]>
Commit | Line | Data |
---|---|---|
5adc2528 AS |
1 | /* |
2 | * Main implementation file for interface to Forwarding Plane Manager. | |
3 | * | |
4 | * Copyright (C) 2012 by Open Source Routing. | |
5 | * Copyright (C) 2012 by Internet Systems Consortium, Inc. ("ISC") | |
6 | * | |
7 | * This file is part of GNU Zebra. | |
8 | * | |
9 | * GNU Zebra is free software; you can redistribute it and/or modify it | |
10 | * under the terms of the GNU General Public License as published by the | |
11 | * Free Software Foundation; either version 2, or (at your option) any | |
12 | * later version. | |
13 | * | |
14 | * GNU Zebra is distributed in the hope that it will be useful, but | |
15 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
16 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
17 | * General Public License for more details. | |
18 | * | |
896014f4 DL |
19 | * You should have received a copy of the GNU General Public License along |
20 | * with this program; see the file COPYING; if not, write to the Free Software | |
21 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA | |
5adc2528 AS |
22 | */ |
23 | ||
24 | #include <zebra.h> | |
25 | ||
26 | #include "log.h" | |
4f8ea50c | 27 | #include "libfrr.h" |
5adc2528 AS |
28 | #include "stream.h" |
29 | #include "thread.h" | |
30 | #include "network.h" | |
31 | #include "command.h" | |
4f8ea50c | 32 | #include "version.h" |
5adc2528 AS |
33 | |
34 | #include "zebra/rib.h" | |
7c551956 DS |
35 | #include "zebra/zserv.h" |
36 | #include "zebra/zebra_ns.h" | |
37 | #include "zebra/zebra_vrf.h" | |
5adc2528 AS |
38 | |
39 | #include "fpm/fpm.h" | |
5adc2528 AS |
40 | #include "zebra_fpm_private.h" |
41 | ||
42 | /* | |
43 | * Interval at which we attempt to connect to the FPM. | |
44 | */ | |
45 | #define ZFPM_CONNECT_RETRY_IVL 5 | |
46 | ||
47 | /* | |
48 | * Sizes of outgoing and incoming stream buffers for writing/reading | |
49 | * FPM messages. | |
50 | */ | |
51 | #define ZFPM_OBUF_SIZE (2 * FPM_MAX_MSG_LEN) | |
52 | #define ZFPM_IBUF_SIZE (FPM_MAX_MSG_LEN) | |
53 | ||
54 | /* | |
55 | * The maximum number of times the FPM socket write callback can call | |
56 | * 'write' before it yields. | |
57 | */ | |
58 | #define ZFPM_MAX_WRITES_PER_RUN 10 | |
59 | ||
60 | /* | |
61 | * Interval over which we collect statistics. | |
62 | */ | |
63 | #define ZFPM_STATS_IVL_SECS 10 | |
64 | ||
65 | /* | |
66 | * Structure that holds state for iterating over all route_node | |
67 | * structures that are candidates for being communicated to the FPM. | |
68 | */ | |
d62a17ae | 69 | typedef struct zfpm_rnodes_iter_t_ { |
70 | rib_tables_iter_t tables_iter; | |
71 | route_table_iter_t iter; | |
5adc2528 AS |
72 | } zfpm_rnodes_iter_t; |
73 | ||
74 | /* | |
75 | * Statistics. | |
76 | */ | |
77 | typedef struct zfpm_stats_t_ { | |
d62a17ae | 78 | unsigned long connect_calls; |
79 | unsigned long connect_no_sock; | |
5adc2528 | 80 | |
d62a17ae | 81 | unsigned long read_cb_calls; |
5adc2528 | 82 | |
d62a17ae | 83 | unsigned long write_cb_calls; |
84 | unsigned long write_calls; | |
85 | unsigned long partial_writes; | |
86 | unsigned long max_writes_hit; | |
87 | unsigned long t_write_yields; | |
5adc2528 | 88 | |
d62a17ae | 89 | unsigned long nop_deletes_skipped; |
90 | unsigned long route_adds; | |
91 | unsigned long route_dels; | |
5adc2528 | 92 | |
d62a17ae | 93 | unsigned long updates_triggered; |
94 | unsigned long redundant_triggers; | |
5adc2528 | 95 | |
d62a17ae | 96 | unsigned long dests_del_after_update; |
5adc2528 | 97 | |
d62a17ae | 98 | unsigned long t_conn_down_starts; |
99 | unsigned long t_conn_down_dests_processed; | |
100 | unsigned long t_conn_down_yields; | |
101 | unsigned long t_conn_down_finishes; | |
5adc2528 | 102 | |
d62a17ae | 103 | unsigned long t_conn_up_starts; |
104 | unsigned long t_conn_up_dests_processed; | |
105 | unsigned long t_conn_up_yields; | |
106 | unsigned long t_conn_up_aborts; | |
107 | unsigned long t_conn_up_finishes; | |
5adc2528 AS |
108 | |
109 | } zfpm_stats_t; | |
110 | ||
111 | /* | |
112 | * States for the FPM state machine. | |
113 | */ | |
114 | typedef enum { | |
115 | ||
d62a17ae | 116 | /* |
117 | * In this state we are not yet ready to connect to the FPM. This | |
118 | * can happen when this module is disabled, or if we're cleaning up | |
119 | * after a connection has gone down. | |
120 | */ | |
121 | ZFPM_STATE_IDLE, | |
122 | ||
123 | /* | |
124 | * Ready to talk to the FPM and periodically trying to connect to | |
125 | * it. | |
126 | */ | |
127 | ZFPM_STATE_ACTIVE, | |
128 | ||
129 | /* | |
130 | * In the middle of bringing up a TCP connection. Specifically, | |
131 | * waiting for a connect() call to complete asynchronously. | |
132 | */ | |
133 | ZFPM_STATE_CONNECTING, | |
134 | ||
135 | /* | |
136 | * TCP connection to the FPM is up. | |
137 | */ | |
138 | ZFPM_STATE_ESTABLISHED | |
5adc2528 AS |
139 | |
140 | } zfpm_state_t; | |
141 | ||
fb0aa886 AS |
142 | /* |
143 | * Message format to be used to communicate with the FPM. | |
144 | */ | |
d62a17ae | 145 | typedef enum { |
146 | ZFPM_MSG_FORMAT_NONE, | |
147 | ZFPM_MSG_FORMAT_NETLINK, | |
148 | ZFPM_MSG_FORMAT_PROTOBUF, | |
fb0aa886 | 149 | } zfpm_msg_format_e; |
5adc2528 AS |
150 | /* |
151 | * Globals. | |
152 | */ | |
d62a17ae | 153 | typedef struct zfpm_glob_t_ { |
154 | ||
155 | /* | |
156 | * True if the FPM module has been enabled. | |
157 | */ | |
158 | int enabled; | |
159 | ||
160 | /* | |
161 | * Message format to be used to communicate with the fpm. | |
162 | */ | |
163 | zfpm_msg_format_e message_format; | |
164 | ||
165 | struct thread_master *master; | |
166 | ||
167 | zfpm_state_t state; | |
168 | ||
169 | in_addr_t fpm_server; | |
170 | /* | |
171 | * Port on which the FPM is running. | |
172 | */ | |
173 | int fpm_port; | |
174 | ||
175 | /* | |
176 | * List of rib_dest_t structures to be processed | |
177 | */ | |
178 | TAILQ_HEAD(zfpm_dest_q, rib_dest_t_) dest_q; | |
179 | ||
180 | /* | |
181 | * Stream socket to the FPM. | |
182 | */ | |
183 | int sock; | |
184 | ||
185 | /* | |
186 | * Buffers for messages to/from the FPM. | |
187 | */ | |
188 | struct stream *obuf; | |
189 | struct stream *ibuf; | |
190 | ||
191 | /* | |
192 | * Threads for I/O. | |
193 | */ | |
194 | struct thread *t_connect; | |
195 | struct thread *t_write; | |
196 | struct thread *t_read; | |
197 | ||
198 | /* | |
199 | * Thread to clean up after the TCP connection to the FPM goes down | |
200 | * and the state that belongs to it. | |
201 | */ | |
202 | struct thread *t_conn_down; | |
203 | ||
204 | struct { | |
205 | zfpm_rnodes_iter_t iter; | |
206 | } t_conn_down_state; | |
