bird/proto/bfd/bfd.c

/*
 *	BIRD -- Bidirectional Forwarding Detection (BFD)
 *
 *	Can be freely distributed and used under the terms of the GNU GPL.
 */

/**
 * DOC: Bidirectional Forwarding Detection
 *
 * The BFD protocol is implemented in three files: |bfd.c| containing the
 * protocol logic and the protocol glue with BIRD core, |packets.c| handling BFD
 * packet processing, RX, TX and protocol sockets. |io.c| then contains generic
 * code for the event loop, threads and event sources (sockets, microsecond
 * timers). This generic code will be merged to the main BIRD I/O code in the
 * future.
 *
 * The BFD implementation uses a separate thread with an internal event loop for
 * handling the protocol logic, which requires high-res and low-latency timing,
 * so it is not affected by the rest of BIRD, which has several low-granularity
 * hooks in the main loop, uses second-based timers and cannot offer good
 * latency. The core of BFD protocol (the code related to BFD sessions,
 * interfaces and packets) runs in the BFD thread, while the rest (the code
 * related to BFD requests, BFD neighbors and the protocol glue) runs in the
 * main thread.
 *
 * BFD sessions are represented by structure &bfd_session that contains a state
 * related to the session and two timers (TX timer for periodic packets and hold
 * timer for session timeout). These sessions are allocated from @session_slab
 * and are accessible by two hash tables, @session_hash_id (by session ID) and
 * @session_hash_ip (by IP addresses of neighbors and associated interfaces).
 * Slab and both hashes are in the main protocol structure &bfd_proto. The
 * protocol logic related to BFD sessions is implemented in internal functions
 * bfd_session_*(), which are expected to be called from the context of BFD
 * thread, and external functions bfd_add_session(), bfd_remove_session() and
 * bfd_reconfigure_session(), which form an interface to the BFD core for the
 * rest and are expected to be called from the context of main thread.
 *
 * Each BFD session has an associated BFD interface, represented by structure
 * &bfd_iface. A BFD interface contains a socket used for TX (the one for RX is
 * shared in &bfd_proto), an interface configuration and reference counter.
 * Compared to interface structures of other protocols, these structures are not
 * created and removed based on interface notification events, but according to
 * the needs of BFD sessions. When a new session is created, it requests a
 * proper BFD interface by function bfd_get_iface(), which either finds an
 * existing one in &iface_list (from &bfd_proto) or allocates a new one. When a
 * session is removed, an associated iface is discharged by bfd_free_iface().
 *
 * BFD requests are the external API for the other protocols. When a protocol
 * wants a BFD session, it calls bfd_request_session(), which creates a
 * structure &bfd_request containing approprite information and an notify hook.
 * This structure is a resource associated with the caller's resource pool. When
 * a BFD protocol is available, a BFD request is submitted to the protocol, an
 * appropriate BFD session is found or created and the request is attached to
 * the session. When a session changes state, all attached requests (and related
 * protocols) are notified. Note that BFD requests do not depend on BFD protocol
 * running. When the BFD protocol is stopped or removed (or not available from
 * beginning), related BFD requests are stored in @bfd_wait_list, where waits
 * for a new protocol.
 *
 * BFD neighbors are just a way to statically configure BFD sessions without
 * requests from other protocol. Structures &bfd_neighbor are part of BFD
 * configuration (like static routes in the static protocol). BFD neighbors are
 * handled by BFD protocol like it is a BFD client -- when a BFD neighbor is
 * ready, the protocol just creates a BFD request like any other protocol.
 *
 * The protocol uses a new generic event loop (structure &birdloop) from |io.c|,
 * which supports sockets, timers and events like the main loop. A birdloop is
 * associated with a thread (field @thread) in which event hooks are executed.
 * Most functions for setting event sources (like sk_start() or tm_start()) must
 * be called from the context of that thread. Birdloop allows to temporarily
 * acquire the context of that thread for the main thread by calling
 * birdloop_enter() and then birdloop_leave(), which also ensures mutual
 * exclusion with all event hooks. Note that resources associated with a
 * birdloop (like timers) should be attached to the independent resource pool,
 * detached from the main resource tree.
 *
 * There are two kinds of interaction between the BFD core (running in the BFD
 * thread) and the rest of BFD (running in the main thread). The first kind are
 * configuration calls from main thread to the BFD thread (like bfd_add_session()).
 * These calls are synchronous and use birdloop_enter() mechanism for mutual
 * exclusion. The second kind is a notification about session changes from the
 * BFD thread to the main thread. This is done in an asynchronous way, sesions
 * with pending notifications are linked (in the BFD thread) to @notify_list in
 * &bfd_proto, and then bfd_notify_hook() in the main thread is activated using
 * bfd_notify_kick() and a pipe. The hook then processes scheduled sessions and
 * calls hooks from associated BFD requests. This @notify_list (and state fields
 * in structure &bfd_session) is protected by a spinlock in &bfd_proto and
 * functions bfd_lock_sessions() / bfd_unlock_sessions().
 *
 * There are few data races (accessing @p->p.debug from TRACE() from the BFD
 * thread and accessing some some private fields of %bfd_session from
 * bfd_show_sessions() from the main thread, but these are harmless (i hope).
 *
 * TODO: document functions and access restrictions for fields in BFD structures.
 *
 * Supported standards:
 * - RFC 5880 - main BFD standard
 * - RFC 5881 - BFD for IP links
 * - RFC 5882 - generic application of BFD
 * - RFC 5883 - BFD for multihop paths
 */

