562a7fa522017ae4126d2d7b76847fbed9f81dc6
[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 *
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 #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 <unistd.h>
40 #include <fcntl.h>
41 #include <config.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
69 static struct lttng_uri *live_uri;
70
71 /*
72 * This pipe is used to inform the worker thread that a command is queued and
73 * ready to be processed.
74 */
75 static int live_conn_pipe[2] = { -1, -1 };
76
77 /* Shared between threads */
78 static int live_dispatch_thread_exit;
79
80 static pthread_t live_listener_thread;
81 static pthread_t live_dispatcher_thread;
82 static pthread_t live_worker_thread;
83
84 /*
85 * Relay command queue.
86 *
87 * The live_thread_listener and live_thread_dispatcher communicate with this
88 * queue.
89 */
90 static struct relay_conn_queue viewer_conn_queue;
91
92 static uint64_t last_relay_viewer_session_id;
93
94 /*
95 * Cleanup the daemon
96 */
97 static
98 void cleanup_relayd_live(void)
99 {
100 DBG("Cleaning up");
101
102 free(live_uri);
103 }
104
105 /*
106 * Receive a request buffer using a given socket, destination allocated buffer
107 * of length size.
108 *
109 * Return the size of the received message or else a negative value on error
110 * with errno being set by recvmsg() syscall.
111 */
112 static
113 ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
114 {
115 ssize_t ret;
116
117 assert(sock);
118 assert(buf);
119
120 ret = sock->ops->recvmsg(sock, buf, size, 0);
121 if (ret < 0 || ret != size) {
122 if (ret == 0) {
123 /* Orderly shutdown. Not necessary to print an error. */
124 DBG("Socket %d did an orderly shutdown", sock->fd);
125 } else {
126 ERR("Relay failed to receive request.");
127 }
128 ret = -1;
129 }
130
131 return ret;
132 }
133
134 /*
135 * Send a response buffer using a given socket, source allocated buffer of
136 * length size.
137 *
138 * Return the size of the sent message or else a negative value on error with
139 * errno being set by sendmsg() syscall.
140 */
141 static
142 ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
143 {
144 ssize_t ret;
145
146 assert(sock);
147 assert(buf);
148
149 ret = sock->ops->sendmsg(sock, buf, size, 0);
150 if (ret < 0) {
151 ERR("Relayd failed to send response.");
152 }
153
154 return ret;
155 }
156
157 /*
158 * Atomically check if new streams got added in one of the sessions attached
159 * and reset the flag to 0.
160 *
161 * Returns 1 if new streams got added, 0 if nothing changed, a negative value
162 * on error.
163 */
164 static
165 int check_new_streams(struct relay_connection *conn)
166 {
167 struct relay_session *session;
168 unsigned long current_val;
169 int ret = 0;
170
171 if (!conn->viewer_session) {
172 goto end;
173 }
174 cds_list_for_each_entry(session,
175 &conn->viewer_session->sessions_head,
176 viewer_session_list) {
177 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
178 ret = current_val;
179 if (ret == 1) {
180 goto end;
181 }
182 }
183
184 end:
185 return ret;
186 }
187
188 /*
189 * Send viewer streams to the given socket. The ignore_sent_flag indicates if
190 * this function should ignore the sent flag or not.
191 *
192 * Return 0 on success or else a negative value.
193 */
194 static
195 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
196 struct relay_session *session, unsigned int ignore_sent_flag)
197 {
198 ssize_t ret;
199 struct lttng_viewer_stream send_stream;
200 struct lttng_ht_iter iter;
201 struct relay_viewer_stream *vstream;
202
203 assert(session);
204
205 rcu_read_lock();
206
207 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
208 stream_n.node) {
209 struct ctf_trace *ctf_trace;
210
211 health_code_update();
212
213 /* Ignore if not the same session. */
214 if (vstream->session_id != session->id ||
215 (!ignore_sent_flag && vstream->sent_flag)) {
216 continue;
217 }
218
219 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
220 vstream->path_name);
221 assert(ctf_trace);
222
223 send_stream.id = htobe64(vstream->stream_handle);
224 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
225 send_stream.metadata_flag = htobe32(vstream->metadata_flag);
226 strncpy(send_stream.path_name, vstream->path_name,
227 sizeof(send_stream.path_name));
228 strncpy(send_stream.channel_name, vstream->channel_name,
229 sizeof(send_stream.channel_name));
230
231 DBG("Sending stream %" PRIu64 " to viewer", vstream->stream_handle);
232 ret = send_response(sock, &send_stream, sizeof(send_stream));
233 if (ret < 0) {
234 goto end_unlock;
235 }
236 vstream->sent_flag = 1;
237 }
238
239 ret = 0;
240
241 end_unlock:
242 rcu_read_unlock();
243 return ret;
244 }
245
246 /*
247 * Create every viewer stream possible for the given session with the seek
248 * type. Three counters *can* be return which are in order the total amount of
249 * viewer stream of the session, the number of unsent stream and the number of
250 * stream created. Those counters can be NULL and thus will be ignored.
251 *
252 * Return 0 on success or else a negative value.
253 */
254 static
255 int make_viewer_streams(struct relay_session *session,
256 enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent,
257 uint32_t *nb_created)
258 {
259 int ret;
260 struct lttng_ht_iter iter;
261 struct ctf_trace *ctf_trace;
262
263 assert(session);
264
265 /*
266 * This is to make sure we create viewer streams for a full received
267 * channel. For instance, if we have 8 streams for a channel that are
268 * concurrently being flagged ready, we can end up creating just a subset
269 * of the 8 streams (the ones that are flagged). This lock avoids this
270 * limbo state.
271 */
272 pthread_mutex_lock(&session->viewer_ready_lock);
273
274 /*
275 * Create viewer streams for relay streams that are ready to be used for a
276 * the given session id only.
277 */
278 rcu_read_lock();
279 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
280 node.node) {
281 struct relay_stream *stream;
282
283 health_code_update();
284
285 if (ctf_trace->invalid_flag) {
286 continue;
287 }
288
289 cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) {
290 struct relay_viewer_stream *vstream;
291
292 if (!stream->viewer_ready) {
293 continue;
294 }
295
296 vstream = viewer_stream_find_by_id(stream->stream_handle);
297 if (!vstream) {
298 vstream = viewer_stream_create(stream, seek_t, ctf_trace);
299 if (!vstream) {
300 ret = -1;
301 goto error_unlock;
302 }
303 /* Acquire reference to ctf_trace. */
304 ctf_trace_get_ref(ctf_trace);
305
306 if (nb_created) {
307 /* Update number of created stream counter. */
308 (*nb_created)++;
309 }
310 } else if (!vstream->sent_flag && nb_unsent) {
311 /* Update number of unsent stream counter. */
312 (*nb_unsent)++;
313 }
314 /* Update number of total stream counter. */
315 if (nb_total) {
316 (*nb_total)++;
317 }
318 }
319 }
320
321 ret = 0;
322
323 error_unlock:
324 rcu_read_unlock();
325 pthread_mutex_unlock(&session->viewer_ready_lock);
326 return ret;
327 }
328
329 int relayd_live_stop(void)
330 {
331 /* Stop dispatch thread */
332 CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
333 futex_nto1_wake(&viewer_conn_queue.futex);
334 return 0;
335 }
336
337 /*
338 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
339 */
340 static
341 int create_thread_poll_set(struct lttng_poll_event *events, int size)
342 {
343 int ret;
344
345 if (events == NULL || size == 0) {
346 ret = -1;
347 goto error;
348 }
349
350 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
351 if (ret < 0) {
352 goto error;
353 }
354
355 /* Add quit pipe */
356 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
357 if (ret < 0) {
358 goto error;
359 }
360
361 return 0;
362
363 error:
364 return ret;
365 }
366
367 /*
368 * Check if the thread quit pipe was triggered.
