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