Backport: relayd: track the live listener thread's epoll fd
[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 #include <common/fd-tracker/utils.h>
57
58 #include "cmd.h"
59 #include "live.h"
60 #include "lttng-relayd.h"
61 #include "utils.h"
62 #include "health-relayd.h"
63 #include "testpoint.h"
64 #include "viewer-stream.h"
65 #include "stream.h"
66 #include "session.h"
67 #include "ctf-trace.h"
68 #include "connection.h"
69 #include "viewer-session.h"
70
71 #define SESSION_BUF_DEFAULT_COUNT 16
72
73 static struct lttng_uri *live_uri;
74
75 /*
76 * This pipe is used to inform the worker thread that a command is queued and
77 * ready to be processed.
78 */
79 static int live_conn_pipe[2] = { -1, -1 };
80
81 /* Shared between threads */
82 static int live_dispatch_thread_exit;
83
84 static pthread_t live_listener_thread;
85 static pthread_t live_dispatcher_thread;
86 static pthread_t live_worker_thread;
87
88 /*
89 * Relay command queue.
90 *
91 * The live_thread_listener and live_thread_dispatcher communicate with this
92 * queue.
93 */
94 static struct relay_conn_queue viewer_conn_queue;
95
96 static uint64_t last_relay_viewer_session_id;
97 static pthread_mutex_t last_relay_viewer_session_id_lock =
98 PTHREAD_MUTEX_INITIALIZER;
99
100 /*
101 * Cleanup the daemon
102 */
103 static
104 void cleanup_relayd_live(void)
105 {
106 DBG("Cleaning up");
107
108 free(live_uri);
109 }
110
111 /*
112 * Receive a request buffer using a given socket, destination allocated buffer
113 * of length size.
114 *
115 * Return the size of the received message or else a negative value on error
116 * with errno being set by recvmsg() syscall.
117 */
118 static
119 ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
120 {
121 ssize_t ret;
122
123 ret = sock->ops->recvmsg(sock, buf, size, 0);
124 if (ret < 0 || ret != size) {
125 if (ret == 0) {
126 /* Orderly shutdown. Not necessary to print an error. */
127 DBG("Socket %d did an orderly shutdown", sock->fd);
128 } else {
129 ERR("Relay failed to receive request.");
130 }
131 ret = -1;
132 }
133
134 return ret;
135 }
136
137 /*
138 * Send a response buffer using a given socket, source allocated buffer of
139 * length size.
140 *
141 * Return the size of the sent message or else a negative value on error with
142 * errno being set by sendmsg() syscall.
143 */
144 static
145 ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
146 {
147 ssize_t ret;
148
149 ret = sock->ops->sendmsg(sock, buf, size, 0);
150 if (ret < 0) {
151 ERR("Relayd failed to send response.");
152 }
153
154 return ret;
155 }
156
157 /*
158 * Atomically check if new streams got added in one of the sessions attached
159 * and reset the flag to 0.
160 *
161 * Returns 1 if new streams got added, 0 if nothing changed, a negative value
162 * on error.
163 */
164 static
165 int check_new_streams(struct relay_connection *conn)
166 {
167 struct relay_session *session;
168 unsigned long current_val;
169 int ret = 0;
170
171 if (!conn->viewer_session) {
172 goto end;
173 }
174 rcu_read_lock();
175 cds_list_for_each_entry_rcu(session,
176 &conn->viewer_session->session_list,
177 viewer_session_node) {
178 if (!session_get(session)) {
179 continue;
180 }
181 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
182 ret = current_val;
183 session_put(session);
184 if (ret == 1) {
185 goto end;
186 }
187 }
188 end:
189 rcu_read_unlock();
190 return ret;
191 }
192
193 /*
194 * Send viewer streams to the given socket. The ignore_sent_flag indicates if
195 * this function should ignore the sent flag or not.
196 *
197 * Return 0 on success or else a negative value.
198 */
199 static
200 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
201 struct relay_session *session, unsigned int ignore_sent_flag)
202 {
203 ssize_t ret;
204 struct lttng_viewer_stream send_stream;
205 struct lttng_ht_iter iter;
206 struct relay_viewer_stream *vstream;
207
208 rcu_read_lock();
209
210 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
211 stream_n.node) {
212 struct ctf_trace *ctf_trace;
213
214 health_code_update();
215
216 if (!viewer_stream_get(vstream)) {
217 continue;
218 }
219
220 pthread_mutex_lock(&vstream->stream->lock);
221 /* Ignore if not the same session. */
222 if (vstream->stream->trace->session->id != session->id ||
223 (!ignore_sent_flag && vstream->sent_flag)) {
224 pthread_mutex_unlock(&vstream->stream->lock);
225 viewer_stream_put(vstream);
226 continue;
227 }
228
229 ctf_trace = vstream->stream->trace;
230 send_stream.id = htobe64(vstream->stream->stream_handle);
231 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
232 send_stream.metadata_flag = htobe32(
233 vstream->stream->is_metadata);
234 if (lttng_strncpy(send_stream.path_name, vstream->path_name,
235 sizeof(send_stream.path_name))) {
236 pthread_mutex_unlock(&vstream->stream->lock);
237 viewer_stream_put(vstream);
238 ret = -1; /* Error. */
239 goto end_unlock;
240 }
241 if (lttng_strncpy(send_stream.channel_name,
242 vstream->channel_name,
243 sizeof(send_stream.channel_name))) {
244 pthread_mutex_unlock(&vstream->stream->lock);
245 viewer_stream_put(vstream);
246 ret = -1; /* Error. */
247 goto end_unlock;
248 }
249
250 DBG("Sending stream %" PRIu64 " to viewer",
251 vstream->stream->stream_handle);
252 vstream->sent_flag = 1;
253 pthread_mutex_unlock(&vstream->stream->lock);
254
255 ret = send_response(sock, &send_stream, sizeof(send_stream));
256 viewer_stream_put(vstream);
257 if (ret < 0) {
258 goto end_unlock;
259 }
260 }
261
262 ret = 0;
263
264 end_unlock:
265 rcu_read_unlock();
266 return ret;
267 }
268
269 /*
270 * Create every viewer stream possible for the given session with the seek
271 * type. Three counters *can* be return which are in order the total amount of
272 * viewer stream of the session, the number of unsent stream and the number of
273 * stream created. Those counters can be NULL and thus will be ignored.
274 *
275 * Return 0 on success or else a negative value.
276 */
277 static
278 int make_viewer_streams(struct relay_session *session,
279 enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent,
280 uint32_t *nb_created, bool *closed)
281 {
282 int ret;
283 struct lttng_ht_iter iter;
284 struct ctf_trace *ctf_trace;
285
286 assert(session);
287
288 /*
289 * Hold the session lock to ensure that we see either none or
290 * all initial streams for a session, but no intermediate state.
291 */
292 pthread_mutex_lock(&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, seek_t);
328 if (!vstream) {
329 ret = -1;
330 ctf_trace_put(ctf_trace);
331 stream_put(stream);
332 goto error_unlock;
333 }
334
335 if (nb_created) {
336 /* Update number of created stream counter. */
337 (*nb_created)++;
338 }
339 /*
340 * Ensure a self-reference is preserved even
341 * after we have put our local reference.
