f28b9c66fa90fb89301287b8af9910d0aac7f5f0
[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_viewer_stream send_stream;
196 struct lttng_ht_iter iter;
197 struct relay_viewer_stream *vstream;
198
199 rcu_read_lock();
200
201 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
202 stream_n.node) {
203 struct ctf_trace *ctf_trace;
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 *session,
272 struct lttng_trace_chunk *viewer_trace_chunk,
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
283 assert(session);
284 ASSERT_LOCKED(session->lock);
285
286 if (!viewer_trace_chunk) {
287 ERR("Internal error: viewer session associated with session \"%s\" has a NULL trace chunk",
288 session->session_name);
289 ret = -1;
290 goto error;
291 }
292
293 if (session->connection_closed) {
294 *closed = true;
295 }
296
297 /*
298 * Create viewer streams for relay streams that are ready to be
299 * used for a the given session id only.
300 */
301 rcu_read_lock();
302 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
303 node.node) {
304 bool trace_has_metadata_stream = false;
305 struct relay_stream *stream;
306
307 health_code_update();
308
309 if (!ctf_trace_get(ctf_trace)) {
310 continue;
311 }
312
313 /*
314 * Iterate over all the streams of the trace to see if we have a
315 * metadata stream.
316 */
317 cds_list_for_each_entry_rcu(
318 stream, &ctf_trace->stream_list, stream_node)
319 {
320 if (stream->is_metadata) {
321 trace_has_metadata_stream = true;
322 break;
323 }
324 }
325
326 /*
327 * If there is no metadata stream in this trace at the moment
328 * and we never sent one to the viewer, skip the trace. We
329 * accept that the viewer will not see this trace at all.
330 */
331 if (!trace_has_metadata_stream &&
332 !ctf_trace->metadata_stream_sent_to_viewer) {
333 ctf_trace_put(ctf_trace);
334 continue;
335 }
336
337 cds_list_for_each_entry_rcu(stream, &ctf_trace->stream_list, stream_node) {
338 struct relay_viewer_stream *vstream;
339
340 if (!stream_get(stream)) {
341 continue;
342 }
343 /*
344 * stream published is protected by the session lock.
345 */
346 if (!stream->published) {
347 goto next;
348 }
349 vstream = viewer_stream_get_by_id(stream->stream_handle);
350 if (!vstream) {
351 /*
352 * Save that we sent the metadata stream to the
353 * viewer. So that we know what trace the viewer
354 * is aware of.
355 */
356 if (stream->is_metadata) {
357 ctf_trace->metadata_stream_sent_to_viewer =
358 true;
359 }
360 vstream = viewer_stream_create(stream,
361 viewer_trace_chunk, seek_t);
362 if (!vstream) {
363 ret = -1;
364 ctf_trace_put(ctf_trace);
365 stream_put(stream);
366 goto error_unlock;
367 }
368
369 if (nb_created) {
370 /* Update number of created stream counter. */
371 (*nb_created)++;
372 }
373 /*
374 * Ensure a self-reference is preserved even
375 * after we have put our local reference.
376 */
377 if (!viewer_stream_get(vstream)) {
378 ERR("Unable to get self-reference on viewer stream, logic error.");
379 abort();
380 }
381 } else {
382 if (!vstream->sent_flag && nb_unsent) {
383 /* Update number of unsent stream counter. */
384 (*nb_unsent)++;
385 }
386 }
387 /* Update number of total stream counter. */
388 if (nb_total) {
389 if (stream->is_metadata) {
390 if (!stream->closed ||
391 stream->metadata_received > vstream->metadata_sent) {
392 (*nb_total)++;
393 }
394 } else {
395 if (!stream->closed ||
396 !(((int64_t) (stream->prev_data_seq - stream->last_net_seq_num)) >= 0)) {
397
398 (*nb_total)++;
399 }
400 }
401 }
402 /* Put local reference. */
403 viewer_stream_put(vstream);
404 next:
405 stream_put(stream);
406 }
407 ctf_trace_put(ctf_trace);
408 }
409
410 ret = 0;
411
412 error_unlock:
413 rcu_read_unlock();
414 error:
415 return ret;
416 }
417
418 int relayd_live_stop(void)
419 {
420 /* Stop dispatch thread */
421 CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
422 futex_nto1_wake(&viewer_conn_queue.futex);
423 return 0;
424 }
425
426 /*
427 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
428 */
429 static
430 int create_named_thread_poll_set(struct lttng_poll_event *events,
431 int size, const char *name)
432 {
433 int ret;
434
435 if (events == NULL || size == 0) {
436 ret = -1;
437 goto error;
438 }
439
440 ret = fd_tracker_util_poll_create(the_fd_tracker,
441 name, events, 1, LTTNG_CLOEXEC);
442
443 /* Add quit pipe */
444 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
445 if (ret < 0) {
446 goto error;
447 }
448
449 return 0;
450
451 error:
452 return ret;
453 }
454
455 /*
456 * Check if the thread quit pipe was triggered.
457 *
458 * Return 1 if it was triggered else 0;
459 */
460 static
461 int check_thread_quit_pipe(int fd, uint32_t events)
462 {
463 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
464 return 1;
465 }
466
467 return 0;
468 }
469
470 static
471 int create_sock(void *data, int *out_fd)
472 {
473 int ret;
474 struct lttcomm_sock *sock = data;
475
476 ret = lttcomm_create_sock(sock);
477 if (ret < 0) {
478 goto end;
479 }
480
481 *out_fd = sock->fd;
482 end:
483 return ret;
484 }
485
486 static
487 int close_sock(void *data, int *in_fd)
488 {
489 struct lttcomm_sock *sock = data;
490
491 return sock->ops->close(sock);
492 }
493
494 static int accept_sock(void *data, int *out_fd)
495 {
496 int ret = 0;
497 /* Socks is an array of in_sock, out_sock. */
498 struct lttcomm_sock **socks = data;
499 struct lttcomm_sock *in_sock = socks[0];
500
501 socks[1] = in_sock->ops->accept(in_sock);
502 if (!socks[1]) {
503 ret = -1;
504 goto end;
505 }
506 *out_fd = socks[1]->fd;
507 end:
508 return ret;
509 }
510
511 static
512 struct lttcomm_sock *accept_live_sock(struct lttcomm_sock *listening_sock,
513 const char *name)
514 {
515 int out_fd, ret;
516 struct lttcomm_sock *socks[2] = { listening_sock, NULL };
517 struct lttcomm_sock *new_sock = NULL;
518
519 ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker, &out_fd,
520 (const char **) &name, 1, accept_sock, &socks);
521 if (ret) {
522 goto end;
523 }
524 new_sock = socks[1];
525 DBG("%s accepted, socket %d", name, new_sock->fd);
526 end:
527 return new_sock;
528 }
529
530 /*
531 * Create and init socket from uri.
532 */
533 static
534 struct lttcomm_sock *init_socket(struct lttng_uri *uri, const char *name)
535 {
536 int ret, sock_fd;
537 struct lttcomm_sock *sock = NULL;
538 char uri_str[LTTNG_PATH_MAX];
539 char *formated_name = NULL;
540
541 sock = lttcomm_alloc_sock_from_uri(uri);
542 if (sock == NULL) {
543 ERR("Allocating socket");
544 goto error;
545 }
546
547 /*
548 * Don't fail to create the socket if the name can't be built as it is
549 * only used for debugging purposes.
550 */
551 ret = uri_to_str_url(uri, uri_str, sizeof(uri_str));
552 uri_str[sizeof(uri_str) - 1] = '\0';
553 if (ret >= 0) {
554 ret = asprintf(&formated_name, "%s socket @ %s", name,
555 uri_str);
556 if (ret < 0) {
557 formated_name = NULL;
558 }
559 }
560
561 ret = fd_tracker_open_unsuspendable_fd(the_fd_tracker, &sock_fd,
562 (const char **) (formated_name ? &formated_name : NULL),
563 1, create_sock, sock);
564 free(formated_name);
565 DBG("Listening on %s socket %d", name, sock->fd);
566
567 ret = sock->ops->bind(sock);
568 if (ret < 0) {
569 PERROR("Failed to bind lttng-live socket");
570 goto error;
571 }
572
573 ret = sock->ops->listen(sock, -1);
574 if (ret < 0) {
575 goto error;
576
577 }
578
579 return sock;
580
581 error:
582 if (sock) {
583 lttcomm_destroy_sock(sock);
584 }
585 return NULL;
586 }
587
588 /*
589 * This thread manages the listening for new connections on the network
590 */
591 static
592 void *thread_listener(void *data)
593 {
594 int i, ret, pollfd, err = -1;
595 uint32_t revents, nb_fd;
596 struct lttng_poll_event events;
597 struct lttcomm_sock *live_control_sock;
598
599 DBG("[thread] Relay live listener started");
600
601 rcu_register_thread();
602 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
603
604 health_code_update();
605
606 live_control_sock = init_socket(live_uri, "Live listener");
607 if (!live_control_sock) {
608 goto error_sock_control;
609 }
610
611 /* Pass 2 as size here for the thread quit pipe and control sockets. */
612 ret = create_named_thread_poll_set(&events, 2,
613 "Live listener thread epoll");
614 if (ret < 0) {
615 goto error_create_poll;
616 }
617
618 /* Add the control socket */
619 ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
620 if (ret < 0) {
621 goto error_poll_add;
622 }
623
624 lttng_relay_notify_ready();
625
626 if (testpoint(relayd_thread_live_listener)) {
627 goto error_testpoint;
628 }
629
630 while (1) {
631 health_code_update();
632
633 DBG("Listener accepting live viewers connections");
634
635 restart:
636 health_poll_entry();
637 ret = lttng_poll_wait(&events, -1);
638 health_poll_exit();
639 if (ret < 0) {
640 /*
641 * Restart interrupted system call.
