Fix: relayd: live: mishandled initial null trace chunk
[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 ret = make_viewer_streams(session,
1169 conn->viewer_session,
1170 LTTNG_VIEWER_SEEK_BEGINNING, &nb_total, &nb_unsent,
1171 &nb_created, &closed);
1172 if (ret < 0) {
1173 goto error_unlock_session;
1174 }
1175 send_streams = 1;
1176 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
1177
1178 /* Only send back the newly created streams with the unsent ones. */
1179 nb_streams = nb_created + nb_unsent;
1180 response.streams_count = htobe32(nb_streams);
1181
1182 /*
1183 * If the session is closed, HUP when there are no more streams
1184 * with data.
1185 */
1186 if (closed && nb_total == 0) {
1187 send_streams = 0;
1188 response.streams_count = 0;
1189 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
1190 goto send_reply_unlock;
1191 }
1192 send_reply_unlock:
1193 pthread_mutex_unlock(&session->lock);
1194
1195 send_reply:
1196 health_code_update();
1197 ret = send_response(conn->sock, &response, sizeof(response));
1198 if (ret < 0) {
1199 goto end_put_session;
1200 }
1201 health_code_update();
1202
1203 /*
1204 * Unknown or empty session, just return gracefully, the viewer
1205 * knows what is happening.
1206 */
1207 if (!send_streams || !nb_streams) {
1208 ret = 0;
1209 goto end_put_session;
1210 }
1211
1212 /*
1213 * Send stream and *DON'T* ignore the sent flag so every viewer
1214 * streams that were not sent from that point will be sent to
1215 * the viewer.
1216 */
1217 ret = send_viewer_streams(conn->sock, session_id, 0);
1218 if (ret < 0) {
1219 goto end_put_session;
1220 }
1221
1222 end_put_session:
1223 if (session) {
1224 session_put(session);
1225 }
1226 error:
1227 return ret;
1228 error_unlock_session:
1229 pthread_mutex_unlock(&session->lock);
1230 session_put(session);
1231 return ret;
1232 }
1233
1234 /*
1235 * Send the viewer the list of current sessions.
1236 */
1237 static
1238 int viewer_attach_session(struct relay_connection *conn)
1239 {
1240 int send_streams = 0;
1241 ssize_t ret;
1242 uint32_t nb_streams = 0;
1243 enum lttng_viewer_seek seek_type;
1244 struct lttng_viewer_attach_session_request request;
1245 struct lttng_viewer_attach_session_response response;
1246 struct relay_session *session = NULL;
1247 enum lttng_viewer_attach_return_code viewer_attach_status;
1248 bool closed = false;
1249 uint64_t session_id;
1250
1251 assert(conn);
1252
1253 health_code_update();
1254
1255 /* Receive the request from the connected client. */
1256 ret = recv_request(conn->sock, &request, sizeof(request));
1257 if (ret < 0) {
1258 goto error;
1259 }
1260
1261 session_id = be64toh(request.session_id);
1262 health_code_update();
1263
1264 memset(&response, 0, sizeof(response));
1265
1266 if (!conn->viewer_session) {
1267 DBG("Client trying to attach before creating a live viewer session");
1268 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1269 goto send_reply;
1270 }
1271
1272 session = session_get_by_id(session_id);
1273 if (!session) {
1274 DBG("Relay session %" PRIu64 " not found", session_id);
1275 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1276 goto send_reply;
1277 }
1278 DBG("Attach session ID %" PRIu64 " received", session_id);
1279
1280 pthread_mutex_lock(&session->lock);
1281 if (session->live_timer == 0) {
1282 DBG("Not live session");
1283 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1284 goto send_reply;
1285 }
1286
1287 send_streams = 1;
1288 viewer_attach_status = viewer_session_attach(conn->viewer_session,
1289 session);
1290 if (viewer_attach_status != LTTNG_VIEWER_ATTACH_OK) {
1291 response.status = htobe32(viewer_attach_status);
1292 goto send_reply;
1293 }
1294
1295 switch (be32toh(request.seek)) {
1296 case LTTNG_VIEWER_SEEK_BEGINNING:
1297 case LTTNG_VIEWER_SEEK_LAST:
1298 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1299 seek_type = be32toh(request.seek);
1300 break;
1301 default:
1302 ERR("Wrong seek parameter");
1303 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1304 send_streams = 0;
1305 goto send_reply;
1306 }
1307
1308 ret = make_viewer_streams(session,
1309 conn->viewer_session, seek_type,
1310 &nb_streams, NULL, NULL, &closed);
1311 if (ret < 0) {
1312 goto end_put_session;
1313 }
1314 pthread_mutex_unlock(&session->lock);
1315 session_put(session);
1316 session = NULL;
1317
1318 response.streams_count = htobe32(nb_streams);
1319 /*
1320 * If the session is closed when the viewer is attaching, it
1321 * means some of the streams may have been concurrently removed,
1322 * so we don't allow the viewer to attach, even if there are
1323 * streams available.
1324 */
1325 if (closed) {
1326 send_streams = 0;
1327 response.streams_count = 0;
1328 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1329 goto send_reply;
1330 }
1331
1332 send_reply:
1333 health_code_update();
1334 ret = send_response(conn->sock, &response, sizeof(response));
1335 if (ret < 0) {
1336 goto end_put_session;
1337 }
1338 health_code_update();
1339
1340 /*
1341 * Unknown or empty session, just return gracefully, the viewer
1342 * knows what is happening.
1343 */
1344 if (!send_streams || !nb_streams) {
1345 ret = 0;
1346 goto end_put_session;
1347 }
1348
1349 /* Send stream and ignore the sent flag. */
1350 ret = send_viewer_streams(conn->sock, session_id, 1);
1351 if (ret < 0) {
1352 goto end_put_session;
1353 }
1354
1355 end_put_session:
1356 if (session) {
1357 pthread_mutex_unlock(&session->lock);
1358 session_put(session);
1359 }
1360 error:
1361 return ret;
1362 }
1363
1364 /*
1365 * Open the index file if needed for the given vstream.
1366 *
1367 * If an index file is successfully opened, the vstream will set it as its
1368 * current index file.
1369 *
1370 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1371 *
1372 * Called with rstream lock held.
