]>
Commit | Line | Data |
---|---|---|
acddc0ed | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
5adc2528 AS |
2 | /* |
3 | * Main implementation file for interface to Forwarding Plane Manager. | |
4 | * | |
5 | * Copyright (C) 2012 by Open Source Routing. | |
6 | * Copyright (C) 2012 by Internet Systems Consortium, Inc. ("ISC") | |
5adc2528 AS |
7 | */ |
8 | ||
9 | #include <zebra.h> | |
10 | ||
11 | #include "log.h" | |
4f8ea50c | 12 | #include "libfrr.h" |
5adc2528 | 13 | #include "stream.h" |
24a58196 | 14 | #include "frrevent.h" |
5adc2528 AS |
15 | #include "network.h" |
16 | #include "command.h" | |
09781197 | 17 | #include "lib/version.h" |
e5218ec8 | 18 | #include "jhash.h" |
5adc2528 AS |
19 | |
20 | #include "zebra/rib.h" | |
7c551956 DS |
21 | #include "zebra/zserv.h" |
22 | #include "zebra/zebra_ns.h" | |
23 | #include "zebra/zebra_vrf.h" | |
67aeb554 | 24 | #include "zebra/zebra_errors.h" |
5adc2528 AS |
25 | |
26 | #include "fpm/fpm.h" | |
5adc2528 | 27 | #include "zebra_fpm_private.h" |
e5218ec8 | 28 | #include "zebra/zebra_router.h" |
a780a738 | 29 | #include "zebra_vxlan_private.h" |
e5218ec8 AD |
30 | |
31 | DEFINE_MTYPE_STATIC(ZEBRA, FPM_MAC_INFO, "FPM_MAC_INFO"); | |
5adc2528 AS |
32 | |
33 | /* | |
34 | * Interval at which we attempt to connect to the FPM. | |
35 | */ | |
36 | #define ZFPM_CONNECT_RETRY_IVL 5 | |
37 | ||
38 | /* | |
39 | * Sizes of outgoing and incoming stream buffers for writing/reading | |
40 | * FPM messages. | |
41 | */ | |
42 | #define ZFPM_OBUF_SIZE (2 * FPM_MAX_MSG_LEN) | |
43 | #define ZFPM_IBUF_SIZE (FPM_MAX_MSG_LEN) | |
44 | ||
45 | /* | |
46 | * The maximum number of times the FPM socket write callback can call | |
47 | * 'write' before it yields. | |
48 | */ | |
49 | #define ZFPM_MAX_WRITES_PER_RUN 10 | |
50 | ||
51 | /* | |
52 | * Interval over which we collect statistics. | |
53 | */ | |
54 | #define ZFPM_STATS_IVL_SECS 10 | |
fbe748e5 AD |
55 | #define FPM_MAX_MAC_MSG_LEN 512 |
56 | ||
1ac88792 | 57 | static void zfpm_iterate_rmac_table(struct hash_bucket *bucket, void *args); |
5adc2528 AS |
58 | |
59 | /* | |
60 | * Structure that holds state for iterating over all route_node | |
61 | * structures that are candidates for being communicated to the FPM. | |
62 | */ | |
332cba05 | 63 | struct zfpm_rnodes_iter { |
d62a17ae | 64 | rib_tables_iter_t tables_iter; |
65 | route_table_iter_t iter; | |
332cba05 | 66 | }; |
5adc2528 AS |
67 | |
68 | /* | |
69 | * Statistics. | |
70 | */ | |
eeaf257b | 71 | struct zfpm_stats { |
d62a17ae | 72 | unsigned long connect_calls; |
73 | unsigned long connect_no_sock; | |
5adc2528 | 74 | |
d62a17ae | 75 | unsigned long read_cb_calls; |
5adc2528 | 76 | |
d62a17ae | 77 | unsigned long write_cb_calls; |
78 | unsigned long write_calls; | |
79 | unsigned long partial_writes; | |
80 | unsigned long max_writes_hit; | |
81 | unsigned long t_write_yields; | |
5adc2528 | 82 | |
d62a17ae | 83 | unsigned long nop_deletes_skipped; |
84 | unsigned long route_adds; | |
85 | unsigned long route_dels; | |
5adc2528 | 86 | |
d62a17ae | 87 | unsigned long updates_triggered; |
88 | unsigned long redundant_triggers; | |
5adc2528 | 89 | |
d62a17ae | 90 | unsigned long dests_del_after_update; |
5adc2528 | 91 | |
d62a17ae | 92 | unsigned long t_conn_down_starts; |
93 | unsigned long t_conn_down_dests_processed; | |
94 | unsigned long t_conn_down_yields; | |
95 | unsigned long t_conn_down_finishes; | |
5adc2528 | 96 | |
d62a17ae | 97 | unsigned long t_conn_up_starts; |
98 | unsigned long t_conn_up_dests_processed; | |
99 | unsigned long t_conn_up_yields; | |
100 | unsigned long t_conn_up_aborts; | |
101 | unsigned long t_conn_up_finishes; | |
eeaf257b | 102 | }; |
5adc2528 AS |
103 | |
104 | /* | |
105 | * States for the FPM state machine. | |
106 | */ | |
1d6a3ee8 | 107 | enum zfpm_state { |
5adc2528 | 108 | |
d62a17ae | 109 | /* |
110 | * In this state we are not yet ready to connect to the FPM. This | |
111 | * can happen when this module is disabled, or if we're cleaning up | |
112 | * after a connection has gone down. | |
113 | */ | |
114 | ZFPM_STATE_IDLE, | |
115 | ||
116 | /* | |
117 | * Ready to talk to the FPM and periodically trying to connect to | |
118 | * it. | |
119 | */ | |
120 | ZFPM_STATE_ACTIVE, | |
121 | ||
122 | /* | |
123 | * In the middle of bringing up a TCP connection. Specifically, | |
124 | * waiting for a connect() call to complete asynchronously. | |
125 | */ | |
126 | ZFPM_STATE_CONNECTING, | |
127 | ||
128 | /* | |
129 | * TCP connection to the FPM is up. | |
130 | */ | |
131 | ZFPM_STATE_ESTABLISHED | |
5adc2528 | 132 | |
1d6a3ee8 | 133 | }; |
5adc2528 | 134 | |
fb0aa886 AS |
135 | /* |
136 | * Message format to be used to communicate with the FPM. | |
137 | */ | |
a78c2b98 | 138 | enum zfpm_msg_format { |
d62a17ae | 139 | ZFPM_MSG_FORMAT_NONE, |
140 | ZFPM_MSG_FORMAT_NETLINK, | |
141 | ZFPM_MSG_FORMAT_PROTOBUF, | |
a78c2b98 DS |
142 | }; |
143 | ||
5adc2528 AS |
144 | /* |
145 | * Globals. | |
146 | */ | |
768e40bd | 147 | struct zfpm_glob { |
d62a17ae | 148 | |
149 | /* | |
150 | * True if the FPM module has been enabled. | |
151 | */ | |
152 | int enabled; | |
153 | ||
154 | /* | |
155 | * Message format to be used to communicate with the fpm. | |
156 | */ | |
a78c2b98 | 157 | enum zfpm_msg_format message_format; |
d62a17ae | 158 | |
cd9d0537 | 159 | struct event_loop *master; |
d62a17ae | 160 | |
1d6a3ee8 | 161 | enum zfpm_state state; |
d62a17ae | 162 | |
163 | in_addr_t fpm_server; | |
164 | /* | |
165 | * Port on which the FPM is running. | |
166 | */ | |
167 | int fpm_port; | |
168 | ||
169 | /* | |
170 | * List of rib_dest_t structures to be processed | |
171 | */ | |
172 | TAILQ_HEAD(zfpm_dest_q, rib_dest_t_) dest_q; | |
173 | ||
e5218ec8 AD |
174 | /* |
175 | * List of fpm_mac_info structures to be processed | |
176 | */ | |
177 | TAILQ_HEAD(zfpm_mac_q, fpm_mac_info_t) mac_q; | |
178 | ||
179 | /* | |
180 | * Hash table of fpm_mac_info_t entries | |
181 | * | |
182 | * While adding fpm_mac_info_t for a MAC to the mac_q, | |
183 | * it is possible that another fpm_mac_info_t node for the this MAC | |
184 | * is already present in the queue. | |
185 | * This is possible in the case of consecutive add->delete operations. | |
186 | * To avoid such duplicate insertions in the mac_q, | |
187 | * define a hash table for fpm_mac_info_t which can be looked up | |
188 | * to see if an fpm_mac_info_t node for a MAC is already present | |
189 | * in the mac_q. | |
190 | */ | |
191 | struct hash *fpm_mac_info_table; | |
192 | ||
d62a17ae | 193 | /* |
194 | * Stream socket to the FPM. | |
195 | */ | |
196 | int sock; | |
197 | ||
198 | /* | |
199 | * Buffers for messages to/from the FPM. | |
200 | */ | |
201 | struct stream *obuf; | |
202 | struct stream *ibuf; | |
203 | ||
204 | /* | |
205 | * Threads for I/O. | |
206 | */ | |
e6685141 DS |
207 | struct event *t_connect; |
208 | struct event *t_write; | |
209 | struct event *t_read; | |
d62a17ae | 210 | |
211 | /* | |
212 | * Thread to clean up after the TCP connection to the FPM goes down | |
213 | * and the state that belongs to it. | |
214 | */ | |
e6685141 | 215 | struct event *t_conn_down; |
d62a17ae | 216 | |
217 | struct { | |
332cba05 | 218 | struct zfpm_rnodes_iter iter; |
d62a17ae | 219 | } t_conn_down_state; |
220 | ||
221 | /* | |
222 | * Thread to take actions once the TCP conn to the FPM comes up, and | |
223 | * the state that belongs to it. | |
224 | */ | |
e6685141 | 225 | struct event *t_conn_up; |
d62a17ae | 226 | |
227 | struct { | |
332cba05 | 228 | struct zfpm_rnodes_iter iter; |
d62a17ae | 229 | } t_conn_up_state; |
230 | ||
231 | unsigned long connect_calls; | |
232 | time_t last_connect_call_time; | |
233 | ||
234 | /* | |
235 | * Stats from the start of the current statistics interval up to | |
236 | * now. These are the counters we typically update in the code. | |
237 | */ | |
eeaf257b | 238 | struct zfpm_stats stats; |
d62a17ae | 239 | |
240 | /* | |
241 | * Statistics that were gathered in the last collection interval. | |
242 | */ | |
eeaf257b | 243 | struct zfpm_stats last_ivl_stats; |
d62a17ae | 244 | |
245 | /* | |
246 | * Cumulative stats from the last clear to the start of the current | |
247 | * statistics interval. | |
248 | */ | |
eeaf257b | 249 | struct zfpm_stats cumulative_stats; |
d62a17ae | 250 | |
251 | /* | |
252 | * Stats interval timer. | |
253 | */ | |
e6685141 | 254 | struct event *t_stats; |
d62a17ae | 255 | |
256 | /* | |
257 | * If non-zero, the last time when statistics were cleared. | |
258 | */ | |
259 | time_t last_stats_clear_time; | |
e840edca KK |
260 | |
261 | /* | |
262 | * Flag to track the MAC dump status to FPM | |
263 | */ | |
264 | bool fpm_mac_dump_done; | |
768e40bd | 265 | }; |
5adc2528 | 266 | |
768e40bd DS |
267 | static struct zfpm_glob zfpm_glob_space; |
268 | static struct zfpm_glob *zfpm_g = &zfpm_glob_space; | |
