Commit | Line | Data |
---|---|---|
b8aa1682 JD |
1 | /* |
2 | * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com> | |
3 | * David Goulet <dgoulet@efficios.com> | |
cd60b05a | 4 | * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com> |
b8aa1682 JD |
5 | * |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License, version 2 only, | |
8 | * as published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, but WITHOUT | |
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
13 | * more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License along | |
16 | * with this program; if not, write to the Free Software Foundation, Inc., | |
17 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. | |
18 | */ | |
19 | ||
20 | #define _GNU_SOURCE | |
21 | #include <getopt.h> | |
22 | #include <grp.h> | |
23 | #include <limits.h> | |
24 | #include <pthread.h> | |
25 | #include <signal.h> | |
26 | #include <stdio.h> | |
27 | #include <stdlib.h> | |
28 | #include <string.h> | |
29 | #include <sys/mman.h> | |
30 | #include <sys/mount.h> | |
31 | #include <sys/resource.h> | |
32 | #include <sys/socket.h> | |
33 | #include <sys/stat.h> | |
34 | #include <sys/types.h> | |
35 | #include <sys/wait.h> | |
173af62f | 36 | #include <inttypes.h> |
b8aa1682 JD |
37 | #include <urcu/futex.h> |
38 | #include <urcu/uatomic.h> | |
39 | #include <unistd.h> | |
40 | #include <fcntl.h> | |
41 | #include <config.h> | |
42 | ||
43 | #include <lttng/lttng.h> | |
44 | #include <common/common.h> | |
45 | #include <common/compat/poll.h> | |
46 | #include <common/compat/socket.h> | |
47 | #include <common/defaults.h> | |
48 | #include <common/futex.h> | |
49 | #include <common/sessiond-comm/sessiond-comm.h> | |
50 | #include <common/sessiond-comm/inet.h> | |
b8aa1682 JD |
51 | #include <common/sessiond-comm/relayd.h> |
52 | #include <common/uri.h> | |
a02de639 | 53 | #include <common/utils.h> |
cd60b05a | 54 | #include <common/config/config.h> |
b8aa1682 | 55 | |
0f907de1 | 56 | #include "cmd.h" |
d3e2ba59 | 57 | #include "ctf-trace.h" |
1c20f0e2 | 58 | #include "index.h" |
0f907de1 | 59 | #include "utils.h" |
b8aa1682 | 60 | #include "lttng-relayd.h" |
d3e2ba59 | 61 | #include "live.h" |
55706a7d | 62 | #include "health-relayd.h" |
b8aa1682 JD |
63 | |
64 | /* command line options */ | |
0f907de1 | 65 | char *opt_output_path; |
b5218ffb | 66 | static int opt_daemon, opt_background; |
095a4ae5 MD |
67 | static struct lttng_uri *control_uri; |
68 | static struct lttng_uri *data_uri; | |
d3e2ba59 | 69 | static struct lttng_uri *live_uri; |
b8aa1682 JD |
70 | |
71 | const char *progname; | |
b8aa1682 | 72 | |
65931c8b | 73 | const char *tracing_group_name = DEFAULT_TRACING_GROUP; |
cd60b05a JG |
74 | static int tracing_group_name_override; |
75 | ||
76 | const char * const config_section_name = "relayd"; | |
65931c8b | 77 | |
b8aa1682 JD |
78 | /* |
79 | * Quit pipe for all threads. This permits a single cancellation point | |
80 | * for all threads when receiving an event on the pipe. | |
81 | */ | |
82 | static int thread_quit_pipe[2] = { -1, -1 }; | |
83 | ||
84 | /* | |
85 | * This pipe is used to inform the worker thread that a command is queued and | |
86 | * ready to be processed. | |
87 | */ | |
88 | static int relay_cmd_pipe[2] = { -1, -1 }; | |
89 | ||
26c9d55e | 90 | /* Shared between threads */ |
b8aa1682 JD |
91 | static int dispatch_thread_exit; |
92 | ||
93 | static pthread_t listener_thread; | |
94 | static pthread_t dispatcher_thread; | |
95 | static pthread_t worker_thread; | |
65931c8b | 96 | static pthread_t health_thread; |
b8aa1682 | 97 | |
095a4ae5 MD |
98 | static uint64_t last_relay_stream_id; |
99 | static uint64_t last_relay_session_id; | |
b8aa1682 JD |
100 | |
101 | /* | |
102 | * Relay command queue. | |
103 | * | |
104 | * The relay_thread_listener and relay_thread_dispatcher communicate with this | |
105 | * queue. | |
106 | */ | |
107 | static struct relay_cmd_queue relay_cmd_queue; | |
108 | ||
109 | /* buffer allocated at startup, used to store the trace data */ | |
095a4ae5 MD |
110 | static char *data_buffer; |
111 | static unsigned int data_buffer_size; | |
b8aa1682 | 112 | |
1c20f0e2 JD |
113 | /* We need those values for the file/dir creation. */ |
114 | static uid_t relayd_uid; | |
115 | static gid_t relayd_gid; | |
116 | ||
d3e2ba59 JD |
117 | /* Global relay stream hash table. */ |
118 | struct lttng_ht *relay_streams_ht; | |
119 | ||
92c6ca54 DG |
120 | /* Global relay viewer stream hash table. */ |
121 | struct lttng_ht *viewer_streams_ht; | |
122 | ||
0a6518b0 DG |
123 | /* Global hash table that stores relay index object. */ |
124 | struct lttng_ht *indexes_ht; | |
125 | ||
55706a7d | 126 | /* Relayd health monitoring */ |
eea7556c | 127 | struct health_app *health_relayd; |
55706a7d | 128 | |
cd60b05a JG |
129 | static struct option long_options[] = { |
130 | { "control-port", 1, 0, 'C', }, | |
131 | { "data-port", 1, 0, 'D', }, | |
8d5c808e | 132 | { "live-port", 1, 0, 'L', }, |
cd60b05a | 133 | { "daemonize", 0, 0, 'd', }, |
b5218ffb | 134 | { "background", 0, 0, 'b', }, |
cd60b05a JG |
135 | { "group", 1, 0, 'g', }, |
136 | { "help", 0, 0, 'h', }, | |
137 | { "output", 1, 0, 'o', }, | |
138 | { "verbose", 0, 0, 'v', }, | |
139 | { "config", 1, 0, 'f' }, | |
140 | { NULL, 0, 0, 0, }, | |
141 | }; | |
142 | ||
143 | static const char *config_ignore_options[] = { "help", "config" }; | |
144 | ||
b8aa1682 JD |
145 | /* |
146 | * usage function on stderr | |
147 | */ | |
148 | static | |
149 | void usage(void) | |
150 | { | |
151 | fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname); | |
994fa64f DG |
152 | fprintf(stderr, " -h, --help Display this usage.\n"); |
153 | fprintf(stderr, " -d, --daemonize Start as a daemon.\n"); | |
b5218ffb | 154 | fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n"); |
994fa64f DG |
155 | fprintf(stderr, " -C, --control-port URL Control port listening.\n"); |
156 | fprintf(stderr, " -D, --data-port URL Data port listening.\n"); | |
8d5c808e | 157 | fprintf(stderr, " -L, --live-port URL Live view port listening.\n"); |
994fa64f DG |
158 | fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n"); |
159 | fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n"); | |
65931c8b | 160 | fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n"); |
cd60b05a | 161 | fprintf(stderr, " -f --config Load daemon configuration file\n"); |
b8aa1682 JD |
162 | } |
163 | ||
cd60b05a JG |
164 | /* |
165 | * Take an option from the getopt output and set it in the right variable to be | |
166 | * used later. | |
167 | * | |
168 | * Return 0 on success else a negative value. | |
169 | */ | |
b8aa1682 | 170 | static |
cd60b05a | 171 | int set_option(int opt, const char *arg, const char *optname) |
b8aa1682 | 172 | { |
cd60b05a JG |
173 | int ret; |
174 | ||
175 | switch (opt) { | |
176 | case 0: | |
177 | fprintf(stderr, "option %s", optname); | |
178 | if (arg) { | |
179 | fprintf(stderr, " with arg %s\n", arg); | |
180 | } | |
181 | break; | |
182 | case 'C': | |
183 | ret = uri_parse(arg, &control_uri); | |
184 | if (ret < 0) { | |
185 | ERR("Invalid control URI specified"); | |
186 | goto end; | |
187 | } | |
188 | if (control_uri->port == 0) { | |
189 | control_uri->port = DEFAULT_NETWORK_CONTROL_PORT; | |
190 | } | |
191 | break; | |
192 | case 'D': | |
193 | ret = uri_parse(arg, &data_uri); | |
194 | if (ret < 0) { | |
195 | ERR("Invalid data URI specified"); | |
196 | goto end; | |
197 | } | |
198 | if (data_uri->port == 0) { | |
199 | data_uri->port = DEFAULT_NETWORK_DATA_PORT; | |
200 | } | |
201 | break; | |
8d5c808e AM |
202 | case 'L': |
203 | ret = uri_parse(arg, &live_uri); | |
204 | if (ret < 0) { | |
205 | ERR("Invalid live URI specified"); | |
206 | goto end; | |
207 | } | |
208 | if (live_uri->port == 0) { | |
209 | live_uri->port = DEFAULT_NETWORK_VIEWER_PORT; | |
210 | } | |
211 | break; | |
cd60b05a JG |
212 | case 'd': |
213 | opt_daemon = 1; | |
214 | break; | |
b5218ffb MD |
215 | case 'b': |
216 | opt_background = 1; | |
217 | break; | |
cd60b05a JG |
218 | case 'g': |
219 | tracing_group_name = strdup(arg); | |
220 | tracing_group_name_override = 1; | |
221 | break; | |
222 | case 'h': | |
223 | usage(); | |
224 | exit(EXIT_FAILURE); | |
225 | case 'o': | |
226 | ret = asprintf(&opt_output_path, "%s", arg); | |
227 | if (ret < 0) { | |
228 | ret = -errno; | |
229 | PERROR("asprintf opt_output_path"); | |
230 | goto end; | |
231 | } | |
232 | break; | |
233 | case 'v': | |
234 | /* Verbose level can increase using multiple -v */ | |
235 | if (arg) { | |
236 | lttng_opt_verbose = config_parse_value(arg); | |
237 | } else { | |
238 | lttng_opt_verbose += 1; | |
239 | } | |
240 | break; | |
241 | default: | |
242 | /* Unknown option or other error. | |
243 | * Error is printed by getopt, just return */ | |
244 | ret = -1; | |
245 | goto end; | |
246 | } | |
247 | ||
248 | /* All good. */ | |
249 | ret = 0; | |
250 | ||
251 | end: | |
252 | return ret; | |
253 | } | |
254 | ||
255 | /* | |
256 | * config_entry_handler_cb used to handle options read from a config file. | |
257 | * See config_entry_handler_cb comment in common/config/config.h for the | |
258 | * return value conventions. | |
259 | */ | |
260 | static | |
261 | int config_entry_handler(const struct config_entry *entry, void *unused) | |
262 | { | |
263 | int ret = 0, i; | |
264 | ||
265 | if (!entry || !entry->name || !entry->value) { | |
266 | ret = -EINVAL; | |
267 | goto end; | |
268 | } | |
269 | ||
270 | /* Check if the option is to be ignored */ | |
271 | for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) { | |
272 | if (!strcmp(entry->name, config_ignore_options[i])) { | |
273 | goto end; | |
274 | } | |
275 | } | |
276 | ||
277 | for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) { | |
278 | /* Ignore if entry name is not fully matched. */ | |
279 | if (strcmp(entry->name, long_options[i].name)) { | |
280 | continue; | |
281 | } | |
282 | ||
283 | /* | |
284 | * If the option takes no argument on the command line, we have to | |
285 | * check if the value is "true". We support non-zero numeric values, | |
286 | * true, on and yes. | |
287 | */ | |
288 | if (!long_options[i].has_arg) { | |
289 | ret = config_parse_value(entry->value); | |
290 | if (ret <= 0) { | |
291 | if (ret) { | |
292 | WARN("Invalid configuration value \"%s\" for option %s", | |
293 | entry->value, entry->name); | |
294 | } | |
295 | /* False, skip boolean config option. */ | |
296 | goto end; | |
297 | } | |
298 | } | |
299 | ||
300 | ret = set_option(long_options[i].val, entry->value, entry->name); | |
301 | goto end; | |
302 | } | |
303 | ||
304 | WARN("Unrecognized option \"%s\" in daemon configuration file.", | |
305 | entry->name); | |
306 | ||
307 | end: | |
308 | return ret; | |
309 | } | |
310 | ||
311 | static | |
312 | int set_options(int argc, char **argv) | |
313 | { | |
314 | int c, ret = 0, option_index = 0; | |
315 | int orig_optopt = optopt, orig_optind = optind; | |
316 | char *default_address, *optstring; | |
317 | const char *config_path = NULL; | |
318 | ||
319 | optstring = utils_generate_optstring(long_options, | |
320 | sizeof(long_options) / sizeof(struct option)); | |
321 | if (!optstring) { | |
322 | ret = -ENOMEM; | |
323 | goto exit; | |
324 | } | |
325 | ||
326 | /* Check for the --config option */ | |
327 | ||
328 | while ((c = getopt_long(argc, argv, optstring, long_options, | |
329 | &option_index)) != -1) { | |
330 | if (c == '?') { | |
331 | ret = -EINVAL; | |
332 | goto exit; | |
333 | } else if (c != 'f') { | |
334 | continue; | |
335 | } | |
336 | ||
337 | config_path = utils_expand_path(optarg); | |
338 | if (!config_path) { | |
339 | ERR("Failed to resolve path: %s", optarg); | |
340 | } | |
341 | } | |
342 | ||
343 | ret = config_get_section_entries(config_path, config_section_name, | |
344 | config_entry_handler, NULL); | |
345 | if (ret) { | |
346 | if (ret > 0) { | |
347 | ERR("Invalid configuration option at line %i", ret); | |
348 | ret = -1; | |
349 | } | |
350 | goto exit; | |
351 | } | |
b8aa1682 | 352 | |
cd60b05a JG |
353 | /* Reset getopt's global state */ |
354 | optopt = orig_optopt; | |
355 | optind = orig_optind; | |
b8aa1682 | 356 | while (1) { |
cd60b05a | 357 | c = getopt_long(argc, argv, optstring, long_options, &option_index); |
b8aa1682 JD |
358 | if (c == -1) { |
359 | break; | |
360 | } | |
361 | ||
cd60b05a JG |
362 | ret = set_option(c, optarg, long_options[option_index].name); |
363 | if (ret < 0) { | |
b8aa1682 JD |
364 | goto exit; |
365 | } | |
366 | } | |
367 | ||
368 | /* assign default values */ | |
369 | if (control_uri == NULL) { | |
370 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
371 | DEFAULT_NETWORK_CONTROL_PORT); | |
372 | if (ret < 0) { | |
373 | PERROR("asprintf default data address"); | |
374 | goto exit; | |
375 | } | |
376 | ||
377 | ret = uri_parse(default_address, &control_uri); | |
378 | free(default_address); | |
379 | if (ret < 0) { | |
380 | ERR("Invalid control URI specified"); | |
381 | goto exit; | |
382 | } | |
383 | } | |
384 | if (data_uri == NULL) { | |
385 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
386 | DEFAULT_NETWORK_DATA_PORT); | |
387 | if (ret < 0) { | |
388 | PERROR("asprintf default data address"); | |
389 | goto exit; | |
390 | } | |
391 | ||
392 | ret = uri_parse(default_address, &data_uri); | |
