88bb7363db962a41e65c8ccff4a6a654cbc666a3
[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 uint64_t session_id, 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 * session must be locked to ensure that we see either none or all initial
275 * streams for a session, but no intermediate state..
276 *
277 * Return 0 on success or else a negative value.
278 */
279 static int make_viewer_streams(struct relay_session *session,
280 struct lttng_trace_chunk *viewer_trace_chunk,
281 enum lttng_viewer_seek seek_t,
282 uint32_t *nb_total,
283 uint32_t *nb_unsent,
284 uint32_t *nb_created,
285 bool *closed)
286 {
287 int ret;
288 struct lttng_ht_iter iter;
289 struct ctf_trace *ctf_trace;
290
291 assert(session);
292 ASSERT_LOCKED(session->lock);
293
294 if (session->connection_closed) {
295 *closed = true;
296 }
297
298 /*
299 * Create viewer streams for relay streams that are ready to be
300 * used for a the given session id only.
301 */
302 rcu_read_lock();
303 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
304 node.node) {
305 struct relay_stream *stream;
306
307 health_code_update();
308
309 if (!ctf_trace_get(ctf_trace)) {
310 continue;
311 }
312
313 cds_list_for_each_entry_rcu(stream, &ctf_trace->stream_list, stream_node) {
314 struct relay_viewer_stream *vstream;
315
316 if (!stream_get(stream)) {
317 continue;
318 }
319 /*
320 * stream published is protected by the session lock.
321 */
322 if (!stream->published) {
323 goto next;
324 }
325 vstream = viewer_stream_get_by_id(stream->stream_handle);
326 if (!vstream) {
327 vstream = viewer_stream_create(stream,
328 viewer_trace_chunk, seek_t);
329 if (!vstream) {
330 ret = -1;
331 ctf_trace_put(ctf_trace);
332 stream_put(stream);
333 goto error_unlock;
334 }
335
336 if (nb_created) {
337 /* Update number of created stream counter. */
338 (*nb_created)++;
339 }
340 /*
341 * Ensure a self-reference is preserved even
342 * after we have put our local reference.
343 */
344 if (!viewer_stream_get(vstream)) {
345 ERR("Unable to get self-reference on viewer stream, logic error.");
346 abort();
347 }
348 } else {
349 if (!vstream->sent_flag && nb_unsent) {
350 /* Update number of unsent stream counter. */
351 (*nb_unsent)++;
352 }
353 }
354 /* Update number of total stream counter. */
355 if (nb_total) {
356 if (stream->is_metadata) {
357 if (!stream->closed ||
358 stream->metadata_received > vstream->metadata_sent) {
359 (*nb_total)++;
360 }
361 } else {
362 if (!stream->closed ||
363 !(((int64_t) (stream->prev_data_seq - stream->last_net_seq_num)) >= 0)) {
364
365 (*nb_total)++;
366 }
367 }
368 }
369 /* Put local reference. */
370 viewer_stream_put(vstream);
371 next:
372 stream_put(stream);
373 }
374 ctf_trace_put(ctf_trace);
375 }
376
377 ret = 0;
378
379 error_unlock:
380 rcu_read_unlock();
381 return ret;
382 }
383
384 int relayd_live_stop(void)
385 {
386 /* Stop dispatch thread */
387 CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
388 futex_nto1_wake(&viewer_conn_queue.futex);
389 return 0;
390 }
391
392 /*
393 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
394 */
395 static
396 int create_thread_poll_set(struct lttng_poll_event *events, int size)
397 {
398 int ret;
399
400 if (events == NULL || size == 0) {
401 ret = -1;
402 goto error;
403 }
404
405 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
406 if (ret < 0) {
407 goto error;
408 }
409
410 /* Add quit pipe */
411 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
412 if (ret < 0) {
413 goto error;
414 }
415
416 return 0;
417
418 error:
419 return ret;
420 }
421
422 /*
423 * Check if the thread quit pipe was triggered.
424 *
425 * Return 1 if it was triggered else 0;
426 */
427 static
428 int check_thread_quit_pipe(int fd, uint32_t events)
429 {
430 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
431 return 1;
432 }
433
434 return 0;
435 }
436
437 /*
438 * Create and init socket from uri.
439 */
440 static
441 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
442 {
443 int ret;
444 struct lttcomm_sock *sock = NULL;
445
446 sock = lttcomm_alloc_sock_from_uri(uri);
447 if (sock == NULL) {
448 ERR("Allocating socket");
449 goto error;
450 }
451
452 ret = lttcomm_create_sock(sock);
453 if (ret < 0) {
454 goto error;
455 }
456 DBG("Listening on sock %d for lttng-live", sock->fd);
457
458 ret = sock->ops->bind(sock);
459 if (ret < 0) {
460 PERROR("Failed to bind lttng-live socket");
461 goto error;
462 }
463
464 ret = sock->ops->listen(sock, -1);
465 if (ret < 0) {
466 goto error;
467
468 }
469
470 return sock;
471
472 error:
473 if (sock) {
474 lttcomm_destroy_sock(sock);
475 }
476 return NULL;
477 }
478
479 /*
480 * This thread manages the listening for new connections on the network
481 */
482 static
483 void *thread_listener(void *data)
484 {
485 int i, ret, pollfd, err = -1;
486 uint32_t revents, nb_fd;
487 struct lttng_poll_event events;
488 struct lttcomm_sock *live_control_sock;
489
490 DBG("[thread] Relay live listener started");
491
492 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
493
494 health_code_update();
495
496 live_control_sock = init_socket(live_uri);
497 if (!live_control_sock) {
498 goto error_sock_control;
499 }
500
501 /* Pass 2 as size here for the thread quit pipe and control sockets. */
502 ret = create_thread_poll_set(&events, 2);
503 if (ret < 0) {
504 goto error_create_poll;
505 }
506
507 /* Add the control socket */
508 ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
509 if (ret < 0) {
510 goto error_poll_add;
511 }
512
513 lttng_relay_notify_ready();
514
515 if (testpoint(relayd_thread_live_listener)) {
516 goto error_testpoint;
517 }
518
519 while (1) {
520 health_code_update();
521
522 DBG("Listener accepting live viewers connections");
523
524 restart:
525 health_poll_entry();
526 ret = lttng_poll_wait(&events, -1);
527 health_poll_exit();
528 if (ret < 0) {
529 /*
530 * Restart interrupted system call.
531 */
532 if (errno == EINTR) {
533 goto restart;
534 }
535 goto error;
536 }
537 nb_fd = ret;
538
539 DBG("Relay new viewer connection received");
540 for (i = 0; i < nb_fd; i++) {
541 health_code_update();
542
543 /* Fetch once the poll data */
544 revents = LTTNG_POLL_GETEV(&events, i);
545 pollfd = LTTNG_POLL_GETFD(&events, i);
546
547 /* Thread quit pipe has been closed. Killing thread. */
548 ret = check_thread_quit_pipe(pollfd, revents);
549 if (ret) {
550 err = 0;
551 goto exit;
552 }
553
554 if (revents & LPOLLIN) {
555 /*
556 * A new connection is requested, therefore a
557 * viewer connection is allocated in this
558 * thread, enqueued to a global queue and
559 * dequeued (and freed) in the worker thread.