207 | ||
208 | /* | |
209 | * Thread to take actions once the TCP conn to the FPM comes up, and | |
210 | * the state that belongs to it. | |
211 | */ | |
212 | struct thread *t_conn_up; | |
213 | ||
214 | struct { | |
215 | zfpm_rnodes_iter_t iter; | |
216 | } t_conn_up_state; | |
217 | ||
218 | unsigned long connect_calls; | |
219 | time_t last_connect_call_time; | |
220 | ||
221 | /* | |
222 | * Stats from the start of the current statistics interval up to | |
223 | * now. These are the counters we typically update in the code. | |
224 | */ | |
225 | zfpm_stats_t stats; | |
226 | ||
227 | /* | |
228 | * Statistics that were gathered in the last collection interval. | |
229 | */ | |
230 | zfpm_stats_t last_ivl_stats; | |
231 | ||
232 | /* | |
233 | * Cumulative stats from the last clear to the start of the current | |
234 | * statistics interval. | |
235 | */ | |
236 | zfpm_stats_t cumulative_stats; | |
237 | ||
238 | /* | |
239 | * Stats interval timer. | |
240 | */ | |
241 | struct thread *t_stats; | |
242 | ||
243 | /* | |
244 | * If non-zero, the last time when statistics were cleared. | |
245 | */ | |
246 | time_t last_stats_clear_time; | |
5adc2528 AS |
247 | |
248 | } zfpm_glob_t; | |
249 | ||
250 | static zfpm_glob_t zfpm_glob_space; | |
251 | static zfpm_glob_t *zfpm_g = &zfpm_glob_space; | |
252 | ||
d62a17ae | 253 | static int zfpm_trigger_update(struct route_node *rn, const char *reason); |
4f8ea50c | 254 | |
d62a17ae | 255 | static int zfpm_read_cb(struct thread *thread); |
256 | static int zfpm_write_cb(struct thread *thread); | |
5adc2528 | 257 | |
d62a17ae | 258 | static void zfpm_set_state(zfpm_state_t state, const char *reason); |
259 | static void zfpm_start_connect_timer(const char *reason); | |
260 | static void zfpm_start_stats_timer(void); | |
5adc2528 AS |
261 | |
262 | /* | |
263 | * zfpm_thread_should_yield | |
264 | */ | |
d62a17ae | 265 | static inline int zfpm_thread_should_yield(struct thread *t) |
5adc2528 | 266 | { |
d62a17ae | 267 | return thread_should_yield(t); |
5adc2528 AS |
268 | } |
269 | ||
270 | /* | |
271 | * zfpm_state_to_str | |
272 | */ | |
d62a17ae | 273 | static const char *zfpm_state_to_str(zfpm_state_t state) |
5adc2528 | 274 | { |
d62a17ae | 275 | switch (state) { |
5adc2528 | 276 | |
d62a17ae | 277 | case ZFPM_STATE_IDLE: |
278 | return "idle"; | |
5adc2528 | 279 | |
d62a17ae | 280 | case ZFPM_STATE_ACTIVE: |
281 | return "active"; | |
5adc2528 | 282 | |
d62a17ae | 283 | case ZFPM_STATE_CONNECTING: |
284 | return "connecting"; | |
5adc2528 | 285 | |
d62a17ae | 286 | case ZFPM_STATE_ESTABLISHED: |
287 | return "established"; | |
5adc2528 | 288 | |
d62a17ae | 289 | default: |
290 | return "unknown"; | |
291 | } | |
5adc2528 AS |
292 | } |
293 | ||
5adc2528 AS |
294 | /* |
295 | * zfpm_get_elapsed_time | |
296 | * | |
297 | * Returns the time elapsed (in seconds) since the given time. | |
298 | */ | |
d62a17ae | 299 | static time_t zfpm_get_elapsed_time(time_t reference) |
5adc2528 | 300 | { |
d62a17ae | 301 | time_t now; |
5adc2528 | 302 | |
d62a17ae | 303 | now = monotime(NULL); |
5adc2528 | 304 | |
d62a17ae | 305 | if (now < reference) { |
306 | assert(0); | |
307 | return 0; | |
308 | } | |
5adc2528 | 309 | |
d62a17ae | 310 | return now - reference; |
5adc2528 AS |
311 | } |
312 | ||
5adc2528 AS |
313 | /* |
314 | * zfpm_rnodes_iter_init | |
315 | */ | |
d62a17ae | 316 | static inline void zfpm_rnodes_iter_init(zfpm_rnodes_iter_t *iter) |
5adc2528 | 317 | { |
d62a17ae | 318 | memset(iter, 0, sizeof(*iter)); |
319 | rib_tables_iter_init(&iter->tables_iter); | |
320 | ||
321 | /* | |
322 | * This is a hack, but it makes implementing 'next' easier by | |
323 | * ensuring that route_table_iter_next() will return NULL the first | |
324 | * time we call it. | |
325 | */ | |
326 | route_table_iter_init(&iter->iter, NULL); | |
327 | route_table_iter_cleanup(&iter->iter); | |
5adc2528 AS |
328 | } |
329 | ||
330 | /* | |
331 | * zfpm_rnodes_iter_next | |
332 | */ | |
d62a17ae | 333 | static inline struct route_node *zfpm_rnodes_iter_next(zfpm_rnodes_iter_t *iter) |
5adc2528 | 334 | { |
d62a17ae | 335 | struct route_node *rn; |
336 | struct route_table *table; | |
5adc2528 | 337 | |
d62a17ae | 338 | while (1) { |
339 | rn = route_table_iter_next(&iter->iter); | |
340 | if (rn) | |
341 | return rn; | |
5adc2528 | 342 | |
d62a17ae | 343 | /* |
344 | * We've made our way through this table, go to the next one. | |
345 | */ | |
346 | route_table_iter_cleanup(&iter->iter); | |
5adc2528 | 347 | |
c6bbea17 | 348 | table = rib_tables_iter_next(&iter->tables_iter); |
5adc2528 | 349 | |
d62a17ae | 350 | if (!table) |
351 | return NULL; | |
5adc2528 | 352 | |
d62a17ae | 353 | route_table_iter_init(&iter->iter, table); |
354 | } | |
5adc2528 | 355 | |
d62a17ae | 356 | return NULL; |
5adc2528 AS |
357 | } |
358 | ||
359 | /* | |
360 | * zfpm_rnodes_iter_pause | |
361 | */ | |
d62a17ae | 362 | static inline void zfpm_rnodes_iter_pause(zfpm_rnodes_iter_t *iter) |
5adc2528 | 363 | { |
d62a17ae | 364 | route_table_iter_pause(&iter->iter); |
5adc2528 AS |
365 | } |
366 | ||
367 | /* | |
368 | * zfpm_rnodes_iter_cleanup | |
369 | */ | |
d62a17ae | 370 | static inline void zfpm_rnodes_iter_cleanup(zfpm_rnodes_iter_t *iter) |
5adc2528 | 371 | { |
d62a17ae | 372 | route_table_iter_cleanup(&iter->iter); |
373 | rib_tables_iter_cleanup(&iter->tables_iter); | |
5adc2528 AS |
374 | } |
375 | ||
376 | /* | |
377 | * zfpm_stats_init | |
378 | * | |
379 | * Initialize a statistics block. | |
380 | */ | |
d62a17ae | 381 | static inline void zfpm_stats_init(zfpm_stats_t *stats) |
5adc2528 | 382 | { |
d62a17ae | 383 | memset(stats, 0, sizeof(*stats)); |
5adc2528 AS |
384 | } |
385 | ||
386 | /* | |
387 | * zfpm_stats_reset | |
388 | */ | |
d62a17ae | 389 | static inline void zfpm_stats_reset(zfpm_stats_t *stats) |
5adc2528 | 390 | { |
d62a17ae | 391 | zfpm_stats_init(stats); |
5adc2528 AS |
392 | } |
393 | ||
394 | /* | |
395 | * zfpm_stats_copy | |
396 | */ | |
d62a17ae | 397 | static inline void zfpm_stats_copy(const zfpm_stats_t *src, zfpm_stats_t *dest) |
5adc2528 | 398 | { |
d62a17ae | 399 | memcpy(dest, src, sizeof(*dest)); |
5adc2528 AS |
400 | } |
401 | ||
402 | /* | |
403 | * zfpm_stats_compose | |
404 | * | |
405 | * Total up the statistics in two stats structures ('s1 and 's2') and | |
406 | * return the result in the third argument, 'result'. Note that the | |
407 | * pointer 'result' may be the same as 's1' or 's2'. | |
408 | * | |
409 | * For simplicity, the implementation below assumes that the stats | |
410 | * structure is composed entirely of counters. This can easily be | |
