projects / lttng-tools.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
Dynamic payload for relayd create session command
[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>
7591bab1 5 * 2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
b8aa1682
JD		 6 *
7 * This program is free software; you can redistribute it and/or modify
8 * it under the terms of the GNU General Public License, version 2 only,
9 * as published by the Free Software Foundation.
10 *
11 * This program is distributed in the hope that it will be useful, but WITHOUT
12 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
14 * more details.
15 *
16 * You should have received a copy of the GNU General Public License along
17 * with this program; if not, write to the Free Software Foundation, Inc.,
18 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
19 */
20
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>
b8aa1682
JD		 42
43#include <lttng/lttng.h>
44#include <common/common.h>
45#include <common/compat/poll.h>
46#include <common/compat/socket.h>
f263b7fd 47#include <common/compat/endian.h>
e8fa9fb0 48#include <common/compat/getenv.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>
d3ecc550 57#include <common/align.h>
f40ef1d5 58#include <common/config/session-config.h>
5312a3ed
JG		 59#include <common/dynamic-buffer.h>
60#include <common/buffer-view.h>
7591bab1 61#include <urcu/rculist.h>
b8aa1682 62
0f907de1 63#include "cmd.h"
d3e2ba59 64#include "ctf-trace.h"
1c20f0e2 65#include "index.h"
0f907de1 66#include "utils.h"
b8aa1682 67#include "lttng-relayd.h"
d3e2ba59 68#include "live.h"
55706a7d 69#include "health-relayd.h"
9b5e0863 70#include "testpoint.h"
2f8f53af 71#include "viewer-stream.h"
2a174661
DG		 72#include "session.h"
73#include "stream.h"
58eb9381 74#include "connection.h"
a44ca2ca 75#include "tracefile-array.h"
f056029c 76#include "tcp_keep_alive.h"
b8aa1682 77
4fc83d94
PP		 78static const char *help_msg =
79#ifdef LTTNG_EMBED_HELP
80#include <lttng-relayd.8.h>
81#else
82NULL
83#endif
84;
85
5569b118
JG		 86enum relay_connection_status {
87 RELAY_CONNECTION_STATUS_OK,
88 /* An error occured while processing an event on the connection. */
89 RELAY_CONNECTION_STATUS_ERROR,
90 /* Connection closed/shutdown cleanly. */
91 RELAY_CONNECTION_STATUS_CLOSED,
92};
93
b8aa1682 94/* command line options */
0f907de1 95char *opt_output_path;
b5218ffb 96static int opt_daemon, opt_background;
3fd27398
MD		 97
98/*
99 * We need to wait for listener and live listener threads, as well as
100 * health check thread, before being ready to signal readiness.
101 */
102#define NR_LTTNG_RELAY_READY 3
103static int lttng_relay_ready = NR_LTTNG_RELAY_READY;
0848dba7
MD		 104
105/* Size of receive buffer. */
106#define RECV_DATA_BUFFER_SIZE 65536
d3ecc550 107#define FILE_COPY_BUFFER_SIZE 65536
0848dba7 108
3fd27398
MD		 109static int recv_child_signal; /* Set to 1 when a SIGUSR1 signal is received. */
110static pid_t child_ppid; /* Internal parent PID use with daemonize. */
111
095a4ae5
MD		 112static struct lttng_uri *control_uri;
113static struct lttng_uri *data_uri;
d3e2ba59 114static struct lttng_uri *live_uri;
b8aa1682
JD		 115
116const char *progname;
b8aa1682 117
65931c8b 118const char *tracing_group_name = DEFAULT_TRACING_GROUP;
cd60b05a
JG		 119static int tracing_group_name_override;
120
121const char * const config_section_name = "relayd";
65931c8b 122
b8aa1682
JD		 123/*
124 * Quit pipe for all threads. This permits a single cancellation point
125 * for all threads when receiving an event on the pipe.
126 */
0b242f62 127int thread_quit_pipe[2] = { -1, -1 };
b8aa1682
JD		 128
129/*
130 * This pipe is used to inform the worker thread that a command is queued and
131 * ready to be processed.
132 */
58eb9381 133static int relay_conn_pipe[2] = { -1, -1 };
b8aa1682 134
26c9d55e 135/* Shared between threads */
b8aa1682
JD		 136static int dispatch_thread_exit;
137
138static pthread_t listener_thread;
139static pthread_t dispatcher_thread;
140static pthread_t worker_thread;
65931c8b 141static pthread_t health_thread;
b8aa1682 142
7591bab1
MD		 143/*
144 * last_relay_stream_id_lock protects last_relay_stream_id increment
145 * atomicity on 32-bit architectures.
146 */
147static pthread_mutex_t last_relay_stream_id_lock = PTHREAD_MUTEX_INITIALIZER;
095a4ae5 148static uint64_t last_relay_stream_id;
b8aa1682
JD		 149
150/*
151 * Relay command queue.
152 *
153 * The relay_thread_listener and relay_thread_dispatcher communicate with this
154 * queue.
155 */
58eb9381 156static struct relay_conn_queue relay_conn_queue;
b8aa1682 157
d3e2ba59
JD		 158/* Global relay stream hash table. */
159struct lttng_ht *relay_streams_ht;
160
92c6ca54
DG		 161/* Global relay viewer stream hash table. */
162struct lttng_ht *viewer_streams_ht;
163
7591bab1
MD		 164/* Global relay sessions hash table. */
165struct lttng_ht *sessions_ht;
0a6518b0 166
55706a7d 167/* Relayd health monitoring */
eea7556c 168struct health_app *health_relayd;
55706a7d 169
cd60b05a
JG		 170static struct option long_options[] = {
171 { "control-port", 1, 0, 'C', },
172 { "data-port", 1, 0, 'D', },
8d5c808e 173 { "live-port", 1, 0, 'L', },
cd60b05a 174 { "daemonize", 0, 0, 'd', },
b5218ffb 175 { "background", 0, 0, 'b', },
cd60b05a
JG		 176 { "group", 1, 0, 'g', },
177 { "help", 0, 0, 'h', },
178 { "output", 1, 0, 'o', },
179 { "verbose", 0, 0, 'v', },
180 { "config", 1, 0, 'f' },
3a904098 181 { "version", 0, 0, 'V' },
cd60b05a
JG		 182 { NULL, 0, 0, 0, },
183};
184
3a904098 185static const char *config_ignore_options[] = { "help", "config", "version" };
cd60b05a 186
cd60b05a
JG		 187/*
188 * Take an option from the getopt output and set it in the right variable to be
189 * used later.
190 *
191 * Return 0 on success else a negative value.
192 */
7591bab1 193static int set_option(int opt, const char *arg, const char *optname)
b8aa1682 194{
cd60b05a
JG		 195 int ret;
196
197 switch (opt) {
198 case 0:
199 fprintf(stderr, "option %s", optname);
200 if (arg) {
201 fprintf(stderr, " with arg %s\n", arg);
202 }
203 break;
204 case 'C':
e8fa9fb0
MD		 205 if (lttng_is_setuid_setgid()) {
206 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
207 "-C, --control-port");
208 } else {
209 ret = uri_parse(arg, &control_uri);
210 if (ret < 0) {
211 ERR("Invalid control URI specified");
212 goto end;
213 }
214 if (control_uri->port == 0) {
215 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
216 }
cd60b05a
JG		 217 }
218 break;
219 case 'D':
e8fa9fb0
MD		 220 if (lttng_is_setuid_setgid()) {
221 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
222 "-D, -data-port");
223 } else {
224 ret = uri_parse(arg, &data_uri);
225 if (ret < 0) {
226 ERR("Invalid data URI specified");
227 goto end;
228 }
229 if (data_uri->port == 0) {
230 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
231 }
cd60b05a
JG		 232 }
233 break;
8d5c808e 234 case 'L':
e8fa9fb0
MD		 235 if (lttng_is_setuid_setgid()) {
236 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
237 "-L, -live-port");
238 } else {
239 ret = uri_parse(arg, &live_uri);
240 if (ret < 0) {
241 ERR("Invalid live URI specified");
242 goto end;
243 }
244 if (live_uri->port == 0) {
245 live_uri->port = DEFAULT_NETWORK_VIEWER_PORT;
246 }
8d5c808e
AM		 247 }
248 break;
cd60b05a
JG		 249 case 'd':
250 opt_daemon = 1;
251 break;
b5218ffb
MD		 252 case 'b':
253 opt_background = 1;
254 break;
cd60b05a 255 case 'g':
e8fa9fb0
MD		 256 if (lttng_is_setuid_setgid()) {
257 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
258 "-g, --group");
259 } else {
260 tracing_group_name = strdup(arg);
261 if (tracing_group_name == NULL) {
262 ret = -errno;
263 PERROR("strdup");
264 goto end;
265 }
266 tracing_group_name_override = 1;
330a40bb 267 }
cd60b05a
JG		 268 break;
269 case 'h':
4fc83d94 270 ret = utils_show_help(8, "lttng-relayd", help_msg);
655b5cc1 271 if (ret) {
4fc83d94 272 ERR("Cannot show --help for `lttng-relayd`");
655b5cc1
PP		 273 perror("exec");
274 }
cd60b05a 275 exit(EXIT_FAILURE);
3a904098
AW		 276 case 'V':
277 fprintf(stdout, "%s\n", VERSION);
278 exit(EXIT_SUCCESS);
cd60b05a 279 case 'o':
e8fa9fb0
MD		 280 if (lttng_is_setuid_setgid()) {
281 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
282 "-o, --output");
283 } else {
284 ret = asprintf(&opt_output_path, "%s", arg);
285 if (ret < 0) {
286 ret = -errno;
287 PERROR("asprintf opt_output_path");
288 goto end;
289 }
cd60b05a
JG		 290 }
291 break;
292 case 'v':
293 /* Verbose level can increase using multiple -v */
294 if (arg) {
295 lttng_opt_verbose = config_parse_value(arg);
296 } else {
849e5b7b
DG		 297 /* Only 3 level of verbosity (-vvv). */
298 if (lttng_opt_verbose < 3) {
299 lttng_opt_verbose += 1;
300 }
cd60b05a
JG		 301 }
302 break;
303 default:
304 /* Unknown option or other error.
305 * Error is printed by getopt, just return */
306 ret = -1;
307 goto end;
308 }
309
310 /* All good. */
311 ret = 0;
312
313end:
314 return ret;
315}
316
317/*
318 * config_entry_handler_cb used to handle options read from a config file.
f40ef1d5 319 * See config_entry_handler_cb comment in common/config/session-config.h for the
cd60b05a
JG		 320 * return value conventions.
321 */
7591bab1 322static int config_entry_handler(const struct config_entry *entry, void *unused)
cd60b05a
JG		 323{
324 int ret = 0, i;
325
326 if (!entry || !entry->name || !entry->value) {
327 ret = -EINVAL;
328 goto end;
329 }
330
331 /* Check if the option is to be ignored */
332 for (i = 0; i < sizeof(config_ignore_options) / sizeof(char *); i++) {
333 if (!strcmp(entry->name, config_ignore_options[i])) {
334 goto end;
335 }
336 }
337
338 for (i = 0; i < (sizeof(long_options) / sizeof(struct option)) - 1; i++) {
339 /* Ignore if entry name is not fully matched. */
340 if (strcmp(entry->name, long_options[i].name)) {
341 continue;
342 }
343
344 /*
7591bab1
MD		 345 * If the option takes no argument on the command line,
346 * we have to check if the value is "true". We support
347 * non-zero numeric values, true, on and yes.
cd60b05a
JG		 348 */
349 if (!long_options[i].has_arg) {
350 ret = config_parse_value(entry->value);
351 if (ret <= 0) {
352 if (ret) {
353 WARN("Invalid configuration value \"%s\" for option %s",
354 entry->value, entry->name);
355 }
356 /* False, skip boolean config option. */
357 goto end;
358 }
359 }
360
361 ret = set_option(long_options[i].val, entry->value, entry->name);
362 goto end;
363 }
364
365 WARN("Unrecognized option \"%s\" in daemon configuration file.",
366 entry->name);
367
368end:
369 return ret;
370}
371
7591bab1 372static int set_options(int argc, char **argv)
cd60b05a 373{
178a0557 374 int c, ret = 0, option_index = 0, retval = 0;
cd60b05a
JG		 375 int orig_optopt = optopt, orig_optind = optind;
376 char *default_address, *optstring;
377 const char *config_path = NULL;
378
379 optstring = utils_generate_optstring(long_options,
380 sizeof(long_options) / sizeof(struct option));
381 if (!optstring) {
178a0557 382 retval = -ENOMEM;
cd60b05a
JG		 383 goto exit;
384 }
385
386 /* Check for the --config option */
387
388 while ((c = getopt_long(argc, argv, optstring, long_options,
389 &option_index)) != -1) {
390 if (c == '?') {
178a0557 391 retval = -EINVAL;
cd60b05a
JG		 392 goto exit;
393 } else if (c != 'f') {
394 continue;
395 }
396
e8fa9fb0
MD		 397 if (lttng_is_setuid_setgid()) {
398 WARN("Getting '%s' argument from setuid/setgid binary refused for security reasons.",
399 "-f, --config");
400 } else {
401 config_path = utils_expand_path(optarg);
402 if (!config_path) {
403 ERR("Failed to resolve path: %s", optarg);
404 }
cd60b05a
JG		 405 }
406 }
407
408 ret = config_get_section_entries(config_path, config_section_name,
409 config_entry_handler, NULL);
410 if (ret) {
411 if (ret > 0) {
412 ERR("Invalid configuration option at line %i", ret);
cd60b05a 413 }
178a0557 414 retval = -1;
cd60b05a
JG		 415 goto exit;
416 }
b8aa1682 417
cd60b05a
JG		 418 /* Reset getopt's global state */
419 optopt = orig_optopt;
420 optind = orig_optind;
b8aa1682 421 while (1) {
cd60b05a 422 c = getopt_long(argc, argv, optstring, long_options, &option_index);
b8aa1682
JD		 423 if (c == -1) {
424 break;
425 }
426
cd60b05a
JG		 427 ret = set_option(c, optarg, long_options[option_index].name);
428 if (ret < 0) {
178a0557 429 retval = -1;
b8aa1682
JD		 430 goto exit;
431 }
432 }
433
434 /* assign default values */
435 if (control_uri == NULL) {
fa91dc52
MD		 436 ret = asprintf(&default_address,
437 "tcp://" DEFAULT_NETWORK_CONTROL_BIND_ADDRESS ":%d",
438 DEFAULT_NETWORK_CONTROL_PORT);
b8aa1682
JD		 439 if (ret < 0) {
440 PERROR("asprintf default data address");
178a0557 441 retval = -1;
b8aa1682
JD		 442 goto exit;
443 }
444
445 ret = uri_parse(default_address, &control_uri);
446 free(default_address);
447 if (ret < 0) {
448 ERR("Invalid control URI specified");
178a0557 449 retval = -1;
b8aa1682
JD		 450 goto exit;
451 }
452 }
453 if (data_uri == NULL) {
fa91dc52
MD		 454 ret = asprintf(&default_address,
455 "tcp://" DEFAULT_NETWORK_DATA_BIND_ADDRESS ":%d",
456 DEFAULT_NETWORK_DATA_PORT);
b8aa1682
JD		 457 if (ret < 0) {
458 PERROR("asprintf default data address");
178a0557 459 retval = -1;
b8aa1682
JD		 460 goto exit;
461 }
462
463 ret = uri_parse(default_address, &data_uri);
464 free(default_address);
465 if (ret < 0) {
466 ERR("Invalid data URI specified");
178a0557 467 retval = -1;
b8aa1682
JD		 468 goto exit;
469 }
470 }
d3e2ba59 471 if (live_uri == NULL) {
fa91dc52
MD		 472 ret = asprintf(&default_address,
473 "tcp://" DEFAULT_NETWORK_VIEWER_BIND_ADDRESS ":%d",
474 DEFAULT_NETWORK_VIEWER_PORT);
d3e2ba59
JD		 475 if (ret < 0) {
476 PERROR("asprintf default viewer control address");
178a0557 477 retval = -1;
d3e2ba59
JD		 478 goto exit;
479 }
480
481 ret = uri_parse(default_address, &live_uri);
482 free(default_address);
483 if (ret < 0) {
484 ERR("Invalid viewer control URI specified");
178a0557 485 retval = -1;
d3e2ba59
JD		 486 goto exit;
487 }
488 }
b8aa1682
JD		 489
490exit:
cd60b05a 491 free(optstring);
178a0557 492 return retval;
b8aa1682
JD		 493}
494
7591bab1
MD		 495static void print_global_objects(void)
496{
497 rcu_register_thread();
498
499 print_viewer_streams();
500 print_relay_streams();
501 print_sessions();
502
503 rcu_unregister_thread();
504}
505
b8aa1682
JD		 506/*
507 * Cleanup the daemon
508 */
7591bab1 509static void relayd_cleanup(void)
b8aa1682 510{
7591bab1
MD		 511 print_global_objects();
512
b8aa1682
JD		 513 DBG("Cleaning up");
514
178a0557
MD		 515 if (viewer_streams_ht)
516 lttng_ht_destroy(viewer_streams_ht);
517 if (relay_streams_ht)
518 lttng_ht_destroy(relay_streams_ht);
7591bab1
MD		 519 if (sessions_ht)
520 lttng_ht_destroy(sessions_ht);
178a0557 521
095a4ae5
MD		 522 /* free the dynamically allocated opt_output_path */
523 free(opt_output_path);
524
a02de639
CB		 525 /* Close thread quit pipes */
526 utils_close_pipe(thread_quit_pipe);
527
710c1f73
DG		 528 uri_free(control_uri);
529 uri_free(data_uri);
8d5c808e 530 /* Live URI is freed in the live thread. */
cd60b05a
JG		 531
532 if (tracing_group_name_override) {
533 free((void *) tracing_group_name);
534 }
b8aa1682
JD		 535}
536
537/*
538 * Write to writable pipe used to notify a thread.
539 */
7591bab1 540static int notify_thread_pipe(int wpipe)
b8aa1682 541{
6cd525e8 542 ssize_t ret;
b8aa1682 543
6cd525e8
MD		 544 ret = lttng_write(wpipe, "!", 1);
545 if (ret < 1) {
b8aa1682 546 PERROR("write poll pipe");
b4aacfdc 547 goto end;
b8aa1682 548 }
b4aacfdc
MD		 549 ret = 0;
550end:
b8aa1682
JD		 551 return ret;
552}
553
7591bab1 554static int notify_health_quit_pipe(int *pipe)
65931c8b 555{
6cd525e8 556 ssize_t ret;
65931c8b 557
6cd525e8
MD		 558 ret = lttng_write(pipe[1], "4", 1);
559 if (ret < 1) {
65931c8b 560 PERROR("write relay health quit");
b4aacfdc 561 goto end;
65931c8b 562 }
b4aacfdc
MD		 563 ret = 0;
564end:
565 return ret;
65931c8b
MD		 566}
567
b8aa1682 568/*
b4aacfdc 569 * Stop all relayd and relayd-live threads.
b8aa1682 570 */
b4aacfdc 571int lttng_relay_stop_threads(void)
b8aa1682 572{
b4aacfdc 573 int retval = 0;
b8aa1682
JD		 574
575 /* Stopping all threads */
576 DBG("Terminating all threads");
b4aacfdc 577 if (notify_thread_pipe(thread_quit_pipe[1])) {
b8aa1682 578 ERR("write error on thread quit pipe");
b4aacfdc 579 retval = -1;
b8aa1682
JD		 580 }
581
b4aacfdc
MD		 582 if (notify_health_quit_pipe(health_quit_pipe)) {
583 ERR("write error on health quit pipe");
584 }
65931c8b 585
b8aa1682 586 /* Dispatch thread */
26c9d55e 587 CMM_STORE_SHARED(dispatch_thread_exit, 1);
58eb9381 588 futex_nto1_wake(&relay_conn_queue.futex);
178a0557 589
b4aacfdc 590 if (relayd_live_stop()) {
178a0557 591 ERR("Error stopping live threads");
b4aacfdc 592 retval = -1;
178a0557 593 }
b4aacfdc 594 return retval;
b8aa1682
JD		 595}
596
597/*
598 * Signal handler for the daemon
599 *
600 * Simply stop all worker threads, leaving main() return gracefully after
601 * joining all threads and calling cleanup().
