Fix: LPOLLHUP and LPOLLERR when there is still data in pipe/socket
[lttng-tools.git] / src / bin / lttng-relayd / live.c
1 /*
2 * Copyright (C) 2013 - Julien Desfossez <jdesfossez@efficios.com>
3 * David Goulet <dgoulet@efficios.com>
4 * 2015 - Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
5 *
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
9 *
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
14 *
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
18 */
19
20 #define _GNU_SOURCE
21 #define _LGPL_SOURCE
22 #include <getopt.h>
23 #include <grp.h>
24 #include <limits.h>
25 #include <pthread.h>
26 #include <signal.h>
27 #include <stdio.h>
28 #include <stdlib.h>
29 #include <string.h>
30 #include <sys/mman.h>
31 #include <sys/mount.h>
32 #include <sys/resource.h>
33 #include <sys/socket.h>
34 #include <sys/stat.h>
35 #include <sys/types.h>
36 #include <sys/wait.h>
37 #include <inttypes.h>
38 #include <urcu/futex.h>
39 #include <urcu/uatomic.h>
40 #include <urcu/rculist.h>
41 #include <unistd.h>
42 #include <fcntl.h>
43 #include <config.h>
44
45 #include <lttng/lttng.h>
46 #include <common/common.h>
47 #include <common/compat/poll.h>
48 #include <common/compat/socket.h>
49 #include <common/compat/endian.h>
50 #include <common/defaults.h>
51 #include <common/futex.h>
52 #include <common/index/index.h>
53 #include <common/sessiond-comm/sessiond-comm.h>
54 #include <common/sessiond-comm/inet.h>
55 #include <common/sessiond-comm/relayd.h>
56 #include <common/uri.h>
57 #include <common/utils.h>
58
59 #include "cmd.h"
60 #include "live.h"
61 #include "lttng-relayd.h"
62 #include "utils.h"
63 #include "health-relayd.h"
64 #include "testpoint.h"
65 #include "viewer-stream.h"
66 #include "stream.h"
67 #include "session.h"
68 #include "ctf-trace.h"
69 #include "connection.h"
70 #include "viewer-session.h"
71
72 #define SESSION_BUF_DEFAULT_COUNT 16
73
74 static struct lttng_uri *live_uri;
75
76 /*
77 * This pipe is used to inform the worker thread that a command is queued and
78 * ready to be processed.
79 */
80 static int live_conn_pipe[2] = { -1, -1 };
81
82 /* Shared between threads */
83 static int live_dispatch_thread_exit;
84
85 static pthread_t live_listener_thread;
86 static pthread_t live_dispatcher_thread;
87 static pthread_t live_worker_thread;
88
89 /*
90 * Relay command queue.
91 *
92 * The live_thread_listener and live_thread_dispatcher communicate with this
93 * queue.
94 */
95 static struct relay_conn_queue viewer_conn_queue;
96
97 static uint64_t last_relay_viewer_session_id;
98 static pthread_mutex_t last_relay_viewer_session_id_lock =
99 PTHREAD_MUTEX_INITIALIZER;
100
101 /*
102 * Cleanup the daemon
103 */
104 static
105 void cleanup_relayd_live(void)
106 {
107 DBG("Cleaning up");
108
109 free(live_uri);
110 }
111
112 /*
113 * Receive a request buffer using a given socket, destination allocated buffer
114 * of length size.
115 *
116 * Return the size of the received message or else a negative value on error
117 * with errno being set by recvmsg() syscall.
118 */
119 static
120 ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
121 {
122 ssize_t ret;
123
124 ret = sock->ops->recvmsg(sock, buf, size, 0);
125 if (ret < 0 || ret != size) {
126 if (ret == 0) {
127 /* Orderly shutdown. Not necessary to print an error. */
128 DBG("Socket %d did an orderly shutdown", sock->fd);
129 } else {
130 ERR("Relay failed to receive request.");
131 }
132 ret = -1;
133 }
134
135 return ret;
136 }
137
138 /*
139 * Send a response buffer using a given socket, source allocated buffer of
140 * length size.
141 *
142 * Return the size of the sent message or else a negative value on error with
143 * errno being set by sendmsg() syscall.
144 */
145 static
146 ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
147 {
148 ssize_t ret;
149
150 ret = sock->ops->sendmsg(sock, buf, size, 0);
151 if (ret < 0) {
152 ERR("Relayd failed to send response.");
153 }
154
155 return ret;
156 }
157
158 /*
159 * Atomically check if new streams got added in one of the sessions attached
160 * and reset the flag to 0.
161 *
162 * Returns 1 if new streams got added, 0 if nothing changed, a negative value
163 * on error.
164 */
165 static
166 int check_new_streams(struct relay_connection *conn)
167 {
168 struct relay_session *session;
169 unsigned long current_val;
170 int ret = 0;
171
172 if (!conn->viewer_session) {
173 goto end;
174 }
175 rcu_read_lock();
176 cds_list_for_each_entry_rcu(session,
177 &conn->viewer_session->session_list,
178 viewer_session_node) {
179 if (!session_get(session)) {
180 continue;
181 }
182 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
183 ret = current_val;
184 session_put(session);
185 if (ret == 1) {
186 goto end;
187 }
188 }
189 end:
190 rcu_read_unlock();
191 return ret;
192 }
193
194 /*
195 * Send viewer streams to the given socket. The ignore_sent_flag indicates if
196 * this function should ignore the sent flag or not.
197 *
198 * Return 0 on success or else a negative value.
199 */
200 static
201 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
202 struct relay_session *session, unsigned int ignore_sent_flag)
203 {
204 ssize_t ret;
205 struct lttng_viewer_stream send_stream;
206 struct lttng_ht_iter iter;
207 struct relay_viewer_stream *vstream;
208
209 rcu_read_lock();
210
211 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
212 stream_n.node) {
213 struct ctf_trace *ctf_trace;
214
215 health_code_update();
216
217 if (!viewer_stream_get(vstream)) {
218 continue;
219 }
220
221 pthread_mutex_lock(&vstream->stream->lock);
222 /* Ignore if not the same session. */
223 if (vstream->stream->trace->session->id != session->id ||
224 (!ignore_sent_flag && vstream->sent_flag)) {
225 pthread_mutex_unlock(&vstream->stream->lock);
226 viewer_stream_put(vstream);
227 continue;
228 }
229
230 ctf_trace = vstream->stream->trace;
231 send_stream.id = htobe64(vstream->stream->stream_handle);
232 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
233 send_stream.metadata_flag = htobe32(
234 vstream->stream->is_metadata);
235 strncpy(send_stream.path_name, vstream->path_name,
236 sizeof(send_stream.path_name));
237 strncpy(send_stream.channel_name, vstream->channel_name,
238 sizeof(send_stream.channel_name));
239
240 DBG("Sending stream %" PRIu64 " to viewer",
241 vstream->stream->stream_handle);
242 vstream->sent_flag = 1;
243 pthread_mutex_unlock(&vstream->stream->lock);
244
245 ret = send_response(sock, &send_stream, sizeof(send_stream));
246 viewer_stream_put(vstream);
247 if (ret < 0) {
248 goto end_unlock;
249 }
250 }
251
252 ret = 0;
253
254 end_unlock:
255 rcu_read_unlock();
256 return ret;
257 }
258
259 /*
260 * Create every viewer stream possible for the given session with the seek
261 * type. Three counters *can* be return which are in order the total amount of
262 * viewer stream of the session, the number of unsent stream and the number of
263 * stream created. Those counters can be NULL and thus will be ignored.
264 *
265 * Return 0 on success or else a negative value.