560 */
561 int val = 1;
562 struct relay_connection *new_conn;
563 struct lttcomm_sock *newsock;
564
565 newsock = live_control_sock->ops->accept(live_control_sock);
566 if (!newsock) {
567 PERROR("accepting control sock");
568 goto error;
569 }
570 DBG("Relay viewer connection accepted socket %d", newsock->fd);
571
572 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
573 sizeof(val));
574 if (ret < 0) {
575 PERROR("setsockopt inet");
576 lttcomm_destroy_sock(newsock);
577 goto error;
578 }
579 new_conn = connection_create(newsock, RELAY_CONNECTION_UNKNOWN);
580 if (!new_conn) {
581 lttcomm_destroy_sock(newsock);
582 goto error;
583 }
584 /* Ownership assumed by the connection. */
585 newsock = NULL;
586
587 /* Enqueue request for the dispatcher thread. */
588 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
589 &new_conn->qnode);
590
591 /*
592 * Wake the dispatch queue futex.
593 * Implicit memory barrier with the
594 * exchange in cds_wfcq_enqueue.
595 */
596 futex_nto1_wake(&viewer_conn_queue.futex);
597 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
598 ERR("socket poll error");
599 goto error;
600 } else {
601 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
602 goto error;
603 }
604 }
605 }
606
607 exit:
608 error:
609 error_poll_add:
610 error_testpoint:
611 lttng_poll_clean(&events);
612 error_create_poll:
613 if (live_control_sock->fd >= 0) {
614 ret = live_control_sock->ops->close(live_control_sock);
615 if (ret) {
616 PERROR("close");
617 }
618 }
619 lttcomm_destroy_sock(live_control_sock);
620 error_sock_control:
621 if (err) {
622 health_error();
623 DBG("Live viewer listener thread exited with error");
624 }
625 health_unregister(health_relayd);
626 DBG("Live viewer listener thread cleanup complete");
627 if (lttng_relay_stop_threads()) {
628 ERR("Error stopping threads");
629 }
630 return NULL;
631 }
632
633 /*
634 * This thread manages the dispatching of the requests to worker threads
635 */
636 static
637 void *thread_dispatcher(void *data)
638 {
639 int err = -1;
640 ssize_t ret;
641 struct cds_wfcq_node *node;
642 struct relay_connection *conn = NULL;
643
644 DBG("[thread] Live viewer relay dispatcher started");
645
646 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
647
648 if (testpoint(relayd_thread_live_dispatcher)) {
649 goto error_testpoint;
650 }
651
652 health_code_update();
653
654 for (;;) {
655 health_code_update();
656
657 /* Atomically prepare the queue futex */
658 futex_nto1_prepare(&viewer_conn_queue.futex);
659
660 if (CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
661 break;
662 }
663
664 do {
665 health_code_update();
666
667 /* Dequeue commands */
668 node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
669 &viewer_conn_queue.tail);
670 if (node == NULL) {
671 DBG("Woken up but nothing in the live-viewer "
672 "relay command queue");
673 /* Continue thread execution */
674 break;
675 }
676 conn = caa_container_of(node, struct relay_connection, qnode);
677 DBG("Dispatching viewer request waiting on sock %d",
678 conn->sock->fd);
679
680 /*
681 * Inform worker thread of the new request. This
682 * call is blocking so we can be assured that
683 * the data will be read at some point in time
684 * or wait to the end of the world :)
685 */
686 ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
687 if (ret < 0) {
688 PERROR("write conn pipe");
689 connection_put(conn);
690 goto error;
691 }
692 } while (node != NULL);
693
694 /* Futex wait on queue. Blocking call on futex() */
695 health_poll_entry();
696 futex_nto1_wait(&viewer_conn_queue.futex);
697 health_poll_exit();
698 }
699
700 /* Normal exit, no error */
701 err = 0;
702
703 error:
704 error_testpoint:
705 if (err) {
706 health_error();
707 ERR("Health error occurred in %s", __func__);
708 }
709 health_unregister(health_relayd);
710 DBG("Live viewer dispatch thread dying");
711 if (lttng_relay_stop_threads()) {
712 ERR("Error stopping threads");
713 }
714 return NULL;
715 }
716
717 /*
718 * Establish connection with the viewer and check the versions.
719 *
720 * Return 0 on success or else negative value.
721 */
722 static
723 int viewer_connect(struct relay_connection *conn)
724 {
725 int ret;
726 struct lttng_viewer_connect reply, msg;
727
728 conn->version_check_done = 1;
729
730 health_code_update();
731
732 DBG("Viewer is establishing a connection to the relayd.");
733
734 ret = recv_request(conn->sock, &msg, sizeof(msg));
735 if (ret < 0) {
736 goto end;
737 }
738
739 health_code_update();
740
741 memset(&reply, 0, sizeof(reply));
742 reply.major = RELAYD_VERSION_COMM_MAJOR;
743 reply.minor = RELAYD_VERSION_COMM_MINOR;
744
745 /* Major versions must be the same */
746 if (reply.major != be32toh(msg.major)) {
747 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
748 reply.major, be32toh(msg.major));
749 ret = -1;
750 goto end;
751 }
752
753 conn->major = reply.major;
754 /* We adapt to the lowest compatible version */
755 if (reply.minor <= be32toh(msg.minor)) {
756 conn->minor = reply.minor;
757 } else {
758 conn->minor = be32toh(msg.minor);
759 }
760
761 if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
762 conn->type = RELAY_VIEWER_COMMAND;
763 } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
764 conn->type = RELAY_VIEWER_NOTIFICATION;
765 } else {
766 ERR("Unknown connection type : %u", be32toh(msg.type));
767 ret = -1;
768 goto end;
769 }
770
771 reply.major = htobe32(reply.major);
772 reply.minor = htobe32(reply.minor);
773 if (conn->type == RELAY_VIEWER_COMMAND) {
774 /*
775 * Increment outside of htobe64 macro, because the argument can
776 * be used more than once within the macro, and thus the
777 * operation may be undefined.
778 */
779 pthread_mutex_lock(&last_relay_viewer_session_id_lock);
780 last_relay_viewer_session_id++;
781 pthread_mutex_unlock(&last_relay_viewer_session_id_lock);
782 reply.viewer_session_id = htobe64(last_relay_viewer_session_id);
783 }
784
785 health_code_update();
786
787 ret = send_response(conn->sock, &reply, sizeof(reply));
788 if (ret < 0) {
789 goto end;
790 }
791
792 health_code_update();
793
794 DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
795 ret = 0;
796
797 end:
798 return ret;
799 }
800
801 /*
802 * Send the viewer the list of current sessions.
803 * We need to create a copy of the hash table content because otherwise
804 * we cannot assume the number of entries stays the same between getting
805 * the number of HT elements and iteration over the HT.
806 *
807 * Return 0 on success or else a negative value.
