Commit | Line | Data |
---|---|---|
b8aa1682 JD |
1 | /* |
2 | * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com> | |
3 | * David Goulet <dgoulet@efficios.com> | |
cd60b05a | 4 | * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com> |
b8aa1682 JD |
5 | * |
6 | * This program is free software; you can redistribute it and/or modify | |
7 | * it under the terms of the GNU General Public License, version 2 only, | |
8 | * as published by the Free Software Foundation. | |
9 | * | |
10 | * This program is distributed in the hope that it will be useful, but WITHOUT | |
11 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
12 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
13 | * more details. | |
14 | * | |
15 | * You should have received a copy of the GNU General Public License along | |
16 | * with this program; if not, write to the Free Software Foundation, Inc., | |
17 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. | |
18 | */ | |
19 | ||
20 | #define _GNU_SOURCE | |
21 | #include <getopt.h> | |
22 | #include <grp.h> | |
23 | #include <limits.h> | |
24 | #include <pthread.h> | |
25 | #include <signal.h> | |
26 | #include <stdio.h> | |
27 | #include <stdlib.h> | |
28 | #include <string.h> | |
29 | #include <sys/mman.h> | |
30 | #include <sys/mount.h> | |
31 | #include <sys/resource.h> | |
32 | #include <sys/socket.h> | |
33 | #include <sys/stat.h> | |
34 | #include <sys/types.h> | |
35 | #include <sys/wait.h> | |
173af62f | 36 | #include <inttypes.h> |
b8aa1682 JD |
37 | #include <urcu/futex.h> |
38 | #include <urcu/uatomic.h> | |
39 | #include <unistd.h> | |
40 | #include <fcntl.h> | |
41 | #include <config.h> | |
42 | ||
43 | #include <lttng/lttng.h> | |
44 | #include <common/common.h> | |
45 | #include <common/compat/poll.h> | |
46 | #include <common/compat/socket.h> | |
47 | #include <common/defaults.h> | |
3fd27398 | 48 | #include <common/daemonize.h> |
b8aa1682 JD |
49 | #include <common/futex.h> |
50 | #include <common/sessiond-comm/sessiond-comm.h> | |
51 | #include <common/sessiond-comm/inet.h> | |
b8aa1682 JD |
52 | #include <common/sessiond-comm/relayd.h> |
53 | #include <common/uri.h> | |
a02de639 | 54 | #include <common/utils.h> |
cd60b05a | 55 | #include <common/config/config.h> |
b8aa1682 | 56 | |
0f907de1 | 57 | #include "cmd.h" |
d3e2ba59 | 58 | #include "ctf-trace.h" |
1c20f0e2 | 59 | #include "index.h" |
0f907de1 | 60 | #include "utils.h" |
b8aa1682 | 61 | #include "lttng-relayd.h" |
d3e2ba59 | 62 | #include "live.h" |
55706a7d | 63 | #include "health-relayd.h" |
9b5e0863 | 64 | #include "testpoint.h" |
2f8f53af | 65 | #include "viewer-stream.h" |
b8aa1682 JD |
66 | |
67 | /* command line options */ | |
0f907de1 | 68 | char *opt_output_path; |
b5218ffb | 69 | static int opt_daemon, opt_background; |
3fd27398 MD |
70 | |
71 | /* | |
72 | * We need to wait for listener and live listener threads, as well as | |
73 | * health check thread, before being ready to signal readiness. | |
74 | */ | |
75 | #define NR_LTTNG_RELAY_READY 3 | |
76 | static int lttng_relay_ready = NR_LTTNG_RELAY_READY; | |
77 | static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */ | |
78 | static pid_t child_ppid; /* Internal parent PID use with daemonize. */ | |
79 | ||
095a4ae5 MD |
80 | static struct lttng_uri *control_uri; |
81 | static struct lttng_uri *data_uri; | |
d3e2ba59 | 82 | static struct lttng_uri *live_uri; |
b8aa1682 JD |
83 | |
84 | const char *progname; | |
b8aa1682 | 85 | |
65931c8b | 86 | const char *tracing_group_name = DEFAULT_TRACING_GROUP; |
cd60b05a JG |
87 | static int tracing_group_name_override; |
88 | ||
89 | const char * const config_section_name = "relayd"; | |
65931c8b | 90 | |
b8aa1682 JD |
91 | /* |
92 | * Quit pipe for all threads. This permits a single cancellation point | |
93 | * for all threads when receiving an event on the pipe. | |
94 | */ | |
0b242f62 | 95 | int thread_quit_pipe[2] = { -1, -1 }; |
b8aa1682 JD |
96 | |
97 | /* | |
98 | * This pipe is used to inform the worker thread that a command is queued and | |
99 | * ready to be processed. | |
100 | */ | |
101 | static int relay_cmd_pipe[2] = { -1, -1 }; | |
102 | ||
26c9d55e | 103 | /* Shared between threads */ |
b8aa1682 JD |
104 | static int dispatch_thread_exit; |
105 | ||
106 | static pthread_t listener_thread; | |
107 | static pthread_t dispatcher_thread; | |
108 | static pthread_t worker_thread; | |
65931c8b | 109 | static pthread_t health_thread; |
b8aa1682 | 110 | |
095a4ae5 MD |
111 | static uint64_t last_relay_stream_id; |
112 | static uint64_t last_relay_session_id; | |
b8aa1682 JD |
113 | |
114 | /* | |
115 | * Relay command queue. | |
116 | * | |
117 | * The relay_thread_listener and relay_thread_dispatcher communicate with this | |
118 | * queue. | |
119 | */ | |
120 | static struct relay_cmd_queue relay_cmd_queue; | |
121 | ||
122 | /* buffer allocated at startup, used to store the trace data */ | |
095a4ae5 MD |
123 | static char *data_buffer; |
124 | static unsigned int data_buffer_size; | |
b8aa1682 | 125 | |
1c20f0e2 JD |
126 | /* We need those values for the file/dir creation. */ |
127 | static uid_t relayd_uid; | |
128 | static gid_t relayd_gid; | |
129 | ||
d3e2ba59 JD |
130 | /* Global relay stream hash table. */ |
131 | struct lttng_ht *relay_streams_ht; | |
132 | ||
92c6ca54 DG |
133 | /* Global relay viewer stream hash table. */ |
134 | struct lttng_ht *viewer_streams_ht; | |
135 | ||
0a6518b0 DG |
136 | /* Global hash table that stores relay index object. */ |
137 | struct lttng_ht *indexes_ht; | |
138 | ||
55706a7d | 139 | /* Relayd health monitoring */ |
eea7556c | 140 | struct health_app *health_relayd; |
55706a7d | 141 | |
cd60b05a JG |
142 | static struct option long_options[] = { |
143 | { "control-port", 1, 0, 'C', }, | |
144 | { "data-port", 1, 0, 'D', }, | |
8d5c808e | 145 | { "live-port", 1, 0, 'L', }, |
cd60b05a | 146 | { "daemonize", 0, 0, 'd', }, |
b5218ffb | 147 | { "background", 0, 0, 'b', }, |
cd60b05a JG |
148 | { "group", 1, 0, 'g', }, |
149 | { "help", 0, 0, 'h', }, | |
150 | { "output", 1, 0, 'o', }, | |
151 | { "verbose", 0, 0, 'v', }, | |
152 | { "config", 1, 0, 'f' }, | |
153 | { NULL, 0, 0, 0, }, | |
154 | }; | |
155 | ||
156 | static const char *config_ignore_options[] = { "help", "config" }; | |
157 | ||
b8aa1682 JD |
158 | /* |
159 | * usage function on stderr | |
160 | */ | |
161 | static | |
162 | void usage(void) | |
163 | { | |
164 | fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname); | |
994fa64f DG |
165 | fprintf(stderr, " -h, --help Display this usage.\n"); |
166 | fprintf(stderr, " -d, --daemonize Start as a daemon.\n"); | |
b5218ffb | 167 | fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n"); |
994fa64f DG |
168 | fprintf(stderr, " -C, --control-port URL Control port listening.\n"); |
169 | fprintf(stderr, " -D, --data-port URL Data port listening.\n"); | |
8d5c808e | 170 | fprintf(stderr, " -L, --live-port URL Live view port listening.\n"); |
994fa64f DG |
171 | fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n"); |
172 | fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n"); | |
65931c8b | 173 | fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n"); |
cd60b05a | 174 | fprintf(stderr, " -f --config Load daemon configuration file\n"); |
b8aa1682 JD |
175 | } |
176 | ||
cd60b05a JG |
177 | /* |
178 | * Take an option from the getopt output and set it in the right variable to be | |
179 | * used later. | |
180 | * | |
181 | * Return 0 on success else a negative value. | |
182 | */ | |
b8aa1682 | 183 | static |
cd60b05a | 184 | int set_option(int opt, const char *arg, const char *optname) |
b8aa1682 | 185 | { |
cd60b05a JG |
186 | int ret; |
187 | ||
188 | switch (opt) { | |
189 | case 0: | |
190 | fprintf(stderr, "option %s", optname); | |
191 | if (arg) { | |
192 | fprintf(stderr, " with arg %s\n", arg); | |
193 | } | |
194 | break; | |
195 | case 'C': | |
196 | ret = uri_parse(arg, &control_uri); | |
197 | if (ret < 0) { | |
198 | ERR("Invalid control URI specified"); | |
199 | goto end; | |
200 | } | |
201 | if (control_uri->port == 0) { | |
202 | control_uri->port = DEFAULT_NETWORK_CONTROL_PORT; | |
203 | } | |
204 | break; | |
205 | case 'D': | |
206 | ret = uri_parse(arg, &data_uri); | |
207 | if (ret < 0) { | |
208 | ERR("Invalid data URI specified"); | |
209 | goto end; | |
210 | } | |
211 | if (data_uri->port == 0) { | |
212 | data_uri->port = DEFAULT_NETWORK_DATA_PORT; | |
213 | } | |
214 | break; | |
8d5c808e AM |
215 | case 'L': |
216 | ret = uri_parse(arg, &live_uri); | |
217 | if (ret < 0) { | |
218 | ERR("Invalid live URI specified"); | |
219 | goto end; | |
220 | } | |
221 | if (live_uri->port == 0) { | |
222 | live_uri->port = DEFAULT_NETWORK_VIEWER_PORT; | |
223 | } | |
224 | break; | |
cd60b05a JG |
225 | case 'd': |
226 | opt_daemon = 1; | |
227 | break; | |
b5218ffb MD |
228 | case 'b': |
229 | opt_background = 1; | |
230 | break; | |
cd60b05a JG |
231 | case 'g': |
232 | tracing_group_name = strdup(arg); | |
233 | tracing_group_name_override = 1; | |
234 | break; | |
235 | case 'h': | |
236 | usage(); | |
237 | exit(EXIT_FAILURE); | |
238 | case 'o': | |
239 | ret = asprintf(&opt_output_path, "%s", arg); | |
240 | if (ret < 0) { | |
241 | ret = -errno; | |
242 | PERROR("asprintf opt_output_path"); | |
243 | goto end; | |
244 | } | |
245 | break; | |
246 | case 'v': | |
247 | /* Verbose level can increase using multiple -v */ | |
248 | if (arg) { | |
249 | lttng_opt_verbose = config_parse_value(arg); | |
250 | } else { | |
251 | lttng_opt_verbose += 1; | |
252 | } | |
253 | break; | |
254 | default: | |
255 | /* Unknown option or other error. | |
256 | * Error is printed by getopt, just return */ | |
257 | ret = -1; | |
258 | goto end; | |
259 | } | |
260 | ||
261 | /* All good. */ | |
262 | ret = 0; | |
263 | ||
264 | end: | |
265 | return ret; | |
266 | } | |
267 | ||
268 | /* | |
269 | * config_entry_handler_cb used to handle options read from a config file. | |
270 | * See config_entry_handler_cb comment in common/config/config.h for the | |
271 | * return value conventions. | |
272 | */ | |
273 | static | |
274 | int config_entry_handler(const struct config_entry *entry, void *unused) | |
275 | { | |
276 | int ret = 0, i; | |
277 | ||
278 | if (!entry || !entry->name || !entry->value) { | |
279 | ret = -EINVAL; | |
280 | goto end; | |
281 | } | |
282 | ||
283 | /* Check if the option is to be ignored */ | |
284 | for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) { | |
285 | if (!strcmp(entry->name, config_ignore_options[i])) { | |
286 | goto end; | |
287 | } | |
288 | } | |
289 | ||
290 | for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) { | |
291 | /* Ignore if entry name is not fully matched. */ | |
292 | if (strcmp(entry->name, long_options[i].name)) { | |
293 | continue; | |
294 | } | |
295 | ||
296 | /* | |
297 | * If the option takes no argument on the command line, we have to | |
298 | * check if the value is "true". We support non-zero numeric values, | |
299 | * true, on and yes. | |
300 | */ | |
301 | if (!long_options[i].has_arg) { | |
302 | ret = config_parse_value(entry->value); | |
303 | if (ret <= 0) { | |
304 | if (ret) { | |
305 | WARN("Invalid configuration value \"%s\" for option %s", | |
306 | entry->value, entry->name); | |
307 | } | |
308 | /* False, skip boolean config option. */ | |
309 | goto end; | |
310 | } | |
311 | } | |
312 | ||
313 | ret = set_option(long_options[i].val, entry->value, entry->name); | |
314 | goto end; | |
315 | } | |
316 | ||
317 | WARN("Unrecognized option \"%s\" in daemon configuration file.", | |
318 | entry->name); | |
319 | ||
320 | end: | |
321 | return ret; | |
322 | } | |
323 | ||
324 | static | |
325 | int set_options(int argc, char **argv) | |
326 | { | |
327 | int c, ret = 0, option_index = 0; | |
328 | int orig_optopt = optopt, orig_optind = optind; | |
329 | char *default_address, *optstring; | |
330 | const char *config_path = NULL; | |
331 | ||
332 | optstring = utils_generate_optstring(long_options, | |
333 | sizeof(long_options) / sizeof(struct option)); | |
334 | if (!optstring) { | |
335 | ret = -ENOMEM; | |
336 | goto exit; | |
337 | } | |
338 | ||
339 | /* Check for the --config option */ | |
340 | ||
341 | while ((c = getopt_long(argc, argv, optstring, long_options, | |
342 | &option_index)) != -1) { | |
343 | if (c == '?') { | |
344 | ret = -EINVAL; | |
345 | goto exit; | |
346 | } else if (c != 'f') { | |
347 | continue; | |
348 | } | |
349 | ||
350 | config_path = utils_expand_path(optarg); | |
351 | if (!config_path) { | |
352 | ERR("Failed to resolve path: %s", optarg); | |
353 | } | |
354 | } | |
355 | ||
356 | ret = config_get_section_entries(config_path, config_section_name, | |
357 | config_entry_handler, NULL); | |
358 | if (ret) { | |
359 | if (ret > 0) { | |
360 | ERR("Invalid configuration option at line %i", ret); | |
361 | ret = -1; | |
362 | } | |
363 | goto exit; | |
364 | } | |
b8aa1682 | 365 | |
cd60b05a JG |
366 | /* Reset getopt's global state */ |
367 | optopt = orig_optopt; | |
368 | optind = orig_optind; | |
b8aa1682 | 369 | while (1) { |
cd60b05a | 370 | c = getopt_long(argc, argv, optstring, long_options, &option_index); |
b8aa1682 JD |
371 | if (c == -1) { |
372 | break; | |
373 | } | |
374 | ||
cd60b05a JG |
375 | ret = set_option(c, optarg, long_options[option_index].name); |
376 | if (ret < 0) { | |
b8aa1682 JD |
377 | goto exit; |
378 | } | |
379 | } | |
380 | ||
381 | /* assign default values */ | |
382 | if (control_uri == NULL) { | |
383 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
384 | DEFAULT_NETWORK_CONTROL_PORT); | |
385 | if (ret < 0) { | |
386 | PERROR("asprintf default data address"); | |
387 | goto exit; | |
388 | } | |
389 | ||
