relayd: track the live viewer worker thread's epoll fd
[deliverable/lttng-tools.git] / src / bin / lttng-relayd / live.c
1 /*
2 * Copyright (C) 2013 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
4 * 2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _LGPL_SOURCE
21 #include <getopt.h>
22 #include <grp.h>
23 #include <limits.h>
24 #include <pthread.h>
25 #include <signal.h>
26 #include <stdio.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <sys/mman.h>
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
33 #include <sys/stat.h>
34 #include <sys/types.h>
35 #include <sys/wait.h>
36 #include <inttypes.h>
37 #include <urcu/futex.h>
38 #include <urcu/uatomic.h>
39 #include <urcu/rculist.h>
40 #include <unistd.h>
41 #include <fcntl.h>
42
43 #include <lttng/lttng.h>
44 #include <common/common.h>
45 #include <common/compat/poll.h>
46 #include <common/compat/socket.h>
47 #include <common/compat/endian.h>
48 #include <common/defaults.h>
49 #include <common/futex.h>
50 #include <common/index/index.h>
51 #include <common/sessiond-comm/sessiond-comm.h>
52 #include <common/sessiond-comm/inet.h>
53 #include <common/sessiond-comm/relayd.h>
54 #include <common/uri.h>
55 #include <common/utils.h>
56 #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 * Check if the thread quit pipe was triggered.
423 *
424 * Return 1 if it was triggered else 0;
425 */
426 static
427 int check_thread_quit_pipe(int fd, uint32_t events)
428 {
429 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
430 return 1;
431 }
432
433 return 0;
434 }
435
436 /*
437 * Create and init socket from uri.
438 */
439 static
440 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
441 {
442 int ret;
443 struct lttcomm_sock *sock = NULL;
444
445 sock = lttcomm_alloc_sock_from_uri(uri);
446 if (sock == NULL) {
447 ERR("Allocating socket");
448 goto error;
449 }
450
451 ret = lttcomm_create_sock(sock);
452 if (ret < 0) {
453 goto error;
454 }
455 DBG("Listening on sock %d for lttng-live", sock->fd);
456
457 ret = sock->ops->bind(sock);
458 if (ret < 0) {
459 PERROR("Failed to bind lttng-live socket");
460 goto error;
461 }
462
463 ret = sock->ops->listen(sock, -1);
464 if (ret < 0) {
465 goto error;
466
467 }
468
469 return sock;
470
471 error:
472 if (sock) {
473 lttcomm_destroy_sock(sock);
474 }
475 return NULL;
476 }
477
478 /*
479 * This thread manages the listening for new connections on the network
480 */
481 static
482 void *thread_listener(void *data)
483 {
484 int i, ret, pollfd, err = -1;
485 uint32_t revents, nb_fd;
486 struct lttng_poll_event events;
487 struct lttcomm_sock *live_control_sock;
488
489 DBG("[thread] Relay live listener started");
490
491 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
492
493 health_code_update();
494
495 live_control_sock = init_socket(live_uri);
496 if (!live_control_sock) {
497 goto error_sock_control;
498 }
499
500 /* Pass 2 as size here for the thread quit pipe and control sockets. */
501 ret = create_named_thread_poll_set(&events, 2,
502 "Live listener thread epoll");
503 if (ret < 0) {
504 goto error_create_poll;
505 }
506
507 /* Add the control socket */
508 ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
509 if (ret < 0) {
510 goto error_poll_add;
511 }
512
513 lttng_relay_notify_ready();
514
515 if (testpoint(relayd_thread_live_listener)) {
516 goto error_testpoint;
517 }
518
519 while (1) {
520 health_code_update();
521
522 DBG("Listener accepting live viewers connections");
523
524 restart:
525 health_poll_entry();
526 ret = lttng_poll_wait(&events, -1);
527 health_poll_exit();
528 if (ret < 0) {
529 /*
530 * Restart interrupted system call.
