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" |
b8aa1682 JD |
63 | |
64 | /* command line options */ | |
0f907de1 | 65 | char *opt_output_path; |
d1a3048a | 66 | static int opt_daemon, opt_background; |
02a6bb53 MD |
67 | |
68 | /* | |
69 | * We need to wait for listener and live listener threads, as well as | |
70 | * health check thread, before being ready to signal readiness. | |
71 | */ | |
72 | #define NR_LTTNG_RELAY_READY 3 | |
73 | static int lttng_relay_ready = NR_LTTNG_RELAY_READY; | |
74 | static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */ | |
75 | static pid_t child_ppid; /* Internal parent PID use with daemonize. */ | |
76 | ||
095a4ae5 MD |
77 | static struct lttng_uri *control_uri; |
78 | static struct lttng_uri *data_uri; | |
d3e2ba59 | 79 | static struct lttng_uri *live_uri; |
b8aa1682 JD |
80 | |
81 | const char *progname; | |
b8aa1682 | 82 | |
65931c8b MD |
83 | const char *tracing_group_name = DEFAULT_TRACING_GROUP; |
84 | ||
b8aa1682 JD |
85 | /* |
86 | * Quit pipe for all threads. This permits a single cancellation point | |
87 | * for all threads when receiving an event on the pipe. | |
88 | */ | |
3557d456 | 89 | int thread_quit_pipe[2] = { -1, -1 }; |
b8aa1682 JD |
90 | |
91 | /* | |
92 | * This pipe is used to inform the worker thread that a command is queued and | |
93 | * ready to be processed. | |
94 | */ | |
95 | static int relay_cmd_pipe[2] = { -1, -1 }; | |
96 | ||
26c9d55e | 97 | /* Shared between threads */ |
b8aa1682 JD |
98 | static int dispatch_thread_exit; |
99 | ||
100 | static pthread_t listener_thread; | |
101 | static pthread_t dispatcher_thread; | |
102 | static pthread_t worker_thread; | |
65931c8b | 103 | static pthread_t health_thread; |
b8aa1682 | 104 | |
095a4ae5 MD |
105 | static uint64_t last_relay_stream_id; |
106 | static uint64_t last_relay_session_id; | |
b8aa1682 JD |
107 | |
108 | /* | |
109 | * Relay command queue. | |
110 | * | |
111 | * The relay_thread_listener and relay_thread_dispatcher communicate with this | |
112 | * queue. | |
113 | */ | |
114 | static struct relay_cmd_queue relay_cmd_queue; | |
115 | ||
116 | /* buffer allocated at startup, used to store the trace data */ | |
095a4ae5 MD |
117 | static char *data_buffer; |
118 | static unsigned int data_buffer_size; | |
b8aa1682 | 119 | |
1c20f0e2 JD |
120 | /* We need those values for the file/dir creation. */ |
121 | static uid_t relayd_uid; | |
122 | static gid_t relayd_gid; | |
123 | ||
d3e2ba59 JD |
124 | /* Global relay stream hash table. */ |
125 | struct lttng_ht *relay_streams_ht; | |
126 | ||
92c6ca54 DG |
127 | /* Global relay viewer stream hash table. */ |
128 | struct lttng_ht *viewer_streams_ht; | |
129 | ||
0a6518b0 DG |
130 | /* Global hash table that stores relay index object. */ |
131 | struct lttng_ht *indexes_ht; | |
132 | ||
55706a7d | 133 | /* Relayd health monitoring */ |
eea7556c | 134 | struct health_app *health_relayd; |
55706a7d | 135 | |
b8aa1682 JD |
136 | /* |
137 | * usage function on stderr | |
138 | */ | |
139 | static | |
140 | void usage(void) | |
141 | { | |
142 | fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname); | |
994fa64f DG |
143 | fprintf(stderr, " -h, --help Display this usage.\n"); |
144 | fprintf(stderr, " -d, --daemonize Start as a daemon.\n"); | |
d1a3048a | 145 | fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n"); |
994fa64f DG |
146 | fprintf(stderr, " -C, --control-port URL Control port listening.\n"); |
147 | fprintf(stderr, " -D, --data-port URL Data port listening.\n"); | |
a6a062a4 | 148 | fprintf(stderr, " -L, --live-port URL Live view port listening.\n"); |
994fa64f DG |
149 | fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n"); |
150 | fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n"); | |
65931c8b | 151 | fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n"); |
b8aa1682 JD |
152 | } |
153 | ||
154 | static | |
155 | int parse_args(int argc, char **argv) | |
156 | { | |
157 | int c; | |
158 | int ret = 0; | |
159 | char *default_address; | |
160 | ||
161 | static struct option long_options[] = { | |
e3678fd8 MD |
162 | { "control-port", 1, 0, 'C', }, |
163 | { "data-port", 1, 0, 'D', }, | |
164 | { "daemonize", 0, 0, 'd', }, | |
65931c8b | 165 | { "group", 1, 0, 'g', }, |
e3678fd8 MD |
166 | { "help", 0, 0, 'h', }, |
167 | { "output", 1, 0, 'o', }, | |
168 | { "verbose", 0, 0, 'v', }, | |
06b55dbb | 169 | { "background", 0, 0, 'b' }, |
095a4ae5 | 170 | { NULL, 0, 0, 0, }, |
b8aa1682 JD |
171 | }; |
172 | ||
173 | while (1) { | |
174 | int option_index = 0; | |
06b55dbb | 175 | c = getopt_long(argc, argv, "dhv" "C:D:L:o:g:b", |
b8aa1682 JD |
176 | long_options, &option_index); |
177 | if (c == -1) { | |
178 | break; | |
179 | } | |
180 | ||
181 | switch (c) { | |
182 | case 0: | |
183 | fprintf(stderr, "option %s", long_options[option_index].name); | |
184 | if (optarg) { | |
185 | fprintf(stderr, " with arg %s\n", optarg); | |
186 | } | |
187 | break; | |
188 | case 'C': | |
189 | ret = uri_parse(optarg, &control_uri); | |
190 | if (ret < 0) { | |
191 | ERR("Invalid control URI specified"); | |
192 | goto exit; | |
193 | } | |
194 | if (control_uri->port == 0) { | |
195 | control_uri->port = DEFAULT_NETWORK_CONTROL_PORT; | |
196 | } | |
197 | break; | |
198 | case 'D': | |
199 | ret = uri_parse(optarg, &data_uri); | |
200 | if (ret < 0) { | |
201 | ERR("Invalid data URI specified"); | |
202 | goto exit; | |
203 | } | |
204 | if (data_uri->port == 0) { | |
205 | data_uri->port = DEFAULT_NETWORK_DATA_PORT; | |
206 | } | |
207 | break; | |
a6a062a4 AM |
208 | case 'L': |
209 | ret = uri_parse(optarg, &live_uri); | |
210 | if (ret < 0) { | |
211 | ERR("Invalid live URI specified"); | |
212 | goto exit; | |
213 | } | |
214 | if (live_uri->port == 0) { | |
215 | live_uri->port = DEFAULT_NETWORK_VIEWER_PORT; | |
216 | } | |
217 | break; | |
b8aa1682 JD |
218 | case 'd': |
219 | opt_daemon = 1; | |
220 | break; | |
06b55dbb DG |
221 | case 'b': |
222 | opt_background = 1; | |
223 | break; | |
65931c8b MD |
224 | case 'g': |
225 | tracing_group_name = optarg; | |
226 | break; | |
b8aa1682 JD |
227 | case 'h': |
228 | usage(); | |
229 | exit(EXIT_FAILURE); | |
230 | case 'o': | |
231 | ret = asprintf(&opt_output_path, "%s", optarg); | |
232 | if (ret < 0) { | |
233 | PERROR("asprintf opt_output_path"); | |
234 | goto exit; | |
235 | } | |
236 | break; | |
237 | case 'v': | |
238 | /* Verbose level can increase using multiple -v */ | |
239 | lttng_opt_verbose += 1; | |
240 | break; | |
241 | default: | |
242 | /* Unknown option or other error. | |
243 | * Error is printed by getopt, just return */ | |
244 | ret = -1; | |
245 | goto exit; | |
246 | } | |
247 | } | |
248 | ||
249 | /* assign default values */ | |
250 | if (control_uri == NULL) { | |
251 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
252 | DEFAULT_NETWORK_CONTROL_PORT); | |
253 | if (ret < 0) { | |
254 | PERROR("asprintf default data address"); | |
255 | goto exit; | |
256 | } | |
257 | ||
258 | ret = uri_parse(default_address, &control_uri); | |
259 | free(default_address); | |
260 | if (ret < 0) { | |
261 | ERR("Invalid control URI specified"); | |
262 | goto exit; | |
263 | } | |
264 | } | |
265 | if (data_uri == NULL) { | |
266 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
267 | DEFAULT_NETWORK_DATA_PORT); | |
268 | if (ret < 0) { | |
269 | PERROR("asprintf default data address"); | |
270 | goto exit; | |
271 | } | |
272 | ||
273 | ret = uri_parse(default_address, &data_uri); | |
274 | free(default_address); | |
275 | if (ret < 0) { | |
276 | ERR("Invalid data URI specified"); | |
277 | goto exit; | |
278 | } | |
279 | } | |
d3e2ba59 JD |
280 | if (live_uri == NULL) { |
281 | ret = asprintf(&default_address, "tcp://0.0.0.0:%d", | |
282 | DEFAULT_NETWORK_VIEWER_PORT); | |
283 | if (ret < 0) { | |
284 | PERROR("asprintf default viewer control address"); | |
285 | goto exit; | |
286 | } | |
287 | ||
288 | ret = uri_parse(default_address, &live_uri); | |
289 | free(default_address); | |
290 | if (ret < 0) { | |
291 | ERR("Invalid viewer control URI specified"); | |
292 | goto exit; | |
293 | } | |
294 | } | |
b8aa1682 JD |
295 | |
296 | exit: | |
297 | return ret; | |
298 | } | |
299 | ||
300 | /* | |
301 | * Cleanup the daemon | |
302 | */ | |
303 | static | |
304 | void cleanup(void) | |
305 | { | |
b8aa1682 JD |
306 | DBG("Cleaning up"); |
307 | ||
095a4ae5 MD |
308 | /* free the dynamically allocated opt_output_path */ |
309 | free(opt_output_path); | |
310 | ||
a02de639 CB |
311 | /* Close thread quit pipes */ |
312 | utils_close_pipe(thread_quit_pipe); | |
313 | ||
710c1f73 DG |
314 | uri_free(control_uri); |
315 | uri_free(data_uri); | |
a6a062a4 | 316 | /* Live URI is freed in the live thread. */ |
b8aa1682 JD |
317 | } |
318 | ||
319 | /* | |
320 | * Write to writable pipe used to notify a thread. | |
321 | */ | |
322 | static | |
323 | int notify_thread_pipe(int wpipe) | |
324 | { | |
6cd525e8 | 325 | ssize_t ret; |
b8aa1682 | 326 | |
6cd525e8 MD |
327 | ret = lttng_write(wpipe, "!", 1); |
328 | if (ret < 1) { | |
b8aa1682 JD |
329 | PERROR("write poll pipe"); |
330 | } | |
331 | ||
332 | return ret; | |
333 | } | |
334 | ||
65931c8b MD |
335 | static void notify_health_quit_pipe(int *pipe) |
336 | { | |
6cd525e8 | 337 | ssize_t ret; |
65931c8b | 338 | |
6cd525e8 MD |
339 | ret = lttng_write(pipe[1], "4", 1); |
340 | if (ret < 1) { | |
65931c8b MD |
341 | PERROR("write relay health quit"); |
342 | } | |
343 | } | |
344 | ||
b8aa1682 JD |
345 | /* |
346 | * Stop all threads by closing the thread quit pipe. | |
347 | */ | |
348 | static | |
349 | void stop_threads(void) | |
350 | { | |
351 | int ret; | |
352 | ||
353 | /* Stopping all threads */ | |
354 | DBG("Terminating all threads"); | |
355 | ret = notify_thread_pipe(thread_quit_pipe[1]); | |
356 | if (ret < 0) { | |
357 | ERR("write error on thread quit pipe"); | |
358 | } | |
359 | ||
65931c8b MD |
360 | notify_health_quit_pipe(health_quit_pipe); |
361 | ||
b8aa1682 | 362 | /* Dispatch thread */ |
26c9d55e | 363 | CMM_STORE_SHARED(dispatch_thread_exit, 1); |
b8aa1682 JD |
364 | futex_nto1_wake(&relay_cmd_queue.futex); |
365 | } | |
366 | ||
367 | /* | |
368 | * Signal handler for the daemon | |
369 | * | |
370 | * Simply stop all worker threads, leaving main() return gracefully after | |
371 | * joining all threads and calling cleanup(). | |
372 | */ | |
373 | static | |
374 | void sighandler(int sig) | |
375 | { | |
376 | switch (sig) { | |
377 | case SIGPIPE: | |
378 | DBG("SIGPIPE caught"); | |
379 | return; | |
380 | case SIGINT: | |
381 | DBG("SIGINT caught"); | |
382 | stop_threads(); | |
383 | break; | |
384 | case SIGTERM: | |
385 | DBG("SIGTERM caught"); | |
386 | stop_threads(); | |
387 | break; | |
02a6bb53 MD |
388 | case SIGUSR1: |
389 | CMM_STORE_SHARED(recv_child_signal, 1); | |
390 | break; | |
b8aa1682 JD |
391 | default: |
392 | break; | |
393 | } | |
394 | } | |
395 | ||
396 | /* | |
397 | * Setup signal handler for : | |
398 | * SIGINT, SIGTERM, SIGPIPE | |
399 | */ | |
400 | static | |
401 | int set_signal_handler(void) | |
402 | { | |
403 | int ret = 0; | |
404 | struct sigaction sa; | |
405 | sigset_t sigset; | |
406 | ||
407 | if ((ret = sigemptyset(&sigset)) < 0) { | |
408 | PERROR("sigemptyset"); | |
409 | return ret; | |
410 | } | |
411 | ||
412 | sa.sa_handler = sighandler; | |
413 | sa.sa_mask = sigset; | |
414 | sa.sa_flags = 0; | |
415 | if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) { | |
416 | PERROR("sigaction"); | |
417 | return ret; | |
418 | } | |
419 | ||
420 | if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) { | |
421 | PERROR("sigaction"); | |
422 | return ret; | |
423 | } | |
424 | ||
425 | if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) { | |
426 | PERROR("sigaction"); | |
427 | return ret; | |
428 | } | |
429 | ||
02a6bb53 MD |
430 | if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) { |
431 | PERROR("sigaction"); | |
432 | return ret; | |
433 | } | |
434 | ||
435 | DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT"); | |
b8aa1682 JD |
436 | |
437 | return ret; | |
438 | } | |
439 | ||
02a6bb53 MD |
440 | void lttng_relay_notify_ready(void) |
441 | { | |
442 | /* Notify the parent of the fork() process that we are ready. */ | |
443 | if (opt_daemon || opt_background) { | |
444 | if (uatomic_sub_return(<tng_relay_ready, 1) == 0) { | |
445 | kill(child_ppid, SIGUSR1); | |
446 | } | |
447 | } | |
448 | } | |
449 | ||
b8aa1682 JD |
450 | /* |
451 | * Init thread quit pipe. | |
452 | * | |
453 | * Return -1 on error or 0 if all pipes are created. | |
454 | */ | |
455 | static | |
456 | int init_thread_quit_pipe(void) | |
457 | { | |
a02de639 | 458 | int ret; |
b8aa1682 | 459 | |
a02de639 | 460 | ret = utils_create_pipe_cloexec(thread_quit_pipe); |
b8aa1682 | 461 | |
b8aa1682 JD |
