Fix: compat poll: add missing empty revents checks
[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 #include <getopt.h>
21 #include <grp.h>
22 #include <limits.h>
23 #include <pthread.h>
24 #include <signal.h>
25 #include <stdio.h>
26 #include <stdlib.h>
27 #include <string.h>
28 #include <sys/mman.h>
29 #include <sys/mount.h>
30 #include <sys/resource.h>
31 #include <sys/socket.h>
32 #include <sys/stat.h>
33 #include <sys/types.h>
34 #include <sys/wait.h>
35 #include <inttypes.h>
36 #include <urcu/futex.h>
37 #include <urcu/uatomic.h>
38 #include <unistd.h>
39 #include <fcntl.h>
40 #include <config.h>
41
42 #include <lttng/lttng.h>
43 #include <common/common.h>
44 #include <common/compat/poll.h>
45 #include <common/compat/socket.h>
46 #include <common/defaults.h>
47 #include <common/futex.h>
48 #include <common/index/index.h>
49 #include <common/sessiond-comm/sessiond-comm.h>
50 #include <common/sessiond-comm/inet.h>
51 #include <common/sessiond-comm/relayd.h>
52 #include <common/uri.h>
53 #include <common/utils.h>
54
55 #include "cmd.h"
56 #include "live.h"
57 #include "lttng-relayd.h"
58 #include "utils.h"
59 #include "health-relayd.h"
60 #include "testpoint.h"
61 #include "viewer-stream.h"
62 #include "stream.h"
63 #include "session.h"
64 #include "ctf-trace.h"
65 #include "connection.h"
66
67 static struct lttng_uri *live_uri;
68
69 /*
70 * This pipe is used to inform the worker thread that a command is queued and
71 * ready to be processed.
72 */
73 static int live_conn_pipe[2] = { -1, -1 };
74
75 /* Shared between threads */
76 static int live_dispatch_thread_exit;
77
78 static pthread_t live_listener_thread;
79 static pthread_t live_dispatcher_thread;
80 static pthread_t live_worker_thread;
81
82 /*
83 * Relay command queue.
84 *
85 * The live_thread_listener and live_thread_dispatcher communicate with this
86 * queue.
87 */
88 static struct relay_conn_queue viewer_conn_queue;
89
90 static uint64_t last_relay_viewer_session_id;
91
92 /*
93 * Cleanup the daemon
94 */
95 static
96 void cleanup(void)
97 {
98 DBG("Cleaning up");
99
100 free(live_uri);
101 }
102
103 /*
104 * Receive a request buffer using a given socket, destination allocated buffer
105 * of length size.
106 *
107 * Return the size of the received message or else a negative value on error
108 * with errno being set by recvmsg() syscall.
109 */
110 static
111 ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
112 {
113 ssize_t ret;
114
115 assert(sock);
116 assert(buf);
117
118 ret = sock->ops->recvmsg(sock, buf, size, 0);
119 if (ret < 0 || ret != size) {
120 if (ret == 0) {
121 /* Orderly shutdown. Not necessary to print an error. */
122 DBG("Socket %d did an orderly shutdown", sock->fd);
123 } else {
124 ERR("Relay failed to receive request.");
125 }
126 ret = -1;
127 }
128
129 return ret;
130 }
131
132 /*
133 * Send a response buffer using a given socket, source allocated buffer of
134 * length size.
135 *
136 * Return the size of the sent message or else a negative value on error with
137 * errno being set by sendmsg() syscall.
138 */
139 static
140 ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
141 {
142 ssize_t ret;
143
144 assert(sock);
145 assert(buf);
146
147 ret = sock->ops->sendmsg(sock, buf, size, 0);
148 if (ret < 0) {
149 ERR("Relayd failed to send response.");
150 }
151
152 return ret;
153 }
154
155 /*
156 * Atomically check if new streams got added in one of the sessions attached
157 * and reset the flag to 0.
158 *
159 * Returns 1 if new streams got added, 0 if nothing changed, a negative value
160 * on error.
161 */
162 static
163 int check_new_streams(struct relay_connection *conn)
164 {
165 struct relay_session *session;
166 unsigned long current_val;
167 int ret = 0;
168
169 if (!conn->viewer_session) {
170 goto end;
171 }
172 cds_list_for_each_entry(session,
173 &conn->viewer_session->sessions_head,
174 viewer_session_list) {
175 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
176 ret = current_val;
177 if (ret == 1) {
178 goto end;
179 }
180 }
181
182 end:
183 return ret;
184 }
185
186 /*
187 * Send viewer streams to the given socket. The ignore_sent_flag indicates if
188 * this function should ignore the sent flag or not.
189 *
190 * Return 0 on success or else a negative value.
191 */
192 static
193 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
194 struct relay_session *session, unsigned int ignore_sent_flag)
195 {
196 ssize_t ret;
197 struct lttng_viewer_stream send_stream;
198 struct lttng_ht_iter iter;
199 struct relay_viewer_stream *vstream;
200
201 assert(session);
202
203 rcu_read_lock();
204
205 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
206 stream_n.node) {
207 struct ctf_trace *ctf_trace;
208
209 health_code_update();
210
211 /* Ignore if not the same session. */
212 if (vstream->session_id != session->id ||
213 (!ignore_sent_flag && vstream->sent_flag)) {
214 continue;
215 }
216
217 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
218 vstream->path_name);
219 assert(ctf_trace);
220
221 send_stream.id = htobe64(vstream->stream_handle);
222 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
223 send_stream.metadata_flag = htobe32(vstream->metadata_flag);
224 strncpy(send_stream.path_name, vstream->path_name,
225 sizeof(send_stream.path_name));
226 strncpy(send_stream.channel_name, vstream->channel_name,
227 sizeof(send_stream.channel_name));
228
229 DBG("Sending stream %" PRIu64 " to viewer", vstream->stream_handle);
230 ret = send_response(sock, &send_stream, sizeof(send_stream));
231 if (ret < 0) {
232 goto end_unlock;
233 }
234 vstream->sent_flag = 1;
235 }
236
237 ret = 0;
238
239 end_unlock:
240 rcu_read_unlock();
241 return ret;
242 }
243
244 /*
245 * Create every viewer stream possible for the given session with the seek
246 * type. Three counters *can* be return which are in order the total amount of
247 * viewer stream of the session, the number of unsent stream and the number of
248 * stream created. Those counters can be NULL and thus will be ignored.
249 *
250 * Return 0 on success or else a negative value.
251 */
252 static
253 int make_viewer_streams(struct relay_session *session,
254 enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent,
255 uint32_t *nb_created)
256 {
257 int ret;
258 struct lttng_ht_iter iter;
259 struct ctf_trace *ctf_trace;
260
261 assert(session);
262
263 /*
264 * This is to make sure we create viewer streams for a full received
265 * channel. For instance, if we have 8 streams for a channel that are
266 * concurrently being flagged ready, we can end up creating just a subset
267 * of the 8 streams (the ones that are flagged). This lock avoids this
268 * limbo state.
269 */
270 pthread_mutex_lock(&session->viewer_ready_lock);
271
272 /*
273 * Create viewer streams for relay streams that are ready to be used for a
274 * the given session id only.
