Relayd data available command support
[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>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19#define _GNU_SOURCE
20#include <getopt.h>
21#include <grp.h>
22#include <limits.h>
23#include <pthread.h>
24#include <signal.h>
25#include <stdio.h>
26#include <stdlib.h>
27#include <string.h>
28#include <sys/mman.h>
29#include <sys/mount.h>
30#include <sys/resource.h>
31#include <sys/socket.h>
32#include <sys/stat.h>
33#include <sys/types.h>
34#include <sys/wait.h>
173af62f 35#include <inttypes.h>
b8aa1682
JD
36#include <urcu/futex.h>
37#include <urcu/uatomic.h>
38#include <unistd.h>
39#include <fcntl.h>
40#include <config.h>
41
42#include <lttng/lttng.h>
43#include <common/common.h>
44#include <common/compat/poll.h>
45#include <common/compat/socket.h>
46#include <common/defaults.h>
47#include <common/futex.h>
48#include <common/sessiond-comm/sessiond-comm.h>
49#include <common/sessiond-comm/inet.h>
50#include <common/hashtable/hashtable.h>
51#include <common/sessiond-comm/relayd.h>
52#include <common/uri.h>
a02de639 53#include <common/utils.h>
b8aa1682
JD
54
55#include "lttng-relayd.h"
56
57/* command line options */
58static int opt_daemon;
59static char *opt_output_path;
095a4ae5
MD
60static struct lttng_uri *control_uri;
61static struct lttng_uri *data_uri;
b8aa1682
JD
62
63const char *progname;
64static int is_root; /* Set to 1 if the daemon is running as root */
65
66/*
67 * Quit pipe for all threads. This permits a single cancellation point
68 * for all threads when receiving an event on the pipe.
69 */
70static int thread_quit_pipe[2] = { -1, -1 };
71
72/*
73 * This pipe is used to inform the worker thread that a command is queued and
74 * ready to be processed.
75 */
76static int relay_cmd_pipe[2] = { -1, -1 };
77
26c9d55e 78/* Shared between threads */
b8aa1682
JD
79static int dispatch_thread_exit;
80
81static pthread_t listener_thread;
82static pthread_t dispatcher_thread;
83static pthread_t worker_thread;
84
095a4ae5
MD
85static uint64_t last_relay_stream_id;
86static uint64_t last_relay_session_id;
b8aa1682
JD
87
88/*
89 * Relay command queue.
90 *
91 * The relay_thread_listener and relay_thread_dispatcher communicate with this
92 * queue.
93 */
94static struct relay_cmd_queue relay_cmd_queue;
95
96/* buffer allocated at startup, used to store the trace data */
095a4ae5
MD
97static char *data_buffer;
98static unsigned int data_buffer_size;
b8aa1682
JD
99
100/*
101 * usage function on stderr
102 */
103static
104void usage(void)
105{
106 fprintf(stderr, "Usage: %s OPTIONS\n\nOptions:\n", progname);
107 fprintf(stderr, " -h, --help Display this usage.\n");
108 fprintf(stderr, " -d, --daemonize Start as a daemon.\n");
109 fprintf(stderr, " -C, --control-port Control port listening (URI)\n");
110 fprintf(stderr, " -D, --data-port Data port listening (URI)\n");
25672d02 111 fprintf(stderr, " -o, --output Output path for traces (PATH)\n");
b8aa1682
JD
112 fprintf(stderr, " -v, --verbose Verbose mode. Activate DBG() macro.\n");
113}
114
115static
116int parse_args(int argc, char **argv)
117{
118 int c;
119 int ret = 0;
120 char *default_address;
121
122 static struct option long_options[] = {
e3678fd8
MD
123 { "control-port", 1, 0, 'C', },
124 { "data-port", 1, 0, 'D', },
125 { "daemonize", 0, 0, 'd', },
126 { "help", 0, 0, 'h', },
127 { "output", 1, 0, 'o', },
128 { "verbose", 0, 0, 'v', },
095a4ae5 129 { NULL, 0, 0, 0, },
b8aa1682
JD
130 };
131
132 while (1) {
133 int option_index = 0;
134 c = getopt_long(argc, argv, "dhv" "C:D:o:",
135 long_options, &option_index);
136 if (c == -1) {
137 break;
138 }
139
140 switch (c) {
141 case 0:
142 fprintf(stderr, "option %s", long_options[option_index].name);
143 if (optarg) {
144 fprintf(stderr, " with arg %s\n", optarg);
145 }
146 break;
147 case 'C':
148 ret = uri_parse(optarg, &control_uri);
149 if (ret < 0) {
150 ERR("Invalid control URI specified");
151 goto exit;
152 }
153 if (control_uri->port == 0) {
154 control_uri->port = DEFAULT_NETWORK_CONTROL_PORT;
155 }
156 break;
157 case 'D':
158 ret = uri_parse(optarg, &data_uri);
159 if (ret < 0) {
160 ERR("Invalid data URI specified");
161 goto exit;
162 }
163 if (data_uri->port == 0) {
164 data_uri->port = DEFAULT_NETWORK_DATA_PORT;
165 }
166 break;
167 case 'd':
168 opt_daemon = 1;
169 break;
170 case 'h':
171 usage();
172 exit(EXIT_FAILURE);
173 case 'o':
174 ret = asprintf(&opt_output_path, "%s", optarg);
175 if (ret < 0) {
176 PERROR("asprintf opt_output_path");
177 goto exit;
178 }
179 break;
180 case 'v':
181 /* Verbose level can increase using multiple -v */
182 lttng_opt_verbose += 1;
183 break;
184 default:
185 /* Unknown option or other error.
186 * Error is printed by getopt, just return */
187 ret = -1;
188 goto exit;
189 }
190 }
191
192 /* assign default values */
193 if (control_uri == NULL) {
194 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
195 DEFAULT_NETWORK_CONTROL_PORT);
196 if (ret < 0) {
197 PERROR("asprintf default data address");
198 goto exit;
199 }
200
201 ret = uri_parse(default_address, &control_uri);
202 free(default_address);
203 if (ret < 0) {
204 ERR("Invalid control URI specified");
205 goto exit;
206 }
207 }
208 if (data_uri == NULL) {
209 ret = asprintf(&default_address, "tcp://0.0.0.0:%d",
210 DEFAULT_NETWORK_DATA_PORT);
211 if (ret < 0) {
212 PERROR("asprintf default data address");
213 goto exit;
214 }
215
216 ret = uri_parse(default_address, &data_uri);
217 free(default_address);
218 if (ret < 0) {
219 ERR("Invalid data URI specified");
220 goto exit;
221 }
222 }
223
224exit:
225 return ret;
226}
227
228/*
229 * Cleanup the daemon
230 */
231static
232void cleanup(void)
233{
b8aa1682
JD
234 DBG("Cleaning up");
235
095a4ae5
MD
236 /* free the dynamically allocated opt_output_path */
237 free(opt_output_path);
238
a02de639
CB
239 /* Close thread quit pipes */
240 utils_close_pipe(thread_quit_pipe);
241
242 /* Close relay cmd pipes */
243 utils_close_pipe(relay_cmd_pipe);
b8aa1682
JD
244}
245
246/*
247 * Write to writable pipe used to notify a thread.
248 */
249static
250int notify_thread_pipe(int wpipe)
251{
252 int ret;
253
6f94560a
MD
254 do {
255 ret = write(wpipe, "!", 1);
256 } while (ret < 0 && errno == EINTR);
b8aa1682
JD
257 if (ret < 0) {
258 PERROR("write poll pipe");
259 }
260
261 return ret;
262}
263
264/*
265 * Stop all threads by closing the thread quit pipe.
266 */
267static
268void stop_threads(void)
269{
270 int ret;
271
272 /* Stopping all threads */
273 DBG("Terminating all threads");
274 ret = notify_thread_pipe(thread_quit_pipe[1]);
275 if (ret < 0) {
276 ERR("write error on thread quit pipe");
277 }
278
279 /* Dispatch thread */
26c9d55e 280 CMM_STORE_SHARED(dispatch_thread_exit, 1);
b8aa1682
JD
281 futex_nto1_wake(&relay_cmd_queue.futex);
282}
283
284/*
285 * Signal handler for the daemon
286 *
287 * Simply stop all worker threads, leaving main() return gracefully after
288 * joining all threads and calling cleanup().