602 */
7591bab1 603static void sighandler(int sig)
b8aa1682
JD		 604{
605 switch (sig) {
b8aa1682
JD		 606 case SIGINT:
607 DBG("SIGINT caught");
b4aacfdc
MD		 608 if (lttng_relay_stop_threads()) {
609 ERR("Error stopping threads");
610 }
b8aa1682
JD		 611 break;
612 case SIGTERM:
613 DBG("SIGTERM caught");
b4aacfdc
MD		 614 if (lttng_relay_stop_threads()) {
615 ERR("Error stopping threads");
616 }
b8aa1682 617 break;
3fd27398
MD		 618 case SIGUSR1:
619 CMM_STORE_SHARED(recv_child_signal, 1);
620 break;
b8aa1682
JD		 621 default:
622 break;
623 }
624}
625
626/*
627 * Setup signal handler for :
628 * SIGINT, SIGTERM, SIGPIPE
629 */
7591bab1 630static int set_signal_handler(void)
b8aa1682
JD		 631{
632 int ret = 0;
633 struct sigaction sa;
634 sigset_t sigset;
635
636 if ((ret = sigemptyset(&sigset)) < 0) {
637 PERROR("sigemptyset");
638 return ret;
639 }
640
b8aa1682
JD		 641 sa.sa_mask = sigset;
642 sa.sa_flags = 0;
0072e5e2
MD		 643
644 sa.sa_handler = sighandler;
b8aa1682
JD		 645 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
646 PERROR("sigaction");
647 return ret;
648 }
649
650 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
651 PERROR("sigaction");
652 return ret;
653 }
654
0072e5e2 655 if ((ret = sigaction(SIGUSR1, &sa, NULL)) < 0) {
b8aa1682
JD		 656 PERROR("sigaction");
657 return ret;
658 }
659
0072e5e2
MD		 660 sa.sa_handler = SIG_IGN;
661 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
3fd27398
MD		 662 PERROR("sigaction");
663 return ret;
664 }
665
666 DBG("Signal handler set for SIGTERM, SIGUSR1, SIGPIPE and SIGINT");
b8aa1682
JD		 667
668 return ret;
669}
670
3fd27398
MD		 671void lttng_relay_notify_ready(void)
672{
673 /* Notify the parent of the fork() process that we are ready. */
674 if (opt_daemon || opt_background) {
675 if (uatomic_sub_return(&lttng_relay_ready, 1) == 0) {
676 kill(child_ppid, SIGUSR1);
677 }
678 }
679}
680
b8aa1682
JD		 681/*
682 * Init thread quit pipe.
683 *
684 * Return -1 on error or 0 if all pipes are created.
685 */
7591bab1 686static int init_thread_quit_pipe(void)
b8aa1682 687{
a02de639 688 int ret;
b8aa1682 689
a02de639 690 ret = utils_create_pipe_cloexec(thread_quit_pipe);
b8aa1682 691
b8aa1682
JD		 692 return ret;
693}
694
695/*
696 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
697 */
7591bab1 698static int create_thread_poll_set(struct lttng_poll_event *events, int size)
b8aa1682
JD		 699{
700 int ret;
701
702 if (events == NULL || size == 0) {
703 ret = -1;
704 goto error;
705 }
706
707 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
708 if (ret < 0) {
709 goto error;
710 }
711
712 /* Add quit pipe */
c7759e6a 713 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
b8aa1682
JD		 714 if (ret < 0) {
715 goto error;
716 }
717
718 return 0;
719
720error:
721 return ret;
722}
723
724/*
725 * Check if the thread quit pipe was triggered.
726 *
727 * Return 1 if it was triggered else 0;
728 */
7591bab1 729static int check_thread_quit_pipe(int fd, uint32_t events)
b8aa1682
JD		 730{
731 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
732 return 1;
733 }
734
735 return 0;
736}
737
738/*
739 * Create and init socket from uri.
740 */
7591bab1 741static struct lttcomm_sock *relay_socket_create(struct lttng_uri *uri)
b8aa1682
JD		 742{
743 int ret;
744 struct lttcomm_sock *sock = NULL;
745
746 sock = lttcomm_alloc_sock_from_uri(uri);
747 if (sock == NULL) {
748 ERR("Allocating socket");
749 goto error;
750 }
751
752 ret = lttcomm_create_sock(sock);
753 if (ret < 0) {
754 goto error;
755 }
756 DBG("Listening on sock %d", sock->fd);
757
758 ret = sock->ops->bind(sock);
759 if (ret < 0) {
2288467f 760 PERROR("Failed to bind socket");
b8aa1682
JD		 761 goto error;
762 }
763
764 ret = sock->ops->listen(sock, -1);
765 if (ret < 0) {
766 goto error;
767
768 }
769
770 return sock;
771
772error:
773 if (sock) {
774 lttcomm_destroy_sock(sock);
775 }
776 return NULL;
777}
778
779/*
780 * This thread manages the listening for new connections on the network
781 */
7591bab1 782static void *relay_thread_listener(void *data)
b8aa1682 783{
095a4ae5 784 int i, ret, pollfd, err = -1;
b8aa1682
JD		 785 uint32_t revents, nb_fd;
786 struct lttng_poll_event events;
787 struct lttcomm_sock *control_sock, *data_sock;
788
b8aa1682
JD		 789 DBG("[thread] Relay listener started");
790
55706a7d
MD		 791 health_register(health_relayd, HEALTH_RELAYD_TYPE_LISTENER);
792
f385ae0a
MD		 793 health_code_update();
794
7591bab1 795 control_sock = relay_socket_create(control_uri);
b8aa1682 796 if (!control_sock) {
095a4ae5 797 goto error_sock_control;
b8aa1682
JD		 798 }
799
7591bab1 800 data_sock = relay_socket_create(data_uri);
b8aa1682 801 if (!data_sock) {
095a4ae5 802 goto error_sock_relay;
b8aa1682
JD		 803 }
804
805 /*
7591bab1
MD		 806 * Pass 3 as size here for the thread quit pipe, control and
807 * data socket.
b8aa1682
JD		 808 */
809 ret = create_thread_poll_set(&events, 3);
810 if (ret < 0) {
811 goto error_create_poll;
812 }
813
814 /* Add the control socket */
815 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
816 if (ret < 0) {
817 goto error_poll_add;
818 }
819
820 /* Add the data socket */
821 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
822 if (ret < 0) {
823 goto error_poll_add;
824 }
825
3fd27398
MD		 826 lttng_relay_notify_ready();
827
9b5e0863
MD		 828 if (testpoint(relayd_thread_listener)) {
829 goto error_testpoint;
830 }
831
b8aa1682 832 while (1) {
f385ae0a
MD		 833 health_code_update();
834
b8aa1682
JD		 835 DBG("Listener accepting connections");
836
b8aa1682 837restart:
f385ae0a 838 health_poll_entry();
b8aa1682 839 ret = lttng_poll_wait(&events, -1);
f385ae0a 840 health_poll_exit();
b8aa1682
JD		 841 if (ret < 0) {
842 /*
843 * Restart interrupted system call.
844 */
845 if (errno == EINTR) {
846 goto restart;
847 }
848 goto error;
849 }
850
0d9c5d77
DG		 851 nb_fd = ret;
852
b8aa1682
JD		 853 DBG("Relay new connection received");
854 for (i = 0; i < nb_fd; i++) {
f385ae0a
MD		 855 health_code_update();
856
b8aa1682
JD		 857 /* Fetch once the poll data */
858 revents = LTTNG_POLL_GETEV(&events, i);
859 pollfd = LTTNG_POLL_GETFD(&events, i);
860
fd20dac9 861 if (!revents) {
7591bab1
MD		 862 /*
863 * No activity for this FD (poll
864 * implementation).
865 */
fd20dac9
MD		 866 continue;
867 }
868
b8aa1682
JD		 869 /* Thread quit pipe has been closed. Killing thread. */
870 ret = check_thread_quit_pipe(pollfd, revents);
871 if (ret) {
095a4ae5
MD		 872 err = 0;
873 goto exit;
b8aa1682
JD		 874 }
875
03e43155 876 if (revents & LPOLLIN) {
4b7f17b2 877 /*
7591bab1
MD		 878 * A new connection is requested, therefore a
879 * sessiond/consumerd connection is allocated in
880 * this thread, enqueued to a global queue and
881 * dequeued (and freed) in the worker thread.
4b7f17b2 882 */
58eb9381
DG		 883 int val = 1;
884 struct relay_connection *new_conn;
4b7f17b2 885 struct lttcomm_sock *newsock;
7591bab1 886 enum connection_type type;
b8aa1682
JD		 887
888 if (pollfd == data_sock->fd) {
7591bab1 889 type = RELAY_DATA;
b8aa1682 890 newsock = data_sock->ops->accept(data_sock);
58eb9381
DG		 891 DBG("Relay data connection accepted, socket %d",
892 newsock->fd);
4b7f17b2
MD		 893 } else {
894 assert(pollfd == control_sock->fd);
7591bab1 895 type = RELAY_CONTROL;
b8aa1682 896 newsock = control_sock->ops->accept(control_sock);
58eb9381
DG		 897 DBG("Relay control connection accepted, socket %d",
898 newsock->fd);
b8aa1682 899 }
58eb9381
DG		 900 if (!newsock) {
901 PERROR("accepting sock");
58eb9381
DG		 902 goto error;
903 }
904
905 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
906 sizeof(val));
b8aa1682
JD		 907 if (ret < 0) {
908 PERROR("setsockopt inet");
4b7f17b2 909 lttcomm_destroy_sock(newsock);
b8aa1682
JD		 910 goto error;
911 }
f056029c
JR		 912
913 ret = socket_apply_keep_alive_config(newsock->fd);
914 if (ret < 0) {
915 ERR("Failed to apply TCP keep-alive configuration on socket (%i)",
916 newsock->fd);
917 lttcomm_destroy_sock(newsock);
918 goto error;
919 }
920
7591bab1
MD		 921 new_conn = connection_create(newsock, type);
922 if (!new_conn) {
923 lttcomm_destroy_sock(newsock);
924 goto error;
925 }
58eb9381
DG		 926
927 /* Enqueue request for the dispatcher thread. */
8bdee6e2
SM		 928 cds_wfcq_enqueue(&relay_conn_queue.head, &relay_conn_queue.tail,
929 &new_conn->qnode);
b8aa1682
JD		 930
931 /*
7591bab1
MD		 932 * Wake the dispatch queue futex.
933 * Implicit memory barrier with the
934 * exchange in cds_wfcq_enqueue.
b8aa1682 935 */
58eb9381 936 futex_nto1_wake(&relay_conn_queue.futex);
03e43155
MD		 937 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
938 ERR("socket poll error");
939 goto error;
940 } else {
941 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
942 goto error;
b8aa1682
JD		 943 }
944 }
945 }
946
095a4ae5 947exit:
b8aa1682
JD		 948error:
949error_poll_add:
9b5e0863 950error_testpoint:
b8aa1682
JD		 951 lttng_poll_clean(&events);
952error_create_poll:
095a4ae5
MD		 953 if (data_sock->fd >= 0) {
954 ret = data_sock->ops->close(data_sock);
b8aa1682
JD		 955 if (ret) {
956 PERROR("close");
957 }
b8aa1682 958 }
095a4ae5
MD		 959 lttcomm_destroy_sock(data_sock);
960error_sock_relay:
961 if (control_sock->fd >= 0) {
962 ret = control_sock->ops->close(control_sock);
b8aa1682
JD		 963 if (ret) {
964 PERROR("close");
965 }
b8aa1682 966 }
095a4ae5
MD		 967 lttcomm_destroy_sock(control_sock);
968error_sock_control:
969 if (err) {
f385ae0a
MD		 970 health_error();
971 ERR("Health error occurred in %s", __func__);
095a4ae5 972 }
55706a7d 973 health_unregister(health_relayd);
b8aa1682 974 DBG("Relay listener thread cleanup complete");
b4aacfdc 975 lttng_relay_stop_threads();
b8aa1682
JD		 976 return NULL;
977}
978
979/*
980 * This thread manages the dispatching of the requests to worker threads
981 */
7591bab1 982static void *relay_thread_dispatcher(void *data)
b8aa1682 983{
6cd525e8
MD		 984 int err = -1;
985 ssize_t ret;
8bdee6e2 986 struct cds_wfcq_node *node;
58eb9381 987 struct relay_connection *new_conn = NULL;
b8aa1682
JD		 988
989 DBG("[thread] Relay dispatcher started");
990
55706a7d
MD		 991 health_register(health_relayd, HEALTH_RELAYD_TYPE_DISPATCHER);
992
9b5e0863
MD		 993 if (testpoint(relayd_thread_dispatcher)) {
994 goto error_testpoint;
995 }
996
f385ae0a
MD		 997 health_code_update();
998
0ed3b1a8 999 for (;;) {
f385ae0a
MD		 1000 health_code_update();
1001
b8aa1682 1002 /* Atomically prepare the queue futex */
58eb9381 1003 futex_nto1_prepare(&relay_conn_queue.futex);
b8aa1682 1004
0ed3b1a8
MD		 1005 if (CMM_LOAD_SHARED(dispatch_thread_exit)) {
1006 break;
1007 }
1008
b8aa1682 1009 do {
f385ae0a
MD		 1010 health_code_update();
1011
b8aa1682 1012 /* Dequeue commands */
8bdee6e2
SM		 1013 node = cds_wfcq_dequeue_blocking(&relay_conn_queue.head,
1014 &relay_conn_queue.tail);
b8aa1682
JD		 1015 if (node == NULL) {
1016 DBG("Woken up but nothing in the relay command queue");
1017 /* Continue thread execution */
1018 break;
1019 }
58eb9381 1020 new_conn = caa_container_of(node, struct relay_connection, qnode);
b8aa1682 1021
58eb9381 1022 DBG("Dispatching request waiting on sock %d", new_conn->sock->fd);
b8aa1682
JD		 1023
1024 /*
7591bab1
MD		 1025 * Inform worker thread of the new request. This
1026 * call is blocking so we can be assured that
1027 * the data will be read at some point in time
1028 * or wait to the end of the world :)
b8aa1682 1029 */
58eb9381
DG		 1030 ret = lttng_write(relay_conn_pipe[1], &new_conn, sizeof(new_conn));
1031 if (ret < 0) {
1032 PERROR("write connection pipe");
7591bab1 1033 connection_put(new_conn);
b8aa1682
JD		 1034 goto error;
1035 }
1036 } while (node != NULL);
1037
1038 /* Futex wait on queue. Blocking call on futex() */
f385ae0a 1039 health_poll_entry();
58eb9381 1040 futex_nto1_wait(&relay_conn_queue.futex);
f385ae0a 1041 health_poll_exit();
b8aa1682
JD		 1042 }
1043
f385ae0a
MD		 1044 /* Normal exit, no error */
1045 err = 0;
1046
b8aa1682 1047error:
9b5e0863 1048error_testpoint:
f385ae0a
MD		 1049 if (err) {
1050 health_error();
1051 ERR("Health error occurred in %s", __func__);
1052 }
55706a7d 1053 health_unregister(health_relayd);
b8aa1682 1054 DBG("Dispatch thread dying");
b4aacfdc 1055 lttng_relay_stop_threads();
b8aa1682
JD		 1056 return NULL;
1057}
1058
b8aa1682 1059/*
7591bab1 1060 * Set index data from the control port to a given index object.
b8aa1682 1061 */
7591bab1 1062static int set_index_control_data(struct relay_index *index,
234cd636
JD		 1063 struct lttcomm_relayd_index *data,
1064 struct relay_connection *conn)
1c20f0e2 1065{
7591bab1 1066 struct ctf_packet_index index_data;
1c20f0e2
JD		 1067
1068 /*
5312a3ed 1069 * The index on disk is encoded in big endian.
1c20f0e2 1070 */
5312a3ed
JG		 1071 index_data.packet_size = htobe64(data->packet_size);
1072 index_data.content_size = htobe64(data->content_size);
1073 index_data.timestamp_begin = htobe64(data->timestamp_begin);
1074 index_data.timestamp_end = htobe64(data->timestamp_end);
1075 index_data.events_discarded = htobe64(data->events_discarded);
1076 index_data.stream_id = htobe64(data->stream_id);
234cd636
JD		 1077
1078 if (conn->minor >= 8) {
5312a3ed
JG		 1079 index->index_data.stream_instance_id = htobe64(data->stream_instance_id);
1080 index->index_data.packet_seq_num = htobe64(data->packet_seq_num);
234cd636
JD		 1081 }
1082
7591bab1 1083 return relay_index_set_data(index, &index_data);
1c20f0e2
JD		 1084}
1085
c5b6f4f0
DG		 1086/*
1087 * Handle the RELAYD_CREATE_SESSION command.
1088 *
1089 * On success, send back the session id or else return a negative value.
1090 */
5312a3ed
JG		 1091static int relay_create_session(const struct lttcomm_relayd_hdr *recv_hdr,
1092 struct relay_connection *conn,
1093 const struct lttng_buffer_view *payload)
c5b6f4f0 1094{
5312a3ed
JG		 1095 int ret = 0;
1096 ssize_t send_ret;
c5b6f4f0
DG		 1097 struct relay_session *session;
1098 struct lttcomm_relayd_status_session reply;
36d2e35d 1099 char session_name[LTTNG_NAME_MAX];
9cf3eb20 1100 char hostname[LTTNG_HOST_NAME_MAX];
7591bab1
MD		 1101 uint32_t live_timer = 0;
1102 bool snapshot = false;
c5b6f4f0 1103
36d2e35d 1104 memset(session_name, 0, LTTNG_NAME_MAX);
9cf3eb20 1105 memset(hostname, 0, LTTNG_HOST_NAME_MAX);
c5b6f4f0
DG		 1106
1107 memset(&reply, 0, sizeof(reply));
1108
f86f6389
JR		 1109 if (conn->minor < 4) {
1110 /* From 2.1 to 2.3 */
1111 ret = 0;
1112 } else if (conn->minor >= 4 && conn->minor < 11) {
1113 /* From 2.4 to 2.10 */
5312a3ed 1114 ret = cmd_create_session_2_4(payload, session_name,
7591bab1 1115 hostname, &live_timer, &snapshot);
f86f6389
JR		 1116 } else {
1117 /* From 2.11 to ... */
1118 ret = cmd_create_session_2_11(payload, session_name,
1119 hostname, &live_timer, &snapshot);
7591bab1 1120 }
f86f6389 1121
7591bab1
MD		 1122 if (ret < 0) {
1123 goto send_reply;
d3e2ba59
JD		 1124 }
1125
7591bab1
MD		 1126 session = session_create(session_name, hostname, live_timer,
1127 snapshot, conn->major, conn->minor);
1128 if (!session) {
1129 ret = -1;
1130 goto send_reply;
1131 }
1132 assert(!conn->session);
1133 conn->session = session;
c5b6f4f0
DG		 1134 DBG("Created session %" PRIu64, session->id);
1135
7591bab1
MD		 1136 reply.session_id = htobe64(session->id);
1137
1138send_reply:
c5b6f4f0
DG		 1139 if (ret < 0) {
1140 reply.ret_code = htobe32(LTTNG_ERR_FATAL);
1141 } else {
1142 reply.ret_code = htobe32(LTTNG_OK);
1143 }
1144
58eb9381 1145 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
5312a3ed
JG		 1146 if (send_ret < (ssize_t) sizeof(reply)) {
1147 ERR("Failed to send \"create session\" command reply (ret = %zd)",
1148 send_ret);
1149 ret = -1;
c5b6f4f0
DG		 1150 }
1151
1152 return ret;
1153}
1154
a4baae1b
JD		 1155/*
1156 * When we have received all the streams and the metadata for a channel,
1157 * we make them visible to the viewer threads.
1158 */
7591bab1 1159static void publish_connection_local_streams(struct relay_connection *conn)
a4baae1b 1160{
7591bab1
MD		 1161 struct relay_stream *stream;
1162 struct relay_session *session = conn->session;
a4baae1b 1163
7591bab1
MD		 1164 /*
1165 * We publish all streams belonging to a session atomically wrt
1166 * session lock.
1167 */
1168 pthread_mutex_lock(&session->lock);
1169 rcu_read_lock();
1170 cds_list_for_each_entry_rcu(stream, &session->recv_list,
1171 recv_node) {
1172 stream_publish(stream);
a4baae1b 1173 }
7591bab1 1174 rcu_read_unlock();
a4baae1b 1175
7591bab1
MD		 1176 /*
1177 * Inform the viewer that there are new streams in the session.