1373 */
1374 static int try_open_index(struct relay_viewer_stream *vstream,
1375 struct relay_stream *rstream)
1376 {
1377 int ret = 0;
1378 const uint32_t connection_major = rstream->trace->session->major;
1379 const uint32_t connection_minor = rstream->trace->session->minor;
1380 enum lttng_trace_chunk_status chunk_status;
1381
1382 if (vstream->index_file) {
1383 goto end;
1384 }
1385
1386 /*
1387 * First time, we open the index file and at least one index is ready.
1388 */
1389 if (rstream->index_received_seqcount == 0 ||
1390 !vstream->stream_file.trace_chunk) {
1391 ret = -ENOENT;
1392 goto end;
1393 }
1394
1395 chunk_status = lttng_index_file_create_from_trace_chunk_read_only(
1396 vstream->stream_file.trace_chunk, rstream->path_name,
1397 rstream->channel_name, rstream->tracefile_size,
1398 vstream->current_tracefile_id,
1399 lttng_to_index_major(connection_major, connection_minor),
1400 lttng_to_index_minor(connection_major, connection_minor),
1401 true, &vstream->index_file);
1402 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1403 if (chunk_status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE) {
1404 ret = -ENOENT;
1405 } else {
1406 ret = -1;
1407 }
1408 }
1409
1410 end:
1411 return ret;
1412 }
1413
1414 /*
1415 * Check the status of the index for the given stream. This function
1416 * updates the index structure if needed and can put (close) the vstream
1417 * in the HUP situation.
1418 *
1419 * Return 0 means that we can proceed with the index. A value of 1 means
1420 * that the index has been updated and is ready to be sent to the
1421 * client. A negative value indicates an error that can't be handled.
1422 *
1423 * Called with rstream lock held.
1424 */
1425 static int check_index_status(struct relay_viewer_stream *vstream,
1426 struct relay_stream *rstream, struct ctf_trace *trace,
1427 struct lttng_viewer_index *index)
1428 {
1429 int ret;
1430
1431 DBG("Check index status: index_received_seqcount %" PRIu64 " "
1432 "index_sent_seqcount %" PRIu64 " "
1433 "for stream %" PRIu64,
1434 rstream->index_received_seqcount,
1435 vstream->index_sent_seqcount,
1436 vstream->stream->stream_handle);
1437 if ((trace->session->connection_closed || rstream->closed)
1438 && rstream->index_received_seqcount
1439 == vstream->index_sent_seqcount) {
1440 /*
1441 * Last index sent and session connection or relay
1442 * stream are closed.
1443 */
1444 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1445 goto hup;
1446 } else if (rstream->beacon_ts_end != -1ULL &&
1447 (rstream->index_received_seqcount == 0 ||
1448 (vstream->index_sent_seqcount != 0 &&
1449 rstream->index_received_seqcount
1450 <= vstream->index_sent_seqcount))) {
1451 /*
1452 * We've received a synchronization beacon and the last index
1453 * available has been sent, the index for now is inactive.
1454 *
1455 * In this case, we have received a beacon which allows us to
1456 * inform the client of a time interval during which we can
1457 * guarantee that there are no events to read (and never will
1458 * be).
1459 *
1460 * The sent seqcount can grow higher than receive seqcount on
1461 * clear because the rotation performed by clear will push
1462 * the index_sent_seqcount ahead (see
1463 * viewer_stream_sync_tracefile_array_tail) and skip over
1464 * packet sequence numbers.
1465 */
1466 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1467 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1468 index->stream_id = htobe64(rstream->ctf_stream_id);
1469 DBG("Check index status: inactive with beacon, for stream %" PRIu64,
1470 vstream->stream->stream_handle);
1471 goto index_ready;
1472 } else if (rstream->index_received_seqcount == 0 ||
1473 (vstream->index_sent_seqcount != 0 &&
1474 rstream->index_received_seqcount
1475 <= vstream->index_sent_seqcount)) {
1476 /*
1477 * This checks whether received <= sent seqcount. In
1478 * this case, we have not received a beacon. Therefore,
1479 * we can only ask the client to retry later.
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_RETRY);
1488 DBG("Check index status: retry for stream %" PRIu64,
1489 vstream->stream->stream_handle);
1490 goto index_ready;
1491 } else if (!tracefile_array_seq_in_file(rstream->tfa,
1492 vstream->current_tracefile_id,
1493 vstream->index_sent_seqcount)) {
1494 /*
1495 * The next index we want to send cannot be read either
1496 * because we need to perform a rotation, or due to
1497 * the producer having overwritten its trace file.
1498 */
1499 DBG("Viewer stream %" PRIu64 " rotation",
1500 vstream->stream->stream_handle);
1501 ret = viewer_stream_rotate(vstream);
1502 if (ret == 1) {
1503 /* EOF across entire stream. */
1504 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1505 goto hup;
1506 }
1507 /*
1508 * If we have been pushed due to overwrite, it
1509 * necessarily means there is data that can be read in
1510 * the stream. If we rotated because we reached the end
1511 * of a tracefile, it means the following tracefile
1512 * needs to contain at least one index, else we would
1513 * have already returned LTTNG_VIEWER_INDEX_RETRY to the
1514 * viewer. The updated index_sent_seqcount needs to
1515 * point to a readable index entry now.
1516 *
1517 * In the case where we "rotate" on a single file, we
1518 * can end up in a case where the requested index is
1519 * still unavailable.
1520 */
1521 if (rstream->tracefile_count == 1 &&
1522 !tracefile_array_seq_in_file(
1523 rstream->tfa,
1524 vstream->current_tracefile_id,
1525 vstream->index_sent_seqcount)) {
1526 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1527 DBG("Check index status: retry: "
1528 "tracefile array sequence number %" PRIu64
1529 " not in file for stream %" PRIu64,
1530 vstream->index_sent_seqcount,
1531 vstream->stream->stream_handle);
1532 goto index_ready;
1533 }
1534 assert(tracefile_array_seq_in_file(rstream->tfa,
1535 vstream->current_tracefile_id,
1536 vstream->index_sent_seqcount));
1537 }
1538 /* ret == 0 means successful so we continue. */
1539 ret = 0;
1540 return ret;
1541
1542 hup:
1543 viewer_stream_put(vstream);
1544 index_ready:
1545 return 1;
1546 }
1547
1548 static
1549 void viewer_stream_rotate_to_trace_chunk(struct relay_viewer_stream *vstream,
1550 struct lttng_trace_chunk *new_trace_chunk)
1551 {
1552 lttng_trace_chunk_put(vstream->stream_file.trace_chunk);
1553
1554 if (new_trace_chunk) {
1555 const bool acquired_reference = lttng_trace_chunk_get(
1556 new_trace_chunk);
1557
1558 assert(acquired_reference);
1559 }
1560
1561 vstream->stream_file.trace_chunk = new_trace_chunk;
1562 viewer_stream_sync_tracefile_array_tail(vstream);
1563 viewer_stream_close_files(vstream);
1564 }
1565
1566 /*
1567 * Send the next index for a stream.
