Fix: relayd: failure to read index entry or stream packet after clear
[lttng-tools.git] / src / bin / lttng-relayd / live.c
1 /*
2 * Copyright (C) 2013 Julien Desfossez <jdesfossez@efficios.com>
3 * Copyright (C) 2013 David Goulet <dgoulet@efficios.com>
4 * Copyright (C) 2015 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5 *
6 * SPDX-License-Identifier: GPL-2.0-only
7 *
8 */
9
10 #define _LGPL_SOURCE
11 #include <fcntl.h>
12 #include <getopt.h>
13 #include <grp.h>
14 #include <inttypes.h>
15 #include <limits.h>
16 #include <pthread.h>
17 #include <signal.h>
18 #include <stdio.h>
19 #include <stdlib.h>
20 #include <string.h>
21 #include <sys/mman.h>
22 #include <sys/mount.h>
23 #include <sys/resource.h>
24 #include <sys/socket.h>
25 #include <sys/stat.h>
26 #include <sys/types.h>
27 #include <sys/wait.h>
28 #include <unistd.h>
29 #include <urcu/futex.h>
30 #include <urcu/rculist.h>
31 #include <urcu/uatomic.h>
32
33 #include <common/common.h>
34 #include <common/compat/endian.h>
35 #include <common/compat/poll.h>
36 #include <common/compat/socket.h>
37 #include <common/defaults.h>
38 #include <common/fd-tracker/utils.h>
39 #include <common/fs-handle.h>
40 #include <common/futex.h>
41 #include <common/index/index.h>
42 #include <common/sessiond-comm/inet.h>
43 #include <common/sessiond-comm/relayd.h>
44 #include <common/sessiond-comm/sessiond-comm.h>
45 #include <common/uri.h>
46 #include <common/utils.h>
47 #include <lttng/lttng.h>
48
49 #include "cmd.h"
50 #include "connection.h"
51 #include "ctf-trace.h"
52 #include "health-relayd.h"
53 #include "live.h"
54 #include "lttng-relayd.h"
55 #include "session.h"
56 #include "stream.h"
57 #include "testpoint.h"
58 #include "utils.h"
59 #include "viewer-session.h"
60 #include "viewer-stream.h"
61
62 #define SESSION_BUF_DEFAULT_COUNT 16
63
64 static struct lttng_uri *live_uri;
65
66 /*
67 * This pipe is used to inform the worker thread that a command is queued and
68 * ready to be processed.
69 */
70 static int live_conn_pipe[2] = { -1, -1 };
71
72 /* Shared between threads */
73 static int live_dispatch_thread_exit;
74
75 static pthread_t live_listener_thread;
76 static pthread_t live_dispatcher_thread;
77 static pthread_t live_worker_thread;
78
79 /*
80 * Relay command queue.
81 *
82 * The live_thread_listener and live_thread_dispatcher communicate with this
83 * queue.
84 */
85 static struct relay_conn_queue viewer_conn_queue;
86
87 static uint64_t last_relay_viewer_session_id;
88 static pthread_mutex_t last_relay_viewer_session_id_lock =
89 PTHREAD_MUTEX_INITIALIZER;
90
91 /*
92 * Cleanup the daemon
93 */
94 static
95 void cleanup_relayd_live(void)
96 {
97 DBG("Cleaning up");
98
99 free(live_uri);
100 }
101
102 /*
103 * Receive a request buffer using a given socket, destination allocated buffer
104 * of length size.
105 *
106 * Return the size of the received message or else a negative value on error
107 * with errno being set by recvmsg() syscall.
108 */
109 static
110 ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
111 {
112 ssize_t ret;
113
114 ret = sock->ops->recvmsg(sock, buf, size, 0);
115 if (ret < 0 || ret != size) {
116 if (ret == 0) {
117 /* Orderly shutdown. Not necessary to print an error. */
118 DBG("Socket %d did an orderly shutdown", sock->fd);
119 } else {
120 ERR("Relay failed to receive request.");
121 }
122 ret = -1;
123 }
124
125 return ret;
126 }
127
128 /*
129 * Send a response buffer using a given socket, source allocated buffer of
130 * length size.
131 *
132 * Return the size of the sent message or else a negative value on error with
133 * errno being set by sendmsg() syscall.
134 */
135 static
136 ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
137 {
138 ssize_t ret;
139
140 ret = sock->ops->sendmsg(sock, buf, size, 0);
141 if (ret < 0) {
142 ERR("Relayd failed to send response.");
143 }
144
145 return ret;
146 }
147
148 /*
149 * Atomically check if new streams got added in one of the sessions attached
150 * and reset the flag to 0.
151 *
152 * Returns 1 if new streams got added, 0 if nothing changed, a negative value
153 * on error.
154 */
155 static
156 int check_new_streams(struct relay_connection *conn)
157 {
158 struct relay_session *session;
159 unsigned long current_val;
160 int ret = 0;
161
162 if (!conn->viewer_session) {
163 goto end;
164 }
165 rcu_read_lock();
166 cds_list_for_each_entry_rcu(session,
167 &conn->viewer_session->session_list,
168 viewer_session_node) {
169 if (!session_get(session)) {
170 continue;
171 }
172 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
173 ret = current_val;
174 session_put(session);
175 if (ret == 1) {
176 goto end;
177 }
178 }
179 end:
180 rcu_read_unlock();
181 return ret;
182 }
183
184 /*
185 * Send viewer streams to the given socket. The ignore_sent_flag indicates if
186 * this function should ignore the sent flag or not.
187 *
188 * Return 0 on success or else a negative value.
189 */
190 static
191 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
192 uint64_t session_id, unsigned int ignore_sent_flag)
193 {
194 ssize_t ret;
195 struct lttng_ht_iter iter;
196 struct relay_viewer_stream *vstream;
197
198 rcu_read_lock();
199
200 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
201 stream_n.node) {
202 struct ctf_trace *ctf_trace;
203 struct lttng_viewer_stream send_stream = {};
204
205 health_code_update();
206
207 if (!viewer_stream_get(vstream)) {
208 continue;
209 }
210
211 pthread_mutex_lock(&vstream->stream->lock);
212 /* Ignore if not the same session. */
213 if (vstream->stream->trace->session->id != session_id ||
214 (!ignore_sent_flag && vstream->sent_flag)) {
215 pthread_mutex_unlock(&vstream->stream->lock);
216 viewer_stream_put(vstream);
217 continue;
218 }
219
220 ctf_trace = vstream->stream->trace;
221 send_stream.id = htobe64(vstream->stream->stream_handle);
222 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
223 send_stream.metadata_flag = htobe32(
224 vstream->stream->is_metadata);
225 if (lttng_strncpy(send_stream.path_name, vstream->path_name,
226 sizeof(send_stream.path_name))) {
227 pthread_mutex_unlock(&vstream->stream->lock);
228 viewer_stream_put(vstream);
229 ret = -1; /* Error. */
230 goto end_unlock;
231 }
232 if (lttng_strncpy(send_stream.channel_name,
233 vstream->channel_name,
234 sizeof(send_stream.channel_name))) {
235 pthread_mutex_unlock(&vstream->stream->lock);
236 viewer_stream_put(vstream);
237 ret = -1; /* Error. */
238 goto end_unlock;
239 }
240
241 DBG("Sending stream %" PRIu64 " to viewer",
242 vstream->stream->stream_handle);
243 vstream->sent_flag = 1;
244 pthread_mutex_unlock(&vstream->stream->lock);
245
246 ret = send_response(sock, &send_stream, sizeof(send_stream));
247 viewer_stream_put(vstream);
248 if (ret < 0) {
249 goto end_unlock;
250 }
251 }
252
253 ret = 0;
254
255 end_unlock:
256 rcu_read_unlock();
257 return ret;
258 }
259
260 /*
261 * Create every viewer stream possible for the given session with the seek
262 * type. Three counters *can* be return which are in order the total amount of
263 * viewer stream of the session, the number of unsent stream and the number of
264 * stream created. Those counters can be NULL and thus will be ignored.
265 *
266 * session must be locked to ensure that we see either none or all initial
267 * streams for a session, but no intermediate state..
268 *
269 * Return 0 on success or else a negative value.
270 */
271 static int make_viewer_streams(struct relay_session *relay_session,
272 struct relay_viewer_session *viewer_session,
273 enum lttng_viewer_seek seek_t,
274 uint32_t *nb_total,
275 uint32_t *nb_unsent,
276 uint32_t *nb_created,
277 bool *closed)
278 {
279 int ret;
280 struct lttng_ht_iter iter;
281 struct ctf_trace *ctf_trace;
282 struct relay_stream *relay_stream = NULL;
283
284 assert(relay_session);
285 ASSERT_LOCKED(relay_session->lock);
286
287 if (relay_session->connection_closed) {
288 *closed = true;
289 }
290
291 /*
292 * Create viewer streams for relay streams that are ready to be
293 * used for a the given session id only.
294 */
295 rcu_read_lock();
296 cds_lfht_for_each_entry (relay_session->ctf_traces_ht->ht, &iter.iter,
297 ctf_trace, node.node) {
298 bool trace_has_metadata_stream = false;
299
300 health_code_update();
301
302 if (!ctf_trace_get(ctf_trace)) {
303 continue;
304 }
305
306 /*
307 * Iterate over all the streams of the trace to see if we have a
308 * metadata stream.
309 */
310 cds_list_for_each_entry_rcu(relay_stream,
311 &ctf_trace->stream_list, stream_node)
312 {
313 bool is_metadata_stream;
314
315 pthread_mutex_lock(&relay_stream->lock);
316 is_metadata_stream = relay_stream->is_metadata;
317 pthread_mutex_unlock(&relay_stream->lock);
318
319 if (is_metadata_stream) {
320 trace_has_metadata_stream = true;
321 break;
322 }
323 }
324
325 relay_stream = NULL;
326
327 /*
328 * If there is no metadata stream in this trace at the moment
329 * and we never sent one to the viewer, skip the trace. We
330 * accept that the viewer will not see this trace at all.
331 */
332 if (!trace_has_metadata_stream &&
333 !ctf_trace->metadata_stream_sent_to_viewer) {
334 ctf_trace_put(ctf_trace);
335 continue;
336 }
337
338 cds_list_for_each_entry_rcu(relay_stream,
339 &ctf_trace->stream_list, stream_node)
340 {
341 struct relay_viewer_stream *viewer_stream;
342
343 if (!stream_get(relay_stream)) {
344 continue;
345 }
346
347 pthread_mutex_lock(&relay_stream->lock);
348 /*
349 * stream published is protected by the session lock.