342 */
343 if (!viewer_stream_get(vstream)) {
344 ERR("Unable to get self-reference on viewer stream, logic error.");
345 abort();
346 }
347 } else {
348 if (!vstream->sent_flag && nb_unsent) {
349 /* Update number of unsent stream counter. */
350 (*nb_unsent)++;
351 }
352 }
353 /* Update number of total stream counter. */
354 if (nb_total) {
355 if (stream->is_metadata) {
356 if (!stream->closed ||
357 stream->metadata_received > vstream->metadata_sent) {
358 (*nb_total)++;
359 }
360 } else {
361 if (!stream->closed ||
362 !(((int64_t) (stream->prev_seq - stream->last_net_seq_num)) >= 0)) {
363
364 (*nb_total)++;
365 }
366 }
367 }
368 /* Put local reference. */
369 viewer_stream_put(vstream);
370 next:
371 stream_put(stream);
372 }
373 ctf_trace_put(ctf_trace);
374 }
375
376 ret = 0;
377
378 error_unlock:
379 rcu_read_unlock();
380 pthread_mutex_unlock(&session->lock);
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_named_thread_poll_set(struct lttng_poll_event *events,
397 int size, const char *name)
398 {
399 int ret;
400
401 if (events == NULL || size == 0) {
402 ret = -1;
403 goto error;
404 }
405
406 ret = fd_tracker_util_poll_create(the_fd_tracker,
407 name, events, 1, LTTNG_CLOEXEC);
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 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
423 */
424 static
425 int create_thread_poll_set(struct lttng_poll_event *events, int size)
426 {
427 return create_named_thread_poll_set(events, size, "Unknown epoll");
428 }
429
430 /*
431 * Check if the thread quit pipe was triggered.
432 *
433 * Return 1 if it was triggered else 0;
434 */
435 static
436 int check_thread_quit_pipe(int fd, uint32_t events)
437 {
438 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
439 return 1;
440 }
441
442 return 0;
443 }
444
445 /*
446 * Create and init socket from uri.
447 */
448 static
449 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
450 {
451 int ret;
452 struct lttcomm_sock *sock = NULL;
453
454 sock = lttcomm_alloc_sock_from_uri(uri);
455 if (sock == NULL) {
456 ERR("Allocating socket");
457 goto error;
458 }
459
460 ret = lttcomm_create_sock(sock);
461 if (ret < 0) {
462 goto error;
463 }
464 DBG("Listening on sock %d for live", sock->fd);
465
466 ret = sock->ops->bind(sock);
467 if (ret < 0) {
468 goto error;
469 }
470
471 ret = sock->ops->listen(sock, -1);
472 if (ret < 0) {
473 goto error;
474
475 }
476
477 return sock;
478
479 error:
480 if (sock) {
481 lttcomm_destroy_sock(sock);
482 }
483 return NULL;
484 }
485
486 /*
487 * This thread manages the listening for new connections on the network
488 */
489 static
490 void *thread_listener(void *data)
491 {
492 int i, ret, pollfd, err = -1;
493 uint32_t revents, nb_fd;
494 struct lttng_poll_event events;
495 struct lttcomm_sock *live_control_sock;
496
497 DBG("[thread] Relay live listener started");
498
499 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
500
501 health_code_update();
502
503 live_control_sock = init_socket(live_uri);
504 if (!live_control_sock) {
505 goto error_sock_control;
506 }
507
508 /* Pass 2 as size here for the thread quit pipe and control sockets. */
509 ret = create_named_thread_poll_set(&events, 2,
510 "Live listener thread epoll");
511 if (ret < 0) {
512 goto error_create_poll;
513 }
514
515 /* Add the control socket */
516 ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
517 if (ret < 0) {
518 goto error_poll_add;
519 }
520
521 lttng_relay_notify_ready();
522
523 if (testpoint(relayd_thread_live_listener)) {
524 goto error_testpoint;
525 }
526
527 while (1) {
528 health_code_update();
529
530 DBG("Listener accepting live viewers connections");
531
532 restart:
533 health_poll_entry();
534 ret = lttng_poll_wait(&events, -1);
535 health_poll_exit();
536 if (ret < 0) {
537 /*
538 * Restart interrupted system call.
539 */
540 if (errno == EINTR) {
541 goto restart;
542 }
543 goto error;
544 }
545 nb_fd = ret;
546
547 DBG("Relay new viewer connection received");
548 for (i = 0; i < nb_fd; i++) {
549 health_code_update();
550
551 /* Fetch once the poll data */
552 revents = LTTNG_POLL_GETEV(&events, i);
553 pollfd = LTTNG_POLL_GETFD(&events, i);
554
555 if (!revents) {
556 /* No activity for this FD (poll implementation). */
557 continue;
558 }
559
560 /* Thread quit pipe has been closed. Killing thread. */
561 ret = check_thread_quit_pipe(pollfd, revents);
562 if (ret) {
563 err = 0;
564 goto exit;
565 }
566
567 if (revents & LPOLLIN) {
568 /*
569 * A new connection is requested, therefore a
570 * viewer connection is allocated in this
571 * thread, enqueued to a global queue and
572 * dequeued (and freed) in the worker thread.
573 */
574 int val = 1;
575 struct relay_connection *new_conn;
576 struct lttcomm_sock *newsock;
577
578 newsock = live_control_sock->ops->accept(live_control_sock);
579 if (!newsock) {
580 PERROR("accepting control sock");
581 goto error;
582 }
583 DBG("Relay viewer connection accepted socket %d", newsock->fd);
584
585 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
586 sizeof(val));
587 if (ret < 0) {
588 PERROR("setsockopt inet");
589 lttcomm_destroy_sock(newsock);
590 goto error;
591 }
592 new_conn = connection_create(newsock, RELAY_CONNECTION_UNKNOWN);
593 if (!new_conn) {
594 lttcomm_destroy_sock(newsock);
595 goto error;
596 }
597 /* Ownership assumed by the connection. */
598 newsock = NULL;
599
600 /* Enqueue request for the dispatcher thread. */
601 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
602 &new_conn->qnode);
603
604 /*
605 * Wake the dispatch queue futex.
606 * Implicit memory barrier with the
607 * exchange in cds_wfcq_enqueue.
608 */
609 futex_nto1_wake(&viewer_conn_queue.futex);
610 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
611 ERR("socket poll error");
612 goto error;
613 } else {
614 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
615 goto error;
616 }
617 }
618 }
619
620 exit:
621 error:
622 error_poll_add:
623 error_testpoint:
624 (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
625 error_create_poll:
626 if (live_control_sock->fd >= 0) {
627 ret = live_control_sock->ops->close(live_control_sock);
628 if (ret) {
629 PERROR("close");
630 }
631 }
632 lttcomm_destroy_sock(live_control_sock);
633 error_sock_control:
634 if (err) {
635 health_error();
636 DBG("Live viewer listener thread exited with error");
637 }
638 health_unregister(health_relayd);
639 DBG("Live viewer listener thread cleanup complete");
640 if (lttng_relay_stop_threads()) {
641 ERR("Error stopping threads");
642 }
643 return NULL;
644 }
645
646 /*
647 * This thread manages the dispatching of the requests to worker threads
648 */
649 static
650 void *thread_dispatcher(void *data)
651 {
652 int err = -1;
653 ssize_t ret;
654 struct cds_wfcq_node *node;
655 struct relay_connection *conn = NULL;
656
657 DBG("[thread] Live viewer relay dispatcher started");
658
659 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
660
661 if (testpoint(relayd_thread_live_dispatcher)) {
662 goto error_testpoint;
663 }
664
665 health_code_update();
666
667 for (;;) {
668 health_code_update();
669
670 /* Atomically prepare the queue futex */
671 futex_nto1_prepare(&viewer_conn_queue.futex);
672
673 if (CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
674 break;
675 }
676
677 do {
678 health_code_update();
679
680 /* Dequeue commands */
681 node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
682 &viewer_conn_queue.tail);
683 if (node == NULL) {
684 DBG("Woken up but nothing in the live-viewer "
685 "relay command queue");
686 /* Continue thread execution */
687 break;
688 }
689 conn = caa_container_of(node, struct relay_connection, qnode);
690 DBG("Dispatching viewer request waiting on sock %d",
691 conn->sock->fd);
692
693 /*
694 * Inform worker thread of the new request. This
695 * call is blocking so we can be assured that
696 * the data will be read at some point in time
697 * or wait to the end of the world :)
698 */
699 ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
700 if (ret < 0) {
701 PERROR("write conn pipe");
702 connection_put(conn);
703 goto error;
704 }
705 } while (node != NULL);
706
707 /* Futex wait on queue. Blocking call on futex() */
708 health_poll_entry();
709 futex_nto1_wait(&viewer_conn_queue.futex);
710 health_poll_exit();
711 }
712
713 /* Normal exit, no error */
714 err = 0;
715
716 error:
717 error_testpoint:
718 if (err) {
719 health_error();
720 ERR("Health error occurred in %s", __func__);
721 }
722 health_unregister(health_relayd);
723 DBG("Live viewer dispatch thread dying");
724 if (lttng_relay_stop_threads()) {
725 ERR("Error stopping threads");
726 }
727 return NULL;
728 }
729
730 /*
731 * Establish connection with the viewer and check the versions.
