+  struct {				/* Split to separate pointers */
+    struct node *head;
+    struct node *null;
+    struct node *tail;
+  };
 } list;
 
+
 #define NODE (node *)
 #define HEAD(list) ((void *)((list).head))
 #define TAIL(list) ((void *)((list).tail))
@@ -64,7 +77,6 @@ typedef struct list {			/* In fact two overlayed nodes */
 void add_tail(list *, node *);
 void add_head(list *, node *);
 void rem_node(node *);
-void rem2_node(node *);
 void add_tail_list(list *, list *);
 void init_list(list *);
 void insert_node(node *, node *);

lib/resource.c

@@ -163,6 +163,7 @@ rfree(void *res)
   if (r->n.next)
     rem_node(&r->n);
   r->class->free(r);
+  r->class = NULL;
   xfree(r);
 }
 
@@ -383,16 +384,9 @@ mb_allocz(pool *p, unsigned size)
 void *
 mb_realloc(void *m, unsigned size)
 {
-  struct mblock *ob = NULL;
+  struct mblock *b = SKIP_BACK(struct mblock, data, m);
 
-  if (m)
+  b = xrealloc(b, sizeof(struct mblock) + size);
-    {
-      ob = SKIP_BACK(struct mblock, data, m);
-      if (ob->r.n.next)
-	rem_node(&ob->r.n);
-    }
-
-  struct mblock *b = xrealloc(ob, sizeof(struct mblock) + size);
   replace_node(&b->r.n, &b->r.n);
   b->size = size;
   return b->data;

lib/socket.h

@@ -27,6 +27,7 @@ typedef struct birdsock {
   struct iface *iface;			/* Interface; specify this for broad/multicast sockets */
 
   byte *rbuf, *rpos;			/* NULL=allocate automatically */
+  uint fast_rx;				/* RX has higher priority in event loop */
   uint rbsize;
   int (*rx_hook)(struct birdsock *, int size); /* NULL=receiving turned off, returns 1 to clear rx buffer */
 

nest/a-path.c

@@ -220,7 +220,7 @@ as_path_get_last(struct adata *path, u32 *orig_as)
 	      p += BS * len;
 	    }
 	  break;
-	default: bug("as_path_get_first: Invalid path segment");
+	default: bug("Invalid path segment");
 	}
     }
 
@@ -229,6 +229,35 @@ as_path_get_last(struct adata *path, u32 *orig_as)
   return found;
 }
 
+u32
+as_path_get_last_nonaggregated(struct adata *path)
+{
+  u8 *p = path->data;
+  u8 *q = p+path->length;
+  u32 res = 0;
+  int len;
+
+  while (p<q)
+    {
+      switch (*p++)
+	{
+	case AS_PATH_SET:
+	  return res;
+
+	case AS_PATH_SEQUENCE:
+	  if (len = *p++)
+	    res = get_as(p + BS * (len - 1));
+	  p += BS * len;
+	  break;
+
+	default: bug("Invalid path segment");
+	}
+    }
+
+  return res;
+}
+
+
 int
 as_path_get_first(struct adata *path, u32 *last_as)
 {

nest/attrs.h

@@ -35,6 +35,7 @@ int as_path_getlen(struct adata *path);
 int as_path_getlen_int(struct adata *path, int bs);
 int as_path_get_first(struct adata *path, u32 *orig_as);
 int as_path_get_last(struct adata *path, u32 *last_as);
+u32 as_path_get_last_nonaggregated(struct adata *path);
 int as_path_contains(struct adata *path, u32 as, int min);
 int as_path_match_set(struct adata *path, struct f_tree *set);
 struct adata *as_path_filter(struct linpool *pool, struct adata *path, struct f_tree *set, u32 key, int pos);

nest/config.Y

@@ -112,7 +112,7 @@ idval:
      else if (($1->class == (SYM_CONSTANT | T_IP)) && ipa_is_ip4(SYM_VAL($1).ip))
        $$ = ipa_to_u32(SYM_VAL($1).ip);
      else
-       cf_error("Number of IPv4 address constant expected");
+       cf_error("Number or IPv4 address constant expected");
    }
  ;
 

nest/proto.c

@@ -264,6 +264,7 @@ channel_stop_export(struct channel *c)
     rt_feed_channel_abort(c);
 
