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" |
2f8f53af | 65 | #include "viewer-stream.h" |
2a174661 DG |
66 | #include "session.h" |
67 | #include "stream.h" | |
b8aa1682 JD |
68 | |
69 | /* command line options */ | |
0f907de1 | 70 | char *opt_output_path; |
b5218ffb | 71 | static int opt_daemon, opt_background; |
3fd27398 MD |
72 | |
73 | /* | |
74 | * We need to wait for listener and live listener threads, as well as | |
75 | * health check thread, before being ready to signal readiness. | |
76 | */ | |
77 | #define NR_LTTNG_RELAY_READY 3 | |
78 | static int lttng_relay_ready = NR_LTTNG_RELAY_READY; | |
79 | static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */ | |
80 | static pid_t child_ppid; /* Internal parent PID use with daemonize. */ | |
81 | ||
095a4ae5 MD |
82 | static struct lttng_uri *control_uri; |
83 | static struct lttng_uri *data_uri; | |
d3e2ba59 | 84 | static struct lttng_uri *live_uri; |
b8aa1682 JD |
85 | |
86 | const char *progname; | |
b8aa1682 | 87 | |
65931c8b | 88 | const char *tracing_group_name = DEFAULT_TRACING_GROUP; |
cd60b05a JG |
89 | static int tracing_group_name_override; |
90 | ||
91 | const char * const config_section_name = "relayd"; | |
65931c8b | 92 | |
b8aa1682 JD |
93 | /* |
94 | * Quit pipe for all threads. This permits a single cancellation point | |
95 | * for all threads when receiving an event on the pipe. | |
96 | */ | |
0b242f62 | 97 | int thread_quit_pipe[2] = { -1, -1 }; |
b8aa1682 JD |
98 | |
99 | /* | |
100 | * This pipe is used to inform the worker thread that a command is queued and | |
101 | * ready to be processed. | |
102 | */ | |
103 | static int relay_cmd_pipe[2] = { -1, -1 }; | |
104 | ||
26c9d55e | 105 | /* Shared between threads */ |
b8aa1682 JD |
106 | static int dispatch_thread_exit; |
107 | ||
108 | static pthread_t listener_thread; | |
109 | static pthread_t dispatcher_thread; | |
110 | static pthread_t worker_thread; | |
65931c8b | 111 | static pthread_t health_thread; |
b8aa1682 | 112 | |
095a4ae5 | 113 | static uint64_t last_relay_stream_id; |
b8aa1682 JD |
114 | |
115 | /* | |
116 | * Relay command queue. | |
117 | * | |
118 | * The relay_thread_listener and relay_thread_dispatcher communicate with this | |
119 | * queue. | |
120 | */ | |
121 | static struct relay_cmd_queue relay_cmd_queue; | |
122 | ||
123 | /* buffer allocated at startup, used to store the trace data */ | |
095a4ae5 MD |
124 | static char *data_buffer; |
125 | static unsigned int data_buffer_size; | |
b8aa1682 | 126 | |
1c20f0e2 JD |
127 | /* We need those values for the file/dir creation. */ |
128 | static uid_t relayd_uid; | |
129 | static gid_t relayd_gid; | |
130 | ||
d3e2ba59 JD |
131 | /* Global relay stream hash table. */ |
132 | struct lttng_ht *relay_streams_ht; | |
133 | ||
92c6ca54 DG |
134 | /* Global relay viewer stream hash table. */ |
135 | struct lttng_ht *viewer_streams_ht; | |
136 | ||
0a6518b0 DG |
137 | /* Global hash table that stores relay index object. */ |
138 | struct lttng_ht *indexes_ht; | |
139 | ||
55706a7d | 140 | /* Relayd health monitoring */ |
eea7556c | 141 | struct health_app *health_relayd; |
55706a7d | 142 | |
cd60b05a JG |
143 | static struct option long_options[] = { |
144 | { "control-port", 1, 0, 'C', }, | |
145 | { "data-port", 1, 0, 'D', }, | |
8d5c808e | 146 | { "live-port", 1, 0, 'L', }, |
cd60b05a | 147 | { "daemonize", 0, 0, 'd', }, |
b5218ffb | 148 | { "background", 0, 0, 'b', }, |
cd60b05a JG |
149 | { "group", 1, 0, 'g', }, |
150 | { "help", 0, 0, 'h', }, | |
151 | { "output", 1, 0, 'o', }, | |
152 | { "verbose", 0, 0, 'v', }, | |
153 | { "config", 1, 0, 'f' }, | |
154 | { NULL, 0, 0, 0, }, | |
155 | }; | |
156 | ||
157 | static const char *config_ignore_options[] = { "help", "config" }; | |
158 | ||
b8aa1682 JD |
159 | /* |
160 | * usage function on stderr | |
161 | */ | |
162 | static | |
163 | void usage(void) | |
164 | { | |
165 | fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname); | |
994fa64f DG |
166 | fprintf(stderr, " -h, --help Display this usage.\n"); |
167 | fprintf(stderr, " -d, --daemonize Start as a daemon.\n"); | |
b5218ffb | 168 | fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n"); |
994fa64f DG |
169 | fprintf(stderr, " -C, --control-port URL Control port listening.\n"); |
170 | fprintf(stderr, " -D, --data-port URL Data port listening.\n"); | |
8d5c808e | 171 | fprintf(stderr, " -L, --live-port URL Live view port listening.\n"); |
994fa64f DG |
172 | fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n"); |
173 | fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n"); | |
65931c8b | 174 | fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n"); |
cd60b05a | 175 | fprintf(stderr, " -f --config Load daemon configuration file\n"); |
b8aa1682 JD |
176 | } |
177 | ||
cd60b05a JG |
178 | /* |
179 | * Take an option from the getopt output and set it in the right variable to be | |
180 | * used later. | |
181 | * | |
182 | * Return 0 on success else a negative value. | |
183 | */ | |
b8aa1682 | 184 | static |
cd60b05a | 185 | int set_option(int opt, const char *arg, const char *optname) |
b8aa1682 | 186 | { |
cd60b05a JG |
187 | int ret; |
188 | ||
189 | switch (opt) { | |
190 | case 0: | |
191 | fprintf(stderr, "option %s", optname); | |
192 | if (arg) { | |
193 | fprintf(stderr, " with arg %s\n", arg); | |
194 | } | |
195 | break; | |
196 | case 'C': | |
197 | ret = uri_parse(arg, &control_uri); | |
198 | if (ret < 0) { | |
199 | ERR("Invalid control URI specified"); | |
200 | goto end; | |
201 | } | |
202 | if (control_uri->port == 0) { | |
203 | control_uri->port = DEFAULT_NETWORK_CONTROL_PORT; | |
204 | } | |
205 | break; | |
206 | case 'D': | |
207 | ret = uri_parse(arg, &data_uri); | |
208 | if (ret < 0) { | |
209 | ERR("Invalid data URI specified"); | |
210 | goto end; | |
211 | } | |
212 | if (data_uri->port == 0) { | |
213 | data_uri->port = DEFAULT_NETWORK_DATA_PORT; | |
214 | } | |
215 | break; | |
8d5c808e AM |
216 | case 'L': |
217 | ret = uri_parse(arg, &live_uri); | |
218 | if (ret < 0) { | |
219 | ERR("Invalid live URI specified"); | |
220 | goto end; | |
221 | } | |
222 | if (live_uri->port == 0) { | |
223 | live_uri->port = DEFAULT_NETWORK_VIEWER_PORT; | |
224 | } | |
225 | break; | |
cd60b05a JG |
226 | case 'd': |
227 | opt_daemon = 1; | |
228 | break; | |
b5218ffb MD |
229 | case 'b': |
230 | opt_background = 1; | |
231 | break; | |
cd60b05a JG |
232 | case 'g': |
233 | tracing_group_name = strdup(arg); | |
234 | tracing_group_name_override = 1; | |
235 | break; | |
236 | case 'h': | |
237 | usage(); | |
238 | exit(EXIT_FAILURE); | |
239 | case 'o': | |
240 | ret = asprintf(&opt_output_path, "%s", arg); | |
241 | if (ret < 0) { | |
242 | ret = -errno; | |
243 | PERROR("asprintf opt_output_path"); | |
244 | goto end; | |
245 | } | |
246 | break; | |
247 | case 'v': | |
248 | /* Verbose level can increase using multiple -v */ | |
249 | if (arg) { | |
250 | lttng_opt_verbose = config_parse_value(arg); | |
251 | } else { | |
252 | lttng_opt_verbose += 1; | |
253 | } | |
254 | break; | |
255 | default: | |
256 | /* Unknown option or other error. | |
257 | * Error is printed by getopt, just return */ | |
258 | ret = -1; | |
259 | goto end; | |
260 | } | |
261 | ||
262 | /* All good. */ | |
263 | ret = 0; | |
264 | ||
265 | end: | |
266 | return ret; | |
267 | } | |
268 | ||
269 | /* | |
270 | * config_entry_handler_cb used to handle options read from a config file. | |
271 | * See config_entry_handler_cb comment in common/config/config.h for the | |
272 | * return value conventions. | |
273 | */ | |
274 | static | |
275 | int config_entry_handler(const struct config_entry *entry, void *unused) | |
276 | { | |
277 | int ret = 0, i; | |
278 | ||
279 | if (!entry || !entry->name || !entry->value) { | |
280 | ret = -EINVAL; | |
281 | goto end; | |
282 | } | |
283 | ||
284 | /* Check if the option is to be ignored */ | |
285 | for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) { | |
286 | if (!strcmp(entry->name, config_ignore_options[i])) { | |
287 | goto end; | |
288 | } | |
289 | } | |
290 | ||
291 | for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) { | |
292 | /* Ignore if entry name is not fully matched. */ | |
293 | if (strcmp(entry->name, long_options[i].name)) { | |
294 | continue; | |
295 | } | |
296 | ||
297 | /* | |
298 | * If the option takes no argument on the command line, we have to | |
299 | * check if the value is "true". We support non-zero numeric values, | |
300 | * true, on and yes. | |
301 | */ | |
302 | if (!long_options[i].has_arg) { | |
303 | ret = config_parse_value(entry->value); | |
304 | if (ret <= 0) { | |
305 | if (ret) { | |
306 | WARN("Invalid configuration value \"%s\" for option %s", | |
307 | entry->value, entry->name); | |
308 | } | |
309 | /* False, skip boolean config option. */ | |
310 | goto end; | |
311 | } | |
312 | } | |
313 | ||
314 | ret = set_option(long_options[i].val, entry->value, entry->name); | |
315 | goto end; | |
316 | } | |
317 | ||
318 | WARN("Unrecognized option \"%s\" in daemon configuration file.", | |
319 | entry->name); | |
320 | ||
321 | end: | |
322 | return ret; | |
323 | } | |
324 | ||
325 | static | |
326 | int set_options(int argc, char **argv) | |
327 | { | |
328 | int c, ret = 0, option_index = 0; | |
329 | int orig_optopt = optopt, orig_optind = optind; | |
330 | char *default_address, *optstring; | |
331 | const char *config_path = NULL; | |
332 | ||
333 | optstring = utils_generate_optstring(long_options, | |
334 | sizeof(long_options) / sizeof(struct option)); | |
335 | if (!optstring) { | |
336 | ret = -ENOMEM; | |
337 | goto exit; | |
338 | } | |
339 | ||
340 | /* Check for the --config option */ | |
341 | ||
342 | while ((c = getopt_long(argc, argv, optstring, long_options, | |
343 | &option_index)) != -1) { | |
344 | if (c == '?') { | |
345 | ret = -EINVAL; | |
346 | goto exit; | |
347 | } else if (c != 'f') { | |
348 | continue; | |
349 | } | |
350 | ||
351 | config_path = utils_expand_path(optarg); | |
352 | if (!config_path) { | |
353 | ERR("Failed to resolve path: %s", optarg); | |
354 | } | |
355 | } | |
356 | ||
357 | ret = config_get_section_entries(config_path, config_section_name, | |
358 | config_entry_handler, NULL); | |
359 | if (ret) { | |
360 | if (ret > 0) { | |
361 | ERR("Invalid configuration option at line %i", ret); | |
362 | ret = -1; | |
363 | } | |
364 | goto exit; | |
365 | } | |
b8aa1682 | 366 | |
cd60b05a JG |
367 | /* Reset getopt's global state */ |
368 | optopt = orig_optopt; | |
369 | optind = orig_optind; | |
b8aa1682 | 370 | while (1) { |
cd60b05a | 371 | c = getopt_long(argc, argv, optstring, long_options, &option_index); |
b8aa1682 JD |
372 | if (c == -1) { |
373 | break; | |
374 | } | |
375 | ||
cd60b05a JG |
376 | ret = set_option(c, optarg, long_options[option_index].name); |
377 | if (ret < 0) { | |
b8aa1682 JD |
378 | goto exit; |
379 | } | |
380 | } | |
381 | ||
382 | /* assign default values */ | |
383 | if (control_uri == NULL) { | |
384 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
385 | DEFAULT_NETWORK_CONTROL_PORT); | |
386 | if (ret < 0) { | |
387 | PERROR("asprintf default data address"); | |
388 | goto exit; | |
389 | } | |
390 | ||
391 | ret = uri_parse(default_address, &control_uri); | |
392 | free(default_address); | |
393 | if (ret < 0) { | |
394 | ERR("Invalid control URI specified"); | |
395 | goto exit; | |
396 | } | |
397 | } | |
398 | if (data_uri == NULL) { | |
399 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
400 | DEFAULT_NETWORK_DATA_PORT); | |
401 | if (ret < 0) { | |
402 | PERROR("asprintf default data address"); | |
403 | goto exit; | |
404 | } | |
405 | ||
406 | ret = uri_parse(default_address, &data_uri); | |
407 | free(default_address); | |
408 | if (ret < 0) { | |
409 | ERR("Invalid data URI specified"); | |
410 | goto exit; | |
411 | } | |
412 | } | |
d3e2ba59 JD |
413 | if (live_uri == NULL) { |
414 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
415 | DEFAULT_NETWORK_VIEWER_PORT); | |
416 | if (ret < 0) { | |
417 | PERROR("asprintf default viewer control address"); | |
418 | goto exit; | |
419 | } | |
420 | ||
421 | ret = uri_parse(default_address, &live_uri); | |
422 | free(default_address); | |
423 | if (ret < 0) { | |
424 | ERR("Invalid viewer control URI specified"); | |
425 | goto exit; | |
426 | } | |
427 | } | |
b8aa1682 JD |
428 | |
429 | exit: | |
cd60b05a | 430 | free(optstring); |
b8aa1682 JD |
431 | return ret; |
432 | } | |
433 | ||
434 | /* | |
435 | * Cleanup the daemon | |
436 | */ | |
437 | static | |
438 | void cleanup(void) | |
439 | { | |
b8aa1682 JD |
440 | DBG("Cleaning up"); |
441 | ||
095a4ae5 MD |
442 | /* free the dynamically allocated opt_output_path */ |
443 | free(opt_output_path); | |
444 | ||
a02de639 CB |
445 | /* Close thread quit pipes */ |
446 | utils_close_pipe(thread_quit_pipe); | |
447 | ||
710c1f73 DG |
448 | uri_free(control_uri); |
449 | uri_free(data_uri); | |