411 | * changed when necessary. | |
412 | */ | |
d62a17ae | 413 | static void zfpm_stats_compose(const zfpm_stats_t *s1, const zfpm_stats_t *s2, |
414 | zfpm_stats_t *result) | |
5adc2528 | 415 | { |
d62a17ae | 416 | const unsigned long *p1, *p2; |
417 | unsigned long *result_p; | |
418 | int i, num_counters; | |
5adc2528 | 419 | |
d62a17ae | 420 | p1 = (const unsigned long *)s1; |
421 | p2 = (const unsigned long *)s2; | |
422 | result_p = (unsigned long *)result; | |
5adc2528 | 423 | |
d62a17ae | 424 | num_counters = (sizeof(zfpm_stats_t) / sizeof(unsigned long)); |
5adc2528 | 425 | |
d62a17ae | 426 | for (i = 0; i < num_counters; i++) { |
427 | result_p[i] = p1[i] + p2[i]; | |
428 | } | |
5adc2528 AS |
429 | } |
430 | ||
431 | /* | |
432 | * zfpm_read_on | |
433 | */ | |
d62a17ae | 434 | static inline void zfpm_read_on(void) |
5adc2528 | 435 | { |
d62a17ae | 436 | assert(!zfpm_g->t_read); |
437 | assert(zfpm_g->sock >= 0); | |
5adc2528 | 438 | |
d62a17ae | 439 | thread_add_read(zfpm_g->master, zfpm_read_cb, 0, zfpm_g->sock, |
440 | &zfpm_g->t_read); | |
5adc2528 AS |
441 | } |
442 | ||
443 | /* | |
444 | * zfpm_write_on | |
445 | */ | |
d62a17ae | 446 | static inline void zfpm_write_on(void) |
5adc2528 | 447 | { |
d62a17ae | 448 | assert(!zfpm_g->t_write); |
449 | assert(zfpm_g->sock >= 0); | |
5adc2528 | 450 | |
d62a17ae | 451 | thread_add_write(zfpm_g->master, zfpm_write_cb, 0, zfpm_g->sock, |
452 | &zfpm_g->t_write); | |
5adc2528 AS |
453 | } |
454 | ||
455 | /* | |
456 | * zfpm_read_off | |
457 | */ | |
d62a17ae | 458 | static inline void zfpm_read_off(void) |
5adc2528 | 459 | { |
d62a17ae | 460 | THREAD_READ_OFF(zfpm_g->t_read); |
5adc2528 AS |
461 | } |
462 | ||
463 | /* | |
464 | * zfpm_write_off | |
465 | */ | |
d62a17ae | 466 | static inline void zfpm_write_off(void) |
5adc2528 | 467 | { |
d62a17ae | 468 | THREAD_WRITE_OFF(zfpm_g->t_write); |
5adc2528 AS |
469 | } |
470 | ||
471 | /* | |
472 | * zfpm_conn_up_thread_cb | |
473 | * | |
474 | * Callback for actions to be taken when the connection to the FPM | |
475 | * comes up. | |
476 | */ | |
d62a17ae | 477 | static int zfpm_conn_up_thread_cb(struct thread *thread) |
5adc2528 | 478 | { |
d62a17ae | 479 | struct route_node *rnode; |
480 | zfpm_rnodes_iter_t *iter; | |
481 | rib_dest_t *dest; | |
5adc2528 | 482 | |
d62a17ae | 483 | zfpm_g->t_conn_up = NULL; |
5adc2528 | 484 | |
d62a17ae | 485 | iter = &zfpm_g->t_conn_up_state.iter; |
5adc2528 | 486 | |
d62a17ae | 487 | if (zfpm_g->state != ZFPM_STATE_ESTABLISHED) { |
488 | zfpm_debug( | |
489 | "Connection not up anymore, conn_up thread aborting"); | |
490 | zfpm_g->stats.t_conn_up_aborts++; | |
491 | goto done; | |
492 | } | |
5adc2528 | 493 | |
d62a17ae | 494 | while ((rnode = zfpm_rnodes_iter_next(iter))) { |
495 | dest = rib_dest_from_rnode(rnode); | |
496 | ||
497 | if (dest) { | |
498 | zfpm_g->stats.t_conn_up_dests_processed++; | |
499 | zfpm_trigger_update(rnode, NULL); | |
500 | } | |
501 | ||
502 | /* | |
503 | * Yield if need be. | |
504 | */ | |
505 | if (!zfpm_thread_should_yield(thread)) | |
506 | continue; | |
507 | ||
508 | zfpm_g->stats.t_conn_up_yields++; | |
509 | zfpm_rnodes_iter_pause(iter); | |
510 | zfpm_g->t_conn_up = NULL; | |
511 | thread_add_timer_msec(zfpm_g->master, zfpm_conn_up_thread_cb, | |
512 | NULL, 0, &zfpm_g->t_conn_up); | |
513 | return 0; | |
5adc2528 AS |
514 | } |
515 | ||
d62a17ae | 516 | zfpm_g->stats.t_conn_up_finishes++; |
517 | ||
518 | done: | |
519 | zfpm_rnodes_iter_cleanup(iter); | |
520 | return 0; | |
5adc2528 AS |
521 | } |
522 | ||
523 | /* | |
524 | * zfpm_connection_up | |
525 | * | |
526 | * Called when the connection to the FPM comes up. | |
527 | */ | |
d62a17ae | 528 | static void zfpm_connection_up(const char *detail) |
5adc2528 | 529 | { |
d62a17ae | 530 | assert(zfpm_g->sock >= 0); |
531 | zfpm_read_on(); | |
532 | zfpm_write_on(); | |
533 | zfpm_set_state(ZFPM_STATE_ESTABLISHED, detail); | |
534 | ||
535 | /* | |
536 | * Start thread to push existing routes to the FPM. | |
537 | */ | |
538 | assert(!zfpm_g->t_conn_up); | |
539 | ||
540 | zfpm_rnodes_iter_init(&zfpm_g->t_conn_up_state.iter); | |
541 | ||
542 | zfpm_debug("Starting conn_up thread"); | |
543 | zfpm_g->t_conn_up = NULL; | |
544 | thread_add_timer_msec(zfpm_g->master, zfpm_conn_up_thread_cb, NULL, 0, | |
545 | &zfpm_g->t_conn_up); | |
546 | zfpm_g->stats.t_conn_up_starts++; | |
5adc2528 AS |
547 | } |
548 | ||
549 | /* | |
550 | * zfpm_connect_check | |
551 | * | |
552 | * Check if an asynchronous connect() to the FPM is complete. | |
553 | */ | |
d62a17ae | 554 | static void zfpm_connect_check(void) |
5adc2528 | 555 | { |
d62a17ae | 556 | int status; |
557 | socklen_t slen; | |
558 | int ret; | |
559 | ||
560 | zfpm_read_off(); | |
561 | zfpm_write_off(); | |
562 | ||
563 | slen = sizeof(status); | |
564 | ret = getsockopt(zfpm_g->sock, SOL_SOCKET, SO_ERROR, (void *)&status, | |
565 | &slen); | |
566 | ||
567 | if (ret >= 0 && status == 0) { | |
568 | zfpm_connection_up("async connect complete"); | |
569 | return; | |
570 | } | |
571 | ||
572 | /* | |
573 | * getsockopt() failed or indicated an error on the socket. | |
574 | */ | |
575 | close(zfpm_g->sock); | |
576 | zfpm_g->sock = -1; | |
577 | ||
578 | zfpm_start_connect_timer("getsockopt() after async connect failed"); | |
579 | return; | |
5adc2528 AS |
580 | } |
581 | ||
582 | /* | |
583 | * zfpm_conn_down_thread_cb | |
584 | * | |
585 | * Callback that is invoked to clean up state after the TCP connection | |
586 | * to the FPM goes down. | |
587 | */ | |
d62a17ae | 588 | static int zfpm_conn_down_thread_cb(struct thread *thread) |
5adc2528 | 589 | { |
d62a17ae | 590 | struct route_node *rnode; |
591 | zfpm_rnodes_iter_t *iter; | |
592 | rib_dest_t *dest; | |
5adc2528 | 593 | |
d62a17ae | 594 | assert(zfpm_g->state == ZFPM_STATE_IDLE); |
5adc2528 | 595 | |
d62a17ae | 596 | zfpm_g->t_conn_down = NULL; |
5adc2528 | 597 | |
d62a17ae | 598 | iter = &zfpm_g->t_conn_down_state.iter; |
5adc2528 | 599 | |
d62a17ae | 600 | while ((rnode = zfpm_rnodes_iter_next(iter))) { |
601 | dest = rib_dest_from_rnode(rnode); | |
5adc2528 | 602 | |
d62a17ae | 603 | if (dest) { |
604 | if (CHECK_FLAG(dest->flags, RIB_DEST_UPDATE_FPM)) { | |
605 | TAILQ_REMOVE(&zfpm_g->dest_q, dest, | |
606 | fpm_q_entries); | |
607 | } | |
608 | ||
609 | UNSET_FLAG(dest->flags, RIB_DEST_UPDATE_FPM); | |
610 | UNSET_FLAG(dest->flags, RIB_DEST_SENT_TO_FPM); | |
5adc2528 | 611 | |
d62a17ae | 612 | zfpm_g->stats.t_conn_down_dests_processed++; |
5adc2528 | 613 | |
d62a17ae | 614 | /* |
615 | * Check if the dest should be deleted. | |
616 | */ | |
617 | rib_gc_dest(rnode); | |
618 | } | |
