relayd: Implement custom EfficiOS session clear
[deliverable/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 lttng-live", sock->fd);
456
457 ret = sock->ops->bind(sock);
458 if (ret < 0) {
459 PERROR("Failed to bind lttng-live socket");
460 goto error;
461 }
462
463 ret = sock->ops->listen(sock, -1);
464 if (ret < 0) {
465 goto error;
466
467 }
468
469 return sock;
470
471 error:
472 if (sock) {
473 lttcomm_destroy_sock(sock);
474 }
475 return NULL;
476 }
477
478 /*
479 * This thread manages the listening for new connections on the network
480 */
481 static
482 void *thread_listener(void *data)
483 {
484 int i, ret, pollfd, err = -1;
485 uint32_t revents, nb_fd;
486 struct lttng_poll_event events;
487 struct lttcomm_sock *live_control_sock;
488
489 DBG("[thread] Relay live listener started");
490
491 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
492
493 health_code_update();
494
495 live_control_sock = init_socket(live_uri);
496 if (!live_control_sock) {
497 goto error_sock_control;
498 }
499
500 /* Pass 2 as size here for the thread quit pipe and control sockets. */
501 ret = create_thread_poll_set(&events, 2);
502 if (ret < 0) {
503 goto error_create_poll;
504 }
505
506 /* Add the control socket */
507 ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
508 if (ret < 0) {
509 goto error_poll_add;
510 }
511
512 lttng_relay_notify_ready();
513
514 if (testpoint(relayd_thread_live_listener)) {
515 goto error_testpoint;
516 }
517
518 while (1) {
519 health_code_update();
520
521 DBG("Listener accepting live viewers connections");
522
523 restart:
524 health_poll_entry();
525 ret = lttng_poll_wait(&events, -1);
526 health_poll_exit();
527 if (ret < 0) {
528 /*
529 * Restart interrupted system call.
530 */
531 if (errno == EINTR) {
532 goto restart;
533 }
534 goto error;
535 }
536 nb_fd = ret;
537
538 DBG("Relay new viewer connection received");
539 for (i = 0; i < nb_fd; i++) {
540 health_code_update();
541
542 /* Fetch once the poll data */
543 revents = LTTNG_POLL_GETEV(&events, i);
544 pollfd = LTTNG_POLL_GETFD(&events, i);
545
546 if (!revents) {
547 /* No activity for this FD (poll implementation). */
548 continue;
549 }
550
551 /* Thread quit pipe has been closed. Killing thread. */
552 ret = check_thread_quit_pipe(pollfd, revents);
553 if (ret) {
554 err = 0;
555 goto exit;
556 }
557
558 if (revents & LPOLLIN) {
559 /*
560 * A new connection is requested, therefore a
561 * viewer connection is allocated in this
562 * thread, enqueued to a global queue and
563 * dequeued (and freed) in the worker thread.
564 */
565 int val = 1;
566 struct relay_connection *new_conn;
567 struct lttcomm_sock *newsock;
568
569 newsock = live_control_sock->ops->accept(live_control_sock);
570 if (!newsock) {
571 PERROR("accepting control sock");
572 goto error;
573 }
574 DBG("Relay viewer connection accepted socket %d", newsock->fd);
575
576 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
577 sizeof(val));
578 if (ret < 0) {
579 PERROR("setsockopt inet");
580 lttcomm_destroy_sock(newsock);
581 goto error;
582 }
583 new_conn = connection_create(newsock, RELAY_CONNECTION_UNKNOWN);
584 if (!new_conn) {
585 lttcomm_destroy_sock(newsock);
586 goto error;
587 }
588 /* Ownership assumed by the connection. */
589 newsock = NULL;
590
591 /* Enqueue request for the dispatcher thread. */
592 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
593 &new_conn->qnode);
594
595 /*
596 * Wake the dispatch queue futex.
597 * Implicit memory barrier with the
598 * exchange in cds_wfcq_enqueue.
599 */
600 futex_nto1_wake(&viewer_conn_queue.futex);
601 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
602 ERR("socket poll error");
603 goto error;
604 } else {
605 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
606 goto error;
607 }
608 }
609 }
610
611 exit:
612 error:
613 error_poll_add:
614 error_testpoint:
615 lttng_poll_clean(&events);
616 error_create_poll:
617 if (live_control_sock->fd >= 0) {
618 ret = live_control_sock->ops->close(live_control_sock);
619 if (ret) {
620 PERROR("close");
621 }
622 }
623 lttcomm_destroy_sock(live_control_sock);
624 error_sock_control:
625 if (err) {
626 health_error();
627 DBG("Live viewer listener thread exited with error");
628 }
629 health_unregister(health_relayd);
630 DBG("Live viewer listener thread cleanup complete");
631 if (lttng_relay_stop_threads()) {
632 ERR("Error stopping threads");
633 }
634 return NULL;
635 }
636
637 /*
638 * This thread manages the dispatching of the requests to worker threads
639 */
640 static
641 void *thread_dispatcher(void *data)
642 {
643 int err = -1;
644 ssize_t ret;
645 struct cds_wfcq_node *node;
646 struct relay_connection *conn = NULL;
647
648 DBG("[thread] Live viewer relay dispatcher started");
649
650 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
651
652 if (testpoint(relayd_thread_live_dispatcher)) {
653 goto error_testpoint;
654 }
655
656 health_code_update();
657
658 for (;;) {
659 health_code_update();
660
661 /* Atomically prepare the queue futex */
662 futex_nto1_prepare(&viewer_conn_queue.futex);
663
664 if (CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
665 break;
666 }
667
668 do {
669 health_code_update();
670
671 /* Dequeue commands */
672 node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
673 &viewer_conn_queue.tail);
674 if (node == NULL) {
675 DBG("Woken up but nothing in the live-viewer "
676 "relay command queue");
677 /* Continue thread execution */
678 break;
679 }
680 conn = caa_container_of(node, struct relay_connection, qnode);
681 DBG("Dispatching viewer request waiting on sock %d",
682 conn->sock->fd);
683
684 /*
685 * Inform worker thread of the new request. This
686 * call is blocking so we can be assured that
687 * the data will be read at some point in time
688 * or wait to the end of the world :)
689 */
690 ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
691 if (ret < 0) {
692 PERROR("write conn pipe");
693 connection_put(conn);
694 goto error;
695 }
696 } while (node != NULL);
697
698 /* Futex wait on queue. Blocking call on futex() */
699 health_poll_entry();
700 futex_nto1_wait(&viewer_conn_queue.futex);
701 health_poll_exit();
702 }
703
704 /* Normal exit, no error */
705 err = 0;
706
707 error:
708 error_testpoint:
709 if (err) {
710 health_error();
711 ERR("Health error occurred in %s", __func__);
712 }
713 health_unregister(health_relayd);
714 DBG("Live viewer dispatch thread dying");
715 if (lttng_relay_stop_threads()) {
716 ERR("Error stopping threads");
717 }
718 return NULL;
719 }
720
721 /*
722 * Establish connection with the viewer and check the versions.
