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


91d870493c9052eef6129c0a0396de9c8cf68ad8
[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 /*
872 * If the session is closed and we have no new streams to send,
873 * it means that the viewer has already received the whole trace
874 * for this session and should now close it.
875 */
876 if (nb_streams == 0 && session->close_flag) {
877 send_streams = 0;
878 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
879 goto send_reply;
880 }
881
882 send_reply:
883 health_code_update();
884 ret = send_response(conn->sock, &response, sizeof(response));
885 if (ret < 0) {
886 goto end_unlock;
887 }
888 health_code_update();
889
890 /*
891 * Unknown or empty session, just return gracefully, the viewer knows what
892 * is happening.
893 */
894 if (!send_streams || !nb_streams) {
895 ret = 0;
896 goto end_unlock;
897 }
898
899 /*
900 * Send stream and *DON'T* ignore the sent flag so every viewer streams
901 * that were not sent from that point will be sent to the viewer.
902 */
903 ret = send_viewer_streams(conn->sock, session, 0);
904 if (ret < 0) {
905 goto end_unlock;
906 }
907
908 end_unlock:
909 rcu_read_unlock();
910 error:
911 return ret;
912 }
913
914 /*
915 * Send the viewer the list of current sessions.
916 */
917 static
918 int viewer_attach_session(struct relay_connection *conn)
919 {
920 int send_streams = 0;
921 ssize_t ret;
922 uint32_t nb_streams = 0;
923 enum lttng_viewer_seek seek_type;
924 struct lttng_viewer_attach_session_request request;
925 struct lttng_viewer_attach_session_response response;
926 struct relay_session *session;
927
928 assert(conn);
929
930 health_code_update();
931
932 /* Receive the request from the connected client. */
933 ret = recv_request(conn->sock, &request, sizeof(request));
934 if (ret < 0) {
935 goto error;
936 }
937
938 health_code_update();
939
940 if (!conn->viewer_session) {
941 DBG("Client trying to attach before creating a live viewer session");
942 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
943 goto send_reply;
944 }
945
946 rcu_read_lock();
947 session = session_find_by_id(conn->sessions_ht,
948 be64toh(request.session_id));
949 if (!session) {
950 DBG("Relay session %" PRIu64 " not found",
951 be64toh(request.session_id));
952 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
953 goto send_reply;
954 }
955 session_viewer_attach(session);
956 DBG("Attach session ID %" PRIu64 " received", be64toh(request.session_id));
957
958 if (uatomic_read(&session->viewer_refcount) > 1) {
959 DBG("Already a viewer attached");
960 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
961 session_viewer_detach(session);
962 goto send_reply;
963 } else if (session->live_timer == 0) {
964 DBG("Not live session");
965 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
966 goto send_reply;
967 } else {
968 send_streams = 1;
969 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
970 conn->session_id = session->id;
971 conn->session = session;
972 }
973
974 switch (be32toh(request.seek)) {
975 case LTTNG_VIEWER_SEEK_BEGINNING:
976 case LTTNG_VIEWER_SEEK_LAST:
977 seek_type = be32toh(request.seek);
978 break;
979 default:
980 ERR("Wrong seek parameter");
981 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
982 send_streams = 0;
983 goto send_reply;
984 }
985
986 if (!send_streams) {
987 goto send_reply;
988 }
989
990 ret = make_viewer_streams(session, seek_type, &nb_streams, NULL, NULL);
991 if (ret < 0) {
992 goto end_unlock;
993 }
994 response.streams_count = htobe32(nb_streams);
995
996 send_reply:
997 health_code_update();
998 ret = send_response(conn->sock, &response, sizeof(response));
999 if (ret < 0) {
1000 goto end_unlock;
1001 }
1002 health_code_update();
1003
1004 /*
1005 * Unknown or empty session, just return gracefully, the viewer knows what
1006 * is happening.
1007 */
1008 if (!send_streams || !nb_streams) {
1009 ret = 0;
1010 goto end_unlock;
1011 }
1012
1013 /* Send stream and ignore the sent flag. */
1014 ret = send_viewer_streams(conn->sock, session, 1);
1015 if (ret < 0) {
1016 goto end_unlock;
1017 }
1018
1019 end_unlock:
1020 rcu_read_unlock();
1021 error:
1022 return ret;
1023 }
1024
1025 /*
1026 * Send the next index for a stream.