350 */
351 if (!relay_stream->published) {
352 goto next;
353 }
354 viewer_stream = viewer_stream_get_by_id(
355 relay_stream->stream_handle);
356 if (!viewer_stream) {
357 struct lttng_trace_chunk *viewer_stream_trace_chunk = NULL;
358
359 /*
360 * Save that we sent the metadata stream to the
361 * viewer. So that we know what trace the viewer
362 * is aware of.
363 */
364 if (relay_stream->is_metadata) {
365 ctf_trace->metadata_stream_sent_to_viewer = true;
366 }
367
368 /*
369 * If a rotation is ongoing, use a copy of the
370 * relay stream's chunk to ensure the stream
371 * files exist.
372 *
373 * Otherwise, the viewer session's current trace
374 * chunk can be used safely.
375 */
376 if ((relay_stream->ongoing_rotation.is_set ||
377 relay_session->ongoing_rotation) &&
378 relay_stream->trace_chunk) {
379 viewer_stream_trace_chunk = lttng_trace_chunk_copy(
380 relay_stream->trace_chunk);
381 if (!viewer_stream_trace_chunk) {
382 ret = -1;
383 ctf_trace_put(ctf_trace);
384 goto error_unlock;
385 }
386 } else {
387 bool reference_acquired;
388
389 /*
390 * Transition the viewer session into the newest trace chunk available.
391 */
392 if (!lttng_trace_chunk_ids_equal(viewer_session->current_trace_chunk,
393 relay_stream->trace_chunk)) {
394
395 ret = viewer_session_set_trace_chunk_copy(
396 viewer_session,
397 relay_stream->trace_chunk);
398 if (ret) {
399 ret = -1;
400 ctf_trace_put(ctf_trace);
401 goto error_unlock;
402 }
403 }
404
405 reference_acquired = lttng_trace_chunk_get(
406 viewer_session->current_trace_chunk);
407 assert(reference_acquired);
408 viewer_stream_trace_chunk =
409 viewer_session->current_trace_chunk;
410 }
411
412 viewer_stream = viewer_stream_create(
413 relay_stream,
414 viewer_stream_trace_chunk,
415 seek_t);
416 lttng_trace_chunk_put(viewer_stream_trace_chunk);
417 viewer_stream_trace_chunk = NULL;
418 if (!viewer_stream) {
419 ret = -1;
420 ctf_trace_put(ctf_trace);
421 goto error_unlock;
422 }
423
424 if (nb_created) {
425 /* Update number of created stream counter. */
426 (*nb_created)++;
427 }
428 /*
429 * Ensure a self-reference is preserved even
430 * after we have put our local reference.
431 */
432 if (!viewer_stream_get(viewer_stream)) {
433 ERR("Unable to get self-reference on viewer stream, logic error.");
434 abort();
435 }
436 } else {
437 if (!viewer_stream->sent_flag && nb_unsent) {
438 /* Update number of unsent stream counter. */
439 (*nb_unsent)++;
440 }
441 }
442 /* Update number of total stream counter. */
443 if (nb_total) {
444 if (relay_stream->is_metadata) {
445 if (!relay_stream->closed ||
446 relay_stream->metadata_received >
447 viewer_stream->metadata_sent) {
448 (*nb_total)++;
449 }
450 } else {
451 if (!relay_stream->closed ||
452 !(((int64_t)(relay_stream->prev_data_seq -
453 relay_stream->last_net_seq_num)) >=
454 0)) {
455 (*nb_total)++;
456 }
457 }
458 }
459 /* Put local reference. */
460 viewer_stream_put(viewer_stream);
461 next:
462 pthread_mutex_unlock(&relay_stream->lock);
463 stream_put(relay_stream);
464 }
465 relay_stream = NULL;
466 ctf_trace_put(ctf_trace);
467 }
468
469 ret = 0;
470
471 error_unlock:
472 rcu_read_unlock();
473
474 if (relay_stream) {
475 pthread_mutex_unlock(&relay_stream->lock);
476 stream_put(relay_stream);
477 }
478
479 return ret;
480 }
481
482 int relayd_live_stop(void)
483 {
484 /* Stop dispatch thread */
485 CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
486 futex_nto1_wake(&viewer_conn_queue.futex);
487 return 0;
488 }
489
490 /*
491 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
492 */
493 static
494 int create_named_thread_poll_set(struct lttng_poll_event *events,
495 int size, const char *name)
496 {
497 int ret;
498
499 if (events == NULL || size == 0) {
500 ret = -1;
501 goto error;
502 }
503
504 ret = fd_tracker_util_poll_create(the_fd_tracker,
505 name, events, 1, LTTNG_CLOEXEC);
506 if (ret) {
507 PERROR("Failed to create \"%s\" poll file descriptor", name);
508 goto error;
509 }
510
511 /* Add quit pipe */
512 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
513 if (ret < 0) {
514 goto error;
515 }
516
517 return 0;
518
519 error:
520 return ret;
521 }
522
523 /*
524 * Check if the thread quit pipe was triggered.
525 *
526 * Return 1 if it was triggered else 0;
527 */
528 static
529 int check_thread_quit_pipe(int fd, uint32_t events)
530 {
531 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
532 return 1;
533 }
534
535 return 0;
536 }
537
538 static
539 int create_sock(void *data, int *out_fd)
540 {
541 int ret;
542 struct lttcomm_sock *sock = data;
543
544 ret = lttcomm_create_sock(sock);
545 if (ret < 0) {
546 goto end;
547 }
548
549 *out_fd = sock->fd;
550 end:
551 return ret;
552 }
553
554 static
555 int close_sock(void *data, int *in_fd)
556 {
557 struct lttcomm_sock *sock = data;
558
559 return sock->ops->close(sock);
560 }
561
562 static int accept_sock(void *data, int *out_fd)
563 {
564 int ret = 0;
565 /* Socks is an array of in_sock, out_sock. */
566 struct lttcomm_sock **socks = data;
567 struct lttcomm_sock *in_sock = socks[0];
568
569 socks[1] = in_sock->ops->accept(in_sock);
570 if (!socks[1]) {
571 ret = -1;
572 goto end;
573 }
574 *out_fd = socks[1]->fd;
575 end:
576 return ret;
577 }
578
579 static
580 struct lttcomm_sock *accept_live_sock(struct lttcomm_sock *listening_sock,
581 const char *name)
582 {
583 int out_fd, ret;
584 struct lttcomm_sock *socks[2] = { listening_sock, NULL };
585 struct lttcomm_sock *new_sock = NULL;
586
587 ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker, &out_fd,
588 (const char **) &name, 1, accept_sock, &socks);
589 if (ret) {
590 goto end;
591 }
592 new_sock = socks[1];
593 DBG("%s accepted, socket %d", name, new_sock->fd);
594 end:
595 return new_sock;
596 }
597
598 /*
599 * Create and init socket from uri.
600 */
601 static
602 struct lttcomm_sock *init_socket(struct lttng_uri *uri, const char *name)
603 {
604 int ret, sock_fd;
605 struct lttcomm_sock *sock = NULL;
606 char uri_str[LTTNG_PATH_MAX];
607 char *formated_name = NULL;
608
609 sock = lttcomm_alloc_sock_from_uri(uri);
610 if (sock == NULL) {
611 ERR("Allocating socket");
612 goto error;
613 }
614
615 /*
616 * Don't fail to create the socket if the name can't be built as it is
617 * only used for debugging purposes.
618 */
619 ret = uri_to_str_url(uri, uri_str, sizeof(uri_str));
620 uri_str[sizeof(uri_str) - 1] = '\0';
621 if (ret >= 0) {
622 ret = asprintf(&formated_name, "%s socket @ %s", name,
623 uri_str);
624 if (ret < 0) {
625 formated_name = NULL;
626 }
627 }
628
629 ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker, &sock_fd,
630 (const char **) (formated_name ? &formated_name : NULL),
631 1, create_sock, sock);
632 if (ret) {
633 PERROR("Failed to create \"%s\" socket",
634 formated_name ?: "Unknown");
635 goto error;
636 }
637 DBG("Listening on %s socket %d", name, sock->fd);
638
639 ret = sock->ops->bind(sock);
640 if (ret < 0) {
641 PERROR("Failed to bind lttng-live socket");
642 goto error;
643 }
644
645 ret = sock->ops->listen(sock, -1);
646 if (ret < 0) {
647 goto error;
648
649 }
650
651 free(formated_name);
652 return sock;
653
654 error:
655 if (sock) {
656 lttcomm_destroy_sock(sock);
657 }
658 free(formated_name);
659 return NULL;
660 }
661
662 /*
663 * This thread manages the listening for new connections on the network
664 */
665 static
666 void *thread_listener(void *data)
667 {
668 int i, ret, pollfd, err = -1;
669 uint32_t revents, nb_fd;
670 struct lttng_poll_event events;
671 struct lttcomm_sock *live_control_sock;
672
673 DBG("[thread] Relay live listener started");
674
675 rcu_register_thread();
676 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
677
678 health_code_update();
679
680 live_control_sock = init_socket(live_uri, "Live listener");
681 if (!live_control_sock) {
682 goto error_sock_control;
683 }
684
685 /* Pass 2 as size here for the thread quit pipe and control sockets. */
686 ret = create_named_thread_poll_set(&events, 2,
687 "Live listener thread epoll");
688 if (ret < 0) {
689 goto error_create_poll;
690 }
691
692 /* Add the control socket */
693 ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
694 if (ret < 0) {
695 goto error_poll_add;
696 }
697
698 lttng_relay_notify_ready();
699
700 if (testpoint(relayd_thread_live_listener)) {
701 goto error_testpoint;
702 }
703
704 while (1) {
705 health_code_update();
706
707 DBG("Listener accepting live viewers connections");
708
709 restart:
710 health_poll_entry();
711 ret = lttng_poll_wait(&events, -1);
712 health_poll_exit();
713 if (ret < 0) {
714 /*
715 * Restart interrupted system call.
