Fix: relayd: failure to open chunk files concurrently with session 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 /*
388 * Transition the viewer session into the newest trace chunk available.
389 */
390 if (!lttng_trace_chunk_ids_equal(viewer_session->current_trace_chunk,
391 relay_stream->trace_chunk)) {
392
393 ret = viewer_session_set_trace_chunk_copy(
394 viewer_session,
395 relay_stream->trace_chunk);
396 if (ret) {
397 ret = -1;
398 ctf_trace_put(ctf_trace);
399 goto error_unlock;
400 }
401 }
402
403 if (relay_stream->trace_chunk) {
404 /*
405 * If the corresponding relay
406 * stream's trace chunk is set,
407 * the viewer stream will be
408 * created under it.
409 *
410 * Note that a relay stream can
411 * have a NULL output trace
412 * chunk (for instance, after a
413 * clear against a stopped
414 * session).
415 */
416 const bool reference_acquired = lttng_trace_chunk_get(
417 viewer_session->current_trace_chunk);
418
419 assert(reference_acquired);
420 viewer_stream_trace_chunk =
421 viewer_session->current_trace_chunk;
422 }
423 }
424
425 viewer_stream = viewer_stream_create(
426 relay_stream,
427 viewer_stream_trace_chunk,
428 seek_t);
429 lttng_trace_chunk_put(viewer_stream_trace_chunk);
430 viewer_stream_trace_chunk = NULL;
431 if (!viewer_stream) {
432 ret = -1;
433 ctf_trace_put(ctf_trace);
434 goto error_unlock;
435 }
436
437 if (nb_created) {
438 /* Update number of created stream counter. */
439 (*nb_created)++;
440 }
441 /*
442 * Ensure a self-reference is preserved even
443 * after we have put our local reference.
444 */
445 if (!viewer_stream_get(viewer_stream)) {
446 ERR("Unable to get self-reference on viewer stream, logic error.");
447 abort();
448 }
449 } else {
450 if (!viewer_stream->sent_flag && nb_unsent) {
451 /* Update number of unsent stream counter. */
452 (*nb_unsent)++;
453 }
454 }
455 /* Update number of total stream counter. */
456 if (nb_total) {
457 if (relay_stream->is_metadata) {
458 if (!relay_stream->closed ||
459 relay_stream->metadata_received >
460 viewer_stream->metadata_sent) {
461 (*nb_total)++;
462 }
463 } else {
464 if (!relay_stream->closed ||
465 !(((int64_t)(relay_stream->prev_data_seq -
466 relay_stream->last_net_seq_num)) >=
467 0)) {
468 (*nb_total)++;
469 }
470 }
471 }
472 /* Put local reference. */
473 viewer_stream_put(viewer_stream);
474 next:
475 pthread_mutex_unlock(&relay_stream->lock);
476 stream_put(relay_stream);
477 }
478 relay_stream = NULL;
479 ctf_trace_put(ctf_trace);
480 }
481
482 ret = 0;
483
484 error_unlock:
485 rcu_read_unlock();
486
487 if (relay_stream) {
488 pthread_mutex_unlock(&relay_stream->lock);
489 stream_put(relay_stream);
490 }
491
492 return ret;
493 }
494
495 int relayd_live_stop(void)
496 {
497 /* Stop dispatch thread */
498 CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
499 futex_nto1_wake(&viewer_conn_queue.futex);
500 return 0;
501 }
502
503 /*
504 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
505 */
506 static
507 int create_named_thread_poll_set(struct lttng_poll_event *events,
508 int size, const char *name)
509 {
510 int ret;
511
512 if (events == NULL || size == 0) {
513 ret = -1;
514 goto error;
515 }
516
517 ret = fd_tracker_util_poll_create(the_fd_tracker,
518 name, events, 1, LTTNG_CLOEXEC);
519 if (ret) {
520 PERROR("Failed to create \"%s\" poll file descriptor", name);
521 goto error;
522 }
523
524 /* Add quit pipe */
525 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
526 if (ret < 0) {
527 goto error;
528 }
529
530 return 0;
531
532 error:
533 return ret;
534 }
535
536 /*
537 * Check if the thread quit pipe was triggered.
538 *
539 * Return 1 if it was triggered else 0;
540 */
541 static
542 int check_thread_quit_pipe(int fd, uint32_t events)
543 {
544 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
545 return 1;
546 }
547
548 return 0;
549 }
550
551 static
552 int create_sock(void *data, int *out_fd)
553 {
554 int ret;
555 struct lttcomm_sock *sock = data;
556
557 ret = lttcomm_create_sock(sock);
558 if (ret < 0) {
559 goto end;
560 }
561
562 *out_fd = sock->fd;
563 end:
564 return ret;
565 }
566
567 static
568 int close_sock(void *data, int *in_fd)
569 {
570 struct lttcomm_sock *sock = data;
571
572 return sock->ops->close(sock);
573 }
574
575 static int accept_sock(void *data, int *out_fd)
576 {
577 int ret = 0;
578 /* Socks is an array of in_sock, out_sock. */
579 struct lttcomm_sock **socks = data;
580 struct lttcomm_sock *in_sock = socks[0];
581
582 socks[1] = in_sock->ops->accept(in_sock);
583 if (!socks[1]) {
584 ret = -1;
585 goto end;
586 }
587 *out_fd = socks[1]->fd;
588 end:
589 return ret;
590 }
591
592 static
593 struct lttcomm_sock *accept_live_sock(struct lttcomm_sock *listening_sock,
594 const char *name)
595 {
596 int out_fd, ret;
597 struct lttcomm_sock *socks[2] = { listening_sock, NULL };
598 struct lttcomm_sock *new_sock = NULL;
599
600 ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker, &out_fd,
601 (const char **) &name, 1, accept_sock, &socks);
602 if (ret) {
603 goto end;
604 }
605 new_sock = socks[1];
606 DBG("%s accepted, socket %d", name, new_sock->fd);
607 end:
608 return new_sock;
609 }
610
611 /*
612 * Create and init socket from uri.
613 */
614 static
615 struct lttcomm_sock *init_socket(struct lttng_uri *uri, const char *name)
616 {
617 int ret, sock_fd;
618 struct lttcomm_sock *sock = NULL;
619 char uri_str[LTTNG_PATH_MAX];
620 char *formated_name = NULL;
621
622 sock = lttcomm_alloc_sock_from_uri(uri);
623 if (sock == NULL) {
624 ERR("Allocating socket");
625 goto error;
626 }
627
628 /*
629 * Don't fail to create the socket if the name can't be built as it is
630 * only used for debugging purposes.
631 */
632 ret = uri_to_str_url(uri, uri_str, sizeof(uri_str));
633 uri_str[sizeof(uri_str) - 1] = '\0';
634 if (ret >= 0) {
635 ret = asprintf(&formated_name, "%s socket @ %s", name,
636 uri_str);
637 if (ret < 0) {
638 formated_name = NULL;
639 }
640 }
641
642 ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker, &sock_fd,
643 (const char **) (formated_name ? &formated_name : NULL),
644 1, create_sock, sock);
645 if (ret) {
646 PERROR("Failed to create \"%s\" socket",
647 formated_name ?: "Unknown");
648 goto error;
649 }
650 DBG("Listening on %s socket %d", name, sock->fd);
651
652 ret = sock->ops->bind(sock);
653 if (ret < 0) {
654 PERROR("Failed to bind lttng-live socket");
655 goto error;
656 }
657
658 ret = sock->ops->listen(sock, -1);
659 if (ret < 0) {
660 goto error;
661
662 }
663
664 free(formated_name);
665 return sock;
666
667 error:
668 if (sock) {
669 lttcomm_destroy_sock(sock);
670 }
671 free(formated_name);
672 return NULL;
673 }
674
675 /*
676 * This thread manages the listening for new connections on the network
677 */
678 static
679 void *thread_listener(void *data)
680 {
681 int i, ret, pollfd, err = -1;
682 uint32_t revents, nb_fd;
683 struct lttng_poll_event events;
684 struct lttcomm_sock *live_control_sock;
685
686 DBG("[thread] Relay live listener started");
687
688 rcu_register_thread();
689 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
690
691 health_code_update();
692
693 live_control_sock = init_socket(live_uri, "Live listener");
694 if (!live_control_sock) {
695 goto error_sock_control;
696 }
697
698 /* Pass 2 as size here for the thread quit pipe and control sockets. */
699 ret = create_named_thread_poll_set(&events, 2,
700 "Live listener thread epoll");
701 if (ret < 0) {
702 goto error_create_poll;
703 }
704
705 /* Add the control socket */
706 ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
707 if (ret < 0) {
708 goto error_poll_add;
709 }
710
711 lttng_relay_notify_ready();
712
713 if (testpoint(relayd_thread_live_listener)) {
714 goto error_testpoint;
715 }
716
717 while (1) {
718 health_code_update();
719
720 DBG("Listener accepting live viewers connections");
721
722 restart:
723 health_poll_entry();
724 ret = lttng_poll_wait(&events, -1);
725 health_poll_exit();
726 if (ret < 0) {
727 /*
728 * Restart interrupted system call.