289 */
290static
291void sighandler(int sig)
292{
293 switch (sig) {
294 case SIGPIPE:
295 DBG("SIGPIPE caught");
296 return;
297 case SIGINT:
298 DBG("SIGINT caught");
299 stop_threads();
300 break;
301 case SIGTERM:
302 DBG("SIGTERM caught");
303 stop_threads();
304 break;
305 default:
306 break;
307 }
308}
309
310/*
311 * Setup signal handler for :
312 * SIGINT, SIGTERM, SIGPIPE
313 */
314static
315int set_signal_handler(void)
316{
317 int ret = 0;
318 struct sigaction sa;
319 sigset_t sigset;
320
321 if ((ret = sigemptyset(&sigset)) < 0) {
322 PERROR("sigemptyset");
323 return ret;
324 }
325
326 sa.sa_handler = sighandler;
327 sa.sa_mask = sigset;
328 sa.sa_flags = 0;
329 if ((ret = sigaction(SIGTERM, &sa, NULL)) < 0) {
330 PERROR("sigaction");
331 return ret;
332 }
333
334 if ((ret = sigaction(SIGINT, &sa, NULL)) < 0) {
335 PERROR("sigaction");
336 return ret;
337 }
338
339 if ((ret = sigaction(SIGPIPE, &sa, NULL)) < 0) {
340 PERROR("sigaction");
341 return ret;
342 }
343
344 DBG("Signal handler set for SIGTERM, SIGPIPE and SIGINT");
345
346 return ret;
347}
348
349/*
350 * Init thread quit pipe.
351 *
352 * Return -1 on error or 0 if all pipes are created.
353 */
354static
355int init_thread_quit_pipe(void)
356{
a02de639 357 int ret;
b8aa1682 358
a02de639 359 ret = utils_create_pipe_cloexec(thread_quit_pipe);
b8aa1682 360
b8aa1682
JD
361 return ret;
362}
363
364/*
365 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
366 */
367static
368int create_thread_poll_set(struct lttng_poll_event *events, int size)
369{
370 int ret;
371
372 if (events == NULL || size == 0) {
373 ret = -1;
374 goto error;
375 }
376
377 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
378 if (ret < 0) {
379 goto error;
380 }
381
382 /* Add quit pipe */
383 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN);
384 if (ret < 0) {
385 goto error;
386 }
387
388 return 0;
389
390error:
391 return ret;
392}
393
394/*
395 * Check if the thread quit pipe was triggered.
396 *
397 * Return 1 if it was triggered else 0;
398 */
399static
400int check_thread_quit_pipe(int fd, uint32_t events)
401{
402 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
403 return 1;
404 }
405
406 return 0;
407}
408
409/*
410 * Create and init socket from uri.
411 */
412static
413struct lttcomm_sock *relay_init_sock(struct lttng_uri *uri)
414{
415 int ret;
416 struct lttcomm_sock *sock = NULL;
417
418 sock = lttcomm_alloc_sock_from_uri(uri);
419 if (sock == NULL) {
420 ERR("Allocating socket");
421 goto error;
422 }
423
424 ret = lttcomm_create_sock(sock);
425 if (ret < 0) {
426 goto error;
427 }
428 DBG("Listening on sock %d", sock->fd);
429
430 ret = sock->ops->bind(sock);
431 if (ret < 0) {
432 goto error;
433 }
434
435 ret = sock->ops->listen(sock, -1);
436 if (ret < 0) {
437 goto error;
438
439 }
440
441 return sock;
442
443error:
444 if (sock) {
445 lttcomm_destroy_sock(sock);
446 }
447 return NULL;
448}
449
173af62f
DG
450/*
451 * Return nonzero if stream needs to be closed.
452 */
453static
454int close_stream_check(struct relay_stream *stream)
455{
456
457 if (stream->close_flag && stream->prev_seq == stream->last_net_seq_num) {
458 return 1;
459 }
460 return 0;
461}
462
b8aa1682
JD
463/*
464 * This thread manages the listening for new connections on the network
465 */
466static
467void *relay_thread_listener(void *data)
468{
095a4ae5 469 int i, ret, pollfd, err = -1;
b8aa1682
JD
470 int val = 1;
471 uint32_t revents, nb_fd;
472 struct lttng_poll_event events;
473 struct lttcomm_sock *control_sock, *data_sock;
474
b8aa1682
JD
475 DBG("[thread] Relay listener started");
476
477 control_sock = relay_init_sock(control_uri);
478 if (!control_sock) {
095a4ae5 479 goto error_sock_control;
b8aa1682
JD
480 }
481
482 data_sock = relay_init_sock(data_uri);
483 if (!data_sock) {
095a4ae5 484 goto error_sock_relay;
b8aa1682
JD
485 }
486
487 /*
488 * Pass 3 as size here for the thread quit pipe, control and data socket.
489 */
490 ret = create_thread_poll_set(&events, 3);
491 if (ret < 0) {
492 goto error_create_poll;
493 }
494
495 /* Add the control socket */
496 ret = lttng_poll_add(&events, control_sock->fd, LPOLLIN | LPOLLRDHUP);
497 if (ret < 0) {
498 goto error_poll_add;
499 }
500
501 /* Add the data socket */
502 ret = lttng_poll_add(&events, data_sock->fd, LPOLLIN | LPOLLRDHUP);
503 if (ret < 0) {
504 goto error_poll_add;
505 }
506
507 while (1) {
508 DBG("Listener accepting connections");
509
510 nb_fd = LTTNG_POLL_GETNB(&events);
511
512restart:
513 ret = lttng_poll_wait(&events, -1);
514 if (ret < 0) {
515 /*
516 * Restart interrupted system call.
517 */
518 if (errno == EINTR) {
519 goto restart;
520 }
521 goto error;
522 }
523
524 DBG("Relay new connection received");
525 for (i = 0; i < nb_fd; i++) {
526 /* Fetch once the poll data */
527 revents = LTTNG_POLL_GETEV(&events, i);
528 pollfd = LTTNG_POLL_GETFD(&events, i);
529
530 /* Thread quit pipe has been closed. Killing thread. */
531 ret = check_thread_quit_pipe(pollfd, revents);
532 if (ret) {
095a4ae5
MD
533 err = 0;
534 goto exit;
b8aa1682
JD
535 }
536
537 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
538 ERR("socket poll error");
539 goto error;
540 } else if (revents & LPOLLIN) {
4b7f17b2
MD
541 /*
542 * Get allocated in this thread,
543 * enqueued to a global queue, dequeued
544 * and freed in the worker thread.
545 */
546 struct relay_command *relay_cmd;
547 struct lttcomm_sock *newsock;
b8aa1682
JD
548
549 relay_cmd = zmalloc(sizeof(struct relay_command));
550 if (relay_cmd == NULL) {
551 PERROR("relay command zmalloc");
552 goto error;
553 }
554
555 if (pollfd == data_sock->fd) {
556 newsock = data_sock->ops->accept(data_sock);
4b7f17b2 557 if (!newsock) {
b8aa1682 558 PERROR("accepting data sock");
4b7f17b2 559 free(relay_cmd);
b8aa1682
JD
560 goto error;
561 }
562 relay_cmd->type = RELAY_DATA;
563 DBG("Relay data connection accepted, socket %d", newsock->fd);
4b7f17b2
MD
564 } else {
565 assert(pollfd == control_sock->fd);
b8aa1682 566 newsock = control_sock->ops->accept(control_sock);
4b7f17b2 567 if (!newsock) {
b8aa1682 568 PERROR("accepting control sock");
4b7f17b2 569 free(relay_cmd);
b8aa1682
JD
570 goto error;
571 }
572 relay_cmd->type = RELAY_CONTROL;
573 DBG("Relay control connection accepted, socket %d", newsock->fd);
574 }
575 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR,
576 &val, sizeof(int));
577 if (ret < 0) {
578 PERROR("setsockopt inet");
4b7f17b2
MD
579 lttcomm_destroy_sock(newsock);
580 free(relay_cmd);
b8aa1682
JD
581 goto error;
582 }
583 relay_cmd->sock = newsock;
584 /*
585 * Lock free enqueue the request.