266 */
267 static
268 int make_viewer_streams(struct relay_session *session,
269 enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent,
270 uint32_t *nb_created, bool *closed)
271 {
272 int ret;
273 struct lttng_ht_iter iter;
274 struct ctf_trace *ctf_trace;
275
276 assert(session);
277
278 /*
279 * Hold the session lock to ensure that we see either none or
280 * all initial streams for a session, but no intermediate state.
281 */
282 pthread_mutex_lock(&session->lock);
283
284 if (session->connection_closed) {
285 *closed = true;
286 }
287
288 /*
289 * Create viewer streams for relay streams that are ready to be
290 * used for a the given session id only.
291 */
292 rcu_read_lock();
293 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
294 node.node) {
295 struct relay_stream *stream;
296
297 health_code_update();
298
299 if (!ctf_trace_get(ctf_trace)) {
300 continue;
301 }
302
303 cds_list_for_each_entry_rcu(stream, &ctf_trace->stream_list, stream_node) {
304 struct relay_viewer_stream *vstream;
305
306 if (!stream_get(stream)) {
307 continue;
308 }
309 /*
310 * stream published is protected by the session lock.
311 */
312 if (!stream->published) {
313 goto next;
314 }
315 vstream = viewer_stream_get_by_id(stream->stream_handle);
316 if (!vstream) {
317 vstream = viewer_stream_create(stream, seek_t);
318 if (!vstream) {
319 ret = -1;
320 ctf_trace_put(ctf_trace);
321 stream_put(stream);
322 goto error_unlock;
323 }
324
325 if (nb_created) {
326 /* Update number of created stream counter. */
327 (*nb_created)++;
328 }
329 /*
330 * Ensure a self-reference is preserved even
331 * after we have put our local reference.
332 */
333 viewer_stream_get(vstream);
334 } else {
335 if (!vstream->sent_flag && nb_unsent) {
336 /* Update number of unsent stream counter. */
337 (*nb_unsent)++;
338 }
339 }
340 /* Update number of total stream counter. */
341 if (nb_total) {
342 if (stream->is_metadata) {
343 if (!stream->closed ||
344 stream->metadata_received > vstream->metadata_sent) {
345 (*nb_total)++;
346 }
347 } else {
348 if (!stream->closed ||
349 !(((int64_t) (stream->prev_seq - stream->last_net_seq_num)) >= 0)) {
350
351 (*nb_total)++;
352 }
353 }
354 }
355 /* Put local reference. */
356 viewer_stream_put(vstream);
357 next:
358 stream_put(stream);
359 }
360 ctf_trace_put(ctf_trace);
361 }
362
363 ret = 0;
364
365 error_unlock:
366 rcu_read_unlock();
367 pthread_mutex_unlock(&session->lock);
368 return ret;
369 }
370
371 int relayd_live_stop(void)
372 {
373 /* Stop dispatch thread */
374 CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
375 futex_nto1_wake(&viewer_conn_queue.futex);
376 return 0;
377 }
378
379 /*
380 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
381 */
382 static
383 int create_thread_poll_set(struct lttng_poll_event *events, int size)
384 {
385 int ret;
386
387 if (events == NULL || size == 0) {
388 ret = -1;
389 goto error;
390 }
391
392 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
393 if (ret < 0) {
394 goto error;
395 }
396
397 /* Add quit pipe */
398 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
399 if (ret < 0) {
400 goto error;
401 }
402
403 return 0;
404
405 error:
406 return ret;
407 }
408
409 /*
410 * Check if the thread quit pipe was triggered.
411 *
412 * Return 1 if it was triggered else 0;
413 */
414 static
415 int check_thread_quit_pipe(int fd, uint32_t events)
416 {
417 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
418 return 1;
419 }
420
421 return 0;
422 }
423
424 /*
425 * Create and init socket from uri.
426 */
427 static
428 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
429 {
430 int ret;
431 struct lttcomm_sock *sock = NULL;
432
433 sock = lttcomm_alloc_sock_from_uri(uri);
434 if (sock == NULL) {
435 ERR("Allocating socket");
436 goto error;
437 }
438
439 ret = lttcomm_create_sock(sock);
440 if (ret < 0) {
441 goto error;
442 }
443 DBG("Listening on sock %d for live", sock->fd);
444
445 ret = sock->ops->bind(sock);
446 if (ret < 0) {
447 goto error;
448 }
449
450 ret = sock->ops->listen(sock, -1);
451 if (ret < 0) {
452 goto error;
453
454 }
455
456 return sock;
457
458 error:
459 if (sock) {
460 lttcomm_destroy_sock(sock);
461 }
462 return NULL;
463 }
464
465 /*
466 * This thread manages the listening for new connections on the network
467 */
468 static
469 void *thread_listener(void *data)
470 {
471 int i, ret, pollfd, err = -1;
472 uint32_t revents, nb_fd;
473 struct lttng_poll_event events;
474 struct lttcomm_sock *live_control_sock;
475
476 DBG("[thread] Relay live listener started");
477
478 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
479
480 health_code_update();
481
482 live_control_sock = init_socket(live_uri);
483 if (!live_control_sock) {
484 goto error_sock_control;
485 }
486
487 /* Pass 2 as size here for the thread quit pipe and control sockets. */
488 ret = create_thread_poll_set(&events, 2);
489 if (ret < 0) {
490 goto error_create_poll;
491 }
492
493 /* Add the control socket */
494 ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
495 if (ret < 0) {
496 goto error_poll_add;
497 }
498
499 lttng_relay_notify_ready();
500
501 if (testpoint(relayd_thread_live_listener)) {
502 goto error_testpoint;
503 }
504
505 while (1) {
506 health_code_update();
507
508 DBG("Listener accepting live viewers connections");
509
510 restart:
511 health_poll_entry();
512 ret = lttng_poll_wait(&events, -1);
513 health_poll_exit();
514 if (ret < 0) {
515 /*
516 * Restart interrupted system call.
517 */
518 if (errno == EINTR) {
519 goto restart;
520 }
521 goto error;
522 }
523 nb_fd = ret;
524
525 DBG("Relay new viewer connection received");
526 for (i = 0; i < nb_fd; i++) {
527 health_code_update();
528
529 /* Fetch once the poll data */
530 revents = LTTNG_POLL_GETEV(&events, i);
531 pollfd = LTTNG_POLL_GETFD(&events, i);
532
533 if (!revents) {
534 /* No activity for this FD (poll implementation). */
535 continue;
536 }
537
538 /* Thread quit pipe has been closed. Killing thread. */
539 ret = check_thread_quit_pipe(pollfd, revents);
540 if (ret) {
541 err = 0;
542 goto exit;
543 }
544
545 if (revents & LPOLLIN) {
546 /*
547 * A new connection is requested, therefore a
548 * viewer connection is allocated in this
549 * thread, enqueued to a global queue and
550 * dequeued (and freed) in the worker thread.
551 */
552 int val = 1;
553 struct relay_connection *new_conn;
554 struct lttcomm_sock *newsock;
555
556 newsock = live_control_sock->ops->accept(live_control_sock);
557 if (!newsock) {
558 PERROR("accepting control sock");
559 goto error;
560 }
561 DBG("Relay viewer connection accepted socket %d", newsock->fd);
562
563 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
564 sizeof(val));
565 if (ret < 0) {
566 PERROR("setsockopt inet");
567 lttcomm_destroy_sock(newsock);
568 goto error;
569 }
570 new_conn = connection_create(newsock, RELAY_CONNECTION_UNKNOWN);
571 if (!new_conn) {
572 lttcomm_destroy_sock(newsock);
573 goto error;
574 }
575 /* Ownership assumed by the connection. */
576 newsock = NULL;
577
578 /* Enqueue request for the dispatcher thread. */
579 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
580 &new_conn->qnode);
581
582 /*
583 * Wake the dispatch queue futex.
584 * Implicit memory barrier with the
585 * exchange in cds_wfcq_enqueue.