1568 *
1569 * Return 0 on success or else a negative value.
1570 */
1571 static
1572 int viewer_get_next_index(struct relay_connection *conn)
1573 {
1574 int ret;
1575 struct lttng_viewer_get_next_index request_index;
1576 struct lttng_viewer_index viewer_index;
1577 struct ctf_packet_index packet_index;
1578 struct relay_viewer_stream *vstream = NULL;
1579 struct relay_stream *rstream = NULL;
1580 struct ctf_trace *ctf_trace = NULL;
1581 struct relay_viewer_stream *metadata_viewer_stream = NULL;
1582
1583 assert(conn);
1584
1585 DBG("Viewer get next index");
1586
1587 memset(&viewer_index, 0, sizeof(viewer_index));
1588 health_code_update();
1589
1590 ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1591 if (ret < 0) {
1592 goto end;
1593 }
1594 health_code_update();
1595
1596 vstream = viewer_stream_get_by_id(be64toh(request_index.stream_id));
1597 if (!vstream) {
1598 DBG("Client requested index of unknown stream id %" PRIu64,
1599 (uint64_t) be64toh(request_index.stream_id));
1600 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1601 goto send_reply;
1602 }
1603
1604 /* Use back. ref. Protected by refcounts. */
1605 rstream = vstream->stream;
1606 ctf_trace = rstream->trace;
1607
1608 /* metadata_viewer_stream may be NULL. */
1609 metadata_viewer_stream =
1610 ctf_trace_get_viewer_metadata_stream(ctf_trace);
1611
1612 pthread_mutex_lock(&rstream->lock);
1613
1614 /*
1615 * The viewer should not ask for index on metadata stream.
1616 */
1617 if (rstream->is_metadata) {
1618 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1619 goto send_reply;
1620 }
1621
1622 if (rstream->ongoing_rotation.is_set) {
1623 /* Rotation is ongoing, try again later. */
1624 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1625 goto send_reply;
1626 }
1627
1628 if (rstream->trace->session->ongoing_rotation) {
1629 /* Rotation is ongoing, try again later. */
1630 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1631 goto send_reply;
1632 }
1633
1634 /*
1635 * Transition the viewer session into the newest trace chunk available.
1636 */
1637 if (!lttng_trace_chunk_ids_equal(
1638 conn->viewer_session->current_trace_chunk,
1639 rstream->trace_chunk)) {
1640 DBG("Relay stream and viewer chunk ids differ");
1641
1642 ret = viewer_session_set_trace_chunk_copy(
1643 conn->viewer_session,
1644 rstream->trace_chunk);
1645 if (ret) {
1646 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1647 goto send_reply;
1648 }
1649 }
1650
1651 /*
1652 * Transition the viewer stream into the latest trace chunk available.
1653 *
1654 * Note that the stream must _not_ rotate in one precise condition:
1655 * the relay stream has rotated to a NULL trace chunk and the viewer
1656 * stream is consuming the trace chunk that was active just before
1657 * that rotation to NULL.
1658 *
1659 * This allows clients to consume all the packets of a trace chunk
1660 * after a session's destruction.
1661 */
1662 if (conn->viewer_session->current_trace_chunk != vstream->stream_file.trace_chunk &&
1663 !(rstream->completed_rotation_count == vstream->last_seen_rotation_count + 1 && !rstream->trace_chunk)) {
1664 DBG("Viewer session and viewer stream chunk differ: "
1665 "vsession chunk %p vstream chunk %p",
1666 conn->viewer_session->current_trace_chunk,
1667 vstream->stream_file.trace_chunk);
1668 viewer_stream_rotate_to_trace_chunk(vstream,
1669 conn->viewer_session->current_trace_chunk);
1670 vstream->last_seen_rotation_count =
1671 rstream->completed_rotation_count;
1672 }
1673
1674 ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1675 if (ret < 0) {
1676 goto error_put;
1677 } else if (ret == 1) {
1678 /*
1679 * We have no index to send and check_index_status has populated
1680 * viewer_index's status.
1681 */
1682 goto send_reply;
1683 }
1684 /* At this point, ret is 0 thus we will be able to read the index. */
1685 assert(!ret);
1686
1687 /* Try to open an index if one is needed for that stream. */
1688 ret = try_open_index(vstream, rstream);
1689 if (ret == -ENOENT) {
1690 if (rstream->closed) {
1691 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1692 goto send_reply;
1693 } else {
1694 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1695 goto send_reply;
1696 }
1697 }
1698 if (ret < 0) {
1699 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1700 goto send_reply;
1701 }
1702
1703 /*
1704 * vstream->stream_fd may be NULL if it has been closed by
1705 * tracefile rotation, or if we are at the beginning of the
1706 * stream. We open the data stream file here to protect against
1707 * overwrite caused by tracefile rotation (in association with
1708 * unlink performed before overwrite).
1709 */
1710 if (!vstream->stream_file.handle) {
1711 char file_path[LTTNG_PATH_MAX];
1712 enum lttng_trace_chunk_status status;
1713 struct fs_handle *fs_handle;
1714
1715 ret = utils_stream_file_path(rstream->path_name,
1716 rstream->channel_name, rstream->tracefile_size,
1717 vstream->current_tracefile_id, NULL, file_path,
1718 sizeof(file_path));
1719 if (ret < 0) {
1720 goto error_put;
1721 }
1722
1723 /*
1724 * It is possible the the file we are trying to open is
1725 * missing if the stream has been closed (application exits with
1726 * per-pid buffers) and a clear command has been performed.