586 */
587 cds_wfq_enqueue(&relay_cmd_queue.queue, &relay_cmd->node);
588
589 /*
590 * Wake the dispatch queue futex. Implicit memory
591 * barrier with the exchange in cds_wfq_enqueue.
592 */
593 futex_nto1_wake(&relay_cmd_queue.futex);
594 }
595 }
596 }
597
095a4ae5 598exit:
b8aa1682
JD
599error:
600error_poll_add:
601 lttng_poll_clean(&events);
602error_create_poll:
095a4ae5
MD
603 if (data_sock->fd >= 0) {
604 ret = data_sock->ops->close(data_sock);
b8aa1682
JD
605 if (ret) {
606 PERROR("close");
607 }
b8aa1682 608 }
095a4ae5
MD
609 lttcomm_destroy_sock(data_sock);
610error_sock_relay:
611 if (control_sock->fd >= 0) {
612 ret = control_sock->ops->close(control_sock);
b8aa1682
JD
613 if (ret) {
614 PERROR("close");
615 }
b8aa1682 616 }
095a4ae5
MD
617 lttcomm_destroy_sock(control_sock);
618error_sock_control:
619 if (err) {
620 DBG("Thread exited with error");
621 }
b8aa1682
JD
622 DBG("Relay listener thread cleanup complete");
623 stop_threads();
b8aa1682
JD
624 return NULL;
625}
626
627/*
628 * This thread manages the dispatching of the requests to worker threads
629 */
630static
631void *relay_thread_dispatcher(void *data)
632{
633 int ret;
634 struct cds_wfq_node *node;
635 struct relay_command *relay_cmd = NULL;
636
637 DBG("[thread] Relay dispatcher started");
638
26c9d55e 639 while (!CMM_LOAD_SHARED(dispatch_thread_exit)) {
b8aa1682
JD
640 /* Atomically prepare the queue futex */
641 futex_nto1_prepare(&relay_cmd_queue.futex);
642
643 do {
644 /* Dequeue commands */
645 node = cds_wfq_dequeue_blocking(&relay_cmd_queue.queue);
646 if (node == NULL) {
647 DBG("Woken up but nothing in the relay command queue");
648 /* Continue thread execution */
649 break;
650 }
651
652 relay_cmd = caa_container_of(node, struct relay_command, node);
653 DBG("Dispatching request waiting on sock %d", relay_cmd->sock->fd);
654
655 /*
656 * Inform worker thread of the new request. This
657 * call is blocking so we can be assured that the data will be read
658 * at some point in time or wait to the end of the world :)
659 */
6f94560a
MD
660 do {
661 ret = write(relay_cmd_pipe[1], relay_cmd,
662 sizeof(struct relay_command));
663 } while (ret < 0 && errno == EINTR);
b8aa1682
JD
664 free(relay_cmd);
665 if (ret < 0) {
666 PERROR("write cmd pipe");
667 goto error;
668 }
669 } while (node != NULL);
670
671 /* Futex wait on queue. Blocking call on futex() */
672 futex_nto1_wait(&relay_cmd_queue.futex);
673 }
674
675error:
676 DBG("Dispatch thread dying");
677 stop_threads();
678 return NULL;
679}
680
681/*
682 * Return the realpath(3) of the path even if the last directory token does not
683 * exist. For example, with /tmp/test1/test2, if test2/ does not exist but the
684 * /tmp/test1 does, the real path is returned. In normal time, realpath(3)
685 * fails if the end point directory does not exist.
686 */
687static
688char *expand_full_path(const char *path)
689{
690 const char *end_path = path;
095a4ae5 691 char *next, *cut_path, *expanded_path, *respath;
b8aa1682
JD
692
693 /* Find last token delimited by '/' */
694 while ((next = strpbrk(end_path + 1, "/"))) {
695 end_path = next;
696 }
697
698 /* Cut last token from original path */
699 cut_path = strndup(path, end_path - path);
700
701 expanded_path = malloc(PATH_MAX);
702 if (expanded_path == NULL) {
095a4ae5
MD
703 respath = NULL;
704 goto end;
b8aa1682
JD
705 }
706
095a4ae5
MD
707 respath = realpath(cut_path, expanded_path);
708 if (respath == NULL) {
b8aa1682
JD
709 switch (errno) {
710 case ENOENT:
711 ERR("%s: No such file or directory", cut_path);
712 break;
713 default:
714 PERROR("realpath");
715 break;
716 }
095a4ae5
MD
717 free(expanded_path);
718 } else {
719 /* Add end part to expanded path */
720 strcat(respath, end_path);
b8aa1682 721 }
095a4ae5 722end:
b8aa1682 723 free(cut_path);
095a4ae5 724 return respath;
b8aa1682
JD
725}
726
727
728/*
729 * config_get_default_path
730 *
731 * Returns the HOME directory path. Caller MUST NOT free(3) the return pointer.
732 */
733static
734char *config_get_default_path(void)
735{
736 return getenv("HOME");
737}
738
739/*
740 * Create recursively directory using the FULL path.
741 */
742static
743int mkdir_recursive(char *path, mode_t mode)
744{
745 char *p, tmp[PATH_MAX];
746 struct stat statbuf;
747 size_t len;
748 int ret;
749
750 ret = snprintf(tmp, sizeof(tmp), "%s", path);
751 if (ret < 0) {
752 PERROR("snprintf mkdir");
753 goto error;
754 }
755
756 len = ret;
757 if (tmp[len - 1] == '/') {
758 tmp[len - 1] = 0;
759 }
760
761 for (p = tmp + 1; *p; p++) {
762 if (*p == '/') {
763 *p = 0;
764 if (tmp[strlen(tmp) - 1] == '.' &&
765 tmp[strlen(tmp) - 2] == '.' &&
766 tmp[strlen(tmp) - 3] == '/') {
767 ERR("Using '/../' is not permitted in the trace path (%s)",
768 tmp);
769 ret = -1;
770 goto error;
771 }
772 ret = stat(tmp, &statbuf);
773 if (ret < 0) {
774 ret = mkdir(tmp, mode);
775 if (ret < 0) {
095a4ae5 776 if (errno != EEXIST) {
b8aa1682
JD
777 PERROR("mkdir recursive");
778 ret = -errno;
779 goto error;
780 }
781 }
782 }
783 *p = '/';
784 }
785 }
786
787 ret = mkdir(tmp, mode);
788 if (ret < 0) {
095a4ae5 789 if (errno != EEXIST) {
b8aa1682
JD
790 PERROR("mkdir recursive last piece");
791 ret = -errno;
792 } else {
793 ret = 0;
794 }
795 }
796
797error:
798 return ret;
799}
800
b8aa1682 801static
095a4ae5 802char *create_output_path_auto(char *path_name)
b8aa1682 803{
095a4ae5 804 int ret;
b8aa1682 805 char *traces_path = NULL;
095a4ae5
MD
806 char *alloc_path = NULL;
807 char *default_path;
b8aa1682 808
095a4ae5
MD
809 default_path = config_get_default_path();
810 if (default_path == NULL) {
811 ERR("Home path not found.\n \
812 Please specify an output path using -o, --output PATH");
813 goto exit;
814 }
815 alloc_path = strdup(default_path);
816 if (alloc_path == NULL) {
817 PERROR("Path allocation");
818 goto exit;
819 }
820 ret = asprintf(&traces_path, "%s/" DEFAULT_TRACE_DIR_NAME
821 "/%s", alloc_path, path_name);
822 if (ret < 0) {
823 PERROR("asprintf trace dir name");
824 goto exit;
b8aa1682 825 }
095a4ae5 826exit:
b8aa1682 827 free(alloc_path);
095a4ae5
MD
828 return traces_path;
829}
830
831static
832char *create_output_path_noauto(char *path_name)
833{
834 int ret;
835 char *traces_path = NULL;
836 char *full_path;
b8aa1682 837
095a4ae5
MD
838 full_path = expand_full_path(opt_output_path);
839 ret = asprintf(&traces_path, "%s/%s", full_path, path_name);
840 if (ret < 0) {
841 PERROR("asprintf trace dir name");
842 goto exit;
843 }
b8aa1682 844exit:
095a4ae5 845 free(full_path);
b8aa1682
JD
846 return traces_path;
847}
848
095a4ae5
MD
849/*
850 * create_output_path: create the output trace directory
851 */
852static
853char *create_output_path(char *path_name)
854{
855 if (opt_output_path == NULL) {
856 return create_output_path_auto(path_name);
857 } else {
858 return create_output_path_noauto(path_name);
859 }
860}
861
9d1bbf21
MD
862static
863void deferred_free_stream(struct rcu_head *head)
864{
865 struct relay_stream *stream =