586 */
587 futex_nto1_wake(&viewer_conn_queue.futex);
588 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
589 ERR("socket poll error");
590 goto error;
591 } else {
592 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
593 goto error;
594 }
595 }
596 }
597
598 exit:
599 error:
600 error_poll_add:
601 error_testpoint:
602 lttng_poll_clean(&events);
603 error_create_poll:
604 if (live_control_sock->fd >= 0) {
605 ret = live_control_sock->ops->close(live_control_sock);
606 if (ret) {
607 PERROR("close");
608 }
609 }
610 lttcomm_destroy_sock(live_control_sock);
611 error_sock_control:
612 if (err) {
613 health_error();
614 DBG("Live viewer listener thread exited with error");
615 }
616 health_unregister(health_relayd);
617 DBG("Live viewer listener thread cleanup complete");
618 if (lttng_relay_stop_threads()) {
619 ERR("Error stopping threads");
620 }
621 return NULL;
622 }
623
624 /*
625 * This thread manages the dispatching of the requests to worker threads
626 */
627 static
628 void *thread_dispatcher(void *data)
629 {
630 int err = -1;
631 ssize_t ret;
632 struct cds_wfcq_node *node;
633 struct relay_connection *conn = NULL;
634
635 DBG("[thread] Live viewer relay dispatcher started");
636
637 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
638
639 if (testpoint(relayd_thread_live_dispatcher)) {
640 goto error_testpoint;
641 }
642
643 health_code_update();
644
645 while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
646 health_code_update();
647
648 /* Atomically prepare the queue futex */
649 futex_nto1_prepare(&viewer_conn_queue.futex);
650
651 do {
652 health_code_update();
653
654 /* Dequeue commands */
655 node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
656 &viewer_conn_queue.tail);
657 if (node == NULL) {
658 DBG("Woken up but nothing in the live-viewer "
659 "relay command queue");
660 /* Continue thread execution */
661 break;
662 }
663 conn = caa_container_of(node, struct relay_connection, qnode);
664 DBG("Dispatching viewer request waiting on sock %d",
665 conn->sock->fd);
666
667 /*
668 * Inform worker thread of the new request. This
669 * call is blocking so we can be assured that
670 * the data will be read at some point in time
671 * or wait to the end of the world :)
672 */
673 ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
674 if (ret < 0) {
675 PERROR("write conn pipe");
676 connection_put(conn);
677 goto error;
678 }
679 } while (node != NULL);
680
681 /* Futex wait on queue. Blocking call on futex() */
682 health_poll_entry();
683 futex_nto1_wait(&viewer_conn_queue.futex);
684 health_poll_exit();
685 }
686
687 /* Normal exit, no error */
688 err = 0;
689
690 error:
691 error_testpoint:
692 if (err) {
693 health_error();
694 ERR("Health error occurred in %s", __func__);
695 }
696 health_unregister(health_relayd);
697 DBG("Live viewer dispatch thread dying");
698 if (lttng_relay_stop_threads()) {
699 ERR("Error stopping threads");
700 }
701 return NULL;
702 }
703
704 /*
705 * Establish connection with the viewer and check the versions.
706 *
707 * Return 0 on success or else negative value.
708 */
709 static
710 int viewer_connect(struct relay_connection *conn)
711 {
712 int ret;
713 struct lttng_viewer_connect reply, msg;
714
715 conn->version_check_done = 1;
716
717 health_code_update();
718
719 DBG("Viewer is establishing a connection to the relayd.");
720
721 ret = recv_request(conn->sock, &msg, sizeof(msg));
722 if (ret < 0) {
723 goto end;
724 }
725
726 health_code_update();
727
728 memset(&reply, 0, sizeof(reply));
729 reply.major = RELAYD_VERSION_COMM_MAJOR;
730 reply.minor = RELAYD_VERSION_COMM_MINOR;
731
732 /* Major versions must be the same */
733 if (reply.major != be32toh(msg.major)) {
734 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
735 reply.major, be32toh(msg.major));
736 ret = -1;
737 goto end;
738 }
739
740 conn->major = reply.major;
741 /* We adapt to the lowest compatible version */
742 if (reply.minor <= be32toh(msg.minor)) {
743 conn->minor = reply.minor;
744 } else {
745 conn->minor = be32toh(msg.minor);
746 }
747
748 if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
749 conn->type = RELAY_VIEWER_COMMAND;
750 } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
751 conn->type = RELAY_VIEWER_NOTIFICATION;
752 } else {
753 ERR("Unknown connection type : %u", be32toh(msg.type));
754 ret = -1;
755 goto end;
756 }
757
758 reply.major = htobe32(reply.major);
759 reply.minor = htobe32(reply.minor);
760 if (conn->type == RELAY_VIEWER_COMMAND) {
761 /*
762 * Increment outside of htobe64 macro, because the argument can
763 * be used more than once within the macro, and thus the
764 * operation may be undefined.
765 */
766 pthread_mutex_lock(&last_relay_viewer_session_id_lock);
767 last_relay_viewer_session_id++;
768 pthread_mutex_unlock(&last_relay_viewer_session_id_lock);
769 reply.viewer_session_id = htobe64(last_relay_viewer_session_id);
770 }
771
772 health_code_update();
773
774 ret = send_response(conn->sock, &reply, sizeof(reply));
775 if (ret < 0) {
776 goto end;
777 }
778
779 health_code_update();
780
781 DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
782 ret = 0;
783
784 end:
785 return ret;
786 }
787
788 /*
789 * Send the viewer the list of current sessions.
790 * We need to create a copy of the hash table content because otherwise
791 * we cannot assume the number of entries stays the same between getting
792 * the number of HT elements and iteration over the HT.
793 *
794 * Return 0 on success or else a negative value.
795 */
796 static
797 int viewer_list_sessions(struct relay_connection *conn)
798 {
799 int ret;
800 struct lttng_viewer_list_sessions session_list;
801 struct lttng_ht_iter iter;
802 struct relay_session *session;
803 struct lttng_viewer_session *send_session_buf = NULL;
804 uint32_t buf_count = SESSION_BUF_DEFAULT_COUNT;
805 uint32_t count = 0;
806
807 DBG("List sessions received");
808
809 send_session_buf = zmalloc(SESSION_BUF_DEFAULT_COUNT * sizeof(*send_session_buf));
810 if (!send_session_buf) {
811 return -1;
812 }
813
814 rcu_read_lock();
815 cds_lfht_for_each_entry(sessions_ht->ht, &iter.iter, session,
816 session_n.node) {
817 struct lttng_viewer_session *send_session;
818
819 health_code_update();
820
821 if (count >= buf_count) {
822 struct lttng_viewer_session *newbuf;
823 uint32_t new_buf_count = buf_count << 1;
824
825 newbuf = realloc(send_session_buf,
826 new_buf_count * sizeof(*send_session_buf));
827 if (!newbuf) {
828 ret = -1;
829 rcu_read_unlock();
830 goto end_free;
831 }
832 send_session_buf = newbuf;
833 buf_count = new_buf_count;
834 }
835 send_session = &send_session_buf[count];
836 strncpy(send_session->session_name, session->session_name,
837 sizeof(send_session->session_name));
838 strncpy(send_session->hostname, session->hostname,
839 sizeof(send_session->hostname));
840 send_session->id = htobe64(session->id);
841 send_session->live_timer = htobe32(session->live_timer);
842 if (session->viewer_attached) {
843 send_session->clients = htobe32(1);
844 } else {
845 send_session->clients = htobe32(0);
846 }
847 send_session->streams = htobe32(session->stream_count);
848 count++;
849 }
850 rcu_read_unlock();
851
852 session_list.sessions_count = htobe32(count);
853
854 health_code_update();
855
856 ret = send_response(conn->sock, &session_list, sizeof(session_list));
857 if (ret < 0) {
858 goto end_free;
859 }
860
861 health_code_update();
862
863 ret = send_response(conn->sock, send_session_buf,
864 count * sizeof(*send_session_buf));
865 if (ret < 0) {
866 goto end_free;
867 }
868 health_code_update();
869
870 ret = 0;
871 end_free:
872 free(send_session_buf);
873 return ret;
874 }
875
876 /*
877 * Send the viewer the list of current streams.