   c->export_state = ES_DOWN;
+  c->stats.exp_routes = 0;
 }
 
 static void
@@ -299,7 +300,7 @@ channel_do_flush(struct channel *c)
 static void
 channel_do_down(struct channel *c)
 {
-  rem2_node(&c->table_node);
+  rem_node(&c->table_node);
   rt_unlock_table(c->table);
   c->proto->active_channels--;
 

nest/route.h

@@ -232,8 +232,8 @@ typedef struct rte {
     struct {				/* Routes generated by krt sync (both temporary and inherited ones) */
       s8 src;				/* Alleged route source (see krt.h) */
       u8 proto;				/* Kernel source protocol ID */
-      u8 type;				/* Kernel route type */
       u8 seen;				/* Seen during last scan */
+      u8 best;				/* Best route in network, propagated to core */
       u32 metric;			/* Kernel metric */
     } krt;
   } u;

proto/bfd/bfd.c

@@ -872,7 +872,7 @@ bfd_notify_hook(sock *sk, int len)
   WALK_LIST_FIRST(s, tmp_list)
   {
     bfd_lock_sessions(p);
-    rem2_node(&s->n);
+    rem_node(&s->n);
     state = s->loc_state;
     diag = s->loc_diag;
     bfd_unlock_sessions(p);

proto/bfd/io.c

@@ -576,7 +576,7 @@ sockets_close_fds(struct birdloop *loop)
   loop->close_scheduled = 0;
 }
 
-int sk_read(sock *s);
+int sk_read(sock *s, int revents);
 int sk_write(sock *s);
 
 static void
@@ -605,7 +605,7 @@ sockets_fire(struct birdloop *loop)
 
     if (pfd->revents & POLLIN)
       while (e && *psk && (*psk)->rx_hook)
-	e = sk_read(*psk);
+	e = sk_read(*psk, 0);
 
     e = 1;
     if (pfd->revents & POLLOUT)

proto/bgp/bgp.c

@@ -374,6 +374,8 @@ bgp_conn_enter_established_state(struct bgp_conn *conn)
   if (ipa_zero(p->source_addr))
     p->source_addr = conn->sk->saddr;
 
+  conn->sk->fast_rx = 0;
+
   p->conn = conn;
   p->last_error_class = 0;
   p->last_error_code = 0;
@@ -666,6 +668,10 @@ bgp_keepalive_timeout(timer *t)
 
   DBG("BGP: Keepalive timer\n");
   bgp_schedule_packet(conn, PKT_KEEPALIVE);
+
+  /* Kick TX a bit faster */
+  if (ev_active(conn->tx_ev))
+    ev_run(conn->tx_ev);
 }
 
 static void
@@ -696,6 +702,7 @@ bgp_setup_sk(struct bgp_conn *conn, sock *s)
 {
   s->data = conn;
   s->err_hook = bgp_sock_err;
+  s->fast_rx = 1;
   conn->sk = s;
 }
 
@@ -813,7 +820,13 @@ bgp_incoming_connection(sock *sk, int dummy UNUSED)
       return 0;
     }
 
-  /* We are in proper state and there is no other incoming connection */
+  /*
+   * BIRD should keep multiple incoming connections in OpenSent state (for
+   * details RFC 4271 8.2.1 par 3), but it keeps just one. Duplicate incoming
+   * connections are rejected istead. The exception is the case where an
+   * incoming connection triggers a graceful restart.
+   */
+
   acc = (p->p.proto_state == PS_START || p->p.proto_state == PS_UP) &&
     (p->start_state >= BSS_CONNECT) && (!p->incoming_conn.sk);
 
@@ -823,6 +836,10 @@ bgp_incoming_connection(sock *sk, int dummy UNUSED)
       bgp_handle_graceful_restart(p);
       bgp_conn_enter_idle_state(p->conn);
       acc = 1;
+
+      /* There might be separate incoming connection in OpenSent state */
+      if (p->incoming_conn.state > BS_ACTIVE)
+	bgp_close_conn(&p->incoming_conn);
     }
 
   BGP_TRACE(D_EVENTS, "Incoming connection from %I%J (port %d) %s",

proto/bgp/packets.c

@@ -163,6 +163,14 @@ bgp_put_cap_rr(struct bgp_proto *p UNUSED, byte *buf)
   return buf;
 }
 