1178 */
1179 if (session->viewer_attached) {
1180 uatomic_set(&session->new_streams, 1);
1181 }
1182 pthread_mutex_unlock(&session->lock);
a4baae1b
JD		 1183}
1184
b8aa1682
JD		 1185/*
1186 * relay_add_stream: allocate a new stream for a session
1187 */
5312a3ed
JG		 1188static int relay_add_stream(const struct lttcomm_relayd_hdr *recv_hdr,
1189 struct relay_connection *conn,
1190 const struct lttng_buffer_view *payload)
b8aa1682 1191{
7591bab1
MD		 1192 int ret;
1193 ssize_t send_ret;
58eb9381 1194 struct relay_session *session = conn->session;
b8aa1682
JD		 1195 struct relay_stream *stream = NULL;
1196 struct lttcomm_relayd_status_stream reply;
4030a636 1197 struct ctf_trace *trace = NULL;
7591bab1
MD		 1198 uint64_t stream_handle = -1ULL;
1199 char *path_name = NULL, *channel_name = NULL;
1200 uint64_t tracefile_size = 0, tracefile_count = 0;
b8aa1682 1201
5312a3ed 1202 if (!session || !conn->version_check_done) {
b8aa1682
JD		 1203 ERR("Trying to add a stream before version check");
1204 ret = -1;
1205 goto end_no_session;
1206 }
1207
7591bab1
MD		 1208 switch (session->minor) {
1209 case 1: /* LTTng sessiond 2.1. Allocates path_name and channel_name. */
5312a3ed 1210 ret = cmd_recv_stream_2_1(payload, &path_name,
7591bab1 1211 &channel_name);
0f907de1 1212 break;
7591bab1 1213 case 2: /* LTTng sessiond 2.2. Allocates path_name and channel_name. */
0f907de1 1214 default:
5312a3ed 1215 ret = cmd_recv_stream_2_2(payload, &path_name,
7591bab1 1216 &channel_name, &tracefile_size, &tracefile_count);
0f907de1
JD		 1217 break;
1218 }
1219 if (ret < 0) {
7591bab1 1220 goto send_reply;
b8aa1682
JD		 1221 }
1222
7591bab1 1223 trace = ctf_trace_get_by_path_or_create(session, path_name);
2a174661 1224 if (!trace) {
7591bab1 1225 goto send_reply;
2a174661 1226 }
7591bab1 1227 /* This stream here has one reference on the trace. */
2a174661 1228
7591bab1
MD		 1229 pthread_mutex_lock(&last_relay_stream_id_lock);
1230 stream_handle = ++last_relay_stream_id;
1231 pthread_mutex_unlock(&last_relay_stream_id_lock);
d3e2ba59 1232
7591bab1
MD		 1233 /* We pass ownership of path_name and channel_name. */
1234 stream = stream_create(trace, stream_handle, path_name,
1235 channel_name, tracefile_size, tracefile_count);
1236 path_name = NULL;
1237 channel_name = NULL;
a4baae1b 1238
2a174661 1239 /*
7591bab1
MD		 1240 * Streams are the owners of their trace. Reference to trace is
1241 * kept within stream_create().
2a174661 1242 */
7591bab1 1243 ctf_trace_put(trace);
d3e2ba59 1244
7591bab1 1245send_reply:
53efb85a 1246 memset(&reply, 0, sizeof(reply));
7591bab1
MD		 1247 reply.handle = htobe64(stream_handle);
1248 if (!stream) {
f73fabfd 1249 reply.ret_code = htobe32(LTTNG_ERR_UNK);
b8aa1682 1250 } else {
f73fabfd 1251 reply.ret_code = htobe32(LTTNG_OK);
b8aa1682 1252 }
5af40280 1253
58eb9381 1254 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
b8aa1682 1255 sizeof(struct lttcomm_relayd_status_stream), 0);
5312a3ed
JG		 1256 if (send_ret < (ssize_t) sizeof(reply)) {
1257 ERR("Failed to send \"add stream\" command reply (ret = %zd)",
1258 send_ret);
1259 ret = -1;
b8aa1682
JD		 1260 }
1261
1262end_no_session:
7591bab1
MD		 1263 free(path_name);
1264 free(channel_name);
0f907de1 1265 return ret;
b8aa1682
JD		 1266}
1267
173af62f
DG		 1268/*
1269 * relay_close_stream: close a specific stream
1270 */
5312a3ed
JG		 1271static int relay_close_stream(const struct lttcomm_relayd_hdr *recv_hdr,
1272 struct relay_connection *conn,
1273 const struct lttng_buffer_view *payload)
173af62f 1274{
5312a3ed
JG		 1275 int ret;
1276 ssize_t send_ret;
58eb9381 1277 struct relay_session *session = conn->session;
173af62f
DG		 1278 struct lttcomm_relayd_close_stream stream_info;
1279 struct lttcomm_relayd_generic_reply reply;
1280 struct relay_stream *stream;
173af62f
DG		 1281
1282 DBG("Close stream received");
1283
5312a3ed 1284 if (!session || !conn->version_check_done) {
173af62f
DG		 1285 ERR("Trying to close a stream before version check");
1286 ret = -1;
1287 goto end_no_session;
1288 }
1289
5312a3ed
JG		 1290 if (payload->size < sizeof(stream_info)) {
1291 ERR("Unexpected payload size in \"relay_close_stream\": expected >= %zu bytes, got %zu bytes",
1292 sizeof(stream_info), payload->size);
173af62f
DG		 1293 ret = -1;
1294 goto end_no_session;
1295 }
5312a3ed
JG		 1296 memcpy(&stream_info, payload->data, sizeof(stream_info));
1297 stream_info.stream_id = be64toh(stream_info.stream_id);
1298 stream_info.last_net_seq_num = be64toh(stream_info.last_net_seq_num);
173af62f 1299
5312a3ed 1300 stream = stream_get_by_id(stream_info.stream_id);
173af62f
DG		 1301 if (!stream) {
1302 ret = -1;
7591bab1 1303 goto end;
173af62f 1304 }
77f7bd85
MD		 1305
1306 /*
1307 * Set last_net_seq_num before the close flag. Required by data
1308 * pending check.
1309 */
7591bab1 1310 pthread_mutex_lock(&stream->lock);
5312a3ed 1311 stream->last_net_seq_num = stream_info.last_net_seq_num;
77f7bd85
MD		 1312 pthread_mutex_unlock(&stream->lock);
1313
bda7c7b9
JG		 1314 /*
1315 * This is one of the conditions which may trigger a stream close
1316 * with the others being:
1317 * 1) A close command is received for a stream
1318 * 2) The control connection owning the stream is closed
1319 * 3) We have received all of the stream's data _after_ a close
1320 * request.
1321 */
1322 try_stream_close(stream);
7591bab1
MD		 1323 if (stream->is_metadata) {
1324 struct relay_viewer_stream *vstream;
173af62f 1325
7591bab1
MD		 1326 vstream = viewer_stream_get_by_id(stream->stream_handle);
1327 if (vstream) {
1328 if (vstream->metadata_sent == stream->metadata_received) {
1329 /*
1330 * Since all the metadata has been sent to the
1331 * viewer and that we have a request to close
1332 * its stream, we can safely teardown the
1333 * corresponding metadata viewer stream.
1334 */
1335 viewer_stream_put(vstream);
1336 }
1337 /* Put local reference. */
1338 viewer_stream_put(vstream);
1339 }
1340 }
7591bab1 1341 stream_put(stream);
5312a3ed 1342 ret = 0;
173af62f 1343
7591bab1 1344end:
53efb85a 1345 memset(&reply, 0, sizeof(reply));
173af62f 1346 if (ret < 0) {
f73fabfd 1347 reply.ret_code = htobe32(LTTNG_ERR_UNK);
173af62f 1348 } else {
f73fabfd 1349 reply.ret_code = htobe32(LTTNG_OK);
173af62f 1350 }
58eb9381 1351 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
173af62f 1352 sizeof(struct lttcomm_relayd_generic_reply), 0);
5312a3ed
JG		 1353 if (send_ret < (ssize_t) sizeof(reply)) {
1354 ERR("Failed to send \"close stream\" command reply (ret = %zd)",
1355 send_ret);
1356 ret = -1;
173af62f
DG		 1357 }
1358
1359end_no_session:
1360 return ret;
1361}
1362
93ec662e
JD		 1363/*
1364 * relay_reset_metadata: reset a metadata stream
1365 */
1366static
5312a3ed
JG		 1367int relay_reset_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1368 struct relay_connection *conn,
1369 const struct lttng_buffer_view *payload)
93ec662e 1370{
5312a3ed
JG		 1371 int ret;
1372 ssize_t send_ret;
93ec662e
JD		 1373 struct relay_session *session = conn->session;
1374 struct lttcomm_relayd_reset_metadata stream_info;
1375 struct lttcomm_relayd_generic_reply reply;
1376 struct relay_stream *stream;
1377
1378 DBG("Reset metadata received");
1379
5312a3ed 1380 if (!session || !conn->version_check_done) {
93ec662e
JD		 1381 ERR("Trying to reset a metadata stream before version check");
1382 ret = -1;
1383 goto end_no_session;
1384 }
1385
5312a3ed
JG		 1386 if (payload->size < sizeof(stream_info)) {
1387 ERR("Unexpected payload size in \"relay_reset_metadata\": expected >= %zu bytes, got %zu bytes",
1388 sizeof(stream_info), payload->size);
93ec662e
JD		 1389 ret = -1;
1390 goto end_no_session;
1391 }
5312a3ed
JG		 1392 memcpy(&stream_info, payload->data, sizeof(stream_info));
1393 stream_info.stream_id = be64toh(stream_info.stream_id);
1394 stream_info.version = be64toh(stream_info.version);
1395
1396 DBG("Update metadata to version %" PRIu64, stream_info.version);
93ec662e
JD		 1397
1398 /* Unsupported for live sessions for now. */
1399 if (session->live_timer != 0) {
1400 ret = -1;
1401 goto end;
1402 }
1403
5312a3ed 1404 stream = stream_get_by_id(stream_info.stream_id);
93ec662e
JD		 1405 if (!stream) {
1406 ret = -1;
1407 goto end;
1408 }
1409 pthread_mutex_lock(&stream->lock);
1410 if (!stream->is_metadata) {
1411 ret = -1;
1412 goto end_unlock;
1413 }
1414
1415 ret = utils_rotate_stream_file(stream->path_name, stream->channel_name,
1416 0, 0, -1, -1, stream->stream_fd->fd, NULL,
1417 &stream->stream_fd->fd);
1418 if (ret < 0) {
1419 ERR("Failed to rotate metadata file %s of channel %s",
1420 stream->path_name, stream->channel_name);
1421 goto end_unlock;
1422 }
1423
1424end_unlock:
1425 pthread_mutex_unlock(&stream->lock);
1426 stream_put(stream);
1427
1428end:
1429 memset(&reply, 0, sizeof(reply));
1430 if (ret < 0) {
1431 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1432 } else {
1433 reply.ret_code = htobe32(LTTNG_OK);
1434 }
1435 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1436 sizeof(struct lttcomm_relayd_generic_reply), 0);
5312a3ed
JG		 1437 if (send_ret < (ssize_t) sizeof(reply)) {
1438 ERR("Failed to send \"reset metadata\" command reply (ret = %zd)",
1439 send_ret);
1440 ret = -1;
93ec662e
JD		 1441 }
1442
1443end_no_session:
1444 return ret;
1445}
1446
b8aa1682
JD		 1447/*
1448 * relay_unknown_command: send -1 if received unknown command
1449 */
7591bab1 1450static void relay_unknown_command(struct relay_connection *conn)
b8aa1682
JD		 1451{
1452 struct lttcomm_relayd_generic_reply reply;
5312a3ed 1453 ssize_t send_ret;
b8aa1682 1454
53efb85a 1455 memset(&reply, 0, sizeof(reply));
f73fabfd 1456 reply.ret_code = htobe32(LTTNG_ERR_UNK);
5312a3ed
JG		 1457 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1458 if (send_ret < sizeof(reply)) {
1459 ERR("Failed to send \"unknown command\" command reply (ret = %zd)", send_ret);
b8aa1682
JD		 1460 }
1461}
1462
1463/*
1464 * relay_start: send an acknowledgment to the client to tell if we are
1465 * ready to receive data. We are ready if a session is established.
1466 */
5312a3ed
JG		 1467static int relay_start(const struct lttcomm_relayd_hdr *recv_hdr,
1468 struct relay_connection *conn,
1469 const struct lttng_buffer_view *payload)
b8aa1682 1470{
5312a3ed
JG		 1471 int ret = 0;
1472 ssize_t send_ret;
b8aa1682 1473 struct lttcomm_relayd_generic_reply reply;
58eb9381 1474 struct relay_session *session = conn->session;
b8aa1682
JD		 1475
1476 if (!session) {
1477 DBG("Trying to start the streaming without a session established");
f73fabfd 1478 ret = htobe32(LTTNG_ERR_UNK);
b8aa1682
JD		 1479 }
1480
53efb85a 1481 memset(&reply, 0, sizeof(reply));
5312a3ed
JG		 1482 reply.ret_code = htobe32(LTTNG_OK);
1483 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1484 sizeof(reply), 0);
1485 if (send_ret < (ssize_t) sizeof(reply)) {
1486 ERR("Failed to send \"relay_start\" command reply (ret = %zd)",
1487 send_ret);
1488 ret = -1;
b8aa1682
JD		 1489 }
1490
1491 return ret;
1492}
1493
1d4dfdef
DG		 1494/*
1495 * Append padding to the file pointed by the file descriptor fd.
1496 */
1497static int write_padding_to_file(int fd, uint32_t size)
1498{
6cd525e8 1499 ssize_t ret = 0;
1d4dfdef
DG		 1500 char *zeros;
1501
1502 if (size == 0) {
1503 goto end;
1504 }
1505
1506 zeros = zmalloc(size);
1507 if (zeros == NULL) {
1508 PERROR("zmalloc zeros for padding");
1509 ret = -1;
1510 goto end;
1511 }
1512
6cd525e8
MD		 1513 ret = lttng_write(fd, zeros, size);
1514 if (ret < size) {
1d4dfdef
DG		 1515 PERROR("write padding to file");
1516 }
1517
e986c7a1
DG		 1518 free(zeros);
1519
1d4dfdef
DG		 1520end:
1521 return ret;
1522}
1523
d3ecc550
JD		 1524/*
1525 * Close the current index file if it is open, and create a new one.
1526 *
1527 * Return 0 on success, -1 on error.
1528 */
1529static
1530int create_rotate_index_file(struct relay_stream *stream)
1531{
1532 int ret;
1533 uint32_t major, minor;
1534
1535 /* Put ref on previous index_file. */
1536 if (stream->index_file) {
1537 lttng_index_file_put(stream->index_file);
1538 stream->index_file = NULL;
1539 }
1540 major = stream->trace->session->major;
1541 minor = stream->trace->session->minor;
1542 stream->index_file = lttng_index_file_create(stream->path_name,
1543 stream->channel_name,
1544 -1, -1, stream->tracefile_size,
1545 tracefile_array_get_file_index_head(stream->tfa),
1546 lttng_to_index_major(major, minor),
1547 lttng_to_index_minor(major, minor));
1548 if (!stream->index_file) {
1549 ret = -1;
1550 goto end;
1551 }
1552
1553 ret = 0;
1554
1555end:
1556 return ret;
1557}
1558
1559static
1560int do_rotate_stream(struct relay_stream *stream)
1561{
1562 int ret;
1563
1564 /* Perform the stream rotation. */
1565 ret = utils_rotate_stream_file(stream->path_name,
1566 stream->channel_name, stream->tracefile_size,
1567 stream->tracefile_count, -1,
1568 -1, stream->stream_fd->fd,
1569 NULL, &stream->stream_fd->fd);
1570 if (ret < 0) {
1571 ERR("Rotating stream output file");
1572 goto end;
1573 }
1574 stream->tracefile_size_current = 0;
1575
1576 /* Rotate also the index if the stream is not a metadata stream. */
1577 if (!stream->is_metadata) {
1578 ret = create_rotate_index_file(stream);
1579 if (ret < 0) {
1580 ERR("Failed to rotate index file");
1581 goto end;
1582 }
1583 }
1584
1585 stream->rotate_at_seq_num = -1ULL;
1586 stream->pos_after_last_complete_data_index = 0;
1587
1588end:
1589 return ret;
1590}
1591
1592/*
1593 * If too much data has been written in a tracefile before we received the
1594 * rotation command, we have to move the excess data to the new tracefile and
1595 * perform the rotation. This can happen because the control and data
1596 * connections are separate, the indexes as well as the commands arrive from
1597 * the control connection and we have no control over the order so we could be
1598 * in a situation where too much data has been received on the data connection
1599 * before the rotation command on the control connection arrives. We don't need
1600 * to update the index because its order is guaranteed with the rotation
1601 * command message.
1602 */
1603static
1604int rotate_truncate_stream(struct relay_stream *stream)
1605{
1606 int ret, new_fd;
a5df8828 1607 off_t lseek_ret;
d3ecc550
JD		 1608 uint64_t diff, pos = 0;
1609 char buf[FILE_COPY_BUFFER_SIZE];
1610
1611 assert(!stream->is_metadata);
1612
1613 assert(stream->tracefile_size_current >
1614 stream->pos_after_last_complete_data_index);
1615 diff = stream->tracefile_size_current -
1616 stream->pos_after_last_complete_data_index;
1617
1618 /* Create the new tracefile. */
1619 new_fd = utils_create_stream_file(stream->path_name,
1620 stream->channel_name,
1621 stream->tracefile_size, stream->tracefile_count,
1622 /* uid */ -1, /* gid */ -1, /* suffix */ NULL);
1623 if (new_fd < 0) {
1624 ERR("Failed to create new stream file at path %s for channel %s",
1625 stream->path_name, stream->channel_name);
1626 ret = -1;
1627 goto end;
1628 }
1629
1630 /*
1631 * Rewind the current tracefile to the position at which the rotation
1632 * should have occured.
1633 */
a5df8828 1634 lseek_ret = lseek(stream->stream_fd->fd,
d3ecc550 1635 stream->pos_after_last_complete_data_index, SEEK_SET);
a5df8828 1636 if (lseek_ret < 0) {
d3ecc550 1637 PERROR("seek truncate stream");
a5df8828 1638 ret = -1;
d3ecc550
JD		 1639 goto end;
1640 }
1641
1642 /* Move data from the old file to the new file. */
1643 while (pos < diff) {
1644 uint64_t count, bytes_left;
1645 ssize_t io_ret;
1646
1647 bytes_left = diff - pos;
1648 count = bytes_left > sizeof(buf) ? sizeof(buf) : bytes_left;
1649 assert(count <= SIZE_MAX);
1650
1651 io_ret = lttng_read(stream->stream_fd->fd, buf, count);
1652 if (io_ret < (ssize_t) count) {
1653 char error_string[256];
1654
1655 snprintf(error_string, sizeof(error_string),
1656 "Failed to read %" PRIu64 " bytes from fd %i in rotate_truncate_stream(), returned %zi",
1657 count, stream->stream_fd->fd, io_ret);
1658 if (io_ret == -1) {
1659 PERROR("%s", error_string);
1660 } else {
1661 ERR("%s", error_string);
1662 }
1663 ret = -1;
1664 goto end;
1665 }
1666
1667 io_ret = lttng_write(new_fd, buf, count);
1668 if (io_ret < (ssize_t) count) {
1669 char error_string[256];
1670
1671 snprintf(error_string, sizeof(error_string),
1672 "Failed to write %" PRIu64 " bytes from fd %i in rotate_truncate_stream(), returned %zi",
1673 count, new_fd, io_ret);
1674 if (io_ret == -1) {
1675 PERROR("%s", error_string);
1676 } else {
1677 ERR("%s", error_string);
1678 }
1679 ret = -1;
1680 goto end;
1681 }
1682
1683 pos += count;
1684 }
1685
1686 /* Truncate the file to get rid of the excess data. */
1687 ret = ftruncate(stream->stream_fd->fd,
1688 stream->pos_after_last_complete_data_index);
1689 if (ret) {
1690 PERROR("ftruncate");
1691 goto end;
1692 }
1693
1694 ret = close(stream->stream_fd->fd);
1695 if (ret < 0) {
1696 PERROR("Closing tracefile");
1697 goto end;
1698 }
1699
1700 ret = create_rotate_index_file(stream);
1701 if (ret < 0) {
1702 ERR("Rotate stream index file");
1703 goto end;
1704 }
1705
1706 /*
1707 * Update the offset and FD of all the eventual indexes created by the
1708 * data connection before the rotation command arrived.
