Sockets: Unified API for main and other loops

Now sk_open() requires an explicit IO loop to open the socket in. Also
specific functions for socket RX pause / resume are added to allow for
BGP corking.

And last but not least, socket reloop is now synchronous to resolve
weird cases of the target loop stopping before actually picking up the
relooped socket. Now the caller must ensure that both loops are locked
while relooping, and this way all sockets always have their respective
loop.

20 files changed, 236 insertions, 220 deletions

diff --git a/lib/io-loop.h b/lib/io-loop.h
index ae58bbee..502d77fc 100644
--- a/lib/io-loop.h
+++ b/lib/io-loop.h
@@ -16,10 +16,6 @@
 
 extern struct birdloop main_birdloop;
 
-void sk_start(sock *s);
-void sk_stop(sock *s);
-void sk_reloop(sock *s, struct birdloop *loop);
-
 /* Start a new birdloop owned by given pool and domain */
 struct birdloop *birdloop_new(pool *p, uint order, const char *name);
 
@@ -58,6 +54,10 @@ struct birdloop_flag_handler {
 void birdloop_flag(struct birdloop *loop, u32 flag);
 void birdloop_flag_set_handler(struct birdloop *, struct birdloop_flag_handler *);
 
+/* Setup sockets */
+void birdloop_add_socket(struct birdloop *, struct birdsock *);
+void birdloop_remove_socket(struct birdloop *, struct birdsock *);
+
 void birdloop_init(void);
 
 /* Yield for a little while. Use only in special cases. */
diff --git a/lib/socket.h b/lib/socket.h
index 5c69482e..4b169581 100644
--- a/lib/socket.h
+++ b/lib/socket.h
@@ -80,17 +80,22 @@ typedef struct birdsock {
   const char *password;			/* Password for MD5 authentication */
   const char *err;			/* Error message */
   struct ssh_sock *ssh;			/* Used in SK_SSH */
-  struct event reloop;			/* Reloop event */
+  struct birdloop *loop;		/* BIRDLoop owning this socket */
 } sock;
 
 sock *sock_new(pool *);			/* Allocate new socket */
 #define sk_new(X) sock_new(X)		/* Wrapper to avoid name collision with OpenSSL */
 
-int sk_open(sock *);			/* Open socket */
+int sk_open(sock *, struct birdloop *);		/* Open socket */
+void sk_reloop(sock *, struct birdloop *);	/* Move socket to another loop. Both loops must be locked. */
+
 int sk_rx_ready(sock *s);
+_Bool sk_tx_pending(sock *s);
 int sk_send(sock *, uint len);		/* Send data, <0=err, >0=ok, 0=sleep */
 int sk_send_to(sock *, uint len, ip_addr to, uint port); /* sk_send to given destination */
 void sk_reallocate(sock *);		/* Free and allocate tbuf & rbuf */
+void sk_pause_rx(struct birdloop *loop, sock *s);
+void sk_resume_rx(struct birdloop *loop, sock *s, int (*hook)(sock *, uint));
 void sk_set_rbsize(sock *s, uint val);	/* Resize RX buffer */
 void sk_set_tbsize(sock *s, uint val);	/* Resize TX buffer, keeping content */
 void sk_set_tbuf(sock *s, void *tbuf);	/* Switch TX buffer, NULL-> return to internal */
@@ -114,6 +119,7 @@ int sk_set_icmp6_filter(sock *s, int p1, int p2);
 void sk_log_error(sock *s, const char *p);
 
 byte * sk_rx_buffer(sock *s, int *len);	/* Temporary */
+sock *sk_next(sock *s);
 
 extern int sk_priority_control;		/* Suggested priority for control traffic, should be sysdep define */
 
@@ -127,11 +133,9 @@ extern int sk_priority_control;		/* Suggested priority for control traffic, shou
 #define SKF_HIGH_PORT	0x20	/* Choose port from high range if possible */
 #define SKF_FREEBIND	0x40	/* Allow socket to bind to a nonlocal address */
 
-#define SKF_THREAD	0x100	/* Socked used in thread, Do not add to main loop */
 #define SKF_TRUNCATED	0x200	/* Received packet was truncated, set by IO layer */
 #define SKF_HDRINCL	0x400	/* Used internally */
 #define SKF_PKTINFO	0x800	/* Used internally */
-#define SKF_PASSIVE_THREAD  0x1000  /* Child sockets used in thread, do not add to main loop */
 
 /*
  *	Socket types		     SA SP DA DP IF  TTL SendTo	(?=may, -=must not, *=must)
diff --git a/proto/babel/packets.c b/proto/babel/packets.c
index d4acc170..f13bb5ba 100644
--- a/proto/babel/packets.c
+++ b/proto/babel/packets.c
@@ -1617,7 +1617,7 @@ babel_open_socket(struct babel_iface *ifa)
   sk->ttl = 1;
   sk->flags = SKF_LADDR_RX;
 