1027 *
1028 * Return 0 on success or else a negative value.
1029 */
1030 static
1031 int viewer_get_next_index(struct relay_connection *conn)
1032 {
1033 int ret;
1034 struct lttng_viewer_get_next_index request_index;
1035 struct lttng_viewer_index viewer_index;
1036 struct ctf_packet_index packet_index;
1037 struct relay_viewer_stream *vstream;
1038 struct relay_stream *rstream;
1039 struct ctf_trace *ctf_trace;
1040 struct relay_session *session;
1041
1042 assert(conn);
1043
1044 DBG("Viewer get next index");
1045
1046 health_code_update();
1047
1048 ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1049 if (ret < 0) {
1050 goto end;
1051 }
1052 health_code_update();
1053
1054 rcu_read_lock();
1055 session = session_find_by_id(conn->sessions_ht, conn->session_id);
1056 if (!session) {
1057 ret = -1;
1058 goto end_unlock;
1059 }
1060
1061 vstream = viewer_stream_find_by_id(be64toh(request_index.stream_id));
1062 if (!vstream) {
1063 ret = -1;
1064 goto end_unlock;
1065 }
1066
1067 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, vstream->path_name);
1068 assert(ctf_trace);
1069
1070 memset(&viewer_index, 0, sizeof(viewer_index));
1071
1072 /*
1073 * The viewer should not ask for index on metadata stream.
1074 */
1075 if (vstream->metadata_flag) {
1076 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1077 goto send_reply;
1078 }
1079
1080 /* First time, we open the index file */
1081 if (vstream->index_read_fd < 0) {
1082 ret = index_open(vstream->path_name, vstream->channel_name,
1083 vstream->tracefile_count, vstream->tracefile_count_current);
1084 if (ret == -ENOENT) {
1085 /*
1086 * The index is created only when the first data packet arrives, it
1087 * might not be ready at the beginning of the session
1088 */
1089 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1090 goto send_reply;
1091 } else if (ret < 0) {
1092 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1093 goto send_reply;
1094 }
1095 vstream->index_read_fd = ret;
1096 }
1097
1098 rstream = stream_find_by_id(relay_streams_ht, vstream->stream_handle);
1099 assert(rstream);
1100
1101 if (!rstream->close_flag) {
1102 if (vstream->abort_flag) {
1103 /* Rotate on abort (overwrite). */
1104 DBG("Viewer rotate because of overwrite");
1105 ret = viewer_stream_rotate(vstream, rstream);
1106 if (ret < 0) {
1107 goto end_unlock;
1108 } else if (ret == 1) {
1109 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1110 viewer_stream_delete(vstream);
1111 viewer_stream_destroy(ctf_trace, vstream);
1112 goto send_reply;
1113 }
1114 /* ret == 0 means successful so we continue. */
1115 }
1116
1117 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1118 if (rstream->tracefile_count_current == vstream->tracefile_count_current) {
1119 if (rstream->beacon_ts_end != -1ULL &&
1120 vstream->last_sent_index == rstream->total_index_received) {
1121 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1122 viewer_index.timestamp_end = htobe64(rstream->beacon_ts_end);
1123 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1124 goto send_reply;
1125 } else if (rstream->total_index_received <= vstream->last_sent_index
1126 && !vstream->close_write_flag) {
1127 /*
1128 * Reader and writer are working in the same tracefile, so we care
1129 * about the number of index received and sent. Otherwise, we read
1130 * up to EOF.
1131 */
1132 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1133 /* No new index to send, retry later. */
1134 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1135 goto send_reply;
1136 }
1137 }
1138 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1139 } else if (rstream->close_flag && vstream->close_write_flag &&
1140 vstream->total_index_received == vstream->last_sent_index) {
1141 /* Last index sent and current tracefile closed in write */
1142 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1143 viewer_stream_delete(vstream);
1144 viewer_stream_destroy(ctf_trace, vstream);
1145 goto send_reply;
1146 } else {
1147 vstream->close_write_flag = 1;
1148 }
1149
1150 if (!ctf_trace->metadata_received ||
1151 ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1152 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1153 }
1154
1155 ret = check_new_streams(vstream->session_id, conn->sessions_ht);
1156 if (ret < 0) {
1157 goto end_unlock;
1158 } else if (ret == 1) {
1159 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1160 }
1161
1162 pthread_mutex_lock(&vstream->overwrite_lock);
1163 if (vstream->abort_flag) {
1164 /*
1165 * The file is being overwritten by the writer, we cannot * use it.
