relayd: rename stream prev_seq to prev_data_seq
[lttng-tools.git] / src / bin / lttng-relayd / live.c
1 /*
2 * Copyright (C) 2013 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
4 * 2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _LGPL_SOURCE
21 #include <getopt.h>
22 #include <grp.h>
23 #include <limits.h>
24 #include <pthread.h>
25 #include <signal.h>
26 #include <stdio.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <sys/mman.h>
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
33 #include <sys/stat.h>
34 #include <sys/types.h>
35 #include <sys/wait.h>
36 #include <inttypes.h>
37 #include <urcu/futex.h>
38 #include <urcu/uatomic.h>
39 #include <urcu/rculist.h>
40 #include <unistd.h>
41 #include <fcntl.h>
42
43 #include <lttng/lttng.h>
44 #include <common/common.h>
45 #include <common/compat/poll.h>
46 #include <common/compat/socket.h>
47 #include <common/compat/endian.h>
48 #include <common/defaults.h>
49 #include <common/futex.h>
50 #include <common/index/index.h>
51 #include <common/sessiond-comm/sessiond-comm.h>
52 #include <common/sessiond-comm/inet.h>
53 #include <common/sessiond-comm/relayd.h>
54 #include <common/uri.h>
55 #include <common/utils.h>
56
57 #include "cmd.h"
58 #include "live.h"
59 #include "lttng-relayd.h"
60 #include "utils.h"
61 #include "health-relayd.h"
62 #include "testpoint.h"
63 #include "viewer-stream.h"
64 #include "stream.h"
65 #include "session.h"
66 #include "ctf-trace.h"
67 #include "connection.h"
68 #include "viewer-session.h"
69
70 #define SESSION_BUF_DEFAULT_COUNT 16
71
72 static struct lttng_uri *live_uri;
73
74 /*
75 * This pipe is used to inform the worker thread that a command is queued and
76 * ready to be processed.
77 */
78 static int live_conn_pipe[2] = { -1, -1 };
79
80 /* Shared between threads */
81 static int live_dispatch_thread_exit;
82
83 static pthread_t live_listener_thread;
84 static pthread_t live_dispatcher_thread;
85 static pthread_t live_worker_thread;
86
87 /*
88 * Relay command queue.
89 *
90 * The live_thread_listener and live_thread_dispatcher communicate with this
91 * queue.
92 */
93 static struct relay_conn_queue viewer_conn_queue;
94
95 static uint64_t last_relay_viewer_session_id;
96 static pthread_mutex_t last_relay_viewer_session_id_lock =
97 PTHREAD_MUTEX_INITIALIZER;
98
99 /*
100 * Cleanup the daemon
101 */
102 static
103 void cleanup_relayd_live(void)
104 {
105 DBG("Cleaning up");
106
107 free(live_uri);
108 }
109
110 /*
111 * Receive a request buffer using a given socket, destination allocated buffer
112 * of length size.
113 *
114 * Return the size of the received message or else a negative value on error
115 * with errno being set by recvmsg() syscall.
116 */
117 static
118 ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
119 {
120 ssize_t ret;
121
122 ret = sock->ops->recvmsg(sock, buf, size, 0);
123 if (ret < 0 || ret != size) {
124 if (ret == 0) {
125 /* Orderly shutdown. Not necessary to print an error. */
126 DBG("Socket %d did an orderly shutdown", sock->fd);
127 } else {
128 ERR("Relay failed to receive request.");
129 }
130 ret = -1;
131 }
132
133 return ret;
134 }
135
136 /*
137 * Send a response buffer using a given socket, source allocated buffer of
138 * length size.
139 *
140 * Return the size of the sent message or else a negative value on error with
141 * errno being set by sendmsg() syscall.
142 */
143 static
144 ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
145 {
146 ssize_t ret;
147
148 ret = sock->ops->sendmsg(sock, buf, size, 0);
149 if (ret < 0) {
150 ERR("Relayd failed to send response.");
151 }
152
153 return ret;
154 }
155
156 /*
157 * Atomically check if new streams got added in one of the sessions attached
158 * and reset the flag to 0.
159 *
160 * Returns 1 if new streams got added, 0 if nothing changed, a negative value
161 * on error.
162 */
163 static
164 int check_new_streams(struct relay_connection *conn)
165 {
166 struct relay_session *session;
167 unsigned long current_val;
168 int ret = 0;
169
170 if (!conn->viewer_session) {
171 goto end;
172 }
173 rcu_read_lock();
174 cds_list_for_each_entry_rcu(session,
175 &conn->viewer_session->session_list,
176 viewer_session_node) {
177 if (!session_get(session)) {
178 continue;
179 }
180 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
181 ret = current_val;
182 session_put(session);
183 if (ret == 1) {
184 goto end;
185 }
186 }
187 end:
188 rcu_read_unlock();
189 return ret;
190 }
191
192 /*
193 * Send viewer streams to the given socket. The ignore_sent_flag indicates if
194 * this function should ignore the sent flag or not.
195 *
196 * Return 0 on success or else a negative value.
197 */
198 static
199 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
200 struct relay_session *session, unsigned int ignore_sent_flag)
201 {
202 ssize_t ret;
203 struct lttng_viewer_stream send_stream;
204 struct lttng_ht_iter iter;
205 struct relay_viewer_stream *vstream;
206
207 rcu_read_lock();
208
209 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
210 stream_n.node) {
211 struct ctf_trace *ctf_trace;
212
213 health_code_update();
214
215 if (!viewer_stream_get(vstream)) {
216 continue;
217 }
218
219 pthread_mutex_lock(&vstream->stream->lock);
220 /* Ignore if not the same session. */
221 if (vstream->stream->trace->session->id != session->id ||
222 (!ignore_sent_flag && vstream->sent_flag)) {
223 pthread_mutex_unlock(&vstream->stream->lock);
224 viewer_stream_put(vstream);
225 continue;
226 }
227
228 ctf_trace = vstream->stream->trace;
229 send_stream.id = htobe64(vstream->stream->stream_handle);
230 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
231 send_stream.metadata_flag = htobe32(
232 vstream->stream->is_metadata);
233 if (lttng_strncpy(send_stream.path_name, vstream->path_name,
234 sizeof(send_stream.path_name))) {
235 pthread_mutex_unlock(&vstream->stream->lock);
236 viewer_stream_put(vstream);
237 ret = -1; /* Error. */
238 goto end_unlock;
239 }
240 if (lttng_strncpy(send_stream.channel_name,
241 vstream->channel_name,
242 sizeof(send_stream.channel_name))) {
243 pthread_mutex_unlock(&vstream->stream->lock);
244 viewer_stream_put(vstream);
245 ret = -1; /* Error. */
246 goto end_unlock;
247 }
248
249 DBG("Sending stream %" PRIu64 " to viewer",
250 vstream->stream->stream_handle);
251 vstream->sent_flag = 1;
252 pthread_mutex_unlock(&vstream->stream->lock);
253
254 ret = send_response(sock, &send_stream, sizeof(send_stream));
255 viewer_stream_put(vstream);
256 if (ret < 0) {
257 goto end_unlock;
258 }
259 }
260
261 ret = 0;
262
263 end_unlock:
264 rcu_read_unlock();
265 return ret;
266 }
267
268 /*
269 * Create every viewer stream possible for the given session with the seek
270 * type. Three counters *can* be return which are in order the total amount of
271 * viewer stream of the session, the number of unsent stream and the number of
272 * stream created. Those counters can be NULL and thus will be ignored.
273 *
274 * Return 0 on success or else a negative value.
275 */
276 static
277 int make_viewer_streams(struct relay_session *session,
278 enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent,
279 uint32_t *nb_created, bool *closed)
280 {
281 int ret;
282 struct lttng_ht_iter iter;
283 struct ctf_trace *ctf_trace;
284
285 assert(session);
286
287 /*
288 * Hold the session lock to ensure that we see either none or
289 * all initial streams for a session, but no intermediate state.
290 */
291 pthread_mutex_lock(&session->lock);
292
293 if (session->connection_closed) {
294 *closed = true;
295 }
296
297 /*
298 * Create viewer streams for relay streams that are ready to be
299 * used for a the given session id only.
