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> | |
3fd27398 | 48 | #include <common/daemonize.h> |
b8aa1682 JD |
49 | #include <common/futex.h> |
50 | #include <common/sessiond-comm/sessiond-comm.h> | |
51 | #include <common/sessiond-comm/inet.h> | |
b8aa1682 JD |
52 | #include <common/sessiond-comm/relayd.h> |
53 | #include <common/uri.h> | |
a02de639 | 54 | #include <common/utils.h> |
cd60b05a | 55 | #include <common/config/config.h> |
b8aa1682 | 56 | |
0f907de1 | 57 | #include "cmd.h" |
d3e2ba59 | 58 | #include "ctf-trace.h" |
1c20f0e2 | 59 | #include "index.h" |
0f907de1 | 60 | #include "utils.h" |
b8aa1682 | 61 | #include "lttng-relayd.h" |
d3e2ba59 | 62 | #include "live.h" |
55706a7d | 63 | #include "health-relayd.h" |
9b5e0863 | 64 | #include "testpoint.h" |
b8aa1682 JD |
65 | |
66 | /* command line options */ | |
0f907de1 | 67 | char *opt_output_path; |
b5218ffb | 68 | static int opt_daemon, opt_background; |
3fd27398 MD |
69 | |
70 | /* | |
71 | * We need to wait for listener and live listener threads, as well as | |
72 | * health check thread, before being ready to signal readiness. | |
73 | */ | |
74 | #define NR_LTTNG_RELAY_READY 3 | |
75 | static int lttng_relay_ready = NR_LTTNG_RELAY_READY; | |
76 | static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */ | |
77 | static pid_t child_ppid; /* Internal parent PID use with daemonize. */ | |
78 | ||
095a4ae5 MD |
79 | static struct lttng_uri *control_uri; |
80 | static struct lttng_uri *data_uri; | |
d3e2ba59 | 81 | static struct lttng_uri *live_uri; |
b8aa1682 JD |
82 | |
83 | const char *progname; | |
b8aa1682 | 84 | |
65931c8b | 85 | const char *tracing_group_name = DEFAULT_TRACING_GROUP; |
cd60b05a JG |
86 | static int tracing_group_name_override; |
87 | ||
88 | const char * const config_section_name = "relayd"; | |
65931c8b | 89 | |
b8aa1682 JD |
90 | /* |
91 | * Quit pipe for all threads. This permits a single cancellation point | |
92 | * for all threads when receiving an event on the pipe. | |
93 | */ | |
0b242f62 | 94 | int thread_quit_pipe[2] = { -1, -1 }; |
b8aa1682 JD |
95 | |
96 | /* | |
97 | * This pipe is used to inform the worker thread that a command is queued and | |
98 | * ready to be processed. | |
99 | */ | |
100 | static int relay_cmd_pipe[2] = { -1, -1 }; | |
101 | ||
26c9d55e | 102 | /* Shared between threads */ |
b8aa1682 JD |
103 | static int dispatch_thread_exit; |
104 | ||
105 | static pthread_t listener_thread; | |
106 | static pthread_t dispatcher_thread; | |
107 | static pthread_t worker_thread; | |
65931c8b | 108 | static pthread_t health_thread; |
b8aa1682 | 109 | |
095a4ae5 MD |
110 | static uint64_t last_relay_stream_id; |
111 | static uint64_t last_relay_session_id; | |
b8aa1682 JD |
112 | |
113 | /* | |
114 | * Relay command queue. | |
115 | * | |
116 | * The relay_thread_listener and relay_thread_dispatcher communicate with this | |
117 | * queue. | |
118 | */ | |
119 | static struct relay_cmd_queue relay_cmd_queue; | |
120 | ||
121 | /* buffer allocated at startup, used to store the trace data */ | |
095a4ae5 MD |
122 | static char *data_buffer; |
123 | static unsigned int data_buffer_size; | |
b8aa1682 | 124 | |
1c20f0e2 JD |
125 | /* We need those values for the file/dir creation. */ |
126 | static uid_t relayd_uid; | |
127 | static gid_t relayd_gid; | |
128 | ||
d3e2ba59 JD |
129 | /* Global relay stream hash table. */ |
130 | struct lttng_ht *relay_streams_ht; | |
131 | ||
92c6ca54 DG |
132 | /* Global relay viewer stream hash table. */ |
133 | struct lttng_ht *viewer_streams_ht; | |
134 | ||
0a6518b0 DG |
135 | /* Global hash table that stores relay index object. */ |
136 | struct lttng_ht *indexes_ht; | |
137 | ||
55706a7d | 138 | /* Relayd health monitoring */ |
eea7556c | 139 | struct health_app *health_relayd; |
55706a7d | 140 | |
cd60b05a JG |
141 | static struct option long_options[] = { |
142 | { "control-port", 1, 0, 'C', }, | |
143 | { "data-port", 1, 0, 'D', }, | |
8d5c808e | 144 | { "live-port", 1, 0, 'L', }, |
cd60b05a | 145 | { "daemonize", 0, 0, 'd', }, |
b5218ffb | 146 | { "background", 0, 0, 'b', }, |
cd60b05a JG |
147 | { "group", 1, 0, 'g', }, |
148 | { "help", 0, 0, 'h', }, | |
149 | { "output", 1, 0, 'o', }, | |
150 | { "verbose", 0, 0, 'v', }, | |
151 | { "config", 1, 0, 'f' }, | |
152 | { NULL, 0, 0, 0, }, | |
153 | }; | |
154 | ||
155 | static const char *config_ignore_options[] = { "help", "config" }; | |
156 | ||
b8aa1682 JD |
157 | /* |
158 | * usage function on stderr | |
159 | */ | |
160 | static | |
161 | void usage(void) | |
162 | { | |
163 | fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname); | |
994fa64f DG |
164 | fprintf(stderr, " -h, --help Display this usage.\n"); |
165 | fprintf(stderr, " -d, --daemonize Start as a daemon.\n"); | |
b5218ffb | 166 | fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n"); |
994fa64f DG |
167 | fprintf(stderr, " -C, --control-port URL Control port listening.\n"); |
168 | fprintf(stderr, " -D, --data-port URL Data port listening.\n"); | |
8d5c808e | 169 | fprintf(stderr, " -L, --live-port URL Live view port listening.\n"); |
994fa64f DG |
170 | fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n"); |
171 | fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n"); | |
65931c8b | 172 | fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n"); |
cd60b05a | 173 | fprintf(stderr, " -f --config Load daemon configuration file\n"); |
b8aa1682 JD |
174 | } |
175 | ||
cd60b05a JG |
176 | /* |
177 | * Take an option from the getopt output and set it in the right variable to be | |
178 | * used later. | |
179 | * | |
180 | * Return 0 on success else a negative value. | |
181 | */ | |
b8aa1682 | 182 | static |
cd60b05a | 183 | int set_option(int opt, const char *arg, const char *optname) |
b8aa1682 | 184 | { |
cd60b05a JG |
185 | int ret; |
186 | ||
187 | switch (opt) { | |
188 | case 0: | |
189 | fprintf(stderr, "option %s", optname); | |
190 | if (arg) { | |
191 | fprintf(stderr, " with arg %s\n", arg); | |
192 | } | |
193 | break; | |
194 | case 'C': | |
195 | ret = uri_parse(arg, &control_uri); | |
196 | if (ret < 0) { | |
197 | ERR("Invalid control URI specified"); | |
198 | goto end; | |
199 | } | |
200 | if (control_uri->port == 0) { | |
201 | control_uri->port = DEFAULT_NETWORK_CONTROL_PORT; | |
202 | } | |
203 | break; | |
204 | case 'D': | |
205 | ret = uri_parse(arg, &data_uri); | |
206 | if (ret < 0) { | |
207 | ERR("Invalid data URI specified"); | |
208 | goto end; | |
209 | } | |
210 | if (data_uri->port == 0) { | |
211 | data_uri->port = DEFAULT_NETWORK_DATA_PORT; | |
212 | } | |
213 | break; | |
8d5c808e AM |
214 | case 'L': |
215 | ret = uri_parse(arg, &live_uri); | |
216 | if (ret < 0) { | |
217 | ERR("Invalid live URI specified"); | |
218 | goto end; | |
219 | } | |
220 | if (live_uri->port == 0) { | |
221 | live_uri->port = DEFAULT_NETWORK_VIEWER_PORT; | |
222 | } | |
223 | break; | |
cd60b05a JG |
224 | case 'd': |
225 | opt_daemon = 1; | |
226 | break; | |
b5218ffb MD |
227 | case 'b': |
228 | opt_background = 1; | |
229 | break; | |
cd60b05a JG |
230 | case 'g': |
231 | tracing_group_name = strdup(arg); | |
232 | tracing_group_name_override = 1; | |
233 | break; | |
234 | case 'h': | |
235 | usage(); | |
236 | exit(EXIT_FAILURE); | |
237 | case 'o': | |
238 | ret = asprintf(&opt_output_path, "%s", arg); | |
239 | if (ret < 0) { | |
240 | ret = -errno; | |
241 | PERROR("asprintf opt_output_path"); | |
242 | goto end; | |
243 | } | |
244 | break; | |
245 | case 'v': | |
246 | /* Verbose level can increase using multiple -v */ | |
247 | if (arg) { | |
248 | lttng_opt_verbose = config_parse_value(arg); | |
249 | } else { | |
250 | lttng_opt_verbose += 1; | |
251 | } | |
252 | break; | |
253 | default: | |
254 | /* Unknown option or other error. | |
255 | * Error is printed by getopt, just return */ | |
256 | ret = -1; | |
257 | goto end; | |
258 | } | |
259 | ||
260 | /* All good. */ | |
261 | ret = 0; | |
262 | ||
263 | end: | |
264 | return ret; | |
265 | } | |
266 | ||
267 | /* | |
268 | * config_entry_handler_cb used to handle options read from a config file. | |
269 | * See config_entry_handler_cb comment in common/config/config.h for the | |
270 | * return value conventions. | |
271 | */ | |
272 | static | |
273 | int config_entry_handler(const struct config_entry *entry, void *unused) | |
274 | { | |
275 | int ret = 0, i; | |
276 | ||
277 | if (!entry || !entry->name || !entry->value) { | |
278 | ret = -EINVAL; | |
279 | goto end; | |
280 | } | |
281 | ||
282 | /* Check if the option is to be ignored */ | |
283 | for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) { | |
284 | if (!strcmp(entry->name, config_ignore_options[i])) { | |
285 | goto end; | |
286 | } | |
287 | } | |
288 | ||
289 | for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) { | |
290 | /* Ignore if entry name is not fully matched. */ | |
291 | if (strcmp(entry->name, long_options[i].name)) { | |
292 | continue; | |
293 | } | |
294 | ||
295 | /* | |
296 | * If the option takes no argument on the command line, we have to | |
297 | * check if the value is "true". We support non-zero numeric values, | |
298 | * true, on and yes. | |
299 | */ | |
300 | if (!long_options[i].has_arg) { | |
301 | ret = config_parse_value(entry->value); | |
302 | if (ret <= 0) { | |
303 | if (ret) { | |
304 | WARN("Invalid configuration value \"%s\" for option %s", | |
305 | entry->value, entry->name); | |
306 | } | |
307 | /* False, skip boolean config option. */ | |
308 | goto end; | |
309 | } | |
310 | } | |
311 | ||
312 | ret = set_option(long_options[i].val, entry->value, entry->name); | |
313 | goto end; | |
314 | } | |
315 | ||
316 | WARN("Unrecognized option \"%s\" in daemon configuration file.", | |
317 | entry->name); | |
318 | ||
319 | end: | |
320 | return ret; | |
321 | } | |
322 | ||
323 | static | |
324 | int set_options(int argc, char **argv) | |
325 | { | |
326 | int c, ret = 0, option_index = 0; | |
327 | int orig_optopt = optopt, orig_optind = optind; | |
328 | char *default_address, *optstring; | |
329 | const char *config_path = NULL; | |
330 | ||
331 | optstring = utils_generate_optstring(long_options, | |
332 | sizeof(long_options) / sizeof(struct option)); | |
333 | if (!optstring) { | |
334 | ret = -ENOMEM; | |
335 | goto exit; | |
336 | } | |
337 | ||
338 | /* Check for the --config option */ | |
339 | ||
340 | while ((c = getopt_long(argc, argv, optstring, long_options, | |
341 | &option_index)) != -1) { | |
342 | if (c == '?') { | |
343 | ret = -EINVAL; | |
344 | goto exit; | |
345 | } else if (c != 'f') { | |
346 | continue; | |
347 | } | |
348 | ||
349 | config_path = utils_expand_path(optarg); | |
350 | if (!config_path) { | |
351 | ERR("Failed to resolve path: %s", optarg); | |
352 | } | |
353 | } | |
354 | ||
355 | ret = config_get_section_entries(config_path, config_section_name, | |
356 | config_entry_handler, NULL); | |
357 | if (ret) { | |
358 | if (ret > 0) { | |
359 | ERR("Invalid configuration option at line %i", ret); | |
360 | ret = -1; | |
361 | } | |
362 | goto exit; | |
363 | } | |
b8aa1682 | 364 | |
cd60b05a JG |
365 | /* Reset getopt's global state */ |
366 | optopt = orig_optopt; | |
367 | optind = orig_optind; | |
b8aa1682 | 368 | while (1) { |
cd60b05a | 369 | c = getopt_long(argc, argv, optstring, long_options, &option_index); |
b8aa1682 JD |
370 | if (c == -1) { |
371 | break; | |
372 | } | |
373 | ||
cd60b05a JG |
374 | ret = set_option(c, optarg, long_options[option_index].name); |
375 | if (ret < 0) { | |
b8aa1682 JD |
376 | goto exit; |
377 | } | |
378 | } | |
379 | ||
380 | /* assign default values */ | |
381 | if (control_uri == NULL) { | |
382 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
383 | DEFAULT_NETWORK_CONTROL_PORT); | |
384 | if (ret < 0) { | |
385 | PERROR("asprintf default data address"); | |
386 | goto exit; | |
387 | } | |
388 | ||
389 | ret = uri_parse(default_address, &control_uri); | |
390 | free(default_address); | |
391 | if (ret < 0) { | |
392 | ERR("Invalid control URI specified"); | |
393 | goto exit; | |
394 | } | |
395 | } | |
396 | if (data_uri == NULL) { | |
397 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
398 | DEFAULT_NETWORK_DATA_PORT); | |
399 | if (ret < 0) { | |
400 | PERROR("asprintf default data address"); | |
401 | goto exit; | |
402 | } | |
403 | ||
404 | ret = uri_parse(default_address, &data_uri); | |
405 | free(default_address); | |
406 | if (ret < 0) { | |
407 | ERR("Invalid data URI specified"); | |
408 | goto exit; | |
409 | } | |
410 | } | |
d3e2ba59 JD |
411 | if (live_uri == NULL) { |
412 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
413 | DEFAULT_NETWORK_VIEWER_PORT); | |
414 | if (ret < 0) { | |
415 | PERROR("asprintf default viewer control address"); | |
416 | goto exit; | |
417 | } | |
418 | ||
419 | ret = uri_parse(default_address, &live_uri); | |
420 | free(default_address); | |
421 | if (ret < 0) { | |
422 | ERR("Invalid viewer control URI specified"); | |
423 | goto exit; | |
424 | } | |
425 | } | |
b8aa1682 JD |
426 | |
427 | exit: | |
cd60b05a | 428 | free(optstring); |
b8aa1682 JD |
429 | return ret; |
430 | } | |
431 | ||
432 | /* | |
433 | * Cleanup the daemon | |
434 | */ | |
435 | static | |
436 | void cleanup(void) | |
437 | { | |
b8aa1682 JD |
438 | DBG("Cleaning up"); |
439 | ||
095a4ae5 MD |
440 | /* free the dynamically allocated opt_output_path */ |
441 | free(opt_output_path); | |
442 | ||
a02de639 CB |
443 | /* Close thread quit pipes */ |
444 | utils_close_pipe(thread_quit_pipe); | |