393 | free(default_address); | |
394 | if (ret < 0) { | |
395 | ERR("Invalid data URI specified"); | |
396 | goto exit; | |
397 | } | |
398 | } | |
d3e2ba59 JD |
399 | if (live_uri == NULL) { |
400 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
401 | DEFAULT_NETWORK_VIEWER_PORT); | |
402 | if (ret < 0) { | |
403 | PERROR("asprintf default viewer control address"); | |
404 | goto exit; | |
405 | } | |
406 | ||
407 | ret = uri_parse(default_address, &live_uri); | |
408 | free(default_address); | |
409 | if (ret < 0) { | |
410 | ERR("Invalid viewer control URI specified"); | |
411 | goto exit; | |
412 | } | |
413 | } | |
b8aa1682 JD |
414 | |
415 | exit: | |
cd60b05a | 416 | free(optstring); |
b8aa1682 JD |
417 | return ret; |
418 | } | |
419 | ||
420 | /* | |
421 | * Cleanup the daemon | |
422 | */ | |
423 | static | |
424 | void cleanup(void) | |
425 | { | |
b8aa1682 JD |
426 | DBG("Cleaning up"); |
427 | ||
095a4ae5 MD |
428 | /* free the dynamically allocated opt_output_path */ |
429 | free(opt_output_path); | |
430 | ||
a02de639 CB |
431 | /* Close thread quit pipes */ |
432 | utils_close_pipe(thread_quit_pipe); | |
433 | ||
710c1f73 DG |
434 | uri_free(control_uri); |
435 | uri_free(data_uri); | |
8d5c808e | 436 | /* Live URI is freed in the live thread. */ |
cd60b05a JG |
437 | |
438 | if (tracing_group_name_override) { | |
439 | free((void *) tracing_group_name); | |
440 | } | |
b8aa1682 JD |
441 | } |
442 | ||
443 | /* | |
444 | * Write to writable pipe used to notify a thread. | |
445 | */ | |
446 | static | |
447 | int notify_thread_pipe(int wpipe) | |
448 | { | |
6cd525e8 | 449 | ssize_t ret; |
b8aa1682 | 450 | |
6cd525e8 MD |
451 | ret = lttng_write(wpipe, "!", 1); |
452 | if (ret < 1) { | |
b8aa1682 JD |
453 | PERROR("write poll pipe"); |
454 | } | |
455 | ||
456 | return ret; | |
457 | } | |
458 | ||
65931c8b MD |
459 | static void notify_health_quit_pipe(int *pipe) |
460 | { | |
6cd525e8 | 461 | ssize_t ret; |
65931c8b | 462 | |
6cd525e8 MD |
463 | ret = lttng_write(pipe[1], "4", 1); |
464 | if (ret < 1) { | |
65931c8b MD |
465 | PERROR("write relay health quit"); |
466 | } | |
467 | } | |
468 | ||
b8aa1682 JD |
469 | /* |
470 | * Stop all threads by closing the thread quit pipe. | |
471 | */ | |
472 | static | |
473 | void stop_threads(void) | |
474 | { | |
475 | int ret; | |
476 | ||
477 | /* Stopping all threads */ | |
478 | DBG("Terminating all threads"); | |
479 | ret = notify_thread_pipe(thread_quit_pipe[1]); | |
480 | if (ret < 0) { | |
481 | ERR("write error on thread quit pipe"); | |
482 | } | |
483 | ||
65931c8b MD |
484 | notify_health_quit_pipe(health_quit_pipe); |
485 | ||
b8aa1682 | 486 | /* Dispatch thread */ |
26c9d55e | 487 | CMM_STORE_SHARED(dispatch_thread_exit, 1); |
b8aa1682 JD |
488 | futex_nto1_wake(&relay_cmd_queue.futex); |
489 | } | |
490 | ||
491 | /* | |
492 | * Signal handler for the daemon | |
493 | * | |
494 | * Simply stop all worker threads, leaving main() return gracefully after | |
495 | * joining all threads and calling cleanup(). | |
496 | */ | |
497 | static | |
498 | void sighandler(int sig) | |
499 | { | |
500 | switch (sig) { | |
501 | case SIGPIPE: | |
502 | DBG("SIGPIPE caught"); | |
503 | return; | |
504 | case SIGINT: | |
505 | DBG("SIGINT caught"); | |
506 | stop_threads(); | |
507 | break; | |
508 | case SIGTERM: | |
509 | DBG("SIGTERM caught"); | |
510 | stop_threads(); | |
511 | break; | |
512 | default: | |
513 | break; | |
514 | } | |
515 | } | |
516 | ||
517 | /* | |
518 | * Setup signal handler for : | |
519 | * SIGINT, SIGTERM, SIGPIPE | |
520 | */ | |
521 | static | |
522 | int set_signal_handler(void) | |
523 | { | |
524 | int ret = 0; | |
525 | struct sigaction sa; | |
526 | sigset_t sigset; | |
527 | ||
528 | if ((ret = sigemptyset(&sigset)) < 0) { | |
529 | PERROR("sigemptyset"); | |
530 | return ret; | |
531 | } | |
532 | ||
533 | sa.sa_handler = sighandler; | |
534 | sa.sa_mask = sigset; | |
535 | sa.sa_flags = 0; | |
536 | if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) { | |
537 | PERROR("sigaction"); | |
538 | return ret; | |
539 | } | |
540 | ||
541 | if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) { | |
542 | PERROR("sigaction"); | |
543 | return ret; | |
544 | } | |
545 | ||
546 | if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) { | |
547 | PERROR("sigaction"); | |
548 | return ret; | |
549 | } | |
550 | ||
551 | DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT"); | |
552 | ||
553 | return ret; | |
554 | } | |
555 | ||
556 | /* | |
557 | * Init thread quit pipe. | |
558 | * | |
559 | * Return -1 on error or 0 if all pipes are created. | |
560 | */ | |
561 | static | |
562 | int init_thread_quit_pipe(void) | |
563 | { | |
a02de639 | 564 | int ret; |
b8aa1682 | 565 | |
a02de639 | 566 | ret = utils_create_pipe_cloexec(thread_quit_pipe); |
b8aa1682 | 567 | |
b8aa1682 JD |
568 | return ret; |
569 | } | |
570 | ||
571 | /* | |
572 | * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set. | |
573 | */ | |
574 | static | |
575 | int create_thread_poll_set(struct lttng_poll_event *events, int size) | |
576 | { | |
577 | int ret; | |
578 | ||
579 | if (events == NULL || size == 0) { | |
580 | ret = -1; | |
581 | goto error; | |
582 | } | |
583 | ||
584 | ret = lttng_poll_create(events, size, LTTNG_CLOEXEC); | |
585 | if (ret < 0) { | |
586 | goto error; | |
587 | } | |
588 | ||
589 | /* Add quit pipe */ | |
590 | ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN); | |
591 | if (ret < 0) { | |
592 | goto error; | |
593 | } | |
594 | ||
595 | return 0; | |
596 | ||
597 | error: | |
598 | return ret; | |
599 | } | |
600 | ||
601 | /* | |
602 | * Check if the thread quit pipe was triggered. | |
603 | * | |
604 | * Return 1 if it was triggered else 0; | |
605 | */ | |
606 | static | |
607 | int check_thread_quit_pipe(int fd, uint32_t events) | |
608 | { | |
609 | if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) { | |
610 | return 1; | |
611 | } | |
612 | ||
613 | return 0; | |
614 | } | |
615 | ||
616 | /* | |
617 | * Create and init socket from uri. | |
618 | */ | |
619 | static | |
620 | struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri) | |
621 | { | |
622 | int ret; | |
623 | struct lttcomm_sock *sock = NULL; | |
624 | ||
625 | sock = lttcomm_alloc_sock_from_uri(uri); | |
626 | if (sock == NULL) { | |
627 | ERR("Allocating socket"); | |
628 | goto error; | |
629 | } | |
630 | ||
631 | ret = lttcomm_create_sock(sock); | |
632 | if (ret < 0) { | |
633 | goto error; | |
634 | } | |
635 | DBG("Listening on sock %d", sock->fd); | |
636 | ||
637 | ret = sock->ops->bind(sock); | |
638 | if (ret < 0) { | |
639 | goto error; | |
640 | } | |
641 | ||
642 | ret = sock->ops->listen(sock, -1); | |
643 | if (ret < 0) { | |
644 | goto error; | |
645 | ||
646 | } | |
647 | ||
648 | return sock; | |
649 | ||
650 | error: | |
651 | if (sock) { | |
652 | lttcomm_destroy_sock(sock); | |
653 | } | |
654 | return NULL; | |
655 | } | |
656 | ||
173af62f DG |
657 | /* |
658 | * Return nonzero if stream needs to be closed. | |
659 | */ | |
660 | static | |
661 | int close_stream_check(struct relay_stream *stream) | |
662 | { | |
663 | ||
664 | if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) { | |
f7079f67 DG |
665 | /* |
666 | * We are about to close the stream so set the data pending flag to 1 | |
667 | * which will make the end data pending command skip the stream which | |
668 | * is now closed and ready. Note that after proceeding to a file close, | |
669 | * the written file is ready for reading. | |
670 | */ | |
671 | stream->data_pending_check_done = 1; | |
173af62f DG |
672 | return 1; |
673 | } | |
674 | return 0; | |
675 | } | |
676 | ||
b8aa1682 JD |
677 | /* |
678 | * This thread manages the listening for new connections on the network | |
679 | */ | |
680 | static | |
681 | void *relay_thread_listener(void *data) | |
682 | { | |
095a4ae5 | 683 | int i, ret, pollfd, err = -1; |
b8aa1682 JD |
684 | int val = 1; |
685 | uint32_t revents, nb_fd; | |
686 | struct lttng_poll_event events; | |
687 | struct lttcomm_sock *control_sock, *data_sock; | |
688 | ||
b8aa1682 JD |
689 | DBG("[thread] Relay listener started"); |
690 | ||
55706a7d MD |
691 | health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER); |
692 | ||
f385ae0a MD |
693 | health_code_update(); |
694 | ||
b8aa1682 JD |
695 | control_sock = relay_init_sock(control_uri); |
696 | if (!control_sock) { | |
095a4ae5 | 697 | goto error_sock_control; |
b8aa1682 JD |
698 | } |
699 | ||
700 | data_sock = relay_init_sock(data_uri); | |
701 | if (!data_sock) { | |
095a4ae5 | 702 | goto error_sock_relay; |
b8aa1682 JD |
703 | } |
704 | ||
705 | /* | |
706 | * Pass 3 as size here for the thread quit pipe, control and data socket. | |
707 | */ | |
708 | ret = create_thread_poll_set(&events, 3); | |
709 | if (ret < 0) { | |
710 | goto error_create_poll; | |
711 | } | |
712 | ||
713 | /* Add the control socket */ | |
714 | ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP); | |
715 | if (ret < 0) { | |
716 | goto error_poll_add; | |
717 | } | |
718 | ||
719 | /* Add the data socket */ | |
720 | ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP); | |
721 | if (ret < 0) { | |
722 | goto error_poll_add; | |
723 | } | |
724 | ||
725 | while (1) { | |
f385ae0a MD |
726 | health_code_update(); |
727 | ||
b8aa1682 JD |
728 | DBG("Listener accepting connections"); |
729 | ||
b8aa1682 | 730 | restart: |
f385ae0a | 731 | health_poll_entry(); |
b8aa1682 | 732 | ret = lttng_poll_wait(&events, -1); |
f385ae0a | 733 | health_poll_exit(); |
b8aa1682 JD |
734 | if (ret < 0) { |
735 | /* | |
736 | * Restart interrupted system call. | |
737 | */ | |
738 | if (errno == EINTR) { | |
739 | goto restart; | |
740 | } | |
741 | goto error; | |
742 | } | |
743 | ||
0d9c5d77 DG |
744 | nb_fd = ret; |
745 | ||
b8aa1682 JD |
746 | DBG("Relay new connection received"); |
747 | for (i = 0; i < nb_fd; i++) { | |
f385ae0a MD |
748 | health_code_update(); |
749 | ||
b8aa1682 JD |
750 | /* Fetch once the poll data */ |
751 | revents = LTTNG_POLL_GETEV(&events, i); | |
752 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
753 | ||
754 | /* Thread quit pipe has been closed. Killing thread. */ | |
755 | ret = check_thread_quit_pipe(pollfd, revents); | |
756 | if (ret) { | |
095a4ae5 MD |
757 | err = 0; |
758 | goto exit; | |
b8aa1682 JD |
759 | } |
760 | ||
761 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
762 | ERR("socket poll error"); | |
763 | goto error; | |
764 | } else if (revents & LPOLLIN) { | |
4b7f17b2 MD |
765 | /* |
766 | * Get allocated in this thread, | |
767 | * enqueued to a global queue, dequeued | |
768 | * and freed in the worker thread. | |
769 | */ | |
770 | struct relay_command *relay_cmd; | |
771 | struct lttcomm_sock *newsock; | |
b8aa1682 JD |
772 | |
773 | relay_cmd = zmalloc(sizeof(struct relay_command)); | |
774 | if (relay_cmd == NULL) { | |
775 | PERROR("relay command zmalloc"); | |
776 | goto error; | |
777 | } | |
778 | ||
779 | if (pollfd == data_sock->fd) { | |
780 | newsock = data_sock->ops->accept(data_sock); | |
4b7f17b2 | 781 | if (!newsock) { |
b8aa1682 | 782 | PERROR("accepting data sock"); |
4b7f17b2 | 783 | free(relay_cmd); |
b8aa1682 JD |
784 | goto error; |
785 | } | |
786 | relay_cmd->type = RELAY_DATA; | |
787 | DBG("Relay data connection accepted, socket %d", newsock->fd); | |
4b7f17b2 MD |
788 | } else { |
789 | assert(pollfd == control_sock->fd); | |
b8aa1682 | 790 | newsock = control_sock->ops->accept(control_sock); |
4b7f17b2 | 791 | if (!newsock) { |
b8aa1682 | 792 | PERROR("accepting control sock"); |
4b7f17b2 | 793 | free(relay_cmd); |
b8aa1682 JD |
794 | goto error; |
795 | } | |
796 | relay_cmd->type = RELAY_CONTROL; | |
797 | DBG("Relay control connection accepted, socket %d", newsock->fd); | |
798 | } | |
799 | ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, | |
800 | &val, sizeof(int)); | |
801 | if (ret < 0) { | |
802 | PERROR("setsockopt inet"); | |
4b7f17b2 MD |
803 | lttcomm_destroy_sock(newsock); |
804 | free(relay_cmd); | |
b8aa1682 JD |
805 | goto error; |
806 | } | |
807 | relay_cmd->sock = newsock; | |
808 | /* | |
809 | * Lock free enqueue the request. | |
810 | */ | |
811 | cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node); | |
812 | ||
813 | /* | |
814 | * Wake the dispatch queue futex. Implicit memory | |
815 | * barrier with the exchange in cds_wfq_enqueue. | |
816 | */ | |
817 | futex_nto1_wake(&relay_cmd_queue.futex); | |
818 | } | |
819 | } | |
820 | } | |
821 | ||
095a4ae5 | 822 | exit: |
b8aa1682 JD |
823 | error: |
824 | error_poll_add: | |
825 | lttng_poll_clean(&events); | |
826 | error_create_poll: | |
095a4ae5 MD |
827 | if (data_sock->fd >= 0) { |
828 | ret = data_sock->ops->close(data_sock); | |
b8aa1682 JD |
829 | if (ret) { |
830 | PERROR("close"); | |
831 | } | |
b8aa1682 | 832 | } |
095a4ae5 MD |
833 | lttcomm_destroy_sock(data_sock); |
834 | error_sock_relay: | |
835 | if (control_sock->fd >= 0) { | |
836 | ret = control_sock->ops->close(control_sock); | |
b8aa1682 JD |
837 | if (ret) { |
838 | PERROR("close"); | |
839 | } | |
b8aa1682 | 840 | } |
095a4ae5 MD |
841 | lttcomm_destroy_sock(control_sock); |
842 | error_sock_control: | |
843 | if (err) { | |
f385ae0a MD |
844 | health_error(); |
845 | ERR("Health error occurred in %s", __func__); | |
095a4ae5 | 846 | } |