300 */
301 rcu_read_lock();
302 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
303 node.node) {
304 struct relay_stream *stream;
305
306 health_code_update();
307
308 if (!ctf_trace_get(ctf_trace)) {
309 continue;
310 }
311
312 cds_list_for_each_entry_rcu(stream, &ctf_trace->stream_list, stream_node) {
313 struct relay_viewer_stream *vstream;
314
315 if (!stream_get(stream)) {
316 continue;
317 }
318 /*
319 * stream published is protected by the session lock.
320 */
321 if (!stream->published) {
322 goto next;
323 }
324 vstream = viewer_stream_get_by_id(stream->stream_handle);
325 if (!vstream) {
326 vstream = viewer_stream_create(stream, seek_t);
327 if (!vstream) {
328 ret = -1;
329 ctf_trace_put(ctf_trace);
330 stream_put(stream);
331 goto error_unlock;
332 }
333
334 if (nb_created) {
335 /* Update number of created stream counter. */
336 (*nb_created)++;
337 }
338 /*
339 * Ensure a self-reference is preserved even
340 * after we have put our local reference.
341 */
342 if (!viewer_stream_get(vstream)) {
343 ERR("Unable to get self-reference on viewer stream, logic error.");
344 abort();
345 }
346 } else {
347 if (!vstream->sent_flag && nb_unsent) {
348 /* Update number of unsent stream counter. */
349 (*nb_unsent)++;
350 }
351 }
352 /* Update number of total stream counter. */
353 if (nb_total) {
354 if (stream->is_metadata) {
355 if (!stream->closed ||
356 stream->metadata_received > vstream->metadata_sent) {
357 (*nb_total)++;
358 }
359 } else {
360 if (!stream->closed ||
361 !(((int64_t) (stream->prev_data_seq - stream->last_net_seq_num)) >= 0)) {
362
363 (*nb_total)++;
364 }
365 }
366 }
367 /* Put local reference. */
368 viewer_stream_put(vstream);
369 next:
370 stream_put(stream);
371 }
372 ctf_trace_put(ctf_trace);
373 }
374
375 ret = 0;
376
377 error_unlock:
378 rcu_read_unlock();
379 pthread_mutex_unlock(&session->lock);
380 return ret;
381 }
382
383 int relayd_live_stop(void)
384 {
385 /* Stop dispatch thread */
386 CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
387 futex_nto1_wake(&viewer_conn_queue.futex);
388 return 0;
389 }
390
391 /*
392 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
393 */
394 static
395 int create_thread_poll_set(struct lttng_poll_event *events, int size)
396 {
397 int ret;
398
399 if (events == NULL || size == 0) {
400 ret = -1;
401 goto error;
402 }
403
404 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
405 if (ret < 0) {
406 goto error;
407 }
408
409 /* Add quit pipe */
410 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
411 if (ret < 0) {
412 goto error;
413 }
414
415 return 0;
416
417 error:
418 return ret;
419 }
420
421 /*
422 * Check if the thread quit pipe was triggered.
423 *
424 * Return 1 if it was triggered else 0;
425 */
426 static
427 int check_thread_quit_pipe(int fd, uint32_t events)
428 {
429 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
430 return 1;
431 }
432
433 return 0;
434 }
435
436 /*
437 * Create and init socket from uri.
438 */
439 static
440 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
441 {
442 int ret;
443 struct lttcomm_sock *sock = NULL;
444
445 sock = lttcomm_alloc_sock_from_uri(uri);
446 if (sock == NULL) {
447 ERR("Allocating socket");
448 goto error;
449 }
450
451 ret = lttcomm_create_sock(sock);
452 if (ret < 0) {
453 goto error;
454 }
455 DBG("Listening on sock %d for lttng-live", sock->fd);
456
457 ret = sock->ops->bind(sock);
458 if (ret < 0) {
459 PERROR("Failed to bind lttng-live socket");
460 goto error;
461 }
462
463 ret = sock->ops->listen(sock, -1);
464 if (ret < 0) {
465 goto error;
466
467 }
468
469 return sock;
470
471 error:
472 if (sock) {
473 lttcomm_destroy_sock(sock);
474 }
475 return NULL;
476 }
477
478 /*
479 * This thread manages the listening for new connections on the network
480 */
481 static
482 void *thread_listener(void *data)
483 {
484 int i, ret, pollfd, err = -1;
485 uint32_t revents, nb_fd;
486 struct lttng_poll_event events;
487 struct lttcomm_sock *live_control_sock;
488
489 DBG("[thread] Relay live listener started");
490
491 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
492
493 health_code_update();
494
495 live_control_sock = init_socket(live_uri);
496 if (!live_control_sock) {
497 goto error_sock_control;
498 }
499
500 /* Pass 2 as size here for the thread quit pipe and control sockets. */
501 ret = create_thread_poll_set(&events, 2);
502 if (ret < 0) {
503 goto error_create_poll;
504 }
505
506 /* Add the control socket */
507 ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
508 if (ret < 0) {
509 goto error_poll_add;
510 }
511
512 lttng_relay_notify_ready();
513
514 if (testpoint(relayd_thread_live_listener)) {
515 goto error_testpoint;
516 }
517
518 while (1) {
519 health_code_update();
520
521 DBG("Listener accepting live viewers connections");
522
523 restart:
524 health_poll_entry();
525 ret = lttng_poll_wait(&events, -1);
526 health_poll_exit();
527 if (ret < 0) {
528 /*
529 * Restart interrupted system call.
530 */
531 if (errno == EINTR) {
532 goto restart;
533 }
534 goto error;
535 }
536 nb_fd = ret;
537
538 DBG("Relay new viewer connection received");
539 for (i = 0; i < nb_fd; i++) {
540 health_code_update();
541
542 /* Fetch once the poll data */
543 revents = LTTNG_POLL_GETEV(&events, i);
544 pollfd = LTTNG_POLL_GETFD(&events, i);
545
546 if (!revents) {
547 /* No activity for this FD (poll implementation). */
548 continue;
549 }
550
551 /* Thread quit pipe has been closed. Killing thread. */
552 ret = check_thread_quit_pipe(pollfd, revents);
553 if (ret) {
554 err = 0;
555 goto exit;
556 }
557
558 if (revents & LPOLLIN) {
559 /*
560 * A new connection is requested, therefore a
561 * viewer connection is allocated in this
562 * thread, enqueued to a global queue and
563 * dequeued (and freed) in the worker thread.
564 */
565 int val = 1;
566 struct relay_connection *new_conn;
567 struct lttcomm_sock *newsock;
568
569 newsock = live_control_sock->ops->accept(live_control_sock);
570 if (!newsock) {
571 PERROR("accepting control sock");
572 goto error;
573 }
574 DBG("Relay viewer connection accepted socket %d", newsock->fd);
575
576 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
577 sizeof(val));
578 if (ret < 0) {
579 PERROR("setsockopt inet");
580 lttcomm_destroy_sock(newsock);
581 goto error;
582 }
583 new_conn = connection_create(newsock, RELAY_CONNECTION_UNKNOWN);
584 if (!new_conn) {
585 lttcomm_destroy_sock(newsock);
586 goto error;
587 }
588 /* Ownership assumed by the connection. */
589 newsock = NULL;
590
591 /* Enqueue request for the dispatcher thread. */
592 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
593 &new_conn->qnode);
594
595 /*
596 * Wake the dispatch queue futex.
597 * Implicit memory barrier with the
598 * exchange in cds_wfcq_enqueue.
