]>
Commit | Line | Data |
---|---|---|
bd8b71e4 QY |
1 | /* |
2 | * BGP Keepalives. | |
03014d48 | 3 | * |
bd8b71e4 QY |
4 | * Implements a producer thread to generate BGP keepalives for peers. |
5 | * ---------------------------------------- | |
03014d48 | 6 | * Copyright (C) 2017 Cumulus Networks, Inc. |
bd8b71e4 | 7 | * Quentin Young |
03014d48 | 8 | * |
bd8b71e4 | 9 | * This file is part of FRRouting. |
03014d48 | 10 | * |
bd8b71e4 QY |
11 | * FRRouting is free software; you can redistribute it and/or modify it under |
12 | * the terms of the GNU General Public License as published by the Free | |
13 | * Software Foundation; either version 2, or (at your option) any later | |
03014d48 QY |
14 | * version. |
15 | * | |
bd8b71e4 QY |
16 | * FRRouting is distributed in the hope that it will be useful, but WITHOUT ANY |
17 | * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS | |
18 | * FOR A PARTICULAR PURPOSE. See the GNU General Public License for more | |
19 | * details. | |
03014d48 | 20 | * |
bd8b71e4 QY |
21 | * You should have received a copy of the GNU General Public License along with |
22 | * FRRouting; see the file COPYING. If not, write to the Free Software | |
23 | * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. | |
03014d48 QY |
24 | */ |
25 | #include <zebra.h> | |
26 | #include <signal.h> | |
27 | #include <sys/time.h> | |
28 | ||
29 | #include "thread.h" | |
30 | #include "log.h" | |
31 | #include "vty.h" | |
32 | #include "monotime.h" | |
bd8b71e4 | 33 | #include "hash.h" |
0ca8b79f | 34 | #include "frr_pthread.h" |
03014d48 QY |
35 | |
36 | #include "bgpd/bgpd.h" | |
37 | #include "bgpd/bgp_keepalives.h" | |
38 | #include "bgpd/bgp_debug.h" | |
39 | #include "bgpd/bgp_attr.h" | |
40 | #include "bgpd/bgp_packet.h" | |
41 | ||
42 | /** | |
43 | * Peer KeepAlive Timer. | |
44 | * Associates a peer with the time of its last keepalive. | |
45 | */ | |
46 | struct pkat { | |
47 | // the peer to send keepalives to | |
48 | struct peer *peer; | |
49 | // absolute time of last keepalive sent | |
50 | struct timeval last; | |
51 | }; | |
52 | ||
53 | /* List of peers we are sending keepalives for, and associated mutex. */ | |
bd8b71e4 QY |
54 | static pthread_mutex_t *peerhash_mtx; |
55 | static pthread_cond_t *peerhash_cond; | |
56 | static struct hash *peerhash; | |
03014d48 QY |
57 | |
58 | /* Thread control flag. */ | |
419dfe6a | 59 | bool bgp_keepalives_thread_run = false; |
03014d48 QY |
60 | |
61 | static struct pkat *pkat_new(struct peer *peer) | |
62 | { | |
63 | struct pkat *pkat = XMALLOC(MTYPE_TMP, sizeof(struct pkat)); | |
64 | pkat->peer = peer; | |
65 | monotime(&pkat->last); | |
66 | return pkat; | |
67 | } | |
68 | ||
69 | static void pkat_del(void *pkat) | |
70 | { | |
71 | XFREE(MTYPE_TMP, pkat); | |
72 | } | |
03014d48 | 73 | |
bd8b71e4 | 74 | |
03014d48 QY |
75 | /* |
76 | * Walks the list of peers, sending keepalives to those that are due for them. | |
77 | * | |
78 | * For any given peer, if the elapsed time since its last keepalive exceeds its | |
79 | * configured keepalive timer, a keepalive is sent to the peer and its | |
80 | * last-sent time is reset. Additionally, If the elapsed time does not exceed | |
81 | * the configured keepalive timer, but the time until the next keepalive is due | |
82 | * is within a hardcoded tolerance, a keepalive is sent as if the configured | |
83 | * timer was exceeded. Doing this helps alleviate nanosecond sleeps between | |
84 | * ticks by grouping together peers who are due for keepalives at roughly the | |