55706a7d | 847 | health_unregister(health_relayd); |
b8aa1682 JD |
848 | DBG("Relay listener thread cleanup complete"); |
849 | stop_threads(); | |
b8aa1682 JD |
850 | return NULL; |
851 | } | |
852 | ||
853 | /* | |
854 | * This thread manages the dispatching of the requests to worker threads | |
855 | */ | |
856 | static | |
857 | void *relay_thread_dispatcher(void *data) | |
858 | { | |
6cd525e8 MD |
859 | int err = -1; |
860 | ssize_t ret; | |
b8aa1682 JD |
861 | struct cds_wfq_node *node; |
862 | struct relay_command *relay_cmd = NULL; | |
863 | ||
864 | DBG("[thread] Relay dispatcher started"); | |
865 | ||
55706a7d MD |
866 | health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER); |
867 | ||
f385ae0a MD |
868 | health_code_update(); |
869 | ||
26c9d55e | 870 | while (!CMM_LOAD_SHARED(dispatch_thread_exit)) { |
f385ae0a MD |
871 | health_code_update(); |
872 | ||
b8aa1682 JD |
873 | /* Atomically prepare the queue futex */ |
874 | futex_nto1_prepare(&relay_cmd_queue.futex); | |
875 | ||
876 | do { | |
f385ae0a MD |
877 | health_code_update(); |
878 | ||
b8aa1682 JD |
879 | /* Dequeue commands */ |
880 | node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue); | |
881 | if (node == NULL) { | |
882 | DBG("Woken up but nothing in the relay command queue"); | |
883 | /* Continue thread execution */ | |
884 | break; | |
885 | } | |
886 | ||
887 | relay_cmd = caa_container_of(node, struct relay_command, node); | |
888 | DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd); | |
889 | ||
890 | /* | |
891 | * Inform worker thread of the new request. This | |
892 | * call is blocking so we can be assured that the data will be read | |
893 | * at some point in time or wait to the end of the world :) | |
894 | */ | |
6cd525e8 MD |
895 | ret = lttng_write(relay_cmd_pipe[1], relay_cmd, |
896 | sizeof(struct relay_command)); | |
b8aa1682 | 897 | free(relay_cmd); |
6cd525e8 | 898 | if (ret < sizeof(struct relay_command)) { |
b8aa1682 JD |
899 | PERROR("write cmd pipe"); |
900 | goto error; | |
901 | } | |
902 | } while (node != NULL); | |
903 | ||
904 | /* Futex wait on queue. Blocking call on futex() */ | |
f385ae0a | 905 | health_poll_entry(); |
b8aa1682 | 906 | futex_nto1_wait(&relay_cmd_queue.futex); |
f385ae0a | 907 | health_poll_exit(); |
b8aa1682 JD |
908 | } |
909 | ||
f385ae0a MD |
910 | /* Normal exit, no error */ |
911 | err = 0; | |
912 | ||
b8aa1682 | 913 | error: |
f385ae0a MD |
914 | if (err) { |
915 | health_error(); | |
916 | ERR("Health error occurred in %s", __func__); | |
917 | } | |
55706a7d | 918 | health_unregister(health_relayd); |
b8aa1682 JD |
919 | DBG("Dispatch thread dying"); |
920 | stop_threads(); | |
921 | return NULL; | |
922 | } | |
923 | ||
de91f48a DG |
924 | /* |
925 | * Get stream from stream id. | |
926 | * Need to be called with RCU read-side lock held. | |
927 | */ | |
d3e2ba59 | 928 | struct relay_stream *relay_stream_find_by_id(uint64_t stream_id) |
de91f48a DG |
929 | { |
930 | struct lttng_ht_node_ulong *node; | |
931 | struct lttng_ht_iter iter; | |
932 | struct relay_stream *ret; | |
933 | ||
d3e2ba59 | 934 | lttng_ht_lookup(relay_streams_ht, |
de91f48a DG |
935 | (void *)((unsigned long) stream_id), |
936 | &iter); | |
937 | node = lttng_ht_iter_get_node_ulong(&iter); | |
938 | if (node == NULL) { | |
939 | DBG("Relay stream %" PRIu64 " not found", stream_id); | |
940 | ret = NULL; | |
941 | goto end; | |
942 | } | |
943 | ||
944 | ret = caa_container_of(node, struct relay_stream, stream_n); | |
945 | ||
946 | end: | |
947 | return ret; | |
948 | } | |
949 | ||
9d1bbf21 MD |
950 | static |
951 | void deferred_free_stream(struct rcu_head *head) | |
952 | { | |
953 | struct relay_stream *stream = | |
954 | caa_container_of(head, struct relay_stream, rcu_node); | |
d3e2ba59 | 955 | |
0f907de1 JD |
956 | free(stream->path_name); |
957 | free(stream->channel_name); | |
9d1bbf21 MD |
958 | free(stream); |
959 | } | |
960 | ||
d3e2ba59 JD |
961 | static |
962 | void deferred_free_session(struct rcu_head *head) | |
963 | { | |
964 | struct relay_session *session = | |
965 | caa_container_of(head, struct relay_session, rcu_node); | |
966 | free(session); | |
967 | } | |
968 | ||
4c80d9a2 DG |
969 | /* |
970 | * Close a given stream. The stream is freed using a call RCU. | |
971 | * | |
972 | * RCU read side lock MUST be acquired. If NO close_stream_check() was called | |
973 | * BEFORE the stream lock MUST be acquired. | |
974 | */ | |
c07d8a78 | 975 | static void destroy_stream(struct relay_stream *stream) |
94d49140 JD |
976 | { |
977 | int delret; | |
978 | struct relay_viewer_stream *vstream; | |
979 | struct lttng_ht_iter iter; | |
980 | ||
981 | assert(stream); | |
94d49140 JD |
982 | |
983 | delret = close(stream->fd); | |
984 | if (delret < 0) { | |
985 | PERROR("close stream"); | |
986 | } | |
987 | ||
988 | if (stream->index_fd >= 0) { | |
989 | delret = close(stream->index_fd); | |
990 | if (delret < 0) { | |
991 | PERROR("close stream index_fd"); | |
992 | } | |
993 | } | |
994 | ||
92c6ca54 | 995 | vstream = live_find_viewer_stream_by_id(stream->stream_handle); |
94d49140 JD |
996 | if (vstream) { |
997 | /* | |
998 | * Set the last good value into the viewer stream. This is done | |
999 | * right before the stream gets deleted from the hash table. The | |
1000 | * lookup failure on the live thread side of a stream indicates | |
1001 | * that the viewer stream index received value should be used. | |
1002 | */ | |
cef0f7d5 | 1003 | pthread_mutex_lock(&stream->viewer_stream_rotation_lock); |
94d49140 | 1004 | vstream->total_index_received = stream->total_index_received; |
a020f610 | 1005 | vstream->tracefile_count_last = stream->tracefile_count_current; |
cef0f7d5 JD |
1006 | vstream->close_write_flag = 1; |
1007 | pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); | |
94d49140 JD |
1008 | } |
1009 | ||
efc8a87f DG |
1010 | /* Cleanup index of that stream. */ |
1011 | relay_index_destroy_by_stream_id(stream->stream_handle); | |
1012 | ||
94d49140 JD |
1013 | iter.iter.node = &stream->stream_n.node; |
1014 | delret = lttng_ht_del(relay_streams_ht, &iter); | |
1015 | assert(!delret); | |
1016 | iter.iter.node = &stream->ctf_trace_node.node; | |
c07d8a78 | 1017 | delret = lttng_ht_del(stream->ctf_traces_ht, &iter); |
94d49140 | 1018 | assert(!delret); |
157df586 JD |
1019 | |
1020 | if (stream->ctf_trace) { | |
1021 | ctf_trace_try_destroy(stream->ctf_trace); | |
1022 | } | |
1023 | ||
94d49140 JD |
1024 | call_rcu(&stream->rcu_node, deferred_free_stream); |
1025 | DBG("Closed tracefile %d from close stream", stream->fd); | |
1026 | } | |
1027 | ||
b8aa1682 JD |
1028 | /* |
1029 | * relay_delete_session: Free all memory associated with a session and | |
1030 | * close all the FDs | |
1031 | */ | |
1032 | static | |
d3e2ba59 JD |
1033 | void relay_delete_session(struct relay_command *cmd, |
1034 | struct lttng_ht *sessions_ht) | |
b8aa1682 JD |
1035 | { |
1036 | struct lttng_ht_iter iter; | |
1037 | struct lttng_ht_node_ulong *node; | |
1038 | struct relay_stream *stream; | |
1039 | int ret; | |
1040 | ||
095a4ae5 | 1041 | if (!cmd->session) { |
b8aa1682 | 1042 | return; |
095a4ae5 | 1043 | } |
b8aa1682 | 1044 | |
77c7c900 | 1045 | DBG("Relay deleting session %" PRIu64, cmd->session->id); |
5b6d8097 | 1046 | |
9d1bbf21 | 1047 | rcu_read_lock(); |
d3e2ba59 | 1048 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) { |
b8aa1682 | 1049 | node = lttng_ht_iter_get_node_ulong(&iter); |
4c80d9a2 DG |
1050 | if (!node) { |
1051 | continue; | |
b8aa1682 | 1052 | } |
4c80d9a2 DG |
1053 | stream = caa_container_of(node, struct relay_stream, stream_n); |
1054 | if (stream->session == cmd->session) { | |
c07d8a78 | 1055 | destroy_stream(stream); |
87b576ec JD |
1056 | cmd->session->stream_count--; |
1057 | assert(cmd->session->stream_count >= 0); | |
4c80d9a2 | 1058 | } |
b8aa1682 | 1059 | } |
4c80d9a2 DG |
1060 | |
1061 | /* Make this session not visible anymore. */ | |
d3e2ba59 JD |
1062 | iter.iter.node = &cmd->session->session_n.node; |
1063 | ret = lttng_ht_del(sessions_ht, &iter); | |
1064 | assert(!ret); | |
4c80d9a2 | 1065 | call_rcu(&cmd->session->rcu_node, deferred_free_session); |
9d1bbf21 | 1066 | rcu_read_unlock(); |
b8aa1682 JD |
1067 | } |
1068 | ||
1c20f0e2 JD |
1069 | /* |
1070 | * Copy index data from the control port to a given index object. | |
1071 | */ | |
1072 | static void copy_index_control_data(struct relay_index *index, | |
1073 | struct lttcomm_relayd_index *data) | |
1074 | { | |
1075 | assert(index); | |
1076 | assert(data); | |
1077 | ||
1078 | /* | |
1079 | * The index on disk is encoded in big endian, so we don't need to convert | |
1080 | * the data received on the network. The data_offset value is NEVER | |
1081 | * modified here and is updated by the data thread. | |
1082 | */ | |
1083 | index->index_data.packet_size = data->packet_size; | |
1084 | index->index_data.content_size = data->content_size; | |
1085 | index->index_data.timestamp_begin = data->timestamp_begin; | |
1086 | index->index_data.timestamp_end = data->timestamp_end; | |
1087 | index->index_data.events_discarded = data->events_discarded; | |
1088 | index->index_data.stream_id = data->stream_id; | |
1089 | } | |
1090 | ||
c5b6f4f0 DG |
1091 | /* |
1092 | * Handle the RELAYD_CREATE_SESSION command. | |
1093 | * | |
1094 | * On success, send back the session id or else return a negative value. | |
1095 | */ | |
1096 | static | |
1097 | int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 JD |
1098 | struct relay_command *cmd, |
1099 | struct lttng_ht *sessions_ht) | |
c5b6f4f0 DG |
1100 | { |
1101 | int ret = 0, send_ret; | |
1102 | struct relay_session *session; | |
1103 | struct lttcomm_relayd_status_session reply; | |
1104 | ||
1105 | assert(recv_hdr); | |
1106 | assert(cmd); | |
1107 | ||
1108 | memset(&reply, 0, sizeof(reply)); | |
1109 | ||
1110 | session = zmalloc(sizeof(struct relay_session)); | |
1111 | if (session == NULL) { | |
1112 | PERROR("relay session zmalloc"); | |
1113 | ret = -1; | |
1114 | goto error; | |
1115 | } | |
1116 | ||
1117 | session->id = ++last_relay_session_id; | |
1118 | session->sock = cmd->sock; | |
7d2f7452 DG |
1119 | session->minor = cmd->minor; |
1120 | session->major = cmd->major; | |
c5b6f4f0 DG |
1121 | cmd->session = session; |
1122 | ||
1123 | reply.session_id = htobe64(session->id); | |
1124 | ||
d3e2ba59 JD |
1125 | switch (cmd->minor) { |
1126 | case 4: /* LTTng sessiond 2.4 */ | |
1127 | default: | |
1128 | ret = cmd_create_session_2_4(cmd, session); | |
1129 | break; | |
1130 | } | |
1131 | ||
1132 | lttng_ht_node_init_ulong(&session->session_n, | |
1133 | (unsigned long) session->id); | |
1134 | lttng_ht_add_unique_ulong(sessions_ht, | |
1135 | &session->session_n); | |
1136 | ||
c5b6f4f0 DG |
1137 | DBG("Created session %" PRIu64, session->id); |
1138 | ||
1139 | error: | |
1140 | if (ret < 0) { | |
1141 | reply.ret_code = htobe32(LTTNG_ERR_FATAL); | |
1142 | } else { | |
1143 | reply.ret_code = htobe32(LTTNG_OK); | |
1144 | } | |
1145 | ||
1146 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1147 | if (send_ret < 0) { | |
1148 | ERR("Relayd sending session id"); | |
4169f5ad | 1149 | ret = send_ret; |
c5b6f4f0 DG |
1150 | } |
1151 | ||
1152 | return ret; | |
1153 | } | |
1154 | ||
a4baae1b JD |
1155 | /* |
1156 | * When we have received all the streams and the metadata for a channel, | |
1157 | * we make them visible to the viewer threads. | |
1158 | */ | |
1159 | static | |
1160 | void set_viewer_ready_flag(struct relay_command *cmd) | |
1161 | { | |
1162 | struct relay_stream_recv_handle *node, *tmp_node; | |
1163 | ||
1164 | cds_list_for_each_entry_safe(node, tmp_node, &cmd->recv_head, node) { | |
1165 | struct relay_stream *stream; | |
1166 | ||
1167 | rcu_read_lock(); | |
1168 | stream = relay_stream_find_by_id(node->id); | |
1169 | if (!stream) { | |
1170 | /* | |
1171 | * Stream is most probably being cleaned up by the data thread thus | |
1172 | * simply continue to the next one. | |
1173 | */ | |
b3bdb105 | 1174 | rcu_read_unlock(); |
a4baae1b JD |
1175 | continue; |
1176 | } | |
1177 | ||
a4baae1b JD |
1178 | stream->viewer_ready = 1; |
1179 | rcu_read_unlock(); | |
1180 | ||
1181 | /* Clean stream handle node. */ | |
1182 | cds_list_del(&node->node); | |
1183 | free(node); | |
1184 | } | |
1185 | ||
a4baae1b JD |
1186 | return; |
1187 | } | |
1188 | ||
1189 | /* | |
1190 | * Add a recv handle node to the connection recv list with the given stream | |
1191 | * handle. A new node is allocated thus must be freed when the node is deleted | |
1192 | * from the list. | |
1193 | */ | |
1194 | static void queue_stream_handle(uint64_t handle, struct relay_command *cmd) | |
1195 | { | |
1196 | struct relay_stream_recv_handle *node; | |
1197 | ||
1198 | assert(cmd); | |
1199 | ||
1200 | node = zmalloc(sizeof(*node)); | |
1201 | if (!node) { | |
1202 | PERROR("zmalloc queue stream handle"); | |
1203 | return; | |
1204 | } | |
1205 | ||
1206 | node->id = handle; | |
1207 | cds_list_add(&node->node, &cmd->recv_head); | |
1208 | } | |
1209 | ||
b8aa1682 JD |
1210 | /* |
1211 | * relay_add_stream: allocate a new stream for a session | |
1212 | */ | |
1213 | static | |
1214 | int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1215 | struct relay_command *cmd, struct lttng_ht *sessions_ht) |
b8aa1682 JD |
1216 | { |
1217 | struct relay_session *session = cmd->session; | |
