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" |
b8aa1682 JD |
64 | |
65 | /* command line options */ | |
0f907de1 | 66 | char *opt_output_path; |
d1a3048a | 67 | static int opt_daemon, opt_background; |
02a6bb53 MD |
68 | |
69 | /* | |
70 | * We need to wait for listener and live listener threads, as well as | |
71 | * health check thread, before being ready to signal readiness. | |
72 | */ | |
73 | #define NR_LTTNG_RELAY_READY 3 | |
74 | static int lttng_relay_ready = NR_LTTNG_RELAY_READY; | |
75 | static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */ | |
76 | static pid_t child_ppid; /* Internal parent PID use with daemonize. */ | |
77 | ||
095a4ae5 MD |
78 | static struct lttng_uri *control_uri; |
79 | static struct lttng_uri *data_uri; | |
d3e2ba59 | 80 | static struct lttng_uri *live_uri; |
b8aa1682 JD |
81 | |
82 | const char *progname; | |
b8aa1682 | 83 | |
65931c8b MD |
84 | const char *tracing_group_name = DEFAULT_TRACING_GROUP; |
85 | ||
b8aa1682 JD |
86 | /* |
87 | * Quit pipe for all threads. This permits a single cancellation point | |
88 | * for all threads when receiving an event on the pipe. | |
89 | */ | |
3557d456 | 90 | int thread_quit_pipe[2] = { -1, -1 }; |
b8aa1682 JD |
91 | |
92 | /* | |
93 | * This pipe is used to inform the worker thread that a command is queued and | |
94 | * ready to be processed. | |
95 | */ | |
96 | static int relay_cmd_pipe[2] = { -1, -1 }; | |
97 | ||
26c9d55e | 98 | /* Shared between threads */ |
b8aa1682 JD |
99 | static int dispatch_thread_exit; |
100 | ||
101 | static pthread_t listener_thread; | |
102 | static pthread_t dispatcher_thread; | |
103 | static pthread_t worker_thread; | |
65931c8b | 104 | static pthread_t health_thread; |
b8aa1682 | 105 | |
095a4ae5 MD |
106 | static uint64_t last_relay_stream_id; |
107 | static uint64_t last_relay_session_id; | |
b8aa1682 JD |
108 | |
109 | /* | |
110 | * Relay command queue. | |
111 | * | |
112 | * The relay_thread_listener and relay_thread_dispatcher communicate with this | |
113 | * queue. | |
114 | */ | |
115 | static struct relay_cmd_queue relay_cmd_queue; | |
116 | ||
117 | /* buffer allocated at startup, used to store the trace data */ | |
095a4ae5 MD |
118 | static char *data_buffer; |
119 | static unsigned int data_buffer_size; | |
b8aa1682 | 120 | |
1c20f0e2 JD |
121 | /* We need those values for the file/dir creation. */ |
122 | static uid_t relayd_uid; | |
123 | static gid_t relayd_gid; | |
124 | ||
d3e2ba59 JD |
125 | /* Global relay stream hash table. */ |
126 | struct lttng_ht *relay_streams_ht; | |
127 | ||
92c6ca54 DG |
128 | /* Global relay viewer stream hash table. */ |
129 | struct lttng_ht *viewer_streams_ht; | |
130 | ||
0a6518b0 DG |
131 | /* Global hash table that stores relay index object. */ |
132 | struct lttng_ht *indexes_ht; | |
133 | ||
55706a7d | 134 | /* Relayd health monitoring */ |
eea7556c | 135 | struct health_app *health_relayd; |
55706a7d | 136 | |
b8aa1682 JD |
137 | /* |
138 | * usage function on stderr | |
139 | */ | |
140 | static | |
141 | void usage(void) | |
142 | { | |
143 | fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname); | |
994fa64f DG |
144 | fprintf(stderr, " -h, --help Display this usage.\n"); |
145 | fprintf(stderr, " -d, --daemonize Start as a daemon.\n"); | |
d1a3048a | 146 | fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n"); |
994fa64f DG |
147 | fprintf(stderr, " -C, --control-port URL Control port listening.\n"); |
148 | fprintf(stderr, " -D, --data-port URL Data port listening.\n"); | |
a6a062a4 | 149 | fprintf(stderr, " -L, --live-port URL Live view port listening.\n"); |
994fa64f DG |
150 | fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n"); |
151 | fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n"); | |
65931c8b | 152 | fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n"); |
b8aa1682 JD |
153 | } |
154 | ||
155 | static | |
156 | int parse_args(int argc, char **argv) | |
157 | { | |
158 | int c; | |
159 | int ret = 0; | |
160 | char *default_address; | |
161 | ||
162 | static struct option long_options[] = { | |
e3678fd8 MD |
163 | { "control-port", 1, 0, 'C', }, |
164 | { "data-port", 1, 0, 'D', }, | |
165 | { "daemonize", 0, 0, 'd', }, | |
65931c8b | 166 | { "group", 1, 0, 'g', }, |
e3678fd8 MD |
167 | { "help", 0, 0, 'h', }, |
168 | { "output", 1, 0, 'o', }, | |
169 | { "verbose", 0, 0, 'v', }, | |
06b55dbb | 170 | { "background", 0, 0, 'b' }, |
095a4ae5 | 171 | { NULL, 0, 0, 0, }, |
b8aa1682 JD |
172 | }; |
173 | ||
174 | while (1) { | |
175 | int option_index = 0; | |
06b55dbb | 176 | c = getopt_long(argc, argv, "dhv" "C:D:L:o:g:b", |
b8aa1682 JD |
177 | long_options, &option_index); |
178 | if (c == -1) { | |
179 | break; | |
180 | } | |
181 | ||
182 | switch (c) { | |
183 | case 0: | |
184 | fprintf(stderr, "option %s", long_options[option_index].name); | |
185 | if (optarg) { | |
186 | fprintf(stderr, " with arg %s\n", optarg); | |
187 | } | |
188 | break; | |
189 | case 'C': | |
190 | ret = uri_parse(optarg, &control_uri); | |
191 | if (ret < 0) { | |
192 | ERR("Invalid control URI specified"); | |
193 | goto exit; | |
194 | } | |
195 | if (control_uri->port == 0) { | |
196 | control_uri->port = DEFAULT_NETWORK_CONTROL_PORT; | |
197 | } | |
198 | break; | |
199 | case 'D': | |
200 | ret = uri_parse(optarg, &data_uri); | |
201 | if (ret < 0) { | |
202 | ERR("Invalid data URI specified"); | |
203 | goto exit; | |
204 | } | |
205 | if (data_uri->port == 0) { | |
206 | data_uri->port = DEFAULT_NETWORK_DATA_PORT; | |
207 | } | |
208 | break; | |
a6a062a4 AM |
209 | case 'L': |
210 | ret = uri_parse(optarg, &live_uri); | |
211 | if (ret < 0) { | |
212 | ERR("Invalid live URI specified"); | |
213 | goto exit; | |
214 | } | |
215 | if (live_uri->port == 0) { | |
216 | live_uri->port = DEFAULT_NETWORK_VIEWER_PORT; | |
217 | } | |
218 | break; | |
b8aa1682 JD |
219 | case 'd': |
220 | opt_daemon = 1; | |
221 | break; | |
06b55dbb DG |
222 | case 'b': |
223 | opt_background = 1; | |
224 | break; | |
65931c8b MD |
225 | case 'g': |
226 | tracing_group_name = optarg; | |
227 | break; | |
b8aa1682 JD |
228 | case 'h': |
229 | usage(); | |
230 | exit(EXIT_FAILURE); | |
231 | case 'o': | |
232 | ret = asprintf(&opt_output_path, "%s", optarg); | |
233 | if (ret < 0) { | |
234 | PERROR("asprintf opt_output_path"); | |
235 | goto exit; | |
236 | } | |
237 | break; | |
238 | case 'v': | |
239 | /* Verbose level can increase using multiple -v */ | |
240 | lttng_opt_verbose += 1; | |
241 | break; | |
242 | default: | |
243 | /* Unknown option or other error. | |
244 | * Error is printed by getopt, just return */ | |
245 | ret = -1; | |
246 | goto exit; | |
247 | } | |
248 | } | |
249 | ||
250 | /* assign default values */ | |
251 | if (control_uri == NULL) { | |
252 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
253 | DEFAULT_NETWORK_CONTROL_PORT); | |
254 | if (ret < 0) { | |
255 | PERROR("asprintf default data address"); | |
256 | goto exit; | |
257 | } | |
258 | ||
259 | ret = uri_parse(default_address, &control_uri); | |
260 | free(default_address); | |
261 | if (ret < 0) { | |
262 | ERR("Invalid control URI specified"); | |
263 | goto exit; | |
264 | } | |
265 | } | |
266 | if (data_uri == NULL) { | |
267 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
268 | DEFAULT_NETWORK_DATA_PORT); | |
269 | if (ret < 0) { | |
270 | PERROR("asprintf default data address"); | |
271 | goto exit; | |
272 | } | |
273 | ||
274 | ret = uri_parse(default_address, &data_uri); | |
275 | free(default_address); | |
276 | if (ret < 0) { | |
277 | ERR("Invalid data URI specified"); | |
278 | goto exit; | |
279 | } | |
280 | } | |
d3e2ba59 JD |
281 | if (live_uri == NULL) { |
282 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
283 | DEFAULT_NETWORK_VIEWER_PORT); | |
284 | if (ret < 0) { | |
285 | PERROR("asprintf default viewer control address"); | |
286 | goto exit; | |
287 | } | |
288 | ||
289 | ret = uri_parse(default_address, &live_uri); | |
290 | free(default_address); | |
291 | if (ret < 0) { | |
292 | ERR("Invalid viewer control URI specified"); | |
293 | goto exit; | |
294 | } | |
295 | } | |
b8aa1682 JD |
296 | |
297 | exit: | |
298 | return ret; | |
299 | } | |
300 | ||
301 | /* | |
302 | * Cleanup the daemon | |
303 | */ | |
304 | static | |
305 | void cleanup(void) | |
306 | { | |
b8aa1682 JD |
307 | DBG("Cleaning up"); |
308 | ||
095a4ae5 MD |
309 | /* free the dynamically allocated opt_output_path */ |
310 | free(opt_output_path); | |
311 | ||
a02de639 CB |
312 | /* Close thread quit pipes */ |
313 | utils_close_pipe(thread_quit_pipe); | |
314 | ||
710c1f73 DG |
315 | uri_free(control_uri); |
316 | uri_free(data_uri); | |
a6a062a4 | 317 | /* Live URI is freed in the live thread. */ |
b8aa1682 JD |
318 | } |
319 | ||
320 | /* | |
321 | * Write to writable pipe used to notify a thread. | |
322 | */ | |
323 | static | |
324 | int notify_thread_pipe(int wpipe) | |
325 | { | |
6cd525e8 | 326 | ssize_t ret; |
b8aa1682 | 327 | |
6cd525e8 MD |
328 | ret = lttng_write(wpipe, "!", 1); |
329 | if (ret < 1) { | |
b8aa1682 JD |
330 | PERROR("write poll pipe"); |
331 | } | |
332 | ||
333 | return ret; | |
334 | } | |
335 | ||
65931c8b MD |
336 | static void notify_health_quit_pipe(int *pipe) |
337 | { | |
6cd525e8 | 338 | ssize_t ret; |
65931c8b | 339 | |
6cd525e8 MD |
340 | ret = lttng_write(pipe[1], "4", 1); |
341 | if (ret < 1) { | |
65931c8b MD |
342 | PERROR("write relay health quit"); |
343 | } | |
344 | } | |
345 | ||
b8aa1682 JD |
346 | /* |
347 | * Stop all threads by closing the thread quit pipe. | |
348 | */ | |
349 | static | |
350 | void stop_threads(void) | |
351 | { | |
352 | int ret; | |
353 | ||
354 | /* Stopping all threads */ | |
355 | DBG("Terminating all threads"); | |
356 | ret = notify_thread_pipe(thread_quit_pipe[1]); | |
357 | if (ret < 0) { | |
358 | ERR("write error on thread quit pipe"); | |
359 | } | |
360 | ||
65931c8b MD |
361 | notify_health_quit_pipe(health_quit_pipe); |
362 | ||
b8aa1682 | 363 | /* Dispatch thread */ |
26c9d55e | 364 | CMM_STORE_SHARED(dispatch_thread_exit, 1); |
b8aa1682 JD |
365 | futex_nto1_wake(&relay_cmd_queue.futex); |
366 | } | |
367 | ||
368 | /* | |
369 | * Signal handler for the daemon | |
370 | * | |
371 | * Simply stop all worker threads, leaving main() return gracefully after | |
372 | * joining all threads and calling cleanup(). | |
373 | */ | |
374 | static | |
375 | void sighandler(int sig) | |
376 | { | |
377 | switch (sig) { | |
378 | case SIGPIPE: | |
379 | DBG("SIGPIPE caught"); | |
380 | return; | |
381 | case SIGINT: | |
382 | DBG("SIGINT caught"); | |
383 | stop_threads(); | |
384 | break; | |
385 | case SIGTERM: | |
386 | DBG("SIGTERM caught"); | |
387 | stop_threads(); | |
388 | break; | |
02a6bb53 MD |
389 | case SIGUSR1: |
390 | CMM_STORE_SHARED(recv_child_signal, 1); | |
391 | break; | |
b8aa1682 JD |
392 | default: |
393 | break; | |
394 | } | |
395 | } | |
396 | ||
397 | /* | |
398 | * Setup signal handler for : | |
399 | * SIGINT, SIGTERM, SIGPIPE | |
400 | */ | |
401 | static | |
402 | int set_signal_handler(void) | |
403 | { | |
404 | int ret = 0; | |
405 | struct sigaction sa; | |
406 | sigset_t sigset; | |
407 | ||
408 | if ((ret = sigemptyset(&sigset)) < 0) { | |
409 | PERROR("sigemptyset"); | |
410 | return ret; | |
411 | } | |
412 | ||
413 | sa.sa_handler = sighandler; | |
414 | sa.sa_mask = sigset; | |
415 | sa.sa_flags = 0; | |
416 | if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) { | |
417 | PERROR("sigaction"); | |
418 | return ret; | |
419 | } | |
420 | ||
421 | if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) { | |
422 | PERROR("sigaction"); | |
423 | return ret; | |
424 | } | |
425 | ||
426 | if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) { | |
427 | PERROR("sigaction"); | |
428 | return ret; | |
429 | } | |
430 | ||
02a6bb53 MD |
431 | if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) { |
432 | PERROR("sigaction"); | |
433 | return ret; | |
434 | } | |
435 | ||
436 | DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT"); | |
b8aa1682 JD |
437 | |
438 | return ret; | |
439 | } | |
440 | ||
02a6bb53 MD |
441 | void lttng_relay_notify_ready(void) |
442 | { | |
443 | /* Notify the parent of the fork() process that we are ready. */ | |
444 | if (opt_daemon || opt_background) { | |
445 | if (uatomic_sub_return(<tng_relay_ready, 1) == 0) { | |
446 | kill(child_ppid, SIGUSR1); | |
447 | } | |
448 | } | |
449 | } | |
450 | ||
b8aa1682 JD |
451 | /* |
452 | * Init thread quit pipe. | |
453 | * | |
454 | * Return -1 on error or 0 if all pipes are created. | |
455 | */ | |
456 | static | |
457 | int init_thread_quit_pipe(void) | |
458 | { | |
a02de639 | 459 | int ret; |
b8aa1682 | 460 | |
a02de639 | 461 | ret = utils_create_pipe_cloexec(thread_quit_pipe); |
b8aa1682 | 462 | |
b8aa1682 JD |
463 | return ret; |
464 | } | |
