Backported to glibc 2.8
[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/compat/endian.h>
47 #include <common/defaults.h>
48 #include <common/futex.h>
49 #include <common/index/index.h>
50 #include <common/sessiond-comm/sessiond-comm.h>
51 #include <common/sessiond-comm/inet.h>
52 #include <common/sessiond-comm/relayd.h>
53 #include <common/uri.h>
54 #include <common/utils.h>
55
56 #include "cmd.h"
57 #include "live.h"
58 #include "lttng-relayd.h"
59 #include "utils.h"
60 #include "health-relayd.h"
61 #include "testpoint.h"
62 #include "viewer-stream.h"
63 #include "stream.h"
64 #include "session.h"
65 #include "ctf-trace.h"
66 #include "connection.h"
67
68 static struct lttng_uri *live_uri;
69
70 /*
71 * This pipe is used to inform the worker thread that a command is queued and
72 * ready to be processed.
73 */
74 static int live_conn_pipe[2] = { -1, -1 };
75
76 /* Shared between threads */
77 static int live_dispatch_thread_exit;
78
79 static pthread_t live_listener_thread;
80 static pthread_t live_dispatcher_thread;
81 static pthread_t live_worker_thread;
82
83 /*
84 * Relay command queue.
85 *
86 * The live_thread_listener and live_thread_dispatcher communicate with this
87 * queue.
88 */
89 static struct relay_conn_queue viewer_conn_queue;
90
91 static uint64_t last_relay_viewer_session_id;
92
93 /*
94 * Cleanup the daemon
95 */
96 static
97 void cleanup(void)
98 {
99 DBG("Cleaning up");
100
101 free(live_uri);
102 }
103
104 /*
105 * Receive a request buffer using a given socket, destination allocated buffer
106 * of length size.
107 *
108 * Return the size of the received message or else a negative value on error
109 * with errno being set by recvmsg() syscall.
110 */
111 static
112 ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
113 {
114 ssize_t ret;
115
116 assert(sock);
117 assert(buf);
118
119 ret = sock->ops->recvmsg(sock, buf, size, 0);
120 if (ret < 0 || ret != size) {
121 if (ret == 0) {
122 /* Orderly shutdown. Not necessary to print an error. */
123 DBG("Socket %d did an orderly shutdown", sock->fd);
124 } else {
125 ERR("Relay failed to receive request.");
126 }
127 ret = -1;
128 }
129
130 return ret;
131 }
132
133 /*
134 * Send a response buffer using a given socket, source allocated buffer of
135 * length size.
136 *
137 * Return the size of the sent message or else a negative value on error with
138 * errno being set by sendmsg() syscall.
139 */
140 static
141 ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
142 {
143 ssize_t ret;
144
145 assert(sock);
146 assert(buf);
147
148 ret = sock->ops->sendmsg(sock, buf, size, 0);
149 if (ret < 0) {
150 ERR("Relayd failed to send response.");
151 }
152
153 return ret;
154 }
155
156 /*
157 * Atomically check if new streams got added in one of the sessions attached
158 * and reset the flag to 0.
159 *
160 * Returns 1 if new streams got added, 0 if nothing changed, a negative value
161 * on error.
162 */
163 static
164 int check_new_streams(struct relay_connection *conn)
165 {
166 struct relay_session *session;
167 unsigned long current_val;
168 int ret = 0;
169
170 if (!conn->viewer_session) {
171 goto end;
172 }
173 cds_list_for_each_entry(session,
174 &conn->viewer_session->sessions_head,
175 viewer_session_list) {
176 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
177 ret = current_val;
178 if (ret == 1) {
179 goto end;
180 }
181 }
182
183 end:
184 return ret;
185 }
186
187 /*
188 * Send viewer streams to the given socket. The ignore_sent_flag indicates if
189 * this function should ignore the sent flag or not.
190 *
191 * Return 0 on success or else a negative value.
192 */
193 static
194 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
195 struct relay_session *session, unsigned int ignore_sent_flag)
196 {
197 ssize_t ret;
198 struct lttng_viewer_stream send_stream;
199 struct lttng_ht_iter iter;
200 struct relay_viewer_stream *vstream;
201
202 assert(session);
203
204 rcu_read_lock();
205
206 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
207 stream_n.node) {
208 struct ctf_trace *ctf_trace;
209
210 health_code_update();
211
212 /* Ignore if not the same session. */
213 if (vstream->session_id != session->id ||
214 (!ignore_sent_flag && vstream->sent_flag)) {
215 continue;
216 }
217
218 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
219 vstream->path_name);
220 assert(ctf_trace);
221
222 send_stream.id = htobe64(vstream->stream_handle);
223 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
224 send_stream.metadata_flag = htobe32(vstream->metadata_flag);
225 strncpy(send_stream.path_name, vstream->path_name,
226 sizeof(send_stream.path_name));
227 strncpy(send_stream.channel_name, vstream->channel_name,
228 sizeof(send_stream.channel_name));
229
230 DBG("Sending stream %" PRIu64 " to viewer", vstream->stream_handle);
231 ret = send_response(sock, &send_stream, sizeof(send_stream));
232 if (ret < 0) {
233 goto end_unlock;
234 }
235 vstream->sent_flag = 1;
236 }
237
238 ret = 0;
239
240 end_unlock:
241 rcu_read_unlock();
242 return ret;
243 }
244
245 /*
246 * Create every viewer stream possible for the given session with the seek
247 * type. Three counters *can* be return which are in order the total amount of
248 * viewer stream of the session, the number of unsent stream and the number of
249 * stream created. Those counters can be NULL and thus will be ignored.
250 *
251 * Return 0 on success or else a negative value.
252 */
253 static
254 int make_viewer_streams(struct relay_session *session,
255 enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent,
256 uint32_t *nb_created)
257 {
258 int ret;
259 struct lttng_ht_iter iter;
260 struct ctf_trace *ctf_trace;
261
262 assert(session);
263
264 /*
265 * This is to make sure we create viewer streams for a full received
266 * channel. For instance, if we have 8 streams for a channel that are
267 * concurrently being flagged ready, we can end up creating just a subset
268 * of the 8 streams (the ones that are flagged). This lock avoids this
269 * limbo state.
270 */
271 pthread_mutex_lock(&session->viewer_ready_lock);
272
273 /*
274 * Create viewer streams for relay streams that are ready to be used for a
275 * the given session id only.
276 */
277 rcu_read_lock();
278 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
279 node.node) {
280 struct relay_stream *stream;
281
282 health_code_update();
283
284 if (ctf_trace->invalid_flag) {
285 continue;
286 }
287
288 cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) {
289 struct relay_viewer_stream *vstream;
290
291 if (!stream->viewer_ready) {
292 continue;
293 }
294
295 vstream = viewer_stream_find_by_id(stream->stream_handle);
296 if (!vstream) {
297 vstream = viewer_stream_create(stream, seek_t, ctf_trace);
298 if (!vstream) {
299 ret = -1;
300 goto error_unlock;
301 }
302 /* Acquire reference to ctf_trace. */
303 ctf_trace_get_ref(ctf_trace);
304
305 if (nb_created) {
306 /* Update number of created stream counter. */
307 (*nb_created)++;
308 }
309 } else if (!vstream->sent_flag && nb_unsent) {
310 /* Update number of unsent stream counter. */
311 (*nb_unsent)++;
312 }
313 /* Update number of total stream counter. */
314 if (nb_total) {
315 (*nb_total)++;
316 }
317 }
318 }
319
320 ret = 0;
321
322 error_unlock:
323 rcu_read_unlock();
324 pthread_mutex_unlock(&session->viewer_ready_lock);
325 return ret;
326 }
327
328 /*
329 * Write to writable pipe used to notify a thread.