1166 */
1167 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1168 pthread_mutex_unlock(&vstream->overwrite_lock);
1169 ret = viewer_stream_rotate(vstream, rstream);
1170 if (ret < 0) {
1171 goto end_unlock;
1172 } else if (ret == 1) {
1173 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1174 viewer_stream_delete(vstream);
1175 viewer_stream_destroy(ctf_trace, vstream);
1176 goto send_reply;
1177 }
1178 goto send_reply;
1179 }
1180
1181 ret = lttng_read(vstream->index_read_fd, &packet_index,
1182 sizeof(packet_index));
1183 pthread_mutex_unlock(&vstream->overwrite_lock);
1184 if (ret < sizeof(packet_index)) {
1185 /*
1186 * The tracefile is closed in write, so we read up to EOF.
1187 */
1188 if (vstream->close_write_flag == 1) {
1189 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1190 /* Rotate on normal EOF */
1191 ret = viewer_stream_rotate(vstream, rstream);
1192 if (ret < 0) {
1193 goto end_unlock;
1194 } else if (ret == 1) {
1195 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1196 viewer_stream_delete(vstream);
1197 viewer_stream_destroy(ctf_trace, vstream);
1198 goto send_reply;
1199 }
1200 } else {
1201 PERROR("Relay reading index file %d", vstream->index_read_fd);
1202 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1203 }
1204 goto send_reply;
1205 } else {
1206 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1207 vstream->last_sent_index++;
1208 }
1209
1210 /*
1211 * Indexes are stored in big endian, no need to switch before sending.
1212 */
1213 viewer_index.offset = packet_index.offset;
1214 viewer_index.packet_size = packet_index.packet_size;
1215 viewer_index.content_size = packet_index.content_size;
1216 viewer_index.timestamp_begin = packet_index.timestamp_begin;
1217 viewer_index.timestamp_end = packet_index.timestamp_end;
1218 viewer_index.events_discarded = packet_index.events_discarded;
1219 viewer_index.stream_id = packet_index.stream_id;
1220
1221 send_reply:
1222 viewer_index.flags = htobe32(viewer_index.flags);
1223 health_code_update();
1224
1225 ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1226 if (ret < 0) {
1227 goto end_unlock;
1228 }
1229 health_code_update();
1230
1231 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1232 vstream->last_sent_index, vstream->stream_handle);
1233
1234 end_unlock:
1235 rcu_read_unlock();
1236
1237 end:
1238 return ret;
1239 }
1240
1241 /*
1242 * Send the next index for a stream
1243 *
1244 * Return 0 on success or else a negative value.
1245 */
1246 static
1247 int viewer_get_packet(struct relay_connection *conn)
1248 {
1249 int ret, send_data = 0;
1250 char *data = NULL;
1251 uint32_t len = 0;
1252 ssize_t read_len;
1253 struct lttng_viewer_get_packet get_packet_info;
1254 struct lttng_viewer_trace_packet reply;
1255 struct relay_viewer_stream *stream;
1256 struct ctf_trace *ctf_trace;
1257
1258 assert(conn);
1259
1260 DBG2("Relay get data packet");
1261
1262 health_code_update();
1263
1264 ret = recv_request(conn->sock, &get_packet_info, sizeof(get_packet_info));
1265 if (ret < 0) {
1266 goto end;
1267 }
1268 health_code_update();
1269
1270 /* From this point on, the error label can be reached. */
1271 memset(&reply, 0, sizeof(reply));
1272
1273 rcu_read_lock();
1274 stream = viewer_stream_find_by_id(be64toh(get_packet_info.stream_id));
1275 if (!stream) {
1276 goto error;
1277 }
1278
1279 ctf_trace = ctf_trace_find_by_path(conn->session->ctf_traces_ht,
1280 stream->path_name);
1281 assert(ctf_trace);
1282
1283 /*
1284 * First time we read this stream, we need open the tracefile, we should
1285 * only arrive here if an index has already been sent to the viewer, so the
1286 * tracefile must exist, if it does not it is a fatal error.