465 | ||
466 | /* | |
467 | * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set. | |
468 | */ | |
469 | static | |
470 | int create_thread_poll_set(struct lttng_poll_event *events, int size) | |
471 | { | |
472 | int ret; | |
473 | ||
474 | if (events == NULL || size == 0) { | |
475 | ret = -1; | |
476 | goto error; | |
477 | } | |
478 | ||
479 | ret = lttng_poll_create(events, size, LTTNG_CLOEXEC); | |
480 | if (ret < 0) { | |
481 | goto error; | |
482 | } | |
483 | ||
484 | /* Add quit pipe */ | |
532ed517 | 485 | ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR); |
b8aa1682 JD |
486 | if (ret < 0) { |
487 | goto error; | |
488 | } | |
489 | ||
490 | return 0; | |
491 | ||
492 | error: | |
493 | return ret; | |
494 | } | |
495 | ||
496 | /* | |
497 | * Check if the thread quit pipe was triggered. | |
498 | * | |
499 | * Return 1 if it was triggered else 0; | |
500 | */ | |
501 | static | |
502 | int check_thread_quit_pipe(int fd, uint32_t events) | |
503 | { | |
504 | if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) { | |
505 | return 1; | |
506 | } | |
507 | ||
508 | return 0; | |
509 | } | |
510 | ||
511 | /* | |
512 | * Create and init socket from uri. | |
513 | */ | |
514 | static | |
515 | struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri) | |
516 | { | |
517 | int ret; | |
518 | struct lttcomm_sock *sock = NULL; | |
519 | ||
520 | sock = lttcomm_alloc_sock_from_uri(uri); | |
521 | if (sock == NULL) { | |
522 | ERR("Allocating socket"); | |
523 | goto error; | |
524 | } | |
525 | ||
526 | ret = lttcomm_create_sock(sock); | |
527 | if (ret < 0) { | |
528 | goto error; | |
529 | } | |
530 | DBG("Listening on sock %d", sock->fd); | |
531 | ||
532 | ret = sock->ops->bind(sock); | |
533 | if (ret < 0) { | |
534 | goto error; | |
535 | } | |
536 | ||
537 | ret = sock->ops->listen(sock, -1); | |
538 | if (ret < 0) { | |
539 | goto error; | |
540 | ||
541 | } | |
542 | ||
543 | return sock; | |
544 | ||
545 | error: | |
546 | if (sock) { | |
547 | lttcomm_destroy_sock(sock); | |
548 | } | |
549 | return NULL; | |
550 | } | |
551 | ||
173af62f DG |
552 | /* |
553 | * Return nonzero if stream needs to be closed. | |
554 | */ | |
555 | static | |
556 | int close_stream_check(struct relay_stream *stream) | |
557 | { | |
558 | ||
559 | if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) { | |
f7079f67 DG |
560 | /* |
561 | * We are about to close the stream so set the data pending flag to 1 | |
562 | * which will make the end data pending command skip the stream which | |
563 | * is now closed and ready. Note that after proceeding to a file close, | |
564 | * the written file is ready for reading. | |
565 | */ | |
566 | stream->data_pending_check_done = 1; | |
173af62f DG |
567 | return 1; |
568 | } | |
569 | return 0; | |
570 | } | |
571 | ||
b8aa1682 JD |
572 | /* |
573 | * This thread manages the listening for new connections on the network | |
574 | */ | |
575 | static | |
576 | void *relay_thread_listener(void *data) | |
577 | { | |
095a4ae5 | 578 | int i, ret, pollfd, err = -1; |
b8aa1682 JD |
579 | int val = 1; |
580 | uint32_t revents, nb_fd; | |
581 | struct lttng_poll_event events; | |
582 | struct lttcomm_sock *control_sock, *data_sock; | |
583 | ||
b8aa1682 JD |
584 | DBG("[thread] Relay listener started"); |
585 | ||
55706a7d MD |
586 | health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER); |
587 | ||
f385ae0a MD |
588 | health_code_update(); |
589 | ||
b8aa1682 JD |
590 | control_sock = relay_init_sock(control_uri); |
591 | if (!control_sock) { | |
095a4ae5 | 592 | goto error_sock_control; |
b8aa1682 JD |
593 | } |
594 | ||
595 | data_sock = relay_init_sock(data_uri); | |
596 | if (!data_sock) { | |
095a4ae5 | 597 | goto error_sock_relay; |
b8aa1682 JD |
598 | } |
599 | ||
600 | /* | |
601 | * Pass 3 as size here for the thread quit pipe, control and data socket. | |
602 | */ | |
603 | ret = create_thread_poll_set(&events, 3); | |
604 | if (ret < 0) { | |
605 | goto error_create_poll; | |
606 | } | |
607 | ||
608 | /* Add the control socket */ | |
609 | ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP); | |
610 | if (ret < 0) { | |
611 | goto error_poll_add; | |
612 | } | |
613 | ||
614 | /* Add the data socket */ | |
615 | ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP); | |
616 | if (ret < 0) { | |
617 | goto error_poll_add; | |
618 | } | |
619 | ||
02a6bb53 MD |
620 | lttng_relay_notify_ready(); |
621 | ||
4e5b8b82 MD |
622 | if (testpoint(relayd_thread_listener)) { |
623 | goto error_testpoint; | |
624 | } | |
625 | ||
b8aa1682 | 626 | while (1) { |
f385ae0a MD |
627 | health_code_update(); |
628 | ||
b8aa1682 JD |
629 | DBG("Listener accepting connections"); |
630 | ||
b8aa1682 | 631 | restart: |
f385ae0a | 632 | health_poll_entry(); |
b8aa1682 | 633 | ret = lttng_poll_wait(&events, -1); |
f385ae0a | 634 | health_poll_exit(); |
b8aa1682 JD |
635 | if (ret < 0) { |
636 | /* | |
637 | * Restart interrupted system call. | |
638 | */ | |
639 | if (errno == EINTR) { | |
640 | goto restart; | |
641 | } | |
642 | goto error; | |
643 | } | |
644 | ||
0d9c5d77 DG |
645 | nb_fd = ret; |
646 | ||
b8aa1682 JD |
647 | DBG("Relay new connection received"); |
648 | for (i = 0; i < nb_fd; i++) { | |
f385ae0a MD |
649 | health_code_update(); |
650 | ||
b8aa1682 JD |
651 | /* Fetch once the poll data */ |
652 | revents = LTTNG_POLL_GETEV(&events, i); | |
653 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
654 | ||
655 | /* Thread quit pipe has been closed. Killing thread. */ | |
656 | ret = check_thread_quit_pipe(pollfd, revents); | |
657 | if (ret) { | |
095a4ae5 MD |
658 | err = 0; |
659 | goto exit; | |
b8aa1682 JD |
660 | } |
661 | ||
662 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
663 | ERR("socket poll error"); | |
664 | goto error; | |
665 | } else if (revents & LPOLLIN) { | |
4b7f17b2 MD |
666 | /* |
667 | * Get allocated in this thread, | |
668 | * enqueued to a global queue, dequeued | |
669 | * and freed in the worker thread. | |
670 | */ | |
671 | struct relay_command *relay_cmd; | |
672 | struct lttcomm_sock *newsock; | |
b8aa1682 JD |
673 | |
674 | relay_cmd = zmalloc(sizeof(struct relay_command)); | |
675 | if (relay_cmd == NULL) { | |
676 | PERROR("relay command zmalloc"); | |
677 | goto error; | |
678 | } | |
679 | ||
680 | if (pollfd == data_sock->fd) { | |
681 | newsock = data_sock->ops->accept(data_sock); | |
4b7f17b2 | 682 | if (!newsock) { |
b8aa1682 | 683 | PERROR("accepting data sock"); |
4b7f17b2 | 684 | free(relay_cmd); |
b8aa1682 JD |
685 | goto error; |
686 | } | |
687 | relay_cmd->type = RELAY_DATA; | |
688 | DBG("Relay data connection accepted, socket %d", newsock->fd); | |
4b7f17b2 MD |
689 | } else { |
690 | assert(pollfd == control_sock->fd); | |
b8aa1682 | 691 | newsock = control_sock->ops->accept(control_sock); |
4b7f17b2 | 692 | if (!newsock) { |
b8aa1682 | 693 | PERROR("accepting control sock"); |
4b7f17b2 | 694 | free(relay_cmd); |
b8aa1682 JD |
695 | goto error; |
696 | } | |
697 | relay_cmd->type = RELAY_CONTROL; | |
698 | DBG("Relay control connection accepted, socket %d", newsock->fd); | |
699 | } | |
700 | ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, | |
701 | &val, sizeof(int)); | |
702 | if (ret < 0) { | |
703 | PERROR("setsockopt inet"); | |
4b7f17b2 MD |
704 | lttcomm_destroy_sock(newsock); |
705 | free(relay_cmd); | |
b8aa1682 JD |
706 | goto error; |
707 | } | |
708 | relay_cmd->sock = newsock; | |
709 | /* | |
710 | * Lock free enqueue the request. | |
711 | */ | |
712 | cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node); | |
713 | ||
714 | /* | |
715 | * Wake the dispatch queue futex. Implicit memory | |
716 | * barrier with the exchange in cds_wfq_enqueue. | |
717 | */ | |
718 | futex_nto1_wake(&relay_cmd_queue.futex); | |
719 | } | |
720 | } | |
721 | } | |
722 | ||
095a4ae5 | 723 | exit: |
b8aa1682 JD |
724 | error: |
725 | error_poll_add: | |
4e5b8b82 | 726 | error_testpoint: |
b8aa1682 JD |
727 | lttng_poll_clean(&events); |
728 | error_create_poll: | |
095a4ae5 MD |
729 | if (data_sock->fd >= 0) { |
730 | ret = data_sock->ops->close(data_sock); | |
b8aa1682 JD |
731 | if (ret) { |
732 | PERROR("close"); | |
733 | } | |
b8aa1682 | 734 | } |
095a4ae5 MD |
735 | lttcomm_destroy_sock(data_sock); |
736 | error_sock_relay: | |
737 | if (control_sock->fd >= 0) { | |
738 | ret = control_sock->ops->close(control_sock); | |
b8aa1682 JD |
739 | if (ret) { |
740 | PERROR("close"); | |
741 | } | |
b8aa1682 | 742 | } |
095a4ae5 MD |
743 | lttcomm_destroy_sock(control_sock); |
744 | error_sock_control: | |
745 | if (err) { | |
f385ae0a MD |
746 | health_error(); |
747 | ERR("Health error occurred in %s", __func__); | |
095a4ae5 | 748 | } |
55706a7d | 749 | health_unregister(health_relayd); |
b8aa1682 JD |
750 | DBG("Relay listener thread cleanup complete"); |
751 | stop_threads(); | |
b8aa1682 JD |
752 | return NULL; |
753 | } | |
754 | ||
755 | /* | |
756 | * This thread manages the dispatching of the requests to worker threads | |
757 | */ | |
758 | static | |
759 | void *relay_thread_dispatcher(void *data) | |
760 | { | |
6cd525e8 MD |
761 | int err = -1; |
762 | ssize_t ret; | |
b8aa1682 JD |
763 | struct cds_wfq_node *node; |
764 | struct relay_command *relay_cmd = NULL; | |
765 | ||
766 | DBG("[thread] Relay dispatcher started"); | |
767 | ||
55706a7d MD |
768 | health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER); |
769 | ||
4e5b8b82 MD |
770 | if (testpoint(relayd_thread_dispatcher)) { |
771 | goto error_testpoint; | |
772 | } | |
773 | ||
f385ae0a MD |
774 | health_code_update(); |
775 | ||
26c9d55e | 776 | while (!CMM_LOAD_SHARED(dispatch_thread_exit)) { |
f385ae0a MD |
777 | health_code_update(); |
778 | ||
b8aa1682 JD |
779 | /* Atomically prepare the queue futex */ |
780 | futex_nto1_prepare(&relay_cmd_queue.futex); | |
781 | ||
782 | do { | |
f385ae0a MD |
783 | health_code_update(); |
784 | ||
b8aa1682 JD |
785 | /* Dequeue commands */ |
786 | node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue); | |
787 | if (node == NULL) { | |
788 | DBG("Woken up but nothing in the relay command queue"); | |
789 | /* Continue thread execution */ | |
790 | break; | |
791 | } | |
792 | ||
793 | relay_cmd = caa_container_of(node, struct relay_command, node); | |
794 | DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd); | |
795 | ||
796 | /* | |
797 | * Inform worker thread of the new request. This | |
798 | * call is blocking so we can be assured that the data will be read | |
799 | * at some point in time or wait to the end of the world :) | |
800 | */ | |
6cd525e8 MD |
801 | ret = lttng_write(relay_cmd_pipe[1], relay_cmd, |
802 | sizeof(struct relay_command)); | |
b8aa1682 | 803 | free(relay_cmd); |
6cd525e8 | 804 | if (ret < sizeof(struct relay_command)) { |
b8aa1682 JD |
805 | PERROR("write cmd pipe"); |
806 | goto error; | |
807 | } | |
808 | } while (node != NULL); | |
809 | ||
810 | /* Futex wait on queue. Blocking call on futex() */ | |
f385ae0a | 811 | health_poll_entry(); |
b8aa1682 | 812 | futex_nto1_wait(&relay_cmd_queue.futex); |
f385ae0a | 813 | health_poll_exit(); |
b8aa1682 JD |
814 | } |
815 | ||
f385ae0a MD |
816 | /* Normal exit, no error */ |
817 | err = 0; | |
818 | ||
b8aa1682 | 819 | error: |
4e5b8b82 | 820 | error_testpoint: |
f385ae0a MD |
821 | if (err) { |
822 | health_error(); | |
823 | ERR("Health error occurred in %s", __func__); | |
824 | } | |
55706a7d | 825 | health_unregister(health_relayd); |
b8aa1682 JD |
826 | DBG("Dispatch thread dying"); |
827 | stop_threads(); | |
828 | return NULL; | |
829 | } | |
830 | ||
de91f48a DG |
831 | /* |
832 | * Get stream from stream id. | |
833 | * Need to be called with RCU read-side lock held. | |
834 | */ | |
d3e2ba59 | 835 | struct relay_stream *relay_stream_find_by_id(uint64_t stream_id) |
de91f48a DG |
836 | { |
837 | struct lttng_ht_node_ulong *node; | |
838 | struct lttng_ht_iter iter; | |
839 | struct relay_stream *ret; | |
840 | ||
d3e2ba59 | 841 | lttng_ht_lookup(relay_streams_ht, |
de91f48a DG |
842 | (void *)((unsigned long) stream_id), |
843 | &iter); | |
844 | node = lttng_ht_iter_get_node_ulong(&iter); | |
845 | if (node == NULL) { | |
846 | DBG("Relay stream %" PRIu64 " not found", stream_id); | |
847 | ret = NULL; | |
848 | goto end; | |
849 | } | |
850 | ||
851 | ret = caa_container_of(node, struct relay_stream, stream_n); | |
852 | ||
853 | end: | |
854 | return ret; | |
855 | } | |
856 | ||
9d1bbf21 MD |
857 | static |
858 | void deferred_free_stream(struct rcu_head *head) | |
859 | { | |
860 | struct relay_stream *stream = | |
861 | caa_container_of(head, struct relay_stream, rcu_node); | |
d3e2ba59 | 862 | |
0f907de1 JD |
863 | free(stream->path_name); |
864 | free(stream->channel_name); | |
9d1bbf21 MD |
865 | free(stream); |
866 | } | |
867 | ||
d3e2ba59 JD |
868 | static |
