Commit | Line | Data |
---|---|---|
b8aa1682 JD |
1 | /* |
2 | * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com> | |
3 | * David Goulet <dgoulet@efficios.com> | |
4 | * | |
5 | * This program is free software; you can redistribute it and/or modify | |
6 | * it under the terms of the GNU General Public License, version 2 only, | |
7 | * as published by the Free Software Foundation. | |
8 | * | |
9 | * This program is distributed in the hope that it will be useful, but WITHOUT | |
10 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for | |
12 | * more details. | |
13 | * | |
14 | * You should have received a copy of the GNU General Public License along | |
15 | * with this program; if not, write to the Free Software Foundation, Inc., | |
16 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. | |
17 | */ | |
18 | ||
19 | #define _GNU_SOURCE | |
20 | #include <getopt.h> | |
21 | #include <grp.h> | |
22 | #include <limits.h> | |
23 | #include <pthread.h> | |
24 | #include <signal.h> | |
25 | #include <stdio.h> | |
26 | #include <stdlib.h> | |
27 | #include <string.h> | |
28 | #include <sys/mman.h> | |
29 | #include <sys/mount.h> | |
30 | #include <sys/resource.h> | |
31 | #include <sys/socket.h> | |
32 | #include <sys/stat.h> | |
33 | #include <sys/types.h> | |
34 | #include <sys/wait.h> | |
173af62f | 35 | #include <inttypes.h> |
b8aa1682 JD |
36 | #include <urcu/futex.h> |
37 | #include <urcu/uatomic.h> | |
38 | #include <unistd.h> | |
39 | #include <fcntl.h> | |
40 | #include <config.h> | |
41 | ||
42 | #include <lttng/lttng.h> | |
43 | #include <common/common.h> | |
44 | #include <common/compat/poll.h> | |
45 | #include <common/compat/socket.h> | |
46 | #include <common/defaults.h> | |
02a6bb53 | 47 | #include <common/daemonize.h> |
b8aa1682 JD |
48 | #include <common/futex.h> |
49 | #include <common/sessiond-comm/sessiond-comm.h> | |
50 | #include <common/sessiond-comm/inet.h> | |
b8aa1682 JD |
51 | #include <common/sessiond-comm/relayd.h> |
52 | #include <common/uri.h> | |
a02de639 | 53 | #include <common/utils.h> |
b8aa1682 | 54 | |
0f907de1 | 55 | #include "cmd.h" |
d3e2ba59 | 56 | #include "ctf-trace.h" |
1c20f0e2 | 57 | #include "index.h" |
0f907de1 | 58 | #include "utils.h" |
b8aa1682 | 59 | #include "lttng-relayd.h" |
d3e2ba59 | 60 | #include "live.h" |
55706a7d | 61 | #include "health-relayd.h" |
4e5b8b82 | 62 | #include "testpoint.h" |
991adae2 | 63 | #include "viewer-stream.h" |
eb702af5 DG |
64 | #include "session.h" |
65 | #include "stream.h" | |
e85cdca9 | 66 | #include "connection.h" |
b8aa1682 JD |
67 | |
68 | /* command line options */ | |
0f907de1 | 69 | char *opt_output_path; |
d1a3048a | 70 | static int opt_daemon, opt_background; |
02a6bb53 MD |
71 | |
72 | /* | |
73 | * We need to wait for listener and live listener threads, as well as | |
74 | * health check thread, before being ready to signal readiness. | |
75 | */ | |
76 | #define NR_LTTNG_RELAY_READY 3 | |
77 | static int lttng_relay_ready = NR_LTTNG_RELAY_READY; | |
78 | static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */ | |
79 | static pid_t child_ppid; /* Internal parent PID use with daemonize. */ | |
80 | ||
095a4ae5 MD |
81 | static struct lttng_uri *control_uri; |
82 | static struct lttng_uri *data_uri; | |
d3e2ba59 | 83 | static struct lttng_uri *live_uri; |
b8aa1682 JD |
84 | |
85 | const char *progname; | |
b8aa1682 | 86 | |
65931c8b MD |
87 | const char *tracing_group_name = DEFAULT_TRACING_GROUP; |
88 | ||
b8aa1682 JD |
89 | /* |
90 | * Quit pipe for all threads. This permits a single cancellation point | |
91 | * for all threads when receiving an event on the pipe. | |
92 | */ | |
3557d456 | 93 | int thread_quit_pipe[2] = { -1, -1 }; |
b8aa1682 JD |
94 | |
95 | /* | |
96 | * This pipe is used to inform the worker thread that a command is queued and | |
97 | * ready to be processed. | |
98 | */ | |
e85cdca9 | 99 | static int relay_conn_pipe[2] = { -1, -1 }; |
b8aa1682 | 100 | |
26c9d55e | 101 | /* Shared between threads */ |
b8aa1682 JD |
102 | static int dispatch_thread_exit; |
103 | ||
104 | static pthread_t listener_thread; | |
105 | static pthread_t dispatcher_thread; | |
106 | static pthread_t worker_thread; | |
65931c8b | 107 | static pthread_t health_thread; |
b8aa1682 | 108 | |
095a4ae5 | 109 | static uint64_t last_relay_stream_id; |
b8aa1682 JD |
110 | |
111 | /* | |
112 | * Relay command queue. | |
113 | * | |
114 | * The relay_thread_listener and relay_thread_dispatcher communicate with this | |
115 | * queue. | |
116 | */ | |
e85cdca9 | 117 | static struct relay_conn_queue relay_conn_queue; |
b8aa1682 JD |
118 | |
119 | /* buffer allocated at startup, used to store the trace data */ | |
095a4ae5 MD |
120 | static char *data_buffer; |
121 | static unsigned int data_buffer_size; | |
b8aa1682 | 122 | |
1c20f0e2 JD |
123 | /* We need those values for the file/dir creation. */ |
124 | static uid_t relayd_uid; | |
125 | static gid_t relayd_gid; | |
126 | ||
d3e2ba59 JD |
127 | /* Global relay stream hash table. */ |
128 | struct lttng_ht *relay_streams_ht; | |
129 | ||
92c6ca54 DG |
130 | /* Global relay viewer stream hash table. */ |
131 | struct lttng_ht *viewer_streams_ht; | |
132 | ||
0a6518b0 DG |
133 | /* Global hash table that stores relay index object. */ |
134 | struct lttng_ht *indexes_ht; | |
135 | ||
55706a7d | 136 | /* Relayd health monitoring */ |
eea7556c | 137 | struct health_app *health_relayd; |
55706a7d | 138 | |
b8aa1682 JD |
139 | /* |
140 | * usage function on stderr | |
141 | */ | |
142 | static | |
143 | void usage(void) | |
144 | { | |
145 | fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname); | |
994fa64f DG |
146 | fprintf(stderr, " -h, --help Display this usage.\n"); |
147 | fprintf(stderr, " -d, --daemonize Start as a daemon.\n"); | |
d1a3048a | 148 | fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n"); |
994fa64f DG |
149 | fprintf(stderr, " -C, --control-port URL Control port listening.\n"); |
150 | fprintf(stderr, " -D, --data-port URL Data port listening.\n"); | |
a6a062a4 | 151 | fprintf(stderr, " -L, --live-port URL Live view port listening.\n"); |
994fa64f DG |
152 | fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n"); |
153 | fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n"); | |
65931c8b | 154 | fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n"); |
b8aa1682 JD |
155 | } |
156 | ||
157 | static | |
158 | int parse_args(int argc, char **argv) | |
159 | { | |
160 | int c; | |
161 | int ret = 0; | |
162 | char *default_address; | |
163 | ||
164 | static struct option long_options[] = { | |
e3678fd8 MD |
165 | { "control-port", 1, 0, 'C', }, |
166 | { "data-port", 1, 0, 'D', }, | |
167 | { "daemonize", 0, 0, 'd', }, | |
65931c8b | 168 | { "group", 1, 0, 'g', }, |
e3678fd8 MD |
169 | { "help", 0, 0, 'h', }, |
170 | { "output", 1, 0, 'o', }, | |
171 | { "verbose", 0, 0, 'v', }, | |
06b55dbb | 172 | { "background", 0, 0, 'b' }, |
095a4ae5 | 173 | { NULL, 0, 0, 0, }, |
b8aa1682 JD |
174 | }; |
175 | ||
176 | while (1) { | |
177 | int option_index = 0; | |
06b55dbb | 178 | c = getopt_long(argc, argv, "dhv" "C:D:L:o:g:b", |
b8aa1682 JD |
179 | long_options, &option_index); |
180 | if (c == -1) { | |
181 | break; | |
182 | } | |
183 | ||
184 | switch (c) { | |
185 | case 0: | |
186 | fprintf(stderr, "option %s", long_options[option_index].name); | |
187 | if (optarg) { | |
188 | fprintf(stderr, " with arg %s\n", optarg); | |
189 | } | |
190 | break; | |
191 | case 'C': | |
192 | ret = uri_parse(optarg, &control_uri); | |
193 | if (ret < 0) { | |
194 | ERR("Invalid control URI specified"); | |
195 | goto exit; | |
196 | } | |
197 | if (control_uri->port == 0) { | |
198 | control_uri->port = DEFAULT_NETWORK_CONTROL_PORT; | |
199 | } | |
200 | break; | |
201 | case 'D': | |
202 | ret = uri_parse(optarg, &data_uri); | |
203 | if (ret < 0) { | |
204 | ERR("Invalid data URI specified"); | |
205 | goto exit; | |
206 | } | |
207 | if (data_uri->port == 0) { | |
208 | data_uri->port = DEFAULT_NETWORK_DATA_PORT; | |
209 | } | |
210 | break; | |
a6a062a4 AM |
211 | case 'L': |
212 | ret = uri_parse(optarg, &live_uri); | |
213 | if (ret < 0) { | |
214 | ERR("Invalid live URI specified"); | |
215 | goto exit; | |
216 | } | |
217 | if (live_uri->port == 0) { | |
218 | live_uri->port = DEFAULT_NETWORK_VIEWER_PORT; | |
219 | } | |
220 | break; | |
b8aa1682 JD |
221 | case 'd': |
222 | opt_daemon = 1; | |
223 | break; | |
06b55dbb DG |
224 | case 'b': |
225 | opt_background = 1; | |
226 | break; | |
65931c8b MD |
227 | case 'g': |
228 | tracing_group_name = optarg; | |
229 | break; | |
b8aa1682 JD |
230 | case 'h': |
231 | usage(); | |
232 | exit(EXIT_FAILURE); | |
233 | case 'o': | |
234 | ret = asprintf(&opt_output_path, "%s", optarg); | |
235 | if (ret < 0) { | |
236 | PERROR("asprintf opt_output_path"); | |
237 | goto exit; | |
238 | } | |
239 | break; | |
240 | case 'v': | |
241 | /* Verbose level can increase using multiple -v */ | |
242 | lttng_opt_verbose += 1; | |
243 | break; | |
244 | default: | |
245 | /* Unknown option or other error. | |
246 | * Error is printed by getopt, just return */ | |
247 | ret = -1; | |
248 | goto exit; | |
249 | } | |
250 | } | |
251 | ||
252 | /* assign default values */ | |
253 | if (control_uri == NULL) { | |
d359cebc MD |
254 | ret = asprintf(&default_address, |
255 | "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d", | |
256 | DEFAULT_NETWORK_CONTROL_PORT); | |
b8aa1682 JD |
257 | if (ret < 0) { |
258 | PERROR("asprintf default data address"); | |
259 | goto exit; | |
260 | } | |
261 | ||
262 | ret = uri_parse(default_address, &control_uri); | |
263 | free(default_address); | |
264 | if (ret < 0) { | |
265 | ERR("Invalid control URI specified"); | |
266 | goto exit; | |
267 | } | |
268 | } | |
269 | if (data_uri == NULL) { | |
d359cebc MD |
270 | ret = asprintf(&default_address, |
271 | "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d", | |
272 | DEFAULT_NETWORK_DATA_PORT); | |
b8aa1682 JD |
273 | if (ret < 0) { |
274 | PERROR("asprintf default data address"); | |
275 | goto exit; | |
276 | } | |
277 | ||
278 | ret = uri_parse(default_address, &data_uri); | |
279 | free(default_address); | |
280 | if (ret < 0) { | |
281 | ERR("Invalid data URI specified"); | |
282 | goto exit; | |
283 | } | |
284 | } | |
d3e2ba59 | 285 | if (live_uri == NULL) { |
d359cebc MD |
286 | ret = asprintf(&default_address, |
287 | "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d", | |
288 | DEFAULT_NETWORK_VIEWER_PORT); | |
d3e2ba59 JD |
289 | if (ret < 0) { |
290 | PERROR("asprintf default viewer control address"); | |
291 | goto exit; | |
292 | } | |
293 | ||
294 | ret = uri_parse(default_address, &live_uri); | |
295 | free(default_address); | |
296 | if (ret < 0) { | |
297 | ERR("Invalid viewer control URI specified"); | |
298 | goto exit; | |
299 | } | |
300 | } | |
b8aa1682 JD |
301 | |
302 | exit: | |
303 | return ret; | |
304 | } | |
305 | ||
306 | /* | |
307 | * Cleanup the daemon | |
308 | */ | |
309 | static | |
310 | void cleanup(void) | |
311 | { | |
b8aa1682 JD |
312 | DBG("Cleaning up"); |
313 | ||
095a4ae5 MD |
314 | /* free the dynamically allocated opt_output_path */ |
315 | free(opt_output_path); | |
316 | ||
a02de639 CB |
317 | /* Close thread quit pipes */ |
318 | utils_close_pipe(thread_quit_pipe); | |
319 | ||
710c1f73 DG |
320 | uri_free(control_uri); |
321 | uri_free(data_uri); | |
a6a062a4 | 322 | /* Live URI is freed in the live thread. */ |
b8aa1682 JD |
323 | } |
324 | ||
325 | /* | |
326 | * Write to writable pipe used to notify a thread. | |
327 | */ | |
328 | static | |
329 | int notify_thread_pipe(int wpipe) | |
330 | { | |
6cd525e8 | 331 | ssize_t ret; |
b8aa1682 | 332 | |
6cd525e8 MD |
333 | ret = lttng_write(wpipe, "!", 1); |
334 | if (ret < 1) { | |
b8aa1682 JD |
335 | PERROR("write poll pipe"); |
336 | } | |
337 | ||
338 | return ret; | |
339 | } | |
340 | ||
65931c8b MD |
341 | static void notify_health_quit_pipe(int *pipe) |
342 | { | |
6cd525e8 | 343 | ssize_t ret; |
65931c8b | 344 | |
6cd525e8 MD |
345 | ret = lttng_write(pipe[1], "4", 1); |
346 | if (ret < 1) { | |
65931c8b MD |
347 | PERROR("write relay health quit"); |
348 | } | |
349 | } | |
350 | ||
b8aa1682 JD |
351 | /* |
352 | * Stop all threads by closing the thread quit pipe. | |
353 | */ | |
354 | static | |
355 | void stop_threads(void) | |
356 | { | |
357 | int ret; | |
358 | ||
359 | /* Stopping all threads */ | |
360 | DBG("Terminating all threads"); | |
361 | ret = notify_thread_pipe(thread_quit_pipe[1]); | |
362 | if (ret < 0) { | |
363 | ERR("write error on thread quit pipe"); | |
364 | } | |
365 | ||
65931c8b MD |