369 *
370 * Return 1 if it was triggered else 0;
371 */
372 static
373 int check_thread_quit_pipe(int fd, uint32_t events)
374 {
375 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
376 return 1;
377 }
378
379 return 0;
380 }
381
382 /*
383 * Create and init socket from uri.
384 */
385 static
386 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
387 {
388 int ret;
389 struct lttcomm_sock *sock = NULL;
390
391 sock = lttcomm_alloc_sock_from_uri(uri);
392 if (sock == NULL) {
393 ERR("Allocating socket");
394 goto error;
395 }
396
397 ret = lttcomm_create_sock(sock);
398 if (ret < 0) {
399 goto error;
400 }
401 DBG("Listening on sock %d for live", sock->fd);
402
403 ret = sock->ops->bind(sock);
404 if (ret < 0) {
405 goto error;
406 }
407
408 ret = sock->ops->listen(sock, -1);
409 if (ret < 0) {
410 goto error;
411
412 }
413
414 return sock;
415
416 error:
417 if (sock) {
418 lttcomm_destroy_sock(sock);
419 }
420 return NULL;
421 }
422
423 /*
424 * This thread manages the listening for new connections on the network
425 */
426 static
427 void *thread_listener(void *data)
428 {
429 int i, ret, pollfd, err = -1;
430 uint32_t revents, nb_fd;
431 struct lttng_poll_event events;
432 struct lttcomm_sock *live_control_sock;
433
434 DBG("[thread] Relay live listener started");
435
436 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
437
438 health_code_update();
439
440 live_control_sock = init_socket(live_uri);
441 if (!live_control_sock) {
442 goto error_sock_control;
443 }
444
445 /* Pass 2 as size here for the thread quit pipe and control sockets. */
446 ret = create_thread_poll_set(&events, 2);
447 if (ret < 0) {
448 goto error_create_poll;
449 }
450
451 /* Add the control socket */
452 ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
453 if (ret < 0) {
454 goto error_poll_add;
455 }
456
457 lttng_relay_notify_ready();
458
459 if (testpoint(relayd_thread_live_listener)) {
460 goto error_testpoint;
461 }
462
463 while (1) {
464 health_code_update();
465
466 DBG("Listener accepting live viewers connections");
467
468 restart:
469 health_poll_entry();
470 ret = lttng_poll_wait(&events, -1);
471 health_poll_exit();
472 if (ret < 0) {
473 /*
474 * Restart interrupted system call.
475 */
476 if (errno == EINTR) {
477 goto restart;
478 }
479 goto error;
480 }
481 nb_fd = ret;
482
483 DBG("Relay new viewer connection received");
484 for (i = 0; i < nb_fd; i++) {
485 health_code_update();
486
487 /* Fetch once the poll data */
488 revents = LTTNG_POLL_GETEV(&events, i);
489 pollfd = LTTNG_POLL_GETFD(&events, i);
490
491 if (!revents) {
492 /* No activity for this FD (poll implementation). */
493 continue;
494 }
495
496 /* Thread quit pipe has been closed. Killing thread. */
497 ret = check_thread_quit_pipe(pollfd, revents);
498 if (ret) {
499 err = 0;
500 goto exit;
501 }
502
503 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
504 ERR("socket poll error");
505 goto error;
506 } else if (revents & LPOLLIN) {
507 /*
508 * Get allocated in this thread, enqueued to a global queue,
509 * dequeued and freed in the worker thread.
510 */
511 int val = 1;
512 struct relay_connection *new_conn;
513 struct lttcomm_sock *newsock;
514
515 new_conn = connection_create();
516 if (!new_conn) {
517 goto error;
518 }
519
520 newsock = live_control_sock->ops->accept(live_control_sock);
521 if (!newsock) {
522 PERROR("accepting control sock");
523 connection_free(new_conn);
524 goto error;
525 }
526 DBG("Relay viewer connection accepted socket %d", newsock->fd);
527
528 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
529 sizeof(val));
530 if (ret < 0) {
531 PERROR("setsockopt inet");
532 lttcomm_destroy_sock(newsock);
533 connection_free(new_conn);
534 goto error;
535 }
536 new_conn->sock = newsock;
537
538 /* Enqueue request for the dispatcher thread. */
539 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
540 &new_conn->qnode);
541
542 /*
543 * Wake the dispatch queue futex. Implicit memory barrier with
544 * the exchange in cds_wfcq_enqueue.
545 */
546 futex_nto1_wake(&viewer_conn_queue.futex);
547 }
548 }
549 }
550
551 exit:
552 error:
553 error_poll_add:
554 error_testpoint:
555 lttng_poll_clean(&events);
556 error_create_poll:
557 if (live_control_sock->fd >= 0) {
558 ret = live_control_sock->ops->close(live_control_sock);
559 if (ret) {
560 PERROR("close");
561 }
562 }
563 lttcomm_destroy_sock(live_control_sock);
564 error_sock_control:
565 if (err) {
566 health_error();
567 DBG("Live viewer listener thread exited with error");
568 }
569 health_unregister(health_relayd);
570 DBG("Live viewer listener thread cleanup complete");
571 if (lttng_relay_stop_threads()) {
572 ERR("Error stopping threads");
573 }
574 return NULL;
575 }
576
577 /*
578 * This thread manages the dispatching of the requests to worker threads
579 */
580 static
581 void *thread_dispatcher(void *data)
582 {
583 int err = -1;
584 ssize_t ret;
585 struct cds_wfcq_node *node;
586 struct relay_connection *conn = NULL;
587
588 DBG("[thread] Live viewer relay dispatcher started");
589
590 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
591
592 if (testpoint(relayd_thread_live_dispatcher)) {
593 goto error_testpoint;
594 }
595
596 health_code_update();
597
598 while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
599 health_code_update();
600
601 /* Atomically prepare the queue futex */
602 futex_nto1_prepare(&viewer_conn_queue.futex);
603
604 do {
605 health_code_update();
606
607 /* Dequeue commands */
608 node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
609 &viewer_conn_queue.tail);
610 if (node == NULL) {
611 DBG("Woken up but nothing in the live-viewer "
612 "relay command queue");
613 /* Continue thread execution */
614 break;
615 }
616 conn = caa_container_of(node, struct relay_connection, qnode);
617 DBG("Dispatching viewer request waiting on sock %d",
618 conn->sock->fd);
619
620 /*
621 * Inform worker thread of the new request. This call is blocking
622 * so we can be assured that the data will be read at some point in
623 * time or wait to the end of the world :)
624 */
625 ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
626 if (ret < 0) {
627 PERROR("write conn pipe");
628 connection_destroy(conn);
629 goto error;
630 }
631 } while (node != NULL);
632
633 /* Futex wait on queue. Blocking call on futex() */
634 health_poll_entry();
635 futex_nto1_wait(&viewer_conn_queue.futex);
636 health_poll_exit();
637 }
638
639 /* Normal exit, no error */
640 err = 0;
641
642 error:
643 error_testpoint:
644 if (err) {
645 health_error();
646 ERR("Health error occurred in %s", __func__);
647 }
648 health_unregister(health_relayd);
649 DBG("Live viewer dispatch thread dying");
650 if (lttng_relay_stop_threads()) {
651 ERR("Error stopping threads");
652 }
653 return NULL;
654 }
655
656 /*
657 * Establish connection with the viewer and check the versions.