#include "bfd.h"


#define HASH_ID_KEY(n)		n->loc_id
#define HASH_ID_NEXT(n)		n->next_id
#define HASH_ID_EQ(a,b)		a == b
#define HASH_ID_FN(k)		k

#define HASH_IP_KEY(n)		n->addr, n->ifindex
#define HASH_IP_NEXT(n)		n->next_ip
#define HASH_IP_EQ(a1,n1,a2,n2)	ipa_equal(a1, a2) && n1 == n2
#define HASH_IP_FN(a,n)		ipa_hash(a) ^ u32_hash(n)

DEFINE_DOMAIN(rtable);
#define BFD_LOCK	LOCK_DOMAIN(rtable, bfd_global.lock)
#define BFD_UNLOCK	UNLOCK_DOMAIN(rtable, bfd_global.lock)
#define BFD_ASSERT_LOCKED	ASSERT_DIE(DOMAIN_IS_LOCKED(rtable, bfd_global.lock))

static struct {
  DOMAIN(rtable) lock;
  list wait_list;
  list proto_list;
} bfd_global;

static struct bfd_session bfd_admin_down = { .loc = { .state = BFD_STATE_ADMIN_DOWN }, };

const char *bfd_state_names[] = { "AdminDown", "Down", "Init", "Up" };

static void bfd_session_set_min_tx(struct bfd_session *s, u32 val);
static struct bfd_iface *bfd_get_iface(struct bfd_proto *p, ip_addr local, struct iface *iface);
static void bfd_free_iface(struct bfd_iface *ifa);
static void bfd_remove_session(struct bfd_proto *p, struct bfd_session *s);
static void bfd_reconfigure_session_hook(void *vsession);


/*
 *	BFD sessions
 */

static inline struct bfd_session_config
bfd_merge_options(const struct bfd_iface_config *cf, const struct bfd_options *opts)
{
  return (struct bfd_session_config) {
    .min_rx_int = opts->min_rx_int ?: cf->min_rx_int,
    .min_tx_int = opts->min_tx_int ?: cf->min_tx_int,
    .idle_tx_int = opts->idle_tx_int ?: cf->idle_tx_int,
    .multiplier = opts->multiplier ?: cf->multiplier,
    .passive = opts->passive_set ? opts->passive : cf->passive
  };
}

static int
bfd_session_update_state(struct bfd_session *s, uint state, uint diag)
{
  struct bfd_proto *p = s->ifa->bfd;
  uint old_state = BFD_LOC_STATE(s).state;

  if (state == old_state)
  {
    if (current_time() > s->last_reqlist_check + 5 S)
    {
      BFD_LOCK;
      if (EMPTY_LIST(s->request_list))
      {
	bfd_remove_session(p, s);
	BFD_UNLOCK;
	return 1;
      }

      s->last_reqlist_check = current_time();
      BFD_UNLOCK;
    }
    return 0;
  }

  TRACE(D_EVENTS, "Session to %I changed state from %s to %s",
	s->addr, bfd_state_names[old_state], bfd_state_names[state]);

  atomic_store_explicit(&s->loc, ((struct bfd_session_state) { .state = state, .diag = diag }), memory_order_release);
  s->last_state_change = current_time();

  if (state == BFD_STATE_UP)
    bfd_session_set_min_tx(s, s->cf.min_tx_int);

  if (old_state == BFD_STATE_UP)
    bfd_session_set_min_tx(s, s->cf.idle_tx_int);

  BFD_LOCK;
  if (EMPTY_LIST(s->request_list))
  {
    bfd_remove_session(p, s);
    BFD_UNLOCK;
    return 1;
  }

  struct bfd_request *req;
  node *nn;
  WALK_LIST2(req, nn, s->request_list, n)
    ev_send_self(&req->event);

