Fix: lttng-live: send HUP reply when per-PID streams are gone
[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 * 2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _LGPL_SOURCE
21 #include <getopt.h>
22 #include <grp.h>
23 #include <limits.h>
24 #include <pthread.h>
25 #include <signal.h>
26 #include <stdio.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <sys/mman.h>
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
33 #include <sys/stat.h>
34 #include <sys/types.h>
35 #include <sys/wait.h>
36 #include <inttypes.h>
37 #include <urcu/futex.h>
38 #include <urcu/uatomic.h>
39 #include <urcu/rculist.h>
40 #include <unistd.h>
41 #include <fcntl.h>
42
43 #include <lttng/lttng.h>
44 #include <common/common.h>
45 #include <common/compat/poll.h>
46 #include <common/compat/socket.h>
47 #include <common/compat/endian.h>
48 #include <common/defaults.h>
49 #include <common/futex.h>
50 #include <common/index/index.h>
51 #include <common/sessiond-comm/sessiond-comm.h>
52 #include <common/sessiond-comm/inet.h>
53 #include <common/sessiond-comm/relayd.h>
54 #include <common/uri.h>
55 #include <common/utils.h>
56
57 #include "cmd.h"
58 #include "live.h"
59 #include "lttng-relayd.h"
60 #include "utils.h"
61 #include "health-relayd.h"
62 #include "testpoint.h"
63 #include "viewer-stream.h"
64 #include "stream.h"
65 #include "session.h"
66 #include "ctf-trace.h"
67 #include "connection.h"
68 #include "viewer-session.h"
69
70 #define SESSION_BUF_DEFAULT_COUNT 16
71
72 static struct lttng_uri *live_uri;
73
74 /*
75 * This pipe is used to inform the worker thread that a command is queued and
76 * ready to be processed.
77 */
78 static int live_conn_pipe[2] = { -1, -1 };
79
80 /* Shared between threads */
81 static int live_dispatch_thread_exit;
82
83 static pthread_t live_listener_thread;
84 static pthread_t live_dispatcher_thread;
85 static pthread_t live_worker_thread;
86
87 /*
88 * Relay command queue.
89 *
90 * The live_thread_listener and live_thread_dispatcher communicate with this
91 * queue.
92 */
93 static struct relay_conn_queue viewer_conn_queue;
94
95 static uint64_t last_relay_viewer_session_id;
96 static pthread_mutex_t last_relay_viewer_session_id_lock =
97 PTHREAD_MUTEX_INITIALIZER;
98
99 /*
100 * Cleanup the daemon
101 */
102 static
103 void cleanup_relayd_live(void)
104 {
105 DBG("Cleaning up");
106
107 free(live_uri);
108 }
109
110 /*
111 * Receive a request buffer using a given socket, destination allocated buffer
112 * of length size.
113 *
114 * Return the size of the received message or else a negative value on error
115 * with errno being set by recvmsg() syscall.
116 */
117 static
118 ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
119 {
120 ssize_t ret;
121
122 ret = sock->ops->recvmsg(sock, buf, size, 0);
123 if (ret < 0 || ret != size) {
124 if (ret == 0) {
125 /* Orderly shutdown. Not necessary to print an error. */
126 DBG("Socket %d did an orderly shutdown", sock->fd);
127 } else {
128 ERR("Relay failed to receive request.");
129 }
130 ret = -1;
131 }
132
133 return ret;
134 }
135
136 /*
137 * Send a response buffer using a given socket, source allocated buffer of
138 * length size.
139 *
140 * Return the size of the sent message or else a negative value on error with
141 * errno being set by sendmsg() syscall.
142 */
143 static
144 ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
145 {
146 ssize_t ret;
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 rcu_read_lock();
174 cds_list_for_each_entry_rcu(session,
175 &conn->viewer_session->session_list,
176 viewer_session_node) {
177 if (!session_get(session)) {
178 continue;
179 }
180 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
181 ret = current_val;
182 session_put(session);
183 if (ret == 1) {
184 goto end;
185 }
186 }
187 end:
188 rcu_read_unlock();
189 return ret;
190 }
191
192 /*
193 * Send viewer streams to the given socket. The ignore_sent_flag indicates if
194 * this function should ignore the sent flag or not.
195 *
196 * Return 0 on success or else a negative value.
197 */
198 static
199 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
200 struct relay_session *session, unsigned int ignore_sent_flag)
201 {
202 ssize_t ret;
203 struct lttng_viewer_stream send_stream;
204 struct lttng_ht_iter iter;
205 struct relay_viewer_stream *vstream;
206
207 rcu_read_lock();
208
209 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
210 stream_n.node) {
211 struct ctf_trace *ctf_trace;
212
213 health_code_update();
214
215 if (!viewer_stream_get(vstream)) {
216 continue;
217 }
218
219 pthread_mutex_lock(&vstream->stream->lock);
220 /* Ignore if not the same session. */
221 if (vstream->stream->trace->session->id != session->id ||
222 (!ignore_sent_flag && vstream->sent_flag)) {
223 pthread_mutex_unlock(&vstream->stream->lock);
224 viewer_stream_put(vstream);
225 continue;
226 }
227
228 ctf_trace = vstream->stream->trace;
229 send_stream.id = htobe64(vstream->stream->stream_handle);
230 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
231 send_stream.metadata_flag = htobe32(
232 vstream->stream->is_metadata);
233 if (lttng_strncpy(send_stream.path_name, vstream->path_name,
234 sizeof(send_stream.path_name))) {
235 pthread_mutex_unlock(&vstream->stream->lock);
236 viewer_stream_put(vstream);
237 ret = -1; /* Error. */
238 goto end_unlock;
239 }
240 if (lttng_strncpy(send_stream.channel_name,
241 vstream->channel_name,
242 sizeof(send_stream.channel_name))) {
243 pthread_mutex_unlock(&vstream->stream->lock);
244 viewer_stream_put(vstream);
245 ret = -1; /* Error. */
246 goto end_unlock;
247 }
248
249 DBG("Sending stream %" PRIu64 " to viewer",
250 vstream->stream->stream_handle);
251 vstream->sent_flag = 1;
252 pthread_mutex_unlock(&vstream->stream->lock);
253
254 ret = send_response(sock, &send_stream, sizeof(send_stream));
255 viewer_stream_put(vstream);
256 if (ret < 0) {
257 goto end_unlock;
258 }
259 }
260
261 ret = 0;
262
263 end_unlock:
264 rcu_read_unlock();
265 return ret;
266 }
267
268 /*
269 * Create every viewer stream possible for the given session with the seek
270 * type. Three counters *can* be return which are in order the total amount of
271 * viewer stream of the session, the number of unsent stream and the number of
272 * stream created. Those counters can be NULL and thus will be ignored.
273 *
274 * Return 0 on success or else a negative value.
275 */
276 static
277 int make_viewer_streams(struct relay_session *session,
278 enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent,
279 uint32_t *nb_created, bool *closed)
280 {
281 int ret;
282 struct lttng_ht_iter iter;
283 struct ctf_trace *ctf_trace;
284
285 assert(session);
286
287 /*
288 * Hold the session lock to ensure that we see either none or
289 * all initial streams for a session, but no intermediate state.
290 */
291 pthread_mutex_lock(&session->lock);
292
293 if (session->connection_closed) {
294 *closed = true;
295 }
296
297 /*
298 * Create viewer streams for relay streams that are ready to be
299 * used for a the given session id only.
300 */
301 rcu_read_lock();
302 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
303 node.node) {
304 struct relay_stream *stream;
305
306 health_code_update();
307
308 if (!ctf_trace_get(ctf_trace)) {
309 continue;
310 }
311
312 cds_list_for_each_entry_rcu(stream, &ctf_trace->stream_list, stream_node) {
313 struct relay_viewer_stream *vstream;
314
315 if (!stream_get(stream)) {
316 continue;
317 }
318 /*
319 * stream published is protected by the session lock.