869 | void deferred_free_session(struct rcu_head *head) | |
870 | { | |
871 | struct relay_session *session = | |
872 | caa_container_of(head, struct relay_session, rcu_node); | |
873 | free(session); | |
874 | } | |
875 | ||
4c80d9a2 DG |
876 | /* |
877 | * Close a given stream. The stream is freed using a call RCU. | |
878 | * | |
879 | * RCU read side lock MUST be acquired. If NO close_stream_check() was called | |
880 | * BEFORE the stream lock MUST be acquired. | |
881 | */ | |
c07d8a78 | 882 | static void destroy_stream(struct relay_stream *stream) |
94d49140 JD |
883 | { |
884 | int delret; | |
885 | struct relay_viewer_stream *vstream; | |
886 | struct lttng_ht_iter iter; | |
887 | ||
888 | assert(stream); | |
94d49140 JD |
889 | |
890 | delret = close(stream->fd); | |
891 | if (delret < 0) { | |
892 | PERROR("close stream"); | |
893 | } | |
894 | ||
895 | if (stream->index_fd >= 0) { | |
896 | delret = close(stream->index_fd); | |
897 | if (delret < 0) { | |
898 | PERROR("close stream index_fd"); | |
899 | } | |
900 | } | |
901 | ||
991adae2 | 902 | vstream = viewer_stream_find_by_id(stream->stream_handle); |
94d49140 JD |
903 | if (vstream) { |
904 | /* | |
905 | * Set the last good value into the viewer stream. This is done | |
906 | * right before the stream gets deleted from the hash table. The | |
907 | * lookup failure on the live thread side of a stream indicates | |
908 | * that the viewer stream index received value should be used. | |
909 | */ | |
cef0f7d5 | 910 | pthread_mutex_lock(&stream->viewer_stream_rotation_lock); |
94d49140 | 911 | vstream->total_index_received = stream->total_index_received; |
a020f610 | 912 | vstream->tracefile_count_last = stream->tracefile_count_current; |
cef0f7d5 JD |
913 | vstream->close_write_flag = 1; |
914 | pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); | |
94d49140 JD |
915 | } |
916 | ||
efc8a87f DG |
917 | /* Cleanup index of that stream. */ |
918 | relay_index_destroy_by_stream_id(stream->stream_handle); | |
919 | ||
94d49140 JD |
920 | iter.iter.node = &stream->stream_n.node; |
921 | delret = lttng_ht_del(relay_streams_ht, &iter); | |
922 | assert(!delret); | |
923 | iter.iter.node = &stream->ctf_trace_node.node; | |
c07d8a78 | 924 | delret = lttng_ht_del(stream->ctf_traces_ht, &iter); |
94d49140 | 925 | assert(!delret); |
157df586 JD |
926 | |
927 | if (stream->ctf_trace) { | |
928 | ctf_trace_try_destroy(stream->ctf_trace); | |
929 | } | |
930 | ||
94d49140 JD |
931 | call_rcu(&stream->rcu_node, deferred_free_stream); |
932 | DBG("Closed tracefile %d from close stream", stream->fd); | |
933 | } | |
934 | ||
b8aa1682 JD |
935 | /* |
936 | * relay_delete_session: Free all memory associated with a session and | |
937 | * close all the FDs | |
938 | */ | |
939 | static | |
d3e2ba59 JD |
940 | void relay_delete_session(struct relay_command *cmd, |
941 | struct lttng_ht *sessions_ht) | |
b8aa1682 JD |
942 | { |
943 | struct lttng_ht_iter iter; | |
944 | struct lttng_ht_node_ulong *node; | |
945 | struct relay_stream *stream; | |
946 | int ret; | |
947 | ||
095a4ae5 | 948 | if (!cmd->session) { |
b8aa1682 | 949 | return; |
095a4ae5 | 950 | } |
b8aa1682 | 951 | |
77c7c900 | 952 | DBG("Relay deleting session %" PRIu64, cmd->session->id); |
5b6d8097 | 953 | |
9d1bbf21 | 954 | rcu_read_lock(); |
d3e2ba59 | 955 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) { |
b8aa1682 | 956 | node = lttng_ht_iter_get_node_ulong(&iter); |
4c80d9a2 DG |
957 | if (!node) { |
958 | continue; | |
b8aa1682 | 959 | } |
4c80d9a2 DG |
960 | stream = caa_container_of(node, struct relay_stream, stream_n); |
961 | if (stream->session == cmd->session) { | |
c07d8a78 | 962 | destroy_stream(stream); |
87b576ec JD |
963 | cmd->session->stream_count--; |
964 | assert(cmd->session->stream_count >= 0); | |
4c80d9a2 | 965 | } |
b8aa1682 | 966 | } |
4c80d9a2 DG |
967 | |
968 | /* Make this session not visible anymore. */ | |
d3e2ba59 JD |
969 | iter.iter.node = &cmd->session->session_n.node; |
970 | ret = lttng_ht_del(sessions_ht, &iter); | |
971 | assert(!ret); | |
4c80d9a2 | 972 | call_rcu(&cmd->session->rcu_node, deferred_free_session); |
9d1bbf21 | 973 | rcu_read_unlock(); |
b8aa1682 JD |
974 | } |
975 | ||
1c20f0e2 JD |
976 | /* |
977 | * Copy index data from the control port to a given index object. | |
978 | */ | |
979 | static void copy_index_control_data(struct relay_index *index, | |
980 | struct lttcomm_relayd_index *data) | |
981 | { | |
982 | assert(index); | |
983 | assert(data); | |
984 | ||
985 | /* | |
986 | * The index on disk is encoded in big endian, so we don't need to convert | |
987 | * the data received on the network. The data_offset value is NEVER | |
988 | * modified here and is updated by the data thread. | |
989 | */ | |
990 | index->index_data.packet_size = data->packet_size; | |
991 | index->index_data.content_size = data->content_size; | |
992 | index->index_data.timestamp_begin = data->timestamp_begin; | |
993 | index->index_data.timestamp_end = data->timestamp_end; | |
994 | index->index_data.events_discarded = data->events_discarded; | |
995 | index->index_data.stream_id = data->stream_id; | |
996 | } | |
997 | ||
c5b6f4f0 DG |
998 | /* |
999 | * Handle the RELAYD_CREATE_SESSION command. | |
1000 | * | |
1001 | * On success, send back the session id or else return a negative value. | |
1002 | */ | |
1003 | static | |
1004 | int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 JD |
1005 | struct relay_command *cmd, |
1006 | struct lttng_ht *sessions_ht) | |
c5b6f4f0 DG |
1007 | { |
1008 | int ret = 0, send_ret; | |
1009 | struct relay_session *session; | |
1010 | struct lttcomm_relayd_status_session reply; | |
1011 | ||
1012 | assert(recv_hdr); | |
1013 | assert(cmd); | |
1014 | ||
1015 | memset(&reply, 0, sizeof(reply)); | |
1016 | ||
1017 | session = zmalloc(sizeof(struct relay_session)); | |
1018 | if (session == NULL) { | |
1019 | PERROR("relay session zmalloc"); | |
1020 | ret = -1; | |
1021 | goto error; | |
1022 | } | |
1023 | ||
1024 | session->id = ++last_relay_session_id; | |
1025 | session->sock = cmd->sock; | |
7d2f7452 DG |
1026 | session->minor = cmd->minor; |
1027 | session->major = cmd->major; | |
091c3237 | 1028 | pthread_mutex_init(&session->viewer_ready_lock, NULL); |
c5b6f4f0 DG |
1029 | cmd->session = session; |
1030 | ||
1031 | reply.session_id = htobe64(session->id); | |
1032 | ||
d3e2ba59 | 1033 | switch (cmd->minor) { |
b0dd7a75 JD |
1034 | case 1: |
1035 | case 2: | |
1036 | case 3: | |
1037 | break; | |
d3e2ba59 JD |
1038 | case 4: /* LTTng sessiond 2.4 */ |
1039 | default: | |
1040 | ret = cmd_create_session_2_4(cmd, session); | |
1041 | break; | |
1042 | } | |
1043 | ||
1044 | lttng_ht_node_init_ulong(&session->session_n, | |
1045 | (unsigned long) session->id); | |
1046 | lttng_ht_add_unique_ulong(sessions_ht, | |
1047 | &session->session_n); | |
1048 | ||
c5b6f4f0 DG |
1049 | DBG("Created session %" PRIu64, session->id); |
1050 | ||
1051 | error: | |
1052 | if (ret < 0) { | |
1053 | reply.ret_code = htobe32(LTTNG_ERR_FATAL); | |
1054 | } else { | |
1055 | reply.ret_code = htobe32(LTTNG_OK); | |
1056 | } | |
1057 | ||
1058 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1059 | if (send_ret < 0) { | |
1060 | ERR("Relayd sending session id"); | |
4169f5ad | 1061 | ret = send_ret; |
c5b6f4f0 DG |
1062 | } |
1063 | ||
1064 | return ret; | |
1065 | } | |
1066 | ||
814fcae4 JD |
1067 | /* |
1068 | * When we have received all the streams and the metadata for a channel, | |
1069 | * we make them visible to the viewer threads. | |
1070 | */ | |
1071 | static | |
1072 | void set_viewer_ready_flag(struct relay_command *cmd) | |
1073 | { | |
1074 | struct relay_stream_recv_handle *node, *tmp_node; | |
1075 | ||
091c3237 JD |
1076 | pthread_mutex_lock(&cmd->session->viewer_ready_lock); |
1077 | ||
814fcae4 JD |
1078 | cds_list_for_each_entry_safe(node, tmp_node, &cmd->recv_head, node) { |
1079 | struct relay_stream *stream; | |
1080 | ||
1081 | rcu_read_lock(); | |
1082 | stream = relay_stream_find_by_id(node->id); | |
1083 | if (!stream) { | |
1084 | /* | |
1085 | * Stream is most probably being cleaned up by the data thread thus | |
1086 | * simply continue to the next one. | |
1087 | */ | |
f188956b | 1088 | rcu_read_unlock(); |
814fcae4 JD |
1089 | continue; |
1090 | } | |
1091 | ||
814fcae4 JD |
1092 | stream->viewer_ready = 1; |
1093 | rcu_read_unlock(); | |
1094 | ||
1095 | /* Clean stream handle node. */ | |
1096 | cds_list_del(&node->node); | |
1097 | free(node); | |
1098 | } | |
1099 | ||
091c3237 | 1100 | pthread_mutex_unlock(&cmd->session->viewer_ready_lock); |
814fcae4 JD |
1101 | return; |
1102 | } | |
1103 | ||
1104 | /* | |
1105 | * Add a recv handle node to the connection recv list with the given stream | |
1106 | * handle. A new node is allocated thus must be freed when the node is deleted | |
1107 | * from the list. | |
1108 | */ | |
1109 | static void queue_stream_handle(uint64_t handle, struct relay_command *cmd) | |
1110 | { | |
1111 | struct relay_stream_recv_handle *node; | |
1112 | ||
1113 | assert(cmd); | |
1114 | ||
1115 | node = zmalloc(sizeof(*node)); | |
1116 | if (!node) { | |
1117 | PERROR("zmalloc queue stream handle"); | |
1118 | return; | |
1119 | } | |
1120 | ||
1121 | node->id = handle; | |
1122 | cds_list_add(&node->node, &cmd->recv_head); | |
1123 | } | |
1124 | ||
b8aa1682 JD |
1125 | /* |
1126 | * relay_add_stream: allocate a new stream for a session | |
1127 | */ | |
1128 | static | |
1129 | int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1130 | struct relay_command *cmd, struct lttng_ht *sessions_ht) |
b8aa1682 JD |
1131 | { |
1132 | struct relay_session *session = cmd->session; | |
b8aa1682 JD |
1133 | struct relay_stream *stream = NULL; |
1134 | struct lttcomm_relayd_status_stream reply; | |
b8aa1682 JD |
1135 | int ret, send_ret; |
1136 | ||
c5b6f4f0 | 1137 | if (!session || cmd->version_check_done == 0) { |
b8aa1682 JD |
1138 | ERR("Trying to add a stream before version check"); |
1139 | ret = -1; | |
1140 | goto end_no_session; | |
1141 | } | |
1142 | ||
b8aa1682 JD |
1143 | stream = zmalloc(sizeof(struct relay_stream)); |
1144 | if (stream == NULL) { | |
1145 | PERROR("relay stream zmalloc"); | |
1146 | ret = -1; | |
1147 | goto end_no_session; | |
1148 | } | |
1149 | ||
0f907de1 JD |
1150 | switch (cmd->minor) { |
1151 | case 1: /* LTTng sessiond 2.1 */ | |
1152 | ret = cmd_recv_stream_2_1(cmd, stream); | |
1153 | break; | |
1154 | case 2: /* LTTng sessiond 2.2 */ | |
1155 | default: | |
1156 | ret = cmd_recv_stream_2_2(cmd, stream); | |
1157 | break; | |
1158 | } | |
1159 | if (ret < 0) { | |
1160 | goto err_free_stream; | |
1161 | } | |
1162 | ||
9d1bbf21 | 1163 | rcu_read_lock(); |
b8aa1682 | 1164 | stream->stream_handle = ++last_relay_stream_id; |
173af62f | 1165 | stream->prev_seq = -1ULL; |
b8aa1682 | 1166 | stream->session = session; |
1c20f0e2 | 1167 | stream->index_fd = -1; |
d3e2ba59 JD |
1168 | stream->read_index_fd = -1; |
1169 | stream->ctf_trace = NULL; | |
1170 | pthread_mutex_init(&stream->lock, NULL); | |
b8aa1682 | 1171 | |
0f907de1 | 1172 | ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG); |
b8aa1682 | 1173 | if (ret < 0) { |
b8aa1682 JD |
1174 | ERR("relay creating output directory"); |
1175 | goto end; | |
1176 | } | |
1177 | ||
be96a7d1 DG |
1178 | /* |
1179 | * No need to use run_as API here because whatever we receives, the relayd | |
1180 | * uses its own credentials for the stream files. | |
1181 | */ | |
0f907de1 | 1182 | ret = utils_create_stream_file(stream->path_name, stream->channel_name, |
1c20f0e2 | 1183 | stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL); |
b8aa1682 | 1184 | if (ret < 0) { |
0f907de1 | 1185 | ERR("Create output file"); |
b8aa1682 JD |
1186 | goto end; |
1187 | } | |
b8aa1682 | 1188 | stream->fd = ret; |
0f907de1 JD |
1189 | if (stream->tracefile_size) { |
1190 | DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name); | |
1191 | } else { | |
1192 | DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name); | |
1193 | } | |
b8aa1682 | 1194 | |
d3e2ba59 JD |
1195 | if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) { |
1196 | stream->metadata_flag = 1; | |
1197 | /* | |
1198 | * When we receive a new metadata stream, we create a new | |
1199 | * ctf_trace and we assign this ctf_trace to all streams with | |
1200 | * the same path. | |
1201 | * | |
1202 | * If later on we receive a new stream for the same ctf_trace, | |
1203 | * we copy the information from the first hit in the HT to the | |
1204 | * new stream. | |
1205 | */ | |
1206 | stream->ctf_trace = ctf_trace_create(); | |
1207 | if (!stream->ctf_trace) { | |
1208 | ret = -1; | |
1209 | goto end; | |
1210 | } | |
1211 | stream->ctf_trace->refcount++; | |
1212 | stream->ctf_trace->metadata_stream = stream; | |
1213 | } | |
1214 | ctf_trace_assign(cmd->ctf_traces_ht, stream); | |