732 *
733 * Return 0 on success or else negative value.
734 */
735 static
736 int viewer_connect(struct relay_connection *conn)
737 {
738 int ret;
739 struct lttng_viewer_connect reply, msg;
740
741 conn->version_check_done = 1;
742
743 health_code_update();
744
745 DBG("Viewer is establishing a connection to the relayd.");
746
747 ret = recv_request(conn->sock, &msg, sizeof(msg));
748 if (ret < 0) {
749 goto end;
750 }
751
752 health_code_update();
753
754 memset(&reply, 0, sizeof(reply));
755 reply.major = RELAYD_VERSION_COMM_MAJOR;
756 reply.minor = RELAYD_VERSION_COMM_MINOR;
757
758 /* Major versions must be the same */
759 if (reply.major != be32toh(msg.major)) {
760 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
761 reply.major, be32toh(msg.major));
762 ret = -1;
763 goto end;
764 }
765
766 conn->major = reply.major;
767 /* We adapt to the lowest compatible version */
768 if (reply.minor <= be32toh(msg.minor)) {
769 conn->minor = reply.minor;
770 } else {
771 conn->minor = be32toh(msg.minor);
772 }
773
774 if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
775 conn->type = RELAY_VIEWER_COMMAND;
776 } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
777 conn->type = RELAY_VIEWER_NOTIFICATION;
778 } else {
779 ERR("Unknown connection type : %u", be32toh(msg.type));
780 ret = -1;
781 goto end;
782 }
783
784 reply.major = htobe32(reply.major);
785 reply.minor = htobe32(reply.minor);
786 if (conn->type == RELAY_VIEWER_COMMAND) {
787 /*
788 * Increment outside of htobe64 macro, because the argument can
789 * be used more than once within the macro, and thus the
790 * operation may be undefined.
791 */
792 pthread_mutex_lock(&last_relay_viewer_session_id_lock);
793 last_relay_viewer_session_id++;
794 pthread_mutex_unlock(&last_relay_viewer_session_id_lock);
795 reply.viewer_session_id = htobe64(last_relay_viewer_session_id);
796 }
797
798 health_code_update();
799
800 ret = send_response(conn->sock, &reply, sizeof(reply));
801 if (ret < 0) {
802 goto end;
803 }
804
805 health_code_update();
806
807 DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
808 ret = 0;
809
810 end:
811 return ret;
812 }
813
814 /*
815 * Send the viewer the list of current sessions.
816 * We need to create a copy of the hash table content because otherwise
817 * we cannot assume the number of entries stays the same between getting
818 * the number of HT elements and iteration over the HT.
819 *
820 * Return 0 on success or else a negative value.
821 */
822 static
823 int viewer_list_sessions(struct relay_connection *conn)
824 {
825 int ret = 0;
826 struct lttng_viewer_list_sessions session_list;
827 struct lttng_ht_iter iter;
828 struct relay_session *session;
829 struct lttng_viewer_session *send_session_buf = NULL;
830 uint32_t buf_count = SESSION_BUF_DEFAULT_COUNT;
831 uint32_t count = 0;
832
833 DBG("List sessions received");
834
835 send_session_buf = zmalloc(SESSION_BUF_DEFAULT_COUNT * sizeof(*send_session_buf));
836 if (!send_session_buf) {
837 return -1;
838 }
839
840 rcu_read_lock();
841 cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, session,
842 session_n.node) {
843 struct lttng_viewer_session *send_session;
844
845 health_code_update();
846
847 if (count >= buf_count) {
848 struct lttng_viewer_session *newbuf;
849 uint32_t new_buf_count = buf_count << 1;
850
851 newbuf = realloc(send_session_buf,
852 new_buf_count * sizeof(*send_session_buf));
853 if (!newbuf) {
854 ret = -1;
855 break;
856 }
857 send_session_buf = newbuf;
858 buf_count = new_buf_count;
859 }
860 send_session = &send_session_buf[count];
861 if (lttng_strncpy(send_session->session_name,
862 session->session_name,
863 sizeof(send_session->session_name))) {
864 ret = -1;
865 break;
866 }
867 if (lttng_strncpy(send_session->hostname, session->hostname,
868 sizeof(send_session->hostname))) {
869 ret = -1;
870 break;
871 }
872 send_session->id = htobe64(session->id);
873 send_session->live_timer = htobe32(session->live_timer);
874 if (session->viewer_attached) {
875 send_session->clients = htobe32(1);
876 } else {
877 send_session->clients = htobe32(0);
878 }
879 send_session->streams = htobe32(session->stream_count);
880 count++;
881 }
882 rcu_read_unlock();
883 if (ret < 0) {
884 goto end_free;
885 }
886
887 session_list.sessions_count = htobe32(count);
888
889 health_code_update();
890
891 ret = send_response(conn->sock, &session_list, sizeof(session_list));
892 if (ret < 0) {
893 goto end_free;
894 }
895
896 health_code_update();
897
898 ret = send_response(conn->sock, send_session_buf,
899 count * sizeof(*send_session_buf));
900 if (ret < 0) {
901 goto end_free;
902 }
903 health_code_update();
904
905 ret = 0;
906 end_free:
907 free(send_session_buf);
908 return ret;
909 }
910
911 /*
912 * Send the viewer the list of current streams.
913 */
914 static
915 int viewer_get_new_streams(struct relay_connection *conn)
916 {
917 int ret, send_streams = 0;
918 uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0, nb_total = 0;
919 struct lttng_viewer_new_streams_request request;
920 struct lttng_viewer_new_streams_response response;
921 struct relay_session *session;
922 uint64_t session_id;
923 bool closed = false;
924
925 assert(conn);
926
927 DBG("Get new streams received");
928
929 health_code_update();
930
931 /* Receive the request from the connected client. */
932 ret = recv_request(conn->sock, &request, sizeof(request));
933 if (ret < 0) {
934 goto error;
935 }
936 session_id = be64toh(request.session_id);
937
938 health_code_update();
939
940 memset(&response, 0, sizeof(response));
941
942 session = session_get_by_id(session_id);
943 if (!session) {
944 DBG("Relay session %" PRIu64 " not found", session_id);
945 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
946 goto send_reply;
947 }
948
949 if (!viewer_session_is_attached(conn->viewer_session, session)) {
950 send_streams = 0;
951 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
952 goto send_reply;
953 }
954
955 send_streams = 1;
956 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
957
958 ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, &nb_total, &nb_unsent,
959 &nb_created, &closed);
960 if (ret < 0) {
961 goto end_put_session;
962 }
963 /* Only send back the newly created streams with the unsent ones. */
964 nb_streams = nb_created + nb_unsent;
965 response.streams_count = htobe32(nb_streams);
966
967 /*
968 * If the session is closed, HUP when there are no more streams
969 * with data.