878 */
879 static
880 int viewer_get_new_streams(struct relay_connection *conn)
881 {
882 int ret, send_streams = 0;
883 uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0, nb_total = 0;
884 struct lttng_viewer_new_streams_request request;
885 struct lttng_viewer_new_streams_response response;
886 struct relay_session *session;
887 uint64_t session_id;
888 bool closed = false;
889
890 assert(conn);
891
892 DBG("Get new streams received");
893
894 health_code_update();
895
896 /* Receive the request from the connected client. */
897 ret = recv_request(conn->sock, &request, sizeof(request));
898 if (ret < 0) {
899 goto error;
900 }
901 session_id = be64toh(request.session_id);
902
903 health_code_update();
904
905 memset(&response, 0, sizeof(response));
906
907 session = session_get_by_id(session_id);
908 if (!session) {
909 DBG("Relay session %" PRIu64 " not found", session_id);
910 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
911 goto send_reply;
912 }
913
914 if (!viewer_session_is_attached(conn->viewer_session, session)) {
915 send_streams = 0;
916 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
917 goto send_reply;
918 }
919
920 send_streams = 1;
921 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
922
923 ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, &nb_total, &nb_unsent,
924 &nb_created, &closed);
925 if (ret < 0) {
926 goto end_put_session;
927 }
928 /* Only send back the newly created streams with the unsent ones. */
929 nb_streams = nb_created + nb_unsent;
930 response.streams_count = htobe32(nb_streams);
931
932 /*
933 * If the session is closed, HUP when there are no more streams
934 * with data.
935 */
936 if (closed && nb_total == 0) {
937 send_streams = 0;
938 response.streams_count = 0;
939 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
940 goto send_reply;
941 }
942
943 send_reply:
944 health_code_update();
945 ret = send_response(conn->sock, &response, sizeof(response));
946 if (ret < 0) {
947 goto end_put_session;
948 }
949 health_code_update();
950
951 /*
952 * Unknown or empty session, just return gracefully, the viewer
953 * knows what is happening.
954 */
955 if (!send_streams || !nb_streams) {
956 ret = 0;
957 goto end_put_session;
958 }
959
960 /*
961 * Send stream and *DON'T* ignore the sent flag so every viewer
962 * streams that were not sent from that point will be sent to
963 * the viewer.
964 */
965 ret = send_viewer_streams(conn->sock, session, 0);
966 if (ret < 0) {
967 goto end_put_session;
968 }
969
970 end_put_session:
971 if (session) {
972 session_put(session);
973 }
974 error:
975 return ret;
976 }
977
978 /*
979 * Send the viewer the list of current sessions.
980 */
981 static
982 int viewer_attach_session(struct relay_connection *conn)
983 {
984 int send_streams = 0;
985 ssize_t ret;
986 uint32_t nb_streams = 0;
987 enum lttng_viewer_seek seek_type;
988 struct lttng_viewer_attach_session_request request;
989 struct lttng_viewer_attach_session_response response;
990 struct relay_session *session = NULL;
991 bool closed = false;
992
993 assert(conn);
994
995 health_code_update();
996
997 /* Receive the request from the connected client. */
998 ret = recv_request(conn->sock, &request, sizeof(request));
999 if (ret < 0) {
1000 goto error;
1001 }
1002
1003 health_code_update();
1004
1005 memset(&response, 0, sizeof(response));
1006
1007 if (!conn->viewer_session) {
1008 DBG("Client trying to attach before creating a live viewer session");
1009 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1010 goto send_reply;
1011 }
1012
1013 session = session_get_by_id(be64toh(request.session_id));
1014 if (!session) {
1015 DBG("Relay session %" PRIu64 " not found",
1016 be64toh(request.session_id));
1017 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1018 goto send_reply;
1019 }
1020 DBG("Attach session ID %" PRIu64 " received",
1021 be64toh(request.session_id));
1022
1023 if (session->live_timer == 0) {
1024 DBG("Not live session");
1025 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1026 goto send_reply;
1027 }
1028
1029 send_streams = 1;
1030 ret = viewer_session_attach(conn->viewer_session, session);
1031 if (ret) {
1032 DBG("Already a viewer attached");
1033 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1034 goto send_reply;
1035 }
1036
1037 switch (be32toh(request.seek)) {
1038 case LTTNG_VIEWER_SEEK_BEGINNING:
1039 case LTTNG_VIEWER_SEEK_LAST:
1040 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1041 seek_type = be32toh(request.seek);
1042 break;
1043 default:
1044 ERR("Wrong seek parameter");
1045 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1046 send_streams = 0;
1047 goto send_reply;
1048 }
1049
1050 ret = make_viewer_streams(session, seek_type, &nb_streams, NULL,
1051 NULL, &closed);
1052 if (ret < 0) {
1053 goto end_put_session;
1054 }
1055 response.streams_count = htobe32(nb_streams);
1056
1057 /*
1058 * If the session is closed when the viewer is attaching, it
1059 * means some of the streams may have been concurrently removed,
1060 * so we don't allow the viewer to attach, even if there are
1061 * streams available.
1062 */
1063 if (closed) {
1064 send_streams = 0;
1065 response.streams_count = 0;
1066 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
1067 goto send_reply;
1068 }
1069
1070 send_reply:
1071 health_code_update();
1072 ret = send_response(conn->sock, &response, sizeof(response));
1073 if (ret < 0) {
1074 goto end_put_session;
1075 }
1076 health_code_update();
1077
1078 /*
1079 * Unknown or empty session, just return gracefully, the viewer
1080 * knows what is happening.
1081 */
1082 if (!send_streams || !nb_streams) {
1083 ret = 0;
1084 goto end_put_session;
1085 }
1086
1087 /* Send stream and ignore the sent flag. */
1088 ret = send_viewer_streams(conn->sock, session, 1);
1089 if (ret < 0) {
1090 goto end_put_session;
1091 }
1092
1093 end_put_session:
1094 if (session) {
1095 session_put(session);
1096 }
1097 error:
1098 return ret;
1099 }
1100
1101 /*
1102 * Open the index file if needed for the given vstream.
1103 *
1104 * If an index file is successfully opened, the vstream index_fd set with
1105 * it.
1106 *
1107 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1108 *
1109 * Called with rstream lock held.
1110 */
1111 static int try_open_index(struct relay_viewer_stream *vstream,
1112 struct relay_stream *rstream)
1113 {
1114 int ret = 0;
1115
1116 if (vstream->index_fd) {
1117 goto end;
1118 }
1119
1120 /*
1121 * First time, we open the index file and at least one index is ready.
1122 */
1123 if (rstream->index_received_seqcount == 0) {
1124 ret = -ENOENT;
1125 goto end;
1126 }
1127 ret = index_open(vstream->path_name, vstream->channel_name,
1128 vstream->stream->tracefile_count,
1129 vstream->current_tracefile_id);
1130 if (ret >= 0) {
1131 vstream->index_fd = stream_fd_create(ret);
1132 if (!vstream->index_fd) {
1133 if (close(ret)) {
1134 PERROR("close");
1135 }
1136 ret = -1;
1137 } else {
1138 ret = 0;
1139 }
1140 goto end;
1141 }
1142
1143 end:
1144 return ret;
1145 }
1146
1147 /*
1148 * Check the status of the index for the given stream. This function
1149 * updates the index structure if needed and can put (close) the vstream
1150 * in the HUP situation.