729 */
730 if (errno == EINTR) {
731 goto restart;
732 }
733 goto error;
734 }
735 nb_fd = ret;
736
737 DBG("Relay new viewer connection received");
738 for (i = 0; i < nb_fd; i++) {
739 health_code_update();
740
741 /* Fetch once the poll data */
742 revents = LTTNG_POLL_GETEV(&events, i);
743 pollfd = LTTNG_POLL_GETFD(&events, i);
744
745 /* Thread quit pipe has been closed. Killing thread. */
746 ret = check_thread_quit_pipe(pollfd, revents);
747 if (ret) {
748 err = 0;
749 goto exit;
750 }
751
752 if (revents & LPOLLIN) {
753 /*
754 * A new connection is requested, therefore a
755 * viewer connection is allocated in this
756 * thread, enqueued to a global queue and
757 * dequeued (and freed) in the worker thread.
758 */
759 int val = 1;
760 struct relay_connection *new_conn;
761 struct lttcomm_sock *newsock;
762
763 newsock = accept_live_sock(live_control_sock,
764 "Live socket to client");
765 if (!newsock) {
766 PERROR("accepting control sock");
767 goto error;
768 }
769 DBG("Relay viewer connection accepted socket %d", newsock->fd);
770
771 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
772 sizeof(val));
773 if (ret < 0) {
774 PERROR("setsockopt inet");
775 lttcomm_destroy_sock(newsock);
776 goto error;
777 }
778 new_conn = connection_create(newsock, RELAY_CONNECTION_UNKNOWN);
779 if (!new_conn) {
780 lttcomm_destroy_sock(newsock);
781 goto error;
782 }
783 /* Ownership assumed by the connection. */
784 newsock = NULL;
785
786 /* Enqueue request for the dispatcher thread. */
787 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
788 &new_conn->qnode);
789
790 /*
791 * Wake the dispatch queue futex.
792 * Implicit memory barrier with the
793 * exchange in cds_wfcq_enqueue.
794 */
795 futex_nto1_wake(&viewer_conn_queue.futex);
796 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
797 ERR("socket poll error");
798 goto error;
799 } else {
800 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
801 goto error;
802 }
803 }
804 }
805
806 exit:
807 error:
808 error_poll_add:
809 error_testpoint:
810 (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
811 error_create_poll:
812 if (live_control_sock->fd >= 0) {
813 int sock_fd = live_control_sock->fd;
814
815 ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker,
816 &sock_fd, 1, close_sock,
817 live_control_sock);
818 if (ret) {
819 PERROR("close");
820 }
821 live_control_sock->fd = -1;
822 }
823 lttcomm_destroy_sock(live_control_sock);
824 error_sock_control:
825 if (err) {
826 health_error();
827 DBG("Live viewer listener thread exited with error");
828 }
829 health_unregister(health_relayd);
830 rcu_unregister_thread();
831 DBG("Live viewer listener thread cleanup complete");
832 if (lttng_relay_stop_threads()) {
833 ERR("Error stopping threads");
834 }
835 return NULL;
836 }
837
838 /*
839 * This thread manages the dispatching of the requests to worker threads
840 */
841 static
842 void *thread_dispatcher(void *data)
843 {
844 int err = -1;
845 ssize_t ret;
846 struct cds_wfcq_node *node;
847 struct relay_connection *conn = NULL;
848
849 DBG("[thread] Live viewer relay dispatcher started");
850
851 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
852
853 if (testpoint(relayd_thread_live_dispatcher)) {
854 goto error_testpoint;
855 }
856
857 health_code_update();
858
859 for (;;) {
860 health_code_update();
861
862 /* Atomically prepare the queue futex */
863 futex_nto1_prepare(&viewer_conn_queue.futex);
864
865 if (CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
866 break;
867 }
868
869 do {
870 health_code_update();
871
872 /* Dequeue commands */
873 node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
874 &viewer_conn_queue.tail);
875 if (node == NULL) {
876 DBG("Woken up but nothing in the live-viewer "
877 "relay command queue");
878 /* Continue thread execution */
879 break;
880 }
881 conn = caa_container_of(node, struct relay_connection, qnode);
882 DBG("Dispatching viewer request waiting on sock %d",
883 conn->sock->fd);
884
885 /*
886 * Inform worker thread of the new request. This
887 * call is blocking so we can be assured that
888 * the data will be read at some point in time
889 * or wait to the end of the world :)
890 */
891 ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
892 if (ret < 0) {
893 PERROR("write conn pipe");
894 connection_put(conn);
895 goto error;
896 }
897 } while (node != NULL);
898
899 /* Futex wait on queue. Blocking call on futex() */
900 health_poll_entry();
901 futex_nto1_wait(&viewer_conn_queue.futex);
902 health_poll_exit();
903 }
904
905 /* Normal exit, no error */
906 err = 0;
907
908 error:
909 error_testpoint:
910 if (err) {
911 health_error();
912 ERR("Health error occurred in %s", __func__);
913 }
914 health_unregister(health_relayd);
915 DBG("Live viewer dispatch thread dying");
916 if (lttng_relay_stop_threads()) {
917 ERR("Error stopping threads");
918 }
919 return NULL;
920 }
921
922 /*
923 * Establish connection with the viewer and check the versions.
924 *
925 * Return 0 on success or else negative value.
926 */
927 static
928 int viewer_connect(struct relay_connection *conn)
929 {
930 int ret;
931 struct lttng_viewer_connect reply, msg;
932
933 conn->version_check_done = 1;
934
935 health_code_update();
936
937 DBG("Viewer is establishing a connection to the relayd.");
938
939 ret = recv_request(conn->sock, &msg, sizeof(msg));
940 if (ret < 0) {
941 goto end;
942 }
943
944 health_code_update();
945
946 memset(&reply, 0, sizeof(reply));
947 reply.major = RELAYD_VERSION_COMM_MAJOR;
948 reply.minor = RELAYD_VERSION_COMM_MINOR;
949
950 /* Major versions must be the same */
951 if (reply.major != be32toh(msg.major)) {
952 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
953 reply.major, be32toh(msg.major));
954 ret = -1;
955 goto end;
956 }
957
958 conn->major = reply.major;
959 /* We adapt to the lowest compatible version */
960 if (reply.minor <= be32toh(msg.minor)) {
961 conn->minor = reply.minor;
962 } else {
963 conn->minor = be32toh(msg.minor);
964 }
965
966 if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
967 conn->type = RELAY_VIEWER_COMMAND;
968 } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
969 conn->type = RELAY_VIEWER_NOTIFICATION;
970 } else {
971 ERR("Unknown connection type : %u", be32toh(msg.type));
972 ret = -1;
973 goto end;
974 }
975
976 reply.major = htobe32(reply.major);
977 reply.minor = htobe32(reply.minor);
978 if (conn->type == RELAY_VIEWER_COMMAND) {
979 /*
980 * Increment outside of htobe64 macro, because the argument can
981 * be used more than once within the macro, and thus the
982 * operation may be undefined.
983 */
984 pthread_mutex_lock(&last_relay_viewer_session_id_lock);
985 last_relay_viewer_session_id++;
986 pthread_mutex_unlock(&last_relay_viewer_session_id_lock);
987 reply.viewer_session_id = htobe64(last_relay_viewer_session_id);
988 }
989
990 health_code_update();
991
992 ret = send_response(conn->sock, &reply, sizeof(reply));
993 if (ret < 0) {
994 goto end;
995 }
996
997 health_code_update();
998
999 DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
1000 ret = 0;
1001
1002 end:
1003 return ret;
1004 }
1005
1006 /*
1007 * Send the viewer the list of current sessions.
1008 * We need to create a copy of the hash table content because otherwise
1009 * we cannot assume the number of entries stays the same between getting
1010 * the number of HT elements and iteration over the HT.
1011 *
1012 * Return 0 on success or else a negative value.