-  if (sk_open(sk) < 0)
+  if (sk_open(sk, p->p.loop) < 0)
     goto err;
 
   if (sk_setup_multicast(sk) < 0)
diff --git a/proto/bfd/bfd.c b/proto/bfd/bfd.c
index 1ef92b18..b6ccdb9a 100644
--- a/proto/bfd/bfd.c
+++ b/proto/bfd/bfd.c
@@ -603,16 +603,10 @@ bfd_free_iface(struct bfd_iface *ifa)
     return;
 
   if (ifa->sk)
-  {
-    sk_stop(ifa->sk);
     rfree(ifa->sk);
-  }
 
   if (ifa->rx)
-  {
-    sk_stop(ifa->rx);
     rfree(ifa->rx);
-  }
 
   rem_node(&ifa->n);
   mb_free(ifa);
@@ -1100,11 +1094,6 @@ bfd_shutdown(struct proto *P)
 
   bfd_drop_requests(p);
 
-  if (p->rx4_1) sk_stop(p->rx4_1);
-  if (p->rx4_m) sk_stop(p->rx4_m);
-  if (p->rx6_1) sk_stop(p->rx6_1);
-  if (p->rx6_m) sk_stop(p->rx6_m);
-
   return PS_DOWN;
 }
 
diff --git a/proto/bfd/packets.c b/proto/bfd/packets.c
index 2200ab09..a22f223b 100644
--- a/proto/bfd/packets.c
+++ b/proto/bfd/packets.c
@@ -430,12 +430,11 @@ bfd_open_rx_sk(struct bfd_proto *p, int multihop, int af)
   /* TODO: configurable ToS and priority */
   sk->tos = IP_PREC_INTERNET_CONTROL;
   sk->priority = sk_priority_control;
-  sk->flags = SKF_THREAD | SKF_LADDR_RX | (!multihop ? SKF_TTL_RX : 0);
+  sk->flags = SKF_LADDR_RX | (!multihop ? SKF_TTL_RX : 0);
 
-  if (sk_open(sk) < 0)
+  if (sk_open(sk, p->p.loop) < 0)
     goto err;
 
-  sk_start(sk);
   return sk;
 
  err:
@@ -462,12 +461,11 @@ bfd_open_rx_sk_bound(struct bfd_proto *p, ip_addr local, struct iface *ifa)
   /* TODO: configurable ToS and priority */
   sk->tos = IP_PREC_INTERNET_CONTROL;
   sk->priority = sk_priority_control;
-  sk->flags = SKF_THREAD | SKF_BIND | (ifa ? SKF_TTL_RX : 0);
+  sk->flags = SKF_BIND | (ifa ? SKF_TTL_RX : 0);
 
-  if (sk_open(sk) < 0)
+  if (sk_open(sk, p->p.loop) < 0)
     goto err;
 
-  sk_start(sk);
   return sk;
 
  err:
@@ -494,12 +492,11 @@ bfd_open_tx_sk(struct bfd_proto *p, ip_addr local, struct iface *ifa)
   sk->tos = IP_PREC_INTERNET_CONTROL;
   sk->priority = sk_priority_control;
   sk->ttl = ifa ? 255 : -1;
-  sk->flags = SKF_THREAD | SKF_BIND | SKF_HIGH_PORT;
+  sk->flags = SKF_BIND | SKF_HIGH_PORT;
 
-  if (sk_open(sk) < 0)
+  if (sk_open(sk, p->p.loop) < 0)
     goto err;
 
-  sk_start(sk);
   return sk;
 
  err:
diff --git a/proto/bgp/bgp.c b/proto/bgp/bgp.c
index 122b0c22..a9028353 100644
--- a/proto/bgp/bgp.c
+++ b/proto/bgp/bgp.c
@@ -263,7 +263,7 @@ bgp_listen_create(void *_ UNUSED)
       sk->rx_hook = bgp_incoming_connection;
       sk->err_hook = bgp_listen_sock_err;
 
-      if (sk_open(sk) < 0)
+      if (sk_open(sk, &main_birdloop) < 0)
       {
 	sk_log_error(sk, p->p.name);
 	log(L_ERR "%s: Cannot open listening socket", p->p.name);
@@ -1203,7 +1203,7 @@ bgp_connect(struct bgp_proto *p)	/* Enter Connect state and start establishing c
   bgp_setup_sk(conn, s);
   bgp_conn_set_state(conn, BS_CONNECT);
 
-  if (sk_open(s) < 0)
+  if (sk_open(s, p->p.loop) < 0)
     goto err;
 
   /* Set minimal receive TTL if needed */
diff --git a/proto/bgp/packets.c b/proto/bgp/packets.c
index 924d6828..c90798a0 100644
--- a/proto/bgp/packets.c
+++ b/proto/bgp/packets.c
@@ -3015,7 +3015,7 @@ bgp_kick_tx(void *vconn)
     ;
 
   if (!max && !ev_active(conn->tx_ev))
-    ev_schedule(conn->tx_ev);
+    proto_send_event(&conn->bgp->p, conn->tx_ev);
 }
 
 void
@@ -3023,13 +3023,14 @@ bgp_tx(sock *sk)
 {
   struct bgp_conn *conn = sk->data;
 