366 | notify_health_quit_pipe(health_quit_pipe); |
367 | ||
b8aa1682 | 368 | /* Dispatch thread */ |
26c9d55e | 369 | CMM_STORE_SHARED(dispatch_thread_exit, 1); |
e85cdca9 | 370 | futex_nto1_wake(&relay_conn_queue.futex); |
b8aa1682 JD |
371 | } |
372 | ||
373 | /* | |
374 | * Signal handler for the daemon | |
375 | * | |
376 | * Simply stop all worker threads, leaving main() return gracefully after | |
377 | * joining all threads and calling cleanup(). | |
378 | */ | |
379 | static | |
380 | void sighandler(int sig) | |
381 | { | |
382 | switch (sig) { | |
383 | case SIGPIPE: | |
384 | DBG("SIGPIPE caught"); | |
385 | return; | |
386 | case SIGINT: | |
387 | DBG("SIGINT caught"); | |
388 | stop_threads(); | |
389 | break; | |
390 | case SIGTERM: | |
391 | DBG("SIGTERM caught"); | |
392 | stop_threads(); | |
393 | break; | |
02a6bb53 MD |
394 | case SIGUSR1: |
395 | CMM_STORE_SHARED(recv_child_signal, 1); | |
396 | break; | |
b8aa1682 JD |
397 | default: |
398 | break; | |
399 | } | |
400 | } | |
401 | ||
402 | /* | |
403 | * Setup signal handler for : | |
404 | * SIGINT, SIGTERM, SIGPIPE | |
405 | */ | |
406 | static | |
407 | int set_signal_handler(void) | |
408 | { | |
409 | int ret = 0; | |
410 | struct sigaction sa; | |
411 | sigset_t sigset; | |
412 | ||
413 | if ((ret = sigemptyset(&sigset)) < 0) { | |
414 | PERROR("sigemptyset"); | |
415 | return ret; | |
416 | } | |
417 | ||
418 | sa.sa_handler = sighandler; | |
419 | sa.sa_mask = sigset; | |
420 | sa.sa_flags = 0; | |
421 | if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) { | |
422 | PERROR("sigaction"); | |
423 | return ret; | |
424 | } | |
425 | ||
426 | if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) { | |
427 | PERROR("sigaction"); | |
428 | return ret; | |
429 | } | |
430 | ||
431 | if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) { | |
432 | PERROR("sigaction"); | |
433 | return ret; | |
434 | } | |
435 | ||
02a6bb53 MD |
436 | if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) { |
437 | PERROR("sigaction"); | |
438 | return ret; | |
439 | } | |
440 | ||
441 | DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT"); | |
b8aa1682 JD |
442 | |
443 | return ret; | |
444 | } | |
445 | ||
02a6bb53 MD |
446 | void lttng_relay_notify_ready(void) |
447 | { | |
448 | /* Notify the parent of the fork() process that we are ready. */ | |
449 | if (opt_daemon || opt_background) { | |
450 | if (uatomic_sub_return(<tng_relay_ready, 1) == 0) { | |
451 | kill(child_ppid, SIGUSR1); | |
452 | } | |
453 | } | |
454 | } | |
455 | ||
b8aa1682 JD |
456 | /* |
457 | * Init thread quit pipe. | |
458 | * | |
459 | * Return -1 on error or 0 if all pipes are created. | |
460 | */ | |
461 | static | |
462 | int init_thread_quit_pipe(void) | |
463 | { | |
a02de639 | 464 | int ret; |
b8aa1682 | 465 | |
a02de639 | 466 | ret = utils_create_pipe_cloexec(thread_quit_pipe); |
b8aa1682 | 467 | |
b8aa1682 JD |
468 | return ret; |
469 | } | |
470 | ||
471 | /* | |
472 | * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set. | |
473 | */ | |
474 | static | |
475 | int create_thread_poll_set(struct lttng_poll_event *events, int size) | |
476 | { | |
477 | int ret; | |
478 | ||
479 | if (events == NULL || size == 0) { | |
480 | ret = -1; | |
481 | goto error; | |
482 | } | |
483 | ||
484 | ret = lttng_poll_create(events, size, LTTNG_CLOEXEC); | |
485 | if (ret < 0) { | |
486 | goto error; | |
487 | } | |
488 | ||
489 | /* Add quit pipe */ | |
532ed517 | 490 | ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR); |
b8aa1682 JD |
491 | if (ret < 0) { |
492 | goto error; | |
493 | } | |
494 | ||
495 | return 0; | |
496 | ||
497 | error: | |
498 | return ret; | |
499 | } | |
500 | ||
501 | /* | |
502 | * Check if the thread quit pipe was triggered. | |
503 | * | |
504 | * Return 1 if it was triggered else 0; | |
505 | */ | |
506 | static | |
507 | int check_thread_quit_pipe(int fd, uint32_t events) | |
508 | { | |
509 | if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) { | |
510 | return 1; | |
511 | } | |
512 | ||
513 | return 0; | |
514 | } | |
515 | ||
516 | /* | |
517 | * Create and init socket from uri. | |
518 | */ | |
519 | static | |
520 | struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri) | |
521 | { | |
522 | int ret; | |
523 | struct lttcomm_sock *sock = NULL; | |
524 | ||
525 | sock = lttcomm_alloc_sock_from_uri(uri); | |
526 | if (sock == NULL) { | |
527 | ERR("Allocating socket"); | |
528 | goto error; | |
529 | } | |
530 | ||
531 | ret = lttcomm_create_sock(sock); | |
532 | if (ret < 0) { | |
533 | goto error; | |
534 | } | |
535 | DBG("Listening on sock %d", sock->fd); | |
536 | ||
537 | ret = sock->ops->bind(sock); | |
538 | if (ret < 0) { | |
539 | goto error; | |
540 | } | |
541 | ||
542 | ret = sock->ops->listen(sock, -1); | |
543 | if (ret < 0) { | |
544 | goto error; | |
545 | ||
546 | } | |
547 | ||
548 | return sock; | |
549 | ||
550 | error: | |
551 | if (sock) { | |
552 | lttcomm_destroy_sock(sock); | |
553 | } | |
554 | return NULL; | |
555 | } | |
556 | ||
173af62f DG |
557 | /* |
558 | * Return nonzero if stream needs to be closed. | |
559 | */ | |
560 | static | |
561 | int close_stream_check(struct relay_stream *stream) | |
562 | { | |
173af62f | 563 | if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) { |
f7079f67 DG |
564 | /* |
565 | * We are about to close the stream so set the data pending flag to 1 | |
566 | * which will make the end data pending command skip the stream which | |
567 | * is now closed and ready. Note that after proceeding to a file close, | |
568 | * the written file is ready for reading. | |
569 | */ | |
570 | stream->data_pending_check_done = 1; | |
173af62f DG |
571 | return 1; |
572 | } | |
573 | return 0; | |
574 | } | |
575 | ||
eb702af5 DG |
576 | static void try_close_stream(struct relay_session *session, |
577 | struct relay_stream *stream) | |
578 | { | |
579 | int ret; | |
580 | struct ctf_trace *ctf_trace; | |
581 | ||
582 | assert(session); | |
583 | assert(stream); | |
584 | ||
585 | if (!close_stream_check(stream)) { | |
586 | /* Can't close it, not ready for that. */ | |
587 | goto end; | |
588 | } | |
589 | ||
590 | ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, | |
591 | stream->path_name); | |
592 | assert(ctf_trace); | |
593 | ||
594 | pthread_mutex_lock(&session->viewer_ready_lock); | |
595 | ctf_trace->invalid_flag = 1; | |
596 | pthread_mutex_unlock(&session->viewer_ready_lock); | |
597 | ||
598 | ret = stream_close(session, stream); | |
50979bcd | 599 | if (ret || session->snapshot) { |
eb702af5 DG |
600 | /* Already close thus the ctf trace is being or has been destroyed. */ |
601 | goto end; | |
602 | } | |
603 | ||
604 | ctf_trace_try_destroy(session, ctf_trace); | |
605 | ||
606 | end: | |
607 | return; | |
608 | } | |
609 | ||
b8aa1682 JD |
610 | /* |
611 | * This thread manages the listening for new connections on the network | |
612 | */ | |
613 | static | |
614 | void *relay_thread_listener(void *data) | |
615 | { | |
095a4ae5 | 616 | int i, ret, pollfd, err = -1; |
b8aa1682 JD |
617 | uint32_t revents, nb_fd; |
618 | struct lttng_poll_event events; | |
619 | struct lttcomm_sock *control_sock, *data_sock; | |
620 | ||
b8aa1682 JD |
621 | DBG("[thread] Relay listener started"); |
622 | ||
55706a7d MD |
623 | health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER); |
624 | ||
f385ae0a MD |
625 | health_code_update(); |
626 | ||
b8aa1682 JD |
627 | control_sock = relay_init_sock(control_uri); |
628 | if (!control_sock) { | |
095a4ae5 | 629 | goto error_sock_control; |
b8aa1682 JD |
630 | } |
631 | ||
632 | data_sock = relay_init_sock(data_uri); | |
633 | if (!data_sock) { | |
095a4ae5 | 634 | goto error_sock_relay; |
b8aa1682 JD |
635 | } |
636 | ||
637 | /* | |
638 | * Pass 3 as size here for the thread quit pipe, control and data socket. | |
639 | */ | |
640 | ret = create_thread_poll_set(&events, 3); | |
641 | if (ret < 0) { | |
642 | goto error_create_poll; | |
643 | } | |
644 | ||
645 | /* Add the control socket */ | |
646 | ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP); | |
647 | if (ret < 0) { | |
648 | goto error_poll_add; | |
649 | } | |
650 | ||
651 | /* Add the data socket */ | |
652 | ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP); | |
653 | if (ret < 0) { | |
654 | goto error_poll_add; | |
655 | } | |
656 | ||
02a6bb53 MD |
657 | lttng_relay_notify_ready(); |
658 | ||
4e5b8b82 MD |
659 | if (testpoint(relayd_thread_listener)) { |
660 | goto error_testpoint; | |
661 | } | |
662 | ||
b8aa1682 | 663 | while (1) { |
f385ae0a MD |
664 | health_code_update(); |
665 | ||
b8aa1682 JD |
666 | DBG("Listener accepting connections"); |
667 | ||
b8aa1682 | 668 | restart: |
f385ae0a | 669 | health_poll_entry(); |
b8aa1682 | 670 | ret = lttng_poll_wait(&events, -1); |
f385ae0a | 671 | health_poll_exit(); |
b8aa1682 JD |
672 | if (ret < 0) { |
673 | /* | |
674 | * Restart interrupted system call. | |
675 | */ | |
676 | if (errno == EINTR) { | |
677 | goto restart; | |
678 | } | |
679 | goto error; | |
680 | } | |
681 | ||
0d9c5d77 DG |
682 | nb_fd = ret; |
683 | ||
b8aa1682 JD |
684 | DBG("Relay new connection received"); |
685 | for (i = 0; i < nb_fd; i++) { | |
f385ae0a MD |
686 | health_code_update(); |
687 | ||
b8aa1682 JD |
688 | /* Fetch once the poll data */ |
689 | revents = LTTNG_POLL_GETEV(&events, i); | |
690 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
691 | ||
692 | /* Thread quit pipe has been closed. Killing thread. */ | |
693 | ret = check_thread_quit_pipe(pollfd, revents); | |
694 | if (ret) { | |
095a4ae5 MD |
695 | err = 0; |
696 | goto exit; | |
b8aa1682 JD |
697 | } |
698 | ||
699 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
700 | ERR("socket poll error"); | |
701 | goto error; | |
702 | } else if (revents & LPOLLIN) { | |
4b7f17b2 | 703 | /* |
e85cdca9 DG |
704 | * Get allocated in this thread, enqueued to a global queue, |
705 | * dequeued and freed in the worker thread. | |
4b7f17b2 | 706 | */ |
e85cdca9 DG |
707 | int val = 1; |
708 | struct relay_connection *new_conn; | |
4b7f17b2 | 709 | struct lttcomm_sock *newsock; |
b8aa1682 | 710 | |
e85cdca9 DG |
711 | new_conn = connection_create(); |
712 | if (!new_conn) { | |
b8aa1682 JD |
713 | goto error; |
714 | } | |
715 | ||
716 | if (pollfd == data_sock->fd) { | |
e85cdca9 | 717 | new_conn->type = RELAY_DATA; |
b8aa1682 | 718 | newsock = data_sock->ops->accept(data_sock); |
e85cdca9 DG |
719 | DBG("Relay data connection accepted, socket %d", |
720 | newsock->fd); | |
4b7f17b2 MD |
721 | } else { |
722 | assert(pollfd == control_sock->fd); | |
e85cdca9 | 723 | new_conn->type = RELAY_CONTROL; |
b8aa1682 | 724 | newsock = control_sock->ops->accept(control_sock); |
e85cdca9 DG |
725 | DBG("Relay control connection accepted, socket %d", |
726 | newsock->fd); | |
b8aa1682 | 727 | } |
e85cdca9 DG |
728 | if (!newsock) { |
729 | PERROR("accepting sock"); | |
730 | connection_free(new_conn); | |
731 | goto error; | |
732 | } | |
733 | ||
734 | ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val, | |
735 | sizeof(val)); | |
b8aa1682 JD |
736 | if (ret < 0) { |
737 | PERROR("setsockopt inet"); | |
4b7f17b2 | 738 | lttcomm_destroy_sock(newsock); |
e85cdca9 | 739 | connection_free(new_conn); |
b8aa1682 JD |
740 | goto error; |
741 | } | |
e85cdca9 DG |
742 | new_conn->sock = newsock; |
743 | ||
744 | /* Enqueue request for the dispatcher thread. */ | |
745 | cds_wfq_enqueue(&relay_conn_queue.queue, &new_conn->qnode); | |
b8aa1682 JD |
746 | |
747 | /* | |
e85cdca9 DG |
748 | * Wake the dispatch queue futex. Implicit memory barrier with |
749 | * the exchange in cds_wfq_enqueue. | |
b8aa1682 | 750 | */ |
e85cdca9 | 751 | futex_nto1_wake(&relay_conn_queue.futex); |
b8aa1682 JD |
752 | } |
753 | } | |
754 | } | |
755 | ||
095a4ae5 | 756 | exit: |
b8aa1682 JD |
757 | error: |
758 | error_poll_add: | |
4e5b8b82 | 759 | error_testpoint: |
b8aa1682 JD |
760 | lttng_poll_clean(&events); |
761 | error_create_poll: | |
095a4ae5 MD |
762 | if (data_sock->fd >= 0) { |
763 | ret = data_sock->ops->close(data_sock); | |
b8aa1682 JD |
764 | if (ret) { |
765 | PERROR("close"); | |
766 | } | |
b8aa1682 | 767 | } |
095a4ae5 MD |
768 | lttcomm_destroy_sock(data_sock); |
769 | error_sock_relay: | |
770 | if (control_sock->fd >= 0) { | |
771 | ret = control_sock->ops->close(control_sock); | |
b8aa1682 JD |
772 | if (ret) { |
773 | PERROR("close"); | |
774 | } | |
b8aa1682 | 775 | } |
095a4ae5 MD |
776 | lttcomm_destroy_sock(control_sock); |
777 | error_sock_control: | |
778 | if (err) { | |
f385ae0a MD |
779 | health_error(); |
780 | ERR("Health error occurred in %s", __func__); | |