1013 */
1014 static
1015 int viewer_list_sessions(struct relay_connection *conn)
1016 {
1017 int ret = 0;
1018 struct lttng_viewer_list_sessions session_list;
1019 struct lttng_ht_iter iter;
1020 struct relay_session *session;
1021 struct lttng_viewer_session *send_session_buf = NULL;
1022 uint32_t buf_count = SESSION_BUF_DEFAULT_COUNT;
1023 uint32_t count = 0;
1024
1025 DBG("List sessions received");
1026
1027 send_session_buf = zmalloc(SESSION_BUF_DEFAULT_COUNT * sizeof(*send_session_buf));
1028 if (!send_session_buf) {
1029 return -1;
1030 }
1031
1032 rcu_read_lock();
1033 cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, session,
1034 session_n.node) {
1035 struct lttng_viewer_session *send_session;
1036
1037 health_code_update();
1038
1039 pthread_mutex_lock(&session->lock);
1040 if (session->connection_closed) {
1041 /* Skip closed session */
1042 goto next_session;
1043 }
1044
1045 if (count >= buf_count) {
1046 struct lttng_viewer_session *newbuf;
1047 uint32_t new_buf_count = buf_count << 1;
1048
1049 newbuf = realloc(send_session_buf,
1050 new_buf_count * sizeof(*send_session_buf));
1051 if (!newbuf) {
1052 ret = -1;
1053 goto break_loop;
1054 }
1055 send_session_buf = newbuf;
1056 buf_count = new_buf_count;
1057 }
1058 send_session = &send_session_buf[count];
1059 if (lttng_strncpy(send_session->session_name,
1060 session->session_name,
1061 sizeof(send_session->session_name))) {
1062 ret = -1;
1063 goto break_loop;
1064 }
1065 if (lttng_strncpy(send_session->hostname, session->hostname,
1066 sizeof(send_session->hostname))) {
1067 ret = -1;
1068 goto break_loop;
1069 }
1070 send_session->id = htobe64(session->id);
1071 send_session->live_timer = htobe32(session->live_timer);
1072 if (session->viewer_attached) {
1073 send_session->clients = htobe32(1);
1074 } else {
1075 send_session->clients = htobe32(0);
1076 }
1077 send_session->streams = htobe32(session->stream_count);
1078 count++;
1079 next_session:
1080 pthread_mutex_unlock(&session->lock);
1081 continue;
1082 break_loop:
1083 pthread_mutex_unlock(&session->lock);
1084 break;
1085 }
1086 rcu_read_unlock();
1087 if (ret < 0) {
1088 goto end_free;
1089 }
1090
1091 session_list.sessions_count = htobe32(count);
1092
1093 health_code_update();
1094
1095 ret = send_response(conn->sock, &session_list, sizeof(session_list));
1096 if (ret < 0) {
1097 goto end_free;
1098 }
1099
1100 health_code_update();
1101
1102 ret = send_response(conn->sock, send_session_buf,
1103 count * sizeof(*send_session_buf));
1104 if (ret < 0) {
1105 goto end_free;
1106 }
1107 health_code_update();
1108
1109 ret = 0;
1110 end_free:
1111 free(send_session_buf);
1112 return ret;
1113 }
1114
1115 /*
1116 * Send the viewer the list of current streams.
1117 */
1118 static
1119 int viewer_get_new_streams(struct relay_connection *conn)
1120 {
1121 int ret, send_streams = 0;
1122 uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0, nb_total = 0;
1123 struct lttng_viewer_new_streams_request request;
1124 struct lttng_viewer_new_streams_response response;
1125 struct relay_session *session = NULL;
1126 uint64_t session_id;
1127 bool closed = false;
1128
1129 assert(conn);
1130
1131 DBG("Get new streams received");
1132
1133 health_code_update();
1134
1135 /* Receive the request from the connected client. */
1136 ret = recv_request(conn->sock, &request, sizeof(request));
1137 if (ret < 0) {
1138 goto error;
1139 }
1140 session_id = be64toh(request.session_id);
1141
1142 health_code_update();
1143
1144 memset(&response, 0, sizeof(response));
1145
1146 session = session_get_by_id(session_id);
1147 if (!session) {
1148 DBG("Relay session %" PRIu64 " not found", session_id);
1149 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
1150 goto send_reply;
1151 }
1152
1153 if (!viewer_session_is_attached(conn->viewer_session, session)) {
1154 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
1155 goto send_reply;
1156 }
1157
1158 /*
1159 * For any new stream, create it with LTTNG_VIEWER_SEEK_BEGINNING since
1160 * that at this point the client is already attached to the session.Aany
1161 * initial stream will have been created with the seek type at attach
1162 * time (for now most readers use the LTTNG_VIEWER_SEEK_LAST on attach).
1163 * Otherwise any event happening in a new stream between the attach and
1164 * a call to viewer_get_new_streams will be "lost" (never received) from
1165 * the viewer's point of view.
1166 */
1167 pthread_mutex_lock(&session->lock);
1168 /*
1169 * If a session rotation is ongoing, do not attempt to open any
1170 * stream, because the chunk can be in an intermediate state
1171 * due to directory renaming.
1172 */
1173 if (session->ongoing_rotation) {
1174 DBG("Relay session %" PRIu64 " rotation ongoing", session_id);
1175 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_NO_NEW);
1176 goto send_reply_unlock;
1177 }
1178 ret = make_viewer_streams(session,
1179 conn->viewer_session,
1180 LTTNG_VIEWER_SEEK_BEGINNING, &nb_total, &nb_unsent,
1181 &nb_created, &closed);
1182 if (ret < 0) {
1183 goto error_unlock_session;
1184 }
1185 send_streams = 1;
1186 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
1187
1188 /* Only send back the newly created streams with the unsent ones. */
1189 nb_streams = nb_created + nb_unsent;
1190 response.streams_count = htobe32(nb_streams);
1191
1192 /*
1193 * If the session is closed, HUP when there are no more streams
1194 * with data.
1195 */
1196 if (closed && nb_total == 0) {
1197 send_streams = 0;
1198 response.streams_count = 0;
1199 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
1200 goto send_reply_unlock;
1201 }
1202 send_reply_unlock:
1203 pthread_mutex_unlock(&session->lock);
1204
1205 send_reply:
1206 health_code_update();
1207 ret = send_response(conn->sock, &response, sizeof(response));
1208 if (ret < 0) {
1209 goto end_put_session;
1210 }
1211 health_code_update();
1212
1213 /*
1214 * Unknown or empty session, just return gracefully, the viewer
1215 * knows what is happening.
1216 */
1217 if (!send_streams || !nb_streams) {
1218 ret = 0;
1219 goto end_put_session;
1220 }
1221
1222 /*
1223 * Send stream and *DON'T* ignore the sent flag so every viewer
1224 * streams that were not sent from that point will be sent to
1225 * the viewer.
1226 */
1227 ret = send_viewer_streams(conn->sock, session_id, 0);
1228 if (ret < 0) {
1229 goto end_put_session;
1230 }
1231
1232 end_put_session:
1233 if (session) {
1234 session_put(session);
1235 }
1236 error:
1237 return ret;
1238 error_unlock_session:
1239 pthread_mutex_unlock(&session->lock);
1240 session_put(session);
1241 return ret;
1242 }
1243
1244 /*
1245 * Send the viewer the list of current sessions.
1246 */
1247 static
1248 int viewer_attach_session(struct relay_connection *conn)
1249 {
1250 int send_streams = 0;
1251 ssize_t ret;
1252 uint32_t nb_streams = 0;
1253 enum lttng_viewer_seek seek_type;
1254 struct lttng_viewer_attach_session_request request;
1255 struct lttng_viewer_attach_session_response response;
1256 struct relay_session *session = NULL;
1257 enum lttng_viewer_attach_return_code viewer_attach_status;
1258 bool closed = false;
1259 uint64_t session_id;
1260
1261 assert(conn);
1262
1263 health_code_update();
1264
1265 /* Receive the request from the connected client. */
1266 ret = recv_request(conn->sock, &request, sizeof(request));
1267 if (ret < 0) {
1268 goto error;
1269 }
1270
1271 session_id = be64toh(request.session_id);
1272 health_code_update();
1273
1274 memset(&response, 0, sizeof(response));
1275
1276 if (!conn->viewer_session) {
1277 DBG("Client trying to attach before creating a live viewer session");
1278 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1279 goto send_reply;
1280 }
1281
1282 session = session_get_by_id(session_id);
1283 if (!session) {
1284 DBG("Relay session %" PRIu64 " not found", session_id);
1285 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1286 goto send_reply;
1287 }
1288 DBG("Attach session ID %" PRIu64 " received", session_id);
1289
1290 pthread_mutex_lock(&session->lock);
1291 if (session->live_timer == 0) {
1292 DBG("Not live session");
1293 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1294 goto send_reply;
1295 }
1296
1297 send_streams = 1;
1298 viewer_attach_status = viewer_session_attach(conn->viewer_session,
1299 session);
1300 if (viewer_attach_status != LTTNG_VIEWER_ATTACH_OK) {
1301 response.status = htobe32(viewer_attach_status);
1302 goto send_reply;
1303 }
1304
1305 switch (be32toh(request.seek)) {
1306 case LTTNG_VIEWER_SEEK_BEGINNING:
1307 case LTTNG_VIEWER_SEEK_LAST:
1308 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1309 seek_type = be32toh(request.seek);
1310 break;
1311 default:
1312 ERR("Wrong seek parameter");
1313 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1314 send_streams = 0;
1315 goto send_reply;
1316 }
1317
1318 /*
1319 * If a session rotation is ongoing, do not attempt to open any
1320 * stream, because the chunk can be in an intermediate state
1321 * due to directory renaming.
1322 */
1323 if (session->ongoing_rotation) {
1324 DBG("Relay session %" PRIu64 " rotation ongoing", session_id);
1325 send_streams = 0;
1326 goto send_reply;
1327 }
1328
1329 ret = make_viewer_streams(session,
1330 conn->viewer_session, seek_type,
1331 &nb_streams, NULL, NULL, &closed);
1332 if (ret < 0) {
1333 goto end_put_session;
1334 }
1335 pthread_mutex_unlock(&session->lock);
1336 session_put(session);
1337 session = NULL;
1338
1339 response.streams_count = htobe32(nb_streams);
1340 /*
1341 * If the session is closed when the viewer is attaching, it
1342 * means some of the streams may have been concurrently removed,
1343 * so we don't allow the viewer to attach, even if there are
1344 * streams available.
