Fix: grab more than one packet for snapshots
[lttng-tools.git] / src / bin / lttng-relayd / main.c
CommitLineData
b8aa1682
JD
1/*
2 * Copyright (C) 2012 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
cd60b05a 4 * 2013 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
b8aa1682
JD
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20#define _GNU_SOURCE
6c1c0768 21#define _LGPL_SOURCE
b8aa1682
JD
22#include <getopt.h>
23#include <grp.h>
24#include <limits.h>
25#include <pthread.h>
26#include <signal.h>
27#include <stdio.h>
28#include <stdlib.h>
29#include <string.h>
30#include <sys/mman.h>
31#include <sys/mount.h>
32#include <sys/resource.h>
33#include <sys/socket.h>
34#include <sys/stat.h>
35#include <sys/types.h>
36#include <sys/wait.h>
173af62f 37#include <inttypes.h>
b8aa1682
JD
38#include <urcu/futex.h>
39#include <urcu/uatomic.h>
40#include <unistd.h>
41#include <fcntl.h>
42#include <config.h>
43
44#include <lttng/lttng.h>
45#include <common/common.h>
46#include <common/compat/poll.h>
47#include <common/compat/socket.h>
f263b7fd 48#include <common/compat/endian.h>
b8aa1682 49#include <common/defaults.h>
3fd27398 50#include <common/daemonize.h>
b8aa1682
JD
51#include <common/futex.h>
52#include <common/sessiond-comm/sessiond-comm.h>
53#include <common/sessiond-comm/inet.h>
b8aa1682
JD
54#include <common/sessiond-comm/relayd.h>
55#include <common/uri.h>
a02de639 56#include <common/utils.h>
cd60b05a 57#include <common/config/config.h>
b8aa1682 58
0f907de1 59#include "cmd.h"
d3e2ba59 60#include "ctf-trace.h"
1c20f0e2 61#include "index.h"
0f907de1 62#include "utils.h"
b8aa1682 63#include "lttng-relayd.h"
d3e2ba59 64#include "live.h"
55706a7d 65#include "health-relayd.h"
9b5e0863 66#include "testpoint.h"
2f8f53af 67#include "viewer-stream.h"
2a174661
DG
68#include "session.h"
69#include "stream.h"
58eb9381 70#include "connection.h"
b8aa1682
JD
71
72/* command line options */
0f907de1 73char *opt_output_path;
b5218ffb 74static int opt_daemon, opt_background;
3fd27398
MD
75
76/*
77 * We need to wait for listener and live listener threads, as well as
78 * health check thread, before being ready to signal readiness.
79 */
80#define NR_LTTNG_RELAY_READY 3
81static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
82static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */
83static pid_t child_ppid; /* Internal parent PID use with daemonize. */
84
095a4ae5
MD
85static struct lttng_uri *control_uri;
86static struct lttng_uri *data_uri;
d3e2ba59 87static struct lttng_uri *live_uri;
b8aa1682
JD
88
89const char *progname;
b8aa1682 90
65931c8b 91const char *tracing_group_name = DEFAULT_TRACING_GROUP;
cd60b05a
JG
92static int tracing_group_name_override;
93
94const char * const config_section_name = "relayd";
65931c8b 95
b8aa1682
JD
96/*
97 * Quit pipe for all threads. This permits a single cancellation point
98 * for all threads when receiving an event on the pipe.
99 */
0b242f62 100int thread_quit_pipe[2] = { -1, -1 };
b8aa1682
JD
101
102/*
103 * This pipe is used to inform the worker thread that a command is queued and
104 * ready to be processed.
105 */
58eb9381 106static int relay_conn_pipe[2] = { -1, -1 };
b8aa1682 107
26c9d55e 108/* Shared between threads */
b8aa1682
JD
109static int dispatch_thread_exit;
110
111static pthread_t listener_thread;
112static pthread_t dispatcher_thread;
113static pthread_t worker_thread;
65931c8b 114static pthread_t health_thread;
b8aa1682 115
095a4ae5 116static uint64_t last_relay_stream_id;
b8aa1682
JD
117
118/*
119 * Relay command queue.
120 *
121 * The relay_thread_listener and relay_thread_dispatcher communicate with this
122 * queue.
123 */
58eb9381 124static struct relay_conn_queue relay_conn_queue;
b8aa1682
JD
125
126/* buffer allocated at startup, used to store the trace data */
095a4ae5
MD
127static char *data_buffer;
128static unsigned int data_buffer_size;
b8aa1682 129
1c20f0e2
JD
130/* We need those values for the file/dir creation. */
131static uid_t relayd_uid;
132static gid_t relayd_gid;
133
d3e2ba59
JD
134/* Global relay stream hash table. */
135struct lttng_ht *relay_streams_ht;
136
92c6ca54
DG
137/* Global relay viewer stream hash table. */
138struct lttng_ht *viewer_streams_ht;
139
0a6518b0
DG
140/* Global hash table that stores relay index object. */
141struct lttng_ht *indexes_ht;
142
55706a7d 143/* Relayd health monitoring */
eea7556c 144struct health_app *health_relayd;
55706a7d 145
cd60b05a
JG
146static struct option long_options[] = {
147 { "control-port", 1, 0, 'C', },
148 { "data-port", 1, 0, 'D', },
8d5c808e 149 { "live-port", 1, 0, 'L', },
cd60b05a 150 { "daemonize", 0, 0, 'd', },
b5218ffb 151 { "background", 0, 0, 'b', },
cd60b05a
JG
152 { "group", 1, 0, 'g', },
153 { "help", 0, 0, 'h', },
154 { "output", 1, 0, 'o', },
155 { "verbose", 0, 0, 'v', },
156 { "config", 1, 0, 'f' },
157 { NULL, 0, 0, 0, },
158};
159
160static const char *config_ignore_options[] = { "help", "config" };
161
b8aa1682
JD
162/*
163 * usage function on stderr
164 */
165static
166void usage(void)
167{
168 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
994fa64f
DG
169 fprintf(stderr, " -h, --help Display this usage.\n");
170 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
b5218ffb 171 fprintf(stderr, " -b, --background Start as a daemon, keeping console open.\n");
994fa64f
DG
172 fprintf(stderr, " -C, --control-port URL Control port listening.\n");
173 fprintf(stderr, " -D, --data-port URL Data port listening.\n");
8d5c808e 174 fprintf(stderr, " -L, --live-port URL Live view port listening.\n");
994fa64f
DG
175 fprintf(stderr, " -o, --output PATH Output path for traces. Must use an absolute path.\n");
176 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
65931c8b 177 fprintf(stderr, " -g, --group NAME Specify the tracing group name. (default: tracing)\n");
cd60b05a 178 fprintf(stderr, " -f --config Load daemon configuration file\n");
b8aa1682
JD
179}
180
cd60b05a
JG
181/*
182 * Take an option from the getopt output and set it in the right variable to be
183 * used later.
184 *
185 * Return 0 on success else a negative value.
186 */
b8aa1682 187static
cd60b05a 188int set_option(int opt, const char *arg, const char *optname)
b8aa1682 189{
cd60b05a
JG
190 int ret;
191
192 switch (opt) {
193 case 0:
194 fprintf(stderr, "option %s", optname);
195 if (arg) {
196 fprintf(stderr, " with arg %s\n", arg);
197 }
198 break;
199 case 'C':
200 ret = uri_parse(arg, &control_uri);
201 if (ret < 0) {
202 ERR("Invalid control URI specified");
203 goto end;
204 }
205 if (control_uri->port == 0) {
206 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
207 }
208 break;
209 case 'D':
210 ret = uri_parse(arg, &data_uri);
211 if (ret < 0) {
212 ERR("Invalid data URI specified");
213 goto end;
214 }
215 if (data_uri->port == 0) {
216 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
217 }
218 break;
8d5c808e
AM
219 case 'L':
220 ret = uri_parse(arg, &live_uri);
221 if (ret < 0) {
222 ERR("Invalid live URI specified");
223 goto end;
224 }
225 if (live_uri->port == 0) {
226 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
227 }
228 break;
cd60b05a
JG
229 case 'd':
230 opt_daemon = 1;
231 break;
b5218ffb
MD
232 case 'b':
233 opt_background = 1;
234 break;
cd60b05a
JG
235 case 'g':
236 tracing_group_name = strdup(arg);
330a40bb
MD
237 if (tracing_group_name == NULL) {
238 ret = -errno;
239 PERROR("strdup");
240 goto end;
241 }
cd60b05a
JG
242 tracing_group_name_override = 1;
243 break;
244 case 'h':
245 usage();
246 exit(EXIT_FAILURE);
247 case 'o':
248 ret = asprintf(&opt_output_path, "%s", arg);
249 if (ret < 0) {
250 ret = -errno;
251 PERROR("asprintf opt_output_path");
252 goto end;
253 }
254 break;
255 case 'v':
256 /* Verbose level can increase using multiple -v */
257 if (arg) {
258 lttng_opt_verbose = config_parse_value(arg);
259 } else {
849e5b7b
DG
260 /* Only 3 level of verbosity (-vvv). */
261 if (lttng_opt_verbose < 3) {
262 lttng_opt_verbose += 1;
263 }
cd60b05a
JG
264 }
265 break;
266 default:
267 /* Unknown option or other error.
268 * Error is printed by getopt, just return */
269 ret = -1;
270 goto end;
271 }
272
273 /* All good. */
274 ret = 0;
275
276end:
277 return ret;
278}
279
280/*
281 * config_entry_handler_cb used to handle options read from a config file.
282 * See config_entry_handler_cb comment in common/config/config.h for the
283 * return value conventions.
284 */
285static
286int config_entry_handler(const struct config_entry *entry, void *unused)
287{
288 int ret = 0, i;
289
290 if (!entry || !entry->name || !entry->value) {
291 ret = -EINVAL;
292 goto end;
293 }
294
295 /* Check if the option is to be ignored */
296 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
297 if (!strcmp(entry->name, config_ignore_options[i])) {
298 goto end;
299 }
300 }
301
302 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
303 /* Ignore if entry name is not fully matched. */
304 if (strcmp(entry->name, long_options[i].name)) {
305 continue;
306 }
307
308 /*
309 * If the option takes no argument on the command line, we have to
310 * check if the value is "true". We support non-zero numeric values,
311 * true, on and yes.
312 */
313 if (!long_options[i].has_arg) {
314 ret = config_parse_value(entry->value);
315 if (ret <= 0) {
316 if (ret) {
317 WARN("Invalid configuration value \"%s\" for option %s",
318 entry->value, entry->name);
319 }
320 /* False, skip boolean config option. */
321 goto end;
322 }
323 }
324
325 ret = set_option(long_options[i].val, entry->value, entry->name);
326 goto end;
327 }
328
329 WARN("Unrecognized option \"%s\" in daemon configuration file.",
330 entry->name);
331
332end:
333 return ret;
334}
335
336static
337int set_options(int argc, char **argv)
338{
178a0557 339 int c, ret = 0, option_index = 0, retval = 0;
cd60b05a
JG
340 int orig_optopt = optopt, orig_optind = optind;
341 char *default_address, *optstring;
342 const char *config_path = NULL;
343
344 optstring = utils_generate_optstring(long_options,
345 sizeof(long_options) / sizeof(struct option));
346 if (!optstring) {
178a0557 347 retval = -ENOMEM;
cd60b05a
JG
348 goto exit;
349 }
350
351 /* Check for the --config option */
352
353 while ((c = getopt_long(argc, argv, optstring, long_options,
354 &option_index)) != -1) {
355 if (c == '?') {
178a0557 356 retval = -EINVAL;
cd60b05a
JG
357 goto exit;
358 } else if (c != 'f') {
359 continue;
360 }
361
362 config_path = utils_expand_path(optarg);
363 if (!config_path) {
364 ERR("Failed to resolve path: %s", optarg);
365 }
366 }
367
368 ret = config_get_section_entries(config_path, config_section_name,
369 config_entry_handler, NULL);
370 if (ret) {
371 if (ret > 0) {
372 ERR("Invalid configuration option at line %i", ret);
cd60b05a 373 }
178a0557 374 retval = -1;
cd60b05a
JG
375 goto exit;
376 }
b8aa1682 377
cd60b05a
JG
378 /* Reset getopt's global state */
379 optopt = orig_optopt;
380 optind = orig_optind;
b8aa1682 381 while (1) {
cd60b05a 382 c = getopt_long(argc, argv, optstring, long_options, &option_index);
b8aa1682
JD
383 if (c == -1) {
384 break;
385 }
386
cd60b05a
JG
387 ret = set_option(c, optarg, long_options[option_index].name);
388 if (ret < 0) {
178a0557 389 retval = -1;
b8aa1682
JD
390 goto exit;
391 }
392 }
393
394 /* assign default values */
395 if (control_uri == NULL) {
fa91dc52
MD
396 ret = asprintf(&default_address,
397 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
398 DEFAULT_NETWORK_CONTROL_PORT);
b8aa1682
JD
399 if (ret < 0) {
400 PERROR("asprintf default data address");
178a0557 401 retval = -1;
b8aa1682
JD
402 goto exit;
403 }
404
405 ret = uri_parse(default_address, &control_uri);
406 free(default_address);
407 if (ret < 0) {
408 ERR("Invalid control URI specified");
178a0557 409 retval = -1;
b8aa1682
JD
410 goto exit;
411 }
412 }
413 if (data_uri == NULL) {
fa91dc52
MD
414 ret = asprintf(&default_address,
415 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
416 DEFAULT_NETWORK_DATA_PORT);
b8aa1682
JD
417 if (ret < 0) {
418 PERROR("asprintf default data address");
178a0557 419 retval = -1;
b8aa1682
JD
420 goto exit;
421 }
422
423 ret = uri_parse(default_address, &data_uri);
424 free(default_address);
425 if (ret < 0) {
426 ERR("Invalid data URI specified");
178a0557 427 retval = -1;
b8aa1682
JD
428 goto exit;
429 }
430 }
d3e2ba59 431 if (live_uri == NULL) {
fa91dc52
MD
432 ret = asprintf(&default_address,
433 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
434 DEFAULT_NETWORK_VIEWER_PORT);
d3e2ba59
JD
435 if (ret < 0) {
436 PERROR("asprintf default viewer control address");
178a0557 437 retval = -1;
d3e2ba59
JD
438 goto exit;
439 }
440
441 ret = uri_parse(default_address, &live_uri);
442 free(default_address);
443 if (ret < 0) {
444 ERR("Invalid viewer control URI specified");
178a0557 445 retval = -1;
d3e2ba59
JD
446 goto exit;
447 }
448 }
b8aa1682
JD
449
450exit:
cd60b05a 451 free(optstring);
178a0557 452 return retval;
b8aa1682
JD
453}
454
455/*
456 * Cleanup the daemon
457 */
458static
178a0557 459void relayd_cleanup(struct relay_local_data *relay_ctx)
b8aa1682 460{
b8aa1682
JD
461 DBG("Cleaning up");
462
178a0557
MD
463 if (viewer_streams_ht)
464 lttng_ht_destroy(viewer_streams_ht);
465 if (relay_streams_ht)
466 lttng_ht_destroy(relay_streams_ht);
467 if (relay_ctx && relay_ctx->sessions_ht)
468 lttng_ht_destroy(relay_ctx->sessions_ht);
469 free(relay_ctx);
470
095a4ae5
MD
471 /* free the dynamically allocated opt_output_path */
472 free(opt_output_path);
473
a02de639
CB
474 /* Close thread quit pipes */
475 utils_close_pipe(thread_quit_pipe);
476
710c1f73
DG
477 uri_free(control_uri);
478 uri_free(data_uri);
8d5c808e 479 /* Live URI is freed in the live thread. */
cd60b05a
JG
480
481 if (tracing_group_name_override) {
482 free((void *) tracing_group_name);
483 }
b8aa1682
JD
484}
485
486/*
487 * Write to writable pipe used to notify a thread.