  BFD_UNLOCK;
  return 0;
}

static void
bfd_session_update_tx_interval(struct bfd_session *s)
{
  u32 tx_int = MAX(s->des_min_tx_int, s->rem_min_rx_int);
  u32 tx_int_l = tx_int - (tx_int / 4);	 // 75 %
  u32 tx_int_h = tx_int - (tx_int / 10); // 90 %

  s->tx_timer->recurrent = tx_int_l;
  s->tx_timer->randomize = tx_int_h - tx_int_l;

  /* Do not set timer if no previous event */
  if (!s->last_tx)
    return;

  /* Set timer relative to last tx_timer event */
  tm_set_in(s->tx_timer, s->last_tx + tx_int_l, s->ifa->bfd->p.loop);
}

static void
bfd_session_update_detection_time(struct bfd_session *s, int kick)
{
  btime timeout = (btime) MAX(s->req_min_rx_int, s->rem_min_tx_int) * s->rem_detect_mult;

  if (kick)
    s->last_rx = current_time();

  if (!s->last_rx)
    return;

  tm_set_in(s->hold_timer, s->last_rx + timeout, s->ifa->bfd->p.loop);
}

static void
bfd_session_control_tx_timer(struct bfd_session *s, int reset)
{
  // if (!s->opened) goto stop;

  if (s->passive && (s->rem_id == 0))
    goto stop;

  if (s->rem_demand_mode &&
      !s->poll_active &&
      (BFD_LOC_STATE(s).state == BFD_STATE_UP) &&
      (s->rem.state == BFD_STATE_UP))
    goto stop;

  if (s->rem_min_rx_int == 0)
    goto stop;

  /* So TX timer should run */
  if (reset || !tm_active(s->tx_timer))
  {
    s->last_tx = 0;
    tm_start_in(s->tx_timer, 0, s->ifa->bfd->p.loop);
  }

  return;

 stop:
  tm_stop(s->tx_timer);
  s->last_tx = 0;
}

static void
bfd_session_request_poll(struct bfd_session *s, u8 request)
{
  /* Not sure about this, but doing poll in this case does not make sense */
  if (s->rem_id == 0)
    return;

  s->poll_scheduled |= request;

  if (s->poll_active)
    return;

  s->poll_active = s->poll_scheduled;
  s->poll_scheduled = 0;

  bfd_session_control_tx_timer(s, 1);
}

static void
bfd_session_terminate_poll(struct bfd_session *s)
{
  u8 poll_done = s->poll_active & ~s->poll_scheduled;

  if (poll_done & BFD_POLL_TX)
    s->des_min_tx_int = s->des_min_tx_new;

  if (poll_done & BFD_POLL_RX)
    s->req_min_rx_int = s->req_min_rx_new;

  s->poll_active = s->poll_scheduled;
  s->poll_scheduled = 0;

  /* Timers are updated by caller - bfd_session_process_ctl() */
}

void
bfd_session_process_ctl(struct bfd_session *s, u8 flags, u32 old_tx_int, u32 old_rx_int)
{
  if (s->poll_active && (flags & BFD_FLAG_FINAL))
    bfd_session_terminate_poll(s);

  if ((s->des_min_tx_int != old_tx_int) || (s->rem_min_rx_int != old_rx_int))
    bfd_session_update_tx_interval(s);

  bfd_session_update_detection_time(s, 1);

  /* Update session state */
  int next_state = 0;
  int diag = BFD_DIAG_NOTHING;

  switch (BFD_LOC_STATE(s).state)
  {
  case BFD_STATE_ADMIN_DOWN:
    return;

  case BFD_STATE_DOWN:
    if (s->rem.state == BFD_STATE_DOWN)		next_state = BFD_STATE_INIT;
    else if (s->rem.state == BFD_STATE_INIT)	next_state = BFD_STATE_UP;
    break;

  case BFD_STATE_INIT:
    if (s->rem.state == BFD_STATE_ADMIN_DOWN)	next_state = BFD_STATE_DOWN, diag = BFD_DIAG_NEIGHBOR_DOWN;
    else if (s->rem.state >= BFD_STATE_INIT)	next_state = BFD_STATE_UP;
    break;

  case BFD_STATE_UP:
    if (s->rem.state <= BFD_STATE_DOWN)		next_state = BFD_STATE_DOWN, diag = BFD_DIAG_NEIGHBOR_DOWN;
    break;
  }

  if (next_state)
    if (bfd_session_update_state(s, next_state, diag))
      return;

  bfd_session_control_tx_timer(s, 0);

  if (flags & BFD_FLAG_POLL)
    bfd_send_ctl(s->ifa->bfd, s, 1);
}

static void
bfd_session_timeout(struct bfd_session *s)
{
  struct bfd_proto *p = s->ifa->bfd;