970 */
971 if (closed && nb_total == 0) {
972 send_streams = 0;
973 response.streams_count = 0;
974 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
975 goto send_reply;
976 }
977
978 send_reply:
979 health_code_update();
980 ret = send_response(conn->sock, &response, sizeof(response));
981 if (ret < 0) {
982 goto end_put_session;
983 }
984 health_code_update();
985
986 /*
987 * Unknown or empty session, just return gracefully, the viewer
988 * knows what is happening.
989 */
990 if (!send_streams || !nb_streams) {
991 ret = 0;
992 goto end_put_session;
993 }
994
995 /*
996 * Send stream and *DON'T* ignore the sent flag so every viewer
997 * streams that were not sent from that point will be sent to
998 * the viewer.
999 */
1000 ret = send_viewer_streams(conn->sock, session, 0);
1001 if (ret < 0) {
1002 goto end_put_session;
1003 }
1004
1005 end_put_session:
1006 if (session) {
1007 session_put(session);
1008 }
1009 error:
1010 return ret;
1011 }
1012
1013 /*
1014 * Send the viewer the list of current sessions.
1015 */
1016 static
1017 int viewer_attach_session(struct relay_connection *conn)
1018 {
1019 int send_streams = 0;
1020 ssize_t ret;
1021 uint32_t nb_streams = 0;
1022 enum lttng_viewer_seek seek_type;
1023 struct lttng_viewer_attach_session_request request;
1024 struct lttng_viewer_attach_session_response response;
1025 struct relay_session *session = NULL;
1026 bool closed = false;
1027
1028 assert(conn);
1029
1030 health_code_update();
1031
1032 /* Receive the request from the connected client. */
1033 ret = recv_request(conn->sock, &request, sizeof(request));
1034 if (ret < 0) {
1035 goto error;
1036 }
1037
1038 health_code_update();
1039
1040 memset(&response, 0, sizeof(response));
1041
1042 if (!conn->viewer_session) {
1043 DBG("Client trying to attach before creating a live viewer session");
1044 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1045 goto send_reply;
1046 }
1047
1048 session = session_get_by_id(be64toh(request.session_id));
1049 if (!session) {
1050 DBG("Relay session %" PRIu64 " not found",
1051 (uint64_t) be64toh(request.session_id));
1052 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1053 goto send_reply;
1054 }
1055 DBG("Attach session ID %" PRIu64 " received",
1056 (uint64_t) be64toh(request.session_id));
1057
1058 if (session->live_timer == 0) {
1059 DBG("Not live session");
1060 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1061 goto send_reply;
1062 }
1063
1064 send_streams = 1;
1065 ret = viewer_session_attach(conn->viewer_session, session);
1066 if (ret) {
1067 DBG("Already a viewer attached");
1068 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1069 goto send_reply;
1070 }
1071
1072 switch (be32toh(request.seek)) {
1073 case LTTNG_VIEWER_SEEK_BEGINNING:
1074 case LTTNG_VIEWER_SEEK_LAST:
1075 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1076 seek_type = be32toh(request.seek);
1077 break;
1078 default:
1079 ERR("Wrong seek parameter");
1080 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1081 send_streams = 0;
1082 goto send_reply;
1083 }
1084
1085 ret = make_viewer_streams(session, seek_type, &nb_streams, NULL,
1086 NULL, &closed);
1087 if (ret < 0) {
1088 goto end_put_session;
1089 }
1090 response.streams_count = htobe32(nb_streams);
1091
1092 /*
1093 * If the session is closed when the viewer is attaching, it
1094 * means some of the streams may have been concurrently removed,
1095 * so we don't allow the viewer to attach, even if there are
1096 * streams available.
1097 */
1098 if (closed) {
1099 send_streams = 0;
1100 response.streams_count = 0;
1101 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
1102 goto send_reply;
1103 }
1104
1105 send_reply:
1106 health_code_update();
1107 ret = send_response(conn->sock, &response, sizeof(response));
1108 if (ret < 0) {
1109 goto end_put_session;
1110 }
1111 health_code_update();
1112
1113 /*
1114 * Unknown or empty session, just return gracefully, the viewer
1115 * knows what is happening.
1116 */
1117 if (!send_streams || !nb_streams) {
1118 ret = 0;
1119 goto end_put_session;
1120 }
1121
1122 /* Send stream and ignore the sent flag. */
1123 ret = send_viewer_streams(conn->sock, session, 1);
1124 if (ret < 0) {
1125 goto end_put_session;
1126 }
1127
1128 end_put_session:
1129 if (session) {
1130 session_put(session);
1131 }
1132 error:
1133 return ret;
1134 }
1135
1136 /*
1137 * Open the index file if needed for the given vstream.
1138 *
1139 * If an index file is successfully opened, the vstream will set it as its
1140 * current index file.
1141 *
1142 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1143 *
1144 * Called with rstream lock held.
1145 */
1146 static int try_open_index(struct relay_viewer_stream *vstream,
1147 struct relay_stream *rstream)
1148 {
1149 int ret = 0;
1150
1151 if (vstream->index_file) {
1152 goto end;
1153 }
1154
1155 /*
1156 * First time, we open the index file and at least one index is ready.
1157 */
1158 if (rstream->index_received_seqcount == 0) {
1159 ret = -ENOENT;
1160 goto end;
1161 }
1162 vstream->index_file = lttng_index_file_open(vstream->path_name,
1163 vstream->channel_name,
1164 vstream->stream->tracefile_count,
1165 vstream->current_tracefile_id);
1166 if (!vstream->index_file) {
1167 ret = -1;
1168 }
1169
1170 end:
1171 return ret;
1172 }
1173
1174 /*
1175 * Check the status of the index for the given stream. This function
1176 * updates the index structure if needed and can put (close) the vstream
1177 * in the HUP situation.
1178 *
1179 * Return 0 means that we can proceed with the index. A value of 1 means
1180 * that the index has been updated and is ready to be sent to the
1181 * client. A negative value indicates an error that can't be handled.
1182 *
1183 * Called with rstream lock held.
1184 */
1185 static int check_index_status(struct relay_viewer_stream *vstream,
1186 struct relay_stream *rstream, struct ctf_trace *trace,
1187 struct lttng_viewer_index *index)
1188 {
1189 int ret;
1190
1191 if (trace->session->connection_closed
1192 && rstream->index_received_seqcount
1193 == vstream->index_sent_seqcount) {
1194 /* Last index sent and session connection is closed. */
1195 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1196 goto hup;
1197 } else if (rstream->beacon_ts_end != -1ULL &&
1198 rstream->index_received_seqcount
1199 == vstream->index_sent_seqcount) {
1200 /*
1201 * We've received a synchronization beacon and the last index
1202 * available has been sent, the index for now is inactive.
1203 *
1204 * In this case, we have received a beacon which allows us to
1205 * inform the client of a time interval during which we can
1206 * guarantee that there are no events to read (and never will
1207 * be).
1208 */
1209 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1210 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1211 index->stream_id = htobe64(rstream->ctf_stream_id);
1212 goto index_ready;
1213 } else if (rstream->index_received_seqcount
1214 == vstream->index_sent_seqcount) {
1215 /*
1216 * This checks whether received == sent seqcount. In
1217 * this case, we have not received a beacon. Therefore,
1218 * we can only ask the client to retry later.
1219 */
1220 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1221 goto index_ready;
1222 } else if (!tracefile_array_seq_in_file(rstream->tfa,
1223 vstream->current_tracefile_id,
1224 vstream->index_sent_seqcount)) {
1225 /*
1226 * The next index we want to send cannot be read either
1227 * because we need to perform a rotation, or due to
1228 * the producer having overwritten its trace file.