599 */
600 futex_nto1_wake(&viewer_conn_queue.futex);
601 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
602 ERR("socket poll error");
603 goto error;
604 } else {
605 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
606 goto error;
607 }
608 }
609 }
610
611 exit:
612 error:
613 error_poll_add:
614 error_testpoint:
615 lttng_poll_clean(&events);
616 error_create_poll:
617 if (live_control_sock->fd >= 0) {
618 ret = live_control_sock->ops->close(live_control_sock);
619 if (ret) {
620 PERROR("close");
621 }
622 }
623 lttcomm_destroy_sock(live_control_sock);
624 error_sock_control:
625 if (err) {
626 health_error();
627 DBG("Live viewer listener thread exited with error");
628 }
629 health_unregister(health_relayd);
630 DBG("Live viewer listener thread cleanup complete");
631 if (lttng_relay_stop_threads()) {
632 ERR("Error stopping threads");
633 }
634 return NULL;
635 }
636
637 /*
638 * This thread manages the dispatching of the requests to worker threads
639 */
640 static
641 void *thread_dispatcher(void *data)
642 {
643 int err = -1;
644 ssize_t ret;
645 struct cds_wfcq_node *node;
646 struct relay_connection *conn = NULL;
647
648 DBG("[thread] Live viewer relay dispatcher started");
649
650 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
651
652 if (testpoint(relayd_thread_live_dispatcher)) {
653 goto error_testpoint;
654 }
655
656 health_code_update();
657
658 for (;;) {
659 health_code_update();
660
661 /* Atomically prepare the queue futex */
662 futex_nto1_prepare(&viewer_conn_queue.futex);
663
664 if (CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
665 break;
666 }
667
668 do {
669 health_code_update();
670
671 /* Dequeue commands */
672 node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
673 &viewer_conn_queue.tail);
674 if (node == NULL) {
675 DBG("Woken up but nothing in the live-viewer "
676 "relay command queue");
677 /* Continue thread execution */
678 break;
679 }
680 conn = caa_container_of(node, struct relay_connection, qnode);
681 DBG("Dispatching viewer request waiting on sock %d",
682 conn->sock->fd);
683
684 /*
685 * Inform worker thread of the new request. This
686 * call is blocking so we can be assured that
687 * the data will be read at some point in time
688 * or wait to the end of the world :)
689 */
690 ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
691 if (ret < 0) {
692 PERROR("write conn pipe");
693 connection_put(conn);
694 goto error;
695 }
696 } while (node != NULL);
697
698 /* Futex wait on queue. Blocking call on futex() */
699 health_poll_entry();
700 futex_nto1_wait(&viewer_conn_queue.futex);
701 health_poll_exit();
702 }
703
704 /* Normal exit, no error */
705 err = 0;
706
707 error:
708 error_testpoint:
709 if (err) {
710 health_error();
711 ERR("Health error occurred in %s", __func__);
712 }
713 health_unregister(health_relayd);
714 DBG("Live viewer dispatch thread dying");
715 if (lttng_relay_stop_threads()) {
716 ERR("Error stopping threads");
717 }
718 return NULL;
719 }
720
721 /*
722 * Establish connection with the viewer and check the versions.
723 *
724 * Return 0 on success or else negative value.
725 */
726 static
727 int viewer_connect(struct relay_connection *conn)
728 {
729 int ret;
730 struct lttng_viewer_connect reply, msg;
731
732 conn->version_check_done = 1;
733
734 health_code_update();
735
736 DBG("Viewer is establishing a connection to the relayd.");
737
738 ret = recv_request(conn->sock, &msg, sizeof(msg));
739 if (ret < 0) {
740 goto end;
741 }
742
743 health_code_update();
744
745 memset(&reply, 0, sizeof(reply));
746 reply.major = RELAYD_VERSION_COMM_MAJOR;
747 reply.minor = RELAYD_VERSION_COMM_MINOR;
748
749 /* Major versions must be the same */
750 if (reply.major != be32toh(msg.major)) {
751 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
752 reply.major, be32toh(msg.major));
753 ret = -1;
754 goto end;
755 }
756
757 conn->major = reply.major;
758 /* We adapt to the lowest compatible version */
759 if (reply.minor <= be32toh(msg.minor)) {
760 conn->minor = reply.minor;
761 } else {
762 conn->minor = be32toh(msg.minor);
763 }
764
765 if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
766 conn->type = RELAY_VIEWER_COMMAND;
767 } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
768 conn->type = RELAY_VIEWER_NOTIFICATION;
769 } else {
770 ERR("Unknown connection type : %u", be32toh(msg.type));
771 ret = -1;
772 goto end;
773 }
774
775 reply.major = htobe32(reply.major);
776 reply.minor = htobe32(reply.minor);
777 if (conn->type == RELAY_VIEWER_COMMAND) {
778 /*
779 * Increment outside of htobe64 macro, because the argument can
780 * be used more than once within the macro, and thus the
781 * operation may be undefined.
782 */
783 pthread_mutex_lock(&last_relay_viewer_session_id_lock);
784 last_relay_viewer_session_id++;
785 pthread_mutex_unlock(&last_relay_viewer_session_id_lock);
786 reply.viewer_session_id = htobe64(last_relay_viewer_session_id);
787 }
788
789 health_code_update();
790
791 ret = send_response(conn->sock, &reply, sizeof(reply));
792 if (ret < 0) {
793 goto end;
794 }
795
796 health_code_update();
797
798 DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
799 ret = 0;
800
801 end:
802 return ret;
803 }
804
805 /*
806 * Send the viewer the list of current sessions.
807 * We need to create a copy of the hash table content because otherwise
808 * we cannot assume the number of entries stays the same between getting
809 * the number of HT elements and iteration over the HT.
810 *
811 * Return 0 on success or else a negative value.
812 */
813 static
814 int viewer_list_sessions(struct relay_connection *conn)
815 {
816 int ret = 0;
817 struct lttng_viewer_list_sessions session_list;
818 struct lttng_ht_iter iter;
819 struct relay_session *session;
820 struct lttng_viewer_session *send_session_buf = NULL;
821 uint32_t buf_count = SESSION_BUF_DEFAULT_COUNT;
822 uint32_t count = 0;
823
824 DBG("List sessions received");
825
826 send_session_buf = zmalloc(SESSION_BUF_DEFAULT_COUNT * sizeof(*send_session_buf));
827 if (!send_session_buf) {
828 return -1;
829 }
830
831 rcu_read_lock();
832 cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, session,
833 session_n.node) {
834 struct lttng_viewer_session *send_session;
835
836 health_code_update();
837
838 if (session->connection_closed) {
839 /* Skip closed session */
840 continue;
841 }
842
843 if (count >= buf_count) {
844 struct lttng_viewer_session *newbuf;
845 uint32_t new_buf_count = buf_count << 1;
846
847 newbuf = realloc(send_session_buf,
848 new_buf_count * sizeof(*send_session_buf));
849 if (!newbuf) {
850 ret = -1;
851 break;
852 }
853 send_session_buf = newbuf;
854 buf_count = new_buf_count;
855 }
856 send_session = &send_session_buf[count];
857 if (lttng_strncpy(send_session->session_name,
858 session->session_name,
859 sizeof(send_session->session_name))) {
860 ret = -1;
861 break;
862 }
863 if (lttng_strncpy(send_session->hostname, session->hostname,
864 sizeof(send_session->hostname))) {
865 ret = -1;
866 break;
867 }
868 send_session->id = htobe64(session->id);
869 send_session->live_timer = htobe32(session->live_timer);
870 if (session->viewer_attached) {
871 send_session->clients = htobe32(1);
872 } else {
873 send_session->clients = htobe32(0);
874 }
875 send_session->streams = htobe32(session->stream_count);
876 count++;
877 }
878 rcu_read_unlock();
879 if (ret < 0) {
880 goto end_free;
881 }
882
883 session_list.sessions_count = htobe32(count);
884
885 health_code_update();
886
887 ret = send_response(conn->sock, &session_list, sizeof(session_list));
888 if (ret < 0) {
889 goto end_free;
890 }
891
892 health_code_update();
893
894 ret = send_response(conn->sock, send_session_buf,
895 count * sizeof(*send_session_buf));
896 if (ret < 0) {
897 goto end_free;
898 }
899 health_code_update();
900
901 ret = 0;
902 end_free:
903 free(send_session_buf);
904 return ret;
905 }
906
907 /*
908 * Send the viewer the list of current streams.