866 caa_container_of(head, struct relay_stream, rcu_node);
867 free(stream);
868}
869
b8aa1682
JD
870/*
871 * relay_delete_session: Free all memory associated with a session and
872 * close all the FDs
873 */
874static
875void relay_delete_session(struct relay_command *cmd, struct lttng_ht *streams_ht)
876{
877 struct lttng_ht_iter iter;
878 struct lttng_ht_node_ulong *node;
879 struct relay_stream *stream;
880 int ret;
881
095a4ae5 882 if (!cmd->session) {
b8aa1682 883 return;
095a4ae5 884 }
b8aa1682 885
77c7c900 886 DBG("Relay deleting session %" PRIu64, cmd->session->id);
b8aa1682
JD
887 free(cmd->session->sock);
888
9d1bbf21 889 rcu_read_lock();
b8aa1682
JD
890 cds_lfht_for_each_entry(streams_ht->ht, &iter.iter, node, node) {
891 node = lttng_ht_iter_get_node_ulong(&iter);
892 if (node) {
893 stream = caa_container_of(node,
894 struct relay_stream, stream_n);
895 if (stream->session == cmd->session) {
896 close(stream->fd);
897 ret = lttng_ht_del(streams_ht, &iter);
898 assert(!ret);
9d1bbf21
MD
899 call_rcu(&stream->rcu_node,
900 deferred_free_stream);
b8aa1682
JD
901 }
902 }
903 }
9d1bbf21 904 rcu_read_unlock();
b8aa1682
JD
905}
906
907/*
908 * relay_add_stream: allocate a new stream for a session
909 */
910static
911int relay_add_stream(struct lttcomm_relayd_hdr *recv_hdr,
912 struct relay_command *cmd, struct lttng_ht *streams_ht)
913{
914 struct relay_session *session = cmd->session;
915 struct lttcomm_relayd_add_stream stream_info;
916 struct relay_stream *stream = NULL;
917 struct lttcomm_relayd_status_stream reply;
918 char *path = NULL, *root_path = NULL;
919 int ret, send_ret;
920
921 if (!session || session->version_check_done == 0) {
922 ERR("Trying to add a stream before version check");
923 ret = -1;
924 goto end_no_session;
925 }
926
927 /* FIXME : use data_size for something ? */
928 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
929 sizeof(struct lttcomm_relayd_add_stream), MSG_WAITALL);
930 if (ret < sizeof(struct lttcomm_relayd_add_stream)) {
931 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
932 ret = -1;
933 goto end_no_session;
934 }
935 stream = zmalloc(sizeof(struct relay_stream));
936 if (stream == NULL) {
937 PERROR("relay stream zmalloc");
938 ret = -1;
939 goto end_no_session;
940 }
941
9d1bbf21 942 rcu_read_lock();
b8aa1682 943 stream->stream_handle = ++last_relay_stream_id;
173af62f 944 stream->prev_seq = -1ULL;
b8aa1682
JD
945 stream->session = session;
946
947 root_path = create_output_path(stream_info.pathname);
948 if (!root_path) {
949 ret = -1;
950 goto end;
951 }
952 ret = mkdir_recursive(root_path, S_IRWXU | S_IRWXG);
953 if (ret < 0) {
b8aa1682
JD
954 ERR("relay creating output directory");
955 goto end;
956 }
957
958 ret = asprintf(&path, "%s/%s", root_path, stream_info.channel_name);
959 if (ret < 0) {
960 PERROR("asprintf stream path");
961 goto end;
962 }
963
964 ret = open(path, O_WRONLY|O_CREAT|O_TRUNC, S_IRWXU|S_IRWXG|S_IRWXO);
965 if (ret < 0) {
966 PERROR("Relay creating trace file");
967 goto end;
968 }
969
970 stream->fd = ret;
971 DBG("Tracefile %s created", path);
972
973 lttng_ht_node_init_ulong(&stream->stream_n,
974 (unsigned long) stream->stream_handle);
975 lttng_ht_add_unique_ulong(streams_ht,
976 &stream->stream_n);
977
978 DBG("Relay new stream added %s", stream_info.channel_name);
979
980end:
981 free(path);
982 free(root_path);
983 /* send the session id to the client or a negative return code on error */
984 if (ret < 0) {
f73fabfd 985 reply.ret_code = htobe32(LTTNG_ERR_UNK);
b8aa1682 986 } else {
f73fabfd 987 reply.ret_code = htobe32(LTTNG_OK);
b8aa1682
JD
988 }
989 reply.handle = htobe64(stream->stream_handle);
990 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
991 sizeof(struct lttcomm_relayd_status_stream), 0);
992 if (send_ret < 0) {
993 ERR("Relay sending stream id");
994 }
9d1bbf21 995 rcu_read_unlock();
b8aa1682
JD
996
997end_no_session:
998 return ret;
999}
1000
173af62f
DG
1001/*
1002 * relay_close_stream: close a specific stream
1003 */
1004static
1005int relay_close_stream(struct lttcomm_relayd_hdr *recv_hdr,
1006 struct relay_command *cmd, struct lttng_ht *streams_ht)
1007{
1008 struct relay_session *session = cmd->session;
1009 struct lttcomm_relayd_close_stream stream_info;
1010 struct lttcomm_relayd_generic_reply reply;
1011 struct relay_stream *stream;
1012 int ret, send_ret;
1013 struct lttng_ht_node_ulong *node;
1014 struct lttng_ht_iter iter;
1015
1016 DBG("Close stream received");
1017
1018 if (!session || session->version_check_done == 0) {
1019 ERR("Trying to close a stream before version check");
1020 ret = -1;
1021 goto end_no_session;
1022 }
1023
1024 ret = cmd->sock->ops->recvmsg(cmd->sock, &stream_info,
1025 sizeof(struct lttcomm_relayd_close_stream), MSG_WAITALL);
1026 if (ret < sizeof(struct lttcomm_relayd_close_stream)) {
1027 ERR("Relay didn't receive valid add_stream struct size : %d", ret);
1028 ret = -1;
1029 goto end_no_session;
1030 }
1031
1032 rcu_read_lock();
1033 lttng_ht_lookup(streams_ht,
1034 (void *)((unsigned long) be64toh(stream_info.stream_id)),
1035 &iter);
1036 node = lttng_ht_iter_get_node_ulong(&iter);
1037 if (node == NULL) {
77c7c900 1038 DBG("Relay stream %" PRIu64 " not found", be64toh(stream_info.stream_id));
173af62f
DG
1039 ret = -1;
1040 goto end_unlock;
1041 }
1042
1043 stream = caa_container_of(node, struct relay_stream, stream_n);
1044 if (!stream) {
1045 ret = -1;
1046 goto end_unlock;
1047 }
1048
8e2583a4 1049 stream->last_net_seq_num = be64toh(stream_info.last_net_seq_num);
173af62f
DG
1050 stream->close_flag = 1;
1051
1052 if (close_stream_check(stream)) {
1053 int delret;
1054
1055 close(stream->fd);
1056 delret = lttng_ht_del(streams_ht, &iter);
1057 assert(!delret);
1058 call_rcu(&stream->rcu_node,
1059 deferred_free_stream);
1060 DBG("Closed tracefile %d from close stream", stream->fd);
1061 }
1062
1063end_unlock:
1064 rcu_read_unlock();
1065
1066 if (ret < 0) {
f73fabfd 1067 reply.ret_code = htobe32(LTTNG_ERR_UNK);
173af62f 1068 } else {
f73fabfd 1069 reply.ret_code = htobe32(LTTNG_OK);
173af62f
DG
1070 }
1071 send_ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1072 sizeof(struct lttcomm_relayd_generic_reply), 0);
1073 if (send_ret < 0) {
1074 ERR("Relay sending stream id");
1075 }
1076
1077end_no_session:
1078 return ret;
1079}
1080
b8aa1682
JD
1081/*
1082 * relay_unknown_command: send -1 if received unknown command
1083 */
1084static
1085void relay_unknown_command(struct relay_command *cmd)
1086{
1087 struct lttcomm_relayd_generic_reply reply;
1088 int ret;
1089
f73fabfd 1090 reply.ret_code = htobe32(LTTNG_ERR_UNK);
b8aa1682
JD
1091 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1092 sizeof(struct lttcomm_relayd_generic_reply), 0);
1093 if (ret < 0) {
1094 ERR("Relay sending unknown command");
1095 }
1096}
1097
1098/*
1099 * relay_start: send an acknowledgment to the client to tell if we are
1100 * ready to receive data. We are ready if a session is established.