320 */
321 if (!stream->published) {
322 goto next;
323 }
324 vstream = viewer_stream_get_by_id(stream->stream_handle);
325 if (!vstream) {
326 vstream = viewer_stream_create(stream, seek_t);
327 if (!vstream) {
328 ret = -1;
329 ctf_trace_put(ctf_trace);
330 stream_put(stream);
331 goto error_unlock;
332 }
333
334 if (nb_created) {
335 /* Update number of created stream counter. */
336 (*nb_created)++;
337 }
338 /*
339 * Ensure a self-reference is preserved even
340 * after we have put our local reference.
341 */
342 if (!viewer_stream_get(vstream)) {
343 ERR("Unable to get self-reference on viewer stream, logic error.");
344 abort();
345 }
346 } else {
347 if (!vstream->sent_flag && nb_unsent) {
348 /* Update number of unsent stream counter. */
349 (*nb_unsent)++;
350 }
351 }
352 /* Update number of total stream counter. */
353 if (nb_total) {
354 if (stream->is_metadata) {
355 if (!stream->closed ||
356 stream->metadata_received > vstream->metadata_sent) {
357 (*nb_total)++;
358 }
359 } else {
360 if (!stream->closed ||
361 !(((int64_t) (stream->prev_seq - stream->last_net_seq_num)) >= 0)) {
362
363 (*nb_total)++;
364 }
365 }
366 }
367 /* Put local reference. */
368 viewer_stream_put(vstream);
369 next:
370 stream_put(stream);
371 }
372 ctf_trace_put(ctf_trace);
373 }
374
375 ret = 0;
376
377 error_unlock:
378 rcu_read_unlock();
379 pthread_mutex_unlock(&session->lock);
380 return ret;
381 }
382
383 int relayd_live_stop(void)
384 {
385 /* Stop dispatch thread */
386 CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
387 futex_nto1_wake(&viewer_conn_queue.futex);
388 return 0;
389 }
390
391 /*
392 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
393 */
394 static
395 int create_thread_poll_set(struct lttng_poll_event *events, int size)
396 {
397 int ret;
398
399 if (events == NULL || size == 0) {
400 ret = -1;
401 goto error;
402 }
403
404 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
405 if (ret < 0) {
406 goto error;
407 }
408
409 /* Add quit pipe */
410 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
411 if (ret < 0) {
412 goto error;
413 }
414
415 return 0;
416
417 error:
418 return ret;
419 }
420
421 /*
422 * Check if the thread quit pipe was triggered.
423 *
424 * Return 1 if it was triggered else 0;
425 */
426 static
427 int check_thread_quit_pipe(int fd, uint32_t events)
428 {
429 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
430 return 1;
431 }
432
433 return 0;
434 }
435
436 /*
437 * Create and init socket from uri.
438 */
439 static
440 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
441 {
442 int ret;
443 struct lttcomm_sock *sock = NULL;
444
445 sock = lttcomm_alloc_sock_from_uri(uri);
446 if (sock == NULL) {
447 ERR("Allocating socket");
448 goto error;
449 }
450
451 ret = lttcomm_create_sock(sock);
452 if (ret < 0) {
453 goto error;
454 }
455 DBG("Listening on sock %d for live", sock->fd);
456
457 ret = sock->ops->bind(sock);
458 if (ret < 0) {
459 goto error;
460 }
461
462 ret = sock->ops->listen(sock, -1);
463 if (ret < 0) {
464 goto error;
465
466 }
467
468 return sock;
469
470 error:
471 if (sock) {
472 lttcomm_destroy_sock(sock);
473 }
474 return NULL;
475 }
476
477 /*
478 * This thread manages the listening for new connections on the network
479 */
480 static
481 void *thread_listener(void *data)
482 {
483 int i, ret, pollfd, err = -1;
484 uint32_t revents, nb_fd;
485 struct lttng_poll_event events;
486 struct lttcomm_sock *live_control_sock;
487
488 DBG("[thread] Relay live listener started");
489
490 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
491
492 health_code_update();
493
494 live_control_sock = init_socket(live_uri);
495 if (!live_control_sock) {
496 goto error_sock_control;
497 }
498
499 /* Pass 2 as size here for the thread quit pipe and control sockets. */
500 ret = create_thread_poll_set(&events, 2);
501 if (ret < 0) {
502 goto error_create_poll;
503 }
504
505 /* Add the control socket */
506 ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
507 if (ret < 0) {
508 goto error_poll_add;
509 }
510
511 lttng_relay_notify_ready();
512
513 if (testpoint(relayd_thread_live_listener)) {
514 goto error_testpoint;
515 }
516
517 while (1) {
518 health_code_update();
519
520 DBG("Listener accepting live viewers connections");
521
522 restart:
523 health_poll_entry();
524 ret = lttng_poll_wait(&events, -1);
525 health_poll_exit();
526 if (ret < 0) {
527 /*
528 * Restart interrupted system call.
529 */
530 if (errno == EINTR) {
531 goto restart;
532 }
533 goto error;
534 }
535 nb_fd = ret;
536
537 DBG("Relay new viewer connection received");
538 for (i = 0; i < nb_fd; i++) {
539 health_code_update();
540
541 /* Fetch once the poll data */
542 revents = LTTNG_POLL_GETEV(&events, i);
543 pollfd = LTTNG_POLL_GETFD(&events, i);
544
545 if (!revents) {
546 /* No activity for this FD (poll implementation). */
547 continue;
548 }
549
550 /* Thread quit pipe has been closed. Killing thread. */
551 ret = check_thread_quit_pipe(pollfd, revents);
552 if (ret) {
553 err = 0;
554 goto exit;
555 }
556
557 if (revents & LPOLLIN) {
558 /*
559 * A new connection is requested, therefore a
560 * viewer connection is allocated in this
561 * thread, enqueued to a global queue and
562 * dequeued (and freed) in the worker thread.
563 */
564 int val = 1;
565 struct relay_connection *new_conn;
566 struct lttcomm_sock *newsock;
567
568 newsock = live_control_sock->ops->accept(live_control_sock);
569 if (!newsock) {
570 PERROR("accepting control sock");
571 goto error;
572 }
573 DBG("Relay viewer connection accepted socket %d", newsock->fd);
574
575 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
576 sizeof(val));
577 if (ret < 0) {
578 PERROR("setsockopt inet");
579 lttcomm_destroy_sock(newsock);
580 goto error;
581 }
582 new_conn = connection_create(newsock, RELAY_CONNECTION_UNKNOWN);
583 if (!new_conn) {
584 lttcomm_destroy_sock(newsock);
585 goto error;
586 }
587 /* Ownership assumed by the connection. */
588 newsock = NULL;
589
590 /* Enqueue request for the dispatcher thread. */
591 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
592 &new_conn->qnode);
593
594 /*
595 * Wake the dispatch queue futex.
596 * Implicit memory barrier with the
597 * exchange in cds_wfcq_enqueue.
