projects / lttng-tools.git / blob
? search:


b2e8e7c8bb1305b07c121172ecb1d2e760339348
[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 the session since the last
157 * check 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(uint64_t session_id, struct lttng_ht *sessions_ht)
164 {
165 int ret;
166 unsigned long current_val;
167 struct relay_session *session;
168
169 assert(sessions_ht);
170
171 session = session_find_by_id(sessions_ht, session_id);
172 if (!session) {
173 DBG("Relay session %" PRIu64 " not found", session_id);
174 ret = -1;
175 goto error;
176 }
177
178 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
179 ret = current_val;
180
181 error:
182 return ret;
183 }
184
185 /*
186 * Send viewer streams to the given socket. The ignore_sent_flag indicates if
187 * this function should ignore the sent flag or not.
188 *
189 * Return 0 on success or else a negative value.
190 */
191 static
192 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
193 struct relay_session *session, unsigned int ignore_sent_flag)
194 {
195 ssize_t ret;
196 struct lttng_viewer_stream send_stream;
197 struct lttng_ht_iter iter;
198 struct relay_viewer_stream *vstream;
199
200 assert(session);
201
202 rcu_read_lock();
203
204 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
205 stream_n.node) {
206 struct ctf_trace *ctf_trace;
207
208 health_code_update();
209
210 /* Ignore if not the same session. */
211 if (vstream->session_id != session->id ||
212 (!ignore_sent_flag && vstream->sent_flag)) {
213 continue;
214 }
215
216 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
217 vstream->path_name);
218 assert(ctf_trace);
219
220 send_stream.id = htobe64(vstream->stream_handle);
221 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
222 send_stream.metadata_flag = htobe32(vstream->metadata_flag);
223 strncpy(send_stream.path_name, vstream->path_name,
224 sizeof(send_stream.path_name));
225 strncpy(send_stream.channel_name, vstream->channel_name,
226 sizeof(send_stream.channel_name));
227
228 DBG("Sending stream %" PRIu64 " to viewer", vstream->stream_handle);
229 ret = send_response(sock, &send_stream, sizeof(send_stream));
230 if (ret < 0) {
231 goto end_unlock;
232 }
233 vstream->sent_flag = 1;
234 }
235
236 ret = 0;
237
238 end_unlock:
239 rcu_read_unlock();
240 return ret;
241 }
242
243 /*
244 * Create every viewer stream possible for the given session with the seek
245 * type. Three counters *can* be return which are in order the total amount of
246 * viewer stream of the session, the number of unsent stream and the number of
247 * stream created. Those counters can be NULL and thus will be ignored.
248 *
249 * Return 0 on success or else a negative value.
250 */
251 static
252 int make_viewer_streams(struct relay_session *session,
253 enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent,
254 uint32_t *nb_created)
255 {
256 int ret;
257 struct lttng_ht_iter iter;
258 struct ctf_trace *ctf_trace;
259
260 assert(session);
261
262 /*
263 * This is to make sure we create viewer streams for a full received
264 * channel. For instance, if we have 8 streams for a channel that are
265 * concurrently being flagged ready, we can end up creating just a subset
266 * of the 8 streams (the ones that are flagged). This lock avoids this
267 * limbo state.
268 */
269 pthread_mutex_lock(&session->viewer_ready_lock);
270
271 /*
272 * Create viewer streams for relay streams that are ready to be used for a
273 * the given session id only.
274 */
275 rcu_read_lock();
276 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
277 node.node) {
278 struct relay_stream *stream;
279
280 health_code_update();
281
282 if (ctf_trace->invalid_flag) {
283 continue;
284 }
285
286 cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) {
287 struct relay_viewer_stream *vstream;
288
289 if (!stream->viewer_ready) {
290 continue;
291 }
292
293 vstream = viewer_stream_find_by_id(stream->stream_handle);
294 if (!vstream) {
295 vstream = viewer_stream_create(stream, seek_t, ctf_trace);
296 if (!vstream) {
297 ret = -1;
298 goto error_unlock;
299 }
300 /* Acquire reference to ctf_trace. */
301 ctf_trace_get_ref(ctf_trace);
302
303 if (nb_created) {
304 /* Update number of created stream counter. */
305 (*nb_created)++;
306 }
307 } else if (!vstream->sent_flag && nb_unsent) {
308 /* Update number of unsent stream counter. */
309 (*nb_unsent)++;
310 }
311 /* Update number of total stream counter. */
312 if (nb_total) {
313 (*nb_total)++;
314 }
315 }
316 }
317
318 ret = 0;
319
320 error_unlock:
321 rcu_read_unlock();
322 pthread_mutex_unlock(&session->viewer_ready_lock);
323 return ret;
324 }
325
326 /*
327 * Write to writable pipe used to notify a thread.
328 */
329 static
330 int notify_thread_pipe(int wpipe)
331 {
332 ssize_t ret;
333
334 ret = lttng_write(wpipe, "!", 1);
335 if (ret < 1) {
336 PERROR("write poll pipe");
337 }
338
339 return (int) ret;
340 }
341
342 /*
343 * Stop all threads by closing the thread quit pipe.
344 */
345 static
346 void stop_threads(void)
347 {
348 int ret;
349
350 /* Stopping all threads */
351 DBG("Terminating all live threads");
352 ret = notify_thread_pipe(live_conn_pipe[1]);
353 if (ret < 0) {
354 ERR("write error on thread quit pipe");
355 }
356
357 /* Dispatch thread */
358 CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
359 futex_nto1_wake(&viewer_conn_queue.futex);
360 }
361
362 /*
363 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
364 */
365 static
366 int create_thread_poll_set(struct lttng_poll_event *events, int size)
367 {
368 int ret;
369
370 if (events == NULL || size == 0) {
371 ret = -1;
372 goto error;
373 }
374
375 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
376 if (ret < 0) {
377 goto error;
378 }
379
380 /* Add quit pipe */
381 ret = lttng_poll_add(events, live_conn_pipe[0], LPOLLIN | LPOLLERR);
382 if (ret < 0) {
383 goto error;
384 }
385
386 return 0;
387
388 error:
389 return ret;
390 }
391
392 /*
393 * Check if the thread quit pipe was triggered.
394 *
395 * Return 1 if it was triggered else 0;
396 */
397 static
398 int check_live_conn_pipe(int fd, uint32_t events)
399 {
400 if (fd == live_conn_pipe[0] && (events & LPOLLIN)) {
401 return 1;
402 }
403
404 return 0;
405 }
406
407 /*
408 * Create and init socket from uri.