462 | return ret; |
463 | } | |
464 | ||
465 | /* | |
466 | * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set. | |
467 | */ | |
468 | static | |
469 | int create_thread_poll_set(struct lttng_poll_event *events, int size) | |
470 | { | |
471 | int ret; | |
472 | ||
473 | if (events == NULL || size == 0) { | |
474 | ret = -1; | |
475 | goto error; | |
476 | } | |
477 | ||
478 | ret = lttng_poll_create(events, size, LTTNG_CLOEXEC); | |
479 | if (ret < 0) { | |
480 | goto error; | |
481 | } | |
482 | ||
483 | /* Add quit pipe */ | |
532ed517 | 484 | ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR); |
b8aa1682 JD |
485 | if (ret < 0) { |
486 | goto error; | |
487 | } | |
488 | ||
489 | return 0; | |
490 | ||
491 | error: | |
492 | return ret; | |
493 | } | |
494 | ||
495 | /* | |
496 | * Check if the thread quit pipe was triggered. | |
497 | * | |
498 | * Return 1 if it was triggered else 0; | |
499 | */ | |
500 | static | |
501 | int check_thread_quit_pipe(int fd, uint32_t events) | |
502 | { | |
503 | if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) { | |
504 | return 1; | |
505 | } | |
506 | ||
507 | return 0; | |
508 | } | |
509 | ||
510 | /* | |
511 | * Create and init socket from uri. | |
512 | */ | |
513 | static | |
514 | struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri) | |
515 | { | |
516 | int ret; | |
517 | struct lttcomm_sock *sock = NULL; | |
518 | ||
519 | sock = lttcomm_alloc_sock_from_uri(uri); | |
520 | if (sock == NULL) { | |
521 | ERR("Allocating socket"); | |
522 | goto error; | |
523 | } | |
524 | ||
525 | ret = lttcomm_create_sock(sock); | |
526 | if (ret < 0) { | |
527 | goto error; | |
528 | } | |
529 | DBG("Listening on sock %d", sock->fd); | |
530 | ||
531 | ret = sock->ops->bind(sock); | |
532 | if (ret < 0) { | |
533 | goto error; | |
534 | } | |
535 | ||
536 | ret = sock->ops->listen(sock, -1); | |
537 | if (ret < 0) { | |
538 | goto error; | |
539 | ||
540 | } | |
541 | ||
542 | return sock; | |
543 | ||
544 | error: | |
545 | if (sock) { | |
546 | lttcomm_destroy_sock(sock); | |
547 | } | |
548 | return NULL; | |
549 | } | |
550 | ||
173af62f DG |
551 | /* |
552 | * Return nonzero if stream needs to be closed. | |
553 | */ | |
554 | static | |
555 | int close_stream_check(struct relay_stream *stream) | |
556 | { | |
557 | ||
558 | if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) { | |
f7079f67 DG |
559 | /* |
560 | * We are about to close the stream so set the data pending flag to 1 | |
561 | * which will make the end data pending command skip the stream which | |
562 | * is now closed and ready. Note that after proceeding to a file close, | |
563 | * the written file is ready for reading. | |
564 | */ | |
565 | stream->data_pending_check_done = 1; | |
173af62f DG |
566 | return 1; |
567 | } | |
568 | return 0; | |
569 | } | |
570 | ||
b8aa1682 JD |
571 | /* |
572 | * This thread manages the listening for new connections on the network | |
573 | */ | |
574 | static | |
575 | void *relay_thread_listener(void *data) | |
576 | { | |
095a4ae5 | 577 | int i, ret, pollfd, err = -1; |
b8aa1682 JD |
578 | int val = 1; |
579 | uint32_t revents, nb_fd; | |
580 | struct lttng_poll_event events; | |
581 | struct lttcomm_sock *control_sock, *data_sock; | |
582 | ||
b8aa1682 JD |
583 | DBG("[thread] Relay listener started"); |
584 | ||
55706a7d MD |
585 | health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER); |
586 | ||
f385ae0a MD |
587 | health_code_update(); |
588 | ||
b8aa1682 JD |
589 | control_sock = relay_init_sock(control_uri); |
590 | if (!control_sock) { | |
095a4ae5 | 591 | goto error_sock_control; |
b8aa1682 JD |
592 | } |
593 | ||
594 | data_sock = relay_init_sock(data_uri); | |
595 | if (!data_sock) { | |
095a4ae5 | 596 | goto error_sock_relay; |
b8aa1682 JD |
597 | } |
598 | ||
599 | /* | |
600 | * Pass 3 as size here for the thread quit pipe, control and data socket. | |
601 | */ | |
602 | ret = create_thread_poll_set(&events, 3); | |
603 | if (ret < 0) { | |
604 | goto error_create_poll; | |
605 | } | |
606 | ||
607 | /* Add the control socket */ | |
608 | ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP); | |
609 | if (ret < 0) { | |
610 | goto error_poll_add; | |
611 | } | |
612 | ||
613 | /* Add the data socket */ | |
614 | ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP); | |
615 | if (ret < 0) { | |
616 | goto error_poll_add; | |
617 | } | |
618 | ||
02a6bb53 MD |
619 | lttng_relay_notify_ready(); |
620 | ||
4e5b8b82 MD |
621 | if (testpoint(relayd_thread_listener)) { |
622 | goto error_testpoint; | |
623 | } | |
624 | ||
b8aa1682 | 625 | while (1) { |
f385ae0a MD |
626 | health_code_update(); |
627 | ||
b8aa1682 JD |
628 | DBG("Listener accepting connections"); |
629 | ||
b8aa1682 | 630 | restart: |
f385ae0a | 631 | health_poll_entry(); |
b8aa1682 | 632 | ret = lttng_poll_wait(&events, -1); |
f385ae0a | 633 | health_poll_exit(); |
b8aa1682 JD |
634 | if (ret < 0) { |
635 | /* | |
636 | * Restart interrupted system call. | |
637 | */ | |
638 | if (errno == EINTR) { | |
639 | goto restart; | |
640 | } | |
641 | goto error; | |
642 | } | |
643 | ||
0d9c5d77 DG |
644 | nb_fd = ret; |
645 | ||
b8aa1682 JD |
646 | DBG("Relay new connection received"); |
647 | for (i = 0; i < nb_fd; i++) { | |
f385ae0a MD |
648 | health_code_update(); |
649 | ||
b8aa1682 JD |
650 | /* Fetch once the poll data */ |
651 | revents = LTTNG_POLL_GETEV(&events, i); | |
652 | pollfd = LTTNG_POLL_GETFD(&events, i); | |
653 | ||
654 | /* Thread quit pipe has been closed. Killing thread. */ | |
655 | ret = check_thread_quit_pipe(pollfd, revents); | |
656 | if (ret) { | |
095a4ae5 MD |
657 | err = 0; |
658 | goto exit; | |
b8aa1682 JD |
659 | } |
660 | ||
661 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
662 | ERR("socket poll error"); | |
663 | goto error; | |
664 | } else if (revents & LPOLLIN) { | |
4b7f17b2 MD |
665 | /* |
666 | * Get allocated in this thread, | |
667 | * enqueued to a global queue, dequeued | |
668 | * and freed in the worker thread. | |
669 | */ | |
670 | struct relay_command *relay_cmd; | |
671 | struct lttcomm_sock *newsock; | |
b8aa1682 JD |
672 | |
673 | relay_cmd = zmalloc(sizeof(struct relay_command)); | |
674 | if (relay_cmd == NULL) { | |
675 | PERROR("relay command zmalloc"); | |
676 | goto error; | |
677 | } | |
678 | ||
679 | if (pollfd == data_sock->fd) { | |
680 | newsock = data_sock->ops->accept(data_sock); | |
4b7f17b2 | 681 | if (!newsock) { |
b8aa1682 | 682 | PERROR("accepting data sock"); |
4b7f17b2 | 683 | free(relay_cmd); |
b8aa1682 JD |
684 | goto error; |
685 | } | |
686 | relay_cmd->type = RELAY_DATA; | |
687 | DBG("Relay data connection accepted, socket %d", newsock->fd); | |
4b7f17b2 MD |
688 | } else { |
689 | assert(pollfd == control_sock->fd); | |
b8aa1682 | 690 | newsock = control_sock->ops->accept(control_sock); |
4b7f17b2 | 691 | if (!newsock) { |
b8aa1682 | 692 | PERROR("accepting control sock"); |
4b7f17b2 | 693 | free(relay_cmd); |
b8aa1682 JD |
694 | goto error; |
695 | } | |
696 | relay_cmd->type = RELAY_CONTROL; | |
697 | DBG("Relay control connection accepted, socket %d", newsock->fd); | |
698 | } | |
699 | ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, | |
700 | &val, sizeof(int)); | |
701 | if (ret < 0) { | |
702 | PERROR("setsockopt inet"); | |
4b7f17b2 MD |
703 | lttcomm_destroy_sock(newsock); |
704 | free(relay_cmd); | |
b8aa1682 JD |
705 | goto error; |
706 | } | |
707 | relay_cmd->sock = newsock; | |
708 | /* | |
709 | * Lock free enqueue the request. | |
710 | */ | |
711 | cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node); | |
712 | ||
713 | /* | |
714 | * Wake the dispatch queue futex. Implicit memory | |
715 | * barrier with the exchange in cds_wfq_enqueue. | |
716 | */ | |
717 | futex_nto1_wake(&relay_cmd_queue.futex); | |
718 | } | |
719 | } | |
720 | } | |
721 | ||
095a4ae5 | 722 | exit: |
b8aa1682 JD |
723 | error: |
724 | error_poll_add: | |
4e5b8b82 | 725 | error_testpoint: |
b8aa1682 JD |
726 | lttng_poll_clean(&events); |
727 | error_create_poll: | |
095a4ae5 MD |
728 | if (data_sock->fd >= 0) { |
729 | ret = data_sock->ops->close(data_sock); | |
b8aa1682 JD |
730 | if (ret) { |
731 | PERROR("close"); | |
732 | } | |
b8aa1682 | 733 | } |
095a4ae5 MD |
734 | lttcomm_destroy_sock(data_sock); |
735 | error_sock_relay: | |
736 | if (control_sock->fd >= 0) { | |
737 | ret = control_sock->ops->close(control_sock); | |
b8aa1682 JD |
738 | if (ret) { |
739 | PERROR("close"); | |
740 | } | |
b8aa1682 | 741 | } |
095a4ae5 MD |
742 | lttcomm_destroy_sock(control_sock); |
743 | error_sock_control: | |
744 | if (err) { | |
f385ae0a MD |
745 | health_error(); |
746 | ERR("Health error occurred in %s", __func__); | |
095a4ae5 | 747 | } |
55706a7d | 748 | health_unregister(health_relayd); |
b8aa1682 JD |
749 | DBG("Relay listener thread cleanup complete"); |
750 | stop_threads(); | |
b8aa1682 JD |
751 | return NULL; |
752 | } | |
753 | ||
754 | /* | |
755 | * This thread manages the dispatching of the requests to worker threads | |
756 | */ | |
757 | static | |
758 | void *relay_thread_dispatcher(void *data) | |
759 | { | |
6cd525e8 MD |
760 | int err = -1; |
761 | ssize_t ret; | |
b8aa1682 JD |
762 | struct cds_wfq_node *node; |
763 | struct relay_command *relay_cmd = NULL; | |
764 | ||
765 | DBG("[thread] Relay dispatcher started"); | |
766 | ||
55706a7d MD |
767 | health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER); |
768 | ||
4e5b8b82 MD |
769 | if (testpoint(relayd_thread_dispatcher)) { |
770 | goto error_testpoint; | |
771 | } | |
772 | ||
f385ae0a MD |
773 | health_code_update(); |
774 | ||
26c9d55e | 775 | while (!CMM_LOAD_SHARED(dispatch_thread_exit)) { |
f385ae0a MD |
776 | health_code_update(); |
777 | ||
b8aa1682 JD |
778 | /* Atomically prepare the queue futex */ |
779 | futex_nto1_prepare(&relay_cmd_queue.futex); | |
780 | ||
781 | do { | |
f385ae0a MD |
782 | health_code_update(); |
783 | ||
b8aa1682 JD |
784 | /* Dequeue commands */ |
785 | node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue); | |
786 | if (node == NULL) { | |
787 | DBG("Woken up but nothing in the relay command queue"); | |
788 | /* Continue thread execution */ | |
789 | break; | |
790 | } | |
791 | ||
792 | relay_cmd = caa_container_of(node, struct relay_command, node); | |
793 | DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd); | |
794 | ||
795 | /* | |
796 | * Inform worker thread of the new request. This | |
797 | * call is blocking so we can be assured that the data will be read | |
798 | * at some point in time or wait to the end of the world :) | |
799 | */ | |
6cd525e8 MD |
800 | ret = lttng_write(relay_cmd_pipe[1], relay_cmd, |
801 | sizeof(struct relay_command)); | |
b8aa1682 | 802 | free(relay_cmd); |
6cd525e8 | 803 | if (ret < sizeof(struct relay_command)) { |
b8aa1682 JD |
804 | PERROR("write cmd pipe"); |
805 | goto error; | |
806 | } | |
807 | } while (node != NULL); | |
808 | ||
809 | /* Futex wait on queue. Blocking call on futex() */ | |
f385ae0a | 810 | health_poll_entry(); |
b8aa1682 | 811 | futex_nto1_wait(&relay_cmd_queue.futex); |
f385ae0a | 812 | health_poll_exit(); |
b8aa1682 JD |
813 | } |
814 | ||
f385ae0a MD |
815 | /* Normal exit, no error */ |
816 | err = 0; | |
817 | ||
b8aa1682 | 818 | error: |
4e5b8b82 | 819 | error_testpoint: |
f385ae0a MD |
820 | if (err) { |
821 | health_error(); | |
822 | ERR("Health error occurred in %s", __func__); | |
823 | } | |
55706a7d | 824 | health_unregister(health_relayd); |
b8aa1682 JD |
825 | DBG("Dispatch thread dying"); |
826 | stop_threads(); | |
827 | return NULL; | |
828 | } | |
829 | ||
de91f48a DG |
830 | /* |
831 | * Get stream from stream id. | |
832 | * Need to be called with RCU read-side lock held. | |
833 | */ | |
d3e2ba59 | 834 | struct relay_stream *relay_stream_find_by_id(uint64_t stream_id) |
de91f48a DG |
835 | { |
836 | struct lttng_ht_node_ulong *node; | |
837 | struct lttng_ht_iter iter; | |
838 | struct relay_stream *ret; | |
839 | ||
d3e2ba59 | 840 | lttng_ht_lookup(relay_streams_ht, |
de91f48a DG |
841 | (void *)((unsigned long) stream_id), |
842 | &iter); | |
843 | node = lttng_ht_iter_get_node_ulong(&iter); | |
844 | if (node == NULL) { | |
845 | DBG("Relay stream %" PRIu64 " not found", stream_id); | |
846 | ret = NULL; | |
847 | goto end; | |
848 | } | |
849 | ||
850 | ret = caa_container_of(node, struct relay_stream, stream_n); | |
851 | ||
852 | end: | |
853 | return ret; | |
854 | } | |
855 | ||
9d1bbf21 MD |
856 | static |
857 | void deferred_free_stream(struct rcu_head *head) | |
858 | { | |
859 | struct relay_stream *stream = | |
860 | caa_container_of(head, struct relay_stream, rcu_node); | |
d3e2ba59 | 861 | |
0f907de1 JD |
862 | free(stream->path_name); |
863 | free(stream->channel_name); | |
9d1bbf21 MD |
864 | free(stream); |
865 | } | |
866 | ||
d3e2ba59 JD |