909 */
910 static
911 int viewer_get_new_streams(struct relay_connection *conn)
912 {
913 int ret, send_streams = 0;
914 uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0, nb_total = 0;
915 struct lttng_viewer_new_streams_request request;
916 struct lttng_viewer_new_streams_response response;
917 struct relay_session *session;
918 uint64_t session_id;
919 bool closed = false;
920
921 assert(conn);
922
923 DBG("Get new streams received");
924
925 health_code_update();
926
927 /* Receive the request from the connected client. */
928 ret = recv_request(conn->sock, &request, sizeof(request));
929 if (ret < 0) {
930 goto error;
931 }
932 session_id = be64toh(request.session_id);
933
934 health_code_update();
935
936 memset(&response, 0, sizeof(response));
937
938 session = session_get_by_id(session_id);
939 if (!session) {
940 DBG("Relay session %" PRIu64 " not found", session_id);
941 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
942 goto send_reply;
943 }
944
945 if (!viewer_session_is_attached(conn->viewer_session, session)) {
946 send_streams = 0;
947 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
948 goto send_reply;
949 }
950
951 send_streams = 1;
952 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
953
954 ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, &nb_total, &nb_unsent,
955 &nb_created, &closed);
956 if (ret < 0) {
957 goto end_put_session;
958 }
959 /* Only send back the newly created streams with the unsent ones. */
960 nb_streams = nb_created + nb_unsent;
961 response.streams_count = htobe32(nb_streams);
962
963 /*
964 * If the session is closed, HUP when there are no more streams
965 * with data.
966 */
967 if (closed && nb_total == 0) {
968 send_streams = 0;
969 response.streams_count = 0;
970 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
971 goto send_reply;
972 }
973
974 send_reply:
975 health_code_update();
976 ret = send_response(conn->sock, &response, sizeof(response));
977 if (ret < 0) {
978 goto end_put_session;
979 }
980 health_code_update();
981
982 /*
983 * Unknown or empty session, just return gracefully, the viewer
984 * knows what is happening.
985 */
986 if (!send_streams || !nb_streams) {
987 ret = 0;
988 goto end_put_session;
989 }
990
991 /*
992 * Send stream and *DON'T* ignore the sent flag so every viewer
993 * streams that were not sent from that point will be sent to
994 * the viewer.
995 */
996 ret = send_viewer_streams(conn->sock, session, 0);
997 if (ret < 0) {
998 goto end_put_session;
999 }
1000
1001 end_put_session:
1002 if (session) {
1003 session_put(session);
1004 }
1005 error:
1006 return ret;
1007 }
1008
1009 /*
1010 * Send the viewer the list of current sessions.
1011 */
1012 static
1013 int viewer_attach_session(struct relay_connection *conn)
1014 {
1015 int send_streams = 0;
1016 ssize_t ret;
1017 uint32_t nb_streams = 0;
1018 enum lttng_viewer_seek seek_type;
1019 struct lttng_viewer_attach_session_request request;
1020 struct lttng_viewer_attach_session_response response;
1021 struct relay_session *session = NULL;
1022 bool closed = false;
1023
1024 assert(conn);
1025
1026 health_code_update();
1027
1028 /* Receive the request from the connected client. */
1029 ret = recv_request(conn->sock, &request, sizeof(request));
1030 if (ret < 0) {
1031 goto error;
1032 }
1033
1034 health_code_update();
1035
1036 memset(&response, 0, sizeof(response));
1037
1038 if (!conn->viewer_session) {
1039 DBG("Client trying to attach before creating a live viewer session");
1040 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1041 goto send_reply;
1042 }
1043
1044 session = session_get_by_id(be64toh(request.session_id));
1045 if (!session) {
1046 DBG("Relay session %" PRIu64 " not found",
1047 (uint64_t) be64toh(request.session_id));
1048 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1049 goto send_reply;
1050 }
1051 DBG("Attach session ID %" PRIu64 " received",
1052 (uint64_t) be64toh(request.session_id));
1053
1054 if (session->live_timer == 0) {
1055 DBG("Not live session");
1056 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1057 goto send_reply;
1058 }
1059
1060 send_streams = 1;
1061 ret = viewer_session_attach(conn->viewer_session, session);
1062 if (ret) {
1063 DBG("Already a viewer attached");
1064 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1065 goto send_reply;
1066 }
1067
1068 switch (be32toh(request.seek)) {
1069 case LTTNG_VIEWER_SEEK_BEGINNING:
1070 case LTTNG_VIEWER_SEEK_LAST:
1071 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1072 seek_type = be32toh(request.seek);
1073 break;
1074 default:
1075 ERR("Wrong seek parameter");
1076 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1077 send_streams = 0;
1078 goto send_reply;
1079 }
1080
1081 ret = make_viewer_streams(session, seek_type, &nb_streams, NULL,
1082 NULL, &closed);
1083 if (ret < 0) {
1084 goto end_put_session;
1085 }
1086 response.streams_count = htobe32(nb_streams);
1087
1088 /*
1089 * If the session is closed when the viewer is attaching, it
1090 * means some of the streams may have been concurrently removed,
1091 * so we don't allow the viewer to attach, even if there are
1092 * streams available.
1093 */
1094 if (closed) {
1095 send_streams = 0;
1096 response.streams_count = 0;
1097 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1098 goto send_reply;
1099 }
1100
1101 send_reply:
1102 health_code_update();
1103 ret = send_response(conn->sock, &response, sizeof(response));
1104 if (ret < 0) {
1105 goto end_put_session;
1106 }
1107 health_code_update();
1108
1109 /*
1110 * Unknown or empty session, just return gracefully, the viewer
1111 * knows what is happening.
1112 */
1113 if (!send_streams || !nb_streams) {
1114 ret = 0;
1115 goto end_put_session;
1116 }
1117
1118 /* Send stream and ignore the sent flag. */
1119 ret = send_viewer_streams(conn->sock, session, 1);
1120 if (ret < 0) {
1121 goto end_put_session;
1122 }
1123
1124 end_put_session:
1125 if (session) {
1126 session_put(session);
1127 }
1128 error:
1129 return ret;
1130 }
1131
1132 /*
1133 * Open the index file if needed for the given vstream.
1134 *
1135 * If an index file is successfully opened, the vstream will set it as its
1136 * current index file.
1137 *
1138 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1139 *
1140 * Called with rstream lock held.
1141 */
1142 static int try_open_index(struct relay_viewer_stream *vstream,
1143 struct relay_stream *rstream)
1144 {
1145 int ret = 0;
1146
1147 if (vstream->index_file) {
1148 goto end;
1149 }
1150
1151 /*
1152 * First time, we open the index file and at least one index is ready.
1153 */
1154 if (rstream->index_received_seqcount == 0) {
1155 ret = -ENOENT;
1156 goto end;
1157 }
1158 vstream->index_file = lttng_index_file_open(vstream->path_name,
1159 vstream->channel_name,
1160 vstream->stream->tracefile_count,
1161 vstream->current_tracefile_id);
1162 if (!vstream->index_file) {
1163 ret = -1;
1164 }
1165
1166 end:
1167 return ret;
1168 }
1169
1170 /*
1171 * Check the status of the index for the given stream. This function
1172 * updates the index structure if needed and can put (close) the vstream
1173 * in the HUP situation.
1174 *
1175 * Return 0 means that we can proceed with the index. A value of 1 means
1176 * that the index has been updated and is ready to be sent to the
1177 * client. A negative value indicates an error that can't be handled.
1178 *
1179 * Called with rstream lock held.
1180 */
1181 static int check_index_status(struct relay_viewer_stream *vstream,
1182 struct relay_stream *rstream, struct ctf_trace *trace,
1183 struct lttng_viewer_index *index)
1184 {
1185 int ret;
1186
1187 if ((trace->session->connection_closed || rstream->closed)
1188 && rstream->index_received_seqcount
1189 == vstream->index_sent_seqcount) {
1190 /*
1191 * Last index sent and session connection or relay
1192 * stream are closed.
1193 */
1194 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1195 goto hup;
1196 } else if (rstream->beacon_ts_end != -1ULL &&
1197 rstream->index_received_seqcount
1198 == vstream->index_sent_seqcount) {
1199 /*
1200 * We've received a synchronization beacon and the last index
1201 * available has been sent, the index for now is inactive.