+  ASSERT_DIE(birdloop_inside(conn->bgp->p.loop));
   DBG("BGP: TX hook\n");
   uint max = 1024;
   while (--max && (bgp_fire_tx(conn) > 0))
     ;
 
   if (!max && !ev_active(conn->tx_ev))
-    ev_schedule(conn->tx_ev);
+    proto_send_event(&conn->bgp->p, conn->tx_ev);
 }
 
 
diff --git a/proto/ospf/iface.c b/proto/ospf/iface.c
index 1919bccb..0aa7fa00 100644
--- a/proto/ospf/iface.c
+++ b/proto/ospf/iface.c
@@ -136,7 +136,7 @@ ospf_sk_open(struct ospf_iface *ifa)
   sk->flags = SKF_LADDR_RX | (ifa->check_ttl ? SKF_TTL_RX : 0);
   sk->ttl = ifa->cf->ttl_security ? 255 : 1;
 
-  if (sk_open(sk) < 0)
+  if (sk_open(sk, p->p.loop) < 0)
     goto err;
 
   /* 12 is an offset of the checksum in an OSPFv3 packet */
@@ -220,7 +220,7 @@ ospf_open_vlink_sk(struct ospf_proto *p)
   sk->data = (void *) p;
   sk->flags = 0;
 
-  if (sk_open(sk) < 0)
+  if (sk_open(sk, p->p.loop) < 0)
     goto err;
 
   /* 12 is an offset of the checksum in an OSPFv3 packet */
diff --git a/proto/radv/packets.c b/proto/radv/packets.c
index 5cd8b2de..c6b565d2 100644
--- a/proto/radv/packets.c
+++ b/proto/radv/packets.c
@@ -493,7 +493,7 @@ radv_sk_open(struct radv_iface *ifa)
   sk->data = ifa;
   sk->flags = SKF_LADDR_RX;
 
-  if (sk_open(sk) < 0)
+  if (sk_open(sk, ifa->ra->p.loop) < 0)
     goto err;
 
   /* We want listen just to ICMPv6 messages of type RS and RA */
diff --git a/proto/rip/packets.c b/proto/rip/packets.c
index 9c3bd7a3..fecdf896 100644
--- a/proto/rip/packets.c
+++ b/proto/rip/packets.c
@@ -1012,7 +1012,7 @@ rip_open_socket(struct rip_iface *ifa)
 
   /* sk->rbsize and sk->tbsize are handled in rip_iface_update_buffers() */
 
-  if (sk_open(sk) < 0)
+  if (sk_open(sk, p->p.loop) < 0)
     goto err;
 
   if (ifa->cf->mode == RIP_IM_MULTICAST)
diff --git a/proto/rpki/ssh_transport.c b/proto/rpki/ssh_transport.c
index 223afa80..425ad460 100644
--- a/proto/rpki/ssh_transport.c
+++ b/proto/rpki/ssh_transport.c
@@ -35,11 +35,9 @@ rpki_tr_ssh_open(struct rpki_tr_sock *tr)
   sk->ssh->subsystem = "rpki-rtr";
   sk->ssh->state = SK_SSH_CONNECT;
 
-  if (sk_open(sk) != 0)
+  if (sk_open(sk, cache->p->p.loop) != 0)
     return RPKI_TR_ERROR;
 
-  sk_start(sk);
-
   return RPKI_TR_SUCCESS;
 }
 
diff --git a/proto/rpki/tcp_transport.c b/proto/rpki/tcp_transport.c
index 4e850c44..ebb8030f 100644
--- a/proto/rpki/tcp_transport.c
+++ b/proto/rpki/tcp_transport.c
@@ -28,11 +28,9 @@ rpki_tr_tcp_open(struct rpki_tr_sock *tr)
 
   sk->type = SK_TCP_ACTIVE;
 
-  if (sk_open(sk) != 0)
+  if (sk_open(sk, tr->cache->p->p.loop) != 0)
     return RPKI_TR_ERROR;
 
-  sk_start(sk);
-
   return RPKI_TR_SUCCESS;
 }
 
diff --git a/proto/rpki/transport.c b/proto/rpki/transport.c
index 4026fca4..81bd6dd8 100644
--- a/proto/rpki/transport.c
+++ b/proto/rpki/transport.c
@@ -85,7 +85,6 @@ rpki_tr_open(struct rpki_tr_sock *tr)
   sk->rbsize = RPKI_RX_BUFFER_SIZE;
   sk->tbsize = RPKI_TX_BUFFER_SIZE;
   sk->tos = IP_PREC_INTERNET_CONTROL;
-  sk->flags |= SKF_THREAD;
   sk->vrf = cache->p->p.vrf;
 
   if (ipa_zero(sk->daddr) && sk->host)
@@ -121,7 +120,6 @@ rpki_tr_close(struct rpki_tr_sock *tr)
 
   if (tr->sk)
   {
-    sk_stop(tr->sk);
     rfree(tr->sk);
     tr->sk = NULL;
   }
diff --git a/sysdep/bsd/krt-sock.c b/sysdep/bsd/krt-sock.c
index e7b21474..094268b7 100644
--- a/sysdep/bsd/krt-sock.c
+++ b/sysdep/bsd/krt-sock.c
@@ -1088,7 +1088,7 @@ krt_sock_open(pool *pool, void *data, int table_id UNUSED)
   sk->fd = fd;
   sk->data = data;
 