095a4ae5 | 781 | } |
55706a7d | 782 | health_unregister(health_relayd); |
b8aa1682 JD |
783 | DBG("Relay listener thread cleanup complete"); |
784 | stop_threads(); | |
b8aa1682 JD |
785 | return NULL; |
786 | } | |
787 | ||
788 | /* | |
789 | * This thread manages the dispatching of the requests to worker threads | |
790 | */ | |
791 | static | |
792 | void *relay_thread_dispatcher(void *data) | |
793 | { | |
6cd525e8 MD |
794 | int err = -1; |
795 | ssize_t ret; | |
b8aa1682 | 796 | struct cds_wfq_node *node; |
e85cdca9 | 797 | struct relay_connection *new_conn = NULL; |
b8aa1682 JD |
798 | |
799 | DBG("[thread] Relay dispatcher started"); | |
800 | ||
55706a7d MD |
801 | health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER); |
802 | ||
4e5b8b82 MD |
803 | if (testpoint(relayd_thread_dispatcher)) { |
804 | goto error_testpoint; | |
805 | } | |
806 | ||
f385ae0a MD |
807 | health_code_update(); |
808 | ||
26c9d55e | 809 | while (!CMM_LOAD_SHARED(dispatch_thread_exit)) { |
f385ae0a MD |
810 | health_code_update(); |
811 | ||
b8aa1682 | 812 | /* Atomically prepare the queue futex */ |
e85cdca9 | 813 | futex_nto1_prepare(&relay_conn_queue.futex); |
b8aa1682 JD |
814 | |
815 | do { | |
f385ae0a MD |
816 | health_code_update(); |
817 | ||
b8aa1682 | 818 | /* Dequeue commands */ |
e85cdca9 | 819 | node = cds_wfq_dequeue_blocking(&relay_conn_queue.queue); |
b8aa1682 JD |
820 | if (node == NULL) { |
821 | DBG("Woken up but nothing in the relay command queue"); | |
822 | /* Continue thread execution */ | |
823 | break; | |
824 | } | |
e85cdca9 | 825 | new_conn = caa_container_of(node, struct relay_connection, qnode); |
b8aa1682 | 826 | |
e85cdca9 | 827 | DBG("Dispatching request waiting on sock %d", new_conn->sock->fd); |
b8aa1682 JD |
828 | |
829 | /* | |
e85cdca9 DG |
830 | * Inform worker thread of the new request. This call is blocking |
831 | * so we can be assured that the data will be read at some point in | |
832 | * time or wait to the end of the world :) | |
b8aa1682 | 833 | */ |
e85cdca9 DG |
834 | ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn)); |
835 | if (ret < 0) { | |
836 | PERROR("write connection pipe"); | |
837 | connection_destroy(new_conn); | |
b8aa1682 JD |
838 | goto error; |
839 | } | |
840 | } while (node != NULL); | |
841 | ||
842 | /* Futex wait on queue. Blocking call on futex() */ | |
f385ae0a | 843 | health_poll_entry(); |
e85cdca9 | 844 | futex_nto1_wait(&relay_conn_queue.futex); |
f385ae0a | 845 | health_poll_exit(); |
b8aa1682 JD |
846 | } |
847 | ||
f385ae0a MD |
848 | /* Normal exit, no error */ |
849 | err = 0; | |
850 | ||
b8aa1682 | 851 | error: |
4e5b8b82 | 852 | error_testpoint: |
f385ae0a MD |
853 | if (err) { |
854 | health_error(); | |
855 | ERR("Health error occurred in %s", __func__); | |
856 | } | |
55706a7d | 857 | health_unregister(health_relayd); |
b8aa1682 JD |
858 | DBG("Dispatch thread dying"); |
859 | stop_threads(); | |
860 | return NULL; | |
861 | } | |
862 | ||
eb702af5 | 863 | static void try_close_streams(struct relay_session *session) |
d3e2ba59 | 864 | { |
eb702af5 | 865 | struct ctf_trace *ctf_trace; |
94d49140 JD |
866 | struct lttng_ht_iter iter; |
867 | ||
eb702af5 | 868 | assert(session); |
94d49140 | 869 | |
eb702af5 DG |
870 | pthread_mutex_lock(&session->viewer_ready_lock); |
871 | rcu_read_lock(); | |
872 | cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace, | |
873 | node.node) { | |
874 | struct relay_stream *stream; | |
94d49140 | 875 | |
eb702af5 DG |
876 | /* Close streams. */ |
877 | cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) { | |
878 | stream_close(session, stream); | |
94d49140 | 879 | } |
94d49140 | 880 | |
eb702af5 DG |
881 | ctf_trace->invalid_flag = 1; |
882 | ctf_trace_try_destroy(session, ctf_trace); | |
157df586 | 883 | } |
eb702af5 DG |
884 | rcu_read_unlock(); |
885 | pthread_mutex_unlock(&session->viewer_ready_lock); | |
94d49140 JD |
886 | } |
887 | ||
b8aa1682 | 888 | /* |
eb702af5 | 889 | * Try to destroy a session within a connection. |
b8aa1682 | 890 | */ |
e85cdca9 | 891 | static void destroy_session(struct relay_session *session, |
d3e2ba59 | 892 | struct lttng_ht *sessions_ht) |
b8aa1682 | 893 | { |
e85cdca9 | 894 | assert(session); |
eb702af5 | 895 | assert(sessions_ht); |
b8aa1682 | 896 | |
eb702af5 | 897 | /* Indicate that this session can be destroyed from now on. */ |
e85cdca9 | 898 | session->close_flag = 1; |
b8aa1682 | 899 | |
e85cdca9 | 900 | try_close_streams(session); |
5b6d8097 | 901 | |
eb702af5 DG |
902 | /* |
903 | * This will try to delete and destroy the session if no viewer is attached | |
904 | * to it meaning the refcount is down to zero. | |
905 | */ | |
e85cdca9 | 906 | session_try_destroy(sessions_ht, session); |
b8aa1682 JD |
907 | } |
908 | ||
1c20f0e2 JD |
909 | /* |
910 | * Copy index data from the control port to a given index object. | |
911 | */ | |
912 | static void copy_index_control_data(struct relay_index *index, | |
913 | struct lttcomm_relayd_index *data) | |
914 | { | |
915 | assert(index); | |
916 | assert(data); | |
917 | ||
918 | /* | |
919 | * The index on disk is encoded in big endian, so we don't need to convert | |
920 | * the data received on the network. The data_offset value is NEVER | |
921 | * modified here and is updated by the data thread. | |
922 | */ | |
923 | index->index_data.packet_size = data->packet_size; | |
924 | index->index_data.content_size = data->content_size; | |
925 | index->index_data.timestamp_begin = data->timestamp_begin; | |
926 | index->index_data.timestamp_end = data->timestamp_end; | |
927 | index->index_data.events_discarded = data->events_discarded; | |
928 | index->index_data.stream_id = data->stream_id; | |
929 | } | |
930 | ||
c5b6f4f0 DG |
931 | /* |
932 | * Handle the RELAYD_CREATE_SESSION command. | |
933 | * | |
934 | * On success, send back the session id or else return a negative value. | |
935 | */ | |
936 | static | |
937 | int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr, | |
e85cdca9 | 938 | struct relay_connection *conn) |
c5b6f4f0 DG |
939 | { |
940 | int ret = 0, send_ret; | |
941 | struct relay_session *session; | |
942 | struct lttcomm_relayd_status_session reply; | |
943 | ||
944 | assert(recv_hdr); | |
e85cdca9 | 945 | assert(conn); |
c5b6f4f0 DG |
946 | |
947 | memset(&reply, 0, sizeof(reply)); | |
948 | ||
eb702af5 DG |
949 | session = session_create(); |
950 | if (!session) { | |
c5b6f4f0 DG |
951 | ret = -1; |
952 | goto error; | |
953 | } | |
e85cdca9 DG |
954 | session->minor = conn->minor; |
955 | session->major = conn->major; | |
956 | conn->session_id = session->id; | |
957 | conn->session = session; | |
c5b6f4f0 DG |
958 | |
959 | reply.session_id = htobe64(session->id); | |
960 | ||
e85cdca9 | 961 | switch (conn->minor) { |
eb702af5 DG |
962 | case 1: |
963 | case 2: | |
964 | case 3: | |
965 | break; | |
966 | case 4: /* LTTng sessiond 2.4 */ | |
967 | default: | |
e85cdca9 | 968 | ret = cmd_create_session_2_4(conn, session); |
d3e2ba59 JD |
969 | } |
970 | ||
e85cdca9 | 971 | lttng_ht_add_unique_u64(conn->sessions_ht, &session->session_n); |
c5b6f4f0 DG |
972 | DBG("Created session %" PRIu64, session->id); |
973 | ||
974 | error: | |
975 | if (ret < 0) { | |
976 | reply.ret_code = htobe32(LTTNG_ERR_FATAL); | |
977 | } else { | |
978 | reply.ret_code = htobe32(LTTNG_OK); | |
979 | } | |
980 | ||
e85cdca9 | 981 | send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); |
c5b6f4f0 DG |
982 | if (send_ret < 0) { |
983 | ERR("Relayd sending session id"); | |
4169f5ad | 984 | ret = send_ret; |
c5b6f4f0 DG |
985 | } |
986 | ||
987 | return ret; | |
988 | } | |
989 | ||
814fcae4 JD |
990 | /* |
991 | * When we have received all the streams and the metadata for a channel, | |
992 | * we make them visible to the viewer threads. | |
993 | */ | |
994 | static | |
e85cdca9 | 995 | void set_viewer_ready_flag(struct relay_connection *conn) |
814fcae4 | 996 | { |
eb702af5 | 997 | struct relay_stream *stream, *tmp_stream; |
814fcae4 | 998 | |
e85cdca9 DG |
999 | pthread_mutex_lock(&conn->session->viewer_ready_lock); |
1000 | cds_list_for_each_entry_safe(stream, tmp_stream, &conn->recv_head, | |
eb702af5 | 1001 | recv_list) { |
814fcae4 | 1002 | stream->viewer_ready = 1; |
eb702af5 | 1003 | cds_list_del(&stream->recv_list); |
814fcae4 | 1004 | } |
e85cdca9 | 1005 | pthread_mutex_unlock(&conn->session->viewer_ready_lock); |
814fcae4 JD |
1006 | return; |
1007 | } | |
1008 | ||
1009 | /* | |
1010 | * Add a recv handle node to the connection recv list with the given stream | |
1011 | * handle. A new node is allocated thus must be freed when the node is deleted | |
1012 | * from the list. | |
1013 | */ | |
e85cdca9 DG |
1014 | static void queue_stream(struct relay_stream *stream, |
1015 | struct relay_connection *conn) | |
814fcae4 | 1016 | { |
e85cdca9 | 1017 | assert(conn); |
eb702af5 | 1018 | assert(stream); |
814fcae4 | 1019 | |
e85cdca9 | 1020 | cds_list_add(&stream->recv_list, &conn->recv_head); |
814fcae4 JD |
1021 | } |
1022 | ||
b8aa1682 JD |
1023 | /* |
1024 | * relay_add_stream: allocate a new stream for a session | |
1025 | */ | |
1026 | static | |
1027 | int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, | |
e85cdca9 | 1028 | struct relay_connection *conn) |
b8aa1682 | 1029 | { |
eb702af5 | 1030 | int ret, send_ret; |
e85cdca9 | 1031 | struct relay_session *session = conn->session; |
b8aa1682 JD |
1032 | struct relay_stream *stream = NULL; |
1033 | struct lttcomm_relayd_status_stream reply; | |
eb702af5 | 1034 | struct ctf_trace *trace; |
b8aa1682 | 1035 | |
e85cdca9 | 1036 | if (!session || conn->version_check_done == 0) { |
b8aa1682 JD |
1037 | ERR("Trying to add a stream before version check"); |
1038 | ret = -1; | |
1039 | goto end_no_session; | |
1040 | } | |
1041 | ||
b8aa1682 JD |
1042 | stream = zmalloc(sizeof(struct relay_stream)); |
1043 | if (stream == NULL) { | |
1044 | PERROR("relay stream zmalloc"); | |
1045 | ret = -1; | |
1046 | goto end_no_session; | |
1047 | } | |
1048 | ||
e85cdca9 | 1049 | switch (conn->minor) { |
0f907de1 | 1050 | case 1: /* LTTng sessiond 2.1 */ |
e85cdca9 | 1051 | ret = cmd_recv_stream_2_1(conn, stream); |
0f907de1 JD |
1052 | break; |
1053 | case 2: /* LTTng sessiond 2.2 */ | |
1054 | default: | |
e85cdca9 | 1055 | ret = cmd_recv_stream_2_2(conn, stream); |
0f907de1 JD |
1056 | break; |
1057 | } | |
1058 | if (ret < 0) { | |
1059 | goto err_free_stream; | |
1060 | } | |
1061 | ||
9d1bbf21 | 1062 | rcu_read_lock(); |
b8aa1682 | 1063 | stream->stream_handle = ++last_relay_stream_id; |
173af62f | 1064 | stream->prev_seq = -1ULL; |
eb702af5 | 1065 | stream->session_id = session->id; |
1c20f0e2 | 1066 | stream->index_fd = -1; |
d3e2ba59 | 1067 | stream->read_index_fd = -1; |
5e372a51 | 1068 | stream->ctf_stream_id = -1ULL; |
eb702af5 | 1069 | lttng_ht_node_init_u64(&stream->node, stream->stream_handle); |
d3e2ba59 | 1070 | pthread_mutex_init(&stream->lock, NULL); |
b8aa1682 | 1071 | |
0f907de1 | 1072 | ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG); |
b8aa1682 | 1073 | if (ret < 0) { |
b8aa1682 JD |
1074 | ERR("relay creating output directory"); |
1075 | goto end; | |
1076 | } | |
1077 | ||
be96a7d1 DG |
1078 | /* |
1079 | * No need to use run_as API here because whatever we receives, the relayd | |
1080 | * uses its own credentials for the stream files. | |
1081 | */ | |
0f907de1 | 1082 | ret = utils_create_stream_file(stream->path_name, stream->channel_name, |
1c20f0e2 | 1083 | stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL); |
b8aa1682 | 1084 | if (ret < 0) { |
0f907de1 | 1085 | ERR("Create output file"); |
b8aa1682 JD |
1086 | goto end; |
1087 | } | |
b8aa1682 | 1088 | stream->fd = ret; |
0f907de1 JD |
1089 | if (stream->tracefile_size) { |
1090 | DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name); | |
1091 | } else { | |
1092 | DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name); | |
1093 | } | |
b8aa1682 | 1094 | |
eb702af5 DG |
1095 | trace = ctf_trace_find_by_path(session->ctf_traces_ht, stream->path_name); |
1096 | if (!trace) { | |
1097 | trace = ctf_trace_create(stream->path_name); | |
1098 | if (!trace) { | |
d3e2ba59 JD |
1099 | ret = -1; |
1100 | goto end; | |
1101 | } | |
eb702af5 DG |
1102 | ctf_trace_add(session->ctf_traces_ht, trace); |
1103 | } | |
1104 | ctf_trace_get_ref(trace); | |
1105 | ||
1106 | if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) { | |
1107 | stream->metadata_flag = 1; | |
1108 | /* Assign quick reference to the metadata stream in the trace. */ | |
1109 | trace->metadata_stream = stream; | |
d3e2ba59 | 1110 | } |
d3e2ba59 | 1111 | |