5adc2528 | 269 | |
d62a17ae | 270 | static int zfpm_trigger_update(struct route_node *rn, const char *reason); |
4f8ea50c | 271 | |
e6685141 DS |
272 | static void zfpm_read_cb(struct event *thread); |
273 | static void zfpm_write_cb(struct event *thread); | |
5adc2528 | 274 | |
1d6a3ee8 | 275 | static void zfpm_set_state(enum zfpm_state state, const char *reason); |
d62a17ae | 276 | static void zfpm_start_connect_timer(const char *reason); |
277 | static void zfpm_start_stats_timer(void); | |
a780a738 | 278 | static void zfpm_mac_info_del(struct fpm_mac_info_t *fpm_mac); |
5adc2528 | 279 | |
316d2d52 NK |
280 | static const char ipv4_ll_buf[16] = "169.254.0.1"; |
281 | union g_addr ipv4ll_gateway; | |
282 | ||
5adc2528 AS |
283 | /* |
284 | * zfpm_thread_should_yield | |
285 | */ | |
e6685141 | 286 | static inline int zfpm_thread_should_yield(struct event *t) |
5adc2528 | 287 | { |
70c35c11 | 288 | return event_should_yield(t); |
5adc2528 AS |
289 | } |
290 | ||
291 | /* | |
292 | * zfpm_state_to_str | |
293 | */ | |
1d6a3ee8 | 294 | static const char *zfpm_state_to_str(enum zfpm_state state) |
5adc2528 | 295 | { |
d62a17ae | 296 | switch (state) { |
5adc2528 | 297 | |
d62a17ae | 298 | case ZFPM_STATE_IDLE: |
299 | return "idle"; | |
5adc2528 | 300 | |
d62a17ae | 301 | case ZFPM_STATE_ACTIVE: |
302 | return "active"; | |
5adc2528 | 303 | |
d62a17ae | 304 | case ZFPM_STATE_CONNECTING: |
305 | return "connecting"; | |
5adc2528 | 306 | |
d62a17ae | 307 | case ZFPM_STATE_ESTABLISHED: |
308 | return "established"; | |
5adc2528 | 309 | |
d62a17ae | 310 | default: |
311 | return "unknown"; | |
312 | } | |
5adc2528 AS |
313 | } |
314 | ||
5adc2528 AS |
315 | /* |
316 | * zfpm_get_elapsed_time | |
317 | * | |
318 | * Returns the time elapsed (in seconds) since the given time. | |
319 | */ | |
d62a17ae | 320 | static time_t zfpm_get_elapsed_time(time_t reference) |
5adc2528 | 321 | { |
d62a17ae | 322 | time_t now; |
5adc2528 | 323 | |
d62a17ae | 324 | now = monotime(NULL); |
5adc2528 | 325 | |
d62a17ae | 326 | if (now < reference) { |
327 | assert(0); | |
328 | return 0; | |
329 | } | |
5adc2528 | 330 | |
d62a17ae | 331 | return now - reference; |
5adc2528 AS |
332 | } |
333 | ||
5adc2528 AS |
334 | /* |
335 | * zfpm_rnodes_iter_init | |
336 | */ | |
332cba05 | 337 | static inline void zfpm_rnodes_iter_init(struct zfpm_rnodes_iter *iter) |
5adc2528 | 338 | { |
d62a17ae | 339 | memset(iter, 0, sizeof(*iter)); |
340 | rib_tables_iter_init(&iter->tables_iter); | |
341 | ||
342 | /* | |
343 | * This is a hack, but it makes implementing 'next' easier by | |
344 | * ensuring that route_table_iter_next() will return NULL the first | |
345 | * time we call it. | |
346 | */ | |
347 | route_table_iter_init(&iter->iter, NULL); | |
348 | route_table_iter_cleanup(&iter->iter); | |
5adc2528 AS |
349 | } |
350 | ||
351 | /* | |
352 | * zfpm_rnodes_iter_next | |
353 | */ | |
332cba05 DS |
354 | static inline struct route_node * |
355 | zfpm_rnodes_iter_next(struct zfpm_rnodes_iter *iter) | |
5adc2528 | 356 | { |
d62a17ae | 357 | struct route_node *rn; |
358 | struct route_table *table; | |
5adc2528 | 359 | |
d62a17ae | 360 | while (1) { |
361 | rn = route_table_iter_next(&iter->iter); | |
362 | if (rn) | |
363 | return rn; | |
5adc2528 | 364 | |
d62a17ae | 365 | /* |
366 | * We've made our way through this table, go to the next one. | |
367 | */ | |
368 | route_table_iter_cleanup(&iter->iter); | |
5adc2528 | 369 | |
c6bbea17 | 370 | table = rib_tables_iter_next(&iter->tables_iter); |
5adc2528 | 371 | |
d62a17ae | 372 | if (!table) |
373 | return NULL; | |
5adc2528 | 374 | |
d62a17ae | 375 | route_table_iter_init(&iter->iter, table); |
376 | } | |
5adc2528 | 377 | |
d62a17ae | 378 | return NULL; |
5adc2528 AS |
379 | } |
380 | ||
381 | /* | |
382 | * zfpm_rnodes_iter_pause | |
383 | */ | |
332cba05 | 384 | static inline void zfpm_rnodes_iter_pause(struct zfpm_rnodes_iter *iter) |
5adc2528 | 385 | { |
d62a17ae | 386 | route_table_iter_pause(&iter->iter); |
5adc2528 AS |
387 | } |
388 | ||
389 | /* | |
390 | * zfpm_rnodes_iter_cleanup | |
391 | */ | |
332cba05 | 392 | static inline void zfpm_rnodes_iter_cleanup(struct zfpm_rnodes_iter *iter) |
5adc2528 | 393 | { |
d62a17ae | 394 | route_table_iter_cleanup(&iter->iter); |
395 | rib_tables_iter_cleanup(&iter->tables_iter); | |
5adc2528 AS |
396 | } |
397 | ||
398 | /* | |
399 | * zfpm_stats_init | |
400 | * | |
401 | * Initialize a statistics block. | |
402 | */ | |
eeaf257b | 403 | static inline void zfpm_stats_init(struct zfpm_stats *stats) |
5adc2528 | 404 | { |
d62a17ae | 405 | memset(stats, 0, sizeof(*stats)); |
5adc2528 AS |
406 | } |
407 | ||
408 | /* | |
409 | * zfpm_stats_reset | |
410 | */ | |
eeaf257b | 411 | static inline void zfpm_stats_reset(struct zfpm_stats *stats) |
5adc2528 | 412 | { |
d62a17ae | 413 | zfpm_stats_init(stats); |
5adc2528 AS |
414 | } |
415 | ||
416 | /* | |
417 | * zfpm_stats_copy | |
418 | */ | |
eeaf257b DS |
419 | static inline void zfpm_stats_copy(const struct zfpm_stats *src, |
420 | struct zfpm_stats *dest) | |
5adc2528 | 421 | { |
d62a17ae | 422 | memcpy(dest, src, sizeof(*dest)); |
5adc2528 AS |
423 | } |
424 | ||
425 | /* | |
426 | * zfpm_stats_compose | |
427 | * | |
428 | * Total up the statistics in two stats structures ('s1 and 's2') and | |
429 | * return the result in the third argument, 'result'. Note that the | |
430 | * pointer 'result' may be the same as 's1' or 's2'. | |
431 | * | |
432 | * For simplicity, the implementation below assumes that the stats | |
433 | * structure is composed entirely of counters. This can easily be | |
434 | * changed when necessary. | |
435 | */ | |
eeaf257b DS |
436 | static void zfpm_stats_compose(const struct zfpm_stats *s1, |
437 | const struct zfpm_stats *s2, | |
438 | struct zfpm_stats *result) | |
5adc2528 | 439 | { |
d62a17ae | 440 | const unsigned long *p1, *p2; |
441 | unsigned long *result_p; | |
442 | int i, num_counters; | |
5adc2528 | 443 | |
d62a17ae | 444 | p1 = (const unsigned long *)s1; |
445 | p2 = (const unsigned long *)s2; | |
446 | result_p = (unsigned long *)result; | |
5adc2528 | 447 | |
eeaf257b | 448 | num_counters = (sizeof(struct zfpm_stats) / sizeof(unsigned long)); |
5adc2528 | 449 | |
d62a17ae | 450 | for (i = 0; i < num_counters; i++) { |
451 | result_p[i] = p1[i] + p2[i]; | |
452 | } | |
5adc2528 AS |
453 | } |
454 | ||
455 | /* | |
456 | * zfpm_read_on | |
457 | */ | |
d62a17ae | 458 | static inline void zfpm_read_on(void) |
5adc2528 | 459 | { |
d62a17ae | 460 | assert(!zfpm_g->t_read); |
461 | assert(zfpm_g->sock >= 0); | |
5adc2528 | 462 | |
907a2395 DS |
463 | event_add_read(zfpm_g->master, zfpm_read_cb, 0, zfpm_g->sock, |
464 | &zfpm_g->t_read); | |
5adc2528 AS |
465 | } |
466 | ||
467 | /* | |
468 | * zfpm_write_on | |
469 | */ | |
d62a17ae | 470 | static inline void zfpm_write_on(void) |
5adc2528 | 471 | { |
d62a17ae | 472 | assert(!zfpm_g->t_write); |
473 | assert(zfpm_g->sock >= 0); | |
5adc2528 | 474 | |
907a2395 DS |
475 | event_add_write(zfpm_g->master, zfpm_write_cb, 0, zfpm_g->sock, |
476 | &zfpm_g->t_write); | |
5adc2528 AS |
477 | } |
478 | ||
479 | /* | |
480 | * zfpm_read_off | |
481 | */ | |
d62a17ae | 482 | static inline void zfpm_read_off(void) |
5adc2528 | 483 | { |
e16d030c | 484 | EVENT_OFF(zfpm_g->t_read); |
5adc2528 AS |
485 | } |
486 | ||
487 | /* | |
488 | * zfpm_write_off | |
489 | */ | |
d62a17ae | 490 | static inline void zfpm_write_off(void) |
5adc2528 | 491 | { |
e16d030c | 492 | EVENT_OFF(zfpm_g->t_write); |
5adc2528 AS |
493 | } |
494 | ||
f0c459f0 DS |
495 | static inline void zfpm_connect_off(void) |
496 | { | |
e16d030c | 497 | EVENT_OFF(zfpm_g->t_connect); |
f0c459f0 DS |
498 | } |
499 | ||
54033432 DS |
500 | static inline void zfpm_conn_down_off(void) |
501 | { | |
502 | EVENT_OFF(zfpm_g->t_conn_down); | |
503 | } | |
504 | ||
5adc2528 AS |
505 | /* |
506 | * zfpm_conn_up_thread_cb | |
507 | * | |
508 | * Callback for actions to be taken when the connection to the FPM | |
509 | * comes up. | |
510 | */ | |
e6685141 | 511 | static void zfpm_conn_up_thread_cb(struct event *thread) |
5adc2528 | 512 | { |
d62a17ae | 513 | struct route_node *rnode; |
332cba05 | 514 | struct zfpm_rnodes_iter *iter; |
d62a17ae | 515 | rib_dest_t *dest; |
5adc2528 | 516 | |
d62a17ae | 517 | iter = &zfpm_g->t_conn_up_state.iter; |
5adc2528 | 518 | |
d62a17ae | 519 | if (zfpm_g->state != ZFPM_STATE_ESTABLISHED) { |
520 | zfpm_debug( | |
521 | "Connection not up anymore, conn_up thread aborting"); | |
522 | zfpm_g->stats.t_conn_up_aborts++; | |
523 | goto done; | |
524 | } | |
5adc2528 | 525 | |
e840edca KK |
526 | if (!zfpm_g->fpm_mac_dump_done) { |
527 | /* Enqueue FPM updates for all the RMAC entries */ | |
528 | hash_iterate(zrouter.l3vni_table, zfpm_iterate_rmac_table, | |
529 | NULL); | |
530 | /* mark dump done so that its not repeated after yield */ | |
531 | zfpm_g->fpm_mac_dump_done = true; | |