445 | ||
710c1f73 DG |
446 | uri_free(control_uri); |
447 | uri_free(data_uri); | |
8d5c808e | 448 | /* Live URI is freed in the live thread. */ |
cd60b05a JG |
449 | |
450 | if (tracing_group_name_override) { | |
451 | free((void *) tracing_group_name); | |
452 | } | |
b8aa1682 JD |
453 | } |
454 | ||
455 | /* | |
456 | * Write to writable pipe used to notify a thread. | |
457 | */ | |
458 | static | |
459 | int notify_thread_pipe(int wpipe) | |
460 | { | |
6cd525e8 | 461 | ssize_t ret; |
b8aa1682 | 462 | |
6cd525e8 MD |
463 | ret = lttng_write(wpipe, "!", 1); |
464 | if (ret < 1) { | |
b8aa1682 JD |
465 | PERROR("write poll pipe"); |
466 | } | |
467 | ||
468 | return ret; | |
469 | } | |
470 | ||
65931c8b MD |
471 | static void notify_health_quit_pipe(int *pipe) |
472 | { | |
6cd525e8 | 473 | ssize_t ret; |
65931c8b | 474 | |
6cd525e8 MD |
475 | ret = lttng_write(pipe[1], "4", 1); |
476 | if (ret < 1) { | |
65931c8b MD |
477 | PERROR("write relay health quit"); |
478 | } | |
479 | } | |
480 | ||
b8aa1682 JD |
481 | /* |
482 | * Stop all threads by closing the thread quit pipe. | |
483 | */ | |
484 | static | |
485 | void stop_threads(void) | |
486 | { | |
487 | int ret; | |
488 | ||
489 | /* Stopping all threads */ | |
490 | DBG("Terminating all threads"); | |
491 | ret = notify_thread_pipe(thread_quit_pipe[1]); | |
492 | if (ret < 0) { | |
493 | ERR("write error on thread quit pipe"); | |
494 | } | |
495 | ||
65931c8b MD |
496 | notify_health_quit_pipe(health_quit_pipe); |
497 | ||
b8aa1682 | 498 | /* Dispatch thread */ |
26c9d55e | 499 | CMM_STORE_SHARED(dispatch_thread_exit, 1); |
b8aa1682 JD |
500 | futex_nto1_wake(&relay_cmd_queue.futex); |
501 | } | |
502 | ||
503 | /* | |
504 | * Signal handler for the daemon | |
505 | * | |
506 | * Simply stop all worker threads, leaving main() return gracefully after | |
507 | * joining all threads and calling cleanup(). | |
508 | */ | |
509 | static | |
510 | void sighandler(int sig) | |
511 | { | |
512 | switch (sig) { | |
513 | case SIGPIPE: | |
514 | DBG("SIGPIPE caught"); | |
515 | return; | |
516 | case SIGINT: | |
517 | DBG("SIGINT caught"); | |
518 | stop_threads(); | |
519 | break; | |
520 | case SIGTERM: | |
521 | DBG("SIGTERM caught"); | |
522 | stop_threads(); | |
523 | break; | |
3fd27398 MD |
524 | case SIGUSR1: |
525 | CMM_STORE_SHARED(recv_child_signal, 1); | |
526 | break; | |
b8aa1682 JD |
527 | default: |
528 | break; | |
529 | } | |
530 | } | |
531 | ||
532 | /* | |
533 | * Setup signal handler for : | |
534 | * SIGINT, SIGTERM, SIGPIPE | |
535 | */ | |
536 | static | |
537 | int set_signal_handler(void) | |
538 | { | |
539 | int ret = 0; | |
540 | struct sigaction sa; | |
541 | sigset_t sigset; | |
542 | ||
543 | if ((ret = sigemptyset(&sigset)) < 0) { | |
544 | PERROR("sigemptyset"); | |
545 | return ret; | |
546 | } | |
547 | ||
548 | sa.sa_handler = sighandler; | |
549 | sa.sa_mask = sigset; | |
550 | sa.sa_flags = 0; | |
551 | if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) { | |
552 | PERROR("sigaction"); | |
553 | return ret; | |
554 | } | |
555 | ||
556 | if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) { | |
557 | PERROR("sigaction"); | |
558 | return ret; | |
559 | } | |
560 | ||
561 | if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) { | |
562 | PERROR("sigaction"); | |
563 | return ret; | |
564 | } | |
565 | ||
3fd27398 MD |
566 | if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) { |
567 | PERROR("sigaction"); | |
568 | return ret; | |
569 | } | |
570 | ||
571 | DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT"); | |
b8aa1682 JD |
572 | |
573 | return ret; | |
574 | } | |
575 | ||
3fd27398 MD |
576 | void lttng_relay_notify_ready(void) |
577 | { | |
578 | /* Notify the parent of the fork() process that we are ready. */ | |
579 | if (opt_daemon || opt_background) { | |
580 | if (uatomic_sub_return(<tng_relay_ready, 1) == 0) { | |
581 | kill(child_ppid, SIGUSR1); | |
582 | } | |
583 | } | |
584 | } | |
585 | ||
b8aa1682 JD |
586 | /* |
587 | * Init thread quit pipe. | |
588 | * | |
589 | * Return -1 on error or 0 if all pipes are created. | |
590 | */ | |
591 | static | |
592 | int init_thread_quit_pipe(void) | |
593 | { | |
a02de639 | 594 | int ret; |
b8aa1682 | 595 | |
a02de639 | 596 | ret = utils_create_pipe_cloexec(thread_quit_pipe); |
b8aa1682 | 597 | |
b8aa1682 JD |
598 | return ret; |
599 | } | |
600 | ||
601 | /* | |
602 | * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set. | |
603 | */ | |
604 | static | |
605 | int create_thread_poll_set(struct lttng_poll_event *events, int size) | |
606 | { | |
607 | int ret; | |
608 | ||
609 | if (events == NULL || size == 0) { | |
610 | ret = -1; | |
611 | goto error; | |
612 | } | |
613 | ||
614 | ret = lttng_poll_create(events, size, LTTNG_CLOEXEC); | |
615 | if (ret < 0) { | |
616 | goto error; | |
617 | } | |
618 | ||
619 | /* Add quit pipe */ | |
c7759e6a | 620 | ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR); |
b8aa1682 JD |
621 | if (ret < 0) { |
622 | goto error; | |
623 | } | |
624 | ||
625 | return 0; | |
626 | ||
627 | error: | |
628 | return ret; | |
629 | } | |
630 | ||
631 | /* | |
632 | * Check if the thread quit pipe was triggered. | |
633 | * | |
634 | * Return 1 if it was triggered else 0; | |
635 | */ | |
636 | static | |
637 | int check_thread_quit_pipe(int fd, uint32_t events) | |
638 | { | |
639 | if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) { | |
640 | return 1; | |
641 | } | |
642 | ||
643 | return 0; | |
644 | } | |
645 | ||
646 | /* | |
647 | * Create and init socket from uri. | |
648 | */ | |
649 | static | |
650 | struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri) | |
651 | { | |
652 | int ret; | |
653 | struct lttcomm_sock *sock = NULL; | |
654 | ||
655 | sock = lttcomm_alloc_sock_from_uri(uri); | |
656 | if (sock == NULL) { | |
657 | ERR("Allocating socket"); | |
658 | goto error; | |
659 | } | |
660 | ||
661 | ret = lttcomm_create_sock(sock); | |
662 | if (ret < 0) { | |
663 | goto error; | |
664 | } | |
665 | DBG("Listening on sock %d", sock->fd); | |
666 | ||
667 | ret = sock->ops->bind(sock); | |
668 | if (ret < 0) { | |
669 | goto error; | |
670 | } | |
671 | ||
672 | ret = sock->ops->listen(sock, -1); | |
673 | if (ret < 0) { | |
674 | goto error; | |
675 | ||
676 | } | |
677 | ||
678 | return sock; | |
679 | ||
680 | error: | |
681 | if (sock) { | |
682 | lttcomm_destroy_sock(sock); | |
683 | } | |
684 | return NULL; | |
685 | } | |
686 | ||
173af62f DG |
687 | /* |
688 | * Return nonzero if stream needs to be closed. | |
689 | */ | |
690 | static | |
691 | int close_stream_check(struct relay_stream *stream) | |
692 | { | |
693 | ||
694 | if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) { | |
f7079f67 DG |
695 | /* |
696 | * We are about to close the stream so set the data pending flag to 1 | |
697 | * which will make the end data pending command skip the stream which | |
698 | * is now closed and ready. Note that after proceeding to a file close, | |
699 | * the written file is ready for reading. | |
700 | */ | |
701 | stream->data_pending_check_done = 1; | |
173af62f DG |
702 | return 1; |
703 | } | |
704 | return 0; | |
705 | } | |
706 | ||
b8aa1682 JD |
707 | /* |
708 | * This thread manages the listening for new connections on the network | |
709 | */ | |
710 | static | |
711 | void *relay_thread_listener(void *data) | |
712 | { | |
095a4ae5 | 713 | int i, ret, pollfd, err = -1; |
b8aa1682 JD |
714 | int val = 1; |
715 | uint32_t revents, nb_fd; | |
716 | struct lttng_poll_event events; | |
717 | struct lttcomm_sock *control_sock, *data_sock; | |
718 | ||
b8aa1682 JD |
719 | DBG("[thread] Relay listener started"); |
720 | ||
55706a7d MD |
721 | health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER); |
722 | ||
f385ae0a MD |
723 | health_code_update(); |
724 | ||
b8aa1682 JD |
725 | control_sock = relay_init_sock(control_uri); |
726 | if (!control_sock) { | |
095a4ae5 | 727 | goto error_sock_control; |
b8aa1682 JD |
728 | } |
729 | ||
730 | data_sock = relay_init_sock(data_uri); | |
731 | if (!data_sock) { | |
095a4ae5 | 732 | goto error_sock_relay; |
b8aa1682 JD |
733 | } |
734 | ||
735 | /* | |
736 | * Pass 3 as size here for the thread quit pipe, control and data socket. | |
737 | */ | |
738 | ret = create_thread_poll_set(&events, 3); | |
739 | if (ret < 0) { | |
740 | goto error_create_poll; | |
741 | } | |
742 | ||
743 | /* Add the control socket */ | |
744 | ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP); | |
745 | if (ret < 0) { | |
746 | goto error_poll_add; | |
747 | } | |
748 | ||
749 | /* Add the data socket */ | |
750 | ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP); | |
751 | if (ret < 0) { | |
752 | goto error_poll_add; | |
753 | } | |
754 | ||
3fd27398 MD |
755 | lttng_relay_notify_ready(); |
756 | ||
9b5e0863 MD |
757 | if (testpoint(relayd_thread_listener)) { |
758 | goto error_testpoint; | |
759 | } | |
760 | ||
b8aa1682 | 761 | while (1) { |
f385ae0a MD |
762 | health_code_update(); |
763 | ||
b8aa1682 JD |
764 | DBG("Listener accepting connections"); |
765 | ||
b8aa1682 | 766 | restart: |
f385ae0a | 767 | health_poll_entry(); |
b8aa1682 | 768 | ret = lttng_poll_wait(&events, -1); |
f385ae0a | 769 | health_poll_exit(); |
b8aa1682 JD |
770 | if (ret < 0) { |
771 | /* | |
772 | * Restart interrupted system call. | |
773 | */ | |
774 | if (errno == EINTR) { | |
775 | goto restart; | |
776 | } | |
777 | goto error; | |
778 | } | |
779 | ||
0d9c5d77 DG |
780 | nb_fd = ret; |
781 | ||
b8aa1682 JD |
782 | DBG("Relay new connection received"); |
783 | for (i = 0; i < nb_fd; i++) { | |
f385ae0a MD |
784 | health_code_update(); |
785 | ||
b8aa1682 JD |
786 | /* Fetch once the poll data */ |
787 | revents = LTTNG_POLL_GETEV(&events, i); | |
788 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
789 | ||
790 | /* Thread quit pipe has been closed. Killing thread. */ | |
791 | ret = check_thread_quit_pipe(pollfd, revents); | |
792 | if (ret) { | |
095a4ae5 MD |
793 | err = 0; |
794 | goto exit; | |
b8aa1682 JD |
795 | } |
796 | ||
797 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
798 | ERR("socket poll error"); | |
799 | goto error; | |
800 | } else if (revents & LPOLLIN) { | |
4b7f17b2 MD |
801 | /* |
802 | * Get allocated in this thread, | |
803 | * enqueued to a global queue, dequeued | |
804 | * and freed in the worker thread. | |
805 | */ | |
806 | struct relay_command *relay_cmd; | |
807 | struct lttcomm_sock *newsock; | |
b8aa1682 JD |
808 | |
809 | relay_cmd = zmalloc(sizeof(struct relay_command)); | |
810 | if (relay_cmd == NULL) { | |
811 | PERROR("relay command zmalloc"); | |
812 | goto error; | |
813 | } | |
814 | ||
815 | if (pollfd == data_sock->fd) { | |
816 | newsock = data_sock->ops->accept(data_sock); | |
4b7f17b2 | 817 | if (!newsock) { |
b8aa1682 | 818 | PERROR("accepting data sock"); |
4b7f17b2 | 819 | free(relay_cmd); |
b8aa1682 JD |
820 | goto error; |
821 | } | |
822 | relay_cmd->type = RELAY_DATA; | |
823 | DBG("Relay data connection accepted, socket %d", newsock->fd); | |
4b7f17b2 MD |
824 | } else { |
825 | assert(pollfd == control_sock->fd); | |
b8aa1682 | 826 | newsock = control_sock->ops->accept(control_sock); |
4b7f17b2 | 827 | if (!newsock) { |
b8aa1682 | 828 | PERROR("accepting control sock"); |
4b7f17b2 | 829 | free(relay_cmd); |
b8aa1682 JD |
830 | goto error; |
831 | } | |
832 | relay_cmd->type = RELAY_CONTROL; | |
833 | DBG("Relay control connection accepted, socket %d", newsock->fd); | |
834 | } | |
835 | ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, | |
836 | &val, sizeof(int)); | |
837 | if (ret < 0) { | |
838 | PERROR("setsockopt inet"); | |
4b7f17b2 MD |
839 | lttcomm_destroy_sock(newsock); |
840 | free(relay_cmd); | |
b8aa1682 JD |
841 | goto error; |
842 | } | |
843 | relay_cmd->sock = newsock; | |
844 | /* | |
845 | * Lock free enqueue the request. | |
846 | */ | |
847 | cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node); | |
848 | ||
849 | /* | |
850 | * Wake the dispatch queue futex. Implicit memory | |
851 | * barrier with the exchange in cds_wfq_enqueue. | |
852 | */ | |
853 | futex_nto1_wake(&relay_cmd_queue.futex); | |
854 | } | |
855 | } | |
856 | } | |
857 | ||
095a4ae5 | 858 | exit: |
b8aa1682 JD |
859 | error: |
860 | error_poll_add: | |
9b5e0863 | 861 | error_testpoint: |
b8aa1682 JD |
862 | lttng_poll_clean(&events); |
863 | error_create_poll: | |
095a4ae5 MD |
864 | if (data_sock->fd >= 0) { |
865 | ret = data_sock->ops->close(data_sock); | |
b8aa1682 JD |
866 | if (ret) { |
867 | PERROR("close"); | |
868 | } | |
b8aa1682 | 869 | } |
095a4ae5 MD |
870 | lttcomm_destroy_sock(data_sock); |
871 | error_sock_relay: | |
872 | if (control_sock->fd >= 0) { | |
873 | ret = control_sock->ops->close(control_sock); | |
b8aa1682 JD |
874 | if (ret) { |
875 | PERROR("close"); | |
876 | } | |
b8aa1682 | 877 | } |
095a4ae5 MD |
878 | lttcomm_destroy_sock(control_sock); |
879 | error_sock_control: | |
880 | if (err) { | |
f385ae0a MD |
881 | health_error(); |
882 | ERR("Health error occurred in %s", __func__); | |
095a4ae5 | 883 | } |
55706a7d | 884 | health_unregister(health_relayd); |
b8aa1682 JD |
885 | DBG("Relay listener thread cleanup complete"); |
886 | stop_threads(); | |
b8aa1682 JD |
887 | return NULL; |
888 | } | |
889 | ||
890 | /* | |
891 | * This thread manages the dispatching of the requests to worker threads | |
892 | */ | |
893 | static | |
894 | void *relay_thread_dispatcher(void *data) | |