+static byte *
+bgp_put_cap_ext_msg(struct bgp_proto *p UNUSED, byte *buf)
+{
+  *buf++ = 6;		/* Capability 6: Support for extended messages */
+  *buf++ = 0;		/* Capability data length */
+  return buf;
+}
+
 static byte *
 bgp_put_cap_gr1(struct bgp_proto *p, byte *buf)
 {
@@ -223,14 +231,6 @@ bgp_put_cap_err(struct bgp_proto *p UNUSED, byte *buf)
   return buf;
 }
 
-static byte *
-bgp_put_cap_ext_msg(struct bgp_proto *p UNUSED, byte *buf)
-{
-  *buf++ = 230;		/* Capability TBD: Support for extended messages */
-  *buf++ = 0;		/* Capability data length */
-  return buf;
-}
-
 
 static byte *
 bgp_create_open(struct bgp_conn *conn, byte *buf)
@@ -827,6 +827,12 @@ bgp_parse_capabilities(struct bgp_conn *conn, byte *opt, int len)
 	  conn->peer_refresh_support = 1;
 	  break;
 
+	case 6: /* Extended message length capability, draft */
+	  if (cl != 0)
+	    goto err;
+	  conn->peer_ext_messages_support = 1;
+	  break;
+
 	case 64: /* Graceful restart capability, RFC 4724 */
 	  if (cl % 4 != 2)
 	    goto err;
@@ -867,12 +873,6 @@ bgp_parse_capabilities(struct bgp_conn *conn, byte *opt, int len)
 	  conn->peer_enhanced_refresh_support = 1;
 	  break;
 
-	case 230: /* Extended message length capability, draft, cap number TBD */
-	  if (cl != 0)
-	    goto err;
-	  conn->peer_ext_messages_support = 1;
-	  break;
-
 	  /* We can safely ignore all other capabilities */
 	}
       len -= 2 + cl;

proto/ospf/iface.c

@@ -595,10 +595,10 @@ ospf_iface_new(struct ospf_area *oa, struct ifa *addr, struct ospf_iface_patt *i
   if (ospf_is_v2(p) && (ifa->type == OSPF_IT_NBMA) && (addr->flags & IA_PEER))
     ifa->type = OSPF_IT_PTMP;
 
-  if ((ifa->type == OSPF_IT_BCAST) && !(iface->flags & if_multi_flag))
+  if ((ifa->type == OSPF_IT_BCAST) && !(iface->flags & if_multi_flag) && !ifa->stub)
     ifa->type = OSPF_IT_NBMA;
 
-  if ((ifa->type == OSPF_IT_PTP) && !(iface->flags & if_multi_flag))
+  if ((ifa->type == OSPF_IT_PTP) && !(iface->flags & if_multi_flag) && !ifa->stub)
     ifa->type = OSPF_IT_PTMP;
 
   if (ifa->type != old_type)

proto/ospf/neighbor.c

@@ -108,6 +108,7 @@ ospf_neigh_down(struct ospf_neighbor *n)
 {
   struct ospf_iface *ifa = n->ifa;
   struct ospf_proto *p = ifa->oa->po;
+  u32 rid = n->rid;
 
   if ((ifa->type == OSPF_IT_NBMA) || (ifa->type == OSPF_IT_PTMP))
   {
@@ -121,7 +122,7 @@ ospf_neigh_down(struct ospf_neighbor *n)
   rem_node(NODE n);
   rfree(n->pool);
 
-  OSPF_TRACE(D_EVENTS, "Neighbor %R on %s removed", n->rid, ifa->ifname);
+  OSPF_TRACE(D_EVENTS, "Neighbor %R on %s removed", rid, ifa->ifname);
 }
 
 /**

proto/ospf/topology.c

@@ -278,7 +278,7 @@ ospf_originate_lsa(struct ospf_proto *p, struct ospf_new_lsa *lsa)
   if (!SNODE_VALID(en))
     s_add_tail(&p->lsal, SNODE en);
 