330 */
331 static
332 int notify_thread_pipe(int wpipe)
333 {
334 ssize_t ret;
335
336 ret = lttng_write(wpipe, "!", 1);
337 if (ret < 1) {
338 PERROR("write poll pipe");
339 }
340
341 return (int) ret;
342 }
343
344 /*
345 * Stop all threads by closing the thread quit pipe.
346 */
347 static
348 void stop_threads(void)
349 {
350 int ret;
351
352 /* Stopping all threads */
353 DBG("Terminating all live threads");
354 ret = notify_thread_pipe(thread_quit_pipe[1]);
355 if (ret < 0) {
356 ERR("write error on thread quit pipe");
357 }
358
359 /* Dispatch thread */
360 CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
361 futex_nto1_wake(&viewer_conn_queue.futex);
362 }
363
364 /*
365 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
366 */
367 static
368 int create_thread_poll_set(struct lttng_poll_event *events, int size)
369 {
370 int ret;
371
372 if (events == NULL || size == 0) {
373 ret = -1;
374 goto error;
375 }
376
377 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
378 if (ret < 0) {
379 goto error;
380 }
381
382 /* Add quit pipe */
383 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
384 if (ret < 0) {
385 goto error;
386 }
387
388 return 0;
389
390 error:
391 return ret;
392 }
393
394 /*
395 * Check if the thread quit pipe was triggered.
396 *
397 * Return 1 if it was triggered else 0;
398 */
399 static
400 int check_thread_quit_pipe(int fd, uint32_t events)
401 {
402 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
403 return 1;
404 }
405
406 return 0;
407 }
408
409 /*
410 * Create and init socket from uri.
411 */
412 static
413 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
414 {
415 int ret;
416 struct lttcomm_sock *sock = NULL;
417
418 sock = lttcomm_alloc_sock_from_uri(uri);
419 if (sock == NULL) {
420 ERR("Allocating socket");
421 goto error;
422 }
423
424 ret = lttcomm_create_sock(sock);
425 if (ret < 0) {
426 goto error;
427 }
428 DBG("Listening on sock %d for live", sock->fd);
429
430 ret = sock->ops->bind(sock);
431 if (ret < 0) {
432 goto error;
433 }
434
435 ret = sock->ops->listen(sock, -1);
436 if (ret < 0) {
437 goto error;
438
439 }
440
441 return sock;
442
443 error:
444 if (sock) {
445 lttcomm_destroy_sock(sock);
446 }
447 return NULL;
448 }
449
450 /*
451 * This thread manages the listening for new connections on the network
452 */
453 static
454 void *thread_listener(void *data)
455 {
456 int i, ret, pollfd, err = -1;
457 uint32_t revents, nb_fd;
458 struct lttng_poll_event events;
459 struct lttcomm_sock *live_control_sock;
460
461 DBG("[thread] Relay live listener started");
462
463 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
464
465 health_code_update();
466
467 live_control_sock = init_socket(live_uri);
468 if (!live_control_sock) {
469 goto error_sock_control;
470 }
471
472 /* Pass 2 as size here for the thread quit pipe and control sockets. */
473 ret = create_thread_poll_set(&events, 2);
474 if (ret < 0) {
475 goto error_create_poll;
476 }
477
478 /* Add the control socket */
479 ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
480 if (ret < 0) {
481 goto error_poll_add;
482 }
483
484 lttng_relay_notify_ready();
485
486 if (testpoint(relayd_thread_live_listener)) {
487 goto error_testpoint;
488 }
489
490 while (1) {
491 health_code_update();
492
493 DBG("Listener accepting live viewers connections");
494
495 restart:
496 health_poll_entry();
497 ret = lttng_poll_wait(&events, -1);
498 health_poll_exit();
499 if (ret < 0) {
500 /*
501 * Restart interrupted system call.
502 */
503 if (errno == EINTR) {
504 goto restart;
505 }
506 goto error;
507 }
508 nb_fd = ret;
509
510 DBG("Relay new viewer connection received");
511 for (i = 0; i < nb_fd; i++) {
512 health_code_update();
513
514 /* Fetch once the poll data */
515 revents = LTTNG_POLL_GETEV(&events, i);
516 pollfd = LTTNG_POLL_GETFD(&events, i);
517
518 /* Thread quit pipe has been closed. Killing thread. */
519 ret = check_thread_quit_pipe(pollfd, revents);
520 if (ret) {
521 err = 0;
522 goto exit;
523 }
524
525 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
526 ERR("socket poll error");
527 goto error;
528 } else if (revents & LPOLLIN) {
529 /*
530 * Get allocated in this thread, enqueued to a global queue,
531 * dequeued and freed in the worker thread.
532 */
533 int val = 1;
534 struct relay_connection *new_conn;
535 struct lttcomm_sock *newsock;
536
537 new_conn = connection_create();
538 if (!new_conn) {
539 goto error;
540 }
541
542 newsock = live_control_sock->ops->accept(live_control_sock);
543 if (!newsock) {
544 PERROR("accepting control sock");
545 connection_free(new_conn);
546 goto error;
547 }
548 DBG("Relay viewer connection accepted socket %d", newsock->fd);
549
550 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
551 sizeof(val));
552 if (ret < 0) {
553 PERROR("setsockopt inet");
554 lttcomm_destroy_sock(newsock);
555 connection_free(new_conn);
556 goto error;
557 }
558 new_conn->sock = newsock;
559
560 /* Enqueue request for the dispatcher thread. */
561 cds_wfq_enqueue(&viewer_conn_queue.queue, &new_conn->qnode);
562
563 /*
564 * Wake the dispatch queue futex. Implicit memory barrier with
565 * the exchange in cds_wfq_enqueue.
566 */
567 futex_nto1_wake(&viewer_conn_queue.futex);
568 }
569 }
570 }
571
572 exit:
573 error:
574 error_poll_add:
575 error_testpoint:
576 lttng_poll_clean(&events);
577 error_create_poll:
578 if (live_control_sock->fd >= 0) {
579 ret = live_control_sock->ops->close(live_control_sock);
580 if (ret) {
581 PERROR("close");
582 }
583 }
584 lttcomm_destroy_sock(live_control_sock);
585 error_sock_control:
586 if (err) {
587 health_error();
588 DBG("Live viewer listener thread exited with error");
589 }
590 health_unregister(health_relayd);
591 DBG("Live viewer listener thread cleanup complete");
592 stop_threads();
593 return NULL;
594 }
595
596 /*
597 * This thread manages the dispatching of the requests to worker threads
598 */
599 static
600 void *thread_dispatcher(void *data)
601 {
602 int err = -1;
603 ssize_t ret;
604 struct cds_wfq_node *node;
605 struct relay_connection *conn = NULL;
606
607 DBG("[thread] Live viewer relay dispatcher started");
608
609 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
610
611 if (testpoint(relayd_thread_live_dispatcher)) {
612 goto error_testpoint;
613 }
614
615 health_code_update();
616
617 while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
618 health_code_update();
619
620 /* Atomically prepare the queue futex */
621 futex_nto1_prepare(&viewer_conn_queue.futex);
622
623 do {
624 health_code_update();
625
626 /* Dequeue commands */
627 node = cds_wfq_dequeue_blocking(&viewer_conn_queue.queue);
628 if (node == NULL) {
629 DBG("Woken up but nothing in the live-viewer "
630 "relay command queue");
631 /* Continue thread execution */
632 break;
633 }
634 conn = caa_container_of(node, struct relay_connection, qnode);
635 DBG("Dispatching viewer request waiting on sock %d",
636 conn->sock->fd);
637
638 /*
639 * Inform worker thread of the new request. This call is blocking
640 * so we can be assured that the data will be read at some point in
641 * time or wait to the end of the world :)
642 */
643 ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
644 if (ret < 0) {
645 PERROR("write conn pipe");
646 connection_destroy(conn);
647 goto error;
648 }
649 } while (node != NULL);
650
651 /* Futex wait on queue. Blocking call on futex() */
652 health_poll_entry();
653 futex_nto1_wait(&viewer_conn_queue.futex);
654 health_poll_exit();
655 }
656
657 /* Normal exit, no error */
658 err = 0;
659
660 error:
661 error_testpoint:
662 if (err) {
663 health_error();
664 ERR("Health error occurred in %s", __func__);
665 }
666 health_unregister(health_relayd);
667 DBG("Live viewer dispatch thread dying");
668 stop_threads();
669 return NULL;
670 }
671
672 /*
673 * Establish connection with the viewer and check the versions.