723 *
724 * Return 0 on success or else negative value.
725 */
726 static
727 int viewer_connect(struct relay_connection *conn)
728 {
729 int ret;
730 struct lttng_viewer_connect reply, msg;
731
732 conn->version_check_done = 1;
733
734 health_code_update();
735
736 DBG("Viewer is establishing a connection to the relayd.");
737
738 ret = recv_request(conn->sock, &msg, sizeof(msg));
739 if (ret < 0) {
740 goto end;
741 }
742
743 health_code_update();
744
745 memset(&reply, 0, sizeof(reply));
746 reply.major = RELAYD_VERSION_COMM_MAJOR;
747 reply.minor = RELAYD_VERSION_COMM_MINOR;
748
749 /* Major versions must be the same */
750 if (reply.major != be32toh(msg.major)) {
751 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
752 reply.major, be32toh(msg.major));
753 ret = -1;
754 goto end;
755 }
756
757 conn->major = reply.major;
758 /* We adapt to the lowest compatible version */
759 if (reply.minor <= be32toh(msg.minor)) {
760 conn->minor = reply.minor;
761 } else {
762 conn->minor = be32toh(msg.minor);
763 }
764
765 if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
766 conn->type = RELAY_VIEWER_COMMAND;
767 } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
768 conn->type = RELAY_VIEWER_NOTIFICATION;
769 } else {
770 ERR("Unknown connection type : %u", be32toh(msg.type));
771 ret = -1;
772 goto end;
773 }
774
775 reply.major = htobe32(reply.major);
776 reply.minor = htobe32(reply.minor);
777 if (conn->type == RELAY_VIEWER_COMMAND) {
778 /*
779 * Increment outside of htobe64 macro, because the argument can
780 * be used more than once within the macro, and thus the
781 * operation may be undefined.
782 */
783 pthread_mutex_lock(&last_relay_viewer_session_id_lock);
784 last_relay_viewer_session_id++;
785 pthread_mutex_unlock(&last_relay_viewer_session_id_lock);
786 reply.viewer_session_id = htobe64(last_relay_viewer_session_id);
787 }
788
789 health_code_update();
790
791 ret = send_response(conn->sock, &reply, sizeof(reply));
792 if (ret < 0) {
793 goto end;
794 }
795
796 health_code_update();
797
798 DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
799 ret = 0;
800
801 end:
802 return ret;
803 }
804
805 /*
806 * Send the viewer the list of current sessions.
807 * We need to create a copy of the hash table content because otherwise
808 * we cannot assume the number of entries stays the same between getting
809 * the number of HT elements and iteration over the HT.
810 *
811 * Return 0 on success or else a negative value.
812 */
813 static
814 int viewer_list_sessions(struct relay_connection *conn)
815 {
816 int ret = 0;
817 struct lttng_viewer_list_sessions session_list;
818 struct lttng_ht_iter iter;
819 struct relay_session *session;
820 struct lttng_viewer_session *send_session_buf = NULL;
821 uint32_t buf_count = SESSION_BUF_DEFAULT_COUNT;
822 uint32_t count = 0;
823
824 DBG("List sessions received");
825
826 send_session_buf = zmalloc(SESSION_BUF_DEFAULT_COUNT * sizeof(*send_session_buf));
827 if (!send_session_buf) {
828 return -1;
829 }
830
831 rcu_read_lock();
832 cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, session,
833 session_n.node) {
834 struct lttng_viewer_session *send_session;
835
836 health_code_update();
837
838 if (session->connection_closed) {
839 /* Skip closed session */
840 continue;
841 }
842
843 if (count >= buf_count) {
844 struct lttng_viewer_session *newbuf;
845 uint32_t new_buf_count = buf_count << 1;
846
847 newbuf = realloc(send_session_buf,
848 new_buf_count * sizeof(*send_session_buf));
849 if (!newbuf) {
850 ret = -1;
851 break;
852 }
853 send_session_buf = newbuf;
854 buf_count = new_buf_count;
855 }
856 send_session = &send_session_buf[count];
857 if (lttng_strncpy(send_session->session_name,
858 session->session_name,
859 sizeof(send_session->session_name))) {
860 ret = -1;
861 break;
862 }
863 if (lttng_strncpy(send_session->hostname, session->hostname,
864 sizeof(send_session->hostname))) {
865 ret = -1;
866 break;
867 }
868 send_session->id = htobe64(session->id);
869 send_session->live_timer = htobe32(session->live_timer);
870 if (session->viewer_attached) {
871 send_session->clients = htobe32(1);
872 } else {
873 send_session->clients = htobe32(0);
874 }
875 send_session->streams = htobe32(session->stream_count);
876 count++;
877 }
878 rcu_read_unlock();
879 if (ret < 0) {
880 goto end_free;
881 }
882
883 session_list.sessions_count = htobe32(count);
884
885 health_code_update();
886
887 ret = send_response(conn->sock, &session_list, sizeof(session_list));
888 if (ret < 0) {
889 goto end_free;
890 }
891
892 health_code_update();
893
894 ret = send_response(conn->sock, send_session_buf,
895 count * sizeof(*send_session_buf));
896 if (ret < 0) {
897 goto end_free;
898 }
899 health_code_update();
900
901 ret = 0;
902 end_free:
903 free(send_session_buf);
904 return ret;
905 }
906
907 /*
908 * Send the viewer the list of current streams.
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, nb_total = 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 bool closed = false;
920
921 assert(conn);
922
923 DBG("Get new streams received");
924
925 health_code_update();
926
927 /* Receive the request from the connected client. */
928 ret = recv_request(conn->sock, &request, sizeof(request));
929 if (ret < 0) {
930 goto error;
931 }
932 session_id = be64toh(request.session_id);
933
934 health_code_update();
935
936 memset(&response, 0, sizeof(response));
937
938 session = session_get_by_id(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 (!viewer_session_is_attached(conn->viewer_session, session)) {
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, &nb_total, &nb_unsent,
955 &nb_created, &closed);
956 if (ret < 0) {
957 goto end_put_session;
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, HUP when there are no more streams
965 * with data.
966 */
967 if (closed && nb_total == 0) {
968 send_streams = 0;
969 response.streams_count = 0;
970 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
971 goto send_reply;
972 }
973
974 send_reply:
975 health_code_update();
976 ret = send_response(conn->sock, &response, sizeof(response));
977 if (ret < 0) {
978 goto end_put_session;
979 }
980 health_code_update();
981
982 /*
983 * Unknown or empty session, just return gracefully, the viewer
984 * knows what is happening.
985 */
986 if (!send_streams || !nb_streams) {
987 ret = 0;
988 goto end_put_session;
989 }
990
991 /*
992 * Send stream and *DON'T* ignore the sent flag so every viewer
993 * streams that were not sent from that point will be sent to
994 * the viewer.