  TRACE(D_EVENTS, "Session to %I expired", s->addr);

  s->rem.state = BFD_STATE_DOWN;
  s->rem_id = 0;
  s->rem_min_tx_int = 0;
  s->rem_min_rx_int = 1;
  s->rem_demand_mode = 0;
  s->rem_detect_mult = 0;
  s->rx_csn_known = 0;

  s->poll_active = 0;
  s->poll_scheduled = 0;

  if (bfd_session_update_state(s, BFD_STATE_DOWN, BFD_DIAG_TIMEOUT))
    return;

  bfd_session_control_tx_timer(s, 1);
}

static void
bfd_session_set_min_tx(struct bfd_session *s, u32 val)
{
  /* Note that des_min_tx_int <= des_min_tx_new */

  if (val == s->des_min_tx_new)
    return;

  s->des_min_tx_new = val;

  /* Postpone timer update if des_min_tx_int increases and the session is up */
  if ((BFD_LOC_STATE(s).state != BFD_STATE_UP) || (val < s->des_min_tx_int))
  {
    s->des_min_tx_int = val;
    bfd_session_update_tx_interval(s);
  }

  bfd_session_request_poll(s, BFD_POLL_TX);
}

static void
bfd_session_set_min_rx(struct bfd_session *s, u32 val)
{
  /* Note that req_min_rx_int >= req_min_rx_new */

  if (val == s->req_min_rx_new)
    return;

  s->req_min_rx_new = val;

  /* Postpone timer update if req_min_rx_int decreases and the session is up */
  if ((BFD_LOC_STATE(s).state != BFD_STATE_UP) || (val > s->req_min_rx_int))
  {
    s->req_min_rx_int = val;
    bfd_session_update_detection_time(s, 0);
  }

  bfd_session_request_poll(s, BFD_POLL_RX);
}

struct bfd_session *
bfd_find_session_by_id(struct bfd_proto *p, u32 id)
{
  ASSERT_DIE(birdloop_inside(p->p.loop));
  return HASH_FIND(p->session_hash_id, HASH_ID, id);
}

struct bfd_session *
bfd_find_session_by_addr(struct bfd_proto *p, ip_addr addr, uint ifindex)
{
  ASSERT_DIE(birdloop_inside(p->p.loop));
  return HASH_FIND(p->session_hash_ip, HASH_IP, addr, ifindex);
}

static void
bfd_tx_timer_hook(timer *t)
{
  struct bfd_session *s = t->data;

  s->last_tx = current_time();
  bfd_send_ctl(s->ifa->bfd, s, 0);
}

static void
bfd_hold_timer_hook(timer *t)
{
  bfd_session_timeout(t->data);
}

static u32
bfd_get_free_id(struct bfd_proto *p)
{
  u32 id;
  for (id = random_u32(); 1; id++)
    if (id && !bfd_find_session_by_id(p, id))
      break;

  return id;
}

static struct bfd_session *
bfd_add_session(struct bfd_proto *p, ip_addr addr, ip_addr local, struct iface *iface, struct bfd_options *opts)
{
  ASSERT_DIE(birdloop_inside(p->p.loop));
  BFD_ASSERT_LOCKED;

  struct bfd_iface *ifa = bfd_get_iface(p, local, iface);

  struct bfd_session *s = sl_allocz(p->session_slab);
  s->addr = addr;
  s->ifa = ifa;
  s->ifindex = iface ? iface->index : 0;
  s->loc_id = bfd_get_free_id(p);

  HASH_INSERT(p->session_hash_id, HASH_ID, s);
  HASH_INSERT(p->session_hash_ip, HASH_IP, s);

  s->cf = bfd_merge_options(ifa->cf, opts);
  s->update_event = (event) {
    .hook = bfd_reconfigure_session_hook,
    .data = s,
    .list = birdloop_event_list(p->p.loop),
  };

  /* Initialization of state variables - see RFC 5880 6.8.1 */
  atomic_store_explicit(&s->loc, ((struct bfd_session_state) { .state = BFD_STATE_DOWN }), memory_order_relaxed);
  s->rem.state = BFD_STATE_DOWN;
  s->des_min_tx_int = s->des_min_tx_new = s->cf.idle_tx_int;
  s->req_min_rx_int = s->req_min_rx_new = s->cf.min_rx_int;
  s->rem_min_rx_int = 1;
  s->detect_mult = s->cf.multiplier;
  s->passive = s->cf.passive;
  s->tx_csn = random_u32();

  s->tx_timer = tm_new_init(p->p.pool, bfd_tx_timer_hook, s, 0, 0);
  s->hold_timer = tm_new_init(p->p.pool, bfd_hold_timer_hook, s, 0, 0);
  bfd_session_update_tx_interval(s);
  bfd_session_control_tx_timer(s, 1);

  init_list(&s->request_list);
  s->last_state_change = current_time();