674 *
675 * Return 0 on success or else negative value.
676 */
677 static
678 int viewer_connect(struct relay_connection *conn)
679 {
680 int ret;
681 struct lttng_viewer_connect reply, msg;
682
683 assert(conn);
684
685 conn->version_check_done = 1;
686
687 health_code_update();
688
689 DBG("Viewer is establishing a connection to the relayd.");
690
691 ret = recv_request(conn->sock, &msg, sizeof(msg));
692 if (ret < 0) {
693 goto end;
694 }
695
696 health_code_update();
697
698 memset(&reply, 0, sizeof(reply));
699 reply.major = RELAYD_VERSION_COMM_MAJOR;
700 reply.minor = RELAYD_VERSION_COMM_MINOR;
701
702 /* Major versions must be the same */
703 if (reply.major != be32toh(msg.major)) {
704 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
705 reply.major, be32toh(msg.major));
706 ret = -1;
707 goto end;
708 }
709
710 conn->major = reply.major;
711 /* We adapt to the lowest compatible version */
712 if (reply.minor <= be32toh(msg.minor)) {
713 conn->minor = reply.minor;
714 } else {
715 conn->minor = be32toh(msg.minor);
716 }
717
718 if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
719 conn->type = RELAY_VIEWER_COMMAND;
720 } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
721 conn->type = RELAY_VIEWER_NOTIFICATION;
722 } else {
723 ERR("Unknown connection type : %u", be32toh(msg.type));
724 ret = -1;
725 goto end;
726 }
727
728 reply.major = htobe32(reply.major);
729 reply.minor = htobe32(reply.minor);
730 if (conn->type == RELAY_VIEWER_COMMAND) {
731 reply.viewer_session_id = htobe64(++last_relay_viewer_session_id);
732 }
733
734 health_code_update();
735
736 ret = send_response(conn->sock, &reply, sizeof(reply));
737 if (ret < 0) {
738 goto end;
739 }
740
741 health_code_update();
742
743 DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
744 ret = 0;
745
746 end:
747 return ret;
748 }
749
750 /*
751 * Send the viewer the list of current sessions.
752 *
753 * Return 0 on success or else a negative value.
754 */
755 static
756 int viewer_list_sessions(struct relay_connection *conn)
757 {
758 int ret;
759 struct lttng_viewer_list_sessions session_list;
760 unsigned long count;
761 long approx_before, approx_after;
762 struct lttng_ht_iter iter;
763 struct lttng_viewer_session send_session;
764 struct relay_session *session;
765
766 DBG("List sessions received");
767
768 rcu_read_lock();
769 cds_lfht_count_nodes(conn->sessions_ht->ht, &approx_before, &count,
770 &approx_after);
771 session_list.sessions_count = htobe32(count);
772
773 health_code_update();
774
775 ret = send_response(conn->sock, &session_list, sizeof(session_list));
776 if (ret < 0) {
777 goto end_unlock;
778 }
779
780 health_code_update();
781
782 cds_lfht_for_each_entry(conn->sessions_ht->ht, &iter.iter, session,
783 session_n.node) {
784 health_code_update();
785
786 strncpy(send_session.session_name, session->session_name,
787 sizeof(send_session.session_name));
788 strncpy(send_session.hostname, session->hostname,
789 sizeof(send_session.hostname));
790 send_session.id = htobe64(session->id);
791 send_session.live_timer = htobe32(session->live_timer);
792 send_session.clients = htobe32(session->viewer_refcount);
793 send_session.streams = htobe32(session->stream_count);
794
795 health_code_update();
796
797 ret = send_response(conn->sock, &send_session, sizeof(send_session));
798 if (ret < 0) {
799 goto end_unlock;
800 }
801 }
802 health_code_update();
803
804 rcu_read_unlock();
805 ret = 0;
806 goto end;
807
808 end_unlock:
809 rcu_read_unlock();
810
811 end:
812 return ret;
813 }
814
815 /*
816 * Check if a connection is attached to a session.
817 * Return 1 if attached, 0 if not attached, a negative value on error.
818 */
819 static
820 int session_attached(struct relay_connection *conn, uint64_t session_id)
821 {
822 struct relay_session *session;
823 int found = 0;
824
825 if (!conn->viewer_session) {
826 goto end;
827 }
828 cds_list_for_each_entry(session,
829 &conn->viewer_session->sessions_head,
830 viewer_session_list) {
831 if (session->id == session_id) {
832 found = 1;
833 goto end;
834 }
835 }
836
837 end:
838 return found;
839 }
840
841 /*
842 * Delete all streams for a specific session ID.
843 */
844 static void destroy_viewer_streams_by_session(struct relay_session *session)
845 {
846 struct relay_viewer_stream *stream;
847 struct lttng_ht_iter iter;
848
849 assert(session);
850
851 rcu_read_lock();
852 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, stream,
853 stream_n.node) {
854 struct ctf_trace *ctf_trace;
855
856 health_code_update();
857 if (stream->session_id != session->id) {
858 continue;
859 }
860
861 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
862 stream->path_name);
863 assert(ctf_trace);
864
865 viewer_stream_delete(stream);
866
867 if (stream->metadata_flag) {
868 ctf_trace->metadata_sent = 0;
869 ctf_trace->viewer_metadata_stream = NULL;
870 }
871
872 viewer_stream_destroy(ctf_trace, stream);
873 }
874 rcu_read_unlock();
875 }
876
877 static void try_destroy_streams(struct relay_session *session)
878 {
879 struct ctf_trace *ctf_trace;
880 struct lttng_ht_iter iter;
881
882 assert(session);
883
884 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
885 node.node) {
886 /* Attempt to destroy the ctf trace of that session. */
887 ctf_trace_try_destroy(session, ctf_trace);
888 }
889 }
890
891 /*
892 * Cleanup a session.
893 */
894 static void cleanup_session(struct relay_connection *conn,
895 struct relay_session *session)
896 {
897 /*
898 * Very important that this is done before destroying the session so we
899 * can put back every viewer stream reference from the ctf_trace.