85 | * same time. This tolerance value is arbitrarily chosen to be 100ms. | |
86 | * | |
87 | * In addition, this function calculates the maximum amount of time that the | |
88 | * keepalive thread can sleep before another tick needs to take place. This is | |
89 | * equivalent to shortest time until a keepalive is due for any one peer. | |
90 | * | |
91 | * @return maximum time to wait until next update (0 if infinity) | |
92 | */ | |
bd8b71e4 | 93 | static void peer_process(struct hash_backet *hb, void *arg) |
03014d48 | 94 | { |
bd8b71e4 QY |
95 | struct pkat *pkat = hb->data; |
96 | ||
97 | struct timeval *next_update = arg; | |
03014d48 | 98 | |
03014d48 QY |
99 | static struct timeval elapsed; // elapsed time since keepalive |
100 | static struct timeval ka = {0}; // peer->v_keepalive as a timeval | |
101 | static struct timeval diff; // ka - elapsed | |
102 | ||
03014d48 QY |
103 | static struct timeval tolerance = {0, 100000}; |
104 | ||
bd8b71e4 QY |
105 | // calculate elapsed time since last keepalive |
106 | monotime_since(&pkat->last, &elapsed); | |
03014d48 | 107 | |
bd8b71e4 QY |
108 | // calculate difference between elapsed time and configured time |
109 | ka.tv_sec = pkat->peer->v_keepalive; | |
110 | timersub(&ka, &elapsed, &diff); | |
03014d48 | 111 | |
bd8b71e4 QY |
112 | int send_keepalive = |
113 | elapsed.tv_sec >= ka.tv_sec || timercmp(&diff, &tolerance, <); | |
03014d48 | 114 | |
bd8b71e4 QY |
115 | if (send_keepalive) { |
116 | if (bgp_debug_neighbor_events(pkat->peer)) | |
117 | zlog_debug("%s [FSM] Timer (keepalive timer expire)", | |
118 | pkat->peer->host); | |
03014d48 | 119 | |
bd8b71e4 QY |
120 | bgp_keepalive_send(pkat->peer); |
121 | monotime(&pkat->last); | |
122 | memset(&elapsed, 0x00, sizeof(struct timeval)); | |
123 | diff = ka; // time until next keepalive == peer keepalive time | |
03014d48 QY |
124 | } |
125 | ||
bd8b71e4 QY |
126 | // if calculated next update for this peer < current delay, use it |
127 | if (next_update->tv_sec <= 0 || timercmp(&diff, next_update, <)) | |
128 | *next_update = diff; | |
129 | } | |
130 | ||
131 | static int peer_hash_cmp(const void *f, const void *s) | |
132 | { | |
133 | const struct pkat *p1 = f; | |
134 | const struct pkat *p2 = s; | |
135 | return p1->peer == p2->peer; | |
136 | } | |
137 | ||
138 | static unsigned int peer_hash_key(void *arg) | |
139 | { | |
140 | struct pkat *pkat = arg; | |
141 | return (uintptr_t)pkat->peer; | |
03014d48 QY |
142 | } |
143 | ||
419dfe6a | 144 | void peer_keepalives_init() |
03014d48 | 145 | { |
bd8b71e4 QY |
146 | peerhash_mtx = XCALLOC(MTYPE_PTHREAD, sizeof(pthread_mutex_t)); |
147 | peerhash_cond = XCALLOC(MTYPE_PTHREAD, sizeof(pthread_cond_t)); | |
03014d48 | 148 | |
419dfe6a | 149 | // initialize mutex |
bd8b71e4 | 150 | pthread_mutex_init(peerhash_mtx, NULL); |
03014d48 QY |
151 | |
152 | // use monotonic clock with condition variable | |
153 | pthread_condattr_t attrs; | |
154 | pthread_condattr_init(&attrs); | |
155 | pthread_condattr_setclock(&attrs, CLOCK_MONOTONIC); | |
bd8b71e4 | 156 | pthread_cond_init(peerhash_cond, &attrs); |
419dfe6a | 157 | pthread_condattr_destroy(&attrs); |
03014d48 | 158 | |
bd8b71e4 QY |
159 | // initialize peer hashtable |
160 | peerhash = hash_create_size(2048, peer_hash_key, peer_hash_cmp); | |
419dfe6a QY |
161 | } |
162 | ||
163 | static void peer_keepalives_finish(void *arg) | |
164 | { | |
165 | bgp_keepalives_thread_run = false; | |
166 | ||
bd8b71e4 QY |
167 | if (peerhash) { |
168 | hash_clean(peerhash, pkat_del); | |
169 | hash_free(peerhash); | |
170 | } | |
419dfe6a | 171 | |