5adc2528 | 619 | |
d62a17ae | 620 | /* |
621 | * Yield if need be. | |
622 | */ | |
623 | if (!zfpm_thread_should_yield(thread)) | |
624 | continue; | |
625 | ||
626 | zfpm_g->stats.t_conn_down_yields++; | |
627 | zfpm_rnodes_iter_pause(iter); | |
628 | zfpm_g->t_conn_down = NULL; | |
629 | thread_add_timer_msec(zfpm_g->master, zfpm_conn_down_thread_cb, | |
630 | NULL, 0, &zfpm_g->t_conn_down); | |
631 | return 0; | |
5adc2528 AS |
632 | } |
633 | ||
d62a17ae | 634 | zfpm_g->stats.t_conn_down_finishes++; |
635 | zfpm_rnodes_iter_cleanup(iter); | |
636 | ||
637 | /* | |
638 | * Start the process of connecting to the FPM again. | |
639 | */ | |
640 | zfpm_start_connect_timer("cleanup complete"); | |
641 | return 0; | |
5adc2528 AS |
642 | } |
643 | ||
644 | /* | |
645 | * zfpm_connection_down | |
646 | * | |
647 | * Called when the connection to the FPM has gone down. | |
648 | */ | |
d62a17ae | 649 | static void zfpm_connection_down(const char *detail) |
5adc2528 | 650 | { |
d62a17ae | 651 | if (!detail) |
652 | detail = "unknown"; | |
5adc2528 | 653 | |
d62a17ae | 654 | assert(zfpm_g->state == ZFPM_STATE_ESTABLISHED); |
5adc2528 | 655 | |
d62a17ae | 656 | zlog_info("connection to the FPM has gone down: %s", detail); |
5adc2528 | 657 | |
d62a17ae | 658 | zfpm_read_off(); |
659 | zfpm_write_off(); | |
5adc2528 | 660 | |
d62a17ae | 661 | stream_reset(zfpm_g->ibuf); |
662 | stream_reset(zfpm_g->obuf); | |
5adc2528 | 663 | |
d62a17ae | 664 | if (zfpm_g->sock >= 0) { |
665 | close(zfpm_g->sock); | |
666 | zfpm_g->sock = -1; | |
667 | } | |
5adc2528 | 668 | |
d62a17ae | 669 | /* |
670 | * Start thread to clean up state after the connection goes down. | |
671 | */ | |
672 | assert(!zfpm_g->t_conn_down); | |
673 | zfpm_debug("Starting conn_down thread"); | |
674 | zfpm_rnodes_iter_init(&zfpm_g->t_conn_down_state.iter); | |
675 | zfpm_g->t_conn_down = NULL; | |
676 | thread_add_timer_msec(zfpm_g->master, zfpm_conn_down_thread_cb, NULL, 0, | |
677 | &zfpm_g->t_conn_down); | |
678 | zfpm_g->stats.t_conn_down_starts++; | |
679 | ||
680 | zfpm_set_state(ZFPM_STATE_IDLE, detail); | |
5adc2528 AS |
681 | } |
682 | ||
683 | /* | |
684 | * zfpm_read_cb | |
685 | */ | |
d62a17ae | 686 | static int zfpm_read_cb(struct thread *thread) |
5adc2528 | 687 | { |
d62a17ae | 688 | size_t already; |
689 | struct stream *ibuf; | |
690 | uint16_t msg_len; | |
691 | fpm_msg_hdr_t *hdr; | |
692 | ||
693 | zfpm_g->stats.read_cb_calls++; | |
694 | zfpm_g->t_read = NULL; | |
695 | ||
696 | /* | |
697 | * Check if async connect is now done. | |
698 | */ | |
699 | if (zfpm_g->state == ZFPM_STATE_CONNECTING) { | |
700 | zfpm_connect_check(); | |
701 | return 0; | |
5adc2528 AS |
702 | } |
703 | ||
d62a17ae | 704 | assert(zfpm_g->state == ZFPM_STATE_ESTABLISHED); |
705 | assert(zfpm_g->sock >= 0); | |
5adc2528 | 706 | |
d62a17ae | 707 | ibuf = zfpm_g->ibuf; |
5adc2528 | 708 | |
d62a17ae | 709 | already = stream_get_endp(ibuf); |
710 | if (already < FPM_MSG_HDR_LEN) { | |
711 | ssize_t nbyte; | |
5adc2528 | 712 | |
d62a17ae | 713 | nbyte = stream_read_try(ibuf, zfpm_g->sock, |
714 | FPM_MSG_HDR_LEN - already); | |
715 | if (nbyte == 0 || nbyte == -1) { | |
716 | zfpm_connection_down("closed socket in read"); | |
717 | return 0; | |
718 | } | |
5adc2528 | 719 | |
d62a17ae | 720 | if (nbyte != (ssize_t)(FPM_MSG_HDR_LEN - already)) |
721 | goto done; | |
5adc2528 | 722 | |
d62a17ae | 723 | already = FPM_MSG_HDR_LEN; |
724 | } | |
5adc2528 | 725 | |
d62a17ae | 726 | stream_set_getp(ibuf, 0); |
5adc2528 | 727 | |
d62a17ae | 728 | hdr = (fpm_msg_hdr_t *)stream_pnt(ibuf); |
5adc2528 | 729 | |
d62a17ae | 730 | if (!fpm_msg_hdr_ok(hdr)) { |
731 | zfpm_connection_down("invalid message header"); | |
732 | return 0; | |
5adc2528 AS |
733 | } |
734 | ||
d62a17ae | 735 | msg_len = fpm_msg_len(hdr); |
5adc2528 | 736 | |
d62a17ae | 737 | /* |
738 | * Read out the rest of the packet. | |
739 | */ | |
740 | if (already < msg_len) { | |
741 | ssize_t nbyte; | |
5adc2528 | 742 | |
d62a17ae | 743 | nbyte = stream_read_try(ibuf, zfpm_g->sock, msg_len - already); |
5adc2528 | 744 | |
d62a17ae | 745 | if (nbyte == 0 || nbyte == -1) { |
746 | zfpm_connection_down("failed to read message"); | |
747 | return 0; | |
748 | } | |
749 | ||
750 | if (nbyte != (ssize_t)(msg_len - already)) | |
751 | goto done; | |
752 | } | |
753 | ||
754 | zfpm_debug("Read out a full fpm message"); | |
755 | ||
756 | /* | |
757 | * Just throw it away for now. | |
758 | */ | |
759 | stream_reset(ibuf); | |
760 | ||
761 | done: | |
762 | zfpm_read_on(); | |
763 | return 0; | |
5adc2528 AS |
764 | } |
765 | ||
766 | /* | |
767 | * zfpm_writes_pending | |
768 | * | |
769 | * Returns TRUE if we may have something to write to the FPM. | |
770 | */ | |
d62a17ae | 771 | static int zfpm_writes_pending(void) |
5adc2528 AS |
772 | { |
773 | ||
d62a17ae | 774 | /* |
775 | * Check if there is any data in the outbound buffer that has not | |
776 | * been written to the socket yet. | |
777 | */ | |
778 | if (stream_get_endp(zfpm_g->obuf) - stream_get_getp(zfpm_g->obuf)) | |
779 | return 1; | |
5adc2528 | 780 | |
d62a17ae | 781 | /* |
782 | * Check if there are any prefixes on the outbound queue. | |
783 | */ | |
784 | if (!TAILQ_EMPTY(&zfpm_g->dest_q)) | |
785 | return 1; | |
5adc2528 | 786 | |
d62a17ae | 787 | return 0; |
5adc2528 AS |
788 | } |
789 | ||
790 | /* | |
791 | * zfpm_encode_route | |
792 | * | |
793 | * Encode a message to the FPM with information about the given route. | |
794 | * | |
795 | * Returns the number of bytes written to the buffer. 0 or a negative | |
796 | * value indicates an error. | |
797 | */ | |
d62a17ae | 798 | static inline int zfpm_encode_route(rib_dest_t *dest, struct route_entry *re, |
799 | char *in_buf, size_t in_buf_len, | |
800 | fpm_msg_type_e *msg_type) | |
5adc2528 | 801 | { |
d62a17ae | 802 | size_t len; |
9bf75362 | 803 | #ifdef HAVE_NETLINK |
d62a17ae | 804 | int cmd; |
9bf75362 | 805 | #endif |
d62a17ae | 806 | len = 0; |
5adc2528 | 807 | |
d62a17ae | 808 | *msg_type = FPM_MSG_TYPE_NONE; |
5adc2528 | 809 | |
d62a17ae | 810 | switch (zfpm_g->message_format) { |
5adc2528 | 811 | |
d62a17ae | 812 | case ZFPM_MSG_FORMAT_PROTOBUF: |
fb0aa886 | 813 | #ifdef HAVE_PROTOBUF |
d62a17ae | 814 | len = zfpm_protobuf_encode_route(dest, re, (uint8_t *)in_buf, |
815 | in_buf_len); | |
816 | *msg_type = FPM_MSG_TYPE_PROTOBUF; | |
fb0aa886 | 817 | #endif |
d62a17ae | 818 | break; |
5adc2528 | 819 | |
d62a17ae | 820 | case ZFPM_MSG_FORMAT_NETLINK: |
fb0aa886 | 821 | #ifdef HAVE_NETLINK |
d62a17ae | 822 | *msg_type = FPM_MSG_TYPE_NETLINK; |
823 | cmd = re ? RTM_NEWROUTE : RTM_DELROUTE; | |