488 */
489static
490int notify_thread_pipe(int wpipe)
491{
6cd525e8 492 ssize_t ret;
b8aa1682 493
6cd525e8
MD
494 ret = lttng_write(wpipe, "!", 1);
495 if (ret < 1) {
b8aa1682 496 PERROR("write poll pipe");
b4aacfdc 497 goto end;
b8aa1682 498 }
b4aacfdc
MD
499 ret = 0;
500end:
b8aa1682
JD
501 return ret;
502}
503
b4aacfdc
MD
504static
505int notify_health_quit_pipe(int *pipe)
65931c8b 506{
6cd525e8 507 ssize_t ret;
65931c8b 508
6cd525e8
MD
509 ret = lttng_write(pipe[1], "4", 1);
510 if (ret < 1) {
65931c8b 511 PERROR("write relay health quit");
b4aacfdc 512 goto end;
65931c8b 513 }
b4aacfdc
MD
514 ret = 0;
515end:
516 return ret;
65931c8b
MD
517}
518
b8aa1682 519/*
b4aacfdc 520 * Stop all relayd and relayd-live threads.
b8aa1682 521 */
b4aacfdc 522int lttng_relay_stop_threads(void)
b8aa1682 523{
b4aacfdc 524 int retval = 0;
b8aa1682
JD
525
526 /* Stopping all threads */
527 DBG("Terminating all threads");
b4aacfdc 528 if (notify_thread_pipe(thread_quit_pipe[1])) {
b8aa1682 529 ERR("write error on thread quit pipe");
b4aacfdc 530 retval = -1;
b8aa1682
JD
531 }
532
b4aacfdc
MD
533 if (notify_health_quit_pipe(health_quit_pipe)) {
534 ERR("write error on health quit pipe");
535 }
65931c8b 536
b8aa1682 537 /* Dispatch thread */
26c9d55e 538 CMM_STORE_SHARED(dispatch_thread_exit, 1);
58eb9381 539 futex_nto1_wake(&relay_conn_queue.futex);
178a0557 540
b4aacfdc 541 if (relayd_live_stop()) {
178a0557 542 ERR("Error stopping live threads");
b4aacfdc 543 retval = -1;
178a0557 544 }
b4aacfdc 545 return retval;
b8aa1682
JD
546}
547
548/*
549 * Signal handler for the daemon
550 *
551 * Simply stop all worker threads, leaving main() return gracefully after
552 * joining all threads and calling cleanup().
553 */
554static
555void sighandler(int sig)
556{
557 switch (sig) {
558 case SIGPIPE:
559 DBG("SIGPIPE caught");
560 return;
561 case SIGINT:
562 DBG("SIGINT caught");
b4aacfdc
MD
563 if (lttng_relay_stop_threads()) {
564 ERR("Error stopping threads");
565 }
b8aa1682
JD
566 break;
567 case SIGTERM:
568 DBG("SIGTERM caught");
b4aacfdc
MD
569 if (lttng_relay_stop_threads()) {
570 ERR("Error stopping threads");
571 }
b8aa1682 572 break;
3fd27398
MD
573 case SIGUSR1:
574 CMM_STORE_SHARED(recv_child_signal, 1);
575 break;
b8aa1682
JD
576 default:
577 break;
578 }
579}
580
581/*
582 * Setup signal handler for :
583 * SIGINT, SIGTERM, SIGPIPE
584 */
585static
586int set_signal_handler(void)
587{
588 int ret = 0;
589 struct sigaction sa;
590 sigset_t sigset;
591
592 if ((ret = sigemptyset(&sigset)) < 0) {
593 PERROR("sigemptyset");
594 return ret;
595 }
596
597 sa.sa_handler = sighandler;
598 sa.sa_mask = sigset;
599 sa.sa_flags = 0;
600 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
601 PERROR("sigaction");
602 return ret;
603 }
604
605 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
606 PERROR("sigaction");
607 return ret;
608 }
609
610 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
611 PERROR("sigaction");
612 return ret;
613 }
614
3fd27398
MD
615 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
616 PERROR("sigaction");
617 return ret;
618 }
619
620 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
b8aa1682
JD
621
622 return ret;
623}
624
3fd27398
MD
625void lttng_relay_notify_ready(void)
626{
627 /* Notify the parent of the fork() process that we are ready. */
628 if (opt_daemon || opt_background) {
629 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
630 kill(child_ppid, SIGUSR1);
631 }
632 }
633}
634
b8aa1682
JD
635/*
636 * Init thread quit pipe.
637 *
638 * Return -1 on error or 0 if all pipes are created.
639 */
640static
641int init_thread_quit_pipe(void)
642{
a02de639 643 int ret;
b8aa1682 644
a02de639 645 ret = utils_create_pipe_cloexec(thread_quit_pipe);
b8aa1682 646
b8aa1682
JD
647 return ret;
648}
649
650/*
651 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
652 */
653static
654int create_thread_poll_set(struct lttng_poll_event *events, int size)
655{
656 int ret;
657
658 if (events == NULL || size == 0) {
659 ret = -1;
660 goto error;
661 }
662
663 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
664 if (ret < 0) {
665 goto error;
666 }
667
668 /* Add quit pipe */
c7759e6a 669 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
b8aa1682
JD
670 if (ret < 0) {
671 goto error;
672 }
673
674 return 0;
675
676error:
677 return ret;
678}
679
680/*
681 * Check if the thread quit pipe was triggered.
682 *
683 * Return 1 if it was triggered else 0;
684 */
685static
686int check_thread_quit_pipe(int fd, uint32_t events)
687{
688 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
689 return 1;
690 }
691
692 return 0;
693}
694
695/*
696 * Create and init socket from uri.
697 */
698static
699struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
700{
701 int ret;
702 struct lttcomm_sock *sock = NULL;
703
704 sock = lttcomm_alloc_sock_from_uri(uri);
705 if (sock == NULL) {
706 ERR("Allocating socket");
707 goto error;
708 }
709
710 ret = lttcomm_create_sock(sock);
711 if (ret < 0) {
712 goto error;
713 }
714 DBG("Listening on sock %d", sock->fd);
715
716 ret = sock->ops->bind(sock);
717 if (ret < 0) {
718 goto error;
719 }
720
721 ret = sock->ops->listen(sock, -1);
722 if (ret < 0) {
723 goto error;
724
725 }
726
727 return sock;
728
729error:
730 if (sock) {
731 lttcomm_destroy_sock(sock);
732 }
733 return NULL;
734}
735
173af62f
DG
736/*
737 * Return nonzero if stream needs to be closed.
738 */
739static
740int close_stream_check(struct relay_stream *stream)
741{
173af62f 742 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
f7079f67
DG
743 /*
744 * We are about to close the stream so set the data pending flag to 1
745 * which will make the end data pending command skip the stream which
746 * is now closed and ready. Note that after proceeding to a file close,
747 * the written file is ready for reading.
748 */
749 stream->data_pending_check_done = 1;
173af62f
DG
750 return 1;
751 }
752 return 0;
753}
754
2a174661
DG
755static void try_close_stream(struct relay_session *session,
756 struct relay_stream *stream)
757{
758 int ret;
759 struct ctf_trace *ctf_trace;
760
761 assert(session);
762 assert(stream);
763
764 if (!close_stream_check(stream)) {
765 /* Can't close it, not ready for that. */
766 goto end;
767 }
768
769 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
770 stream->path_name);
771 assert(ctf_trace);
772
773 pthread_mutex_lock(&session->viewer_ready_lock);
774 ctf_trace->invalid_flag = 1;
775 pthread_mutex_unlock(&session->viewer_ready_lock);
776
777 ret = stream_close(session, stream);
f94b19e6 778 if (ret || session->snapshot) {
2a174661
DG
779 /* Already close thus the ctf trace is being or has been destroyed. */
780 goto end;
781 }
782
783 ctf_trace_try_destroy(session, ctf_trace);
784
785end:
786 return;
787}
788
b8aa1682
JD
789/*
790 * This thread manages the listening for new connections on the network
791 */
792static
793void *relay_thread_listener(void *data)
794{
095a4ae5 795 int i, ret, pollfd, err = -1;
b8aa1682
JD
796 uint32_t revents, nb_fd;
797 struct lttng_poll_event events;
798 struct lttcomm_sock *control_sock, *data_sock;
799
b8aa1682
JD
800 DBG("[thread] Relay listener started");
801
55706a7d
MD
802 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
803
f385ae0a
MD
804 health_code_update();
805
b8aa1682
JD
806 control_sock = relay_init_sock(control_uri);
807 if (!control_sock) {
095a4ae5 808 goto error_sock_control;
b8aa1682
JD
809 }
810
811 data_sock = relay_init_sock(data_uri);
812 if (!data_sock) {
095a4ae5 813 goto error_sock_relay;
b8aa1682
JD
814 }
815
816 /*
817 * Pass 3 as size here for the thread quit pipe, control and data socket.
818 */
819 ret = create_thread_poll_set(&events, 3);
820 if (ret < 0) {
821 goto error_create_poll;
822 }
823
824 /* Add the control socket */
825 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
826 if (ret < 0) {
827 goto error_poll_add;
828 }
829
830 /* Add the data socket */
831 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
832 if (ret < 0) {
833 goto error_poll_add;
834 }
835
3fd27398
MD
836 lttng_relay_notify_ready();
837
9b5e0863
MD
838 if (testpoint(relayd_thread_listener)) {
839 goto error_testpoint;
840 }
841
b8aa1682 842 while (1) {
f385ae0a
MD
843 health_code_update();
844
b8aa1682
JD
845 DBG("Listener accepting connections");
846
b8aa1682 847restart:
f385ae0a 848 health_poll_entry();
b8aa1682 849 ret = lttng_poll_wait(&events, -1);
f385ae0a 850 health_poll_exit();
b8aa1682
JD
851 if (ret < 0) {
852 /*
853 * Restart interrupted system call.
854 */
855 if (errno == EINTR) {
856 goto restart;
857 }
858 goto error;
859 }
860
0d9c5d77
DG
861 nb_fd = ret;
862
b8aa1682
JD
863 DBG("Relay new connection received");
864 for (i = 0; i < nb_fd; i++) {
f385ae0a
MD
865 health_code_update();
866
b8aa1682
JD
867 /* Fetch once the poll data */
868 revents = LTTNG_POLL_GETEV(&events, i);
869 pollfd = LTTNG_POLL_GETFD(&events, i);
870
fd20dac9
MD
871 if (!revents) {
872 /* No activity for this FD (poll implementation). */
873 continue;
874 }
875
b8aa1682
JD
876 /* Thread quit pipe has been closed. Killing thread. */
877 ret = check_thread_quit_pipe(pollfd, revents);
878 if (ret) {
095a4ae5
MD
879 err = 0;
880 goto exit;
b8aa1682
JD
881 }
882
883 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
884 ERR("socket poll error");
885 goto error;
886 } else if (revents & LPOLLIN) {
4b7f17b2 887 /*
58eb9381
DG
888 * Get allocated in this thread, enqueued to a global queue,
889 * dequeued and freed in the worker thread.
