Fix: skip closed session on viewer listing
[lttng-tools.git] / src / bin / lttng-relayd / live.c
1 /*
2 * Copyright (C) 2013 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
4 * 2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _LGPL_SOURCE
21 #include <getopt.h>
22 #include <grp.h>
23 #include <limits.h>
24 #include <pthread.h>
25 #include <signal.h>
26 #include <stdio.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <sys/mman.h>
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
33 #include <sys/stat.h>
34 #include <sys/types.h>
35 #include <sys/wait.h>
36 #include <inttypes.h>
37 #include <urcu/futex.h>
38 #include <urcu/uatomic.h>
39 #include <urcu/rculist.h>
40 #include <unistd.h>
41 #include <fcntl.h>
42
43 #include <lttng/lttng.h>
44 #include <common/common.h>
45 #include <common/compat/poll.h>
46 #include <common/compat/socket.h>
47 #include <common/compat/endian.h>
48 #include <common/defaults.h>
49 #include <common/futex.h>
50 #include <common/index/index.h>
51 #include <common/sessiond-comm/sessiond-comm.h>
52 #include <common/sessiond-comm/inet.h>
53 #include <common/sessiond-comm/relayd.h>
54 #include <common/uri.h>
55 #include <common/utils.h>
56
57 #include "cmd.h"
58 #include "live.h"
59 #include "lttng-relayd.h"
60 #include "utils.h"
61 #include "health-relayd.h"
62 #include "testpoint.h"
63 #include "viewer-stream.h"
64 #include "stream.h"
65 #include "session.h"
66 #include "ctf-trace.h"
67 #include "connection.h"
68 #include "viewer-session.h"
69
70 #define SESSION_BUF_DEFAULT_COUNT 16
71
72 static struct lttng_uri *live_uri;
73
74 /*
75 * This pipe is used to inform the worker thread that a command is queued and
76 * ready to be processed.
77 */
78 static int live_conn_pipe[2] = { -1, -1 };
79
80 /* Shared between threads */
81 static int live_dispatch_thread_exit;
82
83 static pthread_t live_listener_thread;
84 static pthread_t live_dispatcher_thread;
85 static pthread_t live_worker_thread;
86
87 /*
88 * Relay command queue.
89 *
90 * The live_thread_listener and live_thread_dispatcher communicate with this
91 * queue.
92 */
93 static struct relay_conn_queue viewer_conn_queue;
94
95 static uint64_t last_relay_viewer_session_id;
96 static pthread_mutex_t last_relay_viewer_session_id_lock =
97 PTHREAD_MUTEX_INITIALIZER;
98
99 /*
100 * Cleanup the daemon
101 */
102 static
103 void cleanup_relayd_live(void)
104 {
105 DBG("Cleaning up");
106
107 free(live_uri);
108 }
109
110 /*
111 * Receive a request buffer using a given socket, destination allocated buffer
112 * of length size.
113 *
114 * Return the size of the received message or else a negative value on error
115 * with errno being set by recvmsg() syscall.
116 */
117 static
118 ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
119 {
120 ssize_t ret;
121
122 ret = sock->ops->recvmsg(sock, buf, size, 0);
123 if (ret < 0 || ret != size) {
124 if (ret == 0) {
125 /* Orderly shutdown. Not necessary to print an error. */
126 DBG("Socket %d did an orderly shutdown", sock->fd);
127 } else {
128 ERR("Relay failed to receive request.");
129 }
130 ret = -1;
131 }
132
133 return ret;
134 }
135
136 /*
137 * Send a response buffer using a given socket, source allocated buffer of
138 * length size.
139 *
140 * Return the size of the sent message or else a negative value on error with
141 * errno being set by sendmsg() syscall.
142 */
143 static
144 ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
145 {
146 ssize_t ret;
147
148 ret = sock->ops->sendmsg(sock, buf, size, 0);
149 if (ret < 0) {
150 ERR("Relayd failed to send response.");
151 }
152
153 return ret;
154 }
155
156 /*
157 * Atomically check if new streams got added in one of the sessions attached
158 * and reset the flag to 0.
159 *
160 * Returns 1 if new streams got added, 0 if nothing changed, a negative value
161 * on error.
162 */
163 static
164 int check_new_streams(struct relay_connection *conn)
165 {
166 struct relay_session *session;
167 unsigned long current_val;
168 int ret = 0;
169
170 if (!conn->viewer_session) {
171 goto end;
172 }
173 rcu_read_lock();
174 cds_list_for_each_entry_rcu(session,
175 &conn->viewer_session->session_list,
176 viewer_session_node) {
177 if (!session_get(session)) {
178 continue;
179 }
180 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
181 ret = current_val;
182 session_put(session);
183 if (ret == 1) {
184 goto end;
185 }
186 }
187 end:
188 rcu_read_unlock();
189 return ret;
190 }
191
192 /*
193 * Send viewer streams to the given socket. The ignore_sent_flag indicates if
194 * this function should ignore the sent flag or not.
195 *
196 * Return 0 on success or else a negative value.
197 */
198 static
199 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
200 struct relay_session *session, unsigned int ignore_sent_flag)
201 {
202 ssize_t ret;
203 struct lttng_viewer_stream send_stream;
204 struct lttng_ht_iter iter;
205 struct relay_viewer_stream *vstream;
206
207 rcu_read_lock();
208
209 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
210 stream_n.node) {
211 struct ctf_trace *ctf_trace;
212
213 health_code_update();
214
215 if (!viewer_stream_get(vstream)) {
216 continue;
217 }
218
219 pthread_mutex_lock(&vstream->stream->lock);
220 /* Ignore if not the same session. */
221 if (vstream->stream->trace->session->id != session->id ||
222 (!ignore_sent_flag && vstream->sent_flag)) {
223 pthread_mutex_unlock(&vstream->stream->lock);
224 viewer_stream_put(vstream);
225 continue;
226 }
227
228 ctf_trace = vstream->stream->trace;
229 send_stream.id = htobe64(vstream->stream->stream_handle);
230 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
231 send_stream.metadata_flag = htobe32(
232 vstream->stream->is_metadata);
233 if (lttng_strncpy(send_stream.path_name, vstream->path_name,
234 sizeof(send_stream.path_name))) {
235 pthread_mutex_unlock(&vstream->stream->lock);
236 viewer_stream_put(vstream);
237 ret = -1; /* Error. */
238 goto end_unlock;
239 }
240 if (lttng_strncpy(send_stream.channel_name,
241 vstream->channel_name,
242 sizeof(send_stream.channel_name))) {
243 pthread_mutex_unlock(&vstream->stream->lock);
244 viewer_stream_put(vstream);
245 ret = -1; /* Error. */
246 goto end_unlock;
247 }
248
249 DBG("Sending stream %" PRIu64 " to viewer",
250 vstream->stream->stream_handle);
251 vstream->sent_flag = 1;
252 pthread_mutex_unlock(&vstream->stream->lock);
253
254 ret = send_response(sock, &send_stream, sizeof(send_stream));
255 viewer_stream_put(vstream);
256 if (ret < 0) {
257 goto end_unlock;
258 }
259 }
260
261 ret = 0;
262
263 end_unlock:
264 rcu_read_unlock();
265 return ret;
266 }
267
268 /*
269 * Create every viewer stream possible for the given session with the seek
270 * type. Three counters *can* be return which are in order the total amount of
271 * viewer stream of the session, the number of unsent stream and the number of
272 * stream created. Those counters can be NULL and thus will be ignored.
273 *
274 * Return 0 on success or else a negative value.
275 */
276 static
277 int make_viewer_streams(struct relay_session *session,
278 enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent,
279 uint32_t *nb_created, bool *closed)
280 {
281 int ret;
282 struct lttng_ht_iter iter;
283 struct ctf_trace *ctf_trace;
284
285 assert(session);
286
287 /*
288 * Hold the session lock to ensure that we see either none or
289 * all initial streams for a session, but no intermediate state.
290 */
291 pthread_mutex_lock(&session->lock);
292
293 if (session->connection_closed) {
294 *closed = true;
295 }
296
297 /*
298 * Create viewer streams for relay streams that are ready to be
299 * used for a the given session id only.