867 | static |
868 | void deferred_free_session(struct rcu_head *head) | |
869 | { | |
870 | struct relay_session *session = | |
871 | caa_container_of(head, struct relay_session, rcu_node); | |
872 | free(session); | |
873 | } | |
874 | ||
4c80d9a2 DG |
875 | /* |
876 | * Close a given stream. The stream is freed using a call RCU. | |
877 | * | |
878 | * RCU read side lock MUST be acquired. If NO close_stream_check() was called | |
879 | * BEFORE the stream lock MUST be acquired. | |
880 | */ | |
c07d8a78 | 881 | static void destroy_stream(struct relay_stream *stream) |
94d49140 JD |
882 | { |
883 | int delret; | |
884 | struct relay_viewer_stream *vstream; | |
885 | struct lttng_ht_iter iter; | |
886 | ||
887 | assert(stream); | |
94d49140 JD |
888 | |
889 | delret = close(stream->fd); | |
890 | if (delret < 0) { | |
891 | PERROR("close stream"); | |
892 | } | |
893 | ||
894 | if (stream->index_fd >= 0) { | |
895 | delret = close(stream->index_fd); | |
896 | if (delret < 0) { | |
897 | PERROR("close stream index_fd"); | |
898 | } | |
899 | } | |
900 | ||
92c6ca54 | 901 | vstream = live_find_viewer_stream_by_id(stream->stream_handle); |
94d49140 JD |
902 | if (vstream) { |
903 | /* | |
904 | * Set the last good value into the viewer stream. This is done | |
905 | * right before the stream gets deleted from the hash table. The | |
906 | * lookup failure on the live thread side of a stream indicates | |
907 | * that the viewer stream index received value should be used. | |
908 | */ | |
cef0f7d5 | 909 | pthread_mutex_lock(&stream->viewer_stream_rotation_lock); |
94d49140 | 910 | vstream->total_index_received = stream->total_index_received; |
a020f610 | 911 | vstream->tracefile_count_last = stream->tracefile_count_current; |
cef0f7d5 JD |
912 | vstream->close_write_flag = 1; |
913 | pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); | |
94d49140 JD |
914 | } |
915 | ||
efc8a87f DG |
916 | /* Cleanup index of that stream. */ |
917 | relay_index_destroy_by_stream_id(stream->stream_handle); | |
918 | ||
94d49140 JD |
919 | iter.iter.node = &stream->stream_n.node; |
920 | delret = lttng_ht_del(relay_streams_ht, &iter); | |
921 | assert(!delret); | |
922 | iter.iter.node = &stream->ctf_trace_node.node; | |
c07d8a78 | 923 | delret = lttng_ht_del(stream->ctf_traces_ht, &iter); |
94d49140 | 924 | assert(!delret); |
157df586 JD |
925 | |
926 | if (stream->ctf_trace) { | |
927 | ctf_trace_try_destroy(stream->ctf_trace); | |
928 | } | |
929 | ||
94d49140 JD |
930 | call_rcu(&stream->rcu_node, deferred_free_stream); |
931 | DBG("Closed tracefile %d from close stream", stream->fd); | |
932 | } | |
933 | ||
b8aa1682 JD |
934 | /* |
935 | * relay_delete_session: Free all memory associated with a session and | |
936 | * close all the FDs | |
937 | */ | |
938 | static | |
d3e2ba59 JD |
939 | void relay_delete_session(struct relay_command *cmd, |
940 | struct lttng_ht *sessions_ht) | |
b8aa1682 JD |
941 | { |
942 | struct lttng_ht_iter iter; | |
943 | struct lttng_ht_node_ulong *node; | |
944 | struct relay_stream *stream; | |
945 | int ret; | |
946 | ||
095a4ae5 | 947 | if (!cmd->session) { |
b8aa1682 | 948 | return; |
095a4ae5 | 949 | } |
b8aa1682 | 950 | |
77c7c900 | 951 | DBG("Relay deleting session %" PRIu64, cmd->session->id); |
5b6d8097 | 952 | |
9d1bbf21 | 953 | rcu_read_lock(); |
d3e2ba59 | 954 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, node, node) { |
b8aa1682 | 955 | node = lttng_ht_iter_get_node_ulong(&iter); |
4c80d9a2 DG |
956 | if (!node) { |
957 | continue; | |
b8aa1682 | 958 | } |
4c80d9a2 DG |
959 | stream = caa_container_of(node, struct relay_stream, stream_n); |
960 | if (stream->session == cmd->session) { | |
c07d8a78 | 961 | destroy_stream(stream); |
87b576ec JD |
962 | cmd->session->stream_count--; |
963 | assert(cmd->session->stream_count >= 0); | |
4c80d9a2 | 964 | } |
b8aa1682 | 965 | } |
4c80d9a2 DG |
966 | |
967 | /* Make this session not visible anymore. */ | |
d3e2ba59 JD |
968 | iter.iter.node = &cmd->session->session_n.node; |
969 | ret = lttng_ht_del(sessions_ht, &iter); | |
970 | assert(!ret); | |
4c80d9a2 | 971 | call_rcu(&cmd->session->rcu_node, deferred_free_session); |
9d1bbf21 | 972 | rcu_read_unlock(); |
b8aa1682 JD |
973 | } |
974 | ||
1c20f0e2 JD |
975 | /* |
976 | * Copy index data from the control port to a given index object. | |
977 | */ | |
978 | static void copy_index_control_data(struct relay_index *index, | |
979 | struct lttcomm_relayd_index *data) | |
980 | { | |
981 | assert(index); | |
982 | assert(data); | |
983 | ||
984 | /* | |
985 | * The index on disk is encoded in big endian, so we don't need to convert | |
986 | * the data received on the network. The data_offset value is NEVER | |
987 | * modified here and is updated by the data thread. | |
988 | */ | |
989 | index->index_data.packet_size = data->packet_size; | |
990 | index->index_data.content_size = data->content_size; | |
991 | index->index_data.timestamp_begin = data->timestamp_begin; | |
992 | index->index_data.timestamp_end = data->timestamp_end; | |
993 | index->index_data.events_discarded = data->events_discarded; | |
994 | index->index_data.stream_id = data->stream_id; | |
995 | } | |
996 | ||
c5b6f4f0 DG |
997 | /* |
998 | * Handle the RELAYD_CREATE_SESSION command. | |
999 | * | |
1000 | * On success, send back the session id or else return a negative value. | |
1001 | */ | |
1002 | static | |
1003 | int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 JD |
1004 | struct relay_command *cmd, |
1005 | struct lttng_ht *sessions_ht) | |
c5b6f4f0 DG |
1006 | { |
1007 | int ret = 0, send_ret; | |
1008 | struct relay_session *session; | |
1009 | struct lttcomm_relayd_status_session reply; | |
1010 | ||
1011 | assert(recv_hdr); | |
1012 | assert(cmd); | |
1013 | ||
1014 | memset(&reply, 0, sizeof(reply)); | |
1015 | ||
1016 | session = zmalloc(sizeof(struct relay_session)); | |
1017 | if (session == NULL) { | |
1018 | PERROR("relay session zmalloc"); | |
1019 | ret = -1; | |
1020 | goto error; | |
1021 | } | |
1022 | ||
1023 | session->id = ++last_relay_session_id; | |
1024 | session->sock = cmd->sock; | |
7d2f7452 DG |
1025 | session->minor = cmd->minor; |
1026 | session->major = cmd->major; | |
c5b6f4f0 DG |
1027 | cmd->session = session; |
1028 | ||
1029 | reply.session_id = htobe64(session->id); | |
1030 | ||
d3e2ba59 | 1031 | switch (cmd->minor) { |
b0dd7a75 JD |
1032 | case 1: |
1033 | case 2: | |
1034 | case 3: | |
1035 | break; | |
d3e2ba59 JD |
1036 | case 4: /* LTTng sessiond 2.4 */ |
1037 | default: | |
1038 | ret = cmd_create_session_2_4(cmd, session); | |
1039 | break; | |
1040 | } | |
1041 | ||
1042 | lttng_ht_node_init_ulong(&session->session_n, | |
1043 | (unsigned long) session->id); | |
1044 | lttng_ht_add_unique_ulong(sessions_ht, | |
1045 | &session->session_n); | |
1046 | ||
c5b6f4f0 DG |
1047 | DBG("Created session %" PRIu64, session->id); |
1048 | ||
1049 | error: | |
1050 | if (ret < 0) { | |
1051 | reply.ret_code = htobe32(LTTNG_ERR_FATAL); | |
1052 | } else { | |
1053 | reply.ret_code = htobe32(LTTNG_OK); | |
1054 | } | |
1055 | ||
1056 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1057 | if (send_ret < 0) { | |
1058 | ERR("Relayd sending session id"); | |
4169f5ad | 1059 | ret = send_ret; |
c5b6f4f0 DG |
1060 | } |
1061 | ||
1062 | return ret; | |
1063 | } | |
1064 | ||
814fcae4 JD |
1065 | /* |
1066 | * When we have received all the streams and the metadata for a channel, | |
1067 | * we make them visible to the viewer threads. | |
1068 | */ | |
1069 | static | |
1070 | void set_viewer_ready_flag(struct relay_command *cmd) | |
1071 | { | |
1072 | struct relay_stream_recv_handle *node, *tmp_node; | |
1073 | ||
1074 | cds_list_for_each_entry_safe(node, tmp_node, &cmd->recv_head, node) { | |
1075 | struct relay_stream *stream; | |
1076 | ||
1077 | rcu_read_lock(); | |
1078 | stream = relay_stream_find_by_id(node->id); | |
1079 | if (!stream) { | |
1080 | /* | |
1081 | * Stream is most probably being cleaned up by the data thread thus | |
1082 | * simply continue to the next one. | |
1083 | */ | |
f188956b | 1084 | rcu_read_unlock(); |
814fcae4 JD |
1085 | continue; |
1086 | } | |
1087 | ||
814fcae4 JD |
1088 | stream->viewer_ready = 1; |
1089 | rcu_read_unlock(); | |
1090 | ||
1091 | /* Clean stream handle node. */ | |
1092 | cds_list_del(&node->node); | |
1093 | free(node); | |
1094 | } | |
1095 | ||
814fcae4 JD |
1096 | return; |
1097 | } | |
1098 | ||
1099 | /* | |
1100 | * Add a recv handle node to the connection recv list with the given stream | |
1101 | * handle. A new node is allocated thus must be freed when the node is deleted | |
1102 | * from the list. | |
1103 | */ | |
1104 | static void queue_stream_handle(uint64_t handle, struct relay_command *cmd) | |
1105 | { | |
1106 | struct relay_stream_recv_handle *node; | |
1107 | ||
1108 | assert(cmd); | |
1109 | ||
1110 | node = zmalloc(sizeof(*node)); | |
1111 | if (!node) { | |
1112 | PERROR("zmalloc queue stream handle"); | |
1113 | return; | |
1114 | } | |
1115 | ||
1116 | node->id = handle; | |
1117 | cds_list_add(&node->node, &cmd->recv_head); | |
1118 | } | |
1119 | ||
b8aa1682 JD |
1120 | /* |
1121 | * relay_add_stream: allocate a new stream for a session | |
1122 | */ | |
1123 | static | |
1124 | int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1125 | struct relay_command *cmd, struct lttng_ht *sessions_ht) |
b8aa1682 JD |
1126 | { |
1127 | struct relay_session *session = cmd->session; | |
b8aa1682 JD |
1128 | struct relay_stream *stream = NULL; |
1129 | struct lttcomm_relayd_status_stream reply; | |
b8aa1682 JD |
1130 | int ret, send_ret; |
1131 | ||
c5b6f4f0 | 1132 | if (!session || cmd->version_check_done == 0) { |
b8aa1682 JD |
1133 | ERR("Trying to add a stream before version check"); |
1134 | ret = -1; | |
1135 | goto end_no_session; | |
1136 | } | |
1137 | ||
b8aa1682 JD |
1138 | stream = zmalloc(sizeof(struct relay_stream)); |
1139 | if (stream == NULL) { | |
1140 | PERROR("relay stream zmalloc"); | |
1141 | ret = -1; | |
1142 | goto end_no_session; | |
1143 | } | |
1144 | ||
0f907de1 JD |
1145 | switch (cmd->minor) { |
1146 | case 1: /* LTTng sessiond 2.1 */ | |
1147 | ret = cmd_recv_stream_2_1(cmd, stream); | |
1148 | break; | |
1149 | case 2: /* LTTng sessiond 2.2 */ | |
1150 | default: | |
1151 | ret = cmd_recv_stream_2_2(cmd, stream); | |
1152 | break; | |
1153 | } | |
1154 | if (ret < 0) { | |
1155 | goto err_free_stream; | |
1156 | } | |
1157 | ||
9d1bbf21 | 1158 | rcu_read_lock(); |
b8aa1682 | 1159 | stream->stream_handle = ++last_relay_stream_id; |
173af62f | 1160 | stream->prev_seq = -1ULL; |
b8aa1682 | 1161 | stream->session = session; |
1c20f0e2 | 1162 | stream->index_fd = -1; |
d3e2ba59 JD |
1163 | stream->read_index_fd = -1; |
1164 | stream->ctf_trace = NULL; | |
1165 | pthread_mutex_init(&stream->lock, NULL); | |
b8aa1682 | 1166 | |
0f907de1 | 1167 | ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG); |
b8aa1682 | 1168 | if (ret < 0) { |
b8aa1682 JD |
1169 | ERR("relay creating output directory"); |
1170 | goto end; | |
1171 | } | |
1172 | ||
be96a7d1 DG |
1173 | /* |
1174 | * No need to use run_as API here because whatever we receives, the relayd | |
1175 | * uses its own credentials for the stream files. | |
1176 | */ | |
0f907de1 | 1177 | ret = utils_create_stream_file(stream->path_name, stream->channel_name, |
1c20f0e2 | 1178 | stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL); |
b8aa1682 | 1179 | if (ret < 0) { |
0f907de1 | 1180 | ERR("Create output file"); |
b8aa1682 JD |
1181 | goto end; |
1182 | } | |
b8aa1682 | 1183 | stream->fd = ret; |
0f907de1 JD |
1184 | if (stream->tracefile_size) { |
1185 | DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name); | |
1186 | } else { | |
1187 | DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name); | |
1188 | } | |
b8aa1682 | 1189 | |
d3e2ba59 JD |
1190 | if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) { |
1191 | stream->metadata_flag = 1; | |
1192 | /* | |
1193 | * When we receive a new metadata stream, we create a new | |
1194 | * ctf_trace and we assign this ctf_trace to all streams with | |
1195 | * the same path. | |
1196 | * | |
1197 | * If later on we receive a new stream for the same ctf_trace, | |
1198 | * we copy the information from the first hit in the HT to the | |
1199 | * new stream. | |
1200 | */ | |
1201 | stream->ctf_trace = ctf_trace_create(); | |
1202 | if (!stream->ctf_trace) { | |
1203 | ret = -1; | |
1204 | goto end; | |
1205 | } | |
1206 | stream->ctf_trace->refcount++; | |
1207 | stream->ctf_trace->metadata_stream = stream; | |
1208 | } | |
1209 | ctf_trace_assign(cmd->ctf_traces_ht, stream); | |
c07d8a78 | 1210 | stream->ctf_traces_ht = cmd->ctf_traces_ht; |
d3e2ba59 | 1211 | |
814fcae4 JD |
1212 | /* |
1213 | * Add the stream handle in the recv list of the connection. Once the end | |