995 */
996 ret = send_viewer_streams(conn->sock, session, 0);
997 if (ret < 0) {
998 goto end_put_session;
999 }
1000
1001 end_put_session:
1002 if (session) {
1003 session_put(session);
1004 }
1005 error:
1006 return ret;
1007 }
1008
1009 /*
1010 * Send the viewer the list of current sessions.
1011 */
1012 static
1013 int viewer_attach_session(struct relay_connection *conn)
1014 {
1015 int send_streams = 0;
1016 ssize_t ret;
1017 uint32_t nb_streams = 0;
1018 enum lttng_viewer_seek seek_type;
1019 struct lttng_viewer_attach_session_request request;
1020 struct lttng_viewer_attach_session_response response;
1021 struct relay_session *session = NULL;
1022 bool closed = false;
1023
1024 assert(conn);
1025
1026 health_code_update();
1027
1028 /* Receive the request from the connected client. */
1029 ret = recv_request(conn->sock, &request, sizeof(request));
1030 if (ret < 0) {
1031 goto error;
1032 }
1033
1034 health_code_update();
1035
1036 memset(&response, 0, sizeof(response));
1037
1038 if (!conn->viewer_session) {
1039 DBG("Client trying to attach before creating a live viewer session");
1040 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1041 goto send_reply;
1042 }
1043
1044 session = session_get_by_id(be64toh(request.session_id));
1045 if (!session) {
1046 DBG("Relay session %" PRIu64 " not found",
1047 (uint64_t) be64toh(request.session_id));
1048 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1049 goto send_reply;
1050 }
1051 DBG("Attach session ID %" PRIu64 " received",
1052 (uint64_t) be64toh(request.session_id));
1053
1054 if (session->live_timer == 0) {
1055 DBG("Not live session");
1056 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1057 goto send_reply;
1058 }
1059
1060 send_streams = 1;
1061 ret = viewer_session_attach(conn->viewer_session, session);
1062 if (ret) {
1063 DBG("Already a viewer attached");
1064 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1065 goto send_reply;
1066 }
1067
1068 switch (be32toh(request.seek)) {
1069 case LTTNG_VIEWER_SEEK_BEGINNING:
1070 case LTTNG_VIEWER_SEEK_LAST:
1071 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1072 seek_type = be32toh(request.seek);
1073 break;
1074 default:
1075 ERR("Wrong seek parameter");
1076 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1077 send_streams = 0;
1078 goto send_reply;
1079 }
1080
1081 ret = make_viewer_streams(session, seek_type, &nb_streams, NULL,
1082 NULL, &closed);
1083 if (ret < 0) {
1084 goto end_put_session;
1085 }
1086 response.streams_count = htobe32(nb_streams);
1087
1088 /*
1089 * If the session is closed when the viewer is attaching, it
1090 * means some of the streams may have been concurrently removed,
1091 * so we don't allow the viewer to attach, even if there are
1092 * streams available.
1093 */
1094 if (closed) {
1095 send_streams = 0;
1096 response.streams_count = 0;
1097 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1098 goto send_reply;
1099 }
1100
1101 send_reply:
1102 health_code_update();
1103 ret = send_response(conn->sock, &response, sizeof(response));
1104 if (ret < 0) {
1105 goto end_put_session;
1106 }
1107 health_code_update();
1108
1109 /*
1110 * Unknown or empty session, just return gracefully, the viewer
1111 * knows what is happening.
1112 */
1113 if (!send_streams || !nb_streams) {
1114 ret = 0;
1115 goto end_put_session;
1116 }
1117
1118 /* Send stream and ignore the sent flag. */
1119 ret = send_viewer_streams(conn->sock, session, 1);
1120 if (ret < 0) {
1121 goto end_put_session;
1122 }
1123
1124 end_put_session:
1125 if (session) {
1126 session_put(session);
1127 }
1128 error:
1129 return ret;
1130 }
1131
1132 /*
1133 * Open the index file if needed for the given vstream.
1134 *
1135 * If an index file is successfully opened, the vstream will set it as its
1136 * current index file.
1137 *
1138 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1139 *
1140 * Called with rstream lock held.
1141 */
1142 static int try_open_index(struct relay_viewer_stream *vstream,
1143 struct relay_stream *rstream)
1144 {
1145 int ret = 0;
1146
1147 if (vstream->index_file) {
1148 goto end;
1149 }
1150
1151 /*
1152 * Deal with streams that have not received any index so far (or after clear).
1153 */
1154 if (rstream->index_received_seqcount <= rstream->clear_position_index_seqcount) {
1155 ret = -ENOENT;
1156 goto end;
1157 }
1158 vstream->index_file = lttng_index_file_open(vstream->path_name,
1159 vstream->channel_name,
1160 vstream->stream->tracefile_count,
1161 vstream->current_tracefile_id);
1162 if (!vstream->index_file) {
1163 ret = -1;
1164 }
1165
1166 end:
1167 return ret;
1168 }
1169
1170 /*
1171 * Check the status of the index for the given stream. This function
1172 * updates the index structure if needed and can put (close) the vstream
1173 * in the HUP situation.
1174 *
1175 * Return 0 means that we can proceed with the index. A value of 1 means
1176 * that the index has been updated and is ready to be sent to the
1177 * client. A negative value indicates an error that can't be handled.
1178 *
1179 * Called with rstream lock held.
1180 */
1181 static int check_index_status(struct relay_viewer_stream *vstream,
1182 struct relay_stream *rstream, struct ctf_trace *trace,
1183 struct lttng_viewer_index *index)
1184 {
1185 int ret;
1186
1187 DBG("check status: recv %" PRIu64 " sent %" PRIu64 " clear index %" PRIu64 " clear data %" PRIu64 " for stream %" PRIu64,
1188 rstream->index_received_seqcount,
1189 vstream->index_sent_seqcount,
1190 rstream->clear_position_index_seqcount,
1191 rstream->clear_position_data_seqcount, vstream->stream->stream_handle);
1192 if ((trace->session->connection_closed || rstream->closed)
1193 && rstream->index_received_seqcount
1194 == vstream->index_sent_seqcount) {
1195 /*
1196 * Last index sent and session connection or relay
1197 * stream are closed.
1198 */
1199 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1200 goto hup;
1201 } else if (rstream->beacon_ts_end != -1ULL &&
1202 rstream->index_received_seqcount
1203 <= vstream->index_sent_seqcount) {
1204 /*
1205 * We've received a synchronization beacon and the last index
1206 * available has been sent, the index for now is inactive.
1207 *
1208 * In this case, we have received a beacon which allows us to
1209 * inform the client of a time interval during which we can
1210 * guarantee that there are no events to read (and never will
1211 * be).
1212 *
1213 * The sent seqcount can grow higher than receive seqcount on clear.