4b7f17b2 890 */
58eb9381
DG
891 int val = 1;
892 struct relay_connection *new_conn;
4b7f17b2 893 struct lttcomm_sock *newsock;
b8aa1682 894
58eb9381
DG
895 new_conn = connection_create();
896 if (!new_conn) {
b8aa1682
JD
897 goto error;
898 }
899
900 if (pollfd == data_sock->fd) {
58eb9381 901 new_conn->type = RELAY_DATA;
b8aa1682 902 newsock = data_sock->ops->accept(data_sock);
58eb9381
DG
903 DBG("Relay data connection accepted, socket %d",
904 newsock->fd);
4b7f17b2
MD
905 } else {
906 assert(pollfd == control_sock->fd);
58eb9381 907 new_conn->type = RELAY_CONTROL;
b8aa1682 908 newsock = control_sock->ops->accept(control_sock);
58eb9381
DG
909 DBG("Relay control connection accepted, socket %d",
910 newsock->fd);
b8aa1682 911 }
58eb9381
DG
912 if (!newsock) {
913 PERROR("accepting sock");
914 connection_free(new_conn);
915 goto error;
916 }
917
918 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
919 sizeof(val));
b8aa1682
JD
920 if (ret < 0) {
921 PERROR("setsockopt inet");
4b7f17b2 922 lttcomm_destroy_sock(newsock);
58eb9381 923 connection_free(new_conn);
b8aa1682
JD
924 goto error;
925 }
58eb9381
DG
926 new_conn->sock = newsock;
927
928 /* Enqueue request for the dispatcher thread. */
8bdee6e2
SM
929 cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail,
930 &new_conn->qnode);
b8aa1682
JD
931
932 /*
58eb9381 933 * Wake the dispatch queue futex. Implicit memory barrier with
8bdee6e2 934 * the exchange in cds_wfcq_enqueue.
b8aa1682 935 */
58eb9381 936 futex_nto1_wake(&relay_conn_queue.futex);
b8aa1682
JD
937 }
938 }
939 }
940
095a4ae5 941exit:
b8aa1682
JD
942error:
943error_poll_add:
9b5e0863 944error_testpoint:
b8aa1682
JD
945 lttng_poll_clean(&events);
946error_create_poll:
095a4ae5
MD
947 if (data_sock->fd >= 0) {
948 ret = data_sock->ops->close(data_sock);
b8aa1682
JD
949 if (ret) {
950 PERROR("close");
951 }
b8aa1682 952 }
095a4ae5
MD
953 lttcomm_destroy_sock(data_sock);
954error_sock_relay:
955 if (control_sock->fd >= 0) {
956 ret = control_sock->ops->close(control_sock);
b8aa1682
JD
957 if (ret) {
958 PERROR("close");
959 }
b8aa1682 960 }
095a4ae5
MD
961 lttcomm_destroy_sock(control_sock);
962error_sock_control:
963 if (err) {
f385ae0a
MD
964 health_error();
965 ERR("Health error occurred in %s", __func__);
095a4ae5 966 }
55706a7d 967 health_unregister(health_relayd);
b8aa1682 968 DBG("Relay listener thread cleanup complete");
b4aacfdc 969 lttng_relay_stop_threads();
b8aa1682
JD
970 return NULL;
971}
972
973/*
974 * This thread manages the dispatching of the requests to worker threads
975 */
976static
977void *relay_thread_dispatcher(void *data)
978{
6cd525e8
MD
979 int err = -1;
980 ssize_t ret;
8bdee6e2 981 struct cds_wfcq_node *node;
58eb9381 982 struct relay_connection *new_conn = NULL;
b8aa1682
JD
983
984 DBG("[thread] Relay dispatcher started");
985
55706a7d
MD
986 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
987
9b5e0863
MD
988 if (testpoint(relayd_thread_dispatcher)) {
989 goto error_testpoint;
990 }
991
f385ae0a
MD
992 health_code_update();
993
26c9d55e 994 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
f385ae0a
MD
995 health_code_update();
996
b8aa1682 997 /* Atomically prepare the queue futex */
58eb9381 998 futex_nto1_prepare(&relay_conn_queue.futex);
b8aa1682
JD
999
1000 do {
f385ae0a
MD
1001 health_code_update();
1002
b8aa1682 1003 /* Dequeue commands */
8bdee6e2
SM
1004 node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
1005 &relay_conn_queue.tail);
b8aa1682
JD
1006 if (node == NULL) {
1007 DBG("Woken up but nothing in the relay command queue");
1008 /* Continue thread execution */
1009 break;
1010 }
58eb9381 1011 new_conn = caa_container_of(node, struct relay_connection, qnode);
b8aa1682 1012
58eb9381 1013 DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
b8aa1682
JD
1014
1015 /*
58eb9381
DG
1016 * Inform worker thread of the new request. This call is blocking
1017 * so we can be assured that the data will be read at some point in
1018 * time or wait to the end of the world :)
b8aa1682 1019 */
58eb9381
DG
1020 ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn));
1021 if (ret < 0) {
1022 PERROR("write connection pipe");
1023 connection_destroy(new_conn);
b8aa1682
JD
1024 goto error;
1025 }
1026 } while (node != NULL);
1027
1028 /* Futex wait on queue. Blocking call on futex() */
f385ae0a 1029 health_poll_entry();
58eb9381 1030 futex_nto1_wait(&relay_conn_queue.futex);
f385ae0a 1031 health_poll_exit();
b8aa1682
JD
1032 }
1033
f385ae0a
MD
1034 /* Normal exit, no error */
1035 err = 0;
1036
b8aa1682 1037error:
9b5e0863 1038error_testpoint:
f385ae0a
MD
1039 if (err) {
1040 health_error();
1041 ERR("Health error occurred in %s", __func__);
1042 }
55706a7d 1043 health_unregister(health_relayd);
b8aa1682 1044 DBG("Dispatch thread dying");
b4aacfdc 1045 lttng_relay_stop_threads();
b8aa1682
JD
1046 return NULL;
1047}
1048
2a174661 1049static void try_close_streams(struct relay_session *session)
d3e2ba59 1050{
2a174661 1051 struct ctf_trace *ctf_trace;
94d49140
JD
1052 struct lttng_ht_iter iter;
1053
2a174661 1054 assert(session);
94d49140 1055
2a174661
DG
1056 pthread_mutex_lock(&session->viewer_ready_lock);
1057 rcu_read_lock();
1058 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
1059 node.node) {
1060 struct relay_stream *stream;
94d49140 1061
2a174661
DG
1062 /* Close streams. */
1063 cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) {
1064 stream_close(session, stream);
94d49140 1065 }
94d49140 1066
2a174661
DG
1067 ctf_trace->invalid_flag = 1;
1068 ctf_trace_try_destroy(session, ctf_trace);
157df586 1069 }
2a174661
DG
1070 rcu_read_unlock();
1071 pthread_mutex_unlock(&session->viewer_ready_lock);
94d49140
JD
1072}
1073
b8aa1682 1074/*
2a174661 1075 * Try to destroy a session within a connection.
b8aa1682 1076 */
58eb9381 1077static void destroy_session(struct relay_session *session,
d3e2ba59 1078 struct lttng_ht *sessions_ht)
b8aa1682 1079{
58eb9381 1080 assert(session);
2a174661 1081 assert(sessions_ht);
b8aa1682 1082
2a174661 1083 /* Indicate that this session can be destroyed from now on. */
58eb9381 1084 session->close_flag = 1;
b8aa1682 1085
58eb9381 1086 try_close_streams(session);
5b6d8097 1087
2a174661
DG
1088 /*
1089 * This will try to delete and destroy the session if no viewer is attached
1090 * to it meaning the refcount is down to zero.
1091 */
58eb9381 1092 session_try_destroy(sessions_ht, session);
b8aa1682
JD
1093}
1094
1c20f0e2
JD
1095/*
1096 * Copy index data from the control port to a given index object.
1097 */
1098static void copy_index_control_data(struct relay_index *index,
1099 struct lttcomm_relayd_index *data)
1100{
1101 assert(index);
1102 assert(data);
1103
1104 /*
1105 * The index on disk is encoded in big endian, so we don't need to convert
1106 * the data received on the network. The data_offset value is NEVER
1107 * modified here and is updated by the data thread.
1108 */
1109 index->index_data.packet_size = data->packet_size;
1110 index->index_data.content_size = data->content_size;
1111 index->index_data.timestamp_begin = data->timestamp_begin;
1112 index->index_data.timestamp_end = data->timestamp_end;
1113 index->index_data.events_discarded = data->events_discarded;
1114 index->index_data.stream_id = data->stream_id;
1115}
1116
c5b6f4f0
DG
1117/*
1118 * Handle the RELAYD_CREATE_SESSION command.
1119 *
1120 * On success, send back the session id or else return a negative value.
1121 */
1122static
1123int relay_create_session(struct lttcomm_relayd_hdr *recv_hdr,
58eb9381 1124 struct relay_connection *conn)
c5b6f4f0
DG
1125{
1126 int ret = 0, send_ret;
1127 struct relay_session *session;
1128 struct lttcomm_relayd_status_session reply;
1129
1130 assert(recv_hdr);
58eb9381 1131 assert(conn);
c5b6f4f0
DG
1132
1133 memset(&reply, 0, sizeof(reply));
1134
2a174661
DG
1135 session = session_create();
1136 if (!session) {
c5b6f4f0
DG
1137 ret = -1;
1138 goto error;
1139 }
58eb9381
DG
1140 session->minor = conn->minor;
1141 session->major = conn->major;
1142 conn->session_id = session->id;
1143 conn->session = session;
c5b6f4f0
DG
1144
1145 reply.session_id = htobe64(session->id);
1146
58eb9381 1147 switch (conn->minor) {
2a174661
DG
1148 case 1:
1149 case 2:
1150 case 3:
1151 break;
1152 case 4: /* LTTng sessiond 2.4 */
1153 default:
58eb9381 1154 ret = cmd_create_session_2_4(conn, session);
d3e2ba59
JD
1155 }
1156
58eb9381 1157 lttng_ht_add_unique_u64(conn->sessions_ht, &session->session_n);
c5b6f4f0
DG
1158 DBG("Created session %" PRIu64, session->id);
1159
1160error:
1161 if (ret < 0) {
1162 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
1163 } else {
1164 reply.ret_code = htobe32(LTTNG_OK);
1165 }
1166
58eb9381 1167 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
c5b6f4f0
DG
1168 if (send_ret < 0) {
1169 ERR("Relayd sending session id");
4169f5ad 1170 ret = send_ret;
c5b6f4f0
DG
1171 }
1172
1173 return ret;
1174}
1175
a4baae1b
JD
1176/*
1177 * When we have received all the streams and the metadata for a channel,
1178 * we make them visible to the viewer threads.
1179 */
1180static
58eb9381 1181void set_viewer_ready_flag(struct relay_connection *conn)
a4baae1b 1182{
2a174661 1183 struct relay_stream *stream, *tmp_stream;
a4baae1b 1184
58eb9381
DG
1185 pthread_mutex_lock(&conn->session->viewer_ready_lock);
1186 cds_list_for_each_entry_safe(stream, tmp_stream, &conn->recv_head,
2a174661 1187 recv_list) {
a4baae1b 1188 stream->viewer_ready = 1;
2a174661 1189 cds_list_del(&stream->recv_list);
a4baae1b 1190 }
58eb9381 1191 pthread_mutex_unlock(&conn->session->viewer_ready_lock);
a4baae1b
JD
1192 return;
1193}
1194
1195/*
1196 * Add a recv handle node to the connection recv list with the given stream
1197 * handle. A new node is allocated thus must be freed when the node is deleted
1198 * from the list.
1199 */
58eb9381
DG
1200static void queue_stream(struct relay_stream *stream,
1201 struct relay_connection *conn)
a4baae1b 1202{
58eb9381 1203 assert(conn);
2a174661 1204 assert(stream);
a4baae1b 1205
58eb9381 1206 cds_list_add(&stream->recv_list, &conn->recv_head);
a4baae1b
JD
1207}
1208
b8aa1682
JD
1209/*
1210 * relay_add_stream: allocate a new stream for a session
1211 */
1212static
1213int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
58eb9381 1214 struct relay_connection *conn)
b8aa1682 1215{
2a174661 1216 int ret, send_ret;
58eb9381 1217 struct relay_session *session = conn->session;
b8aa1682
JD
1218 struct relay_stream *stream = NULL;
1219 struct lttcomm_relayd_status_stream reply;
2a174661 1220 struct ctf_trace *trace;
b8aa1682 1221
58eb9381 1222 if (!session || conn->version_check_done == 0) {
b8aa1682
JD
1223 ERR("Trying to add a stream before version check");
1224 ret = -1;
1225 goto end_no_session;
1226 }
1227
b8aa1682
JD
1228 stream = zmalloc(sizeof(struct relay_stream));
1229 if (stream == NULL) {
1230 PERROR("relay stream zmalloc");
1231 ret = -1;
1232 goto end_no_session;
1233 }
1234
58eb9381 1235 switch (conn->minor) {
0f907de1 1236 case 1: /* LTTng sessiond 2.1 */
58eb9381 1237 ret = cmd_recv_stream_2_1(conn, stream);
0f907de1
JD
1238 break;
1239 case 2: /* LTTng sessiond 2.2 */
1240 default:
58eb9381 1241 ret = cmd_recv_stream_2_2(conn, stream);
0f907de1
JD
1242 break;
1243 }
1244 if (ret < 0) {
1245 goto err_free_stream;
1246 }
1247
9d1bbf21 1248 rcu_read_lock();
b8aa1682 1249 stream->stream_handle = ++last_relay_stream_id;
173af62f 1250 stream->prev_seq = -1ULL;
2a174661 1251 stream->session_id = session->id;
1c20f0e2 1252 stream->index_fd = -1;
d3e2ba59 1253 stream->read_index_fd = -1;
528f2ffa 1254 stream->ctf_stream_id = -1ULL;
2a174661 1255 lttng_ht_node_init_u64(&stream->node, stream->stream_handle);
d3e2ba59 1256 pthread_mutex_init(&stream->lock, NULL);
b8aa1682 1257
0f907de1 1258 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG);
b8aa1682 1259 if (ret < 0) {
b8aa1682
JD
1260 ERR("relay creating output directory");
1261 goto end;
1262 }
1263
be96a7d1
DG
1264 /*
1265 * No need to use run_as API here because whatever we receives, the relayd
1266 * uses its own credentials for the stream files.