8d5c808e | 450 | /* Live URI is freed in the live thread. */ |
cd60b05a JG |
451 | |
452 | if (tracing_group_name_override) { | |
453 | free((void *) tracing_group_name); | |
454 | } | |
b8aa1682 JD |
455 | } |
456 | ||
457 | /* | |
458 | * Write to writable pipe used to notify a thread. | |
459 | */ | |
460 | static | |
461 | int notify_thread_pipe(int wpipe) | |
462 | { | |
6cd525e8 | 463 | ssize_t ret; |
b8aa1682 | 464 | |
6cd525e8 MD |
465 | ret = lttng_write(wpipe, "!", 1); |
466 | if (ret < 1) { | |
b8aa1682 JD |
467 | PERROR("write poll pipe"); |
468 | } | |
469 | ||
470 | return ret; | |
471 | } | |
472 | ||
65931c8b MD |
473 | static void notify_health_quit_pipe(int *pipe) |
474 | { | |
6cd525e8 | 475 | ssize_t ret; |
65931c8b | 476 | |
6cd525e8 MD |
477 | ret = lttng_write(pipe[1], "4", 1); |
478 | if (ret < 1) { | |
65931c8b MD |
479 | PERROR("write relay health quit"); |
480 | } | |
481 | } | |
482 | ||
b8aa1682 JD |
483 | /* |
484 | * Stop all threads by closing the thread quit pipe. | |
485 | */ | |
486 | static | |
487 | void stop_threads(void) | |
488 | { | |
489 | int ret; | |
490 | ||
491 | /* Stopping all threads */ | |
492 | DBG("Terminating all threads"); | |
493 | ret = notify_thread_pipe(thread_quit_pipe[1]); | |
494 | if (ret < 0) { | |
495 | ERR("write error on thread quit pipe"); | |
496 | } | |
497 | ||
65931c8b MD |
498 | notify_health_quit_pipe(health_quit_pipe); |
499 | ||
b8aa1682 | 500 | /* Dispatch thread */ |
26c9d55e | 501 | CMM_STORE_SHARED(dispatch_thread_exit, 1); |
b8aa1682 JD |
502 | futex_nto1_wake(&relay_cmd_queue.futex); |
503 | } | |
504 | ||
505 | /* | |
506 | * Signal handler for the daemon | |
507 | * | |
508 | * Simply stop all worker threads, leaving main() return gracefully after | |
509 | * joining all threads and calling cleanup(). | |
510 | */ | |
511 | static | |
512 | void sighandler(int sig) | |
513 | { | |
514 | switch (sig) { | |
515 | case SIGPIPE: | |
516 | DBG("SIGPIPE caught"); | |
517 | return; | |
518 | case SIGINT: | |
519 | DBG("SIGINT caught"); | |
520 | stop_threads(); | |
521 | break; | |
522 | case SIGTERM: | |
523 | DBG("SIGTERM caught"); | |
524 | stop_threads(); | |
525 | break; | |
3fd27398 MD |
526 | case SIGUSR1: |
527 | CMM_STORE_SHARED(recv_child_signal, 1); | |
528 | break; | |
b8aa1682 JD |
529 | default: |
530 | break; | |
531 | } | |
532 | } | |
533 | ||
534 | /* | |
535 | * Setup signal handler for : | |
536 | * SIGINT, SIGTERM, SIGPIPE | |
537 | */ | |
538 | static | |
539 | int set_signal_handler(void) | |
540 | { | |
541 | int ret = 0; | |
542 | struct sigaction sa; | |
543 | sigset_t sigset; | |
544 | ||
545 | if ((ret = sigemptyset(&sigset)) < 0) { | |
546 | PERROR("sigemptyset"); | |
547 | return ret; | |
548 | } | |
549 | ||
550 | sa.sa_handler = sighandler; | |
551 | sa.sa_mask = sigset; | |
552 | sa.sa_flags = 0; | |
553 | if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) { | |
554 | PERROR("sigaction"); | |
555 | return ret; | |
556 | } | |
557 | ||
558 | if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) { | |
559 | PERROR("sigaction"); | |
560 | return ret; | |
561 | } | |
562 | ||
563 | if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) { | |
564 | PERROR("sigaction"); | |
565 | return ret; | |
566 | } | |
567 | ||
3fd27398 MD |
568 | if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) { |
569 | PERROR("sigaction"); | |
570 | return ret; | |
571 | } | |
572 | ||
573 | DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT"); | |
b8aa1682 JD |
574 | |
575 | return ret; | |
576 | } | |
577 | ||
3fd27398 MD |
578 | void lttng_relay_notify_ready(void) |
579 | { | |
580 | /* Notify the parent of the fork() process that we are ready. */ | |
581 | if (opt_daemon || opt_background) { | |
582 | if (uatomic_sub_return(<tng_relay_ready, 1) == 0) { | |
583 | kill(child_ppid, SIGUSR1); | |
584 | } | |
585 | } | |
586 | } | |
587 | ||
b8aa1682 JD |
588 | /* |
589 | * Init thread quit pipe. | |
590 | * | |
591 | * Return -1 on error or 0 if all pipes are created. | |
592 | */ | |
593 | static | |
594 | int init_thread_quit_pipe(void) | |
595 | { | |
a02de639 | 596 | int ret; |
b8aa1682 | 597 | |
a02de639 | 598 | ret = utils_create_pipe_cloexec(thread_quit_pipe); |
b8aa1682 | 599 | |
b8aa1682 JD |
600 | return ret; |
601 | } | |
602 | ||
603 | /* | |
604 | * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set. | |
605 | */ | |
606 | static | |
607 | int create_thread_poll_set(struct lttng_poll_event *events, int size) | |
608 | { | |
609 | int ret; | |
610 | ||
611 | if (events == NULL || size == 0) { | |
612 | ret = -1; | |
613 | goto error; | |
614 | } | |
615 | ||
616 | ret = lttng_poll_create(events, size, LTTNG_CLOEXEC); | |
617 | if (ret < 0) { | |
618 | goto error; | |
619 | } | |
620 | ||
621 | /* Add quit pipe */ | |
c7759e6a | 622 | ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR); |
b8aa1682 JD |
623 | if (ret < 0) { |
624 | goto error; | |
625 | } | |
626 | ||
627 | return 0; | |
628 | ||
629 | error: | |
630 | return ret; | |
631 | } | |
632 | ||
633 | /* | |
634 | * Check if the thread quit pipe was triggered. | |
635 | * | |
636 | * Return 1 if it was triggered else 0; | |
637 | */ | |
638 | static | |
639 | int check_thread_quit_pipe(int fd, uint32_t events) | |
640 | { | |
641 | if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) { | |
642 | return 1; | |
643 | } | |
644 | ||
645 | return 0; | |
646 | } | |
647 | ||
648 | /* | |
649 | * Create and init socket from uri. | |
650 | */ | |
651 | static | |
652 | struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri) | |
653 | { | |
654 | int ret; | |
655 | struct lttcomm_sock *sock = NULL; | |
656 | ||
657 | sock = lttcomm_alloc_sock_from_uri(uri); | |
658 | if (sock == NULL) { | |
659 | ERR("Allocating socket"); | |
660 | goto error; | |
661 | } | |
662 | ||
663 | ret = lttcomm_create_sock(sock); | |
664 | if (ret < 0) { | |
665 | goto error; | |
666 | } | |
667 | DBG("Listening on sock %d", sock->fd); | |
668 | ||
669 | ret = sock->ops->bind(sock); | |
670 | if (ret < 0) { | |
671 | goto error; | |
672 | } | |
673 | ||
674 | ret = sock->ops->listen(sock, -1); | |
675 | if (ret < 0) { | |
676 | goto error; | |
677 | ||
678 | } | |
679 | ||
680 | return sock; | |
681 | ||
682 | error: | |
683 | if (sock) { | |
684 | lttcomm_destroy_sock(sock); | |
685 | } | |
686 | return NULL; | |
687 | } | |
688 | ||
173af62f DG |
689 | /* |
690 | * Return nonzero if stream needs to be closed. | |
691 | */ | |
692 | static | |
693 | int close_stream_check(struct relay_stream *stream) | |
694 | { | |
173af62f | 695 | if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) { |
f7079f67 DG |
696 | /* |
697 | * We are about to close the stream so set the data pending flag to 1 | |
698 | * which will make the end data pending command skip the stream which | |
699 | * is now closed and ready. Note that after proceeding to a file close, | |
700 | * the written file is ready for reading. | |
701 | */ | |
702 | stream->data_pending_check_done = 1; | |
173af62f DG |
703 | return 1; |
704 | } | |
705 | return 0; | |
706 | } | |
707 | ||
2a174661 DG |
708 | static void try_close_stream(struct relay_session *session, |
709 | struct relay_stream *stream) | |
710 | { | |
711 | int ret; | |
712 | struct ctf_trace *ctf_trace; | |
713 | ||
714 | assert(session); | |
715 | assert(stream); | |
716 | ||
717 | if (!close_stream_check(stream)) { | |
718 | /* Can't close it, not ready for that. */ | |
719 | goto end; | |
720 | } | |
721 | ||
722 | ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, | |
723 | stream->path_name); | |
724 | assert(ctf_trace); | |
725 | ||
726 | pthread_mutex_lock(&session->viewer_ready_lock); | |
727 | ctf_trace->invalid_flag = 1; | |
728 | pthread_mutex_unlock(&session->viewer_ready_lock); | |
729 | ||
730 | ret = stream_close(session, stream); | |
731 | if (!ret) { | |
732 | /* Already close thus the ctf trace is being or has been destroyed. */ | |
733 | goto end; | |
734 | } | |
735 | ||
736 | ctf_trace_try_destroy(session, ctf_trace); | |
737 | ||
738 | end: | |
739 | return; | |
740 | } | |
741 | ||
b8aa1682 JD |
742 | /* |
743 | * This thread manages the listening for new connections on the network | |
744 | */ | |
745 | static | |
746 | void *relay_thread_listener(void *data) | |
747 | { | |
095a4ae5 | 748 | int i, ret, pollfd, err = -1; |
b8aa1682 JD |
749 | int val = 1; |
750 | uint32_t revents, nb_fd; | |
751 | struct lttng_poll_event events; | |
752 | struct lttcomm_sock *control_sock, *data_sock; | |
753 | ||
b8aa1682 JD |
754 | DBG("[thread] Relay listener started"); |
755 | ||
55706a7d MD |
756 | health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER); |
757 | ||
f385ae0a MD |
758 | health_code_update(); |
759 | ||
b8aa1682 JD |
760 | control_sock = relay_init_sock(control_uri); |
761 | if (!control_sock) { | |
095a4ae5 | 762 | goto error_sock_control; |
b8aa1682 JD |
763 | } |
764 | ||
765 | data_sock = relay_init_sock(data_uri); | |
766 | if (!data_sock) { | |
095a4ae5 | 767 | goto error_sock_relay; |
b8aa1682 JD |
768 | } |
769 | ||
770 | /* | |
771 | * Pass 3 as size here for the thread quit pipe, control and data socket. | |
772 | */ | |
773 | ret = create_thread_poll_set(&events, 3); | |
774 | if (ret < 0) { | |
775 | goto error_create_poll; | |
776 | } | |
777 | ||
778 | /* Add the control socket */ | |
779 | ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP); | |
780 | if (ret < 0) { | |
781 | goto error_poll_add; | |
782 | } | |
783 | ||
784 | /* Add the data socket */ | |
785 | ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP); | |
786 | if (ret < 0) { | |
787 | goto error_poll_add; | |
788 | } | |
789 | ||
3fd27398 MD |
790 | lttng_relay_notify_ready(); |
791 | ||
9b5e0863 MD |
792 | if (testpoint(relayd_thread_listener)) { |
793 | goto error_testpoint; | |
794 | } | |
795 | ||
b8aa1682 | 796 | while (1) { |
f385ae0a MD |
797 | health_code_update(); |
798 | ||
b8aa1682 JD |
799 | DBG("Listener accepting connections"); |
800 | ||
b8aa1682 | 801 | restart: |
f385ae0a | 802 | health_poll_entry(); |
b8aa1682 | 803 | ret = lttng_poll_wait(&events, -1); |
f385ae0a | 804 | health_poll_exit(); |
b8aa1682 JD |
805 | if (ret < 0) { |
806 | /* | |
807 | * Restart interrupted system call. | |
808 | */ | |
809 | if (errno == EINTR) { | |
810 | goto restart; | |
811 | } | |
812 | goto error; | |
813 | } | |
814 | ||
0d9c5d77 DG |
815 | nb_fd = ret; |
816 | ||
b8aa1682 JD |
817 | DBG("Relay new connection received"); |
818 | for (i = 0; i < nb_fd; i++) { | |
f385ae0a MD |
819 | health_code_update(); |
820 | ||
b8aa1682 JD |
821 | /* Fetch once the poll data */ |
822 | revents = LTTNG_POLL_GETEV(&events, i); | |
823 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
824 | ||
825 | /* Thread quit pipe has been closed. Killing thread. */ | |
826 | ret = check_thread_quit_pipe(pollfd, revents); | |
827 | if (ret) { | |
095a4ae5 MD |
828 | err = 0; |
829 | goto exit; | |
b8aa1682 JD |
830 | } |
831 | ||
832 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
833 | ERR("socket poll error"); | |
834 | goto error; | |
835 | } else if (revents & LPOLLIN) { | |
4b7f17b2 MD |
836 | /* |
837 | * Get allocated in this thread, | |
838 | * enqueued to a global queue, dequeued | |
839 | * and freed in the worker thread. | |
840 | */ | |
841 | struct relay_command *relay_cmd; | |
842 | struct lttcomm_sock *newsock; | |
b8aa1682 JD |
843 | |
844 | relay_cmd = zmalloc(sizeof(struct relay_command)); | |
845 | if (relay_cmd == NULL) { | |
846 | PERROR("relay command zmalloc"); | |
847 | goto error; | |
848 | } | |
849 | ||
850 | if (pollfd == data_sock->fd) { | |
851 | newsock = data_sock->ops->accept(data_sock); | |
4b7f17b2 | 852 | if (!newsock) { |
b8aa1682 | 853 | PERROR("accepting data sock"); |
4b7f17b2 | 854 | free(relay_cmd); |
b8aa1682 JD |
855 | goto error; |
856 | } | |
857 | relay_cmd->type = RELAY_DATA; | |
858 | DBG("Relay data connection accepted, socket %d", newsock->fd); | |
4b7f17b2 MD |
859 | } else { |
860 | assert(pollfd == control_sock->fd); | |
b8aa1682 | 861 | newsock = control_sock->ops->accept(control_sock); |
4b7f17b2 | 862 | if (!newsock) { |
b8aa1682 | 863 | PERROR("accepting control sock"); |
4b7f17b2 | 864 | free(relay_cmd); |
b8aa1682 JD |
865 | goto error; |
866 | } | |
867 | relay_cmd->type = RELAY_CONTROL; | |
868 | DBG("Relay control connection accepted, socket %d", newsock->fd); | |
869 | } | |
870 | ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, | |
871 | &val, sizeof(int)); | |
872 | if (ret < 0) { | |
873 | PERROR("setsockopt inet"); | |
4b7f17b2 MD |
874 | lttcomm_destroy_sock(newsock); |
875 | free(relay_cmd); | |
b8aa1682 JD |
876 | goto error; |
877 | } | |
878 | relay_cmd->sock = newsock; | |
879 | /* | |
880 | * Lock free enqueue the request. | |
881 | */ | |
882 | cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node); | |
883 | ||
884 | /* | |