900 */
901 destroy_viewer_streams_by_session(session);
902 try_destroy_streams(session);
903 cds_list_del(&session->viewer_session_list);
904 session_viewer_try_destroy(conn->sessions_ht, session);
905 }
906
907 /*
908 * Send the viewer the list of current sessions.
909 */
910 static
911 int viewer_get_new_streams(struct relay_connection *conn)
912 {
913 int ret, send_streams = 0;
914 uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0;
915 struct lttng_viewer_new_streams_request request;
916 struct lttng_viewer_new_streams_response response;
917 struct relay_session *session;
918 uint64_t session_id;
919
920 assert(conn);
921
922 DBG("Get new streams received");
923
924 health_code_update();
925
926 /* Receive the request from the connected client. */
927 ret = recv_request(conn->sock, &request, sizeof(request));
928 if (ret < 0) {
929 goto error;
930 }
931 session_id = be64toh(request.session_id);
932
933 health_code_update();
934
935 memset(&response, 0, sizeof(response));
936
937 rcu_read_lock();
938 session = session_find_by_id(conn->sessions_ht, session_id);
939 if (!session) {
940 DBG("Relay session %" PRIu64 " not found", session_id);
941 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
942 goto send_reply;
943 }
944
945 if (!session_attached(conn, session_id)) {
946 send_streams = 0;
947 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
948 goto send_reply;
949 }
950
951 send_streams = 1;
952 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
953
954 ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, NULL, &nb_unsent,
955 &nb_created);
956 if (ret < 0) {
957 goto end_unlock;
958 }
959 /* Only send back the newly created streams with the unsent ones. */
960 nb_streams = nb_created + nb_unsent;
961 response.streams_count = htobe32(nb_streams);
962
963 /*
964 * If the session is closed and we have no new streams to send,
965 * it means that the viewer has already received the whole trace
966 * for this session and should now close it.
967 */
968 if (nb_streams == 0 && session->close_flag) {
969 send_streams = 0;
970 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
971 /*
972 * Remove the session from the attached list of the connection
973 * and try to destroy it.
974 */
975 cds_list_del(&session->viewer_session_list);
976 cleanup_session(conn, session);
977 goto send_reply;
978 }
979
980 send_reply:
981 health_code_update();
982 ret = send_response(conn->sock, &response, sizeof(response));
983 if (ret < 0) {
984 goto end_unlock;
985 }
986 health_code_update();
987
988 /*
989 * Unknown or empty session, just return gracefully, the viewer knows what
990 * is happening.
991 */
992 if (!send_streams || !nb_streams) {
993 ret = 0;
994 goto end_unlock;
995 }
996
997 /*
998 * Send stream and *DON'T* ignore the sent flag so every viewer streams
999 * that were not sent from that point will be sent to the viewer.
1000 */
1001 ret = send_viewer_streams(conn->sock, session, 0);
1002 if (ret < 0) {
1003 goto end_unlock;
1004 }
1005
1006 end_unlock:
1007 rcu_read_unlock();
1008 error:
1009 return ret;
1010 }
1011
1012 /*
1013 * Send the viewer the list of current sessions.
1014 */
1015 static
1016 int viewer_attach_session(struct relay_connection *conn)
1017 {
1018 int send_streams = 0;
1019 ssize_t ret;
1020 uint32_t nb_streams = 0;
1021 enum lttng_viewer_seek seek_type;
1022 struct lttng_viewer_attach_session_request request;
1023 struct lttng_viewer_attach_session_response response;
1024 struct relay_session *session;
1025
1026 assert(conn);
1027
1028 health_code_update();
1029
1030 /* Receive the request from the connected client. */
1031 ret = recv_request(conn->sock, &request, sizeof(request));
1032 if (ret < 0) {
1033 goto error;
1034 }
1035
1036 health_code_update();
1037
1038 memset(&response, 0, sizeof(response));
1039
1040 if (!conn->viewer_session) {
1041 DBG("Client trying to attach before creating a live viewer session");
1042 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1043 goto send_reply;
1044 }
1045
1046 rcu_read_lock();
1047 session = session_find_by_id(conn->sessions_ht,
1048 be64toh(request.session_id));
1049 if (!session) {
1050 DBG("Relay session %" PRIu64 " not found",
1051 be64toh(request.session_id));
1052 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1053 goto send_reply;
1054 }
1055 session_viewer_attach(session);
1056 DBG("Attach session ID %" PRIu64 " received", be64toh(request.session_id));
1057
1058 if (uatomic_read(&session->viewer_refcount) > 1) {
1059 DBG("Already a viewer attached");
1060 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1061 session_viewer_detach(session);
1062 goto send_reply;
1063 } else if (session->live_timer == 0) {
1064 DBG("Not live session");
1065 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1066 goto send_reply;
1067 } else {
1068 send_streams = 1;
1069 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1070 cds_list_add(&session->viewer_session_list,
1071 &conn->viewer_session->sessions_head);
1072 }
1073
1074 switch (be32toh(request.seek)) {
1075 case LTTNG_VIEWER_SEEK_BEGINNING:
1076 case LTTNG_VIEWER_SEEK_LAST:
1077 seek_type = be32toh(request.seek);
1078 break;
1079 default:
1080 ERR("Wrong seek parameter");
1081 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1082 send_streams = 0;
1083 goto send_reply;
1084 }
1085
1086 ret = make_viewer_streams(session, seek_type, &nb_streams, NULL, NULL);
1087 if (ret < 0) {
1088 goto end_unlock;
1089 }
1090 response.streams_count = htobe32(nb_streams);
1091
1092 send_reply:
1093 health_code_update();
1094 ret = send_response(conn->sock, &response, sizeof(response));
1095 if (ret < 0) {
1096 goto end_unlock;
1097 }
1098 health_code_update();
1099
1100 /*
1101 * Unknown or empty session, just return gracefully, the viewer knows what
1102 * is happening.
1103 */
1104 if (!send_streams || !nb_streams) {
1105 ret = 0;
1106 goto end_unlock;
1107 }
1108
1109 /* Send stream and ignore the sent flag. */
1110 ret = send_viewer_streams(conn->sock, session, 1);
1111 if (ret < 0) {
1112 goto end_unlock;
1113 }
1114
1115 end_unlock:
1116 rcu_read_unlock();
1117 error:
1118 return ret;
1119 }
1120
1121 /*
1122 * Open the index file if needed for the given vstream.
1123 *
1124 * If an index file is successfully opened, the index_read_fd of the stream is
1125 * set with it.
1126 *
1127 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1128 */
1129 static int try_open_index(struct relay_viewer_stream *vstream,
1130 struct relay_stream *rstream)
1131 {
1132 int ret = 0;
1133
1134 assert(vstream);
1135 assert(rstream);
1136
1137 if (vstream->index_read_fd >= 0) {
1138 goto end;
1139 }
1140
1141 /*
1142 * First time, we open the index file and at least one index is ready. The
1143 * race between the read and write of the total_index_received is
1144 * acceptable here since the client will be notified to simply come back
1145 * and get the next index.
1146 */
1147 if (rstream->total_index_received <= 0) {
1148 ret = -ENOENT;
1149 goto end;
1150 }
1151 ret = index_open(vstream->path_name, vstream->channel_name,
1152 vstream->tracefile_count, vstream->tracefile_count_current);
1153 if (ret >= 0) {
1154 vstream->index_read_fd = ret;
1155 ret = 0;
1156 goto end;
1157 }
1158
1159 end:
1160 return ret;
1161 }
1162
1163 /*
1164 * Check the status of the index for the given stream. This function updates
1165 * the index structure if needed and can destroy the vstream also for the HUP
1166 * situation.