1202 *
1203 * In this case, we have received a beacon which allows us to
1204 * inform the client of a time interval during which we can
1205 * guarantee that there are no events to read (and never will
1206 * be).
1207 */
1208 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1209 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1210 index->stream_id = htobe64(rstream->ctf_stream_id);
1211 goto index_ready;
1212 } else if (rstream->index_received_seqcount
1213 == vstream->index_sent_seqcount) {
1214 /*
1215 * This checks whether received == sent seqcount. In
1216 * this case, we have not received a beacon. Therefore,
1217 * we can only ask the client to retry later.
1218 */
1219 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1220 goto index_ready;
1221 } else if (!tracefile_array_seq_in_file(rstream->tfa,
1222 vstream->current_tracefile_id,
1223 vstream->index_sent_seqcount)) {
1224 /*
1225 * The next index we want to send cannot be read either
1226 * because we need to perform a rotation, or due to
1227 * the producer having overwritten its trace file.
1228 */
1229 DBG("Viewer stream %" PRIu64 " rotation",
1230 vstream->stream->stream_handle);
1231 ret = viewer_stream_rotate(vstream);
1232 if (ret < 0) {
1233 goto end;
1234 } else if (ret == 1) {
1235 /* EOF across entire stream. */
1236 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1237 goto hup;
1238 }
1239 /*
1240 * If we have been pushed due to overwrite, it
1241 * necessarily means there is data that can be read in
1242 * the stream. If we rotated because we reached the end
1243 * of a tracefile, it means the following tracefile
1244 * needs to contain at least one index, else we would
1245 * have already returned LTTNG_VIEWER_INDEX_RETRY to the
1246 * viewer. The updated index_sent_seqcount needs to
1247 * point to a readable index entry now.
1248 *
1249 * In the case where we "rotate" on a single file, we
1250 * can end up in a case where the requested index is
1251 * still unavailable.
1252 */
1253 if (rstream->tracefile_count == 1 &&
1254 !tracefile_array_seq_in_file(
1255 rstream->tfa,
1256 vstream->current_tracefile_id,
1257 vstream->index_sent_seqcount)) {
1258 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1259 goto index_ready;
1260 }
1261 assert(tracefile_array_seq_in_file(rstream->tfa,
1262 vstream->current_tracefile_id,
1263 vstream->index_sent_seqcount));
1264 }
1265 /* ret == 0 means successful so we continue. */
1266 ret = 0;
1267 end:
1268 return ret;
1269
1270 hup:
1271 viewer_stream_put(vstream);
1272 index_ready:
1273 return 1;
1274 }
1275
1276 /*
1277 * Send the next index for a stream.
1278 *
1279 * Return 0 on success or else a negative value.
1280 */
1281 static
1282 int viewer_get_next_index(struct relay_connection *conn)
1283 {
1284 int ret;
1285 struct lttng_viewer_get_next_index request_index;
1286 struct lttng_viewer_index viewer_index;
1287 struct ctf_packet_index packet_index;
1288 struct relay_viewer_stream *vstream = NULL;
1289 struct relay_stream *rstream = NULL;
1290 struct ctf_trace *ctf_trace = NULL;
1291 struct relay_viewer_stream *metadata_viewer_stream = NULL;
1292
1293 assert(conn);
1294
1295 DBG("Viewer get next index");
1296
1297 memset(&viewer_index, 0, sizeof(viewer_index));
1298 health_code_update();
1299
1300 ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1301 if (ret < 0) {
1302 goto end;
1303 }
1304 health_code_update();
1305
1306 vstream = viewer_stream_get_by_id(be64toh(request_index.stream_id));
1307 if (!vstream) {
1308 DBG("Client requested index of unknown stream id %" PRIu64,
1309 (uint64_t) be64toh(request_index.stream_id));
1310 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1311 goto send_reply;
1312 }
1313
1314 /* Use back. ref. Protected by refcounts. */
1315 rstream = vstream->stream;
1316 ctf_trace = rstream->trace;
1317
1318 /* metadata_viewer_stream may be NULL. */
1319 metadata_viewer_stream =
1320 ctf_trace_get_viewer_metadata_stream(ctf_trace);
1321
1322 pthread_mutex_lock(&rstream->lock);
1323
1324 /*
1325 * The viewer should not ask for index on metadata stream.
1326 */
1327 if (rstream->is_metadata) {
1328 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1329 goto send_reply;
1330 }
1331
1332 /* Try to open an index if one is needed for that stream. */
1333 ret = try_open_index(vstream, rstream);
1334 if (ret < 0) {
1335 if (ret == -ENOENT) {
1336 /*
1337 * The index is created only when the first data
1338 * packet arrives, it might not be ready at the
1339 * beginning of the session
1340 */
1341 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1342 } else {
1343 /* Unhandled error. */
1344 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1345 }
1346 goto send_reply;
1347 }
1348
1349 ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1350 if (ret < 0) {
1351 goto error_put;
1352 } else if (ret == 1) {
1353 /*
1354 * We have no index to send and check_index_status has populated
1355 * viewer_index's status.
1356 */
1357 goto send_reply;
1358 }
1359 /* At this point, ret is 0 thus we will be able to read the index. */
1360 assert(!ret);
1361
1362 /*
1363 * vstream->stream_fd may be NULL if it has been closed by
1364 * tracefile rotation, or if we are at the beginning of the
1365 * stream. We open the data stream file here to protect against
1366 * overwrite caused by tracefile rotation (in association with
1367 * unlink performed before overwrite).
1368 */
1369 if (!vstream->stream_fd) {
1370 char fullpath[PATH_MAX];
1371
1372 if (vstream->stream->tracefile_count > 0) {
1373 ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64,
1374 vstream->path_name,
1375 vstream->channel_name,
1376 vstream->current_tracefile_id);
1377 } else {
1378 ret = snprintf(fullpath, PATH_MAX, "%s/%s",
1379 vstream->path_name,
1380 vstream->channel_name);
1381 }
1382 if (ret < 0) {
1383 goto error_put;
1384 }
1385 ret = open(fullpath, O_RDONLY);
1386 if (ret < 0) {
1387 PERROR("Relay opening trace file");
1388 goto error_put;
1389 }
1390 vstream->stream_fd = stream_fd_create(ret);
1391 if (!vstream->stream_fd) {
1392 if (close(ret)) {
1393 PERROR("close");
1394 }
1395 goto error_put;
1396 }
1397 }
1398
1399 ret = check_new_streams(conn);
1400 if (ret < 0) {
1401 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1402 goto send_reply;
1403 } else if (ret == 1) {
1404 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1405 }
1406
1407 ret = lttng_index_file_read(vstream->index_file, &packet_index);
1408 if (ret) {
1409 ERR("Relay error reading index file %d",
1410 vstream->index_file->fd);
1411 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1412 goto send_reply;
1413 } else {
1414 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1415 vstream->index_sent_seqcount++;
1416 }
1417
1418 /*
1419 * Indexes are stored in big endian, no need to switch before sending.