1229 */
1230 DBG("Viewer stream %" PRIu64 " rotation",
1231 vstream->stream->stream_handle);
1232 ret = viewer_stream_rotate(vstream);
1233 if (ret < 0) {
1234 goto end;
1235 } else if (ret == 1) {
1236 /* EOF across entire stream. */
1237 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1238 goto hup;
1239 }
1240 /*
1241 * If we have been pushed due to overwrite, it
1242 * necessarily means there is data that can be read in
1243 * the stream. If we rotated because we reached the end
1244 * of a tracefile, it means the following tracefile
1245 * needs to contain at least one index, else we would
1246 * have already returned LTTNG_VIEWER_INDEX_RETRY to the
1247 * viewer. The updated index_sent_seqcount needs to
1248 * point to a readable index entry now.
1249 *
1250 * In the case where we "rotate" on a single file, we
1251 * can end up in a case where the requested index is
1252 * still unavailable.
1253 */
1254 if (rstream->tracefile_count == 1 &&
1255 !tracefile_array_seq_in_file(
1256 rstream->tfa,
1257 vstream->current_tracefile_id,
1258 vstream->index_sent_seqcount)) {
1259 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1260 goto index_ready;
1261 }
1262 assert(tracefile_array_seq_in_file(rstream->tfa,
1263 vstream->current_tracefile_id,
1264 vstream->index_sent_seqcount));
1265 }
1266 /* ret == 0 means successful so we continue. */
1267 ret = 0;
1268 end:
1269 return ret;
1270
1271 hup:
1272 viewer_stream_put(vstream);
1273 index_ready:
1274 return 1;
1275 }
1276
1277 /*
1278 * Send the next index for a stream.
1279 *
1280 * Return 0 on success or else a negative value.
1281 */
1282 static
1283 int viewer_get_next_index(struct relay_connection *conn)
1284 {
1285 int ret;
1286 struct lttng_viewer_get_next_index request_index;
1287 struct lttng_viewer_index viewer_index;
1288 struct ctf_packet_index packet_index;
1289 struct relay_viewer_stream *vstream = NULL;
1290 struct relay_stream *rstream = NULL;
1291 struct ctf_trace *ctf_trace = NULL;
1292 struct relay_viewer_stream *metadata_viewer_stream = NULL;
1293
1294 assert(conn);
1295
1296 DBG("Viewer get next index");
1297
1298 memset(&viewer_index, 0, sizeof(viewer_index));
1299 health_code_update();
1300
1301 ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1302 if (ret < 0) {
1303 goto end;
1304 }
1305 health_code_update();
1306
1307 vstream = viewer_stream_get_by_id(be64toh(request_index.stream_id));
1308 if (!vstream) {
1309 DBG("Client requested index of unknown stream id %" PRIu64,
1310 (uint64_t) be64toh(request_index.stream_id));
1311 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1312 goto send_reply;
1313 }
1314
1315 /* Use back. ref. Protected by refcounts. */
1316 rstream = vstream->stream;
1317 ctf_trace = rstream->trace;
1318
1319 /* metadata_viewer_stream may be NULL. */
1320 metadata_viewer_stream =
1321 ctf_trace_get_viewer_metadata_stream(ctf_trace);
1322
1323 pthread_mutex_lock(&rstream->lock);
1324
1325 /*
1326 * The viewer should not ask for index on metadata stream.
1327 */
1328 if (rstream->is_metadata) {
1329 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1330 goto send_reply;
1331 }
1332
1333 /* Try to open an index if one is needed for that stream. */
1334 ret = try_open_index(vstream, rstream);
1335 if (ret < 0) {
1336 if (ret == -ENOENT) {
1337 /*
1338 * The index is created only when the first data
1339 * packet arrives, it might not be ready at the
1340 * beginning of the session
1341 */
1342 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1343 } else {
1344 /* Unhandled error. */
1345 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1346 }
1347 goto send_reply;
1348 }
1349
1350 ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1351 if (ret < 0) {
1352 goto error_put;
1353 } else if (ret == 1) {
1354 /*
1355 * We have no index to send and check_index_status has populated
1356 * viewer_index's status.
1357 */
1358 goto send_reply;
1359 }
1360 /* At this point, ret is 0 thus we will be able to read the index. */
1361 assert(!ret);
1362
1363 /*
1364 * vstream->stream_fd may be NULL if it has been closed by
1365 * tracefile rotation, or if we are at the beginning of the
1366 * stream. We open the data stream file here to protect against
1367 * overwrite caused by tracefile rotation (in association with
1368 * unlink performed before overwrite).
1369 */
1370 if (!vstream->stream_fd) {
1371 char fullpath[PATH_MAX];
1372
1373 if (vstream->stream->tracefile_count > 0) {
1374 ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64,
1375 vstream->path_name,
1376 vstream->channel_name,
1377 vstream->current_tracefile_id);
1378 } else {
1379 ret = snprintf(fullpath, PATH_MAX, "%s/%s",
1380 vstream->path_name,
1381 vstream->channel_name);
1382 }
1383 if (ret < 0) {
1384 goto error_put;
1385 }
1386 ret = open(fullpath, O_RDONLY);
1387 if (ret < 0) {
1388 PERROR("Relay opening trace file");
1389 goto error_put;
1390 }
1391 vstream->stream_fd = stream_fd_create(ret);
1392 if (!vstream->stream_fd) {
1393 if (close(ret)) {
1394 PERROR("close");
1395 }
1396 goto error_put;
1397 }
1398 }
1399
1400 ret = check_new_streams(conn);
1401 if (ret < 0) {
1402 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1403 goto send_reply;
1404 } else if (ret == 1) {
1405 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1406 }
1407
1408 ret = lttng_index_file_read(vstream->index_file, &packet_index);
1409 if (ret) {
1410 ERR("Relay error reading index file %d",
1411 vstream->index_file->fd);
1412 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1413 goto send_reply;
1414 } else {
1415 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1416 vstream->index_sent_seqcount++;
1417 }
1418
1419 /*
1420 * Indexes are stored in big endian, no need to switch before sending.
1421 */
1422 DBG("Sending viewer index for stream %" PRIu64 " offset %" PRIu64,
1423 rstream->stream_handle,
1424 (uint64_t) be64toh(packet_index.offset));
1425 viewer_index.offset = packet_index.offset;
1426 viewer_index.packet_size = packet_index.packet_size;
1427 viewer_index.content_size = packet_index.content_size;
1428 viewer_index.timestamp_begin = packet_index.timestamp_begin;
1429 viewer_index.timestamp_end = packet_index.timestamp_end;
1430 viewer_index.events_discarded = packet_index.events_discarded;
1431 viewer_index.stream_id = packet_index.stream_id;
1432
1433 send_reply:
1434 if (rstream) {
1435 pthread_mutex_unlock(&rstream->lock);
1436 }
1437
1438 if (metadata_viewer_stream) {
1439 pthread_mutex_lock(&metadata_viewer_stream->stream->lock);
1440 DBG("get next index metadata check: recv %" PRIu64
1441 " sent %" PRIu64,
1442 metadata_viewer_stream->stream->metadata_received,
1443 metadata_viewer_stream->metadata_sent);
1444 if (!metadata_viewer_stream->stream->metadata_received ||
1445 metadata_viewer_stream->stream->metadata_received >
1446 metadata_viewer_stream->metadata_sent) {
1447 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1448 }
1449 pthread_mutex_unlock(&metadata_viewer_stream->stream->lock);
1450 }
1451
1452 viewer_index.flags = htobe32(viewer_index.flags);
1453 health_code_update();
1454
1455 ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1456 if (ret < 0) {
1457 goto end;
1458 }
1459 health_code_update();
1460
1461 if (vstream) {
1462 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1463 vstream->index_sent_seqcount,
1464 vstream->stream->stream_handle);
1465 }
1466 end:
1467 if (metadata_viewer_stream) {
1468 viewer_stream_put(metadata_viewer_stream);
1469 }
1470 if (vstream) {
1471 viewer_stream_put(vstream);
1472 }
1473 return ret;
1474
1475 error_put:
1476 pthread_mutex_unlock(&rstream->lock);
1477 if (metadata_viewer_stream) {
1478 viewer_stream_put(metadata_viewer_stream);
1479 }
1480 viewer_stream_put(vstream);
1481 return ret;
1482 }
1483
1484 /*
1485 * Send the next index for a stream
1486 *
1487 * Return 0 on success or else a negative value.