824 | len = zfpm_netlink_encode_route(cmd, dest, re, in_buf, | |
825 | in_buf_len); | |
826 | assert(fpm_msg_align(len) == len); | |
827 | *msg_type = FPM_MSG_TYPE_NETLINK; | |
5adc2528 | 828 | #endif /* HAVE_NETLINK */ |
d62a17ae | 829 | break; |
fb0aa886 | 830 | |
d62a17ae | 831 | default: |
832 | break; | |
833 | } | |
fb0aa886 | 834 | |
d62a17ae | 835 | return len; |
5adc2528 AS |
836 | } |
837 | ||
838 | /* | |
839 | * zfpm_route_for_update | |
840 | * | |
f0f77c9a | 841 | * Returns the re that is to be sent to the FPM for a given dest. |
5adc2528 | 842 | */ |
d62a17ae | 843 | struct route_entry *zfpm_route_for_update(rib_dest_t *dest) |
5adc2528 | 844 | { |
5f7a4718 | 845 | return dest->selected_fib; |
5adc2528 AS |
846 | } |
847 | ||
848 | /* | |
849 | * zfpm_build_updates | |
850 | * | |
851 | * Process the outgoing queue and write messages to the outbound | |
852 | * buffer. | |
853 | */ | |
d62a17ae | 854 | static void zfpm_build_updates(void) |
5adc2528 | 855 | { |
d62a17ae | 856 | struct stream *s; |
857 | rib_dest_t *dest; | |
858 | unsigned char *buf, *data, *buf_end; | |
859 | size_t msg_len; | |
860 | size_t data_len; | |
861 | fpm_msg_hdr_t *hdr; | |
862 | struct route_entry *re; | |
863 | int is_add, write_msg; | |
864 | fpm_msg_type_e msg_type; | |
865 | ||
866 | s = zfpm_g->obuf; | |
867 | ||
868 | assert(stream_empty(s)); | |
869 | ||
870 | do { | |
871 | ||
872 | /* | |
873 | * Make sure there is enough space to write another message. | |
874 | */ | |
875 | if (STREAM_WRITEABLE(s) < FPM_MAX_MSG_LEN) | |
876 | break; | |
877 | ||
878 | buf = STREAM_DATA(s) + stream_get_endp(s); | |
879 | buf_end = buf + STREAM_WRITEABLE(s); | |
880 | ||
881 | dest = TAILQ_FIRST(&zfpm_g->dest_q); | |
882 | if (!dest) | |
883 | break; | |
884 | ||
885 | assert(CHECK_FLAG(dest->flags, RIB_DEST_UPDATE_FPM)); | |
886 | ||
887 | hdr = (fpm_msg_hdr_t *)buf; | |
888 | hdr->version = FPM_PROTO_VERSION; | |
889 | ||
890 | data = fpm_msg_data(hdr); | |
891 | ||
892 | re = zfpm_route_for_update(dest); | |
893 | is_add = re ? 1 : 0; | |
894 | ||
895 | write_msg = 1; | |
896 | ||
897 | /* | |
898 | * If this is a route deletion, and we have not sent the route | |
899 | * to | |
900 | * the FPM previously, skip it. | |
901 | */ | |
902 | if (!is_add && !CHECK_FLAG(dest->flags, RIB_DEST_SENT_TO_FPM)) { | |
903 | write_msg = 0; | |
904 | zfpm_g->stats.nop_deletes_skipped++; | |
905 | } | |
906 | ||
907 | if (write_msg) { | |
908 | data_len = zfpm_encode_route(dest, re, (char *)data, | |
909 | buf_end - data, &msg_type); | |
910 | ||
911 | assert(data_len); | |
912 | if (data_len) { | |
913 | hdr->msg_type = msg_type; | |
914 | msg_len = fpm_data_len_to_msg_len(data_len); | |
915 | hdr->msg_len = htons(msg_len); | |
916 | stream_forward_endp(s, msg_len); | |
917 | ||
918 | if (is_add) | |
919 | zfpm_g->stats.route_adds++; | |
920 | else | |
921 | zfpm_g->stats.route_dels++; | |
922 | } | |
923 | } | |
924 | ||
925 | /* | |
926 | * Remove the dest from the queue, and reset the flag. | |
927 | */ | |
928 | UNSET_FLAG(dest->flags, RIB_DEST_UPDATE_FPM); | |
929 | TAILQ_REMOVE(&zfpm_g->dest_q, dest, fpm_q_entries); | |
930 | ||
931 | if (is_add) { | |
932 | SET_FLAG(dest->flags, RIB_DEST_SENT_TO_FPM); | |
933 | } else { | |
934 | UNSET_FLAG(dest->flags, RIB_DEST_SENT_TO_FPM); | |
935 | } | |
936 | ||
937 | /* | |
938 | * Delete the destination if necessary. | |
939 | */ | |
940 | if (rib_gc_dest(dest->rnode)) | |
941 | zfpm_g->stats.dests_del_after_update++; | |
942 | ||
943 | } while (1); | |
5adc2528 AS |
944 | } |
945 | ||
946 | /* | |
947 | * zfpm_write_cb | |
948 | */ | |
d62a17ae | 949 | static int zfpm_write_cb(struct thread *thread) |
5adc2528 | 950 | { |
d62a17ae | 951 | struct stream *s; |
952 | int num_writes; | |
953 | ||
954 | zfpm_g->stats.write_cb_calls++; | |
955 | zfpm_g->t_write = NULL; | |
956 | ||
957 | /* | |
958 | * Check if async connect is now done. | |
959 | */ | |
960 | if (zfpm_g->state == ZFPM_STATE_CONNECTING) { | |
961 | zfpm_connect_check(); | |
962 | return 0; | |
963 | } | |
5adc2528 | 964 | |
d62a17ae | 965 | assert(zfpm_g->state == ZFPM_STATE_ESTABLISHED); |
966 | assert(zfpm_g->sock >= 0); | |
5adc2528 | 967 | |
d62a17ae | 968 | num_writes = 0; |
5adc2528 | 969 | |
d62a17ae | 970 | do { |
971 | int bytes_to_write, bytes_written; | |
5adc2528 | 972 | |
d62a17ae | 973 | s = zfpm_g->obuf; |
5adc2528 | 974 | |
d62a17ae | 975 | /* |
976 | * If the stream is empty, try fill it up with data. | |
977 | */ | |
978 | if (stream_empty(s)) { | |
979 | zfpm_build_updates(); | |
980 | } | |
5adc2528 | 981 | |
d62a17ae | 982 | bytes_to_write = stream_get_endp(s) - stream_get_getp(s); |
983 | if (!bytes_to_write) | |
984 | break; | |
5adc2528 | 985 | |
d62a17ae | 986 | bytes_written = |
2d34fb80 | 987 | write(zfpm_g->sock, stream_pnt(s), bytes_to_write); |
d62a17ae | 988 | zfpm_g->stats.write_calls++; |
989 | num_writes++; | |
5adc2528 | 990 | |
d62a17ae | 991 | if (bytes_written < 0) { |
992 | if (ERRNO_IO_RETRY(errno)) | |
993 | break; | |
5adc2528 | 994 | |
d62a17ae | 995 | zfpm_connection_down("failed to write to socket"); |
996 | return 0; | |
997 | } | |
5adc2528 | 998 | |
d62a17ae | 999 | if (bytes_written != bytes_to_write) { |
5adc2528 | 1000 | |
d62a17ae | 1001 | /* |
1002 | * Partial write. | |
1003 | */ | |
1004 | stream_forward_getp(s, bytes_written); | |
1005 | zfpm_g->stats.partial_writes++; | |
1006 | break; | |
1007 | } | |
5adc2528 | 1008 | |
d62a17ae | 1009 | /* |
1010 | * We've written out the entire contents of the stream. | |
1011 | */ | |
1012 | stream_reset(s); | |
5adc2528 | 1013 | |
d62a17ae | 1014 | if (num_writes >= ZFPM_MAX_WRITES_PER_RUN) { |
1015 | zfpm_g->stats.max_writes_hit++; | |
1016 | break; | |
1017 | } | |
5adc2528 | 1018 | |
d62a17ae | 1019 | if (zfpm_thread_should_yield(thread)) { |
1020 | zfpm_g->stats.t_write_yields++; | |
1021 | break; | |
1022 | } | |
1023 | } while (1); | |
5adc2528 | 1024 | |
d62a17ae | 1025 | if (zfpm_writes_pending()) |
1026 | zfpm_write_on(); | |
5adc2528 | 1027 | |
d62a17ae | 1028 | return 0; |
5adc2528 AS |
1029 | } |
1030 | ||
1031 | /* | |
1032 | * zfpm_connect_cb | |
1033 | */ | |
d62a17ae | 1034 | static int zfpm_connect_cb(struct thread *t) |
5adc2528 | 1035 | { |
d62a17ae | 1036 | int sock, ret; |
1037 | struct sockaddr_in serv; | |
1038 | ||
1039 | zfpm_g->t_connect = NULL; | |
1040 | assert(zfpm_g->state == ZFPM_STATE_ACTIVE); | |
1041 | ||
1042 | sock = socket(AF_INET, SOCK_STREAM, 0); | |
1043 | if (sock < 0) { | |
1044 | zfpm_debug("Failed to create socket for connect(): %s", | |
1045 | strerror(errno)); | |
1046 | zfpm_g->stats.connect_no_sock++; | |