1214 | * stream message is received, this list is emptied and streams are set | |
1215 | * with the viewer ready flag. | |
1216 | */ | |
f188956b JD |
1217 | if (stream->metadata_flag) { |
1218 | stream->viewer_ready = 1; | |
1219 | } else { | |
1220 | queue_stream_handle(stream->stream_handle, cmd); | |
1221 | } | |
814fcae4 | 1222 | |
b8aa1682 JD |
1223 | lttng_ht_node_init_ulong(&stream->stream_n, |
1224 | (unsigned long) stream->stream_handle); | |
d3e2ba59 | 1225 | lttng_ht_add_unique_ulong(relay_streams_ht, |
b8aa1682 JD |
1226 | &stream->stream_n); |
1227 | ||
d3e2ba59 JD |
1228 | lttng_ht_node_init_str(&stream->ctf_trace_node, stream->path_name); |
1229 | lttng_ht_add_str(cmd->ctf_traces_ht, &stream->ctf_trace_node); | |
87b576ec | 1230 | session->stream_count++; |
d3e2ba59 | 1231 | |
1c20f0e2 JD |
1232 | DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name, |
1233 | stream->stream_handle); | |
b8aa1682 JD |
1234 | |
1235 | end: | |
5af40280 | 1236 | reply.handle = htobe64(stream->stream_handle); |
b8aa1682 JD |
1237 | /* send the session id to the client or a negative return code on error */ |
1238 | if (ret < 0) { | |
f73fabfd | 1239 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
5af40280 CB |
1240 | /* stream was not properly added to the ht, so free it */ |
1241 | free(stream); | |
b8aa1682 | 1242 | } else { |
f73fabfd | 1243 | reply.ret_code = htobe32(LTTNG_OK); |
b8aa1682 | 1244 | } |
5af40280 | 1245 | |
b8aa1682 JD |
1246 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, |
1247 | sizeof(struct lttcomm_relayd_status_stream), 0); | |
1248 | if (send_ret < 0) { | |
1249 | ERR("Relay sending stream id"); | |
4169f5ad | 1250 | ret = send_ret; |
b8aa1682 | 1251 | } |
9d1bbf21 | 1252 | rcu_read_unlock(); |
b8aa1682 JD |
1253 | |
1254 | end_no_session: | |
1255 | return ret; | |
0f907de1 JD |
1256 | |
1257 | err_free_stream: | |
1258 | free(stream->path_name); | |
1259 | free(stream->channel_name); | |
1260 | free(stream); | |
1261 | return ret; | |
b8aa1682 JD |
1262 | } |
1263 | ||
173af62f DG |
1264 | /* |
1265 | * relay_close_stream: close a specific stream | |
1266 | */ | |
1267 | static | |
1268 | int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr, | |
92c6ca54 | 1269 | struct relay_command *cmd) |
173af62f | 1270 | { |
94d49140 | 1271 | int ret, send_ret; |
173af62f DG |
1272 | struct relay_session *session = cmd->session; |
1273 | struct lttcomm_relayd_close_stream stream_info; | |
1274 | struct lttcomm_relayd_generic_reply reply; | |
1275 | struct relay_stream *stream; | |
173af62f DG |
1276 | |
1277 | DBG("Close stream received"); | |
1278 | ||
c5b6f4f0 | 1279 | if (!session || cmd->version_check_done == 0) { |
173af62f DG |
1280 | ERR("Trying to close a stream before version check"); |
1281 | ret = -1; | |
1282 | goto end_no_session; | |
1283 | } | |
1284 | ||
1285 | ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info, | |
7c5aef62 | 1286 | sizeof(struct lttcomm_relayd_close_stream), 0); |
173af62f | 1287 | if (ret < sizeof(struct lttcomm_relayd_close_stream)) { |
a6cd2b97 DG |
1288 | if (ret == 0) { |
1289 | /* Orderly shutdown. Not necessary to print an error. */ | |
1290 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1291 | } else { | |
1292 | ERR("Relay didn't receive valid add_stream struct size : %d", ret); | |
1293 | } | |
173af62f DG |
1294 | ret = -1; |
1295 | goto end_no_session; | |
1296 | } | |
1297 | ||
1298 | rcu_read_lock(); | |
d3e2ba59 | 1299 | stream = relay_stream_find_by_id(be64toh(stream_info.stream_id)); |
173af62f DG |
1300 | if (!stream) { |
1301 | ret = -1; | |
1302 | goto end_unlock; | |
1303 | } | |
1304 | ||
8e2583a4 | 1305 | stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num); |
173af62f | 1306 | stream->close_flag = 1; |
87b576ec JD |
1307 | session->stream_count--; |
1308 | assert(session->stream_count >= 0); | |
173af62f DG |
1309 | |
1310 | if (close_stream_check(stream)) { | |
c07d8a78 | 1311 | destroy_stream(stream); |
173af62f DG |
1312 | } |
1313 | ||
1314 | end_unlock: | |
1315 | rcu_read_unlock(); | |
1316 | ||
1317 | if (ret < 0) { | |
f73fabfd | 1318 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
173af62f | 1319 | } else { |
f73fabfd | 1320 | reply.ret_code = htobe32(LTTNG_OK); |
173af62f DG |
1321 | } |
1322 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1323 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1324 | if (send_ret < 0) { | |
1325 | ERR("Relay sending stream id"); | |
4169f5ad | 1326 | ret = send_ret; |
173af62f DG |
1327 | } |
1328 | ||
1329 | end_no_session: | |
1330 | return ret; | |
1331 | } | |
1332 | ||
b8aa1682 JD |
1333 | /* |
1334 | * relay_unknown_command: send -1 if received unknown command | |
1335 | */ | |
1336 | static | |
1337 | void relay_unknown_command(struct relay_command *cmd) | |
1338 | { | |
1339 | struct lttcomm_relayd_generic_reply reply; | |
1340 | int ret; | |
1341 | ||
f73fabfd | 1342 | reply.ret_code = htobe32(LTTNG_ERR_UNK); |
b8aa1682 JD |
1343 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, |
1344 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1345 | if (ret < 0) { | |
1346 | ERR("Relay sending unknown command"); | |
1347 | } | |
1348 | } | |
1349 | ||
1350 | /* | |
1351 | * relay_start: send an acknowledgment to the client to tell if we are | |
1352 | * ready to receive data. We are ready if a session is established. | |
1353 | */ | |
1354 | static | |
1355 | int relay_start(struct lttcomm_relayd_hdr *recv_hdr, | |
1356 | struct relay_command *cmd) | |
1357 | { | |
f73fabfd | 1358 | int ret = htobe32(LTTNG_OK); |
b8aa1682 JD |
1359 | struct lttcomm_relayd_generic_reply reply; |
1360 | struct relay_session *session = cmd->session; | |
1361 | ||
1362 | if (!session) { | |
1363 | DBG("Trying to start the streaming without a session established"); | |
f73fabfd | 1364 | ret = htobe32(LTTNG_ERR_UNK); |
b8aa1682 JD |
1365 | } |
1366 | ||
1367 | reply.ret_code = ret; | |
1368 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1369 | sizeof(struct lttcomm_relayd_generic_reply), 0); | |
1370 | if (ret < 0) { | |
1371 | ERR("Relay sending start ack"); | |
1372 | } | |
1373 | ||
1374 | return ret; | |
1375 | } | |
1376 | ||
1d4dfdef DG |
1377 | /* |
1378 | * Append padding to the file pointed by the file descriptor fd. | |
1379 | */ | |
1380 | static int write_padding_to_file(int fd, uint32_t size) | |
1381 | { | |
6cd525e8 | 1382 | ssize_t ret = 0; |
1d4dfdef DG |
1383 | char *zeros; |
1384 | ||
1385 | if (size == 0) { | |
1386 | goto end; | |
1387 | } | |
1388 | ||
1389 | zeros = zmalloc(size); | |
1390 | if (zeros == NULL) { | |
1391 | PERROR("zmalloc zeros for padding"); | |
1392 | ret = -1; | |
1393 | goto end; | |
1394 | } | |
1395 | ||
6cd525e8 MD |
1396 | ret = lttng_write(fd, zeros, size); |
1397 | if (ret < size) { | |
1d4dfdef DG |
1398 | PERROR("write padding to file"); |
1399 | } | |
1400 | ||
e986c7a1 DG |
1401 | free(zeros); |
1402 | ||
1d4dfdef DG |
1403 | end: |
1404 | return ret; | |
1405 | } | |
1406 | ||
b8aa1682 JD |
1407 | /* |
1408 | * relay_recv_metadata: receive the metada for the session. | |
1409 | */ | |
1410 | static | |
1411 | int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1412 | struct relay_command *cmd) |
b8aa1682 | 1413 | { |
f73fabfd | 1414 | int ret = htobe32(LTTNG_OK); |
6cd525e8 | 1415 | ssize_t size_ret; |
b8aa1682 JD |
1416 | struct relay_session *session = cmd->session; |
1417 | struct lttcomm_relayd_metadata_payload *metadata_struct; | |
1418 | struct relay_stream *metadata_stream; | |
1419 | uint64_t data_size, payload_size; | |
1420 | ||
1421 | if (!session) { | |
1422 | ERR("Metadata sent before version check"); | |
1423 | ret = -1; | |
1424 | goto end; | |
1425 | } | |
1426 | ||
f6416125 MD |
1427 | data_size = payload_size = be64toh(recv_hdr->data_size); |
1428 | if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) { | |
1429 | ERR("Incorrect data size"); | |
1430 | ret = -1; | |
1431 | goto end; | |
1432 | } | |
1433 | payload_size -= sizeof(struct lttcomm_relayd_metadata_payload); | |
1434 | ||
b8aa1682 | 1435 | if (data_buffer_size < data_size) { |
d7b3776f | 1436 | /* In case the realloc fails, we can free the memory */ |
c617c0c6 MD |
1437 | char *tmp_data_ptr; |
1438 | ||
1439 | tmp_data_ptr = realloc(data_buffer, data_size); | |
1440 | if (!tmp_data_ptr) { | |
b8aa1682 | 1441 | ERR("Allocating data buffer"); |
c617c0c6 | 1442 | free(data_buffer); |
b8aa1682 JD |
1443 | ret = -1; |
1444 | goto end; | |
1445 | } | |
c617c0c6 | 1446 | data_buffer = tmp_data_ptr; |
b8aa1682 JD |
1447 | data_buffer_size = data_size; |
1448 | } | |
1449 | memset(data_buffer, 0, data_size); | |
77c7c900 | 1450 | DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size); |
7c5aef62 | 1451 | ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0); |
b8aa1682 | 1452 | if (ret < 0 || ret != data_size) { |
a6cd2b97 DG |
1453 | if (ret == 0) { |
1454 | /* Orderly shutdown. Not necessary to print an error. */ | |
1455 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1456 | } else { | |
1457 | ERR("Relay didn't receive the whole metadata"); | |
1458 | } | |
b8aa1682 | 1459 | ret = -1; |
b8aa1682 JD |
1460 | goto end; |
1461 | } | |
1462 | metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer; | |
9d1bbf21 MD |
1463 | |
1464 | rcu_read_lock(); | |
d3e2ba59 JD |
1465 | metadata_stream = relay_stream_find_by_id( |
1466 | be64toh(metadata_struct->stream_id)); | |
b8aa1682 JD |
1467 | if (!metadata_stream) { |
1468 | ret = -1; | |
9d1bbf21 | 1469 | goto end_unlock; |
b8aa1682 JD |
1470 | } |
1471 | ||
6cd525e8 MD |
1472 | size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload, |
1473 | payload_size); | |
1474 | if (size_ret < payload_size) { | |
b8aa1682 JD |
1475 | ERR("Relay error writing metadata on file"); |
1476 | ret = -1; | |
9d1bbf21 | 1477 | goto end_unlock; |
b8aa1682 | 1478 | } |
1d4dfdef DG |
1479 | |
1480 | ret = write_padding_to_file(metadata_stream->fd, | |
1481 | be32toh(metadata_struct->padding_size)); | |
1482 | if (ret < 0) { | |
1483 | goto end_unlock; | |
1484 | } | |
d3e2ba59 JD |
1485 | metadata_stream->ctf_trace->metadata_received += |
1486 | payload_size + be32toh(metadata_struct->padding_size); | |
1d4dfdef | 1487 | |
b8aa1682 JD |
1488 | DBG2("Relay metadata written"); |
1489 | ||
9d1bbf21 | 1490 | end_unlock: |
6e3c5836 | 1491 | rcu_read_unlock(); |
b8aa1682 JD |
1492 | end: |
1493 | return ret; | |
1494 | } | |
1495 | ||
1496 | /* | |
1497 | * relay_send_version: send relayd version number | |
1498 | */ | |
1499 | static | |
1500 | int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1501 | struct relay_command *cmd, struct lttng_ht *sessions_ht) |
b8aa1682 | 1502 | { |
7f51dcba | 1503 | int ret; |
092b6259 | 1504 | struct lttcomm_relayd_version reply, msg; |
b8aa1682 | 1505 | |
c5b6f4f0 DG |
1506 | assert(cmd); |
1507 | ||
1508 | cmd->version_check_done = 1; | |
b8aa1682 | 1509 | |
092b6259 | 1510 | /* Get version from the other side. */ |
7c5aef62 | 1511 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); |
092b6259 | 1512 | if (ret < 0 || ret != sizeof(msg)) { |
a6cd2b97 DG |
1513 | if (ret == 0) { |
1514 | /* Orderly shutdown. Not necessary to print an error. */ | |
1515 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1516 | } else { | |
1517 | ERR("Relay failed to receive the version values."); | |
1518 | } | |
092b6259 | 1519 | ret = -1; |
092b6259 DG |
1520 | goto end; |
1521 | } | |
1522 | ||
d83a952c MD |
1523 | reply.major = RELAYD_VERSION_COMM_MAJOR; |
1524 | reply.minor = RELAYD_VERSION_COMM_MINOR; | |
d4519fa3 JD |
1525 | |
1526 | /* Major versions must be the same */ | |
1527 | if (reply.major != be32toh(msg.major)) { | |
6151a90f JD |
1528 | DBG("Incompatible major versions (%u vs %u), deleting session", |
1529 | reply.major, be32toh(msg.major)); | |
d3e2ba59 | 1530 | relay_delete_session(cmd, sessions_ht); |
d4519fa3 JD |
1531 | ret = 0; |
1532 | goto end; | |
1533 | } | |
1534 | ||
0f907de1 JD |
1535 | cmd->major = reply.major; |
1536 | /* We adapt to the lowest compatible version */ | |
1537 | if (reply.minor <= be32toh(msg.minor)) { | |
1538 | cmd->minor = reply.minor; | |
1539 | } else { | |
1540 | cmd->minor = be32toh(msg.minor); | |
1541 | } | |
1542 | ||
6151a90f JD |
1543 | reply.major = htobe32(reply.major); |
1544 | reply.minor = htobe32(reply.minor); | |
1545 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, | |
1546 | sizeof(struct lttcomm_relayd_version), 0); | |
1547 | if (ret < 0) { | |
1548 | ERR("Relay sending version"); | |
1549 | } | |
1550 | ||
0f907de1 JD |
1551 | DBG("Version check done using protocol %u.%u", cmd->major, |
1552 | cmd->minor); | |
b8aa1682 JD |
1553 | |
1554 | end: | |
1555 | return ret; | |
1556 | } | |
1557 | ||
c8f59ee5 | 1558 | /* |
6d805429 | 1559 | * Check for data pending for a given stream id from the session daemon. |