1214 */
1215 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1216 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1217 index->stream_id = htobe64(rstream->ctf_stream_id);
1218 DBG("INACTIVE: with beacon, r v recv eq for stream %" PRIu64, vstream->stream->stream_handle);
1219 goto index_ready;
1220 } else if (rstream->index_received_seqcount
1221 <= vstream->index_sent_seqcount) {
1222 /*
1223 * This checks whether received <= sent seqcount. In
1224 * this case, we have not received a beacon. Therefore,
1225 * we can only ask the client to retry later.
1226 *
1227 * The sent seqcount can grow higher than receive seqcount on clear.
1228 */
1229 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1230 DBG("RETRY: r v recv leq for stream %" PRIu64, vstream->stream->stream_handle);
1231 goto index_ready;
1232 } else if (rstream->index_received_seqcount <= rstream->clear_position_index_seqcount ||
1233 rstream->index_received_seqcount <= rstream->clear_position_data_seqcount) {
1234 /*
1235 * Send "retry" reply if a clear operation is in progress.
1236 */
1237 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1238 DBG("RETRY: r <= clear pos for stream %" PRIu64, vstream->stream->stream_handle);
1239 goto index_ready;
1240 } else if (!tracefile_array_seq_in_file(rstream->tfa,
1241 vstream->current_tracefile_id,
1242 vstream->index_sent_seqcount)) {
1243 /*
1244 * The next index we want to send cannot be read either
1245 * because we need to perform a rotation, or due to
1246 * the producer having overwritten its trace file.
1247 */
1248 DBG("Viewer stream %" PRIu64 " rotation",
1249 vstream->stream->stream_handle);
1250 ret = viewer_stream_rotate(vstream);
1251 if (ret < 0) {
1252 goto end;
1253 } else if (ret == 1) {
1254 /* EOF across entire stream. */
1255 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1256 goto hup;
1257 }
1258 /*
1259 * If we have been pushed due to overwrite, it
1260 * necessarily means there is data that can be read in
1261 * the stream. If we rotated because we reached the end
1262 * of a tracefile, it means the following tracefile
1263 * needs to contain at least one index, else we would
1264 * have already returned LTTNG_VIEWER_INDEX_RETRY to the
1265 * viewer. The updated index_sent_seqcount needs to
1266 * point to a readable index entry now.
1267 *
1268 * In the case where we "rotate" on a single file, we
1269 * can end up in a case where the requested index is
1270 * still unavailable.
1271 */
1272 if (rstream->tracefile_count == 1 &&
1273 !tracefile_array_seq_in_file(
1274 rstream->tfa,
1275 vstream->current_tracefile_id,
1276 vstream->index_sent_seqcount)) {
1277 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1278 DBG("RETRY: !tfs seq in file for stream %" PRIu64, vstream->stream->stream_handle);
1279 goto index_ready;
1280 }
1281 assert(tracefile_array_seq_in_file(rstream->tfa,
1282 vstream->current_tracefile_id,
1283 vstream->index_sent_seqcount));
1284 }
1285 /* ret == 0 means successful so we continue. */
1286 ret = 0;
1287 end:
1288 return ret;
1289
1290 hup:
1291 viewer_stream_put(vstream);
1292 index_ready:
1293 return 1;
1294 }
1295
1296 /*
1297 * Send the next index for a stream.
1298 *
1299 * Return 0 on success or else a negative value.
1300 */
1301 static
1302 int viewer_get_next_index(struct relay_connection *conn)
1303 {
1304 int ret;
1305 struct lttng_viewer_get_next_index request_index;
1306 struct lttng_viewer_index viewer_index;
1307 struct ctf_packet_index packet_index;
1308 struct relay_viewer_stream *vstream = NULL;
1309 struct relay_stream *rstream = NULL;
1310 struct ctf_trace *ctf_trace = NULL;
1311 struct relay_viewer_stream *metadata_viewer_stream = NULL;
1312
1313 assert(conn);
1314
1315 DBG("Viewer get next index");
1316
1317 memset(&viewer_index, 0, sizeof(viewer_index));
1318 health_code_update();
1319
1320 ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1321 if (ret < 0) {
1322 goto end;
1323 }
1324 health_code_update();
1325
1326 vstream = viewer_stream_get_by_id(be64toh(request_index.stream_id));
1327 if (!vstream) {
1328 DBG("Client requested index of unknown stream id %" PRIu64,
1329 (uint64_t) be64toh(request_index.stream_id));
1330 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1331 goto send_reply;
1332 }
1333
1334 /* Use back. ref. Protected by refcounts. */
1335 rstream = vstream->stream;
1336 ctf_trace = rstream->trace;
1337
1338 /* metadata_viewer_stream may be NULL. */
1339 metadata_viewer_stream =
1340 ctf_trace_get_viewer_metadata_stream(ctf_trace);
1341
1342 pthread_mutex_lock(&rstream->lock);
1343
1344 /*
1345 * The viewer should not ask for index on metadata stream.
1346 */
1347 if (rstream->is_metadata) {
1348 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1349 goto send_reply;
1350 }
1351
1352 /*
1353 * In case the stream has been cleared, we need to push the viewer
1354 * stream index sent seqcount forward. Note that this can temporarily
1355 * bring the index_sent position beyond the index received position.
1356 */
1357 if (rstream->clear_position_index_seqcount >= vstream->index_sent_seqcount) {
1358 vstream->index_sent_seqcount = rstream->clear_position_index_seqcount;
1359 /*
1360 * Close the currently open index and data files to ensure we
1361 * sync up with the receive side.
1362 */
1363 if (vstream->index_file) {
1364 lttng_index_file_put(vstream->index_file);
1365 vstream->index_file = NULL;
1366 }
1367 if (vstream->stream_fd) {
1368 stream_fd_put(vstream->stream_fd);
1369 vstream->stream_fd = NULL;
1370 }
1371 /*
1372 * In tracefile rotation, we reset the current tracefile to 0.
1373 */
1374 vstream->current_tracefile_id = 0;
1375 }
1376
1377 /* Try to open an index if one is needed for that stream. */
1378 ret = try_open_index(vstream, rstream);
1379 if (ret < 0) {
1380 if (ret == -ENOENT) {
1381 /*
1382 * The index is created only when the first data
1383 * packet arrives, it might not be ready at the
1384 * beginning of the session. Let check_index_status
1385 * deal with inactivity beacons.
1386 */
1387 goto check_status;
1388 } else {
1389 /* Unhandled error. */
1390 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1391 }
1392 goto send_reply;
1393 }
1394
1395 check_status:
1396 ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1397 if (ret < 0) {
1398 goto error_put;
1399 } else if (ret == 1) {
1400 /*
1401 * We have no index to send and check_index_status has populated
1402 * viewer_index's status.