1101 */
1102static
1103int relay_start(struct lttcomm_relayd_hdr *recv_hdr,
1104 struct relay_command *cmd)
1105{
f73fabfd 1106 int ret = htobe32(LTTNG_OK);
b8aa1682
JD
1107 struct lttcomm_relayd_generic_reply reply;
1108 struct relay_session *session = cmd->session;
1109
1110 if (!session) {
1111 DBG("Trying to start the streaming without a session established");
f73fabfd 1112 ret = htobe32(LTTNG_ERR_UNK);
b8aa1682
JD
1113 }
1114
1115 reply.ret_code = ret;
1116 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1117 sizeof(struct lttcomm_relayd_generic_reply), 0);
1118 if (ret < 0) {
1119 ERR("Relay sending start ack");
1120 }
1121
1122 return ret;
1123}
1124
1125/*
1126 * Get stream from stream id.
9d1bbf21 1127 * Need to be called with RCU read-side lock held.
b8aa1682
JD
1128 */
1129static
1130struct relay_stream *relay_stream_from_stream_id(uint64_t stream_id,
1131 struct lttng_ht *streams_ht)
1132{
1133 struct lttng_ht_node_ulong *node;
1134 struct lttng_ht_iter iter;
1135 struct relay_stream *ret;
1136
1137 lttng_ht_lookup(streams_ht,
1138 (void *)((unsigned long) stream_id),
1139 &iter);
1140 node = lttng_ht_iter_get_node_ulong(&iter);
1141 if (node == NULL) {
77c7c900 1142 DBG("Relay stream %" PRIu64 " not found", stream_id);
b8aa1682
JD
1143 ret = NULL;
1144 goto end;
1145 }
1146
1147 ret = caa_container_of(node, struct relay_stream, stream_n);
1148
1149end:
1150 return ret;
1151}
1152
1d4dfdef
DG
1153/*
1154 * Append padding to the file pointed by the file descriptor fd.
1155 */
1156static int write_padding_to_file(int fd, uint32_t size)
1157{
1158 int ret = 0;
1159 char *zeros;
1160
1161 if (size == 0) {
1162 goto end;
1163 }
1164
1165 zeros = zmalloc(size);
1166 if (zeros == NULL) {
1167 PERROR("zmalloc zeros for padding");
1168 ret = -1;
1169 goto end;
1170 }
1171
1172 do {
1173 ret = write(fd, zeros, size);
1174 } while (ret < 0 && errno == EINTR);
1175 if (ret < 0) {
1176 PERROR("write padding to file");
1177 }
1178
1179end:
1180 return ret;
1181}
1182
b8aa1682
JD
1183/*
1184 * relay_recv_metadata: receive the metada for the session.
1185 */
1186static
1187int relay_recv_metadata(struct lttcomm_relayd_hdr *recv_hdr,
1188 struct relay_command *cmd, struct lttng_ht *streams_ht)
1189{
f73fabfd 1190 int ret = htobe32(LTTNG_OK);
b8aa1682
JD
1191 struct relay_session *session = cmd->session;
1192 struct lttcomm_relayd_metadata_payload *metadata_struct;
1193 struct relay_stream *metadata_stream;
1194 uint64_t data_size, payload_size;
1195
1196 if (!session) {
1197 ERR("Metadata sent before version check");
1198 ret = -1;
1199 goto end;
1200 }
1201
f6416125
MD
1202 data_size = payload_size = be64toh(recv_hdr->data_size);
1203 if (data_size < sizeof(struct lttcomm_relayd_metadata_payload)) {
1204 ERR("Incorrect data size");
1205 ret = -1;
1206 goto end;
1207 }
1208 payload_size -= sizeof(struct lttcomm_relayd_metadata_payload);
1209
b8aa1682
JD
1210 if (data_buffer_size < data_size) {
1211 data_buffer = realloc(data_buffer, data_size);
1212 if (!data_buffer) {
1213 ERR("Allocating data buffer");
1214 ret = -1;
1215 goto end;
1216 }
1217 data_buffer_size = data_size;
1218 }
1219 memset(data_buffer, 0, data_size);
77c7c900 1220 DBG2("Relay receiving metadata, waiting for %" PRIu64 " bytes", data_size);
87c1611d
DG
1221 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size,
1222 MSG_WAITALL);
b8aa1682
JD
1223 if (ret < 0 || ret != data_size) {
1224 ret = -1;
1225 ERR("Relay didn't receive the whole metadata");
1226 goto end;
1227 }
1228 metadata_struct = (struct lttcomm_relayd_metadata_payload *) data_buffer;
9d1bbf21
MD
1229
1230 rcu_read_lock();
b8aa1682
JD
1231 metadata_stream = relay_stream_from_stream_id(
1232 be64toh(metadata_struct->stream_id), streams_ht);
1233 if (!metadata_stream) {
1234 ret = -1;
9d1bbf21 1235 goto end_unlock;
b8aa1682
JD
1236 }
1237
6f94560a
MD
1238 do {
1239 ret = write(metadata_stream->fd, metadata_struct->payload,
1240 payload_size);
1241 } while (ret < 0 && errno == EINTR);
87c1611d 1242 if (ret < payload_size) {
b8aa1682
JD
1243 ERR("Relay error writing metadata on file");
1244 ret = -1;
9d1bbf21 1245 goto end_unlock;
b8aa1682 1246 }
1d4dfdef
DG
1247
1248 ret = write_padding_to_file(metadata_stream->fd,
1249 be32toh(metadata_struct->padding_size));
1250 if (ret < 0) {
1251 goto end_unlock;
1252 }
1253
b8aa1682
JD
1254 DBG2("Relay metadata written");
1255
9d1bbf21 1256end_unlock:
6e3c5836 1257 rcu_read_unlock();
b8aa1682
JD
1258end:
1259 return ret;
1260}
1261
1262/*
1263 * relay_send_version: send relayd version number
1264 */
1265static
1266int relay_send_version(struct lttcomm_relayd_hdr *recv_hdr,
1267 struct relay_command *cmd)
1268{
7f51dcba 1269 int ret;
b8aa1682 1270 struct lttcomm_relayd_version reply;
438b505e 1271 struct relay_session *session;
b8aa1682
JD
1272
1273 if (cmd->session == NULL) {
1274 session = zmalloc(sizeof(struct relay_session));
1275 if (session == NULL) {
1276 PERROR("relay session zmalloc");
1277 ret = -1;
1278 goto end;
1279 }
1280 session->id = ++last_relay_session_id;
77c7c900 1281 DBG("Created session %" PRIu64, session->id);
b8aa1682 1282 cmd->session = session;
438b505e
MD
1283 } else {
1284 session = cmd->session;
b8aa1682
JD
1285 }
1286 session->version_check_done = 1;
1287
7f51dcba
MD
1288 ret = sscanf(VERSION, "%u.%u", &reply.major, &reply.minor);
1289 if (ret < 2) {
1290 ERR("Error in scanning version");
1291 ret = -1;
1292 goto end;
1293 }
b8aa1682
JD
1294 reply.major = htobe32(reply.major);
1295 reply.minor = htobe32(reply.minor);
1296 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply,
1297 sizeof(struct lttcomm_relayd_version), 0);
1298 if (ret < 0) {
1299 ERR("Relay sending version");
1300 }
0a6b5085
DG
1301 DBG("Version check done (%u.%u)", be32toh(reply.major),
1302 be32toh(reply.minor));
b8aa1682
JD
1303
1304end:
1305 return ret;
1306}
1307
c8f59ee5
DG
1308/*
1309 * Check for data availability for a given stream id from the session daemon.