c8f59ee5 DG |
1560 | */ |
1561 | static | |
6d805429 | 1562 | int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr, |
d3e2ba59 | 1563 | struct relay_command *cmd) |
c8f59ee5 DG |
1564 | { |
1565 | struct relay_session *session = cmd->session; | |
6d805429 | 1566 | struct lttcomm_relayd_data_pending msg; |
c8f59ee5 DG |
1567 | struct lttcomm_relayd_generic_reply reply; |
1568 | struct relay_stream *stream; | |
1569 | int ret; | |
c8f59ee5 DG |
1570 | uint64_t last_net_seq_num, stream_id; |
1571 | ||
6d805429 | 1572 | DBG("Data pending command received"); |
c8f59ee5 | 1573 | |
c5b6f4f0 | 1574 | if (!session || cmd->version_check_done == 0) { |
c8f59ee5 DG |
1575 | ERR("Trying to check for data before version check"); |
1576 | ret = -1; | |
1577 | goto end_no_session; | |
1578 | } | |
1579 | ||
7c5aef62 | 1580 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); |
c8f59ee5 | 1581 | if (ret < sizeof(msg)) { |
a6cd2b97 DG |
1582 | if (ret == 0) { |
1583 | /* Orderly shutdown. Not necessary to print an error. */ | |
1584 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1585 | } else { | |
1586 | ERR("Relay didn't receive valid data_pending struct size : %d", | |
1587 | ret); | |
1588 | } | |
c8f59ee5 DG |
1589 | ret = -1; |
1590 | goto end_no_session; | |
1591 | } | |
1592 | ||
1593 | stream_id = be64toh(msg.stream_id); | |
1594 | last_net_seq_num = be64toh(msg.last_net_seq_num); | |
1595 | ||
1596 | rcu_read_lock(); | |
d3e2ba59 | 1597 | stream = relay_stream_find_by_id(stream_id); |
de91f48a | 1598 | if (stream == NULL) { |
c8f59ee5 DG |
1599 | ret = -1; |
1600 | goto end_unlock; | |
1601 | } | |
1602 | ||
6d805429 | 1603 | DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64 |
c8f59ee5 DG |
1604 | " and last_seq %" PRIu64, stream_id, stream->prev_seq, |
1605 | last_net_seq_num); | |
1606 | ||
33832e64 | 1607 | /* Avoid wrapping issue */ |
39df6d9f | 1608 | if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) { |
6d805429 | 1609 | /* Data has in fact been written and is NOT pending */ |
c8f59ee5 | 1610 | ret = 0; |
6d805429 DG |
1611 | } else { |
1612 | /* Data still being streamed thus pending */ | |
1613 | ret = 1; | |
c8f59ee5 DG |
1614 | } |
1615 | ||
f7079f67 DG |
1616 | /* Pending check is now done. */ |
1617 | stream->data_pending_check_done = 1; | |
1618 | ||
c8f59ee5 DG |
1619 | end_unlock: |
1620 | rcu_read_unlock(); | |
1621 | ||
1622 | reply.ret_code = htobe32(ret); | |
1623 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1624 | if (ret < 0) { | |
6d805429 | 1625 | ERR("Relay data pending ret code failed"); |
c8f59ee5 DG |
1626 | } |
1627 | ||
1628 | end_no_session: | |
1629 | return ret; | |
1630 | } | |
1631 | ||
1632 | /* | |
1633 | * Wait for the control socket to reach a quiescent state. | |
1634 | * | |
1635 | * Note that for now, when receiving this command from the session daemon, this | |
1636 | * means that every subsequent commands or data received on the control socket | |
1637 | * has been handled. So, this is why we simply return OK here. | |
1638 | */ | |
1639 | static | |
1640 | int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1641 | struct relay_command *cmd) |
c8f59ee5 DG |
1642 | { |
1643 | int ret; | |
ad7051c0 DG |
1644 | uint64_t stream_id; |
1645 | struct relay_stream *stream; | |
1646 | struct lttng_ht_iter iter; | |
1647 | struct lttcomm_relayd_quiescent_control msg; | |
c8f59ee5 DG |
1648 | struct lttcomm_relayd_generic_reply reply; |
1649 | ||
1650 | DBG("Checking quiescent state on control socket"); | |
1651 | ||
ad7051c0 DG |
1652 | if (!cmd->session || cmd->version_check_done == 0) { |
1653 | ERR("Trying to check for data before version check"); | |
1654 | ret = -1; | |
1655 | goto end_no_session; | |
1656 | } | |
1657 | ||
1658 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1659 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1660 | if (ret == 0) { |
1661 | /* Orderly shutdown. Not necessary to print an error. */ | |
1662 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1663 | } else { | |
1664 | ERR("Relay didn't receive valid begin data_pending struct size: %d", | |
1665 | ret); | |
1666 | } | |
ad7051c0 DG |
1667 | ret = -1; |
1668 | goto end_no_session; | |
1669 | } | |
1670 | ||
1671 | stream_id = be64toh(msg.stream_id); | |
1672 | ||
1673 | rcu_read_lock(); | |
d3e2ba59 JD |
1674 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
1675 | stream_n.node) { | |
ad7051c0 DG |
1676 | if (stream->stream_handle == stream_id) { |
1677 | stream->data_pending_check_done = 1; | |
1678 | DBG("Relay quiescent control pending flag set to %" PRIu64, | |
1679 | stream_id); | |
1680 | break; | |
1681 | } | |
1682 | } | |
1683 | rcu_read_unlock(); | |
1684 | ||
c8f59ee5 DG |
1685 | reply.ret_code = htobe32(LTTNG_OK); |
1686 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1687 | if (ret < 0) { | |
6d805429 | 1688 | ERR("Relay data quiescent control ret code failed"); |
c8f59ee5 DG |
1689 | } |
1690 | ||
ad7051c0 | 1691 | end_no_session: |
c8f59ee5 DG |
1692 | return ret; |
1693 | } | |
1694 | ||
f7079f67 DG |
1695 | /* |
1696 | * Initialize a data pending command. This means that a client is about to ask | |
1697 | * for data pending for each stream he/she holds. Simply iterate over all | |
1698 | * streams of a session and set the data_pending_check_done flag. | |
1699 | * | |
1700 | * This command returns to the client a LTTNG_OK code. | |
1701 | */ | |
1702 | static | |
1703 | int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1704 | struct relay_command *cmd) |
f7079f67 DG |
1705 | { |
1706 | int ret; | |
1707 | struct lttng_ht_iter iter; | |
1708 | struct lttcomm_relayd_begin_data_pending msg; | |
1709 | struct lttcomm_relayd_generic_reply reply; | |
1710 | struct relay_stream *stream; | |
1711 | uint64_t session_id; | |
1712 | ||
1713 | assert(recv_hdr); | |
1714 | assert(cmd); | |
f7079f67 DG |
1715 | |
1716 | DBG("Init streams for data pending"); | |
1717 | ||
1718 | if (!cmd->session || cmd->version_check_done == 0) { | |
1719 | ERR("Trying to check for data before version check"); | |
1720 | ret = -1; | |
1721 | goto end_no_session; | |
1722 | } | |
1723 | ||
1724 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1725 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1726 | if (ret == 0) { |
1727 | /* Orderly shutdown. Not necessary to print an error. */ | |
1728 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1729 | } else { | |
1730 | ERR("Relay didn't receive valid begin data_pending struct size: %d", | |
1731 | ret); | |
1732 | } | |
f7079f67 DG |
1733 | ret = -1; |
1734 | goto end_no_session; | |
1735 | } | |
1736 | ||
1737 | session_id = be64toh(msg.session_id); | |
1738 | ||
1739 | /* | |
1740 | * Iterate over all streams to set the begin data pending flag. For now, the | |
1741 | * streams are indexed by stream handle so we have to iterate over all | |
1742 | * streams to find the one associated with the right session_id. | |
1743 | */ | |
1744 | rcu_read_lock(); | |
d3e2ba59 JD |
1745 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
1746 | stream_n.node) { | |
f7079f67 DG |
1747 | if (stream->session->id == session_id) { |
1748 | stream->data_pending_check_done = 0; | |
1749 | DBG("Set begin data pending flag to stream %" PRIu64, | |
1750 | stream->stream_handle); | |
1751 | } | |
1752 | } | |
1753 | rcu_read_unlock(); | |
1754 | ||
1755 | /* All good, send back reply. */ | |
1756 | reply.ret_code = htobe32(LTTNG_OK); | |
1757 | ||
1758 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1759 | if (ret < 0) { | |
1760 | ERR("Relay begin data pending send reply failed"); | |
1761 | } | |
1762 | ||
1763 | end_no_session: | |
1764 | return ret; | |
1765 | } | |
1766 | ||
1767 | /* | |
1768 | * End data pending command. This will check, for a given session id, if each | |
1769 | * stream associated with it has its data_pending_check_done flag set. If not, | |
1770 | * this means that the client lost track of the stream but the data is still | |
1771 | * being streamed on our side. In this case, we inform the client that data is | |
1772 | * inflight. | |
1773 | * | |
1774 | * Return to the client if there is data in flight or not with a ret_code. | |
1775 | */ | |
1776 | static | |
1777 | int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 1778 | struct relay_command *cmd) |
f7079f67 DG |
1779 | { |
1780 | int ret; | |
1781 | struct lttng_ht_iter iter; | |
1782 | struct lttcomm_relayd_end_data_pending msg; | |
1783 | struct lttcomm_relayd_generic_reply reply; | |
1784 | struct relay_stream *stream; | |
1785 | uint64_t session_id; | |
1786 | uint32_t is_data_inflight = 0; | |
1787 | ||
1788 | assert(recv_hdr); | |
1789 | assert(cmd); | |
f7079f67 DG |
1790 | |
1791 | DBG("End data pending command"); | |
1792 | ||
1793 | if (!cmd->session || cmd->version_check_done == 0) { | |
1794 | ERR("Trying to check for data before version check"); | |
1795 | ret = -1; | |
1796 | goto end_no_session; | |
1797 | } | |
1798 | ||
1799 | ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), 0); | |
1800 | if (ret < sizeof(msg)) { | |
a6cd2b97 DG |
1801 | if (ret == 0) { |
1802 | /* Orderly shutdown. Not necessary to print an error. */ | |
1803 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1804 | } else { | |
1805 | ERR("Relay didn't receive valid end data_pending struct size: %d", | |
1806 | ret); | |
1807 | } | |
f7079f67 DG |
1808 | ret = -1; |
1809 | goto end_no_session; | |
1810 | } | |
1811 | ||
1812 | session_id = be64toh(msg.session_id); | |
1813 | ||
1814 | /* Iterate over all streams to see if the begin data pending flag is set. */ | |
1815 | rcu_read_lock(); | |
d3e2ba59 JD |
1816 | cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream, |
1817 | stream_n.node) { | |
f7079f67 DG |
1818 | if (stream->session->id == session_id && |
1819 | !stream->data_pending_check_done) { | |
1820 | is_data_inflight = 1; | |
1821 | DBG("Data is still in flight for stream %" PRIu64, | |
1822 | stream->stream_handle); | |
1823 | break; | |
1824 | } | |
1825 | } | |
1826 | rcu_read_unlock(); | |
1827 | ||
1828 | /* All good, send back reply. */ | |
1829 | reply.ret_code = htobe32(is_data_inflight); | |
1830 | ||
1831 | ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1832 | if (ret < 0) { | |
1833 | ERR("Relay end data pending send reply failed"); | |
1834 | } | |
1835 | ||
1836 | end_no_session: | |
1837 | return ret; | |
1838 | } | |
1839 | ||
1c20f0e2 JD |
1840 | /* |
1841 | * Receive an index for a specific stream. | |
1842 | * | |
1843 | * Return 0 on success else a negative value. | |
1844 | */ | |
1845 | static | |
1846 | int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr, | |
0a6518b0 | 1847 | struct relay_command *cmd) |
1c20f0e2 JD |
1848 | { |
1849 | int ret, send_ret, index_created = 0; | |
1850 | struct relay_session *session = cmd->session; | |
1851 | struct lttcomm_relayd_index index_info; | |
1852 | struct relay_index *index, *wr_index = NULL; | |
1853 | struct lttcomm_relayd_generic_reply reply; | |
1854 | struct relay_stream *stream; | |
1855 | uint64_t net_seq_num; | |
1856 | ||
1857 | assert(cmd); | |
1c20f0e2 JD |
1858 | |
1859 | DBG("Relay receiving index"); | |
1860 | ||
1861 | if (!session || cmd->version_check_done == 0) { | |
1862 | ERR("Trying to close a stream before version check"); | |
1863 | ret = -1; | |
1864 | goto end_no_session; | |
1865 | } | |
1866 | ||
1867 | ret = cmd->sock->ops->recvmsg(cmd->sock, &index_info, | |
1868 | sizeof(index_info), 0); | |
1869 | if (ret < sizeof(index_info)) { | |
1870 | if (ret == 0) { | |
1871 | /* Orderly shutdown. Not necessary to print an error. */ | |
1872 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
1873 | } else { | |
1874 | ERR("Relay didn't receive valid index struct size : %d", ret); | |
1875 | } | |
1876 | ret = -1; | |
1877 | goto end_no_session; | |
1878 | } | |
1879 | ||
1880 | net_seq_num = be64toh(index_info.net_seq_num); | |
1881 | ||
1882 | rcu_read_lock(); | |
d3e2ba59 | 1883 | stream = relay_stream_find_by_id(be64toh(index_info.relay_stream_id)); |
1c20f0e2 JD |
1884 | if (!stream) { |
1885 | ret = -1; | |
1886 | goto end_rcu_unlock; | |
1887 | } | |
1888 | ||
d3e2ba59 JD |
1889 | /* Live beacon handling */ |
1890 | if (index_info.packet_size == 0) { | |
1891 | DBG("Received live beacon for stream %" PRIu64, stream->stream_handle); | |
1892 | ||
1893 | /* | |
1894 | * Only flag a stream inactive when it has already received data. | |
1895 | */ | |
1896 | if (stream->total_index_received > 0) { | |
1897 | stream->beacon_ts_end = be64toh(index_info.timestamp_end); | |
1898 | } | |
1899 | ret = 0; | |
1900 | goto end_rcu_unlock; | |
1901 | } else { | |
1902 | stream->beacon_ts_end = -1ULL; | |
1903 | } | |
1904 | ||
0a6518b0 | 1905 | index = relay_index_find(stream->stream_handle, net_seq_num); |