1287 */
1288 if (stream->read_fd < 0) {
1289 char fullpath[PATH_MAX];
1290
1291 if (stream->tracefile_count > 0) {
1292 ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, stream->path_name,
1293 stream->channel_name,
1294 stream->tracefile_count_current);
1295 } else {
1296 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1297 stream->channel_name);
1298 }
1299 if (ret < 0) {
1300 goto error;
1301 }
1302 ret = open(fullpath, O_RDONLY);
1303 if (ret < 0) {
1304 PERROR("Relay opening trace file");
1305 goto error;
1306 }
1307 stream->read_fd = ret;
1308 }
1309
1310 if (!ctf_trace->metadata_received ||
1311 ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1312 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1313 reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1314 goto send_reply;
1315 }
1316
1317 ret = check_new_streams(stream->session_id, conn->sessions_ht);
1318 if (ret < 0) {
1319 goto end_unlock;
1320 } else if (ret == 1) {
1321 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1322 reply.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1323 goto send_reply;
1324 }
1325
1326 len = be32toh(get_packet_info.len);
1327 data = zmalloc(len);
1328 if (!data) {
1329 PERROR("relay data zmalloc");
1330 goto error;
1331 }
1332
1333 ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
1334 if (ret < 0) {
1335 /*
1336 * If the read fd was closed by the streaming side, the
1337 * abort_flag will be set to 1, otherwise it is an error.
1338 */
1339 if (stream->abort_flag == 0) {
1340 PERROR("lseek");
1341 goto error;
1342 }
1343 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1344 goto send_reply;
1345 }
1346 read_len = lttng_read(stream->read_fd, data, len);
1347 if (read_len < len) {
1348 /*
1349 * If the read fd was closed by the streaming side, the
1350 * abort_flag will be set to 1, otherwise it is an error.
1351 */
1352 if (stream->abort_flag == 0) {
1353 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1354 stream->read_fd,
1355 be64toh(get_packet_info.offset));
1356 goto error;
1357 } else {
1358 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1359 goto send_reply;
1360 }
1361 }
1362 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1363 reply.len = htobe32(len);
1364 send_data = 1;
1365 goto send_reply;
1366
1367 error:
1368 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1369
1370 send_reply:
1371 reply.flags = htobe32(reply.flags);
1372
1373 health_code_update();
1374
1375 ret = send_response(conn->sock, &reply, sizeof(reply));
1376 if (ret < 0) {
1377 goto end_unlock;
1378 }
1379 health_code_update();
1380
1381 if (send_data) {
1382 health_code_update();
1383 ret = send_response(conn->sock, data, len);
1384 if (ret < 0) {
1385 goto end_unlock;
1386 }
1387 health_code_update();
1388 }
1389
1390 DBG("Sent %u bytes for stream %" PRIu64, len,
1391 be64toh(get_packet_info.stream_id));
1392
1393 end_unlock:
1394 free(data);
1395 rcu_read_unlock();
1396
1397 end:
1398 return ret;
1399 }
1400
1401 /*
1402 * Send the session's metadata
1403 *
1404 * Return 0 on success else a negative value.