1310 */
1311static
1312int relay_data_available(struct lttcomm_relayd_hdr *recv_hdr,
1313 struct relay_command *cmd, struct lttng_ht *streams_ht)
1314{
1315 struct relay_session *session = cmd->session;
1316 struct lttcomm_relayd_data_available msg;
1317 struct lttcomm_relayd_generic_reply reply;
1318 struct relay_stream *stream;
1319 int ret;
1320 struct lttng_ht_node_ulong *node;
1321 struct lttng_ht_iter iter;
1322 uint64_t last_net_seq_num, stream_id;
1323
1324 DBG("Data available command received");
1325
1326 if (!session || session->version_check_done == 0) {
1327 ERR("Trying to check for data before version check");
1328 ret = -1;
1329 goto end_no_session;
1330 }
1331
1332 ret = cmd->sock->ops->recvmsg(cmd->sock, &msg, sizeof(msg), MSG_WAITALL);
1333 if (ret < sizeof(msg)) {
1334 ERR("Relay didn't receive valid data_available struct size : %d", ret);
1335 ret = -1;
1336 goto end_no_session;
1337 }
1338
1339 stream_id = be64toh(msg.stream_id);
1340 last_net_seq_num = be64toh(msg.last_net_seq_num);
1341
1342 rcu_read_lock();
1343 lttng_ht_lookup(streams_ht, (void *)((unsigned long) stream_id), &iter);
1344 node = lttng_ht_iter_get_node_ulong(&iter);
1345 if (node == NULL) {
1346 DBG("Relay stream %" PRIu64 " not found", stream_id);
1347 ret = -1;
1348 goto end_unlock;
1349 }
1350
1351 stream = caa_container_of(node, struct relay_stream, stream_n);
1352 assert(stream);
1353
1354 DBG("Data available for stream id %" PRIu64 " prev_seq %" PRIu64
1355 " and last_seq %" PRIu64, stream_id, stream->prev_seq,
1356 last_net_seq_num);
1357
1358 if (stream->prev_seq == -1UL || stream->prev_seq <= last_net_seq_num) {
1359 /* Data has in fact been written and is available */
1360 ret = 1;
1361 } else {
1362 /* Data still being streamed. */
1363 ret = 0;
1364 }
1365
1366end_unlock:
1367 rcu_read_unlock();
1368
1369 reply.ret_code = htobe32(ret);
1370 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1371 if (ret < 0) {
1372 ERR("Relay data available ret code failed");
1373 }
1374
1375end_no_session:
1376 return ret;
1377}
1378
1379/*
1380 * Wait for the control socket to reach a quiescent state.
1381 *
1382 * Note that for now, when receiving this command from the session daemon, this
1383 * means that every subsequent commands or data received on the control socket
1384 * has been handled. So, this is why we simply return OK here.
1385 */
1386static
1387int relay_quiescent_control(struct lttcomm_relayd_hdr *recv_hdr,
1388 struct relay_command *cmd)
1389{
1390 int ret;
1391 struct lttcomm_relayd_generic_reply reply;
1392
1393 DBG("Checking quiescent state on control socket");
1394
1395 reply.ret_code = htobe32(LTTNG_OK);
1396 ret = cmd->sock->ops->sendmsg(cmd->sock, &reply, sizeof(reply), 0);
1397 if (ret < 0) {
1398 ERR("Relay data available ret code failed");
1399 }
1400
1401 return ret;
1402}
1403
b8aa1682
JD
1404/*
1405 * relay_process_control: Process the commands received on the control socket
1406 */
1407static
1408int relay_process_control(struct lttcomm_relayd_hdr *recv_hdr,
1409 struct relay_command *cmd, struct lttng_ht *streams_ht)
1410{
1411 int ret = 0;
1412
1413 switch (be32toh(recv_hdr->cmd)) {
1414 /*
1415 case RELAYD_CREATE_SESSION:
1416 ret = relay_create_session(recv_hdr, cmd);
1417 break;
1418 */
1419 case RELAYD_ADD_STREAM:
1420 ret = relay_add_stream(recv_hdr, cmd, streams_ht);
1421 break;
1422 case RELAYD_START_DATA:
1423 ret = relay_start(recv_hdr, cmd);
1424 break;
1425 case RELAYD_SEND_METADATA:
1426 ret = relay_recv_metadata(recv_hdr, cmd, streams_ht);
1427 break;
1428 case RELAYD_VERSION:
1429 ret = relay_send_version(recv_hdr, cmd);
1430 break;
173af62f
DG
1431 case RELAYD_CLOSE_STREAM:
1432 ret = relay_close_stream(recv_hdr, cmd, streams_ht);
1433 break;
c8f59ee5
DG
1434 case RELAYD_DATA_AVAILABLE:
1435 ret = relay_data_available(recv_hdr, cmd, streams_ht);
1436 break;
1437 case RELAYD_QUIESCENT_CONTROL:
1438 ret = relay_quiescent_control(recv_hdr, cmd);
1439 break;
b8aa1682
JD
1440 case RELAYD_UPDATE_SYNC_INFO:
1441 default:
1442 ERR("Received unknown command (%u)", be32toh(recv_hdr->cmd));
1443 relay_unknown_command(cmd);
1444 ret = -1;
1445 goto end;
1446 }
1447
1448end:
1449 return ret;
1450}
1451
1452/*
1453 * relay_process_data: Process the data received on the data socket
1454 */
1455static
1456int relay_process_data(struct relay_command *cmd, struct lttng_ht *streams_ht)
1457{
1458 int ret = 0;
1459 struct relay_stream *stream;
1460 struct lttcomm_relayd_data_hdr data_hdr;
1461 uint64_t stream_id;
173af62f 1462 uint64_t net_seq_num;
b8aa1682
JD
1463 uint32_t data_size;
1464
1465 ret = cmd->sock->ops->recvmsg(cmd->sock, &data_hdr,
1466 sizeof(struct lttcomm_relayd_data_hdr), MSG_WAITALL);
1467 if (ret <= 0) {
1468 ERR("Connections seems to be closed");
1469 ret = -1;
1470 goto end;
1471 }
1472
1473 stream_id = be64toh(data_hdr.stream_id);
9d1bbf21
MD
1474
1475 rcu_read_lock();
b8aa1682
JD
1476 stream = relay_stream_from_stream_id(stream_id, streams_ht);
1477 if (!stream) {
1478 ret = -1;
9d1bbf21 1479 goto end_unlock;
b8aa1682
JD
1480 }
1481
1482 data_size = be32toh(data_hdr.data_size);
1483 if (data_buffer_size < data_size) {
1484 data_buffer = realloc(data_buffer, data_size);
1485 if (!data_buffer) {
1486 ERR("Allocating data buffer");
1487 ret = -1;
9d1bbf21 1488 goto end_unlock;
b8aa1682
JD
1489 }