532 | } | |
fbe748e5 | 533 | |
d62a17ae | 534 | while ((rnode = zfpm_rnodes_iter_next(iter))) { |
535 | dest = rib_dest_from_rnode(rnode); | |
536 | ||
537 | if (dest) { | |
538 | zfpm_g->stats.t_conn_up_dests_processed++; | |
539 | zfpm_trigger_update(rnode, NULL); | |
540 | } | |
541 | ||
542 | /* | |
543 | * Yield if need be. | |
544 | */ | |
545 | if (!zfpm_thread_should_yield(thread)) | |
546 | continue; | |
547 | ||
548 | zfpm_g->stats.t_conn_up_yields++; | |
549 | zfpm_rnodes_iter_pause(iter); | |
907a2395 DS |
550 | event_add_timer_msec(zfpm_g->master, zfpm_conn_up_thread_cb, |
551 | NULL, 0, &zfpm_g->t_conn_up); | |
cc9f21da | 552 | return; |
5adc2528 AS |
553 | } |
554 | ||
d62a17ae | 555 | zfpm_g->stats.t_conn_up_finishes++; |
556 | ||
557 | done: | |
558 | zfpm_rnodes_iter_cleanup(iter); | |
5adc2528 AS |
559 | } |
560 | ||
561 | /* | |
562 | * zfpm_connection_up | |
563 | * | |
564 | * Called when the connection to the FPM comes up. | |
565 | */ | |
d62a17ae | 566 | static void zfpm_connection_up(const char *detail) |
5adc2528 | 567 | { |
d62a17ae | 568 | assert(zfpm_g->sock >= 0); |
569 | zfpm_read_on(); | |
570 | zfpm_write_on(); | |
571 | zfpm_set_state(ZFPM_STATE_ESTABLISHED, detail); | |
572 | ||
573 | /* | |
574 | * Start thread to push existing routes to the FPM. | |
575 | */ | |
e16d030c | 576 | EVENT_OFF(zfpm_g->t_conn_up); |
d62a17ae | 577 | |
578 | zfpm_rnodes_iter_init(&zfpm_g->t_conn_up_state.iter); | |
e840edca | 579 | zfpm_g->fpm_mac_dump_done = false; |
d62a17ae | 580 | |
581 | zfpm_debug("Starting conn_up thread"); | |
ef1dbba8 | 582 | |
907a2395 DS |
583 | event_add_timer_msec(zfpm_g->master, zfpm_conn_up_thread_cb, NULL, 0, |
584 | &zfpm_g->t_conn_up); | |
d62a17ae | 585 | zfpm_g->stats.t_conn_up_starts++; |
5adc2528 AS |
586 | } |
587 | ||
588 | /* | |
589 | * zfpm_connect_check | |
590 | * | |
591 | * Check if an asynchronous connect() to the FPM is complete. | |
592 | */ | |
d62a17ae | 593 | static void zfpm_connect_check(void) |
5adc2528 | 594 | { |
d62a17ae | 595 | int status; |
596 | socklen_t slen; | |
597 | int ret; | |
598 | ||
599 | zfpm_read_off(); | |
600 | zfpm_write_off(); | |
601 | ||
602 | slen = sizeof(status); | |
603 | ret = getsockopt(zfpm_g->sock, SOL_SOCKET, SO_ERROR, (void *)&status, | |
604 | &slen); | |
605 | ||
606 | if (ret >= 0 && status == 0) { | |
607 | zfpm_connection_up("async connect complete"); | |
608 | return; | |
609 | } | |
610 | ||
611 | /* | |
612 | * getsockopt() failed or indicated an error on the socket. | |
613 | */ | |
614 | close(zfpm_g->sock); | |
615 | zfpm_g->sock = -1; | |
616 | ||
617 | zfpm_start_connect_timer("getsockopt() after async connect failed"); | |
618 | return; | |
5adc2528 AS |
619 | } |
620 | ||
621 | /* | |
622 | * zfpm_conn_down_thread_cb | |
623 | * | |
624 | * Callback that is invoked to clean up state after the TCP connection | |
625 | * to the FPM goes down. | |
626 | */ | |
e6685141 | 627 | static void zfpm_conn_down_thread_cb(struct event *thread) |
5adc2528 | 628 | { |
d62a17ae | 629 | struct route_node *rnode; |
332cba05 | 630 | struct zfpm_rnodes_iter *iter; |
d62a17ae | 631 | rib_dest_t *dest; |
a780a738 | 632 | struct fpm_mac_info_t *mac = NULL; |
5adc2528 | 633 | |
d62a17ae | 634 | assert(zfpm_g->state == ZFPM_STATE_IDLE); |
5adc2528 | 635 | |
a780a738 AD |
636 | /* |
637 | * Delink and free all fpm_mac_info_t nodes | |
638 | * in the mac_q and fpm_mac_info_hash | |
639 | */ | |
640 | while ((mac = TAILQ_FIRST(&zfpm_g->mac_q)) != NULL) | |
641 | zfpm_mac_info_del(mac); | |
642 | ||
d62a17ae | 643 | iter = &zfpm_g->t_conn_down_state.iter; |
5adc2528 | 644 | |
d62a17ae | 645 | while ((rnode = zfpm_rnodes_iter_next(iter))) { |
646 | dest = rib_dest_from_rnode(rnode); | |
5adc2528 | 647 | |
d62a17ae | 648 | if (dest) { |
649 | if (CHECK_FLAG(dest->flags, RIB_DEST_UPDATE_FPM)) { | |
650 | TAILQ_REMOVE(&zfpm_g->dest_q, dest, | |
651 | fpm_q_entries); | |
652 | } | |
653 | ||
654 | UNSET_FLAG(dest->flags, RIB_DEST_UPDATE_FPM); | |
655 | UNSET_FLAG(dest->flags, RIB_DEST_SENT_TO_FPM); | |
5adc2528 | 656 | |
d62a17ae | 657 | zfpm_g->stats.t_conn_down_dests_processed++; |
5adc2528 | 658 | |
d62a17ae | 659 | /* |
660 | * Check if the dest should be deleted. | |
661 | */ | |
662 | rib_gc_dest(rnode); | |
663 | } | |
5adc2528 | 664 | |
d62a17ae | 665 | /* |
666 | * Yield if need be. | |
667 | */ | |
668 | if (!zfpm_thread_should_yield(thread)) | |
669 | continue; | |
670 | ||
671 | zfpm_g->stats.t_conn_down_yields++; | |
672 | zfpm_rnodes_iter_pause(iter); | |
907a2395 DS |
673 | event_add_timer_msec(zfpm_g->master, zfpm_conn_down_thread_cb, |
674 | NULL, 0, &zfpm_g->t_conn_down); | |
cc9f21da | 675 | return; |
5adc2528 AS |
676 | } |
677 | ||
d62a17ae | 678 | zfpm_g->stats.t_conn_down_finishes++; |
679 | zfpm_rnodes_iter_cleanup(iter); | |
680 | ||
681 | /* | |
682 | * Start the process of connecting to the FPM again. | |
683 | */ | |
684 | zfpm_start_connect_timer("cleanup complete"); | |
5adc2528 AS |
685 | } |
686 | ||
687 | /* | |
688 | * zfpm_connection_down | |
689 | * | |
690 | * Called when the connection to the FPM has gone down. | |
691 | */ | |
d62a17ae | 692 | static void zfpm_connection_down(const char *detail) |
5adc2528 | 693 | { |
d62a17ae | 694 | if (!detail) |
695 | detail = "unknown"; | |
5adc2528 | 696 | |
d62a17ae | 697 | assert(zfpm_g->state == ZFPM_STATE_ESTABLISHED); |
5adc2528 | 698 | |
d62a17ae | 699 | zlog_info("connection to the FPM has gone down: %s", detail); |
5adc2528 | 700 | |
d62a17ae | 701 | zfpm_read_off(); |
702 | zfpm_write_off(); | |
5adc2528 | 703 | |
d62a17ae | 704 | stream_reset(zfpm_g->ibuf); |
705 | stream_reset(zfpm_g->obuf); | |
5adc2528 | 706 | |
d62a17ae | 707 | if (zfpm_g->sock >= 0) { |
708 | close(zfpm_g->sock); | |
709 | zfpm_g->sock = -1; | |
710 | } | |
5adc2528 | 711 | |
d62a17ae | 712 | /* |
713 | * Start thread to clean up state after the connection goes down. | |
714 | */ | |
715 | assert(!zfpm_g->t_conn_down); | |
d62a17ae | 716 | zfpm_rnodes_iter_init(&zfpm_g->t_conn_down_state.iter); |
54033432 | 717 | zfpm_conn_down_off(); |
907a2395 DS |
718 | event_add_timer_msec(zfpm_g->master, zfpm_conn_down_thread_cb, NULL, 0, |
719 | &zfpm_g->t_conn_down); | |
d62a17ae | 720 | zfpm_g->stats.t_conn_down_starts++; |
721 | ||
722 | zfpm_set_state(ZFPM_STATE_IDLE, detail); | |
5adc2528 AS |
723 | } |
724 | ||
725 | /* | |
726 | * zfpm_read_cb | |
727 | */ | |
e6685141 | 728 | static void zfpm_read_cb(struct event *thread) |
5adc2528 | 729 | { |
d62a17ae | 730 | size_t already; |
731 | struct stream *ibuf; | |
732 | uint16_t msg_len; | |
733 | fpm_msg_hdr_t *hdr; | |
734 | ||
735 | zfpm_g->stats.read_cb_calls++; | |
d62a17ae | 736 | |
737 | /* | |
738 | * Check if async connect is now done. | |
739 | */ | |
740 | if (zfpm_g->state == ZFPM_STATE_CONNECTING) { | |
741 | zfpm_connect_check(); | |
cc9f21da | 742 | return; |
5adc2528 AS |
743 | } |
744 | ||
d62a17ae | 745 | assert(zfpm_g->state == ZFPM_STATE_ESTABLISHED); |
746 | assert(zfpm_g->sock >= 0); | |
5adc2528 | 747 | |
d62a17ae | 748 | ibuf = zfpm_g->ibuf; |
5adc2528 | 749 | |
d62a17ae | 750 | already = stream_get_endp(ibuf); |
751 | if (already < FPM_MSG_HDR_LEN) { | |
752 | ssize_t nbyte; | |
5adc2528 | 753 | |
d62a17ae | 754 | nbyte = stream_read_try(ibuf, zfpm_g->sock, |
755 | FPM_MSG_HDR_LEN - already); | |
756 | if (nbyte == 0 || nbyte == -1) { | |
677f704d DS |
757 | if (nbyte == -1) { |
758 | char buffer[1024]; | |
759 | ||
772270f3 QY |
760 | snprintf(buffer, sizeof(buffer), |
761 | "closed socket in read(%d): %s", errno, | |
762 | safe_strerror(errno)); | |
677f704d | 763 | zfpm_connection_down(buffer); |
996c9314 | 764 | } else |
677f704d | 765 | zfpm_connection_down("closed socket in read"); |
cc9f21da | 766 | return; |
d62a17ae | 767 | } |
5adc2528 | 768 | |
d62a17ae | 769 | if (nbyte != (ssize_t)(FPM_MSG_HDR_LEN - already)) |
770 | goto done; | |
5adc2528 | 771 | |
d62a17ae | 772 | already = FPM_MSG_HDR_LEN; |
773 | } | |
5adc2528 | 774 | |
d62a17ae | 775 | stream_set_getp(ibuf, 0); |
5adc2528 | 776 | |
d62a17ae | 777 | hdr = (fpm_msg_hdr_t *)stream_pnt(ibuf); |
5adc2528 | 778 | |
d62a17ae | 779 | if (!fpm_msg_hdr_ok(hdr)) { |
780 | zfpm_connection_down("invalid message header"); | |
cc9f21da | 781 | return; |
5adc2528 AS |
782 | } |
783 | ||
d62a17ae | 784 | msg_len = fpm_msg_len(hdr); |
5adc2528 | 785 | |
d62a17ae | 786 | /* |
787 | * Read out the rest of the packet. | |
788 | */ | |
789 | if (already < msg_len) { | |
790 | ssize_t nbyte; | |
5adc2528 | 791 | |
d62a17ae | 792 | nbyte = stream_read_try(ibuf, zfpm_g->sock, msg_len - already); |
5adc2528 | 793 | |
d62a17ae | 794 | if (nbyte == 0 || nbyte == -1) { |
677f704d DS |
795 | if (nbyte == -1) { |
796 | char buffer[1024]; | |
797 | ||
772270f3 QY |
798 | snprintf(buffer, sizeof(buffer), |
799 | "failed to read message(%d) %s", errno, | |
800 | safe_strerror(errno)); | |