c07d8a78 | 1215 | stream->ctf_traces_ht = cmd->ctf_traces_ht; |
d3e2ba59 | 1216 | |
814fcae4 JD |
1217 | /* |
1218 | * Add the stream handle in the recv list of the connection. Once the end | |
1219 | * stream message is received, this list is emptied and streams are set | |
1220 | * with the viewer ready flag. | |
1221 | */ | |
e37dd78f | 1222 | queue_stream_handle(stream->stream_handle, cmd); |
814fcae4 | 1223 | |
b8aa1682 JD |
1224 | lttng_ht_node_init_ulong(&stream->stream_n, |
1225 | (unsigned long) stream->stream_handle); | |
d3e2ba59 | 1226 | lttng_ht_add_unique_ulong(relay_streams_ht, |
b8aa1682 JD |
1227 | &stream->stream_n); |
1228 | ||
d3e2ba59 JD |
1229 | lttng_ht_node_init_str(&stream->ctf_trace_node, stream->path_name); |
1230 | lttng_ht_add_str(cmd->ctf_traces_ht, &stream->ctf_trace_node); | |
87b576ec | 1231 | session->stream_count++; |
d3e2ba59 | 1232 | |
1c20f0e2 JD |
1233 | DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name, |
1234 | stream->stream_handle); | |
b8aa1682 JD |
1235 | |
1236 | end: | |
5af40280 | 1237 | reply.handle = htobe64(stream->stream_handle); |
b8aa1682 JD |
1238 | /* send the session id to the client or a negative return code on error */ |
1239 | if (ret < 0) { | |
f73fabfd | 1240 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
5af40280 CB |
1241 | /* stream was not properly added to the ht, so free it */ |
1242 | free(stream); | |
b8aa1682 | 1243 | } else { |
f73fabfd | 1244 | reply.ret_code = htobe32(LTTNG_OK); |
b8aa1682 | 1245 | } |
5af40280 | 1246 | |
b8aa1682 JD |
1247 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, |
1248 | sizeof(struct lttcomm_relayd_status_stream), 0); | |
1249 | if (send_ret < 0) { | |
1250 | ERR("Relay sending stream id"); | |
4169f5ad | 1251 | ret = send_ret; |
b8aa1682 | 1252 | } |
9d1bbf21 | 1253 | rcu_read_unlock(); |
b8aa1682 JD |
1254 | |
1255 | end_no_session: | |
1256 | return ret; | |
0f907de1 JD |
1257 | |
1258 | err_free_stream: | |
1259 | free(stream->path_name); | |
1260 | free(stream->channel_name); | |
1261 | free(stream); | |
1262 | return ret; | |
b8aa1682 JD |
1263 | } |
1264 | ||
173af62f DG |
1265 | /* |
1266 | * relay_close_stream: close a specific stream | |
1267 | */ | |
1268 | static | |
1269 | int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, | |
92c6ca54 | 1270 | struct relay_command *cmd) |
173af62f | 1271 | { |
94d49140 | 1272 | int ret, send_ret; |
173af62f DG |
1273 | struct relay_session *session = cmd->session; |
1274 | struct lttcomm_relayd_close_stream stream_info; | |
1275 | struct lttcomm_relayd_generic_reply reply; | |
1276 | struct relay_stream *stream; | |
173af62f DG |
1277 | |
1278 | DBG("Close stream received"); | |
1279 | ||
c5b6f4f0 | 1280 | if (!session || cmd->version_check_done == 0) { |
173af62f DG |
1281 | ERR("Trying to close a stream before version check"); |
1282 | ret = -1; | |
1283 | goto end_no_session; | |
1284 | } | |
1285 | ||
1286 | ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info, | |
7c5aef62 | 1287 | sizeof(struct lttcomm_relayd_close_stream), 0); |
173af62f | 1288 | if (ret < sizeof(struct lttcomm_relayd_close_stream)) { |
a6cd2b97 DG |
1289 | if (ret == 0) { |
1290 | /* Orderly shutdown. Not necessary to print an error. */ | |
1291 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1292 | } else { | |
1293 | ERR("Relay didn't receive valid add_stream struct size : %d", ret); | |
1294 | } | |
173af62f DG |
1295 | ret = -1; |
1296 | goto end_no_session; | |
1297 | } | |
1298 | ||
1299 | rcu_read_lock(); | |
d3e2ba59 | 1300 | stream = relay_stream_find_by_id(be64toh(stream_info.stream_id)); |
173af62f DG |
1301 | if (!stream) { |
1302 | ret = -1; | |
1303 | goto end_unlock; | |
1304 | } | |
1305 | ||
8e2583a4 | 1306 | stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num); |
173af62f | 1307 | stream->close_flag = 1; |
87b576ec JD |
1308 | session->stream_count--; |
1309 | assert(session->stream_count >= 0); | |
173af62f DG |
1310 | |
1311 | if (close_stream_check(stream)) { | |
c07d8a78 | 1312 | destroy_stream(stream); |
173af62f DG |
1313 | } |
1314 | ||
1315 | end_unlock: | |
1316 | rcu_read_unlock(); | |
1317 | ||
1318 | if (ret < 0) { | |
f73fabfd | 1319 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
173af62f | 1320 | } else { |
f73fabfd | 1321 | reply.ret_code = htobe32(LTTNG_OK); |
173af62f DG |
1322 | } |
1323 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1324 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1325 | if (send_ret < 0) { | |
1326 | ERR("Relay sending stream id"); | |
4169f5ad | 1327 | ret = send_ret; |
173af62f DG |
1328 | } |
1329 | ||
1330 | end_no_session: | |
1331 | return ret; | |
1332 | } | |
1333 | ||
b8aa1682 JD |
1334 | /* |
1335 | * relay_unknown_command: send -1 if received unknown command | |
1336 | */ | |
1337 | static | |
1338 | void relay_unknown_command(struct relay_command *cmd) | |
1339 | { | |
1340 | struct lttcomm_relayd_generic_reply reply; | |
1341 | int ret; | |
1342 | ||
f73fabfd | 1343 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
b8aa1682 JD |
1344 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, |
1345 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1346 | if (ret < 0) { | |
1347 | ERR("Relay sending unknown command"); | |
1348 | } | |
1349 | } | |
1350 | ||
1351 | /* | |
1352 | * relay_start: send an acknowledgment to the client to tell if we are | |
1353 | * ready to receive data. We are ready if a session is established. | |
1354 | */ | |
1355 | static | |
1356 | int relay_start(struct lttcomm_relayd_hdr *recv_hdr, | |
1357 | struct relay_command *cmd) | |
1358 | { | |
f73fabfd | 1359 | int ret = htobe32(LTTNG_OK); |
b8aa1682 JD |
1360 | struct lttcomm_relayd_generic_reply reply; |
1361 | struct relay_session *session = cmd->session; | |
1362 | ||
1363 | if (!session) { | |
1364 | DBG("Trying to start the streaming without a session established"); | |
f73fabfd | 1365 | ret = htobe32(LTTNG_ERR_UNK); |
b8aa1682 JD |
1366 | } |
1367 | ||
1368 | reply.ret_code = ret; | |
1369 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1370 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1371 | if (ret < 0) { | |
1372 | ERR("Relay sending start ack"); | |
1373 | } | |
1374 | ||
1375 | return ret; | |
1376 | } | |
1377 | ||
1d4dfdef DG |
1378 | /* |
1379 | * Append padding to the file pointed by the file descriptor fd. | |
1380 | */ | |
1381 | static int write_padding_to_file(int fd, uint32_t size) | |
1382 | { | |
6cd525e8 | 1383 | ssize_t ret = 0; |
1d4dfdef DG |
1384 | char *zeros; |
1385 | ||
1386 | if (size == 0) { | |
1387 | goto end; | |
1388 | } | |
1389 | ||
1390 | zeros = zmalloc(size); | |
1391 | if (zeros == NULL) { | |
1392 | PERROR("zmalloc zeros for padding"); | |
1393 | ret = -1; | |
1394 | goto end; | |
1395 | } | |
1396 | ||
6cd525e8 MD |
1397 | ret = lttng_write(fd, zeros, size); |
1398 | if (ret < size) { | |
1d4dfdef DG |
1399 | PERROR("write padding to file"); |
1400 | } | |
1401 | ||
e986c7a1 DG |
1402 | free(zeros); |
1403 | ||
1d4dfdef DG |
1404 | end: |
1405 | return ret; | |
1406 | } | |
1407 | ||
b8aa1682 JD |
1408 | /* |
1409 | * relay_recv_metadata: receive the metada for the session. | |
1410 | */ | |
1411 | static | |
1412 | int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1413 | struct relay_command *cmd) |
b8aa1682 | 1414 | { |
f73fabfd | 1415 | int ret = htobe32(LTTNG_OK); |
6cd525e8 | 1416 | ssize_t size_ret; |
b8aa1682 JD |
1417 | struct relay_session *session = cmd->session; |
1418 | struct lttcomm_relayd_metadata_payload *metadata_struct; | |
1419 | struct relay_stream *metadata_stream; | |
1420 | uint64_t data_size, payload_size; | |
1421 | ||
1422 | if (!session) { | |
1423 | ERR("Metadata sent before version check"); | |
1424 | ret = -1; | |
1425 | goto end; | |
1426 | } | |
1427 | ||
f6416125 MD |
1428 | data_size = payload_size = be64toh(recv_hdr->data_size); |
1429 | if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) { | |
1430 | ERR("Incorrect data size"); | |
1431 | ret = -1; | |
1432 | goto end; | |
1433 | } | |
1434 | payload_size -= sizeof(struct lttcomm_relayd_metadata_payload); | |
1435 | ||
b8aa1682 | 1436 | if (data_buffer_size < data_size) { |
d7b3776f | 1437 | /* In case the realloc fails, we can free the memory */ |
c617c0c6 MD |
1438 | char *tmp_data_ptr; |
1439 | ||
1440 | tmp_data_ptr = realloc(data_buffer, data_size); | |
1441 | if (!tmp_data_ptr) { | |
b8aa1682 | 1442 | ERR("Allocating data buffer"); |
c617c0c6 | 1443 | free(data_buffer); |
b8aa1682 JD |
1444 | ret = -1; |
1445 | goto end; | |
1446 | } | |
c617c0c6 | 1447 | data_buffer = tmp_data_ptr; |
b8aa1682 JD |
1448 | data_buffer_size = data_size; |
1449 | } | |
1450 | memset(data_buffer, 0, data_size); | |
77c7c900 | 1451 | DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size); |
7c5aef62 | 1452 | ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0); |
b8aa1682 | 1453 | if (ret < 0 || ret != data_size) { |
a6cd2b97 DG |
1454 | if (ret == 0) { |
1455 | /* Orderly shutdown. Not necessary to print an error. */ | |
1456 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1457 | } else { | |
1458 | ERR("Relay didn't receive the whole metadata"); | |
1459 | } | |
b8aa1682 | 1460 | ret = -1; |
b8aa1682 JD |
1461 | goto end; |
1462 | } | |
1463 | metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer; | |
9d1bbf21 MD |
1464 | |
1465 | rcu_read_lock(); | |
d3e2ba59 JD |
1466 | metadata_stream = relay_stream_find_by_id( |
1467 | be64toh(metadata_struct->stream_id)); | |
b8aa1682 JD |
1468 | if (!metadata_stream) { |
1469 | ret = -1; | |
9d1bbf21 | 1470 | goto end_unlock; |
b8aa1682 JD |
1471 | } |
1472 | ||
6cd525e8 MD |
1473 | size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload, |
1474 | payload_size); | |
1475 | if (size_ret < payload_size) { | |
b8aa1682 JD |
1476 | ERR("Relay error writing metadata on file"); |
1477 | ret = -1; | |
9d1bbf21 | 1478 | goto end_unlock; |
b8aa1682 | 1479 | } |
1d4dfdef DG |
1480 | |
1481 | ret = write_padding_to_file(metadata_stream->fd, | |
1482 | be32toh(metadata_struct->padding_size)); | |
1483 | if (ret < 0) { | |
1484 | goto end_unlock; | |
1485 | } | |
d3e2ba59 JD |
1486 | metadata_stream->ctf_trace->metadata_received += |
1487 | payload_size + be32toh(metadata_struct->padding_size); | |
1d4dfdef | 1488 | |
b8aa1682 JD |
1489 | DBG2("Relay metadata written"); |
1490 | ||
9d1bbf21 | 1491 | end_unlock: |
6e3c5836 | 1492 | rcu_read_unlock(); |
b8aa1682 JD |
1493 | end: |
1494 | return ret; | |
1495 | } | |
1496 | ||
1497 | /* | |
1498 | * relay_send_version: send relayd version number | |
1499 | */ | |
1500 | static | |
1501 | int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1502 | struct relay_command *cmd, struct lttng_ht *sessions_ht) |
b8aa1682 | 1503 | { |
7f51dcba | 1504 | int ret; |
092b6259 | 1505 | struct lttcomm_relayd_version reply, msg; |
b8aa1682 | 1506 | |
c5b6f4f0 DG |
1507 | assert(cmd); |
1508 | ||
1509 | cmd->version_check_done = 1; | |
b8aa1682 | 1510 | |
092b6259 | 1511 | /* Get version from the other side. */ |
7c5aef62 | 1512 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); |
092b6259 | 1513 | if (ret < 0 || ret != sizeof(msg)) { |
a6cd2b97 DG |
1514 | if (ret == 0) { |
1515 | /* Orderly shutdown. Not necessary to print an error. */ | |
1516 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1517 | } else { | |
1518 | ERR("Relay failed to receive the version values."); | |
1519 | } | |
092b6259 | 1520 | ret = -1; |
092b6259 DG |
1521 | goto end; |
1522 | } | |
1523 | ||
d83a952c MD |
1524 | reply.major = RELAYD_VERSION_COMM_MAJOR; |
1525 | reply.minor = RELAYD_VERSION_COMM_MINOR; | |
d4519fa3 JD |
1526 | |
1527 | /* Major versions must be the same */ | |
1528 | if (reply.major != be32toh(msg.major)) { | |
6151a90f JD |
1529 | DBG("Incompatible major versions (%u vs %u), deleting session", |
1530 | reply.major, be32toh(msg.major)); | |
d3e2ba59 | 1531 | relay_delete_session(cmd, sessions_ht); |
d4519fa3 JD |
1532 | ret = 0; |
1533 | goto end; | |
1534 | } | |
1535 | ||
0f907de1 JD |
1536 | cmd->major = reply.major; |
1537 | /* We adapt to the lowest compatible version */ | |
1538 | if (reply.minor <= be32toh(msg.minor)) { | |
1539 | cmd->minor = reply.minor; | |
1540 | } else { | |
1541 | cmd->minor = be32toh(msg.minor); | |
1542 | } | |
1543 | ||
6151a90f JD |
1544 | reply.major = htobe32(reply.major); |
1545 | reply.minor = htobe32(reply.minor); | |
1546 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1547 | sizeof(struct lttcomm_relayd_version), 0); | |
1548 | if (ret < 0) { | |
1549 | ERR("Relay sending version"); | |
1550 | } | |
1551 | ||
0f907de1 JD |
1552 | DBG("Version check done using protocol %u.%u", cmd->major, |
1553 | cmd->minor); | |
b8aa1682 JD |
1554 | |
1555 | end: | |
1556 | return ret; | |
1557 | } | |
1558 | ||
c8f59ee5 | 1559 | /* |