1345 */
1346 if (closed) {
1347 send_streams = 0;
1348 response.streams_count = 0;
1349 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1350 goto send_reply;
1351 }
1352
1353 send_reply:
1354 health_code_update();
1355 ret = send_response(conn->sock, &response, sizeof(response));
1356 if (ret < 0) {
1357 goto end_put_session;
1358 }
1359 health_code_update();
1360
1361 /*
1362 * Unknown or empty session, just return gracefully, the viewer
1363 * knows what is happening.
1364 */
1365 if (!send_streams || !nb_streams) {
1366 ret = 0;
1367 goto end_put_session;
1368 }
1369
1370 /* Send stream and ignore the sent flag. */
1371 ret = send_viewer_streams(conn->sock, session_id, 1);
1372 if (ret < 0) {
1373 goto end_put_session;
1374 }
1375
1376 end_put_session:
1377 if (session) {
1378 pthread_mutex_unlock(&session->lock);
1379 session_put(session);
1380 }
1381 error:
1382 return ret;
1383 }
1384
1385 /*
1386 * Open the index file if needed for the given vstream.
1387 *
1388 * If an index file is successfully opened, the vstream will set it as its
1389 * current index file.
1390 *
1391 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1392 *
1393 * Called with rstream lock held.
1394 */
1395 static int try_open_index(struct relay_viewer_stream *vstream,
1396 struct relay_stream *rstream)
1397 {
1398 int ret = 0;
1399 const uint32_t connection_major = rstream->trace->session->major;
1400 const uint32_t connection_minor = rstream->trace->session->minor;
1401 enum lttng_trace_chunk_status chunk_status;
1402
1403 if (vstream->index_file) {
1404 goto end;
1405 }
1406
1407 /*
1408 * First time, we open the index file and at least one index is ready.
1409 */
1410 if (rstream->index_received_seqcount == 0 ||
1411 !vstream->stream_file.trace_chunk) {
1412 ret = -ENOENT;
1413 goto end;
1414 }
1415
1416 chunk_status = lttng_index_file_create_from_trace_chunk_read_only(
1417 vstream->stream_file.trace_chunk, rstream->path_name,
1418 rstream->channel_name, rstream->tracefile_size,
1419 vstream->current_tracefile_id,
1420 lttng_to_index_major(connection_major, connection_minor),
1421 lttng_to_index_minor(connection_major, connection_minor),
1422 true, &vstream->index_file);
1423 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1424 if (chunk_status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE) {
1425 ret = -ENOENT;
1426 } else {
1427 ret = -1;
1428 }
1429 }
1430
1431 end:
1432 return ret;
1433 }
1434
1435 /*
1436 * Check the status of the index for the given stream. This function
1437 * updates the index structure if needed and can put (close) the vstream
1438 * in the HUP situation.
1439 *
1440 * Return 0 means that we can proceed with the index. A value of 1 means
1441 * that the index has been updated and is ready to be sent to the
1442 * client. A negative value indicates an error that can't be handled.
1443 *
1444 * Called with rstream lock held.
1445 */
1446 static int check_index_status(struct relay_viewer_stream *vstream,
1447 struct relay_stream *rstream, struct ctf_trace *trace,
1448 struct lttng_viewer_index *index)
1449 {
1450 int ret;
1451
1452 DBG("Check index status: index_received_seqcount %" PRIu64 " "
1453 "index_sent_seqcount %" PRIu64 " "
1454 "for stream %" PRIu64,
1455 rstream->index_received_seqcount,
1456 vstream->index_sent_seqcount,
1457 vstream->stream->stream_handle);
1458 if ((trace->session->connection_closed || rstream->closed)
1459 && rstream->index_received_seqcount
1460 == vstream->index_sent_seqcount) {
1461 /*
1462 * Last index sent and session connection or relay
1463 * stream are closed.
1464 */
1465 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1466 goto hup;
1467 } else if (rstream->beacon_ts_end != -1ULL &&
1468 (rstream->index_received_seqcount == 0 ||
1469 (vstream->index_sent_seqcount != 0 &&
1470 rstream->index_received_seqcount
1471 <= vstream->index_sent_seqcount))) {
1472 /*
1473 * We've received a synchronization beacon and the last index
1474 * available has been sent, the index for now is inactive.
1475 *
1476 * In this case, we have received a beacon which allows us to
1477 * inform the client of a time interval during which we can
1478 * guarantee that there are no events to read (and never will
1479 * be).
1480 *
1481 * The sent seqcount can grow higher than receive seqcount on
1482 * clear because the rotation performed by clear will push
1483 * the index_sent_seqcount ahead (see
1484 * viewer_stream_sync_tracefile_array_tail) and skip over
1485 * packet sequence numbers.
1486 */
1487 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1488 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1489 index->stream_id = htobe64(rstream->ctf_stream_id);
1490 DBG("Check index status: inactive with beacon, for stream %" PRIu64,
1491 vstream->stream->stream_handle);
1492 goto index_ready;
1493 } else if (rstream->index_received_seqcount == 0 ||
1494 (vstream->index_sent_seqcount != 0 &&
1495 rstream->index_received_seqcount
1496 <= vstream->index_sent_seqcount)) {
1497 /*
1498 * This checks whether received <= sent seqcount. In
1499 * this case, we have not received a beacon. Therefore,
1500 * we can only ask the client to retry later.
1501 *
1502 * The sent seqcount can grow higher than receive seqcount on
1503 * clear because the rotation performed by clear will push
1504 * the index_sent_seqcount ahead (see
1505 * viewer_stream_sync_tracefile_array_tail) and skip over
1506 * packet sequence numbers.
1507 */
1508 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1509 DBG("Check index status: retry for stream %" PRIu64,
1510 vstream->stream->stream_handle);
1511 goto index_ready;
1512 } else if (!tracefile_array_seq_in_file(rstream->tfa,
1513 vstream->current_tracefile_id,
1514 vstream->index_sent_seqcount)) {
1515 /*
1516 * The next index we want to send cannot be read either
1517 * because we need to perform a rotation, or due to
1518 * the producer having overwritten its trace file.
1519 */
1520 DBG("Viewer stream %" PRIu64 " rotation",
1521 vstream->stream->stream_handle);
1522 ret = viewer_stream_rotate(vstream);
1523 if (ret == 1) {
1524 /* EOF across entire stream. */
1525 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1526 goto hup;
1527 }
1528 /*
1529 * If we have been pushed due to overwrite, it
1530 * necessarily means there is data that can be read in
1531 * the stream. If we rotated because we reached the end
1532 * of a tracefile, it means the following tracefile
1533 * needs to contain at least one index, else we would
1534 * have already returned LTTNG_VIEWER_INDEX_RETRY to the
1535 * viewer. The updated index_sent_seqcount needs to
1536 * point to a readable index entry now.
1537 *
1538 * In the case where we "rotate" on a single file, we
1539 * can end up in a case where the requested index is
1540 * still unavailable.
1541 */
1542 if (rstream->tracefile_count == 1 &&
1543 !tracefile_array_seq_in_file(
1544 rstream->tfa,
1545 vstream->current_tracefile_id,
1546 vstream->index_sent_seqcount)) {
1547 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1548 DBG("Check index status: retry: "
1549 "tracefile array sequence number %" PRIu64
1550 " not in file for stream %" PRIu64,
1551 vstream->index_sent_seqcount,
1552 vstream->stream->stream_handle);
1553 goto index_ready;
1554 }
1555 assert(tracefile_array_seq_in_file(rstream->tfa,
1556 vstream->current_tracefile_id,
1557 vstream->index_sent_seqcount));
1558 }
1559 /* ret == 0 means successful so we continue. */
1560 ret = 0;
1561 return ret;
1562
1563 hup:
1564 viewer_stream_put(vstream);
1565 index_ready:
1566 return 1;
1567 }
1568
1569 static
1570 void viewer_stream_rotate_to_trace_chunk(struct relay_viewer_stream *vstream,
1571 struct lttng_trace_chunk *new_trace_chunk)
1572 {
1573 lttng_trace_chunk_put(vstream->stream_file.trace_chunk);
1574
1575 if (new_trace_chunk) {
1576 const bool acquired_reference = lttng_trace_chunk_get(
1577 new_trace_chunk);
1578
1579 assert(acquired_reference);
1580 }
1581
1582 vstream->stream_file.trace_chunk = new_trace_chunk;
1583 viewer_stream_sync_tracefile_array_tail(vstream);
1584 viewer_stream_close_files(vstream);
1585 }
1586
1587 /*
1588 * Send the next index for a stream.
1589 *
1590 * Return 0 on success or else a negative value.