814fcae4 | 1112 | /* |
eb702af5 DG |
1113 | * Add the stream in the recv list of the connection. Once the end stream |
1114 | * message is received, this list is emptied and streams are set with the | |
1115 | * viewer ready flag. | |
814fcae4 | 1116 | */ |
e85cdca9 | 1117 | queue_stream(stream, conn); |
814fcae4 | 1118 | |
eb702af5 DG |
1119 | /* |
1120 | * Both in the ctf_trace object and the global stream ht since the data | |
1121 | * side of the relayd does not have the concept of session. | |
1122 | */ | |
1123 | lttng_ht_add_unique_u64(relay_streams_ht, &stream->node); | |
1124 | cds_list_add_tail(&stream->trace_list, &trace->stream_list); | |
b8aa1682 | 1125 | |
87b576ec | 1126 | session->stream_count++; |
d3e2ba59 | 1127 | |
1c20f0e2 JD |
1128 | DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name, |
1129 | stream->stream_handle); | |
b8aa1682 JD |
1130 | |
1131 | end: | |
239f3aec | 1132 | memset(&reply, 0, sizeof(reply)); |
5af40280 | 1133 | reply.handle = htobe64(stream->stream_handle); |
b8aa1682 JD |
1134 | /* send the session id to the client or a negative return code on error */ |
1135 | if (ret < 0) { | |
f73fabfd | 1136 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
5af40280 CB |
1137 | /* stream was not properly added to the ht, so free it */ |
1138 | free(stream); | |
b8aa1682 | 1139 | } else { |
f73fabfd | 1140 | reply.ret_code = htobe32(LTTNG_OK); |
b8aa1682 | 1141 | } |
5af40280 | 1142 | |
e85cdca9 | 1143 | send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, |
b8aa1682 JD |
1144 | sizeof(struct lttcomm_relayd_status_stream), 0); |
1145 | if (send_ret < 0) { | |
1146 | ERR("Relay sending stream id"); | |
4169f5ad | 1147 | ret = send_ret; |
b8aa1682 | 1148 | } |
9d1bbf21 | 1149 | rcu_read_unlock(); |
b8aa1682 JD |
1150 | |
1151 | end_no_session: | |
1152 | return ret; | |
0f907de1 JD |
1153 | |
1154 | err_free_stream: | |
1155 | free(stream->path_name); | |
1156 | free(stream->channel_name); | |
1157 | free(stream); | |
1158 | return ret; | |
b8aa1682 JD |
1159 | } |
1160 | ||
173af62f DG |
1161 | /* |
1162 | * relay_close_stream: close a specific stream | |
1163 | */ | |
1164 | static | |
1165 | int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, | |
e85cdca9 | 1166 | struct relay_connection *conn) |
173af62f | 1167 | { |
94d49140 | 1168 | int ret, send_ret; |
e85cdca9 | 1169 | struct relay_session *session = conn->session; |
173af62f DG |
1170 | struct lttcomm_relayd_close_stream stream_info; |
1171 | struct lttcomm_relayd_generic_reply reply; | |
1172 | struct relay_stream *stream; | |
173af62f DG |
1173 | |
1174 | DBG("Close stream received"); | |
1175 | ||
e85cdca9 | 1176 | if (!session || conn->version_check_done == 0) { |
173af62f DG |
1177 | ERR("Trying to close a stream before version check"); |
1178 | ret = -1; | |
1179 | goto end_no_session; | |
1180 | } | |
1181 | ||
e85cdca9 | 1182 | ret = conn->sock->ops->recvmsg(conn->sock, &stream_info, |
7c5aef62 | 1183 | sizeof(struct lttcomm_relayd_close_stream), 0); |
173af62f | 1184 | if (ret < sizeof(struct lttcomm_relayd_close_stream)) { |
a6cd2b97 DG |
1185 | if (ret == 0) { |
1186 | /* Orderly shutdown. Not necessary to print an error. */ | |
e85cdca9 | 1187 | DBG("Socket %d did an orderly shutdown", conn->sock->fd); |
a6cd2b97 DG |
1188 | } else { |
1189 | ERR("Relay didn't receive valid add_stream struct size : %d", ret); | |
1190 | } | |
173af62f DG |
1191 | ret = -1; |
1192 | goto end_no_session; | |
1193 | } | |
1194 | ||
1195 | rcu_read_lock(); | |
eb702af5 DG |
1196 | stream = stream_find_by_id(relay_streams_ht, |
1197 | be64toh(stream_info.stream_id)); | |
173af62f DG |
1198 | if (!stream) { |
1199 | ret = -1; | |
1200 | goto end_unlock; | |
1201 | } | |
1202 | ||
8e2583a4 | 1203 | stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num); |
173af62f | 1204 | stream->close_flag = 1; |
87b576ec JD |
1205 | session->stream_count--; |
1206 | assert(session->stream_count >= 0); | |
173af62f | 1207 | |
eb702af5 DG |
1208 | /* Check if we can close it or else the data will do it. */ |
1209 | try_close_stream(session, stream); | |
173af62f DG |
1210 | |
1211 | end_unlock: | |
1212 | rcu_read_unlock(); | |
1213 | ||
239f3aec | 1214 | memset(&reply, 0, sizeof(reply)); |
173af62f | 1215 | if (ret < 0) { |
f73fabfd | 1216 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
173af62f | 1217 | } else { |
f73fabfd | 1218 | reply.ret_code = htobe32(LTTNG_OK); |
173af62f | 1219 | } |
e85cdca9 | 1220 | send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, |
173af62f DG |
1221 | sizeof(struct lttcomm_relayd_generic_reply), 0); |
1222 | if (send_ret < 0) { | |
1223 | ERR("Relay sending stream id"); | |
4169f5ad | 1224 | ret = send_ret; |
173af62f DG |
1225 | } |
1226 | ||
1227 | end_no_session: | |
1228 | return ret; | |
1229 | } | |
1230 | ||
b8aa1682 JD |
1231 | /* |
1232 | * relay_unknown_command: send -1 if received unknown command | |
1233 | */ | |
1234 | static | |
e85cdca9 | 1235 | void relay_unknown_command(struct relay_connection *conn) |
b8aa1682 JD |
1236 | { |
1237 | struct lttcomm_relayd_generic_reply reply; | |
1238 | int ret; | |
1239 | ||
239f3aec | 1240 | memset(&reply, 0, sizeof(reply)); |
f73fabfd | 1241 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
e85cdca9 | 1242 | ret = conn->sock->ops->sendmsg(conn->sock, &reply, |
b8aa1682 JD |
1243 | sizeof(struct lttcomm_relayd_generic_reply), 0); |
1244 | if (ret < 0) { | |
1245 | ERR("Relay sending unknown command"); | |
1246 | } | |
1247 | } | |
1248 | ||
1249 | /* | |
1250 | * relay_start: send an acknowledgment to the client to tell if we are | |
1251 | * ready to receive data. We are ready if a session is established. | |
1252 | */ | |
1253 | static | |
1254 | int relay_start(struct lttcomm_relayd_hdr *recv_hdr, | |
e85cdca9 | 1255 | struct relay_connection *conn) |
b8aa1682 | 1256 | { |
f73fabfd | 1257 | int ret = htobe32(LTTNG_OK); |
b8aa1682 | 1258 | struct lttcomm_relayd_generic_reply reply; |
e85cdca9 | 1259 | struct relay_session *session = conn->session; |
b8aa1682 JD |
1260 | |
1261 | if (!session) { | |
1262 | DBG("Trying to start the streaming without a session established"); | |
f73fabfd | 1263 | ret = htobe32(LTTNG_ERR_UNK); |
b8aa1682 JD |
1264 | } |
1265 | ||
239f3aec | 1266 | memset(&reply, 0, sizeof(reply)); |
b8aa1682 | 1267 | reply.ret_code = ret; |
e85cdca9 | 1268 | ret = conn->sock->ops->sendmsg(conn->sock, &reply, |
b8aa1682 JD |
1269 | sizeof(struct lttcomm_relayd_generic_reply), 0); |
1270 | if (ret < 0) { | |
1271 | ERR("Relay sending start ack"); | |
1272 | } | |
1273 | ||
1274 | return ret; | |
1275 | } | |
1276 | ||
1d4dfdef DG |
1277 | /* |
1278 | * Append padding to the file pointed by the file descriptor fd. | |
1279 | */ | |
1280 | static int write_padding_to_file(int fd, uint32_t size) | |
1281 | { | |
6cd525e8 | 1282 | ssize_t ret = 0; |
1d4dfdef DG |
1283 | char *zeros; |
1284 | ||
1285 | if (size == 0) { | |
1286 | goto end; | |
1287 | } | |
1288 | ||
1289 | zeros = zmalloc(size); | |
1290 | if (zeros == NULL) { | |
1291 | PERROR("zmalloc zeros for padding"); | |
1292 | ret = -1; | |
1293 | goto end; | |
1294 | } | |
1295 | ||
6cd525e8 MD |
1296 | ret = lttng_write(fd, zeros, size); |
1297 | if (ret < size) { | |
1d4dfdef DG |
1298 | PERROR("write padding to file"); |
1299 | } | |
1300 | ||
e986c7a1 DG |
1301 | free(zeros); |
1302 | ||
1d4dfdef DG |
1303 | end: |
1304 | return ret; | |
1305 | } | |
1306 | ||
b8aa1682 JD |
1307 | /* |
1308 | * relay_recv_metadata: receive the metada for the session. | |
1309 | */ | |
1310 | static | |
1311 | int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, | |
e85cdca9 | 1312 | struct relay_connection *conn) |
b8aa1682 | 1313 | { |
f73fabfd | 1314 | int ret = htobe32(LTTNG_OK); |
6cd525e8 | 1315 | ssize_t size_ret; |
e85cdca9 | 1316 | struct relay_session *session = conn->session; |
b8aa1682 JD |
1317 | struct lttcomm_relayd_metadata_payload *metadata_struct; |
1318 | struct relay_stream *metadata_stream; | |
1319 | uint64_t data_size, payload_size; | |
eb702af5 | 1320 | struct ctf_trace *ctf_trace; |
b8aa1682 JD |
1321 | |
1322 | if (!session) { | |
1323 | ERR("Metadata sent before version check"); | |
1324 | ret = -1; | |
1325 | goto end; | |
1326 | } | |
1327 | ||
f6416125 MD |
1328 | data_size = payload_size = be64toh(recv_hdr->data_size); |
1329 | if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) { | |
1330 | ERR("Incorrect data size"); | |
1331 | ret = -1; | |
1332 | goto end; | |
1333 | } | |
1334 | payload_size -= sizeof(struct lttcomm_relayd_metadata_payload); | |
1335 | ||
b8aa1682 | 1336 | if (data_buffer_size < data_size) { |
d7b3776f | 1337 | /* In case the realloc fails, we can free the memory */ |
c617c0c6 MD |
1338 | char *tmp_data_ptr; |
1339 | ||
1340 | tmp_data_ptr = realloc(data_buffer, data_size); | |
1341 | if (!tmp_data_ptr) { | |
b8aa1682 | 1342 | ERR("Allocating data buffer"); |
c617c0c6 | 1343 | free(data_buffer); |
b8aa1682 JD |
1344 | ret = -1; |
1345 | goto end; | |
1346 | } | |
c617c0c6 | 1347 | data_buffer = tmp_data_ptr; |
b8aa1682 JD |
1348 | data_buffer_size = data_size; |
1349 | } | |
1350 | memset(data_buffer, 0, data_size); | |
77c7c900 | 1351 | DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size); |
e85cdca9 | 1352 | ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0); |
b8aa1682 | 1353 | if (ret < 0 || ret != data_size) { |
a6cd2b97 DG |
1354 | if (ret == 0) { |
1355 | /* Orderly shutdown. Not necessary to print an error. */ | |
e85cdca9 | 1356 | DBG("Socket %d did an orderly shutdown", conn->sock->fd); |
a6cd2b97 DG |
1357 | } else { |
1358 | ERR("Relay didn't receive the whole metadata"); | |
1359 | } | |
b8aa1682 | 1360 | ret = -1; |
b8aa1682 JD |
1361 | goto end; |
1362 | } | |
1363 | metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer; | |
9d1bbf21 MD |
1364 | |
1365 | rcu_read_lock(); | |
eb702af5 | 1366 | metadata_stream = stream_find_by_id(relay_streams_ht, |
d3e2ba59 | 1367 | be64toh(metadata_struct->stream_id)); |
b8aa1682 JD |
1368 | if (!metadata_stream) { |
1369 | ret = -1; | |
9d1bbf21 | 1370 | goto end_unlock; |
b8aa1682 JD |
1371 | } |
1372 | ||
6cd525e8 MD |
1373 | size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload, |
1374 | payload_size); | |
1375 | if (size_ret < payload_size) { | |
b8aa1682 JD |
1376 | ERR("Relay error writing metadata on file"); |
1377 | ret = -1; | |
9d1bbf21 | 1378 | goto end_unlock; |
b8aa1682 | 1379 | } |
1d4dfdef DG |
1380 | |
1381 | ret = write_padding_to_file(metadata_stream->fd, | |
1382 | be32toh(metadata_struct->padding_size)); | |
1383 | if (ret < 0) { | |
1384 | goto end_unlock; | |
1385 | } | |
eb702af5 DG |
1386 | |
1387 | ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, | |
1388 | metadata_stream->path_name); | |
1389 | assert(ctf_trace); | |
1390 | ctf_trace->metadata_received += | |
d3e2ba59 | 1391 | payload_size + be32toh(metadata_struct->padding_size); |
1d4dfdef | 1392 | |
b8aa1682 JD |
1393 | DBG2("Relay metadata written"); |
1394 | ||
9d1bbf21 | 1395 | end_unlock: |
6e3c5836 | 1396 | rcu_read_unlock(); |
b8aa1682 JD |
1397 | end: |
1398 | return ret; | |
1399 | } | |
1400 | ||
1401 | /* | |
1402 | * relay_send_version: send relayd version number | |
1403 | */ | |
1404 | static | |
1405 | int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr, | |
e85cdca9 | 1406 | struct relay_connection *conn) |
b8aa1682 | 1407 | { |
7f51dcba | 1408 | int ret; |
092b6259 | 1409 | struct lttcomm_relayd_version reply, msg; |
b8aa1682 | 1410 | |
e85cdca9 | 1411 | assert(conn); |
c5b6f4f0 | 1412 | |
e85cdca9 | 1413 | conn->version_check_done = 1; |
b8aa1682 | 1414 | |
092b6259 | 1415 | /* Get version from the other side. */ |
e85cdca9 | 1416 | ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0); |
092b6259 | 1417 | if (ret < 0 || ret != sizeof(msg)) { |
a6cd2b97 DG |
1418 | if (ret == 0) { |
1419 | /* Orderly shutdown. Not necessary to print an error. */ | |
e85cdca9 | 1420 | DBG("Socket %d did an orderly shutdown", conn->sock->fd); |
a6cd2b97 DG |
1421 | } else { |
1422 | ERR("Relay failed to receive the version values."); | |
1423 | } | |
092b6259 | 1424 | ret = -1; |
092b6259 DG |
1425 | goto end; |
1426 | } | |
1427 | ||
239f3aec | 1428 | memset(&reply, 0, sizeof(reply)); |
d83a952c MD |
1429 | reply.major = RELAYD_VERSION_COMM_MAJOR; |
1430 | reply.minor = RELAYD_VERSION_COMM_MINOR; | |
d4519fa3 JD |
1431 | |
1432 | /* Major versions must be the same */ | |
1433 | if (reply.major != be32toh(msg.major)) { | |
6151a90f JD |
1434 | DBG("Incompatible major versions (%u vs %u), deleting session", |
1435 | reply.major, be32toh(msg.major)); | |