598 */
599 futex_nto1_wake(&viewer_conn_queue.futex);
600 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
601 ERR("socket poll error");
602 goto error;
603 } else {
604 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
605 goto error;
606 }
607 }
608 }
609
610 exit:
611 error:
612 error_poll_add:
613 error_testpoint:
614 lttng_poll_clean(&events);
615 error_create_poll:
616 if (live_control_sock->fd >= 0) {
617 ret = live_control_sock->ops->close(live_control_sock);
618 if (ret) {
619 PERROR("close");
620 }
621 }
622 lttcomm_destroy_sock(live_control_sock);
623 error_sock_control:
624 if (err) {
625 health_error();
626 DBG("Live viewer listener thread exited with error");
627 }
628 health_unregister(health_relayd);
629 DBG("Live viewer listener thread cleanup complete");
630 if (lttng_relay_stop_threads()) {
631 ERR("Error stopping threads");
632 }
633 return NULL;
634 }
635
636 /*
637 * This thread manages the dispatching of the requests to worker threads
638 */
639 static
640 void *thread_dispatcher(void *data)
641 {
642 int err = -1;
643 ssize_t ret;
644 struct cds_wfcq_node *node;
645 struct relay_connection *conn = NULL;
646
647 DBG("[thread] Live viewer relay dispatcher started");
648
649 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
650
651 if (testpoint(relayd_thread_live_dispatcher)) {
652 goto error_testpoint;
653 }
654
655 health_code_update();
656
657 while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
658 health_code_update();
659
660 /* Atomically prepare the queue futex */
661 futex_nto1_prepare(&viewer_conn_queue.futex);
662
663 do {
664 health_code_update();
665
666 /* Dequeue commands */
667 node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
668 &viewer_conn_queue.tail);
669 if (node == NULL) {
670 DBG("Woken up but nothing in the live-viewer "
671 "relay command queue");
672 /* Continue thread execution */
673 break;
674 }
675 conn = caa_container_of(node, struct relay_connection, qnode);
676 DBG("Dispatching viewer request waiting on sock %d",
677 conn->sock->fd);
678
679 /*
680 * Inform worker thread of the new request. This
681 * call is blocking so we can be assured that
682 * the data will be read at some point in time
683 * or wait to the end of the world :)
684 */
685 ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
686 if (ret < 0) {
687 PERROR("write conn pipe");
688 connection_put(conn);
689 goto error;
690 }
691 } while (node != NULL);
692
693 /* Futex wait on queue. Blocking call on futex() */
694 health_poll_entry();
695 futex_nto1_wait(&viewer_conn_queue.futex);
696 health_poll_exit();
697 }
698
699 /* Normal exit, no error */
700 err = 0;
701
702 error:
703 error_testpoint:
704 if (err) {
705 health_error();
706 ERR("Health error occurred in %s", __func__);
707 }
708 health_unregister(health_relayd);
709 DBG("Live viewer dispatch thread dying");
710 if (lttng_relay_stop_threads()) {
711 ERR("Error stopping threads");
712 }
713 return NULL;
714 }
715
716 /*
717 * Establish connection with the viewer and check the versions.
718 *
719 * Return 0 on success or else negative value.
720 */
721 static
722 int viewer_connect(struct relay_connection *conn)
723 {
724 int ret;
725 struct lttng_viewer_connect reply, msg;
726
727 conn->version_check_done = 1;
728
729 health_code_update();
730
731 DBG("Viewer is establishing a connection to the relayd.");
732
733 ret = recv_request(conn->sock, &msg, sizeof(msg));
734 if (ret < 0) {
735 goto end;
736 }
737
738 health_code_update();
739
740 memset(&reply, 0, sizeof(reply));
741 reply.major = RELAYD_VERSION_COMM_MAJOR;
742 reply.minor = RELAYD_VERSION_COMM_MINOR;
743
744 /* Major versions must be the same */
745 if (reply.major != be32toh(msg.major)) {
746 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
747 reply.major, be32toh(msg.major));
748 ret = -1;
749 goto end;
750 }
751
752 conn->major = reply.major;
753 /* We adapt to the lowest compatible version */
754 if (reply.minor <= be32toh(msg.minor)) {
755 conn->minor = reply.minor;
756 } else {
757 conn->minor = be32toh(msg.minor);
758 }
759
760 if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
761 conn->type = RELAY_VIEWER_COMMAND;
762 } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
763 conn->type = RELAY_VIEWER_NOTIFICATION;
764 } else {
765 ERR("Unknown connection type : %u", be32toh(msg.type));
766 ret = -1;
767 goto end;
768 }
769
770 reply.major = htobe32(reply.major);
771 reply.minor = htobe32(reply.minor);
772 if (conn->type == RELAY_VIEWER_COMMAND) {
773 /*
774 * Increment outside of htobe64 macro, because the argument can
775 * be used more than once within the macro, and thus the
776 * operation may be undefined.
777 */
778 pthread_mutex_lock(&last_relay_viewer_session_id_lock);
779 last_relay_viewer_session_id++;
780 pthread_mutex_unlock(&last_relay_viewer_session_id_lock);
781 reply.viewer_session_id = htobe64(last_relay_viewer_session_id);
782 }
783
784 health_code_update();
785
786 ret = send_response(conn->sock, &reply, sizeof(reply));
787 if (ret < 0) {
788 goto end;
789 }
790
791 health_code_update();
792
793 DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
794 ret = 0;
795
796 end:
797 return ret;
798 }
799
800 /*
801 * Send the viewer the list of current sessions.
802 * We need to create a copy of the hash table content because otherwise
803 * we cannot assume the number of entries stays the same between getting
804 * the number of HT elements and iteration over the HT.
805 *
806 * Return 0 on success or else a negative value.
807 */
808 static
809 int viewer_list_sessions(struct relay_connection *conn)
810 {
811 int ret = 0;
812 struct lttng_viewer_list_sessions session_list;
813 struct lttng_ht_iter iter;
814 struct relay_session *session;
815 struct lttng_viewer_session *send_session_buf = NULL;
816 uint32_t buf_count = SESSION_BUF_DEFAULT_COUNT;
817 uint32_t count = 0;
818
819 DBG("List sessions received");
820
821 send_session_buf = zmalloc(SESSION_BUF_DEFAULT_COUNT * sizeof(*send_session_buf));
822 if (!send_session_buf) {
823 return -1;
824 }
825
826 rcu_read_lock();
827 cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, session,
828 session_n.node) {
829 struct lttng_viewer_session *send_session;
830
831 health_code_update();
832
833 if (count >= buf_count) {
834 struct lttng_viewer_session *newbuf;
835 uint32_t new_buf_count = buf_count << 1;
836
837 newbuf = realloc(send_session_buf,
838 new_buf_count * sizeof(*send_session_buf));
839 if (!newbuf) {
840 ret = -1;
841 break;
842 }
843 send_session_buf = newbuf;
844 buf_count = new_buf_count;
845 }
846 send_session = &send_session_buf[count];
847 if (lttng_strncpy(send_session->session_name,
848 session->session_name,
849 sizeof(send_session->session_name))) {
850 ret = -1;
851 break;
852 }
853 if (lttng_strncpy(send_session->hostname, session->hostname,
854 sizeof(send_session->hostname))) {
855 ret = -1;
856 break;
857 }
858 send_session->id = htobe64(session->id);
859 send_session->live_timer = htobe32(session->live_timer);
860 if (session->viewer_attached) {
861 send_session->clients = htobe32(1);
862 } else {
863 send_session->clients = htobe32(0);
864 }
865 send_session->streams = htobe32(session->stream_count);
866 count++;
867 }
868 rcu_read_unlock();
869 if (ret < 0) {
870 goto end_free;
871 }
872
873 session_list.sessions_count = htobe32(count);
874
875 health_code_update();
876
877 ret = send_response(conn->sock, &session_list, sizeof(session_list));
878 if (ret < 0) {
879 goto end_free;
880 }
881
882 health_code_update();
883
884 ret = send_response(conn->sock, send_session_buf,
885 count * sizeof(*send_session_buf));
886 if (ret < 0) {
887 goto end_free;
888 }
889 health_code_update();
890
891 ret = 0;
892 end_free:
893 free(send_session_buf);
894 return ret;
895 }
896
897 /*
898 * Send the viewer the list of current streams.