1151 *
1152 * Return 0 means that we can proceed with the index. A value of 1 means
1153 * that the index has been updated and is ready to be sent to the
1154 * client. A negative value indicates an error that can't be handled.
1155 *
1156 * Called with rstream lock held.
1157 */
1158 static int check_index_status(struct relay_viewer_stream *vstream,
1159 struct relay_stream *rstream, struct ctf_trace *trace,
1160 struct lttng_viewer_index *index)
1161 {
1162 int ret;
1163
1164 if (trace->session->connection_closed
1165 && rstream->index_received_seqcount
1166 == vstream->index_sent_seqcount) {
1167 /* Last index sent and session connection is closed. */
1168 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1169 goto hup;
1170 } else if (rstream->beacon_ts_end != -1ULL &&
1171 rstream->index_received_seqcount
1172 == vstream->index_sent_seqcount) {
1173 /*
1174 * We've received a synchronization beacon and the last index
1175 * available has been sent, the index for now is inactive.
1176 *
1177 * In this case, we have received a beacon which allows us to
1178 * inform the client of a time interval during which we can
1179 * guarantee that there are no events to read (and never will
1180 * be).
1181 */
1182 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1183 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1184 index->stream_id = htobe64(rstream->ctf_stream_id);
1185 goto index_ready;
1186 } else if (rstream->index_received_seqcount
1187 == vstream->index_sent_seqcount) {
1188 /*
1189 * This checks whether received == sent seqcount. In
1190 * this case, we have not received a beacon. Therefore,
1191 * we can only ask the client to retry later.
1192 */
1193 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1194 goto index_ready;
1195 } else if (!tracefile_array_seq_in_file(rstream->tfa,
1196 vstream->current_tracefile_id,
1197 vstream->index_sent_seqcount)) {
1198 /*
1199 * The next index we want to send cannot be read either
1200 * because we need to perform a rotation, or due to
1201 * the producer having overwritten its trace file.
1202 */
1203 DBG("Viewer stream %" PRIu64 " rotation",
1204 vstream->stream->stream_handle);
1205 ret = viewer_stream_rotate(vstream);
1206 if (ret < 0) {
1207 goto end;
1208 } else if (ret == 1) {
1209 /* EOF across entire stream. */
1210 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1211 goto hup;
1212 }
1213 /*
1214 * If we have been pushed due to overwrite, it
1215 * necessarily means there is data that can be read in
1216 * the stream. If we rotated because we reached the end
1217 * of a tracefile, it means the following tracefile
1218 * needs to contain at least one index, else we would
1219 * have already returned LTTNG_VIEWER_INDEX_RETRY to the
1220 * viewer. The updated index_sent_seqcount needs to
1221 * point to a readable index entry now.
1222 *
1223 * In the case where we "rotate" on a single file, we
1224 * can end up in a case where the requested index is
1225 * still unavailable.
1226 */
1227 if (rstream->tracefile_count == 1 &&
1228 !tracefile_array_seq_in_file(
1229 rstream->tfa,
1230 vstream->current_tracefile_id,
1231 vstream->index_sent_seqcount)) {
1232 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1233 goto index_ready;
1234 }
1235 assert(tracefile_array_seq_in_file(rstream->tfa,
1236 vstream->current_tracefile_id,
1237 vstream->index_sent_seqcount));
1238 }
1239 /* ret == 0 means successful so we continue. */
1240 ret = 0;
1241 end:
1242 return ret;
1243
1244 hup:
1245 viewer_stream_put(vstream);
1246 index_ready:
1247 return 1;
1248 }
1249
1250 /*
1251 * Send the next index for a stream.
1252 *
1253 * Return 0 on success or else a negative value.
1254 */
1255 static
1256 int viewer_get_next_index(struct relay_connection *conn)
1257 {
1258 int ret;
1259 ssize_t read_ret;
1260 struct lttng_viewer_get_next_index request_index;
1261 struct lttng_viewer_index viewer_index;
1262 struct ctf_packet_index packet_index;
1263 struct relay_viewer_stream *vstream = NULL;
1264 struct relay_stream *rstream = NULL;
1265 struct ctf_trace *ctf_trace = NULL;
1266 struct relay_viewer_stream *metadata_viewer_stream = NULL;
1267
1268 assert(conn);
1269
1270 DBG("Viewer get next index");
1271
1272 memset(&viewer_index, 0, sizeof(viewer_index));
1273 health_code_update();
1274
1275 ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1276 if (ret < 0) {
1277 goto end;
1278 }
1279 health_code_update();
1280
1281 vstream = viewer_stream_get_by_id(be64toh(request_index.stream_id));
1282 if (!vstream) {
1283 DBG("Client requested index of unknown stream id %" PRIu64,
1284 be64toh(request_index.stream_id));
1285 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1286 goto send_reply;
1287 }
1288
1289 /* Use back. ref. Protected by refcounts. */
1290 rstream = vstream->stream;
1291 ctf_trace = rstream->trace;
1292
1293 /* metadata_viewer_stream may be NULL. */
1294 metadata_viewer_stream =
1295 ctf_trace_get_viewer_metadata_stream(ctf_trace);
1296
1297 pthread_mutex_lock(&rstream->lock);
1298
1299 /*
1300 * The viewer should not ask for index on metadata stream.
1301 */
1302 if (rstream->is_metadata) {
1303 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1304 goto send_reply;
1305 }
1306
1307 /* Try to open an index if one is needed for that stream. */
1308 ret = try_open_index(vstream, rstream);
1309 if (ret < 0) {
1310 if (ret == -ENOENT) {
1311 /*
1312 * The index is created only when the first data
1313 * packet arrives, it might not be ready at the
1314 * beginning of the session
1315 */
1316 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1317 } else {
1318 /* Unhandled error. */
1319 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1320 }
1321 goto send_reply;
1322 }
1323
1324 ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1325 if (ret < 0) {
1326 goto error_put;
1327 } else if (ret == 1) {
1328 /*
1329 * We have no index to send and check_index_status has populated
1330 * viewer_index's status.
1331 */
1332 goto send_reply;
1333 }
1334 /* At this point, ret is 0 thus we will be able to read the index. */
1335 assert(!ret);
1336
1337 /*
1338 * vstream->stream_fd may be NULL if it has been closed by
1339 * tracefile rotation, or if we are at the beginning of the
1340 * stream. We open the data stream file here to protect against
1341 * overwrite caused by tracefile rotation (in association with
1342 * unlink performed before overwrite).
1343 */
1344 if (!vstream->stream_fd) {
1345 char fullpath[PATH_MAX];
1346
1347 if (vstream->stream->tracefile_count > 0) {
1348 ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64,
1349 vstream->path_name,
1350 vstream->channel_name,
1351 vstream->current_tracefile_id);
1352 } else {
1353 ret = snprintf(fullpath, PATH_MAX, "%s/%s",
1354 vstream->path_name,
1355 vstream->channel_name);
1356 }
1357 if (ret < 0) {
1358 goto error_put;
1359 }
1360 ret = open(fullpath, O_RDONLY);
1361 if (ret < 0) {
1362 PERROR("Relay opening trace file");
1363 goto error_put;
1364 }
1365 vstream->stream_fd = stream_fd_create(ret);
1366 if (!vstream->stream_fd) {
1367 if (close(ret)) {
1368 PERROR("close");
1369 }
1370 goto error_put;
1371 }
1372 }
1373
1374 ret = check_new_streams(conn);
1375 if (ret < 0) {
1376 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1377 goto send_reply;
1378 } else if (ret == 1) {
1379 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1380 }
1381
1382 read_ret = lttng_read(vstream->index_fd->fd, &packet_index,
1383 sizeof(packet_index));
1384 if (read_ret < sizeof(packet_index)) {
1385 ERR("Relay reading index file %d returned %zd",
1386 vstream->index_fd->fd, read_ret);
1387 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1388 goto send_reply;
1389 } else {
1390 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1391 vstream->index_sent_seqcount++;
1392 }
1393
1394 /*
1395 * Indexes are stored in big endian, no need to switch before sending.