409 */
410 static
411 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
412 {
413 int ret;
414 struct lttcomm_sock *sock = NULL;
415
416 sock = lttcomm_alloc_sock_from_uri(uri);
417 if (sock == NULL) {
418 ERR("Allocating socket");
419 goto error;
420 }
421
422 ret = lttcomm_create_sock(sock);
423 if (ret < 0) {
424 goto error;
425 }
426 DBG("Listening on sock %d for live", sock->fd);
427
428 ret = sock->ops->bind(sock);
429 if (ret < 0) {
430 goto error;
431 }
432
433 ret = sock->ops->listen(sock, -1);
434 if (ret < 0) {
435 goto error;
436
437 }
438
439 return sock;
440
441 error:
442 if (sock) {
443 lttcomm_destroy_sock(sock);
444 }
445 return NULL;
446 }
447
448 /*
449 * This thread manages the listening for new connections on the network
450 */
451 static
452 void *thread_listener(void *data)
453 {
454 int i, ret, pollfd, err = -1;
455 uint32_t revents, nb_fd;
456 struct lttng_poll_event events;
457 struct lttcomm_sock *live_control_sock;
458
459 DBG("[thread] Relay live listener started");
460
461 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
462
463 health_code_update();
464
465 live_control_sock = init_socket(live_uri);
466 if (!live_control_sock) {
467 goto error_sock_control;
468 }
469
470 /* Pass 2 as size here for the thread quit pipe and control sockets. */
471 ret = create_thread_poll_set(&events, 2);
472 if (ret < 0) {
473 goto error_create_poll;
474 }
475
476 /* Add the control socket */
477 ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
478 if (ret < 0) {
479 goto error_poll_add;
480 }
481
482 lttng_relay_notify_ready();
483
484 if (testpoint(relayd_thread_live_listener)) {
485 goto error_testpoint;
486 }
487
488 while (1) {
489 health_code_update();
490
491 DBG("Listener accepting live viewers connections");
492
493 restart:
494 health_poll_entry();
495 ret = lttng_poll_wait(&events, -1);
496 health_poll_exit();
497 if (ret < 0) {
498 /*
499 * Restart interrupted system call.
500 */
501 if (errno == EINTR) {
502 goto restart;
503 }
504 goto error;
505 }
506 nb_fd = ret;
507
508 DBG("Relay new viewer connection received");
509 for (i = 0; i < nb_fd; i++) {
510 health_code_update();
511
512 /* Fetch once the poll data */
513 revents = LTTNG_POLL_GETEV(&events, i);
514 pollfd = LTTNG_POLL_GETFD(&events, i);
515
516 /* Thread quit pipe has been closed. Killing thread. */
517 ret = check_live_conn_pipe(pollfd, revents);
518 if (ret) {
519 err = 0;
520 goto exit;
521 }
522
523 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
524 ERR("socket poll error");
525 goto error;
526 } else if (revents & LPOLLIN) {
527 /*
528 * Get allocated in this thread, enqueued to a global queue,
529 * dequeued and freed in the worker thread.
530 */
531 int val = 1;
532 struct relay_connection *new_conn;
533 struct lttcomm_sock *newsock;
534
535 new_conn = connection_create();
536 if (!new_conn) {
537 goto error;
538 }
539
540 newsock = live_control_sock->ops->accept(live_control_sock);
541 if (!newsock) {
542 PERROR("accepting control sock");
543 connection_free(new_conn);
544 goto error;
545 }
546 DBG("Relay viewer connection accepted socket %d", newsock->fd);
547
548 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
549 sizeof(val));
550 if (ret < 0) {
551 PERROR("setsockopt inet");
552 lttcomm_destroy_sock(newsock);
553 connection_free(new_conn);
554 goto error;
555 }
556 new_conn->sock = newsock;
557
558 /* Enqueue request for the dispatcher thread. */
559 cds_wfq_enqueue(&viewer_conn_queue.queue, &new_conn->qnode);
560
561 /*
562 * Wake the dispatch queue futex. Implicit memory barrier with
563 * the exchange in cds_wfq_enqueue.
564 */
565 futex_nto1_wake(&viewer_conn_queue.futex);
566 }
567 }
568 }
569
570 exit:
571 error:
572 error_poll_add:
573 error_testpoint:
574 lttng_poll_clean(&events);
575 error_create_poll:
576 if (live_control_sock->fd >= 0) {
577 ret = live_control_sock->ops->close(live_control_sock);
578 if (ret) {
579 PERROR("close");
580 }
581 }
582 lttcomm_destroy_sock(live_control_sock);
583 error_sock_control:
584 if (err) {
585 health_error();
586 DBG("Live viewer listener thread exited with error");
587 }
588 health_unregister(health_relayd);
589 DBG("Live viewer listener thread cleanup complete");
590 stop_threads();
591 return NULL;
592 }
593
594 /*
595 * This thread manages the dispatching of the requests to worker threads
596 */
597 static
598 void *thread_dispatcher(void *data)
599 {
600 int err = -1;
601 ssize_t ret;
602 struct cds_wfq_node *node;
603 struct relay_connection *conn = NULL;
604
605 DBG("[thread] Live viewer relay dispatcher started");
606
607 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
608
609 if (testpoint(relayd_thread_live_dispatcher)) {
610 goto error_testpoint;
611 }
612
613 health_code_update();
614
615 while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
616 health_code_update();
617
618 /* Atomically prepare the queue futex */
619 futex_nto1_prepare(&viewer_conn_queue.futex);
620
621 do {
622 health_code_update();
623
624 /* Dequeue commands */
625 node = cds_wfq_dequeue_blocking(&viewer_conn_queue.queue);
626 if (node == NULL) {
627 DBG("Woken up but nothing in the live-viewer "
628 "relay command queue");
629 /* Continue thread execution */
630 break;
631 }
632 conn = caa_container_of(node, struct relay_connection, qnode);
633 DBG("Dispatching viewer request waiting on sock %d",
634 conn->sock->fd);
635
636 /*
637 * Inform worker thread of the new request. This call is blocking
638 * so we can be assured that the data will be read at some point in
639 * time or wait to the end of the world :)
640 */
641 ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
642 if (ret < 0) {
643 PERROR("write conn pipe");
644 connection_destroy(conn);
645 goto error;
646 }
647 } while (node != NULL);
648
649 /* Futex wait on queue. Blocking call on futex() */
650 health_poll_entry();
651 futex_nto1_wait(&viewer_conn_queue.futex);
652 health_poll_exit();
653 }
654
655 /* Normal exit, no error */
656 err = 0;
657
658 error:
659 error_testpoint:
660 if (err) {
661 health_error();
662 ERR("Health error occurred in %s", __func__);
663 }
664 health_unregister(health_relayd);
665 DBG("Live viewer dispatch thread dying");
666 stop_threads();
667 return NULL;
668 }
669
670 /*
671 * Establish connection with the viewer and check the versions.