275 */
276 rcu_read_lock();
277 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
278 node.node) {
279 struct relay_stream *stream;
280
281 health_code_update();
282
283 if (ctf_trace->invalid_flag) {
284 continue;
285 }
286
287 cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) {
288 struct relay_viewer_stream *vstream;
289
290 if (!stream->viewer_ready) {
291 continue;
292 }
293
294 vstream = viewer_stream_find_by_id(stream->stream_handle);
295 if (!vstream) {
296 vstream = viewer_stream_create(stream, seek_t, ctf_trace);
297 if (!vstream) {
298 ret = -1;
299 goto error_unlock;
300 }
301 /* Acquire reference to ctf_trace. */
302 ctf_trace_get_ref(ctf_trace);
303
304 if (nb_created) {
305 /* Update number of created stream counter. */
306 (*nb_created)++;
307 }
308 } else if (!vstream->sent_flag && nb_unsent) {
309 /* Update number of unsent stream counter. */
310 (*nb_unsent)++;
311 }
312 /* Update number of total stream counter. */
313 if (nb_total) {
314 (*nb_total)++;
315 }
316 }
317 }
318
319 ret = 0;
320
321 error_unlock:
322 rcu_read_unlock();
323 pthread_mutex_unlock(&session->viewer_ready_lock);
324 return ret;
325 }
326
327 /*
328 * Write to writable pipe used to notify a thread.
329 */
330 static
331 int notify_thread_pipe(int wpipe)
332 {
333 ssize_t ret;
334
335 ret = lttng_write(wpipe, "!", 1);
336 if (ret < 1) {
337 PERROR("write poll pipe");
338 }
339
340 return (int) ret;
341 }
342
343 /*
344 * Stop all threads by closing the thread quit pipe.
345 */
346 static
347 void stop_threads(void)
348 {
349 int ret;
350
351 /* Stopping all threads */
352 DBG("Terminating all live threads");
353 ret = notify_thread_pipe(thread_quit_pipe[1]);
354 if (ret < 0) {
355 ERR("write error on thread quit pipe");
356 }
357
358 /* Dispatch thread */
359 CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
360 futex_nto1_wake(&viewer_conn_queue.futex);
361 }
362
363 /*
364 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
365 */
366 static
367 int create_thread_poll_set(struct lttng_poll_event *events, int size)
368 {
369 int ret;
370
371 if (events == NULL || size == 0) {
372 ret = -1;
373 goto error;
374 }
375
376 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
377 if (ret < 0) {
378 goto error;
379 }
380
381 /* Add quit pipe */
382 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
383 if (ret < 0) {
384 goto error;
385 }
386
387 return 0;
388
389 error:
390 return ret;
391 }
392
393 /*
394 * Check if the thread quit pipe was triggered.
395 *
396 * Return 1 if it was triggered else 0;
397 */
398 static
399 int check_thread_quit_pipe(int fd, uint32_t events)
400 {
401 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
402 return 1;
403 }
404
405 return 0;
406 }
407
408 /*
409 * Create and init socket from uri.
410 */
411 static
412 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
413 {
414 int ret;
415 struct lttcomm_sock *sock = NULL;
416
417 sock = lttcomm_alloc_sock_from_uri(uri);
418 if (sock == NULL) {
419 ERR("Allocating socket");
420 goto error;
421 }
422
423 ret = lttcomm_create_sock(sock);
424 if (ret < 0) {
425 goto error;
426 }
427 DBG("Listening on sock %d for live", sock->fd);
428
429 ret = sock->ops->bind(sock);
430 if (ret < 0) {
431 goto error;
432 }
433
434 ret = sock->ops->listen(sock, -1);
435 if (ret < 0) {
436 goto error;
437
438 }
439
440 return sock;
441
442 error:
443 if (sock) {
444 lttcomm_destroy_sock(sock);
445 }
446 return NULL;
447 }
448
449 /*
450 * This thread manages the listening for new connections on the network
451 */
452 static
453 void *thread_listener(void *data)
454 {
455 int i, ret, pollfd, err = -1;
456 uint32_t revents, nb_fd;
457 struct lttng_poll_event events;
458 struct lttcomm_sock *live_control_sock;
459
460 DBG("[thread] Relay live listener started");
461
462 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
463
464 health_code_update();
465
466 live_control_sock = init_socket(live_uri);
467 if (!live_control_sock) {
468 goto error_sock_control;
469 }
470
471 /* Pass 2 as size here for the thread quit pipe and control sockets. */
472 ret = create_thread_poll_set(&events, 2);
473 if (ret < 0) {
474 goto error_create_poll;
475 }
476
477 /* Add the control socket */
478 ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
479 if (ret < 0) {
480 goto error_poll_add;
481 }
482
483 lttng_relay_notify_ready();
484
485 if (testpoint(relayd_thread_live_listener)) {
486 goto error_testpoint;
487 }
488
489 while (1) {
490 health_code_update();
491
492 DBG("Listener accepting live viewers connections");
493
494 restart:
495 health_poll_entry();
496 ret = lttng_poll_wait(&events, -1);
497 health_poll_exit();
498 if (ret < 0) {
499 /*
500 * Restart interrupted system call.
501 */
502 if (errno == EINTR) {
503 goto restart;
504 }
505 goto error;
506 }
507 nb_fd = ret;
508
509 DBG("Relay new viewer connection received");
510 for (i = 0; i < nb_fd; i++) {
511 health_code_update();
512
513 /* Fetch once the poll data */
514 revents = LTTNG_POLL_GETEV(&events, i);
515 pollfd = LTTNG_POLL_GETFD(&events, i);
516
517 if (!revents) {
518 /* No activity for this FD (poll implementation). */
519 continue;
520 }
521
522 /* Thread quit pipe has been closed. Killing thread. */
523 ret = check_thread_quit_pipe(pollfd, revents);
524 if (ret) {
525 err = 0;
526 goto exit;
527 }
528
529 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
530 ERR("socket poll error");
531 goto error;
532 } else if (revents & LPOLLIN) {
533 /*
534 * Get allocated in this thread, enqueued to a global queue,
535 * dequeued and freed in the worker thread.
536 */
537 int val = 1;
538 struct relay_connection *new_conn;
539 struct lttcomm_sock *newsock;
540
541 new_conn = connection_create();
542 if (!new_conn) {
543 goto error;
544 }
545
546 newsock = live_control_sock->ops->accept(live_control_sock);
547 if (!newsock) {
548 PERROR("accepting control sock");
549 connection_free(new_conn);
550 goto error;
551 }
552 DBG("Relay viewer connection accepted socket %d", newsock->fd);
553
554 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
555 sizeof(val));
556 if (ret < 0) {
557 PERROR("setsockopt inet");
558 lttcomm_destroy_sock(newsock);
559 connection_free(new_conn);
560 goto error;
561 }
562 new_conn->sock = newsock;
563
564 /* Enqueue request for the dispatcher thread. */
565 cds_wfq_enqueue(&viewer_conn_queue.queue, &new_conn->qnode);
566
567 /*
568 * Wake the dispatch queue futex. Implicit memory barrier with
569 * the exchange in cds_wfq_enqueue.