1267 */
0f907de1 1268 ret = utils_create_stream_file(stream->path_name, stream->channel_name,
1c20f0e2 1269 stream->tracefile_size, 0, relayd_uid, relayd_gid, NULL);
b8aa1682 1270 if (ret < 0) {
0f907de1 1271 ERR("Create output file");
b8aa1682
JD
1272 goto end;
1273 }
b8aa1682 1274 stream->fd = ret;
0f907de1
JD
1275 if (stream->tracefile_size) {
1276 DBG("Tracefile %s/%s_0 created", stream->path_name, stream->channel_name);
1277 } else {
1278 DBG("Tracefile %s/%s created", stream->path_name, stream->channel_name);
1279 }
b8aa1682 1280
2a174661
DG
1281 trace = ctf_trace_find_by_path(session->ctf_traces_ht, stream->path_name);
1282 if (!trace) {
1283 trace = ctf_trace_create(stream->path_name);
1284 if (!trace) {
d3e2ba59
JD
1285 ret = -1;
1286 goto end;
1287 }
2a174661
DG
1288 ctf_trace_add(session->ctf_traces_ht, trace);
1289 }
1290 ctf_trace_get_ref(trace);
1291
1292 if (!strncmp(stream->channel_name, DEFAULT_METADATA_NAME, NAME_MAX)) {
1293 stream->metadata_flag = 1;
1294 /* Assign quick reference to the metadata stream in the trace. */
1295 trace->metadata_stream = stream;
d3e2ba59 1296 }
d3e2ba59 1297
a4baae1b 1298 /*
2a174661
DG
1299 * Add the stream in the recv list of the connection. Once the end stream
1300 * message is received, this list is emptied and streams are set with the
1301 * viewer ready flag.
a4baae1b 1302 */
58eb9381 1303 queue_stream(stream, conn);
a4baae1b 1304
2a174661
DG
1305 /*
1306 * Both in the ctf_trace object and the global stream ht since the data
1307 * side of the relayd does not have the concept of session.
1308 */
1309 lttng_ht_add_unique_u64(relay_streams_ht, &stream->node);
1310 cds_list_add_tail(&stream->trace_list, &trace->stream_list);
b8aa1682 1311
87b576ec 1312 session->stream_count++;
d3e2ba59 1313
1c20f0e2
JD
1314 DBG("Relay new stream added %s with ID %" PRIu64, stream->channel_name,
1315 stream->stream_handle);
b8aa1682
JD
1316
1317end:
53efb85a 1318 memset(&reply, 0, sizeof(reply));
5af40280 1319 reply.handle = htobe64(stream->stream_handle);
b8aa1682
JD
1320 /* send the session id to the client or a negative return code on error */
1321 if (ret < 0) {
f73fabfd 1322 reply.ret_code = htobe32(LTTNG_ERR_UNK);
5af40280
CB
1323 /* stream was not properly added to the ht, so free it */
1324 free(stream);
b8aa1682 1325 } else {
f73fabfd 1326 reply.ret_code = htobe32(LTTNG_OK);
b8aa1682 1327 }
5af40280 1328
58eb9381 1329 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
b8aa1682
JD
1330 sizeof(struct lttcomm_relayd_status_stream), 0);
1331 if (send_ret < 0) {
1332 ERR("Relay sending stream id");
4169f5ad 1333 ret = send_ret;
b8aa1682 1334 }
9d1bbf21 1335 rcu_read_unlock();
b8aa1682
JD
1336
1337end_no_session:
1338 return ret;
0f907de1
JD
1339
1340err_free_stream:
1341 free(stream->path_name);
1342 free(stream->channel_name);
1343 free(stream);
1344 return ret;
b8aa1682
JD
1345}
1346
173af62f
DG
1347/*
1348 * relay_close_stream: close a specific stream
1349 */
1350static
1351int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
58eb9381 1352 struct relay_connection *conn)
173af62f 1353{
94d49140 1354 int ret, send_ret;
58eb9381 1355 struct relay_session *session = conn->session;
173af62f
DG
1356 struct lttcomm_relayd_close_stream stream_info;
1357 struct lttcomm_relayd_generic_reply reply;
1358 struct relay_stream *stream;
173af62f
DG
1359
1360 DBG("Close stream received");
1361
58eb9381 1362 if (!session || conn->version_check_done == 0) {
173af62f
DG
1363 ERR("Trying to close a stream before version check");
1364 ret = -1;
1365 goto end_no_session;
1366 }
1367
58eb9381 1368 ret = conn->sock->ops->recvmsg(conn->sock, &stream_info,
7c5aef62 1369 sizeof(struct lttcomm_relayd_close_stream), 0);
173af62f 1370 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
a6cd2b97
DG
1371 if (ret == 0) {
1372 /* Orderly shutdown. Not necessary to print an error. */
58eb9381 1373 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
a6cd2b97
DG
1374 } else {
1375 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1376 }
173af62f
DG
1377 ret = -1;
1378 goto end_no_session;
1379 }
1380
1381 rcu_read_lock();
2a174661
DG
1382 stream = stream_find_by_id(relay_streams_ht,
1383 be64toh(stream_info.stream_id));
173af62f
DG
1384 if (!stream) {
1385 ret = -1;
1386 goto end_unlock;
1387 }
1388
8e2583a4 1389 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
173af62f 1390 stream->close_flag = 1;
87b576ec 1391 session->stream_count--;
173af62f 1392
2a174661
DG
1393 /* Check if we can close it or else the data will do it. */
1394 try_close_stream(session, stream);
173af62f
DG
1395
1396end_unlock:
1397 rcu_read_unlock();
1398
53efb85a 1399 memset(&reply, 0, sizeof(reply));
173af62f 1400 if (ret < 0) {
f73fabfd 1401 reply.ret_code = htobe32(LTTNG_ERR_UNK);
173af62f 1402 } else {
f73fabfd 1403 reply.ret_code = htobe32(LTTNG_OK);
173af62f 1404 }
58eb9381 1405 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
173af62f
DG
1406 sizeof(struct lttcomm_relayd_generic_reply), 0);
1407 if (send_ret < 0) {
1408 ERR("Relay sending stream id");
4169f5ad 1409 ret = send_ret;
173af62f
DG
1410 }
1411
1412end_no_session:
1413 return ret;
1414}
1415
b8aa1682
JD
1416/*
1417 * relay_unknown_command: send -1 if received unknown command
1418 */
1419static
58eb9381 1420void relay_unknown_command(struct relay_connection *conn)
b8aa1682
JD
1421{
1422 struct lttcomm_relayd_generic_reply reply;
1423 int ret;
1424
53efb85a 1425 memset(&reply, 0, sizeof(reply));
f73fabfd 1426 reply.ret_code = htobe32(LTTNG_ERR_UNK);
58eb9381 1427 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
b8aa1682
JD
1428 sizeof(struct lttcomm_relayd_generic_reply), 0);
1429 if (ret < 0) {
1430 ERR("Relay sending unknown command");
1431 }
1432}
1433
1434/*
1435 * relay_start: send an acknowledgment to the client to tell if we are
1436 * ready to receive data. We are ready if a session is established.
1437 */
1438static
1439int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
58eb9381 1440 struct relay_connection *conn)
b8aa1682 1441{
f73fabfd 1442 int ret = htobe32(LTTNG_OK);
b8aa1682 1443 struct lttcomm_relayd_generic_reply reply;
58eb9381 1444 struct relay_session *session = conn->session;
b8aa1682
JD
1445
1446 if (!session) {
1447 DBG("Trying to start the streaming without a session established");
f73fabfd 1448 ret = htobe32(LTTNG_ERR_UNK);
b8aa1682
JD
1449 }
1450
53efb85a 1451 memset(&reply, 0, sizeof(reply));
b8aa1682 1452 reply.ret_code = ret;
58eb9381 1453 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
b8aa1682
JD
1454 sizeof(struct lttcomm_relayd_generic_reply), 0);
1455 if (ret < 0) {
1456 ERR("Relay sending start ack");
1457 }
1458
1459 return ret;
1460}
1461
1d4dfdef
DG
1462/*
1463 * Append padding to the file pointed by the file descriptor fd.
1464 */
1465static int write_padding_to_file(int fd, uint32_t size)
1466{
6cd525e8 1467 ssize_t ret = 0;
1d4dfdef
DG
1468 char *zeros;
1469
1470 if (size == 0) {
1471 goto end;
1472 }
1473
1474 zeros = zmalloc(size);
1475 if (zeros == NULL) {
1476 PERROR("zmalloc zeros for padding");
1477 ret = -1;
1478 goto end;
1479 }
1480
6cd525e8
MD
1481 ret = lttng_write(fd, zeros, size);
1482 if (ret < size) {
1d4dfdef
DG
1483 PERROR("write padding to file");
1484 }
1485
e986c7a1
DG
1486 free(zeros);
1487
1d4dfdef
DG
1488end:
1489 return ret;
1490}
1491
b8aa1682
JD
1492/*
1493 * relay_recv_metadata: receive the metada for the session.
1494 */
1495static
1496int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
58eb9381 1497 struct relay_connection *conn)
b8aa1682 1498{
f73fabfd 1499 int ret = htobe32(LTTNG_OK);
6cd525e8 1500 ssize_t size_ret;
58eb9381 1501 struct relay_session *session = conn->session;
b8aa1682
JD
1502 struct lttcomm_relayd_metadata_payload *metadata_struct;
1503 struct relay_stream *metadata_stream;
1504 uint64_t data_size, payload_size;
2a174661 1505 struct ctf_trace *ctf_trace;
b8aa1682
JD
1506
1507 if (!session) {
1508 ERR("Metadata sent before version check");
1509 ret = -1;
1510 goto end;
1511 }
1512
f6416125
MD
1513 data_size = payload_size = be64toh(recv_hdr->data_size);
1514 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1515 ERR("Incorrect data size");
1516 ret = -1;
1517 goto end;
1518 }
1519 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1520
b8aa1682 1521 if (data_buffer_size < data_size) {
d7b3776f 1522 /* In case the realloc fails, we can free the memory */
c617c0c6
MD
1523 char *tmp_data_ptr;
1524
1525 tmp_data_ptr = realloc(data_buffer, data_size);
1526 if (!tmp_data_ptr) {
b8aa1682 1527 ERR("Allocating data buffer");
c617c0c6 1528 free(data_buffer);
b8aa1682
JD
1529 ret = -1;
1530 goto end;
1531 }
c617c0c6 1532 data_buffer = tmp_data_ptr;
b8aa1682
JD
1533 data_buffer_size = data_size;
1534 }
1535 memset(data_buffer, 0, data_size);
77c7c900 1536 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
58eb9381 1537 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0);
b8aa1682 1538 if (ret < 0 || ret != data_size) {
a6cd2b97
DG
1539 if (ret == 0) {
1540 /* Orderly shutdown. Not necessary to print an error. */
58eb9381 1541 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
a6cd2b97
DG
1542 } else {
1543 ERR("Relay didn't receive the whole metadata");
1544 }
b8aa1682 1545 ret = -1;
b8aa1682
JD
1546 goto end;
1547 }
1548 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
9d1bbf21
MD
1549
1550 rcu_read_lock();
2a174661 1551 metadata_stream = stream_find_by_id(relay_streams_ht,
d3e2ba59 1552 be64toh(metadata_struct->stream_id));
b8aa1682
JD
1553 if (!metadata_stream) {
1554 ret = -1;
9d1bbf21 1555 goto end_unlock;
b8aa1682
JD
1556 }
1557
6cd525e8
MD
1558 size_ret = lttng_write(metadata_stream->fd, metadata_struct->payload,
1559 payload_size);
1560 if (size_ret < payload_size) {
b8aa1682
JD
1561 ERR("Relay error writing metadata on file");
1562 ret = -1;
9d1bbf21 1563 goto end_unlock;
b8aa1682 1564 }
1d4dfdef
DG
1565
1566 ret = write_padding_to_file(metadata_stream->fd,
1567 be32toh(metadata_struct->padding_size));
1568 if (ret < 0) {
1569 goto end_unlock;
1570 }
2a174661
DG
1571
1572 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1573 metadata_stream->path_name);
1574 assert(ctf_trace);
1575 ctf_trace->metadata_received +=
d3e2ba59 1576 payload_size + be32toh(metadata_struct->padding_size);
1d4dfdef 1577
b8aa1682
JD
1578 DBG2("Relay metadata written");
1579
9d1bbf21 1580end_unlock:
6e3c5836 1581 rcu_read_unlock();
b8aa1682
JD
1582end:
1583 return ret;
1584}
1585
1586/*
1587 * relay_send_version: send relayd version number
1588 */
1589static
1590int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
58eb9381 1591 struct relay_connection *conn)
b8aa1682 1592{
7f51dcba 1593 int ret;
092b6259 1594 struct lttcomm_relayd_version reply, msg;
b8aa1682 1595
58eb9381 1596 assert(conn);
c5b6f4f0 1597
58eb9381 1598 conn->version_check_done = 1;
b8aa1682 1599
092b6259 1600 /* Get version from the other side. */
58eb9381 1601 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
092b6259 1602 if (ret < 0 || ret != sizeof(msg)) {
a6cd2b97
DG
1603 if (ret == 0) {
1604 /* Orderly shutdown. Not necessary to print an error. */
58eb9381 1605 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
a6cd2b97
DG
1606 } else {
1607 ERR("Relay failed to receive the version values.");
1608 }
092b6259 1609 ret = -1;
092b6259
DG
1610 goto end;
1611 }
1612
53efb85a 1613 memset(&reply, 0, sizeof(reply));
d83a952c
MD
1614 reply.major = RELAYD_VERSION_COMM_MAJOR;
1615 reply.minor = RELAYD_VERSION_COMM_MINOR;
d4519fa3
JD
1616
1617 /* Major versions must be the same */
1618 if (reply.major != be32toh(msg.major)) {
6151a90f
JD
1619 DBG("Incompatible major versions (%u vs %u), deleting session",
1620 reply.major, be32toh(msg.major));
58eb9381 1621 destroy_session(conn->session, conn->sessions_ht);
d4519fa3
JD
1622 ret = 0;
1623 goto end;
1624 }
1625
58eb9381 1626 conn->major = reply.major;
0f907de1
JD
1627 /* We adapt to the lowest compatible version */
1628 if (reply.minor <= be32toh(msg.minor)) {
58eb9381 1629 conn->minor = reply.minor;
0f907de1 1630 } else {
58eb9381 1631 conn->minor = be32toh(msg.minor);
0f907de1
JD
1632 }
1633
6151a90f
JD
1634 reply.major = htobe32(reply.major);
1635 reply.minor = htobe32(reply.minor);
58eb9381 1636 ret = conn->sock->ops->sendmsg(conn->sock, &reply,
6151a90f
JD
1637 sizeof(struct lttcomm_relayd_version), 0);
1638 if (ret < 0) {
1639 ERR("Relay sending version");
1640 }
1641
58eb9381
DG
1642 DBG("Version check done using protocol %u.%u", conn->major,
1643 conn->minor);
b8aa1682
JD
1644
1645end:
1646 return ret;
1647}
1648
c8f59ee5 1649/*
6d805429 1650 * Check for data pending for a given stream id from the session daemon.