531 */
532 if (errno == EINTR) {
533 goto restart;
534 }
535 goto error;
536 }
537 nb_fd = ret;
538
539 DBG("Relay new viewer connection received");
540 for (i = 0; i < nb_fd; i++) {
541 health_code_update();
542
543 /* Fetch once the poll data */
544 revents = LTTNG_POLL_GETEV(&events, i);
545 pollfd = LTTNG_POLL_GETFD(&events, i);
546
547 if (!revents) {
548 /* No activity for this FD (poll implementation). */
549 continue;
550 }
551
552 /* Thread quit pipe has been closed. Killing thread. */
553 ret = check_thread_quit_pipe(pollfd, revents);
554 if (ret) {
555 err = 0;
556 goto exit;
557 }
558
559 if (revents & LPOLLIN) {
560 /*
561 * A new connection is requested, therefore a
562 * viewer connection is allocated in this
563 * thread, enqueued to a global queue and
564 * dequeued (and freed) in the worker thread.
565 */
566 int val = 1;
567 struct relay_connection *new_conn;
568 struct lttcomm_sock *newsock;
569
570 newsock = live_control_sock->ops->accept(live_control_sock);
571 if (!newsock) {
572 PERROR("accepting control sock");
573 goto error;
574 }
575 DBG("Relay viewer connection accepted socket %d", newsock->fd);
576
577 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
578 sizeof(val));
579 if (ret < 0) {
580 PERROR("setsockopt inet");
581 lttcomm_destroy_sock(newsock);
582 goto error;
583 }
584 new_conn = connection_create(newsock, RELAY_CONNECTION_UNKNOWN);
585 if (!new_conn) {
586 lttcomm_destroy_sock(newsock);
587 goto error;
588 }
589 /* Ownership assumed by the connection. */
590 newsock = NULL;
591
592 /* Enqueue request for the dispatcher thread. */
593 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
594 &new_conn->qnode);
595
596 /*
597 * Wake the dispatch queue futex.
598 * Implicit memory barrier with the
599 * exchange in cds_wfcq_enqueue.
600 */
601 futex_nto1_wake(&viewer_conn_queue.futex);
602 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
603 ERR("socket poll error");
604 goto error;
605 } else {
606 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
607 goto error;
608 }
609 }
610 }
611
612 exit:
613 error:
614 error_poll_add:
615 error_testpoint:
616 (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
617 error_create_poll:
618 if (live_control_sock->fd >= 0) {
619 ret = live_control_sock->ops->close(live_control_sock);
620 if (ret) {
621 PERROR("close");
622 }
623 }
624 lttcomm_destroy_sock(live_control_sock);
625 error_sock_control:
626 if (err) {
627 health_error();
628 DBG("Live viewer listener thread exited with error");
629 }
630 health_unregister(health_relayd);
631 DBG("Live viewer listener thread cleanup complete");
632 if (lttng_relay_stop_threads()) {
633 ERR("Error stopping threads");
634 }
635 return NULL;
636 }
637
638 /*
639 * This thread manages the dispatching of the requests to worker threads
640 */
641 static
642 void *thread_dispatcher(void *data)
643 {
644 int err = -1;
645 ssize_t ret;
646 struct cds_wfcq_node *node;
647 struct relay_connection *conn = NULL;
648
649 DBG("[thread] Live viewer relay dispatcher started");
650
651 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
652
653 if (testpoint(relayd_thread_live_dispatcher)) {
654 goto error_testpoint;
655 }
656
657 health_code_update();
658
659 for (;;) {
660 health_code_update();
661
662 /* Atomically prepare the queue futex */
663 futex_nto1_prepare(&viewer_conn_queue.futex);
664
665 if (CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
666 break;
667 }
668
669 do {
670 health_code_update();
671
672 /* Dequeue commands */
673 node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
674 &viewer_conn_queue.tail);
675 if (node == NULL) {
676 DBG("Woken up but nothing in the live-viewer "