570 */
571 futex_nto1_wake(&viewer_conn_queue.futex);
572 }
573 }
574 }
575
576 exit:
577 error:
578 error_poll_add:
579 error_testpoint:
580 lttng_poll_clean(&events);
581 error_create_poll:
582 if (live_control_sock->fd >= 0) {
583 ret = live_control_sock->ops->close(live_control_sock);
584 if (ret) {
585 PERROR("close");
586 }
587 }
588 lttcomm_destroy_sock(live_control_sock);
589 error_sock_control:
590 if (err) {
591 health_error();
592 DBG("Live viewer listener thread exited with error");
593 }
594 health_unregister(health_relayd);
595 DBG("Live viewer listener thread cleanup complete");
596 stop_threads();
597 return NULL;
598 }
599
600 /*
601 * This thread manages the dispatching of the requests to worker threads
602 */
603 static
604 void *thread_dispatcher(void *data)
605 {
606 int err = -1;
607 ssize_t ret;
608 struct cds_wfq_node *node;
609 struct relay_connection *conn = NULL;
610
611 DBG("[thread] Live viewer relay dispatcher started");
612
613 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
614
615 if (testpoint(relayd_thread_live_dispatcher)) {
616 goto error_testpoint;
617 }
618
619 health_code_update();
620
621 while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
622 health_code_update();
623
624 /* Atomically prepare the queue futex */
625 futex_nto1_prepare(&viewer_conn_queue.futex);
626
627 do {
628 health_code_update();
629
630 /* Dequeue commands */
631 node = cds_wfq_dequeue_blocking(&viewer_conn_queue.queue);
632 if (node == NULL) {
633 DBG("Woken up but nothing in the live-viewer "
634 "relay command queue");
635 /* Continue thread execution */
636 break;
637 }
638 conn = caa_container_of(node, struct relay_connection, qnode);
639 DBG("Dispatching viewer request waiting on sock %d",
640 conn->sock->fd);
641
642 /*
643 * Inform worker thread of the new request. This call is blocking
644 * so we can be assured that the data will be read at some point in
645 * time or wait to the end of the world :)
646 */
647 ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
648 if (ret < 0) {
649 PERROR("write conn pipe");
650 connection_destroy(conn);
651 goto error;
652 }
653 } while (node != NULL);
654
655 /* Futex wait on queue. Blocking call on futex() */
656 health_poll_entry();
657 futex_nto1_wait(&viewer_conn_queue.futex);
658 health_poll_exit();
659 }
660
661 /* Normal exit, no error */
662 err = 0;
663
664 error:
665 error_testpoint:
666 if (err) {
667 health_error();
668 ERR("Health error occurred in %s", __func__);
669 }
670 health_unregister(health_relayd);
671 DBG("Live viewer dispatch thread dying");
672 stop_threads();
673 return NULL;
674 }
675
676 /*
677 * Establish connection with the viewer and check the versions.
678 *
679 * Return 0 on success or else negative value.
680 */
681 static
682 int viewer_connect(struct relay_connection *conn)
683 {
684 int ret;
685 struct lttng_viewer_connect reply, msg;
686
687 assert(conn);
688
689 conn->version_check_done = 1;
690
691 health_code_update();
692
693 DBG("Viewer is establishing a connection to the relayd.");
694
695 ret = recv_request(conn->sock, &msg, sizeof(msg));
696 if (ret < 0) {
697 goto end;
698 }
699
700 health_code_update();
701
702 memset(&reply, 0, sizeof(reply));
703 reply.major = RELAYD_VERSION_COMM_MAJOR;
704 reply.minor = RELAYD_VERSION_COMM_MINOR;
705
706 /* Major versions must be the same */
707 if (reply.major != be32toh(msg.major)) {
708 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
709 reply.major, be32toh(msg.major));
710 ret = -1;
711 goto end;
712 }
713
714 conn->major = reply.major;
715 /* We adapt to the lowest compatible version */
716 if (reply.minor <= be32toh(msg.minor)) {
717 conn->minor = reply.minor;
718 } else {
719 conn->minor = be32toh(msg.minor);
720 }
721
722 if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
723 conn->type = RELAY_VIEWER_COMMAND;
724 } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
725 conn->type = RELAY_VIEWER_NOTIFICATION;
726 } else {
727 ERR("Unknown connection type : %u", be32toh(msg.type));
728 ret = -1;
729 goto end;
730 }
731
732 reply.major = htobe32(reply.major);
733 reply.minor = htobe32(reply.minor);
734 if (conn->type == RELAY_VIEWER_COMMAND) {
735 reply.viewer_session_id = htobe64(++last_relay_viewer_session_id);
736 }
737
738 health_code_update();
739
740 ret = send_response(conn->sock, &reply, sizeof(reply));
741 if (ret < 0) {
742 goto end;
743 }
744
745 health_code_update();
746
747 DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
748 ret = 0;
749
750 end:
751 return ret;
752 }
753
754 /*
755 * Send the viewer the list of current sessions.
756 *
757 * Return 0 on success or else a negative value.
758 */
759 static
760 int viewer_list_sessions(struct relay_connection *conn)
761 {
762 int ret;
763 struct lttng_viewer_list_sessions session_list;
764 unsigned long count;
765 long approx_before, approx_after;
766 struct lttng_ht_iter iter;
767 struct lttng_viewer_session send_session;
768 struct relay_session *session;
769
770 DBG("List sessions received");
771
772 rcu_read_lock();
773 cds_lfht_count_nodes(conn->sessions_ht->ht, &approx_before, &count,
774 &approx_after);
775 session_list.sessions_count = htobe32(count);
776
777 health_code_update();
778
779 ret = send_response(conn->sock, &session_list, sizeof(session_list));
780 if (ret < 0) {
781 goto end_unlock;
782 }
783
784 health_code_update();
785
786 cds_lfht_for_each_entry(conn->sessions_ht->ht, &iter.iter, session,
787 session_n.node) {
788 health_code_update();
789
790 strncpy(send_session.session_name, session->session_name,
791 sizeof(send_session.session_name));
792 strncpy(send_session.hostname, session->hostname,
793 sizeof(send_session.hostname));
794 send_session.id = htobe64(session->id);
795 send_session.live_timer = htobe32(session->live_timer);
796 send_session.clients = htobe32(session->viewer_refcount);
797 send_session.streams = htobe32(session->stream_count);
798
799 health_code_update();
800
801 ret = send_response(conn->sock, &send_session, sizeof(send_session));
802 if (ret < 0) {
803 goto end_unlock;
804 }
805 }
806 health_code_update();
807
808 rcu_read_unlock();
809 ret = 0;
810 goto end;
811
812 end_unlock:
813 rcu_read_unlock();
814
815 end:
816 return ret;
817 }
818
819 /*
820 * Check if a connection is attached to a session.
821 * Return 1 if attached, 0 if not attached, a negative value on error.
822 */
823 static
824 int session_attached(struct relay_connection *conn, uint64_t session_id)
825 {
826 struct relay_session *session;
827 int found = 0;
828
829 if (!conn->viewer_session) {
830 goto end;
831 }
832 cds_list_for_each_entry(session,
833 &conn->viewer_session->sessions_head,
834 viewer_session_list) {
835 if (session->id == session_id) {
836 found = 1;
837 goto end;
838 }
839 }
840
841 end:
842 return found;
843 }
844
845 /*
846 * Delete all streams for a specific session ID.
847 */
848 static void destroy_viewer_streams_by_session(struct relay_session *session)
849 {
850 struct relay_viewer_stream *stream;
851 struct lttng_ht_iter iter;
852
853 assert(session);
854
855 rcu_read_lock();
856 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, stream,
857 stream_n.node) {
858 struct ctf_trace *ctf_trace;
859
860 health_code_update();
861 if (stream->session_id != session->id) {
862 continue;
863 }
864
865 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
866 stream->path_name);
867 assert(ctf_trace);
868
869 viewer_stream_delete(stream);
870
871 if (stream->metadata_flag) {
872 ctf_trace->metadata_sent = 0;
873 ctf_trace->viewer_metadata_stream = NULL;
874 }
875
876 viewer_stream_destroy(ctf_trace, stream);
877 }
878 rcu_read_unlock();
879 }
880
881 static void try_destroy_streams(struct relay_session *session)
882 {
883 struct ctf_trace *ctf_trace;
884 struct lttng_ht_iter iter;
885
886 assert(session);
887
888 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
889 node.node) {
890 /* Attempt to destroy the ctf trace of that session. */
891 ctf_trace_try_destroy(session, ctf_trace);
892 }
893 }
894
895 /*
896 * Cleanup a session.