390 | ret = uri_parse(default_address, &control_uri); | |
391 | free(default_address); | |
392 | if (ret < 0) { | |
393 | ERR("Invalid control URI specified"); | |
394 | goto exit; | |
395 | } | |
396 | } | |
397 | if (data_uri == NULL) { | |
398 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
399 | DEFAULT_NETWORK_DATA_PORT); | |
400 | if (ret < 0) { | |
401 | PERROR("asprintf default data address"); | |
402 | goto exit; | |
403 | } | |
404 | ||
405 | ret = uri_parse(default_address, &data_uri); | |
406 | free(default_address); | |
407 | if (ret < 0) { | |
408 | ERR("Invalid data URI specified"); | |
409 | goto exit; | |
410 | } | |
411 | } | |
d3e2ba59 JD |
412 | if (live_uri == NULL) { |
413 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
414 | DEFAULT_NETWORK_VIEWER_PORT); | |
415 | if (ret < 0) { | |
416 | PERROR("asprintf default viewer control address"); | |
417 | goto exit; | |
418 | } | |
419 | ||
420 | ret = uri_parse(default_address, &live_uri); | |
421 | free(default_address); | |
422 | if (ret < 0) { | |
423 | ERR("Invalid viewer control URI specified"); | |
424 | goto exit; | |
425 | } | |
426 | } | |
b8aa1682 JD |
427 | |
428 | exit: | |
cd60b05a | 429 | free(optstring); |
b8aa1682 JD |
430 | return ret; |
431 | } | |
432 | ||
433 | /* | |
434 | * Cleanup the daemon | |
435 | */ | |
436 | static | |
437 | void cleanup(void) | |
438 | { | |
b8aa1682 JD |
439 | DBG("Cleaning up"); |
440 | ||
095a4ae5 MD |
441 | /* free the dynamically allocated opt_output_path */ |
442 | free(opt_output_path); | |
443 | ||
a02de639 CB |
444 | /* Close thread quit pipes */ |
445 | utils_close_pipe(thread_quit_pipe); | |
446 | ||
710c1f73 DG |
447 | uri_free(control_uri); |
448 | uri_free(data_uri); | |
8d5c808e | 449 | /* Live URI is freed in the live thread. */ |
cd60b05a JG |
450 | |
451 | if (tracing_group_name_override) { | |
452 | free((void *) tracing_group_name); | |
453 | } | |
b8aa1682 JD |
454 | } |
455 | ||
456 | /* | |
457 | * Write to writable pipe used to notify a thread. | |
458 | */ | |
459 | static | |
460 | int notify_thread_pipe(int wpipe) | |
461 | { | |
6cd525e8 | 462 | ssize_t ret; |
b8aa1682 | 463 | |
6cd525e8 MD |
464 | ret = lttng_write(wpipe, "!", 1); |
465 | if (ret < 1) { | |
b8aa1682 JD |
466 | PERROR("write poll pipe"); |
467 | } | |
468 | ||
469 | return ret; | |
470 | } | |
471 | ||
65931c8b MD |
472 | static void notify_health_quit_pipe(int *pipe) |
473 | { | |
6cd525e8 | 474 | ssize_t ret; |
65931c8b | 475 | |
6cd525e8 MD |
476 | ret = lttng_write(pipe[1], "4", 1); |
477 | if (ret < 1) { | |
65931c8b MD |
478 | PERROR("write relay health quit"); |
479 | } | |
480 | } | |
481 | ||
b8aa1682 JD |
482 | /* |
483 | * Stop all threads by closing the thread quit pipe. | |
484 | */ | |
485 | static | |
486 | void stop_threads(void) | |
487 | { | |
488 | int ret; | |
489 | ||
490 | /* Stopping all threads */ | |
491 | DBG("Terminating all threads"); | |
492 | ret = notify_thread_pipe(thread_quit_pipe[1]); | |
493 | if (ret < 0) { | |
494 | ERR("write error on thread quit pipe"); | |
495 | } | |
496 | ||
65931c8b MD |
497 | notify_health_quit_pipe(health_quit_pipe); |
498 | ||
b8aa1682 | 499 | /* Dispatch thread */ |
26c9d55e | 500 | CMM_STORE_SHARED(dispatch_thread_exit, 1); |
b8aa1682 JD |
501 | futex_nto1_wake(&relay_cmd_queue.futex); |
502 | } | |
503 | ||
504 | /* | |
505 | * Signal handler for the daemon | |
506 | * | |
507 | * Simply stop all worker threads, leaving main() return gracefully after | |
508 | * joining all threads and calling cleanup(). | |
509 | */ | |
510 | static | |
511 | void sighandler(int sig) | |
512 | { | |
513 | switch (sig) { | |
514 | case SIGPIPE: | |
515 | DBG("SIGPIPE caught"); | |
516 | return; | |
517 | case SIGINT: | |
518 | DBG("SIGINT caught"); | |
519 | stop_threads(); | |
520 | break; | |
521 | case SIGTERM: | |
522 | DBG("SIGTERM caught"); | |
523 | stop_threads(); | |
524 | break; | |
3fd27398 MD |
525 | case SIGUSR1: |
526 | CMM_STORE_SHARED(recv_child_signal, 1); | |
527 | break; | |
b8aa1682 JD |
528 | default: |
529 | break; | |
530 | } | |
531 | } | |
532 | ||
533 | /* | |
534 | * Setup signal handler for : | |
535 | * SIGINT, SIGTERM, SIGPIPE | |
536 | */ | |
537 | static | |
538 | int set_signal_handler(void) | |
539 | { | |
540 | int ret = 0; | |
541 | struct sigaction sa; | |
542 | sigset_t sigset; | |
543 | ||
544 | if ((ret = sigemptyset(&sigset)) < 0) { | |
545 | PERROR("sigemptyset"); | |
546 | return ret; | |
547 | } | |
548 | ||
549 | sa.sa_handler = sighandler; | |
550 | sa.sa_mask = sigset; | |
551 | sa.sa_flags = 0; | |
552 | if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) { | |
553 | PERROR("sigaction"); | |
554 | return ret; | |
555 | } | |
556 | ||
557 | if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) { | |
558 | PERROR("sigaction"); | |
559 | return ret; | |
560 | } | |
561 | ||
562 | if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) { | |
563 | PERROR("sigaction"); | |
564 | return ret; | |
565 | } | |
566 | ||
3fd27398 MD |
567 | if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) { |
568 | PERROR("sigaction"); | |
569 | return ret; | |
570 | } | |
571 | ||
572 | DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT"); | |
b8aa1682 JD |
573 | |
574 | return ret; | |
575 | } | |
576 | ||
3fd27398 MD |
577 | void lttng_relay_notify_ready(void) |
578 | { | |
579 | /* Notify the parent of the fork() process that we are ready. */ | |
580 | if (opt_daemon || opt_background) { | |
581 | if (uatomic_sub_return(<tng_relay_ready, 1) == 0) { | |
582 | kill(child_ppid, SIGUSR1); | |
583 | } | |
584 | } | |
585 | } | |
586 | ||
b8aa1682 JD |
587 | /* |
588 | * Init thread quit pipe. | |
589 | * | |
590 | * Return -1 on error or 0 if all pipes are created. | |
591 | */ | |
592 | static | |
593 | int init_thread_quit_pipe(void) | |
594 | { | |
a02de639 | 595 | int ret; |
b8aa1682 | 596 | |
a02de639 | 597 | ret = utils_create_pipe_cloexec(thread_quit_pipe); |
b8aa1682 | 598 | |
b8aa1682 JD |
599 | return ret; |
600 | } | |
601 | ||
602 | /* | |
603 | * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set. | |
604 | */ | |
605 | static | |
606 | int create_thread_poll_set(struct lttng_poll_event *events, int size) | |
607 | { | |
608 | int ret; | |
609 | ||
610 | if (events == NULL || size == 0) { | |
611 | ret = -1; | |
612 | goto error; | |
613 | } | |
614 | ||
615 | ret = lttng_poll_create(events, size, LTTNG_CLOEXEC); | |
616 | if (ret < 0) { | |
617 | goto error; | |
618 | } | |
619 | ||
620 | /* Add quit pipe */ | |
c7759e6a | 621 | ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR); |
b8aa1682 JD |
622 | if (ret < 0) { |
623 | goto error; | |
624 | } | |
625 | ||
626 | return 0; | |
627 | ||
628 | error: | |
629 | return ret; | |
630 | } | |
631 | ||
632 | /* | |
633 | * Check if the thread quit pipe was triggered. | |
634 | * | |
635 | * Return 1 if it was triggered else 0; | |
636 | */ | |
637 | static | |
638 | int check_thread_quit_pipe(int fd, uint32_t events) | |
639 | { | |
640 | if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) { | |
641 | return 1; | |
642 | } | |
643 | ||
644 | return 0; | |
645 | } | |
646 | ||
647 | /* | |
648 | * Create and init socket from uri. | |
649 | */ | |
650 | static | |
651 | struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri) | |
652 | { | |
653 | int ret; | |
654 | struct lttcomm_sock *sock = NULL; | |
655 | ||
656 | sock = lttcomm_alloc_sock_from_uri(uri); | |
657 | if (sock == NULL) { | |
658 | ERR("Allocating socket"); | |
659 | goto error; | |
660 | } | |
661 | ||
662 | ret = lttcomm_create_sock(sock); | |
663 | if (ret < 0) { | |
664 | goto error; | |
665 | } | |
666 | DBG("Listening on sock %d", sock->fd); | |
667 | ||
668 | ret = sock->ops->bind(sock); | |
669 | if (ret < 0) { | |
670 | goto error; | |
671 | } | |
672 | ||
673 | ret = sock->ops->listen(sock, -1); | |
674 | if (ret < 0) { | |
675 | goto error; | |
676 | ||
677 | } | |
678 | ||
679 | return sock; | |
680 | ||
681 | error: | |
682 | if (sock) { | |
683 | lttcomm_destroy_sock(sock); | |
684 | } | |
685 | return NULL; | |
686 | } | |
687 | ||
173af62f DG |
688 | /* |
689 | * Return nonzero if stream needs to be closed. | |
690 | */ | |
691 | static | |
692 | int close_stream_check(struct relay_stream *stream) | |
693 | { | |
694 | ||
695 | if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) { | |
f7079f67 DG |
696 | /* |
697 | * We are about to close the stream so set the data pending flag to 1 | |
698 | * which will make the end data pending command skip the stream which | |
699 | * is now closed and ready. Note that after proceeding to a file close, | |
700 | * the written file is ready for reading. | |
701 | */ | |
702 | stream->data_pending_check_done = 1; | |
173af62f DG |
703 | return 1; |
704 | } | |
705 | return 0; | |
706 | } | |
707 | ||
b8aa1682 JD |
708 | /* |
709 | * This thread manages the listening for new connections on the network | |
710 | */ | |
711 | static | |
712 | void *relay_thread_listener(void *data) | |
713 | { | |
095a4ae5 | 714 | int i, ret, pollfd, err = -1; |
b8aa1682 JD |
715 | int val = 1; |
716 | uint32_t revents, nb_fd; | |
717 | struct lttng_poll_event events; | |
718 | struct lttcomm_sock *control_sock, *data_sock; | |
719 | ||
b8aa1682 JD |
720 | DBG("[thread] Relay listener started"); |
721 | ||
55706a7d MD |
722 | health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER); |
723 | ||
f385ae0a MD |
724 | health_code_update(); |
725 | ||
b8aa1682 JD |
726 | control_sock = relay_init_sock(control_uri); |
727 | if (!control_sock) { | |
095a4ae5 | 728 | goto error_sock_control; |
b8aa1682 JD |
729 | } |
730 | ||
731 | data_sock = relay_init_sock(data_uri); | |
732 | if (!data_sock) { | |
095a4ae5 | 733 | goto error_sock_relay; |
b8aa1682 JD |
734 | } |
735 | ||
736 | /* | |
737 | * Pass 3 as size here for the thread quit pipe, control and data socket. | |
738 | */ | |
739 | ret = create_thread_poll_set(&events, 3); | |
740 | if (ret < 0) { | |
741 | goto error_create_poll; | |
742 | } | |
743 | ||
744 | /* Add the control socket */ | |
745 | ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP); | |
746 | if (ret < 0) { | |
747 | goto error_poll_add; | |
748 | } | |
749 | ||
750 | /* Add the data socket */ | |
751 | ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP); | |
752 | if (ret < 0) { | |
753 | goto error_poll_add; | |
754 | } | |
755 | ||
3fd27398 MD |
756 | lttng_relay_notify_ready(); |
757 | ||
9b5e0863 MD |
758 | if (testpoint(relayd_thread_listener)) { |
759 | goto error_testpoint; | |
760 | } | |
761 | ||
b8aa1682 | 762 | while (1) { |
f385ae0a MD |
763 | health_code_update(); |
764 | ||
b8aa1682 JD |
765 | DBG("Listener accepting connections"); |
766 | ||
b8aa1682 | 767 | restart: |
f385ae0a | 768 | health_poll_entry(); |
b8aa1682 | 769 | ret = lttng_poll_wait(&events, -1); |
f385ae0a | 770 | health_poll_exit(); |
b8aa1682 JD |
771 | if (ret < 0) { |
772 | /* | |
773 | * Restart interrupted system call. | |
774 | */ | |
775 | if (errno == EINTR) { | |
776 | goto restart; | |
777 | } | |
778 | goto error; | |
779 | } | |
780 | ||
0d9c5d77 DG |
781 | nb_fd = ret; |
782 | ||
b8aa1682 JD |
783 | DBG("Relay new connection received"); |
784 | for (i = 0; i < nb_fd; i++) { | |
f385ae0a MD |
785 | health_code_update(); |
786 | ||
b8aa1682 JD |
787 | /* Fetch once the poll data */ |
788 | revents = LTTNG_POLL_GETEV(&events, i); | |
789 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
790 | ||
791 | /* Thread quit pipe has been closed. Killing thread. */ | |
792 | ret = check_thread_quit_pipe(pollfd, revents); | |
793 | if (ret) { | |
095a4ae5 MD |
794 | err = 0; |
795 | goto exit; | |
b8aa1682 JD |
796 | } |
797 | ||
798 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
799 | ERR("socket poll error"); | |
800 | goto error; | |
801 | } else if (revents & LPOLLIN) { | |
4b7f17b2 MD |
802 | /* |
803 | * Get allocated in this thread, | |
804 | * enqueued to a global queue, dequeued | |
805 | * and freed in the worker thread. | |
806 | */ | |
807 | struct relay_command *relay_cmd; | |
808 | struct lttcomm_sock *newsock; | |
b8aa1682 JD |
809 | |
810 | relay_cmd = zmalloc(sizeof(struct relay_command)); | |
811 | if (relay_cmd == NULL) { | |
812 | PERROR("relay command zmalloc"); | |
813 | goto error; | |
814 | } | |
815 | ||
816 | if (pollfd == data_sock->fd) { | |
817 | newsock = data_sock->ops->accept(data_sock); | |
4b7f17b2 | 818 | if (!newsock) { |
b8aa1682 | 819 | PERROR("accepting data sock"); |
4b7f17b2 | 820 | free(relay_cmd); |
b8aa1682 JD |
821 | goto error; |
822 | } | |
823 | relay_cmd->type = RELAY_DATA; | |
824 | DBG("Relay data connection accepted, socket %d", newsock->fd); | |
4b7f17b2 MD |
825 | } else { |
826 | assert(pollfd == control_sock->fd); | |
b8aa1682 | 827 | newsock = control_sock->ops->accept(control_sock); |
4b7f17b2 | 828 | if (!newsock) { |
b8aa1682 | 829 | PERROR("accepting control sock"); |
4b7f17b2 | 830 | free(relay_cmd); |
b8aa1682 JD |
831 | goto error; |
832 | } | |
833 | relay_cmd->type = RELAY_CONTROL; | |
834 | DBG("Relay control connection accepted, socket %d", newsock->fd); | |
835 | } | |
836 | ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, | |
837 | &val, sizeof(int)); | |
838 | if (ret < 0) { | |
839 | PERROR("setsockopt inet"); | |
4b7f17b2 MD |
840 | lttcomm_destroy_sock(newsock); |
841 | free(relay_cmd); | |
b8aa1682 JD |
842 | goto error; |
843 | } | |
844 | relay_cmd->sock = newsock; | |
845 | /* | |
846 | * Lock free enqueue the request. | |
847 | */ | |
848 | cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node); | |
849 | ||
850 | /* | |
851 | * Wake the dispatch queue futex. Implicit memory | |
852 | * barrier with the exchange in cds_wfq_enqueue. | |