1405 */
1406 static
1407 int viewer_get_metadata(struct relay_connection *conn)
1408 {
1409 int ret = 0;
1410 ssize_t read_len;
1411 uint64_t len = 0;
1412 char *data = NULL;
1413 struct lttng_viewer_get_metadata request;
1414 struct lttng_viewer_metadata_packet reply;
1415 struct relay_viewer_stream *stream;
1416 struct ctf_trace *ctf_trace;
1417
1418 assert(conn);
1419
1420 DBG("Relay get metadata");
1421
1422 health_code_update();
1423
1424 ret = recv_request(conn->sock, &request, sizeof(request));
1425 if (ret < 0) {
1426 goto end;
1427 }
1428 health_code_update();
1429
1430 rcu_read_lock();
1431 stream = viewer_stream_find_by_id(be64toh(request.stream_id));
1432 if (!stream || !stream->metadata_flag) {
1433 ERR("Invalid metadata stream");
1434 goto error;
1435 }
1436
1437 ctf_trace = ctf_trace_find_by_path(conn->session->ctf_traces_ht,
1438 stream->path_name);
1439 assert(ctf_trace);
1440 assert(ctf_trace->metadata_sent <= ctf_trace->metadata_received);
1441
1442 len = ctf_trace->metadata_received - ctf_trace->metadata_sent;
1443 if (len == 0) {
1444 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1445 goto send_reply;
1446 }
1447
1448 /* first time, we open the metadata file */
1449 if (stream->read_fd < 0) {
1450 char fullpath[PATH_MAX];
1451
1452 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1453 stream->channel_name);
1454 if (ret < 0) {
1455 goto error;
1456 }
1457 ret = open(fullpath, O_RDONLY);
1458 if (ret < 0) {
1459 PERROR("Relay opening metadata file");
1460 goto error;
1461 }
1462 stream->read_fd = ret;
1463 }
1464
1465 reply.len = htobe64(len);
1466 data = zmalloc(len);
1467 if (!data) {
1468 PERROR("viewer metadata zmalloc");
1469 goto error;
1470 }
1471
1472 read_len = lttng_read(stream->read_fd, data, len);
1473 if (read_len < len) {
1474 PERROR("Relay reading metadata file");
1475 goto error;
1476 }
1477 ctf_trace->metadata_sent += read_len;
1478 reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1479 goto send_reply;
1480
1481 error:
1482 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1483
1484 send_reply:
1485 health_code_update();
1486 ret = send_response(conn->sock, &reply, sizeof(reply));
1487 if (ret < 0) {
1488 goto end_unlock;
1489 }
1490 health_code_update();
1491
1492 if (len > 0) {
1493 ret = send_response(conn->sock, data, len);
1494 if (ret < 0) {
1495 goto end_unlock;
1496 }
1497 }
1498
1499 DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1500 be64toh(request.stream_id));
1501
1502 DBG("Metadata sent");
1503
1504 end_unlock:
1505 free(data);
1506 rcu_read_unlock();
1507 end:
1508 return ret;
1509 }
1510
1511 /*
1512 * Create a viewer session.
1513 *
1514 * Return 0 on success or else a negative value.
1515 */
1516 static
1517 int viewer_create_session(struct relay_connection *conn)
1518 {
1519 int ret;
1520 struct lttng_viewer_create_session_response resp;
1521
1522 DBG("Viewer create session received");
1523
1524 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1525 conn->viewer_session = zmalloc(sizeof(conn->viewer_session));
1526 if (!conn->viewer_session) {
1527 ERR("Allocation viewer session");
1528 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1529 goto send_reply;
1530 }
1531 CDS_INIT_LIST_HEAD(&conn->viewer_session->sessions_head);
1532
1533 send_reply:
1534 health_code_update();
1535 ret = send_response(conn->sock, &resp, sizeof(resp));
1536 if (ret < 0) {
1537 goto end;
1538 }
1539 health_code_update();
1540 ret = 0;
1541
1542 end:
1543 return ret;
1544 }
1545
1546
1547 /*
1548 * live_relay_unknown_command: send -1 if received unknown command
1549 */
1550 static
1551 void live_relay_unknown_command(struct relay_connection *conn)
1552 {
1553 struct lttcomm_relayd_generic_reply reply;
1554
1555 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1556 (void) send_response(conn->sock, &reply, sizeof(reply));
1557 }
1558
1559 /*
1560 * Process the commands received on the control socket
1561 */
1562 static
1563 int process_control(struct lttng_viewer_cmd *recv_hdr,
1564 struct relay_connection *conn)
1565 {
1566 int ret = 0;
1567 uint32_t msg_value;
1568
1569 assert(recv_hdr);
1570 assert(conn);
1571
1572 msg_value = be32toh(recv_hdr->cmd);
1573
1574 /*
1575 * Make sure we've done the version check before any command other then a
1576 * new client connection.