642 */
643 if (errno == EINTR) {
644 goto restart;
645 }
646 goto error;
647 }
648 nb_fd = ret;
649
650 DBG("Relay new viewer connection received");
651 for (i = 0; i < nb_fd; i++) {
652 health_code_update();
653
654 /* Fetch once the poll data */
655 revents = LTTNG_POLL_GETEV(&events, i);
656 pollfd = LTTNG_POLL_GETFD(&events, i);
657
658 /* Thread quit pipe has been closed. Killing thread. */
659 ret = check_thread_quit_pipe(pollfd, revents);
660 if (ret) {
661 err = 0;
662 goto exit;
663 }
664
665 if (revents & LPOLLIN) {
666 /*
667 * A new connection is requested, therefore a
668 * viewer connection is allocated in this
669 * thread, enqueued to a global queue and
670 * dequeued (and freed) in the worker thread.
671 */
672 int val = 1;
673 struct relay_connection *new_conn;
674 struct lttcomm_sock *newsock;
675
676 newsock = accept_live_sock(live_control_sock,
677 "Live socket to client");
678 if (!newsock) {
679 PERROR("accepting control sock");
680 goto error;
681 }
682 DBG("Relay viewer connection accepted socket %d", newsock->fd);
683
684 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
685 sizeof(val));
686 if (ret < 0) {
687 PERROR("setsockopt inet");
688 lttcomm_destroy_sock(newsock);
689 goto error;
690 }
691 new_conn = connection_create(newsock, RELAY_CONNECTION_UNKNOWN);
692 if (!new_conn) {
693 lttcomm_destroy_sock(newsock);
694 goto error;
695 }
696 /* Ownership assumed by the connection. */
697 newsock = NULL;
698
699 /* Enqueue request for the dispatcher thread. */
700 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
701 &new_conn->qnode);
702
703 /*
704 * Wake the dispatch queue futex.
705 * Implicit memory barrier with the
706 * exchange in cds_wfcq_enqueue.
707 */
708 futex_nto1_wake(&viewer_conn_queue.futex);
709 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
710 ERR("socket poll error");
711 goto error;
712 } else {
713 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
714 goto error;
715 }
716 }
717 }
718
719 exit:
720 error:
721 error_poll_add:
722 error_testpoint:
723 (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
724 error_create_poll:
725 if (live_control_sock->fd >= 0) {
726 int sock_fd = live_control_sock->fd;
727
728 ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker,
729 &sock_fd, 1, close_sock,
730 live_control_sock);
731 if (ret) {
732 PERROR("close");
733 }
734 live_control_sock->fd = -1;
735 }
736 lttcomm_destroy_sock(live_control_sock);
737 error_sock_control:
738 if (err) {
739 health_error();
740 DBG("Live viewer listener thread exited with error");
741 }
742 health_unregister(health_relayd);
743 rcu_unregister_thread();
744 DBG("Live viewer listener thread cleanup complete");
745 if (lttng_relay_stop_threads()) {
746 ERR("Error stopping threads");
747 }
748 return NULL;
749 }
750
751 /*
752 * This thread manages the dispatching of the requests to worker threads
753 */
754 static
755 void *thread_dispatcher(void *data)
756 {
757 int err = -1;
758 ssize_t ret;
759 struct cds_wfcq_node *node;
760 struct relay_connection *conn = NULL;
761
762 DBG("[thread] Live viewer relay dispatcher started");
763
764 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
765
766 if (testpoint(relayd_thread_live_dispatcher)) {
767 goto error_testpoint;
768 }
769
770 health_code_update();
771
772 for (;;) {
773 health_code_update();
774
775 /* Atomically prepare the queue futex */
776 futex_nto1_prepare(&viewer_conn_queue.futex);
777
778 if (CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
779 break;
780 }
781
782 do {
783 health_code_update();
784
785 /* Dequeue commands */
786 node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
787 &viewer_conn_queue.tail);
788 if (node == NULL) {
789 DBG("Woken up but nothing in the live-viewer "
790 "relay command queue");
791 /* Continue thread execution */
792 break;
793 }
794 conn = caa_container_of(node, struct relay_connection, qnode);
795 DBG("Dispatching viewer request waiting on sock %d",
796 conn->sock->fd);
797
798 /*
799 * Inform worker thread of the new request. This
800 * call is blocking so we can be assured that
801 * the data will be read at some point in time
802 * or wait to the end of the world :)
803 */
804 ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
805 if (ret < 0) {
806 PERROR("write conn pipe");
807 connection_put(conn);
808 goto error;
809 }
810 } while (node != NULL);
811
812 /* Futex wait on queue. Blocking call on futex() */
813 health_poll_entry();
814 futex_nto1_wait(&viewer_conn_queue.futex);
815 health_poll_exit();
816 }
817
818 /* Normal exit, no error */
819 err = 0;
820
821 error:
822 error_testpoint:
823 if (err) {
824 health_error();
825 ERR("Health error occurred in %s", __func__);
826 }
827 health_unregister(health_relayd);
828 DBG("Live viewer dispatch thread dying");
829 if (lttng_relay_stop_threads()) {
830 ERR("Error stopping threads");
831 }
832 return NULL;
833 }
834
835 /*
836 * Establish connection with the viewer and check the versions.
837 *
838 * Return 0 on success or else negative value.
839 */
840 static
841 int viewer_connect(struct relay_connection *conn)
842 {
843 int ret;
844 struct lttng_viewer_connect reply, msg;
845
846 conn->version_check_done = 1;
847
848 health_code_update();
849
850 DBG("Viewer is establishing a connection to the relayd.");
851
852 ret = recv_request(conn->sock, &msg, sizeof(msg));
853 if (ret < 0) {
854 goto end;
855 }
856
857 health_code_update();
858
859 memset(&reply, 0, sizeof(reply));
860 reply.major = RELAYD_VERSION_COMM_MAJOR;
861 reply.minor = RELAYD_VERSION_COMM_MINOR;
862
863 /* Major versions must be the same */
864 if (reply.major != be32toh(msg.major)) {
865 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
866 reply.major, be32toh(msg.major));
867 ret = -1;
868 goto end;
869 }
870
871 conn->major = reply.major;
872 /* We adapt to the lowest compatible version */
873 if (reply.minor <= be32toh(msg.minor)) {
874 conn->minor = reply.minor;
875 } else {
876 conn->minor = be32toh(msg.minor);
877 }
878
879 if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
880 conn->type = RELAY_VIEWER_COMMAND;
881 } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
882 conn->type = RELAY_VIEWER_NOTIFICATION;
883 } else {
884 ERR("Unknown connection type : %u", be32toh(msg.type));
885 ret = -1;
886 goto end;
887 }
888
889 reply.major = htobe32(reply.major);
890 reply.minor = htobe32(reply.minor);
891 if (conn->type == RELAY_VIEWER_COMMAND) {
892 /*
893 * Increment outside of htobe64 macro, because the argument can
894 * be used more than once within the macro, and thus the
895 * operation may be undefined.
896 */
897 pthread_mutex_lock(&last_relay_viewer_session_id_lock);
898 last_relay_viewer_session_id++;
899 pthread_mutex_unlock(&last_relay_viewer_session_id_lock);
900 reply.viewer_session_id = htobe64(last_relay_viewer_session_id);
901 }
902
903 health_code_update();
904
905 ret = send_response(conn->sock, &reply, sizeof(reply));
906 if (ret < 0) {
907 goto end;
908 }
909
910 health_code_update();
911
912 DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
913 ret = 0;
914
915 end:
916 return ret;
917 }
918
919 /*
920 * Send the viewer the list of current sessions.