677f704d | 801 | zfpm_connection_down(buffer); |
996c9314 | 802 | } else |
677f704d | 803 | zfpm_connection_down("failed to read message"); |
cc9f21da | 804 | return; |
d62a17ae | 805 | } |
806 | ||
807 | if (nbyte != (ssize_t)(msg_len - already)) | |
808 | goto done; | |
809 | } | |
810 | ||
d62a17ae | 811 | /* |
812 | * Just throw it away for now. | |
813 | */ | |
814 | stream_reset(ibuf); | |
815 | ||
816 | done: | |
817 | zfpm_read_on(); | |
5adc2528 AS |
818 | } |
819 | ||
21d814eb AD |
820 | static bool zfpm_updates_pending(void) |
821 | { | |
822 | if (!(TAILQ_EMPTY(&zfpm_g->dest_q)) || !(TAILQ_EMPTY(&zfpm_g->mac_q))) | |
823 | return true; | |
824 | ||
825 | return false; | |
826 | } | |
827 | ||
5adc2528 AS |
828 | /* |
829 | * zfpm_writes_pending | |
830 | * | |
2951a7a4 | 831 | * Returns true if we may have something to write to the FPM. |
5adc2528 | 832 | */ |
d62a17ae | 833 | static int zfpm_writes_pending(void) |
5adc2528 AS |
834 | { |
835 | ||
d62a17ae | 836 | /* |
837 | * Check if there is any data in the outbound buffer that has not | |
838 | * been written to the socket yet. | |
839 | */ | |
840 | if (stream_get_endp(zfpm_g->obuf) - stream_get_getp(zfpm_g->obuf)) | |
841 | return 1; | |
5adc2528 | 842 | |
d62a17ae | 843 | /* |
21d814eb | 844 | * Check if there are any updates scheduled on the outbound queues. |
d62a17ae | 845 | */ |
21d814eb | 846 | if (zfpm_updates_pending()) |
d62a17ae | 847 | return 1; |
5adc2528 | 848 | |
d62a17ae | 849 | return 0; |
5adc2528 AS |
850 | } |
851 | ||
852 | /* | |
853 | * zfpm_encode_route | |
854 | * | |
855 | * Encode a message to the FPM with information about the given route. | |
856 | * | |
857 | * Returns the number of bytes written to the buffer. 0 or a negative | |
858 | * value indicates an error. | |
859 | */ | |
d62a17ae | 860 | static inline int zfpm_encode_route(rib_dest_t *dest, struct route_entry *re, |
861 | char *in_buf, size_t in_buf_len, | |
862 | fpm_msg_type_e *msg_type) | |
5adc2528 | 863 | { |
d62a17ae | 864 | size_t len; |
9bf75362 | 865 | #ifdef HAVE_NETLINK |
d62a17ae | 866 | int cmd; |
9bf75362 | 867 | #endif |
d62a17ae | 868 | len = 0; |
5adc2528 | 869 | |
d62a17ae | 870 | *msg_type = FPM_MSG_TYPE_NONE; |
5adc2528 | 871 | |
d62a17ae | 872 | switch (zfpm_g->message_format) { |
5adc2528 | 873 | |
d62a17ae | 874 | case ZFPM_MSG_FORMAT_PROTOBUF: |
fb0aa886 | 875 | #ifdef HAVE_PROTOBUF |
d62a17ae | 876 | len = zfpm_protobuf_encode_route(dest, re, (uint8_t *)in_buf, |
877 | in_buf_len); | |
878 | *msg_type = FPM_MSG_TYPE_PROTOBUF; | |
fb0aa886 | 879 | #endif |
d62a17ae | 880 | break; |
5adc2528 | 881 | |
d62a17ae | 882 | case ZFPM_MSG_FORMAT_NETLINK: |
fb0aa886 | 883 | #ifdef HAVE_NETLINK |
d62a17ae | 884 | *msg_type = FPM_MSG_TYPE_NETLINK; |
885 | cmd = re ? RTM_NEWROUTE : RTM_DELROUTE; | |
886 | len = zfpm_netlink_encode_route(cmd, dest, re, in_buf, | |
887 | in_buf_len); | |
888 | assert(fpm_msg_align(len) == len); | |
5adc2528 | 889 | #endif /* HAVE_NETLINK */ |
d62a17ae | 890 | break; |
fb0aa886 | 891 | |
a98701f0 | 892 | case ZFPM_MSG_FORMAT_NONE: |
d62a17ae | 893 | break; |
894 | } | |
fb0aa886 | 895 | |
d62a17ae | 896 | return len; |
5adc2528 AS |
897 | } |
898 | ||
899 | /* | |
900 | * zfpm_route_for_update | |
901 | * | |
f0f77c9a | 902 | * Returns the re that is to be sent to the FPM for a given dest. |
5adc2528 | 903 | */ |
d62a17ae | 904 | struct route_entry *zfpm_route_for_update(rib_dest_t *dest) |
5adc2528 | 905 | { |
5f7a4718 | 906 | return dest->selected_fib; |
5adc2528 AS |
907 | } |
908 | ||
909 | /* | |
21d814eb | 910 | * Define an enum for return codes for queue processing functions |
5adc2528 | 911 | * |
21d814eb AD |
912 | * FPM_WRITE_STOP: This return code indicates that the write buffer is full. |
913 | * Stop processing all the queues and empty the buffer by writing its content | |
914 | * to the socket. | |
915 | * | |
916 | * FPM_GOTO_NEXT_Q: This return code indicates that either this queue is | |
917 | * empty or we have processed enough updates from this queue. | |
918 | * So, move on to the next queue. | |
5adc2528 | 919 | */ |
21d814eb AD |
920 | enum { |
921 | FPM_WRITE_STOP = 0, | |
922 | FPM_GOTO_NEXT_Q = 1 | |
923 | }; | |
924 | ||
925 | #define FPM_QUEUE_PROCESS_LIMIT 10000 | |
926 | ||
927 | /* | |
928 | * zfpm_build_route_updates | |
929 | * | |
930 | * Process the dest_q queue and write FPM messages to the outbound buffer. | |
931 | */ | |
932 | static int zfpm_build_route_updates(void) | |
5adc2528 | 933 | { |
d62a17ae | 934 | struct stream *s; |
935 | rib_dest_t *dest; | |
936 | unsigned char *buf, *data, *buf_end; | |
937 | size_t msg_len; | |
938 | size_t data_len; | |
939 | fpm_msg_hdr_t *hdr; | |
940 | struct route_entry *re; | |
941 | int is_add, write_msg; | |
942 | fpm_msg_type_e msg_type; | |
21d814eb | 943 | uint16_t q_limit; |
d62a17ae | 944 | |
21d814eb AD |
945 | if (TAILQ_EMPTY(&zfpm_g->dest_q)) |
946 | return FPM_GOTO_NEXT_Q; | |
d62a17ae | 947 | |
21d814eb AD |
948 | s = zfpm_g->obuf; |
949 | q_limit = FPM_QUEUE_PROCESS_LIMIT; | |
d62a17ae | 950 | |
21d814eb | 951 | do { |
d62a17ae | 952 | /* |
953 | * Make sure there is enough space to write another message. | |
954 | */ | |
955 | if (STREAM_WRITEABLE(s) < FPM_MAX_MSG_LEN) | |
21d814eb | 956 | return FPM_WRITE_STOP; |
d62a17ae | 957 | |
958 | buf = STREAM_DATA(s) + stream_get_endp(s); | |
959 | buf_end = buf + STREAM_WRITEABLE(s); | |
960 | ||
961 | dest = TAILQ_FIRST(&zfpm_g->dest_q); | |
962 | if (!dest) | |
21d814eb | 963 | return FPM_GOTO_NEXT_Q; |
d62a17ae | 964 | |
965 | assert(CHECK_FLAG(dest->flags, RIB_DEST_UPDATE_FPM)); | |
966 | ||
967 | hdr = (fpm_msg_hdr_t *)buf; | |
968 | hdr->version = FPM_PROTO_VERSION; | |
969 | ||
970 | data = fpm_msg_data(hdr); | |
971 | ||
972 | re = zfpm_route_for_update(dest); | |
973 | is_add = re ? 1 : 0; | |
974 | ||
975 | write_msg = 1; | |
976 | ||
977 | /* | |
978 | * If this is a route deletion, and we have not sent the route | |
979 | * to | |
980 | * the FPM previously, skip it. | |
981 | */ | |
982 | if (!is_add && !CHECK_FLAG(dest->flags, RIB_DEST_SENT_TO_FPM)) { | |
983 | write_msg = 0; | |
984 | zfpm_g->stats.nop_deletes_skipped++; | |
985 | } | |
986 | ||
987 | if (write_msg) { | |
988 | data_len = zfpm_encode_route(dest, re, (char *)data, | |
989 | buf_end - data, &msg_type); | |
990 | ||
d62a17ae | 991 | if (data_len) { |
992 | hdr->msg_type = msg_type; | |
993 | msg_len = fpm_data_len_to_msg_len(data_len); | |
994 | hdr->msg_len = htons(msg_len); | |
995 | stream_forward_endp(s, msg_len); | |
996 | ||
997 | if (is_add) | |
998 | zfpm_g->stats.route_adds++; | |
999 | else | |
1000 | zfpm_g->stats.route_dels++; | |
5306e6cf | 1001 | } else { |
1002 | zlog_err("%s: Encoding Prefix: %pRN No valid nexthops", | |
1003 | __func__, dest->rnode); | |
d62a17ae | 1004 | } |
1005 | } | |
1006 | ||
1007 | /* | |
1008 | * Remove the dest from the queue, and reset the flag. | |
1009 | */ | |
1010 | UNSET_FLAG(dest->flags, RIB_DEST_UPDATE_FPM); | |
1011 | TAILQ_REMOVE(&zfpm_g->dest_q, dest, fpm_q_entries); | |
1012 | ||
1013 | if (is_add) { | |
1014 | SET_FLAG(dest->flags, RIB_DEST_SENT_TO_FPM); | |
1015 | } else { | |
1016 | UNSET_FLAG(dest->flags, RIB_DEST_SENT_TO_FPM); | |
1017 | } | |
1018 | ||
1019 | /* | |
1020 | * Delete the destination if necessary. | |
1021 | */ | |
1022 | if (rib_gc_dest(dest->rnode)) | |
1023 | zfpm_g->stats.dests_del_after_update++; | |
1024 | ||
21d814eb AD |
1025 | q_limit--; |
1026 | if (q_limit == 0) { | |
1027 | /* | |
1028 | * We have processed enough updates in this queue. | |
1029 | * Now yield for other queues. | |
1030 | */ | |
1031 | return FPM_GOTO_NEXT_Q; | |
1032 | } | |
c5431822 | 1033 | } while (true); |
21d814eb AD |
1034 | } |
1035 | ||
1036 | /* | |
1037 | * zfpm_encode_mac | |
1038 | * | |
1039 | * Encode a message to FPM with information about the given MAC. | |
1040 | * | |
1041 | * Returns the number of bytes written to the buffer. | |
1042 | */ | |
1043 | static inline int zfpm_encode_mac(struct fpm_mac_info_t *mac, char *in_buf, | |
1044 | size_t in_buf_len, fpm_msg_type_e *msg_type) | |
1045 | { | |
1046 | size_t len = 0; | |
1047 | ||
1048 | *msg_type = FPM_MSG_TYPE_NONE; | |
1049 | ||
1050 | switch (zfpm_g->message_format) { | |
1051 | ||
1052 | case ZFPM_MSG_FORMAT_NONE: | |
1053 | break; | |
1054 | case ZFPM_MSG_FORMAT_NETLINK: | |
9da60d0a AD |
1055 | #ifdef HAVE_NETLINK |
1056 | len = zfpm_netlink_encode_mac(mac, in_buf, in_buf_len); | |
1057 | assert(fpm_msg_align(len) == len); | |
1058 | *msg_type = FPM_MSG_TYPE_NETLINK; | |
1059 | #endif /* HAVE_NETLINK */ | |
21d814eb AD |
1060 | break; |
1061 | case ZFPM_MSG_FORMAT_PROTOBUF: | |
1062 | break; | |
1063 | } | |
1064 | return len; | |
1065 | } | |
1066 | ||
1067 | static int zfpm_build_mac_updates(void) | |
1068 | { | |
1069 | struct stream *s; | |
1070 | struct fpm_mac_info_t *mac; | |