c8f59ee5
DG
1651 */
1652static
6d805429 1653int relay_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
58eb9381 1654 struct relay_connection *conn)
c8f59ee5 1655{
58eb9381 1656 struct relay_session *session = conn->session;
6d805429 1657 struct lttcomm_relayd_data_pending msg;
c8f59ee5
DG
1658 struct lttcomm_relayd_generic_reply reply;
1659 struct relay_stream *stream;
1660 int ret;
c8f59ee5
DG
1661 uint64_t last_net_seq_num, stream_id;
1662
6d805429 1663 DBG("Data pending command received");
c8f59ee5 1664
58eb9381 1665 if (!session || conn->version_check_done == 0) {
c8f59ee5
DG
1666 ERR("Trying to check for data before version check");
1667 ret = -1;
1668 goto end_no_session;
1669 }
1670
58eb9381 1671 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
c8f59ee5 1672 if (ret < sizeof(msg)) {
a6cd2b97
DG
1673 if (ret == 0) {
1674 /* Orderly shutdown. Not necessary to print an error. */
58eb9381 1675 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
a6cd2b97
DG
1676 } else {
1677 ERR("Relay didn't receive valid data_pending struct size : %d",
1678 ret);
1679 }
c8f59ee5
DG
1680 ret = -1;
1681 goto end_no_session;
1682 }
1683
1684 stream_id = be64toh(msg.stream_id);
1685 last_net_seq_num = be64toh(msg.last_net_seq_num);
1686
1687 rcu_read_lock();
2a174661 1688 stream = stream_find_by_id(relay_streams_ht, stream_id);
de91f48a 1689 if (stream == NULL) {
c8f59ee5
DG
1690 ret = -1;
1691 goto end_unlock;
1692 }
1693
6d805429 1694 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
c8f59ee5
DG
1695 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1696 last_net_seq_num);
1697
33832e64 1698 /* Avoid wrapping issue */
39df6d9f 1699 if (((int64_t) (stream->prev_seq - last_net_seq_num)) >= 0) {
6d805429 1700 /* Data has in fact been written and is NOT pending */
c8f59ee5 1701 ret = 0;
6d805429
DG
1702 } else {
1703 /* Data still being streamed thus pending */
1704 ret = 1;
c8f59ee5
DG
1705 }
1706
f7079f67
DG
1707 /* Pending check is now done. */
1708 stream->data_pending_check_done = 1;
1709
c8f59ee5
DG
1710end_unlock:
1711 rcu_read_unlock();
1712
53efb85a 1713 memset(&reply, 0, sizeof(reply));
c8f59ee5 1714 reply.ret_code = htobe32(ret);
58eb9381 1715 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
c8f59ee5 1716 if (ret < 0) {
6d805429 1717 ERR("Relay data pending ret code failed");
c8f59ee5
DG
1718 }
1719
1720end_no_session:
1721 return ret;
1722}
1723
1724/*
1725 * Wait for the control socket to reach a quiescent state.
1726 *
1727 * Note that for now, when receiving this command from the session daemon, this
1728 * means that every subsequent commands or data received on the control socket
1729 * has been handled. So, this is why we simply return OK here.
1730 */
1731static
1732int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
58eb9381 1733 struct relay_connection *conn)
c8f59ee5
DG
1734{
1735 int ret;
ad7051c0
DG
1736 uint64_t stream_id;
1737 struct relay_stream *stream;
1738 struct lttng_ht_iter iter;
1739 struct lttcomm_relayd_quiescent_control msg;
c8f59ee5
DG
1740 struct lttcomm_relayd_generic_reply reply;
1741
1742 DBG("Checking quiescent state on control socket");
1743
58eb9381 1744 if (!conn->session || conn->version_check_done == 0) {
ad7051c0
DG
1745 ERR("Trying to check for data before version check");
1746 ret = -1;
1747 goto end_no_session;
1748 }
1749
58eb9381 1750 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
ad7051c0 1751 if (ret < sizeof(msg)) {
a6cd2b97
DG
1752 if (ret == 0) {
1753 /* Orderly shutdown. Not necessary to print an error. */
58eb9381 1754 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
a6cd2b97
DG
1755 } else {
1756 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1757 ret);
1758 }
ad7051c0
DG
1759 ret = -1;
1760 goto end_no_session;
1761 }
1762
1763 stream_id = be64toh(msg.stream_id);
1764
1765 rcu_read_lock();
d3e2ba59 1766 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2a174661 1767 node.node) {
ad7051c0
DG
1768 if (stream->stream_handle == stream_id) {
1769 stream->data_pending_check_done = 1;
1770 DBG("Relay quiescent control pending flag set to %" PRIu64,
1771 stream_id);
1772 break;
1773 }
1774 }
1775 rcu_read_unlock();
1776
53efb85a 1777 memset(&reply, 0, sizeof(reply));
c8f59ee5 1778 reply.ret_code = htobe32(LTTNG_OK);
58eb9381 1779 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
c8f59ee5 1780 if (ret < 0) {
6d805429 1781 ERR("Relay data quiescent control ret code failed");
c8f59ee5
DG
1782 }
1783
ad7051c0 1784end_no_session:
c8f59ee5
DG
1785 return ret;
1786}
1787
f7079f67
DG
1788/*
1789 * Initialize a data pending command. This means that a client is about to ask
1790 * for data pending for each stream he/she holds. Simply iterate over all
1791 * streams of a session and set the data_pending_check_done flag.
1792 *
1793 * This command returns to the client a LTTNG_OK code.
1794 */
1795static
1796int relay_begin_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
58eb9381 1797 struct relay_connection *conn)
f7079f67
DG
1798{
1799 int ret;
1800 struct lttng_ht_iter iter;
1801 struct lttcomm_relayd_begin_data_pending msg;
1802 struct lttcomm_relayd_generic_reply reply;
1803 struct relay_stream *stream;
1804 uint64_t session_id;
1805
1806 assert(recv_hdr);
58eb9381 1807 assert(conn);
f7079f67
DG
1808
1809 DBG("Init streams for data pending");
1810
58eb9381 1811 if (!conn->session || conn->version_check_done == 0) {
f7079f67
DG
1812 ERR("Trying to check for data before version check");
1813 ret = -1;
1814 goto end_no_session;
1815 }
1816
58eb9381 1817 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
f7079f67 1818 if (ret < sizeof(msg)) {
a6cd2b97
DG
1819 if (ret == 0) {
1820 /* Orderly shutdown. Not necessary to print an error. */
58eb9381 1821 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
a6cd2b97
DG
1822 } else {
1823 ERR("Relay didn't receive valid begin data_pending struct size: %d",
1824 ret);
1825 }
f7079f67
DG
1826 ret = -1;
1827 goto end_no_session;
1828 }
1829
1830 session_id = be64toh(msg.session_id);
1831
1832 /*
1833 * Iterate over all streams to set the begin data pending flag. For now, the
1834 * streams are indexed by stream handle so we have to iterate over all
1835 * streams to find the one associated with the right session_id.
1836 */
1837 rcu_read_lock();
d3e2ba59 1838 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2a174661
DG
1839 node.node) {
1840 if (stream->session_id == session_id) {
f7079f67
DG
1841 stream->data_pending_check_done = 0;
1842 DBG("Set begin data pending flag to stream %" PRIu64,
1843 stream->stream_handle);
1844 }
1845 }
1846 rcu_read_unlock();
1847
53efb85a 1848 memset(&reply, 0, sizeof(reply));
f7079f67
DG
1849 /* All good, send back reply. */
1850 reply.ret_code = htobe32(LTTNG_OK);
1851
58eb9381 1852 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
f7079f67
DG
1853 if (ret < 0) {
1854 ERR("Relay begin data pending send reply failed");
1855 }
1856
1857end_no_session:
1858 return ret;
1859}
1860
1861/*
1862 * End data pending command. This will check, for a given session id, if each
1863 * stream associated with it has its data_pending_check_done flag set. If not,
1864 * this means that the client lost track of the stream but the data is still
1865 * being streamed on our side. In this case, we inform the client that data is
1866 * inflight.
1867 *
1868 * Return to the client if there is data in flight or not with a ret_code.
1869 */
1870static
1871int relay_end_data_pending(struct lttcomm_relayd_hdr *recv_hdr,
58eb9381 1872 struct relay_connection *conn)
f7079f67
DG
1873{
1874 int ret;
1875 struct lttng_ht_iter iter;
1876 struct lttcomm_relayd_end_data_pending msg;
1877 struct lttcomm_relayd_generic_reply reply;
1878 struct relay_stream *stream;
1879 uint64_t session_id;
1880 uint32_t is_data_inflight = 0;
1881
1882 assert(recv_hdr);
58eb9381 1883 assert(conn);
f7079f67
DG
1884
1885 DBG("End data pending command");
1886
58eb9381 1887 if (!conn->session || conn->version_check_done == 0) {
f7079f67
DG
1888 ERR("Trying to check for data before version check");
1889 ret = -1;
1890 goto end_no_session;
1891 }
1892
58eb9381 1893 ret = conn->sock->ops->recvmsg(conn->sock, &msg, sizeof(msg), 0);
f7079f67 1894 if (ret < sizeof(msg)) {
a6cd2b97
DG
1895 if (ret == 0) {
1896 /* Orderly shutdown. Not necessary to print an error. */
58eb9381 1897 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
a6cd2b97
DG
1898 } else {
1899 ERR("Relay didn't receive valid end data_pending struct size: %d",
1900 ret);
1901 }
f7079f67
DG
1902 ret = -1;
1903 goto end_no_session;
1904 }
1905
1906 session_id = be64toh(msg.session_id);
1907
1908 /* Iterate over all streams to see if the begin data pending flag is set. */
1909 rcu_read_lock();
d3e2ba59 1910 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2a174661
DG
1911 node.node) {
1912 if (stream->session_id == session_id &&
f94b19e6 1913 !stream->data_pending_check_done && !stream->terminated_flag) {
f7079f67
DG
1914 is_data_inflight = 1;
1915 DBG("Data is still in flight for stream %" PRIu64,
1916 stream->stream_handle);
1917 break;
1918 }
1919 }
1920 rcu_read_unlock();
1921
53efb85a 1922 memset(&reply, 0, sizeof(reply));
f7079f67
DG
1923 /* All good, send back reply. */
1924 reply.ret_code = htobe32(is_data_inflight);
1925
58eb9381 1926 ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
f7079f67
DG
1927 if (ret < 0) {
1928 ERR("Relay end data pending send reply failed");
1929 }
1930
1931end_no_session:
1932 return ret;
1933}
1934
1c20f0e2
JD
1935/*
1936 * Receive an index for a specific stream.
1937 *
1938 * Return 0 on success else a negative value.