b8aa1682 JD |
1218 | struct relay_stream *stream = NULL; |
1219 | struct lttcomm_relayd_status_stream reply; | |
b8aa1682 JD |
1220 | int ret, send_ret; |
1221 | ||
c5b6f4f0 | 1222 | if (!session || cmd->version_check_done == 0) { |
b8aa1682 JD |
1223 | ERR("Trying to add a stream before version check"); |
1224 | ret = -1; | |
1225 | goto end_no_session; | |
1226 | } | |
1227 | ||
b8aa1682 JD |
1228 | stream = zmalloc(sizeof(struct relay_stream)); |
1229 | if (stream == NULL) { | |
1230 | PERROR("relay stream zmalloc"); | |
1231 | ret = -1; | |
1232 | goto end_no_session; | |
1233 | } | |
1234 | ||
0f907de1 JD |
1235 | switch (cmd->minor) { |
1236 | case 1: /* LTTng sessiond 2.1 */ | |
1237 | ret = cmd_recv_stream_2_1(cmd, stream); | |
1238 | break; | |
1239 | case 2: /* LTTng sessiond 2.2 */ | |
1240 | default: | |
1241 | ret = cmd_recv_stream_2_2(cmd, stream); | |
1242 | break; | |
1243 | } | |
1244 | if (ret < 0) { | |
1245 | goto err_free_stream; | |
1246 | } | |
1247 | ||
9d1bbf21 | 1248 | rcu_read_lock(); |
b8aa1682 | 1249 | stream->stream_handle = ++last_relay_stream_id; |
173af62f | 1250 | stream->prev_seq = -1ULL; |
b8aa1682 | 1251 | stream->session = session; |
1c20f0e2 | 1252 | stream->index_fd = -1; |
d3e2ba59 JD |
1253 | stream->read_index_fd = -1; |
1254 | stream->ctf_trace = NULL; | |
1255 | pthread_mutex_init(&stream->lock, NULL); | |
b8aa1682 | 1256 | |
0f907de1 | 1257 | ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG); |
b8aa1682 | 1258 | if (ret < 0) { |
b8aa1682 JD |
1259 | ERR("relay creating output directory"); |
1260 | goto end; | |
1261 | } | |
1262 | ||
be96a7d1 DG |
1263 | /* |
1264 | * No need to use run_as API here because whatever we receives, the relayd | |
1265 | * uses its own credentials for the stream files. | |
1266 | */ | |
0f907de1 | 1267 | ret = utils_create_stream_file(stream->path_name, stream->channel_name, |
1c20f0e2 | 1268 | stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL); |
b8aa1682 | 1269 | if (ret < 0) { |
0f907de1 | 1270 | ERR("Create output file"); |
b8aa1682 JD |
1271 | goto end; |
1272 | } | |
b8aa1682 | 1273 | stream->fd = ret; |
0f907de1 JD |
1274 | if (stream->tracefile_size) { |
1275 | DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name); | |
1276 | } else { | |
1277 | DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name); | |
1278 | } | |
b8aa1682 | 1279 | |
d3e2ba59 JD |
1280 | if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) { |
1281 | stream->metadata_flag = 1; | |
1282 | /* | |
1283 | * When we receive a new metadata stream, we create a new | |
1284 | * ctf_trace and we assign this ctf_trace to all streams with | |
1285 | * the same path. | |
1286 | * | |
1287 | * If later on we receive a new stream for the same ctf_trace, | |
1288 | * we copy the information from the first hit in the HT to the | |
1289 | * new stream. | |
1290 | */ | |
1291 | stream->ctf_trace = ctf_trace_create(); | |
1292 | if (!stream->ctf_trace) { | |
1293 | ret = -1; | |
1294 | goto end; | |
1295 | } | |
1296 | stream->ctf_trace->refcount++; | |
1297 | stream->ctf_trace->metadata_stream = stream; | |
1298 | } | |
1299 | ctf_trace_assign(cmd->ctf_traces_ht, stream); | |
c07d8a78 | 1300 | stream->ctf_traces_ht = cmd->ctf_traces_ht; |
d3e2ba59 | 1301 | |
a4baae1b JD |
1302 | /* |
1303 | * Add the stream handle in the recv list of the connection. Once the end | |
1304 | * stream message is received, this list is emptied and streams are set | |
1305 | * with the viewer ready flag. | |
1306 | */ | |
b3bdb105 JD |
1307 | if (stream->metadata_flag) { |
1308 | stream->viewer_ready = 1; | |
1309 | } else { | |
1310 | queue_stream_handle(stream->stream_handle, cmd); | |
1311 | } | |
a4baae1b | 1312 | |
b8aa1682 JD |
1313 | lttng_ht_node_init_ulong(&stream->stream_n, |
1314 | (unsigned long) stream->stream_handle); | |
d3e2ba59 | 1315 | lttng_ht_add_unique_ulong(relay_streams_ht, |
b8aa1682 JD |
1316 | &stream->stream_n); |
1317 | ||
d3e2ba59 JD |
1318 | lttng_ht_node_init_str(&stream->ctf_trace_node, stream->path_name); |
1319 | lttng_ht_add_str(cmd->ctf_traces_ht, &stream->ctf_trace_node); | |
87b576ec | 1320 | session->stream_count++; |
d3e2ba59 | 1321 | |
1c20f0e2 JD |
1322 | DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name, |
1323 | stream->stream_handle); | |
b8aa1682 JD |
1324 | |
1325 | end: | |
5af40280 | 1326 | reply.handle = htobe64(stream->stream_handle); |
b8aa1682 JD |
1327 | /* send the session id to the client or a negative return code on error */ |
1328 | if (ret < 0) { | |
f73fabfd | 1329 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
5af40280 CB |
1330 | /* stream was not properly added to the ht, so free it */ |
1331 | free(stream); | |
b8aa1682 | 1332 | } else { |
f73fabfd | 1333 | reply.ret_code = htobe32(LTTNG_OK); |
b8aa1682 | 1334 | } |
5af40280 | 1335 | |
b8aa1682 JD |
1336 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, |
1337 | sizeof(struct lttcomm_relayd_status_stream), 0); | |
1338 | if (send_ret < 0) { | |
1339 | ERR("Relay sending stream id"); | |
4169f5ad | 1340 | ret = send_ret; |
b8aa1682 | 1341 | } |
9d1bbf21 | 1342 | rcu_read_unlock(); |
b8aa1682 JD |
1343 | |
1344 | end_no_session: | |
1345 | return ret; | |
0f907de1 JD |
1346 | |
1347 | err_free_stream: | |
1348 | free(stream->path_name); | |
1349 | free(stream->channel_name); | |
1350 | free(stream); | |
1351 | return ret; | |
b8aa1682 JD |
1352 | } |
1353 | ||
173af62f DG |
1354 | /* |
1355 | * relay_close_stream: close a specific stream | |
1356 | */ | |
1357 | static | |
1358 | int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, | |
92c6ca54 | 1359 | struct relay_command *cmd) |
173af62f | 1360 | { |
94d49140 | 1361 | int ret, send_ret; |
173af62f DG |
1362 | struct relay_session *session = cmd->session; |
1363 | struct lttcomm_relayd_close_stream stream_info; | |
1364 | struct lttcomm_relayd_generic_reply reply; | |
1365 | struct relay_stream *stream; | |
173af62f DG |
1366 | |
1367 | DBG("Close stream received"); | |
1368 | ||
c5b6f4f0 | 1369 | if (!session || cmd->version_check_done == 0) { |
173af62f DG |
1370 | ERR("Trying to close a stream before version check"); |
1371 | ret = -1; | |
1372 | goto end_no_session; | |
1373 | } | |
1374 | ||
1375 | ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info, | |
7c5aef62 | 1376 | sizeof(struct lttcomm_relayd_close_stream), 0); |
173af62f | 1377 | if (ret < sizeof(struct lttcomm_relayd_close_stream)) { |
a6cd2b97 DG |
1378 | if (ret == 0) { |
1379 | /* Orderly shutdown. Not necessary to print an error. */ | |
1380 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1381 | } else { | |
1382 | ERR("Relay didn't receive valid add_stream struct size : %d", ret); | |
1383 | } | |
173af62f DG |
1384 | ret = -1; |
1385 | goto end_no_session; | |
1386 | } | |
1387 | ||
1388 | rcu_read_lock(); | |
d3e2ba59 | 1389 | stream = relay_stream_find_by_id(be64toh(stream_info.stream_id)); |
173af62f DG |
1390 | if (!stream) { |
1391 | ret = -1; | |
1392 | goto end_unlock; | |
1393 | } | |
1394 | ||
8e2583a4 | 1395 | stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num); |
173af62f | 1396 | stream->close_flag = 1; |
87b576ec JD |
1397 | session->stream_count--; |
1398 | assert(session->stream_count >= 0); | |
173af62f DG |
1399 | |
1400 | if (close_stream_check(stream)) { | |
c07d8a78 | 1401 | destroy_stream(stream); |
173af62f DG |
1402 | } |
1403 | ||
1404 | end_unlock: | |
1405 | rcu_read_unlock(); | |
1406 | ||
1407 | if (ret < 0) { | |
f73fabfd | 1408 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
173af62f | 1409 | } else { |
f73fabfd | 1410 | reply.ret_code = htobe32(LTTNG_OK); |
173af62f DG |
1411 | } |
1412 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1413 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1414 | if (send_ret < 0) { | |
1415 | ERR("Relay sending stream id"); | |
4169f5ad | 1416 | ret = send_ret; |
173af62f DG |
1417 | } |
1418 | ||
1419 | end_no_session: | |
1420 | return ret; | |
1421 | } | |
1422 | ||
b8aa1682 JD |
1423 | /* |
1424 | * relay_unknown_command: send -1 if received unknown command | |
1425 | */ | |
1426 | static | |
1427 | void relay_unknown_command(struct relay_command *cmd) | |
1428 | { | |
1429 | struct lttcomm_relayd_generic_reply reply; | |
1430 | int ret; | |
1431 | ||
f73fabfd | 1432 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
b8aa1682 JD |
1433 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, |
1434 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1435 | if (ret < 0) { | |
1436 | ERR("Relay sending unknown command"); | |
1437 | } | |
1438 | } | |
1439 | ||
1440 | /* | |
1441 | * relay_start: send an acknowledgment to the client to tell if we are | |
1442 | * ready to receive data. We are ready if a session is established. | |
1443 | */ | |
1444 | static | |
1445 | int relay_start(struct lttcomm_relayd_hdr *recv_hdr, | |
1446 | struct relay_command *cmd) | |
1447 | { | |
f73fabfd | 1448 | int ret = htobe32(LTTNG_OK); |
b8aa1682 JD |
1449 | struct lttcomm_relayd_generic_reply reply; |
1450 | struct relay_session *session = cmd->session; | |
1451 | ||
1452 | if (!session) { | |
1453 | DBG("Trying to start the streaming without a session established"); | |
f73fabfd | 1454 | ret = htobe32(LTTNG_ERR_UNK); |
b8aa1682 JD |
1455 | } |
1456 | ||
1457 | reply.ret_code = ret; | |
1458 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1459 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1460 | if (ret < 0) { | |
1461 | ERR("Relay sending start ack"); | |
1462 | } | |
1463 | ||
1464 | return ret; | |
1465 | } | |
1466 | ||
1d4dfdef DG |
1467 | /* |
1468 | * Append padding to the file pointed by the file descriptor fd. | |
1469 | */ | |
1470 | static int write_padding_to_file(int fd, uint32_t size) | |
1471 | { | |
6cd525e8 | 1472 | ssize_t ret = 0; |
1d4dfdef DG |
1473 | char *zeros; |
1474 | ||
1475 | if (size == 0) { | |
1476 | goto end; | |
1477 | } | |
1478 | ||
1479 | zeros = zmalloc(size); | |
1480 | if (zeros == NULL) { | |
1481 | PERROR("zmalloc zeros for padding"); | |
1482 | ret = -1; | |
1483 | goto end; | |
1484 | } | |
1485 | ||
6cd525e8 MD |
1486 | ret = lttng_write(fd, zeros, size); |
1487 | if (ret < size) { | |
1d4dfdef DG |
1488 | PERROR("write padding to file"); |
1489 | } | |
1490 | ||
e986c7a1 DG |
1491 | free(zeros); |
1492 | ||
1d4dfdef DG |
1493 | end: |
1494 | return ret; | |
1495 | } | |
1496 | ||
b8aa1682 JD |
1497 | /* |
1498 | * relay_recv_metadata: receive the metada for the session. | |
1499 | */ | |
1500 | static | |
1501 | int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1502 | struct relay_command *cmd) |
b8aa1682 | 1503 | { |
f73fabfd | 1504 | int ret = htobe32(LTTNG_OK); |
6cd525e8 | 1505 | ssize_t size_ret; |
b8aa1682 JD |
1506 | struct relay_session *session = cmd->session; |
1507 | struct lttcomm_relayd_metadata_payload *metadata_struct; | |
1508 | struct relay_stream *metadata_stream; | |
1509 | uint64_t data_size, payload_size; | |
1510 | ||
1511 | if (!session) { | |
1512 | ERR("Metadata sent before version check"); | |
1513 | ret = -1; | |
1514 | goto end; | |
1515 | } | |
1516 | ||
f6416125 MD |
1517 | data_size = payload_size = be64toh(recv_hdr->data_size); |
1518 | if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) { | |
1519 | ERR("Incorrect data size"); | |
1520 | ret = -1; | |
1521 | goto end; | |
1522 | } | |
1523 | payload_size -= sizeof(struct lttcomm_relayd_metadata_payload); | |
1524 | ||
b8aa1682 | 1525 | if (data_buffer_size < data_size) { |
d7b3776f | 1526 | /* In case the realloc fails, we can free the memory */ |
c617c0c6 MD |
1527 | char *tmp_data_ptr; |
1528 | ||
1529 | tmp_data_ptr = realloc(data_buffer, data_size); | |
1530 | if (!tmp_data_ptr) { | |
b8aa1682 | 1531 | ERR("Allocating data buffer"); |
c617c0c6 | 1532 | free(data_buffer); |
b8aa1682 JD |
1533 | ret = -1; |
1534 | goto end; | |
1535 | } | |
c617c0c6 | 1536 | data_buffer = tmp_data_ptr; |
b8aa1682 JD |
1537 | data_buffer_size = data_size; |
1538 | } | |
1539 | memset(data_buffer, 0, data_size); | |
77c7c900 | 1540 | DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size); |
7c5aef62 | 1541 | ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0); |
b8aa1682 | 1542 | if (ret < 0 || ret != data_size) { |
a6cd2b97 DG |
1543 | if (ret == 0) { |
1544 | /* Orderly shutdown. Not necessary to print an error. */ | |
1545 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1546 | } else { | |
1547 | ERR("Relay didn't receive the whole metadata"); | |
1548 | } | |
b8aa1682 | 1549 | ret = -1; |
b8aa1682 JD |
1550 | goto end; |
1551 | } | |
1552 | metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer; | |
9d1bbf21 MD |
1553 | |
1554 | rcu_read_lock(); | |
d3e2ba59 JD |
1555 | metadata_stream = relay_stream_find_by_id( |
1556 | be64toh(metadata_struct->stream_id)); | |
b8aa1682 JD |
1557 | if (!metadata_stream) { |
1558 | ret = -1; | |
9d1bbf21 | 1559 | goto end_unlock; |
b8aa1682 JD |
1560 | } |
1561 | ||
6cd525e8 MD |
1562 | size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload, |
1563 | payload_size); | |
1564 | if (size_ret < payload_size) { | |
b8aa1682 JD |
1565 | ERR("Relay error writing metadata on file"); |
1566 | ret = -1; | |
9d1bbf21 | 1567 | goto end_unlock; |
b8aa1682 | 1568 | } |
1d4dfdef DG |
1569 | |
1570 | ret = write_padding_to_file(metadata_stream->fd, | |