716 */
717 if (errno == EINTR) {
718 goto restart;
719 }
720 goto error;
721 }
722 nb_fd = ret;
723
724 DBG("Relay new viewer connection received");
725 for (i = 0; i < nb_fd; i++) {
726 health_code_update();
727
728 /* Fetch once the poll data */
729 revents = LTTNG_POLL_GETEV(&events, i);
730 pollfd = LTTNG_POLL_GETFD(&events, i);
731
732 /* Thread quit pipe has been closed. Killing thread. */
733 ret = check_thread_quit_pipe(pollfd, revents);
734 if (ret) {
735 err = 0;
736 goto exit;
737 }
738
739 if (revents & LPOLLIN) {
740 /*
741 * A new connection is requested, therefore a
742 * viewer connection is allocated in this
743 * thread, enqueued to a global queue and
744 * dequeued (and freed) in the worker thread.
745 */
746 int val = 1;
747 struct relay_connection *new_conn;
748 struct lttcomm_sock *newsock;
749
750 newsock = accept_live_sock(live_control_sock,
751 "Live socket to client");
752 if (!newsock) {
753 PERROR("accepting control sock");
754 goto error;
755 }
756 DBG("Relay viewer connection accepted socket %d", newsock->fd);
757
758 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
759 sizeof(val));
760 if (ret < 0) {
761 PERROR("setsockopt inet");
762 lttcomm_destroy_sock(newsock);
763 goto error;
764 }
765 new_conn = connection_create(newsock, RELAY_CONNECTION_UNKNOWN);
766 if (!new_conn) {
767 lttcomm_destroy_sock(newsock);
768 goto error;
769 }
770 /* Ownership assumed by the connection. */
771 newsock = NULL;
772
773 /* Enqueue request for the dispatcher thread. */
774 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
775 &new_conn->qnode);
776
777 /*
778 * Wake the dispatch queue futex.
779 * Implicit memory barrier with the
780 * exchange in cds_wfcq_enqueue.
781 */
782 futex_nto1_wake(&viewer_conn_queue.futex);
783 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
784 ERR("socket poll error");
785 goto error;
786 } else {
787 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
788 goto error;
789 }
790 }
791 }
792
793 exit:
794 error:
795 error_poll_add:
796 error_testpoint:
797 (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
798 error_create_poll:
799 if (live_control_sock->fd >= 0) {
800 int sock_fd = live_control_sock->fd;
801
802 ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker,
803 &sock_fd, 1, close_sock,
804 live_control_sock);
805 if (ret) {
806 PERROR("close");
807 }
808 live_control_sock->fd = -1;
809 }
810 lttcomm_destroy_sock(live_control_sock);
811 error_sock_control:
812 if (err) {
813 health_error();
814 DBG("Live viewer listener thread exited with error");
815 }
816 health_unregister(health_relayd);
817 rcu_unregister_thread();
818 DBG("Live viewer listener thread cleanup complete");
819 if (lttng_relay_stop_threads()) {
820 ERR("Error stopping threads");
821 }
822 return NULL;
823 }
824
825 /*
826 * This thread manages the dispatching of the requests to worker threads
827 */
828 static
829 void *thread_dispatcher(void *data)
830 {
831 int err = -1;
832 ssize_t ret;
833 struct cds_wfcq_node *node;
834 struct relay_connection *conn = NULL;
835
836 DBG("[thread] Live viewer relay dispatcher started");
837
838 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
839
840 if (testpoint(relayd_thread_live_dispatcher)) {
841 goto error_testpoint;
842 }
843
844 health_code_update();
845
846 for (;;) {
847 health_code_update();
848
849 /* Atomically prepare the queue futex */
850 futex_nto1_prepare(&viewer_conn_queue.futex);
851
852 if (CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
853 break;
854 }
855
856 do {
857 health_code_update();
858
859 /* Dequeue commands */
860 node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
861 &viewer_conn_queue.tail);
862 if (node == NULL) {
863 DBG("Woken up but nothing in the live-viewer "
864 "relay command queue");
865 /* Continue thread execution */
866 break;
867 }
868 conn = caa_container_of(node, struct relay_connection, qnode);
869 DBG("Dispatching viewer request waiting on sock %d",
870 conn->sock->fd);
871
872 /*
873 * Inform worker thread of the new request. This
874 * call is blocking so we can be assured that
875 * the data will be read at some point in time
876 * or wait to the end of the world :)
877 */
878 ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
879 if (ret < 0) {
880 PERROR("write conn pipe");
881 connection_put(conn);
882 goto error;
883 }
884 } while (node != NULL);
885
886 /* Futex wait on queue. Blocking call on futex() */
887 health_poll_entry();
888 futex_nto1_wait(&viewer_conn_queue.futex);
889 health_poll_exit();
890 }
891
892 /* Normal exit, no error */
893 err = 0;
894
895 error:
896 error_testpoint:
897 if (err) {
898 health_error();
899 ERR("Health error occurred in %s", __func__);
900 }
901 health_unregister(health_relayd);
902 DBG("Live viewer dispatch thread dying");
903 if (lttng_relay_stop_threads()) {
904 ERR("Error stopping threads");
905 }
906 return NULL;
907 }
908
909 /*
910 * Establish connection with the viewer and check the versions.
911 *
912 * Return 0 on success or else negative value.
913 */
914 static
915 int viewer_connect(struct relay_connection *conn)
916 {
917 int ret;
918 struct lttng_viewer_connect reply, msg;
919
920 conn->version_check_done = 1;
921
922 health_code_update();
923
924 DBG("Viewer is establishing a connection to the relayd.");
925
926 ret = recv_request(conn->sock, &msg, sizeof(msg));
927 if (ret < 0) {
928 goto end;
929 }
930
931 health_code_update();
932
933 memset(&reply, 0, sizeof(reply));
934 reply.major = RELAYD_VERSION_COMM_MAJOR;
935 reply.minor = RELAYD_VERSION_COMM_MINOR;
936
937 /* Major versions must be the same */
938 if (reply.major != be32toh(msg.major)) {
939 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
940 reply.major, be32toh(msg.major));
941 ret = -1;
942 goto end;
943 }
944
945 conn->major = reply.major;
946 /* We adapt to the lowest compatible version */
947 if (reply.minor <= be32toh(msg.minor)) {
948 conn->minor = reply.minor;
949 } else {
950 conn->minor = be32toh(msg.minor);
951 }
952
953 if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
954 conn->type = RELAY_VIEWER_COMMAND;
955 } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
956 conn->type = RELAY_VIEWER_NOTIFICATION;
957 } else {
958 ERR("Unknown connection type : %u", be32toh(msg.type));
959 ret = -1;
960 goto end;
961 }
962
963 reply.major = htobe32(reply.major);
964 reply.minor = htobe32(reply.minor);
965 if (conn->type == RELAY_VIEWER_COMMAND) {
966 /*
967 * Increment outside of htobe64 macro, because the argument can
968 * be used more than once within the macro, and thus the
969 * operation may be undefined.
970 */
971 pthread_mutex_lock(&last_relay_viewer_session_id_lock);
972 last_relay_viewer_session_id++;
973 pthread_mutex_unlock(&last_relay_viewer_session_id_lock);
974 reply.viewer_session_id = htobe64(last_relay_viewer_session_id);
975 }
976
977 health_code_update();
978
979 ret = send_response(conn->sock, &reply, sizeof(reply));
980 if (ret < 0) {
981 goto end;
982 }
983
984 health_code_update();
985
986 DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
987 ret = 0;
988
989 end:
990 return ret;
991 }
992
993 /*
994 * Send the viewer the list of current sessions.
995 * We need to create a copy of the hash table content because otherwise
996 * we cannot assume the number of entries stays the same between getting
997 * the number of HT elements and iteration over the HT.
998 *
999 * Return 0 on success or else a negative value.
1000 */
1001 static
1002 int viewer_list_sessions(struct relay_connection *conn)
1003 {
1004 int ret = 0;
1005 struct lttng_viewer_list_sessions session_list;
1006 struct lttng_ht_iter iter;
1007 struct relay_session *session;
1008 struct lttng_viewer_session *send_session_buf = NULL;
1009 uint32_t buf_count = SESSION_BUF_DEFAULT_COUNT;
1010 uint32_t count = 0;
1011
1012 DBG("List sessions received");
1013
1014 send_session_buf = zmalloc(SESSION_BUF_DEFAULT_COUNT * sizeof(*send_session_buf));
1015 if (!send_session_buf) {
1016 return -1;
1017 }
1018
1019 rcu_read_lock();
1020 cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, session,
1021 session_n.node) {
1022 struct lttng_viewer_session *send_session;
1023
1024 health_code_update();
1025
1026 pthread_mutex_lock(&session->lock);
1027 if (session->connection_closed) {
1028 /* Skip closed session */
1029 goto next_session;
1030 }
1031
1032 if (count >= buf_count) {
1033 struct lttng_viewer_session *newbuf;
1034 uint32_t new_buf_count = buf_count << 1;
1035
1036 newbuf = realloc(send_session_buf,
1037 new_buf_count * sizeof(*send_session_buf));
1038 if (!newbuf) {
1039 ret = -1;
1040 goto break_loop;
1041 }
1042 send_session_buf = newbuf;
1043 buf_count = new_buf_count;
1044 }
1045 send_session = &send_session_buf[count];
1046 if (lttng_strncpy(send_session->session_name,
1047 session->session_name,
1048 sizeof(send_session->session_name))) {
1049 ret = -1;
1050 goto break_loop;
1051 }
1052 if (lttng_strncpy(send_session->hostname, session->hostname,
1053 sizeof(send_session->hostname))) {
1054 ret = -1;
1055 goto break_loop;
1056 }
1057 send_session->id = htobe64(session->id);
1058 send_session->live_timer = htobe32(session->live_timer);
1059 if (session->viewer_attached) {
1060 send_session->clients = htobe32(1);
1061 } else {
1062 send_session->clients = htobe32(0);
1063 }
1064 send_session->streams = htobe32(session->stream_count);
1065 count++;
1066 next_session:
1067 pthread_mutex_unlock(&session->lock);
1068 continue;
1069 break_loop:
1070 pthread_mutex_unlock(&session->lock);
1071 break;
1072 }
1073 rcu_read_unlock();
1074 if (ret < 0) {
1075 goto end_free;
1076 }
1077
1078 session_list.sessions_count = htobe32(count);
1079
1080 health_code_update();
1081
1082 ret = send_response(conn->sock, &session_list, sizeof(session_list));
1083 if (ret < 0) {
1084 goto end_free;
1085 }
1086
1087 health_code_update();
1088
1089 ret = send_response(conn->sock, send_session_buf,
1090 count * sizeof(*send_session_buf));
1091 if (ret < 0) {
1092 goto end_free;
1093 }
1094 health_code_update();
1095
1096 ret = 0;
1097 end_free:
1098 free(send_session_buf);
1099 return ret;
1100 }
1101
1102 /*
1103 * Send the viewer the list of current streams.