1591 */
1592 static
1593 int viewer_get_next_index(struct relay_connection *conn)
1594 {
1595 int ret;
1596 struct lttng_viewer_get_next_index request_index;
1597 struct lttng_viewer_index viewer_index;
1598 struct ctf_packet_index packet_index;
1599 struct relay_viewer_stream *vstream = NULL;
1600 struct relay_stream *rstream = NULL;
1601 struct ctf_trace *ctf_trace = NULL;
1602 struct relay_viewer_stream *metadata_viewer_stream = NULL;
1603
1604 assert(conn);
1605
1606 DBG("Viewer get next index");
1607
1608 memset(&viewer_index, 0, sizeof(viewer_index));
1609 health_code_update();
1610
1611 ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1612 if (ret < 0) {
1613 goto end;
1614 }
1615 health_code_update();
1616
1617 vstream = viewer_stream_get_by_id(be64toh(request_index.stream_id));
1618 if (!vstream) {
1619 DBG("Client requested index of unknown stream id %" PRIu64,
1620 (uint64_t) be64toh(request_index.stream_id));
1621 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1622 goto send_reply;
1623 }
1624
1625 /* Use back. ref. Protected by refcounts. */
1626 rstream = vstream->stream;
1627 ctf_trace = rstream->trace;
1628
1629 /* metadata_viewer_stream may be NULL. */
1630 metadata_viewer_stream =
1631 ctf_trace_get_viewer_metadata_stream(ctf_trace);
1632
1633 pthread_mutex_lock(&rstream->lock);
1634
1635 /*
1636 * The viewer should not ask for index on metadata stream.
1637 */
1638 if (rstream->is_metadata) {
1639 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1640 goto send_reply;
1641 }
1642
1643 if (rstream->ongoing_rotation.is_set) {
1644 /* Rotation is ongoing, try again later. */
1645 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1646 goto send_reply;
1647 }
1648
1649 if (rstream->trace->session->ongoing_rotation) {
1650 /* Rotation is ongoing, try again later. */
1651 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1652 goto send_reply;
1653 }
1654
1655 /*
1656 * Transition the viewer session into the newest trace chunk available.
1657 */
1658 if (!lttng_trace_chunk_ids_equal(
1659 conn->viewer_session->current_trace_chunk,
1660 rstream->trace_chunk)) {
1661 DBG("Relay stream and viewer chunk ids differ");
1662
1663 ret = viewer_session_set_trace_chunk_copy(
1664 conn->viewer_session,
1665 rstream->trace_chunk);
1666 if (ret) {
1667 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1668 goto send_reply;
1669 }
1670 }
1671
1672 /*
1673 * Transition the viewer stream into the latest trace chunk available.
1674 *
1675 * Note that the stream must _not_ rotate in one precise condition:
1676 * the relay stream has rotated to a NULL trace chunk and the viewer
1677 * stream is consuming the trace chunk that was active just before
1678 * that rotation to NULL.
1679 *
1680 * This allows clients to consume all the packets of a trace chunk
1681 * after a session's destruction.
1682 */
1683 if (!lttng_trace_chunk_ids_equal(conn->viewer_session->current_trace_chunk, vstream->stream_file.trace_chunk) &&
1684 !(rstream->completed_rotation_count == vstream->last_seen_rotation_count + 1 && !rstream->trace_chunk)) {
1685 DBG("Viewer session and viewer stream chunk IDs differ: "
1686 "vsession chunk %p vstream chunk %p",
1687 conn->viewer_session->current_trace_chunk,
1688 vstream->stream_file.trace_chunk);
1689 viewer_stream_rotate_to_trace_chunk(vstream,
1690 conn->viewer_session->current_trace_chunk);
1691 vstream->last_seen_rotation_count =
1692 rstream->completed_rotation_count;
1693 }
1694
1695 ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1696 if (ret < 0) {
1697 goto error_put;
1698 } else if (ret == 1) {
1699 /*
1700 * We have no index to send and check_index_status has populated
1701 * viewer_index's status.
1702 */
1703 goto send_reply;
1704 }
1705 /* At this point, ret is 0 thus we will be able to read the index. */
1706 assert(!ret);
1707
1708 /* Try to open an index if one is needed for that stream. */
1709 ret = try_open_index(vstream, rstream);
1710 if (ret == -ENOENT) {
1711 if (rstream->closed) {
1712 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1713 goto send_reply;
1714 } else {
1715 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1716 goto send_reply;
1717 }
1718 }
1719 if (ret < 0) {
1720 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1721 goto send_reply;
1722 }
1723
1724 /*
1725 * vstream->stream_fd may be NULL if it has been closed by
1726 * tracefile rotation, or if we are at the beginning of the
1727 * stream. We open the data stream file here to protect against
1728 * overwrite caused by tracefile rotation (in association with
1729 * unlink performed before overwrite).
1730 */
1731 if (!vstream->stream_file.handle) {
1732 char file_path[LTTNG_PATH_MAX];
1733 enum lttng_trace_chunk_status status;
1734 struct fs_handle *fs_handle;
1735
1736 ret = utils_stream_file_path(rstream->path_name,
1737 rstream->channel_name, rstream->tracefile_size,
1738 vstream->current_tracefile_id, NULL, file_path,
1739 sizeof(file_path));
1740 if (ret < 0) {
1741 goto error_put;
1742 }
1743
1744 /*
1745 * It is possible the the file we are trying to open is
1746 * missing if the stream has been closed (application exits with
1747 * per-pid buffers) and a clear command has been performed.
1748 */
1749 status = lttng_trace_chunk_open_fs_handle(
1750 vstream->stream_file.trace_chunk,
1751 file_path, O_RDONLY, 0, &fs_handle, true);
1752 if (status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1753 if (status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE &&
1754 rstream->closed) {
1755 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1756 goto send_reply;
1757 }
1758 PERROR("Failed to open trace file for viewer stream");
1759 goto error_put;
1760 }
1761 vstream->stream_file.handle = fs_handle;
1762 }
1763
1764 ret = check_new_streams(conn);
1765 if (ret < 0) {
1766 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1767 goto send_reply;
1768 } else if (ret == 1) {
1769 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1770 }
1771
1772 ret = lttng_index_file_read(vstream->index_file, &packet_index);
1773 if (ret) {
1774 ERR("Relay error reading index file");
1775 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1776 goto send_reply;
1777 } else {
1778 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1779 vstream->index_sent_seqcount++;
1780 }
1781
1782 /*
1783 * Indexes are stored in big endian, no need to switch before sending.
1784 */
1785 DBG("Sending viewer index for stream %" PRIu64 " offset %" PRIu64,
1786 rstream->stream_handle,
1787 (uint64_t) be64toh(packet_index.offset));
1788 viewer_index.offset = packet_index.offset;
1789 viewer_index.packet_size = packet_index.packet_size;
1790 viewer_index.content_size = packet_index.content_size;
1791 viewer_index.timestamp_begin = packet_index.timestamp_begin;
1792 viewer_index.timestamp_end = packet_index.timestamp_end;
1793 viewer_index.events_discarded = packet_index.events_discarded;
1794 viewer_index.stream_id = packet_index.stream_id;
1795
1796 send_reply:
1797 if (rstream) {
1798 pthread_mutex_unlock(&rstream->lock);
1799 }
1800
1801 if (metadata_viewer_stream) {
1802 pthread_mutex_lock(&metadata_viewer_stream->stream->lock);
1803 DBG("get next index metadata check: recv %" PRIu64
1804 " sent %" PRIu64,
1805 metadata_viewer_stream->stream->metadata_received,
1806 metadata_viewer_stream->metadata_sent);
1807 if (!metadata_viewer_stream->stream->metadata_received ||
1808 metadata_viewer_stream->stream->metadata_received >
1809 metadata_viewer_stream->metadata_sent) {
1810 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1811 }
1812 pthread_mutex_unlock(&metadata_viewer_stream->stream->lock);
1813 }
1814
1815 viewer_index.flags = htobe32(viewer_index.flags);
1816 health_code_update();
1817
1818 ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1819 if (ret < 0) {
1820 goto end;
1821 }
1822 health_code_update();
1823
1824 if (vstream) {
1825 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1826 vstream->index_sent_seqcount,
1827 vstream->stream->stream_handle);
1828 }
1829 end:
1830 if (metadata_viewer_stream) {
1831 viewer_stream_put(metadata_viewer_stream);
1832 }
1833 if (vstream) {
1834 viewer_stream_put(vstream);
1835 }
1836 return ret;
1837
1838 error_put:
1839 pthread_mutex_unlock(&rstream->lock);
1840 if (metadata_viewer_stream) {
1841 viewer_stream_put(metadata_viewer_stream);
1842 }
1843 viewer_stream_put(vstream);
1844 return ret;
1845 }
1846
1847 /*
1848 * Send the next index for a stream
1849 *
1850 * Return 0 on success or else a negative value.