921 * We need to create a copy of the hash table content because otherwise
922 * we cannot assume the number of entries stays the same between getting
923 * the number of HT elements and iteration over the HT.
924 *
925 * Return 0 on success or else a negative value.
926 */
927 static
928 int viewer_list_sessions(struct relay_connection *conn)
929 {
930 int ret = 0;
931 struct lttng_viewer_list_sessions session_list;
932 struct lttng_ht_iter iter;
933 struct relay_session *session;
934 struct lttng_viewer_session *send_session_buf = NULL;
935 uint32_t buf_count = SESSION_BUF_DEFAULT_COUNT;
936 uint32_t count = 0;
937
938 DBG("List sessions received");
939
940 send_session_buf = zmalloc(SESSION_BUF_DEFAULT_COUNT * sizeof(*send_session_buf));
941 if (!send_session_buf) {
942 return -1;
943 }
944
945 rcu_read_lock();
946 cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, session,
947 session_n.node) {
948 struct lttng_viewer_session *send_session;
949
950 health_code_update();
951
952 pthread_mutex_lock(&session->lock);
953 if (session->connection_closed) {
954 /* Skip closed session */
955 goto next_session;
956 }
957 if (!session->current_trace_chunk) {
958 /*
959 * Skip un-attachable session. It is either
960 * being destroyed or has not had a trace
961 * chunk created against it yet.
962 */
963 goto next_session;
964 }
965
966 if (count >= buf_count) {
967 struct lttng_viewer_session *newbuf;
968 uint32_t new_buf_count = buf_count << 1;
969
970 newbuf = realloc(send_session_buf,
971 new_buf_count * sizeof(*send_session_buf));
972 if (!newbuf) {
973 ret = -1;
974 goto break_loop;
975 }
976 send_session_buf = newbuf;
977 buf_count = new_buf_count;
978 }
979 send_session = &send_session_buf[count];
980 if (lttng_strncpy(send_session->session_name,
981 session->session_name,
982 sizeof(send_session->session_name))) {
983 ret = -1;
984 goto break_loop;
985 }
986 if (lttng_strncpy(send_session->hostname, session->hostname,
987 sizeof(send_session->hostname))) {
988 ret = -1;
989 goto break_loop;
990 }
991 send_session->id = htobe64(session->id);
992 send_session->live_timer = htobe32(session->live_timer);
993 if (session->viewer_attached) {
994 send_session->clients = htobe32(1);
995 } else {
996 send_session->clients = htobe32(0);
997 }
998 send_session->streams = htobe32(session->stream_count);
999 count++;
1000 next_session:
1001 pthread_mutex_unlock(&session->lock);
1002 continue;
1003 break_loop:
1004 pthread_mutex_unlock(&session->lock);
1005 break;
1006 }
1007 rcu_read_unlock();
1008 if (ret < 0) {
1009 goto end_free;
1010 }
1011
1012 session_list.sessions_count = htobe32(count);
1013
1014 health_code_update();
1015
1016 ret = send_response(conn->sock, &session_list, sizeof(session_list));
1017 if (ret < 0) {
1018 goto end_free;
1019 }
1020
1021 health_code_update();
1022
1023 ret = send_response(conn->sock, send_session_buf,
1024 count * sizeof(*send_session_buf));
1025 if (ret < 0) {
1026 goto end_free;
1027 }
1028 health_code_update();
1029
1030 ret = 0;
1031 end_free:
1032 free(send_session_buf);
1033 return ret;
1034 }
1035
1036 /*
1037 * Send the viewer the list of current streams.
1038 */
1039 static
1040 int viewer_get_new_streams(struct relay_connection *conn)
1041 {
1042 int ret, send_streams = 0;
1043 uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0, nb_total = 0;
1044 struct lttng_viewer_new_streams_request request;
1045 struct lttng_viewer_new_streams_response response;
1046 struct relay_session *session = NULL;
1047 uint64_t session_id;
1048 bool closed = false;
1049
1050 assert(conn);
1051
1052 DBG("Get new streams received");
1053
1054 health_code_update();
1055
1056 /* Receive the request from the connected client. */
1057 ret = recv_request(conn->sock, &request, sizeof(request));
1058 if (ret < 0) {
1059 goto error;
1060 }
1061 session_id = be64toh(request.session_id);
1062
1063 health_code_update();
1064
1065 memset(&response, 0, sizeof(response));
1066
1067 session = session_get_by_id(session_id);
1068 if (!session) {
1069 DBG("Relay session %" PRIu64 " not found", session_id);
1070 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
1071 goto send_reply;
1072 }
1073
1074 if (!viewer_session_is_attached(conn->viewer_session, session)) {
1075 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
1076 goto send_reply;
1077 }
1078
1079 pthread_mutex_lock(&session->lock);
1080 ret = make_viewer_streams(session,
1081 conn->viewer_session->current_trace_chunk,
1082 LTTNG_VIEWER_SEEK_LAST, &nb_total, &nb_unsent,
1083 &nb_created, &closed);
1084 if (ret < 0) {
1085 goto error_unlock_session;
1086 }
1087 send_streams = 1;
1088 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
1089
1090 /* Only send back the newly created streams with the unsent ones. */
1091 nb_streams = nb_created + nb_unsent;
1092 response.streams_count = htobe32(nb_streams);
1093
1094 /*
1095 * If the session is closed, HUP when there are no more streams
1096 * with data.
1097 */
1098 if (closed && nb_total == 0) {
1099 send_streams = 0;
1100 response.streams_count = 0;
1101 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
1102 goto send_reply_unlock;
1103 }
1104 send_reply_unlock:
1105 pthread_mutex_unlock(&session->lock);
1106
1107 send_reply:
1108 health_code_update();
1109 ret = send_response(conn->sock, &response, sizeof(response));
1110 if (ret < 0) {
1111 goto end_put_session;
1112 }
1113 health_code_update();
1114
1115 /*
1116 * Unknown or empty session, just return gracefully, the viewer
1117 * knows what is happening.
1118 */
1119 if (!send_streams || !nb_streams) {
1120 ret = 0;
1121 goto end_put_session;
1122 }
1123
1124 /*
1125 * Send stream and *DON'T* ignore the sent flag so every viewer
1126 * streams that were not sent from that point will be sent to
1127 * the viewer.
1128 */
1129 ret = send_viewer_streams(conn->sock, session_id, 0);
1130 if (ret < 0) {
1131 goto end_put_session;
1132 }
1133
1134 end_put_session:
1135 if (session) {
1136 session_put(session);
1137 }
1138 error:
1139 return ret;
1140 error_unlock_session:
1141 pthread_mutex_unlock(&session->lock);
1142 session_put(session);
1143 return ret;
1144 }
1145
1146 /*
1147 * Send the viewer the list of current sessions.
1148 */
1149 static
1150 int viewer_attach_session(struct relay_connection *conn)
1151 {
1152 int send_streams = 0;
1153 ssize_t ret;
1154 uint32_t nb_streams = 0;
1155 enum lttng_viewer_seek seek_type;
1156 struct lttng_viewer_attach_session_request request;
1157 struct lttng_viewer_attach_session_response response;
1158 struct relay_session *session = NULL;
1159 enum lttng_viewer_attach_return_code viewer_attach_status;
1160 bool closed = false;
1161 uint64_t session_id;
1162
1163 assert(conn);
1164
1165 health_code_update();
1166
1167 /* Receive the request from the connected client. */
1168 ret = recv_request(conn->sock, &request, sizeof(request));
1169 if (ret < 0) {
1170 goto error;
1171 }
1172
1173 session_id = be64toh(request.session_id);
1174 health_code_update();
1175
1176 memset(&response, 0, sizeof(response));
1177
1178 if (!conn->viewer_session) {
1179 DBG("Client trying to attach before creating a live viewer session");
1180 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1181 goto send_reply;
1182 }
1183
1184 session = session_get_by_id(session_id);
1185 if (!session) {
1186 DBG("Relay session %" PRIu64 " not found", session_id);
1187 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1188 goto send_reply;
1189 }
1190 DBG("Attach session ID %" PRIu64 " received", session_id);
1191
1192 pthread_mutex_lock(&session->lock);
1193 if (!session->current_trace_chunk) {
1194 /*
1195 * Session is either being destroyed or it never had a trace
1196 * chunk created against it.