300 */
301 rcu_read_lock();
302 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
303 node.node) {
304 struct relay_stream *stream;
305
306 health_code_update();
307
308 if (!ctf_trace_get(ctf_trace)) {
309 continue;
310 }
311
312 cds_list_for_each_entry_rcu(stream, &ctf_trace->stream_list, stream_node) {
313 struct relay_viewer_stream *vstream;
314
315 if (!stream_get(stream)) {
316 continue;
317 }
318 /*
319 * stream published is protected by the session lock.
320 */
321 if (!stream->published) {
322 goto next;
323 }
324 vstream = viewer_stream_get_by_id(stream->stream_handle);
325 if (!vstream) {
326 vstream = viewer_stream_create(stream, seek_t);
327 if (!vstream) {
328 ret = -1;
329 ctf_trace_put(ctf_trace);
330 stream_put(stream);
331 goto error_unlock;
332 }
333
334 if (nb_created) {
335 /* Update number of created stream counter. */
336 (*nb_created)++;
337 }
338 /*
339 * Ensure a self-reference is preserved even
340 * after we have put our local reference.
341 */
342 if (!viewer_stream_get(vstream)) {
343 ERR("Unable to get self-reference on viewer stream, logic error.");
344 abort();
345 }
346 } else {
347 if (!vstream->sent_flag && nb_unsent) {
348 /* Update number of unsent stream counter. */
349 (*nb_unsent)++;
350 }
351 }
352 /* Update number of total stream counter. */
353 if (nb_total) {
354 if (stream->is_metadata) {
355 if (!stream->closed ||
356 stream->metadata_received > vstream->metadata_sent) {
357 (*nb_total)++;
358 }
359 } else {
360 if (!stream->closed ||
361 !(((int64_t) (stream->prev_seq - stream->last_net_seq_num)) >= 0)) {
362
363 (*nb_total)++;
364 }
365 }
366 }
367 /* Put local reference. */
368 viewer_stream_put(vstream);
369 next:
370 stream_put(stream);
371 }
372 ctf_trace_put(ctf_trace);
373 }
374
375 ret = 0;
376
377 error_unlock:
378 rcu_read_unlock();
379 pthread_mutex_unlock(&session->lock);
380 return ret;
381 }
382
383 int relayd_live_stop(void)
384 {
385 /* Stop dispatch thread */
386 CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
387 futex_nto1_wake(&viewer_conn_queue.futex);
388 return 0;
389 }
390
391 /*
392 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
393 */
394 static
395 int create_thread_poll_set(struct lttng_poll_event *events, int size)
396 {
397 int ret;
398
399 if (events == NULL || size == 0) {
400 ret = -1;
401 goto error;
402 }
403
404 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
405 if (ret < 0) {
406 goto error;
407 }
408
409 /* Add quit pipe */
410 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
411 if (ret < 0) {
412 goto error;
413 }
414
415 return 0;
416
417 error:
418 return ret;
419 }
420
421 /*
422 * Check if the thread quit pipe was triggered.
423 *
424 * Return 1 if it was triggered else 0;
425 */
426 static
427 int check_thread_quit_pipe(int fd, uint32_t events)
428 {
429 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
430 return 1;
431 }
432
433 return 0;
434 }
435
436 /*
437 * Create and init socket from uri.
438 */
439 static
440 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
441 {
442 int ret;
443 struct lttcomm_sock *sock = NULL;
444
445 sock = lttcomm_alloc_sock_from_uri(uri);
446 if (sock == NULL) {
447 ERR("Allocating socket");
448 goto error;
449 }
450
451 ret = lttcomm_create_sock(sock);
452 if (ret < 0) {
453 goto error;
454 }
455 DBG("Listening on sock %d for live", sock->fd);
456
457 ret = sock->ops->bind(sock);
458 if (ret < 0) {
459 goto error;
460 }
461
462 ret = sock->ops->listen(sock, -1);
463 if (ret < 0) {
464 goto error;
465
466 }
467
468 return sock;
469
470 error:
471 if (sock) {
472 lttcomm_destroy_sock(sock);
473 }
474 return NULL;
475 }
476
477 /*
478 * This thread manages the listening for new connections on the network
479 */
480 static
481 void *thread_listener(void *data)
482 {
483 int i, ret, pollfd, err = -1;
484 uint32_t revents, nb_fd;
485 struct lttng_poll_event events;
486 struct lttcomm_sock *live_control_sock;
487
488 DBG("[thread] Relay live listener started");
489
490 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
491
492 health_code_update();
493
494 live_control_sock = init_socket(live_uri);
495 if (!live_control_sock) {
496 goto error_sock_control;
497 }
498
499 /* Pass 2 as size here for the thread quit pipe and control sockets. */
500 ret = create_thread_poll_set(&events, 2);
501 if (ret < 0) {
502 goto error_create_poll;
503 }
504
505 /* Add the control socket */
506 ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
507 if (ret < 0) {
508 goto error_poll_add;
509 }
510
511 lttng_relay_notify_ready();
512
513 if (testpoint(relayd_thread_live_listener)) {
514 goto error_testpoint;
515 }
516
517 while (1) {
518 health_code_update();
519
520 DBG("Listener accepting live viewers connections");
521
522 restart:
523 health_poll_entry();
524 ret = lttng_poll_wait(&events, -1);
525 health_poll_exit();
526 if (ret < 0) {
527 /*
528 * Restart interrupted system call.
529 */
530 if (errno == EINTR) {
531 goto restart;
532 }
533 goto error;
534 }
535 nb_fd = ret;
536
537 DBG("Relay new viewer connection received");
538 for (i = 0; i < nb_fd; i++) {
539 health_code_update();
540
541 /* Fetch once the poll data */
542 revents = LTTNG_POLL_GETEV(&events, i);
543 pollfd = LTTNG_POLL_GETFD(&events, i);
544
545 if (!revents) {
546 /* No activity for this FD (poll implementation). */
547 continue;
548 }
549
550 /* Thread quit pipe has been closed. Killing thread. */
551 ret = check_thread_quit_pipe(pollfd, revents);
552 if (ret) {
553 err = 0;
554 goto exit;
555 }
556
557 if (revents & LPOLLIN) {
558 /*
559 * A new connection is requested, therefore a
560 * viewer connection is allocated in this
561 * thread, enqueued to a global queue and
562 * dequeued (and freed) in the worker thread.
563 */
564 int val = 1;
565 struct relay_connection *new_conn;
566 struct lttcomm_sock *newsock;
567
568 newsock = live_control_sock->ops->accept(live_control_sock);
569 if (!newsock) {
570 PERROR("accepting control sock");
571 goto error;
572 }
573 DBG("Relay viewer connection accepted socket %d", newsock->fd);
574
575 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
576 sizeof(val));
577 if (ret < 0) {
578 PERROR("setsockopt inet");
579 lttcomm_destroy_sock(newsock);
580 goto error;
581 }
582 new_conn = connection_create(newsock, RELAY_CONNECTION_UNKNOWN);
583 if (!new_conn) {
584 lttcomm_destroy_sock(newsock);
585 goto error;
586 }
587 /* Ownership assumed by the connection. */
588 newsock = NULL;
589
590 /* Enqueue request for the dispatcher thread. */
591 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
592 &new_conn->qnode);
593
594 /*
595 * Wake the dispatch queue futex.
596 * Implicit memory barrier with the
597 * exchange in cds_wfcq_enqueue.