1104 */
1105 static
1106 int viewer_get_new_streams(struct relay_connection *conn)
1107 {
1108 int ret, send_streams = 0;
1109 uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0, nb_total = 0;
1110 struct lttng_viewer_new_streams_request request;
1111 struct lttng_viewer_new_streams_response response;
1112 struct relay_session *session = NULL;
1113 uint64_t session_id;
1114 bool closed = false;
1115
1116 assert(conn);
1117
1118 DBG("Get new streams received");
1119
1120 health_code_update();
1121
1122 /* Receive the request from the connected client. */
1123 ret = recv_request(conn->sock, &request, sizeof(request));
1124 if (ret < 0) {
1125 goto error;
1126 }
1127 session_id = be64toh(request.session_id);
1128
1129 health_code_update();
1130
1131 memset(&response, 0, sizeof(response));
1132
1133 session = session_get_by_id(session_id);
1134 if (!session) {
1135 DBG("Relay session %" PRIu64 " not found", session_id);
1136 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
1137 goto send_reply;
1138 }
1139
1140 if (!viewer_session_is_attached(conn->viewer_session, session)) {
1141 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
1142 goto send_reply;
1143 }
1144
1145 pthread_mutex_lock(&session->lock);
1146 ret = make_viewer_streams(session,
1147 conn->viewer_session,
1148 LTTNG_VIEWER_SEEK_LAST, &nb_total, &nb_unsent,
1149 &nb_created, &closed);
1150 if (ret < 0) {
1151 goto error_unlock_session;
1152 }
1153 send_streams = 1;
1154 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
1155
1156 /* Only send back the newly created streams with the unsent ones. */
1157 nb_streams = nb_created + nb_unsent;
1158 response.streams_count = htobe32(nb_streams);
1159
1160 /*
1161 * If the session is closed, HUP when there are no more streams
1162 * with data.
1163 */
1164 if (closed && nb_total == 0) {
1165 send_streams = 0;
1166 response.streams_count = 0;
1167 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
1168 goto send_reply_unlock;
1169 }
1170 send_reply_unlock:
1171 pthread_mutex_unlock(&session->lock);
1172
1173 send_reply:
1174 health_code_update();
1175 ret = send_response(conn->sock, &response, sizeof(response));
1176 if (ret < 0) {
1177 goto end_put_session;
1178 }
1179 health_code_update();
1180
1181 /*
1182 * Unknown or empty session, just return gracefully, the viewer
1183 * knows what is happening.
1184 */
1185 if (!send_streams || !nb_streams) {
1186 ret = 0;
1187 goto end_put_session;
1188 }
1189
1190 /*
1191 * Send stream and *DON'T* ignore the sent flag so every viewer
1192 * streams that were not sent from that point will be sent to
1193 * the viewer.
1194 */
1195 ret = send_viewer_streams(conn->sock, session_id, 0);
1196 if (ret < 0) {
1197 goto end_put_session;
1198 }
1199
1200 end_put_session:
1201 if (session) {
1202 session_put(session);
1203 }
1204 error:
1205 return ret;
1206 error_unlock_session:
1207 pthread_mutex_unlock(&session->lock);
1208 session_put(session);
1209 return ret;
1210 }
1211
1212 /*
1213 * Send the viewer the list of current sessions.
1214 */
1215 static
1216 int viewer_attach_session(struct relay_connection *conn)
1217 {
1218 int send_streams = 0;
1219 ssize_t ret;
1220 uint32_t nb_streams = 0;
1221 enum lttng_viewer_seek seek_type;
1222 struct lttng_viewer_attach_session_request request;
1223 struct lttng_viewer_attach_session_response response;
1224 struct relay_session *session = NULL;
1225 enum lttng_viewer_attach_return_code viewer_attach_status;
1226 bool closed = false;
1227 uint64_t session_id;
1228
1229 assert(conn);
1230
1231 health_code_update();
1232
1233 /* Receive the request from the connected client. */
1234 ret = recv_request(conn->sock, &request, sizeof(request));
1235 if (ret < 0) {
1236 goto error;
1237 }
1238
1239 session_id = be64toh(request.session_id);
1240 health_code_update();
1241
1242 memset(&response, 0, sizeof(response));
1243
1244 if (!conn->viewer_session) {
1245 DBG("Client trying to attach before creating a live viewer session");
1246 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1247 goto send_reply;
1248 }
1249
1250 session = session_get_by_id(session_id);
1251 if (!session) {
1252 DBG("Relay session %" PRIu64 " not found", session_id);
1253 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1254 goto send_reply;
1255 }
1256 DBG("Attach session ID %" PRIu64 " received", session_id);
1257
1258 pthread_mutex_lock(&session->lock);
1259 if (session->live_timer == 0) {
1260 DBG("Not live session");
1261 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1262 goto send_reply;
1263 }
1264
1265 send_streams = 1;
1266 viewer_attach_status = viewer_session_attach(conn->viewer_session,
1267 session);
1268 if (viewer_attach_status != LTTNG_VIEWER_ATTACH_OK) {
1269 response.status = htobe32(viewer_attach_status);
1270 goto send_reply;
1271 }
1272
1273 switch (be32toh(request.seek)) {
1274 case LTTNG_VIEWER_SEEK_BEGINNING:
1275 case LTTNG_VIEWER_SEEK_LAST:
1276 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1277 seek_type = be32toh(request.seek);
1278 break;
1279 default:
1280 ERR("Wrong seek parameter");
1281 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1282 send_streams = 0;
1283 goto send_reply;
1284 }
1285
1286 ret = make_viewer_streams(session,
1287 conn->viewer_session, seek_type,
1288 &nb_streams, NULL, NULL, &closed);
1289 if (ret < 0) {
1290 goto end_put_session;
1291 }
1292 pthread_mutex_unlock(&session->lock);
1293 session_put(session);
1294 session = NULL;
1295
1296 response.streams_count = htobe32(nb_streams);
1297 /*
1298 * If the session is closed when the viewer is attaching, it
1299 * means some of the streams may have been concurrently removed,
1300 * so we don't allow the viewer to attach, even if there are
1301 * streams available.
1302 */
1303 if (closed) {
1304 send_streams = 0;
1305 response.streams_count = 0;
1306 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1307 goto send_reply;
1308 }
1309
1310 send_reply:
1311 health_code_update();
1312 ret = send_response(conn->sock, &response, sizeof(response));
1313 if (ret < 0) {
1314 goto end_put_session;
1315 }
1316 health_code_update();
1317
1318 /*
1319 * Unknown or empty session, just return gracefully, the viewer
1320 * knows what is happening.
1321 */
1322 if (!send_streams || !nb_streams) {
1323 ret = 0;
1324 goto end_put_session;
1325 }
1326
1327 /* Send stream and ignore the sent flag. */
1328 ret = send_viewer_streams(conn->sock, session_id, 1);
1329 if (ret < 0) {
1330 goto end_put_session;
1331 }
1332
1333 end_put_session:
1334 if (session) {
1335 pthread_mutex_unlock(&session->lock);
1336 session_put(session);
1337 }
1338 error:
1339 return ret;
1340 }
1341
1342 /*
1343 * Open the index file if needed for the given vstream.
1344 *
1345 * If an index file is successfully opened, the vstream will set it as its
1346 * current index file.
1347 *
1348 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1349 *
1350 * Called with rstream lock held.
1351 */
1352 static int try_open_index(struct relay_viewer_stream *vstream,
1353 struct relay_stream *rstream)
1354 {
1355 int ret = 0;
1356 const uint32_t connection_major = rstream->trace->session->major;
1357 const uint32_t connection_minor = rstream->trace->session->minor;
1358 enum lttng_trace_chunk_status chunk_status;
1359
1360 if (vstream->index_file) {
1361 goto end;
1362 }
1363
1364 /*
1365 * First time, we open the index file and at least one index is ready.
1366 */
1367 if (rstream->index_received_seqcount == 0 ||
1368 !vstream->stream_file.trace_chunk) {
1369 ret = -ENOENT;
1370 goto end;
1371 }
1372
1373 chunk_status = lttng_index_file_create_from_trace_chunk_read_only(
1374 vstream->stream_file.trace_chunk, rstream->path_name,
1375 rstream->channel_name, rstream->tracefile_size,
1376 vstream->current_tracefile_id,
1377 lttng_to_index_major(connection_major, connection_minor),
1378 lttng_to_index_minor(connection_major, connection_minor),
1379 true, &vstream->index_file);
1380 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1381 if (chunk_status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE) {
1382 ret = -ENOENT;
1383 } else {
1384 ret = -1;
1385 }
1386 }
1387
1388 end:
1389 return ret;
1390 }
1391
1392 /*
1393 * Check the status of the index for the given stream. This function
1394 * updates the index structure if needed and can put (close) the vstream
1395 * in the HUP situation.
1396 *
1397 * Return 0 means that we can proceed with the index. A value of 1 means
1398 * that the index has been updated and is ready to be sent to the
1399 * client. A negative value indicates an error that can't be handled.
1400 *
1401 * Called with rstream lock held.
1402 */
1403 static int check_index_status(struct relay_viewer_stream *vstream,
1404 struct relay_stream *rstream, struct ctf_trace *trace,
1405 struct lttng_viewer_index *index)
1406 {
1407 int ret;
1408
1409 DBG("Check index status: index_received_seqcount %" PRIu64 " "
1410 "index_sent_seqcount %" PRIu64 " "
1411 "for stream %" PRIu64,
1412 rstream->index_received_seqcount,
1413 vstream->index_sent_seqcount,
1414 vstream->stream->stream_handle);
1415 if ((trace->session->connection_closed || rstream->closed)
1416 && rstream->index_received_seqcount
1417 == vstream->index_sent_seqcount) {
1418 /*
1419 * Last index sent and session connection or relay
1420 * stream are closed.