1577 */
1578 if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1579 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1580 ret = -1;
1581 goto end;
1582 }
1583
1584 switch (msg_value) {
1585 case LTTNG_VIEWER_CONNECT:
1586 ret = viewer_connect(conn);
1587 break;
1588 case LTTNG_VIEWER_LIST_SESSIONS:
1589 ret = viewer_list_sessions(conn);
1590 break;
1591 case LTTNG_VIEWER_ATTACH_SESSION:
1592 ret = viewer_attach_session(conn);
1593 break;
1594 case LTTNG_VIEWER_GET_NEXT_INDEX:
1595 ret = viewer_get_next_index(conn);
1596 break;
1597 case LTTNG_VIEWER_GET_PACKET:
1598 ret = viewer_get_packet(conn);
1599 break;
1600 case LTTNG_VIEWER_GET_METADATA:
1601 ret = viewer_get_metadata(conn);
1602 break;
1603 case LTTNG_VIEWER_GET_NEW_STREAMS:
1604 ret = viewer_get_new_streams(conn);
1605 break;
1606 case LTTNG_VIEWER_CREATE_SESSION:
1607 ret = viewer_create_session(conn);
1608 break;
1609 default:
1610 ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd));
1611 live_relay_unknown_command(conn);
1612 ret = -1;
1613 goto end;
1614 }
1615
1616 end:
1617 return ret;
1618 }
1619
1620 static
1621 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1622 {
1623 int ret;
1624
1625 assert(events);
1626
1627 (void) lttng_poll_del(events, pollfd);
1628
1629 ret = close(pollfd);
1630 if (ret < 0) {
1631 ERR("Closing pollfd %d", pollfd);
1632 }
1633 }
1634
1635 /*
1636 * Delete all streams for a specific session ID.
1637 */
1638 static void destroy_viewer_streams_by_session(struct relay_session *session)
1639 {
1640 struct relay_viewer_stream *stream;
1641 struct lttng_ht_iter iter;
1642
1643 assert(session);
1644
1645 rcu_read_lock();
1646 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, stream,
1647 stream_n.node) {
1648 struct ctf_trace *ctf_trace;
1649
1650 health_code_update();
1651 if (stream->session_id != session->id) {
1652 continue;
1653 }
1654
1655 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1656 stream->path_name);
1657 assert(ctf_trace);
1658
1659 viewer_stream_delete(stream);
1660
1661 if (stream->metadata_flag) {
1662 ctf_trace->metadata_sent = 0;
1663 ctf_trace->viewer_metadata_stream = NULL;
1664 }
1665
1666 viewer_stream_destroy(ctf_trace, stream);
1667 }
1668 rcu_read_unlock();
1669 }
1670
1671 static void try_destroy_streams(struct relay_session *session)
1672 {
1673 struct ctf_trace *ctf_trace;
1674 struct lttng_ht_iter iter;
1675
1676 assert(session);
1677
1678 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
1679 node.node) {
1680 /* Attempt to destroy the ctf trace of that session. */
1681 ctf_trace_try_destroy(session, ctf_trace);
1682 }
1683 }
1684
1685 /*
1686 * Delete and destroy a connection.
1687 *
1688 * RCU read side lock MUST be acquired.
1689 */
1690 static void destroy_connection(struct lttng_ht *relay_connections_ht,
1691 struct relay_connection *conn)
1692 {
1693 struct relay_session *session;
1694
1695 assert(relay_connections_ht);
1696 assert(conn);
1697
1698 DBG("Cleaning connection of session ID %" PRIu64, conn->session_id);
1699
1700 connection_delete(relay_connections_ht, conn);
1701
1702 rcu_read_lock();
1703 session = session_find_by_id(conn->sessions_ht, conn->session_id);
1704 if (session) {
1705 /*
1706 * Very important that this is done before destroying the session so we
1707 * can put back every viewer stream reference from the ctf_trace.
1708 */
1709 destroy_viewer_streams_by_session(session);
1710 try_destroy_streams(session);
1711 session_viewer_try_destroy(conn->sessions_ht, session);
1712 }
1713 rcu_read_unlock();
1714
1715 connection_destroy(conn);
1716 }
1717
1718 /*
1719 * This thread does the actual work
1720 */
1721 static
1722 void *thread_worker(void *data)
1723 {
1724 int ret, err = -1;
1725 uint32_t nb_fd;
1726 struct relay_connection *conn;
1727 struct lttng_poll_event events;
1728 struct lttng_ht *relay_connections_ht;
1729 struct lttng_ht_iter iter;
1730 struct lttng_viewer_cmd recv_hdr;
1731 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
1732 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
1733
1734 DBG("[thread] Live viewer relay worker started");
1735
1736 rcu_register_thread();
1737
1738 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1739
1740 if (testpoint(relayd_thread_live_worker)) {
1741 goto error_testpoint;
1742 }
1743
1744 /* table of connections indexed on socket */
1745 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1746 if (!relay_connections_ht) {
1747 goto relay_connections_ht_error;
1748 }
1749
1750 ret = create_thread_poll_set(&events, 2);
1751 if (ret < 0) {
1752 goto error_poll_create;
1753 }
1754
1755 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1756 if (ret < 0) {
1757 goto error;
1758 }
1759
1760 restart:
1761 while (1) {
1762 int i;
1763
1764 health_code_update();
1765
1766 /* Infinite blocking call, waiting for transmission */
1767 DBG3("Relayd live viewer worker thread polling...");
1768 health_poll_entry();
1769 ret = lttng_poll_wait(&events, -1);
1770 health_poll_exit();
1771 if (ret < 0) {
1772 /*
1773 * Restart interrupted system call.