1851 */
1852 static
1853 int viewer_get_packet(struct relay_connection *conn)
1854 {
1855 int ret;
1856 off_t lseek_ret;
1857 char *reply = NULL;
1858 struct lttng_viewer_get_packet get_packet_info;
1859 struct lttng_viewer_trace_packet reply_header;
1860 struct relay_viewer_stream *vstream = NULL;
1861 uint32_t reply_size = sizeof(reply_header);
1862 uint32_t packet_data_len = 0;
1863 ssize_t read_len;
1864 uint64_t stream_id;
1865
1866 DBG2("Relay get data packet");
1867
1868 health_code_update();
1869
1870 ret = recv_request(conn->sock, &get_packet_info,
1871 sizeof(get_packet_info));
1872 if (ret < 0) {
1873 goto end;
1874 }
1875 health_code_update();
1876
1877 /* From this point on, the error label can be reached. */
1878 memset(&reply_header, 0, sizeof(reply_header));
1879 stream_id = (uint64_t) be64toh(get_packet_info.stream_id);
1880
1881 vstream = viewer_stream_get_by_id(stream_id);
1882 if (!vstream) {
1883 DBG("Client requested packet of unknown stream id %" PRIu64,
1884 stream_id);
1885 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1886 goto send_reply_nolock;
1887 } else {
1888 packet_data_len = be32toh(get_packet_info.len);
1889 reply_size += packet_data_len;
1890 }
1891
1892 reply = zmalloc(reply_size);
1893 if (!reply) {
1894 PERROR("packet reply zmalloc");
1895 reply_size = sizeof(reply_header);
1896 goto error;
1897 }
1898
1899 pthread_mutex_lock(&vstream->stream->lock);
1900 lseek_ret = fs_handle_seek(vstream->stream_file.handle,
1901 be64toh(get_packet_info.offset), SEEK_SET);
1902 if (lseek_ret < 0) {
1903 PERROR("Failed to seek file system handle of viewer stream %" PRIu64
1904 " to offset %" PRIu64,
1905 stream_id,
1906 (uint64_t) be64toh(get_packet_info.offset));
1907 goto error;
1908 }
1909 read_len = fs_handle_read(vstream->stream_file.handle,
1910 reply + sizeof(reply_header), packet_data_len);
1911 if (read_len < packet_data_len) {
1912 PERROR("Failed to read from file system handle of viewer stream id %" PRIu64
1913 ", offset: %" PRIu64,
1914 stream_id,
1915 (uint64_t) be64toh(get_packet_info.offset));
1916 goto error;
1917 }
1918 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1919 reply_header.len = htobe32(packet_data_len);
1920 goto send_reply;
1921
1922 error:
1923 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1924
1925 send_reply:
1926 if (vstream) {
1927 pthread_mutex_unlock(&vstream->stream->lock);
1928 }
1929 send_reply_nolock:
1930
1931 health_code_update();
1932
1933 if (reply) {
1934 memcpy(reply, &reply_header, sizeof(reply_header));
1935 ret = send_response(conn->sock, reply, reply_size);
1936 } else {
1937 /* No reply to send. */
1938 ret = send_response(conn->sock, &reply_header,
1939 reply_size);
1940 }
1941
1942 health_code_update();
1943 if (ret < 0) {
1944 PERROR("sendmsg of packet data failed");
1945 goto end_free;
1946 }
1947
1948 DBG("Sent %u bytes for stream %" PRIu64, reply_size, stream_id);
1949
1950 end_free:
1951 free(reply);
1952 end:
1953 if (vstream) {
1954 viewer_stream_put(vstream);
1955 }
1956 return ret;
1957 }
1958
1959 /*
1960 * Send the session's metadata
1961 *
1962 * Return 0 on success else a negative value.
1963 */
1964 static
1965 int viewer_get_metadata(struct relay_connection *conn)
1966 {
1967 int ret = 0;
1968 int fd = -1;
1969 ssize_t read_len;
1970 uint64_t len = 0;
1971 char *data = NULL;
1972 struct lttng_viewer_get_metadata request;
1973 struct lttng_viewer_metadata_packet reply;
1974 struct relay_viewer_stream *vstream = NULL;
1975
1976 assert(conn);
1977
1978 DBG("Relay get metadata");
1979
1980 health_code_update();
1981
1982 ret = recv_request(conn->sock, &request, sizeof(request));
1983 if (ret < 0) {
1984 goto end;
1985 }
1986 health_code_update();
1987
1988 memset(&reply, 0, sizeof(reply));
1989
1990 vstream = viewer_stream_get_by_id(be64toh(request.stream_id));
1991 if (!vstream) {
1992 /*
1993 * The metadata stream can be closed by a CLOSE command
1994 * just before we attach. It can also be closed by
1995 * per-pid tracing during tracing. Therefore, it is
1996 * possible that we cannot find this viewer stream.
1997 * Reply back to the client with an error if we cannot
1998 * find it.
1999 */
2000 DBG("Client requested metadata of unknown stream id %" PRIu64,
2001 (uint64_t) be64toh(request.stream_id));
2002 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
2003 goto send_reply;
2004 }
2005 pthread_mutex_lock(&vstream->stream->lock);
2006 if (!vstream->stream->is_metadata) {
2007 ERR("Invalid metadata stream");
2008 goto error;
2009 }
2010
2011 if (vstream->metadata_sent >= vstream->stream->metadata_received) {
2012 /*
2013 * The live viewers expect to receive a NO_NEW_METADATA
2014 * status before a stream disappears, otherwise they abort the
2015 * entire live connection when receiving an error status.
2016 *
2017 * Clear feature resets the metadata_sent to 0 until the
2018 * same metadata is received again.
2019 */
2020 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
2021 /*
2022 * The live viewer considers a closed 0 byte metadata stream as
2023 * an error.
2024 */
2025 if (vstream->metadata_sent > 0) {
2026 if (vstream->stream->closed && vstream->stream->no_new_metadata_notified) {
2027 /* Release ownership for the viewer metadata stream. */
2028 viewer_stream_put(vstream);
2029 }
2030 vstream->stream->no_new_metadata_notified = true;
2031 }
2032 goto send_reply;
2033 }
2034
2035 if (vstream->stream->trace_chunk &&
2036 !lttng_trace_chunk_ids_equal(
2037 conn->viewer_session->current_trace_chunk,
2038 vstream->stream->trace_chunk)) {
2039 /* A rotation has occurred on the relay stream. */
2040 DBG("Metadata relay stream and viewer chunk ids differ");
2041
2042 ret = viewer_session_set_trace_chunk_copy(
2043 conn->viewer_session,
2044 vstream->stream->trace_chunk);
2045 if (ret) {
2046 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
2047 goto send_reply;
2048 }
2049 }
2050
2051 if (conn->viewer_session->current_trace_chunk &&
2052 !lttng_trace_chunk_ids_equal(conn->viewer_session->current_trace_chunk,
2053 vstream->stream_file.trace_chunk)) {
2054 bool acquired_reference;
2055
2056 DBG("Viewer session and viewer stream chunk differ: "
2057 "vsession chunk %p vstream chunk %p",
2058 conn->viewer_session->current_trace_chunk,
2059 vstream->stream_file.trace_chunk);
2060 lttng_trace_chunk_put(vstream->stream_file.trace_chunk);
2061 acquired_reference = lttng_trace_chunk_get(conn->viewer_session->current_trace_chunk);
2062 assert(acquired_reference);
2063 vstream->stream_file.trace_chunk =
2064 conn->viewer_session->current_trace_chunk;
2065 viewer_stream_close_files(vstream);
2066 }
2067
2068 len = vstream->stream->metadata_received - vstream->metadata_sent;
2069
2070 if (!vstream->stream_file.trace_chunk) {
2071 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
2072 len = 0;
2073 goto send_reply;
2074 } else if (vstream->stream_file.trace_chunk &&
2075 !vstream->stream_file.handle && len > 0) {
2076 /*
2077 * Either this is the first time the metadata file is read, or a
2078 * rotation of the corresponding relay stream has occurred.
2079 */
2080 struct fs_handle *fs_handle;
2081 char file_path[LTTNG_PATH_MAX];
2082 enum lttng_trace_chunk_status status;
2083 struct relay_stream *rstream = vstream->stream;
2084
2085 ret = utils_stream_file_path(rstream->path_name,
2086 rstream->channel_name, rstream->tracefile_size,
2087 vstream->current_tracefile_id, NULL, file_path,
2088 sizeof(file_path));
2089 if (ret < 0) {
2090 goto error;
2091 }
2092
2093 /*
2094 * It is possible the the metadata file we are trying to open is
2095 * missing if the stream has been closed (application exits with
2096 * per-pid buffers) and a clear command has been performed.
2097 */
2098 status = lttng_trace_chunk_open_fs_handle(
2099 vstream->stream_file.trace_chunk,
2100 file_path, O_RDONLY, 0, &fs_handle, true);
2101 if (status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2102 if (status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE) {
2103 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
2104 len = 0;
2105 if (vstream->stream->closed) {
2106 viewer_stream_put(vstream);
2107 }
2108 goto send_reply;
2109 }
2110 PERROR("Failed to open metadata file for viewer stream");
2111 goto error;
2112 }
2113 vstream->stream_file.handle = fs_handle;
2114
2115 if (vstream->metadata_sent != 0) {
2116 /*
2117 * The client does not expect to receive any metadata
2118 * it has received and metadata files in successive
2119 * chunks must be a strict superset of one another.
2120 *
2121 * Skip the first `metadata_sent` bytes to ensure
2122 * they are not sent a second time to the client.
2123 *
2124 * Baring a block layer error or an internal error,
2125 * this seek should not fail as
2126 * `vstream->stream->metadata_received` is reset when
2127 * a relay stream is rotated. If this is reached, it is
2128 * safe to assume that
2129 * `metadata_received` > `metadata_sent`.