658 *
659 * Return 0 on success or else negative value.
660 */
661 static
662 int viewer_connect(struct relay_connection *conn)
663 {
664 int ret;
665 struct lttng_viewer_connect reply, msg;
666
667 assert(conn);
668
669 conn->version_check_done = 1;
670
671 health_code_update();
672
673 DBG("Viewer is establishing a connection to the relayd.");
674
675 ret = recv_request(conn->sock, &msg, sizeof(msg));
676 if (ret < 0) {
677 goto end;
678 }
679
680 health_code_update();
681
682 memset(&reply, 0, sizeof(reply));
683 reply.major = RELAYD_VERSION_COMM_MAJOR;
684 reply.minor = RELAYD_VERSION_COMM_MINOR;
685
686 /* Major versions must be the same */
687 if (reply.major != be32toh(msg.major)) {
688 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
689 reply.major, be32toh(msg.major));
690 ret = -1;
691 goto end;
692 }
693
694 conn->major = reply.major;
695 /* We adapt to the lowest compatible version */
696 if (reply.minor <= be32toh(msg.minor)) {
697 conn->minor = reply.minor;
698 } else {
699 conn->minor = be32toh(msg.minor);
700 }
701
702 if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
703 conn->type = RELAY_VIEWER_COMMAND;
704 } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
705 conn->type = RELAY_VIEWER_NOTIFICATION;
706 } else {
707 ERR("Unknown connection type : %u", be32toh(msg.type));
708 ret = -1;
709 goto end;
710 }
711
712 reply.major = htobe32(reply.major);
713 reply.minor = htobe32(reply.minor);
714 if (conn->type == RELAY_VIEWER_COMMAND) {
715 /*
716 * Increment outside of htobe64 macro, because can be used more than once
717 * within the macro, and thus the operation may be undefined.
718 */
719 last_relay_viewer_session_id++;
720 reply.viewer_session_id = htobe64(last_relay_viewer_session_id);
721 }
722
723 health_code_update();
724
725 ret = send_response(conn->sock, &reply, sizeof(reply));
726 if (ret < 0) {
727 goto end;
728 }
729
730 health_code_update();
731
732 DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
733 ret = 0;
734
735 end:
736 return ret;
737 }
738
739 /*
740 * Send the viewer the list of current sessions.
741 *
742 * Return 0 on success or else a negative value.
743 */
744 static
745 int viewer_list_sessions(struct relay_connection *conn)
746 {
747 int ret;
748 struct lttng_viewer_list_sessions session_list;
749 unsigned long count;
750 long approx_before, approx_after;
751 struct lttng_ht_iter iter;
752 struct lttng_viewer_session send_session;
753 struct relay_session *session;
754
755 DBG("List sessions received");
756
757 rcu_read_lock();
758 cds_lfht_count_nodes(conn->sessions_ht->ht, &approx_before, &count,
759 &approx_after);
760 session_list.sessions_count = htobe32(count);
761
762 health_code_update();
763
764 ret = send_response(conn->sock, &session_list, sizeof(session_list));
765 if (ret < 0) {
766 goto end_unlock;
767 }
768
769 health_code_update();
770
771 cds_lfht_for_each_entry(conn->sessions_ht->ht, &iter.iter, session,
772 session_n.node) {
773 health_code_update();
774
775 strncpy(send_session.session_name, session->session_name,
776 sizeof(send_session.session_name));
777 strncpy(send_session.hostname, session->hostname,
778 sizeof(send_session.hostname));
779 send_session.id = htobe64(session->id);
780 send_session.live_timer = htobe32(session->live_timer);
781 send_session.clients = htobe32(session->viewer_refcount);
782 send_session.streams = htobe32(session->stream_count);
783
784 health_code_update();
785
786 ret = send_response(conn->sock, &send_session, sizeof(send_session));
787 if (ret < 0) {
788 goto end_unlock;
789 }
790 }
791 health_code_update();
792
793 ret = 0;
794 end_unlock:
795 rcu_read_unlock();
796 return ret;
797 }
798
799 /*
800 * Check if a connection is attached to a session.
801 * Return 1 if attached, 0 if not attached, a negative value on error.
802 */
803 static
804 int session_attached(struct relay_connection *conn, uint64_t session_id)
805 {
806 struct relay_session *session;
807 int found = 0;
808
809 if (!conn->viewer_session) {
810 goto end;
811 }
812 cds_list_for_each_entry(session,
813 &conn->viewer_session->sessions_head,
814 viewer_session_list) {
815 if (session->id == session_id) {
816 found = 1;
817 goto end;
818 }
819 }
820
821 end:
822 return found;
823 }
824
825 /*
826 * Delete all streams for a specific session ID.
827 */
828 static void destroy_viewer_streams_by_session(struct relay_session *session)
829 {
830 struct relay_viewer_stream *stream;
831 struct lttng_ht_iter iter;
832
833 assert(session);
834
835 rcu_read_lock();
836 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, stream,
837 stream_n.node) {
838 struct ctf_trace *ctf_trace;
839
840 health_code_update();
841 if (stream->session_id != session->id) {
842 continue;
843 }
844
845 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
846 stream->path_name);
847 assert(ctf_trace);
848
849 viewer_stream_delete(stream);
850
851 if (stream->metadata_flag) {
852 ctf_trace->metadata_sent = 0;
853 ctf_trace->viewer_metadata_stream = NULL;
854 }
855
856 viewer_stream_destroy(ctf_trace, stream);
857 }
858 rcu_read_unlock();
859 }
860
861 static void try_destroy_streams(struct relay_session *session)
862 {
863 struct ctf_trace *ctf_trace;
864 struct lttng_ht_iter iter;
865
866 assert(session);
867
868 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
869 node.node) {
870 /* Attempt to destroy the ctf trace of that session. */
871 ctf_trace_try_destroy(session, ctf_trace);
872 }
873 }
874
875 /*
876 * Cleanup a session.
877 */
878 static void cleanup_session(struct relay_connection *conn,
879 struct relay_session *session)
880 {
881 /*
882 * Very important that this is done before destroying the session so we
883 * can put back every viewer stream reference from the ctf_trace.
884 */
885 destroy_viewer_streams_by_session(session);
886 try_destroy_streams(session);
887 cds_list_del(&session->viewer_session_list);
888 session_viewer_try_destroy(conn->sessions_ht, session);
889 }
890
891 /*
892 * Send the viewer the list of current sessions.