598 */
599 futex_nto1_wake(&viewer_conn_queue.futex);
600 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
601 ERR("socket poll error");
602 goto error;
603 } else {
604 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
605 goto error;
606 }
607 }
608 }
609
610 exit:
611 error:
612 error_poll_add:
613 error_testpoint:
614 lttng_poll_clean(&events);
615 error_create_poll:
616 if (live_control_sock->fd >= 0) {
617 ret = live_control_sock->ops->close(live_control_sock);
618 if (ret) {
619 PERROR("close");
620 }
621 }
622 lttcomm_destroy_sock(live_control_sock);
623 error_sock_control:
624 if (err) {
625 health_error();
626 DBG("Live viewer listener thread exited with error");
627 }
628 health_unregister(health_relayd);
629 DBG("Live viewer listener thread cleanup complete");
630 if (lttng_relay_stop_threads()) {
631 ERR("Error stopping threads");
632 }
633 return NULL;
634 }
635
636 /*
637 * This thread manages the dispatching of the requests to worker threads
638 */
639 static
640 void *thread_dispatcher(void *data)
641 {
642 int err = -1;
643 ssize_t ret;
644 struct cds_wfcq_node *node;
645 struct relay_connection *conn = NULL;
646
647 DBG("[thread] Live viewer relay dispatcher started");
648
649 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
650
651 if (testpoint(relayd_thread_live_dispatcher)) {
652 goto error_testpoint;
653 }
654
655 health_code_update();
656
657 for (;;) {
658 health_code_update();
659
660 /* Atomically prepare the queue futex */
661 futex_nto1_prepare(&viewer_conn_queue.futex);
662
663 if (CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
664 break;
665 }
666
667 do {
668 health_code_update();
669
670 /* Dequeue commands */
671 node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
672 &viewer_conn_queue.tail);
673 if (node == NULL) {
674 DBG("Woken up but nothing in the live-viewer "
675 "relay command queue");
676 /* Continue thread execution */
677 break;
678 }
679 conn = caa_container_of(node, struct relay_connection, qnode);
680 DBG("Dispatching viewer request waiting on sock %d",
681 conn->sock->fd);
682
683 /*
684 * Inform worker thread of the new request. This
685 * call is blocking so we can be assured that
686 * the data will be read at some point in time
687 * or wait to the end of the world :)
688 */
689 ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
690 if (ret < 0) {
691 PERROR("write conn pipe");
692 connection_put(conn);
693 goto error;
694 }
695 } while (node != NULL);
696
697 /* Futex wait on queue. Blocking call on futex() */
698 health_poll_entry();
699 futex_nto1_wait(&viewer_conn_queue.futex);
700 health_poll_exit();
701 }
702
703 /* Normal exit, no error */
704 err = 0;
705
706 error:
707 error_testpoint:
708 if (err) {
709 health_error();
710 ERR("Health error occurred in %s", __func__);
711 }
712 health_unregister(health_relayd);
713 DBG("Live viewer dispatch thread dying");
714 if (lttng_relay_stop_threads()) {
715 ERR("Error stopping threads");
716 }
717 return NULL;
718 }
719
720 /*
721 * Establish connection with the viewer and check the versions.
722 *
723 * Return 0 on success or else negative value.
724 */
725 static
726 int viewer_connect(struct relay_connection *conn)
727 {
728 int ret;
729 struct lttng_viewer_connect reply, msg;
730
731 conn->version_check_done = 1;
732
733 health_code_update();
734
735 DBG("Viewer is establishing a connection to the relayd.");
736
737 ret = recv_request(conn->sock, &msg, sizeof(msg));
738 if (ret < 0) {
739 goto end;
740 }
741
742 health_code_update();
743
744 memset(&reply, 0, sizeof(reply));
745 reply.major = RELAYD_VERSION_COMM_MAJOR;
746 reply.minor = RELAYD_VERSION_COMM_MINOR;
747
748 /* Major versions must be the same */
749 if (reply.major != be32toh(msg.major)) {
750 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
751 reply.major, be32toh(msg.major));
752 ret = -1;
753 goto end;
754 }
755
756 conn->major = reply.major;
757 /* We adapt to the lowest compatible version */
758 if (reply.minor <= be32toh(msg.minor)) {
759 conn->minor = reply.minor;
760 } else {
761 conn->minor = be32toh(msg.minor);
762 }
763
764 if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
765 conn->type = RELAY_VIEWER_COMMAND;
766 } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
767 conn->type = RELAY_VIEWER_NOTIFICATION;
768 } else {
769 ERR("Unknown connection type : %u", be32toh(msg.type));
770 ret = -1;
771 goto end;
772 }
773
774 reply.major = htobe32(reply.major);
775 reply.minor = htobe32(reply.minor);
776 if (conn->type == RELAY_VIEWER_COMMAND) {
777 /*
778 * Increment outside of htobe64 macro, because the argument can
779 * be used more than once within the macro, and thus the
780 * operation may be undefined.
781 */
782 pthread_mutex_lock(&last_relay_viewer_session_id_lock);
783 last_relay_viewer_session_id++;
784 pthread_mutex_unlock(&last_relay_viewer_session_id_lock);
785 reply.viewer_session_id = htobe64(last_relay_viewer_session_id);
786 }
787
788 health_code_update();
789
790 ret = send_response(conn->sock, &reply, sizeof(reply));
791 if (ret < 0) {
792 goto end;
793 }
794
795 health_code_update();
796
797 DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
798 ret = 0;
799
800 end:
801 return ret;
802 }
803
804 /*
805 * Send the viewer the list of current sessions.
806 * We need to create a copy of the hash table content because otherwise
807 * we cannot assume the number of entries stays the same between getting
808 * the number of HT elements and iteration over the HT.
809 *
810 * Return 0 on success or else a negative value.
811 */
812 static
813 int viewer_list_sessions(struct relay_connection *conn)
814 {
815 int ret = 0;
816 struct lttng_viewer_list_sessions session_list;
817 struct lttng_ht_iter iter;
818 struct relay_session *session;
819 struct lttng_viewer_session *send_session_buf = NULL;
820 uint32_t buf_count = SESSION_BUF_DEFAULT_COUNT;
821 uint32_t count = 0;
822
823 DBG("List sessions received");
824
825 send_session_buf = zmalloc(SESSION_BUF_DEFAULT_COUNT * sizeof(*send_session_buf));
826 if (!send_session_buf) {
827 return -1;
828 }
829
830 rcu_read_lock();
831 cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, session,
832 session_n.node) {
833 struct lttng_viewer_session *send_session;
834
835 health_code_update();
836
837 if (session->connection_closed) {
838 /* Skip closed session */
839 continue;
840 }
841
842 if (count >= buf_count) {
843 struct lttng_viewer_session *newbuf;
844 uint32_t new_buf_count = buf_count << 1;
845
846 newbuf = realloc(send_session_buf,
847 new_buf_count * sizeof(*send_session_buf));
848 if (!newbuf) {
849 ret = -1;
850 break;
851 }
852 send_session_buf = newbuf;
853 buf_count = new_buf_count;
854 }
855 send_session = &send_session_buf[count];
856 if (lttng_strncpy(send_session->session_name,
857 session->session_name,
858 sizeof(send_session->session_name))) {
859 ret = -1;
860 break;
861 }
862 if (lttng_strncpy(send_session->hostname, session->hostname,
863 sizeof(send_session->hostname))) {
864 ret = -1;
865 break;
866 }
867 send_session->id = htobe64(session->id);
868 send_session->live_timer = htobe32(session->live_timer);
869 if (session->viewer_attached) {
870 send_session->clients = htobe32(1);
871 } else {
872 send_session->clients = htobe32(0);
873 }
874 send_session->streams = htobe32(session->stream_count);
875 count++;
876 }
877 rcu_read_unlock();
878 if (ret < 0) {
879 goto end_free;
880 }
881
882 session_list.sessions_count = htobe32(count);
883
884 health_code_update();
885
886 ret = send_response(conn->sock, &session_list, sizeof(session_list));
887 if (ret < 0) {
888 goto end_free;
889 }
890
891 health_code_update();
892
893 ret = send_response(conn->sock, send_session_buf,
894 count * sizeof(*send_session_buf));
895 if (ret < 0) {
896 goto end_free;
897 }
898 health_code_update();
899
900 ret = 0;
901 end_free:
902 free(send_session_buf);
903 return ret;
904 }
905
906 /*
907 * Send the viewer the list of current streams.
