Health check: implement health check query in sessiond and consumerd
[lttng-tools.git] / src / bin / lttng-consumerd / lttng-consumerd.c
1 /*
2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _GNU_SOURCE
20 #include <fcntl.h>
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/ipc.h>
30 #include <sys/resource.h>
31 #include <sys/shm.h>
32 #include <sys/socket.h>
33 #include <sys/stat.h>
34 #include <sys/types.h>
35 #include <urcu/list.h>
36 #include <poll.h>
37 #include <unistd.h>
38 #include <sys/mman.h>
39 #include <assert.h>
40 #include <config.h>
41 #include <urcu/compiler.h>
42 #include <ulimit.h>
43
44 #include <common/defaults.h>
45 #include <common/common.h>
46 #include <common/consumer.h>
47 #include <common/consumer-timer.h>
48 #include <common/compat/poll.h>
49 #include <common/sessiond-comm/sessiond-comm.h>
50 #include <common/utils.h>
51
52 #include "lttng-consumerd.h"
53 #include "health-consumerd.h"
54
55 /* TODO : support UST (all direct kernel-ctl accesses). */
56
57 /* threads (channel handling, poll, metadata, sessiond) */
58
59 static pthread_t channel_thread, data_thread, metadata_thread,
60 sessiond_thread, metadata_timer_thread, health_thread;
61
62 /* to count the number of times the user pressed ctrl+c */
63 static int sigintcount = 0;
64
65 /* Argument variables */
66 int lttng_opt_quiet; /* not static in error.h */
67 int lttng_opt_verbose; /* not static in error.h */
68 static int opt_daemon;
69 static const char *progname;
70 static char command_sock_path[PATH_MAX]; /* Global command socket path */
71 static char error_sock_path[PATH_MAX]; /* Global error path */
72 static enum lttng_consumer_type opt_type = LTTNG_CONSUMER_KERNEL;
73
74 /* the liblttngconsumerd context */
75 static struct lttng_consumer_local_data *ctx;
76
77 /* Consumerd health monitoring */
78 struct health_app *health_consumerd;
79
80 const char *tracing_group_name = DEFAULT_TRACING_GROUP;
81
82 enum lttng_consumer_type lttng_consumer_get_type(void)
83 {
84 if (!ctx) {
85 return LTTNG_CONSUMER_UNKNOWN;
86 }
87 return ctx->type;
88 }
89
90 /*
91 * Signal handler for the daemon
92 */
93 static void sighandler(int sig)
94 {
95 if (sig == SIGINT && sigintcount++ == 0) {
96 DBG("ignoring first SIGINT");
97 return;
98 }
99
100 /*
101 * Ignore SIGPIPE because it should not stop the consumer whenever a
102 * SIGPIPE is catched through a FD operation.
103 */
104 if (sig == SIGPIPE) {
105 return;
106 }
107
108 lttng_consumer_should_exit(ctx);
109 }
110
111 /*
112 * Setup signal handler for :
113 * SIGINT, SIGTERM, SIGPIPE
114 */
115 static int set_signal_handler(void)
116 {
117 int ret = 0;
118 struct sigaction sa;
119 sigset_t sigset;
120
121 if ((ret = sigemptyset(&sigset)) < 0) {
122 perror("sigemptyset");
123 return ret;
124 }
125
126 sa.sa_handler = sighandler;
127 sa.sa_mask = sigset;
128 sa.sa_flags = 0;
129 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
130 perror("sigaction");
131 return ret;
132 }
133
134 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
135 perror("sigaction");
136 return ret;
137 }
138
139 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
140 perror("sigaction");
141 return ret;
142 }
143
144 return ret;
145 }
146
147 /*
148 * Usage function on stream file.