1488 */
1489 static
1490 int viewer_get_packet(struct relay_connection *conn)
1491 {
1492 int ret;
1493 off_t lseek_ret;
1494 char *reply = NULL;
1495 struct lttng_viewer_get_packet get_packet_info;
1496 struct lttng_viewer_trace_packet reply_header;
1497 struct relay_viewer_stream *vstream = NULL;
1498 uint32_t reply_size = sizeof(reply_header);
1499 uint32_t packet_data_len = 0;
1500 ssize_t read_len;
1501
1502 DBG2("Relay get data packet");
1503
1504 health_code_update();
1505
1506 ret = recv_request(conn->sock, &get_packet_info,
1507 sizeof(get_packet_info));
1508 if (ret < 0) {
1509 goto end;
1510 }
1511 health_code_update();
1512
1513 /* From this point on, the error label can be reached. */
1514 memset(&reply_header, 0, sizeof(reply_header));
1515
1516 vstream = viewer_stream_get_by_id(be64toh(get_packet_info.stream_id));
1517 if (!vstream) {
1518 DBG("Client requested packet of unknown stream id %" PRIu64,
1519 (uint64_t) be64toh(get_packet_info.stream_id));
1520 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1521 goto send_reply_nolock;
1522 } else {
1523 packet_data_len = be32toh(get_packet_info.len);
1524 reply_size += packet_data_len;
1525 }
1526
1527 reply = zmalloc(reply_size);
1528 if (!reply) {
1529 PERROR("packet reply zmalloc");
1530 reply_size = sizeof(reply_header);
1531 goto error;
1532 }
1533
1534 pthread_mutex_lock(&vstream->stream->lock);
1535 lseek_ret = lseek(vstream->stream_fd->fd, be64toh(get_packet_info.offset),
1536 SEEK_SET);
1537 if (lseek_ret < 0) {
1538 PERROR("lseek fd %d to offset %" PRIu64, vstream->stream_fd->fd,
1539 (uint64_t) be64toh(get_packet_info.offset));
1540 goto error;
1541 }
1542 read_len = lttng_read(vstream->stream_fd->fd,
1543 reply + sizeof(reply_header),
1544 packet_data_len);
1545 if (read_len < packet_data_len) {
1546 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1547 vstream->stream_fd->fd,
1548 (uint64_t) be64toh(get_packet_info.offset));
1549 goto error;
1550 }
1551 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1552 reply_header.len = htobe32(packet_data_len);
1553 goto send_reply;
1554
1555 error:
1556 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1557
1558 send_reply:
1559 if (vstream) {
1560 pthread_mutex_unlock(&vstream->stream->lock);
1561 }
1562 send_reply_nolock:
1563
1564 health_code_update();
1565
1566 if (reply) {
1567 memcpy(reply, &reply_header, sizeof(reply_header));
1568 ret = send_response(conn->sock, reply, reply_size);
1569 } else {
1570 /* No reply to send. */
1571 ret = send_response(conn->sock, &reply_header,
1572 reply_size);
1573 }
1574
1575 health_code_update();
1576 if (ret < 0) {
1577 PERROR("sendmsg of packet data failed");
1578 goto end_free;
1579 }
1580
1581 DBG("Sent %u bytes for stream %" PRIu64, reply_size,
1582 (uint64_t) be64toh(get_packet_info.stream_id));
1583
1584 end_free:
1585 free(reply);
1586 end:
1587 if (vstream) {
1588 viewer_stream_put(vstream);
1589 }
1590 return ret;
1591 }
1592
1593 /*
1594 * Send the session's metadata
1595 *
1596 * Return 0 on success else a negative value.
1597 */
1598 static
1599 int viewer_get_metadata(struct relay_connection *conn)
1600 {
1601 int ret = 0;
1602 ssize_t read_len;
1603 uint64_t len = 0;
1604 char *data = NULL;
1605 struct lttng_viewer_get_metadata request;
1606 struct lttng_viewer_metadata_packet reply;
1607 struct relay_viewer_stream *vstream = NULL;
1608
1609 assert(conn);
1610
1611 DBG("Relay get metadata");
1612
1613 health_code_update();
1614
1615 ret = recv_request(conn->sock, &request, sizeof(request));
1616 if (ret < 0) {
1617 goto end;
1618 }
1619 health_code_update();
1620
1621 memset(&reply, 0, sizeof(reply));
1622
1623 vstream = viewer_stream_get_by_id(be64toh(request.stream_id));
1624 if (!vstream) {
1625 /*
1626 * The metadata stream can be closed by a CLOSE command
1627 * just before we attach. It can also be closed by
1628 * per-pid tracing during tracing. Therefore, it is
1629 * possible that we cannot find this viewer stream.
1630 * Reply back to the client with an error if we cannot
1631 * find it.
1632 */
1633 DBG("Client requested metadata of unknown stream id %" PRIu64,
1634 (uint64_t) be64toh(request.stream_id));
1635 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1636 goto send_reply;
1637 }
1638 pthread_mutex_lock(&vstream->stream->lock);
1639 if (!vstream->stream->is_metadata) {
1640 ERR("Invalid metadata stream");
1641 goto error;
1642 }
1643
1644 assert(vstream->metadata_sent <= vstream->stream->metadata_received);
1645
1646 len = vstream->stream->metadata_received - vstream->metadata_sent;
1647 if (len == 0) {
1648 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1649 goto send_reply;
1650 }
1651
1652 /* first time, we open the metadata file */
1653 if (!vstream->stream_fd) {
1654 char fullpath[PATH_MAX];
1655
1656 ret = snprintf(fullpath, PATH_MAX, "%s/%s", vstream->path_name,
1657 vstream->channel_name);
1658 if (ret < 0) {
1659 goto error;
1660 }
1661 ret = open(fullpath, O_RDONLY);
1662 if (ret < 0) {
1663 PERROR("Relay opening metadata file");
1664 goto error;
1665 }
1666 vstream->stream_fd = stream_fd_create(ret);
1667 if (!vstream->stream_fd) {
1668 if (close(ret)) {
1669 PERROR("close");
1670 }
1671 goto error;
1672 }
1673 }
1674
1675 reply.len = htobe64(len);
1676 data = zmalloc(len);
1677 if (!data) {
1678 PERROR("viewer metadata zmalloc");
1679 goto error;
1680 }
1681
1682 read_len = lttng_read(vstream->stream_fd->fd, data, len);
1683 if (read_len < len) {
1684 PERROR("Relay reading metadata file");
1685 goto error;
1686 }
1687 vstream->metadata_sent += read_len;
1688 if (vstream->metadata_sent == vstream->stream->metadata_received
1689 && vstream->stream->closed) {
1690 /* Release ownership for the viewer metadata stream. */
1691 viewer_stream_put(vstream);
1692 }
1693
1694 reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1695
1696 goto send_reply;
1697
1698 error:
1699 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1700
1701 send_reply:
1702 health_code_update();
1703 if (vstream) {
1704 pthread_mutex_unlock(&vstream->stream->lock);
1705 }
1706 ret = send_response(conn->sock, &reply, sizeof(reply));
1707 if (ret < 0) {
1708 goto end_free;
1709 }
1710 health_code_update();
1711
1712 if (len > 0) {
1713 ret = send_response(conn->sock, data, len);
1714 if (ret < 0) {
1715 goto end_free;
1716 }
1717 }
1718
1719 DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1720 (uint64_t) be64toh(request.stream_id));
1721
1722 DBG("Metadata sent");
1723
1724 end_free:
1725 free(data);
1726 end:
1727 if (vstream) {
1728 viewer_stream_put(vstream);
1729 }
1730 return ret;
1731 }
1732
1733 /*
1734 * Create a viewer session.