1420 */
1421 DBG("Sending viewer index for stream %" PRIu64 " offset %" PRIu64,
1422 rstream->stream_handle,
1423 (uint64_t) be64toh(packet_index.offset));
1424 viewer_index.offset = packet_index.offset;
1425 viewer_index.packet_size = packet_index.packet_size;
1426 viewer_index.content_size = packet_index.content_size;
1427 viewer_index.timestamp_begin = packet_index.timestamp_begin;
1428 viewer_index.timestamp_end = packet_index.timestamp_end;
1429 viewer_index.events_discarded = packet_index.events_discarded;
1430 viewer_index.stream_id = packet_index.stream_id;
1431
1432 send_reply:
1433 if (rstream) {
1434 pthread_mutex_unlock(&rstream->lock);
1435 }
1436
1437 if (metadata_viewer_stream) {
1438 pthread_mutex_lock(&metadata_viewer_stream->stream->lock);
1439 DBG("get next index metadata check: recv %" PRIu64
1440 " sent %" PRIu64,
1441 metadata_viewer_stream->stream->metadata_received,
1442 metadata_viewer_stream->metadata_sent);
1443 if (!metadata_viewer_stream->stream->metadata_received ||
1444 metadata_viewer_stream->stream->metadata_received >
1445 metadata_viewer_stream->metadata_sent) {
1446 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1447 }
1448 pthread_mutex_unlock(&metadata_viewer_stream->stream->lock);
1449 }
1450
1451 viewer_index.flags = htobe32(viewer_index.flags);
1452 health_code_update();
1453
1454 ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1455 if (ret < 0) {
1456 goto end;
1457 }
1458 health_code_update();
1459
1460 if (vstream) {
1461 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1462 vstream->index_sent_seqcount,
1463 vstream->stream->stream_handle);
1464 }
1465 end:
1466 if (metadata_viewer_stream) {
1467 viewer_stream_put(metadata_viewer_stream);
1468 }
1469 if (vstream) {
1470 viewer_stream_put(vstream);
1471 }
1472 return ret;
1473
1474 error_put:
1475 pthread_mutex_unlock(&rstream->lock);
1476 if (metadata_viewer_stream) {
1477 viewer_stream_put(metadata_viewer_stream);
1478 }
1479 viewer_stream_put(vstream);
1480 return ret;
1481 }
1482
1483 /*
1484 * Send the next index for a stream
1485 *
1486 * Return 0 on success or else a negative value.
1487 */
1488 static
1489 int viewer_get_packet(struct relay_connection *conn)
1490 {
1491 int ret;
1492 off_t lseek_ret;
1493 char *reply = NULL;
1494 struct lttng_viewer_get_packet get_packet_info;
1495 struct lttng_viewer_trace_packet reply_header;
1496 struct relay_viewer_stream *vstream = NULL;
1497 uint32_t reply_size = sizeof(reply_header);
1498 uint32_t packet_data_len = 0;
1499 ssize_t read_len;
1500
1501 DBG2("Relay get data packet");
1502
1503 health_code_update();
1504
1505 ret = recv_request(conn->sock, &get_packet_info,
1506 sizeof(get_packet_info));
1507 if (ret < 0) {
1508 goto end;
1509 }
1510 health_code_update();
1511
1512 /* From this point on, the error label can be reached. */
1513 memset(&reply_header, 0, sizeof(reply_header));
1514
1515 vstream = viewer_stream_get_by_id(be64toh(get_packet_info.stream_id));
1516 if (!vstream) {
1517 DBG("Client requested packet of unknown stream id %" PRIu64,
1518 (uint64_t) be64toh(get_packet_info.stream_id));
1519 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1520 goto send_reply_nolock;
1521 } else {
1522 packet_data_len = be32toh(get_packet_info.len);
1523 reply_size += packet_data_len;
1524 }
1525
1526 reply = zmalloc(reply_size);
1527 if (!reply) {
1528 PERROR("packet reply zmalloc");
1529 reply_size = sizeof(reply_header);
1530 goto error;
1531 }
1532
1533 pthread_mutex_lock(&vstream->stream->lock);
1534 lseek_ret = lseek(vstream->stream_fd->fd, be64toh(get_packet_info.offset),
1535 SEEK_SET);
1536 if (lseek_ret < 0) {
1537 PERROR("lseek fd %d to offset %" PRIu64, vstream->stream_fd->fd,
1538 (uint64_t) be64toh(get_packet_info.offset));
1539 goto error;
1540 }
1541 read_len = lttng_read(vstream->stream_fd->fd,
1542 reply + sizeof(reply_header),
1543 packet_data_len);
1544 if (read_len < packet_data_len) {
1545 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1546 vstream->stream_fd->fd,
1547 (uint64_t) be64toh(get_packet_info.offset));
1548 goto error;
1549 }
1550 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1551 reply_header.len = htobe32(packet_data_len);
1552 goto send_reply;
1553
1554 error:
1555 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1556
1557 send_reply:
1558 if (vstream) {
1559 pthread_mutex_unlock(&vstream->stream->lock);
1560 }
1561 send_reply_nolock:
1562
1563 health_code_update();
1564
1565 if (reply) {
1566 memcpy(reply, &reply_header, sizeof(reply_header));
1567 ret = send_response(conn->sock, reply, reply_size);
1568 } else {
1569 /* No reply to send. */
1570 ret = send_response(conn->sock, &reply_header,
1571 reply_size);
1572 }
1573
1574 health_code_update();
1575 if (ret < 0) {
1576 PERROR("sendmsg of packet data failed");
1577 goto end_free;
1578 }
1579
1580 DBG("Sent %u bytes for stream %" PRIu64, reply_size,
1581 (uint64_t) be64toh(get_packet_info.stream_id));
1582
1583 end_free:
1584 free(reply);
1585 end:
1586 if (vstream) {
1587 viewer_stream_put(vstream);
1588 }
1589 return ret;
1590 }
1591
1592 /*
1593 * Send the session's metadata
1594 *
1595 * Return 0 on success else a negative value.
1596 */
1597 static
1598 int viewer_get_metadata(struct relay_connection *conn)
1599 {
1600 int ret = 0;
1601 ssize_t read_len;
1602 uint64_t len = 0;
1603 char *data = NULL;
1604 struct lttng_viewer_get_metadata request;
1605 struct lttng_viewer_metadata_packet reply;
1606 struct relay_viewer_stream *vstream = NULL;
1607
1608 assert(conn);
1609
1610 DBG("Relay get metadata");
1611
1612 health_code_update();
1613
1614 ret = recv_request(conn->sock, &request, sizeof(request));
1615 if (ret < 0) {
1616 goto end;
1617 }
1618 health_code_update();
1619
1620 memset(&reply, 0, sizeof(reply));
1621
1622 vstream = viewer_stream_get_by_id(be64toh(request.stream_id));
1623 if (!vstream) {
1624 /*
1625 * The metadata stream can be closed by a CLOSE command
1626 * just before we attach. It can also be closed by
1627 * per-pid tracing during tracing. Therefore, it is
1628 * possible that we cannot find this viewer stream.
1629 * Reply back to the client with an error if we cannot
1630 * find it.
1631 */
1632 DBG("Client requested metadata of unknown stream id %" PRIu64,
1633 (uint64_t) be64toh(request.stream_id));
1634 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1635 goto send_reply;
1636 }
1637 pthread_mutex_lock(&vstream->stream->lock);
1638 if (!vstream->stream->is_metadata) {
1639 ERR("Invalid metadata stream");
1640 goto error;
1641 }
1642
1643 assert(vstream->metadata_sent <= vstream->stream->metadata_received);
1644
1645 len = vstream->stream->metadata_received - vstream->metadata_sent;
1646 if (len == 0) {
1647 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1648 goto send_reply;
1649 }
1650
1651 /* first time, we open the metadata file */
1652 if (!vstream->stream_fd) {
1653 char fullpath[PATH_MAX];
1654
1655 ret = snprintf(fullpath, PATH_MAX, "%s/%s", vstream->path_name,
1656 vstream->channel_name);
1657 if (ret < 0) {
1658 goto error;
1659 }
1660 ret = open(fullpath, O_RDONLY);
1661 if (ret < 0) {
1662 PERROR("Relay opening metadata file");
1663 goto error;
1664 }
1665 vstream->stream_fd = stream_fd_create(ret);
1666 if (!vstream->stream_fd) {
1667 if (close(ret)) {
1668 PERROR("close");
1669 }
1670 goto error;
1671 }
1672 }
1673
1674 reply.len = htobe64(len);
1675 data = zmalloc(len);
1676 if (!data) {
1677 PERROR("viewer metadata zmalloc");
1678 goto error;
1679 }
1680
1681 read_len = lttng_read(vstream->stream_fd->fd, data, len);
1682 if (read_len < len) {
1683 PERROR("Relay reading metadata file");
1684 goto error;
1685 }
1686 vstream->metadata_sent += read_len;
1687 if (vstream->metadata_sent == vstream->stream->metadata_received
1688 && vstream->stream->closed) {
1689 /* Release ownership for the viewer metadata stream. */
1690 viewer_stream_put(vstream);
1691 }
1692
1693 reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1694
1695 goto send_reply;
1696
1697 error:
1698 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1699
1700 send_reply:
1701 health_code_update();
1702 if (vstream) {
1703 pthread_mutex_unlock(&vstream->stream->lock);
1704 }
1705 ret = send_response(conn->sock, &reply, sizeof(reply));
1706 if (ret < 0) {
1707 goto end_free;
1708 }
1709 health_code_update();
1710
1711 if (len > 0) {
1712 ret = send_response(conn->sock, data, len);
1713 if (ret < 0) {
1714 goto end_free;
1715 }
1716 }
1717
1718 DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1719 (uint64_t) be64toh(request.stream_id));
1720
1721 DBG("Metadata sent");
1722
1723 end_free:
1724 free(data);
1725 end:
1726 if (vstream) {
1727 viewer_stream_put(vstream);
1728 }
1729 return ret;
1730 }
1731
1732 /*
1733 * Create a viewer session.