897 */
898 static void cleanup_session(struct relay_connection *conn,
899 struct relay_session *session)
900 {
901 /*
902 * Very important that this is done before destroying the session so we
903 * can put back every viewer stream reference from the ctf_trace.
904 */
905 destroy_viewer_streams_by_session(session);
906 try_destroy_streams(session);
907 cds_list_del(&session->viewer_session_list);
908 session_viewer_try_destroy(conn->sessions_ht, session);
909 }
910
911 /*
912 * Send the viewer the list of current sessions.
913 */
914 static
915 int viewer_get_new_streams(struct relay_connection *conn)
916 {
917 int ret, send_streams = 0;
918 uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0;
919 struct lttng_viewer_new_streams_request request;
920 struct lttng_viewer_new_streams_response response;
921 struct relay_session *session;
922 uint64_t session_id;
923
924 assert(conn);
925
926 DBG("Get new streams received");
927
928 health_code_update();
929
930 /* Receive the request from the connected client. */
931 ret = recv_request(conn->sock, &request, sizeof(request));
932 if (ret < 0) {
933 goto error;
934 }
935 session_id = be64toh(request.session_id);
936
937 health_code_update();
938
939 memset(&response, 0, sizeof(response));
940
941 rcu_read_lock();
942 session = session_find_by_id(conn->sessions_ht, session_id);
943 if (!session) {
944 DBG("Relay session %" PRIu64 " not found", session_id);
945 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
946 goto send_reply;
947 }
948
949 if (!session_attached(conn, session_id)) {
950 send_streams = 0;
951 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
952 goto send_reply;
953 }
954
955 send_streams = 1;
956 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
957
958 ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, NULL, &nb_unsent,
959 &nb_created);
960 if (ret < 0) {
961 goto end_unlock;
962 }
963 /* Only send back the newly created streams with the unsent ones. */
964 nb_streams = nb_created + nb_unsent;
965 response.streams_count = htobe32(nb_streams);
966
967 /*
968 * If the session is closed and we have no new streams to send,
969 * it means that the viewer has already received the whole trace
970 * for this session and should now close it.
971 */
972 if (nb_streams == 0 && session->close_flag) {
973 send_streams = 0;
974 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
975 /*
976 * Remove the session from the attached list of the connection
977 * and try to destroy it.
978 */
979 cds_list_del(&session->viewer_session_list);
980 cleanup_session(conn, session);
981 goto send_reply;
982 }
983
984 send_reply:
985 health_code_update();
986 ret = send_response(conn->sock, &response, sizeof(response));
987 if (ret < 0) {
988 goto end_unlock;
989 }
990 health_code_update();
991
992 /*
993 * Unknown or empty session, just return gracefully, the viewer knows what
994 * is happening.
995 */
996 if (!send_streams || !nb_streams) {
997 ret = 0;
998 goto end_unlock;
999 }
1000
1001 /*
1002 * Send stream and *DON'T* ignore the sent flag so every viewer streams
1003 * that were not sent from that point will be sent to the viewer.
1004 */
1005 ret = send_viewer_streams(conn->sock, session, 0);
1006 if (ret < 0) {
1007 goto end_unlock;
1008 }
1009
1010 end_unlock:
1011 rcu_read_unlock();
1012 error:
1013 return ret;
1014 }
1015
1016 /*
1017 * Send the viewer the list of current sessions.
1018 */
1019 static
1020 int viewer_attach_session(struct relay_connection *conn)
1021 {
1022 int send_streams = 0;
1023 ssize_t ret;
1024 uint32_t nb_streams = 0;
1025 enum lttng_viewer_seek seek_type;
1026 struct lttng_viewer_attach_session_request request;
1027 struct lttng_viewer_attach_session_response response;
1028 struct relay_session *session;
1029
1030 assert(conn);
1031
1032 health_code_update();
1033
1034 /* Receive the request from the connected client. */
1035 ret = recv_request(conn->sock, &request, sizeof(request));
1036 if (ret < 0) {
1037 goto error;
1038 }
1039
1040 health_code_update();
1041
1042 memset(&response, 0, sizeof(response));
1043
1044 if (!conn->viewer_session) {
1045 DBG("Client trying to attach before creating a live viewer session");
1046 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1047 goto send_reply;
1048 }
1049
1050 rcu_read_lock();
1051 session = session_find_by_id(conn->sessions_ht,
1052 be64toh(request.session_id));
1053 if (!session) {
1054 DBG("Relay session %" PRIu64 " not found",
1055 be64toh(request.session_id));
1056 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1057 goto send_reply;
1058 }
1059 session_viewer_attach(session);
1060 DBG("Attach session ID %" PRIu64 " received", be64toh(request.session_id));
1061
1062 if (uatomic_read(&session->viewer_refcount) > 1) {
1063 DBG("Already a viewer attached");
1064 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1065 session_viewer_detach(session);
1066 goto send_reply;
1067 } else if (session->live_timer == 0) {
1068 DBG("Not live session");
1069 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1070 goto send_reply;
1071 } else {
1072 send_streams = 1;
1073 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1074 cds_list_add(&session->viewer_session_list,
1075 &conn->viewer_session->sessions_head);
1076 }
1077
1078 switch (be32toh(request.seek)) {
1079 case LTTNG_VIEWER_SEEK_BEGINNING:
1080 case LTTNG_VIEWER_SEEK_LAST:
1081 seek_type = be32toh(request.seek);
1082 break;
1083 default:
1084 ERR("Wrong seek parameter");
1085 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1086 send_streams = 0;
1087 goto send_reply;
1088 }
1089
1090 ret = make_viewer_streams(session, seek_type, &nb_streams, NULL, NULL);
1091 if (ret < 0) {
1092 goto end_unlock;
1093 }
1094 response.streams_count = htobe32(nb_streams);
1095
1096 send_reply:
1097 health_code_update();
1098 ret = send_response(conn->sock, &response, sizeof(response));
1099 if (ret < 0) {
1100 goto end_unlock;
1101 }
1102 health_code_update();
1103
1104 /*
1105 * Unknown or empty session, just return gracefully, the viewer knows what
1106 * is happening.
1107 */
1108 if (!send_streams || !nb_streams) {
1109 ret = 0;
1110 goto end_unlock;
1111 }
1112
1113 /* Send stream and ignore the sent flag. */
1114 ret = send_viewer_streams(conn->sock, session, 1);
1115 if (ret < 0) {
1116 goto end_unlock;
1117 }
1118
1119 end_unlock:
1120 rcu_read_unlock();
1121 error:
1122 return ret;
1123 }
1124
1125 /*
1126 * Open the index file if needed for the given vstream.
1127 *
1128 * If an index file is successfully opened, the index_read_fd of the stream is
1129 * set with it.
1130 *
1131 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1132 */
1133 static int try_open_index(struct relay_viewer_stream *vstream,
1134 struct relay_stream *rstream)
1135 {
1136 int ret = 0;
1137
1138 assert(vstream);
1139 assert(rstream);
1140
1141 if (vstream->index_read_fd >= 0) {
1142 goto end;
1143 }
1144
1145 /*
1146 * First time, we open the index file and at least one index is ready. The
1147 * race between the read and write of the total_index_received is
1148 * acceptable here since the client will be notified to simply come back
1149 * and get the next index.