6d805429 | 1560 | * Check for data pending for a given stream id from the session daemon. |
c8f59ee5 DG |
1561 | */ |
1562 | static | |
6d805429 | 1563 | int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, |
d3e2ba59 | 1564 | struct relay_command *cmd) |
c8f59ee5 DG |
1565 | { |
1566 | struct relay_session *session = cmd->session; | |
6d805429 | 1567 | struct lttcomm_relayd_data_pending msg; |
c8f59ee5 DG |
1568 | struct lttcomm_relayd_generic_reply reply; |
1569 | struct relay_stream *stream; | |
1570 | int ret; | |
c8f59ee5 DG |
1571 | uint64_t last_net_seq_num, stream_id; |
1572 | ||
6d805429 | 1573 | DBG("Data pending command received"); |
c8f59ee5 | 1574 | |
c5b6f4f0 | 1575 | if (!session || cmd->version_check_done == 0) { |
c8f59ee5 DG |
1576 | ERR("Trying to check for data before version check"); |
1577 | ret = -1; | |
1578 | goto end_no_session; | |
1579 | } | |
1580 | ||
7c5aef62 | 1581 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); |
c8f59ee5 | 1582 | if (ret < sizeof(msg)) { |
a6cd2b97 DG |
1583 | if (ret == 0) { |
1584 | /* Orderly shutdown. Not necessary to print an error. */ | |
1585 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1586 | } else { | |
1587 | ERR("Relay didn't receive valid data_pending struct size : %d", | |
1588 | ret); | |
1589 | } | |
c8f59ee5 DG |
1590 | ret = -1; |
1591 | goto end_no_session; | |
1592 | } | |
1593 | ||
1594 | stream_id = be64toh(msg.stream_id); | |
1595 | last_net_seq_num = be64toh(msg.last_net_seq_num); | |
1596 | ||
1597 | rcu_read_lock(); | |
d3e2ba59 | 1598 | stream = relay_stream_find_by_id(stream_id); |
de91f48a | 1599 | if (stream == NULL) { |
c8f59ee5 DG |
1600 | ret = -1; |
1601 | goto end_unlock; | |
1602 | } | |
1603 | ||
6d805429 | 1604 | DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64 |
c8f59ee5 DG |
1605 | " and last_seq %" PRIu64, stream_id, stream->prev_seq, |
1606 | last_net_seq_num); | |
1607 | ||
33832e64 | 1608 | /* Avoid wrapping issue */ |
39df6d9f | 1609 | if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) { |
6d805429 | 1610 | /* Data has in fact been written and is NOT pending */ |
c8f59ee5 | 1611 | ret = 0; |
6d805429 DG |
1612 | } else { |
1613 | /* Data still being streamed thus pending */ | |
1614 | ret = 1; | |
c8f59ee5 DG |
1615 | } |
1616 | ||
f7079f67 DG |
1617 | /* Pending check is now done. */ |
1618 | stream->data_pending_check_done = 1; | |
1619 | ||
c8f59ee5 DG |
1620 | end_unlock: |
1621 | rcu_read_unlock(); | |
1622 | ||
1623 | reply.ret_code = htobe32(ret); | |
1624 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1625 | if (ret < 0) { | |
6d805429 | 1626 | ERR("Relay data pending ret code failed"); |
c8f59ee5 DG |
1627 | } |
1628 | ||
1629 | end_no_session: | |
1630 | return ret; | |
1631 | } | |
1632 | ||
1633 | /* | |
1634 | * Wait for the control socket to reach a quiescent state. | |
1635 | * | |
1636 | * Note that for now, when receiving this command from the session daemon, this | |
1637 | * means that every subsequent commands or data received on the control socket | |
1638 | * has been handled. So, this is why we simply return OK here. | |
1639 | */ | |
1640 | static | |
1641 | int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1642 | struct relay_command *cmd) |
c8f59ee5 DG |
1643 | { |
1644 | int ret; | |
ad7051c0 DG |
1645 | uint64_t stream_id; |
1646 | struct relay_stream *stream; | |
1647 | struct lttng_ht_iter iter; | |
1648 | struct lttcomm_relayd_quiescent_control msg; | |
c8f59ee5 DG |
1649 | struct lttcomm_relayd_generic_reply reply; |
1650 | ||
1651 | DBG("Checking quiescent state on control socket"); | |
1652 | ||
ad7051c0 DG |
1653 | if (!cmd->session || cmd->version_check_done == 0) { |
1654 | ERR("Trying to check for data before version check"); | |
1655 | ret = -1; | |
1656 | goto end_no_session; | |
1657 | } | |
1658 | ||
1659 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1660 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1661 | if (ret == 0) { |
1662 | /* Orderly shutdown. Not necessary to print an error. */ | |
1663 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1664 | } else { | |
1665 | ERR("Relay didn't receive valid begin data_pending struct size: %d", | |
1666 | ret); | |
1667 | } | |
ad7051c0 DG |
1668 | ret = -1; |
1669 | goto end_no_session; | |
1670 | } | |
1671 | ||
1672 | stream_id = be64toh(msg.stream_id); | |
1673 | ||
1674 | rcu_read_lock(); | |
d3e2ba59 JD |
1675 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
1676 | stream_n.node) { | |
ad7051c0 DG |
1677 | if (stream->stream_handle == stream_id) { |
1678 | stream->data_pending_check_done = 1; | |
1679 | DBG("Relay quiescent control pending flag set to %" PRIu64, | |
1680 | stream_id); | |
1681 | break; | |
1682 | } | |
1683 | } | |
1684 | rcu_read_unlock(); | |
1685 | ||
c8f59ee5 DG |
1686 | reply.ret_code = htobe32(LTTNG_OK); |
1687 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1688 | if (ret < 0) { | |
6d805429 | 1689 | ERR("Relay data quiescent control ret code failed"); |
c8f59ee5 DG |
1690 | } |
1691 | ||
ad7051c0 | 1692 | end_no_session: |
c8f59ee5 DG |
1693 | return ret; |
1694 | } | |
1695 | ||
f7079f67 DG |
1696 | /* |
1697 | * Initialize a data pending command. This means that a client is about to ask | |
1698 | * for data pending for each stream he/she holds. Simply iterate over all | |
1699 | * streams of a session and set the data_pending_check_done flag. | |
1700 | * | |
1701 | * This command returns to the client a LTTNG_OK code. | |
1702 | */ | |
1703 | static | |
1704 | int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1705 | struct relay_command *cmd) |
f7079f67 DG |
1706 | { |
1707 | int ret; | |
1708 | struct lttng_ht_iter iter; | |
1709 | struct lttcomm_relayd_begin_data_pending msg; | |
1710 | struct lttcomm_relayd_generic_reply reply; | |
1711 | struct relay_stream *stream; | |
1712 | uint64_t session_id; | |
1713 | ||
1714 | assert(recv_hdr); | |
1715 | assert(cmd); | |
f7079f67 DG |
1716 | |
1717 | DBG("Init streams for data pending"); | |
1718 | ||
1719 | if (!cmd->session || cmd->version_check_done == 0) { | |
1720 | ERR("Trying to check for data before version check"); | |
1721 | ret = -1; | |
1722 | goto end_no_session; | |
1723 | } | |
1724 | ||
1725 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1726 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1727 | if (ret == 0) { |
1728 | /* Orderly shutdown. Not necessary to print an error. */ | |
1729 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1730 | } else { | |
1731 | ERR("Relay didn't receive valid begin data_pending struct size: %d", | |
1732 | ret); | |
1733 | } | |
f7079f67 DG |
1734 | ret = -1; |
1735 | goto end_no_session; | |
1736 | } | |
1737 | ||
1738 | session_id = be64toh(msg.session_id); | |
1739 | ||
1740 | /* | |
1741 | * Iterate over all streams to set the begin data pending flag. For now, the | |
1742 | * streams are indexed by stream handle so we have to iterate over all | |
1743 | * streams to find the one associated with the right session_id. | |
1744 | */ | |
1745 | rcu_read_lock(); | |
d3e2ba59 JD |
1746 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
1747 | stream_n.node) { | |
f7079f67 DG |
1748 | if (stream->session->id == session_id) { |
1749 | stream->data_pending_check_done = 0; | |
1750 | DBG("Set begin data pending flag to stream %" PRIu64, | |
1751 | stream->stream_handle); | |
1752 | } | |
1753 | } | |
1754 | rcu_read_unlock(); | |
1755 | ||
1756 | /* All good, send back reply. */ | |
1757 | reply.ret_code = htobe32(LTTNG_OK); | |
1758 | ||
1759 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1760 | if (ret < 0) { | |
1761 | ERR("Relay begin data pending send reply failed"); | |
1762 | } | |
1763 | ||
1764 | end_no_session: | |
1765 | return ret; | |
1766 | } | |
1767 | ||
1768 | /* | |
1769 | * End data pending command. This will check, for a given session id, if each | |
1770 | * stream associated with it has its data_pending_check_done flag set. If not, | |
1771 | * this means that the client lost track of the stream but the data is still | |
1772 | * being streamed on our side. In this case, we inform the client that data is | |
1773 | * inflight. | |
1774 | * | |
1775 | * Return to the client if there is data in flight or not with a ret_code. | |
1776 | */ | |
1777 | static | |
1778 | int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1779 | struct relay_command *cmd) |
f7079f67 DG |
1780 | { |
1781 | int ret; | |
1782 | struct lttng_ht_iter iter; | |
1783 | struct lttcomm_relayd_end_data_pending msg; | |
1784 | struct lttcomm_relayd_generic_reply reply; | |
1785 | struct relay_stream *stream; | |
1786 | uint64_t session_id; | |
1787 | uint32_t is_data_inflight = 0; | |
1788 | ||
1789 | assert(recv_hdr); | |
1790 | assert(cmd); | |
f7079f67 DG |
1791 | |
1792 | DBG("End data pending command"); | |
1793 | ||
1794 | if (!cmd->session || cmd->version_check_done == 0) { | |
1795 | ERR("Trying to check for data before version check"); | |
1796 | ret = -1; | |
1797 | goto end_no_session; | |
1798 | } | |
1799 | ||
1800 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1801 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1802 | if (ret == 0) { |
1803 | /* Orderly shutdown. Not necessary to print an error. */ | |
1804 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1805 | } else { | |
1806 | ERR("Relay didn't receive valid end data_pending struct size: %d", | |
1807 | ret); | |
1808 | } | |
f7079f67 DG |
1809 | ret = -1; |
1810 | goto end_no_session; | |
1811 | } | |
1812 | ||
1813 | session_id = be64toh(msg.session_id); | |
1814 | ||
1815 | /* Iterate over all streams to see if the begin data pending flag is set. */ | |
1816 | rcu_read_lock(); | |
d3e2ba59 JD |
1817 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
1818 | stream_n.node) { | |
f7079f67 DG |
1819 | if (stream->session->id == session_id && |
1820 | !stream->data_pending_check_done) { | |
1821 | is_data_inflight = 1; | |
1822 | DBG("Data is still in flight for stream %" PRIu64, | |
1823 | stream->stream_handle); | |
1824 | break; | |
1825 | } | |
1826 | } | |
1827 | rcu_read_unlock(); | |
1828 | ||
1829 | /* All good, send back reply. */ | |
1830 | reply.ret_code = htobe32(is_data_inflight); | |
1831 | ||
1832 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1833 | if (ret < 0) { | |
1834 | ERR("Relay end data pending send reply failed"); | |
1835 | } | |
1836 | ||
1837 | end_no_session: | |
1838 | return ret; | |
1839 | } | |
1840 | ||
1c20f0e2 JD |
1841 | /* |
1842 | * Receive an index for a specific stream. | |
1843 | * | |
1844 | * Return 0 on success else a negative value. | |
1845 | */ | |
1846 | static | |
1847 | int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr, | |
0a6518b0 | 1848 | struct relay_command *cmd) |
1c20f0e2 JD |
1849 | { |
1850 | int ret, send_ret, index_created = 0; | |
1851 | struct relay_session *session = cmd->session; | |
1852 | struct lttcomm_relayd_index index_info; | |
1853 | struct relay_index *index, *wr_index = NULL; | |
1854 | struct lttcomm_relayd_generic_reply reply; | |
1855 | struct relay_stream *stream; | |
1856 | uint64_t net_seq_num; | |
1857 | ||
1858 | assert(cmd); | |
1c20f0e2 JD |
1859 | |
1860 | DBG("Relay receiving index"); | |
1861 | ||
1862 | if (!session || cmd->version_check_done == 0) { | |
1863 | ERR("Trying to close a stream before version check"); | |
1864 | ret = -1; | |
1865 | goto end_no_session; | |
1866 | } | |
1867 | ||
1868 | ret = cmd->sock->ops->recvmsg(cmd->sock, &index_info, | |
1869 | sizeof(index_info), 0); | |
1870 | if (ret < sizeof(index_info)) { | |
1871 | if (ret == 0) { | |
1872 | /* Orderly shutdown. Not necessary to print an error. */ | |
1873 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1874 | } else { | |
1875 | ERR("Relay didn't receive valid index struct size : %d", ret); | |
1876 | } | |
1877 | ret = -1; | |
1878 | goto end_no_session; | |
1879 | } | |
1880 | ||
1881 | net_seq_num = be64toh(index_info.net_seq_num); | |
1882 | ||
1883 | rcu_read_lock(); | |
d3e2ba59 | 1884 | stream = relay_stream_find_by_id(be64toh(index_info.relay_stream_id)); |
1c20f0e2 JD |
1885 | if (!stream) { |
1886 | ret = -1; | |
1887 | goto end_rcu_unlock; | |
1888 | } | |
1889 | ||
d3e2ba59 JD |
1890 | /* Live beacon handling */ |
1891 | if (index_info.packet_size == 0) { | |
1892 | DBG("Received live beacon for stream %" PRIu64, stream->stream_handle); | |
1893 | ||
1894 | /* | |
1895 | * Only flag a stream inactive when it has already received data. | |
1896 | */ | |
1897 | if (stream->total_index_received > 0) { | |
1898 | stream->beacon_ts_end = be64toh(index_info.timestamp_end); | |
1899 | } | |
1900 | ret = 0; | |
1901 | goto end_rcu_unlock; | |
1902 | } else { | |
1903 | stream->beacon_ts_end = -1ULL; | |
1904 | } | |
1905 | ||
0a6518b0 | 1906 | index = relay_index_find(stream->stream_handle, net_seq_num); |