e85cdca9 | 1436 | destroy_session(conn->session, conn->sessions_ht); |
d4519fa3 JD |
1437 | ret = 0; |
1438 | goto end; | |
1439 | } | |
1440 | ||
e85cdca9 | 1441 | conn->major = reply.major; |
0f907de1 JD |
1442 | /* We adapt to the lowest compatible version */ |
1443 | if (reply.minor <= be32toh(msg.minor)) { | |
e85cdca9 | 1444 | conn->minor = reply.minor; |
0f907de1 | 1445 | } else { |
e85cdca9 | 1446 | conn->minor = be32toh(msg.minor); |
0f907de1 JD |
1447 | } |
1448 | ||
6151a90f JD |
1449 | reply.major = htobe32(reply.major); |
1450 | reply.minor = htobe32(reply.minor); | |
e85cdca9 | 1451 | ret = conn->sock->ops->sendmsg(conn->sock, &reply, |
6151a90f JD |
1452 | sizeof(struct lttcomm_relayd_version), 0); |
1453 | if (ret < 0) { | |
1454 | ERR("Relay sending version"); | |
1455 | } | |
1456 | ||
e85cdca9 DG |
1457 | DBG("Version check done using protocol %u.%u", conn->major, |
1458 | conn->minor); | |
b8aa1682 JD |
1459 | |
1460 | end: | |
1461 | return ret; | |
1462 | } | |
1463 | ||
c8f59ee5 | 1464 | /* |
6d805429 | 1465 | * Check for data pending for a given stream id from the session daemon. |
c8f59ee5 DG |
1466 | */ |
1467 | static | |
6d805429 | 1468 | int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, |
e85cdca9 | 1469 | struct relay_connection *conn) |
c8f59ee5 | 1470 | { |
e85cdca9 | 1471 | struct relay_session *session = conn->session; |
6d805429 | 1472 | struct lttcomm_relayd_data_pending msg; |
c8f59ee5 DG |
1473 | struct lttcomm_relayd_generic_reply reply; |
1474 | struct relay_stream *stream; | |
1475 | int ret; | |
c8f59ee5 DG |
1476 | uint64_t last_net_seq_num, stream_id; |
1477 | ||
6d805429 | 1478 | DBG("Data pending command received"); |
c8f59ee5 | 1479 | |
e85cdca9 | 1480 | if (!session || conn->version_check_done == 0) { |
c8f59ee5 DG |
1481 | ERR("Trying to check for data before version check"); |
1482 | ret = -1; | |
1483 | goto end_no_session; | |
1484 | } | |
1485 | ||
e85cdca9 | 1486 | ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0); |
c8f59ee5 | 1487 | if (ret < sizeof(msg)) { |
a6cd2b97 DG |
1488 | if (ret == 0) { |
1489 | /* Orderly shutdown. Not necessary to print an error. */ | |
e85cdca9 | 1490 | DBG("Socket %d did an orderly shutdown", conn->sock->fd); |
a6cd2b97 DG |
1491 | } else { |
1492 | ERR("Relay didn't receive valid data_pending struct size : %d", | |
1493 | ret); | |
1494 | } | |
c8f59ee5 DG |
1495 | ret = -1; |
1496 | goto end_no_session; | |
1497 | } | |
1498 | ||
1499 | stream_id = be64toh(msg.stream_id); | |
1500 | last_net_seq_num = be64toh(msg.last_net_seq_num); | |
1501 | ||
1502 | rcu_read_lock(); | |
eb702af5 | 1503 | stream = stream_find_by_id(relay_streams_ht, stream_id); |
de91f48a | 1504 | if (stream == NULL) { |
c8f59ee5 DG |
1505 | ret = -1; |
1506 | goto end_unlock; | |
1507 | } | |
1508 | ||
6d805429 | 1509 | DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64 |
c8f59ee5 DG |
1510 | " and last_seq %" PRIu64, stream_id, stream->prev_seq, |
1511 | last_net_seq_num); | |
1512 | ||
33832e64 | 1513 | /* Avoid wrapping issue */ |
39df6d9f | 1514 | if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) { |
6d805429 | 1515 | /* Data has in fact been written and is NOT pending */ |
c8f59ee5 | 1516 | ret = 0; |
6d805429 DG |
1517 | } else { |
1518 | /* Data still being streamed thus pending */ | |
1519 | ret = 1; | |
c8f59ee5 DG |
1520 | } |
1521 | ||
f7079f67 DG |
1522 | /* Pending check is now done. */ |
1523 | stream->data_pending_check_done = 1; | |
1524 | ||
c8f59ee5 DG |
1525 | end_unlock: |
1526 | rcu_read_unlock(); | |
1527 | ||
239f3aec | 1528 | memset(&reply, 0, sizeof(reply)); |
c8f59ee5 | 1529 | reply.ret_code = htobe32(ret); |
e85cdca9 | 1530 | ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); |
c8f59ee5 | 1531 | if (ret < 0) { |
6d805429 | 1532 | ERR("Relay data pending ret code failed"); |
c8f59ee5 DG |
1533 | } |
1534 | ||
1535 | end_no_session: | |
1536 | return ret; | |
1537 | } | |
1538 | ||
1539 | /* | |
1540 | * Wait for the control socket to reach a quiescent state. | |
1541 | * | |
1542 | * Note that for now, when receiving this command from the session daemon, this | |
1543 | * means that every subsequent commands or data received on the control socket | |
1544 | * has been handled. So, this is why we simply return OK here. | |
1545 | */ | |
1546 | static | |
1547 | int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr, | |
e85cdca9 | 1548 | struct relay_connection *conn) |
c8f59ee5 DG |
1549 | { |
1550 | int ret; | |
ad7051c0 DG |
1551 | uint64_t stream_id; |
1552 | struct relay_stream *stream; | |
1553 | struct lttng_ht_iter iter; | |
1554 | struct lttcomm_relayd_quiescent_control msg; | |
c8f59ee5 DG |
1555 | struct lttcomm_relayd_generic_reply reply; |
1556 | ||
1557 | DBG("Checking quiescent state on control socket"); | |
1558 | ||
e85cdca9 | 1559 | if (!conn->session || conn->version_check_done == 0) { |
ad7051c0 DG |
1560 | ERR("Trying to check for data before version check"); |
1561 | ret = -1; | |
1562 | goto end_no_session; | |
1563 | } | |
1564 | ||
e85cdca9 | 1565 | ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0); |
ad7051c0 | 1566 | if (ret < sizeof(msg)) { |
a6cd2b97 DG |
1567 | if (ret == 0) { |
1568 | /* Orderly shutdown. Not necessary to print an error. */ | |
e85cdca9 | 1569 | DBG("Socket %d did an orderly shutdown", conn->sock->fd); |
a6cd2b97 DG |
1570 | } else { |
1571 | ERR("Relay didn't receive valid begin data_pending struct size: %d", | |
1572 | ret); | |
1573 | } | |
ad7051c0 DG |
1574 | ret = -1; |
1575 | goto end_no_session; | |
1576 | } | |
1577 | ||
1578 | stream_id = be64toh(msg.stream_id); | |
1579 | ||
1580 | rcu_read_lock(); | |
d3e2ba59 | 1581 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
eb702af5 | 1582 | node.node) { |
ad7051c0 DG |
1583 | if (stream->stream_handle == stream_id) { |
1584 | stream->data_pending_check_done = 1; | |
1585 | DBG("Relay quiescent control pending flag set to %" PRIu64, | |
1586 | stream_id); | |
1587 | break; | |
1588 | } | |
1589 | } | |
1590 | rcu_read_unlock(); | |
1591 | ||
239f3aec | 1592 | memset(&reply, 0, sizeof(reply)); |
c8f59ee5 | 1593 | reply.ret_code = htobe32(LTTNG_OK); |
e85cdca9 | 1594 | ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); |
c8f59ee5 | 1595 | if (ret < 0) { |
6d805429 | 1596 | ERR("Relay data quiescent control ret code failed"); |
c8f59ee5 DG |
1597 | } |
1598 | ||
ad7051c0 | 1599 | end_no_session: |
c8f59ee5 DG |
1600 | return ret; |
1601 | } | |
1602 | ||
f7079f67 DG |
1603 | /* |
1604 | * Initialize a data pending command. This means that a client is about to ask | |
1605 | * for data pending for each stream he/she holds. Simply iterate over all | |
1606 | * streams of a session and set the data_pending_check_done flag. | |
1607 | * | |
1608 | * This command returns to the client a LTTNG_OK code. | |
1609 | */ | |
1610 | static | |
1611 | int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, | |
e85cdca9 | 1612 | struct relay_connection *conn) |
f7079f67 DG |
1613 | { |
1614 | int ret; | |
1615 | struct lttng_ht_iter iter; | |
1616 | struct lttcomm_relayd_begin_data_pending msg; | |
1617 | struct lttcomm_relayd_generic_reply reply; | |
1618 | struct relay_stream *stream; | |
1619 | uint64_t session_id; | |
1620 | ||
1621 | assert(recv_hdr); | |
e85cdca9 | 1622 | assert(conn); |
f7079f67 DG |
1623 | |
1624 | DBG("Init streams for data pending"); | |
1625 | ||
e85cdca9 | 1626 | if (!conn->session || conn->version_check_done == 0) { |
f7079f67 DG |
1627 | ERR("Trying to check for data before version check"); |
1628 | ret = -1; | |
1629 | goto end_no_session; | |
1630 | } | |
1631 | ||
e85cdca9 | 1632 | ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0); |
f7079f67 | 1633 | if (ret < sizeof(msg)) { |
a6cd2b97 DG |
1634 | if (ret == 0) { |
1635 | /* Orderly shutdown. Not necessary to print an error. */ | |
e85cdca9 | 1636 | DBG("Socket %d did an orderly shutdown", conn->sock->fd); |
a6cd2b97 DG |
1637 | } else { |
1638 | ERR("Relay didn't receive valid begin data_pending struct size: %d", | |
1639 | ret); | |
1640 | } | |
f7079f67 DG |
1641 | ret = -1; |
1642 | goto end_no_session; | |
1643 | } | |
1644 | ||
1645 | session_id = be64toh(msg.session_id); | |
1646 | ||
1647 | /* | |
1648 | * Iterate over all streams to set the begin data pending flag. For now, the | |
1649 | * streams are indexed by stream handle so we have to iterate over all | |
1650 | * streams to find the one associated with the right session_id. | |
1651 | */ | |
1652 | rcu_read_lock(); | |
d3e2ba59 | 1653 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
eb702af5 DG |
1654 | node.node) { |
1655 | if (stream->session_id == session_id) { | |
f7079f67 DG |
1656 | stream->data_pending_check_done = 0; |
1657 | DBG("Set begin data pending flag to stream %" PRIu64, | |
1658 | stream->stream_handle); | |
1659 | } | |
1660 | } | |
1661 | rcu_read_unlock(); | |
1662 | ||
239f3aec | 1663 | memset(&reply, 0, sizeof(reply)); |
f7079f67 DG |
1664 | /* All good, send back reply. */ |
1665 | reply.ret_code = htobe32(LTTNG_OK); | |
1666 | ||
e85cdca9 | 1667 | ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); |
f7079f67 DG |
1668 | if (ret < 0) { |
1669 | ERR("Relay begin data pending send reply failed"); | |
1670 | } | |
1671 | ||
1672 | end_no_session: | |
1673 | return ret; | |
1674 | } | |
1675 | ||
1676 | /* | |
1677 | * End data pending command. This will check, for a given session id, if each | |
1678 | * stream associated with it has its data_pending_check_done flag set. If not, | |
1679 | * this means that the client lost track of the stream but the data is still | |
1680 | * being streamed on our side. In this case, we inform the client that data is | |
1681 | * inflight. | |
1682 | * | |
1683 | * Return to the client if there is data in flight or not with a ret_code. | |
1684 | */ | |
1685 | static | |
1686 | int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, | |
e85cdca9 | 1687 | struct relay_connection *conn) |
f7079f67 DG |
1688 | { |
1689 | int ret; | |
1690 | struct lttng_ht_iter iter; | |
1691 | struct lttcomm_relayd_end_data_pending msg; | |
1692 | struct lttcomm_relayd_generic_reply reply; | |
1693 | struct relay_stream *stream; | |
1694 | uint64_t session_id; | |
1695 | uint32_t is_data_inflight = 0; | |
1696 | ||
1697 | assert(recv_hdr); | |
e85cdca9 | 1698 | assert(conn); |
f7079f67 DG |
1699 | |
1700 | DBG("End data pending command"); | |
1701 | ||
e85cdca9 | 1702 | if (!conn->session || conn->version_check_done == 0) { |
f7079f67 DG |
1703 | ERR("Trying to check for data before version check"); |
1704 | ret = -1; | |
1705 | goto end_no_session; | |
1706 | } | |
1707 | ||
e85cdca9 | 1708 | ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0); |
f7079f67 | 1709 | if (ret < sizeof(msg)) { |
a6cd2b97 DG |
1710 | if (ret == 0) { |
1711 | /* Orderly shutdown. Not necessary to print an error. */ | |
e85cdca9 | 1712 | DBG("Socket %d did an orderly shutdown", conn->sock->fd); |
a6cd2b97 DG |
1713 | } else { |
1714 | ERR("Relay didn't receive valid end data_pending struct size: %d", | |
1715 | ret); | |
1716 | } | |
f7079f67 DG |
1717 | ret = -1; |
1718 | goto end_no_session; | |
1719 | } | |
1720 | ||
1721 | session_id = be64toh(msg.session_id); | |
1722 | ||
1723 | /* Iterate over all streams to see if the begin data pending flag is set. */ | |
1724 | rcu_read_lock(); | |
d3e2ba59 | 1725 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
eb702af5 DG |
1726 | node.node) { |
1727 | if (stream->session_id == session_id && | |
50979bcd | 1728 | !stream->data_pending_check_done && !stream->terminated_flag) { |
f7079f67 DG |
1729 | is_data_inflight = 1; |
1730 | DBG("Data is still in flight for stream %" PRIu64, | |
1731 | stream->stream_handle); | |
1732 | break; | |
1733 | } | |
1734 | } | |
1735 | rcu_read_unlock(); | |
1736 | ||
239f3aec | 1737 | memset(&reply, 0, sizeof(reply)); |
f7079f67 DG |
1738 | /* All good, send back reply. */ |
1739 | reply.ret_code = htobe32(is_data_inflight); | |
1740 | ||
e85cdca9 | 1741 | ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); |
f7079f67 DG |
1742 | if (ret < 0) { |
1743 | ERR("Relay end data pending send reply failed"); | |
1744 | } | |
1745 | ||
1746 | end_no_session: | |
1747 | return ret; | |
1748 | } | |
1749 | ||
1c20f0e2 JD |
1750 | /* |
1751 | * Receive an index for a specific stream. | |
1752 | * | |
1753 | * Return 0 on success else a negative value. | |
1754 | */ | |
1755 | static | |
1756 | int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr, | |
e85cdca9 | 1757 | struct relay_connection *conn) |
1c20f0e2 JD |
1758 | { |
1759 | int ret, send_ret, index_created = 0; | |