853 | */ | |
854 | futex_nto1_wake(&relay_cmd_queue.futex); | |
855 | } | |
856 | } | |
857 | } | |
858 | ||
095a4ae5 | 859 | exit: |
b8aa1682 JD |
860 | error: |
861 | error_poll_add: | |
9b5e0863 | 862 | error_testpoint: |
b8aa1682 JD |
863 | lttng_poll_clean(&events); |
864 | error_create_poll: | |
095a4ae5 MD |
865 | if (data_sock->fd >= 0) { |
866 | ret = data_sock->ops->close(data_sock); | |
b8aa1682 JD |
867 | if (ret) { |
868 | PERROR("close"); | |
869 | } | |
b8aa1682 | 870 | } |
095a4ae5 MD |
871 | lttcomm_destroy_sock(data_sock); |
872 | error_sock_relay: | |
873 | if (control_sock->fd >= 0) { | |
874 | ret = control_sock->ops->close(control_sock); | |
b8aa1682 JD |
875 | if (ret) { |
876 | PERROR("close"); | |
877 | } | |
b8aa1682 | 878 | } |
095a4ae5 MD |
879 | lttcomm_destroy_sock(control_sock); |
880 | error_sock_control: | |
881 | if (err) { | |
f385ae0a MD |
882 | health_error(); |
883 | ERR("Health error occurred in %s", __func__); | |
095a4ae5 | 884 | } |
55706a7d | 885 | health_unregister(health_relayd); |
b8aa1682 JD |
886 | DBG("Relay listener thread cleanup complete"); |
887 | stop_threads(); | |
b8aa1682 JD |
888 | return NULL; |
889 | } | |
890 | ||
891 | /* | |
892 | * This thread manages the dispatching of the requests to worker threads | |
893 | */ | |
894 | static | |
895 | void *relay_thread_dispatcher(void *data) | |
896 | { | |
6cd525e8 MD |
897 | int err = -1; |
898 | ssize_t ret; | |
b8aa1682 JD |
899 | struct cds_wfq_node *node; |
900 | struct relay_command *relay_cmd = NULL; | |
901 | ||
902 | DBG("[thread] Relay dispatcher started"); | |
903 | ||
55706a7d MD |
904 | health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER); |
905 | ||
9b5e0863 MD |
906 | if (testpoint(relayd_thread_dispatcher)) { |
907 | goto error_testpoint; | |
908 | } | |
909 | ||
f385ae0a MD |
910 | health_code_update(); |
911 | ||
26c9d55e | 912 | while (!CMM_LOAD_SHARED(dispatch_thread_exit)) { |
f385ae0a MD |
913 | health_code_update(); |
914 | ||
b8aa1682 JD |
915 | /* Atomically prepare the queue futex */ |
916 | futex_nto1_prepare(&relay_cmd_queue.futex); | |
917 | ||
918 | do { | |
f385ae0a MD |
919 | health_code_update(); |
920 | ||
b8aa1682 JD |
921 | /* Dequeue commands */ |
922 | node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue); | |
923 | if (node == NULL) { | |
924 | DBG("Woken up but nothing in the relay command queue"); | |
925 | /* Continue thread execution */ | |
926 | break; | |
927 | } | |
928 | ||
929 | relay_cmd = caa_container_of(node, struct relay_command, node); | |
930 | DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd); | |
931 | ||
932 | /* | |
933 | * Inform worker thread of the new request. This | |
934 | * call is blocking so we can be assured that the data will be read | |
935 | * at some point in time or wait to the end of the world :) | |
936 | */ | |
6cd525e8 MD |
937 | ret = lttng_write(relay_cmd_pipe[1], relay_cmd, |
938 | sizeof(struct relay_command)); | |
b8aa1682 | 939 | free(relay_cmd); |
6cd525e8 | 940 | if (ret < sizeof(struct relay_command)) { |
b8aa1682 JD |
941 | PERROR("write cmd pipe"); |
942 | goto error; | |
943 | } | |
944 | } while (node != NULL); | |
945 | ||
946 | /* Futex wait on queue. Blocking call on futex() */ | |
f385ae0a | 947 | health_poll_entry(); |
b8aa1682 | 948 | futex_nto1_wait(&relay_cmd_queue.futex); |
f385ae0a | 949 | health_poll_exit(); |
b8aa1682 JD |
950 | } |
951 | ||
f385ae0a MD |
952 | /* Normal exit, no error */ |
953 | err = 0; | |
954 | ||
b8aa1682 | 955 | error: |
9b5e0863 | 956 | error_testpoint: |
f385ae0a MD |
957 | if (err) { |
958 | health_error(); | |
959 | ERR("Health error occurred in %s", __func__); | |
960 | } | |
55706a7d | 961 | health_unregister(health_relayd); |
b8aa1682 JD |
962 | DBG("Dispatch thread dying"); |
963 | stop_threads(); | |
964 | return NULL; | |
965 | } | |
966 | ||
de91f48a DG |
967 | /* |
968 | * Get stream from stream id. | |
969 | * Need to be called with RCU read-side lock held. | |
970 | */ | |
d3e2ba59 | 971 | struct relay_stream *relay_stream_find_by_id(uint64_t stream_id) |
de91f48a DG |
972 | { |
973 | struct lttng_ht_node_ulong *node; | |
974 | struct lttng_ht_iter iter; | |
975 | struct relay_stream *ret; | |
976 | ||
d3e2ba59 | 977 | lttng_ht_lookup(relay_streams_ht, |
de91f48a DG |
978 | (void *)((unsigned long) stream_id), |
979 | &iter); | |
980 | node = lttng_ht_iter_get_node_ulong(&iter); | |
981 | if (node == NULL) { | |
982 | DBG("Relay stream %" PRIu64 " not found", stream_id); | |
983 | ret = NULL; | |
984 | goto end; | |
985 | } | |
986 | ||
987 | ret = caa_container_of(node, struct relay_stream, stream_n); | |
988 | ||
989 | end: | |
990 | return ret; | |
991 | } | |
992 | ||
9d1bbf21 MD |
993 | static |
994 | void deferred_free_stream(struct rcu_head *head) | |
995 | { | |
996 | struct relay_stream *stream = | |
997 | caa_container_of(head, struct relay_stream, rcu_node); | |
d3e2ba59 | 998 | |
0f907de1 JD |
999 | free(stream->path_name); |
1000 | free(stream->channel_name); | |
9d1bbf21 MD |
1001 | free(stream); |
1002 | } | |
1003 | ||
d3e2ba59 JD |
1004 | static |
1005 | void deferred_free_session(struct rcu_head *head) | |
1006 | { | |
1007 | struct relay_session *session = | |
1008 | caa_container_of(head, struct relay_session, rcu_node); | |
1009 | free(session); | |
1010 | } | |
1011 | ||
4c80d9a2 DG |
1012 | /* |
1013 | * Close a given stream. The stream is freed using a call RCU. | |
1014 | * | |
1015 | * RCU read side lock MUST be acquired. If NO close_stream_check() was called | |
1016 | * BEFORE the stream lock MUST be acquired. | |
1017 | */ | |
c07d8a78 | 1018 | static void destroy_stream(struct relay_stream *stream) |
94d49140 JD |
1019 | { |
1020 | int delret; | |
1021 | struct relay_viewer_stream *vstream; | |
1022 | struct lttng_ht_iter iter; | |
1023 | ||
1024 | assert(stream); | |
94d49140 JD |
1025 | |
1026 | delret = close(stream->fd); | |
1027 | if (delret < 0) { | |
1028 | PERROR("close stream"); | |
1029 | } | |
1030 | ||
1031 | if (stream->index_fd >= 0) { | |
1032 | delret = close(stream->index_fd); | |
1033 | if (delret < 0) { | |
1034 | PERROR("close stream index_fd"); | |
1035 | } | |
1036 | } | |
1037 | ||
2f8f53af | 1038 | vstream = viewer_stream_find_by_id(stream->stream_handle); |
94d49140 JD |
1039 | if (vstream) { |
1040 | /* | |
1041 | * Set the last good value into the viewer stream. This is done | |
1042 | * right before the stream gets deleted from the hash table. The | |
1043 | * lookup failure on the live thread side of a stream indicates | |
1044 | * that the viewer stream index received value should be used. | |
1045 | */ | |
cef0f7d5 | 1046 | pthread_mutex_lock(&stream->viewer_stream_rotation_lock); |
94d49140 | 1047 | vstream->total_index_received = stream->total_index_received; |
a020f610 | 1048 | vstream->tracefile_count_last = stream->tracefile_count_current; |
cef0f7d5 JD |
1049 | vstream->close_write_flag = 1; |
1050 | pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); | |
94d49140 JD |
1051 | } |
1052 | ||
efc8a87f DG |
1053 | /* Cleanup index of that stream. */ |
1054 | relay_index_destroy_by_stream_id(stream->stream_handle); | |
1055 | ||
94d49140 JD |
1056 | iter.iter.node = &stream->stream_n.node; |
1057 | delret = lttng_ht_del(relay_streams_ht, &iter); | |
1058 | assert(!delret); | |
1059 | iter.iter.node = &stream->ctf_trace_node.node; | |
c07d8a78 | 1060 | delret = lttng_ht_del(stream->ctf_traces_ht, &iter); |
94d49140 | 1061 | assert(!delret); |
157df586 JD |
1062 | |
1063 | if (stream->ctf_trace) { | |
1064 | ctf_trace_try_destroy(stream->ctf_trace); | |
1065 | } | |
1066 | ||
94d49140 JD |
1067 | call_rcu(&stream->rcu_node, deferred_free_stream); |
1068 | DBG("Closed tracefile %d from close stream", stream->fd); | |
1069 | } | |
1070 | ||
b8aa1682 JD |
1071 | /* |
1072 | * relay_delete_session: Free all memory associated with a session and | |
1073 | * close all the FDs | |
1074 | */ | |
1075 | static | |
d3e2ba59 JD |
1076 | void relay_delete_session(struct relay_command *cmd, |
1077 | struct lttng_ht *sessions_ht) | |
b8aa1682 JD |
1078 | { |
1079 | struct lttng_ht_iter iter; | |
1080 | struct lttng_ht_node_ulong *node; | |
1081 | struct relay_stream *stream; | |
1082 | int ret; | |
1083 | ||
095a4ae5 | 1084 | if (!cmd->session) { |
b8aa1682 | 1085 | return; |
095a4ae5 | 1086 | } |
b8aa1682 | 1087 | |
77c7c900 | 1088 | DBG("Relay deleting session %" PRIu64, cmd->session->id); |
5b6d8097 | 1089 | |
9d1bbf21 | 1090 | rcu_read_lock(); |
d3e2ba59 | 1091 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) { |
b8aa1682 | 1092 | node = lttng_ht_iter_get_node_ulong(&iter); |
4c80d9a2 DG |
1093 | if (!node) { |
1094 | continue; | |
b8aa1682 | 1095 | } |
4c80d9a2 DG |
1096 | stream = caa_container_of(node, struct relay_stream, stream_n); |
1097 | if (stream->session == cmd->session) { | |
c07d8a78 | 1098 | destroy_stream(stream); |
87b576ec JD |
1099 | cmd->session->stream_count--; |
1100 | assert(cmd->session->stream_count >= 0); | |
4c80d9a2 | 1101 | } |
b8aa1682 | 1102 | } |
4c80d9a2 DG |
1103 | |
1104 | /* Make this session not visible anymore. */ | |
d3e2ba59 JD |
1105 | iter.iter.node = &cmd->session->session_n.node; |
1106 | ret = lttng_ht_del(sessions_ht, &iter); | |
1107 | assert(!ret); | |
4c80d9a2 | 1108 | call_rcu(&cmd->session->rcu_node, deferred_free_session); |
9d1bbf21 | 1109 | rcu_read_unlock(); |
b8aa1682 JD |
1110 | } |
1111 | ||
1c20f0e2 JD |
1112 | /* |
1113 | * Copy index data from the control port to a given index object. | |
1114 | */ | |
1115 | static void copy_index_control_data(struct relay_index *index, | |
1116 | struct lttcomm_relayd_index *data) | |
1117 | { | |
1118 | assert(index); | |
1119 | assert(data); | |
1120 | ||
1121 | /* | |
1122 | * The index on disk is encoded in big endian, so we don't need to convert | |
1123 | * the data received on the network. The data_offset value is NEVER | |
1124 | * modified here and is updated by the data thread. | |
1125 | */ | |
1126 | index->index_data.packet_size = data->packet_size; | |
1127 | index->index_data.content_size = data->content_size; | |
1128 | index->index_data.timestamp_begin = data->timestamp_begin; | |
1129 | index->index_data.timestamp_end = data->timestamp_end; | |
1130 | index->index_data.events_discarded = data->events_discarded; | |
1131 | index->index_data.stream_id = data->stream_id; | |
1132 | } | |
1133 | ||
c5b6f4f0 DG |
1134 | /* |
1135 | * Handle the RELAYD_CREATE_SESSION command. | |
1136 | * | |
1137 | * On success, send back the session id or else return a negative value. | |
1138 | */ | |
1139 | static | |
1140 | int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 JD |
1141 | struct relay_command *cmd, |
1142 | struct lttng_ht *sessions_ht) | |
c5b6f4f0 DG |
1143 | { |
1144 | int ret = 0, send_ret; | |
1145 | struct relay_session *session; | |
1146 | struct lttcomm_relayd_status_session reply; | |
1147 | ||
1148 | assert(recv_hdr); | |
1149 | assert(cmd); | |
1150 | ||
1151 | memset(&reply, 0, sizeof(reply)); | |
1152 | ||
1153 | session = zmalloc(sizeof(struct relay_session)); | |
1154 | if (session == NULL) { | |
1155 | PERROR("relay session zmalloc"); | |
1156 | ret = -1; | |
1157 | goto error; | |
1158 | } | |
1159 | ||
1160 | session->id = ++last_relay_session_id; | |
1161 | session->sock = cmd->sock; | |
7d2f7452 DG |
1162 | session->minor = cmd->minor; |
1163 | session->major = cmd->major; | |
80e8027a | 1164 | pthread_mutex_init(&session->viewer_ready_lock, NULL); |
c5b6f4f0 DG |
1165 | cmd->session = session; |
1166 | ||
1167 | reply.session_id = htobe64(session->id); | |
1168 | ||
d3e2ba59 | 1169 | switch (cmd->minor) { |
719af1bb JD |
1170 | case 1: |
1171 | case 2: | |
1172 | case 3: | |
1173 | break; | |
d3e2ba59 JD |
1174 | case 4: /* LTTng sessiond 2.4 */ |
1175 | default: | |
1176 | ret = cmd_create_session_2_4(cmd, session); | |
1177 | break; | |
1178 | } | |
1179 | ||
1180 | lttng_ht_node_init_ulong(&session->session_n, | |
1181 | (unsigned long) session->id); | |
1182 | lttng_ht_add_unique_ulong(sessions_ht, | |
1183 | &session->session_n); | |
1184 | ||
c5b6f4f0 DG |
1185 | DBG("Created session %" PRIu64, session->id); |
1186 | ||
1187 | error: | |
1188 | if (ret < 0) { | |
1189 | reply.ret_code = htobe32(LTTNG_ERR_FATAL); | |
1190 | } else { | |
1191 | reply.ret_code = htobe32(LTTNG_OK); | |
1192 | } | |
1193 | ||
1194 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1195 | if (send_ret < 0) { | |
1196 | ERR("Relayd sending session id"); | |
4169f5ad | 1197 | ret = send_ret; |
c5b6f4f0 DG |
1198 | } |
1199 | ||
1200 | return ret; | |
1201 | } | |
1202 | ||
a4baae1b JD |
1203 | /* |
1204 | * When we have received all the streams and the metadata for a channel, | |
1205 | * we make them visible to the viewer threads. | |
1206 | */ | |
1207 | static | |
1208 | void set_viewer_ready_flag(struct relay_command *cmd) | |
1209 | { | |
1210 | struct relay_stream_recv_handle *node, *tmp_node; | |
1211 | ||
80e8027a JD |
1212 | pthread_mutex_lock(&cmd->session->viewer_ready_lock); |
1213 | ||
a4baae1b JD |
1214 | cds_list_for_each_entry_safe(node, tmp_node, &cmd->recv_head, node) { |
1215 | struct relay_stream *stream; | |
1216 | ||
1217 | rcu_read_lock(); | |
1218 | stream = relay_stream_find_by_id(node->id); | |
1219 | if (!stream) { | |
1220 | /* | |
1221 | * Stream is most probably being cleaned up by the data thread thus | |
1222 | * simply continue to the next one. | |
1223 | */ | |
b3bdb105 | 1224 | rcu_read_unlock(); |
a4baae1b JD |
1225 | continue; |
1226 | } | |
1227 | ||
a4baae1b JD |
1228 | stream->viewer_ready = 1; |
1229 | rcu_read_unlock(); | |
1230 | ||
1231 | /* Clean stream handle node. */ | |
1232 | cds_list_del(&node->node); | |
