bgpd/bgp_keepalives.c

   1 /* BGP Keepalives.
   2  * Implements a producer thread to generate BGP keepalives for peers.
   3  * Copyright (C) 2017 Cumulus Networks, Inc.
   4  * Quentin Young
   5  *
   6  * This file is part of FRRouting.
   7  *
   8  * FRRouting is free software; you can redistribute it and/or modify it under
   9  * the terms of the GNU General Public License as published by the Free
  10  * Software Foundation; either version 2, or (at your option) any later
  11  * version.
  12  *
  13  * FRRouting is distributed in the hope that it will be useful, but WITHOUT ANY
  14  * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
  15  * FOR A PARTICULAR PURPOSE.  See the GNU General Public License for more
  16  * details.
  17  *
  18  * You should have received a copy of the GNU General Public License along
  19  * with this program; see the file COPYING; if not, write to the Free Software
  20  * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
  21  */
  22
  23 /* clang-format off */
  24 #include <zebra.h>
  25 #include <pthread.h>            // for pthread_mutex_lock, pthread_mutex_unlock
  26
  27 #include "frr_pthread.h"        // for frr_pthread
  28 #include "hash.h"               // for hash, hash_clean, hash_create_size...
  29 #include "log.h"                // for zlog_debug
  30 #include "memory.h"             // for MTYPE_TMP, XFREE, XCALLOC, XMALLOC
  31 #include "monotime.h"           // for monotime, monotime_since
  32
  33 #include "bgpd/bgpd.h"          // for peer, PEER_THREAD_KEEPALIVES_ON, peer...
  34 #include "bgpd/bgp_debug.h"     // for bgp_debug_neighbor_events
  35 #include "bgpd/bgp_packet.h"    // for bgp_keepalive_send
  36 #include "bgpd/bgp_keepalives.h"
  37 /* clang-format on */
  38
  39 /*
  40  * Peer KeepAlive Timer.
  41  * Associates a peer with the time of its last keepalive.
  42  */
  43 struct pkat {
  44         /* the peer to send keepalives to */
  45         struct peer *peer;
  46         /* absolute time of last keepalive sent */
  47         struct timeval last;
  48 };
  49
  50 /* List of peers we are sending keepalives for, and associated mutex. */
  51 static pthread_mutex_t *peerhash_mtx;
  52 static pthread_cond_t *peerhash_cond;
  53 static struct hash *peerhash;
  54
  55 static struct pkat *pkat_new(struct peer *peer)
  56 {
  57         struct pkat *pkat = XMALLOC(MTYPE_TMP, sizeof(struct pkat));
  58         pkat->peer = peer;
  59         monotime(&pkat->last);
  60         return pkat;
  61 }
  62
  63 static void pkat_del(void *pkat)
  64 {
  65         XFREE(MTYPE_TMP, pkat);
  66 }
  67
  68
  69 /*
  70  * Callback for hash_iterate. Determines if a peer needs a keepalive and if so,
  71  * generates and sends it.
  72  *
  73  * For any given peer, if the elapsed time since its last keepalive exceeds its
  74  * configured keepalive timer, a keepalive is sent to the peer and its
  75  * last-sent time is reset. Additionally, If the elapsed time does not exceed
  76  * the configured keepalive timer, but the time until the next keepalive is due
  77  * is within a hardcoded tolerance, a keepalive is sent as if the configured
  78  * timer was exceeded. Doing this helps alleviate nanosecond sleeps between
  79  * ticks by grouping together peers who are due for keepalives at roughly the
  80  * same time. This tolerance value is arbitrarily chosen to be 100ms.
  81  *
  82  * In addition, this function calculates the maximum amount of time that the
  83  * keepalive thread can sleep before another tick needs to take place. This is
  84  * equivalent to shortest time until a keepalive is due for any one peer.