1571 | be32toh(metadata_struct->padding_size)); | |
1572 | if (ret < 0) { | |
1573 | goto end_unlock; | |
1574 | } | |
d3e2ba59 JD |
1575 | metadata_stream->ctf_trace->metadata_received += |
1576 | payload_size + be32toh(metadata_struct->padding_size); | |
1d4dfdef | 1577 | |
b8aa1682 JD |
1578 | DBG2("Relay metadata written"); |
1579 | ||
9d1bbf21 | 1580 | end_unlock: |
6e3c5836 | 1581 | rcu_read_unlock(); |
b8aa1682 JD |
1582 | end: |
1583 | return ret; | |
1584 | } | |
1585 | ||
1586 | /* | |
1587 | * relay_send_version: send relayd version number | |
1588 | */ | |
1589 | static | |
1590 | int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1591 | struct relay_command *cmd, struct lttng_ht *sessions_ht) |
b8aa1682 | 1592 | { |
7f51dcba | 1593 | int ret; |
092b6259 | 1594 | struct lttcomm_relayd_version reply, msg; |
b8aa1682 | 1595 | |
c5b6f4f0 DG |
1596 | assert(cmd); |
1597 | ||
1598 | cmd->version_check_done = 1; | |
b8aa1682 | 1599 | |
092b6259 | 1600 | /* Get version from the other side. */ |
7c5aef62 | 1601 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); |
092b6259 | 1602 | if (ret < 0 || ret != sizeof(msg)) { |
a6cd2b97 DG |
1603 | if (ret == 0) { |
1604 | /* Orderly shutdown. Not necessary to print an error. */ | |
1605 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1606 | } else { | |
1607 | ERR("Relay failed to receive the version values."); | |
1608 | } | |
092b6259 | 1609 | ret = -1; |
092b6259 DG |
1610 | goto end; |
1611 | } | |
1612 | ||
d83a952c MD |
1613 | reply.major = RELAYD_VERSION_COMM_MAJOR; |
1614 | reply.minor = RELAYD_VERSION_COMM_MINOR; | |
d4519fa3 JD |
1615 | |
1616 | /* Major versions must be the same */ | |
1617 | if (reply.major != be32toh(msg.major)) { | |
6151a90f JD |
1618 | DBG("Incompatible major versions (%u vs %u), deleting session", |
1619 | reply.major, be32toh(msg.major)); | |
d3e2ba59 | 1620 | relay_delete_session(cmd, sessions_ht); |
d4519fa3 JD |
1621 | ret = 0; |
1622 | goto end; | |
1623 | } | |
1624 | ||
0f907de1 JD |
1625 | cmd->major = reply.major; |
1626 | /* We adapt to the lowest compatible version */ | |
1627 | if (reply.minor <= be32toh(msg.minor)) { | |
1628 | cmd->minor = reply.minor; | |
1629 | } else { | |
1630 | cmd->minor = be32toh(msg.minor); | |
1631 | } | |
1632 | ||
6151a90f JD |
1633 | reply.major = htobe32(reply.major); |
1634 | reply.minor = htobe32(reply.minor); | |
1635 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1636 | sizeof(struct lttcomm_relayd_version), 0); | |
1637 | if (ret < 0) { | |
1638 | ERR("Relay sending version"); | |
1639 | } | |
1640 | ||
0f907de1 JD |
1641 | DBG("Version check done using protocol %u.%u", cmd->major, |
1642 | cmd->minor); | |
b8aa1682 JD |
1643 | |
1644 | end: | |
1645 | return ret; | |
1646 | } | |
1647 | ||
c8f59ee5 | 1648 | /* |
6d805429 | 1649 | * Check for data pending for a given stream id from the session daemon. |
c8f59ee5 DG |
1650 | */ |
1651 | static | |
6d805429 | 1652 | int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, |
d3e2ba59 | 1653 | struct relay_command *cmd) |
c8f59ee5 DG |
1654 | { |
1655 | struct relay_session *session = cmd->session; | |
6d805429 | 1656 | struct lttcomm_relayd_data_pending msg; |
c8f59ee5 DG |
1657 | struct lttcomm_relayd_generic_reply reply; |
1658 | struct relay_stream *stream; | |
1659 | int ret; | |
c8f59ee5 DG |
1660 | uint64_t last_net_seq_num, stream_id; |
1661 | ||
6d805429 | 1662 | DBG("Data pending command received"); |
c8f59ee5 | 1663 | |
c5b6f4f0 | 1664 | if (!session || cmd->version_check_done == 0) { |
c8f59ee5 DG |
1665 | ERR("Trying to check for data before version check"); |
1666 | ret = -1; | |
1667 | goto end_no_session; | |
1668 | } | |
1669 | ||
7c5aef62 | 1670 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); |
c8f59ee5 | 1671 | if (ret < sizeof(msg)) { |
a6cd2b97 DG |
1672 | if (ret == 0) { |
1673 | /* Orderly shutdown. Not necessary to print an error. */ | |
1674 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1675 | } else { | |
1676 | ERR("Relay didn't receive valid data_pending struct size : %d", | |
1677 | ret); | |
1678 | } | |
c8f59ee5 DG |
1679 | ret = -1; |
1680 | goto end_no_session; | |
1681 | } | |
1682 | ||
1683 | stream_id = be64toh(msg.stream_id); | |
1684 | last_net_seq_num = be64toh(msg.last_net_seq_num); | |
1685 | ||
1686 | rcu_read_lock(); | |
d3e2ba59 | 1687 | stream = relay_stream_find_by_id(stream_id); |
de91f48a | 1688 | if (stream == NULL) { |
c8f59ee5 DG |
1689 | ret = -1; |
1690 | goto end_unlock; | |
1691 | } | |
1692 | ||
6d805429 | 1693 | DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64 |
c8f59ee5 DG |
1694 | " and last_seq %" PRIu64, stream_id, stream->prev_seq, |
1695 | last_net_seq_num); | |
1696 | ||
33832e64 | 1697 | /* Avoid wrapping issue */ |
39df6d9f | 1698 | if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) { |
6d805429 | 1699 | /* Data has in fact been written and is NOT pending */ |
c8f59ee5 | 1700 | ret = 0; |
6d805429 DG |
1701 | } else { |
1702 | /* Data still being streamed thus pending */ | |
1703 | ret = 1; | |
c8f59ee5 DG |
1704 | } |
1705 | ||
f7079f67 DG |
1706 | /* Pending check is now done. */ |
1707 | stream->data_pending_check_done = 1; | |
1708 | ||
c8f59ee5 DG |
1709 | end_unlock: |
1710 | rcu_read_unlock(); | |
1711 | ||
1712 | reply.ret_code = htobe32(ret); | |
1713 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1714 | if (ret < 0) { | |
6d805429 | 1715 | ERR("Relay data pending ret code failed"); |
c8f59ee5 DG |
1716 | } |
1717 | ||
1718 | end_no_session: | |
1719 | return ret; | |
1720 | } | |
1721 | ||
1722 | /* | |
1723 | * Wait for the control socket to reach a quiescent state. | |
1724 | * | |
1725 | * Note that for now, when receiving this command from the session daemon, this | |
1726 | * means that every subsequent commands or data received on the control socket | |
1727 | * has been handled. So, this is why we simply return OK here. | |
1728 | */ | |
1729 | static | |
1730 | int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1731 | struct relay_command *cmd) |
c8f59ee5 DG |
1732 | { |
1733 | int ret; | |
ad7051c0 DG |
1734 | uint64_t stream_id; |
1735 | struct relay_stream *stream; | |
1736 | struct lttng_ht_iter iter; | |
1737 | struct lttcomm_relayd_quiescent_control msg; | |
c8f59ee5 DG |
1738 | struct lttcomm_relayd_generic_reply reply; |
1739 | ||
1740 | DBG("Checking quiescent state on control socket"); | |
1741 | ||
ad7051c0 DG |
1742 | if (!cmd->session || cmd->version_check_done == 0) { |
1743 | ERR("Trying to check for data before version check"); | |
1744 | ret = -1; | |
1745 | goto end_no_session; | |
1746 | } | |
1747 | ||
1748 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1749 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1750 | if (ret == 0) { |
1751 | /* Orderly shutdown. Not necessary to print an error. */ | |
1752 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1753 | } else { | |
1754 | ERR("Relay didn't receive valid begin data_pending struct size: %d", | |
1755 | ret); | |
1756 | } | |
ad7051c0 DG |
1757 | ret = -1; |
1758 | goto end_no_session; | |
1759 | } | |
1760 | ||
1761 | stream_id = be64toh(msg.stream_id); | |
1762 | ||
1763 | rcu_read_lock(); | |
d3e2ba59 JD |
1764 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
1765 | stream_n.node) { | |
ad7051c0 DG |
1766 | if (stream->stream_handle == stream_id) { |
1767 | stream->data_pending_check_done = 1; | |
1768 | DBG("Relay quiescent control pending flag set to %" PRIu64, | |
1769 | stream_id); | |
1770 | break; | |
1771 | } | |
1772 | } | |
1773 | rcu_read_unlock(); | |
1774 | ||
c8f59ee5 DG |
1775 | reply.ret_code = htobe32(LTTNG_OK); |
1776 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1777 | if (ret < 0) { | |
6d805429 | 1778 | ERR("Relay data quiescent control ret code failed"); |
c8f59ee5 DG |
1779 | } |
1780 | ||
ad7051c0 | 1781 | end_no_session: |
c8f59ee5 DG |
1782 | return ret; |
1783 | } | |
1784 | ||
f7079f67 DG |
1785 | /* |
1786 | * Initialize a data pending command. This means that a client is about to ask | |
1787 | * for data pending for each stream he/she holds. Simply iterate over all | |
1788 | * streams of a session and set the data_pending_check_done flag. | |
1789 | * | |
1790 | * This command returns to the client a LTTNG_OK code. | |
1791 | */ | |
1792 | static | |
1793 | int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1794 | struct relay_command *cmd) |
f7079f67 DG |
1795 | { |
1796 | int ret; | |
1797 | struct lttng_ht_iter iter; | |
1798 | struct lttcomm_relayd_begin_data_pending msg; | |
1799 | struct lttcomm_relayd_generic_reply reply; | |
1800 | struct relay_stream *stream; | |
1801 | uint64_t session_id; | |
1802 | ||
1803 | assert(recv_hdr); | |
1804 | assert(cmd); | |
f7079f67 DG |
1805 | |
1806 | DBG("Init streams for data pending"); | |
1807 | ||
1808 | if (!cmd->session || cmd->version_check_done == 0) { | |
1809 | ERR("Trying to check for data before version check"); | |
1810 | ret = -1; | |
1811 | goto end_no_session; | |
1812 | } | |
1813 | ||
1814 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1815 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1816 | if (ret == 0) { |
1817 | /* Orderly shutdown. Not necessary to print an error. */ | |
1818 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1819 | } else { | |
1820 | ERR("Relay didn't receive valid begin data_pending struct size: %d", | |
1821 | ret); | |
1822 | } | |
f7079f67 DG |
1823 | ret = -1; |
1824 | goto end_no_session; | |
1825 | } | |
1826 | ||
1827 | session_id = be64toh(msg.session_id); | |
1828 | ||
1829 | /* | |
1830 | * Iterate over all streams to set the begin data pending flag. For now, the | |
1831 | * streams are indexed by stream handle so we have to iterate over all | |
1832 | * streams to find the one associated with the right session_id. | |
1833 | */ | |
1834 | rcu_read_lock(); | |
d3e2ba59 JD |
1835 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
1836 | stream_n.node) { | |
f7079f67 DG |
1837 | if (stream->session->id == session_id) { |
1838 | stream->data_pending_check_done = 0; | |
1839 | DBG("Set begin data pending flag to stream %" PRIu64, | |
1840 | stream->stream_handle); | |
1841 | } | |
1842 | } | |
1843 | rcu_read_unlock(); | |
1844 | ||
1845 | /* All good, send back reply. */ | |
1846 | reply.ret_code = htobe32(LTTNG_OK); | |
1847 | ||
1848 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1849 | if (ret < 0) { | |
1850 | ERR("Relay begin data pending send reply failed"); | |
1851 | } | |
1852 | ||
1853 | end_no_session: | |
1854 | return ret; | |
1855 | } | |
1856 | ||
1857 | /* | |
1858 | * End data pending command. This will check, for a given session id, if each | |
1859 | * stream associated with it has its data_pending_check_done flag set. If not, | |
1860 | * this means that the client lost track of the stream but the data is still | |
1861 | * being streamed on our side. In this case, we inform the client that data is | |
1862 | * inflight. | |
1863 | * | |
1864 | * Return to the client if there is data in flight or not with a ret_code. | |
1865 | */ | |
1866 | static | |
1867 | int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1868 | struct relay_command *cmd) |
f7079f67 DG |
1869 | { |
1870 | int ret; | |
1871 | struct lttng_ht_iter iter; | |
1872 | struct lttcomm_relayd_end_data_pending msg; | |
1873 | struct lttcomm_relayd_generic_reply reply; | |
1874 | struct relay_stream *stream; | |
1875 | uint64_t session_id; | |
1876 | uint32_t is_data_inflight = 0; | |
1877 | ||
1878 | assert(recv_hdr); | |
1879 | assert(cmd); | |
f7079f67 DG |
1880 | |
1881 | DBG("End data pending command"); | |
1882 | ||
1883 | if (!cmd->session || cmd->version_check_done == 0) { | |
1884 | ERR("Trying to check for data before version check"); | |
1885 | ret = -1; | |
1886 | goto end_no_session; | |
1887 | } | |
1888 | ||
1889 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1890 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1891 | if (ret == 0) { |
1892 | /* Orderly shutdown. Not necessary to print an error. */ | |
1893 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1894 | } else { | |
1895 | ERR("Relay didn't receive valid end data_pending struct size: %d", | |
1896 | ret); | |
1897 | } | |
f7079f67 DG |
1898 | ret = -1; |
1899 | goto end_no_session; | |
1900 | } | |
1901 | ||
1902 | session_id = be64toh(msg.session_id); | |
1903 | ||
1904 | /* Iterate over all streams to see if the begin data pending flag is set. */ | |
1905 | rcu_read_lock(); | |
d3e2ba59 JD |
1906 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
1907 | stream_n.node) { | |
f7079f67 DG |
1908 | if (stream->session->id == session_id && |
1909 | !stream->data_pending_check_done) { | |
1910 | is_data_inflight = 1; | |
1911 | DBG("Data is still in flight for stream %" PRIu64, | |
1912 | stream->stream_handle); | |
1913 | break; | |
1914 | } | |
1915 | } | |
1916 | rcu_read_unlock(); | |
1917 | ||
1918 | /* All good, send back reply. */ | |
1919 | reply.ret_code = htobe32(is_data_inflight); | |
1920 | ||
1921 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1922 | if (ret < 0) { | |
1923 | ERR("Relay end data pending send reply failed"); | |
1924 | } | |
1925 | ||
1926 | end_no_session: | |
1927 | return ret; | |
1928 | } | |
1929 | ||
1c20f0e2 JD |
1930 | /* |
1931 | * Receive an index for a specific stream. | |