1233 | free(node); | |
1234 | } | |
1235 | ||
80e8027a | 1236 | pthread_mutex_unlock(&cmd->session->viewer_ready_lock); |
a4baae1b JD |
1237 | return; |
1238 | } | |
1239 | ||
1240 | /* | |
1241 | * Add a recv handle node to the connection recv list with the given stream | |
1242 | * handle. A new node is allocated thus must be freed when the node is deleted | |
1243 | * from the list. | |
1244 | */ | |
1245 | static void queue_stream_handle(uint64_t handle, struct relay_command *cmd) | |
1246 | { | |
1247 | struct relay_stream_recv_handle *node; | |
1248 | ||
1249 | assert(cmd); | |
1250 | ||
1251 | node = zmalloc(sizeof(*node)); | |
1252 | if (!node) { | |
1253 | PERROR("zmalloc queue stream handle"); | |
1254 | return; | |
1255 | } | |
1256 | ||
1257 | node->id = handle; | |
1258 | cds_list_add(&node->node, &cmd->recv_head); | |
1259 | } | |
1260 | ||
b8aa1682 JD |
1261 | /* |
1262 | * relay_add_stream: allocate a new stream for a session | |
1263 | */ | |
1264 | static | |
1265 | int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1266 | struct relay_command *cmd, struct lttng_ht *sessions_ht) |
b8aa1682 JD |
1267 | { |
1268 | struct relay_session *session = cmd->session; | |
b8aa1682 JD |
1269 | struct relay_stream *stream = NULL; |
1270 | struct lttcomm_relayd_status_stream reply; | |
b8aa1682 JD |
1271 | int ret, send_ret; |
1272 | ||
c5b6f4f0 | 1273 | if (!session || cmd->version_check_done == 0) { |
b8aa1682 JD |
1274 | ERR("Trying to add a stream before version check"); |
1275 | ret = -1; | |
1276 | goto end_no_session; | |
1277 | } | |
1278 | ||
b8aa1682 JD |
1279 | stream = zmalloc(sizeof(struct relay_stream)); |
1280 | if (stream == NULL) { | |
1281 | PERROR("relay stream zmalloc"); | |
1282 | ret = -1; | |
1283 | goto end_no_session; | |
1284 | } | |
1285 | ||
0f907de1 JD |
1286 | switch (cmd->minor) { |
1287 | case 1: /* LTTng sessiond 2.1 */ | |
1288 | ret = cmd_recv_stream_2_1(cmd, stream); | |
1289 | break; | |
1290 | case 2: /* LTTng sessiond 2.2 */ | |
1291 | default: | |
1292 | ret = cmd_recv_stream_2_2(cmd, stream); | |
1293 | break; | |
1294 | } | |
1295 | if (ret < 0) { | |
1296 | goto err_free_stream; | |
1297 | } | |
1298 | ||
9d1bbf21 | 1299 | rcu_read_lock(); |
b8aa1682 | 1300 | stream->stream_handle = ++last_relay_stream_id; |
173af62f | 1301 | stream->prev_seq = -1ULL; |
b8aa1682 | 1302 | stream->session = session; |
1c20f0e2 | 1303 | stream->index_fd = -1; |
d3e2ba59 JD |
1304 | stream->read_index_fd = -1; |
1305 | stream->ctf_trace = NULL; | |
1306 | pthread_mutex_init(&stream->lock, NULL); | |
b8aa1682 | 1307 | |
0f907de1 | 1308 | ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG); |
b8aa1682 | 1309 | if (ret < 0) { |
b8aa1682 JD |
1310 | ERR("relay creating output directory"); |
1311 | goto end; | |
1312 | } | |
1313 | ||
be96a7d1 DG |
1314 | /* |
1315 | * No need to use run_as API here because whatever we receives, the relayd | |
1316 | * uses its own credentials for the stream files. | |
1317 | */ | |
0f907de1 | 1318 | ret = utils_create_stream_file(stream->path_name, stream->channel_name, |
1c20f0e2 | 1319 | stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL); |
b8aa1682 | 1320 | if (ret < 0) { |
0f907de1 | 1321 | ERR("Create output file"); |
b8aa1682 JD |
1322 | goto end; |
1323 | } | |
b8aa1682 | 1324 | stream->fd = ret; |
0f907de1 JD |
1325 | if (stream->tracefile_size) { |
1326 | DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name); | |
1327 | } else { | |
1328 | DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name); | |
1329 | } | |
b8aa1682 | 1330 | |
d3e2ba59 JD |
1331 | if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) { |
1332 | stream->metadata_flag = 1; | |
1333 | /* | |
1334 | * When we receive a new metadata stream, we create a new | |
1335 | * ctf_trace and we assign this ctf_trace to all streams with | |
1336 | * the same path. | |
1337 | * | |
1338 | * If later on we receive a new stream for the same ctf_trace, | |
1339 | * we copy the information from the first hit in the HT to the | |
1340 | * new stream. | |
1341 | */ | |
1342 | stream->ctf_trace = ctf_trace_create(); | |
1343 | if (!stream->ctf_trace) { | |
1344 | ret = -1; | |
1345 | goto end; | |
1346 | } | |
1347 | stream->ctf_trace->refcount++; | |
1348 | stream->ctf_trace->metadata_stream = stream; | |
1349 | } | |
1350 | ctf_trace_assign(cmd->ctf_traces_ht, stream); | |
c07d8a78 | 1351 | stream->ctf_traces_ht = cmd->ctf_traces_ht; |
d3e2ba59 | 1352 | |
a4baae1b JD |
1353 | /* |
1354 | * Add the stream handle in the recv list of the connection. Once the end | |
1355 | * stream message is received, this list is emptied and streams are set | |
1356 | * with the viewer ready flag. | |
1357 | */ | |
601262d6 | 1358 | queue_stream_handle(stream->stream_handle, cmd); |
a4baae1b | 1359 | |
b8aa1682 JD |
1360 | lttng_ht_node_init_ulong(&stream->stream_n, |
1361 | (unsigned long) stream->stream_handle); | |
d3e2ba59 | 1362 | lttng_ht_add_unique_ulong(relay_streams_ht, |
b8aa1682 JD |
1363 | &stream->stream_n); |
1364 | ||
d3e2ba59 JD |
1365 | lttng_ht_node_init_str(&stream->ctf_trace_node, stream->path_name); |
1366 | lttng_ht_add_str(cmd->ctf_traces_ht, &stream->ctf_trace_node); | |
87b576ec | 1367 | session->stream_count++; |
d3e2ba59 | 1368 | |
1c20f0e2 JD |
1369 | DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name, |
1370 | stream->stream_handle); | |
b8aa1682 JD |
1371 | |
1372 | end: | |
5af40280 | 1373 | reply.handle = htobe64(stream->stream_handle); |
b8aa1682 JD |
1374 | /* send the session id to the client or a negative return code on error */ |
1375 | if (ret < 0) { | |
f73fabfd | 1376 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
5af40280 CB |
1377 | /* stream was not properly added to the ht, so free it */ |
1378 | free(stream); | |
b8aa1682 | 1379 | } else { |
f73fabfd | 1380 | reply.ret_code = htobe32(LTTNG_OK); |
b8aa1682 | 1381 | } |
5af40280 | 1382 | |
b8aa1682 JD |
1383 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, |
1384 | sizeof(struct lttcomm_relayd_status_stream), 0); | |
1385 | if (send_ret < 0) { | |
1386 | ERR("Relay sending stream id"); | |
4169f5ad | 1387 | ret = send_ret; |
b8aa1682 | 1388 | } |
9d1bbf21 | 1389 | rcu_read_unlock(); |
b8aa1682 JD |
1390 | |
1391 | end_no_session: | |
1392 | return ret; | |
0f907de1 JD |
1393 | |
1394 | err_free_stream: | |
1395 | free(stream->path_name); | |
1396 | free(stream->channel_name); | |
1397 | free(stream); | |
1398 | return ret; | |
b8aa1682 JD |
1399 | } |
1400 | ||
173af62f DG |
1401 | /* |
1402 | * relay_close_stream: close a specific stream | |
1403 | */ | |
1404 | static | |
1405 | int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, | |
92c6ca54 | 1406 | struct relay_command *cmd) |
173af62f | 1407 | { |
94d49140 | 1408 | int ret, send_ret; |
173af62f DG |
1409 | struct relay_session *session = cmd->session; |
1410 | struct lttcomm_relayd_close_stream stream_info; | |
1411 | struct lttcomm_relayd_generic_reply reply; | |
1412 | struct relay_stream *stream; | |
173af62f DG |
1413 | |
1414 | DBG("Close stream received"); | |
1415 | ||
c5b6f4f0 | 1416 | if (!session || cmd->version_check_done == 0) { |
173af62f DG |
1417 | ERR("Trying to close a stream before version check"); |
1418 | ret = -1; | |
1419 | goto end_no_session; | |
1420 | } | |
1421 | ||
1422 | ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info, | |
7c5aef62 | 1423 | sizeof(struct lttcomm_relayd_close_stream), 0); |
173af62f | 1424 | if (ret < sizeof(struct lttcomm_relayd_close_stream)) { |
a6cd2b97 DG |
1425 | if (ret == 0) { |
1426 | /* Orderly shutdown. Not necessary to print an error. */ | |
1427 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1428 | } else { | |
1429 | ERR("Relay didn't receive valid add_stream struct size : %d", ret); | |
1430 | } | |
173af62f DG |
1431 | ret = -1; |
1432 | goto end_no_session; | |
1433 | } | |
1434 | ||
1435 | rcu_read_lock(); | |
d3e2ba59 | 1436 | stream = relay_stream_find_by_id(be64toh(stream_info.stream_id)); |
173af62f DG |
1437 | if (!stream) { |
1438 | ret = -1; | |
1439 | goto end_unlock; | |
1440 | } | |
1441 | ||
8e2583a4 | 1442 | stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num); |
173af62f | 1443 | stream->close_flag = 1; |
87b576ec JD |
1444 | session->stream_count--; |
1445 | assert(session->stream_count >= 0); | |
173af62f DG |
1446 | |
1447 | if (close_stream_check(stream)) { | |
c07d8a78 | 1448 | destroy_stream(stream); |
173af62f DG |
1449 | } |
1450 | ||
1451 | end_unlock: | |
1452 | rcu_read_unlock(); | |
1453 | ||
1454 | if (ret < 0) { | |
f73fabfd | 1455 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
173af62f | 1456 | } else { |
f73fabfd | 1457 | reply.ret_code = htobe32(LTTNG_OK); |
173af62f DG |
1458 | } |
1459 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1460 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1461 | if (send_ret < 0) { | |
1462 | ERR("Relay sending stream id"); | |
4169f5ad | 1463 | ret = send_ret; |
173af62f DG |
1464 | } |
1465 | ||
1466 | end_no_session: | |
1467 | return ret; | |
1468 | } | |
1469 | ||
b8aa1682 JD |
1470 | /* |
1471 | * relay_unknown_command: send -1 if received unknown command | |
1472 | */ | |
1473 | static | |
1474 | void relay_unknown_command(struct relay_command *cmd) | |
1475 | { | |
1476 | struct lttcomm_relayd_generic_reply reply; | |
1477 | int ret; | |
1478 | ||
f73fabfd | 1479 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
b8aa1682 JD |
1480 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, |
1481 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1482 | if (ret < 0) { | |
1483 | ERR("Relay sending unknown command"); | |
1484 | } | |
1485 | } | |
1486 | ||
1487 | /* | |
1488 | * relay_start: send an acknowledgment to the client to tell if we are | |
1489 | * ready to receive data. We are ready if a session is established. | |
1490 | */ | |
1491 | static | |
1492 | int relay_start(struct lttcomm_relayd_hdr *recv_hdr, | |
1493 | struct relay_command *cmd) | |
1494 | { | |
f73fabfd | 1495 | int ret = htobe32(LTTNG_OK); |
b8aa1682 JD |
1496 | struct lttcomm_relayd_generic_reply reply; |
1497 | struct relay_session *session = cmd->session; | |
1498 | ||
1499 | if (!session) { | |
1500 | DBG("Trying to start the streaming without a session established"); | |
f73fabfd | 1501 | ret = htobe32(LTTNG_ERR_UNK); |
b8aa1682 JD |
1502 | } |
1503 | ||
1504 | reply.ret_code = ret; | |
1505 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1506 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1507 | if (ret < 0) { | |
1508 | ERR("Relay sending start ack"); | |
1509 | } | |
1510 | ||
1511 | return ret; | |
1512 | } | |
1513 | ||
1d4dfdef DG |
1514 | /* |
1515 | * Append padding to the file pointed by the file descriptor fd. | |
1516 | */ | |
1517 | static int write_padding_to_file(int fd, uint32_t size) | |
1518 | { | |
6cd525e8 | 1519 | ssize_t ret = 0; |
1d4dfdef DG |
1520 | char *zeros; |
1521 | ||
1522 | if (size == 0) { | |
1523 | goto end; | |
1524 | } | |
1525 | ||
1526 | zeros = zmalloc(size); | |
1527 | if (zeros == NULL) { | |
1528 | PERROR("zmalloc zeros for padding"); | |
1529 | ret = -1; | |
1530 | goto end; | |
1531 | } | |
1532 | ||
6cd525e8 MD |
1533 | ret = lttng_write(fd, zeros, size); |
1534 | if (ret < size) { | |
1d4dfdef DG |
1535 | PERROR("write padding to file"); |
1536 | } | |
1537 | ||
e986c7a1 DG |
1538 | free(zeros); |
1539 | ||
1d4dfdef DG |
1540 | end: |
1541 | return ret; | |
1542 | } | |
1543 | ||
b8aa1682 JD |
1544 | /* |
1545 | * relay_recv_metadata: receive the metada for the session. | |
1546 | */ | |
1547 | static | |
1548 | int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1549 | struct relay_command *cmd) |
b8aa1682 | 1550 | { |
f73fabfd | 1551 | int ret = htobe32(LTTNG_OK); |
6cd525e8 | 1552 | ssize_t size_ret; |
b8aa1682 JD |
1553 | struct relay_session *session = cmd->session; |
1554 | struct lttcomm_relayd_metadata_payload *metadata_struct; | |
1555 | struct relay_stream *metadata_stream; | |
1556 | uint64_t data_size, payload_size; | |
1557 | ||
1558 | if (!session) { | |
1559 | ERR("Metadata sent before version check"); | |
1560 | ret = -1; | |
1561 | goto end; | |
1562 | } | |
1563 | ||
f6416125 MD |
1564 | data_size = payload_size = be64toh(recv_hdr->data_size); |
1565 | if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) { | |
1566 | ERR("Incorrect data size"); | |
1567 | ret = -1; | |
1568 | goto end; | |
1569 | } | |
1570 | payload_size -= sizeof(struct lttcomm_relayd_metadata_payload); | |
1571 | ||
b8aa1682 | 1572 | if (data_buffer_size < data_size) { |
d7b3776f | 1573 | /* In case the realloc fails, we can free the memory */ |
c617c0c6 MD |
1574 | char *tmp_data_ptr; |
1575 | ||
1576 | tmp_data_ptr = realloc(data_buffer, data_size); | |
1577 | if (!tmp_data_ptr) { | |
b8aa1682 | 1578 | ERR("Allocating data buffer"); |
c617c0c6 | 1579 | free(data_buffer); |
b8aa1682 JD |
1580 | ret = -1; |
1581 | goto end; | |
1582 | } | |
c617c0c6 | 1583 | data_buffer = tmp_data_ptr; |
b8aa1682 JD |
1584 | data_buffer_size = data_size; |
1585 | } | |
1586 | memset(data_buffer, 0, data_size); | |
77c7c900 | 1587 | DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size); |
7c5aef62 | 1588 | ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0); |
b8aa1682 | 1589 | if (ret < 0 || ret != data_size) { |
a6cd2b97 DG |
1590 | if (ret == 0) { |
1591 | /* Orderly shutdown. Not necessary to print an error. */ | |
1592 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1593 | } else { | |
1594 | ERR("Relay didn't receive the whole metadata"); | |
1595 | } | |
b8aa1682 | 1596 | ret = -1; |
b8aa1682 JD |
1597 | goto end; |
1598 | } | |
1599 | metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer; | |