1734 *
1735 * Return 0 on success or else a negative value.
1736 */
1737 static
1738 int viewer_create_session(struct relay_connection *conn)
1739 {
1740 int ret;
1741 struct lttng_viewer_create_session_response resp;
1742
1743 DBG("Viewer create session received");
1744
1745 memset(&resp, 0, sizeof(resp));
1746 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1747 conn->viewer_session = viewer_session_create();
1748 if (!conn->viewer_session) {
1749 ERR("Allocation viewer session");
1750 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1751 goto send_reply;
1752 }
1753
1754 send_reply:
1755 health_code_update();
1756 ret = send_response(conn->sock, &resp, sizeof(resp));
1757 if (ret < 0) {
1758 goto end;
1759 }
1760 health_code_update();
1761 ret = 0;
1762
1763 end:
1764 return ret;
1765 }
1766
1767 /*
1768 * Detach a viewer session.
1769 *
1770 * Return 0 on success or else a negative value.
1771 */
1772 static
1773 int viewer_detach_session(struct relay_connection *conn)
1774 {
1775 int ret;
1776 struct lttng_viewer_detach_session_response response;
1777 struct lttng_viewer_detach_session_request request;
1778 struct relay_session *session = NULL;
1779 uint64_t viewer_session_to_close;
1780
1781 DBG("Viewer detach session received");
1782
1783 assert(conn);
1784
1785 health_code_update();
1786
1787 /* Receive the request from the connected client. */
1788 ret = recv_request(conn->sock, &request, sizeof(request));
1789 if (ret < 0) {
1790 goto end;
1791 }
1792 viewer_session_to_close = be64toh(request.session_id);
1793
1794 if (!conn->viewer_session) {
1795 DBG("Client trying to detach before creating a live viewer session");
1796 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
1797 goto send_reply;
1798 }
1799
1800 health_code_update();
1801
1802 memset(&response, 0, sizeof(response));
1803 DBG("Detaching from session ID %" PRIu64, viewer_session_to_close);
1804
1805 session = session_get_by_id(be64toh(request.session_id));
1806 if (!session) {
1807 DBG("Relay session %" PRIu64 " not found",
1808 (uint64_t) be64toh(request.session_id));
1809 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_UNK);
1810 goto send_reply;
1811 }
1812
1813 ret = viewer_session_is_attached(conn->viewer_session, session);
1814 if (ret != 1) {
1815 DBG("Not attached to this session");
1816 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
1817 goto send_reply_put;
1818 }
1819
1820 viewer_session_close_one_session(conn->viewer_session, session);
1821 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_OK);
1822 DBG("Session %" PRIu64 " detached.", viewer_session_to_close);
1823
1824 send_reply_put:
1825 session_put(session);
1826
1827 send_reply:
1828 health_code_update();
1829 ret = send_response(conn->sock, &response, sizeof(response));
1830 if (ret < 0) {
1831 goto end;
1832 }
1833 health_code_update();
1834 ret = 0;
1835
1836 end:
1837 return ret;
1838 }
1839
1840 /*
1841 * live_relay_unknown_command: send -1 if received unknown command
1842 */
1843 static
1844 void live_relay_unknown_command(struct relay_connection *conn)
1845 {
1846 struct lttcomm_relayd_generic_reply reply;
1847
1848 memset(&reply, 0, sizeof(reply));
1849 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1850 (void) send_response(conn->sock, &reply, sizeof(reply));
1851 }
1852
1853 /*
1854 * Process the commands received on the control socket
1855 */
1856 static
1857 int process_control(struct lttng_viewer_cmd *recv_hdr,
1858 struct relay_connection *conn)
1859 {
1860 int ret = 0;
1861 uint32_t msg_value;
1862
1863 msg_value = be32toh(recv_hdr->cmd);
1864
1865 /*
1866 * Make sure we've done the version check before any command other then a
1867 * new client connection.
1868 */
1869 if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1870 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1871 ret = -1;
1872 goto end;
1873 }
1874
1875 switch (msg_value) {
1876 case LTTNG_VIEWER_CONNECT:
1877 ret = viewer_connect(conn);
1878 break;
1879 case LTTNG_VIEWER_LIST_SESSIONS:
1880 ret = viewer_list_sessions(conn);
1881 break;
1882 case LTTNG_VIEWER_ATTACH_SESSION:
1883 ret = viewer_attach_session(conn);
1884 break;
1885 case LTTNG_VIEWER_GET_NEXT_INDEX:
1886 ret = viewer_get_next_index(conn);
1887 break;
1888 case LTTNG_VIEWER_GET_PACKET:
1889 ret = viewer_get_packet(conn);
1890 break;
1891 case LTTNG_VIEWER_GET_METADATA:
1892 ret = viewer_get_metadata(conn);
1893 break;
1894 case LTTNG_VIEWER_GET_NEW_STREAMS:
1895 ret = viewer_get_new_streams(conn);
1896 break;
1897 case LTTNG_VIEWER_CREATE_SESSION:
1898 ret = viewer_create_session(conn);
1899 break;
1900 case LTTNG_VIEWER_DETACH_SESSION:
1901 ret = viewer_detach_session(conn);
1902 break;
1903 default:
1904 ERR("Received unknown viewer command (%u)",
1905 be32toh(recv_hdr->cmd));
1906 live_relay_unknown_command(conn);
1907 ret = -1;
1908 goto end;
1909 }
1910
1911 end:
1912 return ret;
1913 }
1914
1915 static
1916 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1917 {
1918 int ret;
1919
1920 (void) lttng_poll_del(events, pollfd);
1921
1922 ret = close(pollfd);
1923 if (ret < 0) {
1924 ERR("Closing pollfd %d", pollfd);
1925 }
1926 }
1927
1928 /*
1929 * This thread does the actual work
1930 */
1931 static
1932 void *thread_worker(void *data)
1933 {
1934 int ret, err = -1;
1935 uint32_t nb_fd;
1936 struct lttng_poll_event events;
1937 struct lttng_ht *viewer_connections_ht;
1938 struct lttng_ht_iter iter;
1939 struct lttng_viewer_cmd recv_hdr;
1940 struct relay_connection *destroy_conn;
1941
1942 DBG("[thread] Live viewer relay worker started");
1943
1944 rcu_register_thread();
1945
1946 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1947
1948 if (testpoint(relayd_thread_live_worker)) {
1949 goto error_testpoint;
1950 }
1951
1952 /* table of connections indexed on socket */
1953 viewer_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1954 if (!viewer_connections_ht) {
1955 goto viewer_connections_ht_error;
1956 }
1957
1958 ret = create_thread_poll_set(&events, 2);
1959 if (ret < 0) {
1960 goto error_poll_create;
1961 }
1962
1963 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1964 if (ret < 0) {
1965 goto error;
1966 }
1967
1968 restart:
1969 while (1) {
1970 int i;
1971
1972 health_code_update();
1973
1974 /* Infinite blocking call, waiting for transmission */
1975 DBG3("Relayd live viewer worker thread polling...");
1976 health_poll_entry();
1977 ret = lttng_poll_wait(&events, -1);
1978 health_poll_exit();
1979 if (ret < 0) {
1980 /*
1981 * Restart interrupted system call.