677 "relay command queue");
678 /* Continue thread execution */
679 break;
680 }
681 conn = caa_container_of(node, struct relay_connection, qnode);
682 DBG("Dispatching viewer request waiting on sock %d",
683 conn->sock->fd);
684
685 /*
686 * Inform worker thread of the new request. This
687 * call is blocking so we can be assured that
688 * the data will be read at some point in time
689 * or wait to the end of the world :)
690 */
691 ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
692 if (ret < 0) {
693 PERROR("write conn pipe");
694 connection_put(conn);
695 goto error;
696 }
697 } while (node != NULL);
698
699 /* Futex wait on queue. Blocking call on futex() */
700 health_poll_entry();
701 futex_nto1_wait(&viewer_conn_queue.futex);
702 health_poll_exit();
703 }
704
705 /* Normal exit, no error */
706 err = 0;
707
708 error:
709 error_testpoint:
710 if (err) {
711 health_error();
712 ERR("Health error occurred in %s", __func__);
713 }
714 health_unregister(health_relayd);
715 DBG("Live viewer dispatch thread dying");
716 if (lttng_relay_stop_threads()) {
717 ERR("Error stopping threads");
718 }
719 return NULL;
720 }
721
722 /*
723 * Establish connection with the viewer and check the versions.
724 *
725 * Return 0 on success or else negative value.
726 */
727 static
728 int viewer_connect(struct relay_connection *conn)
729 {
730 int ret;
731 struct lttng_viewer_connect reply, msg;
732
733 conn->version_check_done = 1;
734
735 health_code_update();
736
737 DBG("Viewer is establishing a connection to the relayd.");
738
739 ret = recv_request(conn->sock, &msg, sizeof(msg));
740 if (ret < 0) {
741 goto end;
742 }
743
744 health_code_update();
745
746 memset(&reply, 0, sizeof(reply));
747 reply.major = RELAYD_VERSION_COMM_MAJOR;
748 reply.minor = RELAYD_VERSION_COMM_MINOR;
749
750 /* Major versions must be the same */
751 if (reply.major != be32toh(msg.major)) {
752 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
753 reply.major, be32toh(msg.major));
754 ret = -1;
755 goto end;
756 }
757
758 conn->major = reply.major;
759 /* We adapt to the lowest compatible version */
760 if (reply.minor <= be32toh(msg.minor)) {
761 conn->minor = reply.minor;
762 } else {
763 conn->minor = be32toh(msg.minor);
764 }
765
766 if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
767 conn->type = RELAY_VIEWER_COMMAND;
768 } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
769 conn->type = RELAY_VIEWER_NOTIFICATION;
770 } else {
771 ERR("Unknown connection type : %u", be32toh(msg.type));
772 ret = -1;
773 goto end;
774 }
775
776 reply.major = htobe32(reply.major);
777 reply.minor = htobe32(reply.minor);
778 if (conn->type == RELAY_VIEWER_COMMAND) {
779 /*
780 * Increment outside of htobe64 macro, because the argument can
781 * be used more than once within the macro, and thus the
782 * operation may be undefined.
783 */
784 pthread_mutex_lock(&last_relay_viewer_session_id_lock);
785 last_relay_viewer_session_id++;
786 pthread_mutex_unlock(&last_relay_viewer_session_id_lock);
787 reply.viewer_session_id = htobe64(last_relay_viewer_session_id);
788 }
789
790 health_code_update();
791
792 ret = send_response(conn->sock, &reply, sizeof(reply));
793 if (ret < 0) {
794 goto end;
795 }
796
797 health_code_update();
798
799 DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
800 ret = 0;
801
802 end:
803 return ret;
804 }
805
806 /*
807 * Send the viewer the list of current sessions.
808 * We need to create a copy of the hash table content because otherwise
809 * we cannot assume the number of entries stays the same between getting
810 * the number of HT elements and iteration over the HT.
811 *
812 * Return 0 on success or else a negative value.