808 */
809 static
810 int viewer_list_sessions(struct relay_connection *conn)
811 {
812 int ret = 0;
813 struct lttng_viewer_list_sessions session_list;
814 struct lttng_ht_iter iter;
815 struct relay_session *session;
816 struct lttng_viewer_session *send_session_buf = NULL;
817 uint32_t buf_count = SESSION_BUF_DEFAULT_COUNT;
818 uint32_t count = 0;
819
820 DBG("List sessions received");
821
822 send_session_buf = zmalloc(SESSION_BUF_DEFAULT_COUNT * sizeof(*send_session_buf));
823 if (!send_session_buf) {
824 return -1;
825 }
826
827 rcu_read_lock();
828 cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, session,
829 session_n.node) {
830 struct lttng_viewer_session *send_session;
831
832 health_code_update();
833
834 if (session->connection_closed) {
835 /* Skip closed session */
836 continue;
837 }
838
839 if (count >= buf_count) {
840 struct lttng_viewer_session *newbuf;
841 uint32_t new_buf_count = buf_count << 1;
842
843 newbuf = realloc(send_session_buf,
844 new_buf_count * sizeof(*send_session_buf));
845 if (!newbuf) {
846 ret = -1;
847 break;
848 }
849 send_session_buf = newbuf;
850 buf_count = new_buf_count;
851 }
852 send_session = &send_session_buf[count];
853 if (lttng_strncpy(send_session->session_name,
854 session->session_name,
855 sizeof(send_session->session_name))) {
856 ret = -1;
857 break;
858 }
859 if (lttng_strncpy(send_session->hostname, session->hostname,
860 sizeof(send_session->hostname))) {
861 ret = -1;
862 break;
863 }
864 send_session->id = htobe64(session->id);
865 send_session->live_timer = htobe32(session->live_timer);
866 if (session->viewer_attached) {
867 send_session->clients = htobe32(1);
868 } else {
869 send_session->clients = htobe32(0);
870 }
871 send_session->streams = htobe32(session->stream_count);
872 count++;
873 }
874 rcu_read_unlock();
875 if (ret < 0) {
876 goto end_free;
877 }
878
879 session_list.sessions_count = htobe32(count);
880
881 health_code_update();
882
883 ret = send_response(conn->sock, &session_list, sizeof(session_list));
884 if (ret < 0) {
885 goto end_free;
886 }
887
888 health_code_update();
889
890 ret = send_response(conn->sock, send_session_buf,
891 count * sizeof(*send_session_buf));
892 if (ret < 0) {
893 goto end_free;
894 }
895 health_code_update();
896
897 ret = 0;
898 end_free:
899 free(send_session_buf);
900 return ret;
901 }
902
903 /*
904 * Send the viewer the list of current streams.
905 */
906 static
907 int viewer_get_new_streams(struct relay_connection *conn)
908 {
909 int ret, send_streams = 0;
910 uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0, nb_total = 0;
911 struct lttng_viewer_new_streams_request request;
912 struct lttng_viewer_new_streams_response response;
913 struct relay_session *session = NULL;
914 uint64_t session_id;
915 bool closed = false;
916
917 assert(conn);
918
919 DBG("Get new streams received");
920
921 health_code_update();
922
923 /* Receive the request from the connected client. */
924 ret = recv_request(conn->sock, &request, sizeof(request));
925 if (ret < 0) {
926 goto error;
927 }
928 session_id = be64toh(request.session_id);
929
930 health_code_update();
931
932 memset(&response, 0, sizeof(response));
933
934 session = session_get_by_id(session_id);
935 if (!session) {
936 DBG("Relay session %" PRIu64 " not found", session_id);
937 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
938 goto send_reply;
939 }
940
941 if (!viewer_session_is_attached(conn->viewer_session, session)) {
942 send_streams = 0;
943 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
944 goto send_reply;
945 }
946
947 send_streams = 1;
948 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
949
950 pthread_mutex_lock(&session->lock);
951 if (!conn->viewer_session->current_trace_chunk &&
952 session->current_trace_chunk) {
953 ret = viewer_session_set_trace_chunk(conn->viewer_session,
954 session->current_trace_chunk);
955 if (ret) {
956 goto error_unlock_session;
957 }
958 }
959 ret = make_viewer_streams(session,
960 conn->viewer_session->current_trace_chunk,
961 LTTNG_VIEWER_SEEK_LAST, &nb_total, &nb_unsent,
962 &nb_created, &closed);
963 if (ret < 0) {
964 goto error_unlock_session;
965 }
966 pthread_mutex_unlock(&session->lock);
967
968 /* Only send back the newly created streams with the unsent ones. */
969 nb_streams = nb_created + nb_unsent;
970 response.streams_count = htobe32(nb_streams);
971
972 /*
973 * If the session is closed, HUP when there are no more streams
974 * with data.
975 */
976 if (closed && nb_total == 0) {
977 send_streams = 0;
978 response.streams_count = 0;
979 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
980 goto send_reply;
981 }
982
983 send_reply:
984 health_code_update();
985 ret = send_response(conn->sock, &response, sizeof(response));
986 if (ret < 0) {
987 goto end_put_session;
988 }
989 health_code_update();
990
991 /*
992 * Unknown or empty session, just return gracefully, the viewer
993 * knows what is happening.
994 */
995 if (!send_streams || !nb_streams) {
996 ret = 0;
997 goto end_put_session;
998 }
999
1000 /*
1001 * Send stream and *DON'T* ignore the sent flag so every viewer
1002 * streams that were not sent from that point will be sent to
1003 * the viewer.
1004 */
1005 ret = send_viewer_streams(conn->sock, session_id, 0);
1006 if (ret < 0) {
1007 goto end_put_session;
1008 }
1009
1010 end_put_session:
1011 if (session) {
1012 session_put(session);
1013 }
1014 error:
1015 return ret;
1016 error_unlock_session:
1017 pthread_mutex_unlock(&session->lock);
1018 session_put(session);
1019 return ret;
1020 }
1021
1022 /*
1023 * Send the viewer the list of current sessions.
1024 */
1025 static
1026 int viewer_attach_session(struct relay_connection *conn)
1027 {
1028 int send_streams = 0;
1029 ssize_t ret;
1030 uint32_t nb_streams = 0;
1031 enum lttng_viewer_seek seek_type;
1032 struct lttng_viewer_attach_session_request request;
1033 struct lttng_viewer_attach_session_response response;
1034 struct relay_session *session = NULL;
1035 bool closed = false;
1036 uint64_t session_id;
1037
1038 assert(conn);
1039
1040 health_code_update();
1041
1042 /* Receive the request from the connected client. */
1043 ret = recv_request(conn->sock, &request, sizeof(request));
1044 if (ret < 0) {
1045 goto error;
1046 }
1047
1048 session_id = be64toh(request.session_id);
1049 health_code_update();
1050
1051 memset(&response, 0, sizeof(response));
1052
1053 if (!conn->viewer_session) {
1054 DBG("Client trying to attach before creating a live viewer session");
1055 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1056 goto send_reply;
1057 }
1058
1059 session = session_get_by_id(session_id);
1060 if (!session) {
1061 DBG("Relay session %" PRIu64 " not found", session_id);
1062 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1063 goto send_reply;
1064 }
1065 DBG("Attach session ID %" PRIu64 " received", session_id);
1066
1067 pthread_mutex_lock(&session->lock);
1068 if (session->live_timer == 0) {
1069 DBG("Not live session");
1070 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1071 goto send_reply;
1072 }
1073
1074 send_streams = 1;
1075 ret = viewer_session_attach(conn->viewer_session, session);
1076 if (ret) {
1077 DBG("Already a viewer attached");
1078 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1079 goto send_reply;
1080 }
1081
1082 switch (be32toh(request.seek)) {
1083 case LTTNG_VIEWER_SEEK_BEGINNING:
1084 case LTTNG_VIEWER_SEEK_LAST:
1085 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1086 seek_type = be32toh(request.seek);
1087 break;
1088 default:
1089 ERR("Wrong seek parameter");
1090 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1091 send_streams = 0;
1092 goto send_reply;
1093 }
1094
1095 if (!conn->viewer_session->current_trace_chunk &&
1096 session->current_trace_chunk) {
1097 ret = viewer_session_set_trace_chunk(conn->viewer_session,
1098 session->current_trace_chunk);
1099 if (ret) {
1100 goto end_put_session;
1101 }
1102 }
1103 ret = make_viewer_streams(session,
1104 conn->viewer_session->current_trace_chunk, seek_type,
1105 &nb_streams, NULL, NULL, &closed);
1106 if (ret < 0) {
1107 goto end_put_session;
1108 }
1109 pthread_mutex_unlock(&session->lock);
1110 session_put(session);
1111 session = NULL;
1112
1113 response.streams_count = htobe32(nb_streams);
1114 /*
1115 * If the session is closed when the viewer is attaching, it
1116 * means some of the streams may have been concurrently removed,
1117 * so we don't allow the viewer to attach, even if there are
1118 * streams available.