-  if (en->lsa_body == NULL)
+  if (!en->nf || !en->lsa_body)
     en->nf = lsa->nf;
 
   if (en->nf != lsa->nf)

proto/rip/config.Y

@@ -137,7 +137,7 @@ rip_iface_item:
  | TIMEOUT TIME expr	{ RIP_IFACE->timeout_time = $3; if ($3<=0) cf_error("Timeout time must be positive"); }
  | GARBAGE TIME expr	{ RIP_IFACE->garbage_time = $3; if ($3<=0) cf_error("Garbage time must be positive"); }
  | ECMP WEIGHT expr	{ RIP_IFACE->ecmp_weight = $3 - 1; if (($3<1) || ($3>256)) cf_error("ECMP weight must be in range 1-256"); }
- | RX BUFFER expr	{ RIP_IFACE->rx_buffer = $3; if (($3<256) || ($3>65535)) cf_error("TX length must be in range 256-65535"); }
+ | RX BUFFER expr	{ RIP_IFACE->rx_buffer = $3; if (($3<256) || ($3>65535)) cf_error("RX length must be in range 256-65535"); }
  | TX LENGTH expr	{ RIP_IFACE->tx_length = $3; if (($3<256) || ($3>65535)) cf_error("TX length must be in range 256-65535"); }
  | TX tos		{ RIP_IFACE->tx_tos = $2; }
  | TX PRIORITY expr	{ RIP_IFACE->tx_priority = $3; }

sysdep/bsd/krt-sock.c

@@ -528,9 +528,8 @@ krt_read_route(struct ks_msg *msg, struct krt_proto *p, int scan)
   e->net = net;
   e->u.krt.src = src;
   e->u.krt.proto = src2;
-
+  e->u.krt.seen = 0;
-  /* These are probably too Linux-specific */
+  e->u.krt.best = 0;
-  e->u.krt.type = 0;
   e->u.krt.metric = 0;
 
   if (scan)

sysdep/linux/netlink.c

@@ -1204,7 +1204,8 @@ nl_parse_route(struct nlmsghdr *h, int scan)
   e->net = net;
   e->u.krt.src = src;
   e->u.krt.proto = i->rtm_protocol;
-  e->u.krt.type = i->rtm_type;
+  e->u.krt.seen = 0;
+  e->u.krt.best = 0;
   e->u.krt.metric = 0;
 
   if (a[RTA_PRIORITY])

sysdep/unix/io.c

@@ -19,6 +19,7 @@
 #include <sys/socket.h>
 #include <sys/uio.h>
 #include <sys/un.h>
+#include <poll.h>
 #include <unistd.h>
 #include <fcntl.h>
 #include <errno.h>
@@ -41,12 +42,12 @@
 #include "lib/sysio.h"
 
 /* Maximum number of calls of tx handler for one socket in one
- * select iteration. Should be small enough to not monopolize CPU by
+ * poll iteration. Should be small enough to not monopolize CPU by
  * one protocol instance.
  */
 #define MAX_STEPS 4
 
-/* Maximum number of calls of rx handler for all sockets in one select
+/* Maximum number of calls of rx handler for all sockets in one poll
    iteration. RX callbacks are often much more costly so we limit
    this to gen small latencies */
 #define MAX_RX_STEPS 4
@@ -1023,7 +1024,6 @@ sk_log_error(sock *s, const char *p)
 static list sock_list;
 static struct birdsock *current_sock;
 static struct birdsock *stored_sock;
-static int sock_recalc_fdsets_p;
 
 static inline sock *
 sk_next(sock *s)
@@ -1079,7 +1079,6 @@ sk_free(resource *r)
     if (s == stored_sock)
       stored_sock = sk_next(s);
     rem_node(&s->n);
-    sock_recalc_fdsets_p = 1;
   }
 }
 
@@ -1277,7 +1276,6 @@ static void
 sk_insert(sock *s)
 {
   add_tail(&sock_list, &s->n);
-  sock_recalc_fdsets_p = 1;
 }
 
 static void
@@ -1329,18 +1327,6 @@ sk_passive_connected(sock *s, int type)
       log(L_WARN "SOCK: Cannot get remote IP address for TCP<");
   }
 