e85cdca9 | 1760 | struct relay_session *session = conn->session; |
1c20f0e2 JD |
1761 | struct lttcomm_relayd_index index_info; |
1762 | struct relay_index *index, *wr_index = NULL; | |
1763 | struct lttcomm_relayd_generic_reply reply; | |
1764 | struct relay_stream *stream; | |
1765 | uint64_t net_seq_num; | |
1766 | ||
e85cdca9 | 1767 | assert(conn); |
1c20f0e2 JD |
1768 | |
1769 | DBG("Relay receiving index"); | |
1770 | ||
e85cdca9 | 1771 | if (!session || conn->version_check_done == 0) { |
1c20f0e2 JD |
1772 | ERR("Trying to close a stream before version check"); |
1773 | ret = -1; | |
1774 | goto end_no_session; | |
1775 | } | |
1776 | ||
e85cdca9 | 1777 | ret = conn->sock->ops->recvmsg(conn->sock, &index_info, |
1c20f0e2 JD |
1778 | sizeof(index_info), 0); |
1779 | if (ret < sizeof(index_info)) { | |
1780 | if (ret == 0) { | |
1781 | /* Orderly shutdown. Not necessary to print an error. */ | |
e85cdca9 | 1782 | DBG("Socket %d did an orderly shutdown", conn->sock->fd); |
1c20f0e2 JD |
1783 | } else { |
1784 | ERR("Relay didn't receive valid index struct size : %d", ret); | |
1785 | } | |
1786 | ret = -1; | |
1787 | goto end_no_session; | |
1788 | } | |
1789 | ||
1790 | net_seq_num = be64toh(index_info.net_seq_num); | |
1791 | ||
1792 | rcu_read_lock(); | |
eb702af5 DG |
1793 | stream = stream_find_by_id(relay_streams_ht, |
1794 | be64toh(index_info.relay_stream_id)); | |
1c20f0e2 JD |
1795 | if (!stream) { |
1796 | ret = -1; | |
1797 | goto end_rcu_unlock; | |
1798 | } | |
1799 | ||
d3e2ba59 JD |
1800 | /* Live beacon handling */ |
1801 | if (index_info.packet_size == 0) { | |
1802 | DBG("Received live beacon for stream %" PRIu64, stream->stream_handle); | |
1803 | ||
1804 | /* | |
e36fbffb JD |
1805 | * Only flag a stream inactive when it has already received data |
1806 | * and no indexes are in flight. | |
d3e2ba59 | 1807 | */ |
e36fbffb | 1808 | if (stream->total_index_received > 0 && stream->indexes_in_flight == 0) { |
d3e2ba59 JD |
1809 | stream->beacon_ts_end = be64toh(index_info.timestamp_end); |
1810 | } | |
1811 | ret = 0; | |
1812 | goto end_rcu_unlock; | |
1813 | } else { | |
1814 | stream->beacon_ts_end = -1ULL; | |
1815 | } | |
1816 | ||
0a6518b0 | 1817 | index = relay_index_find(stream->stream_handle, net_seq_num); |
1c20f0e2 JD |
1818 | if (!index) { |
1819 | /* A successful creation will add the object to the HT. */ | |
1820 | index = relay_index_create(stream->stream_handle, net_seq_num); | |
1821 | if (!index) { | |
1822 | goto end_rcu_unlock; | |
1823 | } | |
1824 | index_created = 1; | |
e36fbffb | 1825 | stream->indexes_in_flight++; |
1c20f0e2 JD |
1826 | } |
1827 | ||
1828 | copy_index_control_data(index, &index_info); | |
5e372a51 JD |
1829 | if (stream->ctf_stream_id == -1ULL) { |
1830 | stream->ctf_stream_id = be64toh(index_info.stream_id); | |
1831 | } | |
1c20f0e2 JD |
1832 | |
1833 | if (index_created) { | |
1834 | /* | |
1835 | * Try to add the relay index object to the hash table. If an object | |
1836 | * already exist, destroy back the index created, set the data in this | |
1837 | * object and write it on disk. | |
1838 | */ | |
0a6518b0 | 1839 | relay_index_add(index, &wr_index); |
1c20f0e2 JD |
1840 | if (wr_index) { |
1841 | copy_index_control_data(wr_index, &index_info); | |
1842 | free(index); | |
1843 | } | |
1844 | } else { | |
1845 | /* The index already exists so write it on disk. */ | |
1846 | wr_index = index; | |
1847 | } | |
1848 | ||
1849 | /* Do we have a writable ready index to write on disk. */ | |
1850 | if (wr_index) { | |
0a6518b0 | 1851 | ret = relay_index_write(wr_index->fd, wr_index); |
1c20f0e2 JD |
1852 | if (ret < 0) { |
1853 | goto end_rcu_unlock; | |
1854 | } | |
d3e2ba59 | 1855 | stream->total_index_received++; |
e36fbffb JD |
1856 | stream->indexes_in_flight--; |
1857 | assert(stream->indexes_in_flight >= 0); | |
1c20f0e2 JD |
1858 | } |
1859 | ||
1860 | end_rcu_unlock: | |
1861 | rcu_read_unlock(); | |
1862 | ||
239f3aec | 1863 | memset(&reply, 0, sizeof(reply)); |
1c20f0e2 JD |
1864 | if (ret < 0) { |
1865 | reply.ret_code = htobe32(LTTNG_ERR_UNK); | |
1866 | } else { | |
1867 | reply.ret_code = htobe32(LTTNG_OK); | |
1868 | } | |
e85cdca9 | 1869 | send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); |
1c20f0e2 JD |
1870 | if (send_ret < 0) { |
1871 | ERR("Relay sending close index id reply"); | |
1872 | ret = send_ret; | |
1873 | } | |
1874 | ||
1875 | end_no_session: | |
1876 | return ret; | |
1877 | } | |
1878 | ||
814fcae4 JD |
1879 | /* |
1880 | * Receive the streams_sent message. | |
1881 | * | |
1882 | * Return 0 on success else a negative value. | |
1883 | */ | |
1884 | static | |
1885 | int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr, | |
e85cdca9 | 1886 | struct relay_connection *conn) |
814fcae4 JD |
1887 | { |
1888 | int ret, send_ret; | |
1889 | struct lttcomm_relayd_generic_reply reply; | |
1890 | ||
e85cdca9 | 1891 | assert(conn); |
814fcae4 JD |
1892 | |
1893 | DBG("Relay receiving streams_sent"); | |
1894 | ||
e85cdca9 | 1895 | if (!conn->session || conn->version_check_done == 0) { |
814fcae4 JD |
1896 | ERR("Trying to close a stream before version check"); |
1897 | ret = -1; | |
1898 | goto end_no_session; | |
1899 | } | |
1900 | ||
1901 | /* | |
1902 | * Flag every pending stream in the connection recv list that they are | |
1903 | * ready to be used by the viewer. | |
1904 | */ | |
e85cdca9 | 1905 | set_viewer_ready_flag(conn); |
814fcae4 | 1906 | |
829007d9 JD |
1907 | /* |
1908 | * Inform the viewer that there are new streams in the session. | |
1909 | */ | |
750c7a85 JD |
1910 | if (conn->session->viewer_refcount) { |
1911 | uatomic_set(&conn->session->new_streams, 1); | |
1912 | } | |
829007d9 | 1913 | |
239f3aec | 1914 | memset(&reply, 0, sizeof(reply)); |
814fcae4 | 1915 | reply.ret_code = htobe32(LTTNG_OK); |
e85cdca9 | 1916 | send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0); |
814fcae4 JD |
1917 | if (send_ret < 0) { |
1918 | ERR("Relay sending sent_stream reply"); | |
1919 | ret = send_ret; | |
1920 | } else { | |
1921 | /* Success. */ | |
1922 | ret = 0; | |
1923 | } | |
1924 | ||
1925 | end_no_session: | |
1926 | return ret; | |
1927 | } | |
1928 | ||
b8aa1682 | 1929 | /* |
d3e2ba59 | 1930 | * Process the commands received on the control socket |
b8aa1682 JD |
1931 | */ |
1932 | static | |
1933 | int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, | |
e85cdca9 | 1934 | struct relay_connection *conn) |
b8aa1682 JD |
1935 | { |
1936 | int ret = 0; | |
1937 | ||
1938 | switch (be32toh(recv_hdr->cmd)) { | |
b8aa1682 | 1939 | case RELAYD_CREATE_SESSION: |
e85cdca9 | 1940 | ret = relay_create_session(recv_hdr, conn); |
b8aa1682 | 1941 | break; |
b8aa1682 | 1942 | case RELAYD_ADD_STREAM: |
e85cdca9 | 1943 | ret = relay_add_stream(recv_hdr, conn); |
b8aa1682 JD |
1944 | break; |
1945 | case RELAYD_START_DATA: | |
e85cdca9 | 1946 | ret = relay_start(recv_hdr, conn); |
b8aa1682 JD |
1947 | break; |
1948 | case RELAYD_SEND_METADATA: | |
e85cdca9 | 1949 | ret = relay_recv_metadata(recv_hdr, conn); |
b8aa1682 JD |
1950 | break; |
1951 | case RELAYD_VERSION: | |
e85cdca9 | 1952 | ret = relay_send_version(recv_hdr, conn); |
b8aa1682 | 1953 | break; |
173af62f | 1954 | case RELAYD_CLOSE_STREAM: |
e85cdca9 | 1955 | ret = relay_close_stream(recv_hdr, conn); |
173af62f | 1956 | break; |
6d805429 | 1957 | case RELAYD_DATA_PENDING: |
e85cdca9 | 1958 | ret = relay_data_pending(recv_hdr, conn); |
c8f59ee5 DG |
1959 | break; |
1960 | case RELAYD_QUIESCENT_CONTROL: | |
e85cdca9 | 1961 | ret = relay_quiescent_control(recv_hdr, conn); |
c8f59ee5 | 1962 | break; |
f7079f67 | 1963 | case RELAYD_BEGIN_DATA_PENDING: |
e85cdca9 | 1964 | ret = relay_begin_data_pending(recv_hdr, conn); |
f7079f67 DG |
1965 | break; |
1966 | case RELAYD_END_DATA_PENDING: | |
e85cdca9 | 1967 | ret = relay_end_data_pending(recv_hdr, conn); |
f7079f67 | 1968 | break; |
1c20f0e2 | 1969 | case RELAYD_SEND_INDEX: |
e85cdca9 | 1970 | ret = relay_recv_index(recv_hdr, conn); |
1c20f0e2 | 1971 | break; |
814fcae4 | 1972 | case RELAYD_STREAMS_SENT: |
e85cdca9 | 1973 | ret = relay_streams_sent(recv_hdr, conn); |
814fcae4 | 1974 | break; |
b8aa1682 JD |
1975 | case RELAYD_UPDATE_SYNC_INFO: |
1976 | default: | |
1977 | ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd)); | |
e85cdca9 | 1978 | relay_unknown_command(conn); |
b8aa1682 JD |
1979 | ret = -1; |
1980 | goto end; | |
1981 | } | |
1982 | ||
1983 | end: | |
1984 | return ret; | |
1985 | } | |
1986 | ||
7d2f7452 DG |
1987 | /* |
1988 | * Handle index for a data stream. | |
1989 | * | |
1990 | * RCU read side lock MUST be acquired. | |
1991 | * | |
1992 | * Return 0 on success else a negative value. | |
1993 | */ | |
1994 | static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num, | |
1995 | int rotate_index) | |
1996 | { | |
1997 | int ret = 0, index_created = 0; | |
1998 | uint64_t stream_id, data_offset; | |
1999 | struct relay_index *index, *wr_index = NULL; | |
2000 | ||
2001 | assert(stream); | |
2002 | ||
2003 | stream_id = stream->stream_handle; | |
2004 | /* Get data offset because we are about to update the index. */ | |
2005 | data_offset = htobe64(stream->tracefile_size_current); | |
2006 | ||
2007 | /* | |
2008 | * Lookup for an existing index for that stream id/sequence number. If on | |
2009 | * exists, the control thread already received the data for it thus we need | |
2010 | * to write it on disk. | |
2011 | */ | |
2012 | index = relay_index_find(stream_id, net_seq_num); | |
2013 | if (!index) { | |
2014 | /* A successful creation will add the object to the HT. */ | |
2015 | index = relay_index_create(stream_id, net_seq_num); | |
2016 | if (!index) { | |
2017 | ret = -1; | |
2018 | goto error; | |
2019 | } | |
2020 | index_created = 1; | |
e36fbffb | 2021 | stream->indexes_in_flight++; |
7d2f7452 DG |
2022 | } |
2023 | ||
2024 | if (rotate_index || stream->index_fd < 0) { | |
2025 | index->to_close_fd = stream->index_fd; | |
2026 | ret = index_create_file(stream->path_name, stream->channel_name, | |
2027 | relayd_uid, relayd_gid, stream->tracefile_size, | |
2028 | stream->tracefile_count_current); | |
2029 | if (ret < 0) { | |
2030 | /* This will close the stream's index fd if one. */ | |
2031 | relay_index_free_safe(index); | |
2032 | goto error; | |
2033 | } | |
2034 | stream->index_fd = ret; | |
2035 | } | |
2036 | index->fd = stream->index_fd; | |
2037 | index->index_data.offset = data_offset; | |
2038 | ||
2039 | if (index_created) { | |
2040 | /* | |
2041 | * Try to add the relay index object to the hash table. If an object | |
2042 | * already exist, destroy back the index created and set the data. | |
2043 | */ | |
2044 | relay_index_add(index, &wr_index); | |
2045 | if (wr_index) { | |
2046 | /* Copy back data from the created index. */ | |
2047 | wr_index->fd = index->fd; | |
2048 | wr_index->to_close_fd = index->to_close_fd; | |
2049 | wr_index->index_data.offset = data_offset; | |
2050 | free(index); | |
2051 | } | |
2052 | } else { | |
2053 | /* The index already exists so write it on disk. */ | |
2054 | wr_index = index; | |
2055 | } | |
2056 | ||
2057 | /* Do we have a writable ready index to write on disk. */ | |
2058 | if (wr_index) { | |
2059 | ret = relay_index_write(wr_index->fd, wr_index); | |
2060 | if (ret < 0) { | |
2061 | goto error; | |
2062 | } | |
2063 | stream->total_index_received++; | |
e36fbffb JD |
2064 | stream->indexes_in_flight--; |
2065 | assert(stream->indexes_in_flight >= 0); | |
7d2f7452 DG |
2066 | } |
2067 | ||
2068 | error: | |
2069 | return ret; | |
2070 | } | |
2071 | ||
b8aa1682 JD |
2072 | /* |
2073 | * relay_process_data: Process the data received on the data socket | |
2074 | */ | |
2075 | static | |
e85cdca9 | 2076 | int relay_process_data(struct relay_connection *conn) |
b8aa1682 | 2077 | { |
7d2f7452 | 2078 | int ret = 0, rotate_index = 0; |
6cd525e8 | 2079 | ssize_t size_ret; |
b8aa1682 JD |
2080 | struct relay_stream *stream; |
2081 | struct lttcomm_relayd_data_hdr data_hdr; | |
7d2f7452 | 2082 | uint64_t stream_id; |
173af62f | 2083 | uint64_t net_seq_num; |
b8aa1682 | 2084 | uint32_t data_size; |
eb702af5 | 2085 | struct relay_session *session; |
b8aa1682 | 2086 | |
e85cdca9 DG |
2087 | assert(conn); |
2088 | ||
2089 | ret = conn->sock->ops->recvmsg(conn->sock, &data_hdr, | |
7c5aef62 | 2090 | sizeof(struct lttcomm_relayd_data_hdr), 0); |
b8aa1682 | 2091 | if (ret <= 0) { |
a6cd2b97 DG |
2092 | if (ret == 0) { |
2093 | /* Orderly shutdown. Not necessary to print an error. */ | |
e85cdca9 | 2094 | DBG("Socket %d did an orderly shutdown", conn->sock->fd); |
a6cd2b97 | 2095 | } else { |