1119 */
1120 if (closed) {
1121 send_streams = 0;
1122 response.streams_count = 0;
1123 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1124 goto send_reply;
1125 }
1126
1127 send_reply:
1128 health_code_update();
1129 ret = send_response(conn->sock, &response, sizeof(response));
1130 if (ret < 0) {
1131 goto end_put_session;
1132 }
1133 health_code_update();
1134
1135 /*
1136 * Unknown or empty session, just return gracefully, the viewer
1137 * knows what is happening.
1138 */
1139 if (!send_streams || !nb_streams) {
1140 ret = 0;
1141 goto end_put_session;
1142 }
1143
1144 /* Send stream and ignore the sent flag. */
1145 ret = send_viewer_streams(conn->sock, session_id, 1);
1146 if (ret < 0) {
1147 goto end_put_session;
1148 }
1149
1150 end_put_session:
1151 if (session) {
1152 pthread_mutex_unlock(&session->lock);
1153 session_put(session);
1154 }
1155 error:
1156 return ret;
1157 }
1158
1159 /*
1160 * Open the index file if needed for the given vstream.
1161 *
1162 * If an index file is successfully opened, the vstream will set it as its
1163 * current index file.
1164 *
1165 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1166 *
1167 * Called with rstream lock held.
1168 */
1169 static int try_open_index(struct relay_viewer_stream *vstream,
1170 struct relay_stream *rstream)
1171 {
1172 int ret = 0;
1173 const uint32_t connection_major = rstream->trace->session->major;
1174 const uint32_t connection_minor = rstream->trace->session->minor;
1175
1176 if (vstream->index_file) {
1177 goto end;
1178 }
1179
1180 /*
1181 * First time, we open the index file and at least one index is ready.
1182 */
1183 if (rstream->index_received_seqcount == 0) {
1184 ret = -ENOENT;
1185 goto end;
1186 }
1187 vstream->index_file = lttng_index_file_create_from_trace_chunk_read_only(
1188 vstream->stream_file.trace_chunk, rstream->path_name,
1189 rstream->channel_name, rstream->tracefile_size,
1190 vstream->current_tracefile_id,
1191 lttng_to_index_major(connection_major, connection_minor),
1192 lttng_to_index_minor(connection_major, connection_minor));
1193 if (!vstream->index_file) {
1194 ret = -1;
1195 }
1196
1197 end:
1198 return ret;
1199 }
1200
1201 /*
1202 * Check the status of the index for the given stream. This function
1203 * updates the index structure if needed and can put (close) the vstream
1204 * in the HUP situation.
1205 *
1206 * Return 0 means that we can proceed with the index. A value of 1 means
1207 * that the index has been updated and is ready to be sent to the
1208 * client. A negative value indicates an error that can't be handled.
1209 *
1210 * Called with rstream lock held.
1211 */
1212 static int check_index_status(struct relay_viewer_stream *vstream,
1213 struct relay_stream *rstream, struct ctf_trace *trace,
1214 struct lttng_viewer_index *index)
1215 {
1216 int ret;
1217
1218 if ((trace->session->connection_closed || rstream->closed)
1219 && rstream->index_received_seqcount
1220 == vstream->index_sent_seqcount) {
1221 /*
1222 * Last index sent and session connection or relay
1223 * stream are closed.
1224 */
1225 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1226 goto hup;
1227 } else if (rstream->beacon_ts_end != -1ULL &&
1228 rstream->index_received_seqcount
1229 == vstream->index_sent_seqcount) {
1230 /*
1231 * We've received a synchronization beacon and the last index
1232 * available has been sent, the index for now is inactive.
1233 *
1234 * In this case, we have received a beacon which allows us to
1235 * inform the client of a time interval during which we can
1236 * guarantee that there are no events to read (and never will
1237 * be).
1238 */
1239 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1240 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1241 index->stream_id = htobe64(rstream->ctf_stream_id);
1242 goto index_ready;
1243 } else if (rstream->index_received_seqcount
1244 == vstream->index_sent_seqcount) {
1245 /*
1246 * This checks whether received == sent seqcount. In
1247 * this case, we have not received a beacon. Therefore,
1248 * we can only ask the client to retry later.
1249 */
1250 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1251 goto index_ready;
1252 } else if (!tracefile_array_seq_in_file(rstream->tfa,
1253 vstream->current_tracefile_id,
1254 vstream->index_sent_seqcount)) {
1255 /*
1256 * The next index we want to send cannot be read either
1257 * because we need to perform a rotation, or due to
1258 * the producer having overwritten its trace file.
1259 */
1260 DBG("Viewer stream %" PRIu64 " rotation",
1261 vstream->stream->stream_handle);
1262 ret = viewer_stream_rotate(vstream);
1263 if (ret < 0) {
1264 goto end;
1265 } else if (ret == 1) {
1266 /* EOF across entire stream. */
1267 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1268 goto hup;
1269 }
1270 /*
1271 * If we have been pushed due to overwrite, it
1272 * necessarily means there is data that can be read in
1273 * the stream. If we rotated because we reached the end
1274 * of a tracefile, it means the following tracefile
1275 * needs to contain at least one index, else we would
1276 * have already returned LTTNG_VIEWER_INDEX_RETRY to the
1277 * viewer. The updated index_sent_seqcount needs to
1278 * point to a readable index entry now.
1279 *
1280 * In the case where we "rotate" on a single file, we
1281 * can end up in a case where the requested index is
1282 * still unavailable.
1283 */
1284 if (rstream->tracefile_count == 1 &&
1285 !tracefile_array_seq_in_file(
1286 rstream->tfa,
1287 vstream->current_tracefile_id,
1288 vstream->index_sent_seqcount)) {
1289 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1290 goto index_ready;
1291 }
1292 assert(tracefile_array_seq_in_file(rstream->tfa,
1293 vstream->current_tracefile_id,
1294 vstream->index_sent_seqcount));
1295 }
1296 /* ret == 0 means successful so we continue. */
1297 ret = 0;
1298 end:
1299 return ret;
1300
1301 hup:
1302 viewer_stream_put(vstream);
1303 index_ready:
1304 return 1;
1305 }
1306
1307 /*
1308 * Send the next index for a stream.
1309 *
1310 * Return 0 on success or else a negative value.