1071 | unsigned char *buf, *data, *buf_end; | |
1072 | fpm_msg_hdr_t *hdr; | |
1073 | size_t data_len, msg_len; | |
1074 | fpm_msg_type_e msg_type; | |
1075 | uint16_t q_limit; | |
1076 | ||
1077 | if (TAILQ_EMPTY(&zfpm_g->mac_q)) | |
1078 | return FPM_GOTO_NEXT_Q; | |
1079 | ||
1080 | s = zfpm_g->obuf; | |
1081 | q_limit = FPM_QUEUE_PROCESS_LIMIT; | |
1082 | ||
1083 | do { | |
1084 | /* Make sure there is enough space to write another message. */ | |
1085 | if (STREAM_WRITEABLE(s) < FPM_MAX_MAC_MSG_LEN) | |
1086 | return FPM_WRITE_STOP; | |
1087 | ||
1088 | buf = STREAM_DATA(s) + stream_get_endp(s); | |
1089 | buf_end = buf + STREAM_WRITEABLE(s); | |
1090 | ||
1091 | mac = TAILQ_FIRST(&zfpm_g->mac_q); | |
1092 | if (!mac) | |
1093 | return FPM_GOTO_NEXT_Q; | |
1094 | ||
1095 | /* Check for no-op */ | |
1096 | if (!CHECK_FLAG(mac->fpm_flags, ZEBRA_MAC_UPDATE_FPM)) { | |
1097 | zfpm_g->stats.nop_deletes_skipped++; | |
1098 | zfpm_mac_info_del(mac); | |
1099 | continue; | |
1100 | } | |
1101 | ||
1102 | hdr = (fpm_msg_hdr_t *)buf; | |
1103 | hdr->version = FPM_PROTO_VERSION; | |
1104 | ||
1105 | data = fpm_msg_data(hdr); | |
1106 | data_len = zfpm_encode_mac(mac, (char *)data, buf_end - data, | |
1107 | &msg_type); | |
9da60d0a | 1108 | assert(data_len); |
21d814eb AD |
1109 | |
1110 | hdr->msg_type = msg_type; | |
1111 | msg_len = fpm_data_len_to_msg_len(data_len); | |
1112 | hdr->msg_len = htons(msg_len); | |
1113 | stream_forward_endp(s, msg_len); | |
1114 | ||
1115 | /* Remove the MAC from the queue, and delete it. */ | |
1116 | zfpm_mac_info_del(mac); | |
1117 | ||
1118 | q_limit--; | |
1119 | if (q_limit == 0) { | |
1120 | /* | |
1121 | * We have processed enough updates in this queue. | |
1122 | * Now yield for other queues. | |
1123 | */ | |
1124 | return FPM_GOTO_NEXT_Q; | |
1125 | } | |
d62a17ae | 1126 | } while (1); |
5adc2528 AS |
1127 | } |
1128 | ||
21d814eb AD |
1129 | /* |
1130 | * zfpm_build_updates | |
1131 | * | |
1132 | * Process the outgoing queues and write messages to the outbound | |
1133 | * buffer. | |
1134 | */ | |
1135 | static void zfpm_build_updates(void) | |
1136 | { | |
1137 | struct stream *s; | |
1138 | ||
1139 | s = zfpm_g->obuf; | |
1140 | assert(stream_empty(s)); | |
1141 | ||
1142 | do { | |
1143 | /* | |
1144 | * Stop processing the queues if zfpm_g->obuf is full | |
1145 | * or we do not have more updates to process | |
1146 | */ | |
1147 | if (zfpm_build_mac_updates() == FPM_WRITE_STOP) | |
1148 | break; | |
1149 | if (zfpm_build_route_updates() == FPM_WRITE_STOP) | |
1150 | break; | |
1151 | } while (zfpm_updates_pending()); | |
1152 | } | |
1153 | ||
5adc2528 AS |
1154 | /* |
1155 | * zfpm_write_cb | |
1156 | */ | |
e6685141 | 1157 | static void zfpm_write_cb(struct event *thread) |
5adc2528 | 1158 | { |
d62a17ae | 1159 | struct stream *s; |
1160 | int num_writes; | |
1161 | ||
1162 | zfpm_g->stats.write_cb_calls++; | |
d62a17ae | 1163 | |
1164 | /* | |
1165 | * Check if async connect is now done. | |
1166 | */ | |
1167 | if (zfpm_g->state == ZFPM_STATE_CONNECTING) { | |
1168 | zfpm_connect_check(); | |
cc9f21da | 1169 | return; |
d62a17ae | 1170 | } |
5adc2528 | 1171 | |
d62a17ae | 1172 | assert(zfpm_g->state == ZFPM_STATE_ESTABLISHED); |
1173 | assert(zfpm_g->sock >= 0); | |
5adc2528 | 1174 | |
d62a17ae | 1175 | num_writes = 0; |
5adc2528 | 1176 | |
d62a17ae | 1177 | do { |
1178 | int bytes_to_write, bytes_written; | |
5adc2528 | 1179 | |
d62a17ae | 1180 | s = zfpm_g->obuf; |
5adc2528 | 1181 | |
d62a17ae | 1182 | /* |
1183 | * If the stream is empty, try fill it up with data. | |
1184 | */ | |
1185 | if (stream_empty(s)) { | |
1186 | zfpm_build_updates(); | |
1187 | } | |
5adc2528 | 1188 | |
d62a17ae | 1189 | bytes_to_write = stream_get_endp(s) - stream_get_getp(s); |
1190 | if (!bytes_to_write) | |
1191 | break; | |
5adc2528 | 1192 | |
d62a17ae | 1193 | bytes_written = |
2d34fb80 | 1194 | write(zfpm_g->sock, stream_pnt(s), bytes_to_write); |
d62a17ae | 1195 | zfpm_g->stats.write_calls++; |
1196 | num_writes++; | |
5adc2528 | 1197 | |
d62a17ae | 1198 | if (bytes_written < 0) { |
1199 | if (ERRNO_IO_RETRY(errno)) | |
1200 | break; | |
5adc2528 | 1201 | |
d62a17ae | 1202 | zfpm_connection_down("failed to write to socket"); |
cc9f21da | 1203 | return; |
d62a17ae | 1204 | } |
5adc2528 | 1205 | |
d62a17ae | 1206 | if (bytes_written != bytes_to_write) { |
5adc2528 | 1207 | |
d62a17ae | 1208 | /* |
1209 | * Partial write. | |
1210 | */ | |
1211 | stream_forward_getp(s, bytes_written); | |
1212 | zfpm_g->stats.partial_writes++; | |
1213 | break; | |
1214 | } | |
5adc2528 | 1215 | |
d62a17ae | 1216 | /* |
1217 | * We've written out the entire contents of the stream. | |
1218 | */ | |
1219 | stream_reset(s); | |
5adc2528 | 1220 | |
d62a17ae | 1221 | if (num_writes >= ZFPM_MAX_WRITES_PER_RUN) { |
1222 | zfpm_g->stats.max_writes_hit++; | |
1223 | break; | |
1224 | } | |
5adc2528 | 1225 | |
d62a17ae | 1226 | if (zfpm_thread_should_yield(thread)) { |
1227 | zfpm_g->stats.t_write_yields++; | |
1228 | break; | |
1229 | } | |
1230 | } while (1); | |
5adc2528 | 1231 | |
d62a17ae | 1232 | if (zfpm_writes_pending()) |
1233 | zfpm_write_on(); | |
5adc2528 AS |
1234 | } |
1235 | ||
1236 | /* | |
1237 | * zfpm_connect_cb | |
1238 | */ | |
e6685141 | 1239 | static void zfpm_connect_cb(struct event *t) |
5adc2528 | 1240 | { |
d62a17ae | 1241 | int sock, ret; |
1242 | struct sockaddr_in serv; | |
1243 | ||
d62a17ae | 1244 | assert(zfpm_g->state == ZFPM_STATE_ACTIVE); |
1245 | ||
1246 | sock = socket(AF_INET, SOCK_STREAM, 0); | |
1247 | if (sock < 0) { | |
14a4d9d0 | 1248 | zlog_err("Failed to create socket for connect(): %s", |
d62a17ae | 1249 | strerror(errno)); |
1250 | zfpm_g->stats.connect_no_sock++; | |
cc9f21da | 1251 | return; |
d62a17ae | 1252 | } |
1253 | ||
1254 | set_nonblocking(sock); | |
1255 | ||
1256 | /* Make server socket. */ | |
1257 | memset(&serv, 0, sizeof(serv)); | |
1258 | serv.sin_family = AF_INET; | |
1259 | serv.sin_port = htons(zfpm_g->fpm_port); | |
5adc2528 | 1260 | #ifdef HAVE_STRUCT_SOCKADDR_IN_SIN_LEN |
d62a17ae | 1261 | serv.sin_len = sizeof(struct sockaddr_in); |
5adc2528 | 1262 | #endif /* HAVE_STRUCT_SOCKADDR_IN_SIN_LEN */ |
d62a17ae | 1263 | if (!zfpm_g->fpm_server) |
1264 | serv.sin_addr.s_addr = htonl(INADDR_LOOPBACK); | |
1265 | else | |
1266 | serv.sin_addr.s_addr = (zfpm_g->fpm_server); | |
1267 | ||
1268 | /* | |
1269 | * Connect to the FPM. | |
1270 | */ | |
1271 | zfpm_g->connect_calls++; | |
1272 | zfpm_g->stats.connect_calls++; | |
1273 | zfpm_g->last_connect_call_time = monotime(NULL); | |
1274 | ||
1275 | ret = connect(sock, (struct sockaddr *)&serv, sizeof(serv)); | |
1276 | if (ret >= 0) { | |
1277 | zfpm_g->sock = sock; | |
1278 | zfpm_connection_up("connect succeeded"); | |
cc9f21da | 1279 | return; |
d62a17ae | 1280 | } |
1281 | ||
1282 | if (errno == EINPROGRESS) { | |
1283 | zfpm_g->sock = sock; | |
1284 | zfpm_read_on(); | |
1285 | zfpm_write_on(); | |
1286 | zfpm_set_state(ZFPM_STATE_CONNECTING, | |
1287 | "async connect in progress"); | |
cc9f21da | 1288 | return; |
d62a17ae | 1289 | } |
1290 | ||
1291 | zlog_info("can't connect to FPM %d: %s", sock, safe_strerror(errno)); | |
1292 | close(sock); | |
1293 | ||
1294 | /* | |
1295 | * Restart timer for retrying connection. | |
1296 | */ | |
1297 | zfpm_start_connect_timer("connect() failed"); | |
5adc2528 AS |
1298 | } |
1299 | ||
1300 | /* | |
1301 | * zfpm_set_state | |
1302 | * | |
1303 | * Move state machine into the given state. | |
1304 | */ | |
1d6a3ee8 | 1305 | static void zfpm_set_state(enum zfpm_state state, const char *reason) |
5adc2528 | 1306 | { |
1d6a3ee8 | 1307 | enum zfpm_state cur_state = zfpm_g->state; |
d62a17ae | 1308 | |
1309 | if (!reason) | |
1310 | reason = "Unknown"; | |
1311 | ||
1312 | if (state == cur_state) | |
1313 | return; | |
1314 | ||
1315 | zfpm_debug("beginning state transition %s -> %s. Reason: %s", | |
1316 | zfpm_state_to_str(cur_state), zfpm_state_to_str(state), | |
1317 | reason); | |
1318 | ||
1319 | switch (state) { | |
1320 | ||
1321 | case ZFPM_STATE_IDLE: | |
1322 | assert(cur_state == ZFPM_STATE_ESTABLISHED); | |
1323 | break; | |
1324 | ||
1325 | case ZFPM_STATE_ACTIVE: | |
1326 | assert(cur_state == ZFPM_STATE_IDLE | |
1327 | || cur_state == ZFPM_STATE_CONNECTING); | |
1328 | assert(zfpm_g->t_connect); | |
1329 | break; | |
1330 | ||
1331 | case ZFPM_STATE_CONNECTING: | |
1332 | assert(zfpm_g->sock); | |
1333 | assert(cur_state == ZFPM_STATE_ACTIVE); | |
1334 | assert(zfpm_g->t_read); | |
1335 | assert(zfpm_g->t_write); | |
1336 | break; | |
1337 | ||
1338 | case ZFPM_STATE_ESTABLISHED: | |
1339 | assert(cur_state == ZFPM_STATE_ACTIVE | |
1340 | || cur_state == ZFPM_STATE_CONNECTING); | |
1341 | assert(zfpm_g->sock); | |
1342 | assert(zfpm_g->t_read); | |
1343 | assert(zfpm_g->t_write); | |
1344 | break; | |
1345 | } | |
1346 | ||
1347 | zfpm_g->state = state; | |
5adc2528 AS |
1348 | } |