1150 */
1151 if (rstream->total_index_received <= 0) {
1152 ret = -ENOENT;
1153 goto end;
1154 }
1155 ret = index_open(vstream->path_name, vstream->channel_name,
1156 vstream->tracefile_count, vstream->tracefile_count_current);
1157 if (ret >= 0) {
1158 vstream->index_read_fd = ret;
1159 ret = 0;
1160 goto end;
1161 }
1162
1163 end:
1164 return ret;
1165 }
1166
1167 /*
1168 * Check the status of the index for the given stream. This function updates
1169 * the index structure if needed and can destroy the vstream also for the HUP
1170 * situation.
1171 *
1172 * Return 0 means that we can proceed with the index. A value of 1 means that
1173 * the index has been updated and is ready to be send to the client. A negative
1174 * value indicates an error that can't be handled.
1175 */
1176 static int check_index_status(struct relay_viewer_stream *vstream,
1177 struct relay_stream *rstream, struct ctf_trace *trace,
1178 struct lttng_viewer_index *index)
1179 {
1180 int ret;
1181
1182 assert(vstream);
1183 assert(rstream);
1184 assert(index);
1185 assert(trace);
1186
1187 if (!rstream->close_flag) {
1188 /* Rotate on abort (overwrite). */
1189 if (vstream->abort_flag) {
1190 DBG("Viewer stream %" PRIu64 " rotate because of overwrite",
1191 vstream->stream_handle);
1192 ret = viewer_stream_rotate(vstream, rstream);
1193 if (ret < 0) {
1194 goto error;
1195 } else if (ret == 1) {
1196 /* EOF */
1197 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1198 goto hup;
1199 }
1200 /* ret == 0 means successful so we continue. */
1201 }
1202
1203 /* Check if we are in the same trace file at this point. */
1204 if (rstream->tracefile_count_current == vstream->tracefile_count_current) {
1205 if (rstream->beacon_ts_end != -1ULL &&
1206 vstream->last_sent_index == rstream->total_index_received) {
1207 /*
1208 * We've received a synchronization beacon and the last index
1209 * available has been sent, the index for now is inactive.
1210 */
1211 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1212 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1213 index->stream_id = htobe64(rstream->ctf_stream_id);
1214 goto index_ready;
1215 } else if (rstream->total_index_received <= vstream->last_sent_index
1216 && !vstream->close_write_flag) {
1217 /*
1218 * Reader and writer are working in the same tracefile, so we care
1219 * about the number of index received and sent. Otherwise, we read
1220 * up to EOF.
1221 */
1222 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1223 goto index_ready;
1224 }
1225 }
1226 /* Nothing to do with the index, continue with it. */
1227 ret = 0;
1228 } else if (rstream->close_flag && vstream->close_write_flag &&
1229 vstream->total_index_received == vstream->last_sent_index) {
1230 /* Last index sent and current tracefile closed in write */
1231 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1232 goto hup;
1233 } else {
1234 vstream->close_write_flag = 1;
1235 ret = 0;
1236 }
1237
1238 error:
1239 return ret;
1240
1241 hup:
1242 viewer_stream_delete(vstream);
1243 viewer_stream_destroy(trace, vstream);
1244 index_ready:
1245 return 1;
1246 }
1247
1248 /*
1249 * Send the next index for a stream.
1250 *
1251 * Return 0 on success or else a negative value.
1252 */
1253 static
1254 int viewer_get_next_index(struct relay_connection *conn)
1255 {
1256 int ret;
1257 ssize_t read_ret;
1258 struct lttng_viewer_get_next_index request_index;
1259 struct lttng_viewer_index viewer_index;
1260 struct ctf_packet_index packet_index;
1261 struct relay_viewer_stream *vstream;
1262 struct relay_stream *rstream;
1263 struct ctf_trace *ctf_trace;
1264 struct relay_session *session;
1265
1266 assert(conn);
1267
1268 DBG("Viewer get next index");
1269
1270 health_code_update();
1271
1272 ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1273 if (ret < 0) {
1274 goto end;
1275 }
1276 health_code_update();
1277
1278 rcu_read_lock();
1279 vstream = viewer_stream_find_by_id(be64toh(request_index.stream_id));
1280 if (!vstream) {
1281 ret = -1;
1282 goto end_unlock;
1283 }
1284
1285 session = session_find_by_id(conn->sessions_ht, vstream->session_id);
1286 if (!session) {
1287 ret = -1;
1288 goto end_unlock;
1289 }
1290
1291 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, vstream->path_name);
1292 assert(ctf_trace);
1293
1294 memset(&viewer_index, 0, sizeof(viewer_index));
1295
1296 /*
1297 * The viewer should not ask for index on metadata stream.
1298 */
1299 if (vstream->metadata_flag) {
1300 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1301 goto send_reply;
1302 }
1303
1304 rstream = stream_find_by_id(relay_streams_ht, vstream->stream_handle);
1305 assert(rstream);
1306
1307 /* Try to open an index if one is needed for that stream. */
1308 ret = try_open_index(vstream, rstream);
1309 if (ret < 0) {
1310 if (ret == -ENOENT) {
1311 /*
1312 * The index is created only when the first data packet arrives, it
1313 * might not be ready at the beginning of the session
1314 */
1315 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1316 } else {
1317 /* Unhandled error. */
1318 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1319 }
1320 goto send_reply;
1321 }
1322
1323 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1324 ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1325 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1326 if (ret < 0) {
1327 goto end_unlock;
1328 } else if (ret == 1) {
1329 /*
1330 * This means the viewer index data structure has been populated by the
1331 * check call thus we now send back the reply to the client.
1332 */
1333 goto send_reply;
1334 }
1335 /* At this point, ret MUST be 0 thus we continue with the get. */
1336 assert(!ret);
1337
1338 if (!ctf_trace->metadata_received ||
1339 ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1340 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1341 }
1342
1343 ret = check_new_streams(conn);
1344 if (ret < 0) {
1345 goto end_unlock;
1346 } else if (ret == 1) {
1347 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1348 }
1349
1350 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1351 pthread_mutex_lock(&vstream->overwrite_lock);
1352 if (vstream->abort_flag) {
1353 /* The file is being overwritten by the writer, we cannot use it. */
1354 pthread_mutex_unlock(&vstream->overwrite_lock);
1355 ret = viewer_stream_rotate(vstream, rstream);
1356 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1357 if (ret < 0) {
1358 goto end_unlock;
1359 } else if (ret == 1) {
1360 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1361 viewer_stream_delete(vstream);
1362 viewer_stream_destroy(ctf_trace, vstream);
1363 } else {
1364 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1365 }
1366 goto send_reply;
1367 }
1368
1369 read_ret = lttng_read(vstream->index_read_fd, &packet_index,
1370 sizeof(packet_index));
1371 pthread_mutex_unlock(&vstream->overwrite_lock);
1372 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1373 if (read_ret < 0) {
1374 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1375 viewer_stream_delete(vstream);
1376 viewer_stream_destroy(ctf_trace, vstream);
1377 goto send_reply;
1378 } else if (read_ret < sizeof(packet_index)) {
1379 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1380 if (vstream->close_write_flag) {
1381 ret = viewer_stream_rotate(vstream, rstream);
1382 if (ret < 0) {
1383 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1384 goto end_unlock;
1385 } else if (ret == 1) {
1386 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1387 viewer_stream_delete(vstream);
1388 viewer_stream_destroy(ctf_trace, vstream);
1389 } else {
1390 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1391 }
1392 } else {
1393 ERR("Relay reading index file %d", vstream->index_read_fd);
1394 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1395 }
1396 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1397 goto send_reply;
1398 } else {
1399 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1400 vstream->last_sent_index++;
1401 }
1402
1403 /*
1404 * Indexes are stored in big endian, no need to switch before sending.