1727 */
1728 status = lttng_trace_chunk_open_fs_handle(
1729 vstream->stream_file.trace_chunk,
1730 file_path, O_RDONLY, 0, &fs_handle, true);
1731 if (status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1732 if (status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE &&
1733 rstream->closed) {
1734 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1735 goto send_reply;
1736 }
1737 PERROR("Failed to open trace file for viewer stream");
1738 goto error_put;
1739 }
1740 vstream->stream_file.handle = fs_handle;
1741 }
1742
1743 ret = check_new_streams(conn);
1744 if (ret < 0) {
1745 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1746 goto send_reply;
1747 } else if (ret == 1) {
1748 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1749 }
1750
1751 ret = lttng_index_file_read(vstream->index_file, &packet_index);
1752 if (ret) {
1753 ERR("Relay error reading index file");
1754 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1755 goto send_reply;
1756 } else {
1757 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1758 vstream->index_sent_seqcount++;
1759 }
1760
1761 /*
1762 * Indexes are stored in big endian, no need to switch before sending.
1763 */
1764 DBG("Sending viewer index for stream %" PRIu64 " offset %" PRIu64,
1765 rstream->stream_handle,
1766 (uint64_t) be64toh(packet_index.offset));
1767 viewer_index.offset = packet_index.offset;
1768 viewer_index.packet_size = packet_index.packet_size;
1769 viewer_index.content_size = packet_index.content_size;
1770 viewer_index.timestamp_begin = packet_index.timestamp_begin;
1771 viewer_index.timestamp_end = packet_index.timestamp_end;
1772 viewer_index.events_discarded = packet_index.events_discarded;
1773 viewer_index.stream_id = packet_index.stream_id;
1774
1775 send_reply:
1776 if (rstream) {
1777 pthread_mutex_unlock(&rstream->lock);
1778 }
1779
1780 if (metadata_viewer_stream) {
1781 pthread_mutex_lock(&metadata_viewer_stream->stream->lock);
1782 DBG("get next index metadata check: recv %" PRIu64
1783 " sent %" PRIu64,
1784 metadata_viewer_stream->stream->metadata_received,
1785 metadata_viewer_stream->metadata_sent);
1786 if (!metadata_viewer_stream->stream->metadata_received ||
1787 metadata_viewer_stream->stream->metadata_received >
1788 metadata_viewer_stream->metadata_sent) {
1789 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1790 }
1791 pthread_mutex_unlock(&metadata_viewer_stream->stream->lock);
1792 }
1793
1794 viewer_index.flags = htobe32(viewer_index.flags);
1795 health_code_update();
1796
1797 ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1798 if (ret < 0) {
1799 goto end;
1800 }
1801 health_code_update();
1802
1803 if (vstream) {
1804 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1805 vstream->index_sent_seqcount,
1806 vstream->stream->stream_handle);
1807 }
1808 end:
1809 if (metadata_viewer_stream) {
1810 viewer_stream_put(metadata_viewer_stream);
1811 }
1812 if (vstream) {
1813 viewer_stream_put(vstream);
1814 }
1815 return ret;
1816
1817 error_put:
1818 pthread_mutex_unlock(&rstream->lock);
1819 if (metadata_viewer_stream) {
1820 viewer_stream_put(metadata_viewer_stream);
1821 }
1822 viewer_stream_put(vstream);
1823 return ret;
1824 }
1825
1826 /*
1827 * Send the next index for a stream
1828 *
1829 * Return 0 on success or else a negative value.
1830 */
1831 static
1832 int viewer_get_packet(struct relay_connection *conn)
1833 {
1834 int ret;
1835 off_t lseek_ret;
1836 char *reply = NULL;
1837 struct lttng_viewer_get_packet get_packet_info;
1838 struct lttng_viewer_trace_packet reply_header;
1839 struct relay_viewer_stream *vstream = NULL;
1840 uint32_t reply_size = sizeof(reply_header);
1841 uint32_t packet_data_len = 0;
1842 ssize_t read_len;
1843 uint64_t stream_id;
1844
1845 DBG2("Relay get data packet");
1846
1847 health_code_update();
1848
1849 ret = recv_request(conn->sock, &get_packet_info,
1850 sizeof(get_packet_info));
1851 if (ret < 0) {
1852 goto end;
1853 }
1854 health_code_update();
1855
1856 /* From this point on, the error label can be reached. */
1857 memset(&reply_header, 0, sizeof(reply_header));
1858 stream_id = (uint64_t) be64toh(get_packet_info.stream_id);
1859
1860 vstream = viewer_stream_get_by_id(stream_id);
1861 if (!vstream) {
1862 DBG("Client requested packet of unknown stream id %" PRIu64,
1863 stream_id);
1864 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1865 goto send_reply_nolock;
1866 } else {
1867 packet_data_len = be32toh(get_packet_info.len);
1868 reply_size += packet_data_len;
1869 }
1870
1871 reply = zmalloc(reply_size);
1872 if (!reply) {
1873 PERROR("packet reply zmalloc");
1874 reply_size = sizeof(reply_header);
1875 goto error;
1876 }
1877
1878 pthread_mutex_lock(&vstream->stream->lock);
1879 lseek_ret = fs_handle_seek(vstream->stream_file.handle,
1880 be64toh(get_packet_info.offset), SEEK_SET);
1881 if (lseek_ret < 0) {
1882 PERROR("Failed to seek file system handle of viewer stream %" PRIu64
1883 " to offset %" PRIu64,
1884 stream_id,
1885 (uint64_t) be64toh(get_packet_info.offset));
1886 goto error;
1887 }
1888 read_len = fs_handle_read(vstream->stream_file.handle,
1889 reply + sizeof(reply_header), packet_data_len);
1890 if (read_len < packet_data_len) {
1891 PERROR("Failed to read from file system handle of viewer stream id %" PRIu64
1892 ", offset: %" PRIu64,
1893 stream_id,
1894 (uint64_t) be64toh(get_packet_info.offset));
1895 goto error;
1896 }
1897 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1898 reply_header.len = htobe32(packet_data_len);
1899 goto send_reply;
1900
1901 error:
1902 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1903
1904 send_reply:
1905 if (vstream) {
1906 pthread_mutex_unlock(&vstream->stream->lock);
1907 }
1908 send_reply_nolock:
1909
1910 health_code_update();
1911
1912 if (reply) {
1913 memcpy(reply, &reply_header, sizeof(reply_header));
1914 ret = send_response(conn->sock, reply, reply_size);
1915 } else {
1916 /* No reply to send. */
1917 ret = send_response(conn->sock, &reply_header,
1918 reply_size);
1919 }
1920
1921 health_code_update();
1922 if (ret < 0) {
1923 PERROR("sendmsg of packet data failed");
1924 goto end_free;
1925 }
1926
1927 DBG("Sent %u bytes for stream %" PRIu64, reply_size, stream_id);
1928
1929 end_free:
1930 free(reply);
1931 end:
1932 if (vstream) {
1933 viewer_stream_put(vstream);
1934 }
1935 return ret;
1936 }
1937
1938 /*
1939 * Send the session's metadata
1940 *
1941 * Return 0 on success else a negative value.