9d1bbf21 MD |
1600 | |
1601 | rcu_read_lock(); | |
d3e2ba59 JD |
1602 | metadata_stream = relay_stream_find_by_id( |
1603 | be64toh(metadata_struct->stream_id)); | |
b8aa1682 JD |
1604 | if (!metadata_stream) { |
1605 | ret = -1; | |
9d1bbf21 | 1606 | goto end_unlock; |
b8aa1682 JD |
1607 | } |
1608 | ||
6cd525e8 MD |
1609 | size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload, |
1610 | payload_size); | |
1611 | if (size_ret < payload_size) { | |
b8aa1682 JD |
1612 | ERR("Relay error writing metadata on file"); |
1613 | ret = -1; | |
9d1bbf21 | 1614 | goto end_unlock; |
b8aa1682 | 1615 | } |
1d4dfdef DG |
1616 | |
1617 | ret = write_padding_to_file(metadata_stream->fd, | |
1618 | be32toh(metadata_struct->padding_size)); | |
1619 | if (ret < 0) { | |
1620 | goto end_unlock; | |
1621 | } | |
d3e2ba59 JD |
1622 | metadata_stream->ctf_trace->metadata_received += |
1623 | payload_size + be32toh(metadata_struct->padding_size); | |
1d4dfdef | 1624 | |
b8aa1682 JD |
1625 | DBG2("Relay metadata written"); |
1626 | ||
9d1bbf21 | 1627 | end_unlock: |
6e3c5836 | 1628 | rcu_read_unlock(); |
b8aa1682 JD |
1629 | end: |
1630 | return ret; | |
1631 | } | |
1632 | ||
1633 | /* | |
1634 | * relay_send_version: send relayd version number | |
1635 | */ | |
1636 | static | |
1637 | int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1638 | struct relay_command *cmd, struct lttng_ht *sessions_ht) |
b8aa1682 | 1639 | { |
7f51dcba | 1640 | int ret; |
092b6259 | 1641 | struct lttcomm_relayd_version reply, msg; |
b8aa1682 | 1642 | |
c5b6f4f0 DG |
1643 | assert(cmd); |
1644 | ||
1645 | cmd->version_check_done = 1; | |
b8aa1682 | 1646 | |
092b6259 | 1647 | /* Get version from the other side. */ |
7c5aef62 | 1648 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); |
092b6259 | 1649 | if (ret < 0 || ret != sizeof(msg)) { |
a6cd2b97 DG |
1650 | if (ret == 0) { |
1651 | /* Orderly shutdown. Not necessary to print an error. */ | |
1652 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1653 | } else { | |
1654 | ERR("Relay failed to receive the version values."); | |
1655 | } | |
092b6259 | 1656 | ret = -1; |
092b6259 DG |
1657 | goto end; |
1658 | } | |
1659 | ||
d83a952c MD |
1660 | reply.major = RELAYD_VERSION_COMM_MAJOR; |
1661 | reply.minor = RELAYD_VERSION_COMM_MINOR; | |
d4519fa3 JD |
1662 | |
1663 | /* Major versions must be the same */ | |
1664 | if (reply.major != be32toh(msg.major)) { | |
6151a90f JD |
1665 | DBG("Incompatible major versions (%u vs %u), deleting session", |
1666 | reply.major, be32toh(msg.major)); | |
d3e2ba59 | 1667 | relay_delete_session(cmd, sessions_ht); |
d4519fa3 JD |
1668 | ret = 0; |
1669 | goto end; | |
1670 | } | |
1671 | ||
0f907de1 JD |
1672 | cmd->major = reply.major; |
1673 | /* We adapt to the lowest compatible version */ | |
1674 | if (reply.minor <= be32toh(msg.minor)) { | |
1675 | cmd->minor = reply.minor; | |
1676 | } else { | |
1677 | cmd->minor = be32toh(msg.minor); | |
1678 | } | |
1679 | ||
6151a90f JD |
1680 | reply.major = htobe32(reply.major); |
1681 | reply.minor = htobe32(reply.minor); | |
1682 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1683 | sizeof(struct lttcomm_relayd_version), 0); | |
1684 | if (ret < 0) { | |
1685 | ERR("Relay sending version"); | |
1686 | } | |
1687 | ||
0f907de1 JD |
1688 | DBG("Version check done using protocol %u.%u", cmd->major, |
1689 | cmd->minor); | |
b8aa1682 JD |
1690 | |
1691 | end: | |
1692 | return ret; | |
1693 | } | |
1694 | ||
c8f59ee5 | 1695 | /* |
6d805429 | 1696 | * Check for data pending for a given stream id from the session daemon. |
c8f59ee5 DG |
1697 | */ |
1698 | static | |
6d805429 | 1699 | int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, |
d3e2ba59 | 1700 | struct relay_command *cmd) |
c8f59ee5 DG |
1701 | { |
1702 | struct relay_session *session = cmd->session; | |
6d805429 | 1703 | struct lttcomm_relayd_data_pending msg; |
c8f59ee5 DG |
1704 | struct lttcomm_relayd_generic_reply reply; |
1705 | struct relay_stream *stream; | |
1706 | int ret; | |
c8f59ee5 DG |
1707 | uint64_t last_net_seq_num, stream_id; |
1708 | ||
6d805429 | 1709 | DBG("Data pending command received"); |
c8f59ee5 | 1710 | |
c5b6f4f0 | 1711 | if (!session || cmd->version_check_done == 0) { |
c8f59ee5 DG |
1712 | ERR("Trying to check for data before version check"); |
1713 | ret = -1; | |
1714 | goto end_no_session; | |
1715 | } | |
1716 | ||
7c5aef62 | 1717 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); |
c8f59ee5 | 1718 | if (ret < sizeof(msg)) { |
a6cd2b97 DG |
1719 | if (ret == 0) { |
1720 | /* Orderly shutdown. Not necessary to print an error. */ | |
1721 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1722 | } else { | |
1723 | ERR("Relay didn't receive valid data_pending struct size : %d", | |
1724 | ret); | |
1725 | } | |
c8f59ee5 DG |
1726 | ret = -1; |
1727 | goto end_no_session; | |
1728 | } | |
1729 | ||
1730 | stream_id = be64toh(msg.stream_id); | |
1731 | last_net_seq_num = be64toh(msg.last_net_seq_num); | |
1732 | ||
1733 | rcu_read_lock(); | |
d3e2ba59 | 1734 | stream = relay_stream_find_by_id(stream_id); |
de91f48a | 1735 | if (stream == NULL) { |
c8f59ee5 DG |
1736 | ret = -1; |
1737 | goto end_unlock; | |
1738 | } | |
1739 | ||
6d805429 | 1740 | DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64 |
c8f59ee5 DG |
1741 | " and last_seq %" PRIu64, stream_id, stream->prev_seq, |
1742 | last_net_seq_num); | |
1743 | ||
33832e64 | 1744 | /* Avoid wrapping issue */ |
39df6d9f | 1745 | if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) { |
6d805429 | 1746 | /* Data has in fact been written and is NOT pending */ |
c8f59ee5 | 1747 | ret = 0; |
6d805429 DG |
1748 | } else { |
1749 | /* Data still being streamed thus pending */ | |
1750 | ret = 1; | |
c8f59ee5 DG |
1751 | } |
1752 | ||
f7079f67 DG |
1753 | /* Pending check is now done. */ |
1754 | stream->data_pending_check_done = 1; | |
1755 | ||
c8f59ee5 DG |
1756 | end_unlock: |
1757 | rcu_read_unlock(); | |
1758 | ||
1759 | reply.ret_code = htobe32(ret); | |
1760 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1761 | if (ret < 0) { | |
6d805429 | 1762 | ERR("Relay data pending ret code failed"); |
c8f59ee5 DG |
1763 | } |
1764 | ||
1765 | end_no_session: | |
1766 | return ret; | |
1767 | } | |
1768 | ||
1769 | /* | |
1770 | * Wait for the control socket to reach a quiescent state. | |
1771 | * | |
1772 | * Note that for now, when receiving this command from the session daemon, this | |
1773 | * means that every subsequent commands or data received on the control socket | |
1774 | * has been handled. So, this is why we simply return OK here. | |
1775 | */ | |
1776 | static | |
1777 | int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1778 | struct relay_command *cmd) |
c8f59ee5 DG |
1779 | { |
1780 | int ret; | |
ad7051c0 DG |
1781 | uint64_t stream_id; |
1782 | struct relay_stream *stream; | |
1783 | struct lttng_ht_iter iter; | |
1784 | struct lttcomm_relayd_quiescent_control msg; | |
c8f59ee5 DG |
1785 | struct lttcomm_relayd_generic_reply reply; |
1786 | ||
1787 | DBG("Checking quiescent state on control socket"); | |
1788 | ||
ad7051c0 DG |
1789 | if (!cmd->session || cmd->version_check_done == 0) { |
1790 | ERR("Trying to check for data before version check"); | |
1791 | ret = -1; | |
1792 | goto end_no_session; | |
1793 | } | |
1794 | ||
1795 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1796 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1797 | if (ret == 0) { |
1798 | /* Orderly shutdown. Not necessary to print an error. */ | |
1799 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1800 | } else { | |
1801 | ERR("Relay didn't receive valid begin data_pending struct size: %d", | |
1802 | ret); | |
1803 | } | |
ad7051c0 DG |
1804 | ret = -1; |
1805 | goto end_no_session; | |
1806 | } | |
1807 | ||
1808 | stream_id = be64toh(msg.stream_id); | |
1809 | ||
1810 | rcu_read_lock(); | |
d3e2ba59 JD |
1811 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
1812 | stream_n.node) { | |
ad7051c0 DG |
1813 | if (stream->stream_handle == stream_id) { |
1814 | stream->data_pending_check_done = 1; | |
1815 | DBG("Relay quiescent control pending flag set to %" PRIu64, | |
1816 | stream_id); | |
1817 | break; | |
1818 | } | |
1819 | } | |
1820 | rcu_read_unlock(); | |
1821 | ||
c8f59ee5 DG |
1822 | reply.ret_code = htobe32(LTTNG_OK); |
1823 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1824 | if (ret < 0) { | |
6d805429 | 1825 | ERR("Relay data quiescent control ret code failed"); |
c8f59ee5 DG |
1826 | } |
1827 | ||
ad7051c0 | 1828 | end_no_session: |
c8f59ee5 DG |
1829 | return ret; |
1830 | } | |
1831 | ||
f7079f67 DG |
1832 | /* |
1833 | * Initialize a data pending command. This means that a client is about to ask | |
1834 | * for data pending for each stream he/she holds. Simply iterate over all | |
1835 | * streams of a session and set the data_pending_check_done flag. | |
1836 | * | |
1837 | * This command returns to the client a LTTNG_OK code. | |
1838 | */ | |
1839 | static | |
1840 | int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1841 | struct relay_command *cmd) |
f7079f67 DG |
1842 | { |
1843 | int ret; | |
1844 | struct lttng_ht_iter iter; | |
1845 | struct lttcomm_relayd_begin_data_pending msg; | |
1846 | struct lttcomm_relayd_generic_reply reply; | |
1847 | struct relay_stream *stream; | |
1848 | uint64_t session_id; | |
1849 | ||
1850 | assert(recv_hdr); | |
1851 | assert(cmd); | |
f7079f67 DG |
1852 | |
1853 | DBG("Init streams for data pending"); | |
1854 | ||
1855 | if (!cmd->session || cmd->version_check_done == 0) { | |
1856 | ERR("Trying to check for data before version check"); | |
1857 | ret = -1; | |
1858 | goto end_no_session; | |
1859 | } | |
1860 | ||
1861 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1862 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1863 | if (ret == 0) { |
1864 | /* Orderly shutdown. Not necessary to print an error. */ | |
1865 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1866 | } else { | |
1867 | ERR("Relay didn't receive valid begin data_pending struct size: %d", | |
1868 | ret); | |
1869 | } | |
f7079f67 DG |
1870 | ret = -1; |
1871 | goto end_no_session; | |
1872 | } | |
1873 | ||
1874 | session_id = be64toh(msg.session_id); | |
1875 | ||
1876 | /* | |
1877 | * Iterate over all streams to set the begin data pending flag. For now, the | |
1878 | * streams are indexed by stream handle so we have to iterate over all | |
1879 | * streams to find the one associated with the right session_id. | |
1880 | */ | |
1881 | rcu_read_lock(); | |
d3e2ba59 JD |
1882 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
1883 | stream_n.node) { | |
f7079f67 DG |
1884 | if (stream->session->id == session_id) { |
1885 | stream->data_pending_check_done = 0; | |
1886 | DBG("Set begin data pending flag to stream %" PRIu64, | |
1887 | stream->stream_handle); | |
1888 | } | |
1889 | } | |
1890 | rcu_read_unlock(); | |
1891 | ||
1892 | /* All good, send back reply. */ | |
1893 | reply.ret_code = htobe32(LTTNG_OK); | |
1894 | ||
1895 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1896 | if (ret < 0) { | |
1897 | ERR("Relay begin data pending send reply failed"); | |
1898 | } | |
1899 | ||
1900 | end_no_session: | |
1901 | return ret; | |
1902 | } | |
1903 | ||
1904 | /* | |
1905 | * End data pending command. This will check, for a given session id, if each | |
1906 | * stream associated with it has its data_pending_check_done flag set. If not, | |
1907 | * this means that the client lost track of the stream but the data is still | |
1908 | * being streamed on our side. In this case, we inform the client that data is | |
1909 | * inflight. | |
1910 | * | |
1911 | * Return to the client if there is data in flight or not with a ret_code. | |
1912 | */ | |
1913 | static | |
1914 | int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1915 | struct relay_command *cmd) |
f7079f67 DG |
1916 | { |
1917 | int ret; | |
1918 | struct lttng_ht_iter iter; | |
1919 | struct lttcomm_relayd_end_data_pending msg; | |
1920 | struct lttcomm_relayd_generic_reply reply; | |
1921 | struct relay_stream *stream; | |
1922 | uint64_t session_id; | |
1923 | uint32_t is_data_inflight = 0; | |
1924 | ||
1925 | assert(recv_hdr); | |
1926 | assert(cmd); | |
f7079f67 DG |
1927 | |
1928 | DBG("End data pending command"); | |
1929 | ||
1930 | if (!cmd->session || cmd->version_check_done == 0) { | |
1931 | ERR("Trying to check for data before version check"); | |
1932 | ret = -1; | |
1933 | goto end_no_session; | |
1934 | } | |
1935 | ||
1936 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1937 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1938 | if (ret == 0) { |
1939 | /* Orderly shutdown. Not necessary to print an error. */ | |
1940 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1941 | } else { | |
1942 | ERR("Relay didn't receive valid end data_pending struct size: %d", | |
1943 | ret); | |
1944 | } | |
f7079f67 DG |
1945 | ret = -1; |
1946 | goto end_no_session; | |
1947 | } | |
1948 | ||
1949 | session_id = be64toh(msg.session_id); | |
1950 | ||
1951 | /* Iterate over all streams to see if the begin data pending flag is set. */ | |
1952 | rcu_read_lock(); | |
d3e2ba59 JD |
1953 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
1954 | stream_n.node) { | |
f7079f67 DG |
1955 | if (stream->session->id == session_id && |
1956 | !stream->data_pending_check_done) { | |
1957 | is_data_inflight = 1; | |
1958 | DBG("Data is still in flight for stream %" PRIu64, | |
1959 | stream->stream_handle); | |
1960 | break; | |
1961 | } | |
1962 | } | |
1963 | rcu_read_unlock(); | |
1964 | ||
1965 | /* All good, send back reply. */ | |
1966 | reply.ret_code = htobe32(is_data_inflight); | |