1939 */
1940static
1941int relay_recv_index(struct lttcomm_relayd_hdr *recv_hdr,
58eb9381 1942 struct relay_connection *conn)
1c20f0e2
JD
1943{
1944 int ret, send_ret, index_created = 0;
58eb9381 1945 struct relay_session *session = conn->session;
1c20f0e2
JD
1946 struct lttcomm_relayd_index index_info;
1947 struct relay_index *index, *wr_index = NULL;
1948 struct lttcomm_relayd_generic_reply reply;
1949 struct relay_stream *stream;
1950 uint64_t net_seq_num;
1951
58eb9381 1952 assert(conn);
1c20f0e2
JD
1953
1954 DBG("Relay receiving index");
1955
58eb9381 1956 if (!session || conn->version_check_done == 0) {
1c20f0e2
JD
1957 ERR("Trying to close a stream before version check");
1958 ret = -1;
1959 goto end_no_session;
1960 }
1961
58eb9381 1962 ret = conn->sock->ops->recvmsg(conn->sock, &index_info,
1c20f0e2
JD
1963 sizeof(index_info), 0);
1964 if (ret < sizeof(index_info)) {
1965 if (ret == 0) {
1966 /* Orderly shutdown. Not necessary to print an error. */
58eb9381 1967 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
1c20f0e2
JD
1968 } else {
1969 ERR("Relay didn't receive valid index struct size : %d", ret);
1970 }
1971 ret = -1;
1972 goto end_no_session;
1973 }
1974
1975 net_seq_num = be64toh(index_info.net_seq_num);
1976
1977 rcu_read_lock();
2a174661
DG
1978 stream = stream_find_by_id(relay_streams_ht,
1979 be64toh(index_info.relay_stream_id));
1c20f0e2
JD
1980 if (!stream) {
1981 ret = -1;
1982 goto end_rcu_unlock;
1983 }
1984
d3e2ba59
JD
1985 /* Live beacon handling */
1986 if (index_info.packet_size == 0) {
1987 DBG("Received live beacon for stream %" PRIu64, stream->stream_handle);
1988
1989 /*
6e7241fe
JD
1990 * Only flag a stream inactive when it has already received data
1991 * and no indexes are in flight.
d3e2ba59 1992 */
6e7241fe 1993 if (stream->total_index_received > 0 && stream->indexes_in_flight == 0) {
d3e2ba59
JD
1994 stream->beacon_ts_end = be64toh(index_info.timestamp_end);
1995 }
1996 ret = 0;
1997 goto end_rcu_unlock;
1998 } else {
1999 stream->beacon_ts_end = -1ULL;
2000 }
2001
0a6518b0 2002 index = relay_index_find(stream->stream_handle, net_seq_num);
1c20f0e2
JD
2003 if (!index) {
2004 /* A successful creation will add the object to the HT. */
2005 index = relay_index_create(stream->stream_handle, net_seq_num);
2006 if (!index) {
2007 goto end_rcu_unlock;
2008 }
2009 index_created = 1;
6e7241fe 2010 stream->indexes_in_flight++;
1c20f0e2
JD
2011 }
2012
2013 copy_index_control_data(index, &index_info);
528f2ffa
JD
2014 if (stream->ctf_stream_id == -1ULL) {
2015 stream->ctf_stream_id = be64toh(index_info.stream_id);
2016 }
1c20f0e2
JD
2017
2018 if (index_created) {
2019 /*
2020 * Try to add the relay index object to the hash table. If an object
2021 * already exist, destroy back the index created, set the data in this
2022 * object and write it on disk.
2023 */
0a6518b0 2024 relay_index_add(index, &wr_index);
1c20f0e2
JD
2025 if (wr_index) {
2026 copy_index_control_data(wr_index, &index_info);
2027 free(index);
2028 }
2029 } else {
2030 /* The index already exists so write it on disk. */
2031 wr_index = index;
2032 }
2033
2034 /* Do we have a writable ready index to write on disk. */
2035 if (wr_index) {
0a6518b0 2036 ret = relay_index_write(wr_index->fd, wr_index);
1c20f0e2
JD
2037 if (ret < 0) {
2038 goto end_rcu_unlock;
2039 }
d3e2ba59 2040 stream->total_index_received++;
6e7241fe
JD
2041 stream->indexes_in_flight--;
2042 assert(stream->indexes_in_flight >= 0);
1c20f0e2
JD
2043 }
2044
2045end_rcu_unlock:
2046 rcu_read_unlock();
2047
53efb85a 2048 memset(&reply, 0, sizeof(reply));
1c20f0e2
JD
2049 if (ret < 0) {
2050 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2051 } else {
2052 reply.ret_code = htobe32(LTTNG_OK);
2053 }
58eb9381 2054 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1c20f0e2
JD
2055 if (send_ret < 0) {
2056 ERR("Relay sending close index id reply");
2057 ret = send_ret;
2058 }
2059
2060end_no_session:
2061 return ret;
2062}
2063
a4baae1b
JD
2064/*
2065 * Receive the streams_sent message.
2066 *
2067 * Return 0 on success else a negative value.
2068 */
2069static
2070int relay_streams_sent(struct lttcomm_relayd_hdr *recv_hdr,
58eb9381 2071 struct relay_connection *conn)
a4baae1b
JD
2072{
2073 int ret, send_ret;
2074 struct lttcomm_relayd_generic_reply reply;
2075
58eb9381 2076 assert(conn);
a4baae1b
JD
2077
2078 DBG("Relay receiving streams_sent");
2079
58eb9381 2080 if (!conn->session || conn->version_check_done == 0) {
a4baae1b
JD
2081 ERR("Trying to close a stream before version check");
2082 ret = -1;
2083 goto end_no_session;
2084 }
2085
2086 /*
2087 * Flag every pending stream in the connection recv list that they are
2088 * ready to be used by the viewer.
2089 */
58eb9381 2090 set_viewer_ready_flag(conn);
a4baae1b 2091
4a9daf17
JD
2092 /*
2093 * Inform the viewer that there are new streams in the session.
2094 */
c5141fe5
JD
2095 if (conn->session->viewer_refcount) {
2096 uatomic_set(&conn->session->new_streams, 1);
2097 }
4a9daf17 2098
53efb85a 2099 memset(&reply, 0, sizeof(reply));
a4baae1b 2100 reply.ret_code = htobe32(LTTNG_OK);
58eb9381 2101 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
a4baae1b
JD
2102 if (send_ret < 0) {
2103 ERR("Relay sending sent_stream reply");
2104 ret = send_ret;
2105 } else {
2106 /* Success. */
2107 ret = 0;
2108 }
2109
2110end_no_session:
2111 return ret;
2112}
2113
b8aa1682 2114/*
d3e2ba59 2115 * Process the commands received on the control socket
b8aa1682
JD
2116 */
2117static
2118int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
58eb9381 2119 struct relay_connection *conn)
b8aa1682
JD
2120{
2121 int ret = 0;
2122
2123 switch (be32toh(recv_hdr->cmd)) {
b8aa1682 2124 case RELAYD_CREATE_SESSION:
58eb9381 2125 ret = relay_create_session(recv_hdr, conn);
b8aa1682 2126 break;
b8aa1682 2127 case RELAYD_ADD_STREAM:
58eb9381 2128 ret = relay_add_stream(recv_hdr, conn);
b8aa1682
JD
2129 break;
2130 case RELAYD_START_DATA:
58eb9381 2131 ret = relay_start(recv_hdr, conn);
b8aa1682
JD
2132 break;
2133 case RELAYD_SEND_METADATA:
58eb9381 2134 ret = relay_recv_metadata(recv_hdr, conn);
b8aa1682
JD
2135 break;
2136 case RELAYD_VERSION:
58eb9381 2137 ret = relay_send_version(recv_hdr, conn);
b8aa1682 2138 break;
173af62f 2139 case RELAYD_CLOSE_STREAM:
58eb9381 2140 ret = relay_close_stream(recv_hdr, conn);
173af62f 2141 break;
6d805429 2142 case RELAYD_DATA_PENDING:
58eb9381 2143 ret = relay_data_pending(recv_hdr, conn);
c8f59ee5
DG
2144 break;
2145 case RELAYD_QUIESCENT_CONTROL:
58eb9381 2146 ret = relay_quiescent_control(recv_hdr, conn);
c8f59ee5 2147 break;
f7079f67 2148 case RELAYD_BEGIN_DATA_PENDING:
58eb9381 2149 ret = relay_begin_data_pending(recv_hdr, conn);
f7079f67
DG
2150 break;
2151 case RELAYD_END_DATA_PENDING:
58eb9381 2152 ret = relay_end_data_pending(recv_hdr, conn);
f7079f67 2153 break;
1c20f0e2 2154 case RELAYD_SEND_INDEX:
58eb9381 2155 ret = relay_recv_index(recv_hdr, conn);
1c20f0e2 2156 break;
a4baae1b 2157 case RELAYD_STREAMS_SENT:
58eb9381 2158 ret = relay_streams_sent(recv_hdr, conn);
a4baae1b 2159 break;
b8aa1682
JD
2160 case RELAYD_UPDATE_SYNC_INFO:
2161 default:
2162 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
58eb9381 2163 relay_unknown_command(conn);
b8aa1682
JD
2164 ret = -1;
2165 goto end;
2166 }
2167
2168end:
2169 return ret;
2170}
2171
7d2f7452
DG
2172/*
2173 * Handle index for a data stream.
2174 *
2175 * RCU read side lock MUST be acquired.
2176 *
2177 * Return 0 on success else a negative value.
2178 */
2179static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
2180 int rotate_index)
2181{
2182 int ret = 0, index_created = 0;
2183 uint64_t stream_id, data_offset;
2184 struct relay_index *index, *wr_index = NULL;
2185
2186 assert(stream);
2187
2188 stream_id = stream->stream_handle;
2189 /* Get data offset because we are about to update the index. */
2190 data_offset = htobe64(stream->tracefile_size_current);
2191
2192 /*
2193 * Lookup for an existing index for that stream id/sequence number. If on
2194 * exists, the control thread already received the data for it thus we need
2195 * to write it on disk.
2196 */
2197 index = relay_index_find(stream_id, net_seq_num);
2198 if (!index) {
2199 /* A successful creation will add the object to the HT. */
2200 index = relay_index_create(stream_id, net_seq_num);
2201 if (!index) {
2202 ret = -1;
2203 goto error;
2204 }
2205 index_created = 1;
6e7241fe 2206 stream->indexes_in_flight++;
7d2f7452
DG
2207 }
2208
2209 if (rotate_index || stream->index_fd < 0) {
2210 index->to_close_fd = stream->index_fd;
2211 ret = index_create_file(stream->path_name, stream->channel_name,
2212 relayd_uid, relayd_gid, stream->tracefile_size,
2213 stream->tracefile_count_current);
2214 if (ret < 0) {
2215 /* This will close the stream's index fd if one. */
2216 relay_index_free_safe(index);
2217 goto error;
2218 }
2219 stream->index_fd = ret;
2220 }
2221 index->fd = stream->index_fd;
2222 index->index_data.offset = data_offset;
2223
2224 if (index_created) {
2225 /*
2226 * Try to add the relay index object to the hash table. If an object
2227 * already exist, destroy back the index created and set the data.
2228 */
2229 relay_index_add(index, &wr_index);
2230 if (wr_index) {
2231 /* Copy back data from the created index. */
2232 wr_index->fd = index->fd;
2233 wr_index->to_close_fd = index->to_close_fd;
2234 wr_index->index_data.offset = data_offset;
2235 free(index);
2236 }
2237 } else {
2238 /* The index already exists so write it on disk. */
2239 wr_index = index;
2240 }
2241
2242 /* Do we have a writable ready index to write on disk. */
2243 if (wr_index) {
2244 ret = relay_index_write(wr_index->fd, wr_index);
2245 if (ret < 0) {
2246 goto error;
2247 }
2248 stream->total_index_received++;
6e7241fe
JD
2249 stream->indexes_in_flight--;
2250 assert(stream->indexes_in_flight >= 0);
7d2f7452
DG
2251 }
2252
2253error:
2254 return ret;
2255}
2256
b8aa1682
JD
2257/*
2258 * relay_process_data: Process the data received on the data socket
2259 */
2260static
58eb9381 2261int relay_process_data(struct relay_connection *conn)
b8aa1682 2262{
7d2f7452 2263 int ret = 0, rotate_index = 0;
6cd525e8 2264 ssize_t size_ret;
b8aa1682
JD
2265 struct relay_stream *stream;
2266 struct lttcomm_relayd_data_hdr data_hdr;
7d2f7452 2267 uint64_t stream_id;
173af62f 2268 uint64_t net_seq_num;
b8aa1682 2269 uint32_t data_size;
2a174661 2270 struct relay_session *session;
b8aa1682 2271
58eb9381
DG
2272 assert(conn);
2273
2274 ret = conn->sock->ops->recvmsg(conn->sock, &data_hdr,
7c5aef62 2275 sizeof(struct lttcomm_relayd_data_hdr), 0);
b8aa1682 2276 if (ret <= 0) {
a6cd2b97
DG
2277 if (ret == 0) {
2278 /* Orderly shutdown. Not necessary to print an error. */
58eb9381 2279 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
a6cd2b97 2280 } else {
58eb9381 2281 ERR("Unable to receive data header on sock %d", conn->sock->fd);
a6cd2b97 2282 }
b8aa1682
JD
2283 ret = -1;
2284 goto end;
2285 }
2286
2287 stream_id = be64toh(data_hdr.stream_id);
9d1bbf21
MD
2288
2289 rcu_read_lock();
2a174661 2290 stream = stream_find_by_id(relay_streams_ht, stream_id);
b8aa1682
JD
2291 if (!stream) {
2292 ret = -1;
1c20f0e2 2293 goto end_rcu_unlock;
b8aa1682
JD
2294 }
2295
58eb9381 2296 session = session_find_by_id(conn->sessions_ht, stream->session_id);
2a174661
DG
2297 assert(session);
2298
b8aa1682
JD
2299 data_size = be32toh(data_hdr.data_size);
2300 if (data_buffer_size < data_size) {
c617c0c6
MD
2301 char *tmp_data_ptr;
2302
2303 tmp_data_ptr = realloc(data_buffer, data_size);
2304 if (!tmp_data_ptr) {
b8aa1682 2305 ERR("Allocating data buffer");
c617c0c6 2306 free(data_buffer);
b8aa1682 2307 ret = -1;
1c20f0e2 2308 goto end_rcu_unlock;
b8aa1682 2309 }
c617c0c6 2310 data_buffer = tmp_data_ptr;
b8aa1682
JD
2311 data_buffer_size = data_size;
2312 }
2313 memset(data_buffer, 0, data_size);
2314
173af62f
DG
2315 net_seq_num = be64toh(data_hdr.net_seq_num);
2316
77c7c900 2317 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
173af62f 2318 data_size, stream_id, net_seq_num);
58eb9381 2319 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer, data_size, 0);
b8aa1682 2320 if (ret <= 0) {
a6cd2b97
DG
2321 if (ret == 0) {
2322 /* Orderly shutdown. Not necessary to print an error. */
58eb9381 2323 DBG("Socket %d did an orderly shutdown", conn->sock->fd);
a6cd2b97 2324 }
b8aa1682 2325 ret = -1;
1c20f0e2 2326 goto end_rcu_unlock;
b8aa1682
JD
2327 }
2328
1c20f0e2 2329 /* Check if a rotation is needed. */
0f907de1
JD
2330 if (stream->tracefile_size > 0 &&
2331 (stream->tracefile_size_current + data_size) >
2332 stream->tracefile_size) {
6b6b9a5a
JD
2333 struct relay_viewer_stream *vstream;
2334 uint64_t new_id;
2335
2336 new_id = (stream->tracefile_count_current + 1) %
2337 stream->tracefile_count;
2338 /*
2339 * When we wrap-around back to 0, we start overwriting old
2340 * trace data.
