Refactor relayd main/set_options/cleanup
[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 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 *
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _GNU_SOURCE
20 #define _LGPL_SOURCE
21 #include <getopt.h>
22 #include <grp.h>
23 #include <limits.h>
24 #include <pthread.h>
25 #include <signal.h>
26 #include <stdio.h>
27 #include <stdlib.h>
28 #include <string.h>
29 #include <sys/mman.h>
30 #include <sys/mount.h>
31 #include <sys/resource.h>
32 #include <sys/socket.h>
33 #include <sys/stat.h>
34 #include <sys/types.h>
35 #include <sys/wait.h>
36 #include <inttypes.h>
37 #include <urcu/futex.h>
38 #include <urcu/uatomic.h>
39 #include <unistd.h>
40 #include <fcntl.h>
41 #include <config.h>
42
43 #include <lttng/lttng.h>
44 #include <common/common.h>
45 #include <common/compat/poll.h>
46 #include <common/compat/socket.h>
47 #include <common/compat/endian.h>
48 #include <common/defaults.h>
49 #include <common/futex.h>
50 #include <common/index/index.h>
51 #include <common/sessiond-comm/sessiond-comm.h>
52 #include <common/sessiond-comm/inet.h>
53 #include <common/sessiond-comm/relayd.h>
54 #include <common/uri.h>
55 #include <common/utils.h>
56
57 #include "cmd.h"
58 #include "live.h"
59 #include "lttng-relayd.h"
60 #include "utils.h"
61 #include "health-relayd.h"
62 #include "testpoint.h"
63 #include "viewer-stream.h"
64 #include "stream.h"
65 #include "session.h"
66 #include "ctf-trace.h"
67 #include "connection.h"
68
69 static struct lttng_uri *live_uri;
70
71 /*
72 * This pipe is used to inform the worker thread that a command is queued and
73 * ready to be processed.
74 */
75 static int live_conn_pipe[2] = { -1, -1 };
76
77 /* Shared between threads */
78 static int live_dispatch_thread_exit;
79
80 static pthread_t live_listener_thread;
81 static pthread_t live_dispatcher_thread;
82 static pthread_t live_worker_thread;
83
84 /*
85 * Relay command queue.
86 *
87 * The live_thread_listener and live_thread_dispatcher communicate with this
88 * queue.
89 */
90 static struct relay_conn_queue viewer_conn_queue;
91
92 static uint64_t last_relay_viewer_session_id;
93
94 /*
95 * Cleanup the daemon
96 */
97 static
98 void cleanup_relayd_live(void)
99 {
100 DBG("Cleaning up");
101
102 free(live_uri);
103 }
104
105 /*
106 * Receive a request buffer using a given socket, destination allocated buffer
107 * of length size.
108 *
109 * Return the size of the received message or else a negative value on error
110 * with errno being set by recvmsg() syscall.
111 */
112 static
113 ssize_t recv_request(struct lttcomm_sock *sock, void *buf, size_t size)
114 {
115 ssize_t ret;
116
117 assert(sock);
118 assert(buf);
119
120 ret = sock->ops->recvmsg(sock, buf, size, 0);
121 if (ret < 0 || ret != size) {
122 if (ret == 0) {
123 /* Orderly shutdown. Not necessary to print an error. */
124 DBG("Socket %d did an orderly shutdown", sock->fd);
125 } else {
126 ERR("Relay failed to receive request.");
127 }
128 ret = -1;
129 }
130
131 return ret;
132 }
133
134 /*
135 * Send a response buffer using a given socket, source allocated buffer of
136 * length size.
137 *
138 * Return the size of the sent message or else a negative value on error with
139 * errno being set by sendmsg() syscall.
140 */
141 static
142 ssize_t send_response(struct lttcomm_sock *sock, void *buf, size_t size)
143 {
144 ssize_t ret;
145
146 assert(sock);
147 assert(buf);
148
149 ret = sock->ops->sendmsg(sock, buf, size, 0);
150 if (ret < 0) {
151 ERR("Relayd failed to send response.");
152 }
153
154 return ret;
155 }
156
157 /*
158 * Atomically check if new streams got added in one of the sessions attached
159 * and reset the flag to 0.