  TRACE(D_EVENTS, "Session to %I added", s->addr);

  return s;
}

static void
bfd_remove_session(struct bfd_proto *p, struct bfd_session *s)
{
  ASSERT_DIE(birdloop_inside(p->p.loop));
  BFD_ASSERT_LOCKED;
  ASSERT_DIE(EMPTY_LIST(s->request_list));

  bfd_free_iface(s->ifa);

  rfree(s->tx_timer);
  rfree(s->hold_timer);

  HASH_REMOVE(p->session_hash_id, HASH_ID, s);
  HASH_REMOVE(p->session_hash_ip, HASH_IP, s);

  TRACE(D_EVENTS, "Session to %I removed", s->addr);

  sl_free(p->session_slab, s);
}

static void
bfd_reconfigure_session(struct bfd_proto *p, struct bfd_session *s)
{
  ASSERT_DIE(birdloop_inside(p->p.loop));
  BFD_LOCK;
  if (EMPTY_LIST(s->request_list))
  {
    bfd_remove_session(p, s);
    BFD_UNLOCK;
    return;
  }

  struct bfd_request *req = SKIP_BACK(struct bfd_request, n, HEAD(s->request_list));
  s->cf = bfd_merge_options(s->ifa->cf, &req->opts);

  u32 tx = (BFD_LOC_STATE(s).state == BFD_STATE_UP) ? s->cf.min_tx_int : s->cf.idle_tx_int;
  bfd_session_set_min_tx(s, tx);
  bfd_session_set_min_rx(s, s->cf.min_rx_int);
  s->detect_mult = s->cf.multiplier;
  s->passive = s->cf.passive;

  bfd_session_control_tx_timer(s, 0);

  TRACE(D_EVENTS, "Session to %I reconfigured", s->addr);
  BFD_UNLOCK;
}

static void
bfd_reconfigure_session_hook(void *data)
{
  struct bfd_session *s = data;
  return bfd_reconfigure_session(s->ifa->bfd, s);
}


/*
 *	BFD interfaces
 */

static struct bfd_iface_config bfd_default_iface = {
  .min_rx_int = BFD_DEFAULT_MIN_RX_INT,
  .min_tx_int = BFD_DEFAULT_MIN_TX_INT,
  .idle_tx_int = BFD_DEFAULT_IDLE_TX_INT,
  .multiplier = BFD_DEFAULT_MULTIPLIER
};

static inline struct bfd_iface_config *
bfd_find_iface_config(struct bfd_config *cf, struct iface *iface)
{
  struct bfd_iface_config *ic;

  ic = iface ? (void *) iface_patt_find(&cf->patt_list, iface, NULL) : cf->multihop;

  return ic ? ic : &bfd_default_iface;
}

static struct bfd_iface *
bfd_get_iface(struct bfd_proto *p, ip_addr local, struct iface *iface)
{
  struct bfd_iface *ifa;

  WALK_LIST(ifa, p->iface_list)
    if (ipa_equal(ifa->local, local) && (ifa->iface == iface))
      return ifa->uc++, ifa;

  struct bfd_config *cf = (struct bfd_config *) (p->p.cf);
  struct bfd_iface_config *ic = bfd_find_iface_config(cf, iface);

  ifa = mb_allocz(p->p.pool, sizeof(struct bfd_iface));
  ifa->local = local;
  ifa->iface = iface;
  ifa->cf = ic;
  ifa->bfd = p;

  ifa->sk = bfd_open_tx_sk(p, local, iface);
  ifa->uc = 1;

  add_tail(&p->iface_list, &ifa->n);

  return ifa;
}

static void
bfd_free_iface(struct bfd_iface *ifa)
{
  if (!ifa || --ifa->uc)
    return;

  if (ifa->sk)
  {
    sk_stop(ifa->sk);
    rfree(ifa->sk);
  }

  rem_node(&ifa->n);
  mb_free(ifa);
}

static void
bfd_reconfigure_iface(struct bfd_proto *p UNUSED, struct bfd_iface *ifa, struct bfd_config *nc)
{
  struct bfd_iface_config *new = bfd_find_iface_config(nc, ifa->iface);
  struct bfd_iface_config *old = ifa->cf;

  /* Check options that are handled in bfd_reconfigure_session() */
  ifa->changed =
    (new->min_rx_int != old->min_rx_int) ||
    (new->min_tx_int != old->min_tx_int) ||
    (new->idle_tx_int != old->idle_tx_int) ||
    (new->multiplier != old->multiplier) ||
    (new->passive != old->passive);