-  if (sk_open(sk) < 0)
+  if (sk_open(sk, &main_birdloop) < 0)
     bug("krt-sock: sk_open failed");
 
   return sk;
diff --git a/sysdep/linux/netlink.c b/sysdep/linux/netlink.c
index b18cb899..e8a86ce4 100644
--- a/sysdep/linux/netlink.c
+++ b/sysdep/linux/netlink.c
@@ -2043,7 +2043,7 @@ nl_open_async(void)
   sk->rx_hook = nl_async_hook;
   sk->err_hook = nl_async_err_hook;
   sk->fd = fd;
-  if (sk_open(sk) < 0)
+  if (sk_open(sk, &main_birdloop) < 0)
     bug("Netlink: sk_open failed");
 }
 
diff --git a/sysdep/unix/io-loop.c b/sysdep/unix/io-loop.c
index df162c6e..0e2396f7 100644
--- a/sysdep/unix/io-loop.c
+++ b/sysdep/unix/io-loop.c
@@ -310,59 +310,95 @@ sockets_init(struct birdloop *loop)
   loop->sock_num = 0;
 }
 
-static void
-sockets_add(struct birdloop *loop, sock *s)
+void
+socket_changed(sock *s)
+{
+  struct birdloop *loop = s->loop;
+  ASSERT_DIE(birdloop_inside(loop));
+
+  loop->sock_changed++;
+  birdloop_ping(loop);
+}
+
+void
+birdloop_add_socket(struct birdloop *loop, sock *s)
 {
+  ASSERT_DIE(birdloop_inside(loop));
+  ASSERT_DIE(!s->loop);
+
   LOOP_TRACE(loop, "adding socket %p (total=%d)", s, loop->sock_num);
   add_tail(&loop->sock_list, &s->n);
   loop->sock_num++;
 
+  s->loop = loop;
   s->index = -1;
-  if (loop->thread)
-    atomic_store_explicit(&loop->thread->poll_changed, 1, memory_order_release);
 
-  birdloop_ping(loop);
+  socket_changed(s);
 }
 
+extern sock *stored_sock; /* mainloop hack */
+
 void
-sk_start(sock *s)
+birdloop_remove_socket(struct birdloop *loop, sock *s)
 {
-  ASSERT_DIE(birdloop_current != &main_birdloop);
-  sockets_add(birdloop_current, s);
-}
+  ASSERT_DIE(!enlisted(&s->n) == !s->loop);
 
-static void
-sockets_remove(struct birdloop *loop, sock *s)
-{
-  if (!enlisted(&s->n))
+  if (!s->loop)
     return;
 
+  ASSERT_DIE(birdloop_inside(loop));
+  ASSERT_DIE(s->loop == loop);
+
   /* Decouple the socket from the loop at all. */
   LOOP_TRACE(loop, "removing socket %p (total=%d)", s, loop->sock_num);
 
+  if (loop->sock_active == s)
+    loop->sock_active = sk_next(s);
+
+  if ((loop == &main_birdloop) && (s == stored_sock))
+    stored_sock = sk_next(s);
+
   rem_node(&s->n);
   loop->sock_num--;
-  if (loop->thread)
-    atomic_store_explicit(&loop->thread->poll_changed, 1, memory_order_release);
 
+  socket_changed(s);
+
+  s->loop = NULL;
   s->index = -1;
+}
 
-  /* Close the filedescriptor. If it ever gets into the poll(), it just returns
-   * POLLNVAL for this fd which then is ignored because nobody checks for
-   * that result. Or some other routine opens another fd, getting this number,
-   * yet also in this case poll() at worst spuriously returns and nobody checks
-   * for the result in this fd. No further precaution is needed. */
-  close(s->fd);
+void
+sk_reloop(sock *s, struct birdloop *loop)
+{
+  ASSERT_DIE(birdloop_inside(loop));
+  ASSERT_DIE(birdloop_inside(s->loop));
+
+  if (loop == s->loop)
+    return;
+
+  birdloop_remove_socket(s->loop, s);
+  birdloop_add_socket(loop, s);
+}
+
+void
+sk_pause_rx(struct birdloop *loop, sock *s)
+{
+  ASSERT_DIE(birdloop_inside(loop));
+  s->rx_hook = NULL;
+  socket_changed(s);
 }
 
 void
-sk_stop(sock *s)
+sk_resume_rx(struct birdloop *loop, sock *s, int (*hook)(sock *, uint))
 {
-  sockets_remove(birdloop_current, s);
+  ASSERT_DIE(birdloop_inside(loop));
+  ASSERT_DIE(hook);
+  s->rx_hook = hook;
+  socket_changed(s);
 }
 
 static inline uint sk_want_events(sock *s)
-{ return (s->rx_hook ? POLLIN : 0) | ((s->ttx != s->tpos) ? POLLOUT : 0); }
+{ return (s->rx_hook ? POLLIN : 0) | (sk_tx_pending(s) ? POLLOUT : 0); }
 
 void
 sockets_prepare(struct birdloop *loop, struct pfd *pfd)
@@ -392,40 +428,60 @@ sockets_prepare(struct birdloop *loop, struct pfd *pfd)
 
 int sk_read(sock *s, int revents);
 int sk_write(sock *s);
+void sk_err(sock *s, int revents);
 