1311 */
1312 static
1313 int viewer_get_next_index(struct relay_connection *conn)
1314 {
1315 int ret;
1316 struct lttng_viewer_get_next_index request_index;
1317 struct lttng_viewer_index viewer_index;
1318 struct ctf_packet_index packet_index;
1319 struct relay_viewer_stream *vstream = NULL;
1320 struct relay_stream *rstream = NULL;
1321 struct ctf_trace *ctf_trace = NULL;
1322 struct relay_viewer_stream *metadata_viewer_stream = NULL;
1323
1324 assert(conn);
1325
1326 DBG("Viewer get next index");
1327
1328 memset(&viewer_index, 0, sizeof(viewer_index));
1329 health_code_update();
1330
1331 ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1332 if (ret < 0) {
1333 goto end;
1334 }
1335 health_code_update();
1336
1337 vstream = viewer_stream_get_by_id(be64toh(request_index.stream_id));
1338 if (!vstream) {
1339 DBG("Client requested index of unknown stream id %" PRIu64,
1340 (uint64_t) be64toh(request_index.stream_id));
1341 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1342 goto send_reply;
1343 }
1344
1345 /* Use back. ref. Protected by refcounts. */
1346 rstream = vstream->stream;
1347 ctf_trace = rstream->trace;
1348
1349 /* metadata_viewer_stream may be NULL. */
1350 metadata_viewer_stream =
1351 ctf_trace_get_viewer_metadata_stream(ctf_trace);
1352
1353 pthread_mutex_lock(&rstream->lock);
1354
1355 /*
1356 * The viewer should not ask for index on metadata stream.
1357 */
1358 if (rstream->is_metadata) {
1359 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1360 goto send_reply;
1361 }
1362
1363 /* Try to open an index if one is needed for that stream. */
1364 ret = try_open_index(vstream, rstream);
1365 if (ret < 0) {
1366 if (ret == -ENOENT) {
1367 /*
1368 * The index is created only when the first data
1369 * packet arrives, it might not be ready at the
1370 * beginning of the session
1371 */
1372 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1373 } else {
1374 /* Unhandled error. */
1375 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1376 }
1377 goto send_reply;
1378 }
1379
1380 ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1381 if (ret < 0) {
1382 goto error_put;
1383 } else if (ret == 1) {
1384 /*
1385 * We have no index to send and check_index_status has populated
1386 * viewer_index's status.
1387 */
1388 goto send_reply;
1389 }
1390 /* At this point, ret is 0 thus we will be able to read the index. */
1391 assert(!ret);
1392
1393 /*
1394 * vstream->stream_fd may be NULL if it has been closed by
1395 * tracefile rotation, or if we are at the beginning of the
1396 * stream. We open the data stream file here to protect against
1397 * overwrite caused by tracefile rotation (in association with
1398 * unlink performed before overwrite).
1399 */
1400 if (!vstream->stream_file.fd) {
1401 int fd;
1402 char file_path[LTTNG_PATH_MAX];
1403 enum lttng_trace_chunk_status status;
1404
1405 ret = utils_stream_file_path(rstream->path_name,
1406 rstream->channel_name, rstream->tracefile_size,
1407 vstream->current_tracefile_id, NULL, file_path,
1408 sizeof(file_path));
1409 if (ret < 0) {
1410 goto error_put;
1411 }
1412
1413 status = lttng_trace_chunk_open_file(
1414 vstream->stream_file.trace_chunk,
1415 file_path, O_RDONLY, 0, &fd);
1416 if (status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1417 PERROR("Failed to open trace file for viewer stream");
1418 goto error_put;
1419 }
1420 vstream->stream_file.fd = stream_fd_create(fd);
1421 if (!vstream->stream_file.fd) {
1422 if (close(fd)) {
1423 PERROR("Failed to close viewer stream file");
1424 }
1425 goto error_put;
1426 }
1427 }
1428
1429 ret = check_new_streams(conn);
1430 if (ret < 0) {
1431 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1432 goto send_reply;
1433 } else if (ret == 1) {
1434 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1435 }
1436
1437 ret = lttng_index_file_read(vstream->index_file, &packet_index);
1438 if (ret) {
1439 ERR("Relay error reading index file %d",
1440 vstream->index_file->fd);
1441 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1442 goto send_reply;
1443 } else {
1444 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1445 vstream->index_sent_seqcount++;
1446 }
1447
1448 /*
1449 * Indexes are stored in big endian, no need to switch before sending.
1450 */
1451 DBG("Sending viewer index for stream %" PRIu64 " offset %" PRIu64,
1452 rstream->stream_handle,
1453 (uint64_t) be64toh(packet_index.offset));
1454 viewer_index.offset = packet_index.offset;
1455 viewer_index.packet_size = packet_index.packet_size;
1456 viewer_index.content_size = packet_index.content_size;
1457 viewer_index.timestamp_begin = packet_index.timestamp_begin;
1458 viewer_index.timestamp_end = packet_index.timestamp_end;
1459 viewer_index.events_discarded = packet_index.events_discarded;
1460 viewer_index.stream_id = packet_index.stream_id;
1461
1462 send_reply:
1463 if (rstream) {
1464 pthread_mutex_unlock(&rstream->lock);
1465 }
1466
1467 if (metadata_viewer_stream) {
1468 pthread_mutex_lock(&metadata_viewer_stream->stream->lock);
1469 DBG("get next index metadata check: recv %" PRIu64
1470 " sent %" PRIu64,
1471 metadata_viewer_stream->stream->metadata_received,
1472 metadata_viewer_stream->metadata_sent);
1473 if (!metadata_viewer_stream->stream->metadata_received ||
1474 metadata_viewer_stream->stream->metadata_received >
1475 metadata_viewer_stream->metadata_sent) {
1476 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1477 }
1478 pthread_mutex_unlock(&metadata_viewer_stream->stream->lock);
1479 }
1480
1481 viewer_index.flags = htobe32(viewer_index.flags);
1482 health_code_update();
1483
1484 ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1485 if (ret < 0) {
1486 goto end;
1487 }
1488 health_code_update();
1489
1490 if (vstream) {
1491 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1492 vstream->index_sent_seqcount,
1493 vstream->stream->stream_handle);
1494 }
1495 end:
1496 if (metadata_viewer_stream) {
1497 viewer_stream_put(metadata_viewer_stream);
1498 }
1499 if (vstream) {
1500 viewer_stream_put(vstream);
1501 }
1502 return ret;
1503
1504 error_put:
1505 pthread_mutex_unlock(&rstream->lock);
1506 if (metadata_viewer_stream) {
1507 viewer_stream_put(metadata_viewer_stream);
1508 }
1509 viewer_stream_put(vstream);
1510 return ret;
1511 }
1512
1513 /*
1514 * Send the next index for a stream
1515 *
1516 * Return 0 on success or else a negative value.