160 *
161 * Returns 1 if new streams got added, 0 if nothing changed, a negative value
162 * on error.
163 */
164 static
165 int check_new_streams(struct relay_connection *conn)
166 {
167 struct relay_session *session;
168 unsigned long current_val;
169 int ret = 0;
170
171 if (!conn->viewer_session) {
172 goto end;
173 }
174 cds_list_for_each_entry(session,
175 &conn->viewer_session->sessions_head,
176 viewer_session_list) {
177 current_val = uatomic_cmpxchg(&session->new_streams, 1, 0);
178 ret = current_val;
179 if (ret == 1) {
180 goto end;
181 }
182 }
183
184 end:
185 return ret;
186 }
187
188 /*
189 * Send viewer streams to the given socket. The ignore_sent_flag indicates if
190 * this function should ignore the sent flag or not.
191 *
192 * Return 0 on success or else a negative value.
193 */
194 static
195 ssize_t send_viewer_streams(struct lttcomm_sock *sock,
196 struct relay_session *session, unsigned int ignore_sent_flag)
197 {
198 ssize_t ret;
199 struct lttng_viewer_stream send_stream;
200 struct lttng_ht_iter iter;
201 struct relay_viewer_stream *vstream;
202
203 assert(session);
204
205 rcu_read_lock();
206
207 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, vstream,
208 stream_n.node) {
209 struct ctf_trace *ctf_trace;
210
211 health_code_update();
212
213 /* Ignore if not the same session. */
214 if (vstream->session_id != session->id ||
215 (!ignore_sent_flag && vstream->sent_flag)) {
216 continue;
217 }
218
219 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
220 vstream->path_name);
221 assert(ctf_trace);
222
223 send_stream.id = htobe64(vstream->stream_handle);
224 send_stream.ctf_trace_id = htobe64(ctf_trace->id);
225 send_stream.metadata_flag = htobe32(vstream->metadata_flag);
226 strncpy(send_stream.path_name, vstream->path_name,
227 sizeof(send_stream.path_name));
228 strncpy(send_stream.channel_name, vstream->channel_name,
229 sizeof(send_stream.channel_name));
230
231 DBG("Sending stream %" PRIu64 " to viewer", vstream->stream_handle);
232 ret = send_response(sock, &send_stream, sizeof(send_stream));
233 if (ret < 0) {
234 goto end_unlock;
235 }
236 vstream->sent_flag = 1;
237 }
238
239 ret = 0;
240
241 end_unlock:
242 rcu_read_unlock();
243 return ret;
244 }
245
246 /*
247 * Create every viewer stream possible for the given session with the seek
248 * type. Three counters *can* be return which are in order the total amount of
249 * viewer stream of the session, the number of unsent stream and the number of
250 * stream created. Those counters can be NULL and thus will be ignored.
251 *
252 * Return 0 on success or else a negative value.
253 */
254 static
255 int make_viewer_streams(struct relay_session *session,
256 enum lttng_viewer_seek seek_t, uint32_t *nb_total, uint32_t *nb_unsent,
257 uint32_t *nb_created)
258 {
259 int ret;
260 struct lttng_ht_iter iter;
261 struct ctf_trace *ctf_trace;
262
263 assert(session);
264
265 /*
266 * This is to make sure we create viewer streams for a full received
267 * channel. For instance, if we have 8 streams for a channel that are
268 * concurrently being flagged ready, we can end up creating just a subset
269 * of the 8 streams (the ones that are flagged). This lock avoids this
270 * limbo state.
271 */
272 pthread_mutex_lock(&session->viewer_ready_lock);
273
274 /*
275 * Create viewer streams for relay streams that are ready to be used for a
276 * the given session id only.