  /* This should be probably changed to not access ifa->cf from the BFD thread */
  ifa->cf = new;
}


/*
 *	BFD requests
 */

static void
bfd_request_notify(void *data)
{
  struct bfd_request *req = data;
  struct bfd_session_state old = req->old_state;

  BFD_LOCK; /* Needed to safely access req->session */
  struct bfd_session_state new = atomic_load_explicit(&req->session->loc, memory_order_acquire);
  BFD_UNLOCK;

  if (new.state == old.state)
    return;

  req->state = new.state;
  req->diag = new.diag;
  req->old_state = new;
  req->down = (old.state == BFD_STATE_UP) && (new.state == BFD_STATE_DOWN);

  if (req->hook)
    req->hook(req);
}

static int
bfd_add_request(struct bfd_proto *p, struct bfd_request *req)
{
  BFD_ASSERT_LOCKED;
  ASSERT_DIE(req->session == &bfd_admin_down);

  struct bfd_config *cf = (struct bfd_config *) (p->p.cf);

  if (p->p.vrf && (p->p.vrf != req->vrf))
    return 0;

  if (ipa_is_ip4(req->addr) ? !cf->accept_ipv4 : !cf->accept_ipv6)
    return 0;

  if (req->iface ? !cf->accept_direct : !cf->accept_multihop)
    return 0;

  uint ifindex = req->iface ? req->iface->index : 0;
  struct bfd_session *s = bfd_find_session_by_addr(p, req->addr, ifindex);

  if (!s)
    s = bfd_add_session(p, req->addr, req->local, req->iface, &req->opts);

  rem_node(&req->n);
  add_tail(&s->request_list, &req->n);
  req->session = s;

  ev_send_self(&req->event);

  return 1;
}

static void
bfd_pickup_requests(void *_data UNUSED)
{
  struct bfd_proto *p;
  node *nn;
  WALK_LIST2(p, nn, bfd_global.proto_list, bfd_node)
  {
    birdloop_enter(p->p.loop);
    BFD_LOCK;

    struct bfd_request *req;
    node *rn, *rnxt;
    WALK_LIST2_DELSAFE(req, rn, rnxt, bfd_global.wait_list, n)
      bfd_add_request(p, req);

    BFD_UNLOCK;
    birdloop_ping(p->p.loop);
    birdloop_leave(p->p.loop);
  }
}

static event bfd_pickup_event = { .hook = bfd_pickup_requests };
#define bfd_schedule_pickup()	ev_send(&global_event_list, &bfd_pickup_event)

static void
bfd_take_requests(struct bfd_proto *p)
{
  struct bfd_request *req;
  node *n, *nn;
  BFD_LOCK;
  WALK_LIST2_DELSAFE(req, n, nn, bfd_global.wait_list, n)
    bfd_add_request(p, req);
  BFD_UNLOCK;
}

static void
bfd_drop_requests(struct bfd_proto *p)
{
  node *n;
  BFD_LOCK;
  HASH_WALK_DELSAFE(p->session_hash_id, next_id, s)
  {
    WALK_LIST_FIRST(n, s->request_list)
    {
      struct bfd_request *req = SKIP_BACK(struct bfd_request, n, n);
      rem_node(&req->n);
      add_tail(&bfd_global.wait_list, &req->n);
      req->session = &bfd_admin_down;
      ev_send_self(&req->event);
    }

    bfd_schedule_pickup();
    bfd_remove_session(p, s);
  }
  HASH_WALK_END;
  BFD_UNLOCK;
}

static struct resclass bfd_request_class;

struct bfd_request *
bfd_request_session(pool *p, ip_addr addr, ip_addr local,
		    struct iface *iface, struct iface *vrf,
		    void (*hook)(struct bfd_request *), void *data,
		    struct event_list *list,
		    const struct bfd_options *opts)
{
  struct bfd_request *req = ralloc(p, &bfd_request_class);

  req->addr = addr;
  req->local = local;
  req->iface = iface;
  req->vrf = vrf;

  if (opts)
    req->opts = *opts;

  req->hook = hook;
  req->data = data;
  req->event = (event) {
    .hook = bfd_request_notify,
    .data = req,
    .list = list,
  };

  BFD_LOCK;
  req->session = &bfd_admin_down;
  add_tail(&bfd_global.wait_list, &req->n);
  bfd_schedule_pickup();
  BFD_UNLOCK;

  return req;
}

void
bfd_update_request(struct bfd_request *req, const struct bfd_options *opts)
{
  if (!memcmp(opts, &req->opts, sizeof(const struct bfd_options)))
    return;

  BFD_LOCK;
  req->opts = *opts;

  struct bfd_session *s = req->session;
  if (s != &bfd_admin_down)
    ev_send_self(&s->update_event);