895 | { | |
6cd525e8 MD |
896 | int err = -1; |
897 | ssize_t ret; | |
b8aa1682 JD |
898 | struct cds_wfq_node *node; |
899 | struct relay_command *relay_cmd = NULL; | |
900 | ||
901 | DBG("[thread] Relay dispatcher started"); | |
902 | ||
55706a7d MD |
903 | health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER); |
904 | ||
9b5e0863 MD |
905 | if (testpoint(relayd_thread_dispatcher)) { |
906 | goto error_testpoint; | |
907 | } | |
908 | ||
f385ae0a MD |
909 | health_code_update(); |
910 | ||
26c9d55e | 911 | while (!CMM_LOAD_SHARED(dispatch_thread_exit)) { |
f385ae0a MD |
912 | health_code_update(); |
913 | ||
b8aa1682 JD |
914 | /* Atomically prepare the queue futex */ |
915 | futex_nto1_prepare(&relay_cmd_queue.futex); | |
916 | ||
917 | do { | |
f385ae0a MD |
918 | health_code_update(); |
919 | ||
b8aa1682 JD |
920 | /* Dequeue commands */ |
921 | node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue); | |
922 | if (node == NULL) { | |
923 | DBG("Woken up but nothing in the relay command queue"); | |
924 | /* Continue thread execution */ | |
925 | break; | |
926 | } | |
927 | ||
928 | relay_cmd = caa_container_of(node, struct relay_command, node); | |
929 | DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd); | |
930 | ||
931 | /* | |
932 | * Inform worker thread of the new request. This | |
933 | * call is blocking so we can be assured that the data will be read | |
934 | * at some point in time or wait to the end of the world :) | |
935 | */ | |
6cd525e8 MD |
936 | ret = lttng_write(relay_cmd_pipe[1], relay_cmd, |
937 | sizeof(struct relay_command)); | |
b8aa1682 | 938 | free(relay_cmd); |
6cd525e8 | 939 | if (ret < sizeof(struct relay_command)) { |
b8aa1682 JD |
940 | PERROR("write cmd pipe"); |
941 | goto error; | |
942 | } | |
943 | } while (node != NULL); | |
944 | ||
945 | /* Futex wait on queue. Blocking call on futex() */ | |
f385ae0a | 946 | health_poll_entry(); |
b8aa1682 | 947 | futex_nto1_wait(&relay_cmd_queue.futex); |
f385ae0a | 948 | health_poll_exit(); |
b8aa1682 JD |
949 | } |
950 | ||
f385ae0a MD |
951 | /* Normal exit, no error */ |
952 | err = 0; | |
953 | ||
b8aa1682 | 954 | error: |
9b5e0863 | 955 | error_testpoint: |
f385ae0a MD |
956 | if (err) { |
957 | health_error(); | |
958 | ERR("Health error occurred in %s", __func__); | |
959 | } | |
55706a7d | 960 | health_unregister(health_relayd); |
b8aa1682 JD |
961 | DBG("Dispatch thread dying"); |
962 | stop_threads(); | |
963 | return NULL; | |
964 | } | |
965 | ||
de91f48a DG |
966 | /* |
967 | * Get stream from stream id. | |
968 | * Need to be called with RCU read-side lock held. | |
969 | */ | |
d3e2ba59 | 970 | struct relay_stream *relay_stream_find_by_id(uint64_t stream_id) |
de91f48a DG |
971 | { |
972 | struct lttng_ht_node_ulong *node; | |
973 | struct lttng_ht_iter iter; | |
974 | struct relay_stream *ret; | |
975 | ||
d3e2ba59 | 976 | lttng_ht_lookup(relay_streams_ht, |
de91f48a DG |
977 | (void *)((unsigned long) stream_id), |
978 | &iter); | |
979 | node = lttng_ht_iter_get_node_ulong(&iter); | |
980 | if (node == NULL) { | |
981 | DBG("Relay stream %" PRIu64 " not found", stream_id); | |
982 | ret = NULL; | |
983 | goto end; | |
984 | } | |
985 | ||
986 | ret = caa_container_of(node, struct relay_stream, stream_n); | |
987 | ||
988 | end: | |
989 | return ret; | |
990 | } | |
991 | ||
9d1bbf21 MD |
992 | static |
993 | void deferred_free_stream(struct rcu_head *head) | |
994 | { | |
995 | struct relay_stream *stream = | |
996 | caa_container_of(head, struct relay_stream, rcu_node); | |
d3e2ba59 | 997 | |
0f907de1 JD |
998 | free(stream->path_name); |
999 | free(stream->channel_name); | |
9d1bbf21 MD |
1000 | free(stream); |
1001 | } | |
1002 | ||
d3e2ba59 JD |
1003 | static |
1004 | void deferred_free_session(struct rcu_head *head) | |
1005 | { | |
1006 | struct relay_session *session = | |
1007 | caa_container_of(head, struct relay_session, rcu_node); | |
1008 | free(session); | |
1009 | } | |
1010 | ||
4c80d9a2 DG |
1011 | /* |
1012 | * Close a given stream. The stream is freed using a call RCU. | |
1013 | * | |
1014 | * RCU read side lock MUST be acquired. If NO close_stream_check() was called | |
1015 | * BEFORE the stream lock MUST be acquired. | |
1016 | */ | |
c07d8a78 | 1017 | static void destroy_stream(struct relay_stream *stream) |
94d49140 JD |
1018 | { |
1019 | int delret; | |
1020 | struct relay_viewer_stream *vstream; | |
1021 | struct lttng_ht_iter iter; | |
1022 | ||
1023 | assert(stream); | |
94d49140 JD |
1024 | |
1025 | delret = close(stream->fd); | |
1026 | if (delret < 0) { | |
1027 | PERROR("close stream"); | |
1028 | } | |
1029 | ||
1030 | if (stream->index_fd >= 0) { | |
1031 | delret = close(stream->index_fd); | |
1032 | if (delret < 0) { | |
1033 | PERROR("close stream index_fd"); | |
1034 | } | |
1035 | } | |
1036 | ||
92c6ca54 | 1037 | vstream = live_find_viewer_stream_by_id(stream->stream_handle); |
94d49140 JD |
1038 | if (vstream) { |
1039 | /* | |
1040 | * Set the last good value into the viewer stream. This is done | |
1041 | * right before the stream gets deleted from the hash table. The | |
1042 | * lookup failure on the live thread side of a stream indicates | |
1043 | * that the viewer stream index received value should be used. | |
1044 | */ | |
cef0f7d5 | 1045 | pthread_mutex_lock(&stream->viewer_stream_rotation_lock); |
94d49140 | 1046 | vstream->total_index_received = stream->total_index_received; |
a020f610 | 1047 | vstream->tracefile_count_last = stream->tracefile_count_current; |
cef0f7d5 JD |
1048 | vstream->close_write_flag = 1; |
1049 | pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); | |
94d49140 JD |
1050 | } |
1051 | ||
efc8a87f DG |
1052 | /* Cleanup index of that stream. */ |
1053 | relay_index_destroy_by_stream_id(stream->stream_handle); | |
1054 | ||
94d49140 JD |
1055 | iter.iter.node = &stream->stream_n.node; |
1056 | delret = lttng_ht_del(relay_streams_ht, &iter); | |
1057 | assert(!delret); | |
1058 | iter.iter.node = &stream->ctf_trace_node.node; | |
c07d8a78 | 1059 | delret = lttng_ht_del(stream->ctf_traces_ht, &iter); |
94d49140 | 1060 | assert(!delret); |
157df586 JD |
1061 | |
1062 | if (stream->ctf_trace) { | |
1063 | ctf_trace_try_destroy(stream->ctf_trace); | |
1064 | } | |
1065 | ||
94d49140 JD |
1066 | call_rcu(&stream->rcu_node, deferred_free_stream); |
1067 | DBG("Closed tracefile %d from close stream", stream->fd); | |
1068 | } | |
1069 | ||
b8aa1682 JD |
1070 | /* |
1071 | * relay_delete_session: Free all memory associated with a session and | |
1072 | * close all the FDs | |
1073 | */ | |
1074 | static | |
d3e2ba59 JD |
1075 | void relay_delete_session(struct relay_command *cmd, |
1076 | struct lttng_ht *sessions_ht) | |
b8aa1682 JD |
1077 | { |
1078 | struct lttng_ht_iter iter; | |
1079 | struct lttng_ht_node_ulong *node; | |
1080 | struct relay_stream *stream; | |
1081 | int ret; | |
1082 | ||
095a4ae5 | 1083 | if (!cmd->session) { |
b8aa1682 | 1084 | return; |
095a4ae5 | 1085 | } |
b8aa1682 | 1086 | |
77c7c900 | 1087 | DBG("Relay deleting session %" PRIu64, cmd->session->id); |
5b6d8097 | 1088 | |
9d1bbf21 | 1089 | rcu_read_lock(); |
d3e2ba59 | 1090 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) { |
b8aa1682 | 1091 | node = lttng_ht_iter_get_node_ulong(&iter); |
4c80d9a2 DG |
1092 | if (!node) { |
1093 | continue; | |
b8aa1682 | 1094 | } |
4c80d9a2 DG |
1095 | stream = caa_container_of(node, struct relay_stream, stream_n); |
1096 | if (stream->session == cmd->session) { | |
c07d8a78 | 1097 | destroy_stream(stream); |
87b576ec JD |
1098 | cmd->session->stream_count--; |
1099 | assert(cmd->session->stream_count >= 0); | |
4c80d9a2 | 1100 | } |
b8aa1682 | 1101 | } |
4c80d9a2 DG |
1102 | |
1103 | /* Make this session not visible anymore. */ | |
d3e2ba59 JD |
1104 | iter.iter.node = &cmd->session->session_n.node; |
1105 | ret = lttng_ht_del(sessions_ht, &iter); | |
1106 | assert(!ret); | |
4c80d9a2 | 1107 | call_rcu(&cmd->session->rcu_node, deferred_free_session); |
9d1bbf21 | 1108 | rcu_read_unlock(); |
b8aa1682 JD |
1109 | } |
1110 | ||
1c20f0e2 JD |
1111 | /* |
1112 | * Copy index data from the control port to a given index object. | |
1113 | */ | |
1114 | static void copy_index_control_data(struct relay_index *index, | |
1115 | struct lttcomm_relayd_index *data) | |
1116 | { | |
1117 | assert(index); | |
1118 | assert(data); | |
1119 | ||
1120 | /* | |
1121 | * The index on disk is encoded in big endian, so we don't need to convert | |
1122 | * the data received on the network. The data_offset value is NEVER | |
1123 | * modified here and is updated by the data thread. | |
1124 | */ | |
1125 | index->index_data.packet_size = data->packet_size; | |
1126 | index->index_data.content_size = data->content_size; | |
1127 | index->index_data.timestamp_begin = data->timestamp_begin; | |
1128 | index->index_data.timestamp_end = data->timestamp_end; | |
1129 | index->index_data.events_discarded = data->events_discarded; | |
1130 | index->index_data.stream_id = data->stream_id; | |
1131 | } | |
1132 | ||
c5b6f4f0 DG |
1133 | /* |
1134 | * Handle the RELAYD_CREATE_SESSION command. | |
1135 | * | |
1136 | * On success, send back the session id or else return a negative value. | |
1137 | */ | |
1138 | static | |
1139 | int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 JD |
1140 | struct relay_command *cmd, |
1141 | struct lttng_ht *sessions_ht) | |
c5b6f4f0 DG |
1142 | { |
1143 | int ret = 0, send_ret; | |
1144 | struct relay_session *session; | |
1145 | struct lttcomm_relayd_status_session reply; | |
1146 | ||
1147 | assert(recv_hdr); | |
1148 | assert(cmd); | |
1149 | ||
1150 | memset(&reply, 0, sizeof(reply)); | |
1151 | ||
1152 | session = zmalloc(sizeof(struct relay_session)); | |
1153 | if (session == NULL) { | |
1154 | PERROR("relay session zmalloc"); | |
1155 | ret = -1; | |
1156 | goto error; | |
1157 | } | |
1158 | ||
1159 | session->id = ++last_relay_session_id; | |
1160 | session->sock = cmd->sock; | |
7d2f7452 DG |
1161 | session->minor = cmd->minor; |
1162 | session->major = cmd->major; | |
80e8027a | 1163 | pthread_mutex_init(&session->viewer_ready_lock, NULL); |
c5b6f4f0 DG |
1164 | cmd->session = session; |
1165 | ||
1166 | reply.session_id = htobe64(session->id); | |
1167 | ||
d3e2ba59 | 1168 | switch (cmd->minor) { |
719af1bb JD |
1169 | case 1: |
1170 | case 2: | |
1171 | case 3: | |
1172 | break; | |
d3e2ba59 JD |
1173 | case 4: /* LTTng sessiond 2.4 */ |
1174 | default: | |
1175 | ret = cmd_create_session_2_4(cmd, session); | |
1176 | break; | |
1177 | } | |
1178 | ||
1179 | lttng_ht_node_init_ulong(&session->session_n, | |
1180 | (unsigned long) session->id); | |
1181 | lttng_ht_add_unique_ulong(sessions_ht, | |
1182 | &session->session_n); | |
1183 | ||
c5b6f4f0 DG |
1184 | DBG("Created session %" PRIu64, session->id); |
1185 | ||
1186 | error: | |
1187 | if (ret < 0) { | |
1188 | reply.ret_code = htobe32(LTTNG_ERR_FATAL); | |
1189 | } else { | |
1190 | reply.ret_code = htobe32(LTTNG_OK); | |
1191 | } | |
1192 | ||
1193 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1194 | if (send_ret < 0) { | |
1195 | ERR("Relayd sending session id"); | |
4169f5ad | 1196 | ret = send_ret; |
c5b6f4f0 DG |
1197 | } |
1198 | ||
1199 | return ret; | |
1200 | } | |
1201 | ||
a4baae1b JD |
1202 | /* |
1203 | * When we have received all the streams and the metadata for a channel, | |
1204 | * we make them visible to the viewer threads. | |
1205 | */ | |
1206 | static | |
1207 | void set_viewer_ready_flag(struct relay_command *cmd) | |
1208 | { | |
1209 | struct relay_stream_recv_handle *node, *tmp_node; | |
1210 | ||
80e8027a JD |
1211 | pthread_mutex_lock(&cmd->session->viewer_ready_lock); |
1212 | ||
a4baae1b JD |
1213 | cds_list_for_each_entry_safe(node, tmp_node, &cmd->recv_head, node) { |
1214 | struct relay_stream *stream; | |
1215 | ||
1216 | rcu_read_lock(); | |
1217 | stream = relay_stream_find_by_id(node->id); | |
1218 | if (!stream) { | |
1219 | /* | |
1220 | * Stream is most probably being cleaned up by the data thread thus | |
1221 | * simply continue to the next one. | |
1222 | */ | |
b3bdb105 | 1223 | rcu_read_unlock(); |
a4baae1b JD |
1224 | continue; |
1225 | } | |
1226 | ||
a4baae1b JD |
1227 | stream->viewer_ready = 1; |
1228 | rcu_read_unlock(); | |
1229 | ||
1230 | /* Clean stream handle node. */ | |
1231 | cds_list_del(&node->node); | |
1232 | free(node); | |
1233 | } | |
1234 | ||
80e8027a | 1235 | pthread_mutex_unlock(&cmd->session->viewer_ready_lock); |
a4baae1b JD |
1236 | return; |
1237 | } | |
1238 | ||
1239 | /* | |
1240 | * Add a recv handle node to the connection recv list with the given stream | |
1241 | * handle. A new node is allocated thus must be freed when the node is deleted | |
1242 | * from the list. | |
1243 | */ | |
1244 | static void queue_stream_handle(uint64_t handle, struct relay_command *cmd) | |
1245 | { | |
1246 | struct relay_stream_recv_handle *node; | |
1247 | ||
1248 | assert(cmd); | |
1249 | ||
1250 | node = zmalloc(sizeof(*node)); | |
1251 | if (!node) { | |
1252 | PERROR("zmalloc queue stream handle"); | |
1253 | return; | |
1254 | } | |
1255 | ||
1256 | node->id = handle; | |
1257 | cds_list_add(&node->node, &cmd->recv_head); | |
1258 | } | |
1259 | ||
b8aa1682 JD |
1260 | /* |
1261 | * relay_add_stream: allocate a new stream for a session | |
1262 | */ | |
1263 | static | |
1264 | int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1265 | struct relay_command *cmd, struct lttng_ht *sessions_ht) |
b8aa1682 JD |
1266 | { |