1942 */
1943 static
1944 int viewer_get_metadata(struct relay_connection *conn)
1945 {
1946 int ret = 0;
1947 int fd = -1;
1948 ssize_t read_len;
1949 uint64_t len = 0;
1950 char *data = NULL;
1951 struct lttng_viewer_get_metadata request;
1952 struct lttng_viewer_metadata_packet reply;
1953 struct relay_viewer_stream *vstream = NULL;
1954
1955 assert(conn);
1956
1957 DBG("Relay get metadata");
1958
1959 health_code_update();
1960
1961 ret = recv_request(conn->sock, &request, sizeof(request));
1962 if (ret < 0) {
1963 goto end;
1964 }
1965 health_code_update();
1966
1967 memset(&reply, 0, sizeof(reply));
1968
1969 vstream = viewer_stream_get_by_id(be64toh(request.stream_id));
1970 if (!vstream) {
1971 /*
1972 * The metadata stream can be closed by a CLOSE command
1973 * just before we attach. It can also be closed by
1974 * per-pid tracing during tracing. Therefore, it is
1975 * possible that we cannot find this viewer stream.
1976 * Reply back to the client with an error if we cannot
1977 * find it.
1978 */
1979 DBG("Client requested metadata of unknown stream id %" PRIu64,
1980 (uint64_t) be64toh(request.stream_id));
1981 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1982 goto send_reply;
1983 }
1984 pthread_mutex_lock(&vstream->stream->lock);
1985 if (!vstream->stream->is_metadata) {
1986 ERR("Invalid metadata stream");
1987 goto error;
1988 }
1989
1990 if (vstream->metadata_sent >= vstream->stream->metadata_received) {
1991 /*
1992 * The live viewers expect to receive a NO_NEW_METADATA
1993 * status before a stream disappears, otherwise they abort the
1994 * entire live connection when receiving an error status.
1995 *
1996 * Clear feature resets the metadata_sent to 0 until the
1997 * same metadata is received again.
1998 */
1999 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
2000 /*
2001 * The live viewer considers a closed 0 byte metadata stream as
2002 * an error.
2003 */
2004 if (vstream->metadata_sent > 0) {
2005 vstream->stream->no_new_metadata_notified = true;
2006 if (vstream->stream->closed) {
2007 /* Release ownership for the viewer metadata stream. */
2008 viewer_stream_put(vstream);
2009 }
2010 }
2011 goto send_reply;
2012 }
2013
2014 if (vstream->stream->trace_chunk &&
2015 !lttng_trace_chunk_ids_equal(
2016 conn->viewer_session->current_trace_chunk,
2017 vstream->stream->trace_chunk)) {
2018 /* A rotation has occurred on the relay stream. */
2019 DBG("Metadata relay stream and viewer chunk ids differ");
2020
2021 ret = viewer_session_set_trace_chunk_copy(
2022 conn->viewer_session,
2023 vstream->stream->trace_chunk);
2024 if (ret) {
2025 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
2026 goto send_reply;
2027 }
2028 }
2029
2030 if (conn->viewer_session->current_trace_chunk &&
2031 conn->viewer_session->current_trace_chunk !=
2032 vstream->stream_file.trace_chunk) {
2033 bool acquired_reference;
2034
2035 DBG("Viewer session and viewer stream chunk differ: "
2036 "vsession chunk %p vstream chunk %p",
2037 conn->viewer_session->current_trace_chunk,
2038 vstream->stream_file.trace_chunk);
2039 lttng_trace_chunk_put(vstream->stream_file.trace_chunk);
2040 acquired_reference = lttng_trace_chunk_get(conn->viewer_session->current_trace_chunk);
2041 assert(acquired_reference);
2042 vstream->stream_file.trace_chunk =
2043 conn->viewer_session->current_trace_chunk;
2044 viewer_stream_close_files(vstream);
2045 }
2046
2047 len = vstream->stream->metadata_received - vstream->metadata_sent;
2048
2049 if (!vstream->stream_file.trace_chunk) {
2050 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
2051 len = 0;
2052 goto send_reply;
2053 } else if (vstream->stream_file.trace_chunk &&
2054 !vstream->stream_file.handle && len > 0) {
2055 /*
2056 * Either this is the first time the metadata file is read, or a
2057 * rotation of the corresponding relay stream has occurred.
2058 */
2059 struct fs_handle *fs_handle;
2060 char file_path[LTTNG_PATH_MAX];
2061 enum lttng_trace_chunk_status status;
2062 struct relay_stream *rstream = vstream->stream;
2063
2064 ret = utils_stream_file_path(rstream->path_name,
2065 rstream->channel_name, rstream->tracefile_size,
2066 vstream->current_tracefile_id, NULL, file_path,
2067 sizeof(file_path));
2068 if (ret < 0) {
2069 goto error;
2070 }
2071
2072 /*
2073 * It is possible the the metadata file we are trying to open is
2074 * missing if the stream has been closed (application exits with
2075 * per-pid buffers) and a clear command has been performed.
2076 */
2077 status = lttng_trace_chunk_open_fs_handle(
2078 vstream->stream_file.trace_chunk,
2079 file_path, O_RDONLY, 0, &fs_handle, true);
2080 if (status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2081 if (status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE) {
2082 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
2083 len = 0;
2084 if (vstream->stream->closed) {
2085 viewer_stream_put(vstream);
2086 }
2087 goto send_reply;
2088 }
2089 PERROR("Failed to open metadata file for viewer stream");
2090 goto error;
2091 }
2092 vstream->stream_file.handle = fs_handle;
2093
2094 if (vstream->metadata_sent != 0) {
2095 /*
2096 * The client does not expect to receive any metadata
2097 * it has received and metadata files in successive
2098 * chunks must be a strict superset of one another.