1709 */
1710 ret = relay_index_switch_all_files(stream);
1711 if (ret < 0) {
1712 ERR("Failed to rotate index file");
1713 goto end;
1714 }
1715
1716 stream->stream_fd->fd = new_fd;
1717 stream->tracefile_size_current = diff;
1718 stream->pos_after_last_complete_data_index = 0;
1719 stream->rotate_at_seq_num = -1ULL;
1720
1721 ret = 0;
1722
1723end:
1724 return ret;
1725}
1726
1727/*
1728 * Check if a stream should perform a rotation (for session rotation).
1729 * Must be called with the stream lock held.
1730 *
1731 * Return 0 on success, a negative value on error.
1732 */
1733static
1734int try_rotate_stream(struct relay_stream *stream)
1735{
1736 int ret = 0;
1737
1738 /* No rotation expected. */
1739 if (stream->rotate_at_seq_num == -1ULL) {
1740 goto end;
1741 }
1742
3765c8cc
JG		 1743 if (stream->prev_seq < stream->rotate_at_seq_num ||
1744 stream->prev_seq == -1ULL) {
d3ecc550
JD		 1745 DBG("Stream %" PRIu64 " no yet ready for rotation",
1746 stream->stream_handle);
1747 goto end;
1748 } else if (stream->prev_seq > stream->rotate_at_seq_num) {
1749 DBG("Rotation after too much data has been written in tracefile "
1750 "for stream %" PRIu64 ", need to truncate before "
1751 "rotating", stream->stream_handle);
1752 ret = rotate_truncate_stream(stream);
1753 if (ret) {
1754 ERR("Failed to truncate stream");
1755 goto end;
1756 }
1757 } else {
1758 /* stream->prev_seq == stream->rotate_at_seq_num */
1759 DBG("Stream %" PRIu64 " ready for rotation",
1760 stream->stream_handle);
1761 ret = do_rotate_stream(stream);
1762 }
1763
1764end:
1765 return ret;
1766}
1767
b8aa1682 1768/*
7591bab1 1769 * relay_recv_metadata: receive the metadata for the session.
b8aa1682 1770 */
5312a3ed
JG		 1771static int relay_recv_metadata(const struct lttcomm_relayd_hdr *recv_hdr,
1772 struct relay_connection *conn,
1773 const struct lttng_buffer_view *payload)
b8aa1682 1774{
32d1569c 1775 int ret = 0;
6cd525e8 1776 ssize_t size_ret;
58eb9381 1777 struct relay_session *session = conn->session;
5312a3ed 1778 struct lttcomm_relayd_metadata_payload metadata_payload_header;
b8aa1682 1779 struct relay_stream *metadata_stream;
5312a3ed 1780 uint64_t metadata_payload_size;
b8aa1682
JD		 1781
1782 if (!session) {
1783 ERR("Metadata sent before version check");
1784 ret = -1;
1785 goto end;
1786 }
1787
5312a3ed 1788 if (recv_hdr->data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
f6416125
MD		 1789 ERR("Incorrect data size");
1790 ret = -1;
1791 goto end;
1792 }
5312a3ed
JG		 1793 metadata_payload_size = recv_hdr->data_size -
1794 sizeof(struct lttcomm_relayd_metadata_payload);
f6416125 1795
5312a3ed
JG		 1796 memcpy(&metadata_payload_header, payload->data,
1797 sizeof(metadata_payload_header));
1798 metadata_payload_header.stream_id = be64toh(
1799 metadata_payload_header.stream_id);
1800 metadata_payload_header.padding_size = be32toh(
1801 metadata_payload_header.padding_size);
9d1bbf21 1802
5312a3ed 1803 metadata_stream = stream_get_by_id(metadata_payload_header.stream_id);
b8aa1682
JD		 1804 if (!metadata_stream) {
1805 ret = -1;
7591bab1 1806 goto end;
b8aa1682
JD		 1807 }
1808
7591bab1
MD		 1809 pthread_mutex_lock(&metadata_stream->lock);
1810
5312a3ed
JG		 1811 size_ret = lttng_write(metadata_stream->stream_fd->fd,
1812 payload->data + sizeof(metadata_payload_header),
1813 metadata_payload_size);
1814 if (size_ret < metadata_payload_size) {
b8aa1682
JD		 1815 ERR("Relay error writing metadata on file");
1816 ret = -1;
7591bab1 1817 goto end_put;
b8aa1682 1818 }
1d4dfdef 1819
32d1569c 1820 size_ret = write_padding_to_file(metadata_stream->stream_fd->fd,
5312a3ed 1821 metadata_payload_header.padding_size);
715e6fb1 1822 if (size_ret < (int64_t) metadata_payload_header.padding_size) {
5312a3ed 1823 ret = -1;
7591bab1 1824 goto end_put;
1d4dfdef 1825 }
2a174661 1826
7591bab1 1827 metadata_stream->metadata_received +=
5312a3ed 1828 metadata_payload_size + metadata_payload_header.padding_size;
7591bab1
MD		 1829 DBG2("Relay metadata written. Updated metadata_received %" PRIu64,
1830 metadata_stream->metadata_received);
1d4dfdef 1831
d3ecc550
JD		 1832 ret = try_rotate_stream(metadata_stream);
1833 if (ret < 0) {
1834 goto end_put;
1835 }
1836
7591bab1
MD		 1837end_put:
1838 pthread_mutex_unlock(&metadata_stream->lock);
1839 stream_put(metadata_stream);
b8aa1682
JD		 1840end:
1841 return ret;
1842}
1843
1844/*
1845 * relay_send_version: send relayd version number
1846 */
5312a3ed
JG		 1847static int relay_send_version(const struct lttcomm_relayd_hdr *recv_hdr,
1848 struct relay_connection *conn,
1849 const struct lttng_buffer_view *payload)
b8aa1682 1850{
7f51dcba 1851 int ret;
5312a3ed 1852 ssize_t send_ret;
092b6259 1853 struct lttcomm_relayd_version reply, msg;
87cb6359 1854 bool compatible = true;
b8aa1682 1855
5312a3ed 1856 conn->version_check_done = true;
b8aa1682 1857
092b6259 1858 /* Get version from the other side. */
5312a3ed
JG		 1859 if (payload->size < sizeof(msg)) {
1860 ERR("Unexpected payload size in \"relay_send_version\": expected >= %zu bytes, got %zu bytes",
1861 sizeof(msg), payload->size);
092b6259 1862 ret = -1;
092b6259
DG		 1863 goto end;
1864 }
1865
5312a3ed
JG		 1866 memcpy(&msg, payload->data, sizeof(msg));
1867 msg.major = be32toh(msg.major);
1868 msg.minor = be32toh(msg.minor);
1869
53efb85a 1870 memset(&reply, 0, sizeof(reply));
d83a952c
MD		 1871 reply.major = RELAYD_VERSION_COMM_MAJOR;
1872 reply.minor = RELAYD_VERSION_COMM_MINOR;
d4519fa3
JD		 1873
1874 /* Major versions must be the same */
5312a3ed 1875 if (reply.major != msg.major) {
6151a90f 1876 DBG("Incompatible major versions (%u vs %u), deleting session",
5312a3ed 1877 reply.major, msg.major);
87cb6359 1878 compatible = false;
d4519fa3
JD		 1879 }
1880
58eb9381 1881 conn->major = reply.major;
0f907de1 1882 /* We adapt to the lowest compatible version */
5312a3ed 1883 if (reply.minor <= msg.minor) {
58eb9381 1884 conn->minor = reply.minor;
0f907de1 1885 } else {
5312a3ed 1886 conn->minor = msg.minor;
0f907de1
JD		 1887 }
1888
6151a90f
JD		 1889 reply.major = htobe32(reply.major);
1890 reply.minor = htobe32(reply.minor);
5312a3ed
JG		 1891 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
1892 sizeof(reply), 0);
1893 if (send_ret < (ssize_t) sizeof(reply)) {
1894 ERR("Failed to send \"send version\" command reply (ret = %zd)",
1895 send_ret);
1896 ret = -1;
1897 goto end;
1898 } else {
1899 ret = 0;
6151a90f
JD		 1900 }
1901
87cb6359
JD		 1902 if (!compatible) {
1903 ret = -1;
1904 goto end;
1905 }
1906
58eb9381
DG		 1907 DBG("Version check done using protocol %u.%u", conn->major,
1908 conn->minor);
b8aa1682
JD		 1909
1910end:
1911 return ret;
1912}
1913
c8f59ee5 1914/*
6d805429 1915 * Check for data pending for a given stream id from the session daemon.
c8f59ee5 1916 */
5312a3ed
JG		 1917static int relay_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
1918 struct relay_connection *conn,
1919 const struct lttng_buffer_view *payload)
c8f59ee5 1920{
58eb9381 1921 struct relay_session *session = conn->session;
6d805429 1922 struct lttcomm_relayd_data_pending msg;
c8f59ee5
DG		 1923 struct lttcomm_relayd_generic_reply reply;
1924 struct relay_stream *stream;
5312a3ed 1925 ssize_t send_ret;
c8f59ee5 1926 int ret;
c8f59ee5 1927
6d805429 1928 DBG("Data pending command received");
c8f59ee5 1929
5312a3ed 1930 if (!session || !conn->version_check_done) {
c8f59ee5
DG		 1931 ERR("Trying to check for data before version check");
1932 ret = -1;
1933 goto end_no_session;
1934 }
1935
5312a3ed
JG		 1936 if (payload->size < sizeof(msg)) {
1937 ERR("Unexpected payload size in \"relay_data_pending\": expected >= %zu bytes, got %zu bytes",
1938 sizeof(msg), payload->size);
c8f59ee5
DG		 1939 ret = -1;
1940 goto end_no_session;
1941 }
5312a3ed
JG		 1942 memcpy(&msg, payload->data, sizeof(msg));
1943 msg.stream_id = be64toh(msg.stream_id);
1944 msg.last_net_seq_num = be64toh(msg.last_net_seq_num);
c8f59ee5 1945
5312a3ed 1946 stream = stream_get_by_id(msg.stream_id);
de91f48a 1947 if (stream == NULL) {
c8f59ee5 1948 ret = -1;
7591bab1 1949 goto end;
c8f59ee5
DG		 1950 }
1951
7591bab1
MD		 1952 pthread_mutex_lock(&stream->lock);
1953
6d805429 1954 DBG("Data pending for stream id %" PRIu64 " prev_seq %" PRIu64
5312a3ed
JG		 1955 " and last_seq %" PRIu64, msg.stream_id,
1956 stream->prev_seq, msg.last_net_seq_num);
c8f59ee5 1957
33832e64 1958 /* Avoid wrapping issue */
5312a3ed 1959 if (((int64_t) (stream->prev_seq - msg.last_net_seq_num)) >= 0) {
6d805429 1960 /* Data has in fact been written and is NOT pending */
c8f59ee5 1961 ret = 0;
6d805429
DG		 1962 } else {
1963 /* Data still being streamed thus pending */
1964 ret = 1;
c8f59ee5
DG		 1965 }
1966
7591bab1
MD		 1967 stream->data_pending_check_done = true;
1968 pthread_mutex_unlock(&stream->lock);
f7079f67 1969
7591bab1
MD		 1970 stream_put(stream);
1971end:
c8f59ee5 1972
53efb85a 1973 memset(&reply, 0, sizeof(reply));
c8f59ee5 1974 reply.ret_code = htobe32(ret);
5312a3ed
JG		 1975 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
1976 if (send_ret < (ssize_t) sizeof(reply)) {
1977 ERR("Failed to send \"data pending\" command reply (ret = %zd)",
1978 send_ret);
1979 ret = -1;
c8f59ee5
DG		 1980 }
1981
1982end_no_session:
1983 return ret;
1984}
1985
1986/*
1987 * Wait for the control socket to reach a quiescent state.
1988 *
7591bab1
MD		 1989 * Note that for now, when receiving this command from the session
1990 * daemon, this means that every subsequent commands or data received on
1991 * the control socket has been handled. So, this is why we simply return
1992 * OK here.
c8f59ee5 1993 */
5312a3ed
JG		 1994static int relay_quiescent_control(const struct lttcomm_relayd_hdr *recv_hdr,
1995 struct relay_connection *conn,
1996 const struct lttng_buffer_view *payload)
c8f59ee5
DG		 1997{
1998 int ret;
5312a3ed 1999 ssize_t send_ret;
ad7051c0 2000 struct relay_stream *stream;
ad7051c0 2001 struct lttcomm_relayd_quiescent_control msg;
c8f59ee5
DG		 2002 struct lttcomm_relayd_generic_reply reply;
2003
2004 DBG("Checking quiescent state on control socket");
2005
5312a3ed 2006 if (!conn->session || !conn->version_check_done) {
ad7051c0
DG		 2007 ERR("Trying to check for data before version check");
2008 ret = -1;
2009 goto end_no_session;
2010 }
2011
5312a3ed
JG		 2012 if (payload->size < sizeof(msg)) {
2013 ERR("Unexpected payload size in \"relay_quiescent_control\": expected >= %zu bytes, got %zu bytes",
2014 sizeof(msg), payload->size);
ad7051c0
DG		 2015 ret = -1;
2016 goto end_no_session;
2017 }
5312a3ed
JG		 2018 memcpy(&msg, payload->data, sizeof(msg));
2019 msg.stream_id = be64toh(msg.stream_id);
ad7051c0 2020
5312a3ed 2021 stream = stream_get_by_id(msg.stream_id);
7591bab1
MD		 2022 if (!stream) {
2023 goto reply;
2024 }
2025 pthread_mutex_lock(&stream->lock);
2026 stream->data_pending_check_done = true;
2027 pthread_mutex_unlock(&stream->lock);
5312a3ed
JG		 2028
2029 DBG("Relay quiescent control pending flag set to %" PRIu64, msg.stream_id);
7591bab1
MD		 2030 stream_put(stream);
2031reply:
53efb85a 2032 memset(&reply, 0, sizeof(reply));
c8f59ee5 2033 reply.ret_code = htobe32(LTTNG_OK);
5312a3ed
JG		 2034 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2035 if (send_ret < (ssize_t) sizeof(reply)) {
2036 ERR("Failed to send \"quiescent control\" command reply (ret = %zd)",
2037 send_ret);
2038 ret = -1;
2039 } else {
2040 ret = 0;
c8f59ee5
DG		 2041 }
2042
ad7051c0 2043end_no_session:
c8f59ee5
DG		 2044 return ret;
2045}
2046
f7079f67 2047/*
7591bab1
MD		 2048 * Initialize a data pending command. This means that a consumer is about
2049 * to ask for data pending for each stream it holds. Simply iterate over
2050 * all streams of a session and set the data_pending_check_done flag.
f7079f67
DG		 2051 *
2052 * This command returns to the client a LTTNG_OK code.
2053 */
5312a3ed
JG		 2054static int relay_begin_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2055 struct relay_connection *conn,
2056 const struct lttng_buffer_view *payload)
f7079f67
DG		 2057{
2058 int ret;
5312a3ed 2059 ssize_t send_ret;
f7079f67
DG		 2060 struct lttng_ht_iter iter;
2061 struct lttcomm_relayd_begin_data_pending msg;
2062 struct lttcomm_relayd_generic_reply reply;
2063 struct relay_stream *stream;
f7079f67
DG		 2064
2065 assert(recv_hdr);
58eb9381 2066 assert(conn);
f7079f67
DG		 2067
2068 DBG("Init streams for data pending");
2069
5312a3ed 2070 if (!conn->session || !conn->version_check_done) {
f7079f67
DG		 2071 ERR("Trying to check for data before version check");
2072 ret = -1;
2073 goto end_no_session;
2074 }
2075
5312a3ed
JG		 2076 if (payload->size < sizeof(msg)) {
2077 ERR("Unexpected payload size in \"relay_begin_data_pending\": expected >= %zu bytes, got %zu bytes",
2078 sizeof(msg), payload->size);
f7079f67
DG		 2079 ret = -1;
2080 goto end_no_session;
2081 }
5312a3ed
JG		 2082 memcpy(&msg, payload->data, sizeof(msg));
2083 msg.session_id = be64toh(msg.session_id);
f7079f67
DG		 2084
2085 /*
7591bab1
MD		 2086 * Iterate over all streams to set the begin data pending flag.
2087 * For now, the streams are indexed by stream handle so we have
2088 * to iterate over all streams to find the one associated with
2089 * the right session_id.
f7079f67
DG		 2090 */
2091 rcu_read_lock();
d3e2ba59 2092 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2a174661 2093 node.node) {
7591bab1
MD		 2094 if (!stream_get(stream)) {
2095 continue;
2096 }
5312a3ed 2097 if (stream->trace->session->id == msg.session_id) {
7591bab1
MD		 2098 pthread_mutex_lock(&stream->lock);
2099 stream->data_pending_check_done = false;
2100 pthread_mutex_unlock(&stream->lock);
f7079f67
DG		 2101 DBG("Set begin data pending flag to stream %" PRIu64,
2102 stream->stream_handle);
2103 }
7591bab1 2104 stream_put(stream);
f7079f67
DG		 2105 }
2106 rcu_read_unlock();
2107
53efb85a 2108 memset(&reply, 0, sizeof(reply));
f7079f67
DG		 2109 /* All good, send back reply. */
2110 reply.ret_code = htobe32(LTTNG_OK);
2111
5312a3ed
JG		 2112 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2113 if (send_ret < (ssize_t) sizeof(reply)) {
2114 ERR("Failed to send \"begin data pending\" command reply (ret = %zd)",
2115 send_ret);
2116 ret = -1;
2117 } else {
2118 ret = 0;
f7079f67
DG		 2119 }
2120
2121end_no_session:
2122 return ret;
2123}
2124
2125/*
7591bab1
MD		 2126 * End data pending command. This will check, for a given session id, if
2127 * each stream associated with it has its data_pending_check_done flag
2128 * set. If not, this means that the client lost track of the stream but
2129 * the data is still being streamed on our side. In this case, we inform
2130 * the client that data is in flight.
f7079f67
DG		 2131 *
2132 * Return to the client if there is data in flight or not with a ret_code.
2133 */
5312a3ed
JG		 2134static int relay_end_data_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2135 struct relay_connection *conn,
2136 const struct lttng_buffer_view *payload)
f7079f67
DG		 2137{
2138 int ret;
5312a3ed 2139 ssize_t send_ret;
f7079f67
DG		 2140 struct lttng_ht_iter iter;
2141 struct lttcomm_relayd_end_data_pending msg;
2142 struct lttcomm_relayd_generic_reply reply;
2143 struct relay_stream *stream;
f7079f67
DG		 2144 uint32_t is_data_inflight = 0;
2145
f7079f67
DG		 2146 DBG("End data pending command");
2147
5312a3ed 2148 if (!conn->session || !conn->version_check_done) {
f7079f67
DG		 2149 ERR("Trying to check for data before version check");
2150 ret = -1;
2151 goto end_no_session;
2152 }
2153
5312a3ed
JG		 2154 if (payload->size < sizeof(msg)) {
2155 ERR("Unexpected payload size in \"relay_end_data_pending\": expected >= %zu bytes, got %zu bytes",
2156 sizeof(msg), payload->size);
f7079f67
DG		 2157 ret = -1;
2158 goto end_no_session;
2159 }
5312a3ed
JG		 2160 memcpy(&msg, payload->data, sizeof(msg));
2161 msg.session_id = be64toh(msg.session_id);
f7079f67 2162
7591bab1
MD		 2163 /*
2164 * Iterate over all streams to see if the begin data pending
2165 * flag is set.
2166 */
f7079f67 2167 rcu_read_lock();
d3e2ba59 2168 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2a174661 2169 node.node) {
7591bab1
MD		 2170 if (!stream_get(stream)) {
2171 continue;
2172 }
5312a3ed 2173 if (stream->trace->session->id != msg.session_id) {
7591bab1
MD		 2174 stream_put(stream);
2175 continue;
2176 }
2177 pthread_mutex_lock(&stream->lock);
2178 if (!stream->data_pending_check_done) {
2179 if (!stream->closed || !(((int64_t) (stream->prev_seq - stream->last_net_seq_num)) >= 0)) {
2180 is_data_inflight = 1;
2181 DBG("Data is still in flight for stream %" PRIu64,
2182 stream->stream_handle);
2183 pthread_mutex_unlock(&stream->lock);
2184 stream_put(stream);
2185 break;
2186 }
f7079f67 2187 }
7591bab1
MD		 2188 pthread_mutex_unlock(&stream->lock);
2189 stream_put(stream);
f7079f67
DG		 2190 }
2191 rcu_read_unlock();
2192
53efb85a 2193 memset(&reply, 0, sizeof(reply));
f7079f67
DG		 2194 /* All good, send back reply. */
2195 reply.ret_code = htobe32(is_data_inflight);
2196
5312a3ed
JG		 2197 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2198 if (send_ret < (ssize_t) sizeof(reply)) {
2199 ERR("Failed to send \"end data pending\" command reply (ret = %zd)",
2200 send_ret);
2201 ret = -1;
2202 } else {
2203 ret = 0;
f7079f67
DG		 2204 }
2205
2206end_no_session:
2207 return ret;
2208}
2209
1c20f0e2
JD		 2210/*
2211 * Receive an index for a specific stream.