1421 */
1422 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1423 goto hup;
1424 } else if (rstream->beacon_ts_end != -1ULL &&
1425 (rstream->index_received_seqcount == 0 ||
1426 (vstream->index_sent_seqcount != 0 &&
1427 rstream->index_received_seqcount
1428 <= vstream->index_sent_seqcount))) {
1429 /*
1430 * We've received a synchronization beacon and the last index
1431 * available has been sent, the index for now is inactive.
1432 *
1433 * In this case, we have received a beacon which allows us to
1434 * inform the client of a time interval during which we can
1435 * guarantee that there are no events to read (and never will
1436 * be).
1437 *
1438 * The sent seqcount can grow higher than receive seqcount on
1439 * clear because the rotation performed by clear will push
1440 * the index_sent_seqcount ahead (see
1441 * viewer_stream_sync_tracefile_array_tail) and skip over
1442 * packet sequence numbers.
1443 */
1444 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1445 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1446 index->stream_id = htobe64(rstream->ctf_stream_id);
1447 DBG("Check index status: inactive with beacon, for stream %" PRIu64,
1448 vstream->stream->stream_handle);
1449 goto index_ready;
1450 } else if (rstream->index_received_seqcount == 0 ||
1451 (vstream->index_sent_seqcount != 0 &&
1452 rstream->index_received_seqcount
1453 <= vstream->index_sent_seqcount)) {
1454 /*
1455 * This checks whether received <= sent seqcount. In
1456 * this case, we have not received a beacon. Therefore,
1457 * we can only ask the client to retry later.
1458 *
1459 * The sent seqcount can grow higher than receive seqcount on
1460 * clear because the rotation performed by clear will push
1461 * the index_sent_seqcount ahead (see
1462 * viewer_stream_sync_tracefile_array_tail) and skip over
1463 * packet sequence numbers.
1464 */
1465 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1466 DBG("Check index status: retry for stream %" PRIu64,
1467 vstream->stream->stream_handle);
1468 goto index_ready;
1469 } else if (!tracefile_array_seq_in_file(rstream->tfa,
1470 vstream->current_tracefile_id,
1471 vstream->index_sent_seqcount)) {
1472 /*
1473 * The next index we want to send cannot be read either
1474 * because we need to perform a rotation, or due to
1475 * the producer having overwritten its trace file.
1476 */
1477 DBG("Viewer stream %" PRIu64 " rotation",
1478 vstream->stream->stream_handle);
1479 ret = viewer_stream_rotate(vstream);
1480 if (ret == 1) {
1481 /* EOF across entire stream. */
1482 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1483 goto hup;
1484 }
1485 /*
1486 * If we have been pushed due to overwrite, it
1487 * necessarily means there is data that can be read in
1488 * the stream. If we rotated because we reached the end
1489 * of a tracefile, it means the following tracefile
1490 * needs to contain at least one index, else we would
1491 * have already returned LTTNG_VIEWER_INDEX_RETRY to the
1492 * viewer. The updated index_sent_seqcount needs to
1493 * point to a readable index entry now.
1494 *
1495 * In the case where we "rotate" on a single file, we
1496 * can end up in a case where the requested index is
1497 * still unavailable.
1498 */
1499 if (rstream->tracefile_count == 1 &&
1500 !tracefile_array_seq_in_file(
1501 rstream->tfa,
1502 vstream->current_tracefile_id,
1503 vstream->index_sent_seqcount)) {
1504 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1505 DBG("Check index status: retry: "
1506 "tracefile array sequence number %" PRIu64
1507 " not in file for stream %" PRIu64,
1508 vstream->index_sent_seqcount,
1509 vstream->stream->stream_handle);
1510 goto index_ready;
1511 }
1512 assert(tracefile_array_seq_in_file(rstream->tfa,
1513 vstream->current_tracefile_id,
1514 vstream->index_sent_seqcount));
1515 }
1516 /* ret == 0 means successful so we continue. */
1517 ret = 0;
1518 return ret;
1519
1520 hup:
1521 viewer_stream_put(vstream);
1522 index_ready:
1523 return 1;
1524 }
1525
1526 static
1527 void viewer_stream_rotate_to_trace_chunk(struct relay_viewer_stream *vstream,
1528 struct lttng_trace_chunk *new_trace_chunk)
1529 {
1530 lttng_trace_chunk_put(vstream->stream_file.trace_chunk);
1531
1532 if (new_trace_chunk) {
1533 const bool acquired_reference = lttng_trace_chunk_get(
1534 new_trace_chunk);
1535
1536 assert(acquired_reference);
1537 }
1538
1539 vstream->stream_file.trace_chunk = new_trace_chunk;
1540 viewer_stream_sync_tracefile_array_tail(vstream);
1541 viewer_stream_close_files(vstream);
1542 }
1543
1544 /*
1545 * Send the next index for a stream.
1546 *
1547 * Return 0 on success or else a negative value.
1548 */
1549 static
1550 int viewer_get_next_index(struct relay_connection *conn)
1551 {
1552 int ret;
1553 struct lttng_viewer_get_next_index request_index;
1554 struct lttng_viewer_index viewer_index;
1555 struct ctf_packet_index packet_index;
1556 struct relay_viewer_stream *vstream = NULL;
1557 struct relay_stream *rstream = NULL;
1558 struct ctf_trace *ctf_trace = NULL;
1559 struct relay_viewer_stream *metadata_viewer_stream = NULL;
1560
1561 assert(conn);
1562
1563 DBG("Viewer get next index");
1564
1565 memset(&viewer_index, 0, sizeof(viewer_index));
1566 health_code_update();
1567
1568 ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1569 if (ret < 0) {
1570 goto end;
1571 }
1572 health_code_update();
1573
1574 vstream = viewer_stream_get_by_id(be64toh(request_index.stream_id));
1575 if (!vstream) {
1576 DBG("Client requested index of unknown stream id %" PRIu64,
1577 (uint64_t) be64toh(request_index.stream_id));
1578 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1579 goto send_reply;
1580 }
1581
1582 /* Use back. ref. Protected by refcounts. */
1583 rstream = vstream->stream;
1584 ctf_trace = rstream->trace;
1585
1586 /* metadata_viewer_stream may be NULL. */
1587 metadata_viewer_stream =
1588 ctf_trace_get_viewer_metadata_stream(ctf_trace);
1589
1590 pthread_mutex_lock(&rstream->lock);
1591
1592 /*
1593 * The viewer should not ask for index on metadata stream.
1594 */
1595 if (rstream->is_metadata) {
1596 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1597 goto send_reply;
1598 }
1599
1600 if (rstream->ongoing_rotation.is_set) {
1601 /* Rotation is ongoing, try again later. */
1602 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1603 goto send_reply;
1604 }
1605
1606 if (rstream->trace->session->ongoing_rotation) {
1607 /* Rotation is ongoing, try again later. */
1608 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1609 goto send_reply;
1610 }
1611
1612 /*
1613 * Transition the viewer session into the newest trace chunk available.
1614 */
1615 if (!lttng_trace_chunk_ids_equal(
1616 conn->viewer_session->current_trace_chunk,
1617 rstream->trace_chunk)) {
1618 DBG("Relay stream and viewer chunk ids differ");
1619
1620 ret = viewer_session_set_trace_chunk_copy(
1621 conn->viewer_session,
1622 rstream->trace_chunk);
1623 if (ret) {
1624 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1625 goto send_reply;
1626 }
1627 }
1628
1629 /*
1630 * Transition the viewer stream into the latest trace chunk available.
1631 *
1632 * Note that the stream must _not_ rotate in one precise condition:
1633 * the relay stream has rotated to a NULL trace chunk and the viewer
1634 * stream is consuming the trace chunk that was active just before
1635 * that rotation to NULL.
1636 *
1637 * This allows clients to consume all the packets of a trace chunk
1638 * after a session's destruction.
1639 */
1640 if (conn->viewer_session->current_trace_chunk != vstream->stream_file.trace_chunk &&
1641 !(rstream->completed_rotation_count == vstream->last_seen_rotation_count + 1 && !rstream->trace_chunk)) {
1642 DBG("Viewer session and viewer stream chunk differ: "
1643 "vsession chunk %p vstream chunk %p",
1644 conn->viewer_session->current_trace_chunk,
1645 vstream->stream_file.trace_chunk);
1646 viewer_stream_rotate_to_trace_chunk(vstream,
1647 conn->viewer_session->current_trace_chunk);
1648 vstream->last_seen_rotation_count =
1649 rstream->completed_rotation_count;
1650 }
1651
1652 ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1653 if (ret < 0) {
1654 goto error_put;
1655 } else if (ret == 1) {
1656 /*
1657 * We have no index to send and check_index_status has populated
1658 * viewer_index's status.
1659 */
1660 goto send_reply;
1661 }
1662 /* At this point, ret is 0 thus we will be able to read the index. */
1663 assert(!ret);
1664
1665 /* Try to open an index if one is needed for that stream. */
1666 ret = try_open_index(vstream, rstream);
1667 if (ret == -ENOENT) {
1668 if (rstream->closed) {
1669 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1670 goto send_reply;
1671 } else {
1672 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1673 goto send_reply;
1674 }
1675 }
1676 if (ret < 0) {
1677 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1678 goto send_reply;
1679 }
1680
1681 /*
1682 * vstream->stream_fd may be NULL if it has been closed by
1683 * tracefile rotation, or if we are at the beginning of the
1684 * stream. We open the data stream file here to protect against
1685 * overwrite caused by tracefile rotation (in association with
1686 * unlink performed before overwrite).
1687 */
1688 if (!vstream->stream_file.handle) {
1689 char file_path[LTTNG_PATH_MAX];
1690 enum lttng_trace_chunk_status status;
1691 struct fs_handle *fs_handle;
1692
1693 ret = utils_stream_file_path(rstream->path_name,
1694 rstream->channel_name, rstream->tracefile_size,
1695 vstream->current_tracefile_id, NULL, file_path,
1696 sizeof(file_path));
1697 if (ret < 0) {
1698 goto error_put;
1699 }
1700
1701 /*
1702 * It is possible the the file we are trying to open is
1703 * missing if the stream has been closed (application exits with
1704 * per-pid buffers) and a clear command has been performed.