813 */
814 static
815 int viewer_list_sessions(struct relay_connection *conn)
816 {
817 int ret = 0;
818 struct lttng_viewer_list_sessions session_list;
819 struct lttng_ht_iter iter;
820 struct relay_session *session;
821 struct lttng_viewer_session *send_session_buf = NULL;
822 uint32_t buf_count = SESSION_BUF_DEFAULT_COUNT;
823 uint32_t count = 0;
824
825 DBG("List sessions received");
826
827 send_session_buf = zmalloc(SESSION_BUF_DEFAULT_COUNT * sizeof(*send_session_buf));
828 if (!send_session_buf) {
829 return -1;
830 }
831
832 rcu_read_lock();
833 cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, session,
834 session_n.node) {
835 struct lttng_viewer_session *send_session;
836
837 health_code_update();
838
839 if (count >= buf_count) {
840 struct lttng_viewer_session *newbuf;
841 uint32_t new_buf_count = buf_count << 1;
842
843 newbuf = realloc(send_session_buf,
844 new_buf_count * sizeof(*send_session_buf));
845 if (!newbuf) {
846 ret = -1;
847 break;
848 }
849 send_session_buf = newbuf;
850 buf_count = new_buf_count;
851 }
852 send_session = &send_session_buf[count];
853 if (lttng_strncpy(send_session->session_name,
854 session->session_name,
855 sizeof(send_session->session_name))) {
856 ret = -1;
857 break;
858 }
859 if (lttng_strncpy(send_session->hostname, session->hostname,
860 sizeof(send_session->hostname))) {
861 ret = -1;
862 break;
863 }
864 send_session->id = htobe64(session->id);
865 send_session->live_timer = htobe32(session->live_timer);
866 if (session->viewer_attached) {
867 send_session->clients = htobe32(1);
868 } else {
869 send_session->clients = htobe32(0);
870 }
871 send_session->streams = htobe32(session->stream_count);
872 count++;
873 }
874 rcu_read_unlock();
875 if (ret < 0) {
876 goto end_free;
877 }
878
879 session_list.sessions_count = htobe32(count);
880
881 health_code_update();
882
883 ret = send_response(conn->sock, &session_list, sizeof(session_list));
884 if (ret < 0) {
885 goto end_free;
886 }
887
888 health_code_update();
889
890 ret = send_response(conn->sock, send_session_buf,
891 count * sizeof(*send_session_buf));
892 if (ret < 0) {
893 goto end_free;
894 }
895 health_code_update();
896
897 ret = 0;
898 end_free:
899 free(send_session_buf);
900 return ret;
901 }
902
903 /*
904 * Send the viewer the list of current streams.
905 */
906 static
907 int viewer_get_new_streams(struct relay_connection *conn)
908 {
909 int ret, send_streams = 0;
910 uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0, nb_total = 0;
911 struct lttng_viewer_new_streams_request request;
912 struct lttng_viewer_new_streams_response response;
913 struct relay_session *session;
914 uint64_t session_id;
915 bool closed = false;
916
917 assert(conn);
918
919 DBG("Get new streams received");
920
921 health_code_update();
922
923 /* Receive the request from the connected client. */
924 ret = recv_request(conn->sock, &request, sizeof(request));
925 if (ret < 0) {
926 goto error;
927 }
928 session_id = be64toh(request.session_id);
929
930 health_code_update();
931
932 memset(&response, 0, sizeof(response));
933
934 session = session_get_by_id(session_id);
935 if (!session) {
936 DBG("Relay session %" PRIu64 " not found", session_id);
937 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
938 goto send_reply;
939 }
940
941 if (!viewer_session_is_attached(conn->viewer_session, session)) {
942 send_streams = 0;
943 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
944 goto send_reply;
945 }
946
947 send_streams = 1;
948 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
949
950 ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, &nb_total, &nb_unsent,
951 &nb_created, &closed);
952 if (ret < 0) {
953 goto end_put_session;
954 }
955 /* Only send back the newly created streams with the unsent ones. */
956 nb_streams = nb_created + nb_unsent;
957 response.streams_count = htobe32(nb_streams);
958
959 /*
960 * If the session is closed, HUP when there are no more streams
961 * with data.