2212 *
2213 * Return 0 on success else a negative value.
2214 */
5312a3ed
JG		 2215static int relay_recv_index(const struct lttcomm_relayd_hdr *recv_hdr,
2216 struct relay_connection *conn,
2217 const struct lttng_buffer_view *payload)
1c20f0e2 2218{
5312a3ed
JG		 2219 int ret;
2220 ssize_t send_ret;
58eb9381 2221 struct relay_session *session = conn->session;
1c20f0e2 2222 struct lttcomm_relayd_index index_info;
7591bab1 2223 struct relay_index *index;
1c20f0e2
JD		 2224 struct lttcomm_relayd_generic_reply reply;
2225 struct relay_stream *stream;
f8f3885c 2226 size_t msg_len;
1c20f0e2 2227
58eb9381 2228 assert(conn);
1c20f0e2
JD		 2229
2230 DBG("Relay receiving index");
2231
5312a3ed 2232 if (!session || !conn->version_check_done) {
1c20f0e2
JD		 2233 ERR("Trying to close a stream before version check");
2234 ret = -1;
2235 goto end_no_session;
2236 }
2237
f8f3885c
MD		 2238 msg_len = lttcomm_relayd_index_len(
2239 lttng_to_index_major(conn->major, conn->minor),
2240 lttng_to_index_minor(conn->major, conn->minor));
5312a3ed
JG		 2241 if (payload->size < msg_len) {
2242 ERR("Unexpected payload size in \"relay_recv_index\": expected >= %zu bytes, got %zu bytes",
2243 msg_len, payload->size);
1c20f0e2
JD		 2244 ret = -1;
2245 goto end_no_session;
2246 }
5312a3ed
JG		 2247 memcpy(&index_info, payload->data, msg_len);
2248 index_info.relay_stream_id = be64toh(index_info.relay_stream_id);
2249 index_info.net_seq_num = be64toh(index_info.net_seq_num);
2250 index_info.packet_size = be64toh(index_info.packet_size);
2251 index_info.content_size = be64toh(index_info.content_size);
2252 index_info.timestamp_begin = be64toh(index_info.timestamp_begin);
2253 index_info.timestamp_end = be64toh(index_info.timestamp_end);
2254 index_info.events_discarded = be64toh(index_info.events_discarded);
2255 index_info.stream_id = be64toh(index_info.stream_id);
81df238b
JR		 2256
2257 if (conn->minor >= 8) {
2258 index_info.stream_instance_id =
2259 be64toh(index_info.stream_instance_id);
2260 index_info.packet_seq_num = be64toh(index_info.packet_seq_num);
2261 }
5312a3ed
JG		 2262
2263 stream = stream_get_by_id(index_info.relay_stream_id);
1c20f0e2 2264 if (!stream) {
7591bab1 2265 ERR("stream_get_by_id not found");
1c20f0e2 2266 ret = -1;
7591bab1 2267 goto end;
1c20f0e2 2268 }
7591bab1 2269 pthread_mutex_lock(&stream->lock);
1c20f0e2 2270
d3e2ba59
JD		 2271 /* Live beacon handling */
2272 if (index_info.packet_size == 0) {
7591bab1
MD		 2273 DBG("Received live beacon for stream %" PRIu64,
2274 stream->stream_handle);
d3e2ba59
JD		 2275
2276 /*
7591bab1
MD		 2277 * Only flag a stream inactive when it has already
2278 * received data and no indexes are in flight.
d3e2ba59 2279 */
a44ca2ca 2280 if (stream->index_received_seqcount > 0
7591bab1 2281 && stream->indexes_in_flight == 0) {
5312a3ed 2282 stream->beacon_ts_end = index_info.timestamp_end;
d3e2ba59
JD		 2283 }
2284 ret = 0;
7591bab1 2285 goto end_stream_put;
d3e2ba59
JD		 2286 } else {
2287 stream->beacon_ts_end = -1ULL;
2288 }
2289
528f2ffa 2290 if (stream->ctf_stream_id == -1ULL) {
5312a3ed 2291 stream->ctf_stream_id = index_info.stream_id;
528f2ffa 2292 }
5312a3ed 2293 index = relay_index_get_by_id_or_create(stream, index_info.net_seq_num);
7591bab1
MD		 2294 if (!index) {
2295 ret = -1;
2296 ERR("relay_index_get_by_id_or_create index NULL");
2297 goto end_stream_put;
1c20f0e2 2298 }
234cd636 2299 if (set_index_control_data(index, &index_info, conn)) {
7591bab1
MD		 2300 ERR("set_index_control_data error");
2301 relay_index_put(index);
2302 ret = -1;
2303 goto end_stream_put;
2304 }
2305 ret = relay_index_try_flush(index);
2306 if (ret == 0) {
a44ca2ca
MD		 2307 tracefile_array_commit_seq(stream->tfa);
2308 stream->index_received_seqcount++;
d3ecc550 2309 stream->pos_after_last_complete_data_index += index->total_size;
7591bab1
MD		 2310 } else if (ret > 0) {
2311 /* no flush. */
2312 ret = 0;
2313 } else {
2314 ERR("relay_index_try_flush error %d", ret);
2315 relay_index_put(index);
2316 ret = -1;
1c20f0e2
JD		 2317 }
2318
7591bab1
MD		 2319end_stream_put:
2320 pthread_mutex_unlock(&stream->lock);
2321 stream_put(stream);
2322
2323end:
1c20f0e2 2324
53efb85a 2325 memset(&reply, 0, sizeof(reply));
1c20f0e2
JD		 2326 if (ret < 0) {
2327 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2328 } else {
2329 reply.ret_code = htobe32(LTTNG_OK);
2330 }
58eb9381 2331 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
5312a3ed
JG		 2332 if (send_ret < (ssize_t) sizeof(reply)) {
2333 ERR("Failed to send \"recv index\" command reply (ret = %zd)", send_ret);
2334 ret = -1;
1c20f0e2
JD		 2335 }
2336
2337end_no_session:
2338 return ret;
2339}
2340
a4baae1b
JD		 2341/*
2342 * Receive the streams_sent message.
2343 *
2344 * Return 0 on success else a negative value.
2345 */
5312a3ed
JG		 2346static int relay_streams_sent(const struct lttcomm_relayd_hdr *recv_hdr,
2347 struct relay_connection *conn,
2348 const struct lttng_buffer_view *payload)
a4baae1b 2349{
5312a3ed
JG		 2350 int ret;
2351 ssize_t send_ret;
a4baae1b
JD		 2352 struct lttcomm_relayd_generic_reply reply;
2353
58eb9381 2354 assert(conn);
a4baae1b
JD		 2355
2356 DBG("Relay receiving streams_sent");
2357
5312a3ed 2358 if (!conn->session || !conn->version_check_done) {
a4baae1b
JD		 2359 ERR("Trying to close a stream before version check");
2360 ret = -1;
2361 goto end_no_session;
2362 }
2363
2364 /*
7591bab1
MD		 2365 * Publish every pending stream in the connection recv list which are
2366 * now ready to be used by the viewer.
4a9daf17 2367 */
7591bab1 2368 publish_connection_local_streams(conn);
4a9daf17 2369
53efb85a 2370 memset(&reply, 0, sizeof(reply));
a4baae1b 2371 reply.ret_code = htobe32(LTTNG_OK);
58eb9381 2372 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
5312a3ed
JG		 2373 if (send_ret < (ssize_t) sizeof(reply)) {
2374 ERR("Failed to send \"streams sent\" command reply (ret = %zd)",
2375 send_ret);
2376 ret = -1;
a4baae1b
JD		 2377 } else {
2378 /* Success. */
2379 ret = 0;
2380 }
2381
2382end_no_session:
2383 return ret;
2384}
2385
d3ecc550 2386/*
5312a3ed 2387 * relay_rotate_session_stream: rotate a stream to a new tracefile for the session
d3ecc550
JD		 2388 * rotation feature (not the tracefile rotation feature).
2389 */
5312a3ed
JG		 2390static int relay_rotate_session_stream(const struct lttcomm_relayd_hdr *recv_hdr,
2391 struct relay_connection *conn,
2392 const struct lttng_buffer_view *payload)
d3ecc550 2393{
5312a3ed
JG		 2394 int ret;
2395 ssize_t send_ret;
d3ecc550
JD		 2396 struct relay_session *session = conn->session;
2397 struct lttcomm_relayd_rotate_stream stream_info;
2398 struct lttcomm_relayd_generic_reply reply;
2399 struct relay_stream *stream;
5312a3ed
JG		 2400 size_t header_len;
2401 size_t path_len;
2402 struct lttng_buffer_view new_path_view;
d3ecc550
JD		 2403
2404 DBG("Rotate stream received");
2405
2406 if (!session || !conn->version_check_done) {
2407 ERR("Trying to rotate a stream before version check");
2408 ret = -1;
2409 goto end_no_reply;
2410 }
2411
2412 if (session->major == 2 && session->minor < 11) {
2413 ERR("Unsupported feature before 2.11");
2414 ret = -1;
2415 goto end_no_reply;
2416 }
2417
5312a3ed 2418 header_len = sizeof(struct lttcomm_relayd_rotate_stream);
d3ecc550 2419
5312a3ed
JG		 2420 if (payload->size < header_len) {
2421 ERR("Unexpected payload size in \"relay_rotate_session_stream\": expected >= %zu bytes, got %zu bytes",
2422 header_len, payload->size);
d3ecc550
JD		 2423 ret = -1;
2424 goto end_no_reply;
2425 }
2426
5312a3ed
JG		 2427 memcpy(&stream_info, payload->data, header_len);
2428
2429 /* Convert to host */
2430 stream_info.pathname_length = be32toh(stream_info.pathname_length);
2431 stream_info.stream_id = be64toh(stream_info.stream_id);
2432 stream_info.new_chunk_id = be64toh(stream_info.new_chunk_id);
2433 stream_info.rotate_at_seq_num = be64toh(stream_info.rotate_at_seq_num);
d3ecc550 2434
5312a3ed
JG		 2435 path_len = stream_info.pathname_length;
2436 if (payload->size < header_len + path_len) {
2437 ERR("Unexpected payload size in \"relay_rotate_session_stream\" including path: expected >= %zu bytes, got %zu bytes",
2438 header_len + path_len, payload->size);
2439 ret = -1;
2440 goto end_no_reply;
2441 }
2442
d3ecc550 2443 /* Ensure it fits in local filename length. */
5312a3ed 2444 if (path_len >= LTTNG_PATH_MAX) {
d3ecc550
JD		 2445 ret = -ENAMETOOLONG;
2446 ERR("Length of relay_rotate_session_stream command's path name (%zu bytes) exceeds the maximal allowed length of %i bytes",
5312a3ed 2447 path_len, LTTNG_PATH_MAX);
d3ecc550
JD		 2448 goto end;
2449 }
2450
5312a3ed
JG		 2451 new_path_view = lttng_buffer_view_from_view(payload, header_len,
2452 stream_info.pathname_length);
d3ecc550 2453
5312a3ed
JG		 2454 stream = stream_get_by_id(stream_info.stream_id);
2455 if (!stream) {
d3ecc550 2456 ret = -1;
5312a3ed 2457 goto end;
d3ecc550
JD		 2458 }
2459
2460 pthread_mutex_lock(&stream->lock);
2461
2462 /*
2463 * Update the trace path (just the folder, the stream name does not
2464 * change).
2465 */
2466 free(stream->path_name);
5312a3ed 2467 stream->path_name = create_output_path(new_path_view.data);
d3ecc550
JD		 2468 if (!stream->path_name) {
2469 ERR("Failed to create a new output path");
5312a3ed 2470 ret = -1;
d3ecc550
JD		 2471 goto end_stream_unlock;
2472 }
2473 ret = utils_mkdir_recursive(stream->path_name, S_IRWXU | S_IRWXG,
2474 -1, -1);
2475 if (ret < 0) {
2476 ERR("relay creating output directory");
5312a3ed 2477 ret = -1;
d3ecc550
JD		 2478 goto end_stream_unlock;
2479 }
5312a3ed
JG		 2480
2481 stream->chunk_id = stream_info.new_chunk_id;
d3ecc550
JD		 2482
2483 if (stream->is_metadata) {
2484 /*
2485 * The metadata stream is sent only over the control connection
2486 * so we know we have all the data to perform the stream
2487 * rotation.
2488 */
2489 ret = do_rotate_stream(stream);
2490 } else {
5312a3ed 2491 stream->rotate_at_seq_num = stream_info.rotate_at_seq_num;
d3ecc550
JD		 2492 ret = try_rotate_stream(stream);
2493 }
2494 if (ret < 0) {
2495 goto end_stream_unlock;
2496 }
2497
2498end_stream_unlock:
2499 pthread_mutex_unlock(&stream->lock);
2500 stream_put(stream);
2501end:
2502 memset(&reply, 0, sizeof(reply));
2503 if (ret < 0) {
2504 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2505 } else {
2506 reply.ret_code = htobe32(LTTNG_OK);
2507 }
2508 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
2509 sizeof(struct lttcomm_relayd_generic_reply), 0);
5312a3ed
JG		 2510 if (send_ret < (ssize_t) sizeof(reply)) {
2511 ERR("Failed to send \"rotate session stream\" command reply (ret = %zd)",
2512 send_ret);
2513 ret = -1;
d3ecc550
JD		 2514 }
2515
2516end_no_reply:
d3ecc550
JD		 2517 return ret;
2518}
2519
a1ae2ea5
JD		 2520/*
2521 * relay_mkdir: Create a folder on the disk.
2522 */
5312a3ed
JG		 2523static int relay_mkdir(const struct lttcomm_relayd_hdr *recv_hdr,
2524 struct relay_connection *conn,
2525 const struct lttng_buffer_view *payload)
a1ae2ea5
JD		 2526{
2527 int ret;
a1ae2ea5
JD		 2528 struct relay_session *session = conn->session;
2529 struct lttcomm_relayd_mkdir path_info_header;
a1ae2ea5
JD		 2530 struct lttcomm_relayd_generic_reply reply;
2531 char *path = NULL;
5312a3ed
JG		 2532 size_t header_len;
2533 ssize_t send_ret;
2534 struct lttng_buffer_view path_view;
a1ae2ea5
JD		 2535
2536 if (!session || !conn->version_check_done) {
00fb02ac 2537 ERR("Trying to create a directory before version check");
a1ae2ea5
JD		 2538 ret = -1;
2539 goto end_no_session;
2540 }
2541
2542 if (session->major == 2 && session->minor < 11) {
2543 /*
2544 * This client is not supposed to use this command since
2545 * it predates its introduction.
2546 */
2547 ERR("relay_mkdir command is unsupported before LTTng 2.11");
2548 ret = -1;
2549 goto end_no_session;
2550 }
2551
5312a3ed
JG		 2552 header_len = sizeof(path_info_header);
2553 if (payload->size < header_len) {
2554 ERR("Unexpected payload size in \"relay_mkdir\": expected >= %zu bytes, got %zu bytes",
2555 header_len, payload->size);
a1ae2ea5
JD		 2556 ret = -1;
2557 goto end_no_session;
2558 }
2559
5312a3ed
JG		 2560 memcpy(&path_info_header, payload->data, header_len);
2561
a1ae2ea5
JD		 2562 path_info_header.length = be32toh(path_info_header.length);
2563
5312a3ed
JG		 2564 if (payload->size < header_len + path_info_header.length) {
2565 ERR("Unexpected payload size in \"relay_mkdir\" including path: expected >= %zu bytes, got %zu bytes",
2566 header_len + path_info_header.length, payload->size);
2567 ret = -1;
2568 goto end_no_session;
2569 }
2570
a1ae2ea5
JD		 2571 /* Ensure that it fits in local path length. */
2572 if (path_info_header.length >= LTTNG_PATH_MAX) {
2573 ret = -ENAMETOOLONG;
2574 ERR("Path name argument of mkdir command (%" PRIu32 " bytes) exceeds the maximal length allowed (%d bytes)",
2575 path_info_header.length, LTTNG_PATH_MAX);
2576 goto end;
2577 }
2578
5312a3ed
JG		 2579 path_view = lttng_buffer_view_from_view(payload, header_len,
2580 path_info_header.length);
a1ae2ea5 2581
5312a3ed 2582 path = create_output_path(path_view.data);
a1ae2ea5
JD		 2583 if (!path) {
2584 ERR("Failed to create output path");
2585 ret = -1;
2586 goto end;
2587 }
2588
2589 ret = utils_mkdir_recursive(path, S_IRWXU | S_IRWXG, -1, -1);
2590 if (ret < 0) {
2591 ERR("relay creating output directory");
2592 goto end;
2593 }
2594
2595 ret = 0;
2596
2597end:
2598 memset(&reply, 0, sizeof(reply));
2599 if (ret < 0) {
2600 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2601 } else {
2602 reply.ret_code = htobe32(LTTNG_OK);
2603 }
5312a3ed
JG		 2604 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply, sizeof(reply), 0);
2605 if (send_ret < (ssize_t) sizeof(reply)) {
2606 ERR("Failed to send \"mkdir\" command reply (ret = %zd)", send_ret);
a1ae2ea5
JD		 2607 ret = -1;
2608 }
2609
2610end_no_session:
2611 free(path);
a1ae2ea5
JD		 2612 return ret;
2613}
2614
00fb02ac
JD		 2615static int validate_rotate_rename_path_length(const char *path_type,
2616 uint32_t path_length)
2617{
2618 int ret = 0;
2619
2620 if (path_length > LTTNG_PATH_MAX) {
2621 ret = -ENAMETOOLONG;
2622 ERR("rotate rename \"%s\" path name length (%" PRIu32 " bytes) exceeds the allowed size of %i bytes",
2623 path_type, path_length, LTTNG_PATH_MAX);
2624 } else if (path_length == 0) {
2625 ret = -EINVAL;
2626 ERR("rotate rename \"%s\" path name has an illegal length of 0", path_type);
2627 }
2628 return ret;
2629}
2630
2631/*
2632 * relay_rotate_rename: rename the trace folder after a rotation is
2633 * completed. We are not closing any fd here, just moving the folder, so it
2634 * works even if data is still in-flight.