1405 */
1406 viewer_index.offset = packet_index.offset;
1407 viewer_index.packet_size = packet_index.packet_size;
1408 viewer_index.content_size = packet_index.content_size;
1409 viewer_index.timestamp_begin = packet_index.timestamp_begin;
1410 viewer_index.timestamp_end = packet_index.timestamp_end;
1411 viewer_index.events_discarded = packet_index.events_discarded;
1412 viewer_index.stream_id = packet_index.stream_id;
1413
1414 send_reply:
1415 viewer_index.flags = htobe32(viewer_index.flags);
1416 health_code_update();
1417
1418 ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1419 if (ret < 0) {
1420 goto end_unlock;
1421 }
1422 health_code_update();
1423
1424 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1425 vstream->last_sent_index, vstream->stream_handle);
1426
1427 end_unlock:
1428 rcu_read_unlock();
1429
1430 end:
1431 return ret;
1432 }
1433
1434 /*
1435 * Send the next index for a stream
1436 *
1437 * Return 0 on success or else a negative value.
1438 */
1439 static
1440 int viewer_get_packet(struct relay_connection *conn)
1441 {
1442 int ret, send_data = 0;
1443 char *data = NULL;
1444 uint32_t len = 0;
1445 ssize_t read_len;
1446 struct lttng_viewer_get_packet get_packet_info;
1447 struct lttng_viewer_trace_packet reply;
1448 struct relay_viewer_stream *stream;
1449 struct relay_session *session;
1450 struct ctf_trace *ctf_trace;
1451
1452 assert(conn);
1453
1454 DBG2("Relay get data packet");
1455
1456 health_code_update();
1457
1458 ret = recv_request(conn->sock, &get_packet_info, sizeof(get_packet_info));
1459 if (ret < 0) {
1460 goto end;
1461 }
1462 health_code_update();
1463
1464 /* From this point on, the error label can be reached. */
1465 memset(&reply, 0, sizeof(reply));
1466
1467 rcu_read_lock();
1468 stream = viewer_stream_find_by_id(be64toh(get_packet_info.stream_id));
1469 if (!stream) {
1470 goto error;
1471 }
1472
1473 session = session_find_by_id(conn->sessions_ht, stream->session_id);
1474 if (!session) {
1475 ret = -1;
1476 goto error;
1477 }
1478
1479 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1480 stream->path_name);
1481 assert(ctf_trace);
1482
1483 /*
1484 * First time we read this stream, we need open the tracefile, we should
1485 * only arrive here if an index has already been sent to the viewer, so the
1486 * tracefile must exist, if it does not it is a fatal error.
1487 */
1488 if (stream->read_fd < 0) {
1489 char fullpath[PATH_MAX];
1490
1491 if (stream->tracefile_count > 0) {
1492 ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, stream->path_name,
1493 stream->channel_name,
1494 stream->tracefile_count_current);
1495 } else {
1496 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1497 stream->channel_name);
1498 }
1499 if (ret < 0) {
1500 goto error;
1501 }
1502 ret = open(fullpath, O_RDONLY);
1503 if (ret < 0) {
1504 PERROR("Relay opening trace file");
1505 goto error;
1506 }
1507 stream->read_fd = ret;
1508 }
1509
1510 if (!ctf_trace->metadata_received ||
1511 ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1512 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1513 reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1514 goto send_reply;
1515 }
1516
1517 ret = check_new_streams(conn);
1518 if (ret < 0) {
1519 goto end_unlock;
1520 } else if (ret == 1) {
1521 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1522 reply.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1523 goto send_reply;
1524 }
1525
1526 len = be32toh(get_packet_info.len);
1527 data = zmalloc(len);
1528 if (!data) {
1529 PERROR("relay data zmalloc");
1530 goto error;
1531 }
1532
1533 ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
1534 if (ret < 0) {
1535 /*
1536 * If the read fd was closed by the streaming side, the
1537 * abort_flag will be set to 1, otherwise it is an error.
1538 */
1539 if (stream->abort_flag == 0) {
1540 PERROR("lseek");
1541 goto error;
1542 }
1543 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1544 goto send_reply;
1545 }
1546 read_len = lttng_read(stream->read_fd, data, len);
1547 if (read_len < len) {
1548 /*
1549 * If the read fd was closed by the streaming side, the
1550 * abort_flag will be set to 1, otherwise it is an error.
1551 */
1552 if (stream->abort_flag == 0) {
1553 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1554 stream->read_fd,
1555 be64toh(get_packet_info.offset));
1556 goto error;
1557 } else {
1558 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1559 goto send_reply;
1560 }
1561 }
1562 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1563 reply.len = htobe32(len);
1564 send_data = 1;
1565 goto send_reply;
1566
1567 error:
1568 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1569
1570 send_reply:
1571 reply.flags = htobe32(reply.flags);
1572
1573 health_code_update();
1574
1575 ret = send_response(conn->sock, &reply, sizeof(reply));
1576 if (ret < 0) {
1577 goto end_unlock;
1578 }
1579 health_code_update();
1580
1581 if (send_data) {
1582 health_code_update();
1583 ret = send_response(conn->sock, data, len);
1584 if (ret < 0) {
1585 goto end_unlock;
1586 }
1587 health_code_update();
1588 }
1589
1590 DBG("Sent %u bytes for stream %" PRIu64, len,
1591 be64toh(get_packet_info.stream_id));
1592
1593 end_unlock:
1594 free(data);
1595 rcu_read_unlock();
1596
1597 end:
1598 return ret;
1599 }
1600
1601 /*
1602 * Send the session's metadata
1603 *
1604 * Return 0 on success else a negative value.
1605 */
1606 static
1607 int viewer_get_metadata(struct relay_connection *conn)
1608 {
1609 int ret = 0;
1610 ssize_t read_len;
1611 uint64_t len = 0;
1612 char *data = NULL;
1613 struct lttng_viewer_get_metadata request;
1614 struct lttng_viewer_metadata_packet reply;
1615 struct relay_viewer_stream *stream;
1616 struct ctf_trace *ctf_trace;
1617 struct relay_session *session;
1618
1619 assert(conn);
1620
1621 DBG("Relay get metadata");
1622
1623 health_code_update();
1624
1625 ret = recv_request(conn->sock, &request, sizeof(request));
1626 if (ret < 0) {
1627 goto end;
1628 }
1629 health_code_update();
1630
1631 memset(&reply, 0, sizeof(reply));
1632
1633 rcu_read_lock();
1634 stream = viewer_stream_find_by_id(be64toh(request.stream_id));
1635 if (!stream || !stream->metadata_flag) {
1636 ERR("Invalid metadata stream");
1637 goto error;
1638 }
1639
1640 session = session_find_by_id(conn->sessions_ht, stream->session_id);
1641 if (!session) {
1642 ret = -1;
1643 goto error;
1644 }
1645
1646 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1647 stream->path_name);
1648 assert(ctf_trace);
1649 assert(ctf_trace->metadata_sent <= ctf_trace->metadata_received);
1650
1651 len = ctf_trace->metadata_received - ctf_trace->metadata_sent;
1652 if (len == 0) {
1653 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1654 goto send_reply;
1655 }
1656
1657 /* first time, we open the metadata file */
1658 if (stream->read_fd < 0) {
1659 char fullpath[PATH_MAX];
1660
1661 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1662 stream->channel_name);
1663 if (ret < 0) {
1664 goto error;
1665 }
1666 ret = open(fullpath, O_RDONLY);
1667 if (ret < 0) {
1668 PERROR("Relay opening metadata file");
1669 goto error;
1670 }
1671 stream->read_fd = ret;
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(stream->read_fd, data, len);
1682 if (read_len < len) {
1683 PERROR("Relay reading metadata file");
1684 goto error;
1685 }
1686 ctf_trace->metadata_sent += read_len;
1687 reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1688 goto send_reply;
1689
1690 error:
1691 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1692
1693 send_reply:
1694 health_code_update();
1695 ret = send_response(conn->sock, &reply, sizeof(reply));
1696 if (ret < 0) {
1697 goto end_unlock;
1698 }
1699 health_code_update();
1700
1701 if (len > 0) {
1702 ret = send_response(conn->sock, data, len);
1703 if (ret < 0) {
1704 goto end_unlock;
1705 }
1706 }
1707
1708 DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1709 be64toh(request.stream_id));
1710
1711 DBG("Metadata sent");
1712
1713 end_unlock:
1714 free(data);
1715 rcu_read_unlock();
1716 end:
1717 return ret;
1718 }
1719
1720 /*
1721 * Create a viewer session.