899 */
900 static
901 int viewer_get_new_streams(struct relay_connection *conn)
902 {
903 int ret, send_streams = 0;
904 uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0, nb_total = 0;
905 struct lttng_viewer_new_streams_request request;
906 struct lttng_viewer_new_streams_response response;
907 struct relay_session *session;
908 uint64_t session_id;
909 bool closed = false;
910
911 assert(conn);
912
913 DBG("Get new streams received");
914
915 health_code_update();
916
917 /* Receive the request from the connected client. */
918 ret = recv_request(conn->sock, &request, sizeof(request));
919 if (ret < 0) {
920 goto error;
921 }
922 session_id = be64toh(request.session_id);
923
924 health_code_update();
925
926 memset(&response, 0, sizeof(response));
927
928 session = session_get_by_id(session_id);
929 if (!session) {
930 DBG("Relay session %" PRIu64 " not found", session_id);
931 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
932 goto send_reply;
933 }
934
935 if (!viewer_session_is_attached(conn->viewer_session, session)) {
936 send_streams = 0;
937 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
938 goto send_reply;
939 }
940
941 send_streams = 1;
942 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
943
944 ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, &nb_total, &nb_unsent,
945 &nb_created, &closed);
946 if (ret < 0) {
947 goto end_put_session;
948 }
949 /* Only send back the newly created streams with the unsent ones. */
950 nb_streams = nb_created + nb_unsent;
951 response.streams_count = htobe32(nb_streams);
952
953 /*
954 * If the session is closed, HUP when there are no more streams
955 * with data.
956 */
957 if (closed && nb_total == 0) {
958 send_streams = 0;
959 response.streams_count = 0;
960 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
961 goto send_reply;
962 }
963
964 send_reply:
965 health_code_update();
966 ret = send_response(conn->sock, &response, sizeof(response));
967 if (ret < 0) {
968 goto end_put_session;
969 }
970 health_code_update();
971
972 /*
973 * Unknown or empty session, just return gracefully, the viewer
974 * knows what is happening.
975 */
976 if (!send_streams || !nb_streams) {
977 ret = 0;
978 goto end_put_session;
979 }
980
981 /*
982 * Send stream and *DON'T* ignore the sent flag so every viewer
983 * streams that were not sent from that point will be sent to
984 * the viewer.
985 */
986 ret = send_viewer_streams(conn->sock, session, 0);
987 if (ret < 0) {
988 goto end_put_session;
989 }
990
991 end_put_session:
992 if (session) {
993 session_put(session);
994 }
995 error:
996 return ret;
997 }
998
999 /*
1000 * Send the viewer the list of current sessions.
1001 */
1002 static
1003 int viewer_attach_session(struct relay_connection *conn)
1004 {
1005 int send_streams = 0;
1006 ssize_t ret;
1007 uint32_t nb_streams = 0;
1008 enum lttng_viewer_seek seek_type;
1009 struct lttng_viewer_attach_session_request request;
1010 struct lttng_viewer_attach_session_response response;
1011 struct relay_session *session = NULL;
1012 bool closed = false;
1013
1014 assert(conn);
1015
1016 health_code_update();
1017
1018 /* Receive the request from the connected client. */
1019 ret = recv_request(conn->sock, &request, sizeof(request));
1020 if (ret < 0) {
1021 goto error;
1022 }
1023
1024 health_code_update();
1025
1026 memset(&response, 0, sizeof(response));
1027
1028 if (!conn->viewer_session) {
1029 DBG("Client trying to attach before creating a live viewer session");
1030 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1031 goto send_reply;
1032 }
1033
1034 session = session_get_by_id(be64toh(request.session_id));
1035 if (!session) {
1036 DBG("Relay session %" PRIu64 " not found",
1037 be64toh(request.session_id));
1038 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1039 goto send_reply;
1040 }
1041 DBG("Attach session ID %" PRIu64 " received",
1042 be64toh(request.session_id));
1043
1044 if (session->live_timer == 0) {
1045 DBG("Not live session");
1046 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1047 goto send_reply;
1048 }
1049
1050 send_streams = 1;
1051 ret = viewer_session_attach(conn->viewer_session, session);
1052 if (ret) {
1053 DBG("Already a viewer attached");
1054 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1055 goto send_reply;
1056 }
1057
1058 switch (be32toh(request.seek)) {
1059 case LTTNG_VIEWER_SEEK_BEGINNING:
1060 case LTTNG_VIEWER_SEEK_LAST:
1061 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1062 seek_type = be32toh(request.seek);
1063 break;
1064 default:
1065 ERR("Wrong seek parameter");
1066 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1067 send_streams = 0;
1068 goto send_reply;
1069 }
1070
1071 ret = make_viewer_streams(session, seek_type, &nb_streams, NULL,
1072 NULL, &closed);
1073 if (ret < 0) {
1074 goto end_put_session;
1075 }
1076 response.streams_count = htobe32(nb_streams);
1077
1078 /*
1079 * If the session is closed when the viewer is attaching, it
1080 * means some of the streams may have been concurrently removed,
1081 * so we don't allow the viewer to attach, even if there are
1082 * streams available.
1083 */
1084 if (closed) {
1085 send_streams = 0;
1086 response.streams_count = 0;
1087 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
1088 goto send_reply;
1089 }
1090
1091 send_reply:
1092 health_code_update();
1093 ret = send_response(conn->sock, &response, sizeof(response));
1094 if (ret < 0) {
1095 goto end_put_session;
1096 }
1097 health_code_update();
1098
1099 /*
1100 * Unknown or empty session, just return gracefully, the viewer
1101 * knows what is happening.
1102 */
1103 if (!send_streams || !nb_streams) {
1104 ret = 0;
1105 goto end_put_session;
1106 }
1107
1108 /* Send stream and ignore the sent flag. */
1109 ret = send_viewer_streams(conn->sock, session, 1);
1110 if (ret < 0) {
1111 goto end_put_session;
1112 }
1113
1114 end_put_session:
1115 if (session) {
1116 session_put(session);
1117 }
1118 error:
1119 return ret;
1120 }
1121
1122 /*
1123 * Open the index file if needed for the given vstream.
1124 *
1125 * If an index file is successfully opened, the vstream will set it as its
1126 * current index file.
1127 *
1128 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1129 *
1130 * Called with rstream lock held.
1131 */
1132 static int try_open_index(struct relay_viewer_stream *vstream,
1133 struct relay_stream *rstream)
1134 {
1135 int ret = 0;
1136
1137 if (vstream->index_file) {
1138 goto end;
1139 }
1140
1141 /*
1142 * First time, we open the index file and at least one index is ready.
1143 */
1144 if (rstream->index_received_seqcount == 0) {
1145 ret = -ENOENT;
1146 goto end;
1147 }
1148 vstream->index_file = lttng_index_file_open(vstream->path_name,
1149 vstream->channel_name,
1150 vstream->stream->tracefile_count,
1151 vstream->current_tracefile_id);
1152 if (!vstream->index_file) {
1153 ret = -1;
1154 }
1155
1156 end:
1157 return ret;
1158 }
1159
1160 /*
1161 * Check the status of the index for the given stream. This function
1162 * updates the index structure if needed and can put (close) the vstream
1163 * in the HUP situation.
1164 *
1165 * Return 0 means that we can proceed with the index. A value of 1 means
1166 * that the index has been updated and is ready to be sent to the
1167 * client. A negative value indicates an error that can't be handled.
1168 *
1169 * Called with rstream lock held.
1170 */
1171 static int check_index_status(struct relay_viewer_stream *vstream,
1172 struct relay_stream *rstream, struct ctf_trace *trace,
1173 struct lttng_viewer_index *index)
1174 {
1175 int ret;
1176
1177 if ((trace->session->connection_closed || rstream->closed)
1178 && rstream->index_received_seqcount
1179 == vstream->index_sent_seqcount) {
1180 /*
1181 * Last index sent and session connection or relay
1182 * stream are closed.
1183 */
1184 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1185 goto hup;
1186 } else if (rstream->beacon_ts_end != -1ULL &&
1187 rstream->index_received_seqcount
1188 == vstream->index_sent_seqcount) {
1189 /*
1190 * We've received a synchronization beacon and the last index
1191 * available has been sent, the index for now is inactive.
1192 *
1193 * In this case, we have received a beacon which allows us to
1194 * inform the client of a time interval during which we can
1195 * guarantee that there are no events to read (and never will
1196 * be).