1396 */
1397 DBG("Sending viewer index for stream %" PRIu64 " offset %" PRIu64,
1398 rstream->stream_handle,
1399 be64toh(packet_index.offset));
1400 viewer_index.offset = packet_index.offset;
1401 viewer_index.packet_size = packet_index.packet_size;
1402 viewer_index.content_size = packet_index.content_size;
1403 viewer_index.timestamp_begin = packet_index.timestamp_begin;
1404 viewer_index.timestamp_end = packet_index.timestamp_end;
1405 viewer_index.events_discarded = packet_index.events_discarded;
1406 viewer_index.stream_id = packet_index.stream_id;
1407
1408 send_reply:
1409 if (rstream) {
1410 pthread_mutex_unlock(&rstream->lock);
1411 }
1412
1413 if (metadata_viewer_stream) {
1414 pthread_mutex_lock(&metadata_viewer_stream->stream->lock);
1415 DBG("get next index metadata check: recv %" PRIu64
1416 " sent %" PRIu64,
1417 metadata_viewer_stream->stream->metadata_received,
1418 metadata_viewer_stream->metadata_sent);
1419 if (!metadata_viewer_stream->stream->metadata_received ||
1420 metadata_viewer_stream->stream->metadata_received >
1421 metadata_viewer_stream->metadata_sent) {
1422 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1423 }
1424 pthread_mutex_unlock(&metadata_viewer_stream->stream->lock);
1425 }
1426
1427 viewer_index.flags = htobe32(viewer_index.flags);
1428 health_code_update();
1429
1430 ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1431 if (ret < 0) {
1432 goto end;
1433 }
1434 health_code_update();
1435
1436 if (vstream) {
1437 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1438 vstream->index_sent_seqcount,
1439 vstream->stream->stream_handle);
1440 }
1441 end:
1442 if (metadata_viewer_stream) {
1443 viewer_stream_put(metadata_viewer_stream);
1444 }
1445 if (vstream) {
1446 viewer_stream_put(vstream);
1447 }
1448 return ret;
1449
1450 error_put:
1451 pthread_mutex_unlock(&rstream->lock);
1452 if (metadata_viewer_stream) {
1453 viewer_stream_put(metadata_viewer_stream);
1454 }
1455 viewer_stream_put(vstream);
1456 return ret;
1457 }
1458
1459 /*
1460 * Send the next index for a stream
1461 *
1462 * Return 0 on success or else a negative value.
1463 */
1464 static
1465 int viewer_get_packet(struct relay_connection *conn)
1466 {
1467 int ret, send_data = 0;
1468 char *data = NULL;
1469 uint32_t len = 0;
1470 ssize_t read_len;
1471 struct lttng_viewer_get_packet get_packet_info;
1472 struct lttng_viewer_trace_packet reply;
1473 struct relay_viewer_stream *vstream = NULL;
1474
1475 DBG2("Relay get data packet");
1476
1477 health_code_update();
1478
1479 ret = recv_request(conn->sock, &get_packet_info,
1480 sizeof(get_packet_info));
1481 if (ret < 0) {
1482 goto end;
1483 }
1484 health_code_update();
1485
1486 /* From this point on, the error label can be reached. */
1487 memset(&reply, 0, sizeof(reply));
1488
1489 vstream = viewer_stream_get_by_id(be64toh(get_packet_info.stream_id));
1490 if (!vstream) {
1491 DBG("Client requested packet of unknown stream id %" PRIu64,
1492 be64toh(get_packet_info.stream_id));
1493 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1494 goto send_reply_nolock;
1495 }
1496
1497 pthread_mutex_lock(&vstream->stream->lock);
1498
1499 len = be32toh(get_packet_info.len);
1500 data = zmalloc(len);
1501 if (!data) {
1502 PERROR("relay data zmalloc");
1503 goto error;
1504 }
1505
1506 ret = lseek(vstream->stream_fd->fd, be64toh(get_packet_info.offset),
1507 SEEK_SET);
1508 if (ret < 0) {
1509 PERROR("lseek fd %d to offset %" PRIu64, vstream->stream_fd->fd,
1510 be64toh(get_packet_info.offset));
1511 goto error;
1512 }
1513 read_len = lttng_read(vstream->stream_fd->fd, data, len);
1514 if (read_len < len) {
1515 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1516 vstream->stream_fd->fd,
1517 be64toh(get_packet_info.offset));
1518 goto error;
1519 }
1520 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1521 reply.len = htobe32(len);
1522 send_data = 1;
1523 goto send_reply;
1524
1525 error:
1526 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1527
1528 send_reply:
1529 if (vstream) {
1530 pthread_mutex_unlock(&vstream->stream->lock);
1531 }
1532 send_reply_nolock:
1533 reply.flags = htobe32(reply.flags);
1534
1535 health_code_update();
1536
1537 ret = send_response(conn->sock, &reply, sizeof(reply));
1538 if (ret < 0) {
1539 goto end_free;
1540 }
1541 health_code_update();
1542
1543 if (send_data) {
1544 health_code_update();
1545 ret = send_response(conn->sock, data, len);
1546 if (ret < 0) {
1547 goto end_free;
1548 }
1549 health_code_update();
1550 }
1551
1552 DBG("Sent %u bytes for stream %" PRIu64, len,
1553 be64toh(get_packet_info.stream_id));
1554
1555 end_free:
1556 free(data);
1557 end:
1558 if (vstream) {
1559 viewer_stream_put(vstream);
1560 }
1561 return ret;
1562 }
1563
1564 /*
1565 * Send the session's metadata
1566 *
1567 * Return 0 on success else a negative value.
1568 */
1569 static
1570 int viewer_get_metadata(struct relay_connection *conn)
1571 {
1572 int ret = 0;
1573 ssize_t read_len;
1574 uint64_t len = 0;
1575 char *data = NULL;
1576 struct lttng_viewer_get_metadata request;
1577 struct lttng_viewer_metadata_packet reply;
1578 struct relay_viewer_stream *vstream = NULL;
1579
1580 assert(conn);
1581
1582 DBG("Relay get metadata");
1583
1584 health_code_update();
1585
1586 ret = recv_request(conn->sock, &request, sizeof(request));
1587 if (ret < 0) {
1588 goto end;
1589 }
1590 health_code_update();
1591
1592 memset(&reply, 0, sizeof(reply));
1593
1594 vstream = viewer_stream_get_by_id(be64toh(request.stream_id));
1595 if (!vstream) {
1596 /*
1597 * The metadata stream can be closed by a CLOSE command
1598 * just before we attach. It can also be closed by
1599 * per-pid tracing during tracing. Therefore, it is
1600 * possible that we cannot find this viewer stream.
1601 * Reply back to the client with an error if we cannot
1602 * find it.