1197 */
1198 DBG("Session requested by live client has no current trace chunk, returning unknown session");
1199 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1200 goto send_reply;
1201 }
1202 if (session->live_timer == 0) {
1203 DBG("Not live session");
1204 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1205 goto send_reply;
1206 }
1207
1208 send_streams = 1;
1209 viewer_attach_status = viewer_session_attach(conn->viewer_session,
1210 session);
1211 if (viewer_attach_status != LTTNG_VIEWER_ATTACH_OK) {
1212 response.status = htobe32(viewer_attach_status);
1213 goto send_reply;
1214 }
1215
1216 switch (be32toh(request.seek)) {
1217 case LTTNG_VIEWER_SEEK_BEGINNING:
1218 case LTTNG_VIEWER_SEEK_LAST:
1219 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1220 seek_type = be32toh(request.seek);
1221 break;
1222 default:
1223 ERR("Wrong seek parameter");
1224 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1225 send_streams = 0;
1226 goto send_reply;
1227 }
1228
1229 ret = make_viewer_streams(session,
1230 conn->viewer_session->current_trace_chunk, seek_type,
1231 &nb_streams, NULL, NULL, &closed);
1232 if (ret < 0) {
1233 goto end_put_session;
1234 }
1235 pthread_mutex_unlock(&session->lock);
1236 session_put(session);
1237 session = NULL;
1238
1239 response.streams_count = htobe32(nb_streams);
1240 /*
1241 * If the session is closed when the viewer is attaching, it
1242 * means some of the streams may have been concurrently removed,
1243 * so we don't allow the viewer to attach, even if there are
1244 * streams available.
1245 */
1246 if (closed) {
1247 send_streams = 0;
1248 response.streams_count = 0;
1249 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1250 goto send_reply;
1251 }
1252
1253 send_reply:
1254 health_code_update();
1255 ret = send_response(conn->sock, &response, sizeof(response));
1256 if (ret < 0) {
1257 goto end_put_session;
1258 }
1259 health_code_update();
1260
1261 /*
1262 * Unknown or empty session, just return gracefully, the viewer
1263 * knows what is happening.
1264 */
1265 if (!send_streams || !nb_streams) {
1266 ret = 0;
1267 goto end_put_session;
1268 }
1269
1270 /* Send stream and ignore the sent flag. */
1271 ret = send_viewer_streams(conn->sock, session_id, 1);
1272 if (ret < 0) {
1273 goto end_put_session;
1274 }
1275
1276 end_put_session:
1277 if (session) {
1278 pthread_mutex_unlock(&session->lock);
1279 session_put(session);
1280 }
1281 error:
1282 return ret;
1283 }
1284
1285 /*
1286 * Open the index file if needed for the given vstream.
1287 *
1288 * If an index file is successfully opened, the vstream will set it as its
1289 * current index file.
1290 *
1291 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1292 *
1293 * Called with rstream lock held.
1294 */
1295 static int try_open_index(struct relay_viewer_stream *vstream,
1296 struct relay_stream *rstream)
1297 {
1298 int ret = 0;
1299 const uint32_t connection_major = rstream->trace->session->major;
1300 const uint32_t connection_minor = rstream->trace->session->minor;
1301 enum lttng_trace_chunk_status chunk_status;
1302
1303 if (vstream->index_file) {
1304 goto end;
1305 }
1306
1307 /*
1308 * First time, we open the index file and at least one index is ready.
1309 */
1310 if (rstream->index_received_seqcount == 0) {
1311 ret = -ENOENT;
1312 goto end;
1313 }
1314 chunk_status = lttng_index_file_create_from_trace_chunk_read_only(
1315 vstream->stream_file.trace_chunk, rstream->path_name,
1316 rstream->channel_name, rstream->tracefile_size,
1317 vstream->current_tracefile_id,
1318 lttng_to_index_major(connection_major, connection_minor),
1319 lttng_to_index_minor(connection_major, connection_minor),
1320 true, &vstream->index_file);
1321 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1322 if (chunk_status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE) {
1323 ret = -ENOENT;
1324 } else {
1325 ret = -1;
1326 }
1327 }
1328
1329 end:
1330 return ret;
1331 }
1332
1333 /*
1334 * Check the status of the index for the given stream. This function
1335 * updates the index structure if needed and can put (close) the vstream
1336 * in the HUP situation.
1337 *
1338 * Return 0 means that we can proceed with the index. A value of 1 means
1339 * that the index has been updated and is ready to be sent to the
1340 * client. A negative value indicates an error that can't be handled.
1341 *
1342 * Called with rstream lock held.
1343 */
1344 static int check_index_status(struct relay_viewer_stream *vstream,
1345 struct relay_stream *rstream, struct ctf_trace *trace,
1346 struct lttng_viewer_index *index)
1347 {
1348 int ret;
1349
1350 DBG("Check index status: index_received_seqcount %" PRIu64 " "
1351 "index_sent_seqcount %" PRIu64 " "
1352 "for stream %" PRIu64,
1353 rstream->index_received_seqcount,
1354 vstream->index_sent_seqcount,
1355 vstream->stream->stream_handle);
1356 if ((trace->session->connection_closed || rstream->closed)
1357 && rstream->index_received_seqcount
1358 == vstream->index_sent_seqcount) {
1359 /*
1360 * Last index sent and session connection or relay
1361 * stream are closed.
1362 */
1363 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1364 goto hup;
1365 } else if (rstream->beacon_ts_end != -1ULL &&
1366 (rstream->index_received_seqcount == 0 ||
1367 (vstream->index_sent_seqcount != 0 &&
1368 rstream->index_received_seqcount
1369 <= vstream->index_sent_seqcount))) {
1370 /*
1371 * We've received a synchronization beacon and the last index
1372 * available has been sent, the index for now is inactive.
1373 *
1374 * In this case, we have received a beacon which allows us to
1375 * inform the client of a time interval during which we can
1376 * guarantee that there are no events to read (and never will
1377 * be).
1378 *
1379 * The sent seqcount can grow higher than receive seqcount on
1380 * clear because the rotation performed by clear will push
1381 * the index_sent_seqcount ahead (see
1382 * viewer_stream_sync_tracefile_array_tail) and skip over
1383 * packet sequence numbers.
1384 */
1385 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1386 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1387 index->stream_id = htobe64(rstream->ctf_stream_id);
1388 DBG("Check index status: inactive with beacon, for stream %" PRIu64,
1389 vstream->stream->stream_handle);
1390 goto index_ready;
1391 } else if (rstream->index_received_seqcount == 0 ||
1392 (vstream->index_sent_seqcount != 0 &&
1393 rstream->index_received_seqcount
1394 <= vstream->index_sent_seqcount)) {
1395 /*
1396 * This checks whether received <= sent seqcount. In
1397 * this case, we have not received a beacon. Therefore,
1398 * we can only ask the client to retry later.
1399 *
1400 * The sent seqcount can grow higher than receive seqcount on
1401 * clear because the rotation performed by clear will push
1402 * the index_sent_seqcount ahead (see
1403 * viewer_stream_sync_tracefile_array_tail) and skip over
1404 * packet sequence numbers.
1405 */
1406 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1407 DBG("Check index status: retry for stream %" PRIu64,
1408 vstream->stream->stream_handle);
1409 goto index_ready;
1410 } else if (!tracefile_array_seq_in_file(rstream->tfa,
1411 vstream->current_tracefile_id,
1412 vstream->index_sent_seqcount)) {
1413 /*
1414 * The next index we want to send cannot be read either
1415 * because we need to perform a rotation, or due to
1416 * the producer having overwritten its trace file.
1417 */
1418 DBG("Viewer stream %" PRIu64 " rotation",
1419 vstream->stream->stream_handle);
1420 ret = viewer_stream_rotate(vstream);
1421 if (ret == 1) {
1422 /* EOF across entire stream. */
1423 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1424 goto hup;
1425 }
1426 /*
1427 * If we have been pushed due to overwrite, it
1428 * necessarily means there is data that can be read in
1429 * the stream. If we rotated because we reached the end
1430 * of a tracefile, it means the following tracefile
1431 * needs to contain at least one index, else we would
1432 * have already returned LTTNG_VIEWER_INDEX_RETRY to the
1433 * viewer. The updated index_sent_seqcount needs to
1434 * point to a readable index entry now.
1435 *
1436 * In the case where we "rotate" on a single file, we
1437 * can end up in a case where the requested index is
1438 * still unavailable.
1439 */
1440 if (rstream->tracefile_count == 1 &&
1441 !tracefile_array_seq_in_file(
1442 rstream->tfa,
1443 vstream->current_tracefile_id,
1444 vstream->index_sent_seqcount)) {
1445 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1446 DBG("Check index status: retry: "
1447 "tracefile array sequence number %" PRIu64
1448 " not in file for stream %" PRIu64,
1449 vstream->index_sent_seqcount,
1450 vstream->stream->stream_handle);
1451 goto index_ready;
1452 }
1453 assert(tracefile_array_seq_in_file(rstream->tfa,
1454 vstream->current_tracefile_id,
1455 vstream->index_sent_seqcount));
1456 }
1457 /* ret == 0 means successful so we continue. */
1458 ret = 0;
1459 return ret;
1460
1461 hup:
1462 viewer_stream_put(vstream);
1463 index_ready:
1464 return 1;
1465 }
1466
1467 /*
1468 * Send the next index for a stream.