1774 */
1775 if (errno == EINTR) {
1776 goto restart;
1777 }
1778 goto error;
1779 }
1780
1781 nb_fd = ret;
1782
1783 /*
1784 * Process control. The control connection is prioritised so we don't
1785 * starve it with high throughput tracing data on the data
1786 * connection.
1787 */
1788 for (i = 0; i < nb_fd; i++) {
1789 /* Fetch once the poll data */
1790 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1791 int pollfd = LTTNG_POLL_GETFD(&events, i);
1792
1793 health_code_update();
1794
1795 /* Thread quit pipe has been closed. Killing thread. */
1796 ret = check_live_conn_pipe(pollfd, revents);
1797 if (ret) {
1798 err = 0;
1799 goto exit;
1800 }
1801
1802 /* Inspect the relay conn pipe for new connection */
1803 if (pollfd == live_conn_pipe[0]) {
1804 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1805 ERR("Relay live pipe error");
1806 goto error;
1807 } else if (revents & LPOLLIN) {
1808 ret = lttng_read(live_conn_pipe[0], &conn, sizeof(conn));
1809 if (ret < 0) {
1810 goto error;
1811 }
1812 conn->sessions_ht = sessions_ht;
1813 connection_init(conn);
1814 lttng_poll_add(&events, conn->sock->fd,
1815 LPOLLIN | LPOLLRDHUP);
1816 rcu_read_lock();
1817 lttng_ht_add_unique_ulong(relay_connections_ht,
1818 &conn->sock_n);
1819 rcu_read_unlock();
1820 DBG("Connection socket %d added", conn->sock->fd);
1821 }
1822 } else {
1823 rcu_read_lock();
1824 conn = connection_find_by_sock(relay_connections_ht, pollfd);
1825 /* If not found, there is a synchronization issue. */
1826 assert(conn);
1827
1828 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1829 cleanup_connection_pollfd(&events, pollfd);
1830 destroy_connection(relay_connections_ht, conn);
1831 } else if (revents & LPOLLIN) {
1832 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
1833 sizeof(recv_hdr), 0);
1834 if (ret <= 0) {
1835 /* Connection closed */
1836 cleanup_connection_pollfd(&events, pollfd);
1837 destroy_connection(relay_connections_ht, conn);
1838 DBG("Viewer control conn closed with %d", pollfd);
1839 } else {
1840 ret = process_control(&recv_hdr, conn);
1841 if (ret < 0) {
1842 /* Clear the session on error. */
1843 cleanup_connection_pollfd(&events, pollfd);
1844 destroy_connection(relay_connections_ht, conn);
1845 DBG("Viewer connection closed with %d", pollfd);
1846 }
1847 }
1848 }
1849 rcu_read_unlock();
1850 }
1851 }
1852 }
1853
1854 exit:
1855 error:
1856 lttng_poll_clean(&events);
1857
1858 /* Cleanup reamaining connection object. */
1859 rcu_read_lock();
1860 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn,
1861 sock_n.node) {
1862 health_code_update();
1863 destroy_connection(relay_connections_ht, conn);
1864 }
1865 rcu_read_unlock();
1866 error_poll_create:
1867 lttng_ht_destroy(relay_connections_ht);
1868 relay_connections_ht_error:
1869 /* Close relay conn pipes */
1870 utils_close_pipe(live_conn_pipe);
1871 if (err) {
1872 DBG("Viewer worker thread exited with error");
1873 }
1874 DBG("Viewer worker thread cleanup complete");
1875 error_testpoint:
1876 if (err) {
1877 health_error();
1878 ERR("Health error occurred in %s", __func__);
1879 }
1880 health_unregister(health_relayd);
1881 stop_threads();
1882 rcu_unregister_thread();
1883 return NULL;
1884 }
1885
1886 /*
1887 * Create the relay command pipe to wake thread_manage_apps.