  85  *
  86  * @return maximum time to wait until next update (0 if infinity)
  87  */
  88 static void peer_process(struct hash_backet *hb, void *arg)
  89 {
  90         struct pkat *pkat = hb->data;
  91
  92         struct timeval *next_update = arg;
  93
  94         static struct timeval elapsed;  // elapsed time since keepalive
  95         static struct timeval ka = {0}; // peer->v_keepalive as a timeval
  96         static struct timeval diff;     // ka - elapsed
  97
  98         static struct timeval tolerance = {0, 100000};
  99
 100         /* calculate elapsed time since last keepalive */
 101         monotime_since(&pkat->last, &elapsed);
 102
 103         /* calculate difference between elapsed time and configured time */
 104         ka.tv_sec = pkat->peer->v_keepalive;
 105         timersub(&ka, &elapsed, &diff);
 106
 107         int send_keepalive =
 108                 elapsed.tv_sec >= ka.tv_sec || timercmp(&diff, &tolerance, <);
 109
 110         if (send_keepalive) {
 111                 if (bgp_debug_neighbor_events(pkat->peer))
 112                         zlog_debug("%s [FSM] Timer (keepalive timer expire)",
 113                                    pkat->peer->host);
 114
 115                 bgp_keepalive_send(pkat->peer);
 116                 monotime(&pkat->last);
 117                 memset(&elapsed, 0x00, sizeof(struct timeval));
 118                 diff = ka;
 119         }
 120
 121         /* if calculated next update for this peer < current delay, use it */
 122         if (next_update->tv_sec < 0 || timercmp(&diff, next_update, <))
 123                 *next_update = diff;
 124 }
 125
 126 static int peer_hash_cmp(const void *f, const void *s)
 127 {
 128         const struct pkat *p1 = f;
 129         const struct pkat *p2 = s;
 130         return p1->peer == p2->peer;
 131 }
 132
 133 static unsigned int peer_hash_key(void *arg)
 134 {
 135         struct pkat *pkat = arg;
 136         return (uintptr_t)pkat->peer;
 137 }
 138
 139 /* Cleanup handler / deinitializer. */
 140 static void bgp_keepalives_finish(void *arg)
 141 {
 142         if (peerhash) {
 143                 hash_clean(peerhash, pkat_del);
 144                 hash_free(peerhash);
 145         }
 146
 147         peerhash = NULL;
 148
 149         pthread_mutex_unlock(peerhash_mtx);
 150         pthread_mutex_destroy(peerhash_mtx);
 151         pthread_cond_destroy(peerhash_cond);
 152
 153         XFREE(MTYPE_TMP, peerhash_mtx);
 154         XFREE(MTYPE_TMP, peerhash_cond);
 155 }
 156
 157 /*
 158  * Entry function for peer keepalive generation pthread.
 159  */
 160 void *bgp_keepalives_start(void *arg)
 161 {
 162         struct frr_pthread *fpt = arg;
 163         fpt->master->owner = pthread_self();
 164
 165         struct timeval currtime = {0, 0};
 166         struct timeval aftertime = {0, 0};
 167         struct timeval next_update = {0, 0};
 168         struct timespec next_update_ts = {0, 0};
 169
 170         peerhash_mtx = XCALLOC(MTYPE_TMP, sizeof(pthread_mutex_t));
 171         peerhash_cond = XCALLOC(MTYPE_TMP, sizeof(pthread_cond_t));
 172
 173         /* initialize mutex */
 174         pthread_mutex_init(peerhash_mtx, NULL);
 175
 176         /* use monotonic clock with condition variable */
 177         pthread_condattr_t attrs;
 178         pthread_condattr_init(&attrs);
 179         pthread_condattr_setclock(&attrs, CLOCK_MONOTONIC);
 180         pthread_cond_init(peerhash_cond, &attrs);
 181         pthread_condattr_destroy(&attrs);
 182
 183         frr_pthread_set_name(fpt, NULL, "bgpd_ka");
 184
 185         /* initialize peer hashtable */
 186         peerhash = hash_create_size(2048, peer_hash_key, peer_hash_cmp, NULL);
 187         pthread_mutex_lock(peerhash_mtx);
 188
 189         /* register cleanup handler */
 190         pthread_cleanup_push(&bgp_keepalives_finish, NULL);
 191
 192         /* notify anybody waiting on us that we are done starting up */
 193         frr_pthread_notify_running(fpt);