149 */
150 static void usage(FILE *fp)
151 {
152 fprintf(fp, "Usage: %s OPTIONS\n\nOptions:\n", progname);
153 fprintf(fp, " -h, --help "
154 "Display this usage.\n");
155 fprintf(fp, " -c, --consumerd-cmd-sock PATH "
156 "Specify path for the command socket\n");
157 fprintf(fp, " -e, --consumerd-err-sock PATH "
158 "Specify path for the error socket\n");
159 fprintf(fp, " -d, --daemonize "
160 "Start as a daemon.\n");
161 fprintf(fp, " -q, --quiet "
162 "No output at all.\n");
163 fprintf(fp, " -v, --verbose "
164 "Verbose mode. Activate DBG() macro.\n");
165 fprintf(fp, " -V, --version "
166 "Show version number.\n");
167 fprintf(fp, " -g, --group NAME "
168 "Specify the tracing group name. (default: tracing)\n");
169 fprintf(fp, " -k, --kernel "
170 "Consumer kernel buffers (default).\n");
171 fprintf(fp, " -u, --ust "
172 "Consumer UST buffers.%s\n",
173 #ifdef HAVE_LIBLTTNG_UST_CTL
174 ""
175 #else
176 " (support not compiled in)"
177 #endif
178 );
179 }
180
181 /*
182 * daemon argument parsing
183 */
184 static void parse_args(int argc, char **argv)
185 {
186 int c;
187
188 static struct option long_options[] = {
189 { "consumerd-cmd-sock", 1, 0, 'c' },
190 { "consumerd-err-sock", 1, 0, 'e' },
191 { "daemonize", 0, 0, 'd' },
192 { "group", 1, 0, 'g' },
193 { "help", 0, 0, 'h' },
194 { "quiet", 0, 0, 'q' },
195 { "verbose", 0, 0, 'v' },
196 { "version", 0, 0, 'V' },
197 { "kernel", 0, 0, 'k' },
198 #ifdef HAVE_LIBLTTNG_UST_CTL
199 { "ust", 0, 0, 'u' },
200 #endif
201 { NULL, 0, 0, 0 }
202 };
203
204 while (1) {
205 int option_index = 0;
206 c = getopt_long(argc, argv, "dhqvVku" "c:e:g:", long_options, &option_index);
207 if (c == -1) {
208 break;
209 }
210
211 switch (c) {
212 case 0:
213 fprintf(stderr, "option %s", long_options[option_index].name);
214 if (optarg) {
215 fprintf(stderr, " with arg %s\n", optarg);
216 }
217 break;
218 case 'c':
219 snprintf(command_sock_path, PATH_MAX, "%s", optarg);
220 break;
221 case 'e':
222 snprintf(error_sock_path, PATH_MAX, "%s", optarg);
223 break;
224 case 'd':
225 opt_daemon = 1;
226 break;
227 case 'g':
228 tracing_group_name = optarg;
229 break;
230 case 'h':
231 usage(stdout);
232 exit(EXIT_SUCCESS);
233 case 'q':
234 lttng_opt_quiet = 1;
235 break;
236 case 'v':
237 lttng_opt_verbose = 1;
238 break;
239 case 'V':
240 fprintf(stdout, "%s\n", VERSION);
241 exit(EXIT_SUCCESS);
242 case 'k':
243 opt_type = LTTNG_CONSUMER_KERNEL;
244 break;
245 #ifdef HAVE_LIBLTTNG_UST_CTL
246 case 'u':
247 # if (CAA_BITS_PER_LONG == 64)
248 opt_type = LTTNG_CONSUMER64_UST;
249 # elif (CAA_BITS_PER_LONG == 32)
250 opt_type = LTTNG_CONSUMER32_UST;
251 # else
252 # error "Unknown bitness"
253 # endif
254 break;
255 #endif
256 default:
257 usage(stderr);
258 exit(EXIT_FAILURE);
259 }
260 }
261 }
262
263 /*
264 * Set open files limit to unlimited. This daemon can open a large number of
265 * file descriptors in order to consumer multiple kernel traces.