672 *
673 * Return 0 on success or else negative value.
674 */
675 static
676 int viewer_connect(struct relay_connection *conn)
677 {
678 int ret;
679 struct lttng_viewer_connect reply, msg;
680
681 assert(conn);
682
683 conn->version_check_done = 1;
684
685 health_code_update();
686
687 DBG("Viewer is establishing a connection to the relayd.");
688
689 ret = recv_request(conn->sock, &msg, sizeof(msg));
690 if (ret < 0) {
691 goto end;
692 }
693
694 health_code_update();
695
696 reply.major = RELAYD_VERSION_COMM_MAJOR;
697 reply.minor = RELAYD_VERSION_COMM_MINOR;
698
699 /* Major versions must be the same */
700 if (reply.major != be32toh(msg.major)) {
701 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
702 reply.major, be32toh(msg.major));
703 ret = -1;
704 goto end;
705 }
706
707 conn->major = reply.major;
708 /* We adapt to the lowest compatible version */
709 if (reply.minor <= be32toh(msg.minor)) {
710 conn->minor = reply.minor;
711 } else {
712 conn->minor = be32toh(msg.minor);
713 }
714
715 if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
716 conn->type = RELAY_VIEWER_COMMAND;
717 } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
718 conn->type = RELAY_VIEWER_NOTIFICATION;
719 } else {
720 ERR("Unknown connection type : %u", be32toh(msg.type));
721 ret = -1;
722 goto end;
723 }
724
725 reply.major = htobe32(reply.major);
726 reply.minor = htobe32(reply.minor);
727 if (conn->type == RELAY_VIEWER_COMMAND) {
728 reply.viewer_session_id = htobe64(++last_relay_viewer_session_id);
729 }
730
731 health_code_update();
732
733 ret = send_response(conn->sock, &reply, sizeof(reply));
734 if (ret < 0) {
735 goto end;
736 }
737
738 health_code_update();
739
740 DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
741 ret = 0;
742
743 end:
744 return ret;
745 }
746
747 /*
748 * Send the viewer the list of current sessions.
749 *
750 * Return 0 on success or else a negative value.
751 */
752 static
753 int viewer_list_sessions(struct relay_connection *conn)
754 {
755 int ret;
756 struct lttng_viewer_list_sessions session_list;
757 unsigned long count;
758 long approx_before, approx_after;
759 struct lttng_ht_iter iter;
760 struct lttng_viewer_session send_session;
761 struct relay_session *session;
762
763 DBG("List sessions received");
764
765 rcu_read_lock();
766 cds_lfht_count_nodes(conn->sessions_ht->ht, &approx_before, &count,
767 &approx_after);
768 session_list.sessions_count = htobe32(count);
769
770 health_code_update();
771
772 ret = send_response(conn->sock, &session_list, sizeof(session_list));
773 if (ret < 0) {
774 goto end_unlock;
775 }
776
777 health_code_update();
778
779 cds_lfht_for_each_entry(conn->sessions_ht->ht, &iter.iter, session,
780 session_n.node) {
781 health_code_update();
782
783 strncpy(send_session.session_name, session->session_name,
784 sizeof(send_session.session_name));
785 strncpy(send_session.hostname, session->hostname,
786 sizeof(send_session.hostname));
787 send_session.id = htobe64(session->id);
788 send_session.live_timer = htobe32(session->live_timer);
789 send_session.clients = htobe32(session->viewer_refcount);
790 send_session.streams = htobe32(session->stream_count);
791
792 health_code_update();
793
794 ret = send_response(conn->sock, &send_session, sizeof(send_session));
795 if (ret < 0) {
796 goto end_unlock;
797 }
798 }
799 health_code_update();
800
801 rcu_read_unlock();
802 ret = 0;
803 goto end;
804
805 end_unlock:
806 rcu_read_unlock();
807
808 end:
809 return ret;
810 }
811
812 /*
813 * Send the viewer the list of current sessions.
814 */
815 static
816 int viewer_get_new_streams(struct relay_connection *conn)
817 {
818 int ret, send_streams = 0;
819 uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0;
820 struct lttng_viewer_new_streams_request request;
821 struct lttng_viewer_new_streams_response response;
822 struct relay_session *session;
823
824 assert(conn);
825
826 DBG("Get new streams received");
827
828 health_code_update();
829
830 /* Receive the request from the connected client. */
831 ret = recv_request(conn->sock, &request, sizeof(request));
832 if (ret < 0) {
833 goto error;
834 }
835
836 health_code_update();
837
838 rcu_read_lock();
839 session = session_find_by_id(conn->sessions_ht,
840 be64toh(request.session_id));
841 if (!session) {
842 DBG("Relay session %" PRIu64 " not found",
843 be64toh(request.session_id));
844 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
845 goto send_reply;
846 }
847
848 if (conn->session_id == session->id) {
849 /* We confirmed the viewer is asking for the same session. */
850 send_streams = 1;
851 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
852 } else {
853 send_streams = 0;
854 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
855 goto send_reply;
856 }
857
858 if (!send_streams) {
859 goto send_reply;
860 }
861
862 ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, NULL, &nb_unsent,
863 &nb_created);
864 if (ret < 0) {
865 goto end_unlock;
866 }
867 /* Only send back the newly created streams with the unsent ones. */
868 nb_streams = nb_created + nb_unsent;
869 response.streams_count = htobe32(nb_streams);
870
871 send_reply:
872 health_code_update();
873 ret = send_response(conn->sock, &response, sizeof(response));
874 if (ret < 0) {
875 goto end_unlock;
876 }
877 health_code_update();
878
879 /*
880 * Unknown or empty session, just return gracefully, the viewer knows what
881 * is happening.
882 */
883 if (!send_streams || !nb_streams) {
884 ret = 0;
885 goto end_unlock;
886 }
887
888 /*
889 * Send stream and *DON'T* ignore the sent flag so every viewer streams
890 * that were not sent from that point will be sent to the viewer.
891 */
892 ret = send_viewer_streams(conn->sock, session, 0);
893 if (ret < 0) {
894 goto end_unlock;
895 }
896
897 end_unlock:
898 rcu_read_unlock();
899 error:
900 return ret;
901 }
902
903 /*
904 * Send the viewer the list of current sessions.