1403 */
1404 goto send_reply;
1405 }
1406 /* At this point, ret is 0 thus we will be able to read the index. */
1407 assert(!ret);
1408
1409 /*
1410 * vstream->stream_fd may be NULL if it has been closed by
1411 * tracefile rotation, or if we are at the beginning of the
1412 * stream. We open the data stream file here to protect against
1413 * overwrite caused by tracefile rotation (in association with
1414 * unlink performed before overwrite).
1415 */
1416 if (!vstream->stream_fd) {
1417 char fullpath[PATH_MAX];
1418
1419 if (vstream->stream->tracefile_count > 0) {
1420 ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64,
1421 vstream->path_name,
1422 vstream->channel_name,
1423 vstream->current_tracefile_id);
1424 } else {
1425 ret = snprintf(fullpath, PATH_MAX, "%s/%s",
1426 vstream->path_name,
1427 vstream->channel_name);
1428 }
1429 if (ret < 0) {
1430 goto error_put;
1431 }
1432 ret = open(fullpath, O_RDONLY);
1433 if (ret < 0) {
1434 PERROR("Relay opening trace file");
1435 goto error_put;
1436 }
1437 vstream->stream_fd = stream_fd_create(ret);
1438 if (!vstream->stream_fd) {
1439 if (close(ret)) {
1440 PERROR("close");
1441 }
1442 goto error_put;
1443 }
1444 }
1445
1446 ret = check_new_streams(conn);
1447 if (ret < 0) {
1448 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1449 goto send_reply;
1450 } else if (ret == 1) {
1451 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1452 }
1453
1454 ret = lttng_index_file_read(vstream->index_file, &packet_index);
1455 if (ret) {
1456 ERR("Relay error reading index file %d",
1457 vstream->index_file->fd);
1458 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1459 goto send_reply;
1460 } else {
1461 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1462 vstream->index_sent_seqcount++;
1463 }
1464
1465 /*
1466 * Indexes are stored in big endian, no need to switch before sending.
1467 */
1468 DBG("Sending viewer index for stream %" PRIu64 " offset %" PRIu64,
1469 rstream->stream_handle,
1470 (uint64_t) be64toh(packet_index.offset));
1471 viewer_index.offset = packet_index.offset;
1472 viewer_index.packet_size = packet_index.packet_size;
1473 viewer_index.content_size = packet_index.content_size;
1474 viewer_index.timestamp_begin = packet_index.timestamp_begin;
1475 viewer_index.timestamp_end = packet_index.timestamp_end;
1476 viewer_index.events_discarded = packet_index.events_discarded;
1477 viewer_index.stream_id = packet_index.stream_id;
1478
1479 send_reply:
1480 if (rstream) {
1481 pthread_mutex_unlock(&rstream->lock);
1482 }
1483
1484 if (metadata_viewer_stream) {
1485 pthread_mutex_lock(&metadata_viewer_stream->stream->lock);
1486 DBG("get next index metadata check: recv %" PRIu64
1487 " sent %" PRIu64,
1488 metadata_viewer_stream->stream->metadata_received,
1489 metadata_viewer_stream->metadata_sent);
1490 if (!metadata_viewer_stream->stream->metadata_received ||
1491 metadata_viewer_stream->stream->metadata_received >
1492 metadata_viewer_stream->metadata_sent) {
1493 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1494 }
1495 pthread_mutex_unlock(&metadata_viewer_stream->stream->lock);
1496 }
1497
1498 viewer_index.flags = htobe32(viewer_index.flags);
1499 health_code_update();
1500
1501 ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1502 if (ret < 0) {
1503 goto end;
1504 }
1505 health_code_update();
1506
1507 if (vstream) {
1508 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1509 vstream->index_sent_seqcount,
1510 vstream->stream->stream_handle);
1511 }
1512 end:
1513 if (metadata_viewer_stream) {
1514 viewer_stream_put(metadata_viewer_stream);
1515 }
1516 if (vstream) {
1517 viewer_stream_put(vstream);
1518 }
1519 return ret;
1520
1521 error_put:
1522 pthread_mutex_unlock(&rstream->lock);
1523 if (metadata_viewer_stream) {
1524 viewer_stream_put(metadata_viewer_stream);
1525 }
1526 viewer_stream_put(vstream);
1527 return ret;
1528 }
1529
1530 /*
1531 * Send the next index for a stream
1532 *
1533 * Return 0 on success or else a negative value.
1534 */
1535 static
1536 int viewer_get_packet(struct relay_connection *conn)
1537 {
1538 int ret;
1539 off_t lseek_ret;
1540 char *reply = NULL;
1541 struct lttng_viewer_get_packet get_packet_info;
1542 struct lttng_viewer_trace_packet reply_header;
1543 struct relay_viewer_stream *vstream = NULL;
1544 uint32_t reply_size = sizeof(reply_header);
1545 uint32_t packet_data_len = 0;
1546 ssize_t read_len;
1547
1548 DBG2("Relay get data packet");
1549
1550 health_code_update();
1551
1552 ret = recv_request(conn->sock, &get_packet_info,
1553 sizeof(get_packet_info));
1554 if (ret < 0) {
1555 goto end;
1556 }
1557 health_code_update();
1558
1559 /* From this point on, the error label can be reached. */
1560 memset(&reply_header, 0, sizeof(reply_header));
1561
1562 vstream = viewer_stream_get_by_id(be64toh(get_packet_info.stream_id));
1563 if (!vstream) {
1564 DBG("Client requested packet of unknown stream id %" PRIu64,
1565 (uint64_t) be64toh(get_packet_info.stream_id));
1566 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1567 goto send_reply_nolock;
1568 } else {
1569 packet_data_len = be32toh(get_packet_info.len);
1570 reply_size += packet_data_len;
1571 }
1572
1573 reply = zmalloc(reply_size);
1574 if (!reply) {
1575 PERROR("packet reply zmalloc");
1576 reply_size = sizeof(reply_header);
1577 goto error;
1578 }
1579
1580 pthread_mutex_lock(&vstream->stream->lock);
1581 lseek_ret = lseek(vstream->stream_fd->fd, be64toh(get_packet_info.offset),
1582 SEEK_SET);
1583 if (lseek_ret < 0) {
1584 PERROR("lseek fd %d to offset %" PRIu64, vstream->stream_fd->fd,
1585 (uint64_t) be64toh(get_packet_info.offset));
1586 goto error;
1587 }
1588 read_len = lttng_read(vstream->stream_fd->fd,
1589 reply + sizeof(reply_header),
1590 packet_data_len);
1591 if (read_len < packet_data_len) {
1592 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1593 vstream->stream_fd->fd,
1594 (uint64_t) be64toh(get_packet_info.offset));
1595 goto error;
1596 }
1597 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1598 reply_header.len = htobe32(packet_data_len);
1599 goto send_reply;
1600
1601 error:
1602 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1603
1604 send_reply:
1605 if (vstream) {
1606 pthread_mutex_unlock(&vstream->stream->lock);
1607 }
1608 send_reply_nolock:
1609
1610 health_code_update();
1611
1612 if (reply) {
1613 memcpy(reply, &reply_header, sizeof(reply_header));
1614 ret = send_response(conn->sock, reply, reply_size);
1615 } else {
1616 /* No reply to send. */
1617 ret = send_response(conn->sock, &reply_header,
1618 reply_size);
1619 }
1620
1621 health_code_update();
1622 if (ret < 0) {
1623 PERROR("sendmsg of packet data failed");
1624 goto end_free;
1625 }
1626
1627 DBG("Sent %u bytes for stream %" PRIu64, reply_size,
1628 (uint64_t) be64toh(get_packet_info.stream_id));
1629
1630 end_free:
1631 free(reply);
1632 end:
1633 if (vstream) {
1634 viewer_stream_put(vstream);
1635 }
1636 return ret;
1637 }
1638
1639 /*
1640 * Send the session's metadata
1641 *
1642 * Return 0 on success else a negative value.