1469 *
1470 * Return 0 on success or else a negative value.
1471 */
1472 static
1473 int viewer_get_next_index(struct relay_connection *conn)
1474 {
1475 int ret;
1476 struct lttng_viewer_get_next_index request_index;
1477 struct lttng_viewer_index viewer_index;
1478 struct ctf_packet_index packet_index;
1479 struct relay_viewer_stream *vstream = NULL;
1480 struct relay_stream *rstream = NULL;
1481 struct ctf_trace *ctf_trace = NULL;
1482 struct relay_viewer_stream *metadata_viewer_stream = NULL;
1483
1484 assert(conn);
1485
1486 DBG("Viewer get next index");
1487
1488 memset(&viewer_index, 0, sizeof(viewer_index));
1489 health_code_update();
1490
1491 ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1492 if (ret < 0) {
1493 goto end;
1494 }
1495 health_code_update();
1496
1497 vstream = viewer_stream_get_by_id(be64toh(request_index.stream_id));
1498 if (!vstream) {
1499 DBG("Client requested index of unknown stream id %" PRIu64,
1500 (uint64_t) be64toh(request_index.stream_id));
1501 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1502 goto send_reply;
1503 }
1504
1505 /* Use back. ref. Protected by refcounts. */
1506 rstream = vstream->stream;
1507 ctf_trace = rstream->trace;
1508
1509 /* metadata_viewer_stream may be NULL. */
1510 metadata_viewer_stream =
1511 ctf_trace_get_viewer_metadata_stream(ctf_trace);
1512
1513 pthread_mutex_lock(&rstream->lock);
1514
1515 /*
1516 * The viewer should not ask for index on metadata stream.
1517 */
1518 if (rstream->is_metadata) {
1519 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1520 goto send_reply;
1521 }
1522
1523 if (rstream->ongoing_rotation.is_set) {
1524 /* Rotation is ongoing, try again later. */
1525 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1526 goto send_reply;
1527 }
1528
1529 if (rstream->trace->session->ongoing_rotation) {
1530 /* Rotation is ongoing, try again later. */
1531 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1532 goto send_reply;
1533 }
1534
1535 if (rstream->trace_chunk) {
1536 uint64_t rchunk_id, vchunk_id;
1537
1538 /*
1539 * If the relay stream is not yet closed, ensure the viewer
1540 * chunk matches the relay chunk after clear.
1541 */
1542 if (lttng_trace_chunk_get_id(rstream->trace_chunk,
1543 &rchunk_id) != LTTNG_TRACE_CHUNK_STATUS_OK) {
1544 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1545 goto send_reply;
1546 }
1547 if (lttng_trace_chunk_get_id(
1548 conn->viewer_session->current_trace_chunk,
1549 &vchunk_id) != LTTNG_TRACE_CHUNK_STATUS_OK) {
1550 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1551 goto send_reply;
1552 }
1553
1554 if (rchunk_id != vchunk_id) {
1555 DBG("Relay and viewer chunk ids differ: "
1556 "rchunk_id %" PRIu64 " vchunk_id %" PRIu64,
1557 rchunk_id, vchunk_id);
1558
1559 lttng_trace_chunk_put(
1560 conn->viewer_session->current_trace_chunk);
1561 conn->viewer_session->current_trace_chunk = NULL;
1562 ret = viewer_session_set_trace_chunk_copy(
1563 conn->viewer_session,
1564 rstream->trace_chunk);
1565 if (ret) {
1566 viewer_index.status =
1567 htobe32(LTTNG_VIEWER_INDEX_ERR);
1568 goto send_reply;
1569 }
1570 }
1571 }
1572 if (conn->viewer_session->current_trace_chunk !=
1573 vstream->stream_file.trace_chunk) {
1574 bool acquired_reference;
1575
1576 DBG("Viewer session and viewer stream chunk differ: "
1577 "vsession chunk %p vstream chunk %p",
1578 conn->viewer_session->current_trace_chunk,
1579 vstream->stream_file.trace_chunk);
1580 lttng_trace_chunk_put(vstream->stream_file.trace_chunk);
1581 acquired_reference = lttng_trace_chunk_get(conn->viewer_session->current_trace_chunk);
1582 assert(acquired_reference);
1583 vstream->stream_file.trace_chunk =
1584 conn->viewer_session->current_trace_chunk;
1585 viewer_stream_sync_tracefile_array_tail(vstream);
1586 viewer_stream_close_files(vstream);
1587 }
1588
1589 ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1590 if (ret < 0) {
1591 goto error_put;
1592 } else if (ret == 1) {
1593 /*
1594 * We have no index to send and check_index_status has populated
1595 * viewer_index's status.
1596 */
1597 goto send_reply;
1598 }
1599 /* At this point, ret is 0 thus we will be able to read the index. */
1600 assert(!ret);
1601
1602 /* Try to open an index if one is needed for that stream. */
1603 ret = try_open_index(vstream, rstream);
1604 if (ret == -ENOENT) {
1605 if (rstream->closed) {
1606 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1607 goto send_reply;
1608 } else {
1609 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1610 goto send_reply;
1611 }
1612 }
1613 if (ret < 0) {
1614 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1615 goto send_reply;
1616 }
1617
1618 /*
1619 * vstream->stream_fd may be NULL if it has been closed by
1620 * tracefile rotation, or if we are at the beginning of the
1621 * stream. We open the data stream file here to protect against
1622 * overwrite caused by tracefile rotation (in association with
1623 * unlink performed before overwrite).
1624 */
1625 if (!vstream->stream_file.handle) {
1626 char file_path[LTTNG_PATH_MAX];
1627 enum lttng_trace_chunk_status status;
1628 struct fs_handle *fs_handle;
1629
1630 ret = utils_stream_file_path(rstream->path_name,
1631 rstream->channel_name, rstream->tracefile_size,
1632 vstream->current_tracefile_id, NULL, file_path,
1633 sizeof(file_path));
1634 if (ret < 0) {
1635 goto error_put;
1636 }
1637
1638 /*
1639 * It is possible the the file we are trying to open is
1640 * missing if the stream has been closed (application exits with
1641 * per-pid buffers) and a clear command has been performed.
1642 */
1643 status = lttng_trace_chunk_open_fs_handle(
1644 vstream->stream_file.trace_chunk,
1645 file_path, O_RDONLY, 0, &fs_handle, true);
1646 if (status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1647 if (status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE &&
1648 rstream->closed) {
1649 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1650 goto send_reply;
1651 }
1652 PERROR("Failed to open trace file for viewer stream");
1653 goto error_put;
1654 }
1655 vstream->stream_file.handle = fs_handle;
1656 }
1657
1658 ret = check_new_streams(conn);
1659 if (ret < 0) {
1660 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1661 goto send_reply;
1662 } else if (ret == 1) {
1663 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1664 }
1665
1666 ret = lttng_index_file_read(vstream->index_file, &packet_index);
1667 if (ret) {
1668 ERR("Relay error reading index file");
1669 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1670 goto send_reply;
1671 } else {
1672 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1673 vstream->index_sent_seqcount++;
1674 }
1675
1676 /*
1677 * Indexes are stored in big endian, no need to switch before sending.
1678 */
1679 DBG("Sending viewer index for stream %" PRIu64 " offset %" PRIu64,
1680 rstream->stream_handle,
1681 (uint64_t) be64toh(packet_index.offset));
1682 viewer_index.offset = packet_index.offset;
1683 viewer_index.packet_size = packet_index.packet_size;
1684 viewer_index.content_size = packet_index.content_size;
1685 viewer_index.timestamp_begin = packet_index.timestamp_begin;
1686 viewer_index.timestamp_end = packet_index.timestamp_end;
1687 viewer_index.events_discarded = packet_index.events_discarded;
1688 viewer_index.stream_id = packet_index.stream_id;
1689
1690 send_reply:
1691 if (rstream) {
1692 pthread_mutex_unlock(&rstream->lock);
1693 }
1694
1695 if (metadata_viewer_stream) {
1696 pthread_mutex_lock(&metadata_viewer_stream->stream->lock);
1697 DBG("get next index metadata check: recv %" PRIu64
1698 " sent %" PRIu64,
1699 metadata_viewer_stream->stream->metadata_received,
1700 metadata_viewer_stream->metadata_sent);
1701 if (!metadata_viewer_stream->stream->metadata_received ||
1702 metadata_viewer_stream->stream->metadata_received >
1703 metadata_viewer_stream->metadata_sent) {
1704 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1705 }
1706 pthread_mutex_unlock(&metadata_viewer_stream->stream->lock);
1707 }
1708
1709 viewer_index.flags = htobe32(viewer_index.flags);
1710 health_code_update();
1711
1712 ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1713 if (ret < 0) {
1714 goto end;
1715 }
1716 health_code_update();
1717
1718 if (vstream) {
1719 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1720 vstream->index_sent_seqcount,
1721 vstream->stream->stream_handle);
1722 }
1723 end:
1724 if (metadata_viewer_stream) {
1725 viewer_stream_put(metadata_viewer_stream);
1726 }
1727 if (vstream) {
1728 viewer_stream_put(vstream);
1729 }
1730 return ret;
1731
1732 error_put:
1733 pthread_mutex_unlock(&rstream->lock);
1734 if (metadata_viewer_stream) {
1735 viewer_stream_put(metadata_viewer_stream);
1736 }
1737 viewer_stream_put(vstream);
1738 return ret;
1739 }
1740
1741 /*
1742 * Send the next index for a stream
1743 *
1744 * Return 0 on success or else a negative value.