266 */
267 static void set_ulimit(void)
268 {
269 int ret;
270 struct rlimit lim;
271
272 /* The kernel does not allowed an infinite limit for open files */
273 lim.rlim_cur = 65535;
274 lim.rlim_max = 65535;
275
276 ret = setrlimit(RLIMIT_NOFILE, &lim);
277 if (ret < 0) {
278 PERROR("failed to set open files limit");
279 }
280 }
281
282 /*
283 * main
284 */
285 int main(int argc, char **argv)
286 {
287 int ret = 0;
288 void *status;
289
290 /* Parse arguments */
291 progname = argv[0];
292 parse_args(argc, argv);
293
294 /* Daemonize */
295 if (opt_daemon) {
296 int i;
297
298 /*
299 * fork
300 * child: setsid, close FD 0, 1, 2, chdir /
301 * parent: exit (if fork is successful)
302 */
303 ret = daemon(0, 0);
304 if (ret < 0) {
305 PERROR("daemon");
306 goto error;
307 }
308 /*
309 * We are in the child. Make sure all other file
310 * descriptors are closed, in case we are called with
311 * more opened file descriptors than the standard ones.
312 */
313 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
314 (void) close(i);
315 }
316 }
317
318 /* Set up max poll set size */
319 lttng_poll_set_max_size();
320
321 if (*command_sock_path == '\0') {
322 switch (opt_type) {
323 case LTTNG_CONSUMER_KERNEL:
324 snprintf(command_sock_path, PATH_MAX, DEFAULT_KCONSUMERD_CMD_SOCK_PATH,
325 DEFAULT_LTTNG_RUNDIR);
326 break;
327 case LTTNG_CONSUMER64_UST:
328 snprintf(command_sock_path, PATH_MAX,
329 DEFAULT_USTCONSUMERD64_CMD_SOCK_PATH, DEFAULT_LTTNG_RUNDIR);
330 break;
331 case LTTNG_CONSUMER32_UST:
332 snprintf(command_sock_path, PATH_MAX,
333 DEFAULT_USTCONSUMERD32_CMD_SOCK_PATH, DEFAULT_LTTNG_RUNDIR);
334 break;
335 default:
336 WARN("Unknown consumerd type");
337 goto error;
338 }
339 }
340
341 /* Init */
342 lttng_consumer_init();
343
344 if (!getuid()) {
345 /* Set limit for open files */
346 set_ulimit();
347 }
348
349 health_consumerd = health_app_create(NR_HEALTH_CONSUMERD_TYPES);
350 if (!health_consumerd) {
351 goto error;
352 }
353
354 /* create the consumer instance with and assign the callbacks */
355 ctx = lttng_consumer_create(opt_type, lttng_consumer_read_subbuffer,
356 NULL, lttng_consumer_on_recv_stream, NULL);
357 if (ctx == NULL) {
358 goto error;
359 }
360
361 lttng_consumer_set_command_sock_path(ctx, command_sock_path);
362 if (*error_sock_path == '\0') {
363 switch (opt_type) {
364 case LTTNG_CONSUMER_KERNEL:
365 snprintf(error_sock_path, PATH_MAX, DEFAULT_KCONSUMERD_ERR_SOCK_PATH,
366 DEFAULT_LTTNG_RUNDIR);
367 break;
368 case LTTNG_CONSUMER64_UST:
369 snprintf(error_sock_path, PATH_MAX,
370 DEFAULT_USTCONSUMERD64_ERR_SOCK_PATH, DEFAULT_LTTNG_RUNDIR);
371 break;
372 case LTTNG_CONSUMER32_UST:
373 snprintf(error_sock_path, PATH_MAX,
374 DEFAULT_USTCONSUMERD32_ERR_SOCK_PATH, DEFAULT_LTTNG_RUNDIR);
375 break;
376 default:
377 WARN("Unknown consumerd type");
378 goto error;
379 }
380 }
381
382 if (set_signal_handler() < 0) {
383 goto error;
384 }
385
386 /* Connect to the socket created by lttng-sessiond to report errors */
387 DBG("Connecting to error socket %s", error_sock_path);
388 ret = lttcomm_connect_unix_sock(error_sock_path);
389 /* not a fatal error, but all communication with lttng-sessiond will fail */
390 if (ret < 0) {
391 WARN("Cannot connect to error socket (is lttng-sessiond started?)");
392 }
393 lttng_consumer_set_error_sock(ctx, ret);
394
395 /*
396 * Block RT signals used for UST periodical metadata flush and the live
397 * timer in main, and create a dedicated thread to handle these signals.