-  if (fd >= FD_SETSIZE)
-  {
-    /* FIXME: Call err_hook instead ? */
-    log(L_ERR "SOCK: Incoming connection from %I%J (port %d) %s",
-	t->daddr, ipa_is_link_local(t->daddr) ? t->iface : NULL,
-	t->dport, "rejected due to FD_SETSIZE limit");
-    close(fd);
-    t->fd = -1;
-    rfree(t);
-    return 1;
-  }
-
   if (sk_setup(t) < 0)
   {
     /* FIXME: Call err_hook instead ? */
@@ -1416,9 +1402,6 @@ sk_open(sock *s)
   if (fd < 0)
     ERR("socket");
 
-  if (fd >= FD_SETSIZE)
-    ERR2("FD_SETSIZE limit reached");
-
   s->fd = fd;
 
   if (sk_setup(s) < 0)
@@ -1696,19 +1679,12 @@ sk_maybe_write(sock *s)
 int
 sk_rx_ready(sock *s)
 {
-  fd_set rd, wr;
-  struct timeval timo;
   int rv;
-
+  struct pollfd pfd = { .fd = s->fd };
-  FD_ZERO(&rd);
+  pfd.events |= POLLIN;
-  FD_ZERO(&wr);
-  FD_SET(s->fd, &rd);
-
-  timo.tv_sec = 0;
-  timo.tv_usec = 0;
 
  redo:
-  rv = select(s->fd+1, &rd, &wr, NULL, &timo);
+  rv = poll(&pfd, 1, 0);
 
   if ((rv < 0) && (errno == EINTR || errno == EAGAIN))
     goto redo;
@@ -1777,7 +1753,7 @@ sk_send_full(sock *s, unsigned len, struct iface *ifa,
  /* sk_read() and sk_write() are called from BFD's event loop */
 
 int
-sk_read(sock *s)
+sk_read(sock *s, int revents)
 {
   switch (s->type)
   {
@@ -1796,6 +1772,11 @@ sk_read(sock *s)
       {
 	if (errno != EINTR && errno != EAGAIN)
 	  s->err_hook(s, errno);
+	else if (errno == EAGAIN && !(revents & POLLIN))
+	{
+	  log(L_ERR "Got EAGAIN from read when revents=%x (without POLLIN)", revents);
+	  s->err_hook(s, 0);
+	}
       }
       else if (!c)
 	s->err_hook(s, 0);
@@ -2068,62 +2049,63 @@ static int short_loops = 0;
 void
 io_loop(void)
 {
-  fd_set rd, wr;
+  int poll_tout;
-  struct timeval timo;
   time_t tout;
-  int hi, events;
+  int nfds, events, pout;
   sock *s;
   node *n;
+  int fdmax = 256;
+  struct pollfd *pfd = xmalloc(fdmax * sizeof(struct pollfd));
 
   watchdog_start1();
-  sock_recalc_fdsets_p = 1;
   for(;;)
     {
       events = ev_run_list(&global_event_list);
+    timers:
       update_times();
       tout = tm_first_shot();
       if (tout <= now)
 	{
 	  tm_shot();
-	  continue;
+	  goto timers;
 	}
-      timo.tv_sec = events ? 0 : MIN(tout - now, 3);
+      poll_tout = (events ? 0 : MIN(tout - now, 3)) * 1000; /* Time in milliseconds */
-      timo.tv_usec = 0;
 
       io_close_event();
 
-      if (sock_recalc_fdsets_p)
+      nfds = 0;
-	{
-	  sock_recalc_fdsets_p = 0;
-	  FD_ZERO(&rd);
-	  FD_ZERO(&wr);
-	}
-
-      hi = 0;
       WALK_LIST(n, sock_list)
 	{
+	  pfd[nfds] = (struct pollfd) { .fd = -1 }; /* everything other set to 0 by this */
 	  s = SKIP_BACK(sock, n, n);
 	  if (s->rx_hook)
 	    {
-	      FD_SET(s->fd, &rd);
+	      pfd[nfds].fd = s->fd;
-	      if (s->fd > hi)
+	      pfd[nfds].events |= POLLIN;
-		hi = s->fd;
 	    }
-	  else
-	    FD_CLR(s->fd, &rd);
 	  if (s->tx_hook && s->ttx != s->tpos)
 	    {
-	      FD_SET(s->fd, &wr);
+	      pfd[nfds].fd = s->fd;
-	      if (s->fd > hi)
+	      pfd[nfds].events |= POLLOUT;
-		hi = s->fd;
+	    }
+	  if (pfd[nfds].fd != -1)
+	    {
+	      s->index = nfds;
+	      nfds++;
 	    }
 	  else
-	    FD_CLR(s->fd, &wr);
+	    s->index = -1;
+
+	  if (nfds >= fdmax)
+	    {
+	      fdmax *= 2;
+	      pfd = xrealloc(pfd, fdmax * sizeof(struct pollfd));
+	    }
 	}
 