1735 *
1736 * Return 0 on success or else a negative value.
1737 */
1738 static
1739 int viewer_create_session(struct relay_connection *conn)
1740 {
1741 int ret;
1742 struct lttng_viewer_create_session_response resp;
1743
1744 DBG("Viewer create session received");
1745
1746 memset(&resp, 0, sizeof(resp));
1747 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1748 conn->viewer_session = viewer_session_create();
1749 if (!conn->viewer_session) {
1750 ERR("Allocation viewer session");
1751 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1752 goto send_reply;
1753 }
1754
1755 send_reply:
1756 health_code_update();
1757 ret = send_response(conn->sock, &resp, sizeof(resp));
1758 if (ret < 0) {
1759 goto end;
1760 }
1761 health_code_update();
1762 ret = 0;
1763
1764 end:
1765 return ret;
1766 }
1767
1768 /*
1769 * Detach a viewer session.
1770 *
1771 * Return 0 on success or else a negative value.
1772 */
1773 static
1774 int viewer_detach_session(struct relay_connection *conn)
1775 {
1776 int ret;
1777 struct lttng_viewer_detach_session_response response;
1778 struct lttng_viewer_detach_session_request request;
1779 struct relay_session *session = NULL;
1780 uint64_t viewer_session_to_close;
1781
1782 DBG("Viewer detach session received");
1783
1784 assert(conn);
1785
1786 health_code_update();
1787
1788 /* Receive the request from the connected client. */
1789 ret = recv_request(conn->sock, &request, sizeof(request));
1790 if (ret < 0) {
1791 goto end;
1792 }
1793 viewer_session_to_close = be64toh(request.session_id);
1794
1795 if (!conn->viewer_session) {
1796 DBG("Client trying to detach before creating a live viewer session");
1797 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
1798 goto send_reply;
1799 }
1800
1801 health_code_update();
1802
1803 memset(&response, 0, sizeof(response));
1804 DBG("Detaching from session ID %" PRIu64, viewer_session_to_close);
1805
1806 session = session_get_by_id(be64toh(request.session_id));
1807 if (!session) {
1808 DBG("Relay session %" PRIu64 " not found",
1809 (uint64_t) be64toh(request.session_id));
1810 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_UNK);
1811 goto send_reply;
1812 }
1813
1814 ret = viewer_session_is_attached(conn->viewer_session, session);
1815 if (ret != 1) {
1816 DBG("Not attached to this session");
1817 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
1818 goto send_reply_put;
1819 }
1820
1821 viewer_session_close_one_session(conn->viewer_session, session);
1822 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_OK);
1823 DBG("Session %" PRIu64 " detached.", viewer_session_to_close);
1824
1825 send_reply_put:
1826 session_put(session);
1827
1828 send_reply:
1829 health_code_update();
1830 ret = send_response(conn->sock, &response, sizeof(response));
1831 if (ret < 0) {
1832 goto end;
1833 }
1834 health_code_update();
1835 ret = 0;
1836
1837 end:
1838 return ret;
1839 }
1840
1841 /*
1842 * live_relay_unknown_command: send -1 if received unknown command
1843 */
1844 static
1845 void live_relay_unknown_command(struct relay_connection *conn)
1846 {
1847 struct lttcomm_relayd_generic_reply reply;
1848
1849 memset(&reply, 0, sizeof(reply));
1850 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1851 (void) send_response(conn->sock, &reply, sizeof(reply));
1852 }
1853
1854 /*
1855 * Process the commands received on the control socket
1856 */
1857 static
1858 int process_control(struct lttng_viewer_cmd *recv_hdr,
1859 struct relay_connection *conn)
1860 {
1861 int ret = 0;
1862 uint32_t msg_value;
1863
1864 msg_value = be32toh(recv_hdr->cmd);
1865
1866 /*
1867 * Make sure we've done the version check before any command other then a
1868 * new client connection.
1869 */
1870 if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1871 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1872 ret = -1;
1873 goto end;
1874 }
1875
1876 switch (msg_value) {
1877 case LTTNG_VIEWER_CONNECT:
1878 ret = viewer_connect(conn);
1879 break;
1880 case LTTNG_VIEWER_LIST_SESSIONS:
1881 ret = viewer_list_sessions(conn);
1882 break;
1883 case LTTNG_VIEWER_ATTACH_SESSION:
1884 ret = viewer_attach_session(conn);
1885 break;
1886 case LTTNG_VIEWER_GET_NEXT_INDEX:
1887 ret = viewer_get_next_index(conn);
1888 break;
1889 case LTTNG_VIEWER_GET_PACKET:
1890 ret = viewer_get_packet(conn);
1891 break;
1892 case LTTNG_VIEWER_GET_METADATA:
1893 ret = viewer_get_metadata(conn);
1894 break;
1895 case LTTNG_VIEWER_GET_NEW_STREAMS:
1896 ret = viewer_get_new_streams(conn);
1897 break;
1898 case LTTNG_VIEWER_CREATE_SESSION:
1899 ret = viewer_create_session(conn);
1900 break;
1901 case LTTNG_VIEWER_DETACH_SESSION:
1902 ret = viewer_detach_session(conn);
1903 break;
1904 default:
1905 ERR("Received unknown viewer command (%u)",
1906 be32toh(recv_hdr->cmd));
1907 live_relay_unknown_command(conn);
1908 ret = -1;
1909 goto end;
1910 }
1911
1912 end:
1913 return ret;
1914 }
1915
1916 static
1917 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1918 {
1919 int ret;
1920
1921 (void) lttng_poll_del(events, pollfd);
1922
1923 ret = close(pollfd);
1924 if (ret < 0) {
1925 ERR("Closing pollfd %d", pollfd);
1926 }
1927 }
1928
1929 /*
1930 * This thread does the actual work
1931 */
1932 static
1933 void *thread_worker(void *data)
1934 {
1935 int ret, err = -1;
1936 uint32_t nb_fd;
1937 struct lttng_poll_event events;
1938 struct lttng_ht *viewer_connections_ht;
1939 struct lttng_ht_iter iter;
1940 struct lttng_viewer_cmd recv_hdr;
1941 struct relay_connection *destroy_conn;
1942
1943 DBG("[thread] Live viewer relay worker started");
1944
1945 rcu_register_thread();
1946
1947 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1948
1949 if (testpoint(relayd_thread_live_worker)) {
1950 goto error_testpoint;
1951 }
1952
1953 /* table of connections indexed on socket */
1954 viewer_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1955 if (!viewer_connections_ht) {
1956 goto viewer_connections_ht_error;
1957 }
1958
1959 ret = create_thread_poll_set(&events, 2);
1960 if (ret < 0) {
1961 goto error_poll_create;
1962 }
1963
1964 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1965 if (ret < 0) {
1966 goto error;
1967 }
1968
1969 restart:
1970 while (1) {
1971 int i;
1972
1973 health_code_update();
1974
1975 /* Infinite blocking call, waiting for transmission */
1976 DBG3("Relayd live viewer worker thread polling...");
1977 health_poll_entry();
1978 ret = lttng_poll_wait(&events, -1);
1979 health_poll_exit();
1980 if (ret < 0) {
1981 /*
1982 * Restart interrupted system call.
1983 */
1984 if (errno == EINTR) {
1985 goto restart;
1986 }
1987 goto error;
1988 }
1989
1990 nb_fd = ret;
1991
1992 /*
1993 * Process control. The control connection is prioritised so we don't
1994 * starve it with high throughput tracing data on the data
1995 * connection.