e85cdca9 | 2096 | ERR("Unable to receive data header on sock %d", conn->sock->fd); |
a6cd2b97 | 2097 | } |
b8aa1682 JD |
2098 | ret = -1; |
2099 | goto end; | |
2100 | } | |
2101 | ||
2102 | stream_id = be64toh(data_hdr.stream_id); | |
9d1bbf21 MD |
2103 | |
2104 | rcu_read_lock(); | |
eb702af5 | 2105 | stream = stream_find_by_id(relay_streams_ht, stream_id); |
b8aa1682 JD |
2106 | if (!stream) { |
2107 | ret = -1; | |
1c20f0e2 | 2108 | goto end_rcu_unlock; |
b8aa1682 JD |
2109 | } |
2110 | ||
e85cdca9 | 2111 | session = session_find_by_id(conn->sessions_ht, stream->session_id); |
eb702af5 DG |
2112 | assert(session); |
2113 | ||
b8aa1682 JD |
2114 | data_size = be32toh(data_hdr.data_size); |
2115 | if (data_buffer_size < data_size) { | |
c617c0c6 MD |
2116 | char *tmp_data_ptr; |
2117 | ||
2118 | tmp_data_ptr = realloc(data_buffer, data_size); | |
2119 | if (!tmp_data_ptr) { | |
b8aa1682 | 2120 | ERR("Allocating data buffer"); |
c617c0c6 | 2121 | free(data_buffer); |
b8aa1682 | 2122 | ret = -1; |
1c20f0e2 | 2123 | goto end_rcu_unlock; |
b8aa1682 | 2124 | } |
c617c0c6 | 2125 | data_buffer = tmp_data_ptr; |
b8aa1682 JD |
2126 | data_buffer_size = data_size; |
2127 | } | |
2128 | memset(data_buffer, 0, data_size); | |
2129 | ||
173af62f DG |
2130 | net_seq_num = be64toh(data_hdr.net_seq_num); |
2131 | ||
77c7c900 | 2132 | DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64, |
173af62f | 2133 | data_size, stream_id, net_seq_num); |
e85cdca9 | 2134 | ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0); |
b8aa1682 | 2135 | if (ret <= 0) { |
a6cd2b97 DG |
2136 | if (ret == 0) { |
2137 | /* Orderly shutdown. Not necessary to print an error. */ | |
e85cdca9 | 2138 | DBG("Socket %d did an orderly shutdown", conn->sock->fd); |
a6cd2b97 | 2139 | } |
b8aa1682 | 2140 | ret = -1; |
1c20f0e2 | 2141 | goto end_rcu_unlock; |
b8aa1682 JD |
2142 | } |
2143 | ||
1c20f0e2 | 2144 | /* Check if a rotation is needed. */ |
0f907de1 JD |
2145 | if (stream->tracefile_size > 0 && |
2146 | (stream->tracefile_size_current + data_size) > | |
2147 | stream->tracefile_size) { | |
6b6b9a5a JD |
2148 | struct relay_viewer_stream *vstream; |
2149 | uint64_t new_id; | |
2150 | ||
2151 | new_id = (stream->tracefile_count_current + 1) % | |
2152 | stream->tracefile_count; | |
2153 | /* | |
2154 | * When we wrap-around back to 0, we start overwriting old | |
2155 | * trace data. | |
2156 | */ | |
2157 | if (!stream->tracefile_overwrite && new_id == 0) { | |
2158 | stream->tracefile_overwrite = 1; | |
2159 | } | |
2160 | pthread_mutex_lock(&stream->viewer_stream_rotation_lock); | |
2161 | if (stream->tracefile_overwrite) { | |
2162 | stream->oldest_tracefile_id = | |
2163 | (stream->oldest_tracefile_id + 1) % | |
2164 | stream->tracefile_count; | |
2165 | } | |
991adae2 | 2166 | vstream = viewer_stream_find_by_id(stream->stream_handle); |
6b6b9a5a JD |
2167 | if (vstream) { |
2168 | /* | |
2169 | * The viewer is reading a file about to be | |
2170 | * overwritten. Close the FDs it is | |
2171 | * currently using and let it handle the fault. | |
2172 | */ | |
2173 | if (vstream->tracefile_count_current == new_id) { | |
cef0f7d5 | 2174 | pthread_mutex_lock(&vstream->overwrite_lock); |
6b6b9a5a | 2175 | vstream->abort_flag = 1; |
cef0f7d5 | 2176 | pthread_mutex_unlock(&vstream->overwrite_lock); |
6b6b9a5a JD |
2177 | DBG("Streaming side setting abort_flag on stream %s_%lu\n", |
2178 | stream->channel_name, new_id); | |
2179 | } else if (vstream->tracefile_count_current == | |
2180 | stream->tracefile_count_current) { | |
2181 | /* | |
2182 | * The reader and writer were in the | |
2183 | * same trace file, inform the viewer | |
2184 | * that no new index will ever be added | |
2185 | * to this file. | |
2186 | */ | |
2187 | vstream->close_write_flag = 1; | |
2188 | } | |
2189 | } | |
1c20f0e2 JD |
2190 | ret = utils_rotate_stream_file(stream->path_name, stream->channel_name, |
2191 | stream->tracefile_size, stream->tracefile_count, | |
2192 | relayd_uid, relayd_gid, stream->fd, | |
2193 | &(stream->tracefile_count_current), &stream->fd); | |
cef0f7d5 | 2194 | stream->total_index_received = 0; |
6b6b9a5a | 2195 | pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); |
0f907de1 | 2196 | if (ret < 0) { |
1c20f0e2 JD |
2197 | ERR("Rotating stream output file"); |
2198 | goto end_rcu_unlock; | |
0f907de1 | 2199 | } |
a6976990 DG |
2200 | /* Reset current size because we just perform a stream rotation. */ |
2201 | stream->tracefile_size_current = 0; | |
1c20f0e2 JD |
2202 | rotate_index = 1; |
2203 | } | |
2204 | ||
1c20f0e2 | 2205 | /* |
7d2f7452 DG |
2206 | * Index are handled in protocol version 2.4 and above. Also, snapshot and |
2207 | * index are NOT supported. | |
1c20f0e2 | 2208 | */ |
eb702af5 | 2209 | if (session->minor >= 4 && !session->snapshot) { |
7d2f7452 | 2210 | ret = handle_index_data(stream, net_seq_num, rotate_index); |
1c20f0e2 | 2211 | if (ret < 0) { |
1c20f0e2 JD |
2212 | goto end_rcu_unlock; |
2213 | } | |
1c20f0e2 JD |
2214 | } |
2215 | ||
7d2f7452 | 2216 | /* Write data to stream output fd. */ |
6cd525e8 MD |
2217 | size_ret = lttng_write(stream->fd, data_buffer, data_size); |
2218 | if (size_ret < data_size) { | |
b8aa1682 JD |
2219 | ERR("Relay error writing data to file"); |
2220 | ret = -1; | |
1c20f0e2 | 2221 | goto end_rcu_unlock; |
b8aa1682 | 2222 | } |
1d4dfdef | 2223 | |
5ab7344e JD |
2224 | DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64, |
2225 | ret, stream->stream_handle); | |
2226 | ||
1d4dfdef DG |
2227 | ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size)); |
2228 | if (ret < 0) { | |
1c20f0e2 | 2229 | goto end_rcu_unlock; |
1d4dfdef | 2230 | } |
1c20f0e2 | 2231 | stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size); |
1d4dfdef | 2232 | |
173af62f DG |
2233 | stream->prev_seq = net_seq_num; |
2234 | ||
eb702af5 | 2235 | try_close_stream(session, stream); |
173af62f | 2236 | |
1c20f0e2 | 2237 | end_rcu_unlock: |
9d1bbf21 | 2238 | rcu_read_unlock(); |
b8aa1682 JD |
2239 | end: |
2240 | return ret; | |
2241 | } | |
2242 | ||
2243 | static | |
e85cdca9 | 2244 | void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd) |
b8aa1682 JD |
2245 | { |
2246 | int ret; | |
2247 | ||
e85cdca9 DG |
2248 | assert(events); |
2249 | ||
2250 | (void) lttng_poll_del(events, pollfd); | |
b8aa1682 JD |
2251 | |
2252 | ret = close(pollfd); | |
2253 | if (ret < 0) { | |
2254 | ERR("Closing pollfd %d", pollfd); | |
2255 | } | |
2256 | } | |
2257 | ||
e85cdca9 DG |
2258 | static void destroy_connection(struct lttng_ht *relay_connections_ht, |
2259 | struct relay_connection *conn) | |
9d1bbf21 | 2260 | { |
e85cdca9 DG |
2261 | assert(relay_connections_ht); |
2262 | assert(conn); | |
eb702af5 | 2263 | |
e85cdca9 | 2264 | connection_delete(relay_connections_ht, conn); |
eb702af5 | 2265 | |
e85cdca9 | 2266 | /* For the control socket, we try to destroy the session. */ |
23c1fab9 | 2267 | if (conn->type == RELAY_CONTROL && conn->session) { |
e85cdca9 | 2268 | destroy_session(conn->session, conn->sessions_ht); |
9d1bbf21 | 2269 | } |
5b6d8097 | 2270 | |
e85cdca9 | 2271 | connection_destroy(conn); |
b8aa1682 JD |
2272 | } |
2273 | ||
2274 | /* | |
2275 | * This thread does the actual work | |
2276 | */ | |
2277 | static | |
2278 | void *relay_thread_worker(void *data) | |
2279 | { | |
beaad64c DG |
2280 | int ret, err = -1, last_seen_data_fd = -1; |
2281 | uint32_t nb_fd; | |
e85cdca9 | 2282 | struct relay_connection *conn; |
b8aa1682 JD |
2283 | struct lttng_poll_event events; |
2284 | struct lttng_ht *relay_connections_ht; | |
b8aa1682 | 2285 | struct lttng_ht_iter iter; |
b8aa1682 | 2286 | struct lttcomm_relayd_hdr recv_hdr; |
d3e2ba59 JD |
2287 | struct relay_local_data *relay_ctx = (struct relay_local_data *) data; |
2288 | struct lttng_ht *sessions_ht = relay_ctx->sessions_ht; | |
b8aa1682 JD |
2289 | |
2290 | DBG("[thread] Relay worker started"); | |
2291 | ||
9d1bbf21 MD |
2292 | rcu_register_thread(); |
2293 | ||
55706a7d MD |
2294 | health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER); |
2295 | ||
4e5b8b82 MD |
2296 | if (testpoint(relayd_thread_worker)) { |
2297 | goto error_testpoint; | |
2298 | } | |
2299 | ||
f385ae0a MD |
2300 | health_code_update(); |
2301 | ||
b8aa1682 JD |
2302 | /* table of connections indexed on socket */ |
2303 | relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
095a4ae5 MD |
2304 | if (!relay_connections_ht) { |
2305 | goto relay_connections_ht_error; | |
2306 | } | |
b8aa1682 | 2307 | |
1c20f0e2 JD |
2308 | /* Tables of received indexes indexed by index handle and net_seq_num. */ |
2309 | indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64); | |
2310 | if (!indexes_ht) { | |
2311 | goto indexes_ht_error; | |
2312 | } | |
2313 | ||
b8aa1682 JD |
2314 | ret = create_thread_poll_set(&events, 2); |
2315 | if (ret < 0) { | |
2316 | goto error_poll_create; | |
2317 | } | |
2318 | ||
e85cdca9 | 2319 | ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP); |
b8aa1682 JD |
2320 | if (ret < 0) { |
2321 | goto error; | |
2322 | } | |
2323 | ||
beaad64c | 2324 | restart: |
b8aa1682 | 2325 | while (1) { |
beaad64c DG |
2326 | int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1; |
2327 | ||
f385ae0a MD |
2328 | health_code_update(); |
2329 | ||
b8aa1682 | 2330 | /* Infinite blocking call, waiting for transmission */ |
87c1611d | 2331 | DBG3("Relayd worker thread polling..."); |
f385ae0a | 2332 | health_poll_entry(); |
b8aa1682 | 2333 | ret = lttng_poll_wait(&events, -1); |
f385ae0a | 2334 | health_poll_exit(); |
b8aa1682 JD |
2335 | if (ret < 0) { |
2336 | /* | |
2337 | * Restart interrupted system call. | |
2338 | */ | |
2339 | if (errno == EINTR) { | |
2340 | goto restart; | |
2341 | } | |
2342 | goto error; | |
2343 | } | |
2344 | ||
0d9c5d77 DG |
2345 | nb_fd = ret; |
2346 | ||
beaad64c DG |
2347 | /* |
2348 | * Process control. The control connection is prioritised so we don't | |
2349 | * starve it with high throughout put tracing data on the data | |
2350 | * connection. | |
2351 | */ | |
b8aa1682 JD |
2352 | for (i = 0; i < nb_fd; i++) { |
2353 | /* Fetch once the poll data */ | |
beaad64c DG |
2354 | uint32_t revents = LTTNG_POLL_GETEV(&events, i); |
2355 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
b8aa1682 | 2356 | |
f385ae0a MD |
2357 | health_code_update(); |
2358 | ||
b8aa1682 JD |
2359 | /* Thread quit pipe has been closed. Killing thread. */ |
2360 | ret = check_thread_quit_pipe(pollfd, revents); | |
2361 | if (ret) { | |
095a4ae5 MD |
2362 | err = 0; |
2363 | goto exit; | |
b8aa1682 JD |
2364 | } |
2365 | ||
e85cdca9 DG |
2366 | /* Inspect the relay conn pipe for new connection */ |
2367 | if (pollfd == relay_conn_pipe[0]) { | |
b8aa1682 | 2368 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { |
e85cdca9 | 2369 | ERR("Relay connection pipe error"); |
b8aa1682 JD |
2370 | goto error; |
2371 | } else if (revents & LPOLLIN) { | |
e85cdca9 | 2372 | ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn)); |
b8aa1682 JD |
2373 | if (ret < 0) { |
2374 | goto error; | |
2375 | } | |
e85cdca9 DG |
2376 | conn->sessions_ht = sessions_ht; |
2377 | connection_init(conn); | |
2378 | lttng_poll_add(&events, conn->sock->fd, | |
2379 | LPOLLIN | LPOLLRDHUP); | |
2380 | rcu_read_lock(); | |
2381 | lttng_ht_add_unique_ulong(relay_connections_ht, | |
2382 | &conn->sock_n); | |
9d1bbf21 | 2383 | rcu_read_unlock(); |
e85cdca9 | 2384 | DBG("Connection socket %d added", conn->sock->fd); |
b8aa1682 | 2385 | } |
e85cdca9 DG |
2386 | } else { |
2387 | rcu_read_lock(); | |
2388 | conn = connection_find_by_sock(relay_connections_ht, pollfd); | |
2389 | /* If not found, there is a synchronization issue. */ | |
2390 | assert(conn); | |
2391 | ||
2392 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
2393 | cleanup_connection_pollfd(&events, pollfd); | |
2394 | destroy_connection(relay_connections_ht, conn); | |
beaad64c DG |
2395 | if (last_seen_data_fd == pollfd) { |
2396 | last_seen_data_fd = last_notdel_data_fd; | |
2397 | } | |
b8aa1682 | 2398 | } else if (revents & LPOLLIN) { |
e85cdca9 DG |
2399 | if (conn->type == RELAY_CONTROL) { |
2400 | ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr, | |
2401 | sizeof(recv_hdr), 0); | |
b8aa1682 | 2402 | if (ret <= 0) { |
e85cdca9 DG |
2403 | /* Connection closed */ |
2404 | cleanup_connection_pollfd(&events, pollfd); | |
2405 | destroy_connection(relay_connections_ht, conn); | |
b8aa1682 JD |
2406 | DBG("Control connection closed with %d", pollfd); |
2407 | } else { | |
e85cdca9 | 2408 | ret = relay_process_control(&recv_hdr, conn); |
b8aa1682 | 2409 | if (ret < 0) { |
beaad64c | 2410 | /* Clear the session on error. */ |