1047 | return 0; | |
1048 | } | |
1049 | ||
1050 | set_nonblocking(sock); | |
1051 | ||
1052 | /* Make server socket. */ | |
1053 | memset(&serv, 0, sizeof(serv)); | |
1054 | serv.sin_family = AF_INET; | |
1055 | serv.sin_port = htons(zfpm_g->fpm_port); | |
5adc2528 | 1056 | #ifdef HAVE_STRUCT_SOCKADDR_IN_SIN_LEN |
d62a17ae | 1057 | serv.sin_len = sizeof(struct sockaddr_in); |
5adc2528 | 1058 | #endif /* HAVE_STRUCT_SOCKADDR_IN_SIN_LEN */ |
d62a17ae | 1059 | if (!zfpm_g->fpm_server) |
1060 | serv.sin_addr.s_addr = htonl(INADDR_LOOPBACK); | |
1061 | else | |
1062 | serv.sin_addr.s_addr = (zfpm_g->fpm_server); | |
1063 | ||
1064 | /* | |
1065 | * Connect to the FPM. | |
1066 | */ | |
1067 | zfpm_g->connect_calls++; | |
1068 | zfpm_g->stats.connect_calls++; | |
1069 | zfpm_g->last_connect_call_time = monotime(NULL); | |
1070 | ||
1071 | ret = connect(sock, (struct sockaddr *)&serv, sizeof(serv)); | |
1072 | if (ret >= 0) { | |
1073 | zfpm_g->sock = sock; | |
1074 | zfpm_connection_up("connect succeeded"); | |
1075 | return 1; | |
1076 | } | |
1077 | ||
1078 | if (errno == EINPROGRESS) { | |
1079 | zfpm_g->sock = sock; | |
1080 | zfpm_read_on(); | |
1081 | zfpm_write_on(); | |
1082 | zfpm_set_state(ZFPM_STATE_CONNECTING, | |
1083 | "async connect in progress"); | |
1084 | return 0; | |
1085 | } | |
1086 | ||
1087 | zlog_info("can't connect to FPM %d: %s", sock, safe_strerror(errno)); | |
1088 | close(sock); | |
1089 | ||
1090 | /* | |
1091 | * Restart timer for retrying connection. | |
1092 | */ | |
1093 | zfpm_start_connect_timer("connect() failed"); | |
1094 | return 0; | |
5adc2528 AS |
1095 | } |
1096 | ||
1097 | /* | |
1098 | * zfpm_set_state | |
1099 | * | |
1100 | * Move state machine into the given state. | |
1101 | */ | |
d62a17ae | 1102 | static void zfpm_set_state(zfpm_state_t state, const char *reason) |
5adc2528 | 1103 | { |
d62a17ae | 1104 | zfpm_state_t cur_state = zfpm_g->state; |
1105 | ||
1106 | if (!reason) | |
1107 | reason = "Unknown"; | |
1108 | ||
1109 | if (state == cur_state) | |
1110 | return; | |
1111 | ||
1112 | zfpm_debug("beginning state transition %s -> %s. Reason: %s", | |
1113 | zfpm_state_to_str(cur_state), zfpm_state_to_str(state), | |
1114 | reason); | |
1115 | ||
1116 | switch (state) { | |
1117 | ||
1118 | case ZFPM_STATE_IDLE: | |
1119 | assert(cur_state == ZFPM_STATE_ESTABLISHED); | |
1120 | break; | |
1121 | ||
1122 | case ZFPM_STATE_ACTIVE: | |
1123 | assert(cur_state == ZFPM_STATE_IDLE | |
1124 | || cur_state == ZFPM_STATE_CONNECTING); | |
1125 | assert(zfpm_g->t_connect); | |
1126 | break; | |
1127 | ||
1128 | case ZFPM_STATE_CONNECTING: | |
1129 | assert(zfpm_g->sock); | |
1130 | assert(cur_state == ZFPM_STATE_ACTIVE); | |
1131 | assert(zfpm_g->t_read); | |
1132 | assert(zfpm_g->t_write); | |
1133 | break; | |
1134 | ||
1135 | case ZFPM_STATE_ESTABLISHED: | |
1136 | assert(cur_state == ZFPM_STATE_ACTIVE | |
1137 | || cur_state == ZFPM_STATE_CONNECTING); | |
1138 | assert(zfpm_g->sock); | |
1139 | assert(zfpm_g->t_read); | |
1140 | assert(zfpm_g->t_write); | |
1141 | break; | |
1142 | } | |
1143 | ||
1144 | zfpm_g->state = state; | |
5adc2528 AS |
1145 | } |
1146 | ||
1147 | /* | |
1148 | * zfpm_calc_connect_delay | |
1149 | * | |
1150 | * Returns the number of seconds after which we should attempt to | |
1151 | * reconnect to the FPM. | |
1152 | */ | |
d62a17ae | 1153 | static long zfpm_calc_connect_delay(void) |
5adc2528 | 1154 | { |
d62a17ae | 1155 | time_t elapsed; |
5adc2528 | 1156 | |
d62a17ae | 1157 | /* |
1158 | * Return 0 if this is our first attempt to connect. | |
1159 | */ | |
1160 | if (zfpm_g->connect_calls == 0) { | |
1161 | return 0; | |
1162 | } | |
5adc2528 | 1163 | |
d62a17ae | 1164 | elapsed = zfpm_get_elapsed_time(zfpm_g->last_connect_call_time); |
5adc2528 | 1165 | |
d62a17ae | 1166 | if (elapsed > ZFPM_CONNECT_RETRY_IVL) { |
1167 | return 0; | |
1168 | } | |
5adc2528 | 1169 | |
d62a17ae | 1170 | return ZFPM_CONNECT_RETRY_IVL - elapsed; |
5adc2528 AS |
1171 | } |
1172 | ||
1173 | /* | |
1174 | * zfpm_start_connect_timer | |
1175 | */ | |
d62a17ae | 1176 | static void zfpm_start_connect_timer(const char *reason) |
5adc2528 | 1177 | { |
d62a17ae | 1178 | long delay_secs; |
5adc2528 | 1179 | |
d62a17ae | 1180 | assert(!zfpm_g->t_connect); |
1181 | assert(zfpm_g->sock < 0); | |
5adc2528 | 1182 | |
d62a17ae | 1183 | assert(zfpm_g->state == ZFPM_STATE_IDLE |
1184 | || zfpm_g->state == ZFPM_STATE_ACTIVE | |
1185 | || zfpm_g->state == ZFPM_STATE_CONNECTING); | |
5adc2528 | 1186 | |
d62a17ae | 1187 | delay_secs = zfpm_calc_connect_delay(); |
1188 | zfpm_debug("scheduling connect in %ld seconds", delay_secs); | |
5adc2528 | 1189 | |
d62a17ae | 1190 | thread_add_timer(zfpm_g->master, zfpm_connect_cb, 0, delay_secs, |
1191 | &zfpm_g->t_connect); | |
1192 | zfpm_set_state(ZFPM_STATE_ACTIVE, reason); | |
5adc2528 AS |
1193 | } |
1194 | ||
1195 | /* | |
1196 | * zfpm_is_enabled | |
1197 | * | |
1198 | * Returns TRUE if the zebra FPM module has been enabled. | |
1199 | */ | |
d62a17ae | 1200 | static inline int zfpm_is_enabled(void) |
5adc2528 | 1201 | { |
d62a17ae | 1202 | return zfpm_g->enabled; |
5adc2528 AS |
1203 | } |
1204 | ||
1205 | /* | |
1206 | * zfpm_conn_is_up | |
1207 | * | |
1208 | * Returns TRUE if the connection to the FPM is up. | |
1209 | */ | |
d62a17ae | 1210 | static inline int zfpm_conn_is_up(void) |
5adc2528 | 1211 | { |
d62a17ae | 1212 | if (zfpm_g->state != ZFPM_STATE_ESTABLISHED) |
1213 | return 0; | |
5adc2528 | 1214 | |
d62a17ae | 1215 | assert(zfpm_g->sock >= 0); |
5adc2528 | 1216 | |
d62a17ae | 1217 | return 1; |
5adc2528 AS |
1218 | } |
1219 | ||
1220 | /* | |
1221 | * zfpm_trigger_update | |
1222 | * | |
1223 | * The zebra code invokes this function to indicate that we should | |
1224 | * send an update to the FPM about the given route_node. | |
1225 | */ | |
d62a17ae | 1226 | static int zfpm_trigger_update(struct route_node *rn, const char *reason) |
5adc2528 | 1227 | { |
d62a17ae | 1228 | rib_dest_t *dest; |
1229 | char buf[PREFIX_STRLEN]; | |
1230 | ||
1231 | /* | |
1232 | * Ignore if the connection is down. We will update the FPM about | |
1233 | * all destinations once the connection comes up. | |
1234 | */ | |
1235 | if (!zfpm_conn_is_up()) | |
1236 | return 0; | |
1237 | ||
1238 | dest = rib_dest_from_rnode(rn); | |
1239 | ||
d62a17ae | 1240 | if (CHECK_FLAG(dest->flags, RIB_DEST_UPDATE_FPM)) { |
1241 | zfpm_g->stats.redundant_triggers++; | |
1242 | return 0; | |
1243 | } | |
1244 | ||
1245 | if (reason) { | |
1246 | zfpm_debug("%s triggering update to FPM - Reason: %s", | |
1247 | prefix2str(&rn->p, buf, sizeof(buf)), reason); | |
1248 | } | |