2099 *
2100 * Skip the first `metadata_sent` bytes to ensure
2101 * they are not sent a second time to the client.
2102 *
2103 * Baring a block layer error or an internal error,
2104 * this seek should not fail as
2105 * `vstream->stream->metadata_received` is reset when
2106 * a relay stream is rotated. If this is reached, it is
2107 * safe to assume that
2108 * `metadata_received` > `metadata_sent`.
2109 */
2110 const off_t seek_ret = fs_handle_seek(fs_handle,
2111 vstream->metadata_sent, SEEK_SET);
2112
2113 if (seek_ret < 0) {
2114 PERROR("Failed to seek metadata viewer stream file to `sent` position: pos = %" PRId64,
2115 vstream->metadata_sent);
2116 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
2117 goto send_reply;
2118 }
2119 }
2120 }
2121
2122 reply.len = htobe64(len);
2123 data = zmalloc(len);
2124 if (!data) {
2125 PERROR("viewer metadata zmalloc");
2126 goto error;
2127 }
2128
2129 fd = fs_handle_get_fd(vstream->stream_file.handle);
2130 if (fd < 0) {
2131 ERR("Failed to restore viewer stream file system handle");
2132 goto error;
2133 }
2134 read_len = lttng_read(fd, data, len);
2135 fs_handle_put_fd(vstream->stream_file.handle);
2136 fd = -1;
2137 if (read_len < len) {
2138 if (read_len < 0) {
2139 PERROR("Failed to read metadata file");
2140 goto error;
2141 } else {
2142 /*
2143 * A clear has been performed which prevents the relay
2144 * from sending `len` bytes of metadata.
2145 *
2146 * It is important not to send any metadata if we
2147 * couldn't read all the available metadata in one shot:
2148 * sending partial metadata can cause the client to
2149 * attempt to parse an incomplete (incoherent) metadata
2150 * stream, which would result in an error.
2151 */
2152 const off_t seek_ret = fs_handle_seek(
2153 vstream->stream_file.handle, -read_len,
2154 SEEK_CUR);
2155
2156 DBG("Failed to read metadata: requested = %" PRIu64 ", got = %zd",
2157 len, read_len);
2158 read_len = 0;
2159 len = 0;
2160 if (seek_ret < 0) {
2161 PERROR("Failed to restore metadata file position after partial read");
2162 ret = -1;
2163 goto error;
2164 }
2165 }
2166 }
2167 vstream->metadata_sent += read_len;
2168 reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
2169
2170 goto send_reply;
2171
2172 error:
2173 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
2174
2175 send_reply:
2176 health_code_update();
2177 if (vstream) {
2178 pthread_mutex_unlock(&vstream->stream->lock);
2179 }
2180 ret = send_response(conn->sock, &reply, sizeof(reply));
2181 if (ret < 0) {
2182 goto end_free;
2183 }
2184 health_code_update();
2185
2186 if (len > 0) {
2187 ret = send_response(conn->sock, data, len);
2188 if (ret < 0) {
2189 goto end_free;
2190 }
2191 }
2192
2193 DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
2194 (uint64_t) be64toh(request.stream_id));
2195
2196 DBG("Metadata sent");
2197
2198 end_free:
2199 free(data);
2200 end:
2201 if (vstream) {
2202 viewer_stream_put(vstream);
2203 }
2204 return ret;
2205 }
2206
2207 /*
2208 * Create a viewer session.
2209 *
2210 * Return 0 on success or else a negative value.
2211 */
2212 static
2213 int viewer_create_session(struct relay_connection *conn)
2214 {
2215 int ret;
2216 struct lttng_viewer_create_session_response resp;
2217
2218 DBG("Viewer create session received");
2219
2220 memset(&resp, 0, sizeof(resp));
2221 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
2222 conn->viewer_session = viewer_session_create();
2223 if (!conn->viewer_session) {
2224 ERR("Allocation viewer session");
2225 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
2226 goto send_reply;
2227 }
2228
2229 send_reply:
2230 health_code_update();
2231 ret = send_response(conn->sock, &resp, sizeof(resp));
2232 if (ret < 0) {
2233 goto end;
2234 }
2235 health_code_update();
2236 ret = 0;
2237
2238 end:
2239 return ret;
2240 }
2241
2242 /*
2243 * Detach a viewer session.
2244 *
2245 * Return 0 on success or else a negative value.
2246 */
2247 static
2248 int viewer_detach_session(struct relay_connection *conn)
2249 {
2250 int ret;
2251 struct lttng_viewer_detach_session_response response;
2252 struct lttng_viewer_detach_session_request request;
2253 struct relay_session *session = NULL;
2254 uint64_t viewer_session_to_close;
2255
2256 DBG("Viewer detach session received");
2257
2258 assert(conn);
2259
2260 health_code_update();
2261
2262 /* Receive the request from the connected client. */
2263 ret = recv_request(conn->sock, &request, sizeof(request));
2264 if (ret < 0) {
2265 goto end;
2266 }
2267 viewer_session_to_close = be64toh(request.session_id);
2268
2269 if (!conn->viewer_session) {
2270 DBG("Client trying to detach before creating a live viewer session");
2271 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
2272 goto send_reply;
2273 }
2274
2275 health_code_update();
2276
2277 memset(&response, 0, sizeof(response));
2278 DBG("Detaching from session ID %" PRIu64, viewer_session_to_close);
2279
2280 session = session_get_by_id(be64toh(request.session_id));
2281 if (!session) {
2282 DBG("Relay session %" PRIu64 " not found",
2283 (uint64_t) be64toh(request.session_id));
2284 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_UNK);
2285 goto send_reply;
2286 }
2287
2288 ret = viewer_session_is_attached(conn->viewer_session, session);
2289 if (ret != 1) {
2290 DBG("Not attached to this session");
2291 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
2292 goto send_reply_put;
2293 }
2294
2295 viewer_session_close_one_session(conn->viewer_session, session);
2296 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_OK);
2297 DBG("Session %" PRIu64 " detached.", viewer_session_to_close);
2298
2299 send_reply_put:
2300 session_put(session);
2301
2302 send_reply:
2303 health_code_update();
2304 ret = send_response(conn->sock, &response, sizeof(response));
2305 if (ret < 0) {
2306 goto end;
2307 }
2308 health_code_update();
2309 ret = 0;
2310
2311 end:
2312 return ret;
2313 }
2314
2315 /*
2316 * live_relay_unknown_command: send -1 if received unknown command
2317 */
2318 static
2319 void live_relay_unknown_command(struct relay_connection *conn)
2320 {
2321 struct lttcomm_relayd_generic_reply reply;
2322
2323 memset(&reply, 0, sizeof(reply));
2324 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2325 (void) send_response(conn->sock, &reply, sizeof(reply));
2326 }
2327
2328 /*
2329 * Process the commands received on the control socket
2330 */
2331 static
2332 int process_control(struct lttng_viewer_cmd *recv_hdr,
2333 struct relay_connection *conn)
2334 {
2335 int ret = 0;
2336 uint32_t msg_value;
2337
2338 msg_value = be32toh(recv_hdr->cmd);
2339
2340 /*
2341 * Make sure we've done the version check before any command other then a
2342 * new client connection.