1349 | ||
1350 | /* | |
1351 | * zfpm_calc_connect_delay | |
1352 | * | |
1353 | * Returns the number of seconds after which we should attempt to | |
1354 | * reconnect to the FPM. | |
1355 | */ | |
d62a17ae | 1356 | static long zfpm_calc_connect_delay(void) |
5adc2528 | 1357 | { |
d62a17ae | 1358 | time_t elapsed; |
5adc2528 | 1359 | |
d62a17ae | 1360 | /* |
1361 | * Return 0 if this is our first attempt to connect. | |
1362 | */ | |
1363 | if (zfpm_g->connect_calls == 0) { | |
1364 | return 0; | |
1365 | } | |
5adc2528 | 1366 | |
d62a17ae | 1367 | elapsed = zfpm_get_elapsed_time(zfpm_g->last_connect_call_time); |
5adc2528 | 1368 | |
d62a17ae | 1369 | if (elapsed > ZFPM_CONNECT_RETRY_IVL) { |
1370 | return 0; | |
1371 | } | |
5adc2528 | 1372 | |
d62a17ae | 1373 | return ZFPM_CONNECT_RETRY_IVL - elapsed; |
5adc2528 AS |
1374 | } |
1375 | ||
1376 | /* | |
1377 | * zfpm_start_connect_timer | |
1378 | */ | |
d62a17ae | 1379 | static void zfpm_start_connect_timer(const char *reason) |
5adc2528 | 1380 | { |
d62a17ae | 1381 | long delay_secs; |
5adc2528 | 1382 | |
d62a17ae | 1383 | assert(!zfpm_g->t_connect); |
1384 | assert(zfpm_g->sock < 0); | |
5adc2528 | 1385 | |
d62a17ae | 1386 | assert(zfpm_g->state == ZFPM_STATE_IDLE |
1387 | || zfpm_g->state == ZFPM_STATE_ACTIVE | |
1388 | || zfpm_g->state == ZFPM_STATE_CONNECTING); | |
5adc2528 | 1389 | |
d62a17ae | 1390 | delay_secs = zfpm_calc_connect_delay(); |
1391 | zfpm_debug("scheduling connect in %ld seconds", delay_secs); | |
5adc2528 | 1392 | |
907a2395 DS |
1393 | event_add_timer(zfpm_g->master, zfpm_connect_cb, 0, delay_secs, |
1394 | &zfpm_g->t_connect); | |
d62a17ae | 1395 | zfpm_set_state(ZFPM_STATE_ACTIVE, reason); |
5adc2528 AS |
1396 | } |
1397 | ||
1398 | /* | |
1399 | * zfpm_is_enabled | |
1400 | * | |
2951a7a4 | 1401 | * Returns true if the zebra FPM module has been enabled. |
5adc2528 | 1402 | */ |
d62a17ae | 1403 | static inline int zfpm_is_enabled(void) |
5adc2528 | 1404 | { |
d62a17ae | 1405 | return zfpm_g->enabled; |
5adc2528 AS |
1406 | } |
1407 | ||
1408 | /* | |
1409 | * zfpm_conn_is_up | |
1410 | * | |
2951a7a4 | 1411 | * Returns true if the connection to the FPM is up. |
5adc2528 | 1412 | */ |
d62a17ae | 1413 | static inline int zfpm_conn_is_up(void) |
5adc2528 | 1414 | { |
d62a17ae | 1415 | if (zfpm_g->state != ZFPM_STATE_ESTABLISHED) |
1416 | return 0; | |
5adc2528 | 1417 | |
d62a17ae | 1418 | assert(zfpm_g->sock >= 0); |
5adc2528 | 1419 | |
d62a17ae | 1420 | return 1; |
5adc2528 AS |
1421 | } |
1422 | ||
1423 | /* | |
1424 | * zfpm_trigger_update | |
1425 | * | |
1426 | * The zebra code invokes this function to indicate that we should | |
1427 | * send an update to the FPM about the given route_node. | |
1428 | */ | |
d62a17ae | 1429 | static int zfpm_trigger_update(struct route_node *rn, const char *reason) |
5adc2528 | 1430 | { |
d62a17ae | 1431 | rib_dest_t *dest; |
d62a17ae | 1432 | |
1433 | /* | |
1434 | * Ignore if the connection is down. We will update the FPM about | |
1435 | * all destinations once the connection comes up. | |
1436 | */ | |
1437 | if (!zfpm_conn_is_up()) | |
1438 | return 0; | |
1439 | ||
1440 | dest = rib_dest_from_rnode(rn); | |
1441 | ||
d62a17ae | 1442 | if (CHECK_FLAG(dest->flags, RIB_DEST_UPDATE_FPM)) { |
1443 | zfpm_g->stats.redundant_triggers++; | |
1444 | return 0; | |
1445 | } | |
1446 | ||
1447 | if (reason) { | |
2dbe669b DA |
1448 | zfpm_debug("%pFX triggering update to FPM - Reason: %s", &rn->p, |
1449 | reason); | |
d62a17ae | 1450 | } |
1451 | ||
1452 | SET_FLAG(dest->flags, RIB_DEST_UPDATE_FPM); | |
1453 | TAILQ_INSERT_TAIL(&zfpm_g->dest_q, dest, fpm_q_entries); | |
1454 | zfpm_g->stats.updates_triggered++; | |
1455 | ||
1456 | /* | |
1457 | * Make sure that writes are enabled. | |
1458 | */ | |
1459 | if (zfpm_g->t_write) | |
1460 | return 0; | |
1461 | ||
1462 | zfpm_write_on(); | |
1463 | return 0; | |
5adc2528 AS |
1464 | } |
1465 | ||
0d0f516c | 1466 | /* |
1467 | * zfpm_trigger_remove | |
1468 | * | |
1469 | * The zebra code invokes this function to indicate that we should | |
1470 | * send an remove to the FPM about the given route_node. | |
1471 | */ | |
1472 | ||
1473 | static int zfpm_trigger_remove(struct route_node *rn) | |
1474 | { | |
1475 | rib_dest_t *dest; | |
1476 | ||
1477 | if (!zfpm_conn_is_up()) | |
1478 | return 0; | |
1479 | ||
1480 | dest = rib_dest_from_rnode(rn); | |
1481 | if (!CHECK_FLAG(dest->flags, RIB_DEST_UPDATE_FPM)) | |
1482 | return 0; | |
1483 | ||
1484 | zfpm_debug("%pRN Removing from update queue shutting down", rn); | |
1485 | ||
1486 | UNSET_FLAG(dest->flags, RIB_DEST_UPDATE_FPM); | |
1487 | TAILQ_REMOVE(&zfpm_g->dest_q, dest, fpm_q_entries); | |
1488 | ||
1489 | return 0; | |
1490 | } | |
1491 | ||
e5218ec8 AD |
1492 | /* |
1493 | * Generate Key for FPM MAC info hash entry | |
e5218ec8 AD |
1494 | */ |
1495 | static unsigned int zfpm_mac_info_hash_keymake(const void *p) | |
1496 | { | |
1497 | struct fpm_mac_info_t *fpm_mac = (struct fpm_mac_info_t *)p; | |
1498 | uint32_t mac_key; | |
1499 | ||
1500 | mac_key = jhash(fpm_mac->macaddr.octet, ETH_ALEN, 0xa5a5a55a); | |
1501 | ||
1502 | return jhash_2words(mac_key, fpm_mac->vni, 0); | |
1503 | } | |
1504 | ||
1505 | /* | |
1506 | * Compare function for FPM MAC info hash lookup | |
1507 | */ | |
1508 | static bool zfpm_mac_info_cmp(const void *p1, const void *p2) | |
1509 | { | |
1510 | const struct fpm_mac_info_t *fpm_mac1 = p1; | |
1511 | const struct fpm_mac_info_t *fpm_mac2 = p2; | |
1512 | ||
1513 | if (memcmp(fpm_mac1->macaddr.octet, fpm_mac2->macaddr.octet, ETH_ALEN) | |
1514 | != 0) | |
1515 | return false; | |
e5218ec8 AD |
1516 | if (fpm_mac1->vni != fpm_mac2->vni) |
1517 | return false; | |
1518 | ||
1519 | return true; | |
1520 | } | |
1521 | ||
1522 | /* | |
1523 | * Lookup FPM MAC info hash entry. | |
1524 | */ | |
1525 | static struct fpm_mac_info_t *zfpm_mac_info_lookup(struct fpm_mac_info_t *key) | |
1526 | { | |
1527 | return hash_lookup(zfpm_g->fpm_mac_info_table, key); | |
1528 | } | |
1529 | ||
1530 | /* | |
1531 | * Callback to allocate fpm_mac_info_t structure. | |
1532 | */ | |
1533 | static void *zfpm_mac_info_alloc(void *p) | |
1534 | { | |
1535 | const struct fpm_mac_info_t *key = p; | |
1536 | struct fpm_mac_info_t *fpm_mac; | |
1537 | ||
1538 | fpm_mac = XCALLOC(MTYPE_FPM_MAC_INFO, sizeof(struct fpm_mac_info_t)); | |
1539 | ||
1540 | memcpy(&fpm_mac->macaddr, &key->macaddr, ETH_ALEN); | |
e5218ec8 AD |
1541 | fpm_mac->vni = key->vni; |
1542 | ||
1543 | return (void *)fpm_mac; | |
1544 | } | |
1545 | ||
1546 | /* | |
1547 | * Delink and free fpm_mac_info_t. | |
1548 | */ | |
1549 | static void zfpm_mac_info_del(struct fpm_mac_info_t *fpm_mac) | |
1550 | { | |
1551 | hash_release(zfpm_g->fpm_mac_info_table, fpm_mac); | |
1552 | TAILQ_REMOVE(&zfpm_g->mac_q, fpm_mac, fpm_mac_q_entries); | |
1553 | XFREE(MTYPE_FPM_MAC_INFO, fpm_mac); | |
1554 | } | |
1555 | ||
a780a738 AD |
1556 | /* |
1557 | * zfpm_trigger_rmac_update | |
1558 | * | |
1559 | * Zebra code invokes this function to indicate that we should | |
1560 | * send an update to FPM for given MAC entry. | |
1561 | * | |
1562 | * This function checks if we already have enqueued an update for this RMAC, | |
1563 | * If yes, update the same fpm_mac_info_t. Else, create and enqueue an update. | |
1564 | */ | |
3198b2b3 | 1565 | static int zfpm_trigger_rmac_update(struct zebra_mac *rmac, |
05843a27 | 1566 | struct zebra_l3vni *zl3vni, bool delete, |
3198b2b3 | 1567 | const char *reason) |
a780a738 | 1568 | { |
a780a738 AD |
1569 | struct fpm_mac_info_t *fpm_mac, key; |
1570 | struct interface *vxlan_if, *svi_if; | |
44f7f132 | 1571 | bool mac_found = false; |
a780a738 AD |
1572 | |
1573 | /* | |
1574 | * Ignore if the connection is down. We will update the FPM about | |
1575 | * all destinations once the connection comes up. | |
1576 | */ | |
1577 | if (!zfpm_conn_is_up()) | |
1578 | return 0; | |
1579 | ||
1580 | if (reason) { | |
5e9f9adb DL |
1581 | zfpm_debug("triggering update to FPM - Reason: %s - %pEA", |
1582 | reason, &rmac->macaddr); | |
a780a738 AD |
1583 | } |
1584 | ||
1585 | vxlan_if = zl3vni_map_to_vxlan_if(zl3vni); | |
1586 | svi_if = zl3vni_map_to_svi_if(zl3vni); | |
1587 | ||
6006b807 | 1588 | memset(&key, 0, sizeof(key)); |
a780a738 AD |
1589 | |
1590 | memcpy(&key.macaddr, &rmac->macaddr, ETH_ALEN); | |
a780a738 AD |
1591 | key.vni = zl3vni->vni; |
1592 | ||
1593 | /* Check if this MAC is already present in the queue. */ | |
1594 | fpm_mac = zfpm_mac_info_lookup(&key); | |
1595 | ||
1596 | if (fpm_mac) { | |
44f7f132 | 1597 | mac_found = true; |
a780a738 AD |
1598 | |
1599 | /* | |
44f7f132 AD |
1600 | * If the enqueued op is "add" and current op is "delete", |
1601 | * this is a noop. So, Unset ZEBRA_MAC_UPDATE_FPM flag. | |
1602 | * While processing FPM queue, we will silently delete this | |
1603 | * MAC entry without sending any update for this MAC. | |