908 */
909 static
910 int viewer_get_new_streams(struct relay_connection *conn)
911 {
912 int ret, send_streams = 0;
913 uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0, nb_total = 0;
914 struct lttng_viewer_new_streams_request request;
915 struct lttng_viewer_new_streams_response response;
916 struct relay_session *session;
917 uint64_t session_id;
918 bool closed = false;
919
920 assert(conn);
921
922 DBG("Get new streams received");
923
924 health_code_update();
925
926 /* Receive the request from the connected client. */
927 ret = recv_request(conn->sock, &request, sizeof(request));
928 if (ret < 0) {
929 goto error;
930 }
931 session_id = be64toh(request.session_id);
932
933 health_code_update();
934
935 memset(&response, 0, sizeof(response));
936
937 session = session_get_by_id(session_id);
938 if (!session) {
939 DBG("Relay session %" PRIu64 " not found", session_id);
940 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
941 goto send_reply;
942 }
943
944 if (!viewer_session_is_attached(conn->viewer_session, session)) {
945 send_streams = 0;
946 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
947 goto send_reply;
948 }
949
950 send_streams = 1;
951 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
952
953 ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, &nb_total, &nb_unsent,
954 &nb_created, &closed);
955 if (ret < 0) {
956 goto end_put_session;
957 }
958 /* Only send back the newly created streams with the unsent ones. */
959 nb_streams = nb_created + nb_unsent;
960 response.streams_count = htobe32(nb_streams);
961
962 /*
963 * If the session is closed, HUP when there are no more streams
964 * with data.
965 */
966 if (closed && nb_total == 0) {
967 send_streams = 0;
968 response.streams_count = 0;
969 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
970 goto send_reply;
971 }
972
973 send_reply:
974 health_code_update();
975 ret = send_response(conn->sock, &response, sizeof(response));
976 if (ret < 0) {
977 goto end_put_session;
978 }
979 health_code_update();
980
981 /*
982 * Unknown or empty session, just return gracefully, the viewer
983 * knows what is happening.
984 */
985 if (!send_streams || !nb_streams) {
986 ret = 0;
987 goto end_put_session;
988 }
989
990 /*
991 * Send stream and *DON'T* ignore the sent flag so every viewer
992 * streams that were not sent from that point will be sent to
993 * the viewer.
994 */
995 ret = send_viewer_streams(conn->sock, session, 0);
996 if (ret < 0) {
997 goto end_put_session;
998 }
999
1000 end_put_session:
1001 if (session) {
1002 session_put(session);
1003 }
1004 error:
1005 return ret;
1006 }
1007
1008 /*
1009 * Send the viewer the list of current sessions.
1010 */
1011 static
1012 int viewer_attach_session(struct relay_connection *conn)
1013 {
1014 int send_streams = 0;
1015 ssize_t ret;
1016 uint32_t nb_streams = 0;
1017 enum lttng_viewer_seek seek_type;
1018 struct lttng_viewer_attach_session_request request;
1019 struct lttng_viewer_attach_session_response response;
1020 struct relay_session *session = NULL;
1021 bool closed = false;
1022
1023 assert(conn);
1024
1025 health_code_update();
1026
1027 /* Receive the request from the connected client. */
1028 ret = recv_request(conn->sock, &request, sizeof(request));
1029 if (ret < 0) {
1030 goto error;
1031 }
1032
1033 health_code_update();
1034
1035 memset(&response, 0, sizeof(response));
1036
1037 if (!conn->viewer_session) {
1038 DBG("Client trying to attach before creating a live viewer session");
1039 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1040 goto send_reply;
1041 }
1042
1043 session = session_get_by_id(be64toh(request.session_id));
1044 if (!session) {
1045 DBG("Relay session %" PRIu64 " not found",
1046 (uint64_t) be64toh(request.session_id));
1047 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1048 goto send_reply;
1049 }
1050 DBG("Attach session ID %" PRIu64 " received",
1051 (uint64_t) be64toh(request.session_id));
1052
1053 if (session->live_timer == 0) {
1054 DBG("Not live session");
1055 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1056 goto send_reply;
1057 }
1058
1059 send_streams = 1;
1060 ret = viewer_session_attach(conn->viewer_session, session);
1061 if (ret) {
1062 DBG("Already a viewer attached");
1063 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1064 goto send_reply;
1065 }
1066
1067 switch (be32toh(request.seek)) {
1068 case LTTNG_VIEWER_SEEK_BEGINNING:
1069 case LTTNG_VIEWER_SEEK_LAST:
1070 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1071 seek_type = be32toh(request.seek);
1072 break;
1073 default:
1074 ERR("Wrong seek parameter");
1075 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1076 send_streams = 0;
1077 goto send_reply;
1078 }
1079
1080 ret = make_viewer_streams(session, seek_type, &nb_streams, NULL,
1081 NULL, &closed);
1082 if (ret < 0) {
1083 goto end_put_session;
1084 }
1085 response.streams_count = htobe32(nb_streams);
1086
1087 /*
1088 * If the session is closed when the viewer is attaching, it
1089 * means some of the streams may have been concurrently removed,
1090 * so we don't allow the viewer to attach, even if there are
1091 * streams available.
1092 */
1093 if (closed) {
1094 send_streams = 0;
1095 response.streams_count = 0;
1096 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1097 goto send_reply;
1098 }
1099
1100 send_reply:
1101 health_code_update();
1102 ret = send_response(conn->sock, &response, sizeof(response));
1103 if (ret < 0) {
1104 goto end_put_session;
1105 }
1106 health_code_update();
1107
1108 /*
1109 * Unknown or empty session, just return gracefully, the viewer
1110 * knows what is happening.
1111 */
1112 if (!send_streams || !nb_streams) {
1113 ret = 0;
1114 goto end_put_session;
1115 }
1116
1117 /* Send stream and ignore the sent flag. */
1118 ret = send_viewer_streams(conn->sock, session, 1);
1119 if (ret < 0) {
1120 goto end_put_session;
1121 }
1122
1123 end_put_session:
1124 if (session) {
1125 session_put(session);
1126 }
1127 error:
1128 return ret;
1129 }
1130
1131 /*
1132 * Open the index file if needed for the given vstream.
1133 *
1134 * If an index file is successfully opened, the vstream will set it as its
1135 * current index file.
1136 *
1137 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1138 *
1139 * Called with rstream lock held.
1140 */
1141 static int try_open_index(struct relay_viewer_stream *vstream,
1142 struct relay_stream *rstream)
1143 {
1144 int ret = 0;
1145
1146 if (vstream->index_file) {
1147 goto end;
1148 }
1149
1150 /*
1151 * First time, we open the index file and at least one index is ready.
1152 */
1153 if (rstream->index_received_seqcount == 0) {
1154 ret = -ENOENT;
1155 goto end;
1156 }
1157 vstream->index_file = lttng_index_file_open(vstream->path_name,
1158 vstream->channel_name,
1159 vstream->stream->tracefile_count,
1160 vstream->current_tracefile_id);
1161 if (!vstream->index_file) {
1162 ret = -1;
1163 }
1164
1165 end:
1166 return ret;
1167 }
1168
1169 /*
1170 * Check the status of the index for the given stream. This function
1171 * updates the index structure if needed and can put (close) the vstream
1172 * in the HUP situation.
1173 *
1174 * Return 0 means that we can proceed with the index. A value of 1 means
1175 * that the index has been updated and is ready to be sent to the
1176 * client. A negative value indicates an error that can't be handled.
1177 *
1178 * Called with rstream lock held.
1179 */
1180 static int check_index_status(struct relay_viewer_stream *vstream,
1181 struct relay_stream *rstream, struct ctf_trace *trace,
1182 struct lttng_viewer_index *index)
1183 {
1184 int ret;
1185
1186 if (trace->session->connection_closed
1187 && rstream->index_received_seqcount
1188 == vstream->index_sent_seqcount) {
1189 /* Last index sent and session connection is closed. */
1190 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1191 goto hup;
1192 } else if (rstream->beacon_ts_end != -1ULL &&
1193 rstream->index_received_seqcount
1194 == vstream->index_sent_seqcount) {
1195 /*
1196 * We've received a synchronization beacon and the last index
1197 * available has been sent, the index for now is inactive.
1198 *
1199 * In this case, we have received a beacon which allows us to
1200 * inform the client of a time interval during which we can
1201 * guarantee that there are no events to read (and never will
1202 * be).