885 | * Wake the dispatch queue futex. Implicit memory | |
886 | * barrier with the exchange in cds_wfq_enqueue. | |
887 | */ | |
888 | futex_nto1_wake(&relay_cmd_queue.futex); | |
889 | } | |
890 | } | |
891 | } | |
892 | ||
095a4ae5 | 893 | exit: |
b8aa1682 JD |
894 | error: |
895 | error_poll_add: | |
9b5e0863 | 896 | error_testpoint: |
b8aa1682 JD |
897 | lttng_poll_clean(&events); |
898 | error_create_poll: | |
095a4ae5 MD |
899 | if (data_sock->fd >= 0) { |
900 | ret = data_sock->ops->close(data_sock); | |
b8aa1682 JD |
901 | if (ret) { |
902 | PERROR("close"); | |
903 | } | |
b8aa1682 | 904 | } |
095a4ae5 MD |
905 | lttcomm_destroy_sock(data_sock); |
906 | error_sock_relay: | |
907 | if (control_sock->fd >= 0) { | |
908 | ret = control_sock->ops->close(control_sock); | |
b8aa1682 JD |
909 | if (ret) { |
910 | PERROR("close"); | |
911 | } | |
b8aa1682 | 912 | } |
095a4ae5 MD |
913 | lttcomm_destroy_sock(control_sock); |
914 | error_sock_control: | |
915 | if (err) { | |
f385ae0a MD |
916 | health_error(); |
917 | ERR("Health error occurred in %s", __func__); | |
095a4ae5 | 918 | } |
55706a7d | 919 | health_unregister(health_relayd); |
b8aa1682 JD |
920 | DBG("Relay listener thread cleanup complete"); |
921 | stop_threads(); | |
b8aa1682 JD |
922 | return NULL; |
923 | } | |
924 | ||
925 | /* | |
926 | * This thread manages the dispatching of the requests to worker threads | |
927 | */ | |
928 | static | |
929 | void *relay_thread_dispatcher(void *data) | |
930 | { | |
6cd525e8 MD |
931 | int err = -1; |
932 | ssize_t ret; | |
b8aa1682 JD |
933 | struct cds_wfq_node *node; |
934 | struct relay_command *relay_cmd = NULL; | |
935 | ||
936 | DBG("[thread] Relay dispatcher started"); | |
937 | ||
55706a7d MD |
938 | health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER); |
939 | ||
9b5e0863 MD |
940 | if (testpoint(relayd_thread_dispatcher)) { |
941 | goto error_testpoint; | |
942 | } | |
943 | ||
f385ae0a MD |
944 | health_code_update(); |
945 | ||
26c9d55e | 946 | while (!CMM_LOAD_SHARED(dispatch_thread_exit)) { |
f385ae0a MD |
947 | health_code_update(); |
948 | ||
b8aa1682 JD |
949 | /* Atomically prepare the queue futex */ |
950 | futex_nto1_prepare(&relay_cmd_queue.futex); | |
951 | ||
952 | do { | |
f385ae0a MD |
953 | health_code_update(); |
954 | ||
b8aa1682 JD |
955 | /* Dequeue commands */ |
956 | node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue); | |
957 | if (node == NULL) { | |
958 | DBG("Woken up but nothing in the relay command queue"); | |
959 | /* Continue thread execution */ | |
960 | break; | |
961 | } | |
962 | ||
963 | relay_cmd = caa_container_of(node, struct relay_command, node); | |
964 | DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd); | |
965 | ||
966 | /* | |
967 | * Inform worker thread of the new request. This | |
968 | * call is blocking so we can be assured that the data will be read | |
969 | * at some point in time or wait to the end of the world :) | |
970 | */ | |
6cd525e8 MD |
971 | ret = lttng_write(relay_cmd_pipe[1], relay_cmd, |
972 | sizeof(struct relay_command)); | |
b8aa1682 | 973 | free(relay_cmd); |
6cd525e8 | 974 | if (ret < sizeof(struct relay_command)) { |
b8aa1682 JD |
975 | PERROR("write cmd pipe"); |
976 | goto error; | |
977 | } | |
978 | } while (node != NULL); | |
979 | ||
980 | /* Futex wait on queue. Blocking call on futex() */ | |
f385ae0a | 981 | health_poll_entry(); |
b8aa1682 | 982 | futex_nto1_wait(&relay_cmd_queue.futex); |
f385ae0a | 983 | health_poll_exit(); |
b8aa1682 JD |
984 | } |
985 | ||
f385ae0a MD |
986 | /* Normal exit, no error */ |
987 | err = 0; | |
988 | ||
b8aa1682 | 989 | error: |
9b5e0863 | 990 | error_testpoint: |
f385ae0a MD |
991 | if (err) { |
992 | health_error(); | |
993 | ERR("Health error occurred in %s", __func__); | |
994 | } | |
55706a7d | 995 | health_unregister(health_relayd); |
b8aa1682 JD |
996 | DBG("Dispatch thread dying"); |
997 | stop_threads(); | |
998 | return NULL; | |
999 | } | |
1000 | ||
2a174661 | 1001 | static void try_close_streams(struct relay_session *session) |
d3e2ba59 | 1002 | { |
2a174661 | 1003 | struct ctf_trace *ctf_trace; |
94d49140 JD |
1004 | struct lttng_ht_iter iter; |
1005 | ||
2a174661 | 1006 | assert(session); |
94d49140 | 1007 | |
2a174661 DG |
1008 | pthread_mutex_lock(&session->viewer_ready_lock); |
1009 | rcu_read_lock(); | |
1010 | cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace, | |
1011 | node.node) { | |
1012 | struct relay_stream *stream; | |
94d49140 | 1013 | |
2a174661 DG |
1014 | /* Close streams. */ |
1015 | cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) { | |
1016 | stream_close(session, stream); | |
94d49140 | 1017 | } |
94d49140 | 1018 | |
2a174661 DG |
1019 | ctf_trace->invalid_flag = 1; |
1020 | ctf_trace_try_destroy(session, ctf_trace); | |
157df586 | 1021 | } |
2a174661 DG |
1022 | rcu_read_unlock(); |
1023 | pthread_mutex_unlock(&session->viewer_ready_lock); | |
94d49140 JD |
1024 | } |
1025 | ||
b8aa1682 | 1026 | /* |
2a174661 | 1027 | * Try to destroy a session within a connection. |
b8aa1682 JD |
1028 | */ |
1029 | static | |
d3e2ba59 JD |
1030 | void relay_delete_session(struct relay_command *cmd, |
1031 | struct lttng_ht *sessions_ht) | |
b8aa1682 | 1032 | { |
2a174661 DG |
1033 | assert(cmd); |
1034 | assert(sessions_ht); | |
b8aa1682 | 1035 | |
2a174661 DG |
1036 | /* Indicate that this session can be destroyed from now on. */ |
1037 | cmd->session->close_flag = 1; | |
b8aa1682 | 1038 | |
2a174661 | 1039 | try_close_streams(cmd->session); |
5b6d8097 | 1040 | |
2a174661 DG |
1041 | /* |
1042 | * This will try to delete and destroy the session if no viewer is attached | |
1043 | * to it meaning the refcount is down to zero. | |
1044 | */ | |
1045 | session_try_destroy(sessions_ht, cmd->session); | |
b8aa1682 JD |
1046 | } |
1047 | ||
1c20f0e2 JD |
1048 | /* |
1049 | * Copy index data from the control port to a given index object. | |
1050 | */ | |
1051 | static void copy_index_control_data(struct relay_index *index, | |
1052 | struct lttcomm_relayd_index *data) | |
1053 | { | |
1054 | assert(index); | |
1055 | assert(data); | |
1056 | ||
1057 | /* | |
1058 | * The index on disk is encoded in big endian, so we don't need to convert | |
1059 | * the data received on the network. The data_offset value is NEVER | |
1060 | * modified here and is updated by the data thread. | |
1061 | */ | |
1062 | index->index_data.packet_size = data->packet_size; | |
1063 | index->index_data.content_size = data->content_size; | |
1064 | index->index_data.timestamp_begin = data->timestamp_begin; | |
1065 | index->index_data.timestamp_end = data->timestamp_end; | |
1066 | index->index_data.events_discarded = data->events_discarded; | |
1067 | index->index_data.stream_id = data->stream_id; | |
1068 | } | |
1069 | ||
c5b6f4f0 DG |
1070 | /* |
1071 | * Handle the RELAYD_CREATE_SESSION command. | |
1072 | * | |
1073 | * On success, send back the session id or else return a negative value. | |
1074 | */ | |
1075 | static | |
1076 | int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 JD |
1077 | struct relay_command *cmd, |
1078 | struct lttng_ht *sessions_ht) | |
c5b6f4f0 DG |
1079 | { |
1080 | int ret = 0, send_ret; | |
1081 | struct relay_session *session; | |
1082 | struct lttcomm_relayd_status_session reply; | |
1083 | ||
1084 | assert(recv_hdr); | |
1085 | assert(cmd); | |
1086 | ||
1087 | memset(&reply, 0, sizeof(reply)); | |
1088 | ||
2a174661 DG |
1089 | session = session_create(); |
1090 | if (!session) { | |
c5b6f4f0 DG |
1091 | ret = -1; |
1092 | goto error; | |
1093 | } | |
7d2f7452 DG |
1094 | session->minor = cmd->minor; |
1095 | session->major = cmd->major; | |
2a174661 | 1096 | cmd->session_id = session->id; |
c5b6f4f0 DG |
1097 | cmd->session = session; |
1098 | ||
1099 | reply.session_id = htobe64(session->id); | |
1100 | ||
d3e2ba59 | 1101 | switch (cmd->minor) { |
2a174661 DG |
1102 | case 1: |
1103 | case 2: | |
1104 | case 3: | |
1105 | break; | |
1106 | case 4: /* LTTng sessiond 2.4 */ | |
1107 | default: | |
1108 | ret = cmd_create_session_2_4(cmd, session); | |
1109 | break; | |
d3e2ba59 JD |
1110 | } |
1111 | ||
2a174661 | 1112 | lttng_ht_add_unique_u64(sessions_ht, &session->session_n); |
c5b6f4f0 DG |
1113 | DBG("Created session %" PRIu64, session->id); |
1114 | ||
1115 | error: | |
1116 | if (ret < 0) { | |
1117 | reply.ret_code = htobe32(LTTNG_ERR_FATAL); | |
1118 | } else { | |
1119 | reply.ret_code = htobe32(LTTNG_OK); | |
1120 | } | |
1121 | ||
1122 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1123 | if (send_ret < 0) { | |
1124 | ERR("Relayd sending session id"); | |
4169f5ad | 1125 | ret = send_ret; |
c5b6f4f0 DG |
1126 | } |
1127 | ||
1128 | return ret; | |
1129 | } | |
1130 | ||
a4baae1b JD |
1131 | /* |
1132 | * When we have received all the streams and the metadata for a channel, | |
1133 | * we make them visible to the viewer threads. | |
1134 | */ | |
1135 | static | |
1136 | void set_viewer_ready_flag(struct relay_command *cmd) | |
1137 | { | |
2a174661 | 1138 | struct relay_stream *stream, *tmp_stream; |
a4baae1b | 1139 | |
80e8027a | 1140 | pthread_mutex_lock(&cmd->session->viewer_ready_lock); |
2a174661 DG |
1141 | cds_list_for_each_entry_safe(stream, tmp_stream, &cmd->recv_head, |
1142 | recv_list) { | |
a4baae1b | 1143 | stream->viewer_ready = 1; |
2a174661 | 1144 | cds_list_del(&stream->recv_list); |
a4baae1b | 1145 | } |
80e8027a | 1146 | pthread_mutex_unlock(&cmd->session->viewer_ready_lock); |
a4baae1b JD |
1147 | return; |
1148 | } | |
1149 | ||
1150 | /* | |
1151 | * Add a recv handle node to the connection recv list with the given stream | |
1152 | * handle. A new node is allocated thus must be freed when the node is deleted | |
1153 | * from the list. | |
1154 | */ | |
2a174661 | 1155 | static void queue_stream(struct relay_stream *stream, struct relay_command *cmd) |
a4baae1b | 1156 | { |
a4baae1b | 1157 | assert(cmd); |
2a174661 | 1158 | assert(stream); |
a4baae1b | 1159 | |
2a174661 | 1160 | cds_list_add(&stream->recv_list, &cmd->recv_head); |
a4baae1b JD |
1161 | } |
1162 | ||
b8aa1682 JD |
1163 | /* |
1164 | * relay_add_stream: allocate a new stream for a session | |
1165 | */ | |
1166 | static | |
1167 | int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1168 | struct relay_command *cmd, struct lttng_ht *sessions_ht) |
b8aa1682 | 1169 | { |
2a174661 | 1170 | int ret, send_ret; |
b8aa1682 | 1171 | struct relay_session *session = cmd->session; |
b8aa1682 JD |
1172 | struct relay_stream *stream = NULL; |
1173 | struct lttcomm_relayd_status_stream reply; | |
2a174661 | 1174 | struct ctf_trace *trace; |
b8aa1682 | 1175 | |
c5b6f4f0 | 1176 | if (!session || cmd->version_check_done == 0) { |
b8aa1682 JD |
1177 | ERR("Trying to add a stream before version check"); |
1178 | ret = -1; | |
1179 | goto end_no_session; | |
1180 | } | |
1181 | ||
b8aa1682 JD |
1182 | stream = zmalloc(sizeof(struct relay_stream)); |
1183 | if (stream == NULL) { | |
1184 | PERROR("relay stream zmalloc"); | |
1185 | ret = -1; | |
1186 | goto end_no_session; | |
1187 | } | |
1188 | ||
0f907de1 JD |
1189 | switch (cmd->minor) { |
1190 | case 1: /* LTTng sessiond 2.1 */ | |
1191 | ret = cmd_recv_stream_2_1(cmd, stream); | |
1192 | break; | |
1193 | case 2: /* LTTng sessiond 2.2 */ | |
1194 | default: | |
1195 | ret = cmd_recv_stream_2_2(cmd, stream); | |
1196 | break; | |
1197 | } | |
1198 | if (ret < 0) { | |
1199 | goto err_free_stream; | |
1200 | } | |
1201 | ||
9d1bbf21 | 1202 | rcu_read_lock(); |
b8aa1682 | 1203 | stream->stream_handle = ++last_relay_stream_id; |
173af62f | 1204 | stream->prev_seq = -1ULL; |
2a174661 | 1205 | stream->session_id = session->id; |
1c20f0e2 | 1206 | stream->index_fd = -1; |
d3e2ba59 | 1207 | stream->read_index_fd = -1; |
2a174661 | 1208 | lttng_ht_node_init_u64(&stream->node, stream->stream_handle); |
d3e2ba59 | 1209 | pthread_mutex_init(&stream->lock, NULL); |
b8aa1682 | 1210 | |
0f907de1 | 1211 | ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG); |
b8aa1682 | 1212 | if (ret < 0) { |
b8aa1682 JD |
1213 | ERR("relay creating output directory"); |
1214 | goto end; | |
1215 | } | |
1216 | ||
be96a7d1 DG |
1217 | /* |
1218 | * No need to use run_as API here because whatever we receives, the relayd | |
1219 | * uses its own credentials for the stream files. | |
1220 | */ | |
0f907de1 | 1221 | ret = utils_create_stream_file(stream->path_name, stream->channel_name, |
1c20f0e2 | 1222 | stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL); |
b8aa1682 | 1223 | if (ret < 0) { |
0f907de1 | 1224 | ERR("Create output file"); |
b8aa1682 JD |
1225 | goto end; |
1226 | } | |
b8aa1682 | 1227 | stream->fd = ret; |
0f907de1 JD |
1228 | if (stream->tracefile_size) { |
1229 | DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name); | |
1230 | } else { | |
1231 | DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name); | |
1232 | } | |
b8aa1682 | 1233 | |
2a174661 DG |
