]>
Commit | Line | Data |
---|---|---|
5c0c651c QY |
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 | |
03014d48 | 21 | */ |
5c0c651c | 22 | |
6ee8ea1c | 23 | /* clang-format off */ |
03014d48 | 24 | #include <zebra.h> |
6ee8ea1c | 25 | #include <pthread.h> // for pthread_mutex_lock, pthread_mutex_unlock |
03014d48 | 26 | |
6ee8ea1c QY |
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 | |
03014d48 | 32 | |
6ee8ea1c QY |
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 | |
03014d48 | 36 | #include "bgpd/bgp_keepalives.h" |
6ee8ea1c | 37 | /* clang-format on */ |
03014d48 | 38 | |
a715eab3 | 39 | /* |
03014d48 QY |
40 | * Peer KeepAlive Timer. |
41 | * Associates a peer with the time of its last keepalive. | |
42 | */ | |
43 | struct pkat { | |
a715eab3 | 44 | /* the peer to send keepalives to */ |
03014d48 | 45 | struct peer *peer; |
a715eab3 | 46 | /* absolute time of last keepalive sent */ |
03014d48 QY |
47 | struct timeval last; |
48 | }; | |
49 | ||
50 | /* List of peers we are sending keepalives for, and associated mutex. */ | |
bd8b71e4 QY |
51 | static pthread_mutex_t *peerhash_mtx; |
52 | static pthread_cond_t *peerhash_cond; | |
53 | static struct hash *peerhash; | |
03014d48 | 54 | |
03014d48 QY |
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 | } | |
03014d48 | 67 | |
bd8b71e4 | 68 | |
03014d48 | 69 | /* |
424ab01d QY |
70 | * Callback for hash_iterate. Determines if a peer needs a keepalive and if so, |
71 | * generates and sends it. | |
03014d48 QY |
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 | */ | |
e3b78da8 | 88 | static void peer_process(struct hash_bucket *hb, void *arg) |
03014d48 | 89 | { |
bd8b71e4 QY |
90 | struct pkat *pkat = hb->data; |
91 | ||
92 | struct timeval *next_update = arg; | |
03014d48 | 93 | |
03014d48 QY |
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 | ||
03014d48 QY |
98 | static struct timeval tolerance = {0, 100000}; |
99 | ||
a715eab3 | 100 | /* calculate elapsed time since last keepalive */ |
bd8b71e4 | 101 | monotime_since(&pkat->last, &elapsed); |
03014d48 | 102 | |
a715eab3 | 103 | /* calculate difference between elapsed time and configured time */ |
bd8b71e4 QY |
104 | ka.tv_sec = pkat->peer->v_keepalive; |
105 | timersub(&ka, &elapsed, &diff); | |
03014d48 | 106 | |
bd8b71e4 QY |
107 | int send_keepalive = |
108 | elapsed.tv_sec >= ka.tv_sec || timercmp(&diff, &tolerance, <); | |
03014d48 | 109 | |
bd8b71e4 QY |
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); | |
03014d48 | 114 | |
bd8b71e4 QY |
115 | bgp_keepalive_send(pkat->peer); |
116 | monotime(&pkat->last); | |
117 | memset(&elapsed, 0x00, sizeof(struct timeval)); | |
a715eab3 | 118 | diff = ka; |
03014d48 QY |
119 | } |
120 | ||
a715eab3 | 121 | /* if calculated next update for this peer < current delay, use it */ |
2ccf91b1 | 122 | if (next_update->tv_sec < 0 || timercmp(&diff, next_update, <)) |
bd8b71e4 QY |
123 | *next_update = diff; |
124 | } | |
125 | ||
74df8d6d | 126 | static bool peer_hash_cmp(const void *f, const void *s) |
bd8b71e4 QY |
127 | { |
128 | const struct pkat *p1 = f; | |
129 | const struct pkat *p2 = s; | |
74df8d6d | 130 | |
bd8b71e4 QY |
131 | return p1->peer == p2->peer; |
132 | } | |
133 | ||
134 | static unsigned int peer_hash_key(void *arg) | |
135 | { | |