-static void
+static int
 sockets_fire(struct birdloop *loop)
 {
+  if (EMPTY_LIST(loop->sock_list))
+    return 0;
+
+  int sch = 0;
+
   times_update();
 
   struct pollfd *pfd = loop->thread->pfd->pfd.data;
-  sock *s; node *n, *nxt;
-  WALK_LIST2_DELSAFE(s, n, nxt, loop->sock_list, n)
+  loop->sock_active = SKIP_BACK(sock, n, HEAD(loop->sock_list));
+
+  while (loop->sock_active)
   {
-    if (s->index < 0)
-      continue;
+    sock *s = loop->sock_active;
 
-    int rev = pfd[s->index].revents;
+    int rev;
+    if ((s->index >= 0) && (rev = pfd[s->index].revents) && !(rev & POLLNVAL))
+    {
+      int e = 1;
 
-    if (!rev)
-      continue;
+      if (rev & POLLOUT)
+      {
+	while ((s == loop->sock_active) && (e = sk_write(s)))
+	  ;
 
-    if (rev & POLLNVAL)
-      bug("poll: invalid fd %d", s->fd);
+	if (s != loop->sock_active)
+	  continue;
 
-    int e = 1;
+	if (!sk_tx_pending(s))
+	  sch++;
+      }
 
-    if (rev & POLLIN)
-      while (e && s->rx_hook)
-	e = sk_read(s, rev);
+      if (rev & POLLIN)
+	while (e && (s == loop->sock_active) && s->rx_hook)
+	  e = sk_read(s, rev);
 
-    if (rev & POLLOUT)
-    {
-      atomic_store_explicit(&loop->thread->poll_changed, 1, memory_order_release);
-      while (e = sk_write(s))
-	;
+      if (s != loop->sock_active)
+	continue;
+
+      if (!(rev & (POLLOUT | POLLIN)) && (rev & POLLERR))
+	sk_err(s, rev);
+
+      if (s != loop->sock_active)
+	continue;
     }
+
+    loop->sock_active = sk_next(s);
   }
+
+  return sch;
 }
 
 /*
@@ -547,7 +603,8 @@ bird_thread_main(void *arg)
   thr->meta->thread = thr;
   birdloop_enter(thr->meta);
 
-  u32 refresh_sockets = 1;
+  thr->sock_changed = 1;
+
   struct pfd pfd;
   BUFFER_INIT(pfd.pfd, thr->pool, 16);
   BUFFER_INIT(pfd.loop, thr->pool, 16);
@@ -563,7 +620,7 @@ bird_thread_main(void *arg)
     {
       birdloop_enter(loop);
       if (!EMPTY_LIST(loop->sock_list))
-	refresh_sockets = 1;
+	thr->sock_changed = 1;
       birdloop_leave(loop);
     }
 
@@ -590,10 +647,10 @@ bird_thread_main(void *arg)
     ev_run_list(&thr->priority_events);
 
     /* Do we have to refresh sockets? */
-    refresh_sockets += atomic_exchange_explicit(&thr->poll_changed, 0, memory_order_acq_rel);
-
-    if (refresh_sockets)
+    if (thr->sock_changed)
     {
+      thr->sock_changed = 0;
+
       BUFFER_FLUSH(pfd.pfd);
       BUFFER_FLUSH(pfd.loop);
 
@@ -608,7 +665,6 @@ bird_thread_main(void *arg)
       }
 
       ASSERT_DIE(pfd.loop.used == pfd.pfd.used);
-      refresh_sockets = 0;
     }
     /* Nothing to do in at least 5 seconds, flush local hot page cache */
     else if (timeout > 5000)
@@ -957,6 +1013,15 @@ birdloop_init(void)
 static void
 birdloop_stop_internal(struct birdloop *loop)
 {
+  LOOP_TRACE(loop, "Stopping");
+
+  /* Block incoming pings */
+  u32 ltt = atomic_load_explicit(&loop->thread_transition, memory_order_acquire);
+  while (!atomic_compare_exchange_strong_explicit(
+	&loop->thread_transition, &ltt, LTT_PING,
+	memory_order_acq_rel, memory_order_acquire))
+    ;
+
   /* Flush remaining events */
   ASSERT_DIE(!ev_run_list(&loop->event_list));
 
@@ -965,17 +1030,27 @@ birdloop_stop_internal(struct birdloop *loop)
   while (t = timers_first(&loop->time))
     tm_stop(t);
 
-  /* No sockets allowed */
-  ASSERT_DIE(EMPTY_LIST(loop->sock_list));
+  /* Drop sockets */
+  sock *s;
+  WALK_LIST_FIRST2(s, n, loop->sock_list)
+    birdloop_remove_socket(loop, s);
 
   /* Unschedule from Meta */
   ev_postpone(&loop->event);
   tm_stop(&loop->timer);
 
-  /* Declare loop stopped */
+  /* Remove from thread loop list */
   rem_node(&loop->n);
+  loop->thread = NULL;
+
+  /* Leave the loop context without causing any other fuss */
+  ASSERT_DIE(!ev_active(&loop->event));
+  loop->ping_pending = 0;
   birdloop_leave(loop);
 