2343 */
2344 if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
2345 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
2346 ret = -1;
2347 goto end;
2348 }
2349
2350 switch (msg_value) {
2351 case LTTNG_VIEWER_CONNECT:
2352 ret = viewer_connect(conn);
2353 break;
2354 case LTTNG_VIEWER_LIST_SESSIONS:
2355 ret = viewer_list_sessions(conn);
2356 break;
2357 case LTTNG_VIEWER_ATTACH_SESSION:
2358 ret = viewer_attach_session(conn);
2359 break;
2360 case LTTNG_VIEWER_GET_NEXT_INDEX:
2361 ret = viewer_get_next_index(conn);
2362 break;
2363 case LTTNG_VIEWER_GET_PACKET:
2364 ret = viewer_get_packet(conn);
2365 break;
2366 case LTTNG_VIEWER_GET_METADATA:
2367 ret = viewer_get_metadata(conn);
2368 break;
2369 case LTTNG_VIEWER_GET_NEW_STREAMS:
2370 ret = viewer_get_new_streams(conn);
2371 break;
2372 case LTTNG_VIEWER_CREATE_SESSION:
2373 ret = viewer_create_session(conn);
2374 break;
2375 case LTTNG_VIEWER_DETACH_SESSION:
2376 ret = viewer_detach_session(conn);
2377 break;
2378 default:
2379 ERR("Received unknown viewer command (%u)",
2380 be32toh(recv_hdr->cmd));
2381 live_relay_unknown_command(conn);
2382 ret = -1;
2383 goto end;
2384 }
2385
2386 end:
2387 return ret;
2388 }
2389
2390 static
2391 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
2392 {
2393 int ret;
2394
2395 (void) lttng_poll_del(events, pollfd);
2396
2397 ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker, &pollfd, 1,
2398 fd_tracker_util_close_fd, NULL);
2399 if (ret < 0) {
2400 ERR("Closing pollfd %d", pollfd);
2401 }
2402 }
2403
2404 /*
2405 * This thread does the actual work
2406 */
2407 static
2408 void *thread_worker(void *data)
2409 {
2410 int ret, err = -1;
2411 uint32_t nb_fd;
2412 struct lttng_poll_event events;
2413 struct lttng_ht *viewer_connections_ht;
2414 struct lttng_ht_iter iter;
2415 struct lttng_viewer_cmd recv_hdr;
2416 struct relay_connection *destroy_conn;
2417
2418 DBG("[thread] Live viewer relay worker started");
2419
2420 rcu_register_thread();
2421
2422 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
2423
2424 if (testpoint(relayd_thread_live_worker)) {
2425 goto error_testpoint;
2426 }
2427
2428 /* table of connections indexed on socket */
2429 viewer_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2430 if (!viewer_connections_ht) {
2431 goto viewer_connections_ht_error;
2432 }
2433
2434 ret = create_named_thread_poll_set(&events, 2,
2435 "Live viewer worker thread epoll");
2436 if (ret < 0) {
2437 goto error_poll_create;
2438 }
2439
2440 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
2441 if (ret < 0) {
2442 goto error;
2443 }
2444
2445 restart:
2446 while (1) {
2447 int i;
2448
2449 health_code_update();
2450
2451 /* Infinite blocking call, waiting for transmission */
2452 DBG3("Relayd live viewer worker thread polling...");
2453 health_poll_entry();
2454 ret = lttng_poll_wait(&events, -1);
2455 health_poll_exit();
2456 if (ret < 0) {
2457 /*
2458 * Restart interrupted system call.
2459 */
2460 if (errno == EINTR) {
2461 goto restart;
2462 }
2463 goto error;
2464 }
2465
2466 nb_fd = ret;
2467
2468 /*
2469 * Process control. The control connection is prioritised so we don't
2470 * starve it with high throughput tracing data on the data
2471 * connection.