  BFD_UNLOCK;
}

static void
bfd_request_free(resource *r)
{
  struct bfd_request *req = (struct bfd_request *) r;

  BFD_LOCK;
  rem_node(&req->n);
  BFD_UNLOCK;

  ev_postpone(&req->event);
}

static void
bfd_request_dump(resource *r)
{
  struct bfd_request *req = (struct bfd_request *) r;

  debug("(code %p, data %p)\n", req->hook, req->data);
}

static struct resclass bfd_request_class = {
  "BFD request",
  sizeof(struct bfd_request),
  bfd_request_free,
  bfd_request_dump,
  NULL,
  NULL
};


/*
 *	BFD neighbors
 */

static void
bfd_neigh_notify(struct neighbor *nb)
{
  struct bfd_proto *p = (struct bfd_proto *) nb->proto;
  struct bfd_neighbor *n = nb->data;

  if (!n)
    return;

  if ((nb->scope > 0) && !n->req)
  {
    ip_addr local = ipa_nonzero(n->local) ? n->local : nb->ifa->ip;
    n->req = bfd_request_session(p->p.pool, n->addr, local, nb->iface, p->p.vrf, NULL, NULL, birdloop_event_list(p->p.loop), NULL);
  }

  if ((nb->scope <= 0) && n->req)
  {
    rfree(n->req);
    n->req = NULL;
  }
}

static void
bfd_start_neighbor(struct bfd_proto *p, struct bfd_neighbor *n)
{
  n->active = 1;

  if (n->multihop)
  {
    n->req = bfd_request_session(p->p.pool, n->addr, n->local, NULL, p->p.vrf, NULL, NULL, birdloop_event_list(p->p.loop), NULL);
    return;
  }

  struct neighbor *nb = neigh_find(&p->p, n->addr, n->iface, NEF_STICKY);
  if (!nb)
  {
    log(L_ERR "%s: Invalid remote address %I%J", p->p.name, n->addr, n->iface);
    return;
  }

  if (nb->data)
  {
    log(L_ERR "%s: Duplicate neighbor %I", p->p.name, n->addr);
    return;
  }

  n->neigh = nb;
  nb->data = n;

  if (nb->scope > 0)
    bfd_neigh_notify(nb);
  else
    TRACE(D_EVENTS, "Waiting for %I%J to become my neighbor", n->addr, n->iface);
}

static void
bfd_stop_neighbor(struct bfd_proto *p UNUSED, struct bfd_neighbor *n)
{
  if (n->neigh)
    n->neigh->data = NULL;
  n->neigh = NULL;

  rfree(n->req);
  n->req = NULL;
}

static inline int
bfd_same_neighbor(struct bfd_neighbor *x, struct bfd_neighbor *y)
{
  return ipa_equal(x->addr, y->addr) && ipa_equal(x->local, y->local) &&
    (x->iface == y->iface) && (x->multihop == y->multihop);
}

static void
bfd_reconfigure_neighbors(struct bfd_proto *p, struct bfd_config *new)
{
  struct bfd_config *old = (struct bfd_config *) (p->p.cf);
  struct bfd_neighbor *on, *nn;

  WALK_LIST(on, old->neigh_list)
  {
    WALK_LIST(nn, new->neigh_list)
      if (bfd_same_neighbor(nn, on))
      {
	nn->neigh = on->neigh;
	if (nn->neigh)
	  nn->neigh->data = nn;

	nn->req = on->req;
	nn->active = 1;
	goto next;
      }

    bfd_stop_neighbor(p, on);
  next:;
  }

  WALK_LIST(nn, new->neigh_list)
    if (!nn->active)
      bfd_start_neighbor(p, nn);
}


/*
 *	BFD protocol glue
 */

void
bfd_init_all(void)
{
  bfd_global.lock = DOMAIN_NEW(rtable, "BFD Global");
  init_list(&bfd_global.wait_list);
  init_list(&bfd_global.proto_list);
}

static struct proto *
bfd_init(struct proto_config *c)
{
  struct proto *p = proto_new(c);

  p->neigh_notify = bfd_neigh_notify;

  return p;
}

static int
bfd_start(struct proto *P)
{
  struct bfd_proto *p = (struct bfd_proto *) P;
  struct bfd_config *cf = (struct bfd_config *) (P->cf);

  p->session_slab = sl_new(P->pool, sizeof(struct bfd_session));
  HASH_INIT(p->session_hash_id, P->pool, 8);
  HASH_INIT(p->session_hash_ip, P->pool, 8);

  init_list(&p->iface_list);

  add_tail(&bfd_global.proto_list, &p->bfd_node);

  if (cf->accept_ipv4 && cf->accept_direct)
    p->rx4_1 = bfd_open_rx_sk(p, 0, SK_IPV4);