962 */
963 if (closed && nb_total == 0) {
964 send_streams = 0;
965 response.streams_count = 0;
966 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
967 goto send_reply;
968 }
969
970 send_reply:
971 health_code_update();
972 ret = send_response(conn->sock, &response, sizeof(response));
973 if (ret < 0) {
974 goto end_put_session;
975 }
976 health_code_update();
977
978 /*
979 * Unknown or empty session, just return gracefully, the viewer
980 * knows what is happening.
981 */
982 if (!send_streams || !nb_streams) {
983 ret = 0;
984 goto end_put_session;
985 }
986
987 /*
988 * Send stream and *DON'T* ignore the sent flag so every viewer
989 * streams that were not sent from that point will be sent to
990 * the viewer.
991 */
992 ret = send_viewer_streams(conn->sock, session, 0);
993 if (ret < 0) {
994 goto end_put_session;
995 }
996
997 end_put_session:
998 if (session) {
999 session_put(session);
1000 }
1001 error:
1002 return ret;
1003 }
1004
1005 /*
1006 * Send the viewer the list of current sessions.
1007 */
1008 static
1009 int viewer_attach_session(struct relay_connection *conn)
1010 {
1011 int send_streams = 0;
1012 ssize_t ret;
1013 uint32_t nb_streams = 0;
1014 enum lttng_viewer_seek seek_type;
1015 struct lttng_viewer_attach_session_request request;
1016 struct lttng_viewer_attach_session_response response;
1017 struct relay_session *session = NULL;
1018 bool closed = false;
1019
1020 assert(conn);
1021
1022 health_code_update();
1023
1024 /* Receive the request from the connected client. */
1025 ret = recv_request(conn->sock, &request, sizeof(request));
1026 if (ret < 0) {
1027 goto error;
1028 }
1029
1030 health_code_update();
1031
1032 memset(&response, 0, sizeof(response));
1033
1034 if (!conn->viewer_session) {
1035 DBG("Client trying to attach before creating a live viewer session");
1036 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1037 goto send_reply;
1038 }
1039
1040 session = session_get_by_id(be64toh(request.session_id));
1041 if (!session) {
1042 DBG("Relay session %" PRIu64 " not found",
1043 (uint64_t) be64toh(request.session_id));
1044 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1045 goto send_reply;
1046 }
1047 DBG("Attach session ID %" PRIu64 " received",
1048 (uint64_t) be64toh(request.session_id));
1049
1050 if (session->live_timer == 0) {
1051 DBG("Not live session");
1052 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1053 goto send_reply;
1054 }
1055
1056 send_streams = 1;
1057 ret = viewer_session_attach(conn->viewer_session, session);
1058 if (ret) {
1059 DBG("Already a viewer attached");
1060 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1061 goto send_reply;
1062 }
1063
1064 switch (be32toh(request.seek)) {
1065 case LTTNG_VIEWER_SEEK_BEGINNING:
1066 case LTTNG_VIEWER_SEEK_LAST:
1067 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1068 seek_type = be32toh(request.seek);
1069 break;
1070 default:
1071 ERR("Wrong seek parameter");
1072 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1073 send_streams = 0;
1074 goto send_reply;
1075 }
1076
1077 ret = make_viewer_streams(session, seek_type, &nb_streams, NULL,
1078 NULL, &closed);
1079 if (ret < 0) {
1080 goto end_put_session;
1081 }
1082 response.streams_count = htobe32(nb_streams);
1083
1084 /*
1085 * If the session is closed when the viewer is attaching, it
1086 * means some of the streams may have been concurrently removed,
1087 * so we don't allow the viewer to attach, even if there are
1088 * streams available.