a780a738 | 1604 | */ |
44f7f132 AD |
1605 | if (!CHECK_FLAG(fpm_mac->fpm_flags, ZEBRA_MAC_DELETE_FPM) && |
1606 | delete == 1) { | |
a780a738 AD |
1607 | SET_FLAG(fpm_mac->fpm_flags, ZEBRA_MAC_DELETE_FPM); |
1608 | UNSET_FLAG(fpm_mac->fpm_flags, ZEBRA_MAC_UPDATE_FPM); | |
44f7f132 | 1609 | return 0; |
a780a738 | 1610 | } |
8e3aae66 | 1611 | } else |
44f7f132 AD |
1612 | fpm_mac = hash_get(zfpm_g->fpm_mac_info_table, &key, |
1613 | zfpm_mac_info_alloc); | |
a780a738 | 1614 | |
44f7f132 | 1615 | fpm_mac->r_vtep_ip.s_addr = rmac->fwd_info.r_vtep_ip.s_addr; |
c5431822 | 1616 | fpm_mac->zebra_flags = rmac->flags; |
a780a738 AD |
1617 | fpm_mac->vxlan_if = vxlan_if ? vxlan_if->ifindex : 0; |
1618 | fpm_mac->svi_if = svi_if ? svi_if->ifindex : 0; | |
1619 | ||
1620 | SET_FLAG(fpm_mac->fpm_flags, ZEBRA_MAC_UPDATE_FPM); | |
1621 | if (delete) | |
1622 | SET_FLAG(fpm_mac->fpm_flags, ZEBRA_MAC_DELETE_FPM); | |
44f7f132 AD |
1623 | else |
1624 | UNSET_FLAG(fpm_mac->fpm_flags, ZEBRA_MAC_DELETE_FPM); | |
a780a738 | 1625 | |
44f7f132 AD |
1626 | if (!mac_found) |
1627 | TAILQ_INSERT_TAIL(&zfpm_g->mac_q, fpm_mac, fpm_mac_q_entries); | |
a780a738 AD |
1628 | |
1629 | zfpm_g->stats.updates_triggered++; | |
1630 | ||
a780a738 AD |
1631 | /* If writes are already enabled, return. */ |
1632 | if (zfpm_g->t_write) | |
1633 | return 0; | |
1634 | ||
1635 | zfpm_write_on(); | |
1636 | return 0; | |
1637 | } | |
1638 | ||
fbe748e5 AD |
1639 | /* |
1640 | * This function is called when the FPM connections is established. | |
1641 | * Iterate over all the RMAC entries for the given L3VNI | |
1642 | * and enqueue the RMAC for FPM processing. | |
1643 | */ | |
1ac88792 | 1644 | static void zfpm_trigger_rmac_update_wrapper(struct hash_bucket *bucket, |
fbe748e5 AD |
1645 | void *args) |
1646 | { | |
3198b2b3 | 1647 | struct zebra_mac *zrmac = (struct zebra_mac *)bucket->data; |
05843a27 | 1648 | struct zebra_l3vni *zl3vni = (struct zebra_l3vni *)args; |
fbe748e5 AD |
1649 | |
1650 | zfpm_trigger_rmac_update(zrmac, zl3vni, false, "RMAC added"); | |
1651 | } | |
1652 | ||
1653 | /* | |
1654 | * This function is called when the FPM connections is established. | |
1655 | * This function iterates over all the L3VNIs to trigger | |
1656 | * FPM updates for RMACs currently available. | |
1657 | */ | |
1ac88792 | 1658 | static void zfpm_iterate_rmac_table(struct hash_bucket *bucket, void *args) |
fbe748e5 | 1659 | { |
05843a27 | 1660 | struct zebra_l3vni *zl3vni = (struct zebra_l3vni *)bucket->data; |
fbe748e5 AD |
1661 | |
1662 | hash_iterate(zl3vni->rmac_table, zfpm_trigger_rmac_update_wrapper, | |
1663 | (void *)zl3vni); | |
1664 | } | |
1665 | ||
5adc2528 | 1666 | /* |
eeaf257b | 1667 | * struct zfpm_statsimer_cb |
5adc2528 | 1668 | */ |
e6685141 | 1669 | static void zfpm_stats_timer_cb(struct event *t) |
5adc2528 | 1670 | { |
d62a17ae | 1671 | zfpm_g->t_stats = NULL; |
5adc2528 | 1672 | |
d62a17ae | 1673 | /* |
1674 | * Remember the stats collected in the last interval for display | |
1675 | * purposes. | |
1676 | */ | |
1677 | zfpm_stats_copy(&zfpm_g->stats, &zfpm_g->last_ivl_stats); | |
5adc2528 | 1678 | |
d62a17ae | 1679 | /* |
1680 | * Add the current set of stats into the cumulative statistics. | |
1681 | */ | |
1682 | zfpm_stats_compose(&zfpm_g->cumulative_stats, &zfpm_g->stats, | |
1683 | &zfpm_g->cumulative_stats); | |
5adc2528 | 1684 | |
d62a17ae | 1685 | /* |
1686 | * Start collecting stats afresh over the next interval. | |
1687 | */ | |
1688 | zfpm_stats_reset(&zfpm_g->stats); | |
5adc2528 | 1689 | |
d62a17ae | 1690 | zfpm_start_stats_timer(); |
5adc2528 AS |
1691 | } |
1692 | ||
1693 | /* | |
1694 | * zfpm_stop_stats_timer | |
1695 | */ | |
d62a17ae | 1696 | static void zfpm_stop_stats_timer(void) |
5adc2528 | 1697 | { |
d62a17ae | 1698 | if (!zfpm_g->t_stats) |
1699 | return; | |
5adc2528 | 1700 | |
d62a17ae | 1701 | zfpm_debug("Stopping existing stats timer"); |
e16d030c | 1702 | EVENT_OFF(zfpm_g->t_stats); |
5adc2528 AS |
1703 | } |
1704 | ||
1705 | /* | |
1706 | * zfpm_start_stats_timer | |
1707 | */ | |
d62a17ae | 1708 | void zfpm_start_stats_timer(void) |
5adc2528 | 1709 | { |
d62a17ae | 1710 | assert(!zfpm_g->t_stats); |
5adc2528 | 1711 | |
907a2395 DS |
1712 | event_add_timer(zfpm_g->master, zfpm_stats_timer_cb, 0, |
1713 | ZFPM_STATS_IVL_SECS, &zfpm_g->t_stats); | |
5adc2528 AS |
1714 | } |
1715 | ||
1716 | /* | |
1717 | * Helper macro for zfpm_show_stats() below. | |
1718 | */ | |
d62a17ae | 1719 | #define ZFPM_SHOW_STAT(counter) \ |
1720 | do { \ | |
1721 | vty_out(vty, "%-40s %10lu %16lu\n", #counter, \ | |
1722 | total_stats.counter, zfpm_g->last_ivl_stats.counter); \ | |
1723 | } while (0) | |
5adc2528 AS |
1724 | |
1725 | /* | |
1726 | * zfpm_show_stats | |
1727 | */ | |
d62a17ae | 1728 | static void zfpm_show_stats(struct vty *vty) |
5adc2528 | 1729 | { |
eeaf257b | 1730 | struct zfpm_stats total_stats; |
d62a17ae | 1731 | time_t elapsed; |
1732 | ||
1733 | vty_out(vty, "\n%-40s %10s Last %2d secs\n\n", "Counter", "Total", | |
1734 | ZFPM_STATS_IVL_SECS); | |
1735 | ||
1736 | /* | |
1737 | * Compute the total stats up to this instant. | |
1738 | */ | |
1739 | zfpm_stats_compose(&zfpm_g->cumulative_stats, &zfpm_g->stats, | |
1740 | &total_stats); | |
1741 | ||
1742 | ZFPM_SHOW_STAT(connect_calls); | |
1743 | ZFPM_SHOW_STAT(connect_no_sock); | |
1744 | ZFPM_SHOW_STAT(read_cb_calls); | |
1745 | ZFPM_SHOW_STAT(write_cb_calls); | |
1746 | ZFPM_SHOW_STAT(write_calls); | |
1747 | ZFPM_SHOW_STAT(partial_writes); | |
1748 | ZFPM_SHOW_STAT(max_writes_hit); | |
1749 | ZFPM_SHOW_STAT(t_write_yields); | |
1750 | ZFPM_SHOW_STAT(nop_deletes_skipped); | |
1751 | ZFPM_SHOW_STAT(route_adds); | |
1752 | ZFPM_SHOW_STAT(route_dels); | |
1753 | ZFPM_SHOW_STAT(updates_triggered); | |
d62a17ae | 1754 | ZFPM_SHOW_STAT(redundant_triggers); |
1755 | ZFPM_SHOW_STAT(dests_del_after_update); | |
1756 | ZFPM_SHOW_STAT(t_conn_down_starts); | |
1757 | ZFPM_SHOW_STAT(t_conn_down_dests_processed); | |
1758 | ZFPM_SHOW_STAT(t_conn_down_yields); | |
1759 | ZFPM_SHOW_STAT(t_conn_down_finishes); | |
1760 | ZFPM_SHOW_STAT(t_conn_up_starts); | |
1761 | ZFPM_SHOW_STAT(t_conn_up_dests_processed); | |
1762 | ZFPM_SHOW_STAT(t_conn_up_yields); | |
1763 | ZFPM_SHOW_STAT(t_conn_up_aborts); | |
1764 | ZFPM_SHOW_STAT(t_conn_up_finishes); | |
1765 | ||
1766 | if (!zfpm_g->last_stats_clear_time) | |
1767 | return; | |
1768 | ||
1769 | elapsed = zfpm_get_elapsed_time(zfpm_g->last_stats_clear_time); | |
1770 | ||
1771 | vty_out(vty, "\nStats were cleared %lu seconds ago\n", | |
1772 | (unsigned long)elapsed); | |
5adc2528 AS |
1773 | } |
1774 | ||
1775 | /* | |
1776 | * zfpm_clear_stats | |
1777 | */ | |
d62a17ae | 1778 | static void zfpm_clear_stats(struct vty *vty) |
5adc2528 | 1779 | { |
d62a17ae | 1780 | if (!zfpm_is_enabled()) { |
1781 | vty_out(vty, "The FPM module is not enabled...\n"); | |
1782 | return; | |
1783 | } | |
5adc2528 | 1784 | |
d62a17ae | 1785 | zfpm_stats_reset(&zfpm_g->stats); |
1786 | zfpm_stats_reset(&zfpm_g->last_ivl_stats); | |
1787 | zfpm_stats_reset(&zfpm_g->cumulative_stats); | |
5adc2528 | 1788 | |
d62a17ae | 1789 | zfpm_stop_stats_timer(); |
1790 | zfpm_start_stats_timer(); | |
5adc2528 | 1791 | |
d62a17ae | 1792 | zfpm_g->last_stats_clear_time = monotime(NULL); |
5adc2528 | 1793 | |
d62a17ae | 1794 | vty_out(vty, "Cleared FPM stats\n"); |
5adc2528 AS |
1795 | } |
1796 | ||
1797 | /* | |
1798 | * show_zebra_fpm_stats | |
1799 | */ | |
1800 | DEFUN (show_zebra_fpm_stats, | |
1801 | show_zebra_fpm_stats_cmd, | |
1802 | "show zebra fpm stats", | |
1803 | SHOW_STR | |
41e7fb80 | 1804 | ZEBRA_STR |
5adc2528 AS |
1805 | "Forwarding Path Manager information\n" |
1806 | "Statistics\n") | |
1807 | { | |
d62a17ae | 1808 | zfpm_show_stats(vty); |
1809 | return CMD_SUCCESS; | |
5adc2528 AS |
1810 | } |
1811 | ||
1812 | /* | |
1813 | * clear_zebra_fpm_stats | |
1814 | */ | |
1815 | DEFUN (clear_zebra_fpm_stats, | |
1816 | clear_zebra_fpm_stats_cmd, | |
1817 | "clear zebra fpm stats", | |
1818 | CLEAR_STR | |
41e7fb80 | 1819 | ZEBRA_STR |
5adc2528 AS |
1820 | "Clear Forwarding Path Manager information\n" |
1821 | "Statistics\n") | |
1822 | { | |
d62a17ae | 1823 | zfpm_clear_stats(vty); |
1824 | return CMD_SUCCESS; | |
5adc2528 AS |
1825 | } |
1826 | ||
711ff0ba | 1827 | /* |
d62a17ae | 1828 | * update fpm connection information |
711ff0ba | 1829 | */ |
a0dfca37 | 1830 | DEFUN (fpm_remote_ip, |
e52702f2 | 1831 | fpm_remote_ip_cmd, |
a0dfca37 DL |
1832 | "fpm connection ip A.B.C.D port (1-65535)", |
1833 | "Forwarding Path Manager\n" | |
1834 | "Configure FPM connection\n" | |
1835 | "Connect to IPv4 address\n" | |
1836 | "Connect to IPv4 address\n" | |
1837 | "TCP port number\n" | |
1838 | "TCP port number\n") | |
711ff0ba USK |
1839 | { |
1840 | ||
d62a17ae | 1841 | in_addr_t fpm_server; |
1842 | uint32_t port_no; | |
711ff0ba | 1843 | |
d62a17ae | 1844 | fpm_server = inet_addr(argv[3]->arg); |