1267 | struct relay_session *session = cmd->session; | |
b8aa1682 JD |
1268 | struct relay_stream *stream = NULL; |
1269 | struct lttcomm_relayd_status_stream reply; | |
b8aa1682 JD |
1270 | int ret, send_ret; |
1271 | ||
c5b6f4f0 | 1272 | if (!session || cmd->version_check_done == 0) { |
b8aa1682 JD |
1273 | ERR("Trying to add a stream before version check"); |
1274 | ret = -1; | |
1275 | goto end_no_session; | |
1276 | } | |
1277 | ||
b8aa1682 JD |
1278 | stream = zmalloc(sizeof(struct relay_stream)); |
1279 | if (stream == NULL) { | |
1280 | PERROR("relay stream zmalloc"); | |
1281 | ret = -1; | |
1282 | goto end_no_session; | |
1283 | } | |
1284 | ||
0f907de1 JD |
1285 | switch (cmd->minor) { |
1286 | case 1: /* LTTng sessiond 2.1 */ | |
1287 | ret = cmd_recv_stream_2_1(cmd, stream); | |
1288 | break; | |
1289 | case 2: /* LTTng sessiond 2.2 */ | |
1290 | default: | |
1291 | ret = cmd_recv_stream_2_2(cmd, stream); | |
1292 | break; | |
1293 | } | |
1294 | if (ret < 0) { | |
1295 | goto err_free_stream; | |
1296 | } | |
1297 | ||
9d1bbf21 | 1298 | rcu_read_lock(); |
b8aa1682 | 1299 | stream->stream_handle = ++last_relay_stream_id; |
173af62f | 1300 | stream->prev_seq = -1ULL; |
b8aa1682 | 1301 | stream->session = session; |
1c20f0e2 | 1302 | stream->index_fd = -1; |
d3e2ba59 JD |
1303 | stream->read_index_fd = -1; |
1304 | stream->ctf_trace = NULL; | |
1305 | pthread_mutex_init(&stream->lock, NULL); | |
b8aa1682 | 1306 | |
0f907de1 | 1307 | ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG); |
b8aa1682 | 1308 | if (ret < 0) { |
b8aa1682 JD |
1309 | ERR("relay creating output directory"); |
1310 | goto end; | |
1311 | } | |
1312 | ||
be96a7d1 DG |
1313 | /* |
1314 | * No need to use run_as API here because whatever we receives, the relayd | |
1315 | * uses its own credentials for the stream files. | |
1316 | */ | |
0f907de1 | 1317 | ret = utils_create_stream_file(stream->path_name, stream->channel_name, |
1c20f0e2 | 1318 | stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL); |
b8aa1682 | 1319 | if (ret < 0) { |
0f907de1 | 1320 | ERR("Create output file"); |
b8aa1682 JD |
1321 | goto end; |
1322 | } | |
b8aa1682 | 1323 | stream->fd = ret; |
0f907de1 JD |
1324 | if (stream->tracefile_size) { |
1325 | DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name); | |
1326 | } else { | |
1327 | DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name); | |
1328 | } | |
b8aa1682 | 1329 | |
d3e2ba59 JD |
1330 | if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) { |
1331 | stream->metadata_flag = 1; | |
1332 | /* | |
1333 | * When we receive a new metadata stream, we create a new | |
1334 | * ctf_trace and we assign this ctf_trace to all streams with | |
1335 | * the same path. | |
1336 | * | |
1337 | * If later on we receive a new stream for the same ctf_trace, | |
1338 | * we copy the information from the first hit in the HT to the | |
1339 | * new stream. | |
1340 | */ | |
1341 | stream->ctf_trace = ctf_trace_create(); | |
1342 | if (!stream->ctf_trace) { | |
1343 | ret = -1; | |
1344 | goto end; | |
1345 | } | |
1346 | stream->ctf_trace->refcount++; | |
1347 | stream->ctf_trace->metadata_stream = stream; | |
1348 | } | |
1349 | ctf_trace_assign(cmd->ctf_traces_ht, stream); | |
c07d8a78 | 1350 | stream->ctf_traces_ht = cmd->ctf_traces_ht; |
d3e2ba59 | 1351 | |
a4baae1b JD |
1352 | /* |
1353 | * Add the stream handle in the recv list of the connection. Once the end | |
1354 | * stream message is received, this list is emptied and streams are set | |
1355 | * with the viewer ready flag. | |
1356 | */ | |
601262d6 | 1357 | queue_stream_handle(stream->stream_handle, cmd); |
a4baae1b | 1358 | |
b8aa1682 JD |
1359 | lttng_ht_node_init_ulong(&stream->stream_n, |
1360 | (unsigned long) stream->stream_handle); | |
d3e2ba59 | 1361 | lttng_ht_add_unique_ulong(relay_streams_ht, |
b8aa1682 JD |
1362 | &stream->stream_n); |
1363 | ||
d3e2ba59 JD |
1364 | lttng_ht_node_init_str(&stream->ctf_trace_node, stream->path_name); |
1365 | lttng_ht_add_str(cmd->ctf_traces_ht, &stream->ctf_trace_node); | |
87b576ec | 1366 | session->stream_count++; |
d3e2ba59 | 1367 | |
1c20f0e2 JD |
1368 | DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name, |
1369 | stream->stream_handle); | |
b8aa1682 JD |
1370 | |
1371 | end: | |
5af40280 | 1372 | reply.handle = htobe64(stream->stream_handle); |
b8aa1682 JD |
1373 | /* send the session id to the client or a negative return code on error */ |
1374 | if (ret < 0) { | |
f73fabfd | 1375 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
5af40280 CB |
1376 | /* stream was not properly added to the ht, so free it */ |
1377 | free(stream); | |
b8aa1682 | 1378 | } else { |
f73fabfd | 1379 | reply.ret_code = htobe32(LTTNG_OK); |
b8aa1682 | 1380 | } |
5af40280 | 1381 | |
b8aa1682 JD |
1382 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, |
1383 | sizeof(struct lttcomm_relayd_status_stream), 0); | |
1384 | if (send_ret < 0) { | |
1385 | ERR("Relay sending stream id"); | |
4169f5ad | 1386 | ret = send_ret; |
b8aa1682 | 1387 | } |
9d1bbf21 | 1388 | rcu_read_unlock(); |
b8aa1682 JD |
1389 | |
1390 | end_no_session: | |
1391 | return ret; | |
0f907de1 JD |
1392 | |
1393 | err_free_stream: | |
1394 | free(stream->path_name); | |
1395 | free(stream->channel_name); | |
1396 | free(stream); | |
1397 | return ret; | |
b8aa1682 JD |
1398 | } |
1399 | ||
173af62f DG |
1400 | /* |
1401 | * relay_close_stream: close a specific stream | |
1402 | */ | |
1403 | static | |
1404 | int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, | |
92c6ca54 | 1405 | struct relay_command *cmd) |
173af62f | 1406 | { |
94d49140 | 1407 | int ret, send_ret; |
173af62f DG |
1408 | struct relay_session *session = cmd->session; |
1409 | struct lttcomm_relayd_close_stream stream_info; | |
1410 | struct lttcomm_relayd_generic_reply reply; | |
1411 | struct relay_stream *stream; | |
173af62f DG |
1412 | |
1413 | DBG("Close stream received"); | |
1414 | ||
c5b6f4f0 | 1415 | if (!session || cmd->version_check_done == 0) { |
173af62f DG |
1416 | ERR("Trying to close a stream before version check"); |
1417 | ret = -1; | |
1418 | goto end_no_session; | |
1419 | } | |
1420 | ||
1421 | ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info, | |
7c5aef62 | 1422 | sizeof(struct lttcomm_relayd_close_stream), 0); |
173af62f | 1423 | if (ret < sizeof(struct lttcomm_relayd_close_stream)) { |
a6cd2b97 DG |
1424 | if (ret == 0) { |
1425 | /* Orderly shutdown. Not necessary to print an error. */ | |
1426 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1427 | } else { | |
1428 | ERR("Relay didn't receive valid add_stream struct size : %d", ret); | |
1429 | } | |
173af62f DG |
1430 | ret = -1; |
1431 | goto end_no_session; | |
1432 | } | |
1433 | ||
1434 | rcu_read_lock(); | |
d3e2ba59 | 1435 | stream = relay_stream_find_by_id(be64toh(stream_info.stream_id)); |
173af62f DG |
1436 | if (!stream) { |
1437 | ret = -1; | |
1438 | goto end_unlock; | |
1439 | } | |
1440 | ||
8e2583a4 | 1441 | stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num); |
173af62f | 1442 | stream->close_flag = 1; |
87b576ec JD |
1443 | session->stream_count--; |
1444 | assert(session->stream_count >= 0); | |
173af62f DG |
1445 | |
1446 | if (close_stream_check(stream)) { | |
c07d8a78 | 1447 | destroy_stream(stream); |
173af62f DG |
1448 | } |
1449 | ||
1450 | end_unlock: | |
1451 | rcu_read_unlock(); | |
1452 | ||
1453 | if (ret < 0) { | |
f73fabfd | 1454 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
173af62f | 1455 | } else { |
f73fabfd | 1456 | reply.ret_code = htobe32(LTTNG_OK); |
173af62f DG |
1457 | } |
1458 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1459 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1460 | if (send_ret < 0) { | |
1461 | ERR("Relay sending stream id"); | |
4169f5ad | 1462 | ret = send_ret; |
173af62f DG |
1463 | } |
1464 | ||
1465 | end_no_session: | |
1466 | return ret; | |
1467 | } | |
1468 | ||
b8aa1682 JD |
1469 | /* |
1470 | * relay_unknown_command: send -1 if received unknown command | |
1471 | */ | |
1472 | static | |
1473 | void relay_unknown_command(struct relay_command *cmd) | |
1474 | { | |
1475 | struct lttcomm_relayd_generic_reply reply; | |
1476 | int ret; | |
1477 | ||
f73fabfd | 1478 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
b8aa1682 JD |
1479 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, |
1480 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1481 | if (ret < 0) { | |
1482 | ERR("Relay sending unknown command"); | |
1483 | } | |
1484 | } | |
1485 | ||
1486 | /* | |
1487 | * relay_start: send an acknowledgment to the client to tell if we are | |
1488 | * ready to receive data. We are ready if a session is established. | |
1489 | */ | |
1490 | static | |
1491 | int relay_start(struct lttcomm_relayd_hdr *recv_hdr, | |
1492 | struct relay_command *cmd) | |
1493 | { | |
f73fabfd | 1494 | int ret = htobe32(LTTNG_OK); |
b8aa1682 JD |
1495 | struct lttcomm_relayd_generic_reply reply; |
1496 | struct relay_session *session = cmd->session; | |
1497 | ||
1498 | if (!session) { | |
1499 | DBG("Trying to start the streaming without a session established"); | |
f73fabfd | 1500 | ret = htobe32(LTTNG_ERR_UNK); |
b8aa1682 JD |
1501 | } |
1502 | ||
1503 | reply.ret_code = ret; | |
1504 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1505 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1506 | if (ret < 0) { | |
1507 | ERR("Relay sending start ack"); | |
1508 | } | |
1509 | ||
1510 | return ret; | |
1511 | } | |
1512 | ||
1d4dfdef DG |
1513 | /* |
1514 | * Append padding to the file pointed by the file descriptor fd. | |
1515 | */ | |
1516 | static int write_padding_to_file(int fd, uint32_t size) | |
1517 | { | |
6cd525e8 | 1518 | ssize_t ret = 0; |
1d4dfdef DG |
1519 | char *zeros; |
1520 | ||
1521 | if (size == 0) { | |
1522 | goto end; | |
1523 | } | |
1524 | ||
1525 | zeros = zmalloc(size); | |
1526 | if (zeros == NULL) { | |
1527 | PERROR("zmalloc zeros for padding"); | |
1528 | ret = -1; | |
1529 | goto end; | |
1530 | } | |
1531 | ||
6cd525e8 MD |
1532 | ret = lttng_write(fd, zeros, size); |
1533 | if (ret < size) { | |
1d4dfdef DG |
1534 | PERROR("write padding to file"); |
1535 | } | |
1536 | ||
e986c7a1 DG |
1537 | free(zeros); |
1538 | ||
1d4dfdef DG |
1539 | end: |
1540 | return ret; | |
1541 | } | |
1542 | ||
b8aa1682 JD |
1543 | /* |
1544 | * relay_recv_metadata: receive the metada for the session. | |
1545 | */ | |
1546 | static | |
1547 | int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1548 | struct relay_command *cmd) |
b8aa1682 | 1549 | { |
f73fabfd | 1550 | int ret = htobe32(LTTNG_OK); |
6cd525e8 | 1551 | ssize_t size_ret; |
b8aa1682 JD |
1552 | struct relay_session *session = cmd->session; |
1553 | struct lttcomm_relayd_metadata_payload *metadata_struct; | |
1554 | struct relay_stream *metadata_stream; | |
1555 | uint64_t data_size, payload_size; | |
1556 | ||
1557 | if (!session) { | |
1558 | ERR("Metadata sent before version check"); | |
1559 | ret = -1; | |
1560 | goto end; | |
1561 | } | |
1562 | ||
f6416125 MD |
1563 | data_size = payload_size = be64toh(recv_hdr->data_size); |
1564 | if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) { | |
1565 | ERR("Incorrect data size"); | |
1566 | ret = -1; | |
1567 | goto end; | |
1568 | } | |
1569 | payload_size -= sizeof(struct lttcomm_relayd_metadata_payload); | |
1570 | ||
b8aa1682 | 1571 | if (data_buffer_size < data_size) { |
d7b3776f | 1572 | /* In case the realloc fails, we can free the memory */ |
c617c0c6 MD |
1573 | char *tmp_data_ptr; |
1574 | ||
1575 | tmp_data_ptr = realloc(data_buffer, data_size); | |
1576 | if (!tmp_data_ptr) { | |
b8aa1682 | 1577 | ERR("Allocating data buffer"); |
c617c0c6 | 1578 | free(data_buffer); |
b8aa1682 JD |
1579 | ret = -1; |
1580 | goto end; | |
1581 | } | |
c617c0c6 | 1582 | data_buffer = tmp_data_ptr; |
b8aa1682 JD |
1583 | data_buffer_size = data_size; |
1584 | } | |
1585 | memset(data_buffer, 0, data_size); | |
77c7c900 | 1586 | DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size); |
7c5aef62 | 1587 | ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0); |
b8aa1682 | 1588 | if (ret < 0 || ret != data_size) { |
a6cd2b97 DG |
1589 | if (ret == 0) { |
1590 | /* Orderly shutdown. Not necessary to print an error. */ | |
1591 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1592 | } else { | |
1593 | ERR("Relay didn't receive the whole metadata"); | |
1594 | } | |
b8aa1682 | 1595 | ret = -1; |
b8aa1682 JD |
1596 | goto end; |
1597 | } | |
1598 | metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer; | |
9d1bbf21 MD |
1599 | |
1600 | rcu_read_lock(); | |
d3e2ba59 JD |
1601 | metadata_stream = relay_stream_find_by_id( |
1602 | be64toh(metadata_struct->stream_id)); | |
b8aa1682 JD |
1603 | if (!metadata_stream) { |
1604 | ret = -1; | |
9d1bbf21 | 1605 | goto end_unlock; |
b8aa1682 JD |
1606 | } |
1607 | ||
6cd525e8 MD |
1608 | size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload, |
1609 | payload_size); | |
1610 | if (size_ret < payload_size) { | |
b8aa1682 JD |
1611 | ERR("Relay error writing metadata on file"); |
1612 | ret = -1; | |
9d1bbf21 | 1613 | goto end_unlock; |
b8aa1682 | 1614 | } |
1d4dfdef DG |
1615 | |
1616 | ret = write_padding_to_file(metadata_stream->fd, | |
1617 | be32toh(metadata_struct->padding_size)); | |
1618 | if (ret < 0) { | |
1619 | goto end_unlock; | |
1620 | } | |
d3e2ba59 JD |
1621 | metadata_stream->ctf_trace->metadata_received += |