+  /* Request local socket reload */
+  this_thread->sock_changed++;
+
   /* Tail-call the stopped hook */
   loop->stopped(loop->stop_data);
 }
@@ -989,12 +1064,14 @@ birdloop_run(void *_loop)
   struct birdloop *loop = _loop;
   birdloop_enter(loop);
 
+  LOOP_TRACE(loop, "Regular run");
+
   if (loop->stopped)
     /* Birdloop left inside the helper function */
     return birdloop_stop_internal(loop);
 
   /* Process sockets */
-  sockets_fire(loop);
+  this_thread->sock_changed += sockets_fire(loop);
 
   /* Run timers */
   timers_fire(&loop->time, 0);
@@ -1016,6 +1093,10 @@ birdloop_run(void *_loop)
   else
     tm_stop(&loop->timer);
 
+  /* Collect socket change requests */
+  this_thread->sock_changed += loop->sock_changed;
+  loop->sock_changed = 0;
+
   birdloop_leave(loop);
 }
 
@@ -1074,6 +1155,8 @@ birdloop_new(pool *pp, uint order, const char *name)
 static void
 birdloop_do_stop(struct birdloop *loop, void (*stopped)(void *data), void *data)
 {
+  LOOP_TRACE(loop, "Stop requested");
+
   loop->stopped = stopped;
   loop->stop_data = data;
 
@@ -1100,8 +1183,7 @@ birdloop_stop_self(struct birdloop *loop, void (*stopped)(void *data), void *dat
 void
 birdloop_free(struct birdloop *loop)
 {
-  ASSERT_DIE(loop->links == 0);
-  ASSERT_DIE(birdloop_in_this_thread(loop));
+  ASSERT_DIE(loop->thread == NULL);
 
   domain_free(loop->time.domain);
   rfree(loop->pool);
@@ -1171,20 +1253,6 @@ birdloop_unmask_wakeups(struct birdloop *loop)
 }
 
 void
-birdloop_link(struct birdloop *loop)
-{
-  ASSERT_DIE(birdloop_inside(loop));
-  loop->links++;
-}
-
-void
-birdloop_unlink(struct birdloop *loop)
-{
-  ASSERT_DIE(birdloop_inside(loop));
-  loop->links--;
-}
-
-void
 birdloop_yield(void)
 {
   usleep(100);
diff --git a/sysdep/unix/io-loop.h b/sysdep/unix/io-loop.h
index e606f07e..7ec903af 100644
--- a/sysdep/unix/io-loop.h
+++ b/sysdep/unix/io-loop.h
@@ -22,6 +22,7 @@ struct pfd {
 };
 
 void sockets_prepare(struct birdloop *, struct pfd *);
+void socket_changed(struct birdsock *);
 
 void pipe_new(struct pipe *);
 void pipe_pollin(struct pipe *, struct pfd *);
@@ -40,12 +41,12 @@ struct birdloop
   struct timeloop time;
   event_list event_list;
   list sock_list;
+  struct birdsock *sock_active;
   int sock_num;
+  uint sock_changed;
 
   uint ping_pending;
 
-  uint links;
-
   _Atomic u32 thread_transition;
 #define LTT_PING  1
 #define LTT_MOVE  2
@@ -66,8 +67,6 @@ struct bird_thread
 {
   node n;
 
-  _Atomic u32 poll_changed;
-
   struct pipe wakeup;
   event_list priority_events;
 
@@ -83,6 +82,8 @@ struct bird_thread
   struct pfd *pfd;
 
   event cleanup_event;
+
+  int sock_changed;
 };
 
 #endif
diff --git a/sysdep/unix/io.c b/sysdep/unix/io.c
index 06797096..88d187a4 100644
--- a/sysdep/unix/io.c
+++ b/sysdep/unix/io.c
@@ -722,10 +722,7 @@ sk_log_error(sock *s, const char *p)
  *	Actual struct birdsock code
  */
 
-static struct birdsock *current_sock;
-static struct birdsock *stored_sock;
-
-static inline sock *
+sock *
 sk_next(sock *s)
 {
   if (!s->n.next->next)
@@ -787,6 +784,7 @@ sk_ssh_free(sock *s)
 }
 #endif
 
+
 static void
 sk_free(resource *r)
 {
@@ -799,18 +797,10 @@ sk_free(resource *r)
     sk_ssh_free(s);
 #endif
 
-  if ((s->fd < 0) || (s->flags & SKF_THREAD))
-    return;
-
-  if (s == current_sock)
-    current_sock = sk_next(s);
-  if (s == stored_sock)
-    stored_sock = sk_next(s);
-
-  if (enlisted(&s->n))
-    rem_node(&s->n);
+  if (s->loop)
+    birdloop_remove_socket(s->loop, s);
 
-  if (s->type != SK_SSH && s->type != SK_SSH_ACTIVE)
+  if (s->fd >= 0 && s->type != SK_SSH && s->type != SK_SSH_ACTIVE)
     close(s->fd);
 
   s->fd = -1;
@@ -1023,12 +1013,6 @@ sk_setup(sock *s)
 }
 
 static void
-sk_insert(sock *s)
-{
-  add_tail(&main_birdloop.sock_list, &s->n);
-}
-
-static void
 sk_tcp_connected(sock *s)
 {
   sockaddr sa;
@@ -1101,10 +1085,7 @@ sk_passive_connected(sock *s, int type)
     return 1;
   }
 