       /*
        * Yes, this is racy. But even if the signal comes before this test
-       * and entering select(), it gets caught on the next timer tick.
+       * and entering poll(), it gets caught on the next timer tick.
        */
 
       if (async_config_flag)
@@ -2148,18 +2130,18 @@ io_loop(void)
 	  continue;
 	}
 
-      /* And finally enter select() to find active sockets */
+      /* And finally enter poll() to find active sockets */
       watchdog_stop();
-      hi = select(hi+1, &rd, &wr, NULL, &timo);
+      pout = poll(pfd, nfds, poll_tout);
       watchdog_start();
 
-      if (hi < 0)
+      if (pout < 0)
 	{
 	  if (errno == EINTR || errno == EAGAIN)
 	    continue;
-	  die("select: %m");
+	  die("poll: %m");
 	}
-      if (hi)
+      if (pout)
 	{
 	  /* guaranteed to be non-empty */
 	  current_sock = SKIP_BACK(sock, n, HEAD(sock_list));
@@ -2167,23 +2149,29 @@ io_loop(void)
 	  while (current_sock)
 	    {
 	      sock *s = current_sock;
+	      if (s->index == -1)
+		{
+		  current_sock = sk_next(s);
+		  goto next;
+		}
+
 	      int e;
 	      int steps;
 
 	      steps = MAX_STEPS;
-	      if ((s->type >= SK_MAGIC) && FD_ISSET(s->fd, &rd) && s->rx_hook)
+	      if (s->fast_rx && (pfd[s->index].revents & (POLLIN | POLLHUP | POLLERR)) && s->rx_hook)
 		do
 		  {
 		    steps--;
 		    io_log_event(s->rx_hook, s->data);
-		    e = sk_read(s);
+		    e = sk_read(s, pfd[s->index].revents);
 		    if (s != current_sock)
 		      goto next;
 		  }
 		while (e && s->rx_hook && steps);
 
 	      steps = MAX_STEPS;
-	      if (FD_ISSET(s->fd, &wr))
+	      if (pfd[s->index].revents & POLLOUT)
 		do
 		  {
 		    steps--;
@@ -2210,13 +2198,17 @@ io_loop(void)
 	  while (current_sock && count < MAX_RX_STEPS)
 	    {
 	      sock *s = current_sock;
-	      int e UNUSED;
+	      if (s->index == -1)
+		{
+		  current_sock = sk_next(s);
+		  goto next2;
+		}
 
-	      if ((s->type < SK_MAGIC) && FD_ISSET(s->fd, &rd) && s->rx_hook)
+	      if (!s->fast_rx && (pfd[s->index].revents & (POLLIN | POLLHUP | POLLERR)) && s->rx_hook)
 		{
 		  count++;
 		  io_log_event(s->rx_hook, s->data);
-		  e = sk_read(s);
+		  sk_read(s, pfd[s->index].revents);
 		  if (s != current_sock)
 		      goto next2;
 		}

sysdep/unix/krt.c

@@ -412,46 +412,58 @@ again:
     {
       rte *e, **ee, *best, **pbest, *old_best;
 