1622 | payload_size + be32toh(metadata_struct->padding_size); | |
1d4dfdef | 1623 | |
b8aa1682 JD |
1624 | DBG2("Relay metadata written"); |
1625 | ||
9d1bbf21 | 1626 | end_unlock: |
6e3c5836 | 1627 | rcu_read_unlock(); |
b8aa1682 JD |
1628 | end: |
1629 | return ret; | |
1630 | } | |
1631 | ||
1632 | /* | |
1633 | * relay_send_version: send relayd version number | |
1634 | */ | |
1635 | static | |
1636 | int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1637 | struct relay_command *cmd, struct lttng_ht *sessions_ht) |
b8aa1682 | 1638 | { |
7f51dcba | 1639 | int ret; |
092b6259 | 1640 | struct lttcomm_relayd_version reply, msg; |
b8aa1682 | 1641 | |
c5b6f4f0 DG |
1642 | assert(cmd); |
1643 | ||
1644 | cmd->version_check_done = 1; | |
b8aa1682 | 1645 | |
092b6259 | 1646 | /* Get version from the other side. */ |
7c5aef62 | 1647 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); |
092b6259 | 1648 | if (ret < 0 || ret != sizeof(msg)) { |
a6cd2b97 DG |
1649 | if (ret == 0) { |
1650 | /* Orderly shutdown. Not necessary to print an error. */ | |
1651 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1652 | } else { | |
1653 | ERR("Relay failed to receive the version values."); | |
1654 | } | |
092b6259 | 1655 | ret = -1; |
092b6259 DG |
1656 | goto end; |
1657 | } | |
1658 | ||
d83a952c MD |
1659 | reply.major = RELAYD_VERSION_COMM_MAJOR; |
1660 | reply.minor = RELAYD_VERSION_COMM_MINOR; | |
d4519fa3 JD |
1661 | |
1662 | /* Major versions must be the same */ | |
1663 | if (reply.major != be32toh(msg.major)) { | |
6151a90f JD |
1664 | DBG("Incompatible major versions (%u vs %u), deleting session", |
1665 | reply.major, be32toh(msg.major)); | |
d3e2ba59 | 1666 | relay_delete_session(cmd, sessions_ht); |
d4519fa3 JD |
1667 | ret = 0; |
1668 | goto end; | |
1669 | } | |
1670 | ||
0f907de1 JD |
1671 | cmd->major = reply.major; |
1672 | /* We adapt to the lowest compatible version */ | |
1673 | if (reply.minor <= be32toh(msg.minor)) { | |
1674 | cmd->minor = reply.minor; | |
1675 | } else { | |
1676 | cmd->minor = be32toh(msg.minor); | |
1677 | } | |
1678 | ||
6151a90f JD |
1679 | reply.major = htobe32(reply.major); |
1680 | reply.minor = htobe32(reply.minor); | |
1681 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1682 | sizeof(struct lttcomm_relayd_version), 0); | |
1683 | if (ret < 0) { | |
1684 | ERR("Relay sending version"); | |
1685 | } | |
1686 | ||
0f907de1 JD |
1687 | DBG("Version check done using protocol %u.%u", cmd->major, |
1688 | cmd->minor); | |
b8aa1682 JD |
1689 | |
1690 | end: | |
1691 | return ret; | |
1692 | } | |
1693 | ||
c8f59ee5 | 1694 | /* |
6d805429 | 1695 | * Check for data pending for a given stream id from the session daemon. |
c8f59ee5 DG |
1696 | */ |
1697 | static | |
6d805429 | 1698 | int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, |
d3e2ba59 | 1699 | struct relay_command *cmd) |
c8f59ee5 DG |
1700 | { |
1701 | struct relay_session *session = cmd->session; | |
6d805429 | 1702 | struct lttcomm_relayd_data_pending msg; |
c8f59ee5 DG |
1703 | struct lttcomm_relayd_generic_reply reply; |
1704 | struct relay_stream *stream; | |
1705 | int ret; | |
c8f59ee5 DG |
1706 | uint64_t last_net_seq_num, stream_id; |
1707 | ||
6d805429 | 1708 | DBG("Data pending command received"); |
c8f59ee5 | 1709 | |
c5b6f4f0 | 1710 | if (!session || cmd->version_check_done == 0) { |
c8f59ee5 DG |
1711 | ERR("Trying to check for data before version check"); |
1712 | ret = -1; | |
1713 | goto end_no_session; | |
1714 | } | |
1715 | ||
7c5aef62 | 1716 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); |
c8f59ee5 | 1717 | if (ret < sizeof(msg)) { |
a6cd2b97 DG |
1718 | if (ret == 0) { |
1719 | /* Orderly shutdown. Not necessary to print an error. */ | |
1720 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1721 | } else { | |
1722 | ERR("Relay didn't receive valid data_pending struct size : %d", | |
1723 | ret); | |
1724 | } | |
c8f59ee5 DG |
1725 | ret = -1; |
1726 | goto end_no_session; | |
1727 | } | |
1728 | ||
1729 | stream_id = be64toh(msg.stream_id); | |
1730 | last_net_seq_num = be64toh(msg.last_net_seq_num); | |
1731 | ||
1732 | rcu_read_lock(); | |
d3e2ba59 | 1733 | stream = relay_stream_find_by_id(stream_id); |
de91f48a | 1734 | if (stream == NULL) { |
c8f59ee5 DG |
1735 | ret = -1; |
1736 | goto end_unlock; | |
1737 | } | |
1738 | ||
6d805429 | 1739 | DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64 |
c8f59ee5 DG |
1740 | " and last_seq %" PRIu64, stream_id, stream->prev_seq, |
1741 | last_net_seq_num); | |
1742 | ||
33832e64 | 1743 | /* Avoid wrapping issue */ |
39df6d9f | 1744 | if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) { |
6d805429 | 1745 | /* Data has in fact been written and is NOT pending */ |
c8f59ee5 | 1746 | ret = 0; |
6d805429 DG |
1747 | } else { |
1748 | /* Data still being streamed thus pending */ | |
1749 | ret = 1; | |
c8f59ee5 DG |
1750 | } |
1751 | ||
f7079f67 DG |
1752 | /* Pending check is now done. */ |
1753 | stream->data_pending_check_done = 1; | |
1754 | ||
c8f59ee5 DG |
1755 | end_unlock: |
1756 | rcu_read_unlock(); | |
1757 | ||
1758 | reply.ret_code = htobe32(ret); | |
1759 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1760 | if (ret < 0) { | |
6d805429 | 1761 | ERR("Relay data pending ret code failed"); |
c8f59ee5 DG |
1762 | } |
1763 | ||
1764 | end_no_session: | |
1765 | return ret; | |
1766 | } | |
1767 | ||
1768 | /* | |
1769 | * Wait for the control socket to reach a quiescent state. | |
1770 | * | |
1771 | * Note that for now, when receiving this command from the session daemon, this | |
1772 | * means that every subsequent commands or data received on the control socket | |
1773 | * has been handled. So, this is why we simply return OK here. | |
1774 | */ | |
1775 | static | |
1776 | int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1777 | struct relay_command *cmd) |
c8f59ee5 DG |
1778 | { |
1779 | int ret; | |
ad7051c0 DG |
1780 | uint64_t stream_id; |
1781 | struct relay_stream *stream; | |
1782 | struct lttng_ht_iter iter; | |
1783 | struct lttcomm_relayd_quiescent_control msg; | |
c8f59ee5 DG |
1784 | struct lttcomm_relayd_generic_reply reply; |
1785 | ||
1786 | DBG("Checking quiescent state on control socket"); | |
1787 | ||
ad7051c0 DG |
1788 | if (!cmd->session || cmd->version_check_done == 0) { |
1789 | ERR("Trying to check for data before version check"); | |
1790 | ret = -1; | |
1791 | goto end_no_session; | |
1792 | } | |
1793 | ||
1794 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1795 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1796 | if (ret == 0) { |
1797 | /* Orderly shutdown. Not necessary to print an error. */ | |
1798 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1799 | } else { | |
1800 | ERR("Relay didn't receive valid begin data_pending struct size: %d", | |
1801 | ret); | |
1802 | } | |
ad7051c0 DG |
1803 | ret = -1; |
1804 | goto end_no_session; | |
1805 | } | |
1806 | ||
1807 | stream_id = be64toh(msg.stream_id); | |
1808 | ||
1809 | rcu_read_lock(); | |
d3e2ba59 JD |
1810 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
1811 | stream_n.node) { | |
ad7051c0 DG |
1812 | if (stream->stream_handle == stream_id) { |
1813 | stream->data_pending_check_done = 1; | |
1814 | DBG("Relay quiescent control pending flag set to %" PRIu64, | |
1815 | stream_id); | |
1816 | break; | |
1817 | } | |
1818 | } | |
1819 | rcu_read_unlock(); | |
1820 | ||
c8f59ee5 DG |
1821 | reply.ret_code = htobe32(LTTNG_OK); |
1822 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1823 | if (ret < 0) { | |
6d805429 | 1824 | ERR("Relay data quiescent control ret code failed"); |
c8f59ee5 DG |
1825 | } |
1826 | ||
ad7051c0 | 1827 | end_no_session: |
c8f59ee5 DG |
1828 | return ret; |
1829 | } | |
1830 | ||
f7079f67 DG |
1831 | /* |
1832 | * Initialize a data pending command. This means that a client is about to ask | |
1833 | * for data pending for each stream he/she holds. Simply iterate over all | |
1834 | * streams of a session and set the data_pending_check_done flag. | |
1835 | * | |
1836 | * This command returns to the client a LTTNG_OK code. | |
1837 | */ | |
1838 | static | |
1839 | int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1840 | struct relay_command *cmd) |
f7079f67 DG |
1841 | { |
1842 | int ret; | |
1843 | struct lttng_ht_iter iter; | |
1844 | struct lttcomm_relayd_begin_data_pending msg; | |
1845 | struct lttcomm_relayd_generic_reply reply; | |
1846 | struct relay_stream *stream; | |
1847 | uint64_t session_id; | |
1848 | ||
1849 | assert(recv_hdr); | |
1850 | assert(cmd); | |
f7079f67 DG |
1851 | |
1852 | DBG("Init streams for data pending"); | |
1853 | ||
1854 | if (!cmd->session || cmd->version_check_done == 0) { | |
1855 | ERR("Trying to check for data before version check"); | |
1856 | ret = -1; | |
1857 | goto end_no_session; | |
1858 | } | |
1859 | ||
1860 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1861 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1862 | if (ret == 0) { |
1863 | /* Orderly shutdown. Not necessary to print an error. */ | |
1864 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1865 | } else { | |
1866 | ERR("Relay didn't receive valid begin data_pending struct size: %d", | |
1867 | ret); | |
1868 | } | |
f7079f67 DG |
1869 | ret = -1; |
1870 | goto end_no_session; | |
1871 | } | |
1872 | ||
1873 | session_id = be64toh(msg.session_id); | |
1874 | ||
1875 | /* | |
1876 | * Iterate over all streams to set the begin data pending flag. For now, the | |
1877 | * streams are indexed by stream handle so we have to iterate over all | |
1878 | * streams to find the one associated with the right session_id. | |
1879 | */ | |
1880 | rcu_read_lock(); | |
d3e2ba59 JD |
1881 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
1882 | stream_n.node) { | |
f7079f67 DG |
1883 | if (stream->session->id == session_id) { |
1884 | stream->data_pending_check_done = 0; | |
1885 | DBG("Set begin data pending flag to stream %" PRIu64, | |
1886 | stream->stream_handle); | |
1887 | } | |
1888 | } | |
1889 | rcu_read_unlock(); | |
1890 | ||
1891 | /* All good, send back reply. */ | |
1892 | reply.ret_code = htobe32(LTTNG_OK); | |
1893 | ||
1894 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1895 | if (ret < 0) { | |
1896 | ERR("Relay begin data pending send reply failed"); | |
1897 | } | |
1898 | ||
1899 | end_no_session: | |
1900 | return ret; | |
1901 | } | |
1902 | ||
1903 | /* | |
1904 | * End data pending command. This will check, for a given session id, if each | |
1905 | * stream associated with it has its data_pending_check_done flag set. If not, | |
1906 | * this means that the client lost track of the stream but the data is still | |
1907 | * being streamed on our side. In this case, we inform the client that data is | |
1908 | * inflight. | |
1909 | * | |
1910 | * Return to the client if there is data in flight or not with a ret_code. | |
1911 | */ | |
1912 | static | |
1913 | int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1914 | struct relay_command *cmd) |
f7079f67 DG |
1915 | { |
1916 | int ret; | |
1917 | struct lttng_ht_iter iter; | |
1918 | struct lttcomm_relayd_end_data_pending msg; | |
1919 | struct lttcomm_relayd_generic_reply reply; | |
1920 | struct relay_stream *stream; | |
1921 | uint64_t session_id; | |
1922 | uint32_t is_data_inflight = 0; | |
1923 | ||
1924 | assert(recv_hdr); | |
1925 | assert(cmd); | |
f7079f67 DG |
1926 | |
1927 | DBG("End data pending command"); | |
1928 | ||
1929 | if (!cmd->session || cmd->version_check_done == 0) { | |
1930 | ERR("Trying to check for data before version check"); | |
1931 | ret = -1; | |
1932 | goto end_no_session; | |
1933 | } | |
1934 | ||
1935 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1936 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1937 | if (ret == 0) { |
1938 | /* Orderly shutdown. Not necessary to print an error. */ | |
1939 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1940 | } else { | |
1941 | ERR("Relay didn't receive valid end data_pending struct size: %d", | |
1942 | ret); | |
1943 | } | |
f7079f67 DG |
1944 | ret = -1; |
1945 | goto end_no_session; | |
1946 | } | |
1947 | ||
1948 | session_id = be64toh(msg.session_id); | |
1949 | ||
1950 | /* Iterate over all streams to see if the begin data pending flag is set. */ | |
1951 | rcu_read_lock(); | |
d3e2ba59 JD |
1952 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
1953 | stream_n.node) { | |
f7079f67 DG |
1954 | if (stream->session->id == session_id && |
1955 | !stream->data_pending_check_done) { | |
1956 | is_data_inflight = 1; | |
1957 | DBG("Data is still in flight for stream %" PRIu64, | |
1958 | stream->stream_handle); | |
1959 | break; | |
1960 | } | |
1961 | } | |
1962 | rcu_read_unlock(); | |
1963 | ||
1964 | /* All good, send back reply. */ | |
1965 | reply.ret_code = htobe32(is_data_inflight); | |
1966 | ||
1967 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1968 | if (ret < 0) { | |
1969 | ERR("Relay end data pending send reply failed"); | |
1970 | } | |
1971 | ||
1972 | end_no_session: | |
1973 | return ret; | |
1974 | } | |
1975 | ||
1c20f0e2 JD |
1976 | /* |
1977 | * Receive an index for a specific stream. | |
1978 | * | |
1979 | * Return 0 on success else a negative value. | |
1980 | */ | |
1981 | static | |
1982 | int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr, | |
0a6518b0 | 1983 | struct relay_command *cmd) |
1c20f0e2 JD |
1984 | { |
1985 | int ret, send_ret, index_created = 0; | |
1986 | struct relay_session *session = cmd->session; | |