1234 | trace = ctf_trace_find_by_path(session->ctf_traces_ht, stream->path_name); |
1235 | if (!trace) { | |
1236 | trace = ctf_trace_create(stream->path_name); | |
1237 | if (!trace) { | |
d3e2ba59 JD |
1238 | ret = -1; |
1239 | goto end; | |
1240 | } | |
2a174661 DG |
1241 | ctf_trace_add(session->ctf_traces_ht, trace); |
1242 | } | |
1243 | ctf_trace_get_ref(trace); | |
1244 | ||
1245 | if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) { | |
1246 | stream->metadata_flag = 1; | |
1247 | /* Assign quick reference to the metadata stream in the trace. */ | |
1248 | trace->metadata_stream = stream; | |
d3e2ba59 | 1249 | } |
d3e2ba59 | 1250 | |
a4baae1b | 1251 | /* |
2a174661 DG |
1252 | * Add the stream in the recv list of the connection. Once the end stream |
1253 | * message is received, this list is emptied and streams are set with the | |
1254 | * viewer ready flag. | |
a4baae1b | 1255 | */ |
2a174661 | 1256 | queue_stream(stream, cmd); |
a4baae1b | 1257 | |
2a174661 DG |
1258 | /* |
1259 | * Both in the ctf_trace object and the global stream ht since the data | |
1260 | * side of the relayd does not have the concept of session. | |
1261 | */ | |
1262 | lttng_ht_add_unique_u64(relay_streams_ht, &stream->node); | |
1263 | cds_list_add_tail(&stream->trace_list, &trace->stream_list); | |
b8aa1682 | 1264 | |
87b576ec | 1265 | session->stream_count++; |
d3e2ba59 | 1266 | |
1c20f0e2 JD |
1267 | DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name, |
1268 | stream->stream_handle); | |
b8aa1682 JD |
1269 | |
1270 | end: | |
5af40280 | 1271 | reply.handle = htobe64(stream->stream_handle); |
b8aa1682 JD |
1272 | /* send the session id to the client or a negative return code on error */ |
1273 | if (ret < 0) { | |
f73fabfd | 1274 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
5af40280 CB |
1275 | /* stream was not properly added to the ht, so free it */ |
1276 | free(stream); | |
b8aa1682 | 1277 | } else { |
f73fabfd | 1278 | reply.ret_code = htobe32(LTTNG_OK); |
b8aa1682 | 1279 | } |
5af40280 | 1280 | |
b8aa1682 JD |
1281 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, |
1282 | sizeof(struct lttcomm_relayd_status_stream), 0); | |
1283 | if (send_ret < 0) { | |
1284 | ERR("Relay sending stream id"); | |
4169f5ad | 1285 | ret = send_ret; |
b8aa1682 | 1286 | } |
9d1bbf21 | 1287 | rcu_read_unlock(); |
b8aa1682 JD |
1288 | |
1289 | end_no_session: | |
1290 | return ret; | |
0f907de1 JD |
1291 | |
1292 | err_free_stream: | |
1293 | free(stream->path_name); | |
1294 | free(stream->channel_name); | |
1295 | free(stream); | |
1296 | return ret; | |
b8aa1682 JD |
1297 | } |
1298 | ||
173af62f DG |
1299 | /* |
1300 | * relay_close_stream: close a specific stream | |
1301 | */ | |
1302 | static | |
1303 | int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, | |
92c6ca54 | 1304 | struct relay_command *cmd) |
173af62f | 1305 | { |
94d49140 | 1306 | int ret, send_ret; |
173af62f DG |
1307 | struct relay_session *session = cmd->session; |
1308 | struct lttcomm_relayd_close_stream stream_info; | |
1309 | struct lttcomm_relayd_generic_reply reply; | |
1310 | struct relay_stream *stream; | |
173af62f DG |
1311 | |
1312 | DBG("Close stream received"); | |
1313 | ||
c5b6f4f0 | 1314 | if (!session || cmd->version_check_done == 0) { |
173af62f DG |
1315 | ERR("Trying to close a stream before version check"); |
1316 | ret = -1; | |
1317 | goto end_no_session; | |
1318 | } | |
1319 | ||
1320 | ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info, | |
7c5aef62 | 1321 | sizeof(struct lttcomm_relayd_close_stream), 0); |
173af62f | 1322 | if (ret < sizeof(struct lttcomm_relayd_close_stream)) { |
a6cd2b97 DG |
1323 | if (ret == 0) { |
1324 | /* Orderly shutdown. Not necessary to print an error. */ | |
1325 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1326 | } else { | |
1327 | ERR("Relay didn't receive valid add_stream struct size : %d", ret); | |
1328 | } | |
173af62f DG |
1329 | ret = -1; |
1330 | goto end_no_session; | |
1331 | } | |
1332 | ||
1333 | rcu_read_lock(); | |
2a174661 DG |
1334 | stream = stream_find_by_id(relay_streams_ht, |
1335 | be64toh(stream_info.stream_id)); | |
173af62f DG |
1336 | if (!stream) { |
1337 | ret = -1; | |
1338 | goto end_unlock; | |
1339 | } | |
1340 | ||
8e2583a4 | 1341 | stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num); |
173af62f | 1342 | stream->close_flag = 1; |
87b576ec JD |
1343 | session->stream_count--; |
1344 | assert(session->stream_count >= 0); | |
173af62f | 1345 | |
2a174661 DG |
1346 | /* Check if we can close it or else the data will do it. */ |
1347 | try_close_stream(session, stream); | |
173af62f DG |
1348 | |
1349 | end_unlock: | |
1350 | rcu_read_unlock(); | |
1351 | ||
1352 | if (ret < 0) { | |
f73fabfd | 1353 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
173af62f | 1354 | } else { |
f73fabfd | 1355 | reply.ret_code = htobe32(LTTNG_OK); |
173af62f DG |
1356 | } |
1357 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1358 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1359 | if (send_ret < 0) { | |
1360 | ERR("Relay sending stream id"); | |
4169f5ad | 1361 | ret = send_ret; |
173af62f DG |
1362 | } |
1363 | ||
1364 | end_no_session: | |
1365 | return ret; | |
1366 | } | |
1367 | ||
b8aa1682 JD |
1368 | /* |
1369 | * relay_unknown_command: send -1 if received unknown command | |
1370 | */ | |
1371 | static | |
1372 | void relay_unknown_command(struct relay_command *cmd) | |
1373 | { | |
1374 | struct lttcomm_relayd_generic_reply reply; | |
1375 | int ret; | |
1376 | ||
f73fabfd | 1377 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
b8aa1682 JD |
1378 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, |
1379 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1380 | if (ret < 0) { | |
1381 | ERR("Relay sending unknown command"); | |
1382 | } | |
1383 | } | |
1384 | ||
1385 | /* | |
1386 | * relay_start: send an acknowledgment to the client to tell if we are | |
1387 | * ready to receive data. We are ready if a session is established. | |
1388 | */ | |
1389 | static | |
1390 | int relay_start(struct lttcomm_relayd_hdr *recv_hdr, | |
1391 | struct relay_command *cmd) | |
1392 | { | |
f73fabfd | 1393 | int ret = htobe32(LTTNG_OK); |
b8aa1682 JD |
1394 | struct lttcomm_relayd_generic_reply reply; |
1395 | struct relay_session *session = cmd->session; | |
1396 | ||
1397 | if (!session) { | |
1398 | DBG("Trying to start the streaming without a session established"); | |
f73fabfd | 1399 | ret = htobe32(LTTNG_ERR_UNK); |
b8aa1682 JD |
1400 | } |
1401 | ||
1402 | reply.ret_code = ret; | |
1403 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1404 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1405 | if (ret < 0) { | |
1406 | ERR("Relay sending start ack"); | |
1407 | } | |
1408 | ||
1409 | return ret; | |
1410 | } | |
1411 | ||
1d4dfdef DG |
1412 | /* |
1413 | * Append padding to the file pointed by the file descriptor fd. | |
1414 | */ | |
1415 | static int write_padding_to_file(int fd, uint32_t size) | |
1416 | { | |
6cd525e8 | 1417 | ssize_t ret = 0; |
1d4dfdef DG |
1418 | char *zeros; |
1419 | ||
1420 | if (size == 0) { | |
1421 | goto end; | |
1422 | } | |
1423 | ||
1424 | zeros = zmalloc(size); | |
1425 | if (zeros == NULL) { | |
1426 | PERROR("zmalloc zeros for padding"); | |
1427 | ret = -1; | |
1428 | goto end; | |
1429 | } | |
1430 | ||
6cd525e8 MD |
1431 | ret = lttng_write(fd, zeros, size); |
1432 | if (ret < size) { | |
1d4dfdef DG |
1433 | PERROR("write padding to file"); |
1434 | } | |
1435 | ||
e986c7a1 DG |
1436 | free(zeros); |
1437 | ||
1d4dfdef DG |
1438 | end: |
1439 | return ret; | |
1440 | } | |
1441 | ||
b8aa1682 JD |
1442 | /* |
1443 | * relay_recv_metadata: receive the metada for the session. | |
1444 | */ | |
1445 | static | |
1446 | int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1447 | struct relay_command *cmd) |
b8aa1682 | 1448 | { |
f73fabfd | 1449 | int ret = htobe32(LTTNG_OK); |
6cd525e8 | 1450 | ssize_t size_ret; |
b8aa1682 JD |
1451 | struct relay_session *session = cmd->session; |
1452 | struct lttcomm_relayd_metadata_payload *metadata_struct; | |
1453 | struct relay_stream *metadata_stream; | |
1454 | uint64_t data_size, payload_size; | |
2a174661 | 1455 | struct ctf_trace *ctf_trace; |
b8aa1682 JD |
1456 | |
1457 | if (!session) { | |
1458 | ERR("Metadata sent before version check"); | |
1459 | ret = -1; | |
1460 | goto end; | |
1461 | } | |
1462 | ||
f6416125 MD |
1463 | data_size = payload_size = be64toh(recv_hdr->data_size); |
1464 | if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) { | |
1465 | ERR("Incorrect data size"); | |
1466 | ret = -1; | |
1467 | goto end; | |
1468 | } | |
1469 | payload_size -= sizeof(struct lttcomm_relayd_metadata_payload); | |
1470 | ||
b8aa1682 | 1471 | if (data_buffer_size < data_size) { |
d7b3776f | 1472 | /* In case the realloc fails, we can free the memory */ |
c617c0c6 MD |
1473 | char *tmp_data_ptr; |
1474 | ||
1475 | tmp_data_ptr = realloc(data_buffer, data_size); | |
1476 | if (!tmp_data_ptr) { | |
b8aa1682 | 1477 | ERR("Allocating data buffer"); |
c617c0c6 | 1478 | free(data_buffer); |
b8aa1682 JD |
1479 | ret = -1; |
1480 | goto end; | |
1481 | } | |
c617c0c6 | 1482 | data_buffer = tmp_data_ptr; |
b8aa1682 JD |
1483 | data_buffer_size = data_size; |
1484 | } | |
1485 | memset(data_buffer, 0, data_size); | |
77c7c900 | 1486 | DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size); |
7c5aef62 | 1487 | ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0); |
b8aa1682 | 1488 | if (ret < 0 || ret != data_size) { |
a6cd2b97 DG |
1489 | if (ret == 0) { |
1490 | /* Orderly shutdown. Not necessary to print an error. */ | |
1491 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1492 | } else { | |
1493 | ERR("Relay didn't receive the whole metadata"); | |
1494 | } | |
b8aa1682 | 1495 | ret = -1; |
b8aa1682 JD |
1496 | goto end; |
1497 | } | |
1498 | metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer; | |
9d1bbf21 MD |
1499 | |
1500 | rcu_read_lock(); | |
2a174661 | 1501 | metadata_stream = stream_find_by_id(relay_streams_ht, |
d3e2ba59 | 1502 | be64toh(metadata_struct->stream_id)); |
b8aa1682 JD |
1503 | if (!metadata_stream) { |
1504 | ret = -1; | |
9d1bbf21 | 1505 | goto end_unlock; |
b8aa1682 JD |
1506 | } |
1507 | ||
6cd525e8 MD |
1508 | size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload, |
1509 | payload_size); | |
1510 | if (size_ret < payload_size) { | |
b8aa1682 JD |
1511 | ERR("Relay error writing metadata on file"); |
1512 | ret = -1; | |
9d1bbf21 | 1513 | goto end_unlock; |
b8aa1682 | 1514 | } |
1d4dfdef DG |
1515 | |
1516 | ret = write_padding_to_file(metadata_stream->fd, | |
1517 | be32toh(metadata_struct->padding_size)); | |
1518 | if (ret < 0) { | |
1519 | goto end_unlock; | |
1520 | } | |
2a174661 DG |
1521 | |
1522 | ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, | |
1523 | metadata_stream->path_name); | |
1524 | assert(ctf_trace); | |
1525 | ctf_trace->metadata_received += | |
d3e2ba59 | 1526 | payload_size + be32toh(metadata_struct->padding_size); |
1d4dfdef | 1527 | |
b8aa1682 JD |
1528 | DBG2("Relay metadata written"); |
1529 | ||
9d1bbf21 | 1530 | end_unlock: |
6e3c5836 | 1531 | rcu_read_unlock(); |
b8aa1682 JD |
1532 | end: |
1533 | return ret; | |
1534 | } | |
1535 | ||
1536 | /* | |
1537 | * relay_send_version: send relayd version number | |
1538 | */ | |
1539 | static | |
1540 | int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1541 | struct relay_command *cmd, struct lttng_ht *sessions_ht) |
b8aa1682 | 1542 | { |
7f51dcba | 1543 | int ret; |
092b6259 | 1544 | struct lttcomm_relayd_version reply, msg; |
b8aa1682 | 1545 | |
c5b6f4f0 DG |
1546 | assert(cmd); |
1547 | ||
1548 | cmd->version_check_done = 1; | |
b8aa1682 | 1549 | |
092b6259 | 1550 | /* Get version from the other side. */ |
7c5aef62 | 1551 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); |
092b6259 | 1552 | if (ret < 0 || ret != sizeof(msg)) { |
a6cd2b97 DG |
1553 | if (ret == 0) { |
1554 | /* Orderly shutdown. Not necessary to print an error. */ | |
1555 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1556 | } else { | |
1557 | ERR("Relay failed to receive the version values."); | |
1558 | } | |
092b6259 | 1559 | ret = -1; |
092b6259 DG |
1560 | goto end; |
1561 | } | |
1562 | ||
d83a952c MD |
1563 | reply.major = RELAYD_VERSION_COMM_MAJOR; |
1564 | reply.minor = RELAYD_VERSION_COMM_MINOR; | |
d4519fa3 JD |
1565 | |
1566 | /* Major versions must be the same */ | |
1567 | if (reply.major != be32toh(msg.major)) { | |
6151a90f JD |
1568 | DBG("Incompatible major versions (%u vs %u), deleting session", |
1569 | reply.major, be32toh(msg.major)); | |
d3e2ba59 | 1570 | relay_delete_session(cmd, sessions_ht); |
d4519fa3 JD |
1571 | ret = 0; |
1572 | goto end; | |
1573 | } | |
1574 | ||
0f907de1 JD |
1575 | cmd->major = reply.major; |
1576 | /* We adapt to the lowest compatible version */ | |
1577 | if (reply.minor <= be32toh(msg.minor)) { | |