1517 */
1518 static
1519 int viewer_get_packet(struct relay_connection *conn)
1520 {
1521 int ret;
1522 off_t lseek_ret;
1523 char *reply = NULL;
1524 struct lttng_viewer_get_packet get_packet_info;
1525 struct lttng_viewer_trace_packet reply_header;
1526 struct relay_viewer_stream *vstream = NULL;
1527 uint32_t reply_size = sizeof(reply_header);
1528 uint32_t packet_data_len = 0;
1529 ssize_t read_len;
1530
1531 DBG2("Relay get data packet");
1532
1533 health_code_update();
1534
1535 ret = recv_request(conn->sock, &get_packet_info,
1536 sizeof(get_packet_info));
1537 if (ret < 0) {
1538 goto end;
1539 }
1540 health_code_update();
1541
1542 /* From this point on, the error label can be reached. */
1543 memset(&reply_header, 0, sizeof(reply_header));
1544
1545 vstream = viewer_stream_get_by_id(be64toh(get_packet_info.stream_id));
1546 if (!vstream) {
1547 DBG("Client requested packet of unknown stream id %" PRIu64,
1548 (uint64_t) be64toh(get_packet_info.stream_id));
1549 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1550 goto send_reply_nolock;
1551 } else {
1552 packet_data_len = be32toh(get_packet_info.len);
1553 reply_size += packet_data_len;
1554 }
1555
1556 reply = zmalloc(reply_size);
1557 if (!reply) {
1558 PERROR("packet reply zmalloc");
1559 reply_size = sizeof(reply_header);
1560 goto error;
1561 }
1562
1563 pthread_mutex_lock(&vstream->stream->lock);
1564 lseek_ret = lseek(vstream->stream_file.fd->fd,
1565 be64toh(get_packet_info.offset), SEEK_SET);
1566 if (lseek_ret < 0) {
1567 PERROR("lseek fd %d to offset %" PRIu64,
1568 vstream->stream_file.fd->fd,
1569 (uint64_t) be64toh(get_packet_info.offset));
1570 goto error;
1571 }
1572 read_len = lttng_read(vstream->stream_file.fd->fd,
1573 reply + sizeof(reply_header), packet_data_len);
1574 if (read_len < packet_data_len) {
1575 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1576 vstream->stream_file.fd->fd,
1577 (uint64_t) be64toh(get_packet_info.offset));
1578 goto error;
1579 }
1580 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1581 reply_header.len = htobe32(packet_data_len);
1582 goto send_reply;
1583
1584 error:
1585 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1586
1587 send_reply:
1588 if (vstream) {
1589 pthread_mutex_unlock(&vstream->stream->lock);
1590 }
1591 send_reply_nolock:
1592
1593 health_code_update();
1594
1595 if (reply) {
1596 memcpy(reply, &reply_header, sizeof(reply_header));
1597 ret = send_response(conn->sock, reply, reply_size);
1598 } else {
1599 /* No reply to send. */
1600 ret = send_response(conn->sock, &reply_header,
1601 reply_size);
1602 }
1603
1604 health_code_update();
1605 if (ret < 0) {
1606 PERROR("sendmsg of packet data failed");
1607 goto end_free;
1608 }
1609
1610 DBG("Sent %u bytes for stream %" PRIu64, reply_size,
1611 (uint64_t) be64toh(get_packet_info.stream_id));
1612
1613 end_free:
1614 free(reply);
1615 end:
1616 if (vstream) {
1617 viewer_stream_put(vstream);
1618 }
1619 return ret;
1620 }
1621
1622 /*
1623 * Send the session's metadata
1624 *
1625 * Return 0 on success else a negative value.
1626 */
1627 static
1628 int viewer_get_metadata(struct relay_connection *conn)
1629 {
1630 int ret = 0;
1631 ssize_t read_len;
1632 uint64_t len = 0;
1633 char *data = NULL;
1634 struct lttng_viewer_get_metadata request;
1635 struct lttng_viewer_metadata_packet reply;
1636 struct relay_viewer_stream *vstream = NULL;
1637
1638 assert(conn);
1639
1640 DBG("Relay get metadata");
1641
1642 health_code_update();
1643
1644 ret = recv_request(conn->sock, &request, sizeof(request));
1645 if (ret < 0) {
1646 goto end;
1647 }
1648 health_code_update();
1649
1650 memset(&reply, 0, sizeof(reply));
1651
1652 vstream = viewer_stream_get_by_id(be64toh(request.stream_id));
1653 if (!vstream) {
1654 /*
1655 * The metadata stream can be closed by a CLOSE command
1656 * just before we attach. It can also be closed by
1657 * per-pid tracing during tracing. Therefore, it is
1658 * possible that we cannot find this viewer stream.
1659 * Reply back to the client with an error if we cannot
1660 * find it.
1661 */
1662 DBG("Client requested metadata of unknown stream id %" PRIu64,
1663 (uint64_t) be64toh(request.stream_id));
1664 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1665 goto send_reply;
1666 }
1667 pthread_mutex_lock(&vstream->stream->lock);
1668 if (!vstream->stream->is_metadata) {
1669 ERR("Invalid metadata stream");
1670 goto error;
1671 }
1672
1673 assert(vstream->metadata_sent <= vstream->stream->metadata_received);
1674
1675 len = vstream->stream->metadata_received - vstream->metadata_sent;
1676 if (len == 0) {
1677 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1678 goto send_reply;
1679 }
1680
1681 /* first time, we open the metadata file */
1682 if (!vstream->stream_file.fd) {
1683 int fd;
1684 char file_path[LTTNG_PATH_MAX];
1685 enum lttng_trace_chunk_status status;
1686 struct relay_stream *rstream = vstream->stream;
1687
1688 ret = utils_stream_file_path(rstream->path_name,
1689 rstream->channel_name, rstream->tracefile_size,
1690 vstream->current_tracefile_id, NULL, file_path,
1691 sizeof(file_path));
1692 if (ret < 0) {
1693 goto error;
1694 }
1695
1696 status = lttng_trace_chunk_open_file(
1697 vstream->stream_file.trace_chunk,
1698 file_path, O_RDONLY, 0, &fd);
1699 if (status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1700 PERROR("Failed to open metadata file for viewer stream");
1701 goto error;
1702 }
1703 vstream->stream_file.fd = stream_fd_create(fd);
1704 if (!vstream->stream_file.fd) {
1705 if (close(fd)) {
1706 PERROR("Failed to close viewer metadata file");
1707 }
1708 goto error;
1709 }
1710 }
1711
1712 reply.len = htobe64(len);
1713 data = zmalloc(len);
1714 if (!data) {
1715 PERROR("viewer metadata zmalloc");
1716 goto error;
1717 }
1718
1719 read_len = lttng_read(vstream->stream_file.fd->fd, data, len);
1720 if (read_len < len) {
1721 PERROR("Relay reading metadata file");
1722 goto error;
1723 }
1724 vstream->metadata_sent += read_len;
1725 if (vstream->metadata_sent == vstream->stream->metadata_received
1726 && vstream->stream->closed) {
1727 /* Release ownership for the viewer metadata stream. */
1728 viewer_stream_put(vstream);
1729 }
1730
1731 reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1732
1733 goto send_reply;
1734
1735 error:
1736 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1737
1738 send_reply:
1739 health_code_update();
1740 if (vstream) {
1741 pthread_mutex_unlock(&vstream->stream->lock);
1742 }
1743 ret = send_response(conn->sock, &reply, sizeof(reply));
1744 if (ret < 0) {
1745 goto end_free;
1746 }
1747 health_code_update();
1748
1749 if (len > 0) {
1750 ret = send_response(conn->sock, data, len);
1751 if (ret < 0) {
1752 goto end_free;
1753 }
1754 }
1755
1756 DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1757 (uint64_t) be64toh(request.stream_id));
1758
1759 DBG("Metadata sent");
1760
1761 end_free:
1762 free(data);
1763 end:
1764 if (vstream) {
1765 viewer_stream_put(vstream);
1766 }
1767 return ret;
1768 }
1769
1770 /*
1771 * Create a viewer session.
1772 *
1773 * Return 0 on success or else a negative value.