1089 */
1090 if (closed) {
1091 send_streams = 0;
1092 response.streams_count = 0;
1093 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
1094 goto send_reply;
1095 }
1096
1097 send_reply:
1098 health_code_update();
1099 ret = send_response(conn->sock, &response, sizeof(response));
1100 if (ret < 0) {
1101 goto end_put_session;
1102 }
1103 health_code_update();
1104
1105 /*
1106 * Unknown or empty session, just return gracefully, the viewer
1107 * knows what is happening.
1108 */
1109 if (!send_streams || !nb_streams) {
1110 ret = 0;
1111 goto end_put_session;
1112 }
1113
1114 /* Send stream and ignore the sent flag. */
1115 ret = send_viewer_streams(conn->sock, session, 1);
1116 if (ret < 0) {
1117 goto end_put_session;
1118 }
1119
1120 end_put_session:
1121 if (session) {
1122 session_put(session);
1123 }
1124 error:
1125 return ret;
1126 }
1127
1128 /*
1129 * Open the index file if needed for the given vstream.
1130 *
1131 * If an index file is successfully opened, the vstream will set it as its
1132 * current index file.
1133 *
1134 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1135 *
1136 * Called with rstream lock held.
1137 */
1138 static int try_open_index(struct relay_viewer_stream *vstream,
1139 struct relay_stream *rstream)
1140 {
1141 int ret = 0;
1142
1143 if (vstream->index_file) {
1144 goto end;
1145 }
1146
1147 /*
1148 * First time, we open the index file and at least one index is ready.
1149 */
1150 if (rstream->index_received_seqcount == 0) {
1151 ret = -ENOENT;
1152 goto end;
1153 }
1154 vstream->index_file = lttng_index_file_open(vstream->path_name,
1155 vstream->channel_name,
1156 vstream->stream->tracefile_count,
1157 vstream->current_tracefile_id);
1158 if (!vstream->index_file) {
1159 ret = -1;
1160 }
1161
1162 end:
1163 return ret;
1164 }
1165
1166 /*
1167 * Check the status of the index for the given stream. This function
1168 * updates the index structure if needed and can put (close) the vstream
1169 * in the HUP situation.
1170 *
1171 * Return 0 means that we can proceed with the index. A value of 1 means
1172 * that the index has been updated and is ready to be sent to the
1173 * client. A negative value indicates an error that can't be handled.
1174 *
1175 * Called with rstream lock held.
1176 */
1177 static int check_index_status(struct relay_viewer_stream *vstream,
1178 struct relay_stream *rstream, struct ctf_trace *trace,
1179 struct lttng_viewer_index *index)
1180 {
1181 int ret;
1182
1183 if ((trace->session->connection_closed || rstream->closed)
1184 && rstream->index_received_seqcount
1185 == vstream->index_sent_seqcount) {
1186 /*
1187 * Last index sent and session connection or relay
1188 * stream are closed.
1189 */
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_named_thread_poll_set(&events, 2,
1955 "Live viewer worker thread epoll");
1956 if (ret < 0) {
1957 goto error_poll_create;
1958 }
1959
1960 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1961 if (ret < 0) {
1962 goto error;
1963 }
1964
1965 restart:
1966 while (1) {
1967 int i;
1968
1969 health_code_update();
1970
1971 /* Infinite blocking call, waiting for transmission */
1972 DBG3("Relayd live viewer worker thread polling...");
1973 health_poll_entry();
1974 ret = lttng_poll_wait(&events, -1);
1975 health_poll_exit();
1976 if (ret < 0) {
1977 /*
1978 * Restart interrupted system call.
1979 */
1980 if (errno == EINTR) {
1981 goto restart;
1982 }
1983 goto error;
1984 }
1985
1986 nb_fd = ret;
1987
1988 /*
1989 * Process control. The control connection is prioritised so we don't
1990 * starve it with high throughput tracing data on the data
1991 * connection.