1987 | struct lttcomm_relayd_index index_info; | |
1988 | struct relay_index *index, *wr_index = NULL; | |
1989 | struct lttcomm_relayd_generic_reply reply; | |
1990 | struct relay_stream *stream; | |
1991 | uint64_t net_seq_num; | |
1992 | ||
1993 | assert(cmd); | |
1c20f0e2 JD |
1994 | |
1995 | DBG("Relay receiving index"); | |
1996 | ||
1997 | if (!session || cmd->version_check_done == 0) { | |
1998 | ERR("Trying to close a stream before version check"); | |
1999 | ret = -1; | |
2000 | goto end_no_session; | |
2001 | } | |
2002 | ||
2003 | ret = cmd->sock->ops->recvmsg(cmd->sock, &index_info, | |
2004 | sizeof(index_info), 0); | |
2005 | if (ret < sizeof(index_info)) { | |
2006 | if (ret == 0) { | |
2007 | /* Orderly shutdown. Not necessary to print an error. */ | |
2008 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
2009 | } else { | |
2010 | ERR("Relay didn't receive valid index struct size : %d", ret); | |
2011 | } | |
2012 | ret = -1; | |
2013 | goto end_no_session; | |
2014 | } | |
2015 | ||
2016 | net_seq_num = be64toh(index_info.net_seq_num); | |
2017 | ||
2018 | rcu_read_lock(); | |
d3e2ba59 | 2019 | stream = relay_stream_find_by_id(be64toh(index_info.relay_stream_id)); |
1c20f0e2 JD |
2020 | if (!stream) { |
2021 | ret = -1; | |
2022 | goto end_rcu_unlock; | |
2023 | } | |
2024 | ||
d3e2ba59 JD |
2025 | /* Live beacon handling */ |
2026 | if (index_info.packet_size == 0) { | |
2027 | DBG("Received live beacon for stream %" PRIu64, stream->stream_handle); | |
2028 | ||
2029 | /* | |
2030 | * Only flag a stream inactive when it has already received data. | |
2031 | */ | |
2032 | if (stream->total_index_received > 0) { | |
2033 | stream->beacon_ts_end = be64toh(index_info.timestamp_end); | |
2034 | } | |
2035 | ret = 0; | |
2036 | goto end_rcu_unlock; | |
2037 | } else { | |
2038 | stream->beacon_ts_end = -1ULL; | |
2039 | } | |
2040 | ||
0a6518b0 | 2041 | index = relay_index_find(stream->stream_handle, net_seq_num); |
1c20f0e2 JD |
2042 | if (!index) { |
2043 | /* A successful creation will add the object to the HT. */ | |
2044 | index = relay_index_create(stream->stream_handle, net_seq_num); | |
2045 | if (!index) { | |
2046 | goto end_rcu_unlock; | |
2047 | } | |
2048 | index_created = 1; | |
2049 | } | |
2050 | ||
2051 | copy_index_control_data(index, &index_info); | |
2052 | ||
2053 | if (index_created) { | |
2054 | /* | |
2055 | * Try to add the relay index object to the hash table. If an object | |
2056 | * already exist, destroy back the index created, set the data in this | |
2057 | * object and write it on disk. | |
2058 | */ | |
0a6518b0 | 2059 | relay_index_add(index, &wr_index); |
1c20f0e2 JD |
2060 | if (wr_index) { |
2061 | copy_index_control_data(wr_index, &index_info); | |
2062 | free(index); | |
2063 | } | |
2064 | } else { | |
2065 | /* The index already exists so write it on disk. */ | |
2066 | wr_index = index; | |
2067 | } | |
2068 | ||
2069 | /* Do we have a writable ready index to write on disk. */ | |
2070 | if (wr_index) { | |
2071 | /* Starting at 2.4, create the index file if none available. */ | |
2072 | if (cmd->minor >= 4 && stream->index_fd < 0) { | |
2073 | ret = index_create_file(stream->path_name, stream->channel_name, | |
2074 | relayd_uid, relayd_gid, stream->tracefile_size, | |
2075 | stream->tracefile_count_current); | |
2076 | if (ret < 0) { | |
2077 | goto end_rcu_unlock; | |
2078 | } | |
2079 | stream->index_fd = ret; | |
2080 | } | |
2081 | ||
0a6518b0 | 2082 | ret = relay_index_write(wr_index->fd, wr_index); |
1c20f0e2 JD |
2083 | if (ret < 0) { |
2084 | goto end_rcu_unlock; | |
2085 | } | |
d3e2ba59 | 2086 | stream->total_index_received++; |
1c20f0e2 JD |
2087 | } |
2088 | ||
2089 | end_rcu_unlock: | |
2090 | rcu_read_unlock(); | |
2091 | ||
2092 | if (ret < 0) { | |
2093 | reply.ret_code = htobe32(LTTNG_ERR_UNK); | |
2094 | } else { | |
2095 | reply.ret_code = htobe32(LTTNG_OK); | |
2096 | } | |
2097 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
2098 | if (send_ret < 0) { | |
2099 | ERR("Relay sending close index id reply"); | |
2100 | ret = send_ret; | |
2101 | } | |
2102 | ||
2103 | end_no_session: | |
2104 | return ret; | |
2105 | } | |
2106 | ||
a4baae1b JD |
2107 | /* |
2108 | * Receive the streams_sent message. | |
2109 | * | |
2110 | * Return 0 on success else a negative value. | |
2111 | */ | |
2112 | static | |
2113 | int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr, | |
2114 | struct relay_command *cmd) | |
2115 | { | |
2116 | int ret, send_ret; | |
2117 | struct lttcomm_relayd_generic_reply reply; | |
2118 | ||
2119 | assert(cmd); | |
2120 | ||
2121 | DBG("Relay receiving streams_sent"); | |
2122 | ||
2123 | if (!cmd->session || cmd->version_check_done == 0) { | |
2124 | ERR("Trying to close a stream before version check"); | |
2125 | ret = -1; | |
2126 | goto end_no_session; | |
2127 | } | |
2128 | ||
2129 | /* | |
2130 | * Flag every pending stream in the connection recv list that they are | |
2131 | * ready to be used by the viewer. | |
2132 | */ | |
2133 | set_viewer_ready_flag(cmd); | |
2134 | ||
4a9daf17 JD |
2135 | /* |
2136 | * Inform the viewer that there are new streams in the session. | |
2137 | */ | |
2138 | uatomic_set(&cmd->session->new_streams, 1); | |
2139 | ||
a4baae1b JD |
2140 | reply.ret_code = htobe32(LTTNG_OK); |
2141 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
2142 | if (send_ret < 0) { | |
2143 | ERR("Relay sending sent_stream reply"); | |
2144 | ret = send_ret; | |
2145 | } else { | |
2146 | /* Success. */ | |
2147 | ret = 0; | |
2148 | } | |
2149 | ||
2150 | end_no_session: | |
2151 | return ret; | |
2152 | } | |
2153 | ||
b8aa1682 | 2154 | /* |
d3e2ba59 | 2155 | * Process the commands received on the control socket |
b8aa1682 JD |
2156 | */ |
2157 | static | |
2158 | int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 2159 | struct relay_command *cmd, struct relay_local_data *ctx) |
b8aa1682 JD |
2160 | { |
2161 | int ret = 0; | |
2162 | ||
2163 | switch (be32toh(recv_hdr->cmd)) { | |
b8aa1682 | 2164 | case RELAYD_CREATE_SESSION: |
d3e2ba59 | 2165 | ret = relay_create_session(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 | 2166 | break; |
b8aa1682 | 2167 | case RELAYD_ADD_STREAM: |
d3e2ba59 | 2168 | ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 JD |
2169 | break; |
2170 | case RELAYD_START_DATA: | |
2171 | ret = relay_start(recv_hdr, cmd); | |
2172 | break; | |
2173 | case RELAYD_SEND_METADATA: | |
d3e2ba59 | 2174 | ret = relay_recv_metadata(recv_hdr, cmd); |
b8aa1682 JD |
2175 | break; |
2176 | case RELAYD_VERSION: | |
d3e2ba59 | 2177 | ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 | 2178 | break; |
173af62f | 2179 | case RELAYD_CLOSE_STREAM: |
92c6ca54 | 2180 | ret = relay_close_stream(recv_hdr, cmd); |
173af62f | 2181 | break; |
6d805429 | 2182 | case RELAYD_DATA_PENDING: |
d3e2ba59 | 2183 | ret = relay_data_pending(recv_hdr, cmd); |
c8f59ee5 DG |
2184 | break; |
2185 | case RELAYD_QUIESCENT_CONTROL: | |
d3e2ba59 | 2186 | ret = relay_quiescent_control(recv_hdr, cmd); |
c8f59ee5 | 2187 | break; |
f7079f67 | 2188 | case RELAYD_BEGIN_DATA_PENDING: |
d3e2ba59 | 2189 | ret = relay_begin_data_pending(recv_hdr, cmd); |
f7079f67 DG |
2190 | break; |
2191 | case RELAYD_END_DATA_PENDING: | |
d3e2ba59 | 2192 | ret = relay_end_data_pending(recv_hdr, cmd); |
f7079f67 | 2193 | break; |
1c20f0e2 | 2194 | case RELAYD_SEND_INDEX: |
0a6518b0 | 2195 | ret = relay_recv_index(recv_hdr, cmd); |
1c20f0e2 | 2196 | break; |
a4baae1b JD |
2197 | case RELAYD_STREAMS_SENT: |
2198 | ret = relay_streams_sent(recv_hdr, cmd); | |
2199 | break; | |
b8aa1682 JD |
2200 | case RELAYD_UPDATE_SYNC_INFO: |
2201 | default: | |
2202 | ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd)); | |
2203 | relay_unknown_command(cmd); | |
2204 | ret = -1; | |
2205 | goto end; | |
2206 | } | |
2207 | ||
2208 | end: | |
2209 | return ret; | |
2210 | } | |
2211 | ||
7d2f7452 DG |
2212 | /* |
2213 | * Handle index for a data stream. | |
2214 | * | |
2215 | * RCU read side lock MUST be acquired. | |
2216 | * | |
2217 | * Return 0 on success else a negative value. | |
2218 | */ | |
2219 | static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num, | |
2220 | int rotate_index) | |
2221 | { | |
2222 | int ret = 0, index_created = 0; | |
2223 | uint64_t stream_id, data_offset; | |
2224 | struct relay_index *index, *wr_index = NULL; | |
2225 | ||
2226 | assert(stream); | |
2227 | ||
2228 | stream_id = stream->stream_handle; | |
2229 | /* Get data offset because we are about to update the index. */ | |
2230 | data_offset = htobe64(stream->tracefile_size_current); | |
2231 | ||
2232 | /* | |
2233 | * Lookup for an existing index for that stream id/sequence number. If on | |
2234 | * exists, the control thread already received the data for it thus we need | |
2235 | * to write it on disk. | |
2236 | */ | |
2237 | index = relay_index_find(stream_id, net_seq_num); | |
2238 | if (!index) { | |
2239 | /* A successful creation will add the object to the HT. */ | |
2240 | index = relay_index_create(stream_id, net_seq_num); | |
2241 | if (!index) { | |
2242 | ret = -1; | |
2243 | goto error; | |
2244 | } | |
2245 | index_created = 1; | |
2246 | } | |
2247 | ||
2248 | if (rotate_index || stream->index_fd < 0) { | |
2249 | index->to_close_fd = stream->index_fd; | |
2250 | ret = index_create_file(stream->path_name, stream->channel_name, | |
2251 | relayd_uid, relayd_gid, stream->tracefile_size, | |
2252 | stream->tracefile_count_current); | |
2253 | if (ret < 0) { | |
2254 | /* This will close the stream's index fd if one. */ | |
2255 | relay_index_free_safe(index); | |
2256 | goto error; | |
2257 | } | |
2258 | stream->index_fd = ret; | |
2259 | } | |
2260 | index->fd = stream->index_fd; | |
2261 | index->index_data.offset = data_offset; | |
2262 | ||
2263 | if (index_created) { | |
2264 | /* | |
2265 | * Try to add the relay index object to the hash table. If an object | |
2266 | * already exist, destroy back the index created and set the data. | |
2267 | */ | |
2268 | relay_index_add(index, &wr_index); | |
2269 | if (wr_index) { | |
2270 | /* Copy back data from the created index. */ | |
2271 | wr_index->fd = index->fd; | |
2272 | wr_index->to_close_fd = index->to_close_fd; | |
2273 | wr_index->index_data.offset = data_offset; | |
2274 | free(index); | |
2275 | } | |
2276 | } else { | |
2277 | /* The index already exists so write it on disk. */ | |
2278 | wr_index = index; | |
2279 | } | |
2280 | ||
2281 | /* Do we have a writable ready index to write on disk. */ | |
2282 | if (wr_index) { | |
2283 | ret = relay_index_write(wr_index->fd, wr_index); | |
2284 | if (ret < 0) { | |
2285 | goto error; | |
2286 | } | |
2287 | stream->total_index_received++; | |
2288 | } | |
2289 | ||
2290 | error: | |
2291 | return ret; | |
2292 | } | |
2293 | ||
b8aa1682 JD |
2294 | /* |
2295 | * relay_process_data: Process the data received on the data socket | |
2296 | */ | |
2297 | static | |
0a6518b0 | 2298 | int relay_process_data(struct relay_command *cmd) |
b8aa1682 | 2299 | { |
7d2f7452 | 2300 | int ret = 0, rotate_index = 0; |
6cd525e8 | 2301 | ssize_t size_ret; |
b8aa1682 JD |
2302 | struct relay_stream *stream; |
2303 | struct lttcomm_relayd_data_hdr data_hdr; | |
7d2f7452 | 2304 | uint64_t stream_id; |
173af62f | 2305 | uint64_t net_seq_num; |
b8aa1682 JD |
2306 | uint32_t data_size; |
2307 | ||
2308 | ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr, | |
7c5aef62 | 2309 | sizeof(struct lttcomm_relayd_data_hdr), 0); |
b8aa1682 | 2310 | if (ret <= 0) { |
a6cd2b97 DG |
2311 | if (ret == 0) { |
2312 | /* Orderly shutdown. Not necessary to print an error. */ | |
2313 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
2314 | } else { | |
2315 | ERR("Unable to receive data header on sock %d", cmd->sock->fd); | |
2316 | } | |
b8aa1682 JD |
2317 | ret = -1; |
2318 | goto end; | |
2319 | } | |
2320 | ||
2321 | stream_id = be64toh(data_hdr.stream_id); | |
9d1bbf21 MD |
2322 | |
2323 | rcu_read_lock(); | |
d3e2ba59 | 2324 | stream = relay_stream_find_by_id(stream_id); |
b8aa1682 JD |
2325 | if (!stream) { |
2326 | ret = -1; | |
1c20f0e2 | 2327 | goto end_rcu_unlock; |
b8aa1682 JD |
2328 | } |
2329 | ||
2330 | data_size = be32toh(data_hdr.data_size); | |
2331 | if (data_buffer_size < data_size) { | |
c617c0c6 MD |
2332 | char *tmp_data_ptr; |
2333 | ||
2334 | tmp_data_ptr = realloc(data_buffer, data_size); | |
2335 | if (!tmp_data_ptr) { | |
b8aa1682 | 2336 | ERR("Allocating data buffer"); |
c617c0c6 | 2337 | free(data_buffer); |
b8aa1682 | 2338 | ret = -1; |
1c20f0e2 | 2339 | goto end_rcu_unlock; |
b8aa1682 | 2340 | } |
c617c0c6 | 2341 | data_buffer = tmp_data_ptr; |
b8aa1682 JD |
2342 | data_buffer_size = data_size; |
2343 | } | |
2344 | memset(data_buffer, 0, data_size); | |
2345 | ||
173af62f DG |
2346 | net_seq_num = be64toh(data_hdr.net_seq_num); |
2347 | ||
77c7c900 | 2348 | DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64, |
173af62f | 2349 | data_size, stream_id, net_seq_num); |
7c5aef62 | 2350 | ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0); |
b8aa1682 | 2351 | if (ret <= 0) { |
a6cd2b97 DG |
2352 | if (ret == 0) { |
2353 | /* Orderly shutdown. Not necessary to print an error. */ | |
2354 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
2355 | } | |
b8aa1682 | 2356 | ret = -1; |
1c20f0e2 | 2357 | goto end_rcu_unlock; |
b8aa1682 JD |
2358 | } |
2359 | ||
1c20f0e2 | 2360 | /* Check if a rotation is needed. */ |
0f907de1 JD |
2361 | if (stream->tracefile_size > 0 && |