1722 *
1723 * Return 0 on success or else a negative value.
1724 */
1725 static
1726 int viewer_create_session(struct relay_connection *conn)
1727 {
1728 int ret;
1729 struct lttng_viewer_create_session_response resp;
1730
1731 DBG("Viewer create session received");
1732
1733 memset(&resp, 0, sizeof(resp));
1734 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1735 conn->viewer_session = zmalloc(sizeof(*conn->viewer_session));
1736 if (!conn->viewer_session) {
1737 ERR("Allocation viewer session");
1738 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1739 goto send_reply;
1740 }
1741 CDS_INIT_LIST_HEAD(&conn->viewer_session->sessions_head);
1742
1743 send_reply:
1744 health_code_update();
1745 ret = send_response(conn->sock, &resp, sizeof(resp));
1746 if (ret < 0) {
1747 goto end;
1748 }
1749 health_code_update();
1750 ret = 0;
1751
1752 end:
1753 return ret;
1754 }
1755
1756
1757 /*
1758 * live_relay_unknown_command: send -1 if received unknown command
1759 */
1760 static
1761 void live_relay_unknown_command(struct relay_connection *conn)
1762 {
1763 struct lttcomm_relayd_generic_reply reply;
1764
1765 memset(&reply, 0, sizeof(reply));
1766 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1767 (void) send_response(conn->sock, &reply, sizeof(reply));
1768 }
1769
1770 /*
1771 * Process the commands received on the control socket
1772 */
1773 static
1774 int process_control(struct lttng_viewer_cmd *recv_hdr,
1775 struct relay_connection *conn)
1776 {
1777 int ret = 0;
1778 uint32_t msg_value;
1779
1780 assert(recv_hdr);
1781 assert(conn);
1782
1783 msg_value = be32toh(recv_hdr->cmd);
1784
1785 /*
1786 * Make sure we've done the version check before any command other then a
1787 * new client connection.
1788 */
1789 if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1790 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1791 ret = -1;
1792 goto end;
1793 }
1794
1795 switch (msg_value) {
1796 case LTTNG_VIEWER_CONNECT:
1797 ret = viewer_connect(conn);
1798 break;
1799 case LTTNG_VIEWER_LIST_SESSIONS:
1800 ret = viewer_list_sessions(conn);
1801 break;
1802 case LTTNG_VIEWER_ATTACH_SESSION:
1803 ret = viewer_attach_session(conn);
1804 break;
1805 case LTTNG_VIEWER_GET_NEXT_INDEX:
1806 ret = viewer_get_next_index(conn);
1807 break;
1808 case LTTNG_VIEWER_GET_PACKET:
1809 ret = viewer_get_packet(conn);
1810 break;
1811 case LTTNG_VIEWER_GET_METADATA:
1812 ret = viewer_get_metadata(conn);
1813 break;
1814 case LTTNG_VIEWER_GET_NEW_STREAMS:
1815 ret = viewer_get_new_streams(conn);
1816 break;
1817 case LTTNG_VIEWER_CREATE_SESSION:
1818 ret = viewer_create_session(conn);
1819 break;
1820 default:
1821 ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd));
1822 live_relay_unknown_command(conn);
1823 ret = -1;
1824 goto end;
1825 }
1826
1827 end:
1828 return ret;
1829 }
1830
1831 static
1832 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1833 {
1834 int ret;
1835
1836 assert(events);
1837
1838 (void) lttng_poll_del(events, pollfd);
1839
1840 ret = close(pollfd);
1841 if (ret < 0) {
1842 ERR("Closing pollfd %d", pollfd);
1843 }
1844 }
1845
1846 /*
1847 * Delete and destroy a connection.
1848 *
1849 * RCU read side lock MUST be acquired.
1850 */
1851 static void destroy_connection(struct lttng_ht *relay_connections_ht,
1852 struct relay_connection *conn)
1853 {
1854 struct relay_session *session, *tmp_session;
1855
1856 assert(relay_connections_ht);
1857 assert(conn);
1858
1859 connection_delete(relay_connections_ht, conn);
1860
1861 if (!conn->viewer_session) {
1862 goto end;
1863 }
1864
1865 rcu_read_lock();
1866 cds_list_for_each_entry_safe(session, tmp_session,
1867 &conn->viewer_session->sessions_head,
1868 viewer_session_list) {
1869 DBG("Cleaning connection of session ID %" PRIu64, session->id);
1870 cleanup_session(conn, session);
1871 }
1872 rcu_read_unlock();
1873
1874 end:
1875 connection_destroy(conn);
1876 }
1877
1878 /*
1879 * This thread does the actual work
1880 */
1881 static
1882 void *thread_worker(void *data)
1883 {
1884 int ret, err = -1;
1885 uint32_t nb_fd;
1886 struct relay_connection *conn;
1887 struct lttng_poll_event events;
1888 struct lttng_ht *relay_connections_ht;
1889 struct lttng_ht_iter iter;
1890 struct lttng_viewer_cmd recv_hdr;
1891 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
1892 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
1893
1894 DBG("[thread] Live viewer relay worker started");
1895
1896 rcu_register_thread();
1897
1898 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1899
1900 if (testpoint(relayd_thread_live_worker)) {
1901 goto error_testpoint;
1902 }
1903
1904 /* table of connections indexed on socket */
1905 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1906 if (!relay_connections_ht) {
1907 goto relay_connections_ht_error;
1908 }
1909
1910 ret = create_thread_poll_set(&events, 2);
1911 if (ret < 0) {
1912 goto error_poll_create;
1913 }
1914
1915 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1916 if (ret < 0) {
1917 goto error;
1918 }
1919
1920 restart:
1921 while (1) {
1922 int i;
1923
1924 health_code_update();
1925
1926 /* Infinite blocking call, waiting for transmission */
1927 DBG3("Relayd live viewer worker thread polling...");
1928 health_poll_entry();
1929 ret = lttng_poll_wait(&events, -1);
1930 health_poll_exit();
1931 if (ret < 0) {
1932 /*
1933 * Restart interrupted system call.
1934 */
1935 if (errno == EINTR) {
1936 goto restart;
1937 }
1938 goto error;
1939 }
1940
1941 nb_fd = ret;
1942
1943 /*
1944 * Process control. The control connection is prioritised so we don't
1945 * starve it with high throughput tracing data on the data
1946 * connection.