905 */
906 static
907 int viewer_attach_session(struct relay_connection *conn)
908 {
909 int send_streams = 0;
910 ssize_t ret;
911 uint32_t nb_streams = 0;
912 enum lttng_viewer_seek seek_type;
913 struct lttng_viewer_attach_session_request request;
914 struct lttng_viewer_attach_session_response response;
915 struct relay_session *session;
916
917 assert(conn);
918
919 health_code_update();
920
921 /* Receive the request from the connected client. */
922 ret = recv_request(conn->sock, &request, sizeof(request));
923 if (ret < 0) {
924 goto error;
925 }
926
927 health_code_update();
928
929 if (!conn->viewer_session) {
930 DBG("Client trying to attach before creating a live viewer session");
931 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
932 goto send_reply;
933 }
934
935 rcu_read_lock();
936 session = session_find_by_id(conn->sessions_ht,
937 be64toh(request.session_id));
938 if (!session) {
939 DBG("Relay session %" PRIu64 " not found",
940 be64toh(request.session_id));
941 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
942 goto send_reply;
943 }
944 session_viewer_attach(session);
945 DBG("Attach session ID %" PRIu64 " received", be64toh(request.session_id));
946
947 if (uatomic_read(&session->viewer_refcount) > 1) {
948 DBG("Already a viewer attached");
949 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
950 session_viewer_detach(session);
951 goto send_reply;
952 } else if (session->live_timer == 0) {
953 DBG("Not live session");
954 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
955 goto send_reply;
956 } else {
957 send_streams = 1;
958 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
959 conn->session_id = session->id;
960 conn->session = session;
961 }
962
963 switch (be32toh(request.seek)) {
964 case LTTNG_VIEWER_SEEK_BEGINNING:
965 case LTTNG_VIEWER_SEEK_LAST:
966 seek_type = be32toh(request.seek);
967 break;
968 default:
969 ERR("Wrong seek parameter");
970 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
971 send_streams = 0;
972 goto send_reply;
973 }
974
975 if (!send_streams) {
976 goto send_reply;
977 }
978
979 ret = make_viewer_streams(session, seek_type, &nb_streams, NULL, NULL);
980 if (ret < 0) {
981 goto end_unlock;
982 }
983 response.streams_count = htobe32(nb_streams);
984
985 send_reply:
986 health_code_update();
987 ret = send_response(conn->sock, &response, sizeof(response));
988 if (ret < 0) {
989 goto end_unlock;
990 }
991 health_code_update();
992
993 /*
994 * Unknown or empty session, just return gracefully, the viewer knows what
995 * is happening.
996 */
997 if (!send_streams || !nb_streams) {
998 ret = 0;
999 goto end_unlock;
1000 }
1001
1002 /* Send stream and ignore the sent flag. */
1003 ret = send_viewer_streams(conn->sock, session, 1);
1004 if (ret < 0) {
1005 goto end_unlock;
1006 }
1007
1008 end_unlock:
1009 rcu_read_unlock();
1010 error:
1011 return ret;
1012 }
1013
1014 /*
1015 * Send the next index for a stream.
1016 *
1017 * Return 0 on success or else a negative value.
1018 */
1019 static
1020 int viewer_get_next_index(struct relay_connection *conn)
1021 {
1022 int ret;
1023 struct lttng_viewer_get_next_index request_index;
1024 struct lttng_viewer_index viewer_index;
1025 struct ctf_packet_index packet_index;
1026 struct relay_viewer_stream *vstream;
1027 struct relay_stream *rstream;
1028 struct ctf_trace *ctf_trace;
1029 struct relay_session *session;
1030
1031 assert(conn);
1032
1033 DBG("Viewer get next index");
1034
1035 health_code_update();
1036
1037 ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1038 if (ret < 0) {
1039 goto end;
1040 }
1041 health_code_update();
1042
1043 rcu_read_lock();
1044 session = session_find_by_id(conn->sessions_ht, conn->session_id);
1045 if (!session) {
1046 ret = -1;
1047 goto end_unlock;
1048 }
1049
1050 vstream = viewer_stream_find_by_id(be64toh(request_index.stream_id));
1051 if (!vstream) {
1052 ret = -1;
1053 goto end_unlock;
1054 }
1055
1056 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, vstream->path_name);
1057 assert(ctf_trace);
1058
1059 memset(&viewer_index, 0, sizeof(viewer_index));
1060
1061 /*
1062 * The viewer should not ask for index on metadata stream.
1063 */
1064 if (vstream->metadata_flag) {
1065 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1066 goto send_reply;
1067 }
1068
1069 /* First time, we open the index file */
1070 if (vstream->index_read_fd < 0) {
1071 ret = index_open(vstream->path_name, vstream->channel_name,
1072 vstream->tracefile_count, vstream->tracefile_count_current);
1073 if (ret == -ENOENT) {
1074 /*
1075 * The index is created only when the first data packet arrives, it
1076 * might not be ready at the beginning of the session
1077 */
1078 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1079 goto send_reply;
1080 } else if (ret < 0) {
1081 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1082 goto send_reply;
1083 }
1084 vstream->index_read_fd = ret;
1085 }
1086
1087 rstream = stream_find_by_id(relay_streams_ht, vstream->stream_handle);
1088 assert(rstream);
1089
1090 if (!rstream->close_flag) {
1091 if (vstream->abort_flag) {
1092 /* Rotate on abort (overwrite). */
1093 DBG("Viewer rotate because of overwrite");
1094 ret = viewer_stream_rotate(vstream, rstream);
1095 if (ret < 0) {
1096 goto end_unlock;
1097 } else if (ret == 1) {
1098 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1099 viewer_stream_delete(vstream);
1100 viewer_stream_destroy(ctf_trace, vstream);
1101 goto send_reply;
1102 }
1103 /* ret == 0 means successful so we continue. */
1104 }
1105
1106 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1107 if (rstream->tracefile_count_current == vstream->tracefile_count_current) {
1108 if (rstream->beacon_ts_end != -1ULL &&
1109 vstream->last_sent_index == rstream->total_index_received) {
1110 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1111 viewer_index.timestamp_end = htobe64(rstream->beacon_ts_end);
1112 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1113 goto send_reply;
1114 } else if (rstream->total_index_received <= vstream->last_sent_index
1115 && !vstream->close_write_flag) {
1116 /*
1117 * Reader and writer are working in the same tracefile, so we care
1118 * about the number of index received and sent. Otherwise, we read
1119 * up to EOF.