1774 */
1775 static
1776 int viewer_create_session(struct relay_connection *conn)
1777 {
1778 int ret;
1779 struct lttng_viewer_create_session_response resp;
1780
1781 DBG("Viewer create session received");
1782
1783 memset(&resp, 0, sizeof(resp));
1784 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1785 conn->viewer_session = viewer_session_create();
1786 if (!conn->viewer_session) {
1787 ERR("Allocation viewer session");
1788 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1789 goto send_reply;
1790 }
1791
1792 send_reply:
1793 health_code_update();
1794 ret = send_response(conn->sock, &resp, sizeof(resp));
1795 if (ret < 0) {
1796 goto end;
1797 }
1798 health_code_update();
1799 ret = 0;
1800
1801 end:
1802 return ret;
1803 }
1804
1805 /*
1806 * Detach a viewer session.
1807 *
1808 * Return 0 on success or else a negative value.
1809 */
1810 static
1811 int viewer_detach_session(struct relay_connection *conn)
1812 {
1813 int ret;
1814 struct lttng_viewer_detach_session_response response;
1815 struct lttng_viewer_detach_session_request request;
1816 struct relay_session *session = NULL;
1817 uint64_t viewer_session_to_close;
1818
1819 DBG("Viewer detach session received");
1820
1821 assert(conn);
1822
1823 health_code_update();
1824
1825 /* Receive the request from the connected client. */
1826 ret = recv_request(conn->sock, &request, sizeof(request));
1827 if (ret < 0) {
1828 goto end;
1829 }
1830 viewer_session_to_close = be64toh(request.session_id);
1831
1832 if (!conn->viewer_session) {
1833 DBG("Client trying to detach before creating a live viewer session");
1834 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
1835 goto send_reply;
1836 }
1837
1838 health_code_update();
1839
1840 memset(&response, 0, sizeof(response));
1841 DBG("Detaching from session ID %" PRIu64, viewer_session_to_close);
1842
1843 session = session_get_by_id(be64toh(request.session_id));
1844 if (!session) {
1845 DBG("Relay session %" PRIu64 " not found",
1846 (uint64_t) be64toh(request.session_id));
1847 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_UNK);
1848 goto send_reply;
1849 }
1850
1851 ret = viewer_session_is_attached(conn->viewer_session, session);
1852 if (ret != 1) {
1853 DBG("Not attached to this session");
1854 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
1855 goto send_reply_put;
1856 }
1857
1858 viewer_session_close_one_session(conn->viewer_session, session);
1859 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_OK);
1860 DBG("Session %" PRIu64 " detached.", viewer_session_to_close);
1861
1862 send_reply_put:
1863 session_put(session);
1864
1865 send_reply:
1866 health_code_update();
1867 ret = send_response(conn->sock, &response, sizeof(response));
1868 if (ret < 0) {
1869 goto end;
1870 }
1871 health_code_update();
1872 ret = 0;
1873
1874 end:
1875 return ret;
1876 }
1877
1878 /*
1879 * live_relay_unknown_command: send -1 if received unknown command
1880 */
1881 static
1882 void live_relay_unknown_command(struct relay_connection *conn)
1883 {
1884 struct lttcomm_relayd_generic_reply reply;
1885
1886 memset(&reply, 0, sizeof(reply));
1887 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1888 (void) send_response(conn->sock, &reply, sizeof(reply));
1889 }
1890
1891 /*
1892 * Process the commands received on the control socket
1893 */
1894 static
1895 int process_control(struct lttng_viewer_cmd *recv_hdr,
1896 struct relay_connection *conn)
1897 {
1898 int ret = 0;
1899 uint32_t msg_value;
1900
1901 msg_value = be32toh(recv_hdr->cmd);
1902
1903 /*
1904 * Make sure we've done the version check before any command other then a
1905 * new client connection.
1906 */
1907 if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1908 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1909 ret = -1;
1910 goto end;
1911 }
1912
1913 switch (msg_value) {
1914 case LTTNG_VIEWER_CONNECT:
1915 ret = viewer_connect(conn);
1916 break;
1917 case LTTNG_VIEWER_LIST_SESSIONS:
1918 ret = viewer_list_sessions(conn);
1919 break;
1920 case LTTNG_VIEWER_ATTACH_SESSION:
1921 ret = viewer_attach_session(conn);
1922 break;
1923 case LTTNG_VIEWER_GET_NEXT_INDEX:
1924 ret = viewer_get_next_index(conn);
1925 break;
1926 case LTTNG_VIEWER_GET_PACKET:
1927 ret = viewer_get_packet(conn);
1928 break;
1929 case LTTNG_VIEWER_GET_METADATA:
1930 ret = viewer_get_metadata(conn);
1931 break;
1932 case LTTNG_VIEWER_GET_NEW_STREAMS:
1933 ret = viewer_get_new_streams(conn);
1934 break;
1935 case LTTNG_VIEWER_CREATE_SESSION:
1936 ret = viewer_create_session(conn);
1937 break;
1938 case LTTNG_VIEWER_DETACH_SESSION:
1939 ret = viewer_detach_session(conn);
1940 break;
1941 default:
1942 ERR("Received unknown viewer command (%u)",
1943 be32toh(recv_hdr->cmd));
1944 live_relay_unknown_command(conn);
1945 ret = -1;
1946 goto end;
1947 }
1948
1949 end:
1950 return ret;
1951 }
1952
1953 static
1954 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1955 {
1956 int ret;
1957
1958 (void) lttng_poll_del(events, pollfd);
1959
1960 ret = close(pollfd);
1961 if (ret < 0) {
1962 ERR("Closing pollfd %d", pollfd);
1963 }
1964 }
1965
1966 /*
1967 * This thread does the actual work
1968 */
1969 static
1970 void *thread_worker(void *data)
1971 {
1972 int ret, err = -1;
1973 uint32_t nb_fd;
1974 struct lttng_poll_event events;
1975 struct lttng_ht *viewer_connections_ht;
1976 struct lttng_ht_iter iter;
1977 struct lttng_viewer_cmd recv_hdr;
1978 struct relay_connection *destroy_conn;
1979
1980 DBG("[thread] Live viewer relay worker started");
1981
1982 rcu_register_thread();
1983
1984 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1985
1986 if (testpoint(relayd_thread_live_worker)) {
1987 goto error_testpoint;
1988 }
1989
1990 /* table of connections indexed on socket */
1991 viewer_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1992 if (!viewer_connections_ht) {
1993 goto viewer_connections_ht_error;
1994 }
1995
1996 ret = create_thread_poll_set(&events, 2);
1997 if (ret < 0) {
1998 goto error_poll_create;
1999 }
2000
2001 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
2002 if (ret < 0) {
2003 goto error;
2004 }
2005
2006 restart:
2007 while (1) {
2008 int i;
2009
2010 health_code_update();
2011
2012 /* Infinite blocking call, waiting for transmission */
2013 DBG3("Relayd live viewer worker thread polling...");
2014 health_poll_entry();
2015 ret = lttng_poll_wait(&events, -1);
2016 health_poll_exit();
2017 if (ret < 0) {
2018 /*
2019 * Restart interrupted system call.
2020 */
2021 if (errno == EINTR) {
2022 goto restart;
2023 }
2024 goto error;
2025 }
2026
2027 nb_fd = ret;
2028
2029 /*
2030 * Process control. The control connection is prioritised so we don't
2031 * starve it with high throughput tracing data on the data
2032 * connection.