1197 */
1198 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1199 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1200 index->stream_id = htobe64(rstream->ctf_stream_id);
1201 goto index_ready;
1202 } else if (rstream->index_received_seqcount
1203 == vstream->index_sent_seqcount) {
1204 /*
1205 * This checks whether received == sent seqcount. In
1206 * this case, we have not received a beacon. Therefore,
1207 * we can only ask the client to retry later.
1208 */
1209 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1210 goto index_ready;
1211 } else if (!tracefile_array_seq_in_file(rstream->tfa,
1212 vstream->current_tracefile_id,
1213 vstream->index_sent_seqcount)) {
1214 /*
1215 * The next index we want to send cannot be read either
1216 * because we need to perform a rotation, or due to
1217 * the producer having overwritten its trace file.
1218 */
1219 DBG("Viewer stream %" PRIu64 " rotation",
1220 vstream->stream->stream_handle);
1221 ret = viewer_stream_rotate(vstream);
1222 if (ret < 0) {
1223 goto end;
1224 } else if (ret == 1) {
1225 /* EOF across entire stream. */
1226 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1227 goto hup;
1228 }
1229 /*
1230 * If we have been pushed due to overwrite, it
1231 * necessarily means there is data that can be read in
1232 * the stream. If we rotated because we reached the end
1233 * of a tracefile, it means the following tracefile
1234 * needs to contain at least one index, else we would
1235 * have already returned LTTNG_VIEWER_INDEX_RETRY to the
1236 * viewer. The updated index_sent_seqcount needs to
1237 * point to a readable index entry now.
1238 *
1239 * In the case where we "rotate" on a single file, we
1240 * can end up in a case where the requested index is
1241 * still unavailable.
1242 */
1243 if (rstream->tracefile_count == 1 &&
1244 !tracefile_array_seq_in_file(
1245 rstream->tfa,
1246 vstream->current_tracefile_id,
1247 vstream->index_sent_seqcount)) {
1248 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1249 goto index_ready;
1250 }
1251 assert(tracefile_array_seq_in_file(rstream->tfa,
1252 vstream->current_tracefile_id,
1253 vstream->index_sent_seqcount));
1254 }
1255 /* ret == 0 means successful so we continue. */
1256 ret = 0;
1257 end:
1258 return ret;
1259
1260 hup:
1261 viewer_stream_put(vstream);
1262 index_ready:
1263 return 1;
1264 }
1265
1266 /*
1267 * Send the next index for a stream.
1268 *
1269 * Return 0 on success or else a negative value.
1270 */
1271 static
1272 int viewer_get_next_index(struct relay_connection *conn)
1273 {
1274 int ret;
1275 struct lttng_viewer_get_next_index request_index;
1276 struct lttng_viewer_index viewer_index;
1277 struct ctf_packet_index packet_index;
1278 struct relay_viewer_stream *vstream = NULL;
1279 struct relay_stream *rstream = NULL;
1280 struct ctf_trace *ctf_trace = NULL;
1281 struct relay_viewer_stream *metadata_viewer_stream = NULL;
1282
1283 assert(conn);
1284
1285 DBG("Viewer get next index");
1286
1287 memset(&viewer_index, 0, sizeof(viewer_index));
1288 health_code_update();
1289
1290 ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1291 if (ret < 0) {
1292 goto end;
1293 }
1294 health_code_update();
1295
1296 vstream = viewer_stream_get_by_id(be64toh(request_index.stream_id));
1297 if (!vstream) {
1298 DBG("Client requested index of unknown stream id %" PRIu64,
1299 be64toh(request_index.stream_id));
1300 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1301 goto send_reply;
1302 }
1303
1304 /* Use back. ref. Protected by refcounts. */
1305 rstream = vstream->stream;
1306 ctf_trace = rstream->trace;
1307
1308 /* metadata_viewer_stream may be NULL. */
1309 metadata_viewer_stream =
1310 ctf_trace_get_viewer_metadata_stream(ctf_trace);
1311
1312 pthread_mutex_lock(&rstream->lock);
1313
1314 /*
1315 * The viewer should not ask for index on metadata stream.
1316 */
1317 if (rstream->is_metadata) {
1318 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1319 goto send_reply;
1320 }
1321
1322 /* Try to open an index if one is needed for that stream. */
1323 ret = try_open_index(vstream, rstream);
1324 if (ret < 0) {
1325 if (ret == -ENOENT) {
1326 /*
1327 * The index is created only when the first data
1328 * packet arrives, it might not be ready at the
1329 * beginning of the session
1330 */
1331 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1332 } else {
1333 /* Unhandled error. */
1334 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1335 }
1336 goto send_reply;
1337 }
1338
1339 ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1340 if (ret < 0) {
1341 goto error_put;
1342 } else if (ret == 1) {
1343 /*
1344 * We have no index to send and check_index_status has populated
1345 * viewer_index's status.
1346 */
1347 goto send_reply;
1348 }
1349 /* At this point, ret is 0 thus we will be able to read the index. */
1350 assert(!ret);
1351
1352 /*
1353 * vstream->stream_fd may be NULL if it has been closed by
1354 * tracefile rotation, or if we are at the beginning of the
1355 * stream. We open the data stream file here to protect against
1356 * overwrite caused by tracefile rotation (in association with
1357 * unlink performed before overwrite).
1358 */
1359 if (!vstream->stream_fd) {
1360 char fullpath[PATH_MAX];
1361
1362 if (vstream->stream->tracefile_count > 0) {
1363 ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64,
1364 vstream->path_name,
1365 vstream->channel_name,
1366 vstream->current_tracefile_id);
1367 } else {
1368 ret = snprintf(fullpath, PATH_MAX, "%s/%s",
1369 vstream->path_name,
1370 vstream->channel_name);
1371 }
1372 if (ret < 0) {
1373 goto error_put;
1374 }
1375 ret = open(fullpath, O_RDONLY);
1376 if (ret < 0) {
1377 PERROR("Relay opening trace file");
1378 goto error_put;
1379 }
1380 vstream->stream_fd = stream_fd_create(ret);
1381 if (!vstream->stream_fd) {
1382 if (close(ret)) {
1383 PERROR("close");
1384 }
1385 goto error_put;
1386 }
1387 }
1388
1389 ret = check_new_streams(conn);
1390 if (ret < 0) {
1391 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1392 goto send_reply;
1393 } else if (ret == 1) {
1394 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1395 }
1396
1397 ret = lttng_index_file_read(vstream->index_file, &packet_index);
1398 if (ret) {
1399 ERR("Relay error reading index file %d",
1400 vstream->index_file->fd);
1401 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1402 goto send_reply;
1403 } else {
1404 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1405 vstream->index_sent_seqcount++;
1406 }
1407
1408 /*
1409 * Indexes are stored in big endian, no need to switch before sending.
1410 */
1411 DBG("Sending viewer index for stream %" PRIu64 " offset %" PRIu64,
1412 rstream->stream_handle,
1413 be64toh(packet_index.offset));
1414 viewer_index.offset = packet_index.offset;
1415 viewer_index.packet_size = packet_index.packet_size;
1416 viewer_index.content_size = packet_index.content_size;
1417 viewer_index.timestamp_begin = packet_index.timestamp_begin;
1418 viewer_index.timestamp_end = packet_index.timestamp_end;
1419 viewer_index.events_discarded = packet_index.events_discarded;
1420 viewer_index.stream_id = packet_index.stream_id;
1421
1422 send_reply:
1423 if (rstream) {
1424 pthread_mutex_unlock(&rstream->lock);
1425 }
1426
1427 if (metadata_viewer_stream) {
1428 pthread_mutex_lock(&metadata_viewer_stream->stream->lock);
1429 DBG("get next index metadata check: recv %" PRIu64
1430 " sent %" PRIu64,
1431 metadata_viewer_stream->stream->metadata_received,
1432 metadata_viewer_stream->metadata_sent);
1433 if (!metadata_viewer_stream->stream->metadata_received ||
1434 metadata_viewer_stream->stream->metadata_received >
1435 metadata_viewer_stream->metadata_sent) {
1436 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1437 }
1438 pthread_mutex_unlock(&metadata_viewer_stream->stream->lock);
1439 }
1440
1441 viewer_index.flags = htobe32(viewer_index.flags);
1442 health_code_update();
1443
1444 ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1445 if (ret < 0) {
1446 goto end;
1447 }
1448 health_code_update();
1449
1450 if (vstream) {
1451 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1452 vstream->index_sent_seqcount,
1453 vstream->stream->stream_handle);
1454 }
1455 end:
1456 if (metadata_viewer_stream) {
1457 viewer_stream_put(metadata_viewer_stream);
1458 }
1459 if (vstream) {
1460 viewer_stream_put(vstream);
1461 }
1462 return ret;
1463
1464 error_put:
1465 pthread_mutex_unlock(&rstream->lock);
1466 if (metadata_viewer_stream) {
1467 viewer_stream_put(metadata_viewer_stream);
1468 }
1469 viewer_stream_put(vstream);
1470 return ret;
1471 }
1472
1473 /*
1474 * Send the next index for a stream
1475 *
1476 * Return 0 on success or else a negative value.