1705 */
1706 status = lttng_trace_chunk_open_fs_handle(
1707 vstream->stream_file.trace_chunk,
1708 file_path, O_RDONLY, 0, &fs_handle, true);
1709 if (status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1710 if (status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE &&
1711 rstream->closed) {
1712 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1713 goto send_reply;
1714 }
1715 PERROR("Failed to open trace file for viewer stream");
1716 goto error_put;
1717 }
1718 vstream->stream_file.handle = fs_handle;
1719 }
1720
1721 ret = check_new_streams(conn);
1722 if (ret < 0) {
1723 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1724 goto send_reply;
1725 } else if (ret == 1) {
1726 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1727 }
1728
1729 ret = lttng_index_file_read(vstream->index_file, &packet_index);
1730 if (ret) {
1731 ERR("Relay error reading index file");
1732 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1733 goto send_reply;
1734 } else {
1735 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1736 vstream->index_sent_seqcount++;
1737 }
1738
1739 /*
1740 * Indexes are stored in big endian, no need to switch before sending.
1741 */
1742 DBG("Sending viewer index for stream %" PRIu64 " offset %" PRIu64,
1743 rstream->stream_handle,
1744 (uint64_t) be64toh(packet_index.offset));
1745 viewer_index.offset = packet_index.offset;
1746 viewer_index.packet_size = packet_index.packet_size;
1747 viewer_index.content_size = packet_index.content_size;
1748 viewer_index.timestamp_begin = packet_index.timestamp_begin;
1749 viewer_index.timestamp_end = packet_index.timestamp_end;
1750 viewer_index.events_discarded = packet_index.events_discarded;
1751 viewer_index.stream_id = packet_index.stream_id;
1752
1753 send_reply:
1754 if (rstream) {
1755 pthread_mutex_unlock(&rstream->lock);
1756 }
1757
1758 if (metadata_viewer_stream) {
1759 pthread_mutex_lock(&metadata_viewer_stream->stream->lock);
1760 DBG("get next index metadata check: recv %" PRIu64
1761 " sent %" PRIu64,
1762 metadata_viewer_stream->stream->metadata_received,
1763 metadata_viewer_stream->metadata_sent);
1764 if (!metadata_viewer_stream->stream->metadata_received ||
1765 metadata_viewer_stream->stream->metadata_received >
1766 metadata_viewer_stream->metadata_sent) {
1767 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1768 }
1769 pthread_mutex_unlock(&metadata_viewer_stream->stream->lock);
1770 }
1771
1772 viewer_index.flags = htobe32(viewer_index.flags);
1773 health_code_update();
1774
1775 ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1776 if (ret < 0) {
1777 goto end;
1778 }
1779 health_code_update();
1780
1781 if (vstream) {
1782 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1783 vstream->index_sent_seqcount,
1784 vstream->stream->stream_handle);
1785 }
1786 end:
1787 if (metadata_viewer_stream) {
1788 viewer_stream_put(metadata_viewer_stream);
1789 }
1790 if (vstream) {
1791 viewer_stream_put(vstream);
1792 }
1793 return ret;
1794
1795 error_put:
1796 pthread_mutex_unlock(&rstream->lock);
1797 if (metadata_viewer_stream) {
1798 viewer_stream_put(metadata_viewer_stream);
1799 }
1800 viewer_stream_put(vstream);
1801 return ret;
1802 }
1803
1804 /*
1805 * Send the next index for a stream
1806 *
1807 * Return 0 on success or else a negative value.
1808 */
1809 static
1810 int viewer_get_packet(struct relay_connection *conn)
1811 {
1812 int ret;
1813 off_t lseek_ret;
1814 char *reply = NULL;
1815 struct lttng_viewer_get_packet get_packet_info;
1816 struct lttng_viewer_trace_packet reply_header;
1817 struct relay_viewer_stream *vstream = NULL;
1818 uint32_t reply_size = sizeof(reply_header);
1819 uint32_t packet_data_len = 0;
1820 ssize_t read_len;
1821 uint64_t stream_id;
1822
1823 DBG2("Relay get data packet");
1824
1825 health_code_update();
1826
1827 ret = recv_request(conn->sock, &get_packet_info,
1828 sizeof(get_packet_info));
1829 if (ret < 0) {
1830 goto end;
1831 }
1832 health_code_update();
1833
1834 /* From this point on, the error label can be reached. */
1835 memset(&reply_header, 0, sizeof(reply_header));
1836 stream_id = (uint64_t) be64toh(get_packet_info.stream_id);
1837
1838 vstream = viewer_stream_get_by_id(stream_id);
1839 if (!vstream) {
1840 DBG("Client requested packet of unknown stream id %" PRIu64,
1841 stream_id);
1842 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1843 goto send_reply_nolock;
1844 } else {
1845 packet_data_len = be32toh(get_packet_info.len);
1846 reply_size += packet_data_len;
1847 }
1848
1849 reply = zmalloc(reply_size);
1850 if (!reply) {
1851 PERROR("packet reply zmalloc");
1852 reply_size = sizeof(reply_header);
1853 goto error;
1854 }
1855
1856 pthread_mutex_lock(&vstream->stream->lock);
1857 lseek_ret = fs_handle_seek(vstream->stream_file.handle,
1858 be64toh(get_packet_info.offset), SEEK_SET);
1859 if (lseek_ret < 0) {
1860 PERROR("Failed to seek file system handle of viewer stream %" PRIu64
1861 " to offset %" PRIu64,
1862 stream_id,
1863 (uint64_t) be64toh(get_packet_info.offset));
1864 goto error;
1865 }
1866 read_len = fs_handle_read(vstream->stream_file.handle,
1867 reply + sizeof(reply_header), packet_data_len);
1868 if (read_len < packet_data_len) {
1869 PERROR("Failed to read from file system handle of viewer stream id %" PRIu64
1870 ", offset: %" PRIu64,
1871 stream_id,
1872 (uint64_t) be64toh(get_packet_info.offset));
1873 goto error;
1874 }
1875 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1876 reply_header.len = htobe32(packet_data_len);
1877 goto send_reply;
1878
1879 error:
1880 /* No payload to send on error. */
1881 reply_size = sizeof(reply_header);
1882 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1883
1884 send_reply:
1885 if (vstream) {
1886 pthread_mutex_unlock(&vstream->stream->lock);
1887 }
1888 send_reply_nolock:
1889
1890 health_code_update();
1891
1892 if (reply) {
1893 memcpy(reply, &reply_header, sizeof(reply_header));
1894 ret = send_response(conn->sock, reply, reply_size);
1895 } else {
1896 /* No reply to send. */
1897 ret = send_response(conn->sock, &reply_header,
1898 reply_size);
1899 }
1900
1901 health_code_update();
1902 if (ret < 0) {
1903 PERROR("sendmsg of packet data failed");
1904 goto end_free;
1905 }
1906
1907 DBG("Sent %u bytes for stream %" PRIu64, reply_size, stream_id);
1908
1909 end_free:
1910 free(reply);
1911 end:
1912 if (vstream) {
1913 viewer_stream_put(vstream);
1914 }
1915 return ret;
1916 }
1917
1918 /*
1919 * Send the session's metadata
1920 *
1921 * Return 0 on success else a negative value.
1922 */
1923 static
1924 int viewer_get_metadata(struct relay_connection *conn)
1925 {
1926 int ret = 0;
1927 int fd = -1;
1928 ssize_t read_len;
1929 uint64_t len = 0;
1930 char *data = NULL;
1931 struct lttng_viewer_get_metadata request;
1932 struct lttng_viewer_metadata_packet reply;
1933 struct relay_viewer_stream *vstream = NULL;
1934
1935 assert(conn);
1936
1937 DBG("Relay get metadata");
1938
1939 health_code_update();
1940
1941 ret = recv_request(conn->sock, &request, sizeof(request));
1942 if (ret < 0) {
1943 goto end;
1944 }
1945 health_code_update();
1946
1947 memset(&reply, 0, sizeof(reply));
1948
1949 vstream = viewer_stream_get_by_id(be64toh(request.stream_id));
1950 if (!vstream) {
1951 /*
1952 * The metadata stream can be closed by a CLOSE command
1953 * just before we attach. It can also be closed by
1954 * per-pid tracing during tracing. Therefore, it is
1955 * possible that we cannot find this viewer stream.
1956 * Reply back to the client with an error if we cannot
1957 * find it.
1958 */
1959 DBG("Client requested metadata of unknown stream id %" PRIu64,
1960 (uint64_t) be64toh(request.stream_id));
1961 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1962 goto send_reply;
1963 }
1964 pthread_mutex_lock(&vstream->stream->lock);
1965 if (!vstream->stream->is_metadata) {
1966 ERR("Invalid metadata stream");
1967 goto error;
1968 }
1969
1970 if (vstream->metadata_sent >= vstream->stream->metadata_received) {
1971 /*
1972 * The live viewers expect to receive a NO_NEW_METADATA
1973 * status before a stream disappears, otherwise they abort the
1974 * entire live connection when receiving an error status.
1975 *
1976 * Clear feature resets the metadata_sent to 0 until the
1977 * same metadata is received again.
1978 */
1979 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1980 /*
1981 * The live viewer considers a closed 0 byte metadata stream as
1982 * an error.
1983 */
1984 if (vstream->metadata_sent > 0) {
1985 vstream->stream->no_new_metadata_notified = true;
1986 if (vstream->stream->closed) {
1987 /* Release ownership for the viewer metadata stream. */
1988 viewer_stream_put(vstream);
1989 }
1990 }
1991 goto send_reply;
1992 }
1993
1994 if (vstream->stream->trace_chunk &&
1995 !lttng_trace_chunk_ids_equal(
1996 conn->viewer_session->current_trace_chunk,
1997 vstream->stream->trace_chunk)) {
1998 /* A rotation has occurred on the relay stream. */
1999 DBG("Metadata relay stream and viewer chunk ids differ");
2000
2001 ret = viewer_session_set_trace_chunk_copy(
2002 conn->viewer_session,
2003 vstream->stream->trace_chunk);
2004 if (ret) {
2005 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
2006 goto send_reply;
2007 }
2008 }
2009
2010 if (conn->viewer_session->current_trace_chunk !=
2011 vstream->stream_file.trace_chunk) {
2012 bool acquired_reference;
2013
2014 DBG("Viewer session and viewer stream chunk differ: "
2015 "vsession chunk %p vstream chunk %p",
2016 conn->viewer_session->current_trace_chunk,
2017 vstream->stream_file.trace_chunk);
2018 lttng_trace_chunk_put(vstream->stream_file.trace_chunk);
2019 acquired_reference = lttng_trace_chunk_get(conn->viewer_session->current_trace_chunk);
2020 assert(acquired_reference);
2021 vstream->stream_file.trace_chunk =
2022 conn->viewer_session->current_trace_chunk;
2023 viewer_stream_close_files(vstream);
2024 }
2025
2026 len = vstream->stream->metadata_received - vstream->metadata_sent;
2027
2028 /*
2029 * Either this is the first time the metadata file is read, or a
2030 * rotation of the corresponding relay stream has occured.