1490 data_buffer_size = data_size;
1491 }
1492 memset(data_buffer, 0, data_size);
1493
173af62f
DG
1494 net_seq_num = be64toh(data_hdr.net_seq_num);
1495
77c7c900 1496 DBG3("Receiving data of size %u for stream id %" PRIu64 " seqnum %" PRIu64,
173af62f 1497 data_size, stream_id, net_seq_num);
b8aa1682
JD
1498 ret = cmd->sock->ops->recvmsg(cmd->sock, data_buffer, data_size, MSG_WAITALL);
1499 if (ret <= 0) {
1500 ret = -1;
9d1bbf21 1501 goto end_unlock;
b8aa1682
JD
1502 }
1503
6f94560a
MD
1504 do {
1505 ret = write(stream->fd, data_buffer, data_size);
1506 } while (ret < 0 && errno == EINTR);
b8aa1682
JD
1507 if (ret < data_size) {
1508 ERR("Relay error writing data to file");
1509 ret = -1;
9d1bbf21 1510 goto end_unlock;
b8aa1682 1511 }
1d4dfdef 1512
5ab7344e
JD
1513 DBG2("Relay wrote %d bytes to tracefile for stream id %" PRIu64,
1514 ret, stream->stream_handle);
1515
1d4dfdef
DG
1516 ret = write_padding_to_file(stream->fd, be32toh(data_hdr.padding_size));
1517 if (ret < 0) {
1518 goto end_unlock;
1519 }
1520
173af62f
DG
1521 stream->prev_seq = net_seq_num;
1522
1523 /* Check if we need to close the FD */
1524 if (close_stream_check(stream)) {
1525 struct lttng_ht_iter iter;
1526
1527 close(stream->fd);
1528 iter.iter.node = &stream->stream_n.node;
1529 ret = lttng_ht_del(streams_ht, &iter);
1530 assert(!ret);
1531 call_rcu(&stream->rcu_node,
1532 deferred_free_stream);
1533 DBG("Closed tracefile %d after recv data", stream->fd);
1534 }
1535
9d1bbf21
MD
1536end_unlock:
1537 rcu_read_unlock();
b8aa1682
JD
1538end:
1539 return ret;
1540}
1541
1542static
9d1bbf21 1543void relay_cleanup_poll_connection(struct lttng_poll_event *events, int pollfd)
b8aa1682
JD
1544{
1545 int ret;
1546
b8aa1682
JD
1547 lttng_poll_del(events, pollfd);
1548
1549 ret = close(pollfd);
1550 if (ret < 0) {
1551 ERR("Closing pollfd %d", pollfd);
1552 }
1553}
1554
1555static
1556int relay_add_connection(int fd, struct lttng_poll_event *events,
1557 struct lttng_ht *relay_connections_ht)
1558{
b8aa1682 1559 struct relay_command *relay_connection;
9d1bbf21 1560 int ret;
b8aa1682
JD
1561
1562 relay_connection = zmalloc(sizeof(struct relay_command));
1563 if (relay_connection == NULL) {
1564 PERROR("Relay command zmalloc");
9d1bbf21 1565 goto error;
b8aa1682
JD
1566 }
1567 ret = read(fd, relay_connection, sizeof(struct relay_command));
cdf0f8fc 1568 if (ret < 0 || ret < sizeof(struct relay_command)) {
b8aa1682 1569 PERROR("read relay cmd pipe");
9d1bbf21 1570 goto error_read;
b8aa1682
JD
1571 }
1572
1573 lttng_ht_node_init_ulong(&relay_connection->sock_n,
1574 (unsigned long) relay_connection->sock->fd);
9d1bbf21 1575 rcu_read_lock();
b8aa1682
JD
1576 lttng_ht_add_unique_ulong(relay_connections_ht,
1577 &relay_connection->sock_n);
9d1bbf21
MD
1578 rcu_read_unlock();
1579 return lttng_poll_add(events,
b8aa1682
JD
1580 relay_connection->sock->fd,
1581 LPOLLIN | LPOLLRDHUP);
1582
9d1bbf21
MD
1583error_read:
1584 free(relay_connection);
1585error:
1586 return -1;
1587}
1588
1589static
1590void deferred_free_connection(struct rcu_head *head)
1591{
1592 struct relay_command *relay_connection =
1593 caa_container_of(head, struct relay_command, rcu_node);
1594 free(relay_connection);
1595}
1596
1597static
1598void relay_del_connection(struct lttng_ht *relay_connections_ht,
1599 struct lttng_ht *streams_ht, struct lttng_ht_iter *iter,
1600 struct relay_command *relay_connection)
1601{
1602 int ret;
1603
1604 ret = lttng_ht_del(relay_connections_ht, iter);
1605 assert(!ret);
1606 if (relay_connection->type == RELAY_CONTROL) {
1607 relay_delete_session(relay_connection, streams_ht);
1608 }
1609 call_rcu(&relay_connection->rcu_node,
1610 deferred_free_connection);
b8aa1682
JD
1611}
1612
1613/*
1614 * This thread does the actual work
1615 */
1616static
1617void *relay_thread_worker(void *data)
1618{
095a4ae5 1619 int i, ret, pollfd, err = -1;
b8aa1682
JD
1620 uint32_t revents, nb_fd;
1621 struct relay_command *relay_connection;
1622 struct lttng_poll_event events;
1623 struct lttng_ht *relay_connections_ht;
1624 struct lttng_ht_node_ulong *node;
1625 struct lttng_ht_iter iter;
1626 struct lttng_ht *streams_ht;
1627 struct lttcomm_relayd_hdr recv_hdr;
1628
1629 DBG("[thread] Relay worker started");
1630
9d1bbf21
MD
1631 rcu_register_thread();
1632
b8aa1682
JD
1633 /* table of connections indexed on socket */
1634 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
095a4ae5
MD
1635 if (!relay_connections_ht) {
1636 goto relay_connections_ht_error;
1637 }
b8aa1682
JD
1638
1639 /* tables of streams indexed by stream ID */
1640 streams_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
095a4ae5
MD
1641 if (!streams_ht) {
1642 goto streams_ht_error;
1643 }
b8aa1682
JD
1644
1645 ret = create_thread_poll_set(&events, 2);
1646 if (ret < 0) {
1647 goto error_poll_create;
1648 }
1649
1650 ret = lttng_poll_add(&events, relay_cmd_pipe[0], LPOLLIN | LPOLLRDHUP);
1651 if (ret < 0) {
1652 goto error;
1653 }
1654
1655 while (1) {
1656 /* Zeroed the events structure */
1657 lttng_poll_reset(&events);
1658
1659 nb_fd = LTTNG_POLL_GETNB(&events);
1660
1661 /* Infinite blocking call, waiting for transmission */
1662 restart:
87c1611d 1663 DBG3("Relayd worker thread polling...");
b8aa1682
JD
1664 ret = lttng_poll_wait(&events, -1);
1665 if (ret < 0) {
1666 /*
1667 * Restart interrupted system call.