1578 | cmd->minor = reply.minor; | |
1579 | } else { | |
1580 | cmd->minor = be32toh(msg.minor); | |
1581 | } | |
1582 | ||
6151a90f JD |
1583 | reply.major = htobe32(reply.major); |
1584 | reply.minor = htobe32(reply.minor); | |
1585 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1586 | sizeof(struct lttcomm_relayd_version), 0); | |
1587 | if (ret < 0) { | |
1588 | ERR("Relay sending version"); | |
1589 | } | |
1590 | ||
0f907de1 JD |
1591 | DBG("Version check done using protocol %u.%u", cmd->major, |
1592 | cmd->minor); | |
b8aa1682 JD |
1593 | |
1594 | end: | |
1595 | return ret; | |
1596 | } | |
1597 | ||
c8f59ee5 | 1598 | /* |
6d805429 | 1599 | * Check for data pending for a given stream id from the session daemon. |
c8f59ee5 DG |
1600 | */ |
1601 | static | |
6d805429 | 1602 | int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, |
d3e2ba59 | 1603 | struct relay_command *cmd) |
c8f59ee5 DG |
1604 | { |
1605 | struct relay_session *session = cmd->session; | |
6d805429 | 1606 | struct lttcomm_relayd_data_pending msg; |
c8f59ee5 DG |
1607 | struct lttcomm_relayd_generic_reply reply; |
1608 | struct relay_stream *stream; | |
1609 | int ret; | |
c8f59ee5 DG |
1610 | uint64_t last_net_seq_num, stream_id; |
1611 | ||
6d805429 | 1612 | DBG("Data pending command received"); |
c8f59ee5 | 1613 | |
c5b6f4f0 | 1614 | if (!session || cmd->version_check_done == 0) { |
c8f59ee5 DG |
1615 | ERR("Trying to check for data before version check"); |
1616 | ret = -1; | |
1617 | goto end_no_session; | |
1618 | } | |
1619 | ||
7c5aef62 | 1620 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); |
c8f59ee5 | 1621 | if (ret < sizeof(msg)) { |
a6cd2b97 DG |
1622 | if (ret == 0) { |
1623 | /* Orderly shutdown. Not necessary to print an error. */ | |
1624 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1625 | } else { | |
1626 | ERR("Relay didn't receive valid data_pending struct size : %d", | |
1627 | ret); | |
1628 | } | |
c8f59ee5 DG |
1629 | ret = -1; |
1630 | goto end_no_session; | |
1631 | } | |
1632 | ||
1633 | stream_id = be64toh(msg.stream_id); | |
1634 | last_net_seq_num = be64toh(msg.last_net_seq_num); | |
1635 | ||
1636 | rcu_read_lock(); | |
2a174661 | 1637 | stream = stream_find_by_id(relay_streams_ht, stream_id); |
de91f48a | 1638 | if (stream == NULL) { |
c8f59ee5 DG |
1639 | ret = -1; |
1640 | goto end_unlock; | |
1641 | } | |
1642 | ||
6d805429 | 1643 | DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64 |
c8f59ee5 DG |
1644 | " and last_seq %" PRIu64, stream_id, stream->prev_seq, |
1645 | last_net_seq_num); | |
1646 | ||
33832e64 | 1647 | /* Avoid wrapping issue */ |
39df6d9f | 1648 | if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) { |
6d805429 | 1649 | /* Data has in fact been written and is NOT pending */ |
c8f59ee5 | 1650 | ret = 0; |
6d805429 DG |
1651 | } else { |
1652 | /* Data still being streamed thus pending */ | |
1653 | ret = 1; | |
c8f59ee5 DG |
1654 | } |
1655 | ||
f7079f67 DG |
1656 | /* Pending check is now done. */ |
1657 | stream->data_pending_check_done = 1; | |
1658 | ||
c8f59ee5 DG |
1659 | end_unlock: |
1660 | rcu_read_unlock(); | |
1661 | ||
1662 | reply.ret_code = htobe32(ret); | |
1663 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1664 | if (ret < 0) { | |
6d805429 | 1665 | ERR("Relay data pending ret code failed"); |
c8f59ee5 DG |
1666 | } |
1667 | ||
1668 | end_no_session: | |
1669 | return ret; | |
1670 | } | |
1671 | ||
1672 | /* | |
1673 | * Wait for the control socket to reach a quiescent state. | |
1674 | * | |
1675 | * Note that for now, when receiving this command from the session daemon, this | |
1676 | * means that every subsequent commands or data received on the control socket | |
1677 | * has been handled. So, this is why we simply return OK here. | |
1678 | */ | |
1679 | static | |
1680 | int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1681 | struct relay_command *cmd) |
c8f59ee5 DG |
1682 | { |
1683 | int ret; | |
ad7051c0 DG |
1684 | uint64_t stream_id; |
1685 | struct relay_stream *stream; | |
1686 | struct lttng_ht_iter iter; | |
1687 | struct lttcomm_relayd_quiescent_control msg; | |
c8f59ee5 DG |
1688 | struct lttcomm_relayd_generic_reply reply; |
1689 | ||
1690 | DBG("Checking quiescent state on control socket"); | |
1691 | ||
ad7051c0 DG |
1692 | if (!cmd->session || cmd->version_check_done == 0) { |
1693 | ERR("Trying to check for data before version check"); | |
1694 | ret = -1; | |
1695 | goto end_no_session; | |
1696 | } | |
1697 | ||
1698 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1699 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1700 | if (ret == 0) { |
1701 | /* Orderly shutdown. Not necessary to print an error. */ | |
1702 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1703 | } else { | |
1704 | ERR("Relay didn't receive valid begin data_pending struct size: %d", | |
1705 | ret); | |
1706 | } | |
ad7051c0 DG |
1707 | ret = -1; |
1708 | goto end_no_session; | |
1709 | } | |
1710 | ||
1711 | stream_id = be64toh(msg.stream_id); | |
1712 | ||
1713 | rcu_read_lock(); | |
d3e2ba59 | 1714 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
2a174661 | 1715 | node.node) { |
ad7051c0 DG |
1716 | if (stream->stream_handle == stream_id) { |
1717 | stream->data_pending_check_done = 1; | |
1718 | DBG("Relay quiescent control pending flag set to %" PRIu64, | |
1719 | stream_id); | |
1720 | break; | |
1721 | } | |
1722 | } | |
1723 | rcu_read_unlock(); | |
1724 | ||
c8f59ee5 DG |
1725 | reply.ret_code = htobe32(LTTNG_OK); |
1726 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1727 | if (ret < 0) { | |
6d805429 | 1728 | ERR("Relay data quiescent control ret code failed"); |
c8f59ee5 DG |
1729 | } |
1730 | ||
ad7051c0 | 1731 | end_no_session: |
c8f59ee5 DG |
1732 | return ret; |
1733 | } | |
1734 | ||
f7079f67 DG |
1735 | /* |
1736 | * Initialize a data pending command. This means that a client is about to ask | |
1737 | * for data pending for each stream he/she holds. Simply iterate over all | |
1738 | * streams of a session and set the data_pending_check_done flag. | |
1739 | * | |
1740 | * This command returns to the client a LTTNG_OK code. | |
1741 | */ | |
1742 | static | |
1743 | int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1744 | struct relay_command *cmd) |
f7079f67 DG |
1745 | { |
1746 | int ret; | |
1747 | struct lttng_ht_iter iter; | |
1748 | struct lttcomm_relayd_begin_data_pending msg; | |
1749 | struct lttcomm_relayd_generic_reply reply; | |
1750 | struct relay_stream *stream; | |
1751 | uint64_t session_id; | |
1752 | ||
1753 | assert(recv_hdr); | |
1754 | assert(cmd); | |
f7079f67 DG |
1755 | |
1756 | DBG("Init streams for data pending"); | |
1757 | ||
1758 | if (!cmd->session || cmd->version_check_done == 0) { | |
1759 | ERR("Trying to check for data before version check"); | |
1760 | ret = -1; | |
1761 | goto end_no_session; | |
1762 | } | |
1763 | ||
1764 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1765 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1766 | if (ret == 0) { |
1767 | /* Orderly shutdown. Not necessary to print an error. */ | |
1768 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1769 | } else { | |
1770 | ERR("Relay didn't receive valid begin data_pending struct size: %d", | |
1771 | ret); | |
1772 | } | |
f7079f67 DG |
1773 | ret = -1; |
1774 | goto end_no_session; | |
1775 | } | |
1776 | ||
1777 | session_id = be64toh(msg.session_id); | |
1778 | ||
1779 | /* | |
1780 | * Iterate over all streams to set the begin data pending flag. For now, the | |
1781 | * streams are indexed by stream handle so we have to iterate over all | |
1782 | * streams to find the one associated with the right session_id. | |
1783 | */ | |
1784 | rcu_read_lock(); | |
d3e2ba59 | 1785 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
2a174661 DG |
1786 | node.node) { |
1787 | if (stream->session_id == session_id) { | |
f7079f67 DG |
1788 | stream->data_pending_check_done = 0; |
1789 | DBG("Set begin data pending flag to stream %" PRIu64, | |
1790 | stream->stream_handle); | |
1791 | } | |
1792 | } | |
1793 | rcu_read_unlock(); | |
1794 | ||
1795 | /* All good, send back reply. */ | |
1796 | reply.ret_code = htobe32(LTTNG_OK); | |
1797 | ||
1798 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1799 | if (ret < 0) { | |
1800 | ERR("Relay begin data pending send reply failed"); | |
1801 | } | |
1802 | ||
1803 | end_no_session: | |
1804 | return ret; | |
1805 | } | |
1806 | ||
1807 | /* | |
1808 | * End data pending command. This will check, for a given session id, if each | |
1809 | * stream associated with it has its data_pending_check_done flag set. If not, | |
1810 | * this means that the client lost track of the stream but the data is still | |
1811 | * being streamed on our side. In this case, we inform the client that data is | |
1812 | * inflight. | |
1813 | * | |
1814 | * Return to the client if there is data in flight or not with a ret_code. | |
1815 | */ | |
1816 | static | |
1817 | int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1818 | struct relay_command *cmd) |
f7079f67 DG |
1819 | { |
1820 | int ret; | |
1821 | struct lttng_ht_iter iter; | |
1822 | struct lttcomm_relayd_end_data_pending msg; | |
1823 | struct lttcomm_relayd_generic_reply reply; | |
1824 | struct relay_stream *stream; | |
1825 | uint64_t session_id; | |
1826 | uint32_t is_data_inflight = 0; | |
1827 | ||
1828 | assert(recv_hdr); | |
1829 | assert(cmd); | |
f7079f67 DG |
1830 | |
1831 | DBG("End data pending command"); | |
1832 | ||
1833 | if (!cmd->session || cmd->version_check_done == 0) { | |
1834 | ERR("Trying to check for data before version check"); | |
1835 | ret = -1; | |
1836 | goto end_no_session; | |
1837 | } | |
1838 | ||
1839 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1840 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1841 | if (ret == 0) { |
1842 | /* Orderly shutdown. Not necessary to print an error. */ | |
1843 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1844 | } else { | |
1845 | ERR("Relay didn't receive valid end data_pending struct size: %d", | |
1846 | ret); | |
1847 | } | |
f7079f67 DG |
1848 | ret = -1; |
1849 | goto end_no_session; | |
1850 | } | |
1851 | ||
1852 | session_id = be64toh(msg.session_id); | |
1853 | ||
1854 | /* Iterate over all streams to see if the begin data pending flag is set. */ | |
1855 | rcu_read_lock(); | |
d3e2ba59 | 1856 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
2a174661 DG |
1857 | node.node) { |
1858 | if (stream->session_id == session_id && | |
f7079f67 DG |
1859 | !stream->data_pending_check_done) { |
1860 | is_data_inflight = 1; | |
1861 | DBG("Data is still in flight for stream %" PRIu64, | |
1862 | stream->stream_handle); | |
1863 | break; | |
1864 | } | |
1865 | } | |
1866 | rcu_read_unlock(); | |
1867 | ||
1868 | /* All good, send back reply. */ | |
1869 | reply.ret_code = htobe32(is_data_inflight); | |
1870 | ||
1871 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1872 | if (ret < 0) { | |
1873 | ERR("Relay end data pending send reply failed"); | |
1874 | } | |
1875 | ||
1876 | end_no_session: | |
1877 | return ret; | |
1878 | } | |
1879 | ||
1c20f0e2 JD |
1880 | /* |
1881 | * Receive an index for a specific stream. | |
1882 | * | |
1883 | * Return 0 on success else a negative value. | |
1884 | */ | |
1885 | static | |
1886 | int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr, | |
0a6518b0 | 1887 | struct relay_command *cmd) |
1c20f0e2 JD |
1888 | { |
1889 | int ret, send_ret, index_created = 0; | |
1890 | struct relay_session *session = cmd->session; | |
1891 | struct lttcomm_relayd_index index_info; | |
1892 | struct relay_index *index, *wr_index = NULL; | |
1893 | struct lttcomm_relayd_generic_reply reply; | |
1894 | struct relay_stream *stream; | |
1895 | uint64_t net_seq_num; | |
1896 | ||
1897 | assert(cmd); | |
1c20f0e2 JD |
1898 | |
1899 | DBG("Relay receiving index"); | |
1900 | ||
1901 | if (!session || cmd->version_check_done == 0) { | |
1902 | ERR("Trying to close a stream before version check"); | |
1903 | ret = -1; | |
1904 | goto end_no_session; | |
1905 | } | |
1906 | ||
1907 | ret = cmd->sock->ops->recvmsg(cmd->sock, &index_info, | |
1908 | sizeof(index_info), 0); | |
1909 | if (ret < sizeof(index_info)) { | |
1910 | if (ret == 0) { | |
1911 | /* Orderly shutdown. Not necessary to print an error. */ | |
1912 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1913 | } else { | |
1914 | ERR("Relay didn't receive valid index struct size : %d", ret); | |
1915 | } | |
1916 | ret = -1; | |
1917 | goto end_no_session; | |
1918 | } | |
1919 | ||
1920 | net_seq_num = be64toh(index_info.net_seq_num); | |
1921 | ||
1922 | rcu_read_lock(); | |
2a174661 DG |
1923 | stream = stream_find_by_id(relay_streams_ht, |
1924 | be64toh(index_info.relay_stream_id)); | |
1c20f0e2 JD |
1925 | if (!stream) { |
1926 | ret = -1; | |
1927 | goto end_rcu_unlock; | |
1928 | } | |