136 | struct pkat *pkat = arg; | |
137 | return (uintptr_t)pkat->peer; | |
03014d48 QY |
138 | } |
139 | ||
a715eab3 | 140 | /* Cleanup handler / deinitializer. */ |
b72b6f4f | 141 | static void bgp_keepalives_finish(void *arg) |
419dfe6a | 142 | { |
bd8b71e4 QY |
143 | if (peerhash) { |
144 | hash_clean(peerhash, pkat_del); | |
145 | hash_free(peerhash); | |
146 | } | |
419dfe6a | 147 | |
bd8b71e4 | 148 | peerhash = NULL; |
419dfe6a | 149 | |
bd8b71e4 QY |
150 | pthread_mutex_unlock(peerhash_mtx); |
151 | pthread_mutex_destroy(peerhash_mtx); | |
152 | pthread_cond_destroy(peerhash_cond); | |
419dfe6a | 153 | |
424ab01d QY |
154 | XFREE(MTYPE_TMP, peerhash_mtx); |
155 | XFREE(MTYPE_TMP, peerhash_cond); | |
419dfe6a QY |
156 | } |
157 | ||
a715eab3 | 158 | /* |
419dfe6a | 159 | * Entry function for peer keepalive generation pthread. |
419dfe6a | 160 | */ |
b72b6f4f | 161 | void *bgp_keepalives_start(void *arg) |
419dfe6a | 162 | { |
a715eab3 QY |
163 | struct frr_pthread *fpt = arg; |
164 | fpt->master->owner = pthread_self(); | |
165 | ||
419dfe6a | 166 | struct timeval currtime = {0, 0}; |
bd8b71e4 | 167 | struct timeval aftertime = {0, 0}; |
419dfe6a QY |
168 | struct timeval next_update = {0, 0}; |
169 | struct timespec next_update_ts = {0, 0}; | |
170 | ||
a715eab3 QY |
171 | peerhash_mtx = XCALLOC(MTYPE_TMP, sizeof(pthread_mutex_t)); |
172 | peerhash_cond = XCALLOC(MTYPE_TMP, sizeof(pthread_cond_t)); | |
173 | ||
174 | /* initialize mutex */ | |
175 | pthread_mutex_init(peerhash_mtx, NULL); | |
176 | ||
177 | /* use monotonic clock with condition variable */ | |
178 | pthread_condattr_t attrs; | |
179 | pthread_condattr_init(&attrs); | |
180 | pthread_condattr_setclock(&attrs, CLOCK_MONOTONIC); | |
181 | pthread_cond_init(peerhash_cond, &attrs); | |
182 | pthread_condattr_destroy(&attrs); | |
183 | ||
c80bedb8 DS |
184 | /* |
185 | * We are not using normal FRR pthread mechanics and are | |
186 | * not using fpt_run | |
187 | */ | |
188 | frr_pthread_set_name(fpt); | |
a9198bc1 | 189 | |
a715eab3 QY |
190 | /* initialize peer hashtable */ |
191 | peerhash = hash_create_size(2048, peer_hash_key, peer_hash_cmp, NULL); | |
bd8b71e4 | 192 | pthread_mutex_lock(peerhash_mtx); |
03014d48 | 193 | |
a715eab3 | 194 | /* register cleanup handler */ |
b72b6f4f | 195 | pthread_cleanup_push(&bgp_keepalives_finish, NULL); |
03014d48 | 196 | |
a715eab3 QY |
197 | /* notify anybody waiting on us that we are done starting up */ |
198 | frr_pthread_notify_running(fpt); | |
03014d48 | 199 | |
a715eab3 | 200 | while (atomic_load_explicit(&fpt->running, memory_order_relaxed)) { |
bd8b71e4 QY |
201 | if (peerhash->count > 0) |
202 | pthread_cond_timedwait(peerhash_cond, peerhash_mtx, | |
03014d48 QY |
203 | &next_update_ts); |
204 | else | |
bd8b71e4 | 205 | while (peerhash->count == 0 |
a715eab3 QY |
206 | && atomic_load_explicit(&fpt->running, |
207 | memory_order_relaxed)) | |
bd8b71e4 | 208 | pthread_cond_wait(peerhash_cond, peerhash_mtx); |
03014d48 QY |
209 | |
210 | monotime(&currtime); | |
bd8b71e4 QY |
211 | |
212 | next_update.tv_sec = -1; | |
213 | ||
214 | hash_iterate(peerhash, peer_process, &next_update); | |
215 | if (next_update.tv_sec == -1) | |
216 | memset(&next_update, 0x00, sizeof(next_update)); | |
217 | ||