893 */
894 static
895 int viewer_get_new_streams(struct relay_connection *conn)
896 {
897 int ret, send_streams = 0;
898 uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0;
899 struct lttng_viewer_new_streams_request request;
900 struct lttng_viewer_new_streams_response response;
901 struct relay_session *session;
902 uint64_t session_id;
903
904 assert(conn);
905
906 DBG("Get new streams received");
907
908 health_code_update();
909
910 /* Receive the request from the connected client. */
911 ret = recv_request(conn->sock, &request, sizeof(request));
912 if (ret < 0) {
913 goto error;
914 }
915 session_id = be64toh(request.session_id);
916
917 health_code_update();
918
919 memset(&response, 0, sizeof(response));
920
921 rcu_read_lock();
922 session = session_find_by_id(conn->sessions_ht, session_id);
923 if (!session) {
924 DBG("Relay session %" PRIu64 " not found", session_id);
925 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
926 goto send_reply;
927 }
928
929 if (!session_attached(conn, session_id)) {
930 send_streams = 0;
931 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
932 goto send_reply;
933 }
934
935 send_streams = 1;
936 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
937
938 ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, NULL, &nb_unsent,
939 &nb_created);
940 if (ret < 0) {
941 goto end_unlock;
942 }
943 /* Only send back the newly created streams with the unsent ones. */
944 nb_streams = nb_created + nb_unsent;
945 response.streams_count = htobe32(nb_streams);
946
947 /*
948 * If the session is closed and we have no new streams to send,
949 * it means that the viewer has already received the whole trace
950 * for this session and should now close it.
951 */
952 if (nb_streams == 0 && session->close_flag) {
953 send_streams = 0;
954 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
955 /*
956 * Remove the session from the attached list of the connection
957 * and try to destroy it.
958 */
959 cds_list_del(&session->viewer_session_list);
960 cleanup_session(conn, session);
961 goto send_reply;
962 }
963
964 send_reply:
965 health_code_update();
966 ret = send_response(conn->sock, &response, sizeof(response));
967 if (ret < 0) {
968 goto end_unlock;
969 }
970 health_code_update();
971
972 /*
973 * Unknown or empty session, just return gracefully, the viewer knows what
974 * is happening.
975 */
976 if (!send_streams || !nb_streams) {
977 ret = 0;
978 goto end_unlock;
979 }
980
981 /*
982 * Send stream and *DON'T* ignore the sent flag so every viewer streams
983 * that were not sent from that point will be sent to the viewer.
984 */
985 ret = send_viewer_streams(conn->sock, session, 0);
986 if (ret < 0) {
987 goto end_unlock;
988 }
989
990 end_unlock:
991 rcu_read_unlock();
992 error:
993 return ret;
994 }
995
996 /*
997 * Send the viewer the list of current sessions.
998 */
999 static
1000 int viewer_attach_session(struct relay_connection *conn)
1001 {
1002 int send_streams = 0;
1003 ssize_t ret;
1004 uint32_t nb_streams = 0;
1005 enum lttng_viewer_seek seek_type;
1006 struct lttng_viewer_attach_session_request request;
1007 struct lttng_viewer_attach_session_response response;
1008 struct relay_session *session;
1009
1010 assert(conn);
1011
1012 health_code_update();
1013
1014 /* Receive the request from the connected client. */
1015 ret = recv_request(conn->sock, &request, sizeof(request));
1016 if (ret < 0) {
1017 goto error;
1018 }
1019
1020 health_code_update();
1021
1022 memset(&response, 0, sizeof(response));
1023
1024 if (!conn->viewer_session) {
1025 DBG("Client trying to attach before creating a live viewer session");
1026 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1027 goto send_reply;
1028 }
1029
1030 rcu_read_lock();
1031 session = session_find_by_id(conn->sessions_ht,
1032 be64toh(request.session_id));
1033 if (!session) {
1034 DBG("Relay session %" PRIu64 " not found",
1035 be64toh(request.session_id));
1036 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1037 goto send_reply;
1038 }
1039 session_viewer_attach(session);
1040 DBG("Attach session ID %" PRIu64 " received", be64toh(request.session_id));
1041
1042 if (uatomic_read(&session->viewer_refcount) > 1) {
1043 DBG("Already a viewer attached");
1044 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1045 session_viewer_detach(session);
1046 goto send_reply;
1047 } else if (session->live_timer == 0) {
1048 DBG("Not live session");
1049 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1050 goto send_reply;
1051 } else {
1052 send_streams = 1;
1053 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1054 cds_list_add(&session->viewer_session_list,
1055 &conn->viewer_session->sessions_head);
1056 }
1057
1058 switch (be32toh(request.seek)) {
1059 case LTTNG_VIEWER_SEEK_BEGINNING:
1060 case LTTNG_VIEWER_SEEK_LAST:
1061 seek_type = be32toh(request.seek);
1062 break;
1063 default:
1064 ERR("Wrong seek parameter");
1065 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1066 send_streams = 0;
1067 goto send_reply;
1068 }
1069
1070 ret = make_viewer_streams(session, seek_type, &nb_streams, NULL, NULL);
1071 if (ret < 0) {
1072 goto end_unlock;
1073 }
1074 response.streams_count = htobe32(nb_streams);
1075
1076 send_reply:
1077 health_code_update();
1078 ret = send_response(conn->sock, &response, sizeof(response));
1079 if (ret < 0) {
1080 goto end_unlock;
1081 }
1082 health_code_update();
1083
1084 /*
1085 * Unknown or empty session, just return gracefully, the viewer knows what
1086 * is happening.
1087 */
1088 if (!send_streams || !nb_streams) {
1089 ret = 0;
1090 goto end_unlock;
1091 }
1092
1093 /* Send stream and ignore the sent flag. */
1094 ret = send_viewer_streams(conn->sock, session, 1);
1095 if (ret < 0) {
1096 goto end_unlock;
1097 }
1098
1099 end_unlock:
1100 rcu_read_unlock();
1101 error:
1102 return ret;
1103 }
1104
1105 /*
1106 * Open the index file if needed for the given vstream.
1107 *
1108 * If an index file is successfully opened, the index_read_fd of the stream is
1109 * set with it.
1110 *
1111 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1112 */
1113 static int try_open_index(struct relay_viewer_stream *vstream,
1114 struct relay_stream *rstream)
1115 {
1116 int ret = 0;
1117
1118 assert(vstream);
1119 assert(rstream);
1120
1121 if (vstream->index_read_fd >= 0) {
1122 goto end;
1123 }
1124
1125 /*
1126 * First time, we open the index file and at least one index is ready. The
1127 * race between the read and write of the total_index_received is
1128 * acceptable here since the client will be notified to simply come back
1129 * and get the next index.
1130 */
1131 if (rstream->total_index_received <= 0) {
1132 ret = -ENOENT;
1133 goto end;
1134 }
1135 ret = index_open(vstream->path_name, vstream->channel_name,
1136 vstream->tracefile_count, vstream->tracefile_count_current);
1137 if (ret >= 0) {
1138 vstream->index_read_fd = ret;
1139 ret = 0;
1140 goto end;
1141 }
1142
1143 end:
1144 return ret;
1145 }
1146
1147 /*
1148 * Check the status of the index for the given stream. This function updates
1149 * the index structure if needed and can destroy the vstream also for the HUP
1150 * situation.
1151 *
1152 * Return 0 means that we can proceed with the index. A value of 1 means that
1153 * the index has been updated and is ready to be send to the client. A negative
1154 * value indicates an error that can't be handled.