1929 | ||
d3e2ba59 JD |
1930 | /* Live beacon handling */ |
1931 | if (index_info.packet_size == 0) { | |
1932 | DBG("Received live beacon for stream %" PRIu64, stream->stream_handle); | |
1933 | ||
1934 | /* | |
1935 | * Only flag a stream inactive when it has already received data. | |
1936 | */ | |
1937 | if (stream->total_index_received > 0) { | |
1938 | stream->beacon_ts_end = be64toh(index_info.timestamp_end); | |
1939 | } | |
1940 | ret = 0; | |
1941 | goto end_rcu_unlock; | |
1942 | } else { | |
1943 | stream->beacon_ts_end = -1ULL; | |
1944 | } | |
1945 | ||
0a6518b0 | 1946 | index = relay_index_find(stream->stream_handle, net_seq_num); |
1c20f0e2 JD |
1947 | if (!index) { |
1948 | /* A successful creation will add the object to the HT. */ | |
1949 | index = relay_index_create(stream->stream_handle, net_seq_num); | |
1950 | if (!index) { | |
1951 | goto end_rcu_unlock; | |
1952 | } | |
1953 | index_created = 1; | |
1954 | } | |
1955 | ||
1956 | copy_index_control_data(index, &index_info); | |
1957 | ||
1958 | if (index_created) { | |
1959 | /* | |
1960 | * Try to add the relay index object to the hash table. If an object | |
1961 | * already exist, destroy back the index created, set the data in this | |
1962 | * object and write it on disk. | |
1963 | */ | |
0a6518b0 | 1964 | relay_index_add(index, &wr_index); |
1c20f0e2 JD |
1965 | if (wr_index) { |
1966 | copy_index_control_data(wr_index, &index_info); | |
1967 | free(index); | |
1968 | } | |
1969 | } else { | |
1970 | /* The index already exists so write it on disk. */ | |
1971 | wr_index = index; | |
1972 | } | |
1973 | ||
1974 | /* Do we have a writable ready index to write on disk. */ | |
1975 | if (wr_index) { | |
1976 | /* Starting at 2.4, create the index file if none available. */ | |
1977 | if (cmd->minor >= 4 && stream->index_fd < 0) { | |
1978 | ret = index_create_file(stream->path_name, stream->channel_name, | |
1979 | relayd_uid, relayd_gid, stream->tracefile_size, | |
1980 | stream->tracefile_count_current); | |
1981 | if (ret < 0) { | |
1982 | goto end_rcu_unlock; | |
1983 | } | |
1984 | stream->index_fd = ret; | |
1985 | } | |
1986 | ||
0a6518b0 | 1987 | ret = relay_index_write(wr_index->fd, wr_index); |
1c20f0e2 JD |
1988 | if (ret < 0) { |
1989 | goto end_rcu_unlock; | |
1990 | } | |
d3e2ba59 | 1991 | stream->total_index_received++; |
1c20f0e2 JD |
1992 | } |
1993 | ||
1994 | end_rcu_unlock: | |
1995 | rcu_read_unlock(); | |
1996 | ||
1997 | if (ret < 0) { | |
1998 | reply.ret_code = htobe32(LTTNG_ERR_UNK); | |
1999 | } else { | |
2000 | reply.ret_code = htobe32(LTTNG_OK); | |
2001 | } | |
2002 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
2003 | if (send_ret < 0) { | |
2004 | ERR("Relay sending close index id reply"); | |
2005 | ret = send_ret; | |
2006 | } | |
2007 | ||
2008 | end_no_session: | |
2009 | return ret; | |
2010 | } | |
2011 | ||
a4baae1b JD |
2012 | /* |
2013 | * Receive the streams_sent message. | |
2014 | * | |
2015 | * Return 0 on success else a negative value. | |
2016 | */ | |
2017 | static | |
2018 | int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr, | |
2019 | struct relay_command *cmd) | |
2020 | { | |
2021 | int ret, send_ret; | |
2022 | struct lttcomm_relayd_generic_reply reply; | |
2023 | ||
2024 | assert(cmd); | |
2025 | ||
2026 | DBG("Relay receiving streams_sent"); | |
2027 | ||
2028 | if (!cmd->session || cmd->version_check_done == 0) { | |
2029 | ERR("Trying to close a stream before version check"); | |
2030 | ret = -1; | |
2031 | goto end_no_session; | |
2032 | } | |
2033 | ||
2034 | /* | |
2035 | * Flag every pending stream in the connection recv list that they are | |
2036 | * ready to be used by the viewer. | |
2037 | */ | |
2038 | set_viewer_ready_flag(cmd); | |
2039 | ||
4a9daf17 JD |
2040 | /* |
2041 | * Inform the viewer that there are new streams in the session. | |
2042 | */ | |
2043 | uatomic_set(&cmd->session->new_streams, 1); | |
2044 | ||
a4baae1b JD |
2045 | reply.ret_code = htobe32(LTTNG_OK); |
2046 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
2047 | if (send_ret < 0) { | |
2048 | ERR("Relay sending sent_stream reply"); | |
2049 | ret = send_ret; | |
2050 | } else { | |
2051 | /* Success. */ | |
2052 | ret = 0; | |
2053 | } | |
2054 | ||
2055 | end_no_session: | |
2056 | return ret; | |
2057 | } | |
2058 | ||
b8aa1682 | 2059 | /* |
d3e2ba59 | 2060 | * Process the commands received on the control socket |
b8aa1682 JD |
2061 | */ |
2062 | static | |
2063 | int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 2064 | struct relay_command *cmd, struct relay_local_data *ctx) |
b8aa1682 JD |
2065 | { |
2066 | int ret = 0; | |
2067 | ||
2068 | switch (be32toh(recv_hdr->cmd)) { | |
b8aa1682 | 2069 | case RELAYD_CREATE_SESSION: |
d3e2ba59 | 2070 | ret = relay_create_session(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 | 2071 | break; |
b8aa1682 | 2072 | case RELAYD_ADD_STREAM: |
d3e2ba59 | 2073 | ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 JD |
2074 | break; |
2075 | case RELAYD_START_DATA: | |
2076 | ret = relay_start(recv_hdr, cmd); | |
2077 | break; | |
2078 | case RELAYD_SEND_METADATA: | |
d3e2ba59 | 2079 | ret = relay_recv_metadata(recv_hdr, cmd); |
b8aa1682 JD |
2080 | break; |
2081 | case RELAYD_VERSION: | |
d3e2ba59 | 2082 | ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 | 2083 | break; |
173af62f | 2084 | case RELAYD_CLOSE_STREAM: |
92c6ca54 | 2085 | ret = relay_close_stream(recv_hdr, cmd); |
173af62f | 2086 | break; |
6d805429 | 2087 | case RELAYD_DATA_PENDING: |
d3e2ba59 | 2088 | ret = relay_data_pending(recv_hdr, cmd); |
c8f59ee5 DG |
2089 | break; |
2090 | case RELAYD_QUIESCENT_CONTROL: | |
d3e2ba59 | 2091 | ret = relay_quiescent_control(recv_hdr, cmd); |
c8f59ee5 | 2092 | break; |
f7079f67 | 2093 | case RELAYD_BEGIN_DATA_PENDING: |
d3e2ba59 | 2094 | ret = relay_begin_data_pending(recv_hdr, cmd); |
f7079f67 DG |
2095 | break; |
2096 | case RELAYD_END_DATA_PENDING: | |
d3e2ba59 | 2097 | ret = relay_end_data_pending(recv_hdr, cmd); |
f7079f67 | 2098 | break; |
1c20f0e2 | 2099 | case RELAYD_SEND_INDEX: |
0a6518b0 | 2100 | ret = relay_recv_index(recv_hdr, cmd); |
1c20f0e2 | 2101 | break; |
a4baae1b JD |
2102 | case RELAYD_STREAMS_SENT: |
2103 | ret = relay_streams_sent(recv_hdr, cmd); | |
2104 | break; | |
b8aa1682 JD |
2105 | case RELAYD_UPDATE_SYNC_INFO: |
2106 | default: | |
2107 | ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd)); | |
2108 | relay_unknown_command(cmd); | |
2109 | ret = -1; | |
2110 | goto end; | |
2111 | } | |
2112 | ||
2113 | end: | |
2114 | return ret; | |
2115 | } | |
2116 | ||
7d2f7452 DG |
2117 | /* |
2118 | * Handle index for a data stream. | |
2119 | * | |
2120 | * RCU read side lock MUST be acquired. | |
2121 | * | |
2122 | * Return 0 on success else a negative value. | |
2123 | */ | |
2124 | static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num, | |
2125 | int rotate_index) | |
2126 | { | |
2127 | int ret = 0, index_created = 0; | |
2128 | uint64_t stream_id, data_offset; | |
2129 | struct relay_index *index, *wr_index = NULL; | |
2130 | ||
2131 | assert(stream); | |
2132 | ||
2133 | stream_id = stream->stream_handle; | |
2134 | /* Get data offset because we are about to update the index. */ | |
2135 | data_offset = htobe64(stream->tracefile_size_current); | |
2136 | ||
2137 | /* | |
2138 | * Lookup for an existing index for that stream id/sequence number. If on | |
2139 | * exists, the control thread already received the data for it thus we need | |
2140 | * to write it on disk. | |
2141 | */ | |
2142 | index = relay_index_find(stream_id, net_seq_num); | |
2143 | if (!index) { | |
2144 | /* A successful creation will add the object to the HT. */ | |
2145 | index = relay_index_create(stream_id, net_seq_num); | |
2146 | if (!index) { | |
2147 | ret = -1; | |
2148 | goto error; | |
2149 | } | |
2150 | index_created = 1; | |
2151 | } | |
2152 | ||
2153 | if (rotate_index || stream->index_fd < 0) { | |
2154 | index->to_close_fd = stream->index_fd; | |
2155 | ret = index_create_file(stream->path_name, stream->channel_name, | |
2156 | relayd_uid, relayd_gid, stream->tracefile_size, | |
2157 | stream->tracefile_count_current); | |
2158 | if (ret < 0) { | |
2159 | /* This will close the stream's index fd if one. */ | |
2160 | relay_index_free_safe(index); | |
2161 | goto error; | |
2162 | } | |
2163 | stream->index_fd = ret; | |
2164 | } | |
2165 | index->fd = stream->index_fd; | |
2166 | index->index_data.offset = data_offset; | |
2167 | ||
2168 | if (index_created) { | |
2169 | /* | |
2170 | * Try to add the relay index object to the hash table. If an object | |
2171 | * already exist, destroy back the index created and set the data. | |
2172 | */ | |
2173 | relay_index_add(index, &wr_index); | |
2174 | if (wr_index) { | |
2175 | /* Copy back data from the created index. */ | |
2176 | wr_index->fd = index->fd; | |
2177 | wr_index->to_close_fd = index->to_close_fd; | |
2178 | wr_index->index_data.offset = data_offset; | |
2179 | free(index); | |
2180 | } | |
2181 | } else { | |
2182 | /* The index already exists so write it on disk. */ | |
2183 | wr_index = index; | |
2184 | } | |
2185 | ||
2186 | /* Do we have a writable ready index to write on disk. */ | |
2187 | if (wr_index) { | |
2188 | ret = relay_index_write(wr_index->fd, wr_index); | |
2189 | if (ret < 0) { | |
2190 | goto error; | |
2191 | } | |
2192 | stream->total_index_received++; | |
2193 | } | |
2194 | ||
2195 | error: | |
2196 | return ret; | |
2197 | } | |
2198 | ||
b8aa1682 JD |
2199 | /* |
2200 | * relay_process_data: Process the data received on the data socket | |
2201 | */ | |
2202 | static | |
2a174661 | 2203 | int relay_process_data(struct relay_command *cmd, struct lttng_ht *sessions_ht) |
b8aa1682 | 2204 | { |
7d2f7452 | 2205 | int ret = 0, rotate_index = 0; |
6cd525e8 | 2206 | ssize_t size_ret; |
b8aa1682 JD |
2207 | struct relay_stream *stream; |
2208 | struct lttcomm_relayd_data_hdr data_hdr; | |
7d2f7452 | 2209 | uint64_t stream_id; |
173af62f | 2210 | uint64_t net_seq_num; |
b8aa1682 | 2211 | uint32_t data_size; |
2a174661 | 2212 | struct relay_session *session; |
b8aa1682 JD |
2213 | |
2214 | ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr, | |
7c5aef62 | 2215 | sizeof(struct lttcomm_relayd_data_hdr), 0); |
b8aa1682 | 2216 | if (ret <= 0) { |
a6cd2b97 DG |
2217 | if (ret == 0) { |
2218 | /* Orderly shutdown. Not necessary to print an error. */ | |
2219 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
2220 | } else { | |
2221 | ERR("Unable to receive data header on sock %d", cmd->sock->fd); | |
2222 | } | |
b8aa1682 JD |
2223 | ret = -1; |
2224 | goto end; | |
2225 | } | |
2226 | ||
2227 | stream_id = be64toh(data_hdr.stream_id); | |
9d1bbf21 MD |
2228 | |
2229 | rcu_read_lock(); | |
2a174661 | 2230 | stream = stream_find_by_id(relay_streams_ht, stream_id); |
b8aa1682 JD |
2231 | if (!stream) { |
2232 | ret = -1; | |
1c20f0e2 | 2233 | goto end_rcu_unlock; |
b8aa1682 JD |
2234 | } |
2235 | ||
2a174661 DG |
2236 | session = session_find_by_id(sessions_ht, stream->session_id); |
2237 | assert(session); | |
2238 | ||
b8aa1682 JD |
2239 | data_size = be32toh(data_hdr.data_size); |
2240 | if (data_buffer_size < data_size) { | |
c617c0c6 MD |
2241 | char *tmp_data_ptr; |
2242 | ||
2243 | tmp_data_ptr = realloc(data_buffer, data_size); | |
2244 | if (!tmp_data_ptr) { | |
b8aa1682 | 2245 | ERR("Allocating data buffer"); |
c617c0c6 | 2246 | free(data_buffer); |
b8aa1682 | 2247 | ret = -1; |
1c20f0e2 | 2248 | goto end_rcu_unlock; |
b8aa1682 | 2249 | } |
c617c0c6 | 2250 | data_buffer = tmp_data_ptr; |
b8aa1682 JD |
2251 | data_buffer_size = data_size; |
2252 | } | |
2253 | memset(data_buffer, 0, data_size); | |
2254 | ||
173af62f DG |
2255 | net_seq_num = be64toh(data_hdr.net_seq_num); |
2256 | ||
77c7c900 | 2257 | DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64, |
173af62f | 2258 | data_size, stream_id, net_seq_num); |
7c5aef62 | 2259 | ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0); |
b8aa1682 | 2260 | if (ret <= 0) { |
a6cd2b97 DG |
2261 | if (ret == 0) { |
2262 | /* Orderly shutdown. Not necessary to print an error. */ | |
2263 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
2264 | } | |
b8aa1682 | 2265 | ret = -1; |
1c20f0e2 | 2266 | goto end_rcu_unlock; |
b8aa1682 JD |
2267 | } |
2268 | ||
1c20f0e2 | 2269 | /* Check if a rotation is needed. */ |
0f907de1 JD |
2270 | if (stream->tracefile_size > 0 && |
2271 | (stream->tracefile_size_current + data_size) > | |
2272 | stream->tracefile_size) { | |
6b6b9a5a JD |
2273 | struct relay_viewer_stream *vstream; |
2274 | uint64_t new_id; | |
2275 | ||
2276 | new_id = (stream->tracefile_count_current + 1) % | |
2277 | stream->tracefile_count; | |
2278 | /* | |
2279 | * When we wrap-around back to 0, we start overwriting old | |
2280 | * trace data. | |
2281 | */ | |
2282 | if (!stream->tracefile_overwrite && new_id == 0) { | |
2283 | stream->tracefile_overwrite = 1; | |