1745 */
1746 static
1747 int viewer_get_packet(struct relay_connection *conn)
1748 {
1749 int ret;
1750 off_t lseek_ret;
1751 char *reply = NULL;
1752 struct lttng_viewer_get_packet get_packet_info;
1753 struct lttng_viewer_trace_packet reply_header;
1754 struct relay_viewer_stream *vstream = NULL;
1755 uint32_t reply_size = sizeof(reply_header);
1756 uint32_t packet_data_len = 0;
1757 ssize_t read_len;
1758 uint64_t stream_id;
1759
1760 DBG2("Relay get data packet");
1761
1762 health_code_update();
1763
1764 ret = recv_request(conn->sock, &get_packet_info,
1765 sizeof(get_packet_info));
1766 if (ret < 0) {
1767 goto end;
1768 }
1769 health_code_update();
1770
1771 /* From this point on, the error label can be reached. */
1772 memset(&reply_header, 0, sizeof(reply_header));
1773 stream_id = (uint64_t) be64toh(get_packet_info.stream_id);
1774
1775 vstream = viewer_stream_get_by_id(stream_id);
1776 if (!vstream) {
1777 DBG("Client requested packet of unknown stream id %" PRIu64,
1778 stream_id);
1779 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1780 goto send_reply_nolock;
1781 } else {
1782 packet_data_len = be32toh(get_packet_info.len);
1783 reply_size += packet_data_len;
1784 }
1785
1786 reply = zmalloc(reply_size);
1787 if (!reply) {
1788 PERROR("packet reply zmalloc");
1789 reply_size = sizeof(reply_header);
1790 goto error;
1791 }
1792
1793 pthread_mutex_lock(&vstream->stream->lock);
1794 lseek_ret = fs_handle_seek(vstream->stream_file.handle,
1795 be64toh(get_packet_info.offset), SEEK_SET);
1796 if (lseek_ret < 0) {
1797 PERROR("Failed to seek file system handle of viewer stream %" PRIu64
1798 " to offset %" PRIu64,
1799 stream_id,
1800 (uint64_t) be64toh(get_packet_info.offset));
1801 goto error;
1802 }
1803 read_len = fs_handle_read(vstream->stream_file.handle,
1804 reply + sizeof(reply_header), packet_data_len);
1805 if (read_len < packet_data_len) {
1806 PERROR("Failed to read from file system handle of viewer stream id %" PRIu64
1807 ", offset: %" PRIu64,
1808 stream_id,
1809 (uint64_t) be64toh(get_packet_info.offset));
1810 goto error;
1811 }
1812 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1813 reply_header.len = htobe32(packet_data_len);
1814 goto send_reply;
1815
1816 error:
1817 reply_header.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1818
1819 send_reply:
1820 if (vstream) {
1821 pthread_mutex_unlock(&vstream->stream->lock);
1822 }
1823 send_reply_nolock:
1824
1825 health_code_update();
1826
1827 if (reply) {
1828 memcpy(reply, &reply_header, sizeof(reply_header));
1829 ret = send_response(conn->sock, reply, reply_size);
1830 } else {
1831 /* No reply to send. */
1832 ret = send_response(conn->sock, &reply_header,
1833 reply_size);
1834 }
1835
1836 health_code_update();
1837 if (ret < 0) {
1838 PERROR("sendmsg of packet data failed");
1839 goto end_free;
1840 }
1841
1842 DBG("Sent %u bytes for stream %" PRIu64, reply_size, stream_id);
1843
1844 end_free:
1845 free(reply);
1846 end:
1847 if (vstream) {
1848 viewer_stream_put(vstream);
1849 }
1850 return ret;
1851 }
1852
1853 /*
1854 * Send the session's metadata
1855 *
1856 * Return 0 on success else a negative value.
1857 */
1858 static
1859 int viewer_get_metadata(struct relay_connection *conn)
1860 {
1861 int ret = 0;
1862 int fd = -1;
1863 ssize_t read_len;
1864 uint64_t len = 0;
1865 char *data = NULL;
1866 struct lttng_viewer_get_metadata request;
1867 struct lttng_viewer_metadata_packet reply;
1868 struct relay_viewer_stream *vstream = NULL;
1869
1870 assert(conn);
1871
1872 DBG("Relay get metadata");
1873
1874 health_code_update();
1875
1876 ret = recv_request(conn->sock, &request, sizeof(request));
1877 if (ret < 0) {
1878 goto end;
1879 }
1880 health_code_update();
1881
1882 memset(&reply, 0, sizeof(reply));
1883
1884 vstream = viewer_stream_get_by_id(be64toh(request.stream_id));
1885 if (!vstream) {
1886 /*
1887 * The metadata stream can be closed by a CLOSE command
1888 * just before we attach. It can also be closed by
1889 * per-pid tracing during tracing. Therefore, it is
1890 * possible that we cannot find this viewer stream.
1891 * Reply back to the client with an error if we cannot
1892 * find it.
1893 */
1894 DBG("Client requested metadata of unknown stream id %" PRIu64,
1895 (uint64_t) be64toh(request.stream_id));
1896 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1897 goto send_reply;
1898 }
1899 pthread_mutex_lock(&vstream->stream->lock);
1900 if (!vstream->stream->is_metadata) {
1901 ERR("Invalid metadata stream");
1902 goto error;
1903 }
1904
1905 if (vstream->metadata_sent >= vstream->stream->metadata_received) {
1906 /*
1907 * The live viewers expect to receive a NO_NEW_METADATA
1908 * status before a stream disappears, otherwise they abort the
1909 * entire live connection when receiving an error status.
1910 *
1911 * Clear feature resets the metadata_sent to 0 until the
1912 * same metadata is received again.
1913 */
1914 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1915 /*
1916 * The live viewer considers a closed 0 byte metadata stream as
1917 * an error.
1918 */
1919 if (vstream->metadata_sent > 0) {
1920 vstream->stream->no_new_metadata_notified = true;
1921 if (vstream->stream->closed) {
1922 /* Release ownership for the viewer metadata stream. */
1923 viewer_stream_put(vstream);
1924 }
1925 }
1926 goto send_reply;
1927 }
1928
1929 len = vstream->stream->metadata_received - vstream->metadata_sent;
1930
1931 /* first time, we open the metadata file */
1932 if (!vstream->stream_file.handle) {
1933 struct fs_handle *fs_handle;
1934 char file_path[LTTNG_PATH_MAX];
1935 enum lttng_trace_chunk_status status;
1936 struct relay_stream *rstream = vstream->stream;
1937
1938 ret = utils_stream_file_path(rstream->path_name,
1939 rstream->channel_name, rstream->tracefile_size,
1940 vstream->current_tracefile_id, NULL, file_path,
1941 sizeof(file_path));
1942 if (ret < 0) {
1943 goto error;
1944 }
1945
1946 /*
1947 * It is possible the the metadata file we are trying to open is
1948 * missing if the stream has been closed (application exits with
1949 * per-pid buffers) and a clear command has been performed.