2362 | (stream->tracefile_size_current + data_size) > | |
2363 | stream->tracefile_size) { | |
6b6b9a5a JD |
2364 | struct relay_viewer_stream *vstream; |
2365 | uint64_t new_id; | |
2366 | ||
2367 | new_id = (stream->tracefile_count_current + 1) % | |
2368 | stream->tracefile_count; | |
2369 | /* | |
2370 | * When we wrap-around back to 0, we start overwriting old | |
2371 | * trace data. | |
2372 | */ | |
2373 | if (!stream->tracefile_overwrite && new_id == 0) { | |
2374 | stream->tracefile_overwrite = 1; | |
2375 | } | |
2376 | pthread_mutex_lock(&stream->viewer_stream_rotation_lock); | |
2377 | if (stream->tracefile_overwrite) { | |
2378 | stream->oldest_tracefile_id = | |
2379 | (stream->oldest_tracefile_id + 1) % | |
2380 | stream->tracefile_count; | |
2381 | } | |
2382 | vstream = live_find_viewer_stream_by_id(stream->stream_handle); | |
2383 | if (vstream) { | |
2384 | /* | |
2385 | * The viewer is reading a file about to be | |
2386 | * overwritten. Close the FDs it is | |
2387 | * currently using and let it handle the fault. | |
2388 | */ | |
2389 | if (vstream->tracefile_count_current == new_id) { | |
cef0f7d5 | 2390 | pthread_mutex_lock(&vstream->overwrite_lock); |
6b6b9a5a | 2391 | vstream->abort_flag = 1; |
cef0f7d5 | 2392 | pthread_mutex_unlock(&vstream->overwrite_lock); |
6b6b9a5a JD |
2393 | DBG("Streaming side setting abort_flag on stream %s_%lu\n", |
2394 | stream->channel_name, new_id); | |
2395 | } else if (vstream->tracefile_count_current == | |
2396 | stream->tracefile_count_current) { | |
2397 | /* | |
2398 | * The reader and writer were in the | |
2399 | * same trace file, inform the viewer | |
2400 | * that no new index will ever be added | |
2401 | * to this file. | |
2402 | */ | |
2403 | vstream->close_write_flag = 1; | |
2404 | } | |
2405 | } | |
1c20f0e2 JD |
2406 | ret = utils_rotate_stream_file(stream->path_name, stream->channel_name, |
2407 | stream->tracefile_size, stream->tracefile_count, | |
2408 | relayd_uid, relayd_gid, stream->fd, | |
2409 | &(stream->tracefile_count_current), &stream->fd); | |
cef0f7d5 | 2410 | stream->total_index_received = 0; |
6b6b9a5a | 2411 | pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); |
0f907de1 | 2412 | if (ret < 0) { |
1c20f0e2 JD |
2413 | ERR("Rotating stream output file"); |
2414 | goto end_rcu_unlock; | |
0f907de1 | 2415 | } |
a6976990 DG |
2416 | /* Reset current size because we just perform a stream rotation. */ |
2417 | stream->tracefile_size_current = 0; | |
1c20f0e2 JD |
2418 | rotate_index = 1; |
2419 | } | |
2420 | ||
1c20f0e2 | 2421 | /* |
7d2f7452 DG |
2422 | * Index are handled in protocol version 2.4 and above. Also, snapshot and |
2423 | * index are NOT supported. | |
1c20f0e2 | 2424 | */ |
7d2f7452 DG |
2425 | if (stream->session->minor >= 4 && !stream->session->snapshot) { |
2426 | ret = handle_index_data(stream, net_seq_num, rotate_index); | |
1c20f0e2 | 2427 | if (ret < 0) { |
1c20f0e2 JD |
2428 | goto end_rcu_unlock; |
2429 | } | |
1c20f0e2 JD |
2430 | } |
2431 | ||
7d2f7452 | 2432 | /* Write data to stream output fd. */ |
6cd525e8 MD |
2433 | size_ret = lttng_write(stream->fd, data_buffer, data_size); |
2434 | if (size_ret < data_size) { | |
b8aa1682 JD |
2435 | ERR("Relay error writing data to file"); |
2436 | ret = -1; | |
1c20f0e2 | 2437 | goto end_rcu_unlock; |
b8aa1682 | 2438 | } |
1d4dfdef | 2439 | |
5ab7344e JD |
2440 | DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64, |
2441 | ret, stream->stream_handle); | |
2442 | ||
1d4dfdef DG |
2443 | ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size)); |
2444 | if (ret < 0) { | |
1c20f0e2 | 2445 | goto end_rcu_unlock; |
1d4dfdef | 2446 | } |
1c20f0e2 | 2447 | stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size); |
1d4dfdef | 2448 | |
173af62f DG |
2449 | stream->prev_seq = net_seq_num; |
2450 | ||
2451 | /* Check if we need to close the FD */ | |
2452 | if (close_stream_check(stream)) { | |
c07d8a78 | 2453 | destroy_stream(stream); |
173af62f DG |
2454 | } |
2455 | ||
1c20f0e2 | 2456 | end_rcu_unlock: |
9d1bbf21 | 2457 | rcu_read_unlock(); |
b8aa1682 JD |
2458 | end: |
2459 | return ret; | |
2460 | } | |
2461 | ||
2462 | static | |
9d1bbf21 | 2463 | void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd) |
b8aa1682 JD |
2464 | { |
2465 | int ret; | |
2466 | ||
b8aa1682 JD |
2467 | lttng_poll_del(events, pollfd); |
2468 | ||
2469 | ret = close(pollfd); | |
2470 | if (ret < 0) { | |
2471 | ERR("Closing pollfd %d", pollfd); | |
2472 | } | |
2473 | } | |
2474 | ||
2475 | static | |
2476 | int relay_add_connection(int fd, struct lttng_poll_event *events, | |
2477 | struct lttng_ht *relay_connections_ht) | |
2478 | { | |
b8aa1682 | 2479 | struct relay_command *relay_connection; |
6cd525e8 | 2480 | ssize_t ret; |
b8aa1682 JD |
2481 | |
2482 | relay_connection = zmalloc(sizeof(struct relay_command)); | |
2483 | if (relay_connection == NULL) { | |
2484 | PERROR("Relay command zmalloc"); | |
9d1bbf21 | 2485 | goto error; |
b8aa1682 | 2486 | } |
6cd525e8 MD |
2487 | ret = lttng_read(fd, relay_connection, sizeof(struct relay_command)); |
2488 | if (ret < sizeof(struct relay_command)) { | |
b8aa1682 | 2489 | PERROR("read relay cmd pipe"); |
9d1bbf21 | 2490 | goto error_read; |
b8aa1682 | 2491 | } |
a4baae1b | 2492 | CDS_INIT_LIST_HEAD(&relay_connection->recv_head); |
b8aa1682 | 2493 | |
c07d8a78 DG |
2494 | /* |
2495 | * Only used by the control side and the reference is copied inside each | |
2496 | * stream from that connection. Thus a destroy HT must be done after every | |
2497 | * stream has been destroyed. | |
2498 | */ | |
2499 | if (relay_connection->type == RELAY_CONTROL) { | |
2500 | relay_connection->ctf_traces_ht = lttng_ht_new(0, | |
2501 | LTTNG_HT_TYPE_STRING); | |
2502 | if (!relay_connection->ctf_traces_ht) { | |
2503 | goto error_read; | |
2504 | } | |
d3e2ba59 JD |
2505 | } |
2506 | ||
b8aa1682 JD |
2507 | lttng_ht_node_init_ulong(&relay_connection->sock_n, |
2508 | (unsigned long) relay_connection->sock->fd); | |
9d1bbf21 | 2509 | rcu_read_lock(); |
b8aa1682 JD |
2510 | lttng_ht_add_unique_ulong(relay_connections_ht, |
2511 | &relay_connection->sock_n); | |
9d1bbf21 MD |
2512 | rcu_read_unlock(); |
2513 | return lttng_poll_add(events, | |
b8aa1682 JD |
2514 | relay_connection->sock->fd, |
2515 | LPOLLIN | LPOLLRDHUP); | |
2516 | ||
9d1bbf21 MD |
2517 | error_read: |
2518 | free(relay_connection); | |
2519 | error: | |
2520 | return -1; | |
2521 | } | |
2522 | ||
2523 | static | |
2524 | void deferred_free_connection(struct rcu_head *head) | |
2525 | { | |
2526 | struct relay_command *relay_connection = | |
2527 | caa_container_of(head, struct relay_command, rcu_node); | |
5b6d8097 DG |
2528 | |
2529 | lttcomm_destroy_sock(relay_connection->sock); | |
9d1bbf21 MD |
2530 | free(relay_connection); |
2531 | } | |
2532 | ||
2533 | static | |
2534 | void relay_del_connection(struct lttng_ht *relay_connections_ht, | |
d3e2ba59 JD |
2535 | struct lttng_ht_iter *iter, struct relay_command *relay_connection, |
2536 | struct lttng_ht *sessions_ht) | |
9d1bbf21 MD |
2537 | { |
2538 | int ret; | |
2539 | ||
2540 | ret = lttng_ht_del(relay_connections_ht, iter); | |
2541 | assert(!ret); | |
c07d8a78 | 2542 | |
9d1bbf21 | 2543 | if (relay_connection->type == RELAY_CONTROL) { |
a4baae1b JD |
2544 | struct relay_stream_recv_handle *node, *tmp_node; |
2545 | ||
d3e2ba59 | 2546 | relay_delete_session(relay_connection, sessions_ht); |
c07d8a78 | 2547 | lttng_ht_destroy(relay_connection->ctf_traces_ht); |
a4baae1b JD |
2548 | |
2549 | /* Clean up recv list. */ | |
2550 | cds_list_for_each_entry_safe(node, tmp_node, | |
2551 | &relay_connection->recv_head, node) { | |
2552 | cds_list_del(&node->node); | |
2553 | free(node); | |
2554 | } | |
9d1bbf21 | 2555 | } |
5b6d8097 | 2556 | |
c07d8a78 | 2557 | call_rcu(&relay_connection->rcu_node, deferred_free_connection); |
b8aa1682 JD |
2558 | } |
2559 | ||
2560 | /* | |
2561 | * This thread does the actual work | |
2562 | */ | |
2563 | static | |
2564 | void *relay_thread_worker(void *data) | |
2565 | { | |
beaad64c DG |
2566 | int ret, err = -1, last_seen_data_fd = -1; |
2567 | uint32_t nb_fd; | |
b8aa1682 JD |
2568 | struct relay_command *relay_connection; |
2569 | struct lttng_poll_event events; | |
2570 | struct lttng_ht *relay_connections_ht; | |
2571 | struct lttng_ht_node_ulong *node; | |
2572 | struct lttng_ht_iter iter; | |
b8aa1682 | 2573 | struct lttcomm_relayd_hdr recv_hdr; |
d3e2ba59 JD |
2574 | struct relay_local_data *relay_ctx = (struct relay_local_data *) data; |
2575 | struct lttng_ht *sessions_ht = relay_ctx->sessions_ht; | |
b8aa1682 JD |
2576 | |
2577 | DBG("[thread] Relay worker started"); | |
2578 | ||
9d1bbf21 MD |
2579 | rcu_register_thread(); |
2580 | ||
55706a7d MD |
2581 | health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER); |
2582 | ||
9b5e0863 MD |
2583 | if (testpoint(relayd_thread_worker)) { |
2584 | goto error_testpoint; | |
2585 | } | |
2586 | ||
f385ae0a MD |
2587 | health_code_update(); |
2588 | ||
b8aa1682 JD |
2589 | /* table of connections indexed on socket */ |
2590 | relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
095a4ae5 MD |
2591 | if (!relay_connections_ht) { |
2592 | goto relay_connections_ht_error; | |
2593 | } | |
b8aa1682 | 2594 | |
1c20f0e2 JD |
2595 | /* Tables of received indexes indexed by index handle and net_seq_num. */ |
2596 | indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64); | |
2597 | if (!indexes_ht) { | |
2598 | goto indexes_ht_error; | |
2599 | } | |
2600 | ||
b8aa1682 JD |
2601 | ret = create_thread_poll_set(&events, 2); |
2602 | if (ret < 0) { | |
2603 | goto error_poll_create; | |
2604 | } | |
2605 | ||
2606 | ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP); | |
2607 | if (ret < 0) { | |
2608 | goto error; | |
2609 | } | |
2610 | ||
beaad64c | 2611 | restart: |
b8aa1682 | 2612 | while (1) { |
beaad64c DG |
2613 | int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1; |
2614 | ||
f385ae0a MD |
2615 | health_code_update(); |
2616 | ||
b8aa1682 | 2617 | /* Infinite blocking call, waiting for transmission */ |
87c1611d | 2618 | DBG3("Relayd worker thread polling..."); |
f385ae0a | 2619 | health_poll_entry(); |
b8aa1682 | 2620 | ret = lttng_poll_wait(&events, -1); |
f385ae0a | 2621 | health_poll_exit(); |
b8aa1682 JD |
2622 | if (ret < 0) { |
2623 | /* | |
2624 | * Restart interrupted system call. | |
2625 | */ | |
2626 | if (errno == EINTR) { | |
2627 | goto restart; | |
2628 | } | |
2629 | goto error; | |
2630 | } | |
2631 | ||
0d9c5d77 DG |
2632 | nb_fd = ret; |
2633 | ||
beaad64c DG |
2634 | /* |
2635 | * Process control. The control connection is prioritised so we don't | |
2636 | * starve it with high throughout put tracing data on the data | |
2637 | * connection. | |
2638 | */ | |
b8aa1682 JD |
2639 | for (i = 0; i < nb_fd; i++) { |
2640 | /* Fetch once the poll data */ | |
beaad64c DG |
2641 | uint32_t revents = LTTNG_POLL_GETEV(&events, i); |
2642 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
b8aa1682 | 2643 | |
f385ae0a MD |
2644 | health_code_update(); |
2645 | ||
b8aa1682 JD |
2646 | /* Thread quit pipe has been closed. Killing thread. */ |
2647 | ret = check_thread_quit_pipe(pollfd, revents); | |
2648 | if (ret) { | |
095a4ae5 MD |
2649 | err = 0; |
2650 | goto exit; | |
b8aa1682 JD |
2651 | } |
2652 | ||
2653 | /* Inspect the relay cmd pipe for new connection */ | |
2654 | if (pollfd == relay_cmd_pipe[0]) { | |
2655 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
2656 | ERR("Relay pipe error"); | |
2657 | goto error; | |
2658 | } else if (revents & LPOLLIN) { | |
2659 | DBG("Relay command received"); | |
2660 | ret = relay_add_connection(relay_cmd_pipe[0], | |
2661 | &events, relay_connections_ht); | |
2662 | if (ret < 0) { | |
2663 | goto error; | |
2664 | } | |
2665 | } | |
beaad64c | 2666 | } else if (revents) { |
9d1bbf21 | 2667 | rcu_read_lock(); |
b8aa1682 JD |
2668 | lttng_ht_lookup(relay_connections_ht, |
2669 | (void *)((unsigned long) pollfd), | |
2670 | &iter); | |
2671 | node = lttng_ht_iter_get_node_ulong(&iter); | |
2672 | if (node == NULL) { | |
2673 | DBG2("Relay sock %d not found", pollfd); | |
9d1bbf21 | 2674 | rcu_read_unlock(); |
b8aa1682 JD |
2675 | goto error; |
2676 | } | |
2677 | relay_connection = caa_container_of(node, | |
2678 | struct relay_command, sock_n); | |
2679 | ||
2680 | if (revents & (LPOLLERR)) { | |
2681 | ERR("POLL ERROR"); | |
9d1bbf21 MD |
2682 | relay_cleanup_poll_connection(&events, pollfd); |
2683 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2684 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2685 | if (last_seen_data_fd == pollfd) { |
2686 | last_seen_data_fd = last_notdel_data_fd; | |
2687 | } | |
b8aa1682 JD |
2688 | } else if (revents & (LPOLLHUP | LPOLLRDHUP)) { |
2689 | DBG("Socket %d hung up", pollfd); | |
9d1bbf21 MD |
2690 | relay_cleanup_poll_connection(&events, pollfd); |
2691 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2692 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2693 | if (last_seen_data_fd == pollfd) { |
2694 | last_seen_data_fd = last_notdel_data_fd; | |
2695 | } | |
b8aa1682 JD |
2696 | } else if (revents & LPOLLIN) { |
2697 | /* control socket */ | |
2698 | if (relay_connection->type == RELAY_CONTROL) { | |
2699 | ret = relay_connection->sock->ops->recvmsg( | |
2700 | relay_connection->sock, &recv_hdr, | |
7c5aef62 | 2701 | sizeof(struct lttcomm_relayd_hdr), 0); |
b8aa1682 JD |
2702 | /* connection closed */ |
2703 | if (ret <= 0) { | |