1643 */
1644 static
1645 int viewer_get_metadata(struct relay_connection *conn)
1646 {
1647 int ret = 0;
1648 ssize_t read_len;
1649 uint64_t len = 0;
1650 char *data = NULL;
1651 struct lttng_viewer_get_metadata request;
1652 struct lttng_viewer_metadata_packet reply;
1653 struct relay_viewer_stream *vstream = NULL;
1654
1655 assert(conn);
1656
1657 DBG("Relay get metadata");
1658
1659 health_code_update();
1660
1661 ret = recv_request(conn->sock, &request, sizeof(request));
1662 if (ret < 0) {
1663 goto end;
1664 }
1665 health_code_update();
1666
1667 memset(&reply, 0, sizeof(reply));
1668
1669 vstream = viewer_stream_get_by_id(be64toh(request.stream_id));
1670 if (!vstream) {
1671 /*
1672 * The metadata stream can be closed by a CLOSE command
1673 * just before we attach. It can also be closed by
1674 * per-pid tracing during tracing. Therefore, it is
1675 * possible that we cannot find this viewer stream.
1676 * Reply back to the client with an error if we cannot
1677 * find it.
1678 */
1679 DBG("Client requested metadata of unknown stream id %" PRIu64,
1680 (uint64_t) be64toh(request.stream_id));
1681 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1682 goto send_reply;
1683 }
1684 pthread_mutex_lock(&vstream->stream->lock);
1685 if (!vstream->stream->is_metadata) {
1686 ERR("Invalid metadata stream");
1687 goto error;
1688 }
1689
1690 assert(vstream->metadata_sent <= vstream->stream->metadata_received);
1691
1692 len = vstream->stream->metadata_received - vstream->metadata_sent;
1693 if (len == 0) {
1694 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1695 goto send_reply;
1696 }
1697
1698 /* first time, we open the metadata file */
1699 if (!vstream->stream_fd) {
1700 char fullpath[PATH_MAX];
1701
1702 ret = snprintf(fullpath, PATH_MAX, "%s/%s", vstream->path_name,
1703 vstream->channel_name);
1704 if (ret < 0) {
1705 goto error;
1706 }
1707 ret = open(fullpath, O_RDONLY);
1708 if (ret < 0) {
1709 PERROR("Relay opening metadata file");
1710 goto error;
1711 }
1712 vstream->stream_fd = stream_fd_create(ret);
1713 if (!vstream->stream_fd) {
1714 if (close(ret)) {
1715 PERROR("close");
1716 }
1717 goto error;
1718 }
1719 }
1720
1721 reply.len = htobe64(len);
1722 data = zmalloc(len);
1723 if (!data) {
1724 PERROR("viewer metadata zmalloc");
1725 goto error;
1726 }
1727
1728 read_len = lttng_read(vstream->stream_fd->fd, data, len);
1729 if (read_len < len) {
1730 PERROR("Relay reading metadata file");
1731 goto error;
1732 }
1733 vstream->metadata_sent += read_len;
1734 if (vstream->metadata_sent == vstream->stream->metadata_received
1735 && vstream->stream->closed) {
1736 /* Release ownership for the viewer metadata stream. */
1737 viewer_stream_put(vstream);
1738 }
1739
1740 reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1741
1742 goto send_reply;
1743
1744 error:
1745 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1746
1747 send_reply:
1748 health_code_update();
1749 if (vstream) {
1750 pthread_mutex_unlock(&vstream->stream->lock);
1751 }
1752 ret = send_response(conn->sock, &reply, sizeof(reply));
1753 if (ret < 0) {
1754 goto end_free;
1755 }
1756 health_code_update();
1757
1758 if (len > 0) {
1759 ret = send_response(conn->sock, data, len);
1760 if (ret < 0) {
1761 goto end_free;
1762 }
1763 }
1764
1765 DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1766 (uint64_t) be64toh(request.stream_id));
1767
1768 DBG("Metadata sent");
1769
1770 end_free:
1771 free(data);
1772 end:
1773 if (vstream) {
1774 viewer_stream_put(vstream);
1775 }
1776 return ret;
1777 }
1778
1779 /*
1780 * Create a viewer session.
1781 *
1782 * Return 0 on success or else a negative value.
1783 */
1784 static
1785 int viewer_create_session(struct relay_connection *conn)
1786 {
1787 int ret;
1788 struct lttng_viewer_create_session_response resp;
1789
1790 DBG("Viewer create session received");
1791
1792 memset(&resp, 0, sizeof(resp));
1793 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1794 conn->viewer_session = viewer_session_create();
1795 if (!conn->viewer_session) {
1796 ERR("Allocation viewer session");
1797 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1798 goto send_reply;
1799 }
1800
1801 send_reply:
1802 health_code_update();
1803 ret = send_response(conn->sock, &resp, sizeof(resp));
1804 if (ret < 0) {
1805 goto end;
1806 }
1807 health_code_update();
1808 ret = 0;
1809
1810 end:
1811 return ret;
1812 }
1813
1814 /*
1815 * Detach a viewer session.
1816 *
1817 * Return 0 on success or else a negative value.