1950 */
1951 status = lttng_trace_chunk_open_fs_handle(
1952 vstream->stream_file.trace_chunk,
1953 file_path, O_RDONLY, 0, &fs_handle, true);
1954 if (status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1955 if (status == LTTNG_TRACE_CHUNK_STATUS_NO_FILE) {
1956 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1957 len = 0;
1958 if (vstream->stream->closed) {
1959 viewer_stream_put(vstream);
1960 }
1961 goto send_reply;
1962 }
1963 PERROR("Failed to open metadata file for viewer stream");
1964 goto error;
1965 }
1966 vstream->stream_file.handle = fs_handle;
1967 }
1968
1969 reply.len = htobe64(len);
1970 data = zmalloc(len);
1971 if (!data) {
1972 PERROR("viewer metadata zmalloc");
1973 goto error;
1974 }
1975
1976 fd = fs_handle_get_fd(vstream->stream_file.handle);
1977 if (fd < 0) {
1978 ERR("Failed to restore viewer stream file system handle");
1979 goto error;
1980 }
1981 read_len = lttng_read(fd, data, len);
1982 fs_handle_put_fd(vstream->stream_file.handle);
1983 fd = -1;
1984 if (read_len < len) {
1985 PERROR("Relay reading metadata file");
1986 goto error;
1987 }
1988 vstream->metadata_sent += read_len;
1989 reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1990
1991 goto send_reply;
1992
1993 error:
1994 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1995
1996 send_reply:
1997 health_code_update();
1998 if (vstream) {
1999 pthread_mutex_unlock(&vstream->stream->lock);
2000 }
2001 ret = send_response(conn->sock, &reply, sizeof(reply));
2002 if (ret < 0) {
2003 goto end_free;
2004 }
2005 health_code_update();
2006
2007 if (len > 0) {
2008 ret = send_response(conn->sock, data, len);
2009 if (ret < 0) {
2010 goto end_free;
2011 }
2012 }
2013
2014 DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
2015 (uint64_t) be64toh(request.stream_id));
2016
2017 DBG("Metadata sent");
2018
2019 end_free:
2020 free(data);
2021 end:
2022 if (vstream) {
2023 viewer_stream_put(vstream);
2024 }
2025 return ret;
2026 }
2027
2028 /*
2029 * Create a viewer session.
2030 *
2031 * Return 0 on success or else a negative value.
2032 */
2033 static
2034 int viewer_create_session(struct relay_connection *conn)
2035 {
2036 int ret;
2037 struct lttng_viewer_create_session_response resp;
2038
2039 DBG("Viewer create session received");
2040
2041 memset(&resp, 0, sizeof(resp));
2042 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
2043 conn->viewer_session = viewer_session_create();
2044 if (!conn->viewer_session) {
2045 ERR("Allocation viewer session");
2046 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
2047 goto send_reply;
2048 }
2049
2050 send_reply:
2051 health_code_update();
2052 ret = send_response(conn->sock, &resp, sizeof(resp));
2053 if (ret < 0) {
2054 goto end;
2055 }
2056 health_code_update();
2057 ret = 0;
2058
2059 end:
2060 return ret;
2061 }
2062
2063 /*
2064 * Detach a viewer session.
2065 *
2066 * Return 0 on success or else a negative value.
2067 */
2068 static
2069 int viewer_detach_session(struct relay_connection *conn)
2070 {
2071 int ret;
2072 struct lttng_viewer_detach_session_response response;
2073 struct lttng_viewer_detach_session_request request;
2074 struct relay_session *session = NULL;
2075 uint64_t viewer_session_to_close;
2076
2077 DBG("Viewer detach session received");
2078
2079 assert(conn);
2080
2081 health_code_update();
2082
2083 /* Receive the request from the connected client. */
2084 ret = recv_request(conn->sock, &request, sizeof(request));
2085 if (ret < 0) {
2086 goto end;
2087 }
2088 viewer_session_to_close = be64toh(request.session_id);
2089
2090 if (!conn->viewer_session) {
2091 DBG("Client trying to detach before creating a live viewer session");
2092 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
2093 goto send_reply;
2094 }
2095
2096 health_code_update();
2097
2098 memset(&response, 0, sizeof(response));
2099 DBG("Detaching from session ID %" PRIu64, viewer_session_to_close);
2100
2101 session = session_get_by_id(be64toh(request.session_id));
2102 if (!session) {
2103 DBG("Relay session %" PRIu64 " not found",
2104 (uint64_t) be64toh(request.session_id));
2105 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_UNK);
2106 goto send_reply;
2107 }
2108
2109 ret = viewer_session_is_attached(conn->viewer_session, session);
2110 if (ret != 1) {
2111 DBG("Not attached to this session");
2112 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_ERR);
2113 goto send_reply_put;
2114 }
2115
2116 viewer_session_close_one_session(conn->viewer_session, session);
2117 response.status = htobe32(LTTNG_VIEWER_DETACH_SESSION_OK);
2118 DBG("Session %" PRIu64 " detached.", viewer_session_to_close);
2119
2120 send_reply_put:
2121 session_put(session);
2122
2123 send_reply:
2124 health_code_update();
2125 ret = send_response(conn->sock, &response, sizeof(response));
2126 if (ret < 0) {
2127 goto end;
2128 }
2129 health_code_update();
2130 ret = 0;
2131
2132 end:
2133 return ret;
2134 }
2135
2136 /*
2137 * live_relay_unknown_command: send -1 if received unknown command
2138 */
2139 static
2140 void live_relay_unknown_command(struct relay_connection *conn)
2141 {
2142 struct lttcomm_relayd_generic_reply reply;
2143
2144 memset(&reply, 0, sizeof(reply));
2145 reply.ret_code = htobe32(LTTNG_ERR_UNK);
2146 (void) send_response(conn->sock, &reply, sizeof(reply));
2147 }
2148
2149 /*
2150 * Process the commands received on the control socket
2151 */
2152 static
2153 int process_control(struct lttng_viewer_cmd *recv_hdr,
2154 struct relay_connection *conn)
2155 {
2156 int ret = 0;
2157 uint32_t msg_value;
2158
2159 msg_value = be32toh(recv_hdr->cmd);
2160
2161 /*
2162 * Make sure we've done the version check before any command other then a
2163 * new client connection.
2164 */
2165 if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
2166 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
2167 ret = -1;
2168 goto end;
2169 }
2170
2171 switch (msg_value) {
2172 case LTTNG_VIEWER_CONNECT:
2173 ret = viewer_connect(conn);
2174 break;
2175 case LTTNG_VIEWER_LIST_SESSIONS:
2176 ret = viewer_list_sessions(conn);
2177 break;
2178 case LTTNG_VIEWER_ATTACH_SESSION:
2179 ret = viewer_attach_session(conn);
2180 break;
2181 case LTTNG_VIEWER_GET_NEXT_INDEX:
2182 ret = viewer_get_next_index(conn);
2183 break;
2184 case LTTNG_VIEWER_GET_PACKET:
2185 ret = viewer_get_packet(conn);
2186 break;
2187 case LTTNG_VIEWER_GET_METADATA:
2188 ret = viewer_get_metadata(conn);
2189 break;
2190 case LTTNG_VIEWER_GET_NEW_STREAMS:
2191 ret = viewer_get_new_streams(conn);
2192 break;
2193 case LTTNG_VIEWER_CREATE_SESSION:
2194 ret = viewer_create_session(conn);
2195 break;
2196 case LTTNG_VIEWER_DETACH_SESSION:
2197 ret = viewer_detach_session(conn);
2198 break;
2199 default:
2200 ERR("Received unknown viewer command (%u)",
2201 be32toh(recv_hdr->cmd));
2202 live_relay_unknown_command(conn);
2203 ret = -1;
2204 goto end;
2205 }
2206
2207 end:
2208 return ret;
2209 }
2210
2211 static
2212 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
2213 {
2214 int ret;
2215
2216 (void) lttng_poll_del(events, pollfd);
2217
2218 ret = fd_tracker_close_unsuspendable_fd(the_fd_tracker, &pollfd, 1,
2219 fd_tracker_util_close_fd, NULL);
2220 if (ret < 0) {
2221 ERR("Closing pollfd %d", pollfd);
2222 }
2223 }
2224
2225 /*
2226 * This thread does the actual work
2227 */
2228 static
2229 void *thread_worker(void *data)
2230 {
2231 int ret, err = -1;
2232 uint32_t nb_fd;
2233 struct lttng_poll_event events;
2234 struct lttng_ht *viewer_connections_ht;
2235 struct lttng_ht_iter iter;
2236 struct lttng_viewer_cmd recv_hdr;
2237 struct relay_connection *destroy_conn;
2238
2239 DBG("[thread] Live viewer relay worker started");
2240
2241 rcu_register_thread();
2242
2243 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
2244
2245 if (testpoint(relayd_thread_live_worker)) {
2246 goto error_testpoint;
2247 }
2248
2249 /* table of connections indexed on socket */
2250 viewer_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2251 if (!viewer_connections_ht) {
2252 goto viewer_connections_ht_error;
2253 }
2254
2255 ret = create_named_thread_poll_set(&events, 2,
2256 "Live viewer worker thread epoll");
2257 if (ret < 0) {
2258 goto error_poll_create;
2259 }
2260
2261 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
2262 if (ret < 0) {
2263 goto error;
2264 }
2265
2266 restart:
2267 while (1) {
2268 int i;
2269
2270 health_code_update();
2271
2272 /* Infinite blocking call, waiting for transmission */
2273 DBG3("Relayd live viewer worker thread polling...");
2274 health_poll_entry();
2275 ret = lttng_poll_wait(&events, -1);
2276 health_poll_exit();
2277 if (ret < 0) {
2278 /*
2279 * Restart interrupted system call.