1668 */
1669 if (errno == EINTR) {
1670 goto restart;
1671 }
1672 goto error;
1673 }
1674
1675 for (i = 0; i < nb_fd; i++) {
1676 /* Fetch once the poll data */
1677 revents = LTTNG_POLL_GETEV(&events, i);
1678 pollfd = LTTNG_POLL_GETFD(&events, i);
1679
1680 /* Thread quit pipe has been closed. Killing thread. */
1681 ret = check_thread_quit_pipe(pollfd, revents);
1682 if (ret) {
095a4ae5
MD
1683 err = 0;
1684 goto exit;
b8aa1682
JD
1685 }
1686
1687 /* Inspect the relay cmd pipe for new connection */
1688 if (pollfd == relay_cmd_pipe[0]) {
1689 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1690 ERR("Relay pipe error");
1691 goto error;
1692 } else if (revents & LPOLLIN) {
1693 DBG("Relay command received");
1694 ret = relay_add_connection(relay_cmd_pipe[0],
1695 &events, relay_connections_ht);
1696 if (ret < 0) {
1697 goto error;
1698 }
1699 }
1700 } else if (revents > 0) {
9d1bbf21 1701 rcu_read_lock();
b8aa1682
JD
1702 lttng_ht_lookup(relay_connections_ht,
1703 (void *)((unsigned long) pollfd),
1704 &iter);
1705 node = lttng_ht_iter_get_node_ulong(&iter);
1706 if (node == NULL) {
1707 DBG2("Relay sock %d not found", pollfd);
9d1bbf21 1708 rcu_read_unlock();
b8aa1682
JD
1709 goto error;
1710 }
1711 relay_connection = caa_container_of(node,
1712 struct relay_command, sock_n);
1713
1714 if (revents & (LPOLLERR)) {
1715 ERR("POLL ERROR");
9d1bbf21
MD
1716 relay_cleanup_poll_connection(&events, pollfd);
1717 relay_del_connection(relay_connections_ht,
1718 streams_ht, &iter,
1719 relay_connection);
b8aa1682
JD
1720 } else if (revents & (LPOLLHUP | LPOLLRDHUP)) {
1721 DBG("Socket %d hung up", pollfd);
9d1bbf21
MD
1722 relay_cleanup_poll_connection(&events, pollfd);
1723 relay_del_connection(relay_connections_ht,
1724 streams_ht, &iter,
1725 relay_connection);
b8aa1682
JD
1726 } else if (revents & LPOLLIN) {
1727 /* control socket */
1728 if (relay_connection->type == RELAY_CONTROL) {
1729 ret = relay_connection->sock->ops->recvmsg(
1730 relay_connection->sock, &recv_hdr,
1731 sizeof(struct lttcomm_relayd_hdr), MSG_WAITALL);
1732 /* connection closed */
1733 if (ret <= 0) {
9d1bbf21
MD
1734 relay_cleanup_poll_connection(&events, pollfd);
1735 relay_del_connection(relay_connections_ht,
1736 streams_ht, &iter,
1737 relay_connection);
b8aa1682
JD
1738 DBG("Control connection closed with %d", pollfd);
1739 } else {
1740 if (relay_connection->session) {
77c7c900 1741 DBG2("Relay worker receiving data for session : %" PRIu64,
b8aa1682
JD
1742 relay_connection->session->id);
1743 }
1744 ret = relay_process_control(&recv_hdr,
1745 relay_connection,
1746 streams_ht);
1747 /*
1748 * there was an error in processing a control
1749 * command: clear the session
1750 * */
1751 if (ret < 0) {
9d1bbf21
MD
1752 relay_cleanup_poll_connection(&events, pollfd);
1753 relay_del_connection(relay_connections_ht,
1754 streams_ht, &iter,
1755 relay_connection);
b8aa1682
JD
1756 DBG("Connection closed with %d", pollfd);
1757 }
1758 }
1759 /* data socket */
1760 } else if (relay_connection->type == RELAY_DATA) {
1761 ret = relay_process_data(relay_connection, streams_ht);
1762 /* connection closed */
1763 if (ret < 0) {
9d1bbf21
MD
1764 relay_cleanup_poll_connection(&events, pollfd);
1765 relay_del_connection(relay_connections_ht,
1766 streams_ht, &iter,
1767 relay_connection);
b8aa1682
JD
1768 DBG("Data connection closed with %d", pollfd);
1769 }
1770 }
1771 }
9d1bbf21 1772 rcu_read_unlock();
b8aa1682
JD
1773 }
1774 }
1775 }
1776
095a4ae5 1777exit:
b8aa1682
JD
1778error:
1779 lttng_poll_clean(&events);
1780
1781 /* empty the hash table and free the memory */
9d1bbf21 1782 rcu_read_lock();
b8aa1682
JD
1783 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, node, node) {
1784 node = lttng_ht_iter_get_node_ulong(&iter);
1785 if (node) {
1786 relay_connection = caa_container_of(node,
1787 struct relay_command, sock_n);
9d1bbf21
MD
1788 relay_del_connection(relay_connections_ht,
1789 streams_ht, &iter,
1790 relay_connection);
b8aa1682 1791 }
b8aa1682 1792 }
9d1bbf21 1793 rcu_read_unlock();
b8aa1682 1794error_poll_create:
095a4ae5
MD
1795 lttng_ht_destroy(streams_ht);
1796streams_ht_error:
b8aa1682 1797 lttng_ht_destroy(relay_connections_ht);
095a4ae5
MD
1798relay_connections_ht_error:
1799 if (err) {
1800 DBG("Thread exited with error");
1801 }
b8aa1682 1802 DBG("Worker thread cleanup complete");
095a4ae5 1803 free(data_buffer);
b8aa1682 1804 stop_threads();
9d1bbf21 1805 rcu_unregister_thread();
b8aa1682
JD
1806 return NULL;
1807}
1808
1809/*
1810 * Create the relay command pipe to wake thread_manage_apps.
1811 * Closed in cleanup().
1812 */
1813static int create_relay_cmd_pipe(void)
1814{
a02de639 1815 int ret;
b8aa1682 1816
a02de639 1817 ret = utils_create_pipe_cloexec(relay_cmd_pipe);
b8aa1682 1818
b8aa1682
JD
1819 return ret;
1820}
1821
1822/*
1823 * main
1824 */
1825int main(int argc, char **argv)
1826{
1827 int ret = 0;
1828 void *status;
1829
1830 /* Create thread quit pipe */
1831 if ((ret = init_thread_quit_pipe()) < 0) {
1832 goto error;
1833 }
1834
1835 /* Parse arguments */
1836 progname = argv[0];
1837 if ((ret = parse_args(argc, argv) < 0)) {
a02de639 1838 goto exit;
b8aa1682
JD
1839 }
1840
1841 if ((ret = set_signal_handler()) < 0) {
1842 goto exit;
1843 }
1844
1845 /* Daemonize */
1846 if (opt_daemon) {
1847 ret = daemon(0, 0);
1848 if (ret < 0) {
1849 PERROR("daemon");
a02de639 1850 goto exit;
b8aa1682
JD
1851 }
1852 }
1853
1854 /* Check if daemon is UID = 0 */
1855 is_root = !getuid();
1856
1857 if (!is_root) {
1858 if (control_uri->port < 1024 || data_uri->port < 1024) {
1859 ERR("Need to be root to use ports < 1024");
1860 ret = -1;
a02de639 1861 goto exit;
b8aa1682
JD
1862 }
1863 }
1864
1865 /* Setup the thread apps communication pipe. */
1866 if ((ret = create_relay_cmd_pipe()) < 0) {
1867 goto exit;
1868 }
1869
1870 /* Init relay command queue. */
1871 cds_wfq_init(&relay_cmd_queue.queue);
1872
1873 /* Set up max poll set size */
1874 lttng_poll_set_max_size();
1875
1876 /* Setup the dispatcher thread */
1877 ret = pthread_create(&dispatcher_thread, NULL,
1878 relay_thread_dispatcher, (void *) NULL);
1879 if (ret != 0) {
1880 PERROR("pthread_create dispatcher");
1881 goto exit_dispatcher;
1882 }
1883
1884 /* Setup the worker thread */
1885 ret = pthread_create(&worker_thread, NULL,
1886 relay_thread_worker, (void *) NULL);
1887 if (ret != 0) {
1888 PERROR("pthread_create worker");
1889 goto exit_worker;
1890 }
1891
1892 /* Setup the listener thread */
1893 ret = pthread_create(&listener_thread, NULL,
1894 relay_thread_listener, (void *) NULL);
1895 if (ret != 0) {
1896 PERROR("pthread_create listener");
1897 goto exit_listener;
1898 }
1899
1900exit_listener:
1901 ret = pthread_join(listener_thread, &status);
1902 if (ret != 0) {
1903 PERROR("pthread_join");
1904 goto error; /* join error, exit without cleanup */
1905 }
1906
1907exit_worker:
1908 ret = pthread_join(worker_thread, &status);
1909 if (ret != 0) {
1910 PERROR("pthread_join");
1911 goto error; /* join error, exit without cleanup */
1912 }
1913
1914exit_dispatcher:
1915 ret = pthread_join(dispatcher_thread, &status);
1916 if (ret != 0) {
1917 PERROR("pthread_join");
1918 goto error; /* join error, exit without cleanup */
1919 }
1920
1921exit:
1922 cleanup();
1923 if (!ret) {
1924 exit(EXIT_SUCCESS);
1925 }
a02de639 1926
b8aa1682
JD
1927error:
1928 exit(EXIT_FAILURE);
1929}
This page took 0.125689 seconds and 4 git commands to generate.