1167 *
1168 * Return 0 means that we can proceed with the index. A value of 1 means that
1169 * the index has been updated and is ready to be send to the client. A negative
1170 * value indicates an error that can't be handled.
1171 */
1172 static int check_index_status(struct relay_viewer_stream *vstream,
1173 struct relay_stream *rstream, struct ctf_trace *trace,
1174 struct lttng_viewer_index *index)
1175 {
1176 int ret;
1177
1178 assert(vstream);
1179 assert(rstream);
1180 assert(index);
1181 assert(trace);
1182
1183 if (!rstream->close_flag) {
1184 /* Rotate on abort (overwrite). */
1185 if (vstream->abort_flag) {
1186 DBG("Viewer stream %" PRIu64 " rotate because of overwrite",
1187 vstream->stream_handle);
1188 ret = viewer_stream_rotate(vstream, rstream);
1189 if (ret < 0) {
1190 goto error;
1191 } else if (ret == 1) {
1192 /* EOF */
1193 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1194 goto hup;
1195 }
1196 /* ret == 0 means successful so we continue. */
1197 }
1198
1199 /* Check if we are in the same trace file at this point. */
1200 if (rstream->tracefile_count_current == vstream->tracefile_count_current) {
1201 if (rstream->beacon_ts_end != -1ULL &&
1202 vstream->last_sent_index == rstream->total_index_received) {
1203 /*
1204 * We've received a synchronization beacon and the last index
1205 * available has been sent, the index for now is inactive.
1206 */
1207 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1208 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1209 goto index_ready;
1210 } else if (rstream->total_index_received <= vstream->last_sent_index
1211 && !vstream->close_write_flag) {
1212 /*
1213 * Reader and writer are working in the same tracefile, so we care
1214 * about the number of index received and sent. Otherwise, we read
1215 * up to EOF.
1216 */
1217 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1218 goto index_ready;
1219 }
1220 }
1221 /* Nothing to do with the index, continue with it. */
1222 ret = 0;
1223 } else if (rstream->close_flag && vstream->close_write_flag &&
1224 vstream->total_index_received == vstream->last_sent_index) {
1225 /* Last index sent and current tracefile closed in write */
1226 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1227 goto hup;
1228 } else {
1229 vstream->close_write_flag = 1;
1230 ret = 0;
1231 }
1232
1233 error:
1234 return ret;
1235
1236 hup:
1237 viewer_stream_delete(vstream);
1238 viewer_stream_destroy(trace, vstream);
1239 index_ready:
1240 return 1;
1241 }
1242
1243 /*
1244 * Send the next index for a stream.
1245 *
1246 * Return 0 on success or else a negative value.
1247 */
1248 static
1249 int viewer_get_next_index(struct relay_connection *conn)
1250 {
1251 int ret;
1252 ssize_t read_ret;
1253 struct lttng_viewer_get_next_index request_index;
1254 struct lttng_viewer_index viewer_index;
1255 struct ctf_packet_index packet_index;
1256 struct relay_viewer_stream *vstream;
1257 struct relay_stream *rstream;
1258 struct ctf_trace *ctf_trace;
1259 struct relay_session *session;
1260
1261 assert(conn);
1262
1263 DBG("Viewer get next index");
1264
1265 health_code_update();
1266
1267 ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1268 if (ret < 0) {
1269 goto end;
1270 }
1271 health_code_update();
1272
1273 rcu_read_lock();
1274 vstream = viewer_stream_find_by_id(be64toh(request_index.stream_id));
1275 if (!vstream) {
1276 ret = -1;
1277 goto end_unlock;
1278 }
1279
1280 session = session_find_by_id(conn->sessions_ht, vstream->session_id);
1281 if (!session) {
1282 ret = -1;
1283 goto end_unlock;
1284 }
1285
1286 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, vstream->path_name);
1287 assert(ctf_trace);
1288
1289 memset(&viewer_index, 0, sizeof(viewer_index));
1290
1291 /*
1292 * The viewer should not ask for index on metadata stream.
1293 */
1294 if (vstream->metadata_flag) {
1295 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1296 goto send_reply;
1297 }
1298
1299 rstream = stream_find_by_id(relay_streams_ht, vstream->stream_handle);
1300 assert(rstream);
1301
1302 /* Try to open an index if one is needed for that stream. */
1303 ret = try_open_index(vstream, rstream);
1304 if (ret < 0) {
1305 if (ret == -ENOENT) {
1306 /*
1307 * The index is created only when the first data packet arrives, it
1308 * might not be ready at the beginning of the session
1309 */
1310 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1311 } else {
1312 /* Unhandled error. */
1313 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1314 }
1315 goto send_reply;
1316 }
1317
1318 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1319 ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1320 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1321 if (ret < 0) {
1322 goto end;
1323 } else if (ret == 1) {
1324 /*
1325 * This means the viewer index data structure has been populated by the
1326 * check call thus we now send back the reply to the client.
1327 */
1328 goto send_reply;
1329 }
1330 /* At this point, ret MUST be 0 thus we continue with the get. */
1331 assert(!ret);
1332
1333 if (!ctf_trace->metadata_received ||
1334 ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1335 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1336 }
1337
1338 ret = check_new_streams(conn);
1339 if (ret < 0) {
1340 goto end_unlock;
1341 } else if (ret == 1) {
1342 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1343 }
1344
1345 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1346 pthread_mutex_lock(&vstream->overwrite_lock);
1347 if (vstream->abort_flag) {
1348 /* The file is being overwritten by the writer, we cannot use it. */
1349 pthread_mutex_unlock(&vstream->overwrite_lock);
1350 ret = viewer_stream_rotate(vstream, rstream);
1351 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1352 if (ret < 0) {
1353 goto end_unlock;
1354 } else if (ret == 1) {
1355 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1356 viewer_stream_delete(vstream);
1357 viewer_stream_destroy(ctf_trace, vstream);
1358 } else {
1359 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1360 }
1361 goto send_reply;
1362 }
1363
1364 read_ret = lttng_read(vstream->index_read_fd, &packet_index,
1365 sizeof(packet_index));
1366 pthread_mutex_unlock(&vstream->overwrite_lock);
1367 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1368 if (read_ret < 0) {
1369 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1370 viewer_stream_delete(vstream);
1371 viewer_stream_destroy(ctf_trace, vstream);
1372 goto send_reply;
1373 } else if (read_ret < sizeof(packet_index)) {
1374 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1375 if (vstream->close_write_flag) {
1376 ret = viewer_stream_rotate(vstream, rstream);
1377 if (ret < 0) {
1378 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1379 goto end_unlock;
1380 } else if (ret == 1) {
1381 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1382 viewer_stream_delete(vstream);
1383 viewer_stream_destroy(ctf_trace, vstream);
1384 } else {
1385 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1386 }
1387 } else {
1388 ERR("Relay reading index file %d", vstream->index_read_fd);
1389 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1390 }
1391 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1392 goto send_reply;
1393 } else {
1394 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1395 vstream->last_sent_index++;
1396 }
1397
1398 /*
1399 * Indexes are stored in big endian, no need to switch before sending.