1249 | ||
1250 | SET_FLAG(dest->flags, RIB_DEST_UPDATE_FPM); | |
1251 | TAILQ_INSERT_TAIL(&zfpm_g->dest_q, dest, fpm_q_entries); | |
1252 | zfpm_g->stats.updates_triggered++; | |
1253 | ||
1254 | /* | |
1255 | * Make sure that writes are enabled. | |
1256 | */ | |
1257 | if (zfpm_g->t_write) | |
1258 | return 0; | |
1259 | ||
1260 | zfpm_write_on(); | |
1261 | return 0; | |
5adc2528 AS |
1262 | } |
1263 | ||
1264 | /* | |
1265 | * zfpm_stats_timer_cb | |
1266 | */ | |
d62a17ae | 1267 | static int zfpm_stats_timer_cb(struct thread *t) |
5adc2528 | 1268 | { |
d62a17ae | 1269 | zfpm_g->t_stats = NULL; |
5adc2528 | 1270 | |
d62a17ae | 1271 | /* |
1272 | * Remember the stats collected in the last interval for display | |
1273 | * purposes. | |
1274 | */ | |
1275 | zfpm_stats_copy(&zfpm_g->stats, &zfpm_g->last_ivl_stats); | |
5adc2528 | 1276 | |
d62a17ae | 1277 | /* |
1278 | * Add the current set of stats into the cumulative statistics. | |
1279 | */ | |
1280 | zfpm_stats_compose(&zfpm_g->cumulative_stats, &zfpm_g->stats, | |
1281 | &zfpm_g->cumulative_stats); | |
5adc2528 | 1282 | |
d62a17ae | 1283 | /* |
1284 | * Start collecting stats afresh over the next interval. | |
1285 | */ | |
1286 | zfpm_stats_reset(&zfpm_g->stats); | |
5adc2528 | 1287 | |
d62a17ae | 1288 | zfpm_start_stats_timer(); |
5adc2528 | 1289 | |
d62a17ae | 1290 | return 0; |
5adc2528 AS |
1291 | } |
1292 | ||
1293 | /* | |
1294 | * zfpm_stop_stats_timer | |
1295 | */ | |
d62a17ae | 1296 | static void zfpm_stop_stats_timer(void) |
5adc2528 | 1297 | { |
d62a17ae | 1298 | if (!zfpm_g->t_stats) |
1299 | return; | |
5adc2528 | 1300 | |
d62a17ae | 1301 | zfpm_debug("Stopping existing stats timer"); |
1302 | THREAD_TIMER_OFF(zfpm_g->t_stats); | |
5adc2528 AS |
1303 | } |
1304 | ||
1305 | /* | |
1306 | * zfpm_start_stats_timer | |
1307 | */ | |
d62a17ae | 1308 | void zfpm_start_stats_timer(void) |
5adc2528 | 1309 | { |
d62a17ae | 1310 | assert(!zfpm_g->t_stats); |
5adc2528 | 1311 | |
d62a17ae | 1312 | thread_add_timer(zfpm_g->master, zfpm_stats_timer_cb, 0, |
1313 | ZFPM_STATS_IVL_SECS, &zfpm_g->t_stats); | |
5adc2528 AS |
1314 | } |
1315 | ||
1316 | /* | |
1317 | * Helper macro for zfpm_show_stats() below. | |
1318 | */ | |
d62a17ae | 1319 | #define ZFPM_SHOW_STAT(counter) \ |
1320 | do { \ | |
1321 | vty_out(vty, "%-40s %10lu %16lu\n", #counter, \ | |
1322 | total_stats.counter, zfpm_g->last_ivl_stats.counter); \ | |
1323 | } while (0) | |
5adc2528 AS |
1324 | |
1325 | /* | |
1326 | * zfpm_show_stats | |
1327 | */ | |
d62a17ae | 1328 | static void zfpm_show_stats(struct vty *vty) |
5adc2528 | 1329 | { |
d62a17ae | 1330 | zfpm_stats_t total_stats; |
1331 | time_t elapsed; | |
1332 | ||
1333 | vty_out(vty, "\n%-40s %10s Last %2d secs\n\n", "Counter", "Total", | |
1334 | ZFPM_STATS_IVL_SECS); | |
1335 | ||
1336 | /* | |
1337 | * Compute the total stats up to this instant. | |
1338 | */ | |
1339 | zfpm_stats_compose(&zfpm_g->cumulative_stats, &zfpm_g->stats, | |
1340 | &total_stats); | |
1341 | ||
1342 | ZFPM_SHOW_STAT(connect_calls); | |
1343 | ZFPM_SHOW_STAT(connect_no_sock); | |
1344 | ZFPM_SHOW_STAT(read_cb_calls); | |
1345 | ZFPM_SHOW_STAT(write_cb_calls); | |
1346 | ZFPM_SHOW_STAT(write_calls); | |
1347 | ZFPM_SHOW_STAT(partial_writes); | |
1348 | ZFPM_SHOW_STAT(max_writes_hit); | |
1349 | ZFPM_SHOW_STAT(t_write_yields); | |
1350 | ZFPM_SHOW_STAT(nop_deletes_skipped); | |
1351 | ZFPM_SHOW_STAT(route_adds); | |
1352 | ZFPM_SHOW_STAT(route_dels); | |
1353 | ZFPM_SHOW_STAT(updates_triggered); | |
d62a17ae | 1354 | ZFPM_SHOW_STAT(redundant_triggers); |
1355 | ZFPM_SHOW_STAT(dests_del_after_update); | |
1356 | ZFPM_SHOW_STAT(t_conn_down_starts); | |
1357 | ZFPM_SHOW_STAT(t_conn_down_dests_processed); | |
1358 | ZFPM_SHOW_STAT(t_conn_down_yields); | |
1359 | ZFPM_SHOW_STAT(t_conn_down_finishes); | |
1360 | ZFPM_SHOW_STAT(t_conn_up_starts); | |
1361 | ZFPM_SHOW_STAT(t_conn_up_dests_processed); | |
1362 | ZFPM_SHOW_STAT(t_conn_up_yields); | |
1363 | ZFPM_SHOW_STAT(t_conn_up_aborts); | |
1364 | ZFPM_SHOW_STAT(t_conn_up_finishes); | |
1365 | ||
1366 | if (!zfpm_g->last_stats_clear_time) | |
1367 | return; | |
1368 | ||
1369 | elapsed = zfpm_get_elapsed_time(zfpm_g->last_stats_clear_time); | |
1370 | ||
1371 | vty_out(vty, "\nStats were cleared %lu seconds ago\n", | |
1372 | (unsigned long)elapsed); | |
5adc2528 AS |
1373 | } |
1374 | ||
1375 | /* | |
1376 | * zfpm_clear_stats | |
1377 | */ | |
d62a17ae | 1378 | static void zfpm_clear_stats(struct vty *vty) |
5adc2528 | 1379 | { |
d62a17ae | 1380 | if (!zfpm_is_enabled()) { |
1381 | vty_out(vty, "The FPM module is not enabled...\n"); | |
1382 | return; | |
1383 | } | |
5adc2528 | 1384 | |
d62a17ae | 1385 | zfpm_stats_reset(&zfpm_g->stats); |
1386 | zfpm_stats_reset(&zfpm_g->last_ivl_stats); | |
1387 | zfpm_stats_reset(&zfpm_g->cumulative_stats); | |
5adc2528 | 1388 | |
d62a17ae | 1389 | zfpm_stop_stats_timer(); |
1390 | zfpm_start_stats_timer(); | |
5adc2528 | 1391 | |
d62a17ae | 1392 | zfpm_g->last_stats_clear_time = monotime(NULL); |
5adc2528 | 1393 | |
d62a17ae | 1394 | vty_out(vty, "Cleared FPM stats\n"); |
5adc2528 AS |
1395 | } |
1396 | ||
1397 | /* | |
1398 | * show_zebra_fpm_stats | |
1399 | */ | |
1400 | DEFUN (show_zebra_fpm_stats, | |
1401 | show_zebra_fpm_stats_cmd, | |
1402 | "show zebra fpm stats", | |
1403 | SHOW_STR | |
41e7fb80 | 1404 | ZEBRA_STR |
5adc2528 AS |
1405 | "Forwarding Path Manager information\n" |
1406 | "Statistics\n") | |
1407 | { | |
d62a17ae | 1408 | zfpm_show_stats(vty); |
1409 | return CMD_SUCCESS; | |
5adc2528 AS |
1410 | } |
1411 | ||
1412 | /* | |
1413 | * clear_zebra_fpm_stats | |
1414 | */ | |
1415 | DEFUN (clear_zebra_fpm_stats, | |
1416 | clear_zebra_fpm_stats_cmd, | |
1417 | "clear zebra fpm stats", | |
1418 | CLEAR_STR | |
41e7fb80 | 1419 | ZEBRA_STR |
5adc2528 AS |
1420 | "Clear Forwarding Path Manager information\n" |
1421 | "Statistics\n") | |
1422 | { | |
d62a17ae | 1423 | zfpm_clear_stats(vty); |
1424 | return CMD_SUCCESS; | |
5adc2528 AS |
1425 | } |
1426 | ||
711ff0ba | 1427 | /* |
d62a17ae | 1428 | * update fpm connection information |
711ff0ba | 1429 | */ |
e52702f2 QY |
1430 | DEFUN ( fpm_remote_ip, |
1431 | fpm_remote_ip_cmd, | |
1432 | "fpm connection ip A.B.C.D port (1-65535)", | |
711ff0ba USK |
1433 | "fpm connection remote ip and port\n" |
1434 | "Remote fpm server ip A.B.C.D\n" | |
1435 | "Enter ip ") | |
1436 | { | |
1437 | ||
d62a17ae | 1438 | in_addr_t fpm_server; |
1439 | uint32_t port_no; | |
711ff0ba | 1440 | |
d62a17ae | 1441 | fpm_server = inet_addr(argv[3]->arg); |
1442 | if (fpm_server == INADDR_NONE) | |
1443 | return CMD_ERR_INCOMPLETE; | |
711ff0ba | 1444 | |