1888 * Closed in cleanup().
1889 */
1890 static int create_conn_pipe(void)
1891 {
1892 int ret;
1893
1894 ret = utils_create_pipe_cloexec(live_conn_pipe);
1895
1896 return ret;
1897 }
1898
1899 void live_stop_threads(void)
1900 {
1901 int ret;
1902 void *status;
1903
1904 stop_threads();
1905
1906 ret = pthread_join(live_listener_thread, &status);
1907 if (ret != 0) {
1908 PERROR("pthread_join live listener");
1909 goto error; /* join error, exit without cleanup */
1910 }
1911
1912 ret = pthread_join(live_worker_thread, &status);
1913 if (ret != 0) {
1914 PERROR("pthread_join live worker");
1915 goto error; /* join error, exit without cleanup */
1916 }
1917
1918 ret = pthread_join(live_dispatcher_thread, &status);
1919 if (ret != 0) {
1920 PERROR("pthread_join live dispatcher");
1921 goto error; /* join error, exit without cleanup */
1922 }
1923
1924 cleanup();
1925
1926 error:
1927 return;
1928 }
1929
1930 /*
1931 * main
1932 */
1933 int live_start_threads(struct lttng_uri *uri,
1934 struct relay_local_data *relay_ctx)
1935 {
1936 int ret = 0;
1937 void *status;
1938 int is_root;
1939
1940 assert(uri);
1941 live_uri = uri;
1942
1943 /* Check if daemon is UID = 0 */
1944 is_root = !getuid();
1945
1946 if (!is_root) {
1947 if (live_uri->port < 1024) {
1948 ERR("Need to be root to use ports < 1024");
1949 ret = -1;
1950 goto exit;
1951 }
1952 }
1953
1954 /* Setup the thread apps communication pipe. */
1955 if ((ret = create_conn_pipe()) < 0) {
1956 goto exit;
1957 }
1958
1959 /* Init relay command queue. */
1960 cds_wfq_init(&viewer_conn_queue.queue);
1961
1962 /* Set up max poll set size */
1963 lttng_poll_set_max_size();
1964
1965 /* Setup the dispatcher thread */
1966 ret = pthread_create(&live_dispatcher_thread, NULL,
1967 thread_dispatcher, (void *) NULL);
1968 if (ret != 0) {
1969 PERROR("pthread_create viewer dispatcher");
1970 goto exit_dispatcher;
1971 }
1972
1973 /* Setup the worker thread */
1974 ret = pthread_create(&live_worker_thread, NULL,
1975 thread_worker, relay_ctx);
1976 if (ret != 0) {
1977 PERROR("pthread_create viewer worker");
1978 goto exit_worker;
1979 }
1980
1981 /* Setup the listener thread */
1982 ret = pthread_create(&live_listener_thread, NULL,
1983 thread_listener, (void *) NULL);
1984 if (ret != 0) {
1985 PERROR("pthread_create viewer listener");
1986 goto exit_listener;
1987 }
1988
1989 ret = 0;
1990 goto end;
1991
1992 exit_listener:
1993 ret = pthread_join(live_listener_thread, &status);
1994 if (ret != 0) {
1995 PERROR("pthread_join live listener");
1996 goto error; /* join error, exit without cleanup */
1997 }
1998
1999 exit_worker:
2000 ret = pthread_join(live_worker_thread, &status);
2001 if (ret != 0) {
2002 PERROR("pthread_join live worker");
2003 goto error; /* join error, exit without cleanup */
2004 }
2005
2006 exit_dispatcher:
2007 ret = pthread_join(live_dispatcher_thread, &status);
2008 if (ret != 0) {
2009 PERROR("pthread_join live dispatcher");
2010 goto error; /* join error, exit without cleanup */
2011 }
2012
2013 exit:
2014 cleanup();
2015
2016 end:
2017 error:
2018 return ret;
2019 }
LTTng 2.0 tools and control repository
This page took 0.080531 seconds and 3 git commands to generate.