1845 | if (fpm_server == INADDR_NONE) | |
1846 | return CMD_ERR_INCOMPLETE; | |
711ff0ba | 1847 | |
d62a17ae | 1848 | port_no = atoi(argv[5]->arg); |
1849 | if (port_no < TCP_MIN_PORT || port_no > TCP_MAX_PORT) | |
1850 | return CMD_ERR_INCOMPLETE; | |
711ff0ba | 1851 | |
d62a17ae | 1852 | zfpm_g->fpm_server = fpm_server; |
1853 | zfpm_g->fpm_port = port_no; | |
711ff0ba USK |
1854 | |
1855 | ||
d62a17ae | 1856 | return CMD_SUCCESS; |
711ff0ba USK |
1857 | } |
1858 | ||
a0dfca37 | 1859 | DEFUN (no_fpm_remote_ip, |
e52702f2 | 1860 | no_fpm_remote_ip_cmd, |
a0dfca37 DL |
1861 | "no fpm connection ip A.B.C.D port (1-65535)", |
1862 | NO_STR | |
1863 | "Forwarding Path Manager\n" | |
1864 | "Remove configured FPM connection\n" | |
1865 | "Connect to IPv4 address\n" | |
1866 | "Connect to IPv4 address\n" | |
1867 | "TCP port number\n" | |
1868 | "TCP port number\n") | |
711ff0ba | 1869 | { |
d62a17ae | 1870 | if (zfpm_g->fpm_server != inet_addr(argv[4]->arg) |
1871 | || zfpm_g->fpm_port != atoi(argv[6]->arg)) | |
1872 | return CMD_ERR_NO_MATCH; | |
711ff0ba | 1873 | |
d62a17ae | 1874 | zfpm_g->fpm_server = FPM_DEFAULT_IP; |
1875 | zfpm_g->fpm_port = FPM_DEFAULT_PORT; | |
711ff0ba | 1876 | |
d62a17ae | 1877 | return CMD_SUCCESS; |
711ff0ba | 1878 | } |
711ff0ba | 1879 | |
fb0aa886 AS |
1880 | /* |
1881 | * zfpm_init_message_format | |
1882 | */ | |
d62a17ae | 1883 | static inline void zfpm_init_message_format(const char *format) |
fb0aa886 | 1884 | { |
d62a17ae | 1885 | int have_netlink, have_protobuf; |
fb0aa886 | 1886 | |
fb0aa886 | 1887 | #ifdef HAVE_NETLINK |
d62a17ae | 1888 | have_netlink = 1; |
4b2792b5 | 1889 | #else |
d62a17ae | 1890 | have_netlink = 0; |
fb0aa886 AS |
1891 | #endif |
1892 | ||
1893 | #ifdef HAVE_PROTOBUF | |
d62a17ae | 1894 | have_protobuf = 1; |
4b2792b5 | 1895 | #else |
d62a17ae | 1896 | have_protobuf = 0; |
fb0aa886 AS |
1897 | #endif |
1898 | ||
d62a17ae | 1899 | zfpm_g->message_format = ZFPM_MSG_FORMAT_NONE; |
fb0aa886 | 1900 | |
d62a17ae | 1901 | if (!format) { |
1902 | if (have_netlink) { | |
1903 | zfpm_g->message_format = ZFPM_MSG_FORMAT_NETLINK; | |
1904 | } else if (have_protobuf) { | |
1905 | zfpm_g->message_format = ZFPM_MSG_FORMAT_PROTOBUF; | |
1906 | } | |
1907 | return; | |
fb0aa886 | 1908 | } |
fb0aa886 | 1909 | |
d62a17ae | 1910 | if (!strcmp("netlink", format)) { |
1911 | if (!have_netlink) { | |
1c50c1c0 QY |
1912 | flog_err(EC_ZEBRA_NETLINK_NOT_AVAILABLE, |
1913 | "FPM netlink message format is not available"); | |
d62a17ae | 1914 | return; |
1915 | } | |
1916 | zfpm_g->message_format = ZFPM_MSG_FORMAT_NETLINK; | |
1917 | return; | |
fb0aa886 | 1918 | } |
fb0aa886 | 1919 | |
d62a17ae | 1920 | if (!strcmp("protobuf", format)) { |
1921 | if (!have_protobuf) { | |
af4c2728 | 1922 | flog_err( |
e914ccbe | 1923 | EC_ZEBRA_PROTOBUF_NOT_AVAILABLE, |
d62a17ae | 1924 | "FPM protobuf message format is not available"); |
1925 | return; | |
1926 | } | |
8b9cf71c | 1927 | flog_warn(EC_ZEBRA_PROTOBUF_NOT_AVAILABLE, |
3efd0893 | 1928 | "FPM protobuf message format is deprecated and scheduled to be removed. Please convert to using netlink format or contact dev@lists.frrouting.org with your use case."); |
d62a17ae | 1929 | zfpm_g->message_format = ZFPM_MSG_FORMAT_PROTOBUF; |
1930 | return; | |
fb0aa886 | 1931 | } |
fb0aa886 | 1932 | |
e914ccbe | 1933 | flog_warn(EC_ZEBRA_FPM_FORMAT_UNKNOWN, "Unknown fpm format '%s'", |
9df414fe | 1934 | format); |
fb0aa886 AS |
1935 | } |
1936 | ||
711ff0ba | 1937 | /** |
d62a17ae | 1938 | * fpm_remote_srv_write |
711ff0ba | 1939 | * |
d62a17ae | 1940 | * Module to write remote fpm connection |
711ff0ba USK |
1941 | * |
1942 | * Returns ZERO on success. | |
1943 | */ | |
1944 | ||
d62a17ae | 1945 | static int fpm_remote_srv_write(struct vty *vty) |
711ff0ba | 1946 | { |
d62a17ae | 1947 | struct in_addr in; |
711ff0ba | 1948 | |
d62a17ae | 1949 | in.s_addr = zfpm_g->fpm_server; |
711ff0ba | 1950 | |
9d1c2659 | 1951 | if ((zfpm_g->fpm_server != FPM_DEFAULT_IP |
996c9314 LB |
1952 | && zfpm_g->fpm_server != INADDR_ANY) |
1953 | || (zfpm_g->fpm_port != FPM_DEFAULT_PORT && zfpm_g->fpm_port != 0)) | |
9bcef951 | 1954 | vty_out(vty, "fpm connection ip %pI4 port %d\n", &in, |
d62a17ae | 1955 | zfpm_g->fpm_port); |
711ff0ba | 1956 | |
d62a17ae | 1957 | return 0; |
711ff0ba USK |
1958 | } |
1959 | ||
1960 | ||
612c2c15 | 1961 | static int fpm_remote_srv_write(struct vty *vty); |
4f8ea50c | 1962 | /* Zebra node */ |
62b346ee | 1963 | static struct cmd_node zebra_node = { |
f4b8291f | 1964 | .name = "zebra", |
62b346ee | 1965 | .node = ZEBRA_NODE, |
24389580 | 1966 | .parent_node = CONFIG_NODE, |
62b346ee | 1967 | .prompt = "", |
612c2c15 | 1968 | .config_write = fpm_remote_srv_write, |
62b346ee | 1969 | }; |
4f8ea50c DL |
1970 | |
1971 | ||
5adc2528 AS |
1972 | /** |
1973 | * zfpm_init | |
1974 | * | |
1975 | * One-time initialization of the Zebra FPM module. | |
1976 | * | |
1977 | * @param[in] port port at which FPM is running. | |
2951a7a4 | 1978 | * @param[in] enable true if the zebra FPM module should be enabled |
fb0aa886 | 1979 | * @param[in] format to use to talk to the FPM. Can be 'netink' or 'protobuf'. |
5adc2528 | 1980 | * |
2951a7a4 | 1981 | * Returns true on success. |
5adc2528 | 1982 | */ |
cd9d0537 | 1983 | static int zfpm_init(struct event_loop *master) |
5adc2528 | 1984 | { |
d62a17ae | 1985 | int enable = 1; |
1986 | uint16_t port = 0; | |
1987 | const char *format = THIS_MODULE->load_args; | |
5adc2528 | 1988 | |
d62a17ae | 1989 | memset(zfpm_g, 0, sizeof(*zfpm_g)); |
1990 | zfpm_g->master = master; | |
1991 | TAILQ_INIT(&zfpm_g->dest_q); | |
e5218ec8 AD |
1992 | TAILQ_INIT(&zfpm_g->mac_q); |
1993 | ||
1994 | /* Create hash table for fpm_mac_info_t enties */ | |
1995 | zfpm_g->fpm_mac_info_table = hash_create(zfpm_mac_info_hash_keymake, | |
fbe748e5 AD |
1996 | zfpm_mac_info_cmp, |
1997 | "FPM MAC info hash table"); | |
e5218ec8 | 1998 | |
d62a17ae | 1999 | zfpm_g->sock = -1; |
2000 | zfpm_g->state = ZFPM_STATE_IDLE; | |
5adc2528 | 2001 | |
d62a17ae | 2002 | zfpm_stats_init(&zfpm_g->stats); |
2003 | zfpm_stats_init(&zfpm_g->last_ivl_stats); | |
2004 | zfpm_stats_init(&zfpm_g->cumulative_stats); | |
5adc2528 | 2005 | |
316d2d52 | 2006 | memset(&ipv4ll_gateway, 0, sizeof(ipv4ll_gateway)); |
b51c6597 DS |
2007 | if (inet_pton(AF_INET, ipv4_ll_buf, &ipv4ll_gateway.ipv4) != 1) |
2008 | zlog_warn("inet_pton failed for %s", ipv4_ll_buf); | |
316d2d52 | 2009 | |
612c2c15 | 2010 | install_node(&zebra_node); |
d62a17ae | 2011 | install_element(ENABLE_NODE, &show_zebra_fpm_stats_cmd); |
2012 | install_element(ENABLE_NODE, &clear_zebra_fpm_stats_cmd); | |
2013 | install_element(CONFIG_NODE, &fpm_remote_ip_cmd); | |
2014 | install_element(CONFIG_NODE, &no_fpm_remote_ip_cmd); | |
5adc2528 | 2015 | |
d62a17ae | 2016 | zfpm_init_message_format(format); |
fb0aa886 | 2017 | |
d62a17ae | 2018 | /* |
2019 | * Disable FPM interface if no suitable format is available. | |
2020 | */ | |
2021 | if (zfpm_g->message_format == ZFPM_MSG_FORMAT_NONE) | |
2022 | enable = 0; | |
fb0aa886 | 2023 | |
d62a17ae | 2024 | zfpm_g->enabled = enable; |
5adc2528 | 2025 | |
d62a17ae | 2026 | if (!zfpm_g->fpm_server) |
2027 | zfpm_g->fpm_server = FPM_DEFAULT_IP; | |
711ff0ba | 2028 | |
d62a17ae | 2029 | if (!port) |
2030 | port = FPM_DEFAULT_PORT; | |
5adc2528 | 2031 | |
d62a17ae | 2032 | zfpm_g->fpm_port = port; |
5adc2528 | 2033 | |
d62a17ae | 2034 | zfpm_g->obuf = stream_new(ZFPM_OBUF_SIZE); |
2035 | zfpm_g->ibuf = stream_new(ZFPM_IBUF_SIZE); | |
5adc2528 | 2036 | |
d62a17ae | 2037 | zfpm_start_stats_timer(); |
2038 | zfpm_start_connect_timer("initialized"); | |
2039 | return 0; | |
4f8ea50c | 2040 | } |
5adc2528 | 2041 | |
f0c459f0 DS |
2042 | static int zfpm_fini(void) |
2043 | { | |
2044 | zfpm_write_off(); | |
2045 | zfpm_read_off(); | |
2046 | zfpm_connect_off(); | |
54033432 | 2047 | zfpm_conn_down_off(); |
f0c459f0 DS |
2048 | |
2049 | zfpm_stop_stats_timer(); | |
2050 | ||
2051 | hook_unregister(rib_update, zfpm_trigger_update); | |
5ec001aa DS |
2052 | hook_unregister(zebra_rmac_update, zfpm_trigger_rmac_update); |
2053 | ||
f0c459f0 DS |
2054 | return 0; |
2055 | } | |
2056 | ||
d62a17ae | 2057 | static int zebra_fpm_module_init(void) |
4f8ea50c | 2058 | { |
d62a17ae | 2059 | hook_register(rib_update, zfpm_trigger_update); |
0d0f516c | 2060 | hook_register(rib_shutdown, zfpm_trigger_remove); |
a780a738 | 2061 | hook_register(zebra_rmac_update, zfpm_trigger_rmac_update); |
d62a17ae | 2062 | hook_register(frr_late_init, zfpm_init); |
f0c459f0 | 2063 | hook_register(frr_early_fini, zfpm_fini); |
d62a17ae | 2064 | return 0; |
5adc2528 | 2065 | } |
4f8ea50c | 2066 | |
d62a17ae | 2067 | FRR_MODULE_SETUP(.name = "zebra_fpm", .version = FRR_VERSION, |
2068 | .description = "zebra FPM (Forwarding Plane Manager) module", | |
80413c20 DL |
2069 | .init = zebra_fpm_module_init, |
2070 | ); |