1982 */
1983 if (errno == EINTR) {
1984 goto restart;
1985 }
1986 goto error;
1987 }
1988
1989 nb_fd = ret;
1990
1991 /*
1992 * Process control. The control connection is prioritised so we don't
1993 * starve it with high throughput tracing data on the data
1994 * connection.
1995 */
1996 for (i = 0; i < nb_fd; i++) {
1997 /* Fetch once the poll data */
1998 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1999 int pollfd = LTTNG_POLL_GETFD(&events, i);
2000
2001 health_code_update();
2002
2003 if (!revents) {
2004 /* No activity for this FD (poll implementation). */
2005 continue;
2006 }
2007
2008 /* Thread quit pipe has been closed. Killing thread. */
2009 ret = check_thread_quit_pipe(pollfd, revents);
2010 if (ret) {
2011 err = 0;
2012 goto exit;
2013 }
2014
2015 /* Inspect the relay conn pipe for new connection. */
2016 if (pollfd == live_conn_pipe[0]) {
2017 if (revents & LPOLLIN) {
2018 struct relay_connection *conn;
2019
2020 ret = lttng_read(live_conn_pipe[0],
2021 &conn, sizeof(conn));
2022 if (ret < 0) {
2023 goto error;
2024 }
2025 lttng_poll_add(&events, conn->sock->fd,
2026 LPOLLIN | LPOLLRDHUP);
2027 connection_ht_add(viewer_connections_ht, conn);
2028 DBG("Connection socket %d added to poll", conn->sock->fd);
2029 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2030 ERR("Relay live pipe error");
2031 goto error;
2032 } else {
2033 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2034 goto error;
2035 }
2036 } else {
2037 /* Connection activity. */
2038 struct relay_connection *conn;
2039
2040 conn = connection_get_by_sock(viewer_connections_ht, pollfd);
2041 if (!conn) {
2042 continue;
2043 }
2044
2045 if (revents & LPOLLIN) {
2046 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
2047 sizeof(recv_hdr), 0);
2048 if (ret <= 0) {
2049 /* Connection closed. */
2050 cleanup_connection_pollfd(&events, pollfd);
2051 /* Put "create" ownership reference. */
2052 connection_put(conn);
2053 DBG("Viewer control conn closed with %d", pollfd);
2054 } else {
2055 ret = process_control(&recv_hdr, conn);
2056 if (ret < 0) {
2057 /* Clear the session on error. */
2058 cleanup_connection_pollfd(&events, pollfd);
2059 /* Put "create" ownership reference. */
2060 connection_put(conn);
2061 DBG("Viewer connection closed with %d", pollfd);
2062 }
2063 }
2064 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2065 cleanup_connection_pollfd(&events, pollfd);
2066 /* Put "create" ownership reference. */
2067 connection_put(conn);
2068 } else {
2069 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2070 connection_put(conn);
2071 goto error;
2072 }
2073 /* Put local "get_by_sock" reference. */
2074 connection_put(conn);
2075 }
2076 }
2077 }
2078
2079 exit:
2080 error:
2081 lttng_poll_clean(&events);
2082
2083 /* Cleanup remaining connection object. */
2084 rcu_read_lock();
2085 cds_lfht_for_each_entry(viewer_connections_ht->ht, &iter.iter,
2086 destroy_conn,
2087 sock_n.node) {
2088 health_code_update();
2089 connection_put(destroy_conn);
2090 }
2091 rcu_read_unlock();
2092 error_poll_create:
2093 lttng_ht_destroy(viewer_connections_ht);
2094 viewer_connections_ht_error:
2095 /* Close relay conn pipes */
2096 utils_close_pipe(live_conn_pipe);
2097 if (err) {
2098 DBG("Viewer worker thread exited with error");
2099 }
2100 DBG("Viewer worker thread cleanup complete");
2101 error_testpoint:
2102 if (err) {
2103 health_error();
2104 ERR("Health error occurred in %s", __func__);
2105 }
2106 health_unregister(health_relayd);
2107 if (lttng_relay_stop_threads()) {
2108 ERR("Error stopping threads");
2109 }
2110 rcu_unregister_thread();
2111 return NULL;
2112 }
2113
2114 /*
2115 * Create the relay command pipe to wake thread_manage_apps.
2116 * Closed in cleanup().
2117 */
2118 static int create_conn_pipe(void)
2119 {
2120 return utils_create_pipe_cloexec(live_conn_pipe);
2121 }
2122
2123 int relayd_live_join(void)
2124 {
2125 int ret, retval = 0;
2126 void *status;
2127
2128 ret = pthread_join(live_listener_thread, &status);
2129 if (ret) {
2130 errno = ret;
2131 PERROR("pthread_join live listener");
2132 retval = -1;
2133 }
2134
2135 ret = pthread_join(live_worker_thread, &status);
2136 if (ret) {
2137 errno = ret;
2138 PERROR("pthread_join live worker");
2139 retval = -1;
2140 }
2141
2142 ret = pthread_join(live_dispatcher_thread, &status);
2143 if (ret) {
2144 errno = ret;
2145 PERROR("pthread_join live dispatcher");
2146 retval = -1;
2147 }
2148
2149 cleanup_relayd_live();
2150
2151 return retval;
2152 }
2153
2154 /*
2155 * main
2156 */
2157 int relayd_live_create(struct lttng_uri *uri)
2158 {
2159 int ret = 0, retval = 0;
2160 void *status;
2161 int is_root;
2162
2163 if (!uri) {
2164 retval = -1;
2165 goto exit_init_data;
2166 }
2167 live_uri = uri;
2168
2169 /* Check if daemon is UID = 0 */
2170 is_root = !getuid();
2171
2172 if (!is_root) {
2173 if (live_uri->port < 1024) {
2174 ERR("Need to be root to use ports < 1024");
2175 retval = -1;
2176 goto exit_init_data;
2177 }
2178 }
2179
2180 /* Setup the thread apps communication pipe. */
2181 if (create_conn_pipe()) {
2182 retval = -1;
2183 goto exit_init_data;
2184 }
2185
2186 /* Init relay command queue. */
2187 cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2188
2189 /* Set up max poll set size */
2190 if (lttng_poll_set_max_size()) {
2191 retval = -1;
2192 goto exit_init_data;
2193 }
2194
2195 /* Setup the dispatcher thread */
2196 ret = pthread_create(&live_dispatcher_thread, default_pthread_attr(),
2197 thread_dispatcher, (void *) NULL);
2198 if (ret) {
2199 errno = ret;
2200 PERROR("pthread_create viewer dispatcher");
2201 retval = -1;
2202 goto exit_dispatcher_thread;
2203 }
2204
2205 /* Setup the worker thread */
2206 ret = pthread_create(&live_worker_thread, default_pthread_attr(),
2207 thread_worker, NULL);
2208 if (ret) {
2209 errno = ret;
2210 PERROR("pthread_create viewer worker");
2211 retval = -1;
2212 goto exit_worker_thread;
2213 }
2214
2215 /* Setup the listener thread */
2216 ret = pthread_create(&live_listener_thread, default_pthread_attr(),
2217 thread_listener, (void *) NULL);
2218 if (ret) {
2219 errno = ret;
2220 PERROR("pthread_create viewer listener");
2221 retval = -1;
2222 goto exit_listener_thread;
2223 }
2224
2225 /*
2226 * All OK, started all threads.
2227 */
2228 return retval;
2229
2230 /*
2231 * Join on the live_listener_thread should anything be added after
2232 * the live_listener thread's creation.
2233 */
2234
2235 exit_listener_thread:
2236
2237 ret = pthread_join(live_worker_thread, &status);
2238 if (ret) {
2239 errno = ret;
2240 PERROR("pthread_join live worker");
2241 retval = -1;
2242 }
2243 exit_worker_thread:
2244
2245 ret = pthread_join(live_dispatcher_thread, &status);
2246 if (ret) {
2247 errno = ret;
2248 PERROR("pthread_join live dispatcher");
2249 retval = -1;
2250 }
2251 exit_dispatcher_thread:
2252
2253 exit_init_data:
2254 cleanup_relayd_live();
2255
2256 return retval;
2257 }
This page took 0.07322 seconds and 5 git commands to generate.