1818 */
1819 static
1820 int viewer_detach_session(struct relay_connection *conn)
1821 {
1822 int ret;
1823 struct lttng_viewer_detach_session_response response;
1824 struct lttng_viewer_detach_session_request request;
1825 struct relay_session *session = NULL;
1826 uint64_t viewer_session_to_close;
1827
1828 DBG("Viewer detach session received");
1829
1830 assert(conn);
1831
1832 health_code_update();
1833
1834 /* Receive the request from the connected client. */
1835 ret = recv_request(conn->sock, &request, sizeof(request));
1836 if (ret < 0) {
1837 goto end;
1838 }
1839 viewer_session_to_close = be64toh(request.session_id);
1840
1841 if (!conn->viewer_session) {
1842 DBG("Client trying to detach before creating a live viewer session");
1843 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
1844 goto send_reply;
1845 }
1846
1847 health_code_update();
1848
1849 memset(&response, 0, sizeof(response));
1850 DBG("Detaching from session ID %" PRIu64, viewer_session_to_close);
1851
1852 session = session_get_by_id(be64toh(request.session_id));
1853 if (!session) {
1854 DBG("Relay session %" PRIu64 " not found",
1855 (uint64_t) be64toh(request.session_id));
1856 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_UNK);
1857 goto send_reply;
1858 }
1859
1860 ret = viewer_session_is_attached(conn->viewer_session, session);
1861 if (ret != 1) {
1862 DBG("Not attached to this session");
1863 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
1864 goto send_reply_put;
1865 }
1866
1867 viewer_session_close_one_session(conn->viewer_session, session);
1868 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_OK);
1869 DBG("Session %" PRIu64 " detached.", viewer_session_to_close);
1870
1871 send_reply_put:
1872 session_put(session);
1873
1874 send_reply:
1875 health_code_update();
1876 ret = send_response(conn->sock, &response, sizeof(response));
1877 if (ret < 0) {
1878 goto end;
1879 }
1880 health_code_update();
1881 ret = 0;
1882
1883 end:
1884 return ret;
1885 }
1886
1887 /*
1888 * live_relay_unknown_command: send -1 if received unknown command
1889 */
1890 static
1891 void live_relay_unknown_command(struct relay_connection *conn)
1892 {
1893 struct lttcomm_relayd_generic_reply reply;
1894
1895 memset(&reply, 0, sizeof(reply));
1896 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1897 (void) send_response(conn->sock, &reply, sizeof(reply));
1898 }
1899
1900 /*
1901 * Process the commands received on the control socket
1902 */
1903 static
1904 int process_control(struct lttng_viewer_cmd *recv_hdr,
1905 struct relay_connection *conn)
1906 {
1907 int ret = 0;
1908 uint32_t msg_value;
1909
1910 msg_value = be32toh(recv_hdr->cmd);
1911
1912 /*
1913 * Make sure we've done the version check before any command other then a
1914 * new client connection.
1915 */
1916 if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1917 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1918 ret = -1;
1919 goto end;
1920 }
1921
1922 switch (msg_value) {
1923 case LTTNG_VIEWER_CONNECT:
1924 ret = viewer_connect(conn);
1925 break;
1926 case LTTNG_VIEWER_LIST_SESSIONS:
1927 ret = viewer_list_sessions(conn);
1928 break;
1929 case LTTNG_VIEWER_ATTACH_SESSION:
1930 ret = viewer_attach_session(conn);
1931 break;
1932 case LTTNG_VIEWER_GET_NEXT_INDEX:
1933 ret = viewer_get_next_index(conn);
1934 break;
1935 case LTTNG_VIEWER_GET_PACKET:
1936 ret = viewer_get_packet(conn);
1937 break;
1938 case LTTNG_VIEWER_GET_METADATA:
1939 ret = viewer_get_metadata(conn);
1940 break;
1941 case LTTNG_VIEWER_GET_NEW_STREAMS:
1942 ret = viewer_get_new_streams(conn);
1943 break;
1944 case LTTNG_VIEWER_CREATE_SESSION:
1945 ret = viewer_create_session(conn);
1946 break;
1947 case LTTNG_VIEWER_DETACH_SESSION:
1948 ret = viewer_detach_session(conn);
1949 break;
1950 default:
1951 ERR("Received unknown viewer command (%u)",
1952 be32toh(recv_hdr->cmd));
1953 live_relay_unknown_command(conn);
1954 ret = -1;
1955 goto end;
1956 }
1957
1958 end:
1959 return ret;
1960 }
1961
1962 static
1963 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1964 {
1965 int ret;
1966
1967 (void) lttng_poll_del(events, pollfd);
1968
1969 ret = close(pollfd);
1970 if (ret < 0) {
1971 ERR("Closing pollfd %d", pollfd);
1972 }
1973 }
1974
1975 /*
1976 * This thread does the actual work
1977 */
1978 static
1979 void *thread_worker(void *data)
1980 {
1981 int ret, err = -1;
1982 uint32_t nb_fd;
1983 struct lttng_poll_event events;
1984 struct lttng_ht *viewer_connections_ht;
1985 struct lttng_ht_iter iter;
1986 struct lttng_viewer_cmd recv_hdr;
1987 struct relay_connection *destroy_conn;
1988
1989 DBG("[thread] Live viewer relay worker started");
1990
1991 rcu_register_thread();
1992
1993 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1994
1995 if (testpoint(relayd_thread_live_worker)) {
1996 goto error_testpoint;
1997 }
1998
1999 /* table of connections indexed on socket */
2000 viewer_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2001 if (!viewer_connections_ht) {
2002 goto viewer_connections_ht_error;
2003 }
2004
2005 ret = create_thread_poll_set(&events, 2);
2006 if (ret < 0) {
2007 goto error_poll_create;
2008 }
2009
2010 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
2011 if (ret < 0) {
2012 goto error;
2013 }
2014
2015 restart:
2016 while (1) {
2017 int i;
2018
2019 health_code_update();
2020
2021 /* Infinite blocking call, waiting for transmission */
2022 DBG3("Relayd live viewer worker thread polling...");
2023 health_poll_entry();
2024 ret = lttng_poll_wait(&events, -1);
2025 health_poll_exit();
2026 if (ret < 0) {
2027 /*
2028 * Restart interrupted system call.
2029 */
2030 if (errno == EINTR) {
2031 goto restart;
2032 }
2033 goto error;
2034 }
2035
2036 nb_fd = ret;
2037
2038 /*
2039 * Process control. The control connection is prioritised so we don't
2040 * starve it with high throughput tracing data on the data
2041 * connection.