-  if (s->flags & SKF_PASSIVE_THREAD)
-    t->flags |= SKF_THREAD;
-  else
-    sk_insert(t);
+  birdloop_add_socket(s->loop, t);
 
   sk_alloc_bufs(t);
   s->rx_hook(t, 0);
@@ -1319,6 +1300,7 @@ sk_open_ssh(sock *s)
 
 /**
  * sk_open - open a socket
+ * @loop: loop
  * @s: socket
  *
  * This function takes a socket resource created by sk_new() and
@@ -1328,7 +1310,7 @@ sk_open_ssh(sock *s)
  * Result: 0 for success, -1 for an error.
  */
 int
-sk_open(sock *s)
+sk_open(sock *s, struct birdloop *loop)
 {
   int af = AF_UNSPEC;
   int fd = -1;
@@ -1481,9 +1463,7 @@ sk_open(sock *s)
     sk_alloc_bufs(s);
   }
 
-  if (!(s->flags & SKF_THREAD))
-    sk_insert(s);
-
+  birdloop_add_socket(loop, s);
   return 0;
 
 err:
@@ -1493,7 +1473,7 @@ err:
 }
 
 int
-sk_open_unix(sock *s, char *name)
+sk_open_unix(sock *s, struct birdloop *loop, char *name)
 {
   struct sockaddr_un sa;
   int fd;
@@ -1520,40 +1500,10 @@ sk_open_unix(sock *s, char *name)
     return -1;
 
   s->fd = fd;
-  sk_insert(s);
+  birdloop_add_socket(loop, s);
   return 0;
 }
 
-static void
-sk_reloop_hook(void *_vs)
-{
-  sock *s = _vs;
-  if (birdloop_inside(&main_birdloop))
-  {
-    s->flags &= ~SKF_THREAD;
-    sk_insert(s);
-  }
-  else
-  {
-    s->flags |= SKF_THREAD;
-    sk_start(s);
-  }
-}
-
-void
-sk_reloop(sock *s, struct birdloop *loop)
-{
-  if (enlisted(&s->n))
-    rem_node(&s->n);
-
-  s->reloop = (event) {
-    .hook = sk_reloop_hook,
-    .data = s,
-  };
-
-  ev_send_loop(loop, &s->reloop);
-}
-
 
 #define CMSG_RX_SPACE MAX(CMSG4_SPACE_PKTINFO+CMSG4_SPACE_TTL, \
 			  CMSG6_SPACE_PKTINFO+CMSG6_SPACE_TTL)
@@ -1676,6 +1626,13 @@ sk_recvmsg(sock *s)
 
 static inline void reset_tx_buffer(sock *s) { s->ttx = s->tpos = s->tbuf; }
 
+_Bool
+sk_tx_pending(sock *s)
+{
+  return s->ttx != s->tpos;
+}
+
+
 static int
 sk_maybe_write(sock *s)
 {
@@ -1686,7 +1643,7 @@ sk_maybe_write(sock *s)
   case SK_TCP:
   case SK_MAGIC:
   case SK_UNIX:
-    while (s->ttx != s->tpos)
+    while (sk_tx_pending(s))
     {
       e = write(s->fd, s->ttx, s->tpos - s->ttx);
 
@@ -1708,7 +1665,7 @@ sk_maybe_write(sock *s)
 
 #ifdef HAVE_LIBSSH
   case SK_SSH:
-    while (s->ttx != s->tpos)
+    while (sk_tx_pending(s))
     {
       e = ssh_channel_write(s->ssh->channel, s->ttx, s->tpos - s->ttx);
 
@@ -1791,7 +1748,12 @@ sk_send(sock *s, unsigned len)
 {
   s->ttx = s->tbuf;
   s->tpos = s->tbuf + len;
-  return sk_maybe_write(s);
+
+  int e = sk_maybe_write(s);
+  if (e == 0) /* Trigger thread poll reload to poll this socket's write. */
+    socket_changed(s);
+
+  return e;
 }
 
 /**
@@ -1838,7 +1800,7 @@ call_rx_hook(sock *s, int size)
   if (s->rx_hook(s, size))
   {
     /* We need to be careful since the socket could have been deleted by the hook */
-    if (current_sock == s)
+    if (s->loop->sock_active == s)
       s->rpos = s->rbuf;
   }
 }
@@ -2002,7 +1964,7 @@ sk_write_noflush(sock *s)
 #endif
 
   default:
-    if (s->ttx != s->tpos && sk_maybe_write(s) > 0)
+    if (sk_tx_pending(s) && sk_maybe_write(s) > 0)
     {
       if (s->tx_hook)
 	s->tx_hook(s);
@@ -2224,6 +2186,8 @@ static int short_loops = 0;
 #define SHORT_LOOP_MAX 10
 #define WORK_EVENTS_MAX 10
 
+sock *stored_sock;
+
 void
 io_loop(void)
 {
@@ -2312,17 +2276,13 @@ io_loop(void)
 	  times_update();
 
 	  /* guaranteed to be non-empty */
-	  current_sock = SKIP_BACK(sock, n, HEAD(main_birdloop.sock_list));
+	  main_birdloop.sock_active = SKIP_BACK(sock, n, HEAD(main_birdloop.sock_list));
 