1603 */
1604 DBG("Client requested metadata of unknown stream id %" PRIu64,
1605 be64toh(request.stream_id));
1606 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1607 goto send_reply;
1608 }
1609 pthread_mutex_lock(&vstream->stream->lock);
1610 if (!vstream->stream->is_metadata) {
1611 ERR("Invalid metadata stream");
1612 goto error;
1613 }
1614
1615 assert(vstream->metadata_sent <= vstream->stream->metadata_received);
1616
1617 len = vstream->stream->metadata_received - vstream->metadata_sent;
1618 if (len == 0) {
1619 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1620 goto send_reply;
1621 }
1622
1623 /* first time, we open the metadata file */
1624 if (!vstream->stream_fd) {
1625 char fullpath[PATH_MAX];
1626
1627 ret = snprintf(fullpath, PATH_MAX, "%s/%s", vstream->path_name,
1628 vstream->channel_name);
1629 if (ret < 0) {
1630 goto error;
1631 }
1632 ret = open(fullpath, O_RDONLY);
1633 if (ret < 0) {
1634 PERROR("Relay opening metadata file");
1635 goto error;
1636 }
1637 vstream->stream_fd = stream_fd_create(ret);
1638 if (!vstream->stream_fd) {
1639 if (close(ret)) {
1640 PERROR("close");
1641 }
1642 goto error;
1643 }
1644 }
1645
1646 reply.len = htobe64(len);
1647 data = zmalloc(len);
1648 if (!data) {
1649 PERROR("viewer metadata zmalloc");
1650 goto error;
1651 }
1652
1653 read_len = lttng_read(vstream->stream_fd->fd, data, len);
1654 if (read_len < len) {
1655 PERROR("Relay reading metadata file");
1656 goto error;
1657 }
1658 vstream->metadata_sent += read_len;
1659 if (vstream->metadata_sent == vstream->stream->metadata_received
1660 && vstream->stream->closed) {
1661 /* Release ownership for the viewer metadata stream. */
1662 viewer_stream_put(vstream);
1663 }
1664
1665 reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1666
1667 goto send_reply;
1668
1669 error:
1670 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1671
1672 send_reply:
1673 health_code_update();
1674 if (vstream) {
1675 pthread_mutex_unlock(&vstream->stream->lock);
1676 }
1677 ret = send_response(conn->sock, &reply, sizeof(reply));
1678 if (ret < 0) {
1679 goto end_free;
1680 }
1681 health_code_update();
1682
1683 if (len > 0) {
1684 ret = send_response(conn->sock, data, len);
1685 if (ret < 0) {
1686 goto end_free;
1687 }
1688 }
1689
1690 DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1691 be64toh(request.stream_id));
1692
1693 DBG("Metadata sent");
1694
1695 end_free:
1696 free(data);
1697 end:
1698 if (vstream) {
1699 viewer_stream_put(vstream);
1700 }
1701 return ret;
1702 }
1703
1704 /*
1705 * Create a viewer session.
1706 *
1707 * Return 0 on success or else a negative value.
1708 */
1709 static
1710 int viewer_create_session(struct relay_connection *conn)
1711 {
1712 int ret;
1713 struct lttng_viewer_create_session_response resp;
1714
1715 DBG("Viewer create session received");
1716
1717 memset(&resp, 0, sizeof(resp));
1718 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1719 conn->viewer_session = viewer_session_create();
1720 if (!conn->viewer_session) {
1721 ERR("Allocation viewer session");
1722 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1723 goto send_reply;
1724 }
1725
1726 send_reply:
1727 health_code_update();
1728 ret = send_response(conn->sock, &resp, sizeof(resp));
1729 if (ret < 0) {
1730 goto end;
1731 }
1732 health_code_update();
1733 ret = 0;
1734
1735 end:
1736 return ret;
1737 }
1738
1739
1740 /*
1741 * live_relay_unknown_command: send -1 if received unknown command
1742 */
1743 static
1744 void live_relay_unknown_command(struct relay_connection *conn)
1745 {
1746 struct lttcomm_relayd_generic_reply reply;
1747
1748 memset(&reply, 0, sizeof(reply));
1749 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1750 (void) send_response(conn->sock, &reply, sizeof(reply));
1751 }
1752
1753 /*
1754 * Process the commands received on the control socket
1755 */
1756 static
1757 int process_control(struct lttng_viewer_cmd *recv_hdr,
1758 struct relay_connection *conn)
1759 {
1760 int ret = 0;
1761 uint32_t msg_value;
1762
1763 msg_value = be32toh(recv_hdr->cmd);
1764
1765 /*
1766 * Make sure we've done the version check before any command other then a
1767 * new client connection.
1768 */
1769 if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1770 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1771 ret = -1;
1772 goto end;
1773 }
1774
1775 switch (msg_value) {
1776 case LTTNG_VIEWER_CONNECT:
1777 ret = viewer_connect(conn);
1778 break;
1779 case LTTNG_VIEWER_LIST_SESSIONS:
1780 ret = viewer_list_sessions(conn);
1781 break;
1782 case LTTNG_VIEWER_ATTACH_SESSION:
1783 ret = viewer_attach_session(conn);
1784 break;
1785 case LTTNG_VIEWER_GET_NEXT_INDEX:
1786 ret = viewer_get_next_index(conn);
1787 break;
1788 case LTTNG_VIEWER_GET_PACKET:
1789 ret = viewer_get_packet(conn);
1790 break;
1791 case LTTNG_VIEWER_GET_METADATA:
1792 ret = viewer_get_metadata(conn);
1793 break;
1794 case LTTNG_VIEWER_GET_NEW_STREAMS:
1795 ret = viewer_get_new_streams(conn);
1796 break;
1797 case LTTNG_VIEWER_CREATE_SESSION:
1798 ret = viewer_create_session(conn);
1799 break;
1800 default:
1801 ERR("Received unknown viewer command (%u)",
1802 be32toh(recv_hdr->cmd));
1803 live_relay_unknown_command(conn);
1804 ret = -1;
1805 goto end;
1806 }
1807
1808 end:
1809 return ret;
1810 }
1811
1812 static
1813 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1814 {
1815 int ret;
1816
1817 (void) lttng_poll_del(events, pollfd);
1818
1819 ret = close(pollfd);
1820 if (ret < 0) {
1821 ERR("Closing pollfd %d", pollfd);
1822 }
1823 }
1824
1825 /*
1826 * This thread does the actual work
1827 */
1828 static
1829 void *thread_worker(void *data)
1830 {
1831 int ret, err = -1;
1832 uint32_t nb_fd;
1833 struct lttng_poll_event events;
1834 struct lttng_ht *viewer_connections_ht;
1835 struct lttng_ht_iter iter;
1836 struct lttng_viewer_cmd recv_hdr;
1837 struct relay_connection *destroy_conn;
1838
1839 DBG("[thread] Live viewer relay worker started");
1840
1841 rcu_register_thread();
1842
1843 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1844
1845 if (testpoint(relayd_thread_live_worker)) {
1846 goto error_testpoint;
1847 }
1848
1849 /* table of connections indexed on socket */
1850 viewer_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1851 if (!viewer_connections_ht) {
1852 goto viewer_connections_ht_error;
1853 }
1854
1855 ret = create_thread_poll_set(&events, 2);
1856 if (ret < 0) {
1857 goto error_poll_create;
1858 }
1859
1860 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1861 if (ret < 0) {
1862 goto error;
1863 }
1864
1865 restart:
1866 while (1) {
1867 int i;
1868
1869 health_code_update();
1870
1871 /* Infinite blocking call, waiting for transmission */
1872 DBG3("Relayd live viewer worker thread polling...");
1873 health_poll_entry();
1874 ret = lttng_poll_wait(&events, -1);
1875 health_poll_exit();
1876 if (ret < 0) {
1877 /*
1878 * Restart interrupted system call.
1879 */
1880 if (errno == EINTR) {
1881 goto restart;
1882 }
1883 goto error;
1884 }
1885
1886 nb_fd = ret;
1887
1888 /*
1889 * Process control. The control connection is prioritised so we don't
1890 * starve it with high throughput tracing data on the data
1891 * connection.