2472 */
2473 for (i = 0; i < nb_fd; i++) {
2474 /* Fetch once the poll data */
2475 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2476 int pollfd = LTTNG_POLL_GETFD(&events, i);
2477
2478 health_code_update();
2479
2480 /* Thread quit pipe has been closed. Killing thread. */
2481 ret = check_thread_quit_pipe(pollfd, revents);
2482 if (ret) {
2483 err = 0;
2484 goto exit;
2485 }
2486
2487 /* Inspect the relay conn pipe for new connection. */
2488 if (pollfd == live_conn_pipe[0]) {
2489 if (revents & LPOLLIN) {
2490 struct relay_connection *conn;
2491
2492 ret = lttng_read(live_conn_pipe[0],
2493 &conn, sizeof(conn));
2494 if (ret < 0) {
2495 goto error;
2496 }
2497 ret = lttng_poll_add(&events,
2498 conn->sock->fd,
2499 LPOLLIN | LPOLLRDHUP);
2500 if (ret) {
2501 ERR("Failed to add new live connection file descriptor to poll set");
2502 goto error;
2503 }
2504 connection_ht_add(viewer_connections_ht, conn);
2505 DBG("Connection socket %d added to poll", conn->sock->fd);
2506 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2507 ERR("Relay live pipe error");
2508 goto error;
2509 } else {
2510 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2511 goto error;
2512 }
2513 } else {
2514 /* Connection activity. */
2515 struct relay_connection *conn;
2516
2517 conn = connection_get_by_sock(viewer_connections_ht, pollfd);
2518 if (!conn) {
2519 continue;
2520 }
2521
2522 if (revents & LPOLLIN) {
2523 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
2524 sizeof(recv_hdr), 0);
2525 if (ret <= 0) {
2526 /* Connection closed. */
2527 cleanup_connection_pollfd(&events, pollfd);
2528 /* Put "create" ownership reference. */
2529 connection_put(conn);
2530 DBG("Viewer control conn closed with %d", pollfd);
2531 } else {
2532 ret = process_control(&recv_hdr, conn);
2533 if (ret < 0) {
2534 /* Clear the session on error. */
2535 cleanup_connection_pollfd(&events, pollfd);
2536 /* Put "create" ownership reference. */
2537 connection_put(conn);
2538 DBG("Viewer connection closed with %d", pollfd);
2539 }
2540 }
2541 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2542 cleanup_connection_pollfd(&events, pollfd);
2543 /* Put "create" ownership reference. */
2544 connection_put(conn);
2545 } else {
2546 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2547 connection_put(conn);
2548 goto error;
2549 }
2550 /* Put local "get_by_sock" reference. */
2551 connection_put(conn);
2552 }
2553 }
2554 }
2555
2556 exit:
2557 error:
2558 (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
2559
2560 /* Cleanup remaining connection object. */
2561 rcu_read_lock();
2562 cds_lfht_for_each_entry(viewer_connections_ht->ht, &iter.iter,
2563 destroy_conn,
2564 sock_n.node) {
2565 health_code_update();
2566 connection_put(destroy_conn);
2567 }
2568 rcu_read_unlock();
2569 error_poll_create:
2570 lttng_ht_destroy(viewer_connections_ht);
2571 viewer_connections_ht_error:
2572 /* Close relay conn pipes */
2573 (void) fd_tracker_util_pipe_close(the_fd_tracker, live_conn_pipe);
2574 if (err) {
2575 DBG("Viewer worker thread exited with error");
2576 }
2577 DBG("Viewer worker thread cleanup complete");
2578 error_testpoint:
2579 if (err) {
2580 health_error();
2581 ERR("Health error occurred in %s", __func__);
2582 }
2583 health_unregister(health_relayd);
2584 if (lttng_relay_stop_threads()) {
2585 ERR("Error stopping threads");
2586 }
2587 rcu_unregister_thread();
2588 return NULL;
2589 }
2590
2591 /*
2592 * Create the relay command pipe to wake thread_manage_apps.
2593 * Closed in cleanup().
2594 */
2595 static int create_conn_pipe(void)
2596 {
2597 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
2598 "Live connection pipe", live_conn_pipe);
2599 }
2600
2601 int relayd_live_join(void)
2602 {
2603 int ret, retval = 0;
2604 void *status;
2605
2606 ret = pthread_join(live_listener_thread, &status);
2607 if (ret) {
2608 errno = ret;
2609 PERROR("pthread_join live listener");
2610 retval = -1;
2611 }
2612
2613 ret = pthread_join(live_worker_thread, &status);
2614 if (ret) {
2615 errno = ret;
2616 PERROR("pthread_join live worker");
2617 retval = -1;
2618 }
2619
2620 ret = pthread_join(live_dispatcher_thread, &status);
2621 if (ret) {
2622 errno = ret;
2623 PERROR("pthread_join live dispatcher");
2624 retval = -1;
2625 }
2626
2627 cleanup_relayd_live();
2628
2629 return retval;
2630 }
2631
2632 /*
2633 * main
2634 */
2635 int relayd_live_create(struct lttng_uri *uri)
2636 {
2637 int ret = 0, retval = 0;
2638 void *status;
2639 int is_root;
2640
2641 if (!uri) {
2642 retval = -1;
2643 goto exit_init_data;
2644 }
2645 live_uri = uri;
2646
2647 /* Check if daemon is UID = 0 */
2648 is_root = !getuid();
2649
2650 if (!is_root) {
2651 if (live_uri->port < 1024) {
2652 ERR("Need to be root to use ports < 1024");
2653 retval = -1;
2654 goto exit_init_data;
2655 }
2656 }
2657
2658 /* Setup the thread apps communication pipe. */
2659 if (create_conn_pipe()) {
2660 retval = -1;
2661 goto exit_init_data;
2662 }
2663
2664 /* Init relay command queue. */
2665 cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2666
2667 /* Set up max poll set size */
2668 if (lttng_poll_set_max_size()) {
2669 retval = -1;
2670 goto exit_init_data;
2671 }
2672
2673 /* Setup the dispatcher thread */
2674 ret = pthread_create(&live_dispatcher_thread, default_pthread_attr(),
2675 thread_dispatcher, (void *) NULL);
2676 if (ret) {
2677 errno = ret;
2678 PERROR("pthread_create viewer dispatcher");
2679 retval = -1;
2680 goto exit_dispatcher_thread;
2681 }
2682
2683 /* Setup the worker thread */
2684 ret = pthread_create(&live_worker_thread, default_pthread_attr(),
2685 thread_worker, NULL);
2686 if (ret) {
2687 errno = ret;
2688 PERROR("pthread_create viewer worker");
2689 retval = -1;
2690 goto exit_worker_thread;
2691 }
2692
2693 /* Setup the listener thread */
2694 ret = pthread_create(&live_listener_thread, default_pthread_attr(),
2695 thread_listener, (void *) NULL);
2696 if (ret) {
2697 errno = ret;
2698 PERROR("pthread_create viewer listener");
2699 retval = -1;
2700 goto exit_listener_thread;
2701 }
2702
2703 /*
2704 * All OK, started all threads.
2705 */
2706 return retval;
2707
2708 /*
2709 * Join on the live_listener_thread should anything be added after
2710 * the live_listener thread's creation.
2711 */
2712
2713 exit_listener_thread:
2714
2715 ret = pthread_join(live_worker_thread, &status);
2716 if (ret) {
2717 errno = ret;
2718 PERROR("pthread_join live worker");
2719 retval = -1;
2720 }
2721 exit_worker_thread:
2722
2723 ret = pthread_join(live_dispatcher_thread, &status);
2724 if (ret) {
2725 errno = ret;
2726 PERROR("pthread_join live dispatcher");
2727 retval = -1;
2728 }
2729 exit_dispatcher_thread:
2730
2731 exit_init_data:
2732 cleanup_relayd_live();
2733
2734 return retval;
2735 }
This page took 0.16553 seconds and 5 git commands to generate.