1477 */
1478 static
1479 int viewer_get_packet(struct relay_connection *conn)
1480 {
1481 int ret, send_data = 0;
1482 char *data = NULL;
1483 uint32_t len = 0;
1484 ssize_t read_len;
1485 struct lttng_viewer_get_packet get_packet_info;
1486 struct lttng_viewer_trace_packet reply;
1487 struct relay_viewer_stream *vstream = NULL;
1488
1489 DBG2("Relay get data packet");
1490
1491 health_code_update();
1492
1493 ret = recv_request(conn->sock, &get_packet_info,
1494 sizeof(get_packet_info));
1495 if (ret < 0) {
1496 goto end;
1497 }
1498 health_code_update();
1499
1500 /* From this point on, the error label can be reached. */
1501 memset(&reply, 0, sizeof(reply));
1502
1503 vstream = viewer_stream_get_by_id(be64toh(get_packet_info.stream_id));
1504 if (!vstream) {
1505 DBG("Client requested packet of unknown stream id %" PRIu64,
1506 be64toh(get_packet_info.stream_id));
1507 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1508 goto send_reply_nolock;
1509 }
1510
1511 pthread_mutex_lock(&vstream->stream->lock);
1512
1513 len = be32toh(get_packet_info.len);
1514 data = zmalloc(len);
1515 if (!data) {
1516 PERROR("relay data zmalloc");
1517 goto error;
1518 }
1519
1520 ret = lseek(vstream->stream_fd->fd, be64toh(get_packet_info.offset),
1521 SEEK_SET);
1522 if (ret < 0) {
1523 PERROR("lseek fd %d to offset %" PRIu64, vstream->stream_fd->fd,
1524 be64toh(get_packet_info.offset));
1525 goto error;
1526 }
1527 read_len = lttng_read(vstream->stream_fd->fd, data, len);
1528 if (read_len < len) {
1529 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1530 vstream->stream_fd->fd,
1531 be64toh(get_packet_info.offset));
1532 goto error;
1533 }
1534 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1535 reply.len = htobe32(len);
1536 send_data = 1;
1537 goto send_reply;
1538
1539 error:
1540 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1541
1542 send_reply:
1543 if (vstream) {
1544 pthread_mutex_unlock(&vstream->stream->lock);
1545 }
1546 send_reply_nolock:
1547 reply.flags = htobe32(reply.flags);
1548
1549 health_code_update();
1550
1551 ret = send_response(conn->sock, &reply, sizeof(reply));
1552 if (ret < 0) {
1553 goto end_free;
1554 }
1555 health_code_update();
1556
1557 if (send_data) {
1558 health_code_update();
1559 ret = send_response(conn->sock, data, len);
1560 if (ret < 0) {
1561 goto end_free;
1562 }
1563 health_code_update();
1564 }
1565
1566 DBG("Sent %u bytes for stream %" PRIu64, len,
1567 be64toh(get_packet_info.stream_id));
1568
1569 end_free:
1570 free(data);
1571 end:
1572 if (vstream) {
1573 viewer_stream_put(vstream);
1574 }
1575 return ret;
1576 }
1577
1578 /*
1579 * Send the session's metadata
1580 *
1581 * Return 0 on success else a negative value.
1582 */
1583 static
1584 int viewer_get_metadata(struct relay_connection *conn)
1585 {
1586 int ret = 0;
1587 ssize_t read_len;
1588 uint64_t len = 0;
1589 char *data = NULL;
1590 struct lttng_viewer_get_metadata request;
1591 struct lttng_viewer_metadata_packet reply;
1592 struct relay_viewer_stream *vstream = NULL;
1593
1594 assert(conn);
1595
1596 DBG("Relay get metadata");
1597
1598 health_code_update();
1599
1600 ret = recv_request(conn->sock, &request, sizeof(request));
1601 if (ret < 0) {
1602 goto end;
1603 }
1604 health_code_update();
1605
1606 memset(&reply, 0, sizeof(reply));
1607
1608 vstream = viewer_stream_get_by_id(be64toh(request.stream_id));
1609 if (!vstream) {
1610 /*
1611 * The metadata stream can be closed by a CLOSE command
1612 * just before we attach. It can also be closed by
1613 * per-pid tracing during tracing. Therefore, it is
1614 * possible that we cannot find this viewer stream.
1615 * Reply back to the client with an error if we cannot
1616 * find it.
1617 */
1618 DBG("Client requested metadata of unknown stream id %" PRIu64,
1619 be64toh(request.stream_id));
1620 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1621 goto send_reply;
1622 }
1623 pthread_mutex_lock(&vstream->stream->lock);
1624 if (!vstream->stream->is_metadata) {
1625 ERR("Invalid metadata stream");
1626 goto error;
1627 }
1628
1629 assert(vstream->metadata_sent <= vstream->stream->metadata_received);
1630
1631 len = vstream->stream->metadata_received - vstream->metadata_sent;
1632 if (len == 0) {
1633 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1634 goto send_reply;
1635 }
1636
1637 /* first time, we open the metadata file */
1638 if (!vstream->stream_fd) {
1639 char fullpath[PATH_MAX];
1640
1641 ret = snprintf(fullpath, PATH_MAX, "%s/%s", vstream->path_name,
1642 vstream->channel_name);
1643 if (ret < 0) {
1644 goto error;
1645 }
1646 ret = open(fullpath, O_RDONLY);
1647 if (ret < 0) {
1648 PERROR("Relay opening metadata file");
1649 goto error;
1650 }
1651 vstream->stream_fd = stream_fd_create(ret);
1652 if (!vstream->stream_fd) {
1653 if (close(ret)) {
1654 PERROR("close");
1655 }
1656 goto error;
1657 }
1658 }
1659
1660 reply.len = htobe64(len);
1661 data = zmalloc(len);
1662 if (!data) {
1663 PERROR("viewer metadata zmalloc");
1664 goto error;
1665 }
1666
1667 read_len = lttng_read(vstream->stream_fd->fd, data, len);
1668 if (read_len < len) {
1669 PERROR("Relay reading metadata file");
1670 goto error;
1671 }
1672 vstream->metadata_sent += read_len;
1673 if (vstream->metadata_sent == vstream->stream->metadata_received
1674 && vstream->stream->closed) {
1675 /* Release ownership for the viewer metadata stream. */
1676 viewer_stream_put(vstream);
1677 }
1678
1679 reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1680
1681 goto send_reply;
1682
1683 error:
1684 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1685
1686 send_reply:
1687 health_code_update();
1688 if (vstream) {
1689 pthread_mutex_unlock(&vstream->stream->lock);
1690 }
1691 ret = send_response(conn->sock, &reply, sizeof(reply));
1692 if (ret < 0) {
1693 goto end_free;
1694 }
1695 health_code_update();
1696
1697 if (len > 0) {
1698 ret = send_response(conn->sock, data, len);
1699 if (ret < 0) {
1700 goto end_free;
1701 }
1702 }
1703
1704 DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1705 be64toh(request.stream_id));
1706
1707 DBG("Metadata sent");
1708
1709 end_free:
1710 free(data);
1711 end:
1712 if (vstream) {
1713 viewer_stream_put(vstream);
1714 }
1715 return ret;
1716 }
1717
1718 /*
1719 * Create a viewer session.