1203 */
1204 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1205 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1206 index->stream_id = htobe64(rstream->ctf_stream_id);
1207 goto index_ready;
1208 } else if (rstream->index_received_seqcount
1209 == vstream->index_sent_seqcount) {
1210 /*
1211 * This checks whether received == sent seqcount. In
1212 * this case, we have not received a beacon. Therefore,
1213 * we can only ask the client to retry later.
1214 */
1215 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1216 goto index_ready;
1217 } else if (!tracefile_array_seq_in_file(rstream->tfa,
1218 vstream->current_tracefile_id,
1219 vstream->index_sent_seqcount)) {
1220 /*
1221 * The next index we want to send cannot be read either
1222 * because we need to perform a rotation, or due to
1223 * the producer having overwritten its trace file.
1224 */
1225 DBG("Viewer stream %" PRIu64 " rotation",
1226 vstream->stream->stream_handle);
1227 ret = viewer_stream_rotate(vstream);
1228 if (ret < 0) {
1229 goto end;
1230 } else if (ret == 1) {
1231 /* EOF across entire stream. */
1232 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1233 goto hup;
1234 }
1235 /*
1236 * If we have been pushed due to overwrite, it
1237 * necessarily means there is data that can be read in
1238 * the stream. If we rotated because we reached the end
1239 * of a tracefile, it means the following tracefile
1240 * needs to contain at least one index, else we would
1241 * have already returned LTTNG_VIEWER_INDEX_RETRY to the
1242 * viewer. The updated index_sent_seqcount needs to
1243 * point to a readable index entry now.
1244 *
1245 * In the case where we "rotate" on a single file, we
1246 * can end up in a case where the requested index is
1247 * still unavailable.
1248 */
1249 if (rstream->tracefile_count == 1 &&
1250 !tracefile_array_seq_in_file(
1251 rstream->tfa,
1252 vstream->current_tracefile_id,
1253 vstream->index_sent_seqcount)) {
1254 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1255 goto index_ready;
1256 }
1257 assert(tracefile_array_seq_in_file(rstream->tfa,
1258 vstream->current_tracefile_id,
1259 vstream->index_sent_seqcount));
1260 }
1261 /* ret == 0 means successful so we continue. */
1262 ret = 0;
1263 end:
1264 return ret;
1265
1266 hup:
1267 viewer_stream_put(vstream);
1268 index_ready:
1269 return 1;
1270 }
1271
1272 /*
1273 * Send the next index for a stream.
1274 *
1275 * Return 0 on success or else a negative value.
1276 */
1277 static
1278 int viewer_get_next_index(struct relay_connection *conn)
1279 {
1280 int ret;
1281 struct lttng_viewer_get_next_index request_index;
1282 struct lttng_viewer_index viewer_index;
1283 struct ctf_packet_index packet_index;
1284 struct relay_viewer_stream *vstream = NULL;
1285 struct relay_stream *rstream = NULL;
1286 struct ctf_trace *ctf_trace = NULL;
1287 struct relay_viewer_stream *metadata_viewer_stream = NULL;
1288
1289 assert(conn);
1290
1291 DBG("Viewer get next index");
1292
1293 memset(&viewer_index, 0, sizeof(viewer_index));
1294 health_code_update();
1295
1296 ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1297 if (ret < 0) {
1298 goto end;
1299 }
1300 health_code_update();
1301
1302 vstream = viewer_stream_get_by_id(be64toh(request_index.stream_id));
1303 if (!vstream) {
1304 DBG("Client requested index of unknown stream id %" PRIu64,
1305 (uint64_t) be64toh(request_index.stream_id));
1306 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1307 goto send_reply;
1308 }
1309
1310 /* Use back. ref. Protected by refcounts. */
1311 rstream = vstream->stream;
1312 ctf_trace = rstream->trace;
1313
1314 /* metadata_viewer_stream may be NULL. */
1315 metadata_viewer_stream =
1316 ctf_trace_get_viewer_metadata_stream(ctf_trace);
1317
1318 pthread_mutex_lock(&rstream->lock);
1319
1320 /*
1321 * The viewer should not ask for index on metadata stream.
1322 */
1323 if (rstream->is_metadata) {
1324 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1325 goto send_reply;
1326 }
1327
1328 /* Try to open an index if one is needed for that stream. */
1329 ret = try_open_index(vstream, rstream);
1330 if (ret < 0) {
1331 if (ret == -ENOENT) {
1332 /*
1333 * The index is created only when the first data
1334 * packet arrives, it might not be ready at the
1335 * beginning of the session
1336 */
1337 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1338 } else {
1339 /* Unhandled error. */
1340 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1341 }
1342 goto send_reply;
1343 }
1344
1345 ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1346 if (ret < 0) {
1347 goto error_put;
1348 } else if (ret == 1) {
1349 /*
1350 * We have no index to send and check_index_status has populated
1351 * viewer_index's status.
1352 */
1353 goto send_reply;
1354 }
1355 /* At this point, ret is 0 thus we will be able to read the index. */
1356 assert(!ret);
1357
1358 /*
1359 * vstream->stream_fd may be NULL if it has been closed by
1360 * tracefile rotation, or if we are at the beginning of the
1361 * stream. We open the data stream file here to protect against
1362 * overwrite caused by tracefile rotation (in association with
1363 * unlink performed before overwrite).
1364 */
1365 if (!vstream->stream_fd) {
1366 char fullpath[PATH_MAX];
1367
1368 if (vstream->stream->tracefile_count > 0) {
1369 ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64,
1370 vstream->path_name,
1371 vstream->channel_name,
1372 vstream->current_tracefile_id);
1373 } else {
1374 ret = snprintf(fullpath, PATH_MAX, "%s/%s",
1375 vstream->path_name,
1376 vstream->channel_name);
1377 }
1378 if (ret < 0) {
1379 goto error_put;
1380 }
1381 ret = open(fullpath, O_RDONLY);
1382 if (ret < 0) {
1383 PERROR("Relay opening trace file");
1384 goto error_put;
1385 }
1386 vstream->stream_fd = stream_fd_create(ret);
1387 if (!vstream->stream_fd) {
1388 if (close(ret)) {
1389 PERROR("close");
1390 }
1391 goto error_put;
1392 }
1393 }
1394
1395 ret = check_new_streams(conn);
1396 if (ret < 0) {
1397 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1398 goto send_reply;
1399 } else if (ret == 1) {
1400 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1401 }
1402
1403 ret = lttng_index_file_read(vstream->index_file, &packet_index);
1404 if (ret) {
1405 ERR("Relay error reading index file %d",
1406 vstream->index_file->fd);
1407 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1408 goto send_reply;
1409 } else {
1410 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1411 vstream->index_sent_seqcount++;
1412 }
1413
1414 /*
1415 * Indexes are stored in big endian, no need to switch before sending.
1416 */
1417 DBG("Sending viewer index for stream %" PRIu64 " offset %" PRIu64,
1418 rstream->stream_handle,
1419 (uint64_t) be64toh(packet_index.offset));
1420 viewer_index.offset = packet_index.offset;
1421 viewer_index.packet_size = packet_index.packet_size;
1422 viewer_index.content_size = packet_index.content_size;
1423 viewer_index.timestamp_begin = packet_index.timestamp_begin;
1424 viewer_index.timestamp_end = packet_index.timestamp_end;
1425 viewer_index.events_discarded = packet_index.events_discarded;
1426 viewer_index.stream_id = packet_index.stream_id;
1427
1428 send_reply:
1429 if (rstream) {
1430 pthread_mutex_unlock(&rstream->lock);
1431 }
1432
1433 if (metadata_viewer_stream) {
1434 pthread_mutex_lock(&metadata_viewer_stream->stream->lock);
1435 DBG("get next index metadata check: recv %" PRIu64
1436 " sent %" PRIu64,
1437 metadata_viewer_stream->stream->metadata_received,
1438 metadata_viewer_stream->metadata_sent);
1439 if (!metadata_viewer_stream->stream->metadata_received ||
1440 metadata_viewer_stream->stream->metadata_received >
1441 metadata_viewer_stream->metadata_sent) {
1442 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1443 }
1444 pthread_mutex_unlock(&metadata_viewer_stream->stream->lock);
1445 }
1446
1447 viewer_index.flags = htobe32(viewer_index.flags);
1448 health_code_update();
1449
1450 ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1451 if (ret < 0) {
1452 goto end;
1453 }
1454 health_code_update();
1455
1456 if (vstream) {
1457 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1458 vstream->index_sent_seqcount,
1459 vstream->stream->stream_handle);
1460 }
1461 end:
1462 if (metadata_viewer_stream) {
1463 viewer_stream_put(metadata_viewer_stream);
1464 }
1465 if (vstream) {
1466 viewer_stream_put(vstream);
1467 }
1468 return ret;
1469
1470 error_put:
1471 pthread_mutex_unlock(&rstream->lock);
1472 if (metadata_viewer_stream) {
1473 viewer_stream_put(metadata_viewer_stream);
1474 }
1475 viewer_stream_put(vstream);
1476 return ret;
1477 }
1478
1479 /*
1480 * Send the next index for a stream
1481 *
1482 * Return 0 on success or else a negative value.