-	  while (current_sock)
+	  while (main_birdloop.sock_active)
+	  {
+	    sock *s = main_birdloop.sock_active;
+	    if (s->index != -1)
 	    {
-	      sock *s = current_sock;
-	      if (s->index == -1)
-		{
-		  current_sock = sk_next(s);
-		  goto next;
-		}
-
 	      int e;
 	      int steps;
 
@@ -2333,10 +2293,11 @@ io_loop(void)
 		    steps--;
 		    io_log_event(s->rx_hook, s->data);
 		    e = sk_read(s, pfd.pfd.data[s->index].revents);
-		    if (s != current_sock)
-		      goto next;
 		  }
-		while (e && s->rx_hook && steps);
+		while (e && (main_birdloop.sock_active == s) && s->rx_hook && steps);
+
+	      if (s != main_birdloop.sock_active)
+		continue;
 
 	      steps = MAX_STEPS;
 	      if (pfd.pfd.data[s->index].revents & POLLOUT)
@@ -2345,56 +2306,54 @@ io_loop(void)
 		    steps--;
 		    io_log_event(s->tx_hook, s->data);
 		    e = sk_write(s);
-		    if (s != current_sock)
-		      goto next;
 		  }
-		while (e && steps);
+		while (e && (main_birdloop.sock_active == s) && steps);
 
-	      current_sock = sk_next(s);
-	    next: ;
+	      if (s != main_birdloop.sock_active)
+		continue;
 	    }
 
+	    main_birdloop.sock_active = sk_next(s);
+	  }
+
 	  short_loops++;
 	  if (events && (short_loops < SHORT_LOOP_MAX))
 	    continue;
 	  short_loops = 0;
 
 	  int count = 0;
-	  current_sock = stored_sock;
-	  if (current_sock == NULL)
-	    current_sock = SKIP_BACK(sock, n, HEAD(main_birdloop.sock_list));
+	  main_birdloop.sock_active = stored_sock;
+	  if (main_birdloop.sock_active == NULL)
+	    main_birdloop.sock_active = SKIP_BACK(sock, n, HEAD(main_birdloop.sock_list));
 
-	  while (current_sock && count < MAX_RX_STEPS)
+	  while (main_birdloop.sock_active && count < MAX_RX_STEPS)
 	    {
-	      sock *s = current_sock;
+	      sock *s = main_birdloop.sock_active;
 	      if (s->index == -1)
-		{
-		  current_sock = sk_next(s);
-		  goto next2;
-		}
+		goto next2;
 
 	      if (!s->fast_rx && (pfd.pfd.data[s->index].revents & POLLIN) && s->rx_hook)
 		{
 		  count++;
 		  io_log_event(s->rx_hook, s->data);
 		  sk_read(s, pfd.pfd.data[s->index].revents);
-		  if (s != current_sock)
-		    goto next2;
+		  if (s != main_birdloop.sock_active)
+		    continue;
 		}
 
 	      if (pfd.pfd.data[s->index].revents & (POLLHUP | POLLERR))
 		{
 		  sk_err(s, pfd.pfd.data[s->index].revents);
-		  if (s != current_sock)
-		    goto next2;
+		  if (s != main_birdloop.sock_active)
+		    continue;
 		}
 
-	      current_sock = sk_next(s);
 	    next2: ;
+	      main_birdloop.sock_active = sk_next(s);
 	    }
 
 
-	  stored_sock = current_sock;
+	  stored_sock = main_birdloop.sock_active;
 	}
     }
 }
diff --git a/sysdep/unix/main.c b/sysdep/unix/main.c
index a9ae842a..79db32d3 100644
--- a/sysdep/unix/main.c
+++ b/sysdep/unix/main.c
@@ -542,7 +542,7 @@ cli_init_unix(uid_t use_uid, gid_t use_gid)
   /* Return value intentionally ignored */
   unlink(path_control_socket);
 
-  if (sk_open_unix(s, path_control_socket) < 0)
+  if (sk_open_unix(s, &main_birdloop, path_control_socket) < 0)
     die("Cannot create control socket %s: %m", path_control_socket);
 
   if (use_uid || use_gid)
diff --git a/sysdep/unix/unix.h b/sysdep/unix/unix.h
index e26afe41..606b79cd 100644
--- a/sysdep/unix/unix.h
+++ b/sysdep/unix/unix.h
@@ -9,6 +9,9 @@
 #ifndef _BIRD_UNIX_H_
 #define _BIRD_UNIX_H_
 
+#include "nest/bird.h"
+#include "lib/io-loop.h"
+
 #include <sys/socket.h>
 #include <signal.h>
 
@@ -110,7 +113,7 @@ extern volatile sig_atomic_t async_shutdown_flag;
 void io_init(void);
 void io_loop(void);
 void io_log_dump(void);
-int sk_open_unix(struct birdsock *s, char *name);
+int sk_open_unix(struct birdsock *s, struct birdloop *, char *name);
 struct rfile *rf_open(struct pool *, const char *name, const char *mode);
 void *rf_file(struct rfile *f);
 int rf_fileno(struct rfile *f);