1400 */
1401 viewer_index.offset = packet_index.offset;
1402 viewer_index.packet_size = packet_index.packet_size;
1403 viewer_index.content_size = packet_index.content_size;
1404 viewer_index.timestamp_begin = packet_index.timestamp_begin;
1405 viewer_index.timestamp_end = packet_index.timestamp_end;
1406 viewer_index.events_discarded = packet_index.events_discarded;
1407 viewer_index.stream_id = packet_index.stream_id;
1408
1409 send_reply:
1410 viewer_index.flags = htobe32(viewer_index.flags);
1411 health_code_update();
1412
1413 ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1414 if (ret < 0) {
1415 goto end_unlock;
1416 }
1417 health_code_update();
1418
1419 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1420 vstream->last_sent_index, vstream->stream_handle);
1421
1422 end_unlock:
1423 rcu_read_unlock();
1424
1425 end:
1426 return ret;
1427 }
1428
1429 /*
1430 * Send the next index for a stream
1431 *
1432 * Return 0 on success or else a negative value.
1433 */
1434 static
1435 int viewer_get_packet(struct relay_connection *conn)
1436 {
1437 int ret, send_data = 0;
1438 char *data = NULL;
1439 uint32_t len = 0;
1440 ssize_t read_len;
1441 struct lttng_viewer_get_packet get_packet_info;
1442 struct lttng_viewer_trace_packet reply;
1443 struct relay_viewer_stream *stream;
1444 struct relay_session *session;
1445 struct ctf_trace *ctf_trace;
1446
1447 assert(conn);
1448
1449 DBG2("Relay get data packet");
1450
1451 health_code_update();
1452
1453 ret = recv_request(conn->sock, &get_packet_info, sizeof(get_packet_info));
1454 if (ret < 0) {
1455 goto end;
1456 }
1457 health_code_update();
1458
1459 /* From this point on, the error label can be reached. */
1460 memset(&reply, 0, sizeof(reply));
1461
1462 rcu_read_lock();
1463 stream = viewer_stream_find_by_id(be64toh(get_packet_info.stream_id));
1464 if (!stream) {
1465 goto error;
1466 }
1467
1468 session = session_find_by_id(conn->sessions_ht, stream->session_id);
1469 if (!session) {
1470 ret = -1;
1471 goto error;
1472 }
1473
1474 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1475 stream->path_name);
1476 assert(ctf_trace);
1477
1478 /*
1479 * First time we read this stream, we need open the tracefile, we should
1480 * only arrive here if an index has already been sent to the viewer, so the
1481 * tracefile must exist, if it does not it is a fatal error.
1482 */
1483 if (stream->read_fd < 0) {
1484 char fullpath[PATH_MAX];
1485
1486 if (stream->tracefile_count > 0) {
1487 ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, stream->path_name,
1488 stream->channel_name,
1489 stream->tracefile_count_current);
1490 } else {
1491 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1492 stream->channel_name);
1493 }
1494 if (ret < 0) {
1495 goto error;
1496 }
1497 ret = open(fullpath, O_RDONLY);
1498 if (ret < 0) {
1499 PERROR("Relay opening trace file");
1500 goto error;
1501 }
1502 stream->read_fd = ret;
1503 }
1504
1505 if (!ctf_trace->metadata_received ||
1506 ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1507 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1508 reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1509 goto send_reply;
1510 }
1511
1512 ret = check_new_streams(conn);
1513 if (ret < 0) {
1514 goto end_unlock;
1515 } else if (ret == 1) {
1516 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1517 reply.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1518 goto send_reply;
1519 }
1520
1521 len = be32toh(get_packet_info.len);
1522 data = zmalloc(len);
1523 if (!data) {
1524 PERROR("relay data zmalloc");
1525 goto error;
1526 }
1527
1528 ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
1529 if (ret < 0) {
1530 /*
1531 * If the read fd was closed by the streaming side, the
1532 * abort_flag will be set to 1, otherwise it is an error.
1533 */
1534 if (stream->abort_flag == 0) {
1535 PERROR("lseek");
1536 goto error;
1537 }
1538 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1539 goto send_reply;
1540 }
1541 read_len = lttng_read(stream->read_fd, data, len);
1542 if (read_len < len) {
1543 /*
1544 * If the read fd was closed by the streaming side, the
1545 * abort_flag will be set to 1, otherwise it is an error.
1546 */
1547 if (stream->abort_flag == 0) {
1548 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1549 stream->read_fd,
1550 be64toh(get_packet_info.offset));
1551 goto error;
1552 } else {
1553 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1554 goto send_reply;
1555 }
1556 }
1557 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1558 reply.len = htobe32(len);
1559 send_data = 1;
1560 goto send_reply;
1561
1562 error:
1563 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1564
1565 send_reply:
1566 reply.flags = htobe32(reply.flags);
1567
1568 health_code_update();
1569
1570 ret = send_response(conn->sock, &reply, sizeof(reply));
1571 if (ret < 0) {
1572 goto end_unlock;
1573 }
1574 health_code_update();
1575
1576 if (send_data) {
1577 health_code_update();
1578 ret = send_response(conn->sock, data, len);
1579 if (ret < 0) {
1580 goto end_unlock;
1581 }
1582 health_code_update();
1583 }
1584
1585 DBG("Sent %u bytes for stream %" PRIu64, len,
1586 be64toh(get_packet_info.stream_id));
1587
1588 end_unlock:
1589 free(data);
1590 rcu_read_unlock();
1591
1592 end:
1593 return ret;
1594 }
1595
1596 /*
1597 * Send the session's metadata
1598 *
1599 * Return 0 on success else a negative value.
1600 */
1601 static
1602 int viewer_get_metadata(struct relay_connection *conn)
1603 {
1604 int ret = 0;
1605 ssize_t read_len;
1606 uint64_t len = 0;
1607 char *data = NULL;
1608 struct lttng_viewer_get_metadata request;
1609 struct lttng_viewer_metadata_packet reply;
1610 struct relay_viewer_stream *stream;
1611 struct ctf_trace *ctf_trace;
1612 struct relay_session *session;
1613
1614 assert(conn);
1615
1616 DBG("Relay get metadata");
1617
1618 health_code_update();
1619
1620 ret = recv_request(conn->sock, &request, sizeof(request));
1621 if (ret < 0) {
1622 goto end;
1623 }
1624 health_code_update();
1625
1626 memset(&reply, 0, sizeof(reply));
1627
1628 rcu_read_lock();
1629 stream = viewer_stream_find_by_id(be64toh(request.stream_id));
1630 if (!stream || !stream->metadata_flag) {
1631 ERR("Invalid metadata stream");
1632 goto error;
1633 }
1634
1635 session = session_find_by_id(conn->sessions_ht, stream->session_id);
1636 if (!session) {
1637 ret = -1;
1638 goto error;
1639 }
1640
1641 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1642 stream->path_name);
1643 assert(ctf_trace);
1644 assert(ctf_trace->metadata_sent <= ctf_trace->metadata_received);
1645
1646 len = ctf_trace->metadata_received - ctf_trace->metadata_sent;
1647 if (len == 0) {
1648 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1649 goto send_reply;
1650 }
1651
1652 /* first time, we open the metadata file */
1653 if (stream->read_fd < 0) {
1654 char fullpath[PATH_MAX];
1655
1656 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1657 stream->channel_name);
1658 if (ret < 0) {
1659 goto error;
1660 }
1661 ret = open(fullpath, O_RDONLY);
1662 if (ret < 0) {
1663 PERROR("Relay opening metadata file");
1664 goto error;
1665 }
1666 stream->read_fd = ret;
1667 }
1668
1669 reply.len = htobe64(len);
1670 data = zmalloc(len);
1671 if (!data) {
1672 PERROR("viewer metadata zmalloc");
1673 goto error;
1674 }
1675
1676 read_len = lttng_read(stream->read_fd, data, len);
1677 if (read_len < len) {
1678 PERROR("Relay reading metadata file");
1679 goto error;
1680 }
1681 ctf_trace->metadata_sent += read_len;
1682 reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1683 goto send_reply;
1684
1685 error:
1686 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1687
1688 send_reply:
1689 health_code_update();
1690 ret = send_response(conn->sock, &reply, sizeof(reply));
1691 if (ret < 0) {
1692 goto end_unlock;
1693 }
1694 health_code_update();
1695
1696 if (len > 0) {
1697 ret = send_response(conn->sock, data, len);
1698 if (ret < 0) {
1699 goto end_unlock;
1700 }
1701 }
1702
1703 DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1704 be64toh(request.stream_id));
1705
1706 DBG("Metadata sent");
1707
1708 end_unlock:
1709 free(data);
1710 rcu_read_unlock();
1711 end:
1712 return ret;
1713 }
1714
1715 /*
1716 * Create a viewer session.