2130 */
2131 const off_t seek_ret = fs_handle_seek(fs_handle,
2132 vstream->metadata_sent, SEEK_SET);
2133
2134 if (seek_ret < 0) {
2135 PERROR("Failed to seek metadata viewer stream file to `sent` position: pos = %" PRId64,
2136 vstream->metadata_sent);
2137 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
2138 goto send_reply;
2139 }
2140 }
2141 }
2142
2143 reply.len = htobe64(len);
2144 data = zmalloc(len);
2145 if (!data) {
2146 PERROR("viewer metadata zmalloc");
2147 goto error;
2148 }
2149
2150 fd = fs_handle_get_fd(vstream->stream_file.handle);
2151 if (fd < 0) {
2152 ERR("Failed to restore viewer stream file system handle");
2153 goto error;
2154 }
2155 read_len = lttng_read(fd, data, len);
2156 fs_handle_put_fd(vstream->stream_file.handle);
2157 fd = -1;
2158 if (read_len < len) {
2159 if (read_len < 0) {
2160 PERROR("Failed to read metadata file");
2161 goto error;
2162 } else {
2163 /*
2164 * A clear has been performed which prevents the relay
2165 * from sending `len` bytes of metadata.
2166 *
2167 * It is important not to send any metadata if we
2168 * couldn't read all the available metadata in one shot:
2169 * sending partial metadata can cause the client to
2170 * attempt to parse an incomplete (incoherent) metadata
2171 * stream, which would result in an error.
2172 */
2173 const off_t seek_ret = fs_handle_seek(
2174 vstream->stream_file.handle, -read_len,
2175 SEEK_CUR);
2176
2177 DBG("Failed to read metadata: requested = %" PRIu64 ", got = %zd",
2178 len, read_len);
2179 read_len = 0;
2180 len = 0;
2181 if (seek_ret < 0) {
2182 PERROR("Failed to restore metadata file position after partial read");
2183 ret = -1;
2184 goto error;
2185 }
2186 }
2187 }
2188 vstream->metadata_sent += read_len;
2189 reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
2190
2191 goto send_reply;
2192
2193 error:
2194 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
2195
2196 send_reply:
2197 health_code_update();
2198 if (vstream) {
2199 pthread_mutex_unlock(&vstream->stream->lock);
2200 }
2201 ret = send_response(conn->sock, &reply, sizeof(reply));
2202 if (ret < 0) {
2203 goto end_free;
2204 }
2205 health_code_update();
2206
2207 if (len > 0) {
2208 ret = send_response(conn->sock, data, len);
2209 if (ret < 0) {
2210 goto end_free;
2211 }
2212 }
2213
2214 DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
2215 (uint64_t) be64toh(request.stream_id));
2216
2217 DBG("Metadata sent");
2218
2219 end_free:
2220 free(data);
2221 end:
2222 if (vstream) {
2223 viewer_stream_put(vstream);
2224 }
2225 return ret;
2226 }
2227
2228 /*
2229 * Create a viewer session.
2230 *
2231 * Return 0 on success or else a negative value.
2232 */
2233 static
2234 int viewer_create_session(struct relay_connection *conn)
2235 {
2236 int ret;
2237 struct lttng_viewer_create_session_response resp;
2238
2239 DBG("Viewer create session received");
2240
2241 memset(&resp, 0, sizeof(resp));
2242 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
2243 conn->viewer_session = viewer_session_create();
2244 if (!conn->viewer_session) {
2245 ERR("Allocation viewer session");
2246 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
2247 goto send_reply;
2248 }
2249
2250 send_reply:
2251 health_code_update();
2252 ret = send_response(conn->sock, &resp, sizeof(resp));
2253 if (ret < 0) {
2254 goto end;
2255 }
2256 health_code_update();
2257 ret = 0;
2258
2259 end:
2260 return ret;
2261 }
2262
2263 /*
2264 * Detach a viewer session.
2265 *
2266 * Return 0 on success or else a negative value.
2267 */
2268 static
2269 int viewer_detach_session(struct relay_connection *conn)
2270 {
2271 int ret;
2272 struct lttng_viewer_detach_session_response response;
2273 struct lttng_viewer_detach_session_request request;
2274 struct relay_session *session = NULL;
2275 uint64_t viewer_session_to_close;
2276
2277 DBG("Viewer detach session received");
2278
2279 assert(conn);
2280
2281 health_code_update();
2282
2283 /* Receive the request from the connected client. */
2284 ret = recv_request(conn->sock, &request, sizeof(request));
2285 if (ret < 0) {
2286 goto end;
2287 }
2288 viewer_session_to_close = be64toh(request.session_id);
2289
2290 if (!conn->viewer_session) {
2291 DBG("Client trying to detach before creating a live viewer session");
2292 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
2293 goto send_reply;
2294 }
2295
2296 health_code_update();
2297
2298 memset(&response, 0, sizeof(response));
2299 DBG("Detaching from session ID %" PRIu64, viewer_session_to_close);
2300
2301 session = session_get_by_id(be64toh(request.session_id));
2302 if (!session) {
2303 DBG("Relay session %" PRIu64 " not found",
2304 (uint64_t) be64toh(request.session_id));
2305 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_UNK);
2306 goto send_reply;
2307 }
2308
2309 ret = viewer_session_is_attached(conn->viewer_session, session);
2310 if (ret != 1) {
2311 DBG("Not attached to this session");
2312 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
2313 goto send_reply_put;
2314 }
2315
2316 viewer_session_close_one_session(conn->viewer_session, session);
2317 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_OK);
2318 DBG("Session %" PRIu64 " detached.", viewer_session_to_close);
2319
2320 send_reply_put:
2321 session_put(session);
2322
2323 send_reply:
2324 health_code_update();
2325 ret = send_response(conn->sock, &response, sizeof(response));
2326 if (ret < 0) {
2327 goto end;
2328 }
2329 health_code_update();
2330 ret = 0;
2331
2332 end:
2333 return ret;
2334 }
2335
2336 /*
2337 * live_relay_unknown_command: send -1 if received unknown command
2338 */
2339 static
2340 void live_relay_unknown_command(struct relay_connection *conn)
2341 {
2342 struct lttcomm_relayd_generic_reply reply;
2343
2344 memset(&reply, 0, sizeof(reply));
2345 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2346 (void) send_response(conn->sock, &reply, sizeof(reply));
2347 }
2348
2349 /*
2350 * Process the commands received on the control socket
2351 */
2352 static
2353 int process_control(struct lttng_viewer_cmd *recv_hdr,
2354 struct relay_connection *conn)
2355 {
2356 int ret = 0;
2357 uint32_t msg_value;
2358
2359 msg_value = be32toh(recv_hdr->cmd);
2360
2361 /*
2362 * Make sure we've done the version check before any command other then a
2363 * new client connection.
2364 */
2365 if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
2366 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
2367 ret = -1;
2368 goto end;
2369 }
2370
2371 switch (msg_value) {
2372 case LTTNG_VIEWER_CONNECT:
2373 ret = viewer_connect(conn);
2374 break;
2375 case LTTNG_VIEWER_LIST_SESSIONS:
2376 ret = viewer_list_sessions(conn);
2377 break;
2378 case LTTNG_VIEWER_ATTACH_SESSION:
2379 ret = viewer_attach_session(conn);
2380 break;
2381 case LTTNG_VIEWER_GET_NEXT_INDEX:
2382 ret = viewer_get_next_index(conn);
2383 break;
2384 case LTTNG_VIEWER_GET_PACKET:
2385 ret = viewer_get_packet(conn);
2386 break;
2387 case LTTNG_VIEWER_GET_METADATA:
2388 ret = viewer_get_metadata(conn);
2389 break;
2390 case LTTNG_VIEWER_GET_NEW_STREAMS:
2391 ret = viewer_get_new_streams(conn);
2392 break;
2393 case LTTNG_VIEWER_CREATE_SESSION:
2394 ret = viewer_create_session(conn);
2395 break;
2396 case LTTNG_VIEWER_DETACH_SESSION:
2397 ret = viewer_detach_session(conn);
2398 break;
2399 default:
2400 ERR("Received unknown viewer command (%u)",
2401 be32toh(recv_hdr->cmd));
2402 live_relay_unknown_command(conn);
2403 ret = -1;
2404 goto end;
2405 }
2406
2407 end:
2408 return ret;
2409 }
2410
2411 static
2412 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
2413 {
2414 int ret;
2415
2416 (void) lttng_poll_del(events, pollfd);
2417
2418 ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker, &pollfd, 1,
2419 fd_tracker_util_close_fd, NULL);
2420 if (ret < 0) {
2421 ERR("Closing pollfd %d", pollfd);
2422 }
2423 }
2424
2425 /*
2426 * This thread does the actual work
2427 */
2428 static
2429 void *thread_worker(void *data)
2430 {
2431 int ret, err = -1;
2432 uint32_t nb_fd;
2433 struct lttng_poll_event events;
2434 struct lttng_ht *viewer_connections_ht;
2435 struct lttng_ht_iter iter;
2436 struct lttng_viewer_cmd recv_hdr;
2437 struct relay_connection *destroy_conn;
2438
2439 DBG("[thread] Live viewer relay worker started");
2440
2441 rcu_register_thread();
2442
2443 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
2444
2445 if (testpoint(relayd_thread_live_worker)) {
2446 goto error_testpoint;
2447 }
2448
2449 /* table of connections indexed on socket */
2450 viewer_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2451 if (!viewer_connections_ht) {
2452 goto viewer_connections_ht_error;
2453 }
2454
2455 ret = create_named_thread_poll_set(&events, 2,
2456 "Live viewer worker thread epoll");
2457 if (ret < 0) {
2458 goto error_poll_create;
2459 }
2460
2461 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
2462 if (ret < 0) {
2463 goto error;
2464 }
2465
2466 restart:
2467 while (1) {
2468 int i;
2469
2470 health_code_update();
2471
2472 /* Infinite blocking call, waiting for transmission */
2473 DBG3("Relayd live viewer worker thread polling...");
2474 health_poll_entry();
2475 ret = lttng_poll_wait(&events, -1);
2476 health_poll_exit();
2477 if (ret < 0) {
2478 /*
2479 * Restart interrupted system call.