2341 */
2342 if (!stream->tracefile_overwrite && new_id == 0) {
2343 stream->tracefile_overwrite = 1;
2344 }
2345 pthread_mutex_lock(&stream->viewer_stream_rotation_lock);
2346 if (stream->tracefile_overwrite) {
2347 stream->oldest_tracefile_id =
2348 (stream->oldest_tracefile_id + 1) %
2349 stream->tracefile_count;
2350 }
2f8f53af 2351 vstream = viewer_stream_find_by_id(stream->stream_handle);
6b6b9a5a
JD
2352 if (vstream) {
2353 /*
2354 * The viewer is reading a file about to be
2355 * overwritten. Close the FDs it is
2356 * currently using and let it handle the fault.
2357 */
2358 if (vstream->tracefile_count_current == new_id) {
cef0f7d5 2359 pthread_mutex_lock(&vstream->overwrite_lock);
6b6b9a5a 2360 vstream->abort_flag = 1;
cef0f7d5 2361 pthread_mutex_unlock(&vstream->overwrite_lock);
f48c25b7 2362 DBG("Streaming side setting abort_flag on stream %s_%" PRIu64 "\n",
6b6b9a5a
JD
2363 stream->channel_name, new_id);
2364 } else if (vstream->tracefile_count_current ==
2365 stream->tracefile_count_current) {
2366 /*
2367 * The reader and writer were in the
2368 * same trace file, inform the viewer
2369 * that no new index will ever be added
2370 * to this file.
2371 */
2372 vstream->close_write_flag = 1;
2373 }
2374 }
1c20f0e2
JD
2375 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
2376 stream->tracefile_size, stream->tracefile_count,
2377 relayd_uid, relayd_gid, stream->fd,
2378 &(stream->tracefile_count_current), &stream->fd);
6b6b9a5a 2379 pthread_mutex_unlock(&stream->viewer_stream_rotation_lock);
0f907de1 2380 if (ret < 0) {
1c20f0e2
JD
2381 ERR("Rotating stream output file");
2382 goto end_rcu_unlock;
0f907de1 2383 }
a6976990
DG
2384 /* Reset current size because we just perform a stream rotation. */
2385 stream->tracefile_size_current = 0;
1c20f0e2
JD
2386 rotate_index = 1;
2387 }
2388
1c20f0e2 2389 /*
7d2f7452
DG
2390 * Index are handled in protocol version 2.4 and above. Also, snapshot and
2391 * index are NOT supported.
1c20f0e2 2392 */
2a174661 2393 if (session->minor >= 4 && !session->snapshot) {
7d2f7452 2394 ret = handle_index_data(stream, net_seq_num, rotate_index);
1c20f0e2 2395 if (ret < 0) {
1c20f0e2
JD
2396 goto end_rcu_unlock;
2397 }
1c20f0e2
JD
2398 }
2399
7d2f7452 2400 /* Write data to stream output fd. */
6cd525e8
MD
2401 size_ret = lttng_write(stream->fd, data_buffer, data_size);
2402 if (size_ret < data_size) {
b8aa1682
JD
2403 ERR("Relay error writing data to file");
2404 ret = -1;
1c20f0e2 2405 goto end_rcu_unlock;
b8aa1682 2406 }
1d4dfdef 2407
5ab7344e
JD
2408 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
2409 ret, stream->stream_handle);
2410
1d4dfdef
DG
2411 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
2412 if (ret < 0) {
1c20f0e2 2413 goto end_rcu_unlock;
1d4dfdef 2414 }
1c20f0e2 2415 stream->tracefile_size_current += data_size + be32toh(data_hdr.padding_size);
1d4dfdef 2416
173af62f
DG
2417 stream->prev_seq = net_seq_num;
2418
2a174661 2419 try_close_stream(session, stream);
173af62f 2420
1c20f0e2 2421end_rcu_unlock:
9d1bbf21 2422 rcu_read_unlock();
b8aa1682
JD
2423end:
2424 return ret;
2425}
2426
2427static
58eb9381 2428void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
b8aa1682
JD
2429{
2430 int ret;
2431
58eb9381
DG
2432 assert(events);
2433
2434 (void) lttng_poll_del(events, pollfd);
b8aa1682
JD
2435
2436 ret = close(pollfd);
2437 if (ret < 0) {
2438 ERR("Closing pollfd %d", pollfd);
2439 }
2440}
2441
58eb9381
DG
2442static void destroy_connection(struct lttng_ht *relay_connections_ht,
2443 struct relay_connection *conn)
9d1bbf21 2444{
58eb9381
DG
2445 assert(relay_connections_ht);
2446 assert(conn);
2a174661 2447
58eb9381 2448 connection_delete(relay_connections_ht, conn);
2a174661 2449
58eb9381 2450 /* For the control socket, we try to destroy the session. */
fd6d7293 2451 if (conn->type == RELAY_CONTROL && conn->session) {
58eb9381 2452 destroy_session(conn->session, conn->sessions_ht);
9d1bbf21 2453 }
5b6d8097 2454
58eb9381 2455 connection_destroy(conn);
b8aa1682
JD
2456}
2457
2458/*
2459 * This thread does the actual work
2460 */
2461static
2462void *relay_thread_worker(void *data)
2463{
beaad64c
DG
2464 int ret, err = -1, last_seen_data_fd = -1;
2465 uint32_t nb_fd;
58eb9381 2466 struct relay_connection *conn;
b8aa1682
JD
2467 struct lttng_poll_event events;
2468 struct lttng_ht *relay_connections_ht;
b8aa1682 2469 struct lttng_ht_iter iter;
b8aa1682 2470 struct lttcomm_relayd_hdr recv_hdr;
d3e2ba59
JD
2471 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
2472 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
f233ab30 2473 struct relay_index *index;
b8aa1682
JD
2474
2475 DBG("[thread] Relay worker started");
2476
9d1bbf21
MD
2477 rcu_register_thread();
2478
55706a7d
MD
2479 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
2480
9b5e0863
MD
2481 if (testpoint(relayd_thread_worker)) {
2482 goto error_testpoint;
2483 }
2484
f385ae0a
MD
2485 health_code_update();
2486
b8aa1682
JD
2487 /* table of connections indexed on socket */
2488 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
095a4ae5
MD
2489 if (!relay_connections_ht) {
2490 goto relay_connections_ht_error;
2491 }
b8aa1682 2492
1c20f0e2
JD
2493 /* Tables of received indexes indexed by index handle and net_seq_num. */
2494 indexes_ht = lttng_ht_new(0, LTTNG_HT_TYPE_TWO_U64);
2495 if (!indexes_ht) {
2496 goto indexes_ht_error;
2497 }
2498
b8aa1682
JD
2499 ret = create_thread_poll_set(&events, 2);
2500 if (ret < 0) {
2501 goto error_poll_create;
2502 }
2503
58eb9381 2504 ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
b8aa1682
JD
2505 if (ret < 0) {
2506 goto error;
2507 }
2508
beaad64c 2509restart:
b8aa1682 2510 while (1) {
beaad64c
DG
2511 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
2512
f385ae0a
MD
2513 health_code_update();
2514
b8aa1682 2515 /* Infinite blocking call, waiting for transmission */
87c1611d 2516 DBG3("Relayd worker thread polling...");
f385ae0a 2517 health_poll_entry();
b8aa1682 2518 ret = lttng_poll_wait(&events, -1);
f385ae0a 2519 health_poll_exit();
b8aa1682
JD
2520 if (ret < 0) {
2521 /*
2522 * Restart interrupted system call.
2523 */
2524 if (errno == EINTR) {
2525 goto restart;
2526 }
2527 goto error;
2528 }
2529
0d9c5d77
DG
2530 nb_fd = ret;
2531
beaad64c
DG
2532 /*
2533 * Process control. The control connection is prioritised so we don't
2534 * starve it with high throughout put tracing data on the data
2535 * connection.
2536 */
b8aa1682
JD
2537 for (i = 0; i < nb_fd; i++) {
2538 /* Fetch once the poll data */
beaad64c
DG
2539 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2540 int pollfd = LTTNG_POLL_GETFD(&events, i);
b8aa1682 2541
f385ae0a
MD
2542 health_code_update();
2543
fd20dac9
MD
2544 if (!revents) {
2545 /* No activity for this FD (poll implementation). */
2546 continue;
2547 }
2548
b8aa1682
JD
2549 /* Thread quit pipe has been closed. Killing thread. */
2550 ret = check_thread_quit_pipe(pollfd, revents);
2551 if (ret) {
095a4ae5
MD
2552 err = 0;
2553 goto exit;
b8aa1682
JD
2554 }
2555
58eb9381
DG
2556 /* Inspect the relay conn pipe for new connection */
2557 if (pollfd == relay_conn_pipe[0]) {
b8aa1682 2558 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
58eb9381 2559 ERR("Relay connection pipe error");
b8aa1682
JD
2560 goto error;
2561 } else if (revents & LPOLLIN) {
58eb9381 2562 ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn));
b8aa1682
JD
2563 if (ret < 0) {
2564 goto error;
2565 }
58eb9381
DG
2566 conn->sessions_ht = sessions_ht;
2567 connection_init(conn);
2568 lttng_poll_add(&events, conn->sock->fd,
2569 LPOLLIN | LPOLLRDHUP);
2570 rcu_read_lock();
2571 lttng_ht_add_unique_ulong(relay_connections_ht,
2572 &conn->sock_n);
9d1bbf21 2573 rcu_read_unlock();
58eb9381 2574 DBG("Connection socket %d added", conn->sock->fd);
b8aa1682 2575 }
58eb9381
DG
2576 } else {
2577 rcu_read_lock();
2578 conn = connection_find_by_sock(relay_connections_ht, pollfd);
2579 /* If not found, there is a synchronization issue. */
2580 assert(conn);
2581
2582 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2583 cleanup_connection_pollfd(&events, pollfd);
2584 destroy_connection(relay_connections_ht, conn);
beaad64c
DG
2585 if (last_seen_data_fd == pollfd) {
2586 last_seen_data_fd = last_notdel_data_fd;
2587 }
b8aa1682 2588 } else if (revents & LPOLLIN) {
58eb9381
DG
2589 if (conn->type == RELAY_CONTROL) {
2590 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
2591 sizeof(recv_hdr), 0);
b8aa1682 2592 if (ret <= 0) {
58eb9381
DG
2593 /* Connection closed */
2594 cleanup_connection_pollfd(&events, pollfd);
2595 destroy_connection(relay_connections_ht, conn);
b8aa1682
JD
2596 DBG("Control connection closed with %d", pollfd);
2597 } else {
58eb9381 2598 ret = relay_process_control(&recv_hdr, conn);
b8aa1682 2599 if (ret < 0) {
beaad64c 2600 /* Clear the session on error. */
58eb9381
DG
2601 cleanup_connection_pollfd(&events, pollfd);
2602 destroy_connection(relay_connections_ht, conn);
b8aa1682
JD
2603 DBG("Connection closed with %d", pollfd);
2604 }
beaad64c 2605 seen_control = 1;
b8aa1682 2606 }
beaad64c
DG
2607 } else {
2608 /*
2609 * Flag the last seen data fd not deleted. It will be
2610 * used as the last seen fd if any fd gets deleted in
2611 * this first loop.
2612 */
2613 last_notdel_data_fd = pollfd;
2614 }
58eb9381
DG
2615 } else {
2616 ERR("Unknown poll events %u for sock %d", revents, pollfd);
beaad64c
DG
2617 }
2618 rcu_read_unlock();
2619 }
2620 }
2621
2622 /*
2623 * The last loop handled a control request, go back to poll to make
2624 * sure we prioritise the control socket.