1717 *
1718 * Return 0 on success or else a negative value.
1719 */
1720 static
1721 int viewer_create_session(struct relay_connection *conn)
1722 {
1723 int ret;
1724 struct lttng_viewer_create_session_response resp;
1725
1726 DBG("Viewer create session received");
1727
1728 memset(&resp, 0, sizeof(resp));
1729 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1730 conn->viewer_session = zmalloc(sizeof(*conn->viewer_session));
1731 if (!conn->viewer_session) {
1732 ERR("Allocation viewer session");
1733 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1734 goto send_reply;
1735 }
1736 CDS_INIT_LIST_HEAD(&conn->viewer_session->sessions_head);
1737
1738 send_reply:
1739 health_code_update();
1740 ret = send_response(conn->sock, &resp, sizeof(resp));
1741 if (ret < 0) {
1742 goto end;
1743 }
1744 health_code_update();
1745 ret = 0;
1746
1747 end:
1748 return ret;
1749 }
1750
1751
1752 /*
1753 * live_relay_unknown_command: send -1 if received unknown command
1754 */
1755 static
1756 void live_relay_unknown_command(struct relay_connection *conn)
1757 {
1758 struct lttcomm_relayd_generic_reply reply;
1759
1760 memset(&reply, 0, sizeof(reply));
1761 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1762 (void) send_response(conn->sock, &reply, sizeof(reply));
1763 }
1764
1765 /*
1766 * Process the commands received on the control socket
1767 */
1768 static
1769 int process_control(struct lttng_viewer_cmd *recv_hdr,
1770 struct relay_connection *conn)
1771 {
1772 int ret = 0;
1773 uint32_t msg_value;
1774
1775 assert(recv_hdr);
1776 assert(conn);
1777
1778 msg_value = be32toh(recv_hdr->cmd);
1779
1780 /*
1781 * Make sure we've done the version check before any command other then a
1782 * new client connection.
1783 */
1784 if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1785 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1786 ret = -1;
1787 goto end;
1788 }
1789
1790 switch (msg_value) {
1791 case LTTNG_VIEWER_CONNECT:
1792 ret = viewer_connect(conn);
1793 break;
1794 case LTTNG_VIEWER_LIST_SESSIONS:
1795 ret = viewer_list_sessions(conn);
1796 break;
1797 case LTTNG_VIEWER_ATTACH_SESSION:
1798 ret = viewer_attach_session(conn);
1799 break;
1800 case LTTNG_VIEWER_GET_NEXT_INDEX:
1801 ret = viewer_get_next_index(conn);
1802 break;
1803 case LTTNG_VIEWER_GET_PACKET:
1804 ret = viewer_get_packet(conn);
1805 break;
1806 case LTTNG_VIEWER_GET_METADATA:
1807 ret = viewer_get_metadata(conn);
1808 break;
1809 case LTTNG_VIEWER_GET_NEW_STREAMS:
1810 ret = viewer_get_new_streams(conn);
1811 break;
1812 case LTTNG_VIEWER_CREATE_SESSION:
1813 ret = viewer_create_session(conn);
1814 break;
1815 default:
1816 ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd));
1817 live_relay_unknown_command(conn);
1818 ret = -1;
1819 goto end;
1820 }
1821
1822 end:
1823 return ret;
1824 }
1825
1826 static
1827 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1828 {
1829 int ret;
1830
1831 assert(events);
1832
1833 (void) lttng_poll_del(events, pollfd);
1834
1835 ret = close(pollfd);
1836 if (ret < 0) {
1837 ERR("Closing pollfd %d", pollfd);
1838 }
1839 }
1840
1841 /*
1842 * Delete and destroy a connection.
1843 *
1844 * RCU read side lock MUST be acquired.
1845 */
1846 static void destroy_connection(struct lttng_ht *relay_connections_ht,
1847 struct relay_connection *conn)
1848 {
1849 struct relay_session *session, *tmp_session;
1850
1851 assert(relay_connections_ht);
1852 assert(conn);
1853
1854 connection_delete(relay_connections_ht, conn);
1855
1856 if (!conn->viewer_session) {
1857 goto end;
1858 }
1859
1860 rcu_read_lock();
1861 cds_list_for_each_entry_safe(session, tmp_session,
1862 &conn->viewer_session->sessions_head,
1863 viewer_session_list) {
1864 DBG("Cleaning connection of session ID %" PRIu64, session->id);
1865 cleanup_session(conn, session);
1866 }
1867 rcu_read_unlock();
1868
1869 end:
1870 connection_destroy(conn);
1871 }
1872
1873 /*
1874 * This thread does the actual work
1875 */
1876 static
1877 void *thread_worker(void *data)
1878 {
1879 int ret, err = -1;
1880 uint32_t nb_fd;
1881 struct relay_connection *conn;
1882 struct lttng_poll_event events;
1883 struct lttng_ht *relay_connections_ht;
1884 struct lttng_ht_iter iter;
1885 struct lttng_viewer_cmd recv_hdr;
1886 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
1887 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
1888
1889 DBG("[thread] Live viewer relay worker started");
1890
1891 rcu_register_thread();
1892
1893 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1894
1895 if (testpoint(relayd_thread_live_worker)) {
1896 goto error_testpoint;
1897 }
1898
1899 /* table of connections indexed on socket */
1900 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1901 if (!relay_connections_ht) {
1902 goto relay_connections_ht_error;
1903 }
1904
1905 ret = create_thread_poll_set(&events, 2);
1906 if (ret < 0) {
1907 goto error_poll_create;
1908 }
1909
1910 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1911 if (ret < 0) {
1912 goto error;
1913 }
1914
1915 restart:
1916 while (1) {
1917 int i;
1918
1919 health_code_update();
1920
1921 /* Infinite blocking call, waiting for transmission */
1922 DBG3("Relayd live viewer worker thread polling...");
1923 health_poll_entry();
1924 ret = lttng_poll_wait(&events, -1);
1925 health_poll_exit();
1926 if (ret < 0) {
1927 /*
1928 * Restart interrupted system call.
1929 */
1930 if (errno == EINTR) {
1931 goto restart;
1932 }
1933 goto error;
1934 }
1935
1936 nb_fd = ret;
1937
1938 /*
1939 * Process control. The control connection is prioritised so we don't
1940 * starve it with high throughput tracing data on the data
1941 * connection.