2635 */
5312a3ed
JG		 2636static int relay_rotate_rename(const struct lttcomm_relayd_hdr *recv_hdr,
2637 struct relay_connection *conn,
2638 const struct lttng_buffer_view *payload)
00fb02ac
JD		 2639{
2640 int ret;
5312a3ed 2641 ssize_t send_ret;
00fb02ac
JD		 2642 struct relay_session *session = conn->session;
2643 struct lttcomm_relayd_generic_reply reply;
2644 struct lttcomm_relayd_rotate_rename header;
5312a3ed 2645 size_t header_len;
00fb02ac 2646 size_t received_paths_size;
00fb02ac 2647 char *complete_old_path = NULL, *complete_new_path = NULL;
5312a3ed
JG		 2648 struct lttng_buffer_view old_path_view;
2649 struct lttng_buffer_view new_path_view;
00fb02ac
JD		 2650
2651 if (!session || !conn->version_check_done) {
2652 ERR("Trying to rename a trace folder before version check");
2653 ret = -1;
2654 goto end_no_reply;
2655 }
2656
2657 if (session->major == 2 && session->minor < 11) {
2658 ERR("relay_rotate_rename command is unsupported before LTTng 2.11");
2659 ret = -1;
2660 goto end_no_reply;
2661 }
2662
5312a3ed
JG		 2663 header_len = sizeof(header);
2664 if (payload->size < header_len) {
2665 ERR("Unexpected payload size in \"relay_rotate_rename\": expected >= %zu bytes, got %zu bytes",
2666 header_len, payload->size);
00fb02ac
JD		 2667 ret = -1;
2668 goto end_no_reply;
2669 }
2670
5312a3ed
JG		 2671 memcpy(&header, payload->data, header_len);
2672
00fb02ac
JD		 2673 header.old_path_length = be32toh(header.old_path_length);
2674 header.new_path_length = be32toh(header.new_path_length);
2675 received_paths_size = header.old_path_length + header.new_path_length;
2676
5312a3ed
JG		 2677 if (payload->size < header_len + received_paths_size) {
2678 ERR("Unexpected payload size in \"relay_rotate_rename\" including paths: expected >= %zu bytes, got %zu bytes",
2679 header_len, payload->size);
2680 ret = -1;
2681 goto end_no_reply;
2682 }
2683
00fb02ac
JD		 2684 /* Ensure the paths don't exceed their allowed size. */
2685 ret = validate_rotate_rename_path_length("old", header.old_path_length);
2686 if (ret) {
2687 goto end;
2688 }
2689 ret = validate_rotate_rename_path_length("new", header.new_path_length);
2690 if (ret) {
2691 goto end;
2692 }
2693
5312a3ed
JG		 2694 old_path_view = lttng_buffer_view_from_view(payload, header_len,
2695 header.old_path_length);
2696 new_path_view = lttng_buffer_view_from_view(payload,
2697 header_len + header.old_path_length,
2698 header.new_path_length);
00fb02ac
JD		 2699
2700 /* Validate that both paths received are NULL terminated. */
5312a3ed 2701 if (old_path_view.data[old_path_view.size - 1] != '\0') {
00fb02ac
JD		 2702 ERR("relay_rotate_rename command's \"old\" path is invalid (not NULL terminated)");
2703 ret = -1;
2704 goto end;
2705 }
5312a3ed 2706 if (new_path_view.data[new_path_view.size - 1] != '\0') {
00fb02ac
JD		 2707 ERR("relay_rotate_rename command's \"new\" path is invalid (not NULL terminated)");
2708 ret = -1;
2709 goto end;
2710 }
2711
5312a3ed 2712 complete_old_path = create_output_path(old_path_view.data);
00fb02ac
JD		 2713 if (!complete_old_path) {
2714 ERR("Failed to build old output path in rotate_rename command");
2715 ret = -1;
2716 goto end;
2717 }
2718
5312a3ed 2719 complete_new_path = create_output_path(new_path_view.data);
00fb02ac
JD		 2720 if (!complete_new_path) {
2721 ERR("Failed to build new output path in rotate_rename command");
2722 ret = -1;
2723 goto end;
2724 }
2725
2726 ret = utils_mkdir_recursive(complete_new_path, S_IRWXU | S_IRWXG,
2727 -1, -1);
2728 if (ret < 0) {
2729 ERR("Failed to mkdir() rotate_rename's \"new\" output directory at \"%s\"",
2730 complete_new_path);
2731 goto end;
2732 }
2733
2734 /*
2735 * If a domain has not yet created its channel, the domain-specific
2736 * folder might not exist, but this is not an error.
2737 */
2738 ret = rename(complete_old_path, complete_new_path);
2739 if (ret < 0 && errno != ENOENT) {
2740 PERROR("Renaming chunk in rotate_rename command from \"%s\" to \"%s\"",
2741 complete_old_path, complete_new_path);
2742 goto end;
2743 }
2744 ret = 0;
2745
2746end:
2747 memset(&reply, 0, sizeof(reply));
2748 if (ret < 0) {
2749 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2750 } else {
2751 reply.ret_code = htobe32(LTTNG_OK);
2752 }
5312a3ed
JG		 2753 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
2754 sizeof(reply), 0);
2755 if (send_ret < sizeof(reply)) {
2756 ERR("Failed to send \"rotate rename\" command reply (ret = %zd)",
2757 send_ret);
00fb02ac
JD		 2758 ret = -1;
2759 }
2760
2761end_no_reply:
00fb02ac
JD		 2762 free(complete_old_path);
2763 free(complete_new_path);
2764 return ret;
2765}
2766
e2acc8d2
JD		 2767/*
2768 * Check if all the streams in the session have completed the last rotation.
2769 * The chunk_id value is used to distinguish the cases where a stream was
2770 * closed on the consumerd before the rotation started but it still active on
2771 * the relayd, and the case where a stream appeared on the consumerd/relayd
2772 * after the last rotation started (in that case, it is already writing in the
2773 * new chunk folder).
2774 */
2775static
5312a3ed
JG		 2776int relay_rotate_pending(const struct lttcomm_relayd_hdr *recv_hdr,
2777 struct relay_connection *conn,
2778 const struct lttng_buffer_view *payload)
e2acc8d2
JD		 2779{
2780 struct relay_session *session = conn->session;
2781 struct lttcomm_relayd_rotate_pending msg;
d88744a4 2782 struct lttcomm_relayd_rotate_pending_reply reply;
e2acc8d2
JD		 2783 struct lttng_ht_iter iter;
2784 struct relay_stream *stream;
4f5fb4c3 2785 int ret = 0;
5312a3ed 2786 ssize_t send_ret;
e2acc8d2
JD		 2787 uint64_t chunk_id;
2788 bool rotate_pending = false;
2789
2790 DBG("Rotate pending command received");
2791
2792 if (!session || !conn->version_check_done) {
2793 ERR("Trying to check for data before version check");
2794 ret = -1;
2795 goto end_no_reply;
2796 }
2797
2798 if (session->major == 2 && session->minor < 11) {
2799 ERR("Unsupported feature before 2.11");
2800 ret = -1;
2801 goto end_no_reply;
2802 }
2803
5312a3ed
JG		 2804 if (payload->size < sizeof(msg)) {
2805 ERR("Unexpected payload size in \"relay_rotate_pending\": expected >= %zu bytes, got %zu bytes",
2806 sizeof(msg), payload->size);
e2acc8d2
JD		 2807 ret = -1;
2808 goto end_no_reply;
2809 }
2810
5312a3ed
JG		 2811 memcpy(&msg, payload->data, sizeof(msg));
2812
e2acc8d2 2813 chunk_id = be64toh(msg.chunk_id);
5312a3ed 2814
e2acc8d2
JD		 2815 DBG("Evaluating rotate pending for chunk id %" PRIu64, chunk_id);
2816
2817 /*
2818 * Iterate over all the streams in the session and check if they are
2819 * still waiting for data to perform their rotation.
2820 */
2821 rcu_read_lock();
2822 cds_lfht_for_each_entry(relay_streams_ht->ht, &iter.iter, stream,
2823 node.node) {
2824 if (!stream_get(stream)) {
2825 continue;
2826 }
2827 if (stream->trace->session != session) {
2828 stream_put(stream);
2829 continue;
2830 }
2831 pthread_mutex_lock(&stream->lock);
2832 if (stream->rotate_at_seq_num != -1ULL) {
2833 /* We have not yet performed the rotation. */
2834 rotate_pending = true;
2835 DBG("Stream %" PRIu64 " is still rotating",
2836 stream->stream_handle);
2837 } else if (stream->chunk_id < chunk_id) {
2838 /*
2839 * Stream closed on the consumer but still active on the
2840 * relay.
2841 */
2842 rotate_pending = true;
2843 DBG("Stream %" PRIu64 " did not exist on the consumer "
2844 "when the last rotation started, but is"
2845 "still waiting for data before getting"
2846 "closed",
2847 stream->stream_handle);
2848 }
2849 pthread_mutex_unlock(&stream->lock);
2850 stream_put(stream);
2851 if (rotate_pending) {
2852 goto send_reply;
2853 }
2854 }
2855
2856send_reply:
2857 rcu_read_unlock();
2858 memset(&reply, 0, sizeof(reply));
d88744a4
JD		 2859 reply.generic.ret_code = htobe32((uint32_t) LTTNG_OK);
2860 reply.is_pending = (uint8_t) !!rotate_pending;
5312a3ed 2861 send_ret = conn->sock->ops->sendmsg(conn->sock, &reply,
e2acc8d2 2862 sizeof(reply), 0);
5312a3ed
JG		 2863 if (send_ret < (ssize_t) sizeof(reply)) {
2864 ERR("Failed to send \"rotate pending\" command reply (ret = %zd)",
2865 send_ret);
e2acc8d2
JD		 2866 ret = -1;
2867 }
2868
2869end_no_reply:
2870 return ret;
2871}
2872
5312a3ed
JG		 2873#define DBG_CMD(cmd_name, conn) \
2874 DBG3("Processing \"%s\" command for socket %i", cmd_name, conn->sock->fd);
2875
2876static int relay_process_control_command(struct relay_connection *conn,
2877 const struct lttcomm_relayd_hdr *header,
2878 const struct lttng_buffer_view *payload)
b8aa1682
JD		 2879{
2880 int ret = 0;
2881
5312a3ed 2882 switch (header->cmd) {
b8aa1682 2883 case RELAYD_CREATE_SESSION:
5312a3ed
JG		 2884 DBG_CMD("RELAYD_CREATE_SESSION", conn);
2885 ret = relay_create_session(header, conn, payload);
b8aa1682 2886 break;
b8aa1682 2887 case RELAYD_ADD_STREAM:
5312a3ed
JG		 2888 DBG_CMD("RELAYD_ADD_STREAM", conn);
2889 ret = relay_add_stream(header, conn, payload);
b8aa1682
JD		 2890 break;
2891 case RELAYD_START_DATA:
5312a3ed
JG		 2892 DBG_CMD("RELAYD_START_DATA", conn);
2893 ret = relay_start(header, conn, payload);
b8aa1682
JD		 2894 break;
2895 case RELAYD_SEND_METADATA:
5312a3ed
JG		 2896 DBG_CMD("RELAYD_SEND_METADATA", conn);
2897 ret = relay_recv_metadata(header, conn, payload);
b8aa1682
JD		 2898 break;
2899 case RELAYD_VERSION:
5312a3ed
JG		 2900 DBG_CMD("RELAYD_VERSION", conn);
2901 ret = relay_send_version(header, conn, payload);
b8aa1682 2902 break;
173af62f 2903 case RELAYD_CLOSE_STREAM:
5312a3ed
JG		 2904 DBG_CMD("RELAYD_CLOSE_STREAM", conn);
2905 ret = relay_close_stream(header, conn, payload);
173af62f 2906 break;
6d805429 2907 case RELAYD_DATA_PENDING:
5312a3ed
JG		 2908 DBG_CMD("RELAYD_DATA_PENDING", conn);
2909 ret = relay_data_pending(header, conn, payload);
c8f59ee5
DG		 2910 break;
2911 case RELAYD_QUIESCENT_CONTROL:
5312a3ed
JG		 2912 DBG_CMD("RELAYD_QUIESCENT_CONTROL", conn);
2913 ret = relay_quiescent_control(header, conn, payload);
c8f59ee5 2914 break;
f7079f67 2915 case RELAYD_BEGIN_DATA_PENDING:
5312a3ed
JG		 2916 DBG_CMD("RELAYD_BEGIN_DATA_PENDING", conn);
2917 ret = relay_begin_data_pending(header, conn, payload);
f7079f67
DG		 2918 break;
2919 case RELAYD_END_DATA_PENDING:
5312a3ed
JG		 2920 DBG_CMD("RELAYD_END_DATA_PENDING", conn);
2921 ret = relay_end_data_pending(header, conn, payload);
f7079f67 2922 break;
1c20f0e2 2923 case RELAYD_SEND_INDEX:
5312a3ed
JG		 2924 DBG_CMD("RELAYD_SEND_INDEX", conn);
2925 ret = relay_recv_index(header, conn, payload);
1c20f0e2 2926 break;
a4baae1b 2927 case RELAYD_STREAMS_SENT:
5312a3ed
JG		 2928 DBG_CMD("RELAYD_STREAMS_SENT", conn);
2929 ret = relay_streams_sent(header, conn, payload);
a4baae1b 2930 break;
93ec662e 2931 case RELAYD_RESET_METADATA:
5312a3ed
JG		 2932 DBG_CMD("RELAYD_RESET_METADATA", conn);
2933 ret = relay_reset_metadata(header, conn, payload);
93ec662e 2934 break;
d3ecc550 2935 case RELAYD_ROTATE_STREAM:
5312a3ed
JG		 2936 DBG_CMD("RELAYD_ROTATE_STREAM", conn);
2937 ret = relay_rotate_session_stream(header, conn, payload);
d3ecc550 2938 break;
00fb02ac 2939 case RELAYD_ROTATE_RENAME:
5312a3ed
JG		 2940 DBG_CMD("RELAYD_ROTATE_RENAME", conn);
2941 ret = relay_rotate_rename(header, conn, payload);
00fb02ac 2942 break;
e2acc8d2 2943 case RELAYD_ROTATE_PENDING:
5312a3ed
JG		 2944 DBG_CMD("RELAYD_ROTATE_PENDING", conn);
2945 ret = relay_rotate_pending(header, conn, payload);
e2acc8d2 2946 break;
a1ae2ea5 2947 case RELAYD_MKDIR:
5312a3ed
JG		 2948 DBG_CMD("RELAYD_MKDIR", conn);
2949 ret = relay_mkdir(header, conn, payload);
a1ae2ea5 2950 break;
b8aa1682
JD		 2951 case RELAYD_UPDATE_SYNC_INFO:
2952 default:
5312a3ed 2953 ERR("Received unknown command (%u)", header->cmd);
58eb9381 2954 relay_unknown_command(conn);
b8aa1682
JD		 2955 ret = -1;
2956 goto end;
2957 }
2958
2959end:
2960 return ret;
2961}
2962
5569b118
JG		 2963static enum relay_connection_status relay_process_control_receive_payload(
2964 struct relay_connection *conn)
5312a3ed
JG		 2965{
2966 int ret = 0;
5569b118 2967 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
5312a3ed
JG		 2968 struct lttng_dynamic_buffer *reception_buffer =
2969 &conn->protocol.ctrl.reception_buffer;
2970 struct ctrl_connection_state_receive_payload *state =
2971 &conn->protocol.ctrl.state.receive_payload;
2972 struct lttng_buffer_view payload_view;
2973
2974 if (state->left_to_receive == 0) {
2975 /* Short-circuit for payload-less commands. */
2976 goto reception_complete;
2977 }
2978
2979 ret = conn->sock->ops->recvmsg(conn->sock,
2980 reception_buffer->data + state->received,
2981 state->left_to_receive, MSG_DONTWAIT);
2982 if (ret < 0) {
5569b118
JG		 2983 if (errno != EAGAIN && errno != EWOULDBLOCK) {
2984 PERROR("Unable to receive command payload on sock %d",
2985 conn->sock->fd);
2986 status = RELAY_CONNECTION_STATUS_ERROR;
2987 }
5312a3ed
JG		 2988 goto end;
2989 } else if (ret == 0) {
2990 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
5569b118 2991 status = RELAY_CONNECTION_STATUS_CLOSED;
5312a3ed
JG		 2992 goto end;
2993 }
2994
2995 assert(ret > 0);
2996 assert(ret <= state->left_to_receive);
2997
2998 state->left_to_receive -= ret;
2999 state->received += ret;
3000
3001 if (state->left_to_receive > 0) {
3002 /*
3003 * Can't transition to the protocol's next state, wait to
3004 * receive the rest of the header.
3005 */
3006 DBG3("Partial reception of control connection protocol payload (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3007 state->received, state->left_to_receive,
3008 conn->sock->fd);
5312a3ed
JG		 3009 goto end;
3010 }
3011
3012reception_complete:
3013 DBG("Done receiving control command payload: fd = %i, payload size = %" PRIu64 " bytes",
3014 conn->sock->fd, state->received);
3015 /*
3016 * The payload required to process the command has been received.
3017 * A view to the reception buffer is forwarded to the various
3018 * commands and the state of the control is reset on success.
3019 *
3020 * Commands are responsible for sending their reply to the peer.
3021 */
3022 payload_view = lttng_buffer_view_from_dynamic_buffer(reception_buffer,
3023 0, -1);
3024 ret = relay_process_control_command(conn,
3025 &state->header, &payload_view);
3026 if (ret < 0) {
5569b118 3027 status = RELAY_CONNECTION_STATUS_ERROR;
5312a3ed
JG		 3028 goto end;
3029 }
3030
3031 ret = connection_reset_protocol_state(conn);
5569b118
JG		 3032 if (ret) {
3033 status = RELAY_CONNECTION_STATUS_ERROR;
3034 }
5312a3ed 3035end:
5569b118 3036 return status;
5312a3ed
JG		 3037}
3038
5569b118
JG		 3039static enum relay_connection_status relay_process_control_receive_header(
3040 struct relay_connection *conn)
5312a3ed
JG		 3041{
3042 int ret = 0;
5569b118 3043 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
5312a3ed
JG		 3044 struct lttcomm_relayd_hdr header;
3045 struct lttng_dynamic_buffer *reception_buffer =
3046 &conn->protocol.ctrl.reception_buffer;
3047 struct ctrl_connection_state_receive_header *state =
3048 &conn->protocol.ctrl.state.receive_header;
3049
3050 assert(state->left_to_receive != 0);
3051
3052 ret = conn->sock->ops->recvmsg(conn->sock,
3053 reception_buffer->data + state->received,
3054 state->left_to_receive, MSG_DONTWAIT);
3055 if (ret < 0) {
5569b118
JG		 3056 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3057 PERROR("Unable to receive control command header on sock %d",
3058 conn->sock->fd);
3059 status = RELAY_CONNECTION_STATUS_ERROR;
3060 }
5312a3ed
JG		 3061 goto end;
3062 } else if (ret == 0) {
3063 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
5569b118 3064 status = RELAY_CONNECTION_STATUS_CLOSED;
5312a3ed
JG		 3065 goto end;
3066 }
3067
3068 assert(ret > 0);
3069 assert(ret <= state->left_to_receive);
3070
3071 state->left_to_receive -= ret;
3072 state->received += ret;
3073
3074 if (state->left_to_receive > 0) {
3075 /*
3076 * Can't transition to the protocol's next state, wait to
3077 * receive the rest of the header.
3078 */
3079 DBG3("Partial reception of control connection protocol header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3080 state->received, state->left_to_receive,
3081 conn->sock->fd);
5312a3ed
JG		 3082 goto end;
3083 }
3084
3085 /* Transition to next state: receiving the command's payload. */
3086 conn->protocol.ctrl.state_id =
3087 CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD;
3088 memcpy(&header, reception_buffer->data, sizeof(header));
3089 header.circuit_id = be64toh(header.circuit_id);
3090 header.data_size = be64toh(header.data_size);
3091 header.cmd = be32toh(header.cmd);
3092 header.cmd_version = be32toh(header.cmd_version);
3093 memcpy(&conn->protocol.ctrl.state.receive_payload.header,
3094 &header, sizeof(header));
3095
3096 DBG("Done receiving control command header: fd = %i, cmd = %" PRIu32 ", cmd_version = %" PRIu32 ", payload size = %" PRIu64 " bytes",
3097 conn->sock->fd, header.cmd, header.cmd_version,
3098 header.data_size);
3099
715e6fb1 3100 if (header.data_size > DEFAULT_NETWORK_RELAYD_CTRL_MAX_PAYLOAD_SIZE) {
5312a3ed
JG		 3101 ERR("Command header indicates a payload (%" PRIu64 " bytes) that exceeds the maximal payload size allowed on a control connection.",
3102 header.data_size);
5569b118 3103 status = RELAY_CONNECTION_STATUS_ERROR;
5312a3ed
JG		 3104 goto end;
3105 }
3106
3107 conn->protocol.ctrl.state.receive_payload.left_to_receive =
3108 header.data_size;
3109 conn->protocol.ctrl.state.receive_payload.received = 0;
3110 ret = lttng_dynamic_buffer_set_size(reception_buffer,
3111 header.data_size);
3112 if (ret) {
5569b118 3113 status = RELAY_CONNECTION_STATUS_ERROR;
5312a3ed
JG		 3114 goto end;
3115 }
3116
3117 if (header.data_size == 0) {
3118 /*
3119 * Manually invoke the next state as the poll loop
3120 * will not wake-up to allow us to proceed further.
3121 */
5569b118 3122 status = relay_process_control_receive_payload(conn);
5312a3ed
JG		 3123 }
3124end:
5569b118 3125 return status;
5312a3ed
JG		 3126}
3127
3128/*
3129 * Process the commands received on the control socket
3130 */
5569b118
JG		 3131static enum relay_connection_status relay_process_control(
3132 struct relay_connection *conn)
5312a3ed 3133{
5569b118 3134 enum relay_connection_status status;
5312a3ed
JG		 3135
3136 switch (conn->protocol.ctrl.state_id) {
3137 case CTRL_CONNECTION_STATE_RECEIVE_HEADER:
5569b118 3138 status = relay_process_control_receive_header(conn);
5312a3ed
JG		 3139 break;
3140 case CTRL_CONNECTION_STATE_RECEIVE_PAYLOAD:
5569b118 3141 status = relay_process_control_receive_payload(conn);
5312a3ed
JG		 3142 break;
3143 default:
3144 ERR("Unknown control connection protocol state encountered.");
3145 abort();
3146 }
3147
5569b118 3148 return status;
5312a3ed
JG		 3149}
3150
7d2f7452
DG		 3151/*
3152 * Handle index for a data stream.