2480 */
2481 if (errno == EINTR) {
2482 goto restart;
2483 }
2484 goto error;
2485 }
2486
2487 nb_fd = ret;
2488
2489 /*
2490 * Process control. The control connection is prioritised so we don't
2491 * starve it with high throughput tracing data on the data
2492 * connection.
2493 */
2494 for (i = 0; i < nb_fd; i++) {
2495 /* Fetch once the poll data */
2496 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2497 int pollfd = LTTNG_POLL_GETFD(&events, i);
2498
2499 health_code_update();
2500
2501 /* Thread quit pipe has been closed. Killing thread. */
2502 ret = check_thread_quit_pipe(pollfd, revents);
2503 if (ret) {
2504 err = 0;
2505 goto exit;
2506 }
2507
2508 /* Inspect the relay conn pipe for new connection. */
2509 if (pollfd == live_conn_pipe[0]) {
2510 if (revents & LPOLLIN) {
2511 struct relay_connection *conn;
2512
2513 ret = lttng_read(live_conn_pipe[0],
2514 &conn, sizeof(conn));
2515 if (ret < 0) {
2516 goto error;
2517 }
2518 ret = lttng_poll_add(&events,
2519 conn->sock->fd,
2520 LPOLLIN | LPOLLRDHUP);
2521 if (ret) {
2522 ERR("Failed to add new live connection file descriptor to poll set");
2523 goto error;
2524 }
2525 connection_ht_add(viewer_connections_ht, conn);
2526 DBG("Connection socket %d added to poll", conn->sock->fd);
2527 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2528 ERR("Relay live pipe error");
2529 goto error;
2530 } else {
2531 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2532 goto error;
2533 }
2534 } else {
2535 /* Connection activity. */
2536 struct relay_connection *conn;
2537
2538 conn = connection_get_by_sock(viewer_connections_ht, pollfd);
2539 if (!conn) {
2540 continue;
2541 }
2542
2543 if (revents & LPOLLIN) {
2544 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
2545 sizeof(recv_hdr), 0);
2546 if (ret <= 0) {
2547 /* Connection closed. */
2548 cleanup_connection_pollfd(&events, pollfd);
2549 /* Put "create" ownership reference. */
2550 connection_put(conn);
2551 DBG("Viewer control conn closed with %d", pollfd);
2552 } else {
2553 ret = process_control(&recv_hdr, conn);
2554 if (ret < 0) {
2555 /* Clear the session on error. */
2556 cleanup_connection_pollfd(&events, pollfd);
2557 /* Put "create" ownership reference. */
2558 connection_put(conn);
2559 DBG("Viewer connection closed with %d", pollfd);
2560 }
2561 }
2562 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2563 cleanup_connection_pollfd(&events, pollfd);
2564 /* Put "create" ownership reference. */
2565 connection_put(conn);
2566 } else {
2567 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2568 connection_put(conn);
2569 goto error;
2570 }
2571 /* Put local "get_by_sock" reference. */
2572 connection_put(conn);
2573 }
2574 }
2575 }
2576
2577 exit:
2578 error:
2579 (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
2580
2581 /* Cleanup remaining connection object. */
2582 rcu_read_lock();
2583 cds_lfht_for_each_entry(viewer_connections_ht->ht, &iter.iter,
2584 destroy_conn,
2585 sock_n.node) {
2586 health_code_update();
2587 connection_put(destroy_conn);
2588 }
2589 rcu_read_unlock();
2590 error_poll_create:
2591 lttng_ht_destroy(viewer_connections_ht);
2592 viewer_connections_ht_error:
2593 /* Close relay conn pipes */
2594 (void) fd_tracker_util_pipe_close(the_fd_tracker, live_conn_pipe);
2595 if (err) {
2596 DBG("Viewer worker thread exited with error");
2597 }
2598 DBG("Viewer worker thread cleanup complete");
2599 error_testpoint:
2600 if (err) {
2601 health_error();
2602 ERR("Health error occurred in %s", __func__);
2603 }
2604 health_unregister(health_relayd);
2605 if (lttng_relay_stop_threads()) {
2606 ERR("Error stopping threads");
2607 }
2608 rcu_unregister_thread();
2609 return NULL;
2610 }
2611
2612 /*
2613 * Create the relay command pipe to wake thread_manage_apps.
2614 * Closed in cleanup().
2615 */
2616 static int create_conn_pipe(void)
2617 {
2618 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
2619 "Live connection pipe", live_conn_pipe);
2620 }
2621
2622 int relayd_live_join(void)
2623 {
2624 int ret, retval = 0;
2625 void *status;
2626
2627 ret = pthread_join(live_listener_thread, &status);
2628 if (ret) {
2629 errno = ret;
2630 PERROR("pthread_join live listener");
2631 retval = -1;
2632 }
2633
2634 ret = pthread_join(live_worker_thread, &status);
2635 if (ret) {
2636 errno = ret;
2637 PERROR("pthread_join live worker");
2638 retval = -1;
2639 }
2640
2641 ret = pthread_join(live_dispatcher_thread, &status);
2642 if (ret) {
2643 errno = ret;
2644 PERROR("pthread_join live dispatcher");
2645 retval = -1;
2646 }
2647
2648 cleanup_relayd_live();
2649
2650 return retval;
2651 }
2652
2653 /*
2654 * main
2655 */
2656 int relayd_live_create(struct lttng_uri *uri)
2657 {
2658 int ret = 0, retval = 0;
2659 void *status;
2660 int is_root;
2661
2662 if (!uri) {
2663 retval = -1;
2664 goto exit_init_data;
2665 }
2666 live_uri = uri;
2667
2668 /* Check if daemon is UID = 0 */
2669 is_root = !getuid();
2670
2671 if (!is_root) {
2672 if (live_uri->port < 1024) {
2673 ERR("Need to be root to use ports < 1024");
2674 retval = -1;
2675 goto exit_init_data;
2676 }
2677 }
2678
2679 /* Setup the thread apps communication pipe. */
2680 if (create_conn_pipe()) {
2681 retval = -1;
2682 goto exit_init_data;
2683 }
2684
2685 /* Init relay command queue. */
2686 cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2687
2688 /* Set up max poll set size */
2689 if (lttng_poll_set_max_size()) {
2690 retval = -1;
2691 goto exit_init_data;
2692 }
2693
2694 /* Setup the dispatcher thread */
2695 ret = pthread_create(&live_dispatcher_thread, default_pthread_attr(),
2696 thread_dispatcher, (void *) NULL);
2697 if (ret) {
2698 errno = ret;
2699 PERROR("pthread_create viewer dispatcher");
2700 retval = -1;
2701 goto exit_dispatcher_thread;
2702 }
2703
2704 /* Setup the worker thread */
2705 ret = pthread_create(&live_worker_thread, default_pthread_attr(),
2706 thread_worker, NULL);
2707 if (ret) {
2708 errno = ret;
2709 PERROR("pthread_create viewer worker");
2710 retval = -1;
2711 goto exit_worker_thread;
2712 }
2713
2714 /* Setup the listener thread */
2715 ret = pthread_create(&live_listener_thread, default_pthread_attr(),
2716 thread_listener, (void *) NULL);
2717 if (ret) {
2718 errno = ret;
2719 PERROR("pthread_create viewer listener");
2720 retval = -1;
2721 goto exit_listener_thread;
2722 }
2723
2724 /*
2725 * All OK, started all threads.
2726 */
2727 return retval;
2728
2729 /*
2730 * Join on the live_listener_thread should anything be added after
2731 * the live_listener thread's creation.
2732 */
2733
2734 exit_listener_thread:
2735
2736 ret = pthread_join(live_worker_thread, &status);
2737 if (ret) {
2738 errno = ret;
2739 PERROR("pthread_join live worker");
2740 retval = -1;
2741 }
2742 exit_worker_thread:
2743
2744 ret = pthread_join(live_dispatcher_thread, &status);
2745 if (ret) {
2746 errno = ret;
2747 PERROR("pthread_join live dispatcher");
2748 retval = -1;
2749 }
2750 exit_dispatcher_thread:
2751
2752 exit_init_data:
2753 cleanup_relayd_live();
2754
2755 return retval;
2756 }
This page took 0.124855 seconds and 5 git commands to generate.