1c20f0e2 JD |
1907 | if (!index) { |
1908 | /* A successful creation will add the object to the HT. */ | |
1909 | index = relay_index_create(stream->stream_handle, net_seq_num); | |
1910 | if (!index) { | |
1911 | goto end_rcu_unlock; | |
1912 | } | |
1913 | index_created = 1; | |
1914 | } | |
1915 | ||
1916 | copy_index_control_data(index, &index_info); | |
1917 | ||
1918 | if (index_created) { | |
1919 | /* | |
1920 | * Try to add the relay index object to the hash table. If an object | |
1921 | * already exist, destroy back the index created, set the data in this | |
1922 | * object and write it on disk. | |
1923 | */ | |
0a6518b0 | 1924 | relay_index_add(index, &wr_index); |
1c20f0e2 JD |
1925 | if (wr_index) { |
1926 | copy_index_control_data(wr_index, &index_info); | |
1927 | free(index); | |
1928 | } | |
1929 | } else { | |
1930 | /* The index already exists so write it on disk. */ | |
1931 | wr_index = index; | |
1932 | } | |
1933 | ||
1934 | /* Do we have a writable ready index to write on disk. */ | |
1935 | if (wr_index) { | |
1936 | /* Starting at 2.4, create the index file if none available. */ | |
1937 | if (cmd->minor >= 4 && stream->index_fd < 0) { | |
1938 | ret = index_create_file(stream->path_name, stream->channel_name, | |
1939 | relayd_uid, relayd_gid, stream->tracefile_size, | |
1940 | stream->tracefile_count_current); | |
1941 | if (ret < 0) { | |
1942 | goto end_rcu_unlock; | |
1943 | } | |
1944 | stream->index_fd = ret; | |
1945 | } | |
1946 | ||
0a6518b0 | 1947 | ret = relay_index_write(wr_index->fd, wr_index); |
1c20f0e2 JD |
1948 | if (ret < 0) { |
1949 | goto end_rcu_unlock; | |
1950 | } | |
d3e2ba59 | 1951 | stream->total_index_received++; |
1c20f0e2 JD |
1952 | } |
1953 | ||
1954 | end_rcu_unlock: | |
1955 | rcu_read_unlock(); | |
1956 | ||
1957 | if (ret < 0) { | |
1958 | reply.ret_code = htobe32(LTTNG_ERR_UNK); | |
1959 | } else { | |
1960 | reply.ret_code = htobe32(LTTNG_OK); | |
1961 | } | |
1962 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1963 | if (send_ret < 0) { | |
1964 | ERR("Relay sending close index id reply"); | |
1965 | ret = send_ret; | |
1966 | } | |
1967 | ||
1968 | end_no_session: | |
1969 | return ret; | |
1970 | } | |
1971 | ||
814fcae4 JD |
1972 | /* |
1973 | * Receive the streams_sent message. | |
1974 | * | |
1975 | * Return 0 on success else a negative value. | |
1976 | */ | |
1977 | static | |
1978 | int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr, | |
1979 | struct relay_command *cmd) | |
1980 | { | |
1981 | int ret, send_ret; | |
1982 | struct lttcomm_relayd_generic_reply reply; | |
1983 | ||
1984 | assert(cmd); | |
1985 | ||
1986 | DBG("Relay receiving streams_sent"); | |
1987 | ||
1988 | if (!cmd->session || cmd->version_check_done == 0) { | |
1989 | ERR("Trying to close a stream before version check"); | |
1990 | ret = -1; | |
1991 | goto end_no_session; | |
1992 | } | |
1993 | ||
1994 | /* | |
1995 | * Flag every pending stream in the connection recv list that they are | |
1996 | * ready to be used by the viewer. | |
1997 | */ | |
1998 | set_viewer_ready_flag(cmd); | |
1999 | ||
829007d9 JD |
2000 | /* |
2001 | * Inform the viewer that there are new streams in the session. | |
2002 | */ | |
2003 | uatomic_set(&cmd->session->new_streams, 1); | |
2004 | ||
814fcae4 JD |
2005 | reply.ret_code = htobe32(LTTNG_OK); |
2006 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
2007 | if (send_ret < 0) { | |
2008 | ERR("Relay sending sent_stream reply"); | |
2009 | ret = send_ret; | |
2010 | } else { | |
2011 | /* Success. */ | |
2012 | ret = 0; | |
2013 | } | |
2014 | ||
2015 | end_no_session: | |
2016 | return ret; | |
2017 | } | |
2018 | ||
b8aa1682 | 2019 | /* |
d3e2ba59 | 2020 | * Process the commands received on the control socket |
b8aa1682 JD |
2021 | */ |
2022 | static | |
2023 | int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 2024 | struct relay_command *cmd, struct relay_local_data *ctx) |
b8aa1682 JD |
2025 | { |
2026 | int ret = 0; | |
2027 | ||
2028 | switch (be32toh(recv_hdr->cmd)) { | |
b8aa1682 | 2029 | case RELAYD_CREATE_SESSION: |
d3e2ba59 | 2030 | ret = relay_create_session(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 | 2031 | break; |
b8aa1682 | 2032 | case RELAYD_ADD_STREAM: |
d3e2ba59 | 2033 | ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 JD |
2034 | break; |
2035 | case RELAYD_START_DATA: | |
2036 | ret = relay_start(recv_hdr, cmd); | |
2037 | break; | |
2038 | case RELAYD_SEND_METADATA: | |
d3e2ba59 | 2039 | ret = relay_recv_metadata(recv_hdr, cmd); |
b8aa1682 JD |
2040 | break; |
2041 | case RELAYD_VERSION: | |
d3e2ba59 | 2042 | ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 | 2043 | break; |
173af62f | 2044 | case RELAYD_CLOSE_STREAM: |
92c6ca54 | 2045 | ret = relay_close_stream(recv_hdr, cmd); |
173af62f | 2046 | break; |
6d805429 | 2047 | case RELAYD_DATA_PENDING: |
d3e2ba59 | 2048 | ret = relay_data_pending(recv_hdr, cmd); |
c8f59ee5 DG |
2049 | break; |
2050 | case RELAYD_QUIESCENT_CONTROL: | |
d3e2ba59 | 2051 | ret = relay_quiescent_control(recv_hdr, cmd); |
c8f59ee5 | 2052 | break; |
f7079f67 | 2053 | case RELAYD_BEGIN_DATA_PENDING: |
d3e2ba59 | 2054 | ret = relay_begin_data_pending(recv_hdr, cmd); |
f7079f67 DG |
2055 | break; |
2056 | case RELAYD_END_DATA_PENDING: | |
d3e2ba59 | 2057 | ret = relay_end_data_pending(recv_hdr, cmd); |
f7079f67 | 2058 | break; |
1c20f0e2 | 2059 | case RELAYD_SEND_INDEX: |
0a6518b0 | 2060 | ret = relay_recv_index(recv_hdr, cmd); |
1c20f0e2 | 2061 | break; |
814fcae4 JD |
2062 | case RELAYD_STREAMS_SENT: |
2063 | ret = relay_streams_sent(recv_hdr, cmd); | |
2064 | break; | |
b8aa1682 JD |
2065 | case RELAYD_UPDATE_SYNC_INFO: |
2066 | default: | |
2067 | ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd)); | |
2068 | relay_unknown_command(cmd); | |
2069 | ret = -1; | |
2070 | goto end; | |
2071 | } | |
2072 | ||
2073 | end: | |
2074 | return ret; | |
2075 | } | |
2076 | ||
7d2f7452 DG |
2077 | /* |
2078 | * Handle index for a data stream. | |
2079 | * | |
2080 | * RCU read side lock MUST be acquired. | |
2081 | * | |
2082 | * Return 0 on success else a negative value. | |
2083 | */ | |
2084 | static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num, | |
2085 | int rotate_index) | |
2086 | { | |
2087 | int ret = 0, index_created = 0; | |
2088 | uint64_t stream_id, data_offset; | |
2089 | struct relay_index *index, *wr_index = NULL; | |
2090 | ||
2091 | assert(stream); | |
2092 | ||
2093 | stream_id = stream->stream_handle; | |
2094 | /* Get data offset because we are about to update the index. */ | |
2095 | data_offset = htobe64(stream->tracefile_size_current); | |
2096 | ||
2097 | /* | |
2098 | * Lookup for an existing index for that stream id/sequence number. If on | |
2099 | * exists, the control thread already received the data for it thus we need | |
2100 | * to write it on disk. | |
2101 | */ | |
2102 | index = relay_index_find(stream_id, net_seq_num); | |
2103 | if (!index) { | |
2104 | /* A successful creation will add the object to the HT. */ | |
2105 | index = relay_index_create(stream_id, net_seq_num); | |
2106 | if (!index) { | |
2107 | ret = -1; | |
2108 | goto error; | |
2109 | } | |
2110 | index_created = 1; | |
2111 | } | |
2112 | ||
2113 | if (rotate_index || stream->index_fd < 0) { | |
2114 | index->to_close_fd = stream->index_fd; | |
2115 | ret = index_create_file(stream->path_name, stream->channel_name, | |
2116 | relayd_uid, relayd_gid, stream->tracefile_size, | |
2117 | stream->tracefile_count_current); | |
2118 | if (ret < 0) { | |
2119 | /* This will close the stream's index fd if one. */ | |
2120 | relay_index_free_safe(index); | |
2121 | goto error; | |
2122 | } | |
2123 | stream->index_fd = ret; | |
2124 | } | |
2125 | index->fd = stream->index_fd; | |
2126 | index->index_data.offset = data_offset; | |
2127 | ||
2128 | if (index_created) { | |
2129 | /* | |
2130 | * Try to add the relay index object to the hash table. If an object | |
2131 | * already exist, destroy back the index created and set the data. | |
2132 | */ | |
2133 | relay_index_add(index, &wr_index); | |
2134 | if (wr_index) { | |
2135 | /* Copy back data from the created index. */ | |
2136 | wr_index->fd = index->fd; | |
2137 | wr_index->to_close_fd = index->to_close_fd; | |
2138 | wr_index->index_data.offset = data_offset; | |
2139 | free(index); | |
2140 | } | |
2141 | } else { | |
2142 | /* The index already exists so write it on disk. */ | |
2143 | wr_index = index; | |
2144 | } | |
2145 | ||
2146 | /* Do we have a writable ready index to write on disk. */ | |
2147 | if (wr_index) { | |
2148 | ret = relay_index_write(wr_index->fd, wr_index); | |
2149 | if (ret < 0) { | |
2150 | goto error; | |
2151 | } | |
2152 | stream->total_index_received++; | |
2153 | } | |
2154 | ||
2155 | error: | |
2156 | return ret; | |
2157 | } | |
2158 | ||
b8aa1682 JD |
2159 | /* |
2160 | * relay_process_data: Process the data received on the data socket | |
2161 | */ | |
2162 | static | |
0a6518b0 | 2163 | int relay_process_data(struct relay_command *cmd) |
b8aa1682 | 2164 | { |
7d2f7452 | 2165 | int ret = 0, rotate_index = 0; |
6cd525e8 | 2166 | ssize_t size_ret; |
b8aa1682 JD |
2167 | struct relay_stream *stream; |
2168 | struct lttcomm_relayd_data_hdr data_hdr; | |
7d2f7452 | 2169 | uint64_t stream_id; |
173af62f | 2170 | uint64_t net_seq_num; |
b8aa1682 JD |
2171 | uint32_t data_size; |
2172 | ||
2173 | ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr, | |
7c5aef62 | 2174 | sizeof(struct lttcomm_relayd_data_hdr), 0); |
b8aa1682 | 2175 | if (ret <= 0) { |
a6cd2b97 DG |
2176 | if (ret == 0) { |
2177 | /* Orderly shutdown. Not necessary to print an error. */ | |
2178 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
2179 | } else { | |
2180 | ERR("Unable to receive data header on sock %d", cmd->sock->fd); | |
2181 | } | |
b8aa1682 JD |
2182 | ret = -1; |
2183 | goto end; | |
2184 | } | |
2185 | ||
2186 | stream_id = be64toh(data_hdr.stream_id); | |
9d1bbf21 MD |
2187 | |
2188 | rcu_read_lock(); | |
d3e2ba59 | 2189 | stream = relay_stream_find_by_id(stream_id); |
b8aa1682 JD |
2190 | if (!stream) { |
2191 | ret = -1; | |
1c20f0e2 | 2192 | goto end_rcu_unlock; |
b8aa1682 JD |
2193 | } |
2194 | ||
2195 | data_size = be32toh(data_hdr.data_size); | |
2196 | if (data_buffer_size < data_size) { | |
c617c0c6 MD |
2197 | char *tmp_data_ptr; |
2198 | ||
2199 | tmp_data_ptr = realloc(data_buffer, data_size); | |
2200 | if (!tmp_data_ptr) { | |
b8aa1682 | 2201 | ERR("Allocating data buffer"); |
c617c0c6 | 2202 | free(data_buffer); |
b8aa1682 | 2203 | ret = -1; |
1c20f0e2 | 2204 | goto end_rcu_unlock; |
b8aa1682 | 2205 | } |
c617c0c6 | 2206 | data_buffer = tmp_data_ptr; |
b8aa1682 JD |
2207 | data_buffer_size = data_size; |
2208 | } | |
2209 | memset(data_buffer, 0, data_size); | |
2210 | ||
173af62f DG |
2211 | net_seq_num = be64toh(data_hdr.net_seq_num); |
2212 | ||
77c7c900 | 2213 | DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64, |
173af62f | 2214 | data_size, stream_id, net_seq_num); |
7c5aef62 | 2215 | ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0); |
b8aa1682 | 2216 | if (ret <= 0) { |
a6cd2b97 DG |
2217 | if (ret == 0) { |
2218 | /* Orderly shutdown. Not necessary to print an error. */ | |
2219 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
2220 | } | |
b8aa1682 | 2221 | ret = -1; |
1c20f0e2 | 2222 | goto end_rcu_unlock; |
b8aa1682 JD |
2223 | } |
2224 | ||
1c20f0e2 | 2225 | /* Check if a rotation is needed. */ |
0f907de1 JD |
2226 | if (stream->tracefile_size > 0 && |
2227 | (stream->tracefile_size_current + data_size) > | |
2228 | stream->tracefile_size) { | |
6b6b9a5a JD |
2229 | struct relay_viewer_stream *vstream; |
2230 | uint64_t new_id; | |
2231 | ||
2232 | new_id = (stream->tracefile_count_current + 1) % | |
2233 | stream->tracefile_count; | |
2234 | /* | |
2235 | * When we wrap-around back to 0, we start overwriting old | |
2236 | * trace data. | |
2237 | */ | |
2238 | if (!stream->tracefile_overwrite && new_id == 0) { | |
2239 | stream->tracefile_overwrite = 1; | |
2240 | } | |
2241 | pthread_mutex_lock(&stream->viewer_stream_rotation_lock); | |
2242 | if (stream->tracefile_overwrite) { | |
2243 | stream->oldest_tracefile_id = | |
2244 | (stream->oldest_tracefile_id + 1) % | |
2245 | stream->tracefile_count; | |
2246 | } | |
991adae2 | 2247 | vstream = viewer_stream_find_by_id(stream->stream_handle); |
6b6b9a5a JD |
2248 | if (vstream) { |
2249 | /* | |
2250 | * The viewer is reading a file about to be | |
2251 | * overwritten. Close the FDs it is | |
2252 | * currently using and let it handle the fault. | |
2253 | */ | |
2254 | if (vstream->tracefile_count_current == new_id) { | |
cef0f7d5 | 2255 | pthread_mutex_lock(&vstream->overwrite_lock); |
6b6b9a5a | 2256 | vstream->abort_flag = 1; |
cef0f7d5 | 2257 | pthread_mutex_unlock(&vstream->overwrite_lock); |