1967 | ||
1968 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1969 | if (ret < 0) { | |
1970 | ERR("Relay end data pending send reply failed"); | |
1971 | } | |
1972 | ||
1973 | end_no_session: | |
1974 | return ret; | |
1975 | } | |
1976 | ||
1c20f0e2 JD |
1977 | /* |
1978 | * Receive an index for a specific stream. | |
1979 | * | |
1980 | * Return 0 on success else a negative value. | |
1981 | */ | |
1982 | static | |
1983 | int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr, | |
0a6518b0 | 1984 | struct relay_command *cmd) |
1c20f0e2 JD |
1985 | { |
1986 | int ret, send_ret, index_created = 0; | |
1987 | struct relay_session *session = cmd->session; | |
1988 | struct lttcomm_relayd_index index_info; | |
1989 | struct relay_index *index, *wr_index = NULL; | |
1990 | struct lttcomm_relayd_generic_reply reply; | |
1991 | struct relay_stream *stream; | |
1992 | uint64_t net_seq_num; | |
1993 | ||
1994 | assert(cmd); | |
1c20f0e2 JD |
1995 | |
1996 | DBG("Relay receiving index"); | |
1997 | ||
1998 | if (!session || cmd->version_check_done == 0) { | |
1999 | ERR("Trying to close a stream before version check"); | |
2000 | ret = -1; | |
2001 | goto end_no_session; | |
2002 | } | |
2003 | ||
2004 | ret = cmd->sock->ops->recvmsg(cmd->sock, &index_info, | |
2005 | sizeof(index_info), 0); | |
2006 | if (ret < sizeof(index_info)) { | |
2007 | if (ret == 0) { | |
2008 | /* Orderly shutdown. Not necessary to print an error. */ | |
2009 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
2010 | } else { | |
2011 | ERR("Relay didn't receive valid index struct size : %d", ret); | |
2012 | } | |
2013 | ret = -1; | |
2014 | goto end_no_session; | |
2015 | } | |
2016 | ||
2017 | net_seq_num = be64toh(index_info.net_seq_num); | |
2018 | ||
2019 | rcu_read_lock(); | |
d3e2ba59 | 2020 | stream = relay_stream_find_by_id(be64toh(index_info.relay_stream_id)); |
1c20f0e2 JD |
2021 | if (!stream) { |
2022 | ret = -1; | |
2023 | goto end_rcu_unlock; | |
2024 | } | |
2025 | ||
d3e2ba59 JD |
2026 | /* Live beacon handling */ |
2027 | if (index_info.packet_size == 0) { | |
2028 | DBG("Received live beacon for stream %" PRIu64, stream->stream_handle); | |
2029 | ||
2030 | /* | |
2031 | * Only flag a stream inactive when it has already received data. | |
2032 | */ | |
2033 | if (stream->total_index_received > 0) { | |
2034 | stream->beacon_ts_end = be64toh(index_info.timestamp_end); | |
2035 | } | |
2036 | ret = 0; | |
2037 | goto end_rcu_unlock; | |
2038 | } else { | |
2039 | stream->beacon_ts_end = -1ULL; | |
2040 | } | |
2041 | ||
0a6518b0 | 2042 | index = relay_index_find(stream->stream_handle, net_seq_num); |
1c20f0e2 JD |
2043 | if (!index) { |
2044 | /* A successful creation will add the object to the HT. */ | |
2045 | index = relay_index_create(stream->stream_handle, net_seq_num); | |
2046 | if (!index) { | |
2047 | goto end_rcu_unlock; | |
2048 | } | |
2049 | index_created = 1; | |
2050 | } | |
2051 | ||
2052 | copy_index_control_data(index, &index_info); | |
2053 | ||
2054 | if (index_created) { | |
2055 | /* | |
2056 | * Try to add the relay index object to the hash table. If an object | |
2057 | * already exist, destroy back the index created, set the data in this | |
2058 | * object and write it on disk. | |
2059 | */ | |
0a6518b0 | 2060 | relay_index_add(index, &wr_index); |
1c20f0e2 JD |
2061 | if (wr_index) { |
2062 | copy_index_control_data(wr_index, &index_info); | |
2063 | free(index); | |
2064 | } | |
2065 | } else { | |
2066 | /* The index already exists so write it on disk. */ | |
2067 | wr_index = index; | |
2068 | } | |
2069 | ||
2070 | /* Do we have a writable ready index to write on disk. */ | |
2071 | if (wr_index) { | |
2072 | /* Starting at 2.4, create the index file if none available. */ | |
2073 | if (cmd->minor >= 4 && stream->index_fd < 0) { | |
2074 | ret = index_create_file(stream->path_name, stream->channel_name, | |
2075 | relayd_uid, relayd_gid, stream->tracefile_size, | |
2076 | stream->tracefile_count_current); | |
2077 | if (ret < 0) { | |
2078 | goto end_rcu_unlock; | |
2079 | } | |
2080 | stream->index_fd = ret; | |
2081 | } | |
2082 | ||
0a6518b0 | 2083 | ret = relay_index_write(wr_index->fd, wr_index); |
1c20f0e2 JD |
2084 | if (ret < 0) { |
2085 | goto end_rcu_unlock; | |
2086 | } | |
d3e2ba59 | 2087 | stream->total_index_received++; |
1c20f0e2 JD |
2088 | } |
2089 | ||
2090 | end_rcu_unlock: | |
2091 | rcu_read_unlock(); | |
2092 | ||
2093 | if (ret < 0) { | |
2094 | reply.ret_code = htobe32(LTTNG_ERR_UNK); | |
2095 | } else { | |
2096 | reply.ret_code = htobe32(LTTNG_OK); | |
2097 | } | |
2098 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
2099 | if (send_ret < 0) { | |
2100 | ERR("Relay sending close index id reply"); | |
2101 | ret = send_ret; | |
2102 | } | |
2103 | ||
2104 | end_no_session: | |
2105 | return ret; | |
2106 | } | |
2107 | ||
a4baae1b JD |
2108 | /* |
2109 | * Receive the streams_sent message. | |
2110 | * | |
2111 | * Return 0 on success else a negative value. | |
2112 | */ | |
2113 | static | |
2114 | int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr, | |
2115 | struct relay_command *cmd) | |
2116 | { | |
2117 | int ret, send_ret; | |
2118 | struct lttcomm_relayd_generic_reply reply; | |
2119 | ||
2120 | assert(cmd); | |
2121 | ||
2122 | DBG("Relay receiving streams_sent"); | |
2123 | ||
2124 | if (!cmd->session || cmd->version_check_done == 0) { | |
2125 | ERR("Trying to close a stream before version check"); | |
2126 | ret = -1; | |
2127 | goto end_no_session; | |
2128 | } | |
2129 | ||
2130 | /* | |
2131 | * Flag every pending stream in the connection recv list that they are | |
2132 | * ready to be used by the viewer. | |
2133 | */ | |
2134 | set_viewer_ready_flag(cmd); | |
2135 | ||
4a9daf17 JD |
2136 | /* |
2137 | * Inform the viewer that there are new streams in the session. | |
2138 | */ | |
2139 | uatomic_set(&cmd->session->new_streams, 1); | |
2140 | ||
a4baae1b JD |
2141 | reply.ret_code = htobe32(LTTNG_OK); |
2142 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
2143 | if (send_ret < 0) { | |
2144 | ERR("Relay sending sent_stream reply"); | |
2145 | ret = send_ret; | |
2146 | } else { | |
2147 | /* Success. */ | |
2148 | ret = 0; | |
2149 | } | |
2150 | ||
2151 | end_no_session: | |
2152 | return ret; | |
2153 | } | |
2154 | ||
b8aa1682 | 2155 | /* |
d3e2ba59 | 2156 | * Process the commands received on the control socket |
b8aa1682 JD |
2157 | */ |
2158 | static | |
2159 | int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 2160 | struct relay_command *cmd, struct relay_local_data *ctx) |
b8aa1682 JD |
2161 | { |
2162 | int ret = 0; | |
2163 | ||
2164 | switch (be32toh(recv_hdr->cmd)) { | |
b8aa1682 | 2165 | case RELAYD_CREATE_SESSION: |
d3e2ba59 | 2166 | ret = relay_create_session(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 | 2167 | break; |
b8aa1682 | 2168 | case RELAYD_ADD_STREAM: |
d3e2ba59 | 2169 | ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 JD |
2170 | break; |
2171 | case RELAYD_START_DATA: | |
2172 | ret = relay_start(recv_hdr, cmd); | |
2173 | break; | |
2174 | case RELAYD_SEND_METADATA: | |
d3e2ba59 | 2175 | ret = relay_recv_metadata(recv_hdr, cmd); |
b8aa1682 JD |
2176 | break; |
2177 | case RELAYD_VERSION: | |
d3e2ba59 | 2178 | ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 | 2179 | break; |
173af62f | 2180 | case RELAYD_CLOSE_STREAM: |
92c6ca54 | 2181 | ret = relay_close_stream(recv_hdr, cmd); |
173af62f | 2182 | break; |
6d805429 | 2183 | case RELAYD_DATA_PENDING: |
d3e2ba59 | 2184 | ret = relay_data_pending(recv_hdr, cmd); |
c8f59ee5 DG |
2185 | break; |
2186 | case RELAYD_QUIESCENT_CONTROL: | |
d3e2ba59 | 2187 | ret = relay_quiescent_control(recv_hdr, cmd); |
c8f59ee5 | 2188 | break; |
f7079f67 | 2189 | case RELAYD_BEGIN_DATA_PENDING: |
d3e2ba59 | 2190 | ret = relay_begin_data_pending(recv_hdr, cmd); |
f7079f67 DG |
2191 | break; |
2192 | case RELAYD_END_DATA_PENDING: | |
d3e2ba59 | 2193 | ret = relay_end_data_pending(recv_hdr, cmd); |
f7079f67 | 2194 | break; |
1c20f0e2 | 2195 | case RELAYD_SEND_INDEX: |
0a6518b0 | 2196 | ret = relay_recv_index(recv_hdr, cmd); |
1c20f0e2 | 2197 | break; |
a4baae1b JD |
2198 | case RELAYD_STREAMS_SENT: |
2199 | ret = relay_streams_sent(recv_hdr, cmd); | |
2200 | break; | |
b8aa1682 JD |
2201 | case RELAYD_UPDATE_SYNC_INFO: |
2202 | default: | |
2203 | ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd)); | |
2204 | relay_unknown_command(cmd); | |
2205 | ret = -1; | |
2206 | goto end; | |
2207 | } | |
2208 | ||
2209 | end: | |
2210 | return ret; | |
2211 | } | |
2212 | ||
7d2f7452 DG |
2213 | /* |
2214 | * Handle index for a data stream. | |
2215 | * | |
2216 | * RCU read side lock MUST be acquired. | |
2217 | * | |
2218 | * Return 0 on success else a negative value. | |
2219 | */ | |
2220 | static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num, | |
2221 | int rotate_index) | |
2222 | { | |
2223 | int ret = 0, index_created = 0; | |
2224 | uint64_t stream_id, data_offset; | |
2225 | struct relay_index *index, *wr_index = NULL; | |
2226 | ||
2227 | assert(stream); | |
2228 | ||
2229 | stream_id = stream->stream_handle; | |
2230 | /* Get data offset because we are about to update the index. */ | |
2231 | data_offset = htobe64(stream->tracefile_size_current); | |
2232 | ||
2233 | /* | |
2234 | * Lookup for an existing index for that stream id/sequence number. If on | |
2235 | * exists, the control thread already received the data for it thus we need | |
2236 | * to write it on disk. | |
2237 | */ | |
2238 | index = relay_index_find(stream_id, net_seq_num); | |
2239 | if (!index) { | |
2240 | /* A successful creation will add the object to the HT. */ | |
2241 | index = relay_index_create(stream_id, net_seq_num); | |
2242 | if (!index) { | |
2243 | ret = -1; | |
2244 | goto error; | |
2245 | } | |
2246 | index_created = 1; | |
2247 | } | |
2248 | ||
2249 | if (rotate_index || stream->index_fd < 0) { | |
2250 | index->to_close_fd = stream->index_fd; | |
2251 | ret = index_create_file(stream->path_name, stream->channel_name, | |
2252 | relayd_uid, relayd_gid, stream->tracefile_size, | |
2253 | stream->tracefile_count_current); | |
2254 | if (ret < 0) { | |
2255 | /* This will close the stream's index fd if one. */ | |
2256 | relay_index_free_safe(index); | |
2257 | goto error; | |
2258 | } | |
2259 | stream->index_fd = ret; | |
2260 | } | |
2261 | index->fd = stream->index_fd; | |
2262 | index->index_data.offset = data_offset; | |
2263 | ||
2264 | if (index_created) { | |
2265 | /* | |
2266 | * Try to add the relay index object to the hash table. If an object | |
2267 | * already exist, destroy back the index created and set the data. | |
2268 | */ | |
2269 | relay_index_add(index, &wr_index); | |
2270 | if (wr_index) { | |
2271 | /* Copy back data from the created index. */ | |
2272 | wr_index->fd = index->fd; | |
2273 | wr_index->to_close_fd = index->to_close_fd; | |
2274 | wr_index->index_data.offset = data_offset; | |
2275 | free(index); | |
2276 | } | |
2277 | } else { | |
2278 | /* The index already exists so write it on disk. */ | |
2279 | wr_index = index; | |
2280 | } | |
2281 | ||
2282 | /* Do we have a writable ready index to write on disk. */ | |
2283 | if (wr_index) { | |
2284 | ret = relay_index_write(wr_index->fd, wr_index); | |
2285 | if (ret < 0) { | |
2286 | goto error; | |
2287 | } | |
2288 | stream->total_index_received++; | |
2289 | } | |
2290 | ||
2291 | error: | |
2292 | return ret; | |
2293 | } | |
2294 | ||
b8aa1682 JD |
2295 | /* |
2296 | * relay_process_data: Process the data received on the data socket | |
2297 | */ | |
2298 | static | |
0a6518b0 | 2299 | int relay_process_data(struct relay_command *cmd) |
b8aa1682 | 2300 | { |
7d2f7452 | 2301 | int ret = 0, rotate_index = 0; |
6cd525e8 | 2302 | ssize_t size_ret; |
b8aa1682 JD |
2303 | struct relay_stream *stream; |
2304 | struct lttcomm_relayd_data_hdr data_hdr; | |
7d2f7452 | 2305 | uint64_t stream_id; |
173af62f | 2306 | uint64_t net_seq_num; |
b8aa1682 JD |
2307 | uint32_t data_size; |
2308 | ||
2309 | ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr, | |
7c5aef62 | 2310 | sizeof(struct lttcomm_relayd_data_hdr), 0); |
b8aa1682 | 2311 | if (ret <= 0) { |
a6cd2b97 DG |
2312 | if (ret == 0) { |
2313 | /* Orderly shutdown. Not necessary to print an error. */ | |
2314 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
2315 | } else { | |
2316 | ERR("Unable to receive data header on sock %d", cmd->sock->fd); | |
2317 | } | |
b8aa1682 JD |
2318 | ret = -1; |
2319 | goto end; | |
2320 | } | |
2321 | ||
2322 | stream_id = be64toh(data_hdr.stream_id); | |
9d1bbf21 MD |
2323 | |
2324 | rcu_read_lock(); | |
d3e2ba59 | 2325 | stream = relay_stream_find_by_id(stream_id); |
b8aa1682 JD |
2326 | if (!stream) { |
2327 | ret = -1; | |
1c20f0e2 | 2328 | goto end_rcu_unlock; |
b8aa1682 JD |
2329 | } |
2330 | ||
2331 | data_size = be32toh(data_hdr.data_size); | |
2332 | if (data_buffer_size < data_size) { | |
c617c0c6 MD |
2333 | char *tmp_data_ptr; |
2334 | ||
2335 | tmp_data_ptr = realloc(data_buffer, data_size); | |
2336 | if (!tmp_data_ptr) { | |
b8aa1682 | 2337 | ERR("Allocating data buffer"); |
c617c0c6 | 2338 | free(data_buffer); |
b8aa1682 | 2339 | ret = -1; |
1c20f0e2 | 2340 | goto end_rcu_unlock; |
b8aa1682 | 2341 | } |
c617c0c6 | 2342 | data_buffer = tmp_data_ptr; |
b8aa1682 JD |
2343 | data_buffer_size = data_size; |
2344 | } | |
2345 | memset(data_buffer, 0, data_size); | |
2346 | ||
173af62f DG |
2347 | net_seq_num = be64toh(data_hdr.net_seq_num); |
2348 | ||
77c7c900 | 2349 | DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64, |
173af62f | 2350 | data_size, stream_id, net_seq_num); |
7c5aef62 | 2351 | ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0); |