1155 */
1156 static int check_index_status(struct relay_viewer_stream *vstream,
1157 struct relay_stream *rstream, struct ctf_trace *trace,
1158 struct lttng_viewer_index *index)
1159 {
1160 int ret;
1161
1162 assert(vstream);
1163 assert(rstream);
1164 assert(index);
1165 assert(trace);
1166
1167 if (!rstream->close_flag) {
1168 /* Rotate on abort (overwrite). */
1169 if (vstream->abort_flag) {
1170 DBG("Viewer stream %" PRIu64 " rotate because of overwrite",
1171 vstream->stream_handle);
1172 ret = viewer_stream_rotate(vstream, rstream);
1173 if (ret < 0) {
1174 goto error;
1175 } else if (ret == 1) {
1176 /* EOF */
1177 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1178 goto hup;
1179 }
1180 /* ret == 0 means successful so we continue. */
1181 }
1182
1183 /* Check if we are in the same trace file at this point. */
1184 if (rstream->tracefile_count_current == vstream->tracefile_count_current) {
1185 if (rstream->beacon_ts_end != -1ULL &&
1186 vstream->last_sent_index == rstream->total_index_received) {
1187 /*
1188 * We've received a synchronization beacon and the last index
1189 * available has been sent, the index for now is inactive.
1190 */
1191 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1192 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1193 index->stream_id = htobe64(rstream->ctf_stream_id);
1194 goto index_ready;
1195 } else if (rstream->total_index_received <= vstream->last_sent_index
1196 && !vstream->close_write_flag) {
1197 /*
1198 * Reader and writer are working in the same tracefile, so we care
1199 * about the number of index received and sent. Otherwise, we read
1200 * up to EOF.
1201 */
1202 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1203 goto index_ready;
1204 }
1205 }
1206 /* Nothing to do with the index, continue with it. */
1207 ret = 0;
1208 } else if (rstream->close_flag && vstream->close_write_flag &&
1209 vstream->total_index_received == vstream->last_sent_index) {
1210 /* Last index sent and current tracefile closed in write */
1211 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1212 goto hup;
1213 } else {
1214 vstream->close_write_flag = 1;
1215 ret = 0;
1216 }
1217
1218 error:
1219 return ret;
1220
1221 hup:
1222 viewer_stream_delete(vstream);
1223 viewer_stream_destroy(trace, vstream);
1224 index_ready:
1225 return 1;
1226 }
1227
1228 /*
1229 * Send the next index for a stream.
1230 *
1231 * Return 0 on success or else a negative value.
1232 */
1233 static
1234 int viewer_get_next_index(struct relay_connection *conn)
1235 {
1236 int ret;
1237 ssize_t read_ret;
1238 struct lttng_viewer_get_next_index request_index;
1239 struct lttng_viewer_index viewer_index;
1240 struct ctf_packet_index packet_index;
1241 struct relay_viewer_stream *vstream;
1242 struct relay_stream *rstream;
1243 struct ctf_trace *ctf_trace;
1244 struct relay_session *session;
1245
1246 assert(conn);
1247
1248 DBG("Viewer get next index");
1249
1250 health_code_update();
1251
1252 ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1253 if (ret < 0) {
1254 goto end;
1255 }
1256 health_code_update();
1257
1258 rcu_read_lock();
1259 vstream = viewer_stream_find_by_id(be64toh(request_index.stream_id));
1260 if (!vstream) {
1261 ret = -1;
1262 goto end_unlock;
1263 }
1264
1265 session = session_find_by_id(conn->sessions_ht, vstream->session_id);
1266 if (!session) {
1267 ret = -1;
1268 goto end_unlock;
1269 }
1270
1271 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, vstream->path_name);
1272 assert(ctf_trace);
1273
1274 memset(&viewer_index, 0, sizeof(viewer_index));
1275
1276 /*
1277 * The viewer should not ask for index on metadata stream.
1278 */
1279 if (vstream->metadata_flag) {
1280 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1281 goto send_reply;
1282 }
1283
1284 rstream = stream_find_by_id(relay_streams_ht, vstream->stream_handle);
1285 assert(rstream);
1286
1287 /* Try to open an index if one is needed for that stream. */
1288 ret = try_open_index(vstream, rstream);
1289 if (ret < 0) {
1290 if (ret == -ENOENT) {
1291 /*
1292 * The index is created only when the first data packet arrives, it
1293 * might not be ready at the beginning of the session
1294 */
1295 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1296 } else {
1297 /* Unhandled error. */
1298 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1299 }
1300 goto send_reply;
1301 }
1302
1303 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1304 ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1305 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1306 if (ret < 0) {
1307 goto end_unlock;
1308 } else if (ret == 1) {
1309 /*
1310 * This means the viewer index data structure has been populated by the
1311 * check call thus we now send back the reply to the client.
1312 */
1313 goto send_reply;
1314 }
1315 /* At this point, ret MUST be 0 thus we continue with the get. */
1316 assert(!ret);
1317
1318 if (!ctf_trace->metadata_received ||
1319 ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1320 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1321 }
1322
1323 ret = check_new_streams(conn);
1324 if (ret < 0) {
1325 goto end_unlock;
1326 } else if (ret == 1) {
1327 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1328 }
1329
1330 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1331 pthread_mutex_lock(&vstream->overwrite_lock);
1332 if (vstream->abort_flag) {
1333 /* The file is being overwritten by the writer, we cannot use it. */
1334 pthread_mutex_unlock(&vstream->overwrite_lock);
1335 ret = viewer_stream_rotate(vstream, rstream);
1336 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1337 if (ret < 0) {
1338 goto end_unlock;
1339 } else if (ret == 1) {
1340 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1341 viewer_stream_delete(vstream);
1342 viewer_stream_destroy(ctf_trace, vstream);
1343 } else {
1344 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1345 }
1346 goto send_reply;
1347 }
1348
1349 read_ret = lttng_read(vstream->index_read_fd, &packet_index,
1350 sizeof(packet_index));
1351 pthread_mutex_unlock(&vstream->overwrite_lock);
1352 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1353 if (read_ret < 0) {
1354 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1355 viewer_stream_delete(vstream);
1356 viewer_stream_destroy(ctf_trace, vstream);
1357 goto send_reply;
1358 } else if (read_ret < sizeof(packet_index)) {
1359 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1360 if (vstream->close_write_flag) {
1361 ret = viewer_stream_rotate(vstream, rstream);
1362 if (ret < 0) {
1363 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1364 goto end_unlock;
1365 } else if (ret == 1) {
1366 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1367 viewer_stream_delete(vstream);
1368 viewer_stream_destroy(ctf_trace, vstream);
1369 } else {
1370 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1371 }
1372 } else {
1373 ERR("Relay reading index file %d", vstream->index_read_fd);
1374 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1375 }
1376 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1377 goto send_reply;
1378 } else {
1379 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1380 vstream->last_sent_index++;
1381 }
1382
1383 /*
1384 * Indexes are stored in big endian, no need to switch before sending.
1385 */
1386 viewer_index.offset = packet_index.offset;
1387 viewer_index.packet_size = packet_index.packet_size;
1388 viewer_index.content_size = packet_index.content_size;
1389 viewer_index.timestamp_begin = packet_index.timestamp_begin;
1390 viewer_index.timestamp_end = packet_index.timestamp_end;
1391 viewer_index.events_discarded = packet_index.events_discarded;
1392 viewer_index.stream_id = packet_index.stream_id;
1393
1394 send_reply:
1395 viewer_index.flags = htobe32(viewer_index.flags);
1396 health_code_update();
1397
1398 ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1399 if (ret < 0) {
1400 goto end_unlock;
1401 }
1402 health_code_update();
1403
1404 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1405 vstream->last_sent_index, vstream->stream_handle);
1406
1407 end_unlock:
1408 rcu_read_unlock();
1409
1410 end:
1411 return ret;
1412 }
1413
1414 /*
1415 * Send the next index for a stream
1416 *
1417 * Return 0 on success or else a negative value.