1942 */
1943 for (i = 0; i < nb_fd; i++) {
1944 /* Fetch once the poll data */
1945 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1946 int pollfd = LTTNG_POLL_GETFD(&events, i);
1947
1948 health_code_update();
1949
1950 /* Thread quit pipe has been closed. Killing thread. */
1951 ret = check_thread_quit_pipe(pollfd, revents);
1952 if (ret) {
1953 err = 0;
1954 goto exit;
1955 }
1956
1957 /* Inspect the relay conn pipe for new connection */
1958 if (pollfd == live_conn_pipe[0]) {
1959 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1960 ERR("Relay live pipe error");
1961 goto error;
1962 } else if (revents & LPOLLIN) {
1963 ret = lttng_read(live_conn_pipe[0], &conn, sizeof(conn));
1964 if (ret < 0) {
1965 goto error;
1966 }
1967 conn->sessions_ht = sessions_ht;
1968 connection_init(conn);
1969 lttng_poll_add(&events, conn->sock->fd,
1970 LPOLLIN | LPOLLRDHUP);
1971 rcu_read_lock();
1972 lttng_ht_add_unique_ulong(relay_connections_ht,
1973 &conn->sock_n);
1974 rcu_read_unlock();
1975 DBG("Connection socket %d added", conn->sock->fd);
1976 }
1977 } else {
1978 rcu_read_lock();
1979 conn = connection_find_by_sock(relay_connections_ht, pollfd);
1980 /* If not found, there is a synchronization issue. */
1981 assert(conn);
1982
1983 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1984 cleanup_connection_pollfd(&events, pollfd);
1985 destroy_connection(relay_connections_ht, conn);
1986 } else if (revents & LPOLLIN) {
1987 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
1988 sizeof(recv_hdr), 0);
1989 if (ret <= 0) {
1990 /* Connection closed */
1991 cleanup_connection_pollfd(&events, pollfd);
1992 destroy_connection(relay_connections_ht, conn);
1993 DBG("Viewer control conn closed with %d", pollfd);
1994 } else {
1995 ret = process_control(&recv_hdr, conn);
1996 if (ret < 0) {
1997 /* Clear the session on error. */
1998 cleanup_connection_pollfd(&events, pollfd);
1999 destroy_connection(relay_connections_ht, conn);
2000 DBG("Viewer connection closed with %d", pollfd);
2001 }
2002 }
2003 }
2004 rcu_read_unlock();
2005 }
2006 }
2007 }
2008
2009 exit:
2010 error:
2011 lttng_poll_clean(&events);
2012
2013 /* Cleanup reamaining connection object. */
2014 rcu_read_lock();
2015 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn,
2016 sock_n.node) {
2017 health_code_update();
2018 destroy_connection(relay_connections_ht, conn);
2019 }
2020 rcu_read_unlock();
2021 error_poll_create:
2022 lttng_ht_destroy(relay_connections_ht);
2023 relay_connections_ht_error:
2024 /* Close relay conn pipes */
2025 utils_close_pipe(live_conn_pipe);
2026 if (err) {
2027 DBG("Viewer worker thread exited with error");
2028 }
2029 DBG("Viewer worker thread cleanup complete");
2030 error_testpoint:
2031 if (err) {
2032 health_error();
2033 ERR("Health error occurred in %s", __func__);
2034 }
2035 health_unregister(health_relayd);
2036 stop_threads();
2037 rcu_unregister_thread();
2038 return NULL;
2039 }
2040
2041 /*
2042 * Create the relay command pipe to wake thread_manage_apps.
2043 * Closed in cleanup().
2044 */
2045 static int create_conn_pipe(void)
2046 {
2047 int ret;
2048
2049 ret = utils_create_pipe_cloexec(live_conn_pipe);
2050
2051 return ret;
2052 }
2053
2054 void live_stop_threads(void)
2055 {
2056 int ret;
2057 void *status;
2058
2059 stop_threads();
2060
2061 ret = pthread_join(live_listener_thread, &status);
2062 if (ret != 0) {
2063 PERROR("pthread_join live listener");
2064 goto error; /* join error, exit without cleanup */
2065 }
2066
2067 ret = pthread_join(live_worker_thread, &status);
2068 if (ret != 0) {
2069 PERROR("pthread_join live worker");
2070 goto error; /* join error, exit without cleanup */
2071 }
2072
2073 ret = pthread_join(live_dispatcher_thread, &status);
2074 if (ret != 0) {
2075 PERROR("pthread_join live dispatcher");
2076 goto error; /* join error, exit without cleanup */
2077 }
2078
2079 cleanup();
2080
2081 error:
2082 return;
2083 }
2084
2085 /*
2086 * main
2087 */
2088 int live_start_threads(struct lttng_uri *uri,
2089 struct relay_local_data *relay_ctx)
2090 {
2091 int ret = 0;
2092 void *status;
2093 int is_root;
2094
2095 assert(uri);
2096 live_uri = uri;
2097
2098 /* Check if daemon is UID = 0 */
2099 is_root = !getuid();
2100
2101 if (!is_root) {
2102 if (live_uri->port < 1024) {
2103 ERR("Need to be root to use ports < 1024");
2104 ret = -1;
2105 goto exit;
2106 }
2107 }
2108
2109 /* Setup the thread apps communication pipe. */
2110 if ((ret = create_conn_pipe()) < 0) {
2111 goto exit;
2112 }
2113
2114 /* Init relay command queue. */
2115 cds_wfq_init(&viewer_conn_queue.queue);
2116
2117 /* Set up max poll set size */
2118 lttng_poll_set_max_size();
2119
2120 /* Setup the dispatcher thread */
2121 ret = pthread_create(&live_dispatcher_thread, NULL,
2122 thread_dispatcher, (void *) NULL);
2123 if (ret != 0) {
2124 PERROR("pthread_create viewer dispatcher");
2125 goto exit_dispatcher;
2126 }
2127
2128 /* Setup the worker thread */
2129 ret = pthread_create(&live_worker_thread, NULL,
2130 thread_worker, relay_ctx);
2131 if (ret != 0) {
2132 PERROR("pthread_create viewer worker");
2133 goto exit_worker;
2134 }
2135
2136 /* Setup the listener thread */
2137 ret = pthread_create(&live_listener_thread, NULL,
2138 thread_listener, (void *) NULL);
2139 if (ret != 0) {
2140 PERROR("pthread_create viewer listener");
2141 goto exit_listener;
2142 }
2143
2144 ret = 0;
2145 goto end;
2146
2147 exit_listener:
2148 ret = pthread_join(live_listener_thread, &status);
2149 if (ret != 0) {
2150 PERROR("pthread_join live listener");
2151 goto error; /* join error, exit without cleanup */
2152 }
2153
2154 exit_worker:
2155 ret = pthread_join(live_worker_thread, &status);
2156 if (ret != 0) {
2157 PERROR("pthread_join live worker");
2158 goto error; /* join error, exit without cleanup */
2159 }
2160
2161 exit_dispatcher:
2162 ret = pthread_join(live_dispatcher_thread, &status);
2163 if (ret != 0) {
2164 PERROR("pthread_join live dispatcher");
2165 goto error; /* join error, exit without cleanup */
2166 }
2167
2168 exit:
2169 cleanup();
2170
2171 end:
2172 error:
2173 return ret;
2174 }
This page took 0.122286 seconds and 6 git commands to generate.