1720 *
1721 * Return 0 on success or else a negative value.
1722 */
1723 static
1724 int viewer_create_session(struct relay_connection *conn)
1725 {
1726 int ret;
1727 struct lttng_viewer_create_session_response resp;
1728
1729 DBG("Viewer create session received");
1730
1731 memset(&resp, 0, sizeof(resp));
1732 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1733 conn->viewer_session = viewer_session_create();
1734 if (!conn->viewer_session) {
1735 ERR("Allocation viewer session");
1736 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1737 goto send_reply;
1738 }
1739
1740 send_reply:
1741 health_code_update();
1742 ret = send_response(conn->sock, &resp, sizeof(resp));
1743 if (ret < 0) {
1744 goto end;
1745 }
1746 health_code_update();
1747 ret = 0;
1748
1749 end:
1750 return ret;
1751 }
1752
1753 /*
1754 * Detach a viewer session.
1755 *
1756 * Return 0 on success or else a negative value.
1757 */
1758 static
1759 int viewer_detach_session(struct relay_connection *conn)
1760 {
1761 int ret;
1762 struct lttng_viewer_detach_session_response response;
1763 struct lttng_viewer_detach_session_request request;
1764 struct relay_session *session = NULL;
1765 uint64_t viewer_session_to_close;
1766
1767 DBG("Viewer detach session received");
1768
1769 assert(conn);
1770
1771 health_code_update();
1772
1773 /* Receive the request from the connected client. */
1774 ret = recv_request(conn->sock, &request, sizeof(request));
1775 if (ret < 0) {
1776 goto end;
1777 }
1778 viewer_session_to_close = be64toh(request.session_id);
1779
1780 if (!conn->viewer_session) {
1781 DBG("Client trying to detach before creating a live viewer session");
1782 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
1783 goto send_reply;
1784 }
1785
1786 health_code_update();
1787
1788 memset(&response, 0, sizeof(response));
1789 DBG("Detaching from session ID %" PRIu64, viewer_session_to_close);
1790
1791 session = session_get_by_id(be64toh(request.session_id));
1792 if (!session) {
1793 DBG("Relay session %" PRIu64 " not found",
1794 be64toh(request.session_id));
1795 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_UNK);
1796 goto send_reply;
1797 }
1798
1799 ret = viewer_session_is_attached(conn->viewer_session, session);
1800 if (ret != 1) {
1801 DBG("Not attached to this session");
1802 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
1803 goto send_reply_put;
1804 }
1805
1806 viewer_session_close_one_session(conn->viewer_session, session);
1807 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_OK);
1808 DBG("Session %" PRIu64 " detached.", viewer_session_to_close);
1809
1810 send_reply_put:
1811 session_put(session);
1812
1813 send_reply:
1814 health_code_update();
1815 ret = send_response(conn->sock, &response, sizeof(response));
1816 if (ret < 0) {
1817 goto end;
1818 }
1819 health_code_update();
1820 ret = 0;
1821
1822 end:
1823 return ret;
1824 }
1825
1826 /*
1827 * live_relay_unknown_command: send -1 if received unknown command
1828 */
1829 static
1830 void live_relay_unknown_command(struct relay_connection *conn)
1831 {
1832 struct lttcomm_relayd_generic_reply reply;
1833
1834 memset(&reply, 0, sizeof(reply));
1835 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1836 (void) send_response(conn->sock, &reply, sizeof(reply));
1837 }
1838
1839 /*
1840 * Process the commands received on the control socket
1841 */
1842 static
1843 int process_control(struct lttng_viewer_cmd *recv_hdr,
1844 struct relay_connection *conn)
1845 {
1846 int ret = 0;
1847 uint32_t msg_value;
1848
1849 msg_value = be32toh(recv_hdr->cmd);
1850
1851 /*
1852 * Make sure we've done the version check before any command other then a
1853 * new client connection.
1854 */
1855 if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1856 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1857 ret = -1;
1858 goto end;
1859 }
1860
1861 switch (msg_value) {
1862 case LTTNG_VIEWER_CONNECT:
1863 ret = viewer_connect(conn);
1864 break;
1865 case LTTNG_VIEWER_LIST_SESSIONS:
1866 ret = viewer_list_sessions(conn);
1867 break;
1868 case LTTNG_VIEWER_ATTACH_SESSION:
1869 ret = viewer_attach_session(conn);
1870 break;
1871 case LTTNG_VIEWER_GET_NEXT_INDEX:
1872 ret = viewer_get_next_index(conn);
1873 break;
1874 case LTTNG_VIEWER_GET_PACKET:
1875 ret = viewer_get_packet(conn);
1876 break;
1877 case LTTNG_VIEWER_GET_METADATA:
1878 ret = viewer_get_metadata(conn);
1879 break;
1880 case LTTNG_VIEWER_GET_NEW_STREAMS:
1881 ret = viewer_get_new_streams(conn);
1882 break;
1883 case LTTNG_VIEWER_CREATE_SESSION:
1884 ret = viewer_create_session(conn);
1885 break;
1886 case LTTNG_VIEWER_DETACH_SESSION:
1887 ret = viewer_detach_session(conn);
1888 break;
1889 default:
1890 ERR("Received unknown viewer command (%u)",
1891 be32toh(recv_hdr->cmd));
1892 live_relay_unknown_command(conn);
1893 ret = -1;
1894 goto end;
1895 }
1896
1897 end:
1898 return ret;
1899 }
1900
1901 static
1902 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1903 {
1904 int ret;
1905
1906 (void) lttng_poll_del(events, pollfd);
1907
1908 ret = close(pollfd);
1909 if (ret < 0) {
1910 ERR("Closing pollfd %d", pollfd);
1911 }
1912 }
1913
1914 /*
1915 * This thread does the actual work
1916 */
1917 static
1918 void *thread_worker(void *data)
1919 {
1920 int ret, err = -1;
1921 uint32_t nb_fd;
1922 struct lttng_poll_event events;
1923 struct lttng_ht *viewer_connections_ht;
1924 struct lttng_ht_iter iter;
1925 struct lttng_viewer_cmd recv_hdr;
1926 struct relay_connection *destroy_conn;
1927
1928 DBG("[thread] Live viewer relay worker started");
1929
1930 rcu_register_thread();
1931
1932 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1933
1934 if (testpoint(relayd_thread_live_worker)) {
1935 goto error_testpoint;
1936 }
1937
1938 /* table of connections indexed on socket */
1939 viewer_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1940 if (!viewer_connections_ht) {
1941 goto viewer_connections_ht_error;
1942 }
1943
1944 ret = create_thread_poll_set(&events, 2);
1945 if (ret < 0) {
1946 goto error_poll_create;
1947 }
1948
1949 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1950 if (ret < 0) {
1951 goto error;
1952 }
1953
1954 restart:
1955 while (1) {
1956 int i;
1957
1958 health_code_update();
1959
1960 /* Infinite blocking call, waiting for transmission */
1961 DBG3("Relayd live viewer worker thread polling...");
1962 health_poll_entry();
1963 ret = lttng_poll_wait(&events, -1);
1964 health_poll_exit();
1965 if (ret < 0) {
1966 /*
1967 * Restart interrupted system call.
1968 */
1969 if (errno == EINTR) {
1970 goto restart;
1971 }
1972 goto error;
1973 }
1974
1975 nb_fd = ret;
1976
1977 /*
1978 * Process control. The control connection is prioritised so we don't
1979 * starve it with high throughput tracing data on the data
1980 * connection.