3153 *
7591bab1 3154 * Called with the stream lock held.
7d2f7452
DG		 3155 *
3156 * Return 0 on success else a negative value.
3157 */
3158static int handle_index_data(struct relay_stream *stream, uint64_t net_seq_num,
5312a3ed 3159 bool rotate_index, bool *flushed, uint64_t total_size)
7d2f7452 3160{
7591bab1
MD		 3161 int ret = 0;
3162 uint64_t data_offset;
3163 struct relay_index *index;
7d2f7452 3164
7d2f7452
DG		 3165 /* Get data offset because we are about to update the index. */
3166 data_offset = htobe64(stream->tracefile_size_current);
3167
65d66b19
MD		 3168 DBG("handle_index_data: stream %" PRIu64 " net_seq_num %" PRIu64 " data offset %" PRIu64,
3169 stream->stream_handle, net_seq_num, stream->tracefile_size_current);
7591bab1 3170
7d2f7452 3171 /*
7591bab1
MD		 3172 * Lookup for an existing index for that stream id/sequence
3173 * number. If it exists, the control thread has already received the
3174 * data for it, thus we need to write it to disk.
7d2f7452 3175 */
7591bab1 3176 index = relay_index_get_by_id_or_create(stream, net_seq_num);
7d2f7452 3177 if (!index) {
7591bab1
MD		 3178 ret = -1;
3179 goto end;
7d2f7452
DG		 3180 }
3181
f8f3885c 3182 if (rotate_index || !stream->index_file) {
d3ecc550
JD		 3183 ret = create_rotate_index_file(stream);
3184 if (ret < 0) {
3185 ERR("Failed to rotate index");
7591bab1
MD		 3186 /* Put self-ref for this index due to error. */
3187 relay_index_put(index);
f8f3885c 3188 index = NULL;
7591bab1 3189 goto end;
7d2f7452 3190 }
7d2f7452
DG		 3191 }
3192
f8f3885c 3193 if (relay_index_set_file(index, stream->index_file, data_offset)) {
7591bab1
MD		 3194 ret = -1;
3195 /* Put self-ref for this index due to error. */
3196 relay_index_put(index);
f8f3885c 3197 index = NULL;
7591bab1 3198 goto end;
7d2f7452
DG		 3199 }
3200
7591bab1
MD		 3201 ret = relay_index_try_flush(index);
3202 if (ret == 0) {
a44ca2ca
MD		 3203 tracefile_array_commit_seq(stream->tfa);
3204 stream->index_received_seqcount++;
d3ecc550 3205 *flushed = true;
7591bab1 3206 } else if (ret > 0) {
d3ecc550 3207 index->total_size = total_size;
7591bab1
MD		 3208 /* No flush. */
3209 ret = 0;
3210 } else {
3211 /* Put self-ref for this index due to error. */
3212 relay_index_put(index);
f8f3885c 3213 index = NULL;
7591bab1
MD		 3214 ret = -1;
3215 }
3216end:
7d2f7452
DG		 3217 return ret;
3218}
3219
5569b118
JG		 3220static enum relay_connection_status relay_process_data_receive_header(
3221 struct relay_connection *conn)
b8aa1682 3222{
5312a3ed 3223 int ret;
5569b118 3224 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
5312a3ed
JG		 3225 struct data_connection_state_receive_header *state =
3226 &conn->protocol.data.state.receive_header;
3227 struct lttcomm_relayd_data_hdr header;
b8aa1682 3228 struct relay_stream *stream;
5312a3ed
JG		 3229
3230 assert(state->left_to_receive != 0);
3231
3232 ret = conn->sock->ops->recvmsg(conn->sock,
3233 state->header_reception_buffer + state->received,
3234 state->left_to_receive, MSG_DONTWAIT);
3235 if (ret < 0) {
5569b118
JG		 3236 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3237 PERROR("Unable to receive data header on sock %d", conn->sock->fd);
3238 status = RELAY_CONNECTION_STATUS_ERROR;
3239 }
5312a3ed
JG		 3240 goto end;
3241 } else if (ret == 0) {
3242 /* Orderly shutdown. Not necessary to print an error. */
3243 DBG("Socket %d performed an orderly shutdown (received EOF)", conn->sock->fd);
5569b118 3244 status = RELAY_CONNECTION_STATUS_CLOSED;
b8aa1682
JD		 3245 goto end;
3246 }
3247
5312a3ed
JG		 3248 assert(ret > 0);
3249 assert(ret <= state->left_to_receive);
3250
3251 state->left_to_receive -= ret;
3252 state->received += ret;
3253
3254 if (state->left_to_receive > 0) {
3255 /*
3256 * Can't transition to the protocol's next state, wait to
3257 * receive the rest of the header.
3258 */
3259 DBG3("Partial reception of data connection header (received %" PRIu64 " bytes, %" PRIu64 " bytes left to receive, fd = %i)",
3260 state->received, state->left_to_receive,
3261 conn->sock->fd);
7591bab1 3262 goto end;
b8aa1682 3263 }
b8aa1682 3264
5312a3ed
JG		 3265 /* Transition to next state: receiving the payload. */
3266 conn->protocol.data.state_id = DATA_CONNECTION_STATE_RECEIVE_PAYLOAD;
173af62f 3267
5312a3ed
JG		 3268 memcpy(&header, state->header_reception_buffer, sizeof(header));
3269 header.circuit_id = be64toh(header.circuit_id);
3270 header.stream_id = be64toh(header.stream_id);
3271 header.data_size = be32toh(header.data_size);
3272 header.net_seq_num = be64toh(header.net_seq_num);
3273 header.padding_size = be32toh(header.padding_size);
3274 memcpy(&conn->protocol.data.state.receive_payload.header, &header, sizeof(header));
3275
3276 conn->protocol.data.state.receive_payload.left_to_receive =
3277 header.data_size;
3278 conn->protocol.data.state.receive_payload.received = 0;
3279 conn->protocol.data.state.receive_payload.rotate_index = false;
3280
3281 DBG("Received data connection header on fd %i: circuit_id = %" PRIu64 ", stream_id = %" PRIu64 ", data_size = %" PRIu32 ", net_seq_num = %" PRIu64 ", padding_size = %" PRIu32,
3282 conn->sock->fd, header.circuit_id,
3283 header.stream_id, header.data_size,
3284 header.net_seq_num, header.padding_size);
3285
3286 stream = stream_get_by_id(header.stream_id);
3287 if (!stream) {
3288 DBG("relay_process_data_receive_payload: Cannot find stream %" PRIu64,
3289 header.stream_id);
5569b118
JG		 3290 /* Protocol error. */
3291 status = RELAY_CONNECTION_STATUS_ERROR;
5312a3ed
JG		 3292 goto end;
3293 }
b8aa1682 3294
7591bab1
MD		 3295 pthread_mutex_lock(&stream->lock);
3296
1c20f0e2 3297 /* Check if a rotation is needed. */
0f907de1 3298 if (stream->tracefile_size > 0 &&
5312a3ed 3299 (stream->tracefile_size_current + header.data_size) >
0f907de1 3300 stream->tracefile_size) {
a44ca2ca
MD		 3301 uint64_t old_id, new_id;
3302
3303 old_id = tracefile_array_get_file_index_head(stream->tfa);
3304 tracefile_array_file_rotate(stream->tfa);
3305
3306 /* new_id is updated by utils_rotate_stream_file. */
3307 new_id = old_id;
6b6b9a5a 3308
7591bab1
MD		 3309 ret = utils_rotate_stream_file(stream->path_name,
3310 stream->channel_name, stream->tracefile_size,
3311 stream->tracefile_count, -1,
3312 -1, stream->stream_fd->fd,
a44ca2ca 3313 &new_id, &stream->stream_fd->fd);
0f907de1 3314 if (ret < 0) {
5312a3ed 3315 ERR("Failed to rotate stream output file");
5569b118 3316 status = RELAY_CONNECTION_STATUS_ERROR;
7591bab1
MD		 3317 goto end_stream_unlock;
3318 }
5312a3ed 3319
7591bab1
MD		 3320 /*
3321 * Reset current size because we just performed a stream
3322 * rotation.
3323 */
a6976990 3324 stream->tracefile_size_current = 0;
5312a3ed 3325 conn->protocol.data.state.receive_payload.rotate_index = true;
1c20f0e2
JD		 3326 }
3327
5312a3ed
JG		 3328end_stream_unlock:
3329 pthread_mutex_unlock(&stream->lock);
3330 stream_put(stream);
3331end:
5569b118 3332 return status;
5312a3ed
JG		 3333}
3334
5569b118
JG		 3335static enum relay_connection_status relay_process_data_receive_payload(
3336 struct relay_connection *conn)
5312a3ed
JG		 3337{
3338 int ret;
5569b118 3339 enum relay_connection_status status = RELAY_CONNECTION_STATUS_OK;
5312a3ed
JG		 3340 struct relay_stream *stream;
3341 struct data_connection_state_receive_payload *state =
3342 &conn->protocol.data.state.receive_payload;
3343 const size_t chunk_size = RECV_DATA_BUFFER_SIZE;
3344 char data_buffer[chunk_size];
3345 bool partial_recv = false;
3346 bool new_stream = false, close_requested = false, index_flushed = false;
3347 uint64_t left_to_receive = state->left_to_receive;
3348 struct relay_session *session;
3349
fd0f1e3e
JR		 3350 DBG3("Receiving data for stream id %" PRIu64 " seqnum %" PRIu64 ", %" PRIu64" bytes received, %" PRIu64 " bytes left to receive",
3351 state->header.stream_id, state->header.net_seq_num,
3352 state->received, left_to_receive);
3353
5312a3ed
JG		 3354 stream = stream_get_by_id(state->header.stream_id);
3355 if (!stream) {
5569b118 3356 /* Protocol error. */
fd0f1e3e 3357 ERR("relay_process_data_receive_payload: cannot find stream %" PRIu64,
5312a3ed 3358 state->header.stream_id);
5569b118 3359 status = RELAY_CONNECTION_STATUS_ERROR;
5312a3ed 3360 goto end;
1c20f0e2
JD		 3361 }
3362
5312a3ed
JG		 3363 pthread_mutex_lock(&stream->lock);
3364 session = stream->trace->session;
fd0f1e3e
JR		 3365 if (!conn->session) {
3366 ret = connection_set_session(conn, session);
3367 if (ret) {
3368 status = RELAY_CONNECTION_STATUS_ERROR;
3369 goto end_stream_unlock;
3370 }
3371 }
5312a3ed
JG		 3372
3373 /*
3374 * The size of the "chunk" received on any iteration is bounded by:
3375 * - the data left to receive,
3376 * - the data immediately available on the socket,
3377 * - the on-stack data buffer
3378 */
3379 while (left_to_receive > 0 && !partial_recv) {
3380 ssize_t write_ret;
3381 size_t recv_size = min(left_to_receive, chunk_size);
3382
3383 ret = conn->sock->ops->recvmsg(conn->sock, data_buffer,
3384 recv_size, MSG_DONTWAIT);
3385 if (ret < 0) {
5569b118
JG		 3386 if (errno != EAGAIN && errno != EWOULDBLOCK) {
3387 PERROR("Socket %d error", conn->sock->fd);
3388 status = RELAY_CONNECTION_STATUS_ERROR;
3389 }
0848dba7 3390 goto end_stream_unlock;
5312a3ed
JG		 3391 } else if (ret == 0) {
3392 /* No more data ready to be consumed on socket. */
3393 DBG3("No more data ready for consumption on data socket of stream id %" PRIu64,
3394 state->header.stream_id);
5569b118 3395 status = RELAY_CONNECTION_STATUS_CLOSED;
5312a3ed
JG		 3396 break;
3397 } else if (ret < (int) recv_size) {
3398 /*
3399 * All the data available on the socket has been
3400 * consumed.
3401 */
3402 partial_recv = true;
0848dba7
MD		 3403 }
3404
5312a3ed
JG		 3405 recv_size = ret;
3406
0848dba7 3407 /* Write data to stream output fd. */
5312a3ed 3408 write_ret = lttng_write(stream->stream_fd->fd, data_buffer,
0848dba7 3409 recv_size);
5312a3ed 3410 if (write_ret < (ssize_t) recv_size) {
0848dba7 3411 ERR("Relay error writing data to file");
5569b118 3412 status = RELAY_CONNECTION_STATUS_ERROR;
0848dba7
MD		 3413 goto end_stream_unlock;
3414 }
3415
5312a3ed
JG		 3416 left_to_receive -= recv_size;
3417 state->received += recv_size;
3418 state->left_to_receive = left_to_receive;
3419
0848dba7 3420 DBG2("Relay wrote %zd bytes to tracefile for stream id %" PRIu64,
5312a3ed
JG		 3421 write_ret, stream->stream_handle);
3422 }
3423
3424 if (state->left_to_receive > 0) {
3425 /*
3426 * Did not receive all the data expected, wait for more data to
3427 * become available on the socket.
3428 */
3429 DBG3("Partial receive on data connection of stream id %" PRIu64 ", %" PRIu64 " bytes received, %" PRIu64 " bytes left to receive",
3430 state->header.stream_id, state->received,
3431 state->left_to_receive);
5312a3ed 3432 goto end_stream_unlock;
0848dba7 3433 }
5ab7344e 3434
7591bab1 3435 ret = write_padding_to_file(stream->stream_fd->fd,
5312a3ed 3436 state->header.padding_size);
46f4eaa5 3437 if ((int64_t) ret < (int64_t) state->header.padding_size) {
65d66b19 3438 ERR("write_padding_to_file: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
5312a3ed
JG		 3439 stream->stream_handle,
3440 state->header.net_seq_num, ret);
5569b118 3441 status = RELAY_CONNECTION_STATUS_ERROR;
7591bab1 3442 goto end_stream_unlock;
1d4dfdef 3443 }
5312a3ed
JG		 3444
3445
3446 if (session->minor >= 4 && !session->snapshot) {
3447 ret = handle_index_data(stream, state->header.net_seq_num,
3448 state->rotate_index, &index_flushed, state->header.data_size + state->header.padding_size);
3449 if (ret < 0) {
3450 ERR("handle_index_data: fail stream %" PRIu64 " net_seq_num %" PRIu64 " ret %d",
3451 stream->stream_handle,
3452 state->header.net_seq_num, ret);
5569b118 3453 status = RELAY_CONNECTION_STATUS_ERROR;
5312a3ed
JG		 3454 goto end_stream_unlock;
3455 }
3456 }
3457
3458 stream->tracefile_size_current += state->header.data_size +
3459 state->header.padding_size;
3460
c0bae11d
MD		 3461 if (stream->prev_seq == -1ULL) {
3462 new_stream = true;
3463 }
d3ecc550
JD		 3464 if (index_flushed) {
3465 stream->pos_after_last_complete_data_index =
3466 stream->tracefile_size_current;
3467 }
c0bae11d 3468
5312a3ed
JG		 3469 stream->prev_seq = state->header.net_seq_num;
3470
3471 /*
3472 * Resetting the protocol state (to RECEIVE_HEADER) will trash the
3473 * contents of *state which are aliased (union) to the same location as
3474 * the new state. Don't use it beyond this point.
3475 */
3476 connection_reset_protocol_state(conn);
3477 state = NULL;
173af62f 3478
d3ecc550
JD		 3479 ret = try_rotate_stream(stream);
3480 if (ret < 0) {
5569b118 3481 status = RELAY_CONNECTION_STATUS_ERROR;
d3ecc550
JD		 3482 goto end_stream_unlock;
3483 }
3484
7591bab1 3485end_stream_unlock:
bda7c7b9 3486 close_requested = stream->close_requested;
7591bab1 3487 pthread_mutex_unlock(&stream->lock);
5312a3ed 3488 if (close_requested && left_to_receive == 0) {
bda7c7b9
JG		 3489 try_stream_close(stream);
3490 }
3491
c0bae11d
MD		 3492 if (new_stream) {
3493 pthread_mutex_lock(&session->lock);
3494 uatomic_set(&session->new_streams, 1);
3495 pthread_mutex_unlock(&session->lock);
3496 }
5312a3ed 3497
7591bab1 3498 stream_put(stream);
b8aa1682 3499end:
5569b118 3500 return status;
b8aa1682
JD		 3501}
3502
5312a3ed
JG		 3503/*
3504 * relay_process_data: Process the data received on the data socket
3505 */
5569b118
JG		 3506static enum relay_connection_status relay_process_data(
3507 struct relay_connection *conn)
5312a3ed 3508{
5569b118 3509 enum relay_connection_status status;
5312a3ed
JG		 3510
3511 switch (conn->protocol.data.state_id) {
3512 case DATA_CONNECTION_STATE_RECEIVE_HEADER:
5569b118 3513 status = relay_process_data_receive_header(conn);
5312a3ed
JG		 3514 break;
3515 case DATA_CONNECTION_STATE_RECEIVE_PAYLOAD:
5569b118 3516 status = relay_process_data_receive_payload(conn);
5312a3ed
JG		 3517 break;
3518 default:
3519 ERR("Unexpected data connection communication state.");
3520 abort();
3521 }
3522
5569b118 3523 return status;
5312a3ed
JG		 3524}
3525
7591bab1 3526static void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
b8aa1682
JD		 3527{
3528 int ret;
3529
58eb9381 3530 (void) lttng_poll_del(events, pollfd);
b8aa1682
JD		 3531
3532 ret = close(pollfd);
3533 if (ret < 0) {
3534 ERR("Closing pollfd %d", pollfd);
3535 }
3536}
3537
7591bab1
MD		 3538static void relay_thread_close_connection(struct lttng_poll_event *events,
3539 int pollfd, struct relay_connection *conn)
9d1bbf21 3540{
7591bab1 3541 const char *type_str;
2a174661 3542
7591bab1
MD		 3543 switch (conn->type) {
3544 case RELAY_DATA:
3545 type_str = "Data";
3546 break;
3547 case RELAY_CONTROL:
3548 type_str = "Control";
3549 break;
3550 case RELAY_VIEWER_COMMAND:
3551 type_str = "Viewer Command";
3552 break;
3553 case RELAY_VIEWER_NOTIFICATION:
3554 type_str = "Viewer Notification";
3555 break;
3556 default:
3557 type_str = "Unknown";
9d1bbf21 3558 }
7591bab1
MD		 3559 cleanup_connection_pollfd(events, pollfd);
3560 connection_put(conn);
3561 DBG("%s connection closed with %d", type_str, pollfd);
b8aa1682
JD		 3562}
3563
3564/*
3565 * This thread does the actual work
3566 */
7591bab1 3567static void *relay_thread_worker(void *data)
b8aa1682 3568{
beaad64c
DG		 3569 int ret, err = -1, last_seen_data_fd = -1;
3570 uint32_t nb_fd;
b8aa1682
JD		 3571 struct lttng_poll_event events;
3572 struct lttng_ht *relay_connections_ht;
b8aa1682 3573 struct lttng_ht_iter iter;
90e7d72f 3574 struct relay_connection *destroy_conn = NULL;
b8aa1682
JD		 3575
3576 DBG("[thread] Relay worker started");
3577
9d1bbf21
MD		 3578 rcu_register_thread();
3579
55706a7d
MD		 3580 health_register(health_relayd, HEALTH_RELAYD_TYPE_WORKER);
3581
9b5e0863
MD		 3582 if (testpoint(relayd_thread_worker)) {
3583 goto error_testpoint;
3584 }
3585
f385ae0a
MD		 3586 health_code_update();
3587
b8aa1682
JD		 3588 /* table of connections indexed on socket */
3589 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
095a4ae5
MD		 3590 if (!relay_connections_ht) {
3591 goto relay_connections_ht_error;
3592 }
b8aa1682 3593
b8aa1682
JD		 3594 ret = create_thread_poll_set(&events, 2);
3595 if (ret < 0) {
3596 goto error_poll_create;
3597 }
3598
58eb9381 3599 ret = lttng_poll_add(&events, relay_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
b8aa1682
JD		 3600 if (ret < 0) {
3601 goto error;
3602 }
3603
beaad64c 3604restart:
b8aa1682 3605 while (1) {
beaad64c
DG		 3606 int idx = -1, i, seen_control = 0, last_notdel_data_fd = -1;
3607
f385ae0a
MD		 3608 health_code_update();
3609
b8aa1682 3610 /* Infinite blocking call, waiting for transmission */
87c1611d 3611 DBG3("Relayd worker thread polling...");
f385ae0a 3612 health_poll_entry();
b8aa1682 3613 ret = lttng_poll_wait(&events, -1);
f385ae0a 3614 health_poll_exit();
b8aa1682
JD		 3615 if (ret < 0) {
3616 /*
3617 * Restart interrupted system call.