  if (cf->accept_ipv4 && cf->accept_multihop)
    p->rx4_m = bfd_open_rx_sk(p, 1, SK_IPV4);

  if (cf->accept_ipv6 && cf->accept_direct)
    p->rx6_1 = bfd_open_rx_sk(p, 0, SK_IPV6);

  if (cf->accept_ipv6 && cf->accept_multihop)
    p->rx6_m = bfd_open_rx_sk(p, 1, SK_IPV6);

  bfd_take_requests(p);

  struct bfd_neighbor *n;
  WALK_LIST(n, cf->neigh_list)
    bfd_start_neighbor(p, n);

  return PS_UP;
}

static int
bfd_shutdown(struct proto *P)
{
  struct bfd_proto *p = (struct bfd_proto *) P;
  struct bfd_config *cf = (struct bfd_config *) (p->p.cf);

  rem_node(&p->bfd_node);

  struct bfd_neighbor *bn;
  WALK_LIST(bn, cf->neigh_list)
    bfd_stop_neighbor(p, bn);

  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;
}

static int
bfd_reconfigure(struct proto *P, struct proto_config *c)
{
  struct bfd_proto *p = (struct bfd_proto *) P;
  struct bfd_config *old = (struct bfd_config *) (P->cf);
  struct bfd_config *new = (struct bfd_config *) c;
  struct bfd_iface *ifa;

  ASSERT_DIE(birdloop_inside(P->loop));

  /* TODO: Improve accept reconfiguration */
  if ((new->accept_ipv4 != old->accept_ipv4) ||
      (new->accept_ipv6 != old->accept_ipv6) ||
      (new->accept_direct != old->accept_direct) ||
      (new->accept_multihop != old->accept_multihop))
    return 0;

  birdloop_mask_wakeups(p->p.loop);

  WALK_LIST(ifa, p->iface_list)
    bfd_reconfigure_iface(p, ifa, new);

  HASH_WALK_DELSAFE(p->session_hash_id, next_id, s)
  {
    if (s->ifa->changed)
      bfd_reconfigure_session(p, s);
  }
  HASH_WALK_DELSAFE_END;

  bfd_reconfigure_neighbors(p, new);

  birdloop_unmask_wakeups(p->p.loop);

  return 1;
}

static void
bfd_copy_config(struct proto_config *dest, struct proto_config *src UNUSED)
{
  struct bfd_config *d = (struct bfd_config *) dest;
  // struct bfd_config *s = (struct bfd_config *) src;

  /* We clean up patt_list and neigh_list, neighbors and ifaces are non-sharable */
  init_list(&d->patt_list);
  init_list(&d->neigh_list);
}

void
bfd_show_sessions(struct proto *P)
{
  byte tbuf[TM_DATETIME_BUFFER_SIZE];
  struct bfd_proto *p = (struct bfd_proto *) P;
  btime tx_int, timeout;
  const char *ifname;

  birdloop_enter(P->loop);
  if (p->p.proto_state != PS_UP)
  {
    cli_msg(-1020, "%s: is not up", p->p.name);
    birdloop_leave(P->loop);
    return;
  }

  cli_msg(-1020, "%s:", p->p.name);
  cli_msg(-1020, "%-25s %-10s %-10s %-12s  %8s %8s",
	  "IP address", "Interface", "State", "Since", "Interval", "Timeout");

  HASH_WALK(p->session_hash_id, next_id, s)
  {
    uint state = BFD_LOC_STATE(s).state;
    ifname = (s->ifa && s->ifa->iface) ? s->ifa->iface->name : "---";
    tx_int = s->last_tx ? MAX(s->des_min_tx_int, s->rem_min_rx_int) : 0;
    timeout = (btime) MAX(s->req_min_rx_int, s->rem_min_tx_int) * s->rem_detect_mult;

    state = (state < 4) ? state : 0;
    tm_format_time(tbuf, &config->tf_proto, s->last_state_change);

    cli_msg(-1020, "%-25I %-10s %-10s %-12s  %7t  %7t",
	    s->addr, ifname, bfd_state_names[state], tbuf, tx_int, timeout);
  }
  HASH_WALK_END;

  birdloop_leave(P->loop);
}


struct protocol proto_bfd = {
  .name =		"BFD",
  .template =		"bfd%d",
  .class =		PROTOCOL_BFD,
  .proto_size =		sizeof(struct bfd_proto),
  .config_size =	sizeof(struct bfd_config),
  .init =		bfd_init,
  .start =		bfd_start,
  .shutdown =		bfd_shutdown,
  .reconfigure =	bfd_reconfigure,
  .copy_config =	bfd_copy_config,
};