1483 */
1484 static
1485 int viewer_get_packet(struct relay_connection *conn)
1486 {
1487 int ret;
1488 off_t lseek_ret;
1489 char *reply = NULL;
1490 struct lttng_viewer_get_packet get_packet_info;
1491 struct lttng_viewer_trace_packet reply_header;
1492 struct relay_viewer_stream *vstream = NULL;
1493 uint32_t reply_size = sizeof(reply_header);
1494 uint32_t packet_data_len = 0;
1495 ssize_t read_len;
1496
1497 DBG2("Relay get data packet");
1498
1499 health_code_update();
1500
1501 ret = recv_request(conn->sock, &get_packet_info,
1502 sizeof(get_packet_info));
1503 if (ret < 0) {
1504 goto end;
1505 }
1506 health_code_update();
1507
1508 /* From this point on, the error label can be reached. */
1509 memset(&reply_header, 0, sizeof(reply_header));
1510
1511 vstream = viewer_stream_get_by_id(be64toh(get_packet_info.stream_id));
1512 if (!vstream) {
1513 DBG("Client requested packet of unknown stream id %" PRIu64,
1514 (uint64_t) be64toh(get_packet_info.stream_id));
1515 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1516 goto send_reply_nolock;
1517 } else {
1518 packet_data_len = be32toh(get_packet_info.len);
1519 reply_size += packet_data_len;
1520 }
1521
1522 reply = zmalloc(reply_size);
1523 if (!reply) {
1524 PERROR("packet reply zmalloc");
1525 reply_size = sizeof(reply_header);
1526 goto error;
1527 }
1528
1529 pthread_mutex_lock(&vstream->stream->lock);
1530 lseek_ret = lseek(vstream->stream_fd->fd, be64toh(get_packet_info.offset),
1531 SEEK_SET);
1532 if (lseek_ret < 0) {
1533 PERROR("lseek fd %d to offset %" PRIu64, vstream->stream_fd->fd,
1534 (uint64_t) be64toh(get_packet_info.offset));
1535 goto error;
1536 }
1537 read_len = lttng_read(vstream->stream_fd->fd,
1538 reply + sizeof(reply_header),
1539 packet_data_len);
1540 if (read_len < packet_data_len) {
1541 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1542 vstream->stream_fd->fd,
1543 (uint64_t) be64toh(get_packet_info.offset));
1544 goto error;
1545 }
1546 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1547 reply_header.len = htobe32(packet_data_len);
1548 goto send_reply;
1549
1550 error:
1551 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1552
1553 send_reply:
1554 if (vstream) {
1555 pthread_mutex_unlock(&vstream->stream->lock);
1556 }
1557 send_reply_nolock:
1558
1559 health_code_update();
1560
1561 if (reply) {
1562 memcpy(reply, &reply_header, sizeof(reply_header));
1563 ret = send_response(conn->sock, reply, reply_size);
1564 } else {
1565 /* No reply to send. */
1566 ret = send_response(conn->sock, &reply_header,
1567 reply_size);
1568 }
1569
1570 health_code_update();
1571 if (ret < 0) {
1572 PERROR("sendmsg of packet data failed");
1573 goto end_free;
1574 }
1575
1576 DBG("Sent %u bytes for stream %" PRIu64, reply_size,
1577 (uint64_t) be64toh(get_packet_info.stream_id));
1578
1579 end_free:
1580 free(reply);
1581 end:
1582 if (vstream) {
1583 viewer_stream_put(vstream);
1584 }
1585 return ret;
1586 }
1587
1588 /*
1589 * Send the session's metadata
1590 *
1591 * Return 0 on success else a negative value.
1592 */
1593 static
1594 int viewer_get_metadata(struct relay_connection *conn)
1595 {
1596 int ret = 0;
1597 ssize_t read_len;
1598 uint64_t len = 0;
1599 char *data = NULL;
1600 struct lttng_viewer_get_metadata request;
1601 struct lttng_viewer_metadata_packet reply;
1602 struct relay_viewer_stream *vstream = NULL;
1603
1604 assert(conn);
1605
1606 DBG("Relay get metadata");
1607
1608 health_code_update();
1609
1610 ret = recv_request(conn->sock, &request, sizeof(request));
1611 if (ret < 0) {
1612 goto end;
1613 }
1614 health_code_update();
1615
1616 memset(&reply, 0, sizeof(reply));
1617
1618 vstream = viewer_stream_get_by_id(be64toh(request.stream_id));
1619 if (!vstream) {
1620 /*
1621 * The metadata stream can be closed by a CLOSE command
1622 * just before we attach. It can also be closed by
1623 * per-pid tracing during tracing. Therefore, it is
1624 * possible that we cannot find this viewer stream.
1625 * Reply back to the client with an error if we cannot
1626 * find it.
1627 */
1628 DBG("Client requested metadata of unknown stream id %" PRIu64,
1629 (uint64_t) be64toh(request.stream_id));
1630 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1631 goto send_reply;
1632 }
1633 pthread_mutex_lock(&vstream->stream->lock);
1634 if (!vstream->stream->is_metadata) {
1635 ERR("Invalid metadata stream");
1636 goto error;
1637 }
1638
1639 assert(vstream->metadata_sent <= vstream->stream->metadata_received);
1640
1641 len = vstream->stream->metadata_received - vstream->metadata_sent;
1642 if (len == 0) {
1643 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1644 goto send_reply;
1645 }
1646
1647 /* first time, we open the metadata file */
1648 if (!vstream->stream_fd) {
1649 char fullpath[PATH_MAX];
1650
1651 ret = snprintf(fullpath, PATH_MAX, "%s/%s", vstream->path_name,
1652 vstream->channel_name);
1653 if (ret < 0) {
1654 goto error;
1655 }
1656 ret = open(fullpath, O_RDONLY);
1657 if (ret < 0) {
1658 PERROR("Relay opening metadata file");
1659 goto error;
1660 }
1661 vstream->stream_fd = stream_fd_create(ret);
1662 if (!vstream->stream_fd) {
1663 if (close(ret)) {
1664 PERROR("close");
1665 }
1666 goto error;
1667 }
1668 }
1669
1670 reply.len = htobe64(len);
1671 data = zmalloc(len);
1672 if (!data) {
1673 PERROR("viewer metadata zmalloc");
1674 goto error;
1675 }
1676
1677 read_len = lttng_read(vstream->stream_fd->fd, data, len);
1678 if (read_len < len) {
1679 PERROR("Relay reading metadata file");
1680 goto error;
1681 }
1682 vstream->metadata_sent += read_len;
1683 if (vstream->metadata_sent == vstream->stream->metadata_received
1684 && vstream->stream->closed) {
1685 /* Release ownership for the viewer metadata stream. */
1686 viewer_stream_put(vstream);
1687 }
1688
1689 reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1690
1691 goto send_reply;
1692
1693 error:
1694 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1695
1696 send_reply:
1697 health_code_update();
1698 if (vstream) {
1699 pthread_mutex_unlock(&vstream->stream->lock);
1700 }
1701 ret = send_response(conn->sock, &reply, sizeof(reply));
1702 if (ret < 0) {
1703 goto end_free;
1704 }
1705 health_code_update();
1706
1707 if (len > 0) {
1708 ret = send_response(conn->sock, data, len);
1709 if (ret < 0) {
1710 goto end_free;
1711 }
1712 }
1713
1714 DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1715 (uint64_t) be64toh(request.stream_id));
1716
1717 DBG("Metadata sent");
1718
1719 end_free:
1720 free(data);
1721 end:
1722 if (vstream) {
1723 viewer_stream_put(vstream);
1724 }
1725 return ret;
1726 }
1727
1728 /*
1729 * Create a viewer session.