2033 */
2034 for (i = 0; i < nb_fd; i++) {
2035 /* Fetch once the poll data */
2036 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2037 int pollfd = LTTNG_POLL_GETFD(&events, i);
2038
2039 health_code_update();
2040
2041 /* Thread quit pipe has been closed. Killing thread. */
2042 ret = check_thread_quit_pipe(pollfd, revents);
2043 if (ret) {
2044 err = 0;
2045 goto exit;
2046 }
2047
2048 /* Inspect the relay conn pipe for new connection. */
2049 if (pollfd == live_conn_pipe[0]) {
2050 if (revents & LPOLLIN) {
2051 struct relay_connection *conn;
2052
2053 ret = lttng_read(live_conn_pipe[0],
2054 &conn, sizeof(conn));
2055 if (ret < 0) {
2056 goto error;
2057 }
2058 lttng_poll_add(&events, conn->sock->fd,
2059 LPOLLIN | LPOLLRDHUP);
2060 connection_ht_add(viewer_connections_ht, conn);
2061 DBG("Connection socket %d added to poll", conn->sock->fd);
2062 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2063 ERR("Relay live pipe error");
2064 goto error;
2065 } else {
2066 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2067 goto error;
2068 }
2069 } else {
2070 /* Connection activity. */
2071 struct relay_connection *conn;
2072
2073 conn = connection_get_by_sock(viewer_connections_ht, pollfd);
2074 if (!conn) {
2075 continue;
2076 }
2077
2078 if (revents & LPOLLIN) {
2079 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
2080 sizeof(recv_hdr), 0);
2081 if (ret <= 0) {
2082 /* Connection closed. */
2083 cleanup_connection_pollfd(&events, pollfd);
2084 /* Put "create" ownership reference. */
2085 connection_put(conn);
2086 DBG("Viewer control conn closed with %d", pollfd);
2087 } else {
2088 ret = process_control(&recv_hdr, conn);
2089 if (ret < 0) {
2090 /* Clear the session on error. */
2091 cleanup_connection_pollfd(&events, pollfd);
2092 /* Put "create" ownership reference. */
2093 connection_put(conn);
2094 DBG("Viewer connection closed with %d", pollfd);
2095 }
2096 }
2097 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2098 cleanup_connection_pollfd(&events, pollfd);
2099 /* Put "create" ownership reference. */
2100 connection_put(conn);
2101 } else {
2102 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2103 connection_put(conn);
2104 goto error;
2105 }
2106 /* Put local "get_by_sock" reference. */
2107 connection_put(conn);
2108 }
2109 }
2110 }
2111
2112 exit:
2113 error:
2114 lttng_poll_clean(&events);
2115
2116 /* Cleanup remaining connection object. */
2117 rcu_read_lock();
2118 cds_lfht_for_each_entry(viewer_connections_ht->ht, &iter.iter,
2119 destroy_conn,
2120 sock_n.node) {
2121 health_code_update();
2122 connection_put(destroy_conn);
2123 }
2124 rcu_read_unlock();
2125 error_poll_create:
2126 lttng_ht_destroy(viewer_connections_ht);
2127 viewer_connections_ht_error:
2128 /* Close relay conn pipes */
2129 utils_close_pipe(live_conn_pipe);
2130 if (err) {
2131 DBG("Viewer worker thread exited with error");
2132 }
2133 DBG("Viewer worker thread cleanup complete");
2134 error_testpoint:
2135 if (err) {
2136 health_error();
2137 ERR("Health error occurred in %s", __func__);
2138 }
2139 health_unregister(health_relayd);
2140 if (lttng_relay_stop_threads()) {
2141 ERR("Error stopping threads");
2142 }
2143 rcu_unregister_thread();
2144 return NULL;
2145 }
2146
2147 /*
2148 * Create the relay command pipe to wake thread_manage_apps.
2149 * Closed in cleanup().
2150 */
2151 static int create_conn_pipe(void)
2152 {
2153 return utils_create_pipe_cloexec(live_conn_pipe);
2154 }
2155
2156 int relayd_live_join(void)
2157 {
2158 int ret, retval = 0;
2159 void *status;
2160
2161 ret = pthread_join(live_listener_thread, &status);
2162 if (ret) {
2163 errno = ret;
2164 PERROR("pthread_join live listener");
2165 retval = -1;
2166 }
2167
2168 ret = pthread_join(live_worker_thread, &status);
2169 if (ret) {
2170 errno = ret;
2171 PERROR("pthread_join live worker");
2172 retval = -1;
2173 }
2174
2175 ret = pthread_join(live_dispatcher_thread, &status);
2176 if (ret) {
2177 errno = ret;
2178 PERROR("pthread_join live dispatcher");
2179 retval = -1;
2180 }
2181
2182 cleanup_relayd_live();
2183
2184 return retval;
2185 }
2186
2187 /*
2188 * main
2189 */
2190 int relayd_live_create(struct lttng_uri *uri)
2191 {
2192 int ret = 0, retval = 0;
2193 void *status;
2194 int is_root;
2195
2196 if (!uri) {
2197 retval = -1;
2198 goto exit_init_data;
2199 }
2200 live_uri = uri;
2201
2202 /* Check if daemon is UID = 0 */
2203 is_root = !getuid();
2204
2205 if (!is_root) {
2206 if (live_uri->port < 1024) {
2207 ERR("Need to be root to use ports < 1024");
2208 retval = -1;
2209 goto exit_init_data;
2210 }
2211 }
2212
2213 /* Setup the thread apps communication pipe. */
2214 if (create_conn_pipe()) {
2215 retval = -1;
2216 goto exit_init_data;
2217 }
2218
2219 /* Init relay command queue. */
2220 cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2221
2222 /* Set up max poll set size */
2223 if (lttng_poll_set_max_size()) {
2224 retval = -1;
2225 goto exit_init_data;
2226 }
2227
2228 /* Setup the dispatcher thread */
2229 ret = pthread_create(&live_dispatcher_thread, default_pthread_attr(),
2230 thread_dispatcher, (void *) NULL);
2231 if (ret) {
2232 errno = ret;
2233 PERROR("pthread_create viewer dispatcher");
2234 retval = -1;
2235 goto exit_dispatcher_thread;
2236 }
2237
2238 /* Setup the worker thread */
2239 ret = pthread_create(&live_worker_thread, default_pthread_attr(),
2240 thread_worker, NULL);
2241 if (ret) {
2242 errno = ret;
2243 PERROR("pthread_create viewer worker");
2244 retval = -1;
2245 goto exit_worker_thread;
2246 }
2247
2248 /* Setup the listener thread */
2249 ret = pthread_create(&live_listener_thread, default_pthread_attr(),
2250 thread_listener, (void *) NULL);
2251 if (ret) {
2252 errno = ret;
2253 PERROR("pthread_create viewer listener");
2254 retval = -1;
2255 goto exit_listener_thread;
2256 }
2257
2258 /*
2259 * All OK, started all threads.
2260 */
2261 return retval;
2262
2263 /*
2264 * Join on the live_listener_thread should anything be added after
2265 * the live_listener thread's creation.
2266 */
2267
2268 exit_listener_thread:
2269
2270 ret = pthread_join(live_worker_thread, &status);
2271 if (ret) {
2272 errno = ret;
2273 PERROR("pthread_join live worker");
2274 retval = -1;
2275 }
2276 exit_worker_thread:
2277
2278 ret = pthread_join(live_dispatcher_thread, &status);
2279 if (ret) {
2280 errno = ret;
2281 PERROR("pthread_join live dispatcher");
2282 retval = -1;
2283 }
2284 exit_dispatcher_thread:
2285
2286 exit_init_data:
2287 cleanup_relayd_live();
2288
2289 return retval;
2290 }
This page took 0.128873 seconds and 3 git commands to generate.