-      old_best = n->routes;
+      /*
+       * Note that old_best may be NULL even if there was an old best route in
+       * the previous step, because it might be replaced in krt_learn_scan().
+       * But in that case there is a new valid best route.
+       */
+
+      old_best = NULL;
       best = NULL;
       pbest = NULL;
       ee = &n->routes;
       while (e = *ee)
 	{
+	  if (e->u.krt.best)
+	    old_best = e;
+
 	  if (!e->u.krt.seen)
 	    {
 	      *ee = e->next;
 	      rte_free(e);
 	      continue;
 	    }
+
 	  if (!best || best->u.krt.metric > e->u.krt.metric)
 	    {
 	      best = e;
 	      pbest = ee;
 	    }
+
 	  e->u.krt.seen = 0;
+	  e->u.krt.best = 0;
 	  ee = &e->next;
 	}
       if (!n->routes)
 	{
 	  DBG("%I/%d: deleting\n", n->n.prefix, n->n.pxlen);
 	  if (old_best)
-	    {
 	    krt_learn_announce_delete(p, n);
-	      n->n.flags &= ~KRF_INSTALLED;
+
-	    }
 	  FIB_ITERATE_PUT(&fit);
 	  fib_delete(fib, n);
 	  goto again;
 	}
+
+      best->u.krt.best = 1;
       *pbest = best->next;
       best->next = n->routes;
       n->routes = best;
-      if (best != old_best || !(n->n.flags & KRF_INSTALLED) || p->reload)
+
+      if ((best != old_best) || p->reload)
 	{
 	  DBG("%I/%d: announcing (metric=%d)\n", n->n.prefix, n->n.pxlen, best->u.krt.metric);
 	  krt_learn_announce_update(p, best);
-	  n->n.flags |= KRF_INSTALLED;
 	}
       else
 	DBG("%I/%d: uptodate (metric=%d)\n", n->n.prefix, n->n.pxlen, best->u.krt.metric);
@@ -510,31 +522,31 @@ krt_learn_async(struct krt_proto *p, rte *e, int new)
   best = n->routes;
   bestp = &n->routes;
   for(gg=&n->routes; g=*gg; gg=&g->next)
+  {
     if (best->u.krt.metric > g->u.krt.metric)
       {
 	best = g;
 	bestp = gg;
       }
+
+    g->u.krt.best = 0;
+  }
+
   if (best)
     {
+      best->u.krt.best = 1;
       *bestp = best->next;
       best->next = n->routes;
       n->routes = best;
     }
+
   if (best != old_best)
     {
       DBG("krt_learn_async: distributing change\n");
       if (best)
-	{
 	krt_learn_announce_update(p, best);
-	  n->n.flags |= KRF_INSTALLED;
-	}
       else
-	{
-	  n->routes = NULL;
 	krt_learn_announce_delete(p, n);
-	  n->n.flags &= ~KRF_INSTALLED;
-	}
     }
 }
 
@@ -559,7 +571,7 @@ krt_dump(struct proto *P)
 static void
 krt_dump_attrs(rte *e)
 {
-  debug(" [m=%d,p=%d,t=%d]", e->u.krt.metric, e->u.krt.proto, e->u.krt.type);
+  debug(" [m=%d,p=%d]", e->u.krt.metric, e->u.krt.proto);
 }
 
 #endif

sysdep/unix/log.c

@@ -284,17 +284,18 @@ log_switch(int debug, list *l, char *new_syslog_name)
   current_log_list = l;
 
 #ifdef HAVE_SYSLOG
-  if (current_syslog_name && new_syslog_name &&
+  char *old_syslog_name = current_syslog_name;
-      !strcmp(current_syslog_name, new_syslog_name))
+  current_syslog_name = new_syslog_name;
+
+  if (old_syslog_name && new_syslog_name &&
+      !strcmp(old_syslog_name, new_syslog_name))
     return;
 
-  if (current_syslog_name)
+  if (old_syslog_name)
     closelog();
 
   if (new_syslog_name)
     openlog(new_syslog_name, LOG_CONS | LOG_NDELAY, LOG_DAEMON);
-
-  current_syslog_name = new_syslog_name;
 #endif
 }
 

sysdep/unix/main.c

@@ -450,6 +450,7 @@ cli_connect(sock *s, int size UNUSED)
   s->err_hook = cli_err;
   s->data = c = cli_new(s);
   s->pool = c->pool;		/* We need to have all the socket buffers allocated in the cli pool */
+  s->fast_rx = 1;
   c->rx_pos = c->rx_buf;
   c->rx_aux = NULL;
   rmove(s, c->pool);
@@ -466,6 +467,7 @@ cli_init_unix(uid_t use_uid, gid_t use_gid)
   s->type = SK_UNIX_PASSIVE;
   s->rx_hook = cli_connect;
   s->rbsize = 1024;
+  s->fast_rx = 1;
 
   /* Return value intentionally ignored */
   unlink(path_control_socket);