2042 */
2043 for (i = 0; i < nb_fd; i++) {
2044 /* Fetch once the poll data */
2045 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2046 int pollfd = LTTNG_POLL_GETFD(&events, i);
2047
2048 health_code_update();
2049
2050 if (!revents) {
2051 /* No activity for this FD (poll implementation). */
2052 continue;
2053 }
2054
2055 /* Thread quit pipe has been closed. Killing thread. */
2056 ret = check_thread_quit_pipe(pollfd, revents);
2057 if (ret) {
2058 err = 0;
2059 goto exit;
2060 }
2061
2062 /* Inspect the relay conn pipe for new connection. */
2063 if (pollfd == live_conn_pipe[0]) {
2064 if (revents & LPOLLIN) {
2065 struct relay_connection *conn;
2066
2067 ret = lttng_read(live_conn_pipe[0],
2068 &conn, sizeof(conn));
2069 if (ret < 0) {
2070 goto error;
2071 }
2072 lttng_poll_add(&events, conn->sock->fd,
2073 LPOLLIN | LPOLLRDHUP);
2074 connection_ht_add(viewer_connections_ht, conn);
2075 DBG("Connection socket %d added to poll", conn->sock->fd);
2076 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2077 ERR("Relay live pipe error");
2078 goto error;
2079 } else {
2080 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2081 goto error;
2082 }
2083 } else {
2084 /* Connection activity. */
2085 struct relay_connection *conn;
2086
2087 conn = connection_get_by_sock(viewer_connections_ht, pollfd);
2088 if (!conn) {
2089 continue;
2090 }
2091
2092 if (revents & LPOLLIN) {
2093 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
2094 sizeof(recv_hdr), 0);
2095 if (ret <= 0) {
2096 /* Connection closed. */
2097 cleanup_connection_pollfd(&events, pollfd);
2098 /* Put "create" ownership reference. */
2099 connection_put(conn);
2100 DBG("Viewer control conn closed with %d", pollfd);
2101 } else {
2102 ret = process_control(&recv_hdr, conn);
2103 if (ret < 0) {
2104 /* Clear the session on error. */
2105 cleanup_connection_pollfd(&events, pollfd);
2106 /* Put "create" ownership reference. */
2107 connection_put(conn);
2108 DBG("Viewer connection closed with %d", pollfd);
2109 }
2110 }
2111 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2112 cleanup_connection_pollfd(&events, pollfd);
2113 /* Put "create" ownership reference. */
2114 connection_put(conn);
2115 } else {
2116 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2117 connection_put(conn);
2118 goto error;
2119 }
2120 /* Put local "get_by_sock" reference. */
2121 connection_put(conn);
2122 }
2123 }
2124 }
2125
2126 exit:
2127 error:
2128 lttng_poll_clean(&events);
2129
2130 /* Cleanup reamaining connection object. */
2131 rcu_read_lock();
2132 cds_lfht_for_each_entry(viewer_connections_ht->ht, &iter.iter,
2133 destroy_conn,
2134 sock_n.node) {
2135 health_code_update();
2136 connection_put(destroy_conn);
2137 }
2138 rcu_read_unlock();
2139 error_poll_create:
2140 lttng_ht_destroy(viewer_connections_ht);
2141 viewer_connections_ht_error:
2142 /* Close relay conn pipes */
2143 utils_close_pipe(live_conn_pipe);
2144 if (err) {
2145 DBG("Viewer worker thread exited with error");
2146 }
2147 DBG("Viewer worker thread cleanup complete");
2148 error_testpoint:
2149 if (err) {
2150 health_error();
2151 ERR("Health error occurred in %s", __func__);
2152 }
2153 health_unregister(health_relayd);
2154 if (lttng_relay_stop_threads()) {
2155 ERR("Error stopping threads");
2156 }
2157 rcu_unregister_thread();
2158 return NULL;
2159 }
2160
2161 /*
2162 * Create the relay command pipe to wake thread_manage_apps.
2163 * Closed in cleanup().
2164 */
2165 static int create_conn_pipe(void)
2166 {
2167 return utils_create_pipe_cloexec(live_conn_pipe);
2168 }
2169
2170 int relayd_live_join(void)
2171 {
2172 int ret, retval = 0;
2173 void *status;
2174
2175 ret = pthread_join(live_listener_thread, &status);
2176 if (ret) {
2177 errno = ret;
2178 PERROR("pthread_join live listener");
2179 retval = -1;
2180 }
2181
2182 ret = pthread_join(live_worker_thread, &status);
2183 if (ret) {
2184 errno = ret;
2185 PERROR("pthread_join live worker");
2186 retval = -1;
2187 }
2188
2189 ret = pthread_join(live_dispatcher_thread, &status);
2190 if (ret) {
2191 errno = ret;
2192 PERROR("pthread_join live dispatcher");
2193 retval = -1;
2194 }
2195
2196 cleanup_relayd_live();
2197
2198 return retval;
2199 }
2200
2201 /*
2202 * main
2203 */
2204 int relayd_live_create(struct lttng_uri *uri)
2205 {
2206 int ret = 0, retval = 0;
2207 void *status;
2208 int is_root;
2209
2210 if (!uri) {
2211 retval = -1;
2212 goto exit_init_data;
2213 }
2214 live_uri = uri;
2215
2216 /* Check if daemon is UID = 0 */
2217 is_root = !getuid();
2218
2219 if (!is_root) {
2220 if (live_uri->port < 1024) {
2221 ERR("Need to be root to use ports < 1024");
2222 retval = -1;
2223 goto exit_init_data;
2224 }
2225 }
2226
2227 /* Setup the thread apps communication pipe. */
2228 if (create_conn_pipe()) {
2229 retval = -1;
2230 goto exit_init_data;
2231 }
2232
2233 /* Init relay command queue. */
2234 cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2235
2236 /* Set up max poll set size */
2237 if (lttng_poll_set_max_size()) {
2238 retval = -1;
2239 goto exit_init_data;
2240 }
2241
2242 /* Setup the dispatcher thread */
2243 ret = pthread_create(&live_dispatcher_thread, default_pthread_attr(),
2244 thread_dispatcher, (void *) NULL);
2245 if (ret) {
2246 errno = ret;
2247 PERROR("pthread_create viewer dispatcher");
2248 retval = -1;
2249 goto exit_dispatcher_thread;
2250 }
2251
2252 /* Setup the worker thread */
2253 ret = pthread_create(&live_worker_thread, default_pthread_attr(),
2254 thread_worker, NULL);
2255 if (ret) {
2256 errno = ret;
2257 PERROR("pthread_create viewer worker");
2258 retval = -1;
2259 goto exit_worker_thread;
2260 }
2261
2262 /* Setup the listener thread */
2263 ret = pthread_create(&live_listener_thread, default_pthread_attr(),
2264 thread_listener, (void *) NULL);
2265 if (ret) {
2266 errno = ret;
2267 PERROR("pthread_create viewer listener");
2268 retval = -1;
2269 goto exit_listener_thread;
2270 }
2271
2272 /*
2273 * All OK, started all threads.
2274 */
2275 return retval;
2276
2277 /*
2278 * Join on the live_listener_thread should anything be added after
2279 * the live_listener thread's creation.
2280 */
2281
2282 exit_listener_thread:
2283
2284 ret = pthread_join(live_worker_thread, &status);
2285 if (ret) {
2286 errno = ret;
2287 PERROR("pthread_join live worker");
2288 retval = -1;
2289 }
2290 exit_worker_thread:
2291
2292 ret = pthread_join(live_dispatcher_thread, &status);
2293 if (ret) {
2294 errno = ret;
2295 PERROR("pthread_join live dispatcher");
2296 retval = -1;
2297 }
2298 exit_dispatcher_thread:
2299
2300 exit_init_data:
2301 cleanup_relayd_live();
2302
2303 return retval;
2304 }
This page took 0.080294 seconds and 5 git commands to generate.