b8aa1682 | 2352 | if (ret <= 0) { |
a6cd2b97 DG |
2353 | if (ret == 0) { |
2354 | /* Orderly shutdown. Not necessary to print an error. */ | |
2355 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
2356 | } | |
b8aa1682 | 2357 | ret = -1; |
1c20f0e2 | 2358 | goto end_rcu_unlock; |
b8aa1682 JD |
2359 | } |
2360 | ||
1c20f0e2 | 2361 | /* Check if a rotation is needed. */ |
0f907de1 JD |
2362 | if (stream->tracefile_size > 0 && |
2363 | (stream->tracefile_size_current + data_size) > | |
2364 | stream->tracefile_size) { | |
6b6b9a5a JD |
2365 | struct relay_viewer_stream *vstream; |
2366 | uint64_t new_id; | |
2367 | ||
2368 | new_id = (stream->tracefile_count_current + 1) % | |
2369 | stream->tracefile_count; | |
2370 | /* | |
2371 | * When we wrap-around back to 0, we start overwriting old | |
2372 | * trace data. | |
2373 | */ | |
2374 | if (!stream->tracefile_overwrite && new_id == 0) { | |
2375 | stream->tracefile_overwrite = 1; | |
2376 | } | |
2377 | pthread_mutex_lock(&stream->viewer_stream_rotation_lock); | |
2378 | if (stream->tracefile_overwrite) { | |
2379 | stream->oldest_tracefile_id = | |
2380 | (stream->oldest_tracefile_id + 1) % | |
2381 | stream->tracefile_count; | |
2382 | } | |
2f8f53af | 2383 | vstream = viewer_stream_find_by_id(stream->stream_handle); |
6b6b9a5a JD |
2384 | if (vstream) { |
2385 | /* | |
2386 | * The viewer is reading a file about to be | |
2387 | * overwritten. Close the FDs it is | |
2388 | * currently using and let it handle the fault. | |
2389 | */ | |
2390 | if (vstream->tracefile_count_current == new_id) { | |
cef0f7d5 | 2391 | pthread_mutex_lock(&vstream->overwrite_lock); |
6b6b9a5a | 2392 | vstream->abort_flag = 1; |
cef0f7d5 | 2393 | pthread_mutex_unlock(&vstream->overwrite_lock); |
6b6b9a5a JD |
2394 | DBG("Streaming side setting abort_flag on stream %s_%lu\n", |
2395 | stream->channel_name, new_id); | |
2396 | } else if (vstream->tracefile_count_current == | |
2397 | stream->tracefile_count_current) { | |
2398 | /* | |
2399 | * The reader and writer were in the | |
2400 | * same trace file, inform the viewer | |
2401 | * that no new index will ever be added | |
2402 | * to this file. | |
2403 | */ | |
2404 | vstream->close_write_flag = 1; | |
2405 | } | |
2406 | } | |
1c20f0e2 JD |
2407 | ret = utils_rotate_stream_file(stream->path_name, stream->channel_name, |
2408 | stream->tracefile_size, stream->tracefile_count, | |
2409 | relayd_uid, relayd_gid, stream->fd, | |
2410 | &(stream->tracefile_count_current), &stream->fd); | |
cef0f7d5 | 2411 | stream->total_index_received = 0; |
6b6b9a5a | 2412 | pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); |
0f907de1 | 2413 | if (ret < 0) { |
1c20f0e2 JD |
2414 | ERR("Rotating stream output file"); |
2415 | goto end_rcu_unlock; | |
0f907de1 | 2416 | } |
a6976990 DG |
2417 | /* Reset current size because we just perform a stream rotation. */ |
2418 | stream->tracefile_size_current = 0; | |
1c20f0e2 JD |
2419 | rotate_index = 1; |
2420 | } | |
2421 | ||
1c20f0e2 | 2422 | /* |
7d2f7452 DG |
2423 | * Index are handled in protocol version 2.4 and above. Also, snapshot and |
2424 | * index are NOT supported. | |
1c20f0e2 | 2425 | */ |
7d2f7452 DG |
2426 | if (stream->session->minor >= 4 && !stream->session->snapshot) { |
2427 | ret = handle_index_data(stream, net_seq_num, rotate_index); | |
1c20f0e2 | 2428 | if (ret < 0) { |
1c20f0e2 JD |
2429 | goto end_rcu_unlock; |
2430 | } | |
1c20f0e2 JD |
2431 | } |
2432 | ||
7d2f7452 | 2433 | /* Write data to stream output fd. */ |
6cd525e8 MD |
2434 | size_ret = lttng_write(stream->fd, data_buffer, data_size); |
2435 | if (size_ret < data_size) { | |
b8aa1682 JD |
2436 | ERR("Relay error writing data to file"); |
2437 | ret = -1; | |
1c20f0e2 | 2438 | goto end_rcu_unlock; |
b8aa1682 | 2439 | } |
1d4dfdef | 2440 | |
5ab7344e JD |
2441 | DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64, |
2442 | ret, stream->stream_handle); | |
2443 | ||
1d4dfdef DG |
2444 | ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size)); |
2445 | if (ret < 0) { | |
1c20f0e2 | 2446 | goto end_rcu_unlock; |
1d4dfdef | 2447 | } |
1c20f0e2 | 2448 | stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size); |
1d4dfdef | 2449 | |
173af62f DG |
2450 | stream->prev_seq = net_seq_num; |
2451 | ||
2452 | /* Check if we need to close the FD */ | |
2453 | if (close_stream_check(stream)) { | |
c07d8a78 | 2454 | destroy_stream(stream); |
173af62f DG |
2455 | } |
2456 | ||
1c20f0e2 | 2457 | end_rcu_unlock: |
9d1bbf21 | 2458 | rcu_read_unlock(); |
b8aa1682 JD |
2459 | end: |
2460 | return ret; | |
2461 | } | |
2462 | ||
2463 | static | |
9d1bbf21 | 2464 | void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd) |
b8aa1682 JD |
2465 | { |
2466 | int ret; | |
2467 | ||
b8aa1682 JD |
2468 | lttng_poll_del(events, pollfd); |
2469 | ||
2470 | ret = close(pollfd); | |
2471 | if (ret < 0) { | |
2472 | ERR("Closing pollfd %d", pollfd); | |
2473 | } | |
2474 | } | |
2475 | ||
2476 | static | |
2477 | int relay_add_connection(int fd, struct lttng_poll_event *events, | |
2478 | struct lttng_ht *relay_connections_ht) | |
2479 | { | |
b8aa1682 | 2480 | struct relay_command *relay_connection; |
6cd525e8 | 2481 | ssize_t ret; |
b8aa1682 JD |
2482 | |
2483 | relay_connection = zmalloc(sizeof(struct relay_command)); | |
2484 | if (relay_connection == NULL) { | |
2485 | PERROR("Relay command zmalloc"); | |
9d1bbf21 | 2486 | goto error; |
b8aa1682 | 2487 | } |
6cd525e8 MD |
2488 | ret = lttng_read(fd, relay_connection, sizeof(struct relay_command)); |
2489 | if (ret < sizeof(struct relay_command)) { | |
b8aa1682 | 2490 | PERROR("read relay cmd pipe"); |
9d1bbf21 | 2491 | goto error_read; |
b8aa1682 | 2492 | } |
a4baae1b | 2493 | CDS_INIT_LIST_HEAD(&relay_connection->recv_head); |
b8aa1682 | 2494 | |
c07d8a78 DG |
2495 | /* |
2496 | * Only used by the control side and the reference is copied inside each | |
2497 | * stream from that connection. Thus a destroy HT must be done after every | |
2498 | * stream has been destroyed. | |
2499 | */ | |
2500 | if (relay_connection->type == RELAY_CONTROL) { | |
2501 | relay_connection->ctf_traces_ht = lttng_ht_new(0, | |
2502 | LTTNG_HT_TYPE_STRING); | |
2503 | if (!relay_connection->ctf_traces_ht) { | |
2504 | goto error_read; | |
2505 | } | |
d3e2ba59 JD |
2506 | } |
2507 | ||
b8aa1682 JD |
2508 | lttng_ht_node_init_ulong(&relay_connection->sock_n, |
2509 | (unsigned long) relay_connection->sock->fd); | |
9d1bbf21 | 2510 | rcu_read_lock(); |
b8aa1682 JD |
2511 | lttng_ht_add_unique_ulong(relay_connections_ht, |
2512 | &relay_connection->sock_n); | |
9d1bbf21 MD |
2513 | rcu_read_unlock(); |
2514 | return lttng_poll_add(events, | |
b8aa1682 JD |
2515 | relay_connection->sock->fd, |
2516 | LPOLLIN | LPOLLRDHUP); | |
2517 | ||
9d1bbf21 MD |
2518 | error_read: |
2519 | free(relay_connection); | |
2520 | error: | |
2521 | return -1; | |
2522 | } | |
2523 | ||
2524 | static | |
2525 | void deferred_free_connection(struct rcu_head *head) | |
2526 | { | |
2527 | struct relay_command *relay_connection = | |
2528 | caa_container_of(head, struct relay_command, rcu_node); | |
5b6d8097 DG |
2529 | |
2530 | lttcomm_destroy_sock(relay_connection->sock); | |
9d1bbf21 MD |
2531 | free(relay_connection); |
2532 | } | |
2533 | ||
2534 | static | |
2535 | void relay_del_connection(struct lttng_ht *relay_connections_ht, | |
d3e2ba59 JD |
2536 | struct lttng_ht_iter *iter, struct relay_command *relay_connection, |
2537 | struct lttng_ht *sessions_ht) | |
9d1bbf21 MD |
2538 | { |
2539 | int ret; | |
2540 | ||
2541 | ret = lttng_ht_del(relay_connections_ht, iter); | |
2542 | assert(!ret); | |
c07d8a78 | 2543 | |
9d1bbf21 | 2544 | if (relay_connection->type == RELAY_CONTROL) { |
a4baae1b JD |
2545 | struct relay_stream_recv_handle *node, *tmp_node; |
2546 | ||
d3e2ba59 | 2547 | relay_delete_session(relay_connection, sessions_ht); |
c07d8a78 | 2548 | lttng_ht_destroy(relay_connection->ctf_traces_ht); |
a4baae1b JD |
2549 | |
2550 | /* Clean up recv list. */ | |
2551 | cds_list_for_each_entry_safe(node, tmp_node, | |
2552 | &relay_connection->recv_head, node) { | |
2553 | cds_list_del(&node->node); | |
2554 | free(node); | |
2555 | } | |
9d1bbf21 | 2556 | } |
5b6d8097 | 2557 | |
c07d8a78 | 2558 | call_rcu(&relay_connection->rcu_node, deferred_free_connection); |
b8aa1682 JD |
2559 | } |
2560 | ||
2561 | /* | |
2562 | * This thread does the actual work | |
2563 | */ | |
2564 | static | |
2565 | void *relay_thread_worker(void *data) | |
2566 | { | |
beaad64c DG |
2567 | int ret, err = -1, last_seen_data_fd = -1; |
2568 | uint32_t nb_fd; | |
b8aa1682 JD |
2569 | struct relay_command *relay_connection; |
2570 | struct lttng_poll_event events; | |
2571 | struct lttng_ht *relay_connections_ht; | |
2572 | struct lttng_ht_node_ulong *node; | |
2573 | struct lttng_ht_iter iter; | |
b8aa1682 | 2574 | struct lttcomm_relayd_hdr recv_hdr; |
d3e2ba59 JD |
2575 | struct relay_local_data *relay_ctx = (struct relay_local_data *) data; |
2576 | struct lttng_ht *sessions_ht = relay_ctx->sessions_ht; | |
b8aa1682 JD |
2577 | |
2578 | DBG("[thread] Relay worker started"); | |
2579 | ||
9d1bbf21 MD |
2580 | rcu_register_thread(); |
2581 | ||
55706a7d MD |
2582 | health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER); |
2583 | ||
9b5e0863 MD |
2584 | if (testpoint(relayd_thread_worker)) { |
2585 | goto error_testpoint; | |
2586 | } | |
2587 | ||
f385ae0a MD |
2588 | health_code_update(); |
2589 | ||
b8aa1682 JD |
2590 | /* table of connections indexed on socket */ |
2591 | relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
095a4ae5 MD |
2592 | if (!relay_connections_ht) { |
2593 | goto relay_connections_ht_error; | |
2594 | } | |
b8aa1682 | 2595 | |
1c20f0e2 JD |
2596 | /* Tables of received indexes indexed by index handle and net_seq_num. */ |
2597 | indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64); | |
2598 | if (!indexes_ht) { | |
2599 | goto indexes_ht_error; | |
2600 | } | |
2601 | ||
b8aa1682 JD |
2602 | ret = create_thread_poll_set(&events, 2); |
2603 | if (ret < 0) { | |
2604 | goto error_poll_create; | |
2605 | } | |
2606 | ||
2607 | ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP); | |
2608 | if (ret < 0) { | |
2609 | goto error; | |
2610 | } | |
2611 | ||
beaad64c | 2612 | restart: |
b8aa1682 | 2613 | while (1) { |
beaad64c DG |
2614 | int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1; |
2615 | ||
f385ae0a MD |
2616 | health_code_update(); |
2617 | ||
b8aa1682 | 2618 | /* Infinite blocking call, waiting for transmission */ |
87c1611d | 2619 | DBG3("Relayd worker thread polling..."); |
f385ae0a | 2620 | health_poll_entry(); |
b8aa1682 | 2621 | ret = lttng_poll_wait(&events, -1); |
f385ae0a | 2622 | health_poll_exit(); |
b8aa1682 JD |
2623 | if (ret < 0) { |
2624 | /* | |
2625 | * Restart interrupted system call. | |
2626 | */ | |
2627 | if (errno == EINTR) { | |
2628 | goto restart; | |
2629 | } | |
2630 | goto error; | |
2631 | } | |
2632 | ||
0d9c5d77 DG |
2633 | nb_fd = ret; |
2634 | ||
beaad64c DG |
2635 | /* |
2636 | * Process control. The control connection is prioritised so we don't | |
2637 | * starve it with high throughout put tracing data on the data | |
2638 | * connection. | |
2639 | */ | |
b8aa1682 JD |
2640 | for (i = 0; i < nb_fd; i++) { |
2641 | /* Fetch once the poll data */ | |
beaad64c DG |
2642 | uint32_t revents = LTTNG_POLL_GETEV(&events, i); |
2643 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
b8aa1682 | 2644 | |
f385ae0a MD |
2645 | health_code_update(); |
2646 | ||
b8aa1682 JD |
2647 | /* Thread quit pipe has been closed. Killing thread. */ |
2648 | ret = check_thread_quit_pipe(pollfd, revents); | |
2649 | if (ret) { | |
095a4ae5 MD |
2650 | err = 0; |
2651 | goto exit; | |
b8aa1682 JD |
2652 | } |
2653 | ||
2654 | /* Inspect the relay cmd pipe for new connection */ | |
2655 | if (pollfd == relay_cmd_pipe[0]) { | |
2656 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
2657 | ERR("Relay pipe error"); | |
2658 | goto error; | |
2659 | } else if (revents & LPOLLIN) { | |
2660 | DBG("Relay command received"); | |
2661 | ret = relay_add_connection(relay_cmd_pipe[0], | |
2662 | &events, relay_connections_ht); | |
2663 | if (ret < 0) { | |
2664 | goto error; | |
2665 | } | |
2666 | } | |
beaad64c | 2667 | } else if (revents) { |
9d1bbf21 | 2668 | rcu_read_lock(); |
b8aa1682 JD |
2669 | lttng_ht_lookup(relay_connections_ht, |
2670 | (void *)((unsigned long) pollfd), | |
2671 | &iter); | |
2672 | node = lttng_ht_iter_get_node_ulong(&iter); | |
2673 | if (node == NULL) { | |
2674 | DBG2("Relay sock %d not found", pollfd); | |
9d1bbf21 | 2675 | rcu_read_unlock(); |
b8aa1682 JD |
2676 | goto error; |
2677 | } | |
2678 | relay_connection = caa_container_of(node, | |
2679 | struct relay_command, sock_n); | |
2680 | ||
2681 | if (revents & (LPOLLERR)) { | |
2682 | ERR("POLL ERROR"); | |
9d1bbf21 MD |
2683 | relay_cleanup_poll_connection(&events, pollfd); |
2684 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2685 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2686 | if (last_seen_data_fd == pollfd) { |
2687 | last_seen_data_fd = last_notdel_data_fd; | |
2688 | } | |
b8aa1682 JD |
2689 | } else if (revents & (LPOLLHUP | LPOLLRDHUP)) { |
2690 | DBG("Socket %d hung up", pollfd); | |
9d1bbf21 MD |
2691 | relay_cleanup_poll_connection(&events, pollfd); |
2692 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2693 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2694 | if (last_seen_data_fd == pollfd) { |
2695 | last_seen_data_fd = last_notdel_data_fd; | |
2696 | } | |
b8aa1682 JD |
2697 | } else if (revents & LPOLLIN) { |
2698 | /* control socket */ | |
2699 | if (relay_connection->type == RELAY_CONTROL) { | |