e85cdca9 DG |
2411 | cleanup_connection_pollfd(&events, pollfd); |
2412 | destroy_connection(relay_connections_ht, conn); | |
b8aa1682 JD |
2413 | DBG("Connection closed with %d", pollfd); |
2414 | } | |
beaad64c | 2415 | seen_control = 1; |
b8aa1682 | 2416 | } |
beaad64c DG |
2417 | } else { |
2418 | /* | |
2419 | * Flag the last seen data fd not deleted. It will be | |
2420 | * used as the last seen fd if any fd gets deleted in | |
2421 | * this first loop. | |
2422 | */ | |
2423 | last_notdel_data_fd = pollfd; | |
2424 | } | |
e85cdca9 DG |
2425 | } else { |
2426 | ERR("Unknown poll events %u for sock %d", revents, pollfd); | |
beaad64c DG |
2427 | } |
2428 | rcu_read_unlock(); | |
2429 | } | |
2430 | } | |
2431 | ||
2432 | /* | |
2433 | * The last loop handled a control request, go back to poll to make | |
2434 | * sure we prioritise the control socket. | |
2435 | */ | |
2436 | if (seen_control) { | |
2437 | continue; | |
2438 | } | |
2439 | ||
2440 | if (last_seen_data_fd >= 0) { | |
2441 | for (i = 0; i < nb_fd; i++) { | |
2442 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
f385ae0a MD |
2443 | |
2444 | health_code_update(); | |
2445 | ||
beaad64c DG |
2446 | if (last_seen_data_fd == pollfd) { |
2447 | idx = i; | |
2448 | break; | |
2449 | } | |
2450 | } | |
2451 | } | |
2452 | ||
2453 | /* Process data connection. */ | |
2454 | for (i = idx + 1; i < nb_fd; i++) { | |
2455 | /* Fetch the poll data. */ | |
2456 | uint32_t revents = LTTNG_POLL_GETEV(&events, i); | |
2457 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
2458 | ||
f385ae0a MD |
2459 | health_code_update(); |
2460 | ||
beaad64c | 2461 | /* Skip the command pipe. It's handled in the first loop. */ |
e85cdca9 | 2462 | if (pollfd == relay_conn_pipe[0]) { |
beaad64c DG |
2463 | continue; |
2464 | } | |
2465 | ||
2466 | if (revents) { | |
2467 | rcu_read_lock(); | |
e85cdca9 DG |
2468 | conn = connection_find_by_sock(relay_connections_ht, pollfd); |
2469 | if (!conn) { | |
beaad64c DG |
2470 | /* Skip it. Might be removed before. */ |
2471 | rcu_read_unlock(); | |
2472 | continue; | |
2473 | } | |
beaad64c DG |
2474 | |
2475 | if (revents & LPOLLIN) { | |
e85cdca9 | 2476 | if (conn->type != RELAY_DATA) { |
beaad64c DG |
2477 | continue; |
2478 | } | |
2479 | ||
e85cdca9 DG |
2480 | ret = relay_process_data(conn); |
2481 | /* Connection closed */ | |
beaad64c | 2482 | if (ret < 0) { |
e85cdca9 DG |
2483 | cleanup_connection_pollfd(&events, pollfd); |
2484 | destroy_connection(relay_connections_ht, conn); | |
beaad64c DG |
2485 | DBG("Data connection closed with %d", pollfd); |
2486 | /* | |
2487 | * Every goto restart call sets the last seen fd where | |
2488 | * here we don't really care since we gracefully | |
2489 | * continue the loop after the connection is deleted. | |
2490 | */ | |
2491 | } else { | |
2492 | /* Keep last seen port. */ | |
2493 | last_seen_data_fd = pollfd; | |
2494 | rcu_read_unlock(); | |
2495 | goto restart; | |
b8aa1682 JD |
2496 | } |
2497 | } | |
9d1bbf21 | 2498 | rcu_read_unlock(); |
b8aa1682 JD |
2499 | } |
2500 | } | |
beaad64c | 2501 | last_seen_data_fd = -1; |
b8aa1682 JD |
2502 | } |
2503 | ||
f385ae0a MD |
2504 | /* Normal exit, no error */ |
2505 | ret = 0; | |
2506 | ||
095a4ae5 | 2507 | exit: |
b8aa1682 JD |
2508 | error: |
2509 | lttng_poll_clean(&events); | |
2510 | ||
e85cdca9 | 2511 | /* Cleanup reamaining connection object. */ |
9d1bbf21 | 2512 | rcu_read_lock(); |
e85cdca9 DG |
2513 | cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn, |
2514 | sock_n.node) { | |
f385ae0a | 2515 | health_code_update(); |
e85cdca9 | 2516 | destroy_connection(relay_connections_ht, conn); |
b8aa1682 | 2517 | } |
94d49140 | 2518 | rcu_read_unlock(); |
7d2f7452 DG |
2519 | error_poll_create: |
2520 | lttng_ht_destroy(indexes_ht); | |
1c20f0e2 | 2521 | indexes_ht_error: |
b8aa1682 | 2522 | lttng_ht_destroy(relay_connections_ht); |
095a4ae5 | 2523 | relay_connections_ht_error: |
e85cdca9 DG |
2524 | /* Close relay conn pipes */ |
2525 | utils_close_pipe(relay_conn_pipe); | |
095a4ae5 MD |
2526 | if (err) { |
2527 | DBG("Thread exited with error"); | |
2528 | } | |
b8aa1682 | 2529 | DBG("Worker thread cleanup complete"); |
095a4ae5 | 2530 | free(data_buffer); |
4e5b8b82 | 2531 | error_testpoint: |
f385ae0a MD |
2532 | if (err) { |
2533 | health_error(); | |
2534 | ERR("Health error occurred in %s", __func__); | |
2535 | } | |
2536 | health_unregister(health_relayd); | |
9d1bbf21 | 2537 | rcu_unregister_thread(); |
f385ae0a | 2538 | stop_threads(); |
b8aa1682 JD |
2539 | return NULL; |
2540 | } | |
2541 | ||
2542 | /* | |
2543 | * Create the relay command pipe to wake thread_manage_apps. | |
2544 | * Closed in cleanup(). | |
2545 | */ | |
e85cdca9 | 2546 | static int create_relay_conn_pipe(void) |
b8aa1682 | 2547 | { |
a02de639 | 2548 | int ret; |
b8aa1682 | 2549 | |
e85cdca9 | 2550 | ret = utils_create_pipe_cloexec(relay_conn_pipe); |
b8aa1682 | 2551 | |
b8aa1682 JD |
2552 | return ret; |
2553 | } | |
2554 | ||
2555 | /* | |
2556 | * main | |
2557 | */ | |
2558 | int main(int argc, char **argv) | |
2559 | { | |
2560 | int ret = 0; | |
2561 | void *status; | |
d3e2ba59 | 2562 | struct relay_local_data *relay_ctx; |
b8aa1682 | 2563 | |
b8aa1682 JD |
2564 | /* Parse arguments */ |
2565 | progname = argv[0]; | |
c617c0c6 | 2566 | if ((ret = parse_args(argc, argv)) < 0) { |
a02de639 | 2567 | goto exit; |
b8aa1682 JD |
2568 | } |
2569 | ||
2570 | if ((ret = set_signal_handler()) < 0) { | |
2571 | goto exit; | |
2572 | } | |
2573 | ||
4d513a50 DG |
2574 | /* Try to create directory if -o, --output is specified. */ |
2575 | if (opt_output_path) { | |
994fa64f DG |
2576 | if (*opt_output_path != '/') { |
2577 | ERR("Please specify an absolute path for -o, --output PATH"); | |
2578 | goto exit; | |
2579 | } | |
2580 | ||
4d513a50 DG |
2581 | ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG); |
2582 | if (ret < 0) { | |
2583 | ERR("Unable to create %s", opt_output_path); | |
2584 | goto exit; | |
2585 | } | |
2586 | } | |
2587 | ||
b8aa1682 | 2588 | /* Daemonize */ |
d1a3048a | 2589 | if (opt_daemon || opt_background) { |
02a6bb53 MD |
2590 | int i; |
2591 | ||
2592 | ret = lttng_daemonize(&child_ppid, &recv_child_signal, | |
2593 | !opt_background); | |
b8aa1682 | 2594 | if (ret < 0) { |
a02de639 | 2595 | goto exit; |
b8aa1682 | 2596 | } |
02a6bb53 MD |
2597 | |
2598 | /* | |
2599 | * We are in the child. Make sure all other file | |
2600 | * descriptors are closed, in case we are called with | |
2601 | * more opened file descriptors than the standard ones. | |
2602 | */ | |
2603 | for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) { | |
2604 | (void) close(i); | |
2605 | } | |
2606 | } | |
2607 | ||
2608 | /* Create thread quit pipe */ | |
2609 | if ((ret = init_thread_quit_pipe()) < 0) { | |
2610 | goto error; | |
b8aa1682 JD |
2611 | } |
2612 | ||
1c20f0e2 JD |
2613 | /* We need those values for the file/dir creation. */ |
2614 | relayd_uid = getuid(); | |
2615 | relayd_gid = getgid(); | |
b8aa1682 | 2616 | |
1c20f0e2 JD |
2617 | /* Check if daemon is UID = 0 */ |
2618 | if (relayd_uid == 0) { | |
a6a062a4 AM |
2619 | if (control_uri->port < 1024 || data_uri->port < 1024 || |
2620 | live_uri->port < 1024) { | |
b8aa1682 JD |
2621 | ERR("Need to be root to use ports < 1024"); |
2622 | ret = -1; | |
a02de639 | 2623 | goto exit; |
b8aa1682 JD |
2624 | } |
2625 | } | |
2626 | ||
2627 | /* Setup the thread apps communication pipe. */ | |
e85cdca9 | 2628 | if ((ret = create_relay_conn_pipe()) < 0) { |
b8aa1682 JD |
2629 | goto exit; |
2630 | } | |
2631 | ||
2632 | /* Init relay command queue. */ | |
e85cdca9 | 2633 | cds_wfq_init(&relay_conn_queue.queue); |
b8aa1682 JD |
2634 | |
2635 | /* Set up max poll set size */ | |
2636 | lttng_poll_set_max_size(); | |
2637 | ||
554831e7 MD |
2638 | /* Initialize communication library */ |
2639 | lttcomm_init(); | |
d5a5e58f | 2640 | lttcomm_inet_init(); |
554831e7 | 2641 | |
d3e2ba59 JD |
2642 | relay_ctx = zmalloc(sizeof(struct relay_local_data)); |
2643 | if (!relay_ctx) { | |
2644 | PERROR("relay_ctx"); | |
2645 | goto exit; | |
2646 | } | |
2647 | ||
2648 | /* tables of sessions indexed by session ID */ | |
eb702af5 | 2649 | relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
d3e2ba59 JD |
2650 | if (!relay_ctx->sessions_ht) { |
2651 | goto exit_relay_ctx_sessions; | |
2652 | } | |
2653 | ||
2654 | /* tables of streams indexed by stream ID */ | |
eb702af5 | 2655 | relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
d3e2ba59 JD |
2656 | if (!relay_streams_ht) { |
2657 | goto exit_relay_ctx_streams; | |
2658 | } | |
2659 | ||
2660 | /* tables of streams indexed by stream ID */ | |
92c6ca54 DG |
2661 | viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
2662 | if (!viewer_streams_ht) { | |
d3e2ba59 JD |
2663 | goto exit_relay_ctx_viewer_streams; |
2664 | } | |
2665 | ||
55706a7d MD |
2666 | /* Initialize thread health monitoring */ |
2667 | health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES); | |
2668 | if (!health_relayd) { | |
2669 | PERROR("health_app_create error"); | |
2670 | goto exit_health_app_create; | |
2671 | } | |
2672 | ||
65931c8b MD |
2673 | ret = utils_create_pipe(health_quit_pipe); |
2674 | if (ret < 0) { | |
2675 | goto error_health_pipe; | |
2676 | } | |
2677 | ||
2678 | /* Create thread to manage the client socket */ | |
2679 | ret = pthread_create(&health_thread, NULL, | |
2680 | thread_manage_health, (void *) NULL); | |
2681 | if (ret != 0) { | |
2682 | PERROR("pthread_create health"); | |
2683 | goto health_error; | |
2684 | } | |
2685 | ||
b8aa1682 JD |
2686 | /* Setup the dispatcher thread */ |
2687 | ret = pthread_create(&dispatcher_thread, NULL, | |
2688 | relay_thread_dispatcher, (void *) NULL); | |
2689 | if (ret != 0) { | |
2690 | PERROR("pthread_create dispatcher"); | |
2691 | goto exit_dispatcher; | |
2692 | } | |
2693 | ||
2694 | /* Setup the worker thread */ | |
2695 | ret = pthread_create(&worker_thread, NULL, | |
d3e2ba59 | 2696 | relay_thread_worker, (void *) relay_ctx); |
b8aa1682 JD |
2697 | if (ret != 0) { |
2698 | PERROR("pthread_create worker"); | |
2699 | goto exit_worker; | |
2700 | } | |
2701 | ||
2702 | /* Setup the listener thread */ | |
2703 | ret = pthread_create(&listener_thread, NULL, | |
2704 | relay_thread_listener, (void *) NULL); | |
2705 | if (ret != 0) { | |
2706 | PERROR("pthread_create listener"); | |
2707 | goto exit_listener; | |
2708 | } | |
2709 | ||
3557d456 | 2710 | ret = live_start_threads(live_uri, relay_ctx); |
d3e2ba59 JD |
2711 | if (ret != 0) { |
2712 | ERR("Starting live viewer threads"); | |
50138f51 | 2713 | goto exit_live; |
d3e2ba59 JD |
2714 | } |
2715 | ||
50138f51 | 2716 | exit_live: |
b8aa1682 JD |
2717 | ret = pthread_join(listener_thread, &status); |
2718 | if (ret != 0) { | |
2719 | PERROR("pthread_join"); | |
2720 | goto error; /* join error, exit without cleanup */ | |
2721 | } | |
2722 | ||
50138f51 | 2723 | exit_listener: |
b8aa1682 JD |
2724 | ret = pthread_join(worker_thread, &status); |
2725 | if (ret != 0) { | |
2726 | PERROR("pthread_join"); | |
2727 | goto error; /* join error, exit without cleanup */ | |
2728 | } | |
2729 | ||
50138f51 | 2730 | exit_worker: |
b8aa1682 JD |
2731 | ret = pthread_join(dispatcher_thread, &status); |
2732 | if (ret != 0) { | |
2733 | PERROR("pthread_join"); | |
2734 | goto error; /* join error, exit without cleanup */ | |
2735 | } | |
42415026 | 2736 | |
50138f51 | 2737 | exit_dispatcher: |
65931c8b MD |
2738 | ret = pthread_join(health_thread, &status); |
2739 | if (ret != 0) { | |
2740 | PERROR("pthread_join health thread"); | |
2741 | goto error; /* join error, exit without cleanup */ | |
2742 | } | |
2743 | ||
bd11e201 MD |
2744 | /* |
2745 | * Stop live threads only after joining other threads. | |
2746 | */ | |
2747 | live_stop_threads(); | |
2748 | ||
65931c8b MD |
2749 | health_error: |
2750 | utils_close_pipe(health_quit_pipe); | |
2751 | ||
2752 | error_health_pipe: | |
55706a7d MD |
2753 | health_app_destroy(health_relayd); |
2754 | ||
2755 | exit_health_app_create: | |
92c6ca54 | 2756 | lttng_ht_destroy(viewer_streams_ht); |
d3e2ba59 JD |
2757 | |
2758 | exit_relay_ctx_viewer_streams: | |
2759 | lttng_ht_destroy(relay_streams_ht); | |
2760 | ||
2761 | exit_relay_ctx_streams: | |
2762 | lttng_ht_destroy(relay_ctx->sessions_ht); | |
2763 | ||
2764 | exit_relay_ctx_sessions: | |
2765 | free(relay_ctx); | |
b8aa1682 JD |
2766 | |
2767 | exit: | |
2768 | cleanup(); | |
2769 | if (!ret) { | |
2770 | exit(EXIT_SUCCESS); | |
2771 | } | |
a02de639 | 2772 | |
b8aa1682 JD |
2773 | error: |
2774 | exit(EXIT_FAILURE); | |
2775 | } |