1992 */
1993 for (i = 0; i < nb_fd; i++) {
1994 /* Fetch once the poll data */
1995 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1996 int pollfd = LTTNG_POLL_GETFD(&events, i);
1997
1998 health_code_update();
1999
2000 if (!revents) {
2001 /* No activity for this FD (poll implementation). */
2002 continue;
2003 }
2004
2005 /* Thread quit pipe has been closed. Killing thread. */
2006 ret = check_thread_quit_pipe(pollfd, revents);
2007 if (ret) {
2008 err = 0;
2009 goto exit;
2010 }
2011
2012 /* Inspect the relay conn pipe for new connection. */
2013 if (pollfd == live_conn_pipe[0]) {
2014 if (revents & LPOLLIN) {
2015 struct relay_connection *conn;
2016
2017 ret = lttng_read(live_conn_pipe[0],
2018 &conn, sizeof(conn));
2019 if (ret < 0) {
2020 goto error;
2021 }
2022 lttng_poll_add(&events, conn->sock->fd,
2023 LPOLLIN | LPOLLRDHUP);
2024 connection_ht_add(viewer_connections_ht, conn);
2025 DBG("Connection socket %d added to poll", conn->sock->fd);
2026 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2027 ERR("Relay live pipe error");
2028 goto error;
2029 } else {
2030 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2031 goto error;
2032 }
2033 } else {
2034 /* Connection activity. */
2035 struct relay_connection *conn;
2036
2037 conn = connection_get_by_sock(viewer_connections_ht, pollfd);
2038 if (!conn) {
2039 continue;
2040 }
2041
2042 if (revents & LPOLLIN) {
2043 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
2044 sizeof(recv_hdr), 0);
2045 if (ret <= 0) {
2046 /* Connection closed. */
2047 cleanup_connection_pollfd(&events, pollfd);
2048 /* Put "create" ownership reference. */
2049 connection_put(conn);
2050 DBG("Viewer control conn closed with %d", pollfd);
2051 } else {
2052 ret = process_control(&recv_hdr, conn);
2053 if (ret < 0) {
2054 /* Clear the session on error. */
2055 cleanup_connection_pollfd(&events, pollfd);
2056 /* Put "create" ownership reference. */
2057 connection_put(conn);
2058 DBG("Viewer connection closed with %d", pollfd);
2059 }
2060 }
2061 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2062 cleanup_connection_pollfd(&events, pollfd);
2063 /* Put "create" ownership reference. */
2064 connection_put(conn);
2065 } else {
2066 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2067 connection_put(conn);
2068 goto error;
2069 }
2070 /* Put local "get_by_sock" reference. */
2071 connection_put(conn);
2072 }
2073 }
2074 }
2075
2076 exit:
2077 error:
2078 (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
2079
2080 /* Cleanup remaining connection object. */
2081 rcu_read_lock();
2082 cds_lfht_for_each_entry(viewer_connections_ht->ht, &iter.iter,
2083 destroy_conn,
2084 sock_n.node) {
2085 health_code_update();
2086 connection_put(destroy_conn);
2087 }
2088 rcu_read_unlock();
2089 error_poll_create:
2090 lttng_ht_destroy(viewer_connections_ht);
2091 viewer_connections_ht_error:
2092 /* Close relay conn pipes */
2093 (void) fd_tracker_util_pipe_close(the_fd_tracker, live_conn_pipe);
2094 if (err) {
2095 DBG("Viewer worker thread exited with error");
2096 }
2097 DBG("Viewer worker thread cleanup complete");
2098 error_testpoint:
2099 if (err) {
2100 health_error();
2101 ERR("Health error occurred in %s", __func__);
2102 }
2103 health_unregister(health_relayd);
2104 if (lttng_relay_stop_threads()) {
2105 ERR("Error stopping threads");
2106 }
2107 rcu_unregister_thread();
2108 return NULL;
2109 }
2110
2111 /*
2112 * Create the relay command pipe to wake thread_manage_apps.
2113 * Closed in cleanup().