1730 *
1731 * Return 0 on success or else a negative value.
1732 */
1733 static
1734 int viewer_create_session(struct relay_connection *conn)
1735 {
1736 int ret;
1737 struct lttng_viewer_create_session_response resp;
1738
1739 DBG("Viewer create session received");
1740
1741 memset(&resp, 0, sizeof(resp));
1742 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1743 conn->viewer_session = viewer_session_create();
1744 if (!conn->viewer_session) {
1745 ERR("Allocation viewer session");
1746 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1747 goto send_reply;
1748 }
1749
1750 send_reply:
1751 health_code_update();
1752 ret = send_response(conn->sock, &resp, sizeof(resp));
1753 if (ret < 0) {
1754 goto end;
1755 }
1756 health_code_update();
1757 ret = 0;
1758
1759 end:
1760 return ret;
1761 }
1762
1763 /*
1764 * Detach a viewer session.
1765 *
1766 * Return 0 on success or else a negative value.
1767 */
1768 static
1769 int viewer_detach_session(struct relay_connection *conn)
1770 {
1771 int ret;
1772 struct lttng_viewer_detach_session_response response;
1773 struct lttng_viewer_detach_session_request request;
1774 struct relay_session *session = NULL;
1775 uint64_t viewer_session_to_close;
1776
1777 DBG("Viewer detach session received");
1778
1779 assert(conn);
1780
1781 health_code_update();
1782
1783 /* Receive the request from the connected client. */
1784 ret = recv_request(conn->sock, &request, sizeof(request));
1785 if (ret < 0) {
1786 goto end;
1787 }
1788 viewer_session_to_close = be64toh(request.session_id);
1789
1790 if (!conn->viewer_session) {
1791 DBG("Client trying to detach before creating a live viewer session");
1792 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
1793 goto send_reply;
1794 }
1795
1796 health_code_update();
1797
1798 memset(&response, 0, sizeof(response));
1799 DBG("Detaching from session ID %" PRIu64, viewer_session_to_close);
1800
1801 session = session_get_by_id(be64toh(request.session_id));
1802 if (!session) {
1803 DBG("Relay session %" PRIu64 " not found",
1804 (uint64_t) be64toh(request.session_id));
1805 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_UNK);
1806 goto send_reply;
1807 }
1808
1809 ret = viewer_session_is_attached(conn->viewer_session, session);
1810 if (ret != 1) {
1811 DBG("Not attached to this session");
1812 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
1813 goto send_reply_put;
1814 }
1815
1816 viewer_session_close_one_session(conn->viewer_session, session);
1817 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_OK);
1818 DBG("Session %" PRIu64 " detached.", viewer_session_to_close);
1819
1820 send_reply_put:
1821 session_put(session);
1822
1823 send_reply:
1824 health_code_update();
1825 ret = send_response(conn->sock, &response, sizeof(response));
1826 if (ret < 0) {
1827 goto end;
1828 }
1829 health_code_update();
1830 ret = 0;
1831
1832 end:
1833 return ret;
1834 }
1835
1836 /*
1837 * live_relay_unknown_command: send -1 if received unknown command
1838 */
1839 static
1840 void live_relay_unknown_command(struct relay_connection *conn)
1841 {
1842 struct lttcomm_relayd_generic_reply reply;
1843
1844 memset(&reply, 0, sizeof(reply));
1845 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1846 (void) send_response(conn->sock, &reply, sizeof(reply));
1847 }
1848
1849 /*
1850 * Process the commands received on the control socket
1851 */
1852 static
1853 int process_control(struct lttng_viewer_cmd *recv_hdr,
1854 struct relay_connection *conn)
1855 {
1856 int ret = 0;
1857 uint32_t msg_value;
1858
1859 msg_value = be32toh(recv_hdr->cmd);
1860
1861 /*
1862 * Make sure we've done the version check before any command other then a
1863 * new client connection.
1864 */
1865 if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1866 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1867 ret = -1;
1868 goto end;
1869 }
1870
1871 switch (msg_value) {
1872 case LTTNG_VIEWER_CONNECT:
1873 ret = viewer_connect(conn);
1874 break;
1875 case LTTNG_VIEWER_LIST_SESSIONS:
1876 ret = viewer_list_sessions(conn);
1877 break;
1878 case LTTNG_VIEWER_ATTACH_SESSION:
1879 ret = viewer_attach_session(conn);
1880 break;
1881 case LTTNG_VIEWER_GET_NEXT_INDEX:
1882 ret = viewer_get_next_index(conn);
1883 break;
1884 case LTTNG_VIEWER_GET_PACKET:
1885 ret = viewer_get_packet(conn);
1886 break;
1887 case LTTNG_VIEWER_GET_METADATA:
1888 ret = viewer_get_metadata(conn);
1889 break;
1890 case LTTNG_VIEWER_GET_NEW_STREAMS:
1891 ret = viewer_get_new_streams(conn);
1892 break;
1893 case LTTNG_VIEWER_CREATE_SESSION:
1894 ret = viewer_create_session(conn);
1895 break;
1896 case LTTNG_VIEWER_DETACH_SESSION:
1897 ret = viewer_detach_session(conn);
1898 break;
1899 default:
1900 ERR("Received unknown viewer command (%u)",
1901 be32toh(recv_hdr->cmd));
1902 live_relay_unknown_command(conn);
1903 ret = -1;
1904 goto end;
1905 }
1906
1907 end:
1908 return ret;
1909 }
1910
1911 static
1912 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1913 {
1914 int ret;
1915
1916 (void) lttng_poll_del(events, pollfd);
1917
1918 ret = close(pollfd);
1919 if (ret < 0) {
1920 ERR("Closing pollfd %d", pollfd);
1921 }
1922 }
1923
1924 /*
1925 * This thread does the actual work
1926 */
1927 static
1928 void *thread_worker(void *data)
1929 {
1930 int ret, err = -1;
1931 uint32_t nb_fd;
1932 struct lttng_poll_event events;
1933 struct lttng_ht *viewer_connections_ht;
1934 struct lttng_ht_iter iter;
1935 struct lttng_viewer_cmd recv_hdr;
1936 struct relay_connection *destroy_conn;
1937
1938 DBG("[thread] Live viewer relay worker started");
1939
1940 rcu_register_thread();
1941
1942 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1943
1944 if (testpoint(relayd_thread_live_worker)) {
1945 goto error_testpoint;
1946 }
1947
1948 /* table of connections indexed on socket */
1949 viewer_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1950 if (!viewer_connections_ht) {
1951 goto viewer_connections_ht_error;
1952 }
1953
1954 ret = create_thread_poll_set(&events, 2);
1955 if (ret < 0) {
1956 goto error_poll_create;
1957 }
1958
1959 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1960 if (ret < 0) {
1961 goto error;
1962 }
1963
1964 restart:
1965 while (1) {
1966 int i;
1967
1968 health_code_update();
1969
1970 /* Infinite blocking call, waiting for transmission */
1971 DBG3("Relayd live viewer worker thread polling...");
1972 health_poll_entry();
1973 ret = lttng_poll_wait(&events, -1);
1974 health_poll_exit();
1975 if (ret < 0) {
1976 /*
1977 * Restart interrupted system call.