2280 */
2281 if (errno == EINTR) {
2282 goto restart;
2283 }
2284 goto error;
2285 }
2286
2287 nb_fd = ret;
2288
2289 /*
2290 * Process control. The control connection is prioritised so we don't
2291 * starve it with high throughput tracing data on the data
2292 * connection.
2293 */
2294 for (i = 0; i < nb_fd; i++) {
2295 /* Fetch once the poll data */
2296 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
2297 int pollfd = LTTNG_POLL_GETFD(&events, i);
2298
2299 health_code_update();
2300
2301 /* Thread quit pipe has been closed. Killing thread. */
2302 ret = check_thread_quit_pipe(pollfd, revents);
2303 if (ret) {
2304 err = 0;
2305 goto exit;
2306 }
2307
2308 /* Inspect the relay conn pipe for new connection. */
2309 if (pollfd == live_conn_pipe[0]) {
2310 if (revents & LPOLLIN) {
2311 struct relay_connection *conn;
2312
2313 ret = lttng_read(live_conn_pipe[0],
2314 &conn, sizeof(conn));
2315 if (ret < 0) {
2316 goto error;
2317 }
2318 ret = lttng_poll_add(&events,
2319 conn->sock->fd,
2320 LPOLLIN | LPOLLRDHUP);
2321 if (ret) {
2322 ERR("Failed to add new live connection file descriptor to poll set");
2323 goto error;
2324 }
2325 connection_ht_add(viewer_connections_ht, conn);
2326 DBG("Connection socket %d added to poll", conn->sock->fd);
2327 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2328 ERR("Relay live pipe error");
2329 goto error;
2330 } else {
2331 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2332 goto error;
2333 }
2334 } else {
2335 /* Connection activity. */
2336 struct relay_connection *conn;
2337
2338 conn = connection_get_by_sock(viewer_connections_ht, pollfd);
2339 if (!conn) {
2340 continue;
2341 }
2342
2343 if (revents & LPOLLIN) {
2344 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
2345 sizeof(recv_hdr), 0);
2346 if (ret <= 0) {
2347 /* Connection closed. */
2348 cleanup_connection_pollfd(&events, pollfd);
2349 /* Put "create" ownership reference. */
2350 connection_put(conn);
2351 DBG("Viewer control conn closed with %d", pollfd);
2352 } else {
2353 ret = process_control(&recv_hdr, conn);
2354 if (ret < 0) {
2355 /* Clear the session on error. */
2356 cleanup_connection_pollfd(&events, pollfd);
2357 /* Put "create" ownership reference. */
2358 connection_put(conn);
2359 DBG("Viewer connection closed with %d", pollfd);
2360 }
2361 }
2362 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
2363 cleanup_connection_pollfd(&events, pollfd);
2364 /* Put "create" ownership reference. */
2365 connection_put(conn);
2366 } else {
2367 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2368 connection_put(conn);
2369 goto error;
2370 }
2371 /* Put local "get_by_sock" reference. */
2372 connection_put(conn);
2373 }
2374 }
2375 }
2376
2377 exit:
2378 error:
2379 (void) fd_tracker_util_poll_clean(the_fd_tracker, &events);
2380
2381 /* Cleanup remaining connection object. */
2382 rcu_read_lock();
2383 cds_lfht_for_each_entry(viewer_connections_ht->ht, &iter.iter,
2384 destroy_conn,
2385 sock_n.node) {
2386 health_code_update();
2387 connection_put(destroy_conn);
2388 }
2389 rcu_read_unlock();
2390 error_poll_create:
2391 lttng_ht_destroy(viewer_connections_ht);
2392 viewer_connections_ht_error:
2393 /* Close relay conn pipes */
2394 (void) fd_tracker_util_pipe_close(the_fd_tracker, live_conn_pipe);
2395 if (err) {
2396 DBG("Viewer worker thread exited with error");
2397 }
2398 DBG("Viewer worker thread cleanup complete");
2399 error_testpoint:
2400 if (err) {
2401 health_error();
2402 ERR("Health error occurred in %s", __func__);
2403 }
2404 health_unregister(health_relayd);
2405 if (lttng_relay_stop_threads()) {
2406 ERR("Error stopping threads");
2407 }
2408 rcu_unregister_thread();
2409 return NULL;
2410 }
2411
2412 /*
2413 * Create the relay command pipe to wake thread_manage_apps.
2414 * Closed in cleanup().
2415 */
2416 static int create_conn_pipe(void)
2417 {
2418 return fd_tracker_util_pipe_open_cloexec(the_fd_tracker,
2419 "Live connection pipe", live_conn_pipe);
2420 }
2421
2422 int relayd_live_join(void)
2423 {
2424 int ret, retval = 0;
2425 void *status;
2426
2427 ret = pthread_join(live_listener_thread, &status);
2428 if (ret) {
2429 errno = ret;
2430 PERROR("pthread_join live listener");
2431 retval = -1;
2432 }
2433
2434 ret = pthread_join(live_worker_thread, &status);
2435 if (ret) {
2436 errno = ret;
2437 PERROR("pthread_join live worker");
2438 retval = -1;
2439 }
2440
2441 ret = pthread_join(live_dispatcher_thread, &status);
2442 if (ret) {
2443 errno = ret;
2444 PERROR("pthread_join live dispatcher");
2445 retval = -1;
2446 }
2447
2448 cleanup_relayd_live();
2449
2450 return retval;
2451 }
2452
2453 /*
2454 * main
2455 */
2456 int relayd_live_create(struct lttng_uri *uri)
2457 {
2458 int ret = 0, retval = 0;
2459 void *status;
2460 int is_root;
2461
2462 if (!uri) {
2463 retval = -1;
2464 goto exit_init_data;
2465 }
2466 live_uri = uri;
2467
2468 /* Check if daemon is UID = 0 */
2469 is_root = !getuid();
2470
2471 if (!is_root) {
2472 if (live_uri->port < 1024) {
2473 ERR("Need to be root to use ports < 1024");
2474 retval = -1;
2475 goto exit_init_data;
2476 }
2477 }
2478
2479 /* Setup the thread apps communication pipe. */
2480 if (create_conn_pipe()) {
2481 retval = -1;
2482 goto exit_init_data;
2483 }
2484
2485 /* Init relay command queue. */
2486 cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2487
2488 /* Set up max poll set size */
2489 if (lttng_poll_set_max_size()) {
2490 retval = -1;
2491 goto exit_init_data;
2492 }
2493
2494 /* Setup the dispatcher thread */
2495 ret = pthread_create(&live_dispatcher_thread, default_pthread_attr(),
2496 thread_dispatcher, (void *) NULL);
2497 if (ret) {
2498 errno = ret;
2499 PERROR("pthread_create viewer dispatcher");
2500 retval = -1;
2501 goto exit_dispatcher_thread;
2502 }
2503
2504 /* Setup the worker thread */
2505 ret = pthread_create(&live_worker_thread, default_pthread_attr(),
2506 thread_worker, NULL);
2507 if (ret) {
2508 errno = ret;
2509 PERROR("pthread_create viewer worker");
2510 retval = -1;
2511 goto exit_worker_thread;
2512 }
2513
2514 /* Setup the listener thread */
2515 ret = pthread_create(&live_listener_thread, default_pthread_attr(),
2516 thread_listener, (void *) NULL);
2517 if (ret) {
2518 errno = ret;
2519 PERROR("pthread_create viewer listener");
2520 retval = -1;
2521 goto exit_listener_thread;
2522 }
2523
2524 /*
2525 * All OK, started all threads.
2526 */
2527 return retval;
2528
2529 /*
2530 * Join on the live_listener_thread should anything be added after
2531 * the live_listener thread's creation.
2532 */
2533
2534 exit_listener_thread:
2535
2536 ret = pthread_join(live_worker_thread, &status);
2537 if (ret) {
2538 errno = ret;
2539 PERROR("pthread_join live worker");
2540 retval = -1;
2541 }
2542 exit_worker_thread:
2543
2544 ret = pthread_join(live_dispatcher_thread, &status);
2545 if (ret) {
2546 errno = ret;
2547 PERROR("pthread_join live dispatcher");
2548 retval = -1;
2549 }
2550 exit_dispatcher_thread:
2551
2552 exit_init_data:
2553 cleanup_relayd_live();
2554
2555 return retval;
2556 }
This page took 0.134195 seconds and 3 git commands to generate.