1932 | * | |
1933 | * Return 0 on success else a negative value. | |
1934 | */ | |
1935 | static | |
1936 | int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr, | |
0a6518b0 | 1937 | struct relay_command *cmd) |
1c20f0e2 JD |
1938 | { |
1939 | int ret, send_ret, index_created = 0; | |
1940 | struct relay_session *session = cmd->session; | |
1941 | struct lttcomm_relayd_index index_info; | |
1942 | struct relay_index *index, *wr_index = NULL; | |
1943 | struct lttcomm_relayd_generic_reply reply; | |
1944 | struct relay_stream *stream; | |
1945 | uint64_t net_seq_num; | |
1946 | ||
1947 | assert(cmd); | |
1c20f0e2 JD |
1948 | |
1949 | DBG("Relay receiving index"); | |
1950 | ||
1951 | if (!session || cmd->version_check_done == 0) { | |
1952 | ERR("Trying to close a stream before version check"); | |
1953 | ret = -1; | |
1954 | goto end_no_session; | |
1955 | } | |
1956 | ||
1957 | ret = cmd->sock->ops->recvmsg(cmd->sock, &index_info, | |
1958 | sizeof(index_info), 0); | |
1959 | if (ret < sizeof(index_info)) { | |
1960 | if (ret == 0) { | |
1961 | /* Orderly shutdown. Not necessary to print an error. */ | |
1962 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1963 | } else { | |
1964 | ERR("Relay didn't receive valid index struct size : %d", ret); | |
1965 | } | |
1966 | ret = -1; | |
1967 | goto end_no_session; | |
1968 | } | |
1969 | ||
1970 | net_seq_num = be64toh(index_info.net_seq_num); | |
1971 | ||
1972 | rcu_read_lock(); | |
d3e2ba59 | 1973 | stream = relay_stream_find_by_id(be64toh(index_info.relay_stream_id)); |
1c20f0e2 JD |
1974 | if (!stream) { |
1975 | ret = -1; | |
1976 | goto end_rcu_unlock; | |
1977 | } | |
1978 | ||
d3e2ba59 JD |
1979 | /* Live beacon handling */ |
1980 | if (index_info.packet_size == 0) { | |
1981 | DBG("Received live beacon for stream %" PRIu64, stream->stream_handle); | |
1982 | ||
1983 | /* | |
1984 | * Only flag a stream inactive when it has already received data. | |
1985 | */ | |
1986 | if (stream->total_index_received > 0) { | |
1987 | stream->beacon_ts_end = be64toh(index_info.timestamp_end); | |
1988 | } | |
1989 | ret = 0; | |
1990 | goto end_rcu_unlock; | |
1991 | } else { | |
1992 | stream->beacon_ts_end = -1ULL; | |
1993 | } | |
1994 | ||
0a6518b0 | 1995 | index = relay_index_find(stream->stream_handle, net_seq_num); |
1c20f0e2 JD |
1996 | if (!index) { |
1997 | /* A successful creation will add the object to the HT. */ | |
1998 | index = relay_index_create(stream->stream_handle, net_seq_num); | |
1999 | if (!index) { | |
2000 | goto end_rcu_unlock; | |
2001 | } | |
2002 | index_created = 1; | |
2003 | } | |
2004 | ||
2005 | copy_index_control_data(index, &index_info); | |
2006 | ||
2007 | if (index_created) { | |
2008 | /* | |
2009 | * Try to add the relay index object to the hash table. If an object | |
2010 | * already exist, destroy back the index created, set the data in this | |
2011 | * object and write it on disk. | |
2012 | */ | |
0a6518b0 | 2013 | relay_index_add(index, &wr_index); |
1c20f0e2 JD |
2014 | if (wr_index) { |
2015 | copy_index_control_data(wr_index, &index_info); | |
2016 | free(index); | |
2017 | } | |
2018 | } else { | |
2019 | /* The index already exists so write it on disk. */ | |
2020 | wr_index = index; | |
2021 | } | |
2022 | ||
2023 | /* Do we have a writable ready index to write on disk. */ | |
2024 | if (wr_index) { | |
2025 | /* Starting at 2.4, create the index file if none available. */ | |
2026 | if (cmd->minor >= 4 && stream->index_fd < 0) { | |
2027 | ret = index_create_file(stream->path_name, stream->channel_name, | |
2028 | relayd_uid, relayd_gid, stream->tracefile_size, | |
2029 | stream->tracefile_count_current); | |
2030 | if (ret < 0) { | |
2031 | goto end_rcu_unlock; | |
2032 | } | |
2033 | stream->index_fd = ret; | |
2034 | } | |
2035 | ||
0a6518b0 | 2036 | ret = relay_index_write(wr_index->fd, wr_index); |
1c20f0e2 JD |
2037 | if (ret < 0) { |
2038 | goto end_rcu_unlock; | |
2039 | } | |
d3e2ba59 | 2040 | stream->total_index_received++; |
1c20f0e2 JD |
2041 | } |
2042 | ||
2043 | end_rcu_unlock: | |
2044 | rcu_read_unlock(); | |
2045 | ||
2046 | if (ret < 0) { | |
2047 | reply.ret_code = htobe32(LTTNG_ERR_UNK); | |
2048 | } else { | |
2049 | reply.ret_code = htobe32(LTTNG_OK); | |
2050 | } | |
2051 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
2052 | if (send_ret < 0) { | |
2053 | ERR("Relay sending close index id reply"); | |
2054 | ret = send_ret; | |
2055 | } | |
2056 | ||
2057 | end_no_session: | |
2058 | return ret; | |
2059 | } | |
2060 | ||
a4baae1b JD |
2061 | /* |
2062 | * Receive the streams_sent message. | |
2063 | * | |
2064 | * Return 0 on success else a negative value. | |
2065 | */ | |
2066 | static | |
2067 | int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr, | |
2068 | struct relay_command *cmd) | |
2069 | { | |
2070 | int ret, send_ret; | |
2071 | struct lttcomm_relayd_generic_reply reply; | |
2072 | ||
2073 | assert(cmd); | |
2074 | ||
2075 | DBG("Relay receiving streams_sent"); | |
2076 | ||
2077 | if (!cmd->session || cmd->version_check_done == 0) { | |
2078 | ERR("Trying to close a stream before version check"); | |
2079 | ret = -1; | |
2080 | goto end_no_session; | |
2081 | } | |
2082 | ||
2083 | /* | |
2084 | * Flag every pending stream in the connection recv list that they are | |
2085 | * ready to be used by the viewer. | |
2086 | */ | |
2087 | set_viewer_ready_flag(cmd); | |
2088 | ||
2089 | reply.ret_code = htobe32(LTTNG_OK); | |
2090 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
2091 | if (send_ret < 0) { | |
2092 | ERR("Relay sending sent_stream reply"); | |
2093 | ret = send_ret; | |
2094 | } else { | |
2095 | /* Success. */ | |
2096 | ret = 0; | |
2097 | } | |
2098 | ||
2099 | end_no_session: | |
2100 | return ret; | |
2101 | } | |
2102 | ||
b8aa1682 | 2103 | /* |
d3e2ba59 | 2104 | * Process the commands received on the control socket |
b8aa1682 JD |
2105 | */ |
2106 | static | |
2107 | int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 2108 | struct relay_command *cmd, struct relay_local_data *ctx) |
b8aa1682 JD |
2109 | { |
2110 | int ret = 0; | |
2111 | ||
2112 | switch (be32toh(recv_hdr->cmd)) { | |
b8aa1682 | 2113 | case RELAYD_CREATE_SESSION: |
d3e2ba59 | 2114 | ret = relay_create_session(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 | 2115 | break; |
b8aa1682 | 2116 | case RELAYD_ADD_STREAM: |
d3e2ba59 | 2117 | ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 JD |
2118 | break; |
2119 | case RELAYD_START_DATA: | |
2120 | ret = relay_start(recv_hdr, cmd); | |
2121 | break; | |
2122 | case RELAYD_SEND_METADATA: | |
d3e2ba59 | 2123 | ret = relay_recv_metadata(recv_hdr, cmd); |
b8aa1682 JD |
2124 | break; |
2125 | case RELAYD_VERSION: | |
d3e2ba59 | 2126 | ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 | 2127 | break; |
173af62f | 2128 | case RELAYD_CLOSE_STREAM: |
92c6ca54 | 2129 | ret = relay_close_stream(recv_hdr, cmd); |
173af62f | 2130 | break; |
6d805429 | 2131 | case RELAYD_DATA_PENDING: |
d3e2ba59 | 2132 | ret = relay_data_pending(recv_hdr, cmd); |
c8f59ee5 DG |
2133 | break; |
2134 | case RELAYD_QUIESCENT_CONTROL: | |
d3e2ba59 | 2135 | ret = relay_quiescent_control(recv_hdr, cmd); |
c8f59ee5 | 2136 | break; |
f7079f67 | 2137 | case RELAYD_BEGIN_DATA_PENDING: |
d3e2ba59 | 2138 | ret = relay_begin_data_pending(recv_hdr, cmd); |
f7079f67 DG |
2139 | break; |
2140 | case RELAYD_END_DATA_PENDING: | |
d3e2ba59 | 2141 | ret = relay_end_data_pending(recv_hdr, cmd); |
f7079f67 | 2142 | break; |
1c20f0e2 | 2143 | case RELAYD_SEND_INDEX: |
0a6518b0 | 2144 | ret = relay_recv_index(recv_hdr, cmd); |
1c20f0e2 | 2145 | break; |
a4baae1b JD |
2146 | case RELAYD_STREAMS_SENT: |
2147 | ret = relay_streams_sent(recv_hdr, cmd); | |
2148 | break; | |
b8aa1682 JD |
2149 | case RELAYD_UPDATE_SYNC_INFO: |
2150 | default: | |
2151 | ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd)); | |
2152 | relay_unknown_command(cmd); | |
2153 | ret = -1; | |
2154 | goto end; | |
2155 | } | |
2156 | ||
2157 | end: | |
2158 | return ret; | |
2159 | } | |
2160 | ||
7d2f7452 DG |
2161 | /* |
2162 | * Handle index for a data stream. | |
2163 | * | |
2164 | * RCU read side lock MUST be acquired. | |
2165 | * | |
2166 | * Return 0 on success else a negative value. | |
2167 | */ | |
2168 | static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num, | |
2169 | int rotate_index) | |
2170 | { | |
2171 | int ret = 0, index_created = 0; | |
2172 | uint64_t stream_id, data_offset; | |
2173 | struct relay_index *index, *wr_index = NULL; | |
2174 | ||
2175 | assert(stream); | |
2176 | ||
2177 | stream_id = stream->stream_handle; | |
2178 | /* Get data offset because we are about to update the index. */ | |
2179 | data_offset = htobe64(stream->tracefile_size_current); | |
2180 | ||
2181 | /* | |
2182 | * Lookup for an existing index for that stream id/sequence number. If on | |
2183 | * exists, the control thread already received the data for it thus we need | |
2184 | * to write it on disk. | |
2185 | */ | |
2186 | index = relay_index_find(stream_id, net_seq_num); | |
2187 | if (!index) { | |
2188 | /* A successful creation will add the object to the HT. */ | |
2189 | index = relay_index_create(stream_id, net_seq_num); | |
2190 | if (!index) { | |
2191 | ret = -1; | |
2192 | goto error; | |
2193 | } | |
2194 | index_created = 1; | |
2195 | } | |
2196 | ||
2197 | if (rotate_index || stream->index_fd < 0) { | |
2198 | index->to_close_fd = stream->index_fd; | |
2199 | ret = index_create_file(stream->path_name, stream->channel_name, | |
2200 | relayd_uid, relayd_gid, stream->tracefile_size, | |
2201 | stream->tracefile_count_current); | |
2202 | if (ret < 0) { | |
2203 | /* This will close the stream's index fd if one. */ | |
2204 | relay_index_free_safe(index); | |
2205 | goto error; | |
2206 | } | |
2207 | stream->index_fd = ret; | |
2208 | } | |
2209 | index->fd = stream->index_fd; | |
2210 | index->index_data.offset = data_offset; | |
2211 | ||
2212 | if (index_created) { | |
2213 | /* | |
2214 | * Try to add the relay index object to the hash table. If an object | |
2215 | * already exist, destroy back the index created and set the data. | |
2216 | */ | |
2217 | relay_index_add(index, &wr_index); | |
2218 | if (wr_index) { | |
2219 | /* Copy back data from the created index. */ | |
2220 | wr_index->fd = index->fd; | |
2221 | wr_index->to_close_fd = index->to_close_fd; | |
2222 | wr_index->index_data.offset = data_offset; | |
2223 | free(index); | |
2224 | } | |
2225 | } else { | |
2226 | /* The index already exists so write it on disk. */ | |
2227 | wr_index = index; | |
2228 | } | |
2229 | ||
2230 | /* Do we have a writable ready index to write on disk. */ | |
2231 | if (wr_index) { | |
2232 | ret = relay_index_write(wr_index->fd, wr_index); | |
2233 | if (ret < 0) { | |
2234 | goto error; | |
2235 | } | |
2236 | stream->total_index_received++; | |
2237 | } | |
2238 | ||
2239 | error: | |
2240 | return ret; | |
2241 | } | |
2242 | ||
b8aa1682 JD |
2243 | /* |
2244 | * relay_process_data: Process the data received on the data socket | |
2245 | */ | |
2246 | static | |
0a6518b0 | 2247 | int relay_process_data(struct relay_command *cmd) |
b8aa1682 | 2248 | { |
7d2f7452 | 2249 | int ret = 0, rotate_index = 0; |
6cd525e8 | 2250 | ssize_t size_ret; |
b8aa1682 JD |
2251 | struct relay_stream *stream; |
2252 | struct lttcomm_relayd_data_hdr data_hdr; | |
7d2f7452 | 2253 | uint64_t stream_id; |
173af62f | 2254 | uint64_t net_seq_num; |
b8aa1682 JD |
2255 | uint32_t data_size; |
2256 | ||
2257 | ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr, | |
7c5aef62 | 2258 | sizeof(struct lttcomm_relayd_data_hdr), 0); |
b8aa1682 | 2259 | if (ret <= 0) { |
a6cd2b97 DG |
2260 | if (ret == 0) { |
2261 | /* Orderly shutdown. Not necessary to print an error. */ | |
2262 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
2263 | } else { | |
2264 | ERR("Unable to receive data header on sock %d", cmd->sock->fd); | |
2265 | } | |
b8aa1682 JD |
2266 | ret = -1; |
2267 | goto end; | |
2268 | } | |
2269 | ||
2270 | stream_id = be64toh(data_hdr.stream_id); | |
9d1bbf21 MD |
2271 | |
2272 | rcu_read_lock(); | |
d3e2ba59 | 2273 | stream = relay_stream_find_by_id(stream_id); |
b8aa1682 JD |
2274 | if (!stream) { |
2275 | ret = -1; | |
1c20f0e2 | 2276 | goto end_rcu_unlock; |
b8aa1682 JD |
2277 | } |
2278 | ||
2279 | data_size = be32toh(data_hdr.data_size); | |
2280 | if (data_buffer_size < data_size) { | |
c617c0c6 MD |
2281 | char *tmp_data_ptr; |
2282 | ||
2283 | tmp_data_ptr = realloc(data_buffer, data_size); | |
2284 | if (!tmp_data_ptr) { | |
b8aa1682 | 2285 | ERR("Allocating data buffer"); |
c617c0c6 | 2286 | free(data_buffer); |
b8aa1682 | 2287 | ret = -1; |
1c20f0e2 | 2288 | goto end_rcu_unlock; |
b8aa1682 | 2289 | } |
c617c0c6 | 2290 | data_buffer = tmp_data_ptr; |
b8aa1682 JD |
2291 | data_buffer_size = data_size; |
2292 | } | |
2293 | memset(data_buffer, 0, data_size); | |
2294 | ||
173af62f DG |
2295 | net_seq_num = be64toh(data_hdr.net_seq_num); |
2296 | ||
77c7c900 | 2297 | DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64, |
173af62f | 2298 | data_size, stream_id, net_seq_num); |
7c5aef62 | 2299 | ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0); |
b8aa1682 | 2300 | if (ret <= 0) { |
a6cd2b97 DG |
2301 | if (ret == 0) { |
2302 | /* Orderly shutdown. Not necessary to print an error. */ | |