1120 */
1121 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1122 /* No new index to send, retry later. */
1123 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1124 goto send_reply;
1125 }
1126 }
1127 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1128 } else if (rstream->close_flag && vstream->close_write_flag &&
1129 vstream->total_index_received == vstream->last_sent_index) {
1130 /* Last index sent and current tracefile closed in write */
1131 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1132 viewer_stream_delete(vstream);
1133 viewer_stream_destroy(ctf_trace, vstream);
1134 goto send_reply;
1135 } else {
1136 vstream->close_write_flag = 1;
1137 }
1138
1139 if (!ctf_trace->metadata_received ||
1140 ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1141 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1142 }
1143
1144 ret = check_new_streams(vstream->session_id, conn->sessions_ht);
1145 if (ret < 0) {
1146 goto end_unlock;
1147 } else if (ret == 1) {
1148 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1149 }
1150
1151 pthread_mutex_lock(&vstream->overwrite_lock);
1152 if (vstream->abort_flag) {
1153 /*
1154 * The file is being overwritten by the writer, we cannot * use it.
1155 */
1156 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1157 pthread_mutex_unlock(&vstream->overwrite_lock);
1158 ret = viewer_stream_rotate(vstream, rstream);
1159 if (ret < 0) {
1160 goto end_unlock;
1161 } else if (ret == 1) {
1162 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1163 viewer_stream_delete(vstream);
1164 viewer_stream_destroy(ctf_trace, vstream);
1165 goto send_reply;
1166 }
1167 goto send_reply;
1168 }
1169
1170 ret = lttng_read(vstream->index_read_fd, &packet_index,
1171 sizeof(packet_index));
1172 pthread_mutex_unlock(&vstream->overwrite_lock);
1173 if (ret < sizeof(packet_index)) {
1174 /*
1175 * The tracefile is closed in write, so we read up to EOF.
1176 */
1177 if (vstream->close_write_flag == 1) {
1178 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1179 /* Rotate on normal EOF */
1180 ret = viewer_stream_rotate(vstream, rstream);
1181 if (ret < 0) {
1182 goto end_unlock;
1183 } else if (ret == 1) {
1184 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1185 viewer_stream_delete(vstream);
1186 viewer_stream_destroy(ctf_trace, vstream);
1187 goto send_reply;
1188 }
1189 } else {
1190 PERROR("Relay reading index file %d", vstream->index_read_fd);
1191 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1192 }
1193 goto send_reply;
1194 } else {
1195 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1196 vstream->last_sent_index++;
1197 }
1198
1199 /*
1200 * Indexes are stored in big endian, no need to switch before sending.
1201 */
1202 viewer_index.offset = packet_index.offset;
1203 viewer_index.packet_size = packet_index.packet_size;
1204 viewer_index.content_size = packet_index.content_size;
1205 viewer_index.timestamp_begin = packet_index.timestamp_begin;
1206 viewer_index.timestamp_end = packet_index.timestamp_end;
1207 viewer_index.events_discarded = packet_index.events_discarded;
1208 viewer_index.stream_id = packet_index.stream_id;
1209
1210 send_reply:
1211 viewer_index.flags = htobe32(viewer_index.flags);
1212 health_code_update();
1213
1214 ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1215 if (ret < 0) {
1216 goto end_unlock;
1217 }
1218 health_code_update();
1219
1220 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1221 vstream->last_sent_index, vstream->stream_handle);
1222
1223 end_unlock:
1224 rcu_read_unlock();
1225
1226 end:
1227 return ret;
1228 }
1229
1230 /*
1231 * Send the next index for a stream
1232 *
1233 * Return 0 on success or else a negative value.
1234 */
1235 static
1236 int viewer_get_packet(struct relay_connection *conn)
1237 {
1238 int ret, send_data = 0;
1239 char *data = NULL;
1240 uint32_t len = 0;
1241 ssize_t read_len;
1242 struct lttng_viewer_get_packet get_packet_info;
1243 struct lttng_viewer_trace_packet reply;
1244 struct relay_viewer_stream *stream;
1245 struct ctf_trace *ctf_trace;
1246
1247 assert(conn);
1248
1249 DBG2("Relay get data packet");
1250
1251 health_code_update();
1252
1253 ret = recv_request(conn->sock, &get_packet_info, sizeof(get_packet_info));
1254 if (ret < 0) {
1255 goto end;
1256 }
1257 health_code_update();
1258
1259 /* From this point on, the error label can be reached. */
1260 memset(&reply, 0, sizeof(reply));
1261
1262 rcu_read_lock();
1263 stream = viewer_stream_find_by_id(be64toh(get_packet_info.stream_id));
1264 if (!stream) {
1265 goto error;
1266 }
1267
1268 ctf_trace = ctf_trace_find_by_path(conn->session->ctf_traces_ht,
1269 stream->path_name);
1270 assert(ctf_trace);
1271
1272 /*
1273 * First time we read this stream, we need open the tracefile, we should
1274 * only arrive here if an index has already been sent to the viewer, so the
1275 * tracefile must exist, if it does not it is a fatal error.
1276 */
1277 if (stream->read_fd < 0) {
1278 char fullpath[PATH_MAX];
1279
1280 if (stream->tracefile_count > 0) {
1281 ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, stream->path_name,
1282 stream->channel_name,
1283 stream->tracefile_count_current);
1284 } else {
1285 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1286 stream->channel_name);
1287 }
1288 if (ret < 0) {
1289 goto error;
1290 }
1291 ret = open(fullpath, O_RDONLY);
1292 if (ret < 0) {
1293 PERROR("Relay opening trace file");
1294 goto error;
1295 }
1296 stream->read_fd = ret;
1297 }
1298
1299 if (!ctf_trace->metadata_received ||
1300 ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1301 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1302 reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1303 goto send_reply;
1304 }
1305
1306 ret = check_new_streams(stream->session_id, conn->sessions_ht);
1307 if (ret < 0) {
1308 goto end_unlock;
1309 } else if (ret == 1) {
1310 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1311 reply.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1312 goto send_reply;
1313 }
1314
1315 len = be32toh(get_packet_info.len);
1316 data = zmalloc(len);
1317 if (!data) {
1318 PERROR("relay data zmalloc");
1319 goto error;
1320 }
1321
1322 ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
1323 if (ret < 0) {
1324 /*
1325 * If the read fd was closed by the streaming side, the
1326 * abort_flag will be set to 1, otherwise it is an error.
1327 */
1328 if (stream->abort_flag == 0) {
1329 PERROR("lseek");
1330 goto error;
1331 }
1332 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1333 goto send_reply;
1334 }
1335 read_len = lttng_read(stream->read_fd, data, len);
1336 if (read_len < len) {
1337 /*
1338 * If the read fd was closed by the streaming side, the
1339 * abort_flag will be set to 1, otherwise it is an error.