1892 */
1893 for (i = 0; i < nb_fd; i++) {
1894 /* Fetch once the poll data */
1895 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1896 int pollfd = LTTNG_POLL_GETFD(&events, i);
1897
1898 health_code_update();
1899
1900 if (!revents) {
1901 /* No activity for this FD (poll implementation). */
1902 continue;
1903 }
1904
1905 /* Thread quit pipe has been closed. Killing thread. */
1906 ret = check_thread_quit_pipe(pollfd, revents);
1907 if (ret) {
1908 err = 0;
1909 goto exit;
1910 }
1911
1912 /* Inspect the relay conn pipe for new connection. */
1913 if (pollfd == live_conn_pipe[0]) {
1914 if (revents & LPOLLIN) {
1915 struct relay_connection *conn;
1916
1917 ret = lttng_read(live_conn_pipe[0],
1918 &conn, sizeof(conn));
1919 if (ret < 0) {
1920 goto error;
1921 }
1922 lttng_poll_add(&events, conn->sock->fd,
1923 LPOLLIN | LPOLLRDHUP);
1924 connection_ht_add(viewer_connections_ht, conn);
1925 DBG("Connection socket %d added to poll", conn->sock->fd);
1926 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1927 ERR("Relay live pipe error");
1928 goto error;
1929 } else {
1930 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1931 goto error;
1932 }
1933 } else {
1934 /* Connection activity. */
1935 struct relay_connection *conn;
1936
1937 conn = connection_get_by_sock(viewer_connections_ht, pollfd);
1938 if (!conn) {
1939 continue;
1940 }
1941
1942 if (revents & LPOLLIN) {
1943 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
1944 sizeof(recv_hdr), 0);
1945 if (ret <= 0) {
1946 /* Connection closed. */
1947 cleanup_connection_pollfd(&events, pollfd);
1948 /* Put "create" ownership reference. */
1949 connection_put(conn);
1950 DBG("Viewer control conn closed with %d", pollfd);
1951 } else {
1952 ret = process_control(&recv_hdr, conn);
1953 if (ret < 0) {
1954 /* Clear the session on error. */
1955 cleanup_connection_pollfd(&events, pollfd);
1956 /* Put "create" ownership reference. */
1957 connection_put(conn);
1958 DBG("Viewer connection closed with %d", pollfd);
1959 }
1960 }
1961 } else if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1962 cleanup_connection_pollfd(&events, pollfd);
1963 /* Put "create" ownership reference. */
1964 connection_put(conn);
1965 } else {
1966 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
1967 connection_put(conn);
1968 goto error;
1969 }
1970 /* Put local "get_by_sock" reference. */
1971 connection_put(conn);
1972 }
1973 }
1974 }
1975
1976 exit:
1977 error:
1978 lttng_poll_clean(&events);
1979
1980 /* Cleanup reamaining connection object. */
1981 rcu_read_lock();
1982 cds_lfht_for_each_entry(viewer_connections_ht->ht, &iter.iter,
1983 destroy_conn,
1984 sock_n.node) {
1985 health_code_update();
1986 connection_put(destroy_conn);
1987 }
1988 rcu_read_unlock();
1989 error_poll_create:
1990 lttng_ht_destroy(viewer_connections_ht);
1991 viewer_connections_ht_error:
1992 /* Close relay conn pipes */
1993 utils_close_pipe(live_conn_pipe);
1994 if (err) {
1995 DBG("Viewer worker thread exited with error");
1996 }
1997 DBG("Viewer worker thread cleanup complete");
1998 error_testpoint:
1999 if (err) {
2000 health_error();
2001 ERR("Health error occurred in %s", __func__);
2002 }
2003 health_unregister(health_relayd);
2004 if (lttng_relay_stop_threads()) {
2005 ERR("Error stopping threads");
2006 }
2007 rcu_unregister_thread();
2008 return NULL;
2009 }
2010
2011 /*
2012 * Create the relay command pipe to wake thread_manage_apps.
2013 * Closed in cleanup().
2014 */
2015 static int create_conn_pipe(void)
2016 {
2017 return utils_create_pipe_cloexec(live_conn_pipe);
2018 }
2019
2020 int relayd_live_join(void)
2021 {
2022 int ret, retval = 0;
2023 void *status;
2024
2025 ret = pthread_join(live_listener_thread, &status);
2026 if (ret) {
2027 errno = ret;
2028 PERROR("pthread_join live listener");
2029 retval = -1;
2030 }
2031
2032 ret = pthread_join(live_worker_thread, &status);
2033 if (ret) {
2034 errno = ret;
2035 PERROR("pthread_join live worker");
2036 retval = -1;
2037 }
2038
2039 ret = pthread_join(live_dispatcher_thread, &status);
2040 if (ret) {
2041 errno = ret;
2042 PERROR("pthread_join live dispatcher");
2043 retval = -1;
2044 }
2045
2046 cleanup_relayd_live();
2047
2048 return retval;
2049 }
2050
2051 /*
2052 * main
2053 */
2054 int relayd_live_create(struct lttng_uri *uri)
2055 {
2056 int ret = 0, retval = 0;
2057 void *status;
2058 int is_root;
2059
2060 if (!uri) {
2061 retval = -1;
2062 goto exit_init_data;
2063 }
2064 live_uri = uri;
2065
2066 /* Check if daemon is UID = 0 */
2067 is_root = !getuid();
2068
2069 if (!is_root) {
2070 if (live_uri->port < 1024) {
2071 ERR("Need to be root to use ports < 1024");
2072 retval = -1;
2073 goto exit_init_data;
2074 }
2075 }
2076
2077 /* Setup the thread apps communication pipe. */
2078 if (create_conn_pipe()) {
2079 retval = -1;
2080 goto exit_init_data;
2081 }
2082
2083 /* Init relay command queue. */
2084 cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2085
2086 /* Set up max poll set size */
2087 if (lttng_poll_set_max_size()) {
2088 retval = -1;
2089 goto exit_init_data;
2090 }
2091
2092 /* Setup the dispatcher thread */
2093 ret = pthread_create(&live_dispatcher_thread, NULL,
2094 thread_dispatcher, (void *) NULL);
2095 if (ret) {
2096 errno = ret;
2097 PERROR("pthread_create viewer dispatcher");
2098 retval = -1;
2099 goto exit_dispatcher_thread;
2100 }
2101
2102 /* Setup the worker thread */
2103 ret = pthread_create(&live_worker_thread, NULL,
2104 thread_worker, NULL);
2105 if (ret) {
2106 errno = ret;
2107 PERROR("pthread_create viewer worker");
2108 retval = -1;
2109 goto exit_worker_thread;
2110 }
2111
2112 /* Setup the listener thread */
2113 ret = pthread_create(&live_listener_thread, NULL,
2114 thread_listener, (void *) NULL);
2115 if (ret) {
2116 errno = ret;
2117 PERROR("pthread_create viewer listener");
2118 retval = -1;
2119 goto exit_listener_thread;
2120 }
2121
2122 /*
2123 * All OK, started all threads.
2124 */
2125 return retval;
2126
2127 /*
2128 * Join on the live_listener_thread should anything be added after
2129 * the live_listener thread's creation.
2130 */
2131
2132 exit_listener_thread:
2133
2134 ret = pthread_join(live_worker_thread, &status);
2135 if (ret) {
2136 errno = ret;
2137 PERROR("pthread_join live worker");
2138 retval = -1;
2139 }
2140 exit_worker_thread:
2141
2142 ret = pthread_join(live_dispatcher_thread, &status);
2143 if (ret) {
2144 errno = ret;
2145 PERROR("pthread_join live dispatcher");
2146 retval = -1;
2147 }
2148 exit_dispatcher_thread:
2149
2150 exit_init_data:
2151 cleanup_relayd_live();
2152
2153 return retval;
2154 }
This page took 0.132065 seconds and 4 git commands to generate.