1996 */
1997 for (i = 0; i < nb_fd; i++) {
1998 /* Fetch once the poll data */
1999 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2000 int pollfd = LTTNG_POLL_GETFD(&events, i);
2001
2002 health_code_update();
2003
2004 if (!revents) {
2005 /* No activity for this FD (poll implementation). */
2006 continue;
2007 }
2008
2009 /* Thread quit pipe has been closed. Killing thread. */
2010 ret = check_thread_quit_pipe(pollfd, revents);
2011 if (ret) {
2012 err = 0;
2013 goto exit;
2014 }
2015
2016 /* Inspect the relay conn pipe for new connection. */
2017 if (pollfd == live_conn_pipe[0]) {
2018 if (revents & LPOLLIN) {
2019 struct relay_connection *conn;
2020
2021 ret = lttng_read(live_conn_pipe[0],
2022 &conn, sizeof(conn));
2023 if (ret < 0) {
2024 goto error;
2025 }
2026 lttng_poll_add(&events, conn->sock->fd,
2027 LPOLLIN | LPOLLRDHUP);
2028 connection_ht_add(viewer_connections_ht, conn);
2029 DBG("Connection socket %d added to poll", conn->sock->fd);
2030 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2031 ERR("Relay live pipe error");
2032 goto error;
2033 } else {
2034 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2035 goto error;
2036 }
2037 } else {
2038 /* Connection activity. */
2039 struct relay_connection *conn;
2040
2041 conn = connection_get_by_sock(viewer_connections_ht, pollfd);
2042 if (!conn) {
2043 continue;
2044 }
2045
2046 if (revents & LPOLLIN) {
2047 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
2048 sizeof(recv_hdr), 0);
2049 if (ret <= 0) {
2050 /* Connection closed. */
2051 cleanup_connection_pollfd(&events, pollfd);
2052 /* Put "create" ownership reference. */
2053 connection_put(conn);
2054 DBG("Viewer control conn closed with %d", pollfd);
2055 } else {
2056 ret = process_control(&recv_hdr, conn);
2057 if (ret < 0) {
2058 /* Clear the session on error. */
2059 cleanup_connection_pollfd(&events, pollfd);
2060 /* Put "create" ownership reference. */
2061 connection_put(conn);
2062 DBG("Viewer connection closed with %d", pollfd);
2063 }
2064 }
2065 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2066 cleanup_connection_pollfd(&events, pollfd);
2067 /* Put "create" ownership reference. */
2068 connection_put(conn);
2069 } else {
2070 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2071 connection_put(conn);
2072 goto error;
2073 }
2074 /* Put local "get_by_sock" reference. */
2075 connection_put(conn);
2076 }
2077 }
2078 }
2079
2080 exit:
2081 error:
2082 lttng_poll_clean(&events);
2083
2084 /* Cleanup reamaining connection object. */
2085 rcu_read_lock();
2086 cds_lfht_for_each_entry(viewer_connections_ht->ht, &iter.iter,
2087 destroy_conn,
2088 sock_n.node) {
2089 health_code_update();
2090 connection_put(destroy_conn);
2091 }
2092 rcu_read_unlock();
2093 error_poll_create:
2094 lttng_ht_destroy(viewer_connections_ht);
2095 viewer_connections_ht_error:
2096 /* Close relay conn pipes */
2097 (void) fd_tracker_util_pipe_close(the_fd_tracker, live_conn_pipe);
2098 if (err) {
2099 DBG("Viewer worker thread exited with error");
2100 }
2101 DBG("Viewer worker thread cleanup complete");
2102 error_testpoint:
2103 if (err) {
2104 health_error();
2105 ERR("Health error occurred in %s", __func__);
2106 }
2107 health_unregister(health_relayd);
2108 if (lttng_relay_stop_threads()) {
2109 ERR("Error stopping threads");
2110 }
2111 rcu_unregister_thread();
2112 return NULL;
2113 }
2114
2115 /*
2116 * Create the relay command pipe to wake thread_manage_apps.
2117 * Closed in cleanup().
2118 */
2119 static int create_conn_pipe(void)
2120 {
2121 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
2122 "Live connection pipe", live_conn_pipe);
2123 }
2124
2125 int relayd_live_join(void)
2126 {
2127 int ret, retval = 0;
2128 void *status;
2129
2130 ret = pthread_join(live_listener_thread, &status);
2131 if (ret) {
2132 errno = ret;
2133 PERROR("pthread_join live listener");
2134 retval = -1;
2135 }
2136
2137 ret = pthread_join(live_worker_thread, &status);
2138 if (ret) {
2139 errno = ret;
2140 PERROR("pthread_join live worker");
2141 retval = -1;
2142 }
2143
2144 ret = pthread_join(live_dispatcher_thread, &status);
2145 if (ret) {
2146 errno = ret;
2147 PERROR("pthread_join live dispatcher");
2148 retval = -1;
2149 }
2150
2151 cleanup_relayd_live();
2152
2153 return retval;
2154 }
2155
2156 /*
2157 * main
2158 */
2159 int relayd_live_create(struct lttng_uri *uri)
2160 {
2161 int ret = 0, retval = 0;
2162 void *status;
2163 int is_root;
2164
2165 if (!uri) {
2166 retval = -1;
2167 goto exit_init_data;
2168 }
2169 live_uri = uri;
2170
2171 /* Check if daemon is UID = 0 */
2172 is_root = !getuid();
2173
2174 if (!is_root) {
2175 if (live_uri->port < 1024) {
2176 ERR("Need to be root to use ports < 1024");
2177 retval = -1;
2178 goto exit_init_data;
2179 }
2180 }
2181
2182 /* Setup the thread apps communication pipe. */
2183 if (create_conn_pipe()) {
2184 retval = -1;
2185 goto exit_init_data;
2186 }
2187
2188 /* Init relay command queue. */
2189 cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2190
2191 /* Set up max poll set size */
2192 if (lttng_poll_set_max_size()) {
2193 retval = -1;
2194 goto exit_init_data;
2195 }
2196
2197 /* Setup the dispatcher thread */
2198 ret = pthread_create(&live_dispatcher_thread, default_pthread_attr(),
2199 thread_dispatcher, (void *) NULL);
2200 if (ret) {
2201 errno = ret;
2202 PERROR("pthread_create viewer dispatcher");
2203 retval = -1;
2204 goto exit_dispatcher_thread;
2205 }
2206
2207 /* Setup the worker thread */
2208 ret = pthread_create(&live_worker_thread, default_pthread_attr(),
2209 thread_worker, NULL);
2210 if (ret) {
2211 errno = ret;
2212 PERROR("pthread_create viewer worker");
2213 retval = -1;
2214 goto exit_worker_thread;
2215 }
2216
2217 /* Setup the listener thread */
2218 ret = pthread_create(&live_listener_thread, default_pthread_attr(),
2219 thread_listener, (void *) NULL);
2220 if (ret) {
2221 errno = ret;
2222 PERROR("pthread_create viewer listener");
2223 retval = -1;
2224 goto exit_listener_thread;
2225 }
2226
2227 /*
2228 * All OK, started all threads.
2229 */
2230 return retval;
2231
2232 /*
2233 * Join on the live_listener_thread should anything be added after
2234 * the live_listener thread's creation.
2235 */
2236
2237 exit_listener_thread:
2238
2239 ret = pthread_join(live_worker_thread, &status);
2240 if (ret) {
2241 errno = ret;
2242 PERROR("pthread_join live worker");
2243 retval = -1;
2244 }
2245 exit_worker_thread:
2246
2247 ret = pthread_join(live_dispatcher_thread, &status);
2248 if (ret) {
2249 errno = ret;
2250 PERROR("pthread_join live dispatcher");
2251 retval = -1;
2252 }
2253 exit_dispatcher_thread:
2254
2255 exit_init_data:
2256 cleanup_relayd_live();
2257
2258 return retval;
2259 }
This page took 0.127176 seconds and 5 git commands to generate.