1418 */
1419 static
1420 int viewer_get_packet(struct relay_connection *conn)
1421 {
1422 int ret, send_data = 0;
1423 char *data = NULL;
1424 uint32_t len = 0;
1425 ssize_t read_len;
1426 struct lttng_viewer_get_packet get_packet_info;
1427 struct lttng_viewer_trace_packet reply;
1428 struct relay_viewer_stream *stream;
1429 struct relay_session *session;
1430 struct ctf_trace *ctf_trace;
1431
1432 assert(conn);
1433
1434 DBG2("Relay get data packet");
1435
1436 health_code_update();
1437
1438 ret = recv_request(conn->sock, &get_packet_info, sizeof(get_packet_info));
1439 if (ret < 0) {
1440 goto end;
1441 }
1442 health_code_update();
1443
1444 /* From this point on, the error label can be reached. */
1445 memset(&reply, 0, sizeof(reply));
1446
1447 rcu_read_lock();
1448 stream = viewer_stream_find_by_id(be64toh(get_packet_info.stream_id));
1449 if (!stream) {
1450 goto error;
1451 }
1452
1453 session = session_find_by_id(conn->sessions_ht, stream->session_id);
1454 if (!session) {
1455 ret = -1;
1456 goto error;
1457 }
1458
1459 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1460 stream->path_name);
1461 assert(ctf_trace);
1462
1463 /*
1464 * First time we read this stream, we need open the tracefile, we should
1465 * only arrive here if an index has already been sent to the viewer, so the
1466 * tracefile must exist, if it does not it is a fatal error.
1467 */
1468 if (stream->read_fd < 0) {
1469 char fullpath[PATH_MAX];
1470
1471 if (stream->tracefile_count > 0) {
1472 ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, stream->path_name,
1473 stream->channel_name,
1474 stream->tracefile_count_current);
1475 } else {
1476 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1477 stream->channel_name);
1478 }
1479 if (ret < 0) {
1480 goto error;
1481 }
1482 ret = open(fullpath, O_RDONLY);
1483 if (ret < 0) {
1484 PERROR("Relay opening trace file");
1485 goto error;
1486 }
1487 stream->read_fd = ret;
1488 }
1489
1490 if (!ctf_trace->metadata_received ||
1491 ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1492 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1493 reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1494 goto send_reply;
1495 }
1496
1497 ret = check_new_streams(conn);
1498 if (ret < 0) {
1499 goto end_unlock;
1500 } else if (ret == 1) {
1501 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1502 reply.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1503 goto send_reply;
1504 }
1505
1506 len = be32toh(get_packet_info.len);
1507 data = zmalloc(len);
1508 if (!data) {
1509 PERROR("relay data zmalloc");
1510 goto error;
1511 }
1512
1513 ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
1514 if (ret < 0) {
1515 /*
1516 * If the read fd was closed by the streaming side, the
1517 * abort_flag will be set to 1, otherwise it is an error.
1518 */
1519 if (stream->abort_flag == 0) {
1520 PERROR("lseek");
1521 goto error;
1522 }
1523 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1524 goto send_reply;
1525 }
1526 read_len = lttng_read(stream->read_fd, data, len);
1527 if (read_len < len) {
1528 /*
1529 * If the read fd was closed by the streaming side, the
1530 * abort_flag will be set to 1, otherwise it is an error.
1531 */
1532 if (stream->abort_flag == 0) {
1533 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1534 stream->read_fd,
1535 be64toh(get_packet_info.offset));
1536 goto error;
1537 } else {
1538 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1539 goto send_reply;
1540 }
1541 }
1542 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1543 reply.len = htobe32(len);
1544 send_data = 1;
1545 goto send_reply;
1546
1547 error:
1548 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1549
1550 send_reply:
1551 reply.flags = htobe32(reply.flags);
1552
1553 health_code_update();
1554
1555 ret = send_response(conn->sock, &reply, sizeof(reply));
1556 if (ret < 0) {
1557 goto end_unlock;
1558 }
1559 health_code_update();
1560
1561 if (send_data) {
1562 health_code_update();
1563 ret = send_response(conn->sock, data, len);
1564 if (ret < 0) {
1565 goto end_unlock;
1566 }
1567 health_code_update();
1568 }
1569
1570 DBG("Sent %u bytes for stream %" PRIu64, len,
1571 be64toh(get_packet_info.stream_id));
1572
1573 end_unlock:
1574 free(data);
1575 rcu_read_unlock();
1576
1577 end:
1578 return ret;
1579 }
1580
1581 /*
1582 * Send the session's metadata
1583 *
1584 * Return 0 on success else a negative value.
1585 */
1586 static
1587 int viewer_get_metadata(struct relay_connection *conn)
1588 {
1589 int ret = 0;
1590 ssize_t read_len;
1591 uint64_t len = 0;
1592 char *data = NULL;
1593 struct lttng_viewer_get_metadata request;
1594 struct lttng_viewer_metadata_packet reply;
1595 struct relay_viewer_stream *stream;
1596 struct ctf_trace *ctf_trace;
1597 struct relay_session *session;
1598
1599 assert(conn);
1600
1601 DBG("Relay get metadata");
1602
1603 health_code_update();
1604
1605 ret = recv_request(conn->sock, &request, sizeof(request));
1606 if (ret < 0) {
1607 goto end;
1608 }
1609 health_code_update();
1610
1611 memset(&reply, 0, sizeof(reply));
1612
1613 rcu_read_lock();
1614 stream = viewer_stream_find_by_id(be64toh(request.stream_id));
1615 if (!stream || !stream->metadata_flag) {
1616 ERR("Invalid metadata stream");
1617 goto error;
1618 }
1619
1620 session = session_find_by_id(conn->sessions_ht, stream->session_id);
1621 if (!session) {
1622 ret = -1;
1623 goto error;
1624 }
1625
1626 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1627 stream->path_name);
1628 assert(ctf_trace);
1629 assert(ctf_trace->metadata_sent <= ctf_trace->metadata_received);
1630
1631 len = ctf_trace->metadata_received - ctf_trace->metadata_sent;
1632 if (len == 0) {
1633 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1634 goto send_reply;
1635 }
1636
1637 /* first time, we open the metadata file */
1638 if (stream->read_fd < 0) {
1639 char fullpath[PATH_MAX];
1640
1641 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1642 stream->channel_name);
1643 if (ret < 0) {
1644 goto error;
1645 }
1646 ret = open(fullpath, O_RDONLY);
1647 if (ret < 0) {
1648 PERROR("Relay opening metadata file");
1649 goto error;
1650 }
1651 stream->read_fd = ret;
1652 }
1653
1654 reply.len = htobe64(len);
1655 data = zmalloc(len);
1656 if (!data) {
1657 PERROR("viewer metadata zmalloc");
1658 goto error;
1659 }
1660
1661 read_len = lttng_read(stream->read_fd, data, len);
1662 if (read_len < len) {
1663 PERROR("Relay reading metadata file");
1664 goto error;
1665 }
1666 ctf_trace->metadata_sent += read_len;
1667 reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1668 goto send_reply;
1669
1670 error:
1671 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1672
1673 send_reply:
1674 health_code_update();
1675 ret = send_response(conn->sock, &reply, sizeof(reply));
1676 if (ret < 0) {
1677 goto end_unlock;
1678 }
1679 health_code_update();
1680
1681 if (len > 0) {
1682 ret = send_response(conn->sock, data, len);
1683 if (ret < 0) {
1684 goto end_unlock;
1685 }
1686 }
1687
1688 DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1689 be64toh(request.stream_id));
1690
1691 DBG("Metadata sent");
1692
1693 end_unlock:
1694 free(data);
1695 rcu_read_unlock();
1696 end:
1697 return ret;
1698 }
1699
1700 /*
1701 * Create a viewer session.