2303 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
2304 | } | |
b8aa1682 | 2305 | ret = -1; |
1c20f0e2 | 2306 | goto end_rcu_unlock; |
b8aa1682 JD |
2307 | } |
2308 | ||
1c20f0e2 | 2309 | /* Check if a rotation is needed. */ |
0f907de1 JD |
2310 | if (stream->tracefile_size > 0 && |
2311 | (stream->tracefile_size_current + data_size) > | |
2312 | stream->tracefile_size) { | |
6b6b9a5a JD |
2313 | struct relay_viewer_stream *vstream; |
2314 | uint64_t new_id; | |
2315 | ||
2316 | new_id = (stream->tracefile_count_current + 1) % | |
2317 | stream->tracefile_count; | |
2318 | /* | |
2319 | * When we wrap-around back to 0, we start overwriting old | |
2320 | * trace data. | |
2321 | */ | |
2322 | if (!stream->tracefile_overwrite && new_id == 0) { | |
2323 | stream->tracefile_overwrite = 1; | |
2324 | } | |
2325 | pthread_mutex_lock(&stream->viewer_stream_rotation_lock); | |
2326 | if (stream->tracefile_overwrite) { | |
2327 | stream->oldest_tracefile_id = | |
2328 | (stream->oldest_tracefile_id + 1) % | |
2329 | stream->tracefile_count; | |
2330 | } | |
2331 | vstream = live_find_viewer_stream_by_id(stream->stream_handle); | |
2332 | if (vstream) { | |
2333 | /* | |
2334 | * The viewer is reading a file about to be | |
2335 | * overwritten. Close the FDs it is | |
2336 | * currently using and let it handle the fault. | |
2337 | */ | |
2338 | if (vstream->tracefile_count_current == new_id) { | |
cef0f7d5 | 2339 | pthread_mutex_lock(&vstream->overwrite_lock); |
6b6b9a5a | 2340 | vstream->abort_flag = 1; |
cef0f7d5 | 2341 | pthread_mutex_unlock(&vstream->overwrite_lock); |
6b6b9a5a JD |
2342 | DBG("Streaming side setting abort_flag on stream %s_%lu\n", |
2343 | stream->channel_name, new_id); | |
2344 | } else if (vstream->tracefile_count_current == | |
2345 | stream->tracefile_count_current) { | |
2346 | /* | |
2347 | * The reader and writer were in the | |
2348 | * same trace file, inform the viewer | |
2349 | * that no new index will ever be added | |
2350 | * to this file. | |
2351 | */ | |
2352 | vstream->close_write_flag = 1; | |
2353 | } | |
2354 | } | |
1c20f0e2 JD |
2355 | ret = utils_rotate_stream_file(stream->path_name, stream->channel_name, |
2356 | stream->tracefile_size, stream->tracefile_count, | |
2357 | relayd_uid, relayd_gid, stream->fd, | |
2358 | &(stream->tracefile_count_current), &stream->fd); | |
cef0f7d5 | 2359 | stream->total_index_received = 0; |
6b6b9a5a | 2360 | pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); |
0f907de1 | 2361 | if (ret < 0) { |
1c20f0e2 JD |
2362 | ERR("Rotating stream output file"); |
2363 | goto end_rcu_unlock; | |
0f907de1 | 2364 | } |
a6976990 DG |
2365 | /* Reset current size because we just perform a stream rotation. */ |
2366 | stream->tracefile_size_current = 0; | |
1c20f0e2 JD |
2367 | rotate_index = 1; |
2368 | } | |
2369 | ||
1c20f0e2 | 2370 | /* |
7d2f7452 DG |
2371 | * Index are handled in protocol version 2.4 and above. Also, snapshot and |
2372 | * index are NOT supported. | |
1c20f0e2 | 2373 | */ |
7d2f7452 DG |
2374 | if (stream->session->minor >= 4 && !stream->session->snapshot) { |
2375 | ret = handle_index_data(stream, net_seq_num, rotate_index); | |
1c20f0e2 | 2376 | if (ret < 0) { |
1c20f0e2 JD |
2377 | goto end_rcu_unlock; |
2378 | } | |
1c20f0e2 JD |
2379 | } |
2380 | ||
7d2f7452 | 2381 | /* Write data to stream output fd. */ |
6cd525e8 MD |
2382 | size_ret = lttng_write(stream->fd, data_buffer, data_size); |
2383 | if (size_ret < data_size) { | |
b8aa1682 JD |
2384 | ERR("Relay error writing data to file"); |
2385 | ret = -1; | |
1c20f0e2 | 2386 | goto end_rcu_unlock; |
b8aa1682 | 2387 | } |
1d4dfdef | 2388 | |
5ab7344e JD |
2389 | DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64, |
2390 | ret, stream->stream_handle); | |
2391 | ||
1d4dfdef DG |
2392 | ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size)); |
2393 | if (ret < 0) { | |
1c20f0e2 | 2394 | goto end_rcu_unlock; |
1d4dfdef | 2395 | } |
1c20f0e2 | 2396 | stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size); |
1d4dfdef | 2397 | |
173af62f DG |
2398 | stream->prev_seq = net_seq_num; |
2399 | ||
2400 | /* Check if we need to close the FD */ | |
2401 | if (close_stream_check(stream)) { | |
c07d8a78 | 2402 | destroy_stream(stream); |
173af62f DG |
2403 | } |
2404 | ||
1c20f0e2 | 2405 | end_rcu_unlock: |
9d1bbf21 | 2406 | rcu_read_unlock(); |
b8aa1682 JD |
2407 | end: |
2408 | return ret; | |
2409 | } | |
2410 | ||
2411 | static | |
9d1bbf21 | 2412 | void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd) |
b8aa1682 JD |
2413 | { |
2414 | int ret; | |
2415 | ||
b8aa1682 JD |
2416 | lttng_poll_del(events, pollfd); |
2417 | ||
2418 | ret = close(pollfd); | |
2419 | if (ret < 0) { | |
2420 | ERR("Closing pollfd %d", pollfd); | |
2421 | } | |
2422 | } | |
2423 | ||
2424 | static | |
2425 | int relay_add_connection(int fd, struct lttng_poll_event *events, | |
2426 | struct lttng_ht *relay_connections_ht) | |
2427 | { | |
b8aa1682 | 2428 | struct relay_command *relay_connection; |
6cd525e8 | 2429 | ssize_t ret; |
b8aa1682 JD |
2430 | |
2431 | relay_connection = zmalloc(sizeof(struct relay_command)); | |
2432 | if (relay_connection == NULL) { | |
2433 | PERROR("Relay command zmalloc"); | |
9d1bbf21 | 2434 | goto error; |
b8aa1682 | 2435 | } |
6cd525e8 MD |
2436 | ret = lttng_read(fd, relay_connection, sizeof(struct relay_command)); |
2437 | if (ret < sizeof(struct relay_command)) { | |
b8aa1682 | 2438 | PERROR("read relay cmd pipe"); |
9d1bbf21 | 2439 | goto error_read; |
b8aa1682 | 2440 | } |
a4baae1b | 2441 | CDS_INIT_LIST_HEAD(&relay_connection->recv_head); |
b8aa1682 | 2442 | |
c07d8a78 DG |
2443 | /* |
2444 | * Only used by the control side and the reference is copied inside each | |
2445 | * stream from that connection. Thus a destroy HT must be done after every | |
2446 | * stream has been destroyed. | |
2447 | */ | |
2448 | if (relay_connection->type == RELAY_CONTROL) { | |
2449 | relay_connection->ctf_traces_ht = lttng_ht_new(0, | |
2450 | LTTNG_HT_TYPE_STRING); | |
2451 | if (!relay_connection->ctf_traces_ht) { | |
2452 | goto error_read; | |
2453 | } | |
d3e2ba59 JD |
2454 | } |
2455 | ||
b8aa1682 JD |
2456 | lttng_ht_node_init_ulong(&relay_connection->sock_n, |
2457 | (unsigned long) relay_connection->sock->fd); | |
9d1bbf21 | 2458 | rcu_read_lock(); |
b8aa1682 JD |
2459 | lttng_ht_add_unique_ulong(relay_connections_ht, |
2460 | &relay_connection->sock_n); | |
9d1bbf21 MD |
2461 | rcu_read_unlock(); |
2462 | return lttng_poll_add(events, | |
b8aa1682 JD |
2463 | relay_connection->sock->fd, |
2464 | LPOLLIN | LPOLLRDHUP); | |
2465 | ||
9d1bbf21 MD |
2466 | error_read: |
2467 | free(relay_connection); | |
2468 | error: | |
2469 | return -1; | |
2470 | } | |
2471 | ||
2472 | static | |
2473 | void deferred_free_connection(struct rcu_head *head) | |
2474 | { | |
2475 | struct relay_command *relay_connection = | |
2476 | caa_container_of(head, struct relay_command, rcu_node); | |
5b6d8097 DG |
2477 | |
2478 | lttcomm_destroy_sock(relay_connection->sock); | |
9d1bbf21 MD |
2479 | free(relay_connection); |
2480 | } | |
2481 | ||
2482 | static | |
2483 | void relay_del_connection(struct lttng_ht *relay_connections_ht, | |
d3e2ba59 JD |
2484 | struct lttng_ht_iter *iter, struct relay_command *relay_connection, |
2485 | struct lttng_ht *sessions_ht) | |
9d1bbf21 MD |
2486 | { |
2487 | int ret; | |
2488 | ||
2489 | ret = lttng_ht_del(relay_connections_ht, iter); | |
2490 | assert(!ret); | |
c07d8a78 | 2491 | |
9d1bbf21 | 2492 | if (relay_connection->type == RELAY_CONTROL) { |
a4baae1b JD |
2493 | struct relay_stream_recv_handle *node, *tmp_node; |
2494 | ||
d3e2ba59 | 2495 | relay_delete_session(relay_connection, sessions_ht); |
c07d8a78 | 2496 | lttng_ht_destroy(relay_connection->ctf_traces_ht); |
a4baae1b JD |
2497 | |
2498 | /* Clean up recv list. */ | |
2499 | cds_list_for_each_entry_safe(node, tmp_node, | |
2500 | &relay_connection->recv_head, node) { | |
2501 | cds_list_del(&node->node); | |
2502 | free(node); | |
2503 | } | |
9d1bbf21 | 2504 | } |
5b6d8097 | 2505 | |
c07d8a78 | 2506 | call_rcu(&relay_connection->rcu_node, deferred_free_connection); |
b8aa1682 JD |
2507 | } |
2508 | ||
2509 | /* | |
2510 | * This thread does the actual work | |
2511 | */ | |
2512 | static | |
2513 | void *relay_thread_worker(void *data) | |
2514 | { | |
beaad64c DG |
2515 | int ret, err = -1, last_seen_data_fd = -1; |
2516 | uint32_t nb_fd; | |
b8aa1682 JD |
2517 | struct relay_command *relay_connection; |
2518 | struct lttng_poll_event events; | |
2519 | struct lttng_ht *relay_connections_ht; | |
2520 | struct lttng_ht_node_ulong *node; | |
2521 | struct lttng_ht_iter iter; | |
b8aa1682 | 2522 | struct lttcomm_relayd_hdr recv_hdr; |
d3e2ba59 JD |
2523 | struct relay_local_data *relay_ctx = (struct relay_local_data *) data; |
2524 | struct lttng_ht *sessions_ht = relay_ctx->sessions_ht; | |
b8aa1682 JD |
2525 | |
2526 | DBG("[thread] Relay worker started"); | |
2527 | ||
9d1bbf21 MD |
2528 | rcu_register_thread(); |
2529 | ||
55706a7d MD |
2530 | health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER); |
2531 | ||
f385ae0a MD |
2532 | health_code_update(); |
2533 | ||
b8aa1682 JD |
2534 | /* table of connections indexed on socket */ |
2535 | relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
095a4ae5 MD |
2536 | if (!relay_connections_ht) { |
2537 | goto relay_connections_ht_error; | |
2538 | } | |
b8aa1682 | 2539 | |
1c20f0e2 JD |
2540 | /* Tables of received indexes indexed by index handle and net_seq_num. */ |
2541 | indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64); | |
2542 | if (!indexes_ht) { | |
2543 | goto indexes_ht_error; | |
2544 | } | |
2545 | ||
b8aa1682 JD |
2546 | ret = create_thread_poll_set(&events, 2); |
2547 | if (ret < 0) { | |
2548 | goto error_poll_create; | |
2549 | } | |
2550 | ||
2551 | ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP); | |
2552 | if (ret < 0) { | |
2553 | goto error; | |
2554 | } | |
2555 | ||
beaad64c | 2556 | restart: |
b8aa1682 | 2557 | while (1) { |
beaad64c DG |
2558 | int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1; |
2559 | ||
f385ae0a MD |
2560 | health_code_update(); |
2561 | ||
b8aa1682 | 2562 | /* Infinite blocking call, waiting for transmission */ |
87c1611d | 2563 | DBG3("Relayd worker thread polling..."); |
f385ae0a | 2564 | health_poll_entry(); |
b8aa1682 | 2565 | ret = lttng_poll_wait(&events, -1); |
f385ae0a | 2566 | health_poll_exit(); |
b8aa1682 JD |
2567 | if (ret < 0) { |
2568 | /* | |
2569 | * Restart interrupted system call. | |
2570 | */ | |
2571 | if (errno == EINTR) { | |
2572 | goto restart; | |
2573 | } | |
2574 | goto error; | |
2575 | } | |
2576 | ||
0d9c5d77 DG |
2577 | nb_fd = ret; |
2578 | ||
beaad64c DG |
2579 | /* |
2580 | * Process control. The control connection is prioritised so we don't | |
2581 | * starve it with high throughout put tracing data on the data | |
2582 | * connection. | |
2583 | */ | |
b8aa1682 JD |
2584 | for (i = 0; i < nb_fd; i++) { |
2585 | /* Fetch once the poll data */ | |
beaad64c DG |
2586 | uint32_t revents = LTTNG_POLL_GETEV(&events, i); |
2587 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
b8aa1682 | 2588 | |
f385ae0a MD |
2589 | health_code_update(); |
2590 | ||
b8aa1682 JD |
2591 | /* Thread quit pipe has been closed. Killing thread. */ |
2592 | ret = check_thread_quit_pipe(pollfd, revents); | |
2593 | if (ret) { | |
095a4ae5 MD |
2594 | err = 0; |
2595 | goto exit; | |
b8aa1682 JD |
2596 | } |
2597 | ||
2598 | /* Inspect the relay cmd pipe for new connection */ | |
2599 | if (pollfd == relay_cmd_pipe[0]) { | |
2600 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
2601 | ERR("Relay pipe error"); | |
2602 | goto error; | |
2603 | } else if (revents & LPOLLIN) { | |
2604 | DBG("Relay command received"); | |
2605 | ret = relay_add_connection(relay_cmd_pipe[0], | |
2606 | &events, relay_connections_ht); | |
2607 | if (ret < 0) { | |
2608 | goto error; | |
2609 | } | |
2610 | } | |
beaad64c | 2611 | } else if (revents) { |
9d1bbf21 | 2612 | rcu_read_lock(); |
b8aa1682 JD |
2613 | lttng_ht_lookup(relay_connections_ht, |
2614 | (void *)((unsigned long) pollfd), | |
2615 | &iter); | |
2616 | node = lttng_ht_iter_get_node_ulong(&iter); | |
2617 | if (node == NULL) { | |
2618 | DBG2("Relay sock %d not found", pollfd); | |
9d1bbf21 | 2619 | rcu_read_unlock(); |
b8aa1682 JD |
2620 | goto error; |
2621 | } | |
2622 | relay_connection = caa_container_of(node, | |
2623 | struct relay_command, sock_n); | |
2624 | ||
2625 | if (revents & (LPOLLERR)) { | |
2626 | ERR("POLL ERROR"); | |
9d1bbf21 MD |
2627 | relay_cleanup_poll_connection(&events, pollfd); |
2628 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2629 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2630 | if (last_seen_data_fd == pollfd) { |
2631 | last_seen_data_fd = last_notdel_data_fd; | |
2632 | } | |
b8aa1682 JD |
2633 | } else if (revents & (LPOLLHUP | LPOLLRDHUP)) { |
2634 | DBG("Socket %d hung up", pollfd); | |
9d1bbf21 MD |
2635 | relay_cleanup_poll_connection(&events, pollfd); |
2636 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2637 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2638 | if (last_seen_data_fd == pollfd) { |
2639 | last_seen_data_fd = last_notdel_data_fd; | |
2640 | } | |
b8aa1682 JD |
2641 | } else if (revents & LPOLLIN) { |