1340 */
1341 if (stream->abort_flag == 0) {
1342 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1343 stream->read_fd,
1344 be64toh(get_packet_info.offset));
1345 goto error;
1346 } else {
1347 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1348 goto send_reply;
1349 }
1350 }
1351 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1352 reply.len = htobe32(len);
1353 send_data = 1;
1354 goto send_reply;
1355
1356 error:
1357 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1358
1359 send_reply:
1360 reply.flags = htobe32(reply.flags);
1361
1362 health_code_update();
1363
1364 ret = send_response(conn->sock, &reply, sizeof(reply));
1365 if (ret < 0) {
1366 goto end_unlock;
1367 }
1368 health_code_update();
1369
1370 if (send_data) {
1371 health_code_update();
1372 ret = send_response(conn->sock, data, len);
1373 if (ret < 0) {
1374 goto end_unlock;
1375 }
1376 health_code_update();
1377 }
1378
1379 DBG("Sent %u bytes for stream %" PRIu64, len,
1380 be64toh(get_packet_info.stream_id));
1381
1382 end_unlock:
1383 free(data);
1384 rcu_read_unlock();
1385
1386 end:
1387 return ret;
1388 }
1389
1390 /*
1391 * Send the session's metadata
1392 *
1393 * Return 0 on success else a negative value.
1394 */
1395 static
1396 int viewer_get_metadata(struct relay_connection *conn)
1397 {
1398 int ret = 0;
1399 ssize_t read_len;
1400 uint64_t len = 0;
1401 char *data = NULL;
1402 struct lttng_viewer_get_metadata request;
1403 struct lttng_viewer_metadata_packet reply;
1404 struct relay_viewer_stream *stream;
1405 struct ctf_trace *ctf_trace;
1406
1407 assert(conn);
1408
1409 DBG("Relay get metadata");
1410
1411 health_code_update();
1412
1413 ret = recv_request(conn->sock, &request, sizeof(request));
1414 if (ret < 0) {
1415 goto end;
1416 }
1417 health_code_update();
1418
1419 rcu_read_lock();
1420 stream = viewer_stream_find_by_id(be64toh(request.stream_id));
1421 if (!stream || !stream->metadata_flag) {
1422 ERR("Invalid metadata stream");
1423 goto error;
1424 }
1425
1426 ctf_trace = ctf_trace_find_by_path(conn->session->ctf_traces_ht,
1427 stream->path_name);
1428 assert(ctf_trace);
1429 assert(ctf_trace->metadata_sent <= ctf_trace->metadata_received);
1430
1431 len = ctf_trace->metadata_received - ctf_trace->metadata_sent;
1432 if (len == 0) {
1433 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1434 goto send_reply;
1435 }
1436
1437 /* first time, we open the metadata file */
1438 if (stream->read_fd < 0) {
1439 char fullpath[PATH_MAX];
1440
1441 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1442 stream->channel_name);
1443 if (ret < 0) {
1444 goto error;
1445 }
1446 ret = open(fullpath, O_RDONLY);
1447 if (ret < 0) {
1448 PERROR("Relay opening metadata file");
1449 goto error;
1450 }
1451 stream->read_fd = ret;
1452 }
1453
1454 reply.len = htobe64(len);
1455 data = zmalloc(len);
1456 if (!data) {
1457 PERROR("viewer metadata zmalloc");
1458 goto error;
1459 }
1460
1461 read_len = lttng_read(stream->read_fd, data, len);
1462 if (read_len < len) {
1463 PERROR("Relay reading metadata file");
1464 goto error;
1465 }
1466 ctf_trace->metadata_sent += read_len;
1467 reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1468 goto send_reply;
1469
1470 error:
1471 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1472
1473 send_reply:
1474 health_code_update();
1475 ret = send_response(conn->sock, &reply, sizeof(reply));
1476 if (ret < 0) {
1477 goto end_unlock;
1478 }
1479 health_code_update();
1480
1481 if (len > 0) {
1482 ret = send_response(conn->sock, data, len);
1483 if (ret < 0) {
1484 goto end_unlock;
1485 }
1486 }
1487
1488 DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1489 be64toh(request.stream_id));
1490
1491 DBG("Metadata sent");
1492
1493 end_unlock:
1494 free(data);
1495 rcu_read_unlock();
1496 end:
1497 return ret;
1498 }
1499
1500 /*
1501 * Create a viewer session.
1502 *
1503 * Return 0 on success or else a negative value.
1504 */
1505 static
1506 int viewer_create_session(struct relay_connection *conn)
1507 {
1508 int ret;
1509 struct lttng_viewer_create_session_response resp;
1510
1511 DBG("Viewer create session received");
1512
1513 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1514 conn->viewer_session = zmalloc(sizeof(conn->viewer_session));
1515 if (!conn->viewer_session) {
1516 ERR("Allocation viewer session");
1517 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1518 goto send_reply;
1519 }
1520 CDS_INIT_LIST_HEAD(&conn->viewer_session->sessions_head);
1521
1522 send_reply:
1523 health_code_update();
1524 ret = send_response(conn->sock, &resp, sizeof(resp));
1525 if (ret < 0) {
1526 goto end;
1527 }
1528 health_code_update();
1529 ret = 0;
1530
1531 end:
1532 return ret;
1533 }
1534
1535
1536 /*
1537 * live_relay_unknown_command: send -1 if received unknown command
1538 */
1539 static
1540 void live_relay_unknown_command(struct relay_connection *conn)
1541 {
1542 struct lttcomm_relayd_generic_reply reply;
1543
1544 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1545 (void) send_response(conn->sock, &reply, sizeof(reply));
1546 }
1547
1548 /*
1549 * Process the commands received on the control socket
1550 */
1551 static
1552 int process_control(struct lttng_viewer_cmd *recv_hdr,
1553 struct relay_connection *conn)
1554 {
1555 int ret = 0;
1556 uint32_t msg_value;
1557
1558 assert(recv_hdr);
1559 assert(conn);
1560
1561 msg_value = be32toh(recv_hdr->cmd);
1562
1563 /*
1564 * Make sure we've done the version check before any command other then a
1565 * new client connection.