1c20f0e2 JD |
1906 | if (!index) { |
1907 | /* A successful creation will add the object to the HT. */ | |
1908 | index = relay_index_create(stream->stream_handle, net_seq_num); | |
1909 | if (!index) { | |
1910 | goto end_rcu_unlock; | |
1911 | } | |
1912 | index_created = 1; | |
1913 | } | |
1914 | ||
1915 | copy_index_control_data(index, &index_info); | |
1916 | ||
1917 | if (index_created) { | |
1918 | /* | |
1919 | * Try to add the relay index object to the hash table. If an object | |
1920 | * already exist, destroy back the index created, set the data in this | |
1921 | * object and write it on disk. | |
1922 | */ | |
0a6518b0 | 1923 | relay_index_add(index, &wr_index); |
1c20f0e2 JD |
1924 | if (wr_index) { |
1925 | copy_index_control_data(wr_index, &index_info); | |
1926 | free(index); | |
1927 | } | |
1928 | } else { | |
1929 | /* The index already exists so write it on disk. */ | |
1930 | wr_index = index; | |
1931 | } | |
1932 | ||
1933 | /* Do we have a writable ready index to write on disk. */ | |
1934 | if (wr_index) { | |
1935 | /* Starting at 2.4, create the index file if none available. */ | |
1936 | if (cmd->minor >= 4 && stream->index_fd < 0) { | |
1937 | ret = index_create_file(stream->path_name, stream->channel_name, | |
1938 | relayd_uid, relayd_gid, stream->tracefile_size, | |
1939 | stream->tracefile_count_current); | |
1940 | if (ret < 0) { | |
1941 | goto end_rcu_unlock; | |
1942 | } | |
1943 | stream->index_fd = ret; | |
1944 | } | |
1945 | ||
0a6518b0 | 1946 | ret = relay_index_write(wr_index->fd, wr_index); |
1c20f0e2 JD |
1947 | if (ret < 0) { |
1948 | goto end_rcu_unlock; | |
1949 | } | |
d3e2ba59 | 1950 | stream->total_index_received++; |
1c20f0e2 JD |
1951 | } |
1952 | ||
1953 | end_rcu_unlock: | |
1954 | rcu_read_unlock(); | |
1955 | ||
1956 | if (ret < 0) { | |
1957 | reply.ret_code = htobe32(LTTNG_ERR_UNK); | |
1958 | } else { | |
1959 | reply.ret_code = htobe32(LTTNG_OK); | |
1960 | } | |
1961 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
1962 | if (send_ret < 0) { | |
1963 | ERR("Relay sending close index id reply"); | |
1964 | ret = send_ret; | |
1965 | } | |
1966 | ||
1967 | end_no_session: | |
1968 | return ret; | |
1969 | } | |
1970 | ||
814fcae4 JD |
1971 | /* |
1972 | * Receive the streams_sent message. | |
1973 | * | |
1974 | * Return 0 on success else a negative value. | |
1975 | */ | |
1976 | static | |
1977 | int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr, | |
1978 | struct relay_command *cmd) | |
1979 | { | |
1980 | int ret, send_ret; | |
1981 | struct lttcomm_relayd_generic_reply reply; | |
1982 | ||
1983 | assert(cmd); | |
1984 | ||
1985 | DBG("Relay receiving streams_sent"); | |
1986 | ||
1987 | if (!cmd->session || cmd->version_check_done == 0) { | |
1988 | ERR("Trying to close a stream before version check"); | |
1989 | ret = -1; | |
1990 | goto end_no_session; | |
1991 | } | |
1992 | ||
1993 | /* | |
1994 | * Flag every pending stream in the connection recv list that they are | |
1995 | * ready to be used by the viewer. | |
1996 | */ | |
1997 | set_viewer_ready_flag(cmd); | |
1998 | ||
1999 | reply.ret_code = htobe32(LTTNG_OK); | |
2000 | send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0); | |
2001 | if (send_ret < 0) { | |
2002 | ERR("Relay sending sent_stream reply"); | |
2003 | ret = send_ret; | |
2004 | } else { | |
2005 | /* Success. */ | |
2006 | ret = 0; | |
2007 | } | |
2008 | ||
2009 | end_no_session: | |
2010 | return ret; | |
2011 | } | |
2012 | ||
b8aa1682 | 2013 | /* |
d3e2ba59 | 2014 | * Process the commands received on the control socket |
b8aa1682 JD |
2015 | */ |
2016 | static | |
2017 | int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr, | |
d3e2ba59 | 2018 | struct relay_command *cmd, struct relay_local_data *ctx) |
b8aa1682 JD |
2019 | { |
2020 | int ret = 0; | |
2021 | ||
2022 | switch (be32toh(recv_hdr->cmd)) { | |
b8aa1682 | 2023 | case RELAYD_CREATE_SESSION: |
d3e2ba59 | 2024 | ret = relay_create_session(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 | 2025 | break; |
b8aa1682 | 2026 | case RELAYD_ADD_STREAM: |
d3e2ba59 | 2027 | ret = relay_add_stream(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 JD |
2028 | break; |
2029 | case RELAYD_START_DATA: | |
2030 | ret = relay_start(recv_hdr, cmd); | |
2031 | break; | |
2032 | case RELAYD_SEND_METADATA: | |
d3e2ba59 | 2033 | ret = relay_recv_metadata(recv_hdr, cmd); |
b8aa1682 JD |
2034 | break; |
2035 | case RELAYD_VERSION: | |
d3e2ba59 | 2036 | ret = relay_send_version(recv_hdr, cmd, ctx->sessions_ht); |
b8aa1682 | 2037 | break; |
173af62f | 2038 | case RELAYD_CLOSE_STREAM: |
92c6ca54 | 2039 | ret = relay_close_stream(recv_hdr, cmd); |
173af62f | 2040 | break; |
6d805429 | 2041 | case RELAYD_DATA_PENDING: |
d3e2ba59 | 2042 | ret = relay_data_pending(recv_hdr, cmd); |
c8f59ee5 DG |
2043 | break; |
2044 | case RELAYD_QUIESCENT_CONTROL: | |
d3e2ba59 | 2045 | ret = relay_quiescent_control(recv_hdr, cmd); |
c8f59ee5 | 2046 | break; |
f7079f67 | 2047 | case RELAYD_BEGIN_DATA_PENDING: |
d3e2ba59 | 2048 | ret = relay_begin_data_pending(recv_hdr, cmd); |
f7079f67 DG |
2049 | break; |
2050 | case RELAYD_END_DATA_PENDING: | |
d3e2ba59 | 2051 | ret = relay_end_data_pending(recv_hdr, cmd); |
f7079f67 | 2052 | break; |
1c20f0e2 | 2053 | case RELAYD_SEND_INDEX: |
0a6518b0 | 2054 | ret = relay_recv_index(recv_hdr, cmd); |
1c20f0e2 | 2055 | break; |
814fcae4 JD |
2056 | case RELAYD_STREAMS_SENT: |
2057 | ret = relay_streams_sent(recv_hdr, cmd); | |
2058 | break; | |
b8aa1682 JD |
2059 | case RELAYD_UPDATE_SYNC_INFO: |
2060 | default: | |
2061 | ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd)); | |
2062 | relay_unknown_command(cmd); | |
2063 | ret = -1; | |
2064 | goto end; | |
2065 | } | |
2066 | ||
2067 | end: | |
2068 | return ret; | |
2069 | } | |
2070 | ||
7d2f7452 DG |
2071 | /* |
2072 | * Handle index for a data stream. | |
2073 | * | |
2074 | * RCU read side lock MUST be acquired. | |
2075 | * | |
2076 | * Return 0 on success else a negative value. | |
2077 | */ | |
2078 | static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num, | |
2079 | int rotate_index) | |
2080 | { | |
2081 | int ret = 0, index_created = 0; | |
2082 | uint64_t stream_id, data_offset; | |
2083 | struct relay_index *index, *wr_index = NULL; | |
2084 | ||
2085 | assert(stream); | |
2086 | ||
2087 | stream_id = stream->stream_handle; | |
2088 | /* Get data offset because we are about to update the index. */ | |
2089 | data_offset = htobe64(stream->tracefile_size_current); | |
2090 | ||
2091 | /* | |
2092 | * Lookup for an existing index for that stream id/sequence number. If on | |
2093 | * exists, the control thread already received the data for it thus we need | |
2094 | * to write it on disk. | |
2095 | */ | |
2096 | index = relay_index_find(stream_id, net_seq_num); | |
2097 | if (!index) { | |
2098 | /* A successful creation will add the object to the HT. */ | |
2099 | index = relay_index_create(stream_id, net_seq_num); | |
2100 | if (!index) { | |
2101 | ret = -1; | |
2102 | goto error; | |
2103 | } | |
2104 | index_created = 1; | |
2105 | } | |
2106 | ||
2107 | if (rotate_index || stream->index_fd < 0) { | |
2108 | index->to_close_fd = stream->index_fd; | |
2109 | ret = index_create_file(stream->path_name, stream->channel_name, | |
2110 | relayd_uid, relayd_gid, stream->tracefile_size, | |
2111 | stream->tracefile_count_current); | |
2112 | if (ret < 0) { | |
2113 | /* This will close the stream's index fd if one. */ | |
2114 | relay_index_free_safe(index); | |
2115 | goto error; | |
2116 | } | |
2117 | stream->index_fd = ret; | |
2118 | } | |
2119 | index->fd = stream->index_fd; | |
2120 | index->index_data.offset = data_offset; | |
2121 | ||
2122 | if (index_created) { | |
2123 | /* | |
2124 | * Try to add the relay index object to the hash table. If an object | |
2125 | * already exist, destroy back the index created and set the data. | |
2126 | */ | |
2127 | relay_index_add(index, &wr_index); | |
2128 | if (wr_index) { | |
2129 | /* Copy back data from the created index. */ | |
2130 | wr_index->fd = index->fd; | |
2131 | wr_index->to_close_fd = index->to_close_fd; | |
2132 | wr_index->index_data.offset = data_offset; | |
2133 | free(index); | |
2134 | } | |
2135 | } else { | |
2136 | /* The index already exists so write it on disk. */ | |
2137 | wr_index = index; | |
2138 | } | |
2139 | ||
2140 | /* Do we have a writable ready index to write on disk. */ | |
2141 | if (wr_index) { | |
2142 | ret = relay_index_write(wr_index->fd, wr_index); | |
2143 | if (ret < 0) { | |
2144 | goto error; | |
2145 | } | |
2146 | stream->total_index_received++; | |
2147 | } | |
2148 | ||
2149 | error: | |
2150 | return ret; | |
2151 | } | |
2152 | ||
b8aa1682 JD |
2153 | /* |
2154 | * relay_process_data: Process the data received on the data socket | |
2155 | */ | |
2156 | static | |
0a6518b0 | 2157 | int relay_process_data(struct relay_command *cmd) |
b8aa1682 | 2158 | { |
7d2f7452 | 2159 | int ret = 0, rotate_index = 0; |
6cd525e8 | 2160 | ssize_t size_ret; |
b8aa1682 JD |
2161 | struct relay_stream *stream; |
2162 | struct lttcomm_relayd_data_hdr data_hdr; | |
7d2f7452 | 2163 | uint64_t stream_id; |
173af62f | 2164 | uint64_t net_seq_num; |
b8aa1682 JD |
2165 | uint32_t data_size; |
2166 | ||
2167 | ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr, | |
7c5aef62 | 2168 | sizeof(struct lttcomm_relayd_data_hdr), 0); |
b8aa1682 | 2169 | if (ret <= 0) { |
a6cd2b97 DG |
2170 | if (ret == 0) { |
2171 | /* Orderly shutdown. Not necessary to print an error. */ | |
2172 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
2173 | } else { | |
2174 | ERR("Unable to receive data header on sock %d", cmd->sock->fd); | |
2175 | } | |
b8aa1682 JD |
2176 | ret = -1; |
2177 | goto end; | |
2178 | } | |
2179 | ||
2180 | stream_id = be64toh(data_hdr.stream_id); | |
9d1bbf21 MD |
2181 | |
2182 | rcu_read_lock(); | |
d3e2ba59 | 2183 | stream = relay_stream_find_by_id(stream_id); |
b8aa1682 JD |
2184 | if (!stream) { |
2185 | ret = -1; | |
1c20f0e2 | 2186 | goto end_rcu_unlock; |
b8aa1682 JD |
2187 | } |
2188 | ||
2189 | data_size = be32toh(data_hdr.data_size); | |
2190 | if (data_buffer_size < data_size) { | |
c617c0c6 MD |
2191 | char *tmp_data_ptr; |
2192 | ||
2193 | tmp_data_ptr = realloc(data_buffer, data_size); | |
2194 | if (!tmp_data_ptr) { | |
b8aa1682 | 2195 | ERR("Allocating data buffer"); |
c617c0c6 | 2196 | free(data_buffer); |
b8aa1682 | 2197 | ret = -1; |
1c20f0e2 | 2198 | goto end_rcu_unlock; |
b8aa1682 | 2199 | } |
c617c0c6 | 2200 | data_buffer = tmp_data_ptr; |
b8aa1682 JD |
2201 | data_buffer_size = data_size; |
2202 | } | |
2203 | memset(data_buffer, 0, data_size); | |
2204 | ||
173af62f DG |
2205 | net_seq_num = be64toh(data_hdr.net_seq_num); |
2206 | ||
77c7c900 | 2207 | DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64, |
173af62f | 2208 | data_size, stream_id, net_seq_num); |
7c5aef62 | 2209 | ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, 0); |
b8aa1682 | 2210 | if (ret <= 0) { |
a6cd2b97 DG |
2211 | if (ret == 0) { |
2212 | /* Orderly shutdown. Not necessary to print an error. */ | |
2213 | DBG("Socket %d did an orderly shutdown", cmd->sock->fd); | |
2214 | } | |
b8aa1682 | 2215 | ret = -1; |
1c20f0e2 | 2216 | goto end_rcu_unlock; |
b8aa1682 JD |
2217 | } |
2218 | ||
1c20f0e2 | 2219 | /* Check if a rotation is needed. */ |
0f907de1 JD |
2220 | if (stream->tracefile_size > 0 && |
2221 | (stream->tracefile_size_current + data_size) > | |
2222 | stream->tracefile_size) { | |
6b6b9a5a JD |
2223 | struct relay_viewer_stream *vstream; |
2224 | uint64_t new_id; | |
2225 | ||
2226 | new_id = (stream->tracefile_count_current + 1) % | |
2227 | stream->tracefile_count; | |
2228 | /* | |
2229 | * When we wrap-around back to 0, we start overwriting old | |
2230 | * trace data. | |
2231 | */ | |
2232 | if (!stream->tracefile_overwrite && new_id == 0) { | |
2233 | stream->tracefile_overwrite = 1; | |
2234 | } | |
2235 | pthread_mutex_lock(&stream->viewer_stream_rotation_lock); | |
2236 | if (stream->tracefile_overwrite) { | |
2237 | stream->oldest_tracefile_id = | |
2238 | (stream->oldest_tracefile_id + 1) % | |
2239 | stream->tracefile_count; | |
2240 | } | |
2241 | vstream = live_find_viewer_stream_by_id(stream->stream_handle); | |
2242 | if (vstream) { | |
2243 | /* | |
2244 | * The viewer is reading a file about to be | |
2245 | * overwritten. Close the FDs it is | |
2246 | * currently using and let it handle the fault. | |
2247 | */ | |
2248 | if (vstream->tracefile_count_current == new_id) { | |
cef0f7d5 | 2249 | pthread_mutex_lock(&vstream->overwrite_lock); |
6b6b9a5a | 2250 | vstream->abort_flag = 1; |
cef0f7d5 | 2251 | pthread_mutex_unlock(&vstream->overwrite_lock); |
6b6b9a5a JD |
2252 | DBG("Streaming side setting abort_flag on stream %s_%lu\n", |