2031 */
2032 if (!vstream->stream_file.handle && len > 0) {
2033 struct fs_handle *fs_handle;
2034 char file_path[LTTNG_PATH_MAX];
2035 enum lttng_trace_chunk_status status;
2036 struct relay_stream *rstream = vstream->stream;
2037
2038 ret = utils_stream_file_path(rstream->path_name,
2039 rstream->channel_name, rstream->tracefile_size,
2040 vstream->current_tracefile_id, NULL, file_path,
2041 sizeof(file_path));
2042 if (ret < 0) {
2043 goto error;
2044 }
2045
2046 /*
2047 * It is possible the the metadata file we are trying to open is
2048 * missing if the stream has been closed (application exits with
2049 * per-pid buffers) and a clear command has been performed.
2050 */
2051 status = lttng_trace_chunk_open_fs_handle(
2052 vstream->stream_file.trace_chunk,
2053 file_path, O_RDONLY, 0, &fs_handle, true);
2054 if (status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2055 if (status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE) {
2056 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
2057 len = 0;
2058 if (vstream->stream->closed) {
2059 viewer_stream_put(vstream);
2060 }
2061 goto send_reply;
2062 }
2063 PERROR("Failed to open metadata file for viewer stream");
2064 goto error;
2065 }
2066 vstream->stream_file.handle = fs_handle;
2067
2068 if (vstream->metadata_sent != 0) {
2069 /*
2070 * The client does not expect to receive any metadata
2071 * it has received and metadata files in successive
2072 * chunks must be a strict superset of one another.
2073 *
2074 * Skip the first `metadata_sent` bytes to ensure
2075 * they are not sent a second time to the client.
2076 *
2077 * Baring a block layer error or an internal error,
2078 * this seek should not fail as
2079 * `vstream->stream->metadata_received` is reset when
2080 * a relay stream is rotated. If this is reached, it is
2081 * safe to assume that
2082 * `metadata_received` > `metadata_sent`.
2083 */
2084 const off_t seek_ret = fs_handle_seek(fs_handle,
2085 vstream->metadata_sent, SEEK_SET);
2086
2087 if (seek_ret < 0) {
2088 PERROR("Failed to seek metadata viewer stream file to `sent` position: pos = %" PRId64,
2089 vstream->metadata_sent);
2090 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
2091 goto send_reply;
2092 }
2093 }
2094 }
2095
2096 reply.len = htobe64(len);
2097 data = zmalloc(len);
2098 if (!data) {
2099 PERROR("viewer metadata zmalloc");
2100 goto error;
2101 }
2102
2103 fd = fs_handle_get_fd(vstream->stream_file.handle);
2104 if (fd < 0) {
2105 ERR("Failed to restore viewer stream file system handle");
2106 goto error;
2107 }
2108 read_len = lttng_read(fd, data, len);
2109 fs_handle_put_fd(vstream->stream_file.handle);
2110 fd = -1;
2111 if (read_len < len) {
2112 if (read_len < 0) {
2113 PERROR("Failed to read metadata file");
2114 goto error;
2115 } else {
2116 /*
2117 * A clear has been performed which prevents the relay
2118 * from sending `len` bytes of metadata.
2119 *
2120 * It is important not to send any metadata if we
2121 * couldn't read all the available metadata in one shot:
2122 * sending partial metadata can cause the client to
2123 * attempt to parse an incomplete (incoherent) metadata
2124 * stream, which would result in an error.
2125 */
2126 const off_t seek_ret = fs_handle_seek(
2127 vstream->stream_file.handle, -read_len,
2128 SEEK_CUR);
2129
2130 DBG("Failed to read metadata: requested = %" PRIu64 ", got = %zd",
2131 len, read_len);
2132 read_len = 0;
2133 len = 0;
2134 if (seek_ret < 0) {
2135 PERROR("Failed to restore metadata file position after partial read");
2136 ret = -1;
2137 goto error;
2138 }
2139 }
2140 }
2141 vstream->metadata_sent += read_len;
2142 reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
2143
2144 goto send_reply;
2145
2146 error:
2147 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
2148
2149 send_reply:
2150 health_code_update();
2151 if (vstream) {
2152 pthread_mutex_unlock(&vstream->stream->lock);
2153 }
2154 ret = send_response(conn->sock, &reply, sizeof(reply));
2155 if (ret < 0) {
2156 goto end_free;
2157 }
2158 health_code_update();
2159
2160 if (len > 0) {
2161 ret = send_response(conn->sock, data, len);
2162 if (ret < 0) {
2163 goto end_free;
2164 }
2165 }
2166
2167 DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
2168 (uint64_t) be64toh(request.stream_id));
2169
2170 DBG("Metadata sent");
2171
2172 end_free:
2173 free(data);
2174 end:
2175 if (vstream) {
2176 viewer_stream_put(vstream);
2177 }
2178 return ret;
2179 }
2180
2181 /*
2182 * Create a viewer session.
2183 *
2184 * Return 0 on success or else a negative value.
2185 */
2186 static
2187 int viewer_create_session(struct relay_connection *conn)
2188 {
2189 int ret;
2190 struct lttng_viewer_create_session_response resp;
2191
2192 DBG("Viewer create session received");
2193
2194 memset(&resp, 0, sizeof(resp));
2195 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
2196 conn->viewer_session = viewer_session_create();
2197 if (!conn->viewer_session) {
2198 ERR("Allocation viewer session");
2199 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
2200 goto send_reply;
2201 }
2202
2203 send_reply:
2204 health_code_update();
2205 ret = send_response(conn->sock, &resp, sizeof(resp));
2206 if (ret < 0) {
2207 goto end;
2208 }
2209 health_code_update();
2210 ret = 0;
2211
2212 end:
2213 return ret;
2214 }
2215
2216 /*
2217 * Detach a viewer session.
2218 *
2219 * Return 0 on success or else a negative value.
2220 */
2221 static
2222 int viewer_detach_session(struct relay_connection *conn)
2223 {
2224 int ret;
2225 struct lttng_viewer_detach_session_response response;
2226 struct lttng_viewer_detach_session_request request;
2227 struct relay_session *session = NULL;
2228 uint64_t viewer_session_to_close;
2229
2230 DBG("Viewer detach session received");
2231
2232 assert(conn);
2233
2234 health_code_update();
2235
2236 /* Receive the request from the connected client. */
2237 ret = recv_request(conn->sock, &request, sizeof(request));
2238 if (ret < 0) {
2239 goto end;
2240 }
2241 viewer_session_to_close = be64toh(request.session_id);
2242
2243 if (!conn->viewer_session) {
2244 DBG("Client trying to detach before creating a live viewer session");
2245 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
2246 goto send_reply;
2247 }
2248
2249 health_code_update();
2250
2251 memset(&response, 0, sizeof(response));
2252 DBG("Detaching from session ID %" PRIu64, viewer_session_to_close);
2253
2254 session = session_get_by_id(be64toh(request.session_id));
2255 if (!session) {
2256 DBG("Relay session %" PRIu64 " not found",
2257 (uint64_t) be64toh(request.session_id));
2258 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_UNK);
2259 goto send_reply;
2260 }
2261
2262 ret = viewer_session_is_attached(conn->viewer_session, session);
2263 if (ret != 1) {
2264 DBG("Not attached to this session");
2265 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
2266 goto send_reply_put;
2267 }
2268
2269 viewer_session_close_one_session(conn->viewer_session, session);
2270 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_OK);
2271 DBG("Session %" PRIu64 " detached.", viewer_session_to_close);
2272
2273 send_reply_put:
2274 session_put(session);
2275
2276 send_reply:
2277 health_code_update();
2278 ret = send_response(conn->sock, &response, sizeof(response));
2279 if (ret < 0) {
2280 goto end;
2281 }
2282 health_code_update();
2283 ret = 0;
2284
2285 end:
2286 return ret;
2287 }
2288
2289 /*
2290 * live_relay_unknown_command: send -1 if received unknown command
2291 */
2292 static
2293 void live_relay_unknown_command(struct relay_connection *conn)
2294 {
2295 struct lttcomm_relayd_generic_reply reply;
2296
2297 memset(&reply, 0, sizeof(reply));
2298 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2299 (void) send_response(conn->sock, &reply, sizeof(reply));
2300 }
2301
2302 /*
2303 * Process the commands received on the control socket
2304 */
2305 static
2306 int process_control(struct lttng_viewer_cmd *recv_hdr,
2307 struct relay_connection *conn)
2308 {
2309 int ret = 0;
2310 uint32_t msg_value;
2311
2312 msg_value = be32toh(recv_hdr->cmd);
2313
2314 /*
2315 * Make sure we've done the version check before any command other then a
2316 * new client connection.