1566 */
1567 if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1568 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1569 ret = -1;
1570 goto end;
1571 }
1572
1573 switch (msg_value) {
1574 case LTTNG_VIEWER_CONNECT:
1575 ret = viewer_connect(conn);
1576 break;
1577 case LTTNG_VIEWER_LIST_SESSIONS:
1578 ret = viewer_list_sessions(conn);
1579 break;
1580 case LTTNG_VIEWER_ATTACH_SESSION:
1581 ret = viewer_attach_session(conn);
1582 break;
1583 case LTTNG_VIEWER_GET_NEXT_INDEX:
1584 ret = viewer_get_next_index(conn);
1585 break;
1586 case LTTNG_VIEWER_GET_PACKET:
1587 ret = viewer_get_packet(conn);
1588 break;
1589 case LTTNG_VIEWER_GET_METADATA:
1590 ret = viewer_get_metadata(conn);
1591 break;
1592 case LTTNG_VIEWER_GET_NEW_STREAMS:
1593 ret = viewer_get_new_streams(conn);
1594 break;
1595 case LTTNG_VIEWER_CREATE_SESSION:
1596 ret = viewer_create_session(conn);
1597 break;
1598 default:
1599 ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd));
1600 live_relay_unknown_command(conn);
1601 ret = -1;
1602 goto end;
1603 }
1604
1605 end:
1606 return ret;
1607 }
1608
1609 static
1610 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1611 {
1612 int ret;
1613
1614 assert(events);
1615
1616 (void) lttng_poll_del(events, pollfd);
1617
1618 ret = close(pollfd);
1619 if (ret < 0) {
1620 ERR("Closing pollfd %d", pollfd);
1621 }
1622 }
1623
1624 /*
1625 * Delete all streams for a specific session ID.
1626 */
1627 static void destroy_viewer_streams_by_session(struct relay_session *session)
1628 {
1629 struct relay_viewer_stream *stream;
1630 struct lttng_ht_iter iter;
1631
1632 assert(session);
1633
1634 rcu_read_lock();
1635 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, stream,
1636 stream_n.node) {
1637 struct ctf_trace *ctf_trace;
1638
1639 health_code_update();
1640 if (stream->session_id != session->id) {
1641 continue;
1642 }
1643
1644 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1645 stream->path_name);
1646 assert(ctf_trace);
1647
1648 viewer_stream_delete(stream);
1649
1650 if (stream->metadata_flag) {
1651 ctf_trace->metadata_sent = 0;
1652 ctf_trace->viewer_metadata_stream = NULL;
1653 }
1654
1655 viewer_stream_destroy(ctf_trace, stream);
1656 }
1657 rcu_read_unlock();
1658 }
1659
1660 static void try_destroy_streams(struct relay_session *session)
1661 {
1662 struct ctf_trace *ctf_trace;
1663 struct lttng_ht_iter iter;
1664
1665 assert(session);
1666
1667 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
1668 node.node) {
1669 /* Attempt to destroy the ctf trace of that session. */
1670 ctf_trace_try_destroy(session, ctf_trace);
1671 }
1672 }
1673
1674 /*
1675 * Delete and destroy a connection.
1676 *
1677 * RCU read side lock MUST be acquired.
1678 */
1679 static void destroy_connection(struct lttng_ht *relay_connections_ht,
1680 struct relay_connection *conn)
1681 {
1682 struct relay_session *session;
1683
1684 assert(relay_connections_ht);
1685 assert(conn);
1686
1687 DBG("Cleaning connection of session ID %" PRIu64, conn->session_id);
1688
1689 connection_delete(relay_connections_ht, conn);
1690
1691 rcu_read_lock();
1692 session = session_find_by_id(conn->sessions_ht, conn->session_id);
1693 if (session) {
1694 /*
1695 * Very important that this is done before destroying the session so we
1696 * can put back every viewer stream reference from the ctf_trace.
1697 */
1698 destroy_viewer_streams_by_session(session);
1699 try_destroy_streams(session);
1700 session_viewer_try_destroy(conn->sessions_ht, session);
1701 }
1702 rcu_read_unlock();
1703
1704 connection_destroy(conn);
1705 }
1706
1707 /*
1708 * This thread does the actual work
1709 */
1710 static
1711 void *thread_worker(void *data)
1712 {
1713 int ret, err = -1;
1714 uint32_t nb_fd;
1715 struct relay_connection *conn;
1716 struct lttng_poll_event events;
1717 struct lttng_ht *relay_connections_ht;
1718 struct lttng_ht_iter iter;
1719 struct lttng_viewer_cmd recv_hdr;
1720 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
1721 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
1722
1723 DBG("[thread] Live viewer relay worker started");
1724
1725 rcu_register_thread();
1726
1727 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1728
1729 if (testpoint(relayd_thread_live_worker)) {
1730 goto error_testpoint;
1731 }
1732
1733 /* table of connections indexed on socket */
1734 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1735 if (!relay_connections_ht) {
1736 goto relay_connections_ht_error;
1737 }
1738
1739 ret = create_thread_poll_set(&events, 2);
1740 if (ret < 0) {
1741 goto error_poll_create;
1742 }
1743
1744 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1745 if (ret < 0) {
1746 goto error;
1747 }
1748
1749 restart:
1750 while (1) {
1751 int i;
1752
1753 health_code_update();
1754
1755 /* Infinite blocking call, waiting for transmission */
1756 DBG3("Relayd live viewer worker thread polling...");
1757 health_poll_entry();
1758 ret = lttng_poll_wait(&events, -1);
1759 health_poll_exit();
1760 if (ret < 0) {
1761 /*
1762 * Restart interrupted system call.
1763 */
1764 if (errno == EINTR) {
1765 goto restart;
1766 }
1767 goto error;
1768 }
1769
1770 nb_fd = ret;
1771
1772 /*
1773 * Process control. The control connection is prioritised so we don't
1774 * starve it with high throughput tracing data on the data
1775 * connection.