2253 | stream->channel_name, new_id); | |
2254 | } else if (vstream->tracefile_count_current == | |
2255 | stream->tracefile_count_current) { | |
2256 | /* | |
2257 | * The reader and writer were in the | |
2258 | * same trace file, inform the viewer | |
2259 | * that no new index will ever be added | |
2260 | * to this file. | |
2261 | */ | |
2262 | vstream->close_write_flag = 1; | |
2263 | } | |
2264 | } | |
1c20f0e2 JD |
2265 | ret = utils_rotate_stream_file(stream->path_name, stream->channel_name, |
2266 | stream->tracefile_size, stream->tracefile_count, | |
2267 | relayd_uid, relayd_gid, stream->fd, | |
2268 | &(stream->tracefile_count_current), &stream->fd); | |
cef0f7d5 | 2269 | stream->total_index_received = 0; |
6b6b9a5a | 2270 | pthread_mutex_unlock(&stream->viewer_stream_rotation_lock); |
0f907de1 | 2271 | if (ret < 0) { |
1c20f0e2 JD |
2272 | ERR("Rotating stream output file"); |
2273 | goto end_rcu_unlock; | |
0f907de1 | 2274 | } |
a6976990 DG |
2275 | /* Reset current size because we just perform a stream rotation. */ |
2276 | stream->tracefile_size_current = 0; | |
1c20f0e2 JD |
2277 | rotate_index = 1; |
2278 | } | |
2279 | ||
1c20f0e2 | 2280 | /* |
7d2f7452 DG |
2281 | * Index are handled in protocol version 2.4 and above. Also, snapshot and |
2282 | * index are NOT supported. | |
1c20f0e2 | 2283 | */ |
7d2f7452 DG |
2284 | if (stream->session->minor >= 4 && !stream->session->snapshot) { |
2285 | ret = handle_index_data(stream, net_seq_num, rotate_index); | |
1c20f0e2 | 2286 | if (ret < 0) { |
1c20f0e2 JD |
2287 | goto end_rcu_unlock; |
2288 | } | |
1c20f0e2 JD |
2289 | } |
2290 | ||
7d2f7452 | 2291 | /* Write data to stream output fd. */ |
6cd525e8 MD |
2292 | size_ret = lttng_write(stream->fd, data_buffer, data_size); |
2293 | if (size_ret < data_size) { | |
b8aa1682 JD |
2294 | ERR("Relay error writing data to file"); |
2295 | ret = -1; | |
1c20f0e2 | 2296 | goto end_rcu_unlock; |
b8aa1682 | 2297 | } |
1d4dfdef | 2298 | |
5ab7344e JD |
2299 | DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64, |
2300 | ret, stream->stream_handle); | |
2301 | ||
1d4dfdef DG |
2302 | ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size)); |
2303 | if (ret < 0) { | |
1c20f0e2 | 2304 | goto end_rcu_unlock; |
1d4dfdef | 2305 | } |
1c20f0e2 | 2306 | stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size); |
1d4dfdef | 2307 | |
173af62f DG |
2308 | stream->prev_seq = net_seq_num; |
2309 | ||
2310 | /* Check if we need to close the FD */ | |
2311 | if (close_stream_check(stream)) { | |
c07d8a78 | 2312 | destroy_stream(stream); |
173af62f DG |
2313 | } |
2314 | ||
1c20f0e2 | 2315 | end_rcu_unlock: |
9d1bbf21 | 2316 | rcu_read_unlock(); |
b8aa1682 JD |
2317 | end: |
2318 | return ret; | |
2319 | } | |
2320 | ||
2321 | static | |
9d1bbf21 | 2322 | void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd) |
b8aa1682 JD |
2323 | { |
2324 | int ret; | |
2325 | ||
b8aa1682 JD |
2326 | lttng_poll_del(events, pollfd); |
2327 | ||
2328 | ret = close(pollfd); | |
2329 | if (ret < 0) { | |
2330 | ERR("Closing pollfd %d", pollfd); | |
2331 | } | |
2332 | } | |
2333 | ||
2334 | static | |
2335 | int relay_add_connection(int fd, struct lttng_poll_event *events, | |
2336 | struct lttng_ht *relay_connections_ht) | |
2337 | { | |
b8aa1682 | 2338 | struct relay_command *relay_connection; |
6cd525e8 | 2339 | ssize_t ret; |
b8aa1682 JD |
2340 | |
2341 | relay_connection = zmalloc(sizeof(struct relay_command)); | |
2342 | if (relay_connection == NULL) { | |
2343 | PERROR("Relay command zmalloc"); | |
9d1bbf21 | 2344 | goto error; |
b8aa1682 | 2345 | } |
6cd525e8 MD |
2346 | ret = lttng_read(fd, relay_connection, sizeof(struct relay_command)); |
2347 | if (ret < sizeof(struct relay_command)) { | |
b8aa1682 | 2348 | PERROR("read relay cmd pipe"); |
9d1bbf21 | 2349 | goto error_read; |
b8aa1682 | 2350 | } |
814fcae4 | 2351 | CDS_INIT_LIST_HEAD(&relay_connection->recv_head); |
b8aa1682 | 2352 | |
c07d8a78 DG |
2353 | /* |
2354 | * Only used by the control side and the reference is copied inside each | |
2355 | * stream from that connection. Thus a destroy HT must be done after every | |
2356 | * stream has been destroyed. | |
2357 | */ | |
2358 | if (relay_connection->type == RELAY_CONTROL) { | |
2359 | relay_connection->ctf_traces_ht = lttng_ht_new(0, | |
2360 | LTTNG_HT_TYPE_STRING); | |
2361 | if (!relay_connection->ctf_traces_ht) { | |
2362 | goto error_read; | |
2363 | } | |
d3e2ba59 JD |
2364 | } |
2365 | ||
b8aa1682 JD |
2366 | lttng_ht_node_init_ulong(&relay_connection->sock_n, |
2367 | (unsigned long) relay_connection->sock->fd); | |
9d1bbf21 | 2368 | rcu_read_lock(); |
b8aa1682 JD |
2369 | lttng_ht_add_unique_ulong(relay_connections_ht, |
2370 | &relay_connection->sock_n); | |
9d1bbf21 MD |
2371 | rcu_read_unlock(); |
2372 | return lttng_poll_add(events, | |
b8aa1682 JD |
2373 | relay_connection->sock->fd, |
2374 | LPOLLIN | LPOLLRDHUP); | |
2375 | ||
9d1bbf21 MD |
2376 | error_read: |
2377 | free(relay_connection); | |
2378 | error: | |
2379 | return -1; | |
2380 | } | |
2381 | ||
2382 | static | |
2383 | void deferred_free_connection(struct rcu_head *head) | |
2384 | { | |
2385 | struct relay_command *relay_connection = | |
2386 | caa_container_of(head, struct relay_command, rcu_node); | |
5b6d8097 DG |
2387 | |
2388 | lttcomm_destroy_sock(relay_connection->sock); | |
9d1bbf21 MD |
2389 | free(relay_connection); |
2390 | } | |
2391 | ||
2392 | static | |
2393 | void relay_del_connection(struct lttng_ht *relay_connections_ht, | |
d3e2ba59 JD |
2394 | struct lttng_ht_iter *iter, struct relay_command *relay_connection, |
2395 | struct lttng_ht *sessions_ht) | |
9d1bbf21 MD |
2396 | { |
2397 | int ret; | |
2398 | ||
2399 | ret = lttng_ht_del(relay_connections_ht, iter); | |
2400 | assert(!ret); | |
c07d8a78 | 2401 | |
9d1bbf21 | 2402 | if (relay_connection->type == RELAY_CONTROL) { |
814fcae4 JD |
2403 | struct relay_stream_recv_handle *node, *tmp_node; |
2404 | ||
d3e2ba59 | 2405 | relay_delete_session(relay_connection, sessions_ht); |
c07d8a78 | 2406 | lttng_ht_destroy(relay_connection->ctf_traces_ht); |
814fcae4 JD |
2407 | |
2408 | /* Clean up recv list. */ | |
2409 | cds_list_for_each_entry_safe(node, tmp_node, | |
2410 | &relay_connection->recv_head, node) { | |
2411 | cds_list_del(&node->node); | |
2412 | free(node); | |
2413 | } | |
9d1bbf21 | 2414 | } |
5b6d8097 | 2415 | |
c07d8a78 | 2416 | call_rcu(&relay_connection->rcu_node, deferred_free_connection); |
b8aa1682 JD |
2417 | } |
2418 | ||
2419 | /* | |
2420 | * This thread does the actual work | |
2421 | */ | |
2422 | static | |
2423 | void *relay_thread_worker(void *data) | |
2424 | { | |
beaad64c DG |
2425 | int ret, err = -1, last_seen_data_fd = -1; |
2426 | uint32_t nb_fd; | |
b8aa1682 JD |
2427 | struct relay_command *relay_connection; |
2428 | struct lttng_poll_event events; | |
2429 | struct lttng_ht *relay_connections_ht; | |
2430 | struct lttng_ht_node_ulong *node; | |
2431 | struct lttng_ht_iter iter; | |
b8aa1682 | 2432 | struct lttcomm_relayd_hdr recv_hdr; |
d3e2ba59 JD |
2433 | struct relay_local_data *relay_ctx = (struct relay_local_data *) data; |
2434 | struct lttng_ht *sessions_ht = relay_ctx->sessions_ht; | |
b8aa1682 JD |
2435 | |
2436 | DBG("[thread] Relay worker started"); | |
2437 | ||
9d1bbf21 MD |
2438 | rcu_register_thread(); |
2439 | ||
55706a7d MD |
2440 | health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER); |
2441 | ||
4e5b8b82 MD |
2442 | if (testpoint(relayd_thread_worker)) { |
2443 | goto error_testpoint; | |
2444 | } | |
2445 | ||
f385ae0a MD |
2446 | health_code_update(); |
2447 | ||
b8aa1682 JD |
2448 | /* table of connections indexed on socket */ |
2449 | relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
095a4ae5 MD |
2450 | if (!relay_connections_ht) { |
2451 | goto relay_connections_ht_error; | |
2452 | } | |
b8aa1682 | 2453 | |
1c20f0e2 JD |
2454 | /* Tables of received indexes indexed by index handle and net_seq_num. */ |
2455 | indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64); | |
2456 | if (!indexes_ht) { | |
2457 | goto indexes_ht_error; | |
2458 | } | |
2459 | ||
b8aa1682 JD |
2460 | ret = create_thread_poll_set(&events, 2); |
2461 | if (ret < 0) { | |
2462 | goto error_poll_create; | |
2463 | } | |
2464 | ||
2465 | ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP); | |
2466 | if (ret < 0) { | |
2467 | goto error; | |
2468 | } | |
2469 | ||
beaad64c | 2470 | restart: |
b8aa1682 | 2471 | while (1) { |
beaad64c DG |
2472 | int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1; |
2473 | ||
f385ae0a MD |
2474 | health_code_update(); |
2475 | ||
b8aa1682 | 2476 | /* Infinite blocking call, waiting for transmission */ |
87c1611d | 2477 | DBG3("Relayd worker thread polling..."); |
f385ae0a | 2478 | health_poll_entry(); |
b8aa1682 | 2479 | ret = lttng_poll_wait(&events, -1); |
f385ae0a | 2480 | health_poll_exit(); |
b8aa1682 JD |
2481 | if (ret < 0) { |
2482 | /* | |
2483 | * Restart interrupted system call. | |
2484 | */ | |
2485 | if (errno == EINTR) { | |
2486 | goto restart; | |
2487 | } | |
2488 | goto error; | |
2489 | } | |
2490 | ||
0d9c5d77 DG |
2491 | nb_fd = ret; |
2492 | ||
beaad64c DG |
2493 | /* |
2494 | * Process control. The control connection is prioritised so we don't | |
2495 | * starve it with high throughout put tracing data on the data | |
2496 | * connection. | |
2497 | */ | |
b8aa1682 JD |
2498 | for (i = 0; i < nb_fd; i++) { |
2499 | /* Fetch once the poll data */ | |
beaad64c DG |
2500 | uint32_t revents = LTTNG_POLL_GETEV(&events, i); |
2501 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
b8aa1682 | 2502 | |
f385ae0a MD |
2503 | health_code_update(); |
2504 | ||
b8aa1682 JD |
2505 | /* Thread quit pipe has been closed. Killing thread. */ |
2506 | ret = check_thread_quit_pipe(pollfd, revents); | |
2507 | if (ret) { | |
095a4ae5 MD |
2508 | err = 0; |
2509 | goto exit; | |
b8aa1682 JD |
2510 | } |
2511 | ||
2512 | /* Inspect the relay cmd pipe for new connection */ | |
2513 | if (pollfd == relay_cmd_pipe[0]) { | |
2514 | if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) { | |
2515 | ERR("Relay pipe error"); | |
2516 | goto error; | |
2517 | } else if (revents & LPOLLIN) { | |
2518 | DBG("Relay command received"); | |
2519 | ret = relay_add_connection(relay_cmd_pipe[0], | |
2520 | &events, relay_connections_ht); | |
2521 | if (ret < 0) { | |
2522 | goto error; | |
2523 | } | |
2524 | } | |
beaad64c | 2525 | } else if (revents) { |
9d1bbf21 | 2526 | rcu_read_lock(); |
b8aa1682 JD |
2527 | lttng_ht_lookup(relay_connections_ht, |
2528 | (void *)((unsigned long) pollfd), | |
2529 | &iter); | |
2530 | node = lttng_ht_iter_get_node_ulong(&iter); | |
2531 | if (node == NULL) { | |
2532 | DBG2("Relay sock %d not found", pollfd); | |
9d1bbf21 | 2533 | rcu_read_unlock(); |
b8aa1682 JD |
2534 | goto error; |
2535 | } | |
2536 | relay_connection = caa_container_of(node, | |
2537 | struct relay_command, sock_n); | |
2538 | ||
2539 | if (revents & (LPOLLERR)) { | |
2540 | ERR("POLL ERROR"); | |
9d1bbf21 MD |
2541 | relay_cleanup_poll_connection(&events, pollfd); |
2542 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2543 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2544 | if (last_seen_data_fd == pollfd) { |
2545 | last_seen_data_fd = last_notdel_data_fd; | |
2546 | } | |
b8aa1682 JD |
2547 | } else if (revents & (LPOLLHUP | LPOLLRDHUP)) { |
2548 | DBG("Socket %d hung up", pollfd); | |
9d1bbf21 MD |
2549 | relay_cleanup_poll_connection(&events, pollfd); |
2550 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2551 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2552 | if (last_seen_data_fd == pollfd) { |
2553 | last_seen_data_fd = last_notdel_data_fd; | |
2554 | } | |
b8aa1682 JD |
2555 | } else if (revents & LPOLLIN) { |
2556 | /* control socket */ | |
2557 | if (relay_connection->type == RELAY_CONTROL) { | |
2558 | ret = relay_connection->sock->ops->recvmsg( | |
2559 | relay_connection->sock, &recv_hdr, | |
7c5aef62 | 2560 | sizeof(struct lttcomm_relayd_hdr), 0); |
b8aa1682 JD |
2561 | /* connection closed */ |
2562 | if (ret <= 0) { | |
9d1bbf21 MD |
2563 | relay_cleanup_poll_connection(&events, pollfd); |
2564 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2565 | &iter, relay_connection, sessions_ht); |
b8aa1682 JD |
2566 | DBG("Control connection closed with %d", pollfd); |
2567 | } else { | |
2568 | if (relay_connection->session) { | |
77c7c900 | 2569 | DBG2("Relay worker receiving data for session : %" PRIu64, |
b8aa1682 JD |
2570 | relay_connection->session->id); |
2571 | } | |
2572 | ret = relay_process_control(&recv_hdr, | |
d3e2ba59 | 2573 | relay_connection, relay_ctx); |
b8aa1682 | 2574 | if (ret < 0) { |
beaad64c | 2575 | /* Clear the session on error. */ |
9d1bbf21 MD |
2576 | relay_cleanup_poll_connection(&events, pollfd); |