2284 | } | |
2285 | pthread_mutex_lock(&stream->viewer_stream_rotation_lock); | |
2286 | if (stream->tracefile_overwrite) { | |
2287 | stream->oldest_tracefile_id = | |
2288 | (stream->oldest_tracefile_id + 1) % | |
2289 | stream->tracefile_count; | |
2290 | } | |
2f8f53af | 2291 | vstream = viewer_stream_find_by_id(stream->stream_handle); |
6b6b9a5a JD |
2292 | if (vstream) { |
2293 | /* | |
2294 | * The viewer is reading a file about to be | |
2295 | * overwritten. Close the FDs it is | |
2296 | * currently using and let it handle the fault. | |
2297 | */ | |
2298 | if (vstream->tracefile_count_current == new_id) { | |
cef0f7d5 | 2299 | pthread_mutex_lock(&vstream->overwrite_lock); |
6b6b9a5a | 2300 | vstream->abort_flag = 1; |
cef0f7d5 | 2301 | pthread_mutex_unlock(&vstream->overwrite_lock); |
6b6b9a5a JD |
2302 | DBG("Streaming side setting abort_flag on stream %s_%lu\n", |
2303 | stream->channel_name, new_id); | |
2304 | } else if (vstream->tracefile_count_current == | |
2305 | stream->tracefile_count_current) { | |
2306 | /* | |
2307 | * The reader and writer were in the | |
2308 | * same trace file, inform the viewer | |
2309 | * that no new index will ever be added | |
2310 | * to this file. | |
2311 | */ | |
2312 | vstream->close_write_flag = 1; | |
2313 | } | |
2314 | } | |
1c20f0e2 JD |
2315 | ret = utils_rotate_stream_file(stream->path_name, stream->channel_name, |
2316 | stream->tracefile_size, stream->tracefile_count, | |
2317 | relayd_uid, relayd_gid, stream->fd, | |
2318 | &(stream->tracefile_count_current), &stream->fd); | |
cef0f7d5 | 2319 | stream->total_index_received = 0; |
6b6b9a5a | 2320 | pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); |
0f907de1 | 2321 | if (ret < 0) { |
1c20f0e2 JD |
2322 | ERR("Rotating stream output file"); |
2323 | goto end_rcu_unlock; | |
0f907de1 | 2324 | } |
a6976990 DG |
2325 | /* Reset current size because we just perform a stream rotation. */ |
2326 | stream->tracefile_size_current = 0; | |
1c20f0e2 JD |
2327 | rotate_index = 1; |
2328 | } | |
2329 | ||
1c20f0e2 | 2330 | /* |
7d2f7452 DG |
2331 | * Index are handled in protocol version 2.4 and above. Also, snapshot and |
2332 | * index are NOT supported. | |
1c20f0e2 | 2333 | */ |
2a174661 | 2334 | if (session->minor >= 4 && !session->snapshot) { |
7d2f7452 | 2335 | ret = handle_index_data(stream, net_seq_num, rotate_index); |
1c20f0e2 | 2336 | if (ret < 0) { |
1c20f0e2 JD |
2337 | goto end_rcu_unlock; |
2338 | } | |
1c20f0e2 JD |
2339 | } |
2340 | ||
7d2f7452 | 2341 | /* Write data to stream output fd. */ |
6cd525e8 MD |
2342 | size_ret = lttng_write(stream->fd, data_buffer, data_size); |
2343 | if (size_ret < data_size) { | |
b8aa1682 JD |
2344 | ERR("Relay error writing data to file"); |
2345 | ret = -1; | |
1c20f0e2 | 2346 | goto end_rcu_unlock; |
b8aa1682 | 2347 | } |
1d4dfdef | 2348 | |
5ab7344e JD |
2349 | DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64, |
2350 | ret, stream->stream_handle); | |
2351 | ||
1d4dfdef DG |
2352 | ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size)); |
2353 | if (ret < 0) { | |
1c20f0e2 | 2354 | goto end_rcu_unlock; |
1d4dfdef | 2355 | } |
1c20f0e2 | 2356 | stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size); |
1d4dfdef | 2357 | |
173af62f DG |
2358 | stream->prev_seq = net_seq_num; |
2359 | ||
2a174661 | 2360 | try_close_stream(session, stream); |
173af62f | 2361 | |
1c20f0e2 | 2362 | end_rcu_unlock: |
9d1bbf21 | 2363 | rcu_read_unlock(); |
b8aa1682 JD |
2364 | end: |
2365 | return ret; | |
2366 | } | |
2367 | ||
2368 | static | |
9d1bbf21 | 2369 | void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd) |
b8aa1682 JD |
2370 | { |
2371 | int ret; | |
2372 | ||
b8aa1682 JD |
2373 | lttng_poll_del(events, pollfd); |
2374 | ||
2375 | ret = close(pollfd); | |
2376 | if (ret < 0) { | |
2377 | ERR("Closing pollfd %d", pollfd); | |
2378 | } | |
2379 | } | |
2380 | ||
2381 | static | |
2382 | int relay_add_connection(int fd, struct lttng_poll_event *events, | |
2383 | struct lttng_ht *relay_connections_ht) | |
2384 | { | |
b8aa1682 | 2385 | struct relay_command *relay_connection; |
6cd525e8 | 2386 | ssize_t ret; |
b8aa1682 JD |
2387 | |
2388 | relay_connection = zmalloc(sizeof(struct relay_command)); | |
2389 | if (relay_connection == NULL) { | |
2390 | PERROR("Relay command zmalloc"); | |
9d1bbf21 | 2391 | goto error; |
b8aa1682 | 2392 | } |
6cd525e8 MD |
2393 | ret = lttng_read(fd, relay_connection, sizeof(struct relay_command)); |
2394 | if (ret < sizeof(struct relay_command)) { | |
b8aa1682 | 2395 | PERROR("read relay cmd pipe"); |
9d1bbf21 | 2396 | goto error_read; |
b8aa1682 | 2397 | } |
a4baae1b | 2398 | CDS_INIT_LIST_HEAD(&relay_connection->recv_head); |
b8aa1682 JD |
2399 | |
2400 | lttng_ht_node_init_ulong(&relay_connection->sock_n, | |
2401 | (unsigned long) relay_connection->sock->fd); | |
9d1bbf21 | 2402 | rcu_read_lock(); |
b8aa1682 JD |
2403 | lttng_ht_add_unique_ulong(relay_connections_ht, |
2404 | &relay_connection->sock_n); | |
9d1bbf21 MD |
2405 | rcu_read_unlock(); |
2406 | return lttng_poll_add(events, | |
b8aa1682 JD |
2407 | relay_connection->sock->fd, |
2408 | LPOLLIN | LPOLLRDHUP); | |
2409 | ||
9d1bbf21 MD |
2410 | error_read: |
2411 | free(relay_connection); | |
2412 | error: | |
2413 | return -1; | |
2414 | } | |
2415 | ||
2416 | static | |
2417 | void deferred_free_connection(struct rcu_head *head) | |
2418 | { | |
2419 | struct relay_command *relay_connection = | |
2420 | caa_container_of(head, struct relay_command, rcu_node); | |
5b6d8097 DG |
2421 | |
2422 | lttcomm_destroy_sock(relay_connection->sock); | |
9d1bbf21 MD |
2423 | free(relay_connection); |
2424 | } | |
2425 | ||
2426 | static | |
2427 | void relay_del_connection(struct lttng_ht *relay_connections_ht, | |
d3e2ba59 JD |
2428 | struct lttng_ht_iter *iter, struct relay_command *relay_connection, |
2429 | struct lttng_ht *sessions_ht) | |
9d1bbf21 MD |
2430 | { |
2431 | int ret; | |
2432 | ||
2433 | ret = lttng_ht_del(relay_connections_ht, iter); | |
2434 | assert(!ret); | |
c07d8a78 | 2435 | |
9d1bbf21 | 2436 | if (relay_connection->type == RELAY_CONTROL) { |
2a174661 | 2437 | struct relay_stream *stream, *tmp_stream; |
a4baae1b JD |
2438 | |
2439 | /* Clean up recv list. */ | |
2a174661 DG |
2440 | cds_list_for_each_entry_safe(stream, tmp_stream, |
2441 | &relay_connection->recv_head, recv_list) { | |
2442 | cds_list_del(&stream->recv_list); | |
a4baae1b | 2443 | } |
2a174661 DG |
2444 | |
2445 | relay_delete_session(relay_connection, sessions_ht); | |
2446 | ||
9d1bbf21 | 2447 | } |
5b6d8097 | 2448 | |
c07d8a78 | 2449 | call_rcu(&relay_connection->rcu_node, deferred_free_connection); |
b8aa1682 JD |
2450 | } |
2451 | ||
2452 | /* | |
2453 | * This thread does the actual work | |
2454 | */ | |
2455 | static | |
2456 | void *relay_thread_worker(void *data) | |
2457 | { | |
beaad64c DG |
2458 | int ret, err = -1, last_seen_data_fd = -1; |
2459 | uint32_t nb_fd; | |
b8aa1682 JD |
2460 | struct relay_command *relay_connection; |
2461 | struct lttng_poll_event events; | |
2462 | struct lttng_ht *relay_connections_ht; | |
2463 | struct lttng_ht_node_ulong *node; | |
2464 | struct lttng_ht_iter iter; | |
b8aa1682 | 2465 | struct lttcomm_relayd_hdr recv_hdr; |
d3e2ba59 JD |
2466 | struct relay_local_data *relay_ctx = (struct relay_local_data *) data; |
2467 | struct lttng_ht *sessions_ht = relay_ctx->sessions_ht; | |
b8aa1682 JD |
2468 | |
2469 | DBG("[thread] Relay worker started"); | |
2470 | ||
9d1bbf21 MD |
2471 | rcu_register_thread(); |
2472 | ||
55706a7d MD |
2473 | health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER); |
2474 | ||
9b5e0863 MD |
2475 | if (testpoint(relayd_thread_worker)) { |
2476 | goto error_testpoint; | |
2477 | } | |
2478 | ||
f385ae0a MD |
2479 | health_code_update(); |
2480 | ||
b8aa1682 JD |
2481 | /* table of connections indexed on socket */ |
2482 | relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
095a4ae5 MD |
2483 | if (!relay_connections_ht) { |
2484 | goto relay_connections_ht_error; | |
2485 | } | |
b8aa1682 | 2486 | |
1c20f0e2 JD |
2487 | /* Tables of received indexes indexed by index handle and net_seq_num. */ |
2488 | indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64); | |
2489 | if (!indexes_ht) { | |
2490 | goto indexes_ht_error; | |
2491 | } | |
2492 | ||
b8aa1682 JD |
2493 | ret = create_thread_poll_set(&events, 2); |
2494 | if (ret < 0) { | |
2495 | goto error_poll_create; | |
2496 | } | |
2497 | ||
2498 | ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP); | |
2499 | if (ret < 0) { | |
2500 | goto error; | |
2501 | } | |
2502 | ||
beaad64c | 2503 | restart: |
b8aa1682 | 2504 | while (1) { |
beaad64c DG |
2505 | int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1; |
2506 | ||
f385ae0a MD |
2507 | health_code_update(); |
2508 | ||
b8aa1682 | 2509 | /* Infinite blocking call, waiting for transmission */ |
87c1611d | 2510 | DBG3("Relayd worker thread polling..."); |
f385ae0a | 2511 | health_poll_entry(); |
b8aa1682 | 2512 | ret = lttng_poll_wait(&events, -1); |
f385ae0a | 2513 | health_poll_exit(); |
b8aa1682 JD |
2514 | if (ret < 0) { |
2515 | /* | |
2516 | * Restart interrupted system call. | |
2517 | */ | |
2518 | if (errno == EINTR) { | |
2519 | goto restart; | |
2520 | } | |
2521 | goto error; | |
2522 | } | |
2523 | ||
0d9c5d77 DG |
2524 | nb_fd = ret; |
2525 | ||
beaad64c DG |
2526 | /* |
2527 | * Process control. The control connection is prioritised so we don't | |
2528 | * starve it with high throughout put tracing data on the data | |
2529 | * connection. | |
2530 | */ | |
b8aa1682 JD |
2531 | for (i = 0; i < nb_fd; i++) { |
2532 | /* Fetch once the poll data */ | |
beaad64c DG |
2533 | uint32_t revents = LTTNG_POLL_GETEV(&events, i); |
2534 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
b8aa1682 | 2535 | |
f385ae0a MD |
2536 | health_code_update(); |
2537 | ||
b8aa1682 JD |
2538 | /* Thread quit pipe has been closed. Killing thread. */ |
2539 | ret = check_thread_quit_pipe(pollfd, revents); | |
2540 | if (ret) { | |
095a4ae5 MD |
2541 | err = 0; |
2542 | goto exit; | |
b8aa1682 JD |
2543 | } |
2544 | ||
2545 | /* Inspect the relay cmd pipe for new connection */ | |
2546 | if (pollfd == relay_cmd_pipe[0]) { | |
2547 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
2548 | ERR("Relay pipe error"); | |
2549 | goto error; | |
2550 | } else if (revents & LPOLLIN) { | |
2551 | DBG("Relay command received"); | |
2552 | ret = relay_add_connection(relay_cmd_pipe[0], | |
2553 | &events, relay_connections_ht); | |
2554 | if (ret < 0) { | |
2555 | goto error; | |
2556 | } | |
2557 | } | |
beaad64c | 2558 | } else if (revents) { |
9d1bbf21 | 2559 | rcu_read_lock(); |
b8aa1682 JD |
2560 | lttng_ht_lookup(relay_connections_ht, |
2561 | (void *)((unsigned long) pollfd), | |
2562 | &iter); | |
2563 | node = lttng_ht_iter_get_node_ulong(&iter); | |
2564 | if (node == NULL) { | |
2565 | DBG2("Relay sock %d not found", pollfd); | |
9d1bbf21 | 2566 | rcu_read_unlock(); |
b8aa1682 JD |
2567 | goto error; |
2568 | } | |
2569 | relay_connection = caa_container_of(node, | |
2570 | struct relay_command, sock_n); | |
2571 | ||
2572 | if (revents & (LPOLLERR)) { | |
2573 | ERR("POLL ERROR"); | |
9d1bbf21 MD |
2574 | relay_cleanup_poll_connection(&events, pollfd); |
2575 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2576 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2577 | if (last_seen_data_fd == pollfd) { |
2578 | last_seen_data_fd = last_notdel_data_fd; | |
2579 | } | |
b8aa1682 JD |
2580 | } else if (revents & (LPOLLHUP | LPOLLRDHUP)) { |
2581 | DBG("Socket %d hung up", pollfd); | |
9d1bbf21 MD |
2582 | relay_cleanup_poll_connection(&events, pollfd); |
2583 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2584 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2585 | if (last_seen_data_fd == pollfd) { |
2586 | last_seen_data_fd = last_notdel_data_fd; | |
2587 | } | |
b8aa1682 JD |
2588 | } else if (revents & LPOLLIN) { |
2589 | /* control socket */ | |
2590 | if (relay_connection->type == RELAY_CONTROL) { | |
2591 | ret = relay_connection->sock->ops->recvmsg( | |
2592 | relay_connection->sock, &recv_hdr, | |
7c5aef62 | 2593 | sizeof(struct lttcomm_relayd_hdr), 0); |
b8aa1682 JD |
2594 | /* connection closed */ |
2595 | if (ret <= 0) { | |
9d1bbf21 MD |
2596 | relay_cleanup_poll_connection(&events, pollfd); |
2597 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2598 | &iter, relay_connection, sessions_ht); |
b8aa1682 JD |
2599 | DBG("Control connection closed with %d", pollfd); |
2600 | } else { | |
2601 | if (relay_connection->session) { | |
77c7c900 | 2602 | DBG2("Relay worker receiving data for session : %" PRIu64, |
b8aa1682 JD |
2603 | relay_connection->session->id); |
2604 | } | |
2605 | ret = relay_process_control(&recv_hdr, | |
d3e2ba59 | 2606 | relay_connection, relay_ctx); |
b8aa1682 | 2607 | if (ret < 0) { |
beaad64c | 2608 | /* Clear the session on error. */ |