 194
 195         while (atomic_load_explicit(&fpt->running, memory_order_relaxed)) {
 196                 if (peerhash->count > 0)
 197                         pthread_cond_timedwait(peerhash_cond, peerhash_mtx,
 198                                                &next_update_ts);
 199                 else
 200                         while (peerhash->count == 0
 201                                && atomic_load_explicit(&fpt->running,
 202                                                        memory_order_relaxed))
 203                                 pthread_cond_wait(peerhash_cond, peerhash_mtx);
 204
 205                 monotime(&currtime);
 206
 207                 next_update.tv_sec = -1;
 208
 209                 hash_iterate(peerhash, peer_process, &next_update);
 210                 if (next_update.tv_sec == -1)
 211                         memset(&next_update, 0x00, sizeof(next_update));
 212
 213                 monotime_since(&currtime, &aftertime);
 214
 215                 timeradd(&currtime, &next_update, &next_update);
 216                 TIMEVAL_TO_TIMESPEC(&next_update, &next_update_ts);
 217         }
 218
 219         /* clean up */
 220         pthread_cleanup_pop(1);
 221
 222         return NULL;
 223 }
 224
 225 /* --- thread external functions ------------------------------------------- */
 226
 227 void bgp_keepalives_on(struct peer *peer)
 228 {
 229         if (CHECK_FLAG(peer->thread_flags, PEER_THREAD_KEEPALIVES_ON))
 230                 return;
 231
 232         struct frr_pthread *fpt = bgp_pth_ka;
 233         assert(fpt->running);
 234
 235         /* placeholder bucket data to use for fast key lookups */
 236         static struct pkat holder = {0};
 237
 238         /*
 239          * We need to ensure that bgp_keepalives_init was called first
 240          */
 241         assert(peerhash_mtx);
 242
 243         pthread_mutex_lock(peerhash_mtx);
 244         {
 245                 holder.peer = peer;
 246                 if (!hash_lookup(peerhash, &holder)) {
 247                         struct pkat *pkat = pkat_new(peer);
 248                         hash_get(peerhash, pkat, hash_alloc_intern);
 249                         peer_lock(peer);
 250                 }
 251                 SET_FLAG(peer->thread_flags, PEER_THREAD_KEEPALIVES_ON);
 252         }
 253         pthread_mutex_unlock(peerhash_mtx);
 254         bgp_keepalives_wake();
 255 }
 256
 257 void bgp_keepalives_off(struct peer *peer)
 258 {
 259         if (!CHECK_FLAG(peer->thread_flags, PEER_THREAD_KEEPALIVES_ON))
 260                 return;
 261
 262         struct frr_pthread *fpt = bgp_pth_ka;
 263         assert(fpt->running);
 264
 265         /* placeholder bucket data to use for fast key lookups */
 266         static struct pkat holder = {0};
 267
 268         /*
 269          * We need to ensure that bgp_keepalives_init was called first
 270          */
 271         assert(peerhash_mtx);
 272
 273         pthread_mutex_lock(peerhash_mtx);
 274         {
 275                 holder.peer = peer;
 276                 struct pkat *res = hash_release(peerhash, &holder);
 277                 if (res) {
 278                         pkat_del(res);
 279                         peer_unlock(peer);
 280                 }
 281                 UNSET_FLAG(peer->thread_flags, PEER_THREAD_KEEPALIVES_ON);
 282         }
 283         pthread_mutex_unlock(peerhash_mtx);
 284 }
 285
 286 void bgp_keepalives_wake()
 287 {
 288         pthread_mutex_lock(peerhash_mtx);
 289         {
 290                 pthread_cond_signal(peerhash_cond);
 291         }
 292         pthread_mutex_unlock(peerhash_mtx);
 293 }
 294
 295 int bgp_keepalives_stop(struct frr_pthread *fpt, void **result)
 296 {
 297         assert(fpt->running);
 298
 299         atomic_store_explicit(&fpt->running, false, memory_order_relaxed);
 300         bgp_keepalives_wake();
 301
 302         pthread_join(fpt->thread, result);
 303         return 0;
 304 }