1776 */
1777 for (i = 0; i < nb_fd; i++) {
1778 /* Fetch once the poll data */
1779 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1780 int pollfd = LTTNG_POLL_GETFD(&events, i);
1781
1782 health_code_update();
1783
1784 /* Thread quit pipe has been closed. Killing thread. */
1785 ret = check_live_conn_pipe(pollfd, revents);
1786 if (ret) {
1787 err = 0;
1788 goto exit;
1789 }
1790
1791 /* Inspect the relay conn pipe for new connection */
1792 if (pollfd == live_conn_pipe[0]) {
1793 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1794 ERR("Relay live pipe error");
1795 goto error;
1796 } else if (revents & LPOLLIN) {
1797 ret = lttng_read(live_conn_pipe[0], &conn, sizeof(conn));
1798 if (ret < 0) {
1799 goto error;
1800 }
1801 conn->sessions_ht = sessions_ht;
1802 connection_init(conn);
1803 lttng_poll_add(&events, conn->sock->fd,
1804 LPOLLIN | LPOLLRDHUP);
1805 rcu_read_lock();
1806 lttng_ht_add_unique_ulong(relay_connections_ht,
1807 &conn->sock_n);
1808 rcu_read_unlock();
1809 DBG("Connection socket %d added", conn->sock->fd);
1810 }
1811 } else {
1812 rcu_read_lock();
1813 conn = connection_find_by_sock(relay_connections_ht, pollfd);
1814 /* If not found, there is a synchronization issue. */
1815 assert(conn);
1816
1817 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1818 cleanup_connection_pollfd(&events, pollfd);
1819 destroy_connection(relay_connections_ht, conn);
1820 } else if (revents & LPOLLIN) {
1821 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
1822 sizeof(recv_hdr), 0);
1823 if (ret <= 0) {
1824 /* Connection closed */
1825 cleanup_connection_pollfd(&events, pollfd);
1826 destroy_connection(relay_connections_ht, conn);
1827 DBG("Viewer control conn closed with %d", pollfd);
1828 } else {
1829 ret = process_control(&recv_hdr, conn);
1830 if (ret < 0) {
1831 /* Clear the session on error. */
1832 cleanup_connection_pollfd(&events, pollfd);
1833 destroy_connection(relay_connections_ht, conn);
1834 DBG("Viewer connection closed with %d", pollfd);
1835 }
1836 }
1837 }
1838 rcu_read_unlock();
1839 }
1840 }
1841 }
1842
1843 exit:
1844 error:
1845 lttng_poll_clean(&events);
1846
1847 /* Cleanup reamaining connection object. */
1848 rcu_read_lock();
1849 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn,
1850 sock_n.node) {
1851 health_code_update();
1852 destroy_connection(relay_connections_ht, conn);
1853 }
1854 rcu_read_unlock();
1855 error_poll_create:
1856 lttng_ht_destroy(relay_connections_ht);
1857 relay_connections_ht_error:
1858 /* Close relay conn pipes */
1859 utils_close_pipe(live_conn_pipe);
1860 if (err) {
1861 DBG("Viewer worker thread exited with error");
1862 }
1863 DBG("Viewer worker thread cleanup complete");
1864 error_testpoint:
1865 if (err) {
1866 health_error();
1867 ERR("Health error occurred in %s", __func__);
1868 }
1869 health_unregister(health_relayd);
1870 stop_threads();
1871 rcu_unregister_thread();
1872 return NULL;
1873 }
1874
1875 /*
1876 * Create the relay command pipe to wake thread_manage_apps.
1877 * Closed in cleanup().
1878 */
1879 static int create_conn_pipe(void)
1880 {
1881 int ret;
1882
1883 ret = utils_create_pipe_cloexec(live_conn_pipe);
1884
1885 return ret;
1886 }
1887
1888 void live_stop_threads(void)
1889 {
1890 int ret;
1891 void *status;
1892
1893 stop_threads();
1894
1895 ret = pthread_join(live_listener_thread, &status);
1896 if (ret != 0) {
1897 PERROR("pthread_join live listener");
1898 goto error; /* join error, exit without cleanup */
1899 }
1900
1901 ret = pthread_join(live_worker_thread, &status);
1902 if (ret != 0) {
1903 PERROR("pthread_join live worker");
1904 goto error; /* join error, exit without cleanup */
1905 }
1906
1907 ret = pthread_join(live_dispatcher_thread, &status);
1908 if (ret != 0) {
1909 PERROR("pthread_join live dispatcher");
1910 goto error; /* join error, exit without cleanup */
1911 }
1912
1913 cleanup();
1914
1915 error:
1916 return;
1917 }
1918
1919 /*
1920 * main
1921 */
1922 int live_start_threads(struct lttng_uri *uri,
1923 struct relay_local_data *relay_ctx)
1924 {
1925 int ret = 0;
1926 void *status;
1927 int is_root;
1928
1929 assert(uri);
1930 live_uri = uri;
1931
1932 /* Check if daemon is UID = 0 */
1933 is_root = !getuid();
1934
1935 if (!is_root) {
1936 if (live_uri->port < 1024) {
1937 ERR("Need to be root to use ports < 1024");
1938 ret = -1;
1939 goto exit;
1940 }
1941 }
1942
1943 /* Setup the thread apps communication pipe. */
1944 if ((ret = create_conn_pipe()) < 0) {
1945 goto exit;
1946 }
1947
1948 /* Init relay command queue. */
1949 cds_wfq_init(&viewer_conn_queue.queue);
1950
1951 /* Set up max poll set size */
1952 lttng_poll_set_max_size();
1953
1954 /* Setup the dispatcher thread */
1955 ret = pthread_create(&live_dispatcher_thread, NULL,
1956 thread_dispatcher, (void *) NULL);
1957 if (ret != 0) {
1958 PERROR("pthread_create viewer dispatcher");
1959 goto exit_dispatcher;
1960 }
1961
1962 /* Setup the worker thread */
1963 ret = pthread_create(&live_worker_thread, NULL,
1964 thread_worker, relay_ctx);
1965 if (ret != 0) {
1966 PERROR("pthread_create viewer worker");
1967 goto exit_worker;
1968 }
1969
1970 /* Setup the listener thread */
1971 ret = pthread_create(&live_listener_thread, NULL,
1972 thread_listener, (void *) NULL);
1973 if (ret != 0) {
1974 PERROR("pthread_create viewer listener");
1975 goto exit_listener;
1976 }
1977
1978 ret = 0;
1979 goto end;
1980
1981 exit_listener:
1982 ret = pthread_join(live_listener_thread, &status);
1983 if (ret != 0) {
1984 PERROR("pthread_join live listener");
1985 goto error; /* join error, exit without cleanup */
1986 }
1987
1988 exit_worker:
1989 ret = pthread_join(live_worker_thread, &status);
1990 if (ret != 0) {
1991 PERROR("pthread_join live worker");
1992 goto error; /* join error, exit without cleanup */
1993 }
1994
1995 exit_dispatcher:
1996 ret = pthread_join(live_dispatcher_thread, &status);
1997 if (ret != 0) {
1998 PERROR("pthread_join live dispatcher");
1999 goto error; /* join error, exit without cleanup */
2000 }
2001
2002 exit:
2003 cleanup();
2004
2005 end:
2006 error:
2007 return ret;
2008 }
LTTng 2.0 tools and control repository
This page took 0.112382 seconds and 3 git commands to generate.