1702 *
1703 * Return 0 on success or else a negative value.
1704 */
1705 static
1706 int viewer_create_session(struct relay_connection *conn)
1707 {
1708 int ret;
1709 struct lttng_viewer_create_session_response resp;
1710
1711 DBG("Viewer create session received");
1712
1713 memset(&resp, 0, sizeof(resp));
1714 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1715 conn->viewer_session = zmalloc(sizeof(*conn->viewer_session));
1716 if (!conn->viewer_session) {
1717 ERR("Allocation viewer session");
1718 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1719 goto send_reply;
1720 }
1721 CDS_INIT_LIST_HEAD(&conn->viewer_session->sessions_head);
1722
1723 send_reply:
1724 health_code_update();
1725 ret = send_response(conn->sock, &resp, sizeof(resp));
1726 if (ret < 0) {
1727 goto end;
1728 }
1729 health_code_update();
1730 ret = 0;
1731
1732 end:
1733 return ret;
1734 }
1735
1736
1737 /*
1738 * live_relay_unknown_command: send -1 if received unknown command
1739 */
1740 static
1741 void live_relay_unknown_command(struct relay_connection *conn)
1742 {
1743 struct lttcomm_relayd_generic_reply reply;
1744
1745 memset(&reply, 0, sizeof(reply));
1746 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1747 (void) send_response(conn->sock, &reply, sizeof(reply));
1748 }
1749
1750 /*
1751 * Process the commands received on the control socket
1752 */
1753 static
1754 int process_control(struct lttng_viewer_cmd *recv_hdr,
1755 struct relay_connection *conn)
1756 {
1757 int ret = 0;
1758 uint32_t msg_value;
1759
1760 assert(recv_hdr);
1761 assert(conn);
1762
1763 msg_value = be32toh(recv_hdr->cmd);
1764
1765 /*
1766 * Make sure we've done the version check before any command other then a
1767 * new client connection.
1768 */
1769 if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1770 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1771 ret = -1;
1772 goto end;
1773 }
1774
1775 switch (msg_value) {
1776 case LTTNG_VIEWER_CONNECT:
1777 ret = viewer_connect(conn);
1778 break;
1779 case LTTNG_VIEWER_LIST_SESSIONS:
1780 ret = viewer_list_sessions(conn);
1781 break;
1782 case LTTNG_VIEWER_ATTACH_SESSION:
1783 ret = viewer_attach_session(conn);
1784 break;
1785 case LTTNG_VIEWER_GET_NEXT_INDEX:
1786 ret = viewer_get_next_index(conn);
1787 break;
1788 case LTTNG_VIEWER_GET_PACKET:
1789 ret = viewer_get_packet(conn);
1790 break;
1791 case LTTNG_VIEWER_GET_METADATA:
1792 ret = viewer_get_metadata(conn);
1793 break;
1794 case LTTNG_VIEWER_GET_NEW_STREAMS:
1795 ret = viewer_get_new_streams(conn);
1796 break;
1797 case LTTNG_VIEWER_CREATE_SESSION:
1798 ret = viewer_create_session(conn);
1799 break;
1800 default:
1801 ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd));
1802 live_relay_unknown_command(conn);
1803 ret = -1;
1804 goto end;
1805 }
1806
1807 end:
1808 return ret;
1809 }
1810
1811 static
1812 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1813 {
1814 int ret;
1815
1816 assert(events);
1817
1818 (void) lttng_poll_del(events, pollfd);
1819
1820 ret = close(pollfd);
1821 if (ret < 0) {
1822 ERR("Closing pollfd %d", pollfd);
1823 }
1824 }
1825
1826 /*
1827 * Delete and destroy a connection.
1828 *
1829 * RCU read side lock MUST be acquired.
1830 */
1831 static void destroy_connection(struct lttng_ht *relay_connections_ht,
1832 struct relay_connection *conn)
1833 {
1834 struct relay_session *session, *tmp_session;
1835
1836 assert(relay_connections_ht);
1837 assert(conn);
1838
1839 connection_delete(relay_connections_ht, conn);
1840
1841 if (!conn->viewer_session) {
1842 goto end;
1843 }
1844
1845 rcu_read_lock();
1846 cds_list_for_each_entry_safe(session, tmp_session,
1847 &conn->viewer_session->sessions_head,
1848 viewer_session_list) {
1849 DBG("Cleaning connection of session ID %" PRIu64, session->id);
1850 cleanup_session(conn, session);
1851 }
1852 rcu_read_unlock();
1853
1854 end:
1855 connection_destroy(conn);
1856 }
1857
1858 /*
1859 * This thread does the actual work
1860 */
1861 static
1862 void *thread_worker(void *data)
1863 {
1864 int ret, err = -1;
1865 uint32_t nb_fd;
1866 struct lttng_poll_event events;
1867 struct lttng_ht *relay_connections_ht;
1868 struct lttng_ht_iter iter;
1869 struct lttng_viewer_cmd recv_hdr;
1870 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
1871 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
1872 struct relay_connection *destroy_conn;
1873
1874 DBG("[thread] Live viewer relay worker started");
1875
1876 rcu_register_thread();
1877
1878 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1879
1880 if (testpoint(relayd_thread_live_worker)) {
1881 goto error_testpoint;
1882 }
1883
1884 /* table of connections indexed on socket */
1885 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1886 if (!relay_connections_ht) {
1887 goto relay_connections_ht_error;
1888 }
1889
1890 ret = create_thread_poll_set(&events, 2);
1891 if (ret < 0) {
1892 goto error_poll_create;
1893 }
1894
1895 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1896 if (ret < 0) {
1897 goto error;
1898 }
1899
1900 restart:
1901 while (1) {
1902 int i;
1903
1904 health_code_update();
1905
1906 /* Infinite blocking call, waiting for transmission */
1907 DBG3("Relayd live viewer worker thread polling...");
1908 health_poll_entry();
1909 ret = lttng_poll_wait(&events, -1);
1910 health_poll_exit();
1911 if (ret < 0) {
1912 /*
1913 * Restart interrupted system call.
1914 */
1915 if (errno == EINTR) {
1916 goto restart;
1917 }
1918 goto error;
1919 }
1920
1921 nb_fd = ret;
1922
1923 /*
1924 * Process control. The control connection is prioritised so we don't
1925 * starve it with high throughput tracing data on the data
1926 * connection.