1947 */
1948 for (i = 0; i < nb_fd; i++) {
1949 /* Fetch once the poll data */
1950 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1951 int pollfd = LTTNG_POLL_GETFD(&events, i);
1952
1953 health_code_update();
1954
1955 if (!revents) {
1956 /* No activity for this FD (poll implementation). */
1957 continue;
1958 }
1959
1960 /* Thread quit pipe has been closed. Killing thread. */
1961 ret = check_thread_quit_pipe(pollfd, revents);
1962 if (ret) {
1963 err = 0;
1964 goto exit;
1965 }
1966
1967 /* Inspect the relay conn pipe for new connection */
1968 if (pollfd == live_conn_pipe[0]) {
1969 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1970 ERR("Relay live pipe error");
1971 goto error;
1972 } else if (revents & LPOLLIN) {
1973 ret = lttng_read(live_conn_pipe[0], &conn, sizeof(conn));
1974 if (ret < 0) {
1975 goto error;
1976 }
1977 conn->sessions_ht = sessions_ht;
1978 connection_init(conn);
1979 lttng_poll_add(&events, conn->sock->fd,
1980 LPOLLIN | LPOLLRDHUP);
1981 rcu_read_lock();
1982 lttng_ht_add_unique_ulong(relay_connections_ht,
1983 &conn->sock_n);
1984 rcu_read_unlock();
1985 DBG("Connection socket %d added", conn->sock->fd);
1986 }
1987 } else {
1988 rcu_read_lock();
1989 conn = connection_find_by_sock(relay_connections_ht, pollfd);
1990 /* If not found, there is a synchronization issue. */
1991 assert(conn);
1992
1993 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1994 cleanup_connection_pollfd(&events, pollfd);
1995 destroy_connection(relay_connections_ht, conn);
1996 } else if (revents & LPOLLIN) {
1997 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
1998 sizeof(recv_hdr), 0);
1999 if (ret <= 0) {
2000 /* Connection closed */
2001 cleanup_connection_pollfd(&events, pollfd);
2002 destroy_connection(relay_connections_ht, conn);
2003 DBG("Viewer control conn closed with %d", pollfd);
2004 } else {
2005 ret = process_control(&recv_hdr, conn);
2006 if (ret < 0) {
2007 /* Clear the session on error. */
2008 cleanup_connection_pollfd(&events, pollfd);
2009 destroy_connection(relay_connections_ht, conn);
2010 DBG("Viewer connection closed with %d", pollfd);
2011 }
2012 }
2013 }
2014 rcu_read_unlock();
2015 }
2016 }
2017 }
2018
2019 exit:
2020 error:
2021 lttng_poll_clean(&events);
2022
2023 /* Cleanup reamaining connection object. */
2024 rcu_read_lock();
2025 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn,
2026 sock_n.node) {
2027 health_code_update();
2028 destroy_connection(relay_connections_ht, conn);
2029 }
2030 rcu_read_unlock();
2031 error_poll_create:
2032 lttng_ht_destroy(relay_connections_ht);
2033 relay_connections_ht_error:
2034 /* Close relay conn pipes */
2035 utils_close_pipe(live_conn_pipe);
2036 if (err) {
2037 DBG("Viewer worker thread exited with error");
2038 }
2039 DBG("Viewer worker thread cleanup complete");
2040 error_testpoint:
2041 if (err) {
2042 health_error();
2043 ERR("Health error occurred in %s", __func__);
2044 }
2045 health_unregister(health_relayd);
2046 stop_threads();
2047 rcu_unregister_thread();
2048 return NULL;
2049 }
2050
2051 /*
2052 * Create the relay command pipe to wake thread_manage_apps.
2053 * Closed in cleanup().
2054 */
2055 static int create_conn_pipe(void)
2056 {
2057 int ret;
2058
2059 ret = utils_create_pipe_cloexec(live_conn_pipe);
2060
2061 return ret;
2062 }
2063
2064 void live_stop_threads(void)
2065 {
2066 int ret;
2067 void *status;
2068
2069 stop_threads();
2070
2071 ret = pthread_join(live_listener_thread, &status);
2072 if (ret != 0) {
2073 PERROR("pthread_join live listener");
2074 goto error; /* join error, exit without cleanup */
2075 }
2076
2077 ret = pthread_join(live_worker_thread, &status);
2078 if (ret != 0) {
2079 PERROR("pthread_join live worker");
2080 goto error; /* join error, exit without cleanup */
2081 }
2082
2083 ret = pthread_join(live_dispatcher_thread, &status);
2084 if (ret != 0) {
2085 PERROR("pthread_join live dispatcher");
2086 goto error; /* join error, exit without cleanup */
2087 }
2088
2089 cleanup();
2090
2091 error:
2092 return;
2093 }
2094
2095 /*
2096 * main
2097 */
2098 int live_start_threads(struct lttng_uri *uri,
2099 struct relay_local_data *relay_ctx)
2100 {
2101 int ret = 0;
2102 void *status;
2103 int is_root;
2104
2105 assert(uri);
2106 live_uri = uri;
2107
2108 /* Check if daemon is UID = 0 */
2109 is_root = !getuid();
2110
2111 if (!is_root) {
2112 if (live_uri->port < 1024) {
2113 ERR("Need to be root to use ports < 1024");
2114 ret = -1;
2115 goto exit;
2116 }
2117 }
2118
2119 /* Setup the thread apps communication pipe. */
2120 if ((ret = create_conn_pipe()) < 0) {
2121 goto exit;
2122 }
2123
2124 /* Init relay command queue. */
2125 cds_wfq_init(&viewer_conn_queue.queue);
2126
2127 /* Set up max poll set size */
2128 lttng_poll_set_max_size();
2129
2130 /* Setup the dispatcher thread */
2131 ret = pthread_create(&live_dispatcher_thread, NULL,
2132 thread_dispatcher, (void *) NULL);
2133 if (ret != 0) {
2134 PERROR("pthread_create viewer dispatcher");
2135 goto exit_dispatcher;
2136 }
2137
2138 /* Setup the worker thread */
2139 ret = pthread_create(&live_worker_thread, NULL,
2140 thread_worker, relay_ctx);
2141 if (ret != 0) {
2142 PERROR("pthread_create viewer worker");
2143 goto exit_worker;
2144 }
2145
2146 /* Setup the listener thread */
2147 ret = pthread_create(&live_listener_thread, NULL,
2148 thread_listener, (void *) NULL);
2149 if (ret != 0) {
2150 PERROR("pthread_create viewer listener");
2151 goto exit_listener;
2152 }
2153
2154 ret = 0;
2155 goto end;
2156
2157 exit_listener:
2158 ret = pthread_join(live_listener_thread, &status);
2159 if (ret != 0) {
2160 PERROR("pthread_join live listener");
2161 goto error; /* join error, exit without cleanup */
2162 }
2163
2164 exit_worker:
2165 ret = pthread_join(live_worker_thread, &status);
2166 if (ret != 0) {
2167 PERROR("pthread_join live worker");
2168 goto error; /* join error, exit without cleanup */
2169 }
2170
2171 exit_dispatcher:
2172 ret = pthread_join(live_dispatcher_thread, &status);
2173 if (ret != 0) {
2174 PERROR("pthread_join live dispatcher");
2175 goto error; /* join error, exit without cleanup */
2176 }
2177
2178 exit:
2179 cleanup();
2180
2181 end:
2182 error:
2183 return ret;
2184 }
This page took 0.073641 seconds and 4 git commands to generate.