398 */
399 consumer_signal_init();
400
401 ctx->type = opt_type;
402
403 /* Initialize communication library */
404 lttcomm_init();
405
406 ret = utils_create_pipe(health_quit_pipe);
407 if (ret < 0) {
408 goto error_health_pipe;
409 }
410
411 /* Create thread to manage the client socket */
412 ret = pthread_create(&health_thread, NULL,
413 thread_manage_health, (void *) NULL);
414 if (ret != 0) {
415 PERROR("pthread_create health");
416 goto health_error;
417 }
418
419 /* Create thread to manage channels */
420 ret = pthread_create(&channel_thread, NULL, consumer_thread_channel_poll,
421 (void *) ctx);
422 if (ret != 0) {
423 perror("pthread_create");
424 goto channel_error;
425 }
426
427 /* Create thread to manage the polling/writing of trace metadata */
428 ret = pthread_create(&metadata_thread, NULL, consumer_thread_metadata_poll,
429 (void *) ctx);
430 if (ret != 0) {
431 perror("pthread_create");
432 goto metadata_error;
433 }
434
435 /* Create thread to manage the polling/writing of trace data */
436 ret = pthread_create(&data_thread, NULL, consumer_thread_data_poll,
437 (void *) ctx);
438 if (ret != 0) {
439 perror("pthread_create");
440 goto data_error;
441 }
442
443 /* Create the thread to manage the receive of fd */
444 ret = pthread_create(&sessiond_thread, NULL, consumer_thread_sessiond_poll,
445 (void *) ctx);
446 if (ret != 0) {
447 perror("pthread_create");
448 goto sessiond_error;
449 }
450
451 /*
452 * Create the thread to manage the UST metadata periodic timer and
453 * live timer.
454 */
455 ret = pthread_create(&metadata_timer_thread, NULL,
456 consumer_timer_thread, (void *) ctx);
457 if (ret != 0) {
458 perror("pthread_create");
459 goto metadata_timer_error;
460 }
461
462 ret = pthread_detach(metadata_timer_thread);
463 if (ret) {
464 errno = ret;
465 perror("pthread_detach");
466 }
467
468 metadata_timer_error:
469 ret = pthread_join(sessiond_thread, &status);
470 if (ret != 0) {
471 perror("pthread_join");
472 goto error;
473 }
474
475 sessiond_error:
476 ret = pthread_join(data_thread, &status);
477 if (ret != 0) {
478 perror("pthread_join");
479 goto error;
480 }
481
482 data_error:
483 ret = pthread_join(metadata_thread, &status);
484 if (ret != 0) {
485 perror("pthread_join");
486 goto error;
487 }
488
489 metadata_error:
490 ret = pthread_join(channel_thread, &status);
491 if (ret != 0) {
492 perror("pthread_join");
493 goto error;
494 }
495
496 channel_error:
497 ret = pthread_join(health_thread, &status);
498 if (ret != 0) {
499 PERROR("pthread_join health thread");
500 goto error; /* join error, exit without cleanup */
501 }
502
503 health_error:
504 utils_close_pipe(health_quit_pipe);
505
506 error_health_pipe:
507 if (!ret) {
508 ret = EXIT_SUCCESS;
509 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_EXIT_SUCCESS);
510 goto end;
511 }
512
513 error:
514 ret = EXIT_FAILURE;
515 if (ctx) {
516 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_EXIT_FAILURE);
517 }
518
519 end:
520 lttng_consumer_destroy(ctx);
521 lttng_consumer_cleanup();
522 if (health_consumerd) {
523 health_app_destroy(health_consumerd);
524 }
525
526 return ret;
527 }
This page took 0.040417 seconds and 4 git commands to generate.