1927 */
1928 for (i = 0; i < nb_fd; i++) {
1929 /* Fetch once the poll data */
1930 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1931 int pollfd = LTTNG_POLL_GETFD(&events, i);
1932
1933 health_code_update();
1934
1935 if (!revents) {
1936 /* No activity for this FD (poll implementation). */
1937 continue;
1938 }
1939
1940 /* Thread quit pipe has been closed. Killing thread. */
1941 ret = check_thread_quit_pipe(pollfd, revents);
1942 if (ret) {
1943 err = 0;
1944 goto exit;
1945 }
1946
1947 /* Inspect the relay conn pipe for new connection */
1948 if (pollfd == live_conn_pipe[0]) {
1949 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1950 ERR("Relay live pipe error");
1951 goto error;
1952 } else if (revents & LPOLLIN) {
1953 struct relay_connection *conn;
1954
1955 ret = lttng_read(live_conn_pipe[0], &conn, sizeof(conn));
1956 if (ret < 0) {
1957 goto error;
1958 }
1959 conn->sessions_ht = sessions_ht;
1960 connection_init(conn);
1961 lttng_poll_add(&events, conn->sock->fd,
1962 LPOLLIN | LPOLLRDHUP);
1963 rcu_read_lock();
1964 lttng_ht_add_unique_ulong(relay_connections_ht,
1965 &conn->sock_n);
1966 rcu_read_unlock();
1967 DBG("Connection socket %d added", conn->sock->fd);
1968 }
1969 } else {
1970 struct relay_connection *conn;
1971
1972 rcu_read_lock();
1973 conn = connection_find_by_sock(relay_connections_ht, pollfd);
1974 /* If not found, there is a synchronization issue. */
1975 assert(conn);
1976
1977 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1978 cleanup_connection_pollfd(&events, pollfd);
1979 destroy_connection(relay_connections_ht, conn);
1980 } else if (revents & LPOLLIN) {
1981 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
1982 sizeof(recv_hdr), 0);
1983 if (ret <= 0) {
1984 /* Connection closed */
1985 cleanup_connection_pollfd(&events, pollfd);
1986 destroy_connection(relay_connections_ht, conn);
1987 DBG("Viewer control conn closed with %d", pollfd);
1988 } else {
1989 ret = process_control(&recv_hdr, conn);
1990 if (ret < 0) {
1991 /* Clear the session on error. */
1992 cleanup_connection_pollfd(&events, pollfd);
1993 destroy_connection(relay_connections_ht, conn);
1994 DBG("Viewer connection closed with %d", pollfd);
1995 }
1996 }
1997 }
1998 rcu_read_unlock();
1999 }
2000 }
2001 }
2002
2003 exit:
2004 error:
2005 lttng_poll_clean(&events);
2006
2007 /* Cleanup reamaining connection object. */
2008 rcu_read_lock();
2009 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter,
2010 destroy_conn,
2011 sock_n.node) {
2012 health_code_update();
2013 destroy_connection(relay_connections_ht, destroy_conn);
2014 }
2015 rcu_read_unlock();
2016 error_poll_create:
2017 lttng_ht_destroy(relay_connections_ht);
2018 relay_connections_ht_error:
2019 /* Close relay conn pipes */
2020 utils_close_pipe(live_conn_pipe);
2021 if (err) {
2022 DBG("Viewer worker thread exited with error");
2023 }
2024 DBG("Viewer worker thread cleanup complete");
2025 error_testpoint:
2026 if (err) {
2027 health_error();
2028 ERR("Health error occurred in %s", __func__);
2029 }
2030 health_unregister(health_relayd);
2031 if (lttng_relay_stop_threads()) {
2032 ERR("Error stopping threads");
2033 }
2034 rcu_unregister_thread();
2035 return NULL;
2036 }
2037
2038 /*
2039 * Create the relay command pipe to wake thread_manage_apps.
2040 * Closed in cleanup().
2041 */
2042 static int create_conn_pipe(void)
2043 {
2044 return utils_create_pipe_cloexec(live_conn_pipe);
2045 }
2046
2047 int relayd_live_join(void)
2048 {
2049 int ret, retval = 0;
2050 void *status;
2051
2052 ret = pthread_join(live_listener_thread, &status);
2053 if (ret) {
2054 errno = ret;
2055 PERROR("pthread_join live listener");
2056 retval = -1;
2057 }
2058
2059 ret = pthread_join(live_worker_thread, &status);
2060 if (ret) {
2061 errno = ret;
2062 PERROR("pthread_join live worker");
2063 retval = -1;
2064 }
2065
2066 ret = pthread_join(live_dispatcher_thread, &status);
2067 if (ret) {
2068 errno = ret;
2069 PERROR("pthread_join live dispatcher");
2070 retval = -1;
2071 }
2072
2073 cleanup_relayd_live();
2074
2075 return retval;
2076 }
2077
2078 /*
2079 * main
2080 */
2081 int relayd_live_create(struct lttng_uri *uri,
2082 struct relay_local_data *relay_ctx)
2083 {
2084 int ret = 0, retval = 0;
2085 void *status;
2086 int is_root;
2087
2088 if (!uri) {
2089 retval = -1;
2090 goto exit_init_data;
2091 }
2092 live_uri = uri;
2093
2094 /* Check if daemon is UID = 0 */
2095 is_root = !getuid();
2096
2097 if (!is_root) {
2098 if (live_uri->port < 1024) {
2099 ERR("Need to be root to use ports < 1024");
2100 retval = -1;
2101 goto exit_init_data;
2102 }
2103 }
2104
2105 /* Setup the thread apps communication pipe. */
2106 if (create_conn_pipe()) {
2107 retval = -1;
2108 goto exit_init_data;
2109 }
2110
2111 /* Init relay command queue. */
2112 cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2113
2114 /* Set up max poll set size */
2115 if (lttng_poll_set_max_size()) {
2116 retval = -1;
2117 goto exit_init_data;
2118 }
2119
2120 /* Setup the dispatcher thread */
2121 ret = pthread_create(&live_dispatcher_thread, NULL,
2122 thread_dispatcher, (void *) NULL);
2123 if (ret) {
2124 errno = ret;
2125 PERROR("pthread_create viewer dispatcher");
2126 retval = -1;
2127 goto exit_dispatcher_thread;
2128 }
2129
2130 /* Setup the worker thread */
2131 ret = pthread_create(&live_worker_thread, NULL,
2132 thread_worker, relay_ctx);
2133 if (ret) {
2134 errno = ret;
2135 PERROR("pthread_create viewer worker");
2136 retval = -1;
2137 goto exit_worker_thread;
2138 }
2139
2140 /* Setup the listener thread */
2141 ret = pthread_create(&live_listener_thread, NULL,
2142 thread_listener, (void *) NULL);
2143 if (ret) {
2144 errno = ret;
2145 PERROR("pthread_create viewer listener");
2146 retval = -1;
2147 goto exit_listener_thread;
2148 }
2149
2150 /*
2151 * All OK, started all threads.
2152 */
2153 return retval;
2154
2155 /*
2156 * Join on the live_listener_thread should anything be added after
2157 * the live_listener thread's creation.
2158 */
2159
2160 exit_listener_thread:
2161
2162 ret = pthread_join(live_worker_thread, &status);
2163 if (ret) {
2164 errno = ret;
2165 PERROR("pthread_join live worker");
2166 retval = -1;
2167 }
2168 exit_worker_thread:
2169
2170 ret = pthread_join(live_dispatcher_thread, &status);
2171 if (ret) {
2172 errno = ret;
2173 PERROR("pthread_join live dispatcher");
2174 retval = -1;
2175 }
2176 exit_dispatcher_thread:
2177
2178 exit_init_data:
2179 cleanup_relayd_live();
2180
2181 return retval;
2182 }
This page took 0.131774 seconds and 3 git commands to generate.