bd8b71e4 | 172 | peerhash = NULL; |
419dfe6a | 173 | |
bd8b71e4 QY |
174 | pthread_mutex_unlock(peerhash_mtx); |
175 | pthread_mutex_destroy(peerhash_mtx); | |
176 | pthread_cond_destroy(peerhash_cond); | |
419dfe6a | 177 | |
bd8b71e4 QY |
178 | XFREE(MTYPE_PTHREAD, peerhash_mtx); |
179 | XFREE(MTYPE_PTHREAD, peerhash_cond); | |
419dfe6a QY |
180 | } |
181 | ||
182 | /** | |
183 | * Entry function for peer keepalive generation pthread. | |
184 | * | |
185 | * peer_keepalives_init() must be called prior to this. | |
186 | */ | |
187 | void *peer_keepalives_start(void *arg) | |
188 | { | |
189 | struct timeval currtime = {0, 0}; | |
bd8b71e4 | 190 | struct timeval aftertime = {0, 0}; |
419dfe6a QY |
191 | struct timeval next_update = {0, 0}; |
192 | struct timespec next_update_ts = {0, 0}; | |
193 | ||
bd8b71e4 | 194 | pthread_mutex_lock(peerhash_mtx); |
03014d48 | 195 | |
419dfe6a QY |
196 | // register cleanup handler |
197 | pthread_cleanup_push(&peer_keepalives_finish, NULL); | |
03014d48 QY |
198 | |
199 | bgp_keepalives_thread_run = true; | |
200 | ||
201 | while (bgp_keepalives_thread_run) { | |
bd8b71e4 QY |
202 | if (peerhash->count > 0) |
203 | pthread_cond_timedwait(peerhash_cond, peerhash_mtx, | |
03014d48 QY |
204 | &next_update_ts); |
205 | else | |
bd8b71e4 | 206 | while (peerhash->count == 0 |
03014d48 | 207 | && bgp_keepalives_thread_run) |
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 | ||
224 | // clean up | |
225 | pthread_cleanup_pop(1); | |
226 | ||
227 | return NULL; | |
228 | } | |
229 | ||
230 | /* --- thread external functions ------------------------------------------- */ | |
231 | ||
232 | void peer_keepalives_on(struct peer *peer) | |
233 | { | |
bd8b71e4 QY |
234 | /* placeholder bucket data to use for fast key lookups */ |
235 | static struct pkat holder = {0}; | |
236 | ||
237 | pthread_mutex_lock(peerhash_mtx); | |
03014d48 | 238 | { |
bd8b71e4 QY |
239 | holder.peer = peer; |
240 | if (!hash_lookup(peerhash, &holder)) { | |
241 | struct pkat *pkat = pkat_new(peer); | |
242 | hash_get(peerhash, pkat, hash_alloc_intern); | |
243 | peer_lock(peer); | |
244 | } | |
49507a6f | 245 | SET_FLAG(peer->thread_flags, PEER_THREAD_KEEPALIVES_ON); |
03014d48 | 246 | } |
bd8b71e4 | 247 | pthread_mutex_unlock(peerhash_mtx); |
03014d48 QY |
248 | peer_keepalives_wake(); |
249 | } | |
250 | ||
251 | void peer_keepalives_off(struct peer *peer) | |
252 | { | |
bd8b71e4 QY |
253 | /* placeholder bucket data to use for fast key lookups */ |
254 | static struct pkat holder = {0}; | |
49507a6f | 255 | |
bd8b71e4 QY |
256 | pthread_mutex_lock(peerhash_mtx); |
257 | { | |
258 | holder.peer = peer; | |
259 | struct pkat *res = hash_release(peerhash, &holder); | |
260 | if (res) { | |
261 | pkat_del(res); | |
262 | peer_unlock(peer); | |
263 | } | |
49507a6f | 264 | UNSET_FLAG(peer->thread_flags, PEER_THREAD_KEEPALIVES_ON); |
03014d48 | 265 | } |
bd8b71e4 | 266 | pthread_mutex_unlock(peerhash_mtx); |
03014d48 QY |
267 | } |
268 | ||
269 | void peer_keepalives_wake() | |
270 | { | |
bd8b71e4 | 271 | pthread_mutex_lock(peerhash_mtx); |
03014d48 | 272 | { |
bd8b71e4 | 273 | pthread_cond_signal(peerhash_cond); |
03014d48 | 274 | } |
bd8b71e4 | 275 | pthread_mutex_unlock(peerhash_mtx); |
03014d48 | 276 | } |
0ca8b79f QY |
277 | |
278 | int peer_keepalives_stop(void **result) | |
279 | { | |
280 | struct frr_pthread *fpt = frr_pthread_get(PTHREAD_KEEPALIVES); | |
281 | bgp_keepalives_thread_run = false; | |
282 | peer_keepalives_wake(); | |
283 | pthread_join(fpt->thread, result); | |
284 | return 0; | |
285 | } |