2114 */
2115 static int create_conn_pipe(void)
2116 {
2117 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
2118 "Live connection pipe", live_conn_pipe);
2119 }
2120
2121 int relayd_live_join(void)
2122 {
2123 int ret, retval = 0;
2124 void *status;
2125
2126 ret = pthread_join(live_listener_thread, &status);
2127 if (ret) {
2128 errno = ret;
2129 PERROR("pthread_join live listener");
2130 retval = -1;
2131 }
2132
2133 ret = pthread_join(live_worker_thread, &status);
2134 if (ret) {
2135 errno = ret;
2136 PERROR("pthread_join live worker");
2137 retval = -1;
2138 }
2139
2140 ret = pthread_join(live_dispatcher_thread, &status);
2141 if (ret) {
2142 errno = ret;
2143 PERROR("pthread_join live dispatcher");
2144 retval = -1;
2145 }
2146
2147 cleanup_relayd_live();
2148
2149 return retval;
2150 }
2151
2152 /*
2153 * main
2154 */
2155 int relayd_live_create(struct lttng_uri *uri)
2156 {
2157 int ret = 0, retval = 0;
2158 void *status;
2159 int is_root;
2160
2161 if (!uri) {
2162 retval = -1;
2163 goto exit_init_data;
2164 }
2165 live_uri = uri;
2166
2167 /* Check if daemon is UID = 0 */
2168 is_root = !getuid();
2169
2170 if (!is_root) {
2171 if (live_uri->port < 1024) {
2172 ERR("Need to be root to use ports < 1024");
2173 retval = -1;
2174 goto exit_init_data;
2175 }
2176 }
2177
2178 /* Setup the thread apps communication pipe. */
2179 if (create_conn_pipe()) {
2180 retval = -1;
2181 goto exit_init_data;
2182 }
2183
2184 /* Init relay command queue. */
2185 cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2186
2187 /* Set up max poll set size */
2188 if (lttng_poll_set_max_size()) {
2189 retval = -1;
2190 goto exit_init_data;
2191 }
2192
2193 /* Setup the dispatcher thread */
2194 ret = pthread_create(&live_dispatcher_thread, default_pthread_attr(),
2195 thread_dispatcher, (void *) NULL);
2196 if (ret) {
2197 errno = ret;
2198 PERROR("pthread_create viewer dispatcher");
2199 retval = -1;
2200 goto exit_dispatcher_thread;
2201 }
2202
2203 /* Setup the worker thread */
2204 ret = pthread_create(&live_worker_thread, default_pthread_attr(),
2205 thread_worker, NULL);
2206 if (ret) {
2207 errno = ret;
2208 PERROR("pthread_create viewer worker");
2209 retval = -1;
2210 goto exit_worker_thread;
2211 }
2212
2213 /* Setup the listener thread */
2214 ret = pthread_create(&live_listener_thread, default_pthread_attr(),
2215 thread_listener, (void *) NULL);
2216 if (ret) {
2217 errno = ret;
2218 PERROR("pthread_create viewer listener");
2219 retval = -1;
2220 goto exit_listener_thread;
2221 }
2222
2223 /*
2224 * All OK, started all threads.
2225 */
2226 return retval;
2227
2228 /*
2229 * Join on the live_listener_thread should anything be added after
2230 * the live_listener thread's creation.
2231 */
2232
2233 exit_listener_thread:
2234
2235 ret = pthread_join(live_worker_thread, &status);
2236 if (ret) {
2237 errno = ret;
2238 PERROR("pthread_join live worker");
2239 retval = -1;
2240 }
2241 exit_worker_thread:
2242
2243 ret = pthread_join(live_dispatcher_thread, &status);
2244 if (ret) {
2245 errno = ret;
2246 PERROR("pthread_join live dispatcher");
2247 retval = -1;
2248 }
2249 exit_dispatcher_thread:
2250
2251 exit_init_data:
2252 cleanup_relayd_live();
2253
2254 return retval;
2255 }
This page took 0.104006 seconds and 6 git commands to generate.