1981 */
1982 for (i = 0; i < nb_fd; i++) {
1983 /* Fetch once the poll data */
1984 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1985 int pollfd = LTTNG_POLL_GETFD(&events, i);
1986
1987 health_code_update();
1988
1989 if (!revents) {
1990 /* No activity for this FD (poll implementation). */
1991 continue;
1992 }
1993
1994 /* Thread quit pipe has been closed. Killing thread. */
1995 ret = check_thread_quit_pipe(pollfd, revents);
1996 if (ret) {
1997 err = 0;
1998 goto exit;
1999 }
2000
2001 /* Inspect the relay conn pipe for new connection. */
2002 if (pollfd == live_conn_pipe[0]) {
2003 if (revents & LPOLLIN) {
2004 struct relay_connection *conn;
2005
2006 ret = lttng_read(live_conn_pipe[0],
2007 &conn, sizeof(conn));
2008 if (ret < 0) {
2009 goto error;
2010 }
2011 lttng_poll_add(&events, conn->sock->fd,
2012 LPOLLIN | LPOLLRDHUP);
2013 connection_ht_add(viewer_connections_ht, conn);
2014 DBG("Connection socket %d added to poll", conn->sock->fd);
2015 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2016 ERR("Relay live pipe error");
2017 goto error;
2018 } else {
2019 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2020 goto error;
2021 }
2022 } else {
2023 /* Connection activity. */
2024 struct relay_connection *conn;
2025
2026 conn = connection_get_by_sock(viewer_connections_ht, pollfd);
2027 if (!conn) {
2028 continue;
2029 }
2030
2031 if (revents & LPOLLIN) {
2032 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
2033 sizeof(recv_hdr), 0);
2034 if (ret <= 0) {
2035 /* Connection closed. */
2036 cleanup_connection_pollfd(&events, pollfd);
2037 /* Put "create" ownership reference. */
2038 connection_put(conn);
2039 DBG("Viewer control conn closed with %d", pollfd);
2040 } else {
2041 ret = process_control(&recv_hdr, conn);
2042 if (ret < 0) {
2043 /* Clear the session on error. */
2044 cleanup_connection_pollfd(&events, pollfd);
2045 /* Put "create" ownership reference. */
2046 connection_put(conn);
2047 DBG("Viewer connection closed with %d", pollfd);
2048 }
2049 }
2050 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2051 cleanup_connection_pollfd(&events, pollfd);
2052 /* Put "create" ownership reference. */
2053 connection_put(conn);
2054 } else {
2055 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2056 connection_put(conn);
2057 goto error;
2058 }
2059 /* Put local "get_by_sock" reference. */
2060 connection_put(conn);
2061 }
2062 }
2063 }
2064
2065 exit:
2066 error:
2067 lttng_poll_clean(&events);
2068
2069 /* Cleanup reamaining connection object. */
2070 rcu_read_lock();
2071 cds_lfht_for_each_entry(viewer_connections_ht->ht, &iter.iter,
2072 destroy_conn,
2073 sock_n.node) {
2074 health_code_update();
2075 connection_put(destroy_conn);
2076 }
2077 rcu_read_unlock();
2078 error_poll_create:
2079 lttng_ht_destroy(viewer_connections_ht);
2080 viewer_connections_ht_error:
2081 /* Close relay conn pipes */
2082 utils_close_pipe(live_conn_pipe);
2083 if (err) {
2084 DBG("Viewer worker thread exited with error");
2085 }
2086 DBG("Viewer worker thread cleanup complete");
2087 error_testpoint:
2088 if (err) {
2089 health_error();
2090 ERR("Health error occurred in %s", __func__);
2091 }
2092 health_unregister(health_relayd);
2093 if (lttng_relay_stop_threads()) {
2094 ERR("Error stopping threads");
2095 }
2096 rcu_unregister_thread();
2097 return NULL;
2098 }
2099
2100 /*
2101 * Create the relay command pipe to wake thread_manage_apps.
2102 * Closed in cleanup().
2103 */
2104 static int create_conn_pipe(void)
2105 {
2106 return utils_create_pipe_cloexec(live_conn_pipe);
2107 }
2108
2109 int relayd_live_join(void)
2110 {
2111 int ret, retval = 0;
2112 void *status;
2113
2114 ret = pthread_join(live_listener_thread, &status);
2115 if (ret) {
2116 errno = ret;
2117 PERROR("pthread_join live listener");
2118 retval = -1;
2119 }
2120
2121 ret = pthread_join(live_worker_thread, &status);
2122 if (ret) {
2123 errno = ret;
2124 PERROR("pthread_join live worker");
2125 retval = -1;
2126 }
2127
2128 ret = pthread_join(live_dispatcher_thread, &status);
2129 if (ret) {
2130 errno = ret;
2131 PERROR("pthread_join live dispatcher");
2132 retval = -1;
2133 }
2134
2135 cleanup_relayd_live();
2136
2137 return retval;
2138 }
2139
2140 /*
2141 * main
2142 */
2143 int relayd_live_create(struct lttng_uri *uri)
2144 {
2145 int ret = 0, retval = 0;
2146 void *status;
2147 int is_root;
2148
2149 if (!uri) {
2150 retval = -1;
2151 goto exit_init_data;
2152 }
2153 live_uri = uri;
2154
2155 /* Check if daemon is UID = 0 */
2156 is_root = !getuid();
2157
2158 if (!is_root) {
2159 if (live_uri->port < 1024) {
2160 ERR("Need to be root to use ports < 1024");
2161 retval = -1;
2162 goto exit_init_data;
2163 }
2164 }
2165
2166 /* Setup the thread apps communication pipe. */
2167 if (create_conn_pipe()) {
2168 retval = -1;
2169 goto exit_init_data;
2170 }
2171
2172 /* Init relay command queue. */
2173 cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2174
2175 /* Set up max poll set size */
2176 if (lttng_poll_set_max_size()) {
2177 retval = -1;
2178 goto exit_init_data;
2179 }
2180
2181 /* Setup the dispatcher thread */
2182 ret = pthread_create(&live_dispatcher_thread, default_pthread_attr(),
2183 thread_dispatcher, (void *) NULL);
2184 if (ret) {
2185 errno = ret;
2186 PERROR("pthread_create viewer dispatcher");
2187 retval = -1;
2188 goto exit_dispatcher_thread;
2189 }
2190
2191 /* Setup the worker thread */
2192 ret = pthread_create(&live_worker_thread, default_pthread_attr(),
2193 thread_worker, NULL);
2194 if (ret) {
2195 errno = ret;
2196 PERROR("pthread_create viewer worker");
2197 retval = -1;
2198 goto exit_worker_thread;
2199 }
2200
2201 /* Setup the listener thread */
2202 ret = pthread_create(&live_listener_thread, default_pthread_attr(),
2203 thread_listener, (void *) NULL);
2204 if (ret) {
2205 errno = ret;
2206 PERROR("pthread_create viewer listener");
2207 retval = -1;
2208 goto exit_listener_thread;
2209 }
2210
2211 /*
2212 * All OK, started all threads.
2213 */
2214 return retval;
2215
2216 /*
2217 * Join on the live_listener_thread should anything be added after
2218 * the live_listener thread's creation.
2219 */
2220
2221 exit_listener_thread:
2222
2223 ret = pthread_join(live_worker_thread, &status);
2224 if (ret) {
2225 errno = ret;
2226 PERROR("pthread_join live worker");
2227 retval = -1;
2228 }
2229 exit_worker_thread:
2230
2231 ret = pthread_join(live_dispatcher_thread, &status);
2232 if (ret) {
2233 errno = ret;
2234 PERROR("pthread_join live dispatcher");
2235 retval = -1;
2236 }
2237 exit_dispatcher_thread:
2238
2239 exit_init_data:
2240 cleanup_relayd_live();
2241
2242 return retval;
2243 }
This page took 0.138882 seconds and 4 git commands to generate.