9d1bbf21 MD |
2704 | relay_cleanup_poll_connection(&events, pollfd); |
2705 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2706 | &iter, relay_connection, sessions_ht); |
b8aa1682 JD |
2707 | DBG("Control connection closed with %d", pollfd); |
2708 | } else { | |
2709 | if (relay_connection->session) { | |
77c7c900 | 2710 | DBG2("Relay worker receiving data for session : %" PRIu64, |
b8aa1682 JD |
2711 | relay_connection->session->id); |
2712 | } | |
2713 | ret = relay_process_control(&recv_hdr, | |
d3e2ba59 | 2714 | relay_connection, relay_ctx); |
b8aa1682 | 2715 | if (ret < 0) { |
beaad64c | 2716 | /* Clear the session on error. */ |
9d1bbf21 MD |
2717 | relay_cleanup_poll_connection(&events, pollfd); |
2718 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2719 | &iter, relay_connection, sessions_ht); |
b8aa1682 JD |
2720 | DBG("Connection closed with %d", pollfd); |
2721 | } | |
beaad64c | 2722 | seen_control = 1; |
b8aa1682 | 2723 | } |
beaad64c DG |
2724 | } else { |
2725 | /* | |
2726 | * Flag the last seen data fd not deleted. It will be | |
2727 | * used as the last seen fd if any fd gets deleted in | |
2728 | * this first loop. | |
2729 | */ | |
2730 | last_notdel_data_fd = pollfd; | |
2731 | } | |
2732 | } | |
2733 | rcu_read_unlock(); | |
2734 | } | |
2735 | } | |
2736 | ||
2737 | /* | |
2738 | * The last loop handled a control request, go back to poll to make | |
2739 | * sure we prioritise the control socket. | |
2740 | */ | |
2741 | if (seen_control) { | |
2742 | continue; | |
2743 | } | |
2744 | ||
2745 | if (last_seen_data_fd >= 0) { | |
2746 | for (i = 0; i < nb_fd; i++) { | |
2747 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
f385ae0a MD |
2748 | |
2749 | health_code_update(); | |
2750 | ||
beaad64c DG |
2751 | if (last_seen_data_fd == pollfd) { |
2752 | idx = i; | |
2753 | break; | |
2754 | } | |
2755 | } | |
2756 | } | |
2757 | ||
2758 | /* Process data connection. */ | |
2759 | for (i = idx + 1; i < nb_fd; i++) { | |
2760 | /* Fetch the poll data. */ | |
2761 | uint32_t revents = LTTNG_POLL_GETEV(&events, i); | |
2762 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
2763 | ||
f385ae0a MD |
2764 | health_code_update(); |
2765 | ||
beaad64c DG |
2766 | /* Skip the command pipe. It's handled in the first loop. */ |
2767 | if (pollfd == relay_cmd_pipe[0]) { | |
2768 | continue; | |
2769 | } | |
2770 | ||
2771 | if (revents) { | |
2772 | rcu_read_lock(); | |
2773 | lttng_ht_lookup(relay_connections_ht, | |
2774 | (void *)((unsigned long) pollfd), | |
2775 | &iter); | |
2776 | node = lttng_ht_iter_get_node_ulong(&iter); | |
2777 | if (node == NULL) { | |
2778 | /* Skip it. Might be removed before. */ | |
2779 | rcu_read_unlock(); | |
2780 | continue; | |
2781 | } | |
2782 | relay_connection = caa_container_of(node, | |
2783 | struct relay_command, sock_n); | |
2784 | ||
2785 | if (revents & LPOLLIN) { | |
2786 | if (relay_connection->type != RELAY_DATA) { | |
2787 | continue; | |
2788 | } | |
2789 | ||
0a6518b0 | 2790 | ret = relay_process_data(relay_connection); |
beaad64c DG |
2791 | /* connection closed */ |
2792 | if (ret < 0) { | |
2793 | relay_cleanup_poll_connection(&events, pollfd); | |
2794 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2795 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2796 | DBG("Data connection closed with %d", pollfd); |
2797 | /* | |
2798 | * Every goto restart call sets the last seen fd where | |
2799 | * here we don't really care since we gracefully | |
2800 | * continue the loop after the connection is deleted. | |
2801 | */ | |
2802 | } else { | |
2803 | /* Keep last seen port. */ | |
2804 | last_seen_data_fd = pollfd; | |
2805 | rcu_read_unlock(); | |
2806 | goto restart; | |
b8aa1682 JD |
2807 | } |
2808 | } | |
9d1bbf21 | 2809 | rcu_read_unlock(); |
b8aa1682 JD |
2810 | } |
2811 | } | |
beaad64c | 2812 | last_seen_data_fd = -1; |
b8aa1682 JD |
2813 | } |
2814 | ||
f385ae0a MD |
2815 | /* Normal exit, no error */ |
2816 | ret = 0; | |
2817 | ||
095a4ae5 | 2818 | exit: |
b8aa1682 JD |
2819 | error: |
2820 | lttng_poll_clean(&events); | |
2821 | ||
2822 | /* empty the hash table and free the memory */ | |
9d1bbf21 | 2823 | rcu_read_lock(); |
b8aa1682 | 2824 | cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) { |
f385ae0a MD |
2825 | health_code_update(); |
2826 | ||
b8aa1682 JD |
2827 | node = lttng_ht_iter_get_node_ulong(&iter); |
2828 | if (node) { | |
2829 | relay_connection = caa_container_of(node, | |
2830 | struct relay_command, sock_n); | |
9d1bbf21 | 2831 | relay_del_connection(relay_connections_ht, |
d3e2ba59 | 2832 | &iter, relay_connection, sessions_ht); |
b8aa1682 | 2833 | } |
b8aa1682 | 2834 | } |
94d49140 | 2835 | rcu_read_unlock(); |
7d2f7452 DG |
2836 | error_poll_create: |
2837 | lttng_ht_destroy(indexes_ht); | |
1c20f0e2 | 2838 | indexes_ht_error: |
b8aa1682 | 2839 | lttng_ht_destroy(relay_connections_ht); |
095a4ae5 | 2840 | relay_connections_ht_error: |
6620da75 DG |
2841 | /* Close relay cmd pipes */ |
2842 | utils_close_pipe(relay_cmd_pipe); | |
095a4ae5 MD |
2843 | if (err) { |
2844 | DBG("Thread exited with error"); | |
2845 | } | |
b8aa1682 | 2846 | DBG("Worker thread cleanup complete"); |
095a4ae5 | 2847 | free(data_buffer); |
9b5e0863 | 2848 | error_testpoint: |
f385ae0a MD |
2849 | if (err) { |
2850 | health_error(); | |
2851 | ERR("Health error occurred in %s", __func__); | |
2852 | } | |
2853 | health_unregister(health_relayd); | |
9d1bbf21 | 2854 | rcu_unregister_thread(); |
f385ae0a | 2855 | stop_threads(); |
b8aa1682 JD |
2856 | return NULL; |
2857 | } | |
2858 | ||
2859 | /* | |
2860 | * Create the relay command pipe to wake thread_manage_apps. | |
2861 | * Closed in cleanup(). | |
2862 | */ | |
2863 | static int create_relay_cmd_pipe(void) | |
2864 | { | |
a02de639 | 2865 | int ret; |
b8aa1682 | 2866 | |
a02de639 | 2867 | ret = utils_create_pipe_cloexec(relay_cmd_pipe); |
b8aa1682 | 2868 | |
b8aa1682 JD |
2869 | return ret; |
2870 | } | |
2871 | ||
2872 | /* | |
2873 | * main | |
2874 | */ | |
2875 | int main(int argc, char **argv) | |
2876 | { | |
2877 | int ret = 0; | |
2878 | void *status; | |
d3e2ba59 | 2879 | struct relay_local_data *relay_ctx; |
b8aa1682 | 2880 | |
b8aa1682 JD |
2881 | /* Parse arguments */ |
2882 | progname = argv[0]; | |
cd60b05a | 2883 | if ((ret = set_options(argc, argv)) < 0) { |
a02de639 | 2884 | goto exit; |
b8aa1682 JD |
2885 | } |
2886 | ||
2887 | if ((ret = set_signal_handler()) < 0) { | |
2888 | goto exit; | |
2889 | } | |
2890 | ||
4d513a50 DG |
2891 | /* Try to create directory if -o, --output is specified. */ |
2892 | if (opt_output_path) { | |
994fa64f DG |
2893 | if (*opt_output_path != '/') { |
2894 | ERR("Please specify an absolute path for -o, --output PATH"); | |
2895 | goto exit; | |
2896 | } | |
2897 | ||
4d513a50 DG |
2898 | ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG); |
2899 | if (ret < 0) { | |
2900 | ERR("Unable to create %s", opt_output_path); | |
2901 | goto exit; | |
2902 | } | |
2903 | } | |
2904 | ||
b8aa1682 | 2905 | /* Daemonize */ |
b5218ffb | 2906 | if (opt_daemon || opt_background) { |
3fd27398 MD |
2907 | int i; |
2908 | ||
2909 | ret = lttng_daemonize(&child_ppid, &recv_child_signal, | |
2910 | !opt_background); | |
b8aa1682 | 2911 | if (ret < 0) { |
a02de639 | 2912 | goto exit; |
b8aa1682 | 2913 | } |
3fd27398 MD |
2914 | |
2915 | /* | |
2916 | * We are in the child. Make sure all other file | |
2917 | * descriptors are closed, in case we are called with | |
2918 | * more opened file descriptors than the standard ones. | |
2919 | */ | |
2920 | for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) { | |
2921 | (void) close(i); | |
2922 | } | |
2923 | } | |
2924 | ||
2925 | /* Create thread quit pipe */ | |
2926 | if ((ret = init_thread_quit_pipe()) < 0) { | |
2927 | goto error; | |
b8aa1682 JD |
2928 | } |
2929 | ||
1c20f0e2 JD |
2930 | /* We need those values for the file/dir creation. */ |
2931 | relayd_uid = getuid(); | |
2932 | relayd_gid = getgid(); | |
b8aa1682 | 2933 | |
1c20f0e2 JD |
2934 | /* Check if daemon is UID = 0 */ |
2935 | if (relayd_uid == 0) { | |
8d5c808e | 2936 | if (control_uri->port < 1024 || data_uri->port < 1024 || live_uri->port < 1024) { |
b8aa1682 JD |
2937 | ERR("Need to be root to use ports < 1024"); |
2938 | ret = -1; | |
a02de639 | 2939 | goto exit; |
b8aa1682 JD |
2940 | } |
2941 | } | |
2942 | ||
2943 | /* Setup the thread apps communication pipe. */ | |
2944 | if ((ret = create_relay_cmd_pipe()) < 0) { | |
2945 | goto exit; | |
2946 | } | |
2947 | ||
2948 | /* Init relay command queue. */ | |
2949 | cds_wfq_init(&relay_cmd_queue.queue); | |
2950 | ||
2951 | /* Set up max poll set size */ | |
2952 | lttng_poll_set_max_size(); | |
2953 | ||
554831e7 MD |
2954 | /* Initialize communication library */ |
2955 | lttcomm_init(); | |
87e45c13 | 2956 | lttcomm_inet_init(); |
554831e7 | 2957 | |
d3e2ba59 JD |
2958 | relay_ctx = zmalloc(sizeof(struct relay_local_data)); |
2959 | if (!relay_ctx) { | |
2960 | PERROR("relay_ctx"); | |
2961 | goto exit; | |
2962 | } | |
2963 | ||
2964 | /* tables of sessions indexed by session ID */ | |
2965 | relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
2966 | if (!relay_ctx->sessions_ht) { | |
2967 | goto exit_relay_ctx_sessions; | |
2968 | } | |
2969 | ||
2970 | /* tables of streams indexed by stream ID */ | |
2971 | relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
2972 | if (!relay_streams_ht) { | |
2973 | goto exit_relay_ctx_streams; | |
2974 | } | |
2975 | ||
2976 | /* tables of streams indexed by stream ID */ | |
92c6ca54 DG |
2977 | viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
2978 | if (!viewer_streams_ht) { | |
d3e2ba59 JD |
2979 | goto exit_relay_ctx_viewer_streams; |
2980 | } | |
2981 | ||
55706a7d MD |
2982 | /* Initialize thread health monitoring */ |
2983 | health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES); | |
2984 | if (!health_relayd) { | |
2985 | PERROR("health_app_create error"); | |
2986 | goto exit_health_app_create; | |
2987 | } | |
2988 | ||
65931c8b MD |
2989 | ret = utils_create_pipe(health_quit_pipe); |
2990 | if (ret < 0) { | |
2991 | goto error_health_pipe; | |
2992 | } | |
2993 | ||
2994 | /* Create thread to manage the client socket */ | |
2995 | ret = pthread_create(&health_thread, NULL, | |
2996 | thread_manage_health, (void *) NULL); | |
2997 | if (ret != 0) { | |
2998 | PERROR("pthread_create health"); | |
2999 | goto health_error; | |
3000 | } | |
3001 | ||
b8aa1682 JD |
3002 | /* Setup the dispatcher thread */ |
3003 | ret = pthread_create(&dispatcher_thread, NULL, | |
3004 | relay_thread_dispatcher, (void *) NULL); | |
3005 | if (ret != 0) { | |
3006 | PERROR("pthread_create dispatcher"); | |
3007 | goto exit_dispatcher; | |
3008 | } | |
3009 | ||
3010 | /* Setup the worker thread */ | |
3011 | ret = pthread_create(&worker_thread, NULL, | |
d3e2ba59 | 3012 | relay_thread_worker, (void *) relay_ctx); |
b8aa1682 JD |
3013 | if (ret != 0) { |
3014 | PERROR("pthread_create worker"); | |
3015 | goto exit_worker; | |
3016 | } | |
3017 | ||
3018 | /* Setup the listener thread */ | |
3019 | ret = pthread_create(&listener_thread, NULL, | |
3020 | relay_thread_listener, (void *) NULL); | |
3021 | if (ret != 0) { | |
3022 | PERROR("pthread_create listener"); | |
3023 | goto exit_listener; | |
3024 | } | |
3025 | ||
0b242f62 | 3026 | ret = live_start_threads(live_uri, relay_ctx); |
d3e2ba59 JD |
3027 | if (ret != 0) { |
3028 | ERR("Starting live viewer threads"); | |
50138f51 | 3029 | goto exit_live; |
d3e2ba59 JD |
3030 | } |
3031 | ||
50138f51 | 3032 | exit_live: |
b8aa1682 JD |
3033 | ret = pthread_join(listener_thread, &status); |
3034 | if (ret != 0) { | |
3035 | PERROR("pthread_join"); | |
3036 | goto error; /* join error, exit without cleanup */ | |
3037 | } | |
3038 | ||
50138f51 | 3039 | exit_listener: |
b8aa1682 JD |
3040 | ret = pthread_join(worker_thread, &status); |
3041 | if (ret != 0) { | |
3042 | PERROR("pthread_join"); | |
3043 | goto error; /* join error, exit without cleanup */ | |
3044 | } | |
3045 | ||
50138f51 | 3046 | exit_worker: |
b8aa1682 JD |
3047 | ret = pthread_join(dispatcher_thread, &status); |
3048 | if (ret != 0) { | |
3049 | PERROR("pthread_join"); | |
3050 | goto error; /* join error, exit without cleanup */ | |
3051 | } | |
42415026 | 3052 | |
50138f51 | 3053 | exit_dispatcher: |
65931c8b MD |
3054 | ret = pthread_join(health_thread, &status); |
3055 | if (ret != 0) { | |
3056 | PERROR("pthread_join health thread"); | |
3057 | goto error; /* join error, exit without cleanup */ | |
3058 | } | |
3059 | ||
bd11e201 MD |
3060 | /* |
3061 | * Stop live threads only after joining other threads. | |
3062 | */ | |
3063 | live_stop_threads(); | |
3064 | ||
65931c8b MD |
3065 | health_error: |
3066 | utils_close_pipe(health_quit_pipe); | |
3067 | ||
3068 | error_health_pipe: | |
55706a7d MD |
3069 | health_app_destroy(health_relayd); |
3070 | ||
3071 | exit_health_app_create: | |
92c6ca54 | 3072 | lttng_ht_destroy(viewer_streams_ht); |
d3e2ba59 JD |
3073 | |
3074 | exit_relay_ctx_viewer_streams: | |
3075 | lttng_ht_destroy(relay_streams_ht); | |
3076 | ||
3077 | exit_relay_ctx_streams: | |
3078 | lttng_ht_destroy(relay_ctx->sessions_ht); | |
3079 | ||
3080 | exit_relay_ctx_sessions: | |
3081 | free(relay_ctx); | |
b8aa1682 JD |
3082 | |
3083 | exit: | |
3084 | cleanup(); | |
3085 | if (!ret) { | |
3086 | exit(EXIT_SUCCESS); | |
3087 | } | |
a02de639 | 3088 | |
b8aa1682 JD |
3089 | error: |
3090 | exit(EXIT_FAILURE); | |
3091 | } |