218 | monotime_since(&currtime, &aftertime); | |
219 | ||
03014d48 QY |
220 | timeradd(&currtime, &next_update, &next_update); |
221 | TIMEVAL_TO_TIMESPEC(&next_update, &next_update_ts); | |
222 | } | |
223 | ||
a715eab3 | 224 | /* clean up */ |
03014d48 QY |
225 | pthread_cleanup_pop(1); |
226 | ||
227 | return NULL; | |
228 | } | |
229 | ||
230 | /* --- thread external functions ------------------------------------------- */ | |
231 | ||
b72b6f4f | 232 | void bgp_keepalives_on(struct peer *peer) |
03014d48 | 233 | { |
096476dd QY |
234 | if (CHECK_FLAG(peer->thread_flags, PEER_THREAD_KEEPALIVES_ON)) |
235 | return; | |
236 | ||
1ac267a2 | 237 | struct frr_pthread *fpt = bgp_pth_ka; |
a715eab3 QY |
238 | assert(fpt->running); |
239 | ||
bd8b71e4 QY |
240 | /* placeholder bucket data to use for fast key lookups */ |
241 | static struct pkat holder = {0}; | |
242 | ||
68ede9c4 DS |
243 | /* |
244 | * We need to ensure that bgp_keepalives_init was called first | |
245 | */ | |
246 | assert(peerhash_mtx); | |
934af458 | 247 | |
bd8b71e4 | 248 | pthread_mutex_lock(peerhash_mtx); |
03014d48 | 249 | { |
bd8b71e4 QY |
250 | holder.peer = peer; |
251 | if (!hash_lookup(peerhash, &holder)) { | |
252 | struct pkat *pkat = pkat_new(peer); | |
253 | hash_get(peerhash, pkat, hash_alloc_intern); | |
254 | peer_lock(peer); | |
255 | } | |
49507a6f | 256 | SET_FLAG(peer->thread_flags, PEER_THREAD_KEEPALIVES_ON); |
03014d48 | 257 | } |
bd8b71e4 | 258 | pthread_mutex_unlock(peerhash_mtx); |
b72b6f4f | 259 | bgp_keepalives_wake(); |
03014d48 QY |
260 | } |
261 | ||
b72b6f4f | 262 | void bgp_keepalives_off(struct peer *peer) |
03014d48 | 263 | { |
096476dd QY |
264 | if (!CHECK_FLAG(peer->thread_flags, PEER_THREAD_KEEPALIVES_ON)) |
265 | return; | |
266 | ||
1ac267a2 | 267 | struct frr_pthread *fpt = bgp_pth_ka; |
a715eab3 QY |
268 | assert(fpt->running); |
269 | ||
bd8b71e4 QY |
270 | /* placeholder bucket data to use for fast key lookups */ |
271 | static struct pkat holder = {0}; | |
49507a6f | 272 | |
68ede9c4 DS |
273 | /* |
274 | * We need to ensure that bgp_keepalives_init was called first | |
275 | */ | |
276 | assert(peerhash_mtx); | |
934af458 | 277 | |
bd8b71e4 QY |
278 | pthread_mutex_lock(peerhash_mtx); |
279 | { | |
280 | holder.peer = peer; | |
281 | struct pkat *res = hash_release(peerhash, &holder); | |
282 | if (res) { | |
283 | pkat_del(res); | |
284 | peer_unlock(peer); | |
285 | } | |
49507a6f | 286 | UNSET_FLAG(peer->thread_flags, PEER_THREAD_KEEPALIVES_ON); |
03014d48 | 287 | } |
bd8b71e4 | 288 | pthread_mutex_unlock(peerhash_mtx); |
03014d48 QY |
289 | } |
290 | ||
4d762f26 | 291 | void bgp_keepalives_wake(void) |
03014d48 | 292 | { |
bd8b71e4 | 293 | pthread_mutex_lock(peerhash_mtx); |
03014d48 | 294 | { |
bd8b71e4 | 295 | pthread_cond_signal(peerhash_cond); |
03014d48 | 296 | } |
bd8b71e4 | 297 | pthread_mutex_unlock(peerhash_mtx); |
03014d48 | 298 | } |
0ca8b79f | 299 | |
a715eab3 | 300 | int bgp_keepalives_stop(struct frr_pthread *fpt, void **result) |
0ca8b79f | 301 | { |
a715eab3 QY |
302 | assert(fpt->running); |
303 | ||
304 | atomic_store_explicit(&fpt->running, false, memory_order_relaxed); | |
b72b6f4f | 305 | bgp_keepalives_wake(); |
a715eab3 | 306 | |
0ca8b79f QY |
307 | pthread_join(fpt->thread, result); |
308 | return 0; | |
309 | } |