2577 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2578 | &iter, relay_connection, sessions_ht); |
b8aa1682 JD |
2579 | DBG("Connection closed with %d", pollfd); |
2580 | } | |
beaad64c | 2581 | seen_control = 1; |
b8aa1682 | 2582 | } |
beaad64c DG |
2583 | } else { |
2584 | /* | |
2585 | * Flag the last seen data fd not deleted. It will be | |
2586 | * used as the last seen fd if any fd gets deleted in | |
2587 | * this first loop. | |
2588 | */ | |
2589 | last_notdel_data_fd = pollfd; | |
2590 | } | |
2591 | } | |
2592 | rcu_read_unlock(); | |
2593 | } | |
2594 | } | |
2595 | ||
2596 | /* | |
2597 | * The last loop handled a control request, go back to poll to make | |
2598 | * sure we prioritise the control socket. | |
2599 | */ | |
2600 | if (seen_control) { | |
2601 | continue; | |
2602 | } | |
2603 | ||
2604 | if (last_seen_data_fd >= 0) { | |
2605 | for (i = 0; i < nb_fd; i++) { | |
2606 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
f385ae0a MD |
2607 | |
2608 | health_code_update(); | |
2609 | ||
beaad64c DG |
2610 | if (last_seen_data_fd == pollfd) { |
2611 | idx = i; | |
2612 | break; | |
2613 | } | |
2614 | } | |
2615 | } | |
2616 | ||
2617 | /* Process data connection. */ | |
2618 | for (i = idx + 1; i < nb_fd; i++) { | |
2619 | /* Fetch the poll data. */ | |
2620 | uint32_t revents = LTTNG_POLL_GETEV(&events, i); | |
2621 | int pollfd = LTTNG_POLL_GETFD(&events, i); | |
2622 | ||
f385ae0a MD |
2623 | health_code_update(); |
2624 | ||
beaad64c DG |
2625 | /* Skip the command pipe. It's handled in the first loop. */ |
2626 | if (pollfd == relay_cmd_pipe[0]) { | |
2627 | continue; | |
2628 | } | |
2629 | ||
2630 | if (revents) { | |
2631 | rcu_read_lock(); | |
2632 | lttng_ht_lookup(relay_connections_ht, | |
2633 | (void *)((unsigned long) pollfd), | |
2634 | &iter); | |
2635 | node = lttng_ht_iter_get_node_ulong(&iter); | |
2636 | if (node == NULL) { | |
2637 | /* Skip it. Might be removed before. */ | |
2638 | rcu_read_unlock(); | |
2639 | continue; | |
2640 | } | |
2641 | relay_connection = caa_container_of(node, | |
2642 | struct relay_command, sock_n); | |
2643 | ||
2644 | if (revents & LPOLLIN) { | |
2645 | if (relay_connection->type != RELAY_DATA) { | |
2646 | continue; | |
2647 | } | |
2648 | ||
0a6518b0 | 2649 | ret = relay_process_data(relay_connection); |
beaad64c DG |
2650 | /* connection closed */ |
2651 | if (ret < 0) { | |
2652 | relay_cleanup_poll_connection(&events, pollfd); | |
2653 | relay_del_connection(relay_connections_ht, | |
d3e2ba59 | 2654 | &iter, relay_connection, sessions_ht); |
beaad64c DG |
2655 | DBG("Data connection closed with %d", pollfd); |
2656 | /* | |
2657 | * Every goto restart call sets the last seen fd where | |
2658 | * here we don't really care since we gracefully | |
2659 | * continue the loop after the connection is deleted. | |
2660 | */ | |
2661 | } else { | |
2662 | /* Keep last seen port. */ | |
2663 | last_seen_data_fd = pollfd; | |
2664 | rcu_read_unlock(); | |
2665 | goto restart; | |
b8aa1682 JD |
2666 | } |
2667 | } | |
9d1bbf21 | 2668 | rcu_read_unlock(); |
b8aa1682 JD |
2669 | } |
2670 | } | |
beaad64c | 2671 | last_seen_data_fd = -1; |
b8aa1682 JD |
2672 | } |
2673 | ||
f385ae0a MD |
2674 | /* Normal exit, no error */ |
2675 | ret = 0; | |
2676 | ||
095a4ae5 | 2677 | exit: |
b8aa1682 JD |
2678 | error: |
2679 | lttng_poll_clean(&events); | |
2680 | ||
2681 | /* empty the hash table and free the memory */ | |
9d1bbf21 | 2682 | rcu_read_lock(); |
b8aa1682 | 2683 | cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) { |
f385ae0a MD |
2684 | health_code_update(); |
2685 | ||
b8aa1682 JD |
2686 | node = lttng_ht_iter_get_node_ulong(&iter); |
2687 | if (node) { | |
2688 | relay_connection = caa_container_of(node, | |
2689 | struct relay_command, sock_n); | |
9d1bbf21 | 2690 | relay_del_connection(relay_connections_ht, |
d3e2ba59 | 2691 | &iter, relay_connection, sessions_ht); |
b8aa1682 | 2692 | } |
b8aa1682 | 2693 | } |
94d49140 | 2694 | rcu_read_unlock(); |
7d2f7452 DG |
2695 | error_poll_create: |
2696 | lttng_ht_destroy(indexes_ht); | |
1c20f0e2 | 2697 | indexes_ht_error: |
b8aa1682 | 2698 | lttng_ht_destroy(relay_connections_ht); |
095a4ae5 | 2699 | relay_connections_ht_error: |
6620da75 DG |
2700 | /* Close relay cmd pipes */ |
2701 | utils_close_pipe(relay_cmd_pipe); | |
095a4ae5 MD |
2702 | if (err) { |
2703 | DBG("Thread exited with error"); | |
2704 | } | |
b8aa1682 | 2705 | DBG("Worker thread cleanup complete"); |
095a4ae5 | 2706 | free(data_buffer); |
4e5b8b82 | 2707 | error_testpoint: |
f385ae0a MD |
2708 | if (err) { |
2709 | health_error(); | |
2710 | ERR("Health error occurred in %s", __func__); | |
2711 | } | |
2712 | health_unregister(health_relayd); | |
9d1bbf21 | 2713 | rcu_unregister_thread(); |
f385ae0a | 2714 | stop_threads(); |
b8aa1682 JD |
2715 | return NULL; |
2716 | } | |
2717 | ||
2718 | /* | |
2719 | * Create the relay command pipe to wake thread_manage_apps. | |
2720 | * Closed in cleanup(). | |
2721 | */ | |
2722 | static int create_relay_cmd_pipe(void) | |
2723 | { | |
a02de639 | 2724 | int ret; |
b8aa1682 | 2725 | |
a02de639 | 2726 | ret = utils_create_pipe_cloexec(relay_cmd_pipe); |
b8aa1682 | 2727 | |
b8aa1682 JD |
2728 | return ret; |
2729 | } | |
2730 | ||
2731 | /* | |
2732 | * main | |
2733 | */ | |
2734 | int main(int argc, char **argv) | |
2735 | { | |
2736 | int ret = 0; | |
2737 | void *status; | |
d3e2ba59 | 2738 | struct relay_local_data *relay_ctx; |
b8aa1682 | 2739 | |
b8aa1682 JD |
2740 | /* Parse arguments */ |
2741 | progname = argv[0]; | |
c617c0c6 | 2742 | if ((ret = parse_args(argc, argv)) < 0) { |
a02de639 | 2743 | goto exit; |
b8aa1682 JD |
2744 | } |
2745 | ||
2746 | if ((ret = set_signal_handler()) < 0) { | |
2747 | goto exit; | |
2748 | } | |
2749 | ||
4d513a50 DG |
2750 | /* Try to create directory if -o, --output is specified. */ |
2751 | if (opt_output_path) { | |
994fa64f DG |
2752 | if (*opt_output_path != '/') { |
2753 | ERR("Please specify an absolute path for -o, --output PATH"); | |
2754 | goto exit; | |
2755 | } | |
2756 | ||
4d513a50 DG |
2757 | ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG); |
2758 | if (ret < 0) { | |
2759 | ERR("Unable to create %s", opt_output_path); | |
2760 | goto exit; | |
2761 | } | |
2762 | } | |
2763 | ||
b8aa1682 | 2764 | /* Daemonize */ |
d1a3048a | 2765 | if (opt_daemon || opt_background) { |
02a6bb53 MD |
2766 | int i; |
2767 | ||
2768 | ret = lttng_daemonize(&child_ppid, &recv_child_signal, | |
2769 | !opt_background); | |
b8aa1682 | 2770 | if (ret < 0) { |
a02de639 | 2771 | goto exit; |
b8aa1682 | 2772 | } |
02a6bb53 MD |
2773 | |
2774 | /* | |
2775 | * We are in the child. Make sure all other file | |
2776 | * descriptors are closed, in case we are called with | |
2777 | * more opened file descriptors than the standard ones. | |
2778 | */ | |
2779 | for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) { | |
2780 | (void) close(i); | |
2781 | } | |
2782 | } | |
2783 | ||
2784 | /* Create thread quit pipe */ | |
2785 | if ((ret = init_thread_quit_pipe()) < 0) { | |
2786 | goto error; | |
b8aa1682 JD |
2787 | } |
2788 | ||
1c20f0e2 JD |
2789 | /* We need those values for the file/dir creation. */ |
2790 | relayd_uid = getuid(); | |
2791 | relayd_gid = getgid(); | |
b8aa1682 | 2792 | |
1c20f0e2 JD |
2793 | /* Check if daemon is UID = 0 */ |
2794 | if (relayd_uid == 0) { | |
a6a062a4 AM |
2795 | if (control_uri->port < 1024 || data_uri->port < 1024 || |
2796 | live_uri->port < 1024) { | |
b8aa1682 JD |
2797 | ERR("Need to be root to use ports < 1024"); |
2798 | ret = -1; | |
a02de639 | 2799 | goto exit; |
b8aa1682 JD |
2800 | } |
2801 | } | |
2802 | ||
2803 | /* Setup the thread apps communication pipe. */ | |
2804 | if ((ret = create_relay_cmd_pipe()) < 0) { | |
2805 | goto exit; | |
2806 | } | |
2807 | ||
2808 | /* Init relay command queue. */ | |
2809 | cds_wfq_init(&relay_cmd_queue.queue); | |
2810 | ||
2811 | /* Set up max poll set size */ | |
2812 | lttng_poll_set_max_size(); | |
2813 | ||
554831e7 MD |
2814 | /* Initialize communication library */ |
2815 | lttcomm_init(); | |
d5a5e58f | 2816 | lttcomm_inet_init(); |
554831e7 | 2817 | |
d3e2ba59 JD |
2818 | relay_ctx = zmalloc(sizeof(struct relay_local_data)); |
2819 | if (!relay_ctx) { | |
2820 | PERROR("relay_ctx"); | |
2821 | goto exit; | |
2822 | } | |
2823 | ||
2824 | /* tables of sessions indexed by session ID */ | |
2825 | relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
2826 | if (!relay_ctx->sessions_ht) { | |
2827 | goto exit_relay_ctx_sessions; | |
2828 | } | |
2829 | ||
2830 | /* tables of streams indexed by stream ID */ | |
2831 | relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG); | |
2832 | if (!relay_streams_ht) { | |
2833 | goto exit_relay_ctx_streams; | |
2834 | } | |
2835 | ||
2836 | /* tables of streams indexed by stream ID */ | |
92c6ca54 DG |
2837 | viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64); |
2838 | if (!viewer_streams_ht) { | |
d3e2ba59 JD |
2839 | goto exit_relay_ctx_viewer_streams; |
2840 | } | |
2841 | ||
55706a7d MD |
2842 | /* Initialize thread health monitoring */ |
2843 | health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES); | |
2844 | if (!health_relayd) { | |
2845 | PERROR("health_app_create error"); | |
2846 | goto exit_health_app_create; | |
2847 | } | |
2848 | ||
65931c8b MD |
2849 | ret = utils_create_pipe(health_quit_pipe); |
2850 | if (ret < 0) { | |
2851 | goto error_health_pipe; | |
2852 | } | |
2853 | ||
2854 | /* Create thread to manage the client socket */ | |
2855 | ret = pthread_create(&health_thread, NULL, | |
2856 | thread_manage_health, (void *) NULL); | |
2857 | if (ret != 0) { | |
2858 | PERROR("pthread_create health"); | |
2859 | goto health_error; | |
2860 | } | |
2861 | ||
b8aa1682 JD |
2862 | /* Setup the dispatcher thread */ |
2863 | ret = pthread_create(&dispatcher_thread, NULL, | |
2864 | relay_thread_dispatcher, (void *) NULL); | |
2865 | if (ret != 0) { | |
2866 | PERROR("pthread_create dispatcher"); | |
2867 | goto exit_dispatcher; | |
2868 | } | |
2869 | ||
2870 | /* Setup the worker thread */ | |
2871 | ret = pthread_create(&worker_thread, NULL, | |
d3e2ba59 | 2872 | relay_thread_worker, (void *) relay_ctx); |
b8aa1682 JD |
2873 | if (ret != 0) { |
2874 | PERROR("pthread_create worker"); | |
2875 | goto exit_worker; | |
2876 | } | |
2877 | ||
2878 | /* Setup the listener thread */ | |
2879 | ret = pthread_create(&listener_thread, NULL, | |
2880 | relay_thread_listener, (void *) NULL); | |
2881 | if (ret != 0) { | |
2882 | PERROR("pthread_create listener"); | |
2883 | goto exit_listener; | |
2884 | } | |
2885 | ||
3557d456 | 2886 | ret = live_start_threads(live_uri, relay_ctx); |
d3e2ba59 JD |
2887 | if (ret != 0) { |
2888 | ERR("Starting live viewer threads"); | |
50138f51 | 2889 | goto exit_live; |
d3e2ba59 JD |
2890 | } |
2891 | ||
50138f51 | 2892 | exit_live: |
b8aa1682 JD |
2893 | ret = pthread_join(listener_thread, &status); |
2894 | if (ret != 0) { | |
2895 | PERROR("pthread_join"); | |
2896 | goto error; /* join error, exit without cleanup */ | |
2897 | } | |
2898 | ||
50138f51 | 2899 | exit_listener: |
b8aa1682 JD |
2900 | ret = pthread_join(worker_thread, &status); |
2901 | if (ret != 0) { | |
2902 | PERROR("pthread_join"); | |
2903 | goto error; /* join error, exit without cleanup */ | |
2904 | } | |
2905 | ||
50138f51 | 2906 | exit_worker: |
b8aa1682 JD |
2907 | ret = pthread_join(dispatcher_thread, &status); |
2908 | if (ret != 0) { | |
2909 | PERROR("pthread_join"); | |
2910 | goto error; /* join error, exit without cleanup */ | |
2911 | } | |
42415026 | 2912 | |
50138f51 | 2913 | exit_dispatcher: |
65931c8b MD |
2914 | ret = pthread_join(health_thread, &status); |
2915 | if (ret != 0) { | |
2916 | PERROR("pthread_join health thread"); | |
2917 | goto error; /* join error, exit without cleanup */ | |
2918 | } | |
2919 | ||
bd11e201 MD |
2920 | /* |
2921 | * Stop live threads only after joining other threads. | |
2922 | */ | |
2923 | live_stop_threads(); | |
2924 | ||
65931c8b MD |
2925 | health_error: |
2926 | utils_close_pipe(health_quit_pipe); | |
2927 | ||
2928 | error_health_pipe: | |
55706a7d MD |
2929 | health_app_destroy(health_relayd); |
2930 | ||
2931 | exit_health_app_create: | |
92c6ca54 | 2932 | lttng_ht_destroy(viewer_streams_ht); |
d3e2ba59 JD |
2933 | |
2934 | exit_relay_ctx_viewer_streams: | |
2935 | lttng_ht_destroy(relay_streams_ht); | |
2936 | ||
2937 | exit_relay_ctx_streams: | |
2938 | lttng_ht_destroy(relay_ctx->sessions_ht); | |
2939 | ||
2940 | exit_relay_ctx_sessions: | |
2941 | free(relay_ctx); | |
b8aa1682 JD |
2942 | |
2943 | exit: | |
2944 | cleanup(); | |
2945 | if (!ret) { | |
2946 | exit(EXIT_SUCCESS); | |
2947 | } | |
a02de639 | 2948 | |
b8aa1682 JD |
2949 | error: |
2950 | exit(EXIT_FAILURE); | |
2951 | } |