2625 */
2626 if (seen_control) {
2627 continue;
2628 }
2629
2630 if (last_seen_data_fd >= 0) {
2631 for (i = 0; i < nb_fd; i++) {
2632 int pollfd = LTTNG_POLL_GETFD(&events, i);
f385ae0a
MD
2633
2634 health_code_update();
2635
beaad64c
DG
2636 if (last_seen_data_fd == pollfd) {
2637 idx = i;
2638 break;
2639 }
2640 }
2641 }
2642
2643 /* Process data connection. */
2644 for (i = idx + 1; i < nb_fd; i++) {
2645 /* Fetch the poll data. */
2646 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2647 int pollfd = LTTNG_POLL_GETFD(&events, i);
2648
f385ae0a
MD
2649 health_code_update();
2650
fd20dac9
MD
2651 if (!revents) {
2652 /* No activity for this FD (poll implementation). */
2653 continue;
2654 }
2655
beaad64c 2656 /* Skip the command pipe. It's handled in the first loop. */
58eb9381 2657 if (pollfd == relay_conn_pipe[0]) {
beaad64c
DG
2658 continue;
2659 }
2660
fd20dac9
MD
2661 rcu_read_lock();
2662 conn = connection_find_by_sock(relay_connections_ht, pollfd);
2663 if (!conn) {
2664 /* Skip it. Might be removed before. */
2665 rcu_read_unlock();
2666 continue;
2667 }
2668
2669 if (revents & LPOLLIN) {
2670 if (conn->type != RELAY_DATA) {
beaad64c
DG
2671 rcu_read_unlock();
2672 continue;
2673 }
beaad64c 2674
fd20dac9
MD
2675 ret = relay_process_data(conn);
2676 /* Connection closed */
2677 if (ret < 0) {
2678 cleanup_connection_pollfd(&events, pollfd);
2679 destroy_connection(relay_connections_ht, conn);
2680 DBG("Data connection closed with %d", pollfd);
2681 /*
2682 * Every goto restart call sets the last seen fd where
2683 * here we don't really care since we gracefully
2684 * continue the loop after the connection is deleted.
2685 */
2686 } else {
2687 /* Keep last seen port. */
2688 last_seen_data_fd = pollfd;
2689 rcu_read_unlock();
2690 goto restart;
b8aa1682
JD
2691 }
2692 }
fd20dac9 2693 rcu_read_unlock();
b8aa1682 2694 }
beaad64c 2695 last_seen_data_fd = -1;
b8aa1682
JD
2696 }
2697
f385ae0a
MD
2698 /* Normal exit, no error */
2699 ret = 0;
2700
095a4ae5 2701exit:
b8aa1682
JD
2702error:
2703 lttng_poll_clean(&events);
2704
58eb9381 2705 /* Cleanup reamaining connection object. */
9d1bbf21 2706 rcu_read_lock();
58eb9381
DG
2707 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn,
2708 sock_n.node) {
f385ae0a 2709 health_code_update();
58eb9381 2710 destroy_connection(relay_connections_ht, conn);
b8aa1682 2711 }
94d49140 2712 rcu_read_unlock();
7d2f7452 2713error_poll_create:
f233ab30
JD
2714 rcu_read_lock();
2715 cds_lfht_for_each_entry(indexes_ht->ht, &iter.iter, index,
2716 index_n.node) {
2717 health_code_update();
2718 relay_index_delete(index);
2719 relay_index_free_safe(index);
2720 }
2721 rcu_read_unlock();
7d2f7452 2722 lttng_ht_destroy(indexes_ht);
1c20f0e2 2723indexes_ht_error:
b8aa1682 2724 lttng_ht_destroy(relay_connections_ht);
095a4ae5 2725relay_connections_ht_error:
58eb9381
DG
2726 /* Close relay conn pipes */
2727 utils_close_pipe(relay_conn_pipe);
095a4ae5
MD
2728 if (err) {
2729 DBG("Thread exited with error");
2730 }
b8aa1682 2731 DBG("Worker thread cleanup complete");
095a4ae5 2732 free(data_buffer);
9b5e0863 2733error_testpoint:
f385ae0a
MD
2734 if (err) {
2735 health_error();
2736 ERR("Health error occurred in %s", __func__);
2737 }
2738 health_unregister(health_relayd);
9d1bbf21 2739 rcu_unregister_thread();
b4aacfdc 2740 lttng_relay_stop_threads();
b8aa1682
JD
2741 return NULL;
2742}
2743
2744/*
2745 * Create the relay command pipe to wake thread_manage_apps.
2746 * Closed in cleanup().
2747 */
58eb9381 2748static int create_relay_conn_pipe(void)
b8aa1682 2749{
a02de639 2750 int ret;
b8aa1682 2751
58eb9381 2752 ret = utils_create_pipe_cloexec(relay_conn_pipe);
b8aa1682 2753
b8aa1682
JD
2754 return ret;
2755}
2756
2757/*
2758 * main
2759 */
2760int main(int argc, char **argv)
2761{
178a0557 2762 int ret = 0, retval = 0;
b8aa1682 2763 void *status;
178a0557 2764 struct relay_local_data *relay_ctx = NULL;
b8aa1682 2765
b8aa1682
JD
2766 /* Parse arguments */
2767 progname = argv[0];
178a0557
MD
2768 if (set_options(argc, argv)) {
2769 retval = -1;
2770 goto exit_options;
b8aa1682
JD
2771 }
2772
178a0557
MD
2773 if (set_signal_handler()) {
2774 retval = -1;
2775 goto exit_options;
b8aa1682
JD
2776 }
2777
4d513a50
DG
2778 /* Try to create directory if -o, --output is specified. */
2779 if (opt_output_path) {
994fa64f
DG
2780 if (*opt_output_path != '/') {
2781 ERR("Please specify an absolute path for -o, --output PATH");
178a0557
MD
2782 retval = -1;
2783 goto exit_options;
994fa64f
DG
2784 }
2785
4d513a50
DG
2786 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG);
2787 if (ret < 0) {
2788 ERR("Unable to create %s", opt_output_path);
178a0557
MD
2789 retval = -1;
2790 goto exit_options;
4d513a50
DG
2791 }
2792 }
2793
b8aa1682 2794 /* Daemonize */
b5218ffb 2795 if (opt_daemon || opt_background) {
3fd27398
MD
2796 int i;
2797
2798 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
2799 !opt_background);
b8aa1682 2800 if (ret < 0) {
178a0557
MD
2801 retval = -1;
2802 goto exit_options;
b8aa1682 2803 }
3fd27398
MD
2804
2805 /*
2806 * We are in the child. Make sure all other file
2807 * descriptors are closed, in case we are called with
2808 * more opened file descriptors than the standard ones.
2809 */
2810 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
2811 (void) close(i);
2812 }
2813 }
2814
178a0557
MD
2815
2816 /* Initialize thread health monitoring */
2817 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
2818 if (!health_relayd) {
2819 PERROR("health_app_create error");
2820 retval = -1;
2821 goto exit_health_app_create;
2822 }
2823
3fd27398 2824 /* Create thread quit pipe */
178a0557
MD
2825 if (init_thread_quit_pipe()) {
2826 retval = -1;
2827 goto exit_init_data;
b8aa1682
JD
2828 }
2829
1c20f0e2
JD
2830 /* We need those values for the file/dir creation. */
2831 relayd_uid = getuid();
2832 relayd_gid = getgid();
b8aa1682 2833
1c20f0e2
JD
2834 /* Check if daemon is UID = 0 */
2835 if (relayd_uid == 0) {
8d5c808e 2836 if (control_uri->port < 1024 || data_uri->port < 1024 || live_uri->port < 1024) {
b8aa1682 2837 ERR("Need to be root to use ports < 1024");
178a0557
MD
2838 retval = -1;
2839 goto exit_init_data;
b8aa1682
JD
2840 }
2841 }
2842
2843 /* Setup the thread apps communication pipe. */
178a0557
MD
2844 if (create_relay_conn_pipe()) {
2845 retval = -1;
2846 goto exit_init_data;
b8aa1682
JD
2847 }
2848
2849 /* Init relay command queue. */
8bdee6e2 2850 cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
b8aa1682
JD
2851
2852 /* Set up max poll set size */
25b397f9
MD
2853 if (lttng_poll_set_max_size()) {
2854 retval = -1;
2855 goto exit_init_data;
2856 }
b8aa1682 2857
554831e7
MD
2858 /* Initialize communication library */
2859 lttcomm_init();
87e45c13 2860 lttcomm_inet_init();
554831e7 2861
d3e2ba59
JD
2862 relay_ctx = zmalloc(sizeof(struct relay_local_data));
2863 if (!relay_ctx) {
2864 PERROR("relay_ctx");
178a0557
MD
2865 retval = -1;
2866 goto exit_init_data;
d3e2ba59
JD
2867 }
2868
2869 /* tables of sessions indexed by session ID */
2a174661 2870 relay_ctx->sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
d3e2ba59 2871 if (!relay_ctx->sessions_ht) {
178a0557
MD
2872 retval = -1;
2873 goto exit_init_data;
d3e2ba59
JD
2874 }
2875
2876 /* tables of streams indexed by stream ID */
2a174661 2877 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
d3e2ba59 2878 if (!relay_streams_ht) {
178a0557
MD
2879 retval = -1;
2880 goto exit_init_data;
d3e2ba59
JD
2881 }
2882
2883 /* tables of streams indexed by stream ID */
92c6ca54
DG
2884 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2885 if (!viewer_streams_ht) {
178a0557
MD
2886 retval = -1;
2887 goto exit_init_data;
55706a7d
MD
2888 }
2889
65931c8b 2890 ret = utils_create_pipe(health_quit_pipe);
178a0557
MD
2891 if (ret) {
2892 retval = -1;
2893 goto exit_health_quit_pipe;
65931c8b
MD
2894 }
2895
2896 /* Create thread to manage the client socket */
2897 ret = pthread_create(&health_thread, NULL,
2898 thread_manage_health, (void *) NULL);
178a0557
MD
2899 if (ret) {
2900 errno = ret;
65931c8b 2901 PERROR("pthread_create health");
178a0557
MD
2902 retval = -1;
2903 goto exit_health_thread;
65931c8b
MD
2904 }
2905
b8aa1682
JD
2906 /* Setup the dispatcher thread */
2907 ret = pthread_create(&dispatcher_thread, NULL,
2908 relay_thread_dispatcher, (void *) NULL);
178a0557
MD
2909 if (ret) {
2910 errno = ret;
b8aa1682 2911 PERROR("pthread_create dispatcher");
178a0557
MD
2912 retval = -1;
2913 goto exit_dispatcher_thread;
b8aa1682
JD
2914 }
2915
2916 /* Setup the worker thread */
2917 ret = pthread_create(&worker_thread, NULL,
d3e2ba59 2918 relay_thread_worker, (void *) relay_ctx);
178a0557
MD
2919 if (ret) {
2920 errno = ret;
b8aa1682 2921 PERROR("pthread_create worker");
178a0557
MD
2922 retval = -1;
2923 goto exit_worker_thread;
b8aa1682
JD
2924 }
2925
2926 /* Setup the listener thread */
2927 ret = pthread_create(&listener_thread, NULL,
2928 relay_thread_listener, (void *) NULL);
178a0557
MD
2929 if (ret) {
2930 errno = ret;
b8aa1682 2931 PERROR("pthread_create listener");
178a0557
MD
2932 retval = -1;
2933 goto exit_listener_thread;
b8aa1682
JD
2934 }
2935
178a0557
MD
2936 ret = relayd_live_create(live_uri, relay_ctx);
2937 if (ret) {
d3e2ba59 2938 ERR("Starting live viewer threads");
178a0557 2939 retval = -1;
50138f51 2940 goto exit_live;
d3e2ba59
JD
2941 }
2942
178a0557
MD
2943 /*
2944 * This is where we start awaiting program completion (e.g. through
2945 * signal that asks threads to teardown).
2946 */
2947
2948 ret = relayd_live_join();
2949 if (ret) {
2950 retval = -1;
2951 }
50138f51 2952exit_live:
178a0557 2953
b8aa1682 2954 ret = pthread_join(listener_thread, &status);
178a0557
MD
2955 if (ret) {
2956 errno = ret;
2957 PERROR("pthread_join listener_thread");
2958 retval = -1;
b8aa1682
JD
2959 }
2960
178a0557 2961exit_listener_thread:
b8aa1682 2962 ret = pthread_join(worker_thread, &status);
178a0557
MD
2963 if (ret) {
2964 errno = ret;
2965 PERROR("pthread_join worker_thread");
2966 retval = -1;
b8aa1682
JD
2967 }
2968
178a0557 2969exit_worker_thread:
b8aa1682 2970 ret = pthread_join(dispatcher_thread, &status);
178a0557
MD
2971 if (ret) {
2972 errno = ret;
2973 PERROR("pthread_join dispatcher_thread");
2974 retval = -1;
b8aa1682 2975 }
178a0557 2976exit_dispatcher_thread:
42415026 2977
65931c8b 2978 ret = pthread_join(health_thread, &status);
178a0557
MD
2979 if (ret) {
2980 errno = ret;
2981 PERROR("pthread_join health_thread");
2982 retval = -1;
65931c8b 2983 }
178a0557 2984exit_health_thread:
65931c8b 2985
65931c8b 2986 utils_close_pipe(health_quit_pipe);
178a0557 2987exit_health_quit_pipe:
65931c8b 2988
178a0557 2989exit_init_data:
55706a7d 2990 health_app_destroy(health_relayd);
55706a7d 2991exit_health_app_create:
178a0557
MD
2992exit_options:
2993 relayd_cleanup(relay_ctx);
d3e2ba59 2994
178a0557 2995 if (!retval) {
b8aa1682 2996 exit(EXIT_SUCCESS);
178a0557
MD
2997 } else {
2998 exit(EXIT_FAILURE);
b8aa1682 2999 }
b8aa1682 3000}
This page took 0.359608 seconds and 4 git commands to generate.