6b6b9a5a JD |
2258 | DBG("Streaming side setting abort_flag on stream %s_%lu\n", |
2259 | stream->channel_name, new_id); | |
2260 | } else if (vstream->tracefile_count_current == | |
2261 | stream->tracefile_count_current) { | |
2262 | /* | |
2263 | * The reader and writer were in the | |
2264 | * same trace file, inform the viewer | |
2265 | * that no new index will ever be added | |
2266 | * to this file. | |
2267 | */ | |
2268 | vstream->close_write_flag = 1; | |
2269 | } | |
2270 | } | |
1c20f0e2 JD |
2271 | ret = utils_rotate_stream_file(stream->path_name, stream->channel_name, |
2272 | stream->tracefile_size, stream->tracefile_count, | |
2273 | relayd_uid, relayd_gid, stream->fd, | |
2274 | &(stream->tracefile_count_current), &stream->fd); | |
cef0f7d5 | 2275 | stream->total_index_received = 0; |
6b6b9a5a | 2276 | pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); |
0f907de1 | 2277 | if (ret < 0) { |
1c20f0e2 JD |
2278 | ERR("Rotating stream output file"); |
2279 | goto end_rcu_unlock; | |
0f907de1 | 2280 | } |
a6976990 DG |
2281 | /* Reset current size because we just perform a stream rotation. */ |
2282 | stream->tracefile_size_current = 0; | |
1c20f0e2 JD |
2283 | rotate_index = 1; |
2284 | } | |
2285 | ||
1c20f0e2 | 2286 | /* |
7d2f7452 DG |
2287 | * Index are handled in protocol version 2.4 and above. Also, snapshot and |
2288 | * index are NOT supported. | |
1c20f0e2 | 2289 | */ |
7d2f7452 DG |
2290 | if (stream->session->minor >= 4 && !stream->session->snapshot) { |
2291 | ret = handle_index_data(stream, net_seq_num, rotate_index); | |
1c20f0e2 | 2292 | if (ret < 0) { |
1c20f0e2 JD |
2293 | goto end_rcu_unlock; |
2294 | } | |
1c20f0e2 JD |
2295 | } |
2296 | ||
7d2f7452 | 2297 | /* Write data to stream output fd. */ |
6cd525e8 MD |
2298 | size_ret = lttng_write(stream->fd, data_buffer, data_size); |
2299 | if (size_ret < data_size) { | |
b8aa1682 JD |
2300 | ERR("Relay error writing data to file"); |
2301 | ret = -1; | |
1c20f0e2 | 2302 | goto end_rcu_unlock; |
b8aa1682 | 2303 | } |
1d4dfdef | 2304 | |
5ab7344e JD |
2305 | DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64, |
2306 | ret, stream->stream_handle); | |
2307 | ||
1d4dfdef DG |
2308 | ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size)); |
2309 | if (ret < 0) { | |
1c20f0e2 | 2310 | goto end_rcu_unlock; |
1d4dfdef | 2311 | } |
1c20f0e2 | 2312 | stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size); |
1d4dfdef | 2313 | |
173af62f DG |
2314 | stream->prev_seq = net_seq_num; |
2315 | ||
2316 | /* Check if we need to close the FD */ | |
2317 | if (close_stream_check(stream)) { | |
c07d8a78 | 2318 | destroy_stream(stream); |
173af62f DG |
2319 | } |
2320 | ||
1c20f0e2 | 2321 | end_rcu_unlock: |
9d1bbf21 | 2322 | rcu_read_unlock(); |
b8aa1682 JD |
2323 | end: |
2324 | return ret; | |
2325 | } | |
2326 | ||
2327 | static | |
9d1bbf21 | 2328 | void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd) |
b8aa1682 JD |
2329 | { |
2330 | int ret; | |
2331 | ||
b8aa1682 JD |
2332 | lttng_poll_del(events, pollfd); |
2333 | ||
2334 | ret = close(pollfd); | |
2335 | if (ret < 0) { | |
2336 | ERR("Closing pollfd %d", pollfd); | |
2337 | } | |
2338 | } | |
2339 | ||
2340 | static | |
2341 | int relay_add_connection(int fd, struct lttng_poll_event *events, | |
2342 | struct lttng_ht *relay_connections_ht) | |
2343 | { | |
b8aa1682 | 2344 | struct relay_command *relay_connection; |
6cd525e8 | 2345 | ssize_t ret; |
b8aa1682 JD |
2346 | |
2347 | relay_connection = zmalloc(sizeof(struct relay_command)); | |
2348 | if (relay_connection == NULL) { | |
2349 | PERROR("Relay command zmalloc"); | |
9d1bbf21 | 2350 | goto error; |
b8aa1682 | 2351 | } |
6cd525e8 MD |
2352 | ret = lttng_read(fd, relay_connection, sizeof(struct relay_command)); |
2353 | if (ret < sizeof(struct relay_command)) { | |
b8aa1682 | 2354 | PERROR("read relay cmd pipe"); |
9d1bbf21 | 2355 | goto error_read; |
b8aa1682 | 2356 | } |
814fcae4 | 2357 | CDS_INIT_LIST_HEAD(&relay_connection->recv_head); |
b8aa1682 | 2358 | |
c07d8a78 DG |
2359 | /* |
2360 | * Only used by the control side and the reference is copied inside each | |
2361 | * stream from that connection. Thus a destroy HT must be done after every | |
2362 | * stream has been destroyed. | |
2363 | */ | |
2364 | if (relay_connection->type == RELAY_CONTROL) { | |
2365 | relay_connection->ctf_traces_ht = lttng_ht_new(0, | |
2366 | LTTNG_HT_TYPE_STRING); | |
2367 | if (!relay_connection->ctf_traces_ht) { | |
2368 | goto error_read; | |
2369 | } | |
d3e2ba59 JD |
2370 | } |
2371 | ||
b8aa1682 JD |
2372 | lttng_ht_node_init_ulong(&relay_connection->sock_n, |
2373 | (unsigned long) relay_connection->sock->fd); | |
9d1bbf21 | 2374 | rcu_read_lock(); |
b8aa1682 JD |
2375 | lttng_ht_add_unique_ulong(relay_connections_ht, |
2376 | &relay_connection->sock_n); | |
9d1bbf21 MD |
2377 | rcu_read_unlock(); |
2378 | return lttng_poll_add(events, | |
b8aa1682 JD |
2379 | relay_connection->sock->fd, |
2380 | LPOLLIN | LPOLLRDHUP); | |
2381 | ||
9d1bbf21 MD |
2382 | error_read: |
2383 | free(relay_connection); | |
2384 | error: | |
2385 | return -1; | |
2386 | } | |
2387 | ||
2388 | static | |
2389 | void deferred_free_connection(struct rcu_head *head) | |
2390 | { | |
2391 | struct relay_command *relay_connection = | |
2392 | caa_container_of(head, struct relay_command, rcu_node); | |
5b6d8097 DG |
2393 | |
2394 | lttcomm_destroy_sock(relay_connection->sock); | |
9d1bbf21 MD |
2395 | free(relay_connection); |
2396 | } | |
2397 | ||
2398 | static | |
2399 | void relay_del_connection(struct lttng_ht *relay_connections_ht, | |
d3e2ba59 JD |
2400 | struct lttng_ht_iter *iter, struct relay_command *relay_connection, |
2401 | struct lttng_ht *sessions_ht) | |
9d1bbf21 MD |
2402 | { |
2403 | int ret; | |
2404 | ||
2405 | ret = lttng_ht_del(relay_connections_ht, iter); | |
2406 | assert(!ret); | |
c07d8a78 | 2407 | |
9d1bbf21 | 2408 | if (relay_connection->type == RELAY_CONTROL) { |
814fcae4 JD |
2409 | struct relay_stream_recv_handle *node, *tmp_node; |
2410 | ||
d3e2ba59 | 2411 | relay_delete_session(relay_connection, sessions_ht); |
c07d8a78 | 2412 | lttng_ht_destroy(relay_connection->ctf_traces_ht); |
814fcae4 JD |
2413 | |
2414 | /* Clean up recv list. */ | |
2415 | cds_list_for_each_entry_safe(node, tmp_node, | |
2416 | &relay_connection->recv_head, node) { | |
2417 | cds_list_del(&node->node); | |
2418 | free(node); | |
2419 | } | |
9d1bbf21 | 2420 | } |
5b6d8097 | 2421 | |
c07d8a78 | 2422 | call_rcu(&relay_connection->rcu_node, deferred_free_connection); |
b8aa1682 JD |
2423 | } |
2424 | ||
2425 | /* | |
2426 | * This thread does the actual work | |
2427 | */ | |
2428 | static | |
2429 | void *relay_thread_worker(void *data) | |
2430 | { | |
beaad64c DG |
2431 | int ret, err = -1, last_seen_data_fd = -1; |
2432 | uint32_t nb_fd; | |
b8aa1682 JD |
2433 | struct relay_command *relay_connection; |
2434 | struct lttng_poll_event events; | |
2435 | struct lttng_ht *relay_connections_ht; | |
2436 | struct lttng_ht_node_ulong *node; | |
2437 | struct lttng_ht_iter iter; | |
b8aa1682 | 2438 | struct lttcomm_relayd_hdr recv_hdr; |
d3e2ba59 JD |
2439 | struct relay_local_data *relay_ctx = (struct relay_local_data *) data; |
2440 | struct lttng_ht *sessions_ht = relay_ctx->sessions_ht; | |
b8aa1682 JD |
2441 | |
2442 | DBG("[thread] Relay worker started"); | |
2443 | ||
9d1bbf21 MD |
2444 | rcu_register_thread(); |
2445 | ||
55706a7d MD |
2446 | health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER); |
2447 | ||
4e5b8b82 MD |
2448 | if (testpoint(relayd_thread_worker)) { |
2449 | goto error_testpoint; | |
2450 | } | |
2451 | ||
f385ae0a MD |
2452 | health_code_update(); |
2453 | ||
b8aa1682 JD |
2454 | /* table of connections indexed on socket */ |
2455 | relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
095a4ae5 MD |
2456 | if (!relay_connections_ht) { |
2457 | goto relay_connections_ht_error; | |
2458 | } | |
b8aa1682 | 2459 | |
1c20f0e2 JD |
2460 | /* Tables of received indexes indexed by index handle and net_seq_num. */ |
2461 | indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64); | |
2462 | if (!indexes_ht) { | |
2463 | goto indexes_ht_error; | |
2464 | } | |
2465 | ||
b8aa1682 JD |
2466 | ret = create_thread_poll_set(&events, 2); |
2467 | if (ret < 0) { | |
2468 | goto error_poll_create; | |
2469 | } | |
2470 | ||
2471 | ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP); | |
2472 | if (ret < 0) { | |
2473 | goto error; | |
2474 | } | |
2475 | ||
beaad64c | 2476 | restart: |
b8aa1682 | 2477 | while (1) { |
beaad64c DG |
2478 | int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1; |
2479 | ||
f385ae0a MD |
2480 | health_code_update(); |
2481 | ||
b8aa1682 | 2482 | /* Infinite blocking call, waiting for transmission */ |
87c1611d | 2483 | DBG3("Relayd worker thread polling..."); |
f385ae0a | 2484 | health_poll_entry(); |
b8aa1682 | 2485 | ret = lttng_poll_wait(&events, -1); |
f385ae0a | 2486 | health_poll_exit(); |
b8aa1682 JD |
2487 | if (ret < 0) { |
2488 | /* | |
2489 | * Restart interrupted system call. | |
2490 | */ | |
2491 | if (errno == EINTR) { | |
2492 | goto restart; | |
2493 | } | |
2494 | goto error; | |
2495 | } | |
2496 | ||
0d9c5d77 DG |
2497 | nb_fd = ret; |
2498 | ||
beaad64c DG |
2499 | /* |
2500 | * Process control. The control connection is prioritised so we don't | |
2501 | * starve it with high throughout put tracing data on the data | |
2502 | * connection. | |
2503 | */ | |
b8aa1682 JD |
2504 | for (i = 0; i < nb_fd; i++) { |
2505 | /* Fetch once the poll data */ | |
beaad64c DG |
2506 | uint32_t revents = LTTNG_POLL_GETEV(&events, i); |
2507 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
b8aa1682 | 2508 | |
f385ae0a MD |
2509 | health_code_update(); |
2510 | ||
b8aa1682 JD |
2511 | /* Thread quit pipe has been closed. Killing thread. */ |
2512 | ret = check_thread_quit_pipe(pollfd, revents); | |
2513 | if (ret) { | |
095a4ae5 MD |
2514 | err = 0; |
2515 | goto exit; | |
b8aa1682 JD |
2516 | } |
2517 | ||
2518 | /* Inspect the relay cmd pipe for new connection */ | |
2519 | if (pollfd == relay_cmd_pipe[0]) { | |
2520 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
2521 | ERR("Relay pipe error"); | |
2522 | goto error; | |
2523 | } else if (revents & LPOLLIN) { | |
2524 | DBG("Relay command received"); | |
2525 | ret = relay_add_connection(relay_cmd_pipe[0], | |
2526 | &events, relay_connections_ht); | |
2527 | if (ret < 0) { | |
2528 | goto error; | |
2529 | } | |
2530 | } | |
beaad64c | 2531 | } else if (revents) { |
9d1bbf21 | 2532 | rcu_read_lock(); |
b8aa1682 JD |
2533 | lttng_ht_lookup(relay_connections_ht, |
2534 | (void *)((unsigned long) pollfd), | |
2535 | &iter); | |
2536 | node = lttng_ht_iter_get_node_ulong(&iter); | |
2537 | if (node == NULL) { | |
2538 | DBG2("Relay sock %d not found", pollfd); | |
9d1bbf21 | 2539 | rcu_read_unlock(); |
b8aa1682 JD |
2540 | goto error; |
2541 | } | |
2542 | relay_connection = caa_container_of(node, | |
2543 | struct relay_command, sock_n); | |
2544 | ||
2545 | if (revents & (LPOLLERR)) { | |
2546 | ERR("POLL ERROR"); | |
9d1bbf21 MD |
2547 | relay_cleanup_poll_connection(&events, pollfd); |
2548 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2549 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2550 | if (last_seen_data_fd == pollfd) { |
2551 | last_seen_data_fd = last_notdel_data_fd; | |
2552 | } | |
b8aa1682 JD |
2553 | } else if (revents & (LPOLLHUP | LPOLLRDHUP)) { |
2554 | DBG("Socket %d hung up", pollfd); | |
9d1bbf21 MD |
2555 | relay_cleanup_poll_connection(&events, pollfd); |
2556 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2557 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2558 | if (last_seen_data_fd == pollfd) { |
2559 | last_seen_data_fd = last_notdel_data_fd; | |
2560 | } | |
b8aa1682 JD |
2561 | } else if (revents & LPOLLIN) { |
2562 | /* control socket */ | |
2563 | if (relay_connection->type == RELAY_CONTROL) { | |
2564 | ret = relay_connection->sock->ops->recvmsg( | |
2565 | relay_connection->sock, &recv_hdr, | |
7c5aef62 | 2566 | sizeof(struct lttcomm_relayd_hdr), 0); |
b8aa1682 JD |
2567 | /* connection closed */ |
2568 | if (ret <= 0) { | |
9d1bbf21 MD |
2569 | relay_cleanup_poll_connection(&events, pollfd); |
2570 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2571 | &iter, relay_connection, sessions_ht); |
b8aa1682 JD |
2572 | DBG("Control connection closed with %d", pollfd); |
2573 | } else { | |
2574 | if (relay_connection->session) { | |
77c7c900 | 2575 | DBG2("Relay worker receiving data for session : %" PRIu64, |
b8aa1682 JD |
2576 | relay_connection->session->id); |
2577 | } | |
2578 | ret = relay_process_control(&recv_hdr, | |
d3e2ba59 | 2579 | relay_connection, relay_ctx); |
b8aa1682 | 2580 | if (ret < 0) { |
beaad64c | 2581 | /* Clear the session on error. */ |
9d1bbf21 MD |