9d1bbf21 MD |
2609 | relay_cleanup_poll_connection(&events, pollfd); |
2610 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2611 | &iter, relay_connection, sessions_ht); |
b8aa1682 JD |
2612 | DBG("Connection closed with %d", pollfd); |
2613 | } | |
beaad64c | 2614 | seen_control = 1; |
b8aa1682 | 2615 | } |
beaad64c DG |
2616 | } else { |
2617 | /* | |
2618 | * Flag the last seen data fd not deleted. It will be | |
2619 | * used as the last seen fd if any fd gets deleted in | |
2620 | * this first loop. | |
2621 | */ | |
2622 | last_notdel_data_fd = pollfd; | |
2623 | } | |
2624 | } | |
2625 | rcu_read_unlock(); | |
2626 | } | |
2627 | } | |
2628 | ||
2629 | /* | |
2630 | * The last loop handled a control request, go back to poll to make | |
2631 | * sure we prioritise the control socket. | |
2632 | */ | |
2633 | if (seen_control) { | |
2634 | continue; | |
2635 | } | |
2636 | ||
2637 | if (last_seen_data_fd >= 0) { | |
2638 | for (i = 0; i < nb_fd; i++) { | |
2639 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
f385ae0a MD |
2640 | |
2641 | health_code_update(); | |
2642 | ||
beaad64c DG |
2643 | if (last_seen_data_fd == pollfd) { |
2644 | idx = i; | |
2645 | break; | |
2646 | } | |
2647 | } | |
2648 | } | |
2649 | ||
2650 | /* Process data connection. */ | |
2651 | for (i = idx + 1; i < nb_fd; i++) { | |
2652 | /* Fetch the poll data. */ | |
2653 | uint32_t revents = LTTNG_POLL_GETEV(&events, i); | |
2654 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
2655 | ||
f385ae0a MD |
2656 | health_code_update(); |
2657 | ||
beaad64c DG |
2658 | /* Skip the command pipe. It's handled in the first loop. */ |
2659 | if (pollfd == relay_cmd_pipe[0]) { | |
2660 | continue; | |
2661 | } | |
2662 | ||
2663 | if (revents) { | |
2664 | rcu_read_lock(); | |
2665 | lttng_ht_lookup(relay_connections_ht, | |
2666 | (void *)((unsigned long) pollfd), | |
2667 | &iter); | |
2668 | node = lttng_ht_iter_get_node_ulong(&iter); | |
2669 | if (node == NULL) { | |
2670 | /* Skip it. Might be removed before. */ | |
2671 | rcu_read_unlock(); | |
2672 | continue; | |
2673 | } | |
2674 | relay_connection = caa_container_of(node, | |
2675 | struct relay_command, sock_n); | |
2676 | ||
2677 | if (revents & LPOLLIN) { | |
2678 | if (relay_connection->type != RELAY_DATA) { | |
2679 | continue; | |
2680 | } | |
2681 | ||
2a174661 DG |
2682 | ret = relay_process_data(relay_connection, |
2683 | sessions_ht); | |
beaad64c DG |
2684 | /* connection closed */ |
2685 | if (ret < 0) { | |
2686 | relay_cleanup_poll_connection(&events, pollfd); | |
2687 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2688 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2689 | DBG("Data connection closed with %d", pollfd); |
2690 | /* | |
2691 | * Every goto restart call sets the last seen fd where | |
2692 | * here we don't really care since we gracefully | |
2693 | * continue the loop after the connection is deleted. | |
2694 | */ | |
2695 | } else { | |
2696 | /* Keep last seen port. */ | |
2697 | last_seen_data_fd = pollfd; | |
2698 | rcu_read_unlock(); | |
2699 | goto restart; | |
b8aa1682 JD |
2700 | } |
2701 | } | |
9d1bbf21 | 2702 | rcu_read_unlock(); |
b8aa1682 JD |
2703 | } |
2704 | } | |
beaad64c | 2705 | last_seen_data_fd = -1; |
b8aa1682 JD |
2706 | } |
2707 | ||
f385ae0a MD |
2708 | /* Normal exit, no error */ |
2709 | ret = 0; | |
2710 | ||
095a4ae5 | 2711 | exit: |
b8aa1682 JD |
2712 | error: |
2713 | lttng_poll_clean(&events); | |
2714 | ||
2715 | /* empty the hash table and free the memory */ | |
9d1bbf21 | 2716 | rcu_read_lock(); |
b8aa1682 | 2717 | cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) { |
f385ae0a MD |
2718 | health_code_update(); |
2719 | ||
b8aa1682 JD |
2720 | node = lttng_ht_iter_get_node_ulong(&iter); |
2721 | if (node) { | |
2722 | relay_connection = caa_container_of(node, | |
2723 | struct relay_command, sock_n); | |
9d1bbf21 | 2724 | relay_del_connection(relay_connections_ht, |
d3e2ba59 | 2725 | &iter, relay_connection, sessions_ht); |
b8aa1682 | 2726 | } |
b8aa1682 | 2727 | } |
94d49140 | 2728 | rcu_read_unlock(); |
7d2f7452 DG |
2729 | error_poll_create: |
2730 | lttng_ht_destroy(indexes_ht); | |
1c20f0e2 | 2731 | indexes_ht_error: |
b8aa1682 | 2732 | lttng_ht_destroy(relay_connections_ht); |
095a4ae5 | 2733 | relay_connections_ht_error: |
6620da75 DG |
2734 | /* Close relay cmd pipes */ |
2735 | utils_close_pipe(relay_cmd_pipe); | |
095a4ae5 MD |
2736 | if (err) { |
2737 | DBG("Thread exited with error"); | |
2738 | } | |
b8aa1682 | 2739 | DBG("Worker thread cleanup complete"); |
095a4ae5 | 2740 | free(data_buffer); |
9b5e0863 | 2741 | error_testpoint: |
f385ae0a MD |
2742 | if (err) { |
2743 | health_error(); | |
2744 | ERR("Health error occurred in %s", __func__); | |
2745 | } | |
2746 | health_unregister(health_relayd); | |
9d1bbf21 | 2747 | rcu_unregister_thread(); |
f385ae0a | 2748 | stop_threads(); |
b8aa1682 JD |
2749 | return NULL; |
2750 | } | |
2751 | ||
2752 | /* | |
2753 | * Create the relay command pipe to wake thread_manage_apps. | |
2754 | * Closed in cleanup(). | |
2755 | */ | |
2756 | static int create_relay_cmd_pipe(void) | |
2757 | { | |
a02de639 | 2758 | int ret; |
b8aa1682 | 2759 | |
a02de639 | 2760 | ret = utils_create_pipe_cloexec(relay_cmd_pipe); |
b8aa1682 | 2761 | |
b8aa1682 JD |
2762 | return ret; |
2763 | } | |
2764 | ||
2765 | /* | |
2766 | * main | |
2767 | */ | |
2768 | int main(int argc, char **argv) | |
2769 | { | |
2770 | int ret = 0; | |
2771 | void *status; | |
d3e2ba59 | 2772 | struct relay_local_data *relay_ctx; |
b8aa1682 | 2773 | |
b8aa1682 JD |
2774 | /* Parse arguments */ |
2775 | progname = argv[0]; | |
cd60b05a | 2776 | if ((ret = set_options(argc, argv)) < 0) { |
a02de639 | 2777 | goto exit; |
b8aa1682 JD |
2778 | } |
2779 | ||
2780 | if ((ret = set_signal_handler()) < 0) { | |
2781 | goto exit; | |
2782 | } | |
2783 | ||
4d513a50 DG |
2784 | /* Try to create directory if -o, --output is specified. */ |
2785 | if (opt_output_path) { | |
994fa64f DG |
2786 | if (*opt_output_path != '/') { |
2787 | ERR("Please specify an absolute path for -o, --output PATH"); | |
2788 | goto exit; | |
2789 | } | |
2790 | ||
4d513a50 DG |
2791 | ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG); |
2792 | if (ret < 0) { | |
2793 | ERR("Unable to create %s", opt_output_path); | |
2794 | goto exit; | |
2795 | } | |
2796 | } | |
2797 | ||
b8aa1682 | 2798 | /* Daemonize */ |
b5218ffb | 2799 | if (opt_daemon || opt_background) { |
3fd27398 MD |
2800 | int i; |
2801 | ||
2802 | ret = lttng_daemonize(&child_ppid, &recv_child_signal, | |
2803 | !opt_background); | |
b8aa1682 | 2804 | if (ret < 0) { |
a02de639 | 2805 | goto exit; |
b8aa1682 | 2806 | } |
3fd27398 MD |
2807 | |
2808 | /* | |
2809 | * We are in the child. Make sure all other file | |
2810 | * descriptors are closed, in case we are called with | |
2811 | * more opened file descriptors than the standard ones. | |
2812 | */ | |
2813 | for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) { | |
2814 | (void) close(i); | |
2815 | } | |
2816 | } | |
2817 | ||
2818 | /* Create thread quit pipe */ | |
2819 | if ((ret = init_thread_quit_pipe()) < 0) { | |
2820 | goto error; | |
b8aa1682 JD |
2821 | } |
2822 | ||
1c20f0e2 JD |
2823 | /* We need those values for the file/dir creation. */ |
2824 | relayd_uid = getuid(); | |
2825 | relayd_gid = getgid(); | |
b8aa1682 | 2826 | |
1c20f0e2 JD |
2827 | /* Check if daemon is UID = 0 */ |
2828 | if (relayd_uid == 0) { | |
8d5c808e | 2829 | if (control_uri->port < 1024 || data_uri->port < 1024 || live_uri->port < 1024) { |
b8aa1682 JD |
2830 | ERR("Need to be root to use ports < 1024"); |
2831 | ret = -1; | |
a02de639 | 2832 | goto exit; |
b8aa1682 JD |
2833 | } |
2834 | } | |
2835 | ||
2836 | /* Setup the thread apps communication pipe. */ | |
2837 | if ((ret = create_relay_cmd_pipe()) < 0) { | |
2838 | goto exit; | |
2839 | } | |
2840 | ||
2841 | /* Init relay command queue. */ | |
2842 | cds_wfq_init(&relay_cmd_queue.queue); | |
2843 | ||
2844 | /* Set up max poll set size */ | |
2845 | lttng_poll_set_max_size(); | |
2846 | ||
554831e7 MD |
2847 | /* Initialize communication library */ |
2848 | lttcomm_init(); | |
87e45c13 | 2849 | lttcomm_inet_init(); |
554831e7 | 2850 | |
d3e2ba59 JD |
2851 | relay_ctx = zmalloc(sizeof(struct relay_local_data)); |
2852 | if (!relay_ctx) { | |
2853 | PERROR("relay_ctx"); | |
2854 | goto exit; | |
2855 | } | |
2856 | ||
2857 | /* tables of sessions indexed by session ID */ | |
2a174661 | 2858 | relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
d3e2ba59 JD |
2859 | if (!relay_ctx->sessions_ht) { |
2860 | goto exit_relay_ctx_sessions; | |
2861 | } | |
2862 | ||
2863 | /* tables of streams indexed by stream ID */ | |
2a174661 | 2864 | relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
d3e2ba59 JD |
2865 | if (!relay_streams_ht) { |
2866 | goto exit_relay_ctx_streams; | |
2867 | } | |
2868 | ||
2869 | /* tables of streams indexed by stream ID */ | |
92c6ca54 DG |
2870 | viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
2871 | if (!viewer_streams_ht) { | |
d3e2ba59 JD |
2872 | goto exit_relay_ctx_viewer_streams; |
2873 | } | |
2874 | ||
55706a7d MD |
2875 | /* Initialize thread health monitoring */ |
2876 | health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES); | |
2877 | if (!health_relayd) { | |
2878 | PERROR("health_app_create error"); | |
2879 | goto exit_health_app_create; | |
2880 | } | |
2881 | ||
65931c8b MD |
2882 | ret = utils_create_pipe(health_quit_pipe); |
2883 | if (ret < 0) { | |
2884 | goto error_health_pipe; | |
2885 | } | |
2886 | ||
2887 | /* Create thread to manage the client socket */ | |
2888 | ret = pthread_create(&health_thread, NULL, | |
2889 | thread_manage_health, (void *) NULL); | |
2890 | if (ret != 0) { | |
2891 | PERROR("pthread_create health"); | |
2892 | goto health_error; | |
2893 | } | |
2894 | ||
b8aa1682 JD |
2895 | /* Setup the dispatcher thread */ |
2896 | ret = pthread_create(&dispatcher_thread, NULL, | |
2897 | relay_thread_dispatcher, (void *) NULL); | |
2898 | if (ret != 0) { | |
2899 | PERROR("pthread_create dispatcher"); | |
2900 | goto exit_dispatcher; | |
2901 | } | |
2902 | ||
2903 | /* Setup the worker thread */ | |
2904 | ret = pthread_create(&worker_thread, NULL, | |
d3e2ba59 | 2905 | relay_thread_worker, (void *) relay_ctx); |
b8aa1682 JD |
2906 | if (ret != 0) { |
2907 | PERROR("pthread_create worker"); | |
2908 | goto exit_worker; | |
2909 | } | |
2910 | ||
2911 | /* Setup the listener thread */ | |
2912 | ret = pthread_create(&listener_thread, NULL, | |
2913 | relay_thread_listener, (void *) NULL); | |
2914 | if (ret != 0) { | |
2915 | PERROR("pthread_create listener"); | |
2916 | goto exit_listener; | |
2917 | } | |
2918 | ||
0b242f62 | 2919 | ret = live_start_threads(live_uri, relay_ctx); |
d3e2ba59 JD |
2920 | if (ret != 0) { |
2921 | ERR("Starting live viewer threads"); | |
50138f51 | 2922 | goto exit_live; |
d3e2ba59 JD |
2923 | } |
2924 | ||
50138f51 | 2925 | exit_live: |
b8aa1682 JD |
2926 | ret = pthread_join(listener_thread, &status); |
2927 | if (ret != 0) { | |
2928 | PERROR("pthread_join"); | |
2929 | goto error; /* join error, exit without cleanup */ | |
2930 | } | |
2931 | ||
50138f51 | 2932 | exit_listener: |
b8aa1682 JD |
2933 | ret = pthread_join(worker_thread, &status); |
2934 | if (ret != 0) { | |
2935 | PERROR("pthread_join"); | |
2936 | goto error; /* join error, exit without cleanup */ | |
2937 | } | |
2938 | ||
50138f51 | 2939 | exit_worker: |
b8aa1682 JD |
2940 | ret = pthread_join(dispatcher_thread, &status); |
2941 | if (ret != 0) { | |
2942 | PERROR("pthread_join"); | |
2943 | goto error; /* join error, exit without cleanup */ | |
2944 | } | |
42415026 | 2945 | |
50138f51 | 2946 | exit_dispatcher: |
65931c8b MD |
2947 | ret = pthread_join(health_thread, &status); |
2948 | if (ret != 0) { | |
2949 | PERROR("pthread_join health thread"); | |
2950 | goto error; /* join error, exit without cleanup */ | |
2951 | } | |
2952 | ||
bd11e201 MD |
2953 | /* |
2954 | * Stop live threads only after joining other threads. | |
2955 | */ | |
2956 | live_stop_threads(); | |
2957 | ||
65931c8b MD |
2958 | health_error: |
2959 | utils_close_pipe(health_quit_pipe); | |
2960 | ||
2961 | error_health_pipe: | |
55706a7d MD |
2962 | health_app_destroy(health_relayd); |
2963 | ||
2964 | exit_health_app_create: | |
92c6ca54 | 2965 | lttng_ht_destroy(viewer_streams_ht); |
d3e2ba59 JD |
2966 | |
2967 | exit_relay_ctx_viewer_streams: | |
2968 | lttng_ht_destroy(relay_streams_ht); | |
2969 | ||
2970 | exit_relay_ctx_streams: | |
2971 | lttng_ht_destroy(relay_ctx->sessions_ht); | |
2972 | ||
2973 | exit_relay_ctx_sessions: | |
2974 | free(relay_ctx); | |
b8aa1682 JD |
2975 | |
2976 | exit: | |
2977 | cleanup(); | |
2978 | if (!ret) { | |
2979 | exit(EXIT_SUCCESS); | |
2980 | } | |
a02de639 | 2981 | |
b8aa1682 JD |
2982 | error: |
2983 | exit(EXIT_FAILURE); | |
2984 | } |