2317 */
2318 if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
2319 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
2320 ret = -1;
2321 goto end;
2322 }
2323
2324 switch (msg_value) {
2325 case LTTNG_VIEWER_CONNECT:
2326 ret = viewer_connect(conn);
2327 break;
2328 case LTTNG_VIEWER_LIST_SESSIONS:
2329 ret = viewer_list_sessions(conn);
2330 break;
2331 case LTTNG_VIEWER_ATTACH_SESSION:
2332 ret = viewer_attach_session(conn);
2333 break;
2334 case LTTNG_VIEWER_GET_NEXT_INDEX:
2335 ret = viewer_get_next_index(conn);
2336 break;
2337 case LTTNG_VIEWER_GET_PACKET:
2338 ret = viewer_get_packet(conn);
2339 break;
2340 case LTTNG_VIEWER_GET_METADATA:
2341 ret = viewer_get_metadata(conn);
2342 break;
2343 case LTTNG_VIEWER_GET_NEW_STREAMS:
2344 ret = viewer_get_new_streams(conn);
2345 break;
2346 case LTTNG_VIEWER_CREATE_SESSION:
2347 ret = viewer_create_session(conn);
2348 break;
2349 case LTTNG_VIEWER_DETACH_SESSION:
2350 ret = viewer_detach_session(conn);
2351 break;
2352 default:
2353 ERR("Received unknown viewer command (%u)",
2354 be32toh(recv_hdr->cmd));
2355 live_relay_unknown_command(conn);
2356 ret = -1;
2357 goto end;
2358 }
2359
2360 end:
2361 return ret;
2362 }
2363
2364 static
2365 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
2366 {
2367 int ret;
2368
2369 (void) lttng_poll_del(events, pollfd);
2370
2371 ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker, &pollfd, 1,
2372 fd_tracker_util_close_fd, NULL);
2373 if (ret < 0) {
2374 ERR("Closing pollfd %d", pollfd);
2375 }
2376 }
2377
2378 /*
2379 * This thread does the actual work
2380 */
2381 static
2382 void *thread_worker(void *data)
2383 {
2384 int ret, err = -1;
2385 uint32_t nb_fd;
2386 struct lttng_poll_event events;
2387 struct lttng_ht *viewer_connections_ht;
2388 struct lttng_ht_iter iter;
2389 struct lttng_viewer_cmd recv_hdr;
2390 struct relay_connection *destroy_conn;
2391
2392 DBG("[thread] Live viewer relay worker started");
2393
2394 rcu_register_thread();
2395
2396 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
2397
2398 if (testpoint(relayd_thread_live_worker)) {
2399 goto error_testpoint;
2400 }
2401
2402 /* table of connections indexed on socket */
2403 viewer_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2404 if (!viewer_connections_ht) {
2405 goto viewer_connections_ht_error;
2406 }
2407
2408 ret = create_named_thread_poll_set(&events, 2,
2409 "Live viewer worker thread epoll");
2410 if (ret < 0) {
2411 goto error_poll_create;
2412 }
2413
2414 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
2415 if (ret < 0) {
2416 goto error;
2417 }
2418
2419 restart:
2420 while (1) {
2421 int i;
2422
2423 health_code_update();
2424
2425 /* Infinite blocking call, waiting for transmission */
2426 DBG3("Relayd live viewer worker thread polling...");
2427 health_poll_entry();
2428 ret = lttng_poll_wait(&events, -1);
2429 health_poll_exit();
2430 if (ret < 0) {
2431 /*
2432 * Restart interrupted system call.
2433 */
2434 if (errno == EINTR) {
2435 goto restart;
2436 }
2437 goto error;
2438 }
2439
2440 nb_fd = ret;
2441
2442 /*
2443 * Process control. The control connection is prioritised so we don't
2444 * starve it with high throughput tracing data on the data
2445 * connection.
2446 */
2447 for (i = 0; i < nb_fd; i++) {
2448 /* Fetch once the poll data */
2449 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2450 int pollfd = LTTNG_POLL_GETFD(&events, i);
2451
2452 health_code_update();
2453
2454 /* Thread quit pipe has been closed. Killing thread. */
2455 ret = check_thread_quit_pipe(pollfd, revents);
2456 if (ret) {
2457 err = 0;
2458 goto exit;
2459 }
2460
2461 /* Inspect the relay conn pipe for new connection. */
2462 if (pollfd == live_conn_pipe[0]) {
2463 if (revents & LPOLLIN) {
2464 struct relay_connection *conn;
2465
2466 ret = lttng_read(live_conn_pipe[0],
2467 &conn, sizeof(conn));
2468 if (ret < 0) {
2469 goto error;
2470 }
2471 ret = lttng_poll_add(&events,
2472 conn->sock->fd,
2473 LPOLLIN | LPOLLRDHUP);
2474 if (ret) {
2475 ERR("Failed to add new live connection file descriptor to poll set");
2476 goto error;
2477 }
2478 connection_ht_add(viewer_connections_ht, conn);
2479 DBG("Connection socket %d added to poll", conn->sock->fd);
2480 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2481 ERR("Relay live pipe error");
2482 goto error;
2483 } else {
2484 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2485 goto error;
2486 }
2487 } else {
2488 /* Connection activity. */
2489 struct relay_connection *conn;
2490
2491 conn = connection_get_by_sock(viewer_connections_ht, pollfd);
2492 if (!conn) {
2493 continue;
2494 }
2495
2496 if (revents & LPOLLIN) {
2497 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
2498 sizeof(recv_hdr), 0);
2499 if (ret <= 0) {
2500 /* Connection closed. */
2501 cleanup_connection_pollfd(&events, pollfd);
2502 /* Put "create" ownership reference. */
2503 connection_put(conn);
2504 DBG("Viewer control conn closed with %d", pollfd);
2505 } else {
2506 ret = process_control(&recv_hdr, conn);
2507 if (ret < 0) {
2508 /* Clear the session on error. */
2509 cleanup_connection_pollfd(&events, pollfd);
2510 /* Put "create" ownership reference. */
2511 connection_put(conn);
2512 DBG("Viewer connection closed with %d", pollfd);
2513 }
2514 }
2515 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2516 cleanup_connection_pollfd(&events, pollfd);
2517 /* Put "create" ownership reference. */
2518 connection_put(conn);
2519 } else {
2520 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2521 connection_put(conn);
2522 goto error;
2523 }
2524 /* Put local "get_by_sock" reference. */
2525 connection_put(conn);
2526 }
2527 }
2528 }
2529
2530 exit:
2531 error:
2532 (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
2533
2534 /* Cleanup remaining connection object. */
2535 rcu_read_lock();
2536 cds_lfht_for_each_entry(viewer_connections_ht->ht, &iter.iter,
2537 destroy_conn,
2538 sock_n.node) {
2539 health_code_update();
2540 connection_put(destroy_conn);
2541 }
2542 rcu_read_unlock();
2543 error_poll_create:
2544 lttng_ht_destroy(viewer_connections_ht);
2545 viewer_connections_ht_error:
2546 /* Close relay conn pipes */
2547 (void) fd_tracker_util_pipe_close(the_fd_tracker, live_conn_pipe);
2548 if (err) {
2549 DBG("Viewer worker thread exited with error");
2550 }
2551 DBG("Viewer worker thread cleanup complete");
2552 error_testpoint:
2553 if (err) {
2554 health_error();
2555 ERR("Health error occurred in %s", __func__);
2556 }
2557 health_unregister(health_relayd);
2558 if (lttng_relay_stop_threads()) {
2559 ERR("Error stopping threads");
2560 }
2561 rcu_unregister_thread();
2562 return NULL;
2563 }
2564
2565 /*
2566 * Create the relay command pipe to wake thread_manage_apps.
2567 * Closed in cleanup().
2568 */
2569 static int create_conn_pipe(void)
2570 {
2571 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
2572 "Live connection pipe", live_conn_pipe);
2573 }
2574
2575 int relayd_live_join(void)
2576 {
2577 int ret, retval = 0;
2578 void *status;
2579
2580 ret = pthread_join(live_listener_thread, &status);
2581 if (ret) {
2582 errno = ret;
2583 PERROR("pthread_join live listener");
2584 retval = -1;
2585 }
2586
2587 ret = pthread_join(live_worker_thread, &status);
2588 if (ret) {
2589 errno = ret;
2590 PERROR("pthread_join live worker");
2591 retval = -1;
2592 }
2593
2594 ret = pthread_join(live_dispatcher_thread, &status);
2595 if (ret) {
2596 errno = ret;
2597 PERROR("pthread_join live dispatcher");
2598 retval = -1;
2599 }
2600
2601 cleanup_relayd_live();
2602
2603 return retval;
2604 }
2605
2606 /*
2607 * main
2608 */
2609 int relayd_live_create(struct lttng_uri *uri)
2610 {
2611 int ret = 0, retval = 0;
2612 void *status;
2613 int is_root;
2614
2615 if (!uri) {
2616 retval = -1;
2617 goto exit_init_data;
2618 }
2619 live_uri = uri;
2620
2621 /* Check if daemon is UID = 0 */
2622 is_root = !getuid();
2623
2624 if (!is_root) {
2625 if (live_uri->port < 1024) {
2626 ERR("Need to be root to use ports < 1024");
2627 retval = -1;
2628 goto exit_init_data;
2629 }
2630 }
2631
2632 /* Setup the thread apps communication pipe. */
2633 if (create_conn_pipe()) {
2634 retval = -1;
2635 goto exit_init_data;
2636 }
2637
2638 /* Init relay command queue. */
2639 cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2640
2641 /* Set up max poll set size */
2642 if (lttng_poll_set_max_size()) {
2643 retval = -1;
2644 goto exit_init_data;
2645 }
2646
2647 /* Setup the dispatcher thread */
2648 ret = pthread_create(&live_dispatcher_thread, default_pthread_attr(),
2649 thread_dispatcher, (void *) NULL);
2650 if (ret) {
2651 errno = ret;
2652 PERROR("pthread_create viewer dispatcher");
2653 retval = -1;
2654 goto exit_dispatcher_thread;
2655 }
2656
2657 /* Setup the worker thread */
2658 ret = pthread_create(&live_worker_thread, default_pthread_attr(),
2659 thread_worker, NULL);
2660 if (ret) {
2661 errno = ret;
2662 PERROR("pthread_create viewer worker");
2663 retval = -1;
2664 goto exit_worker_thread;
2665 }
2666
2667 /* Setup the listener thread */
2668 ret = pthread_create(&live_listener_thread, default_pthread_attr(),
2669 thread_listener, (void *) NULL);
2670 if (ret) {
2671 errno = ret;
2672 PERROR("pthread_create viewer listener");
2673 retval = -1;
2674 goto exit_listener_thread;
2675 }
2676
2677 /*
2678 * All OK, started all threads.
2679 */
2680 return retval;
2681
2682 /*
2683 * Join on the live_listener_thread should anything be added after
2684 * the live_listener thread's creation.
2685 */
2686
2687 exit_listener_thread:
2688
2689 ret = pthread_join(live_worker_thread, &status);
2690 if (ret) {
2691 errno = ret;
2692 PERROR("pthread_join live worker");
2693 retval = -1;
2694 }
2695 exit_worker_thread:
2696
2697 ret = pthread_join(live_dispatcher_thread, &status);
2698 if (ret) {
2699 errno = ret;
2700 PERROR("pthread_join live dispatcher");
2701 retval = -1;
2702 }
2703 exit_dispatcher_thread:
2704
2705 exit_init_data:
2706 cleanup_relayd_live();
2707
2708 return retval;
2709 }
This page took 0.151817 seconds and 4 git commands to generate.