3618 */
3619 if (errno == EINTR) {
3620 goto restart;
3621 }
3622 goto error;
3623 }
3624
0d9c5d77
DG		 3625 nb_fd = ret;
3626
beaad64c 3627 /*
7591bab1
MD		 3628 * Process control. The control connection is
3629 * prioritized so we don't starve it with high
3630 * throughput tracing data on the data connection.
beaad64c 3631 */
b8aa1682
JD		 3632 for (i = 0; i < nb_fd; i++) {
3633 /* Fetch once the poll data */
beaad64c
DG		 3634 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
3635 int pollfd = LTTNG_POLL_GETFD(&events, i);
b8aa1682 3636
f385ae0a
MD		 3637 health_code_update();
3638
fd20dac9 3639 if (!revents) {
7591bab1
MD		 3640 /*
3641 * No activity for this FD (poll
3642 * implementation).
3643 */
fd20dac9
MD		 3644 continue;
3645 }
3646
b8aa1682
JD		 3647 /* Thread quit pipe has been closed. Killing thread. */
3648 ret = check_thread_quit_pipe(pollfd, revents);
3649 if (ret) {
095a4ae5
MD		 3650 err = 0;
3651 goto exit;
b8aa1682
JD		 3652 }
3653
58eb9381
DG		 3654 /* Inspect the relay conn pipe for new connection */
3655 if (pollfd == relay_conn_pipe[0]) {
03e43155 3656 if (revents & LPOLLIN) {
90e7d72f
JG		 3657 struct relay_connection *conn;
3658
58eb9381 3659 ret = lttng_read(relay_conn_pipe[0], &conn, sizeof(conn));
b8aa1682
JD		 3660 if (ret < 0) {
3661 goto error;
3662 }
58eb9381
DG		 3663 lttng_poll_add(&events, conn->sock->fd,
3664 LPOLLIN | LPOLLRDHUP);
7591bab1 3665 connection_ht_add(relay_connections_ht, conn);
58eb9381 3666 DBG("Connection socket %d added", conn->sock->fd);
03e43155
MD		 3667 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3668 ERR("Relay connection pipe error");
3669 goto error;
3670 } else {
3671 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
3672 goto error;
b8aa1682 3673 }
58eb9381 3674 } else {
90e7d72f
JG		 3675 struct relay_connection *ctrl_conn;
3676
7591bab1 3677 ctrl_conn = connection_get_by_sock(relay_connections_ht, pollfd);
58eb9381 3678 /* If not found, there is a synchronization issue. */
90e7d72f 3679 assert(ctrl_conn);
58eb9381 3680
03e43155
MD		 3681 if (ctrl_conn->type == RELAY_DATA) {
3682 if (revents & LPOLLIN) {
beaad64c
DG		 3683 /*
3684 * Flag the last seen data fd not deleted. It will be
3685 * used as the last seen fd if any fd gets deleted in
3686 * this first loop.
3687 */
3688 last_notdel_data_fd = pollfd;
3689 }
03e43155
MD		 3690 goto put_ctrl_connection;
3691 }
3692 assert(ctrl_conn->type == RELAY_CONTROL);
3693
3694 if (revents & LPOLLIN) {
5569b118
JG		 3695 enum relay_connection_status status;
3696
3697 status = relay_process_control(ctrl_conn);
3698 if (status != RELAY_CONNECTION_STATUS_OK) {
fd0f1e3e
JR		 3699 /*
3700 * On socket error flag the session as aborted to force
3701 * the cleanup of its stream otherwise it can leak
3702 * during the lifetime of the relayd.
3703 *
3704 * This prevents situations in which streams can be
3705 * left opened because an index was received, the
3706 * control connection is closed, and the data
3707 * connection is closed (uncleanly) before the packet's
3708 * data provided.
3709 *
3710 * Since the control connection encountered an error,
3711 * it is okay to be conservative and close the
3712 * session right now as we can't rely on the protocol
3713 * being respected anymore.
3714 */
3715 if (status == RELAY_CONNECTION_STATUS_ERROR) {
3716 session_abort(ctrl_conn->session);
3717 }
3718
5569b118 3719 /* Clear the connection on error or close. */
5312a3ed
JG		 3720 relay_thread_close_connection(&events,
3721 pollfd,
03e43155 3722 ctrl_conn);
03e43155 3723 }
5312a3ed 3724 seen_control = 1;
03e43155
MD		 3725 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3726 relay_thread_close_connection(&events,
3727 pollfd, ctrl_conn);
3728 if (last_seen_data_fd == pollfd) {
3729 last_seen_data_fd = last_notdel_data_fd;
3730 }
58eb9381 3731 } else {
03e43155
MD		 3732 ERR("Unexpected poll events %u for control sock %d",
3733 revents, pollfd);
3734 connection_put(ctrl_conn);
3735 goto error;
beaad64c 3736 }
03e43155 3737 put_ctrl_connection:
7591bab1 3738 connection_put(ctrl_conn);
beaad64c
DG		 3739 }
3740 }
3741
3742 /*
3743 * The last loop handled a control request, go back to poll to make
3744 * sure we prioritise the control socket.
3745 */
3746 if (seen_control) {
3747 continue;
3748 }
3749
3750 if (last_seen_data_fd >= 0) {
3751 for (i = 0; i < nb_fd; i++) {
3752 int pollfd = LTTNG_POLL_GETFD(&events, i);
f385ae0a
MD		 3753
3754 health_code_update();
3755
beaad64c
DG		 3756 if (last_seen_data_fd == pollfd) {
3757 idx = i;
3758 break;
3759 }
3760 }
3761 }
3762
3763 /* Process data connection. */
3764 for (i = idx + 1; i < nb_fd; i++) {
3765 /* Fetch the poll data. */
3766 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
3767 int pollfd = LTTNG_POLL_GETFD(&events, i);
90e7d72f 3768 struct relay_connection *data_conn;
beaad64c 3769
f385ae0a
MD		 3770 health_code_update();
3771
fd20dac9
MD		 3772 if (!revents) {
3773 /* No activity for this FD (poll implementation). */
3774 continue;
3775 }
3776
beaad64c 3777 /* Skip the command pipe. It's handled in the first loop. */
58eb9381 3778 if (pollfd == relay_conn_pipe[0]) {
beaad64c
DG		 3779 continue;
3780 }
3781
7591bab1 3782 data_conn = connection_get_by_sock(relay_connections_ht, pollfd);
90e7d72f 3783 if (!data_conn) {
fd20dac9 3784 /* Skip it. Might be removed before. */
fd20dac9
MD		 3785 continue;
3786 }
03e43155
MD		 3787 if (data_conn->type == RELAY_CONTROL) {
3788 goto put_data_connection;
3789 }
3790 assert(data_conn->type == RELAY_DATA);
fd20dac9
MD		 3791
3792 if (revents & LPOLLIN) {
5569b118
JG		 3793 enum relay_connection_status status;
3794
3795 status = relay_process_data(data_conn);
3796 /* Connection closed or error. */
3797 if (status != RELAY_CONNECTION_STATUS_OK) {
fd0f1e3e
JR		 3798 /*
3799 * On socket error flag the session as aborted to force
3800 * the cleanup of its stream otherwise it can leak
3801 * during the lifetime of the relayd.
3802 *
3803 * This prevents situations in which streams can be
3804 * left opened because an index was received, the
3805 * control connection is closed, and the data
3806 * connection is closed (uncleanly) before the packet's
3807 * data provided.
3808 *
3809 * Since the data connection encountered an error,
3810 * it is okay to be conservative and close the
3811 * session right now as we can't rely on the protocol
3812 * being respected anymore.
3813 */
3814 if (status == RELAY_CONNECTION_STATUS_ERROR) {
3815 session_abort(data_conn->session);
3816 }
7591bab1 3817 relay_thread_close_connection(&events, pollfd,
03e43155 3818 data_conn);
fd20dac9
MD		 3819 /*
3820 * Every goto restart call sets the last seen fd where
3821 * here we don't really care since we gracefully
3822 * continue the loop after the connection is deleted.
3823 */
3824 } else {
3825 /* Keep last seen port. */
3826 last_seen_data_fd = pollfd;
7591bab1 3827 connection_put(data_conn);
fd20dac9 3828 goto restart;
b8aa1682 3829 }
03e43155
MD		 3830 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
3831 relay_thread_close_connection(&events, pollfd,
3832 data_conn);
3833 } else {
3834 ERR("Unknown poll events %u for data sock %d",
3835 revents, pollfd);
b8aa1682 3836 }
03e43155 3837 put_data_connection:
7591bab1 3838 connection_put(data_conn);
b8aa1682 3839 }
beaad64c 3840 last_seen_data_fd = -1;
b8aa1682
JD		 3841 }
3842
f385ae0a
MD		 3843 /* Normal exit, no error */
3844 ret = 0;
3845
095a4ae5 3846exit:
b8aa1682 3847error:
71efa8ef 3848 /* Cleanup remaining connection object. */
9d1bbf21 3849 rcu_read_lock();
90e7d72f
JG		 3850 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter,
3851 destroy_conn,
58eb9381 3852 sock_n.node) {
f385ae0a 3853 health_code_update();
98ba050e 3854
fd0f1e3e 3855 session_abort(destroy_conn->session);
98ba050e 3856
7591bab1
MD		 3857 /*
3858 * No need to grab another ref, because we own
3859 * destroy_conn.
3860 */
3861 relay_thread_close_connection(&events, destroy_conn->sock->fd,
3862 destroy_conn);
b8aa1682 3863 }
94d49140 3864 rcu_read_unlock();
7591bab1
MD		 3865
3866 lttng_poll_clean(&events);
7d2f7452 3867error_poll_create:
b8aa1682 3868 lttng_ht_destroy(relay_connections_ht);
095a4ae5 3869relay_connections_ht_error:
58eb9381
DG		 3870 /* Close relay conn pipes */
3871 utils_close_pipe(relay_conn_pipe);
095a4ae5
MD		 3872 if (err) {
3873 DBG("Thread exited with error");
3874 }
b8aa1682 3875 DBG("Worker thread cleanup complete");
9b5e0863 3876error_testpoint:
f385ae0a
MD		 3877 if (err) {
3878 health_error();
3879 ERR("Health error occurred in %s", __func__);
3880 }
3881 health_unregister(health_relayd);
9d1bbf21 3882 rcu_unregister_thread();
b4aacfdc 3883 lttng_relay_stop_threads();
b8aa1682
JD		 3884 return NULL;
3885}
3886
3887/*
3888 * Create the relay command pipe to wake thread_manage_apps.
3889 * Closed in cleanup().
3890 */
58eb9381 3891static int create_relay_conn_pipe(void)
b8aa1682 3892{
a02de639 3893 int ret;
b8aa1682 3894
58eb9381 3895 ret = utils_create_pipe_cloexec(relay_conn_pipe);
b8aa1682 3896
b8aa1682
JD		 3897 return ret;
3898}
3899
3900/*
3901 * main
3902 */
3903int main(int argc, char **argv)
3904{
178a0557 3905 int ret = 0, retval = 0;
b8aa1682
JD		 3906 void *status;
3907
b8aa1682
JD		 3908 /* Parse arguments */
3909 progname = argv[0];
178a0557
MD		 3910 if (set_options(argc, argv)) {
3911 retval = -1;
3912 goto exit_options;
b8aa1682
JD		 3913 }
3914
178a0557
MD		 3915 if (set_signal_handler()) {
3916 retval = -1;
3917 goto exit_options;
b8aa1682
JD		 3918 }
3919
4d513a50
DG		 3920 /* Try to create directory if -o, --output is specified. */
3921 if (opt_output_path) {
994fa64f
DG		 3922 if (*opt_output_path != '/') {
3923 ERR("Please specify an absolute path for -o, --output PATH");
178a0557
MD		 3924 retval = -1;
3925 goto exit_options;
994fa64f
DG		 3926 }
3927
d77dded2
JG		 3928 ret = utils_mkdir_recursive(opt_output_path, S_IRWXU | S_IRWXG,
3929 -1, -1);
4d513a50
DG		 3930 if (ret < 0) {
3931 ERR("Unable to create %s", opt_output_path);
178a0557
MD		 3932 retval = -1;
3933 goto exit_options;
4d513a50
DG		 3934 }
3935 }
3936
b8aa1682 3937 /* Daemonize */
b5218ffb 3938 if (opt_daemon || opt_background) {
3fd27398
MD		 3939 int i;
3940
3941 ret = lttng_daemonize(&child_ppid, &recv_child_signal,
3942 !opt_background);
b8aa1682 3943 if (ret < 0) {
178a0557
MD		 3944 retval = -1;
3945 goto exit_options;
b8aa1682 3946 }
3fd27398
MD		 3947
3948 /*
3949 * We are in the child. Make sure all other file
3950 * descriptors are closed, in case we are called with
3951 * more opened file descriptors than the standard ones.
3952 */
3953 for (i = 3; i < sysconf(_SC_OPEN_MAX); i++) {
3954 (void) close(i);
3955 }
3956 }
3957
178a0557
MD		 3958 /* Initialize thread health monitoring */
3959 health_relayd = health_app_create(NR_HEALTH_RELAYD_TYPES);
3960 if (!health_relayd) {
3961 PERROR("health_app_create error");
3962 retval = -1;
3963 goto exit_health_app_create;
3964 }
3965
3fd27398 3966 /* Create thread quit pipe */
178a0557
MD		 3967 if (init_thread_quit_pipe()) {
3968 retval = -1;
3969 goto exit_init_data;
b8aa1682
JD		 3970 }
3971
b8aa1682 3972 /* Setup the thread apps communication pipe. */
178a0557
MD		 3973 if (create_relay_conn_pipe()) {
3974 retval = -1;
3975 goto exit_init_data;
b8aa1682
JD		 3976 }
3977
3978 /* Init relay command queue. */
8bdee6e2 3979 cds_wfcq_init(&relay_conn_queue.head, &relay_conn_queue.tail);
b8aa1682 3980
554831e7
MD		 3981 /* Initialize communication library */
3982 lttcomm_init();
87e45c13 3983 lttcomm_inet_init();
554831e7 3984
d3e2ba59 3985 /* tables of sessions indexed by session ID */
7591bab1
MD		 3986 sessions_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3987 if (!sessions_ht) {
178a0557
MD		 3988 retval = -1;
3989 goto exit_init_data;
d3e2ba59
JD		 3990 }
3991
3992 /* tables of streams indexed by stream ID */
2a174661 3993 relay_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
d3e2ba59 3994 if (!relay_streams_ht) {
178a0557
MD		 3995 retval = -1;
3996 goto exit_init_data;
d3e2ba59
JD		 3997 }
3998
3999 /* tables of streams indexed by stream ID */
92c6ca54
DG		 4000 viewer_streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
4001 if (!viewer_streams_ht) {
178a0557
MD		 4002 retval = -1;
4003 goto exit_init_data;
55706a7d
MD		 4004 }
4005
65931c8b 4006 ret = utils_create_pipe(health_quit_pipe);
178a0557
MD		 4007 if (ret) {
4008 retval = -1;
4009 goto exit_health_quit_pipe;
65931c8b
MD		 4010 }
4011
4012 /* Create thread to manage the client socket */
1a1a34b4 4013 ret = pthread_create(&health_thread, default_pthread_attr(),
65931c8b 4014 thread_manage_health, (void *) NULL);
178a0557
MD		 4015 if (ret) {
4016 errno = ret;
65931c8b 4017 PERROR("pthread_create health");
178a0557
MD		 4018 retval = -1;
4019 goto exit_health_thread;
65931c8b
MD		 4020 }
4021
b8aa1682 4022 /* Setup the dispatcher thread */
1a1a34b4 4023 ret = pthread_create(&dispatcher_thread, default_pthread_attr(),
b8aa1682 4024 relay_thread_dispatcher, (void *) NULL);
178a0557
MD		 4025 if (ret) {
4026 errno = ret;
b8aa1682 4027 PERROR("pthread_create dispatcher");
178a0557
MD		 4028 retval = -1;
4029 goto exit_dispatcher_thread;
b8aa1682
JD		 4030 }
4031
4032 /* Setup the worker thread */
1a1a34b4 4033 ret = pthread_create(&worker_thread, default_pthread_attr(),
7591bab1 4034 relay_thread_worker, NULL);
178a0557
MD		 4035 if (ret) {
4036 errno = ret;
b8aa1682 4037 PERROR("pthread_create worker");
178a0557
MD		 4038 retval = -1;
4039 goto exit_worker_thread;
b8aa1682
JD		 4040 }
4041
4042 /* Setup the listener thread */
1a1a34b4 4043 ret = pthread_create(&listener_thread, default_pthread_attr(),
b8aa1682 4044 relay_thread_listener, (void *) NULL);
178a0557
MD		 4045 if (ret) {
4046 errno = ret;
b8aa1682 4047 PERROR("pthread_create listener");
178a0557
MD		 4048 retval = -1;
4049 goto exit_listener_thread;
b8aa1682
JD		 4050 }
4051
7591bab1 4052 ret = relayd_live_create(live_uri);
178a0557 4053 if (ret) {
d3e2ba59 4054 ERR("Starting live viewer threads");
178a0557 4055 retval = -1;
50138f51 4056 goto exit_live;
d3e2ba59
JD		 4057 }
4058
178a0557
MD		 4059 /*
4060 * This is where we start awaiting program completion (e.g. through
4061 * signal that asks threads to teardown).
4062 */
4063
4064 ret = relayd_live_join();
4065 if (ret) {
4066 retval = -1;
4067 }
50138f51 4068exit_live:
178a0557 4069
b8aa1682 4070 ret = pthread_join(listener_thread, &status);
178a0557
MD		 4071 if (ret) {
4072 errno = ret;
4073 PERROR("pthread_join listener_thread");
4074 retval = -1;
b8aa1682
JD		 4075 }
4076
178a0557 4077exit_listener_thread:
b8aa1682 4078 ret = pthread_join(worker_thread, &status);
178a0557
MD		 4079 if (ret) {
4080 errno = ret;
4081 PERROR("pthread_join worker_thread");
4082 retval = -1;
b8aa1682
JD		 4083 }
4084
178a0557 4085exit_worker_thread:
b8aa1682 4086 ret = pthread_join(dispatcher_thread, &status);
178a0557
MD		 4087 if (ret) {
4088 errno = ret;
4089 PERROR("pthread_join dispatcher_thread");
4090 retval = -1;
b8aa1682 4091 }
178a0557 4092exit_dispatcher_thread:
42415026 4093
65931c8b 4094 ret = pthread_join(health_thread, &status);
178a0557
MD		 4095 if (ret) {
4096 errno = ret;
4097 PERROR("pthread_join health_thread");
4098 retval = -1;
65931c8b 4099 }
178a0557 4100exit_health_thread:
65931c8b 4101
65931c8b 4102 utils_close_pipe(health_quit_pipe);
178a0557 4103exit_health_quit_pipe:
65931c8b 4104
178a0557 4105exit_init_data:
55706a7d 4106 health_app_destroy(health_relayd);
55706a7d 4107exit_health_app_create:
178a0557 4108exit_options:
4d62fbf8
MD		 4109 /*
4110 * Wait for all pending call_rcu work to complete before tearing
4111 * down data structures. call_rcu worker may be trying to
4112 * perform lookups in those structures.
4113 */
4114 rcu_barrier();
7591bab1
MD		 4115 relayd_cleanup();
4116
4117 /* Ensure all prior call_rcu are done. */
4118 rcu_barrier();
d3e2ba59 4119
178a0557 4120 if (!retval) {
b8aa1682 4121 exit(EXIT_SUCCESS);
178a0557
MD		 4122 } else {
4123 exit(EXIT_FAILURE);
b8aa1682 4124 }
b8aa1682 4125}
LTTng 2.0 tools and control repository
This page took 0.272879 seconds and 4 git commands to generate.