d62a17ae | 1445 | port_no = atoi(argv[5]->arg); |
1446 | if (port_no < TCP_MIN_PORT || port_no > TCP_MAX_PORT) | |
1447 | return CMD_ERR_INCOMPLETE; | |
711ff0ba | 1448 | |
d62a17ae | 1449 | zfpm_g->fpm_server = fpm_server; |
1450 | zfpm_g->fpm_port = port_no; | |
711ff0ba USK |
1451 | |
1452 | ||
d62a17ae | 1453 | return CMD_SUCCESS; |
711ff0ba USK |
1454 | } |
1455 | ||
e52702f2 QY |
1456 | DEFUN ( no_fpm_remote_ip, |
1457 | no_fpm_remote_ip_cmd, | |
1458 | "no fpm connection ip A.B.C.D port (1-65535)", | |
711ff0ba USK |
1459 | "fpm connection remote ip and port\n" |
1460 | "Connection\n" | |
1461 | "Remote fpm server ip A.B.C.D\n" | |
1462 | "Enter ip ") | |
1463 | { | |
d62a17ae | 1464 | if (zfpm_g->fpm_server != inet_addr(argv[4]->arg) |
1465 | || zfpm_g->fpm_port != atoi(argv[6]->arg)) | |
1466 | return CMD_ERR_NO_MATCH; | |
711ff0ba | 1467 | |
d62a17ae | 1468 | zfpm_g->fpm_server = FPM_DEFAULT_IP; |
1469 | zfpm_g->fpm_port = FPM_DEFAULT_PORT; | |
711ff0ba | 1470 | |
d62a17ae | 1471 | return CMD_SUCCESS; |
711ff0ba | 1472 | } |
711ff0ba | 1473 | |
fb0aa886 AS |
1474 | /* |
1475 | * zfpm_init_message_format | |
1476 | */ | |
d62a17ae | 1477 | static inline void zfpm_init_message_format(const char *format) |
fb0aa886 | 1478 | { |
d62a17ae | 1479 | int have_netlink, have_protobuf; |
fb0aa886 | 1480 | |
fb0aa886 | 1481 | #ifdef HAVE_NETLINK |
d62a17ae | 1482 | have_netlink = 1; |
4b2792b5 | 1483 | #else |
d62a17ae | 1484 | have_netlink = 0; |
fb0aa886 AS |
1485 | #endif |
1486 | ||
1487 | #ifdef HAVE_PROTOBUF | |
d62a17ae | 1488 | have_protobuf = 1; |
4b2792b5 | 1489 | #else |
d62a17ae | 1490 | have_protobuf = 0; |
fb0aa886 AS |
1491 | #endif |
1492 | ||
d62a17ae | 1493 | zfpm_g->message_format = ZFPM_MSG_FORMAT_NONE; |
fb0aa886 | 1494 | |
d62a17ae | 1495 | if (!format) { |
1496 | if (have_netlink) { | |
1497 | zfpm_g->message_format = ZFPM_MSG_FORMAT_NETLINK; | |
1498 | } else if (have_protobuf) { | |
1499 | zfpm_g->message_format = ZFPM_MSG_FORMAT_PROTOBUF; | |
1500 | } | |
1501 | return; | |
fb0aa886 | 1502 | } |
fb0aa886 | 1503 | |
d62a17ae | 1504 | if (!strcmp("netlink", format)) { |
1505 | if (!have_netlink) { | |
1506 | zlog_err("FPM netlink message format is not available"); | |
1507 | return; | |
1508 | } | |
1509 | zfpm_g->message_format = ZFPM_MSG_FORMAT_NETLINK; | |
1510 | return; | |
fb0aa886 | 1511 | } |
fb0aa886 | 1512 | |
d62a17ae | 1513 | if (!strcmp("protobuf", format)) { |
1514 | if (!have_protobuf) { | |
1515 | zlog_err( | |
1516 | "FPM protobuf message format is not available"); | |
1517 | return; | |
1518 | } | |
1519 | zfpm_g->message_format = ZFPM_MSG_FORMAT_PROTOBUF; | |
1520 | return; | |
fb0aa886 | 1521 | } |
fb0aa886 | 1522 | |
d62a17ae | 1523 | zlog_warn("Unknown fpm format '%s'", format); |
fb0aa886 AS |
1524 | } |
1525 | ||
711ff0ba | 1526 | /** |
d62a17ae | 1527 | * fpm_remote_srv_write |
711ff0ba | 1528 | * |
d62a17ae | 1529 | * Module to write remote fpm connection |
711ff0ba USK |
1530 | * |
1531 | * Returns ZERO on success. | |
1532 | */ | |
1533 | ||
d62a17ae | 1534 | static int fpm_remote_srv_write(struct vty *vty) |
711ff0ba | 1535 | { |
d62a17ae | 1536 | struct in_addr in; |
711ff0ba | 1537 | |
d62a17ae | 1538 | in.s_addr = zfpm_g->fpm_server; |
711ff0ba | 1539 | |
9d1c2659 DL |
1540 | if ((zfpm_g->fpm_server != FPM_DEFAULT_IP |
1541 | && zfpm_g->fpm_server != INADDR_ANY) | |
1542 | || (zfpm_g->fpm_port != FPM_DEFAULT_PORT | |
1543 | && zfpm_g->fpm_port != 0)) | |
d62a17ae | 1544 | vty_out(vty, "fpm connection ip %s port %d\n", inet_ntoa(in), |
1545 | zfpm_g->fpm_port); | |
711ff0ba | 1546 | |
d62a17ae | 1547 | return 0; |
711ff0ba USK |
1548 | } |
1549 | ||
1550 | ||
4f8ea50c | 1551 | /* Zebra node */ |
d62a17ae | 1552 | static struct cmd_node zebra_node = {ZEBRA_NODE, "", 1}; |
4f8ea50c DL |
1553 | |
1554 | ||
5adc2528 AS |
1555 | /** |
1556 | * zfpm_init | |
1557 | * | |
1558 | * One-time initialization of the Zebra FPM module. | |
1559 | * | |
1560 | * @param[in] port port at which FPM is running. | |
1561 | * @param[in] enable TRUE if the zebra FPM module should be enabled | |
fb0aa886 | 1562 | * @param[in] format to use to talk to the FPM. Can be 'netink' or 'protobuf'. |
5adc2528 AS |
1563 | * |
1564 | * Returns TRUE on success. | |
1565 | */ | |
d62a17ae | 1566 | static int zfpm_init(struct thread_master *master) |
5adc2528 | 1567 | { |
d62a17ae | 1568 | int enable = 1; |
1569 | uint16_t port = 0; | |
1570 | const char *format = THIS_MODULE->load_args; | |
5adc2528 | 1571 | |
d62a17ae | 1572 | memset(zfpm_g, 0, sizeof(*zfpm_g)); |
1573 | zfpm_g->master = master; | |
1574 | TAILQ_INIT(&zfpm_g->dest_q); | |
1575 | zfpm_g->sock = -1; | |
1576 | zfpm_g->state = ZFPM_STATE_IDLE; | |
5adc2528 | 1577 | |
d62a17ae | 1578 | zfpm_stats_init(&zfpm_g->stats); |
1579 | zfpm_stats_init(&zfpm_g->last_ivl_stats); | |
1580 | zfpm_stats_init(&zfpm_g->cumulative_stats); | |
5adc2528 | 1581 | |
d62a17ae | 1582 | install_node(&zebra_node, fpm_remote_srv_write); |
1583 | install_element(ENABLE_NODE, &show_zebra_fpm_stats_cmd); | |
1584 | install_element(ENABLE_NODE, &clear_zebra_fpm_stats_cmd); | |
1585 | install_element(CONFIG_NODE, &fpm_remote_ip_cmd); | |
1586 | install_element(CONFIG_NODE, &no_fpm_remote_ip_cmd); | |
5adc2528 | 1587 | |
d62a17ae | 1588 | zfpm_init_message_format(format); |
fb0aa886 | 1589 | |
d62a17ae | 1590 | /* |
1591 | * Disable FPM interface if no suitable format is available. | |
1592 | */ | |
1593 | if (zfpm_g->message_format == ZFPM_MSG_FORMAT_NONE) | |
1594 | enable = 0; | |
fb0aa886 | 1595 | |
d62a17ae | 1596 | zfpm_g->enabled = enable; |
5adc2528 | 1597 | |
d62a17ae | 1598 | if (!zfpm_g->fpm_server) |
1599 | zfpm_g->fpm_server = FPM_DEFAULT_IP; | |
711ff0ba | 1600 | |
d62a17ae | 1601 | if (!port) |
1602 | port = FPM_DEFAULT_PORT; | |
5adc2528 | 1603 | |
d62a17ae | 1604 | zfpm_g->fpm_port = port; |
5adc2528 | 1605 | |
d62a17ae | 1606 | zfpm_g->obuf = stream_new(ZFPM_OBUF_SIZE); |
1607 | zfpm_g->ibuf = stream_new(ZFPM_IBUF_SIZE); | |
5adc2528 | 1608 | |
d62a17ae | 1609 | zfpm_start_stats_timer(); |
1610 | zfpm_start_connect_timer("initialized"); | |
1611 | return 0; | |
4f8ea50c | 1612 | } |
5adc2528 | 1613 | |
d62a17ae | 1614 | static int zebra_fpm_module_init(void) |
4f8ea50c | 1615 | { |
d62a17ae | 1616 | hook_register(rib_update, zfpm_trigger_update); |
1617 | hook_register(frr_late_init, zfpm_init); | |
1618 | return 0; | |
5adc2528 | 1619 | } |
4f8ea50c | 1620 | |
d62a17ae | 1621 | FRR_MODULE_SETUP(.name = "zebra_fpm", .version = FRR_VERSION, |
1622 | .description = "zebra FPM (Forwarding Plane Manager) module", | |
1623 | .init = zebra_fpm_module_init, ) |