2700 | ret = relay_connection->sock->ops->recvmsg( | |
2701 | relay_connection->sock, &recv_hdr, | |
7c5aef62 | 2702 | sizeof(struct lttcomm_relayd_hdr), 0); |
b8aa1682 JD |
2703 | /* connection closed */ |
2704 | if (ret <= 0) { | |
9d1bbf21 MD |
2705 | relay_cleanup_poll_connection(&events, pollfd); |
2706 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2707 | &iter, relay_connection, sessions_ht); |
b8aa1682 JD |
2708 | DBG("Control connection closed with %d", pollfd); |
2709 | } else { | |
2710 | if (relay_connection->session) { | |
77c7c900 | 2711 | DBG2("Relay worker receiving data for session : %" PRIu64, |
b8aa1682 JD |
2712 | relay_connection->session->id); |
2713 | } | |
2714 | ret = relay_process_control(&recv_hdr, | |
d3e2ba59 | 2715 | relay_connection, relay_ctx); |
b8aa1682 | 2716 | if (ret < 0) { |
beaad64c | 2717 | /* Clear the session on error. */ |
9d1bbf21 MD |
2718 | relay_cleanup_poll_connection(&events, pollfd); |
2719 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2720 | &iter, relay_connection, sessions_ht); |
b8aa1682 JD |
2721 | DBG("Connection closed with %d", pollfd); |
2722 | } | |
beaad64c | 2723 | seen_control = 1; |
b8aa1682 | 2724 | } |
beaad64c DG |
2725 | } else { |
2726 | /* | |
2727 | * Flag the last seen data fd not deleted. It will be | |
2728 | * used as the last seen fd if any fd gets deleted in | |
2729 | * this first loop. | |
2730 | */ | |
2731 | last_notdel_data_fd = pollfd; | |
2732 | } | |
2733 | } | |
2734 | rcu_read_unlock(); | |
2735 | } | |
2736 | } | |
2737 | ||
2738 | /* | |
2739 | * The last loop handled a control request, go back to poll to make | |
2740 | * sure we prioritise the control socket. | |
2741 | */ | |
2742 | if (seen_control) { | |
2743 | continue; | |
2744 | } | |
2745 | ||
2746 | if (last_seen_data_fd >= 0) { | |
2747 | for (i = 0; i < nb_fd; i++) { | |
2748 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
f385ae0a MD |
2749 | |
2750 | health_code_update(); | |
2751 | ||
beaad64c DG |
2752 | if (last_seen_data_fd == pollfd) { |
2753 | idx = i; | |
2754 | break; | |
2755 | } | |
2756 | } | |
2757 | } | |
2758 | ||
2759 | /* Process data connection. */ | |
2760 | for (i = idx + 1; i < nb_fd; i++) { | |
2761 | /* Fetch the poll data. */ | |
2762 | uint32_t revents = LTTNG_POLL_GETEV(&events, i); | |
2763 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
2764 | ||
f385ae0a MD |
2765 | health_code_update(); |
2766 | ||
beaad64c DG |
2767 | /* Skip the command pipe. It's handled in the first loop. */ |
2768 | if (pollfd == relay_cmd_pipe[0]) { | |
2769 | continue; | |
2770 | } | |
2771 | ||
2772 | if (revents) { | |
2773 | rcu_read_lock(); | |
2774 | lttng_ht_lookup(relay_connections_ht, | |
2775 | (void *)((unsigned long) pollfd), | |
2776 | &iter); | |
2777 | node = lttng_ht_iter_get_node_ulong(&iter); | |
2778 | if (node == NULL) { | |
2779 | /* Skip it. Might be removed before. */ | |
2780 | rcu_read_unlock(); | |
2781 | continue; | |
2782 | } | |
2783 | relay_connection = caa_container_of(node, | |
2784 | struct relay_command, sock_n); | |
2785 | ||
2786 | if (revents & LPOLLIN) { | |
2787 | if (relay_connection->type != RELAY_DATA) { | |
2788 | continue; | |
2789 | } | |
2790 | ||
0a6518b0 | 2791 | ret = relay_process_data(relay_connection); |
beaad64c DG |
2792 | /* connection closed */ |
2793 | if (ret < 0) { | |
2794 | relay_cleanup_poll_connection(&events, pollfd); | |
2795 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2796 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2797 | DBG("Data connection closed with %d", pollfd); |
2798 | /* | |
2799 | * Every goto restart call sets the last seen fd where | |
2800 | * here we don't really care since we gracefully | |
2801 | * continue the loop after the connection is deleted. | |
2802 | */ | |
2803 | } else { | |
2804 | /* Keep last seen port. */ | |
2805 | last_seen_data_fd = pollfd; | |
2806 | rcu_read_unlock(); | |
2807 | goto restart; | |
b8aa1682 JD |
2808 | } |
2809 | } | |
9d1bbf21 | 2810 | rcu_read_unlock(); |
b8aa1682 JD |
2811 | } |
2812 | } | |
beaad64c | 2813 | last_seen_data_fd = -1; |
b8aa1682 JD |
2814 | } |
2815 | ||
f385ae0a MD |
2816 | /* Normal exit, no error */ |
2817 | ret = 0; | |
2818 | ||
095a4ae5 | 2819 | exit: |
b8aa1682 JD |
2820 | error: |
2821 | lttng_poll_clean(&events); | |
2822 | ||
2823 | /* empty the hash table and free the memory */ | |
9d1bbf21 | 2824 | rcu_read_lock(); |
b8aa1682 | 2825 | cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) { |
f385ae0a MD |
2826 | health_code_update(); |
2827 | ||
b8aa1682 JD |
2828 | node = lttng_ht_iter_get_node_ulong(&iter); |
2829 | if (node) { | |
2830 | relay_connection = caa_container_of(node, | |
2831 | struct relay_command, sock_n); | |
9d1bbf21 | 2832 | relay_del_connection(relay_connections_ht, |
d3e2ba59 | 2833 | &iter, relay_connection, sessions_ht); |
b8aa1682 | 2834 | } |
b8aa1682 | 2835 | } |
94d49140 | 2836 | rcu_read_unlock(); |
7d2f7452 DG |
2837 | error_poll_create: |
2838 | lttng_ht_destroy(indexes_ht); | |
1c20f0e2 | 2839 | indexes_ht_error: |
b8aa1682 | 2840 | lttng_ht_destroy(relay_connections_ht); |
095a4ae5 | 2841 | relay_connections_ht_error: |
6620da75 DG |
2842 | /* Close relay cmd pipes */ |
2843 | utils_close_pipe(relay_cmd_pipe); | |
095a4ae5 MD |
2844 | if (err) { |
2845 | DBG("Thread exited with error"); | |
2846 | } | |
b8aa1682 | 2847 | DBG("Worker thread cleanup complete"); |
095a4ae5 | 2848 | free(data_buffer); |
9b5e0863 | 2849 | error_testpoint: |
f385ae0a MD |
2850 | if (err) { |
2851 | health_error(); | |
2852 | ERR("Health error occurred in %s", __func__); | |
2853 | } | |
2854 | health_unregister(health_relayd); | |
9d1bbf21 | 2855 | rcu_unregister_thread(); |
f385ae0a | 2856 | stop_threads(); |
b8aa1682 JD |
2857 | return NULL; |
2858 | } | |
2859 | ||
2860 | /* | |
2861 | * Create the relay command pipe to wake thread_manage_apps. | |
2862 | * Closed in cleanup(). | |
2863 | */ | |
2864 | static int create_relay_cmd_pipe(void) | |
2865 | { | |
a02de639 | 2866 | int ret; |
b8aa1682 | 2867 | |
a02de639 | 2868 | ret = utils_create_pipe_cloexec(relay_cmd_pipe); |
b8aa1682 | 2869 | |
b8aa1682 JD |
2870 | return ret; |
2871 | } | |
2872 | ||
2873 | /* | |
2874 | * main | |
2875 | */ | |
2876 | int main(int argc, char **argv) | |
2877 | { | |
2878 | int ret = 0; | |
2879 | void *status; | |
d3e2ba59 | 2880 | struct relay_local_data *relay_ctx; |
b8aa1682 | 2881 | |
b8aa1682 JD |
2882 | /* Parse arguments */ |
2883 | progname = argv[0]; | |
cd60b05a | 2884 | if ((ret = set_options(argc, argv)) < 0) { |
a02de639 | 2885 | goto exit; |
b8aa1682 JD |
2886 | } |
2887 | ||
2888 | if ((ret = set_signal_handler()) < 0) { | |
2889 | goto exit; | |
2890 | } | |
2891 | ||
4d513a50 DG |
2892 | /* Try to create directory if -o, --output is specified. */ |
2893 | if (opt_output_path) { | |
994fa64f DG |
2894 | if (*opt_output_path != '/') { |
2895 | ERR("Please specify an absolute path for -o, --output PATH"); | |
2896 | goto exit; | |
2897 | } | |
2898 | ||
4d513a50 DG |
2899 | ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG); |
2900 | if (ret < 0) { | |
2901 | ERR("Unable to create %s", opt_output_path); | |
2902 | goto exit; | |
2903 | } | |
2904 | } | |
2905 | ||
b8aa1682 | 2906 | /* Daemonize */ |
b5218ffb | 2907 | if (opt_daemon || opt_background) { |
3fd27398 MD |
2908 | int i; |
2909 | ||
2910 | ret = lttng_daemonize(&child_ppid, &recv_child_signal, | |
2911 | !opt_background); | |
b8aa1682 | 2912 | if (ret < 0) { |
a02de639 | 2913 | goto exit; |
b8aa1682 | 2914 | } |
3fd27398 MD |
2915 | |
2916 | /* | |
2917 | * We are in the child. Make sure all other file | |
2918 | * descriptors are closed, in case we are called with | |
2919 | * more opened file descriptors than the standard ones. | |
2920 | */ | |
2921 | for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) { | |
2922 | (void) close(i); | |
2923 | } | |
2924 | } | |
2925 | ||
2926 | /* Create thread quit pipe */ | |
2927 | if ((ret = init_thread_quit_pipe()) < 0) { | |
2928 | goto error; | |
b8aa1682 JD |
2929 | } |
2930 | ||
1c20f0e2 JD |
2931 | /* We need those values for the file/dir creation. */ |
2932 | relayd_uid = getuid(); | |
2933 | relayd_gid = getgid(); | |
b8aa1682 | 2934 | |
1c20f0e2 JD |
2935 | /* Check if daemon is UID = 0 */ |
2936 | if (relayd_uid == 0) { | |
8d5c808e | 2937 | if (control_uri->port < 1024 || data_uri->port < 1024 || live_uri->port < 1024) { |
b8aa1682 JD |
2938 | ERR("Need to be root to use ports < 1024"); |
2939 | ret = -1; | |
a02de639 | 2940 | goto exit; |
b8aa1682 JD |
2941 | } |
2942 | } | |
2943 | ||
2944 | /* Setup the thread apps communication pipe. */ | |
2945 | if ((ret = create_relay_cmd_pipe()) < 0) { | |
2946 | goto exit; | |
2947 | } | |
2948 | ||
2949 | /* Init relay command queue. */ | |
2950 | cds_wfq_init(&relay_cmd_queue.queue); | |
2951 | ||
2952 | /* Set up max poll set size */ | |
2953 | lttng_poll_set_max_size(); | |
2954 | ||
554831e7 MD |
2955 | /* Initialize communication library */ |
2956 | lttcomm_init(); | |
87e45c13 | 2957 | lttcomm_inet_init(); |
554831e7 | 2958 | |
d3e2ba59 JD |
2959 | relay_ctx = zmalloc(sizeof(struct relay_local_data)); |
2960 | if (!relay_ctx) { | |
2961 | PERROR("relay_ctx"); | |
2962 | goto exit; | |
2963 | } | |
2964 | ||
2965 | /* tables of sessions indexed by session ID */ | |
2966 | relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
2967 | if (!relay_ctx->sessions_ht) { | |
2968 | goto exit_relay_ctx_sessions; | |
2969 | } | |
2970 | ||
2971 | /* tables of streams indexed by stream ID */ | |
2972 | relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
2973 | if (!relay_streams_ht) { | |
2974 | goto exit_relay_ctx_streams; | |
2975 | } | |
2976 | ||
2977 | /* tables of streams indexed by stream ID */ | |
92c6ca54 DG |
2978 | viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
2979 | if (!viewer_streams_ht) { | |
d3e2ba59 JD |
2980 | goto exit_relay_ctx_viewer_streams; |
2981 | } | |
2982 | ||
55706a7d MD |
2983 | /* Initialize thread health monitoring */ |
2984 | health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES); | |
2985 | if (!health_relayd) { | |
2986 | PERROR("health_app_create error"); | |
2987 | goto exit_health_app_create; | |
2988 | } | |
2989 | ||
65931c8b MD |
2990 | ret = utils_create_pipe(health_quit_pipe); |
2991 | if (ret < 0) { | |
2992 | goto error_health_pipe; | |
2993 | } | |
2994 | ||
2995 | /* Create thread to manage the client socket */ | |
2996 | ret = pthread_create(&health_thread, NULL, | |
2997 | thread_manage_health, (void *) NULL); | |
2998 | if (ret != 0) { | |
2999 | PERROR("pthread_create health"); | |
3000 | goto health_error; | |
3001 | } | |
3002 | ||
b8aa1682 JD |
3003 | /* Setup the dispatcher thread */ |
3004 | ret = pthread_create(&dispatcher_thread, NULL, | |
3005 | relay_thread_dispatcher, (void *) NULL); | |
3006 | if (ret != 0) { | |
3007 | PERROR("pthread_create dispatcher"); | |
3008 | goto exit_dispatcher; | |
3009 | } | |
3010 | ||
3011 | /* Setup the worker thread */ | |
3012 | ret = pthread_create(&worker_thread, NULL, | |
d3e2ba59 | 3013 | relay_thread_worker, (void *) relay_ctx); |
b8aa1682 JD |
3014 | if (ret != 0) { |
3015 | PERROR("pthread_create worker"); | |
3016 | goto exit_worker; | |
3017 | } | |
3018 | ||
3019 | /* Setup the listener thread */ | |
3020 | ret = pthread_create(&listener_thread, NULL, | |
3021 | relay_thread_listener, (void *) NULL); | |
3022 | if (ret != 0) { | |
3023 | PERROR("pthread_create listener"); | |
3024 | goto exit_listener; | |
3025 | } | |
3026 | ||
0b242f62 | 3027 | ret = live_start_threads(live_uri, relay_ctx); |
d3e2ba59 JD |
3028 | if (ret != 0) { |
3029 | ERR("Starting live viewer threads"); | |
50138f51 | 3030 | goto exit_live; |
d3e2ba59 JD |
3031 | } |
3032 | ||
50138f51 | 3033 | exit_live: |
b8aa1682 JD |
3034 | ret = pthread_join(listener_thread, &status); |
3035 | if (ret != 0) { | |
3036 | PERROR("pthread_join"); | |
3037 | goto error; /* join error, exit without cleanup */ | |
3038 | } | |
3039 | ||
50138f51 | 3040 | exit_listener: |
b8aa1682 JD |
3041 | ret = pthread_join(worker_thread, &status); |
3042 | if (ret != 0) { | |
3043 | PERROR("pthread_join"); | |
3044 | goto error; /* join error, exit without cleanup */ | |
3045 | } | |
3046 | ||
50138f51 | 3047 | exit_worker: |
b8aa1682 JD |
3048 | ret = pthread_join(dispatcher_thread, &status); |
3049 | if (ret != 0) { | |
3050 | PERROR("pthread_join"); | |
3051 | goto error; /* join error, exit without cleanup */ | |
3052 | } | |
42415026 | 3053 | |
50138f51 | 3054 | exit_dispatcher: |
65931c8b MD |
3055 | ret = pthread_join(health_thread, &status); |
3056 | if (ret != 0) { | |
3057 | PERROR("pthread_join health thread"); | |
3058 | goto error; /* join error, exit without cleanup */ | |
3059 | } | |
3060 | ||
bd11e201 MD |
3061 | /* |
3062 | * Stop live threads only after joining other threads. | |
3063 | */ | |
3064 | live_stop_threads(); | |
3065 | ||
65931c8b MD |
3066 | health_error: |
3067 | utils_close_pipe(health_quit_pipe); | |
3068 | ||
3069 | error_health_pipe: | |
55706a7d MD |
3070 | health_app_destroy(health_relayd); |
3071 | ||
3072 | exit_health_app_create: | |
92c6ca54 | 3073 | lttng_ht_destroy(viewer_streams_ht); |
d3e2ba59 JD |
3074 | |
3075 | exit_relay_ctx_viewer_streams: | |
3076 | lttng_ht_destroy(relay_streams_ht); | |
3077 | ||
3078 | exit_relay_ctx_streams: | |
3079 | lttng_ht_destroy(relay_ctx->sessions_ht); | |
3080 | ||
3081 | exit_relay_ctx_sessions: | |
3082 | free(relay_ctx); | |
b8aa1682 JD |
3083 | |
3084 | exit: | |
3085 | cleanup(); | |
3086 | if (!ret) { | |
3087 | exit(EXIT_SUCCESS); | |
3088 | } | |
a02de639 | 3089 | |
b8aa1682 JD |
3090 | error: |
3091 | exit(EXIT_FAILURE); | |
3092 | } |