2642 | /* control socket */ | |
2643 | if (relay_connection->type == RELAY_CONTROL) { | |
2644 | ret = relay_connection->sock->ops->recvmsg( | |
2645 | relay_connection->sock, &recv_hdr, | |
7c5aef62 | 2646 | sizeof(struct lttcomm_relayd_hdr), 0); |
b8aa1682 JD |
2647 | /* connection closed */ |
2648 | if (ret <= 0) { | |
9d1bbf21 MD |
2649 | relay_cleanup_poll_connection(&events, pollfd); |
2650 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2651 | &iter, relay_connection, sessions_ht); |
b8aa1682 JD |
2652 | DBG("Control connection closed with %d", pollfd); |
2653 | } else { | |
2654 | if (relay_connection->session) { | |
77c7c900 | 2655 | DBG2("Relay worker receiving data for session : %" PRIu64, |
b8aa1682 JD |
2656 | relay_connection->session->id); |
2657 | } | |
2658 | ret = relay_process_control(&recv_hdr, | |
d3e2ba59 | 2659 | relay_connection, relay_ctx); |
b8aa1682 | 2660 | if (ret < 0) { |
beaad64c | 2661 | /* Clear the session on error. */ |
9d1bbf21 MD |
2662 | relay_cleanup_poll_connection(&events, pollfd); |
2663 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2664 | &iter, relay_connection, sessions_ht); |
b8aa1682 JD |
2665 | DBG("Connection closed with %d", pollfd); |
2666 | } | |
beaad64c | 2667 | seen_control = 1; |
b8aa1682 | 2668 | } |
beaad64c DG |
2669 | } else { |
2670 | /* | |
2671 | * Flag the last seen data fd not deleted. It will be | |
2672 | * used as the last seen fd if any fd gets deleted in | |
2673 | * this first loop. | |
2674 | */ | |
2675 | last_notdel_data_fd = pollfd; | |
2676 | } | |
2677 | } | |
2678 | rcu_read_unlock(); | |
2679 | } | |
2680 | } | |
2681 | ||
2682 | /* | |
2683 | * The last loop handled a control request, go back to poll to make | |
2684 | * sure we prioritise the control socket. | |
2685 | */ | |
2686 | if (seen_control) { | |
2687 | continue; | |
2688 | } | |
2689 | ||
2690 | if (last_seen_data_fd >= 0) { | |
2691 | for (i = 0; i < nb_fd; i++) { | |
2692 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
f385ae0a MD |
2693 | |
2694 | health_code_update(); | |
2695 | ||
beaad64c DG |
2696 | if (last_seen_data_fd == pollfd) { |
2697 | idx = i; | |
2698 | break; | |
2699 | } | |
2700 | } | |
2701 | } | |
2702 | ||
2703 | /* Process data connection. */ | |
2704 | for (i = idx + 1; i < nb_fd; i++) { | |
2705 | /* Fetch the poll data. */ | |
2706 | uint32_t revents = LTTNG_POLL_GETEV(&events, i); | |
2707 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
2708 | ||
f385ae0a MD |
2709 | health_code_update(); |
2710 | ||
beaad64c DG |
2711 | /* Skip the command pipe. It's handled in the first loop. */ |
2712 | if (pollfd == relay_cmd_pipe[0]) { | |
2713 | continue; | |
2714 | } | |
2715 | ||
2716 | if (revents) { | |
2717 | rcu_read_lock(); | |
2718 | lttng_ht_lookup(relay_connections_ht, | |
2719 | (void *)((unsigned long) pollfd), | |
2720 | &iter); | |
2721 | node = lttng_ht_iter_get_node_ulong(&iter); | |
2722 | if (node == NULL) { | |
2723 | /* Skip it. Might be removed before. */ | |
2724 | rcu_read_unlock(); | |
2725 | continue; | |
2726 | } | |
2727 | relay_connection = caa_container_of(node, | |
2728 | struct relay_command, sock_n); | |
2729 | ||
2730 | if (revents & LPOLLIN) { | |
2731 | if (relay_connection->type != RELAY_DATA) { | |
2732 | continue; | |
2733 | } | |
2734 | ||
0a6518b0 | 2735 | ret = relay_process_data(relay_connection); |
beaad64c DG |
2736 | /* connection closed */ |
2737 | if (ret < 0) { | |
2738 | relay_cleanup_poll_connection(&events, pollfd); | |
2739 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2740 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2741 | DBG("Data connection closed with %d", pollfd); |
2742 | /* | |
2743 | * Every goto restart call sets the last seen fd where | |
2744 | * here we don't really care since we gracefully | |
2745 | * continue the loop after the connection is deleted. | |
2746 | */ | |
2747 | } else { | |
2748 | /* Keep last seen port. */ | |
2749 | last_seen_data_fd = pollfd; | |
2750 | rcu_read_unlock(); | |
2751 | goto restart; | |
b8aa1682 JD |
2752 | } |
2753 | } | |
9d1bbf21 | 2754 | rcu_read_unlock(); |
b8aa1682 JD |
2755 | } |
2756 | } | |
beaad64c | 2757 | last_seen_data_fd = -1; |
b8aa1682 JD |
2758 | } |
2759 | ||
f385ae0a MD |
2760 | /* Normal exit, no error */ |
2761 | ret = 0; | |
2762 | ||
095a4ae5 | 2763 | exit: |
b8aa1682 JD |
2764 | error: |
2765 | lttng_poll_clean(&events); | |
2766 | ||
2767 | /* empty the hash table and free the memory */ | |
9d1bbf21 | 2768 | rcu_read_lock(); |
b8aa1682 | 2769 | cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) { |
f385ae0a MD |
2770 | health_code_update(); |
2771 | ||
b8aa1682 JD |
2772 | node = lttng_ht_iter_get_node_ulong(&iter); |
2773 | if (node) { | |
2774 | relay_connection = caa_container_of(node, | |
2775 | struct relay_command, sock_n); | |
9d1bbf21 | 2776 | relay_del_connection(relay_connections_ht, |
d3e2ba59 | 2777 | &iter, relay_connection, sessions_ht); |
b8aa1682 | 2778 | } |
b8aa1682 | 2779 | } |
94d49140 | 2780 | rcu_read_unlock(); |
7d2f7452 DG |
2781 | error_poll_create: |
2782 | lttng_ht_destroy(indexes_ht); | |
1c20f0e2 | 2783 | indexes_ht_error: |
b8aa1682 | 2784 | lttng_ht_destroy(relay_connections_ht); |
095a4ae5 | 2785 | relay_connections_ht_error: |
6620da75 DG |
2786 | /* Close relay cmd pipes */ |
2787 | utils_close_pipe(relay_cmd_pipe); | |
095a4ae5 MD |
2788 | if (err) { |
2789 | DBG("Thread exited with error"); | |
2790 | } | |
b8aa1682 | 2791 | DBG("Worker thread cleanup complete"); |
095a4ae5 | 2792 | free(data_buffer); |
f385ae0a MD |
2793 | if (err) { |
2794 | health_error(); | |
2795 | ERR("Health error occurred in %s", __func__); | |
2796 | } | |
2797 | health_unregister(health_relayd); | |
9d1bbf21 | 2798 | rcu_unregister_thread(); |
f385ae0a | 2799 | stop_threads(); |
b8aa1682 JD |
2800 | return NULL; |
2801 | } | |
2802 | ||
2803 | /* | |
2804 | * Create the relay command pipe to wake thread_manage_apps. | |
2805 | * Closed in cleanup(). | |
2806 | */ | |
2807 | static int create_relay_cmd_pipe(void) | |
2808 | { | |
a02de639 | 2809 | int ret; |
b8aa1682 | 2810 | |
a02de639 | 2811 | ret = utils_create_pipe_cloexec(relay_cmd_pipe); |
b8aa1682 | 2812 | |
b8aa1682 JD |
2813 | return ret; |
2814 | } | |
2815 | ||
2816 | /* | |
2817 | * main | |
2818 | */ | |
2819 | int main(int argc, char **argv) | |
2820 | { | |
2821 | int ret = 0; | |
2822 | void *status; | |
d3e2ba59 | 2823 | struct relay_local_data *relay_ctx; |
b8aa1682 JD |
2824 | |
2825 | /* Create thread quit pipe */ | |
2826 | if ((ret = init_thread_quit_pipe()) < 0) { | |
2827 | goto error; | |
2828 | } | |
2829 | ||
2830 | /* Parse arguments */ | |
2831 | progname = argv[0]; | |
cd60b05a | 2832 | if ((ret = set_options(argc, argv)) < 0) { |
a02de639 | 2833 | goto exit; |
b8aa1682 JD |
2834 | } |
2835 | ||
2836 | if ((ret = set_signal_handler()) < 0) { | |
2837 | goto exit; | |
2838 | } | |
2839 | ||
4d513a50 DG |
2840 | /* Try to create directory if -o, --output is specified. */ |
2841 | if (opt_output_path) { | |
994fa64f DG |
2842 | if (*opt_output_path != '/') { |
2843 | ERR("Please specify an absolute path for -o, --output PATH"); | |
2844 | goto exit; | |
2845 | } | |
2846 | ||
4d513a50 DG |
2847 | ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG); |
2848 | if (ret < 0) { | |
2849 | ERR("Unable to create %s", opt_output_path); | |
2850 | goto exit; | |
2851 | } | |
2852 | } | |
2853 | ||
b8aa1682 | 2854 | /* Daemonize */ |
b5218ffb MD |
2855 | if (opt_daemon || opt_background) { |
2856 | ret = daemon(0, opt_background); | |
b8aa1682 JD |
2857 | if (ret < 0) { |
2858 | PERROR("daemon"); | |
a02de639 | 2859 | goto exit; |
b8aa1682 JD |
2860 | } |
2861 | } | |
2862 | ||
1c20f0e2 JD |
2863 | /* We need those values for the file/dir creation. */ |
2864 | relayd_uid = getuid(); | |
2865 | relayd_gid = getgid(); | |
b8aa1682 | 2866 | |
1c20f0e2 JD |
2867 | /* Check if daemon is UID = 0 */ |
2868 | if (relayd_uid == 0) { | |
8d5c808e | 2869 | if (control_uri->port < 1024 || data_uri->port < 1024 || live_uri->port < 1024) { |
b8aa1682 JD |
2870 | ERR("Need to be root to use ports < 1024"); |
2871 | ret = -1; | |
a02de639 | 2872 | goto exit; |
b8aa1682 JD |
2873 | } |
2874 | } | |
2875 | ||
2876 | /* Setup the thread apps communication pipe. */ | |
2877 | if ((ret = create_relay_cmd_pipe()) < 0) { | |
2878 | goto exit; | |
2879 | } | |
2880 | ||
2881 | /* Init relay command queue. */ | |
2882 | cds_wfq_init(&relay_cmd_queue.queue); | |
2883 | ||
2884 | /* Set up max poll set size */ | |
2885 | lttng_poll_set_max_size(); | |
2886 | ||
554831e7 MD |
2887 | /* Initialize communication library */ |
2888 | lttcomm_init(); | |
87e45c13 | 2889 | lttcomm_inet_init(); |
554831e7 | 2890 | |
d3e2ba59 JD |
2891 | relay_ctx = zmalloc(sizeof(struct relay_local_data)); |
2892 | if (!relay_ctx) { | |
2893 | PERROR("relay_ctx"); | |
2894 | goto exit; | |
2895 | } | |
2896 | ||
2897 | /* tables of sessions indexed by session ID */ | |
2898 | relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
2899 | if (!relay_ctx->sessions_ht) { | |
2900 | goto exit_relay_ctx_sessions; | |
2901 | } | |
2902 | ||
2903 | /* tables of streams indexed by stream ID */ | |
2904 | relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
2905 | if (!relay_streams_ht) { | |
2906 | goto exit_relay_ctx_streams; | |
2907 | } | |
2908 | ||
2909 | /* tables of streams indexed by stream ID */ | |
92c6ca54 DG |
2910 | viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
2911 | if (!viewer_streams_ht) { | |
d3e2ba59 JD |
2912 | goto exit_relay_ctx_viewer_streams; |
2913 | } | |
2914 | ||
55706a7d MD |
2915 | /* Initialize thread health monitoring */ |
2916 | health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES); | |
2917 | if (!health_relayd) { | |
2918 | PERROR("health_app_create error"); | |
2919 | goto exit_health_app_create; | |
2920 | } | |
2921 | ||
65931c8b MD |
2922 | ret = utils_create_pipe(health_quit_pipe); |
2923 | if (ret < 0) { | |
2924 | goto error_health_pipe; | |
2925 | } | |
2926 | ||
2927 | /* Create thread to manage the client socket */ | |
2928 | ret = pthread_create(&health_thread, NULL, | |
2929 | thread_manage_health, (void *) NULL); | |
2930 | if (ret != 0) { | |
2931 | PERROR("pthread_create health"); | |
2932 | goto health_error; | |
2933 | } | |
2934 | ||
b8aa1682 JD |
2935 | /* Setup the dispatcher thread */ |
2936 | ret = pthread_create(&dispatcher_thread, NULL, | |
2937 | relay_thread_dispatcher, (void *) NULL); | |
2938 | if (ret != 0) { | |
2939 | PERROR("pthread_create dispatcher"); | |
2940 | goto exit_dispatcher; | |
2941 | } | |
2942 | ||
2943 | /* Setup the worker thread */ | |
2944 | ret = pthread_create(&worker_thread, NULL, | |
d3e2ba59 | 2945 | relay_thread_worker, (void *) relay_ctx); |
b8aa1682 JD |
2946 | if (ret != 0) { |
2947 | PERROR("pthread_create worker"); | |
2948 | goto exit_worker; | |
2949 | } | |
2950 | ||
2951 | /* Setup the listener thread */ | |
2952 | ret = pthread_create(&listener_thread, NULL, | |
2953 | relay_thread_listener, (void *) NULL); | |
2954 | if (ret != 0) { | |
2955 | PERROR("pthread_create listener"); | |
2956 | goto exit_listener; | |
2957 | } | |
2958 | ||
42415026 | 2959 | ret = live_start_threads(live_uri, relay_ctx, thread_quit_pipe); |
d3e2ba59 JD |
2960 | if (ret != 0) { |
2961 | ERR("Starting live viewer threads"); | |
50138f51 | 2962 | goto exit_live; |
d3e2ba59 JD |
2963 | } |
2964 | ||
50138f51 | 2965 | exit_live: |
b8aa1682 JD |
2966 | ret = pthread_join(listener_thread, &status); |
2967 | if (ret != 0) { | |
2968 | PERROR("pthread_join"); | |
2969 | goto error; /* join error, exit without cleanup */ | |
2970 | } | |
2971 | ||
50138f51 | 2972 | exit_listener: |
b8aa1682 JD |
2973 | ret = pthread_join(worker_thread, &status); |
2974 | if (ret != 0) { | |
2975 | PERROR("pthread_join"); | |
2976 | goto error; /* join error, exit without cleanup */ | |
2977 | } | |
2978 | ||
50138f51 | 2979 | exit_worker: |
b8aa1682 JD |
2980 | ret = pthread_join(dispatcher_thread, &status); |
2981 | if (ret != 0) { | |
2982 | PERROR("pthread_join"); | |
2983 | goto error; /* join error, exit without cleanup */ | |
2984 | } | |
42415026 | 2985 | |
50138f51 | 2986 | exit_dispatcher: |
65931c8b MD |
2987 | ret = pthread_join(health_thread, &status); |
2988 | if (ret != 0) { | |
2989 | PERROR("pthread_join health thread"); | |
2990 | goto error; /* join error, exit without cleanup */ | |
2991 | } | |
2992 | ||
bd11e201 MD |
2993 | /* |
2994 | * Stop live threads only after joining other threads. | |
2995 | */ | |
2996 | live_stop_threads(); | |
2997 | ||
65931c8b MD |
2998 | health_error: |
2999 | utils_close_pipe(health_quit_pipe); | |
3000 | ||
3001 | error_health_pipe: | |
55706a7d MD |
3002 | health_app_destroy(health_relayd); |
3003 | ||
3004 | exit_health_app_create: | |
92c6ca54 | 3005 | lttng_ht_destroy(viewer_streams_ht); |
d3e2ba59 JD |
3006 | |
3007 | exit_relay_ctx_viewer_streams: | |
3008 | lttng_ht_destroy(relay_streams_ht); | |
3009 | ||
3010 | exit_relay_ctx_streams: | |
3011 | lttng_ht_destroy(relay_ctx->sessions_ht); | |
3012 | ||
3013 | exit_relay_ctx_sessions: | |
3014 | free(relay_ctx); | |
b8aa1682 JD |
3015 | |
3016 | exit: | |
3017 | cleanup(); | |
3018 | if (!ret) { | |
3019 | exit(EXIT_SUCCESS); | |
3020 | } | |
a02de639 | 3021 | |
b8aa1682 JD |
3022 | error: |
3023 | exit(EXIT_FAILURE); | |
3024 | } |