1978 */
1979 if (errno == EINTR) {
1980 goto restart;
1981 }
1982 goto error;
1983 }
1984
1985 nb_fd = ret;
1986
1987 /*
1988 * Process control. The control connection is prioritised so we don't
1989 * starve it with high throughput tracing data on the data
1990 * connection.
1991 */
1992 for (i = 0; i < nb_fd; i++) {
1993 /* Fetch once the poll data */
1994 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1995 int pollfd = LTTNG_POLL_GETFD(&events, i);
1996
1997 health_code_update();
1998
1999 if (!revents) {
2000 /* No activity for this FD (poll implementation). */
2001 continue;
2002 }
2003
2004 /* Thread quit pipe has been closed. Killing thread. */
2005 ret = check_thread_quit_pipe(pollfd, revents);
2006 if (ret) {
2007 err = 0;
2008 goto exit;
2009 }
2010
2011 /* Inspect the relay conn pipe for new connection. */
2012 if (pollfd == live_conn_pipe[0]) {
2013 if (revents & LPOLLIN) {
2014 struct relay_connection *conn;
2015
2016 ret = lttng_read(live_conn_pipe[0],
2017 &conn, sizeof(conn));
2018 if (ret < 0) {
2019 goto error;
2020 }
2021 lttng_poll_add(&events, conn->sock->fd,
2022 LPOLLIN | LPOLLRDHUP);
2023 connection_ht_add(viewer_connections_ht, conn);
2024 DBG("Connection socket %d added to poll", conn->sock->fd);
2025 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2026 ERR("Relay live pipe error");
2027 goto error;
2028 } else {
2029 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2030 goto error;
2031 }
2032 } else {
2033 /* Connection activity. */
2034 struct relay_connection *conn;
2035
2036 conn = connection_get_by_sock(viewer_connections_ht, pollfd);
2037 if (!conn) {
2038 continue;
2039 }
2040
2041 if (revents & LPOLLIN) {
2042 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
2043 sizeof(recv_hdr), 0);
2044 if (ret <= 0) {
2045 /* Connection closed. */
2046 cleanup_connection_pollfd(&events, pollfd);
2047 /* Put "create" ownership reference. */
2048 connection_put(conn);
2049 DBG("Viewer control conn closed with %d", pollfd);
2050 } else {
2051 ret = process_control(&recv_hdr, conn);
2052 if (ret < 0) {
2053 /* Clear the session on error. */
2054 cleanup_connection_pollfd(&events, pollfd);
2055 /* Put "create" ownership reference. */
2056 connection_put(conn);
2057 DBG("Viewer connection closed with %d", pollfd);
2058 }
2059 }
2060 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2061 cleanup_connection_pollfd(&events, pollfd);
2062 /* Put "create" ownership reference. */
2063 connection_put(conn);
2064 } else {
2065 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2066 connection_put(conn);
2067 goto error;
2068 }
2069 /* Put local "get_by_sock" reference. */
2070 connection_put(conn);
2071 }
2072 }
2073 }
2074
2075 exit:
2076 error:
2077 lttng_poll_clean(&events);
2078
2079 /* Cleanup reamaining connection object. */
2080 rcu_read_lock();
2081 cds_lfht_for_each_entry(viewer_connections_ht->ht, &iter.iter,
2082 destroy_conn,
2083 sock_n.node) {
2084 health_code_update();
2085 connection_put(destroy_conn);
2086 }
2087 rcu_read_unlock();
2088 error_poll_create:
2089 lttng_ht_destroy(viewer_connections_ht);
2090 viewer_connections_ht_error:
2091 /* Close relay conn pipes */
2092 utils_close_pipe(live_conn_pipe);
2093 if (err) {
2094 DBG("Viewer worker thread exited with error");
2095 }
2096 DBG("Viewer worker thread cleanup complete");
2097 error_testpoint:
2098 if (err) {
2099 health_error();
2100 ERR("Health error occurred in %s", __func__);
2101 }
2102 health_unregister(health_relayd);
2103 if (lttng_relay_stop_threads()) {
2104 ERR("Error stopping threads");
2105 }
2106 rcu_unregister_thread();
2107 return NULL;
2108 }
2109
2110 /*
2111 * Create the relay command pipe to wake thread_manage_apps.
2112 * Closed in cleanup().
2113 */
2114 static int create_conn_pipe(void)
2115 {
2116 return utils_create_pipe_cloexec(live_conn_pipe);
2117 }
2118
2119 int relayd_live_join(void)
2120 {
2121 int ret, retval = 0;
2122 void *status;
2123
2124 ret = pthread_join(live_listener_thread, &status);
2125 if (ret) {
2126 errno = ret;
2127 PERROR("pthread_join live listener");
2128 retval = -1;
2129 }
2130
2131 ret = pthread_join(live_worker_thread, &status);
2132 if (ret) {
2133 errno = ret;
2134 PERROR("pthread_join live worker");
2135 retval = -1;
2136 }
2137
2138 ret = pthread_join(live_dispatcher_thread, &status);
2139 if (ret) {
2140 errno = ret;
2141 PERROR("pthread_join live dispatcher");
2142 retval = -1;
2143 }
2144
2145 cleanup_relayd_live();
2146
2147 return retval;
2148 }
2149
2150 /*
2151 * main
2152 */
2153 int relayd_live_create(struct lttng_uri *uri)
2154 {
2155 int ret = 0, retval = 0;
2156 void *status;
2157 int is_root;
2158
2159 if (!uri) {
2160 retval = -1;
2161 goto exit_init_data;
2162 }
2163 live_uri = uri;
2164
2165 /* Check if daemon is UID = 0 */
2166 is_root = !getuid();
2167
2168 if (!is_root) {
2169 if (live_uri->port < 1024) {
2170 ERR("Need to be root to use ports < 1024");
2171 retval = -1;
2172 goto exit_init_data;
2173 }
2174 }
2175
2176 /* Setup the thread apps communication pipe. */
2177 if (create_conn_pipe()) {
2178 retval = -1;
2179 goto exit_init_data;
2180 }
2181
2182 /* Init relay command queue. */
2183 cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2184
2185 /* Set up max poll set size */
2186 if (lttng_poll_set_max_size()) {
2187 retval = -1;
2188 goto exit_init_data;
2189 }
2190
2191 /* Setup the dispatcher thread */
2192 ret = pthread_create(&live_dispatcher_thread, default_pthread_attr(),
2193 thread_dispatcher, (void *) NULL);
2194 if (ret) {
2195 errno = ret;
2196 PERROR("pthread_create viewer dispatcher");
2197 retval = -1;
2198 goto exit_dispatcher_thread;
2199 }
2200
2201 /* Setup the worker thread */
2202 ret = pthread_create(&live_worker_thread, default_pthread_attr(),
2203 thread_worker, NULL);
2204 if (ret) {
2205 errno = ret;
2206 PERROR("pthread_create viewer worker");
2207 retval = -1;
2208 goto exit_worker_thread;
2209 }
2210
2211 /* Setup the listener thread */
2212 ret = pthread_create(&live_listener_thread, default_pthread_attr(),
2213 thread_listener, (void *) NULL);
2214 if (ret) {
2215 errno = ret;
2216 PERROR("pthread_create viewer listener");
2217 retval = -1;
2218 goto exit_listener_thread;
2219 }
2220
2221 /*
2222 * All OK, started all threads.
2223 */
2224 return retval;
2225
2226 /*
2227 * Join on the live_listener_thread should anything be added after
2228 * the live_listener thread's creation.
2229 */
2230
2231 exit_listener_thread:
2232
2233 ret = pthread_join(live_worker_thread, &status);
2234 if (ret) {
2235 errno = ret;
2236 PERROR("pthread_join live worker");
2237 retval = -1;
2238 }
2239 exit_worker_thread:
2240
2241 ret = pthread_join(live_dispatcher_thread, &status);
2242 if (ret) {
2243 errno = ret;
2244 PERROR("pthread_join live dispatcher");
2245 retval = -1;
2246 }
2247 exit_dispatcher_thread:
2248
2249 exit_init_data:
2250 cleanup_relayd_live();
2251
2252 return retval;
2253 }
This page took 0.075265 seconds and 4 git commands to generate.