2582 | relay_cleanup_poll_connection(&events, pollfd); |
2583 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2584 | &iter, relay_connection, sessions_ht); |
b8aa1682 JD |
2585 | DBG("Connection closed with %d", pollfd); |
2586 | } | |
beaad64c | 2587 | seen_control = 1; |
b8aa1682 | 2588 | } |
beaad64c DG |
2589 | } else { |
2590 | /* | |
2591 | * Flag the last seen data fd not deleted. It will be | |
2592 | * used as the last seen fd if any fd gets deleted in | |
2593 | * this first loop. | |
2594 | */ | |
2595 | last_notdel_data_fd = pollfd; | |
2596 | } | |
2597 | } | |
2598 | rcu_read_unlock(); | |
2599 | } | |
2600 | } | |
2601 | ||
2602 | /* | |
2603 | * The last loop handled a control request, go back to poll to make | |
2604 | * sure we prioritise the control socket. | |
2605 | */ | |
2606 | if (seen_control) { | |
2607 | continue; | |
2608 | } | |
2609 | ||
2610 | if (last_seen_data_fd >= 0) { | |
2611 | for (i = 0; i < nb_fd; i++) { | |
2612 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
f385ae0a MD |
2613 | |
2614 | health_code_update(); | |
2615 | ||
beaad64c DG |
2616 | if (last_seen_data_fd == pollfd) { |
2617 | idx = i; | |
2618 | break; | |
2619 | } | |
2620 | } | |
2621 | } | |
2622 | ||
2623 | /* Process data connection. */ | |
2624 | for (i = idx + 1; i < nb_fd; i++) { | |
2625 | /* Fetch the poll data. */ | |
2626 | uint32_t revents = LTTNG_POLL_GETEV(&events, i); | |
2627 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
2628 | ||
f385ae0a MD |
2629 | health_code_update(); |
2630 | ||
beaad64c DG |
2631 | /* Skip the command pipe. It's handled in the first loop. */ |
2632 | if (pollfd == relay_cmd_pipe[0]) { | |
2633 | continue; | |
2634 | } | |
2635 | ||
2636 | if (revents) { | |
2637 | rcu_read_lock(); | |
2638 | lttng_ht_lookup(relay_connections_ht, | |
2639 | (void *)((unsigned long) pollfd), | |
2640 | &iter); | |
2641 | node = lttng_ht_iter_get_node_ulong(&iter); | |
2642 | if (node == NULL) { | |
2643 | /* Skip it. Might be removed before. */ | |
2644 | rcu_read_unlock(); | |
2645 | continue; | |
2646 | } | |
2647 | relay_connection = caa_container_of(node, | |
2648 | struct relay_command, sock_n); | |
2649 | ||
2650 | if (revents & LPOLLIN) { | |
2651 | if (relay_connection->type != RELAY_DATA) { | |
2652 | continue; | |
2653 | } | |
2654 | ||
0a6518b0 | 2655 | ret = relay_process_data(relay_connection); |
beaad64c DG |
2656 | /* connection closed */ |
2657 | if (ret < 0) { | |
2658 | relay_cleanup_poll_connection(&events, pollfd); | |
2659 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2660 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2661 | DBG("Data connection closed with %d", pollfd); |
2662 | /* | |
2663 | * Every goto restart call sets the last seen fd where | |
2664 | * here we don't really care since we gracefully | |
2665 | * continue the loop after the connection is deleted. | |
2666 | */ | |
2667 | } else { | |
2668 | /* Keep last seen port. */ | |
2669 | last_seen_data_fd = pollfd; | |
2670 | rcu_read_unlock(); | |
2671 | goto restart; | |
b8aa1682 JD |
2672 | } |
2673 | } | |
9d1bbf21 | 2674 | rcu_read_unlock(); |
b8aa1682 JD |
2675 | } |
2676 | } | |
beaad64c | 2677 | last_seen_data_fd = -1; |
b8aa1682 JD |
2678 | } |
2679 | ||
f385ae0a MD |
2680 | /* Normal exit, no error */ |
2681 | ret = 0; | |
2682 | ||
095a4ae5 | 2683 | exit: |
b8aa1682 JD |
2684 | error: |
2685 | lttng_poll_clean(&events); | |
2686 | ||
2687 | /* empty the hash table and free the memory */ | |
9d1bbf21 | 2688 | rcu_read_lock(); |
b8aa1682 | 2689 | cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) { |
f385ae0a MD |
2690 | health_code_update(); |
2691 | ||
b8aa1682 JD |
2692 | node = lttng_ht_iter_get_node_ulong(&iter); |
2693 | if (node) { | |
2694 | relay_connection = caa_container_of(node, | |
2695 | struct relay_command, sock_n); | |
9d1bbf21 | 2696 | relay_del_connection(relay_connections_ht, |
d3e2ba59 | 2697 | &iter, relay_connection, sessions_ht); |
b8aa1682 | 2698 | } |
b8aa1682 | 2699 | } |
94d49140 | 2700 | rcu_read_unlock(); |
7d2f7452 DG |
2701 | error_poll_create: |
2702 | lttng_ht_destroy(indexes_ht); | |
1c20f0e2 | 2703 | indexes_ht_error: |
b8aa1682 | 2704 | lttng_ht_destroy(relay_connections_ht); |
095a4ae5 | 2705 | relay_connections_ht_error: |
6620da75 DG |
2706 | /* Close relay cmd pipes */ |
2707 | utils_close_pipe(relay_cmd_pipe); | |
095a4ae5 MD |
2708 | if (err) { |
2709 | DBG("Thread exited with error"); | |
2710 | } | |
b8aa1682 | 2711 | DBG("Worker thread cleanup complete"); |
095a4ae5 | 2712 | free(data_buffer); |
4e5b8b82 | 2713 | error_testpoint: |
f385ae0a MD |
2714 | if (err) { |
2715 | health_error(); | |
2716 | ERR("Health error occurred in %s", __func__); | |
2717 | } | |
2718 | health_unregister(health_relayd); | |
9d1bbf21 | 2719 | rcu_unregister_thread(); |
f385ae0a | 2720 | stop_threads(); |
b8aa1682 JD |
2721 | return NULL; |
2722 | } | |
2723 | ||
2724 | /* | |
2725 | * Create the relay command pipe to wake thread_manage_apps. | |
2726 | * Closed in cleanup(). | |
2727 | */ | |
2728 | static int create_relay_cmd_pipe(void) | |
2729 | { | |
a02de639 | 2730 | int ret; |
b8aa1682 | 2731 | |
a02de639 | 2732 | ret = utils_create_pipe_cloexec(relay_cmd_pipe); |
b8aa1682 | 2733 | |
b8aa1682 JD |
2734 | return ret; |
2735 | } | |
2736 | ||
2737 | /* | |
2738 | * main | |
2739 | */ | |
2740 | int main(int argc, char **argv) | |
2741 | { | |
2742 | int ret = 0; | |
2743 | void *status; | |
d3e2ba59 | 2744 | struct relay_local_data *relay_ctx; |
b8aa1682 | 2745 | |
b8aa1682 JD |
2746 | /* Parse arguments */ |
2747 | progname = argv[0]; | |
c617c0c6 | 2748 | if ((ret = parse_args(argc, argv)) < 0) { |
a02de639 | 2749 | goto exit; |
b8aa1682 JD |
2750 | } |
2751 | ||
2752 | if ((ret = set_signal_handler()) < 0) { | |
2753 | goto exit; | |
2754 | } | |
2755 | ||
4d513a50 DG |
2756 | /* Try to create directory if -o, --output is specified. */ |
2757 | if (opt_output_path) { | |
994fa64f DG |
2758 | if (*opt_output_path != '/') { |
2759 | ERR("Please specify an absolute path for -o, --output PATH"); | |
2760 | goto exit; | |
2761 | } | |
2762 | ||
4d513a50 DG |
2763 | ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG); |
2764 | if (ret < 0) { | |
2765 | ERR("Unable to create %s", opt_output_path); | |
2766 | goto exit; | |
2767 | } | |
2768 | } | |
2769 | ||
b8aa1682 | 2770 | /* Daemonize */ |
d1a3048a | 2771 | if (opt_daemon || opt_background) { |
02a6bb53 MD |
2772 | int i; |
2773 | ||
2774 | ret = lttng_daemonize(&child_ppid, &recv_child_signal, | |
2775 | !opt_background); | |
b8aa1682 | 2776 | if (ret < 0) { |
a02de639 | 2777 | goto exit; |
b8aa1682 | 2778 | } |
02a6bb53 MD |
2779 | |
2780 | /* | |
2781 | * We are in the child. Make sure all other file | |
2782 | * descriptors are closed, in case we are called with | |
2783 | * more opened file descriptors than the standard ones. | |
2784 | */ | |
2785 | for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) { | |
2786 | (void) close(i); | |
2787 | } | |
2788 | } | |
2789 | ||
2790 | /* Create thread quit pipe */ | |
2791 | if ((ret = init_thread_quit_pipe()) < 0) { | |
2792 | goto error; | |
b8aa1682 JD |
2793 | } |
2794 | ||
1c20f0e2 JD |
2795 | /* We need those values for the file/dir creation. */ |
2796 | relayd_uid = getuid(); | |
2797 | relayd_gid = getgid(); | |
b8aa1682 | 2798 | |
1c20f0e2 JD |
2799 | /* Check if daemon is UID = 0 */ |
2800 | if (relayd_uid == 0) { | |
a6a062a4 AM |
2801 | if (control_uri->port < 1024 || data_uri->port < 1024 || |
2802 | live_uri->port < 1024) { | |
b8aa1682 JD |
2803 | ERR("Need to be root to use ports < 1024"); |
2804 | ret = -1; | |
a02de639 | 2805 | goto exit; |
b8aa1682 JD |
2806 | } |
2807 | } | |
2808 | ||
2809 | /* Setup the thread apps communication pipe. */ | |
2810 | if ((ret = create_relay_cmd_pipe()) < 0) { | |
2811 | goto exit; | |
2812 | } | |
2813 | ||
2814 | /* Init relay command queue. */ | |
2815 | cds_wfq_init(&relay_cmd_queue.queue); | |
2816 | ||
2817 | /* Set up max poll set size */ | |
2818 | lttng_poll_set_max_size(); | |
2819 | ||
554831e7 MD |
2820 | /* Initialize communication library */ |
2821 | lttcomm_init(); | |
d5a5e58f | 2822 | lttcomm_inet_init(); |
554831e7 | 2823 | |
d3e2ba59 JD |
2824 | relay_ctx = zmalloc(sizeof(struct relay_local_data)); |
2825 | if (!relay_ctx) { | |
2826 | PERROR("relay_ctx"); | |
2827 | goto exit; | |
2828 | } | |
2829 | ||
2830 | /* tables of sessions indexed by session ID */ | |
2831 | relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
2832 | if (!relay_ctx->sessions_ht) { | |
2833 | goto exit_relay_ctx_sessions; | |
2834 | } | |
2835 | ||
2836 | /* tables of streams indexed by stream ID */ | |
2837 | relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
2838 | if (!relay_streams_ht) { | |
2839 | goto exit_relay_ctx_streams; | |
2840 | } | |
2841 | ||
2842 | /* tables of streams indexed by stream ID */ | |
92c6ca54 DG |
2843 | viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
2844 | if (!viewer_streams_ht) { | |
d3e2ba59 JD |
2845 | goto exit_relay_ctx_viewer_streams; |
2846 | } | |
2847 | ||
55706a7d MD |
2848 | /* Initialize thread health monitoring */ |
2849 | health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES); | |
2850 | if (!health_relayd) { | |
2851 | PERROR("health_app_create error"); | |
2852 | goto exit_health_app_create; | |
2853 | } | |
2854 | ||
65931c8b MD |
2855 | ret = utils_create_pipe(health_quit_pipe); |
2856 | if (ret < 0) { | |
2857 | goto error_health_pipe; | |
2858 | } | |
2859 | ||
2860 | /* Create thread to manage the client socket */ | |
2861 | ret = pthread_create(&health_thread, NULL, | |
2862 | thread_manage_health, (void *) NULL); | |
2863 | if (ret != 0) { | |
2864 | PERROR("pthread_create health"); | |
2865 | goto health_error; | |
2866 | } | |
2867 | ||
b8aa1682 JD |
2868 | /* Setup the dispatcher thread */ |
2869 | ret = pthread_create(&dispatcher_thread, NULL, | |
2870 | relay_thread_dispatcher, (void *) NULL); | |
2871 | if (ret != 0) { | |
2872 | PERROR("pthread_create dispatcher"); | |
2873 | goto exit_dispatcher; | |
2874 | } | |
2875 | ||
2876 | /* Setup the worker thread */ | |
2877 | ret = pthread_create(&worker_thread, NULL, | |
d3e2ba59 | 2878 | relay_thread_worker, (void *) relay_ctx); |
b8aa1682 JD |
2879 | if (ret != 0) { |
2880 | PERROR("pthread_create worker"); | |
2881 | goto exit_worker; | |
2882 | } | |
2883 | ||
2884 | /* Setup the listener thread */ | |
2885 | ret = pthread_create(&listener_thread, NULL, | |
2886 | relay_thread_listener, (void *) NULL); | |
2887 | if (ret != 0) { | |
2888 | PERROR("pthread_create listener"); | |
2889 | goto exit_listener; | |
2890 | } | |
2891 | ||
3557d456 | 2892 | ret = live_start_threads(live_uri, relay_ctx); |
d3e2ba59 JD |
2893 | if (ret != 0) { |
2894 | ERR("Starting live viewer threads"); | |
50138f51 | 2895 | goto exit_live; |
d3e2ba59 JD |
2896 | } |
2897 | ||
50138f51 | 2898 | exit_live: |
b8aa1682 JD |
2899 | ret = pthread_join(listener_thread, &status); |
2900 | if (ret != 0) { | |
2901 | PERROR("pthread_join"); | |
2902 | goto error; /* join error, exit without cleanup */ | |
2903 | } | |
2904 | ||
50138f51 | 2905 | exit_listener: |
b8aa1682 JD |
2906 | ret = pthread_join(worker_thread, &status); |
2907 | if (ret != 0) { | |
2908 | PERROR("pthread_join"); | |
2909 | goto error; /* join error, exit without cleanup */ | |
2910 | } | |
2911 | ||
50138f51 | 2912 | exit_worker: |
b8aa1682 JD |
2913 | ret = pthread_join(dispatcher_thread, &status); |
2914 | if (ret != 0) { | |
2915 | PERROR("pthread_join"); | |
2916 | goto error; /* join error, exit without cleanup */ | |
2917 | } | |
42415026 | 2918 | |
50138f51 | 2919 | exit_dispatcher: |
65931c8b MD |
2920 | ret = pthread_join(health_thread, &status); |
2921 | if (ret != 0) { | |
2922 | PERROR("pthread_join health thread"); | |
2923 | goto error; /* join error, exit without cleanup */ | |
2924 | } | |
2925 | ||
bd11e201 MD |
2926 | /* |
2927 | * Stop live threads only after joining other threads. | |
2928 | */ | |
2929 | live_stop_threads(); | |
2930 | ||
65931c8b MD |
2931 | health_error: |
2932 | utils_close_pipe(health_quit_pipe); | |
2933 | ||
2934 | error_health_pipe: | |
55706a7d MD |
2935 | health_app_destroy(health_relayd); |
2936 | ||
2937 | exit_health_app_create: | |
92c6ca54 | 2938 | lttng_ht_destroy(viewer_streams_ht); |
d3e2ba59 JD |
2939 | |
2940 | exit_relay_ctx_viewer_streams: | |
2941 | lttng_ht_destroy(relay_streams_ht); | |
2942 | ||
2943 | exit_relay_ctx_streams: | |
2944 | lttng_ht_destroy(relay_ctx->sessions_ht); | |
2945 | ||
2946 | exit_relay_ctx_sessions: | |
2947 | free(relay_ctx); | |
b8aa1682 JD |
2948 | |
2949 | exit: | |
2950 | cleanup(); | |
2951 | if (!ret) { | |
2952 | exit(EXIT_SUCCESS); | |
2953 | } | |
a02de639 | 2954 | |
b8aa1682 JD |
2955 | error: |
2956 | exit(EXIT_FAILURE); | |
2957 | } |