277 */
278 rcu_read_lock();
279 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
280 node.node) {
281 struct relay_stream *stream;
282
283 health_code_update();
284
285 if (ctf_trace->invalid_flag) {
286 continue;
287 }
288
289 cds_list_for_each_entry(stream, &ctf_trace->stream_list, trace_list) {
290 struct relay_viewer_stream *vstream;
291
292 if (!stream->viewer_ready) {
293 continue;
294 }
295
296 vstream = viewer_stream_find_by_id(stream->stream_handle);
297 if (!vstream) {
298 vstream = viewer_stream_create(stream, seek_t, ctf_trace);
299 if (!vstream) {
300 ret = -1;
301 goto error_unlock;
302 }
303 /* Acquire reference to ctf_trace. */
304 ctf_trace_get_ref(ctf_trace);
305
306 if (nb_created) {
307 /* Update number of created stream counter. */
308 (*nb_created)++;
309 }
310 } else if (!vstream->sent_flag && nb_unsent) {
311 /* Update number of unsent stream counter. */
312 (*nb_unsent)++;
313 }
314 /* Update number of total stream counter. */
315 if (nb_total) {
316 (*nb_total)++;
317 }
318 }
319 }
320
321 ret = 0;
322
323 error_unlock:
324 rcu_read_unlock();
325 pthread_mutex_unlock(&session->viewer_ready_lock);
326 return ret;
327 }
328
329 /*
330 * Write to writable pipe used to notify a thread.
331 */
332 static
333 int notify_thread_pipe(int wpipe)
334 {
335 ssize_t ret;
336
337 ret = lttng_write(wpipe, "!", 1);
338 if (ret < 1) {
339 PERROR("write poll pipe");
340 }
341
342 return (int) ret;
343 }
344
345 /*
346 * Stop all threads by closing the thread quit pipe.
347 */
348 static
349 int stop_threads(void)
350 {
351 int ret, retval = 0;
352
353 /* Stopping all threads */
354 DBG("Terminating all live threads");
355 ret = notify_thread_pipe(thread_quit_pipe[1]);
356 if (ret < 0) {
357 ERR("write error on thread quit pipe");
358 retval = -1;
359 }
360
361 /* Dispatch thread */
362 CMM_STORE_SHARED(live_dispatch_thread_exit, 1);
363 futex_nto1_wake(&viewer_conn_queue.futex);
364 return retval;
365 }
366
367 /*
368 * Create a poll set with O_CLOEXEC and add the thread quit pipe to the set.
369 */
370 static
371 int create_thread_poll_set(struct lttng_poll_event *events, int size)
372 {
373 int ret;
374
375 if (events == NULL || size == 0) {
376 ret = -1;
377 goto error;
378 }
379
380 ret = lttng_poll_create(events, size, LTTNG_CLOEXEC);
381 if (ret < 0) {
382 goto error;
383 }
384
385 /* Add quit pipe */
386 ret = lttng_poll_add(events, thread_quit_pipe[0], LPOLLIN | LPOLLERR);
387 if (ret < 0) {
388 goto error;
389 }
390
391 return 0;
392
393 error:
394 return ret;
395 }
396
397 /*
398 * Check if the thread quit pipe was triggered.
399 *
400 * Return 1 if it was triggered else 0;
401 */
402 static
403 int check_thread_quit_pipe(int fd, uint32_t events)
404 {
405 if (fd == thread_quit_pipe[0] && (events & LPOLLIN)) {
406 return 1;
407 }
408
409 return 0;
410 }
411
412 /*
413 * Create and init socket from uri.
414 */
415 static
416 struct lttcomm_sock *init_socket(struct lttng_uri *uri)
417 {
418 int ret;
419 struct lttcomm_sock *sock = NULL;
420
421 sock = lttcomm_alloc_sock_from_uri(uri);
422 if (sock == NULL) {
423 ERR("Allocating socket");
424 goto error;
425 }
426
427 ret = lttcomm_create_sock(sock);
428 if (ret < 0) {
429 goto error;
430 }
431 DBG("Listening on sock %d for live", sock->fd);
432
433 ret = sock->ops->bind(sock);
434 if (ret < 0) {
435 goto error;
436 }
437
438 ret = sock->ops->listen(sock, -1);
439 if (ret < 0) {
440 goto error;
441
442 }
443
444 return sock;
445
446 error:
447 if (sock) {
448 lttcomm_destroy_sock(sock);
449 }
450 return NULL;
451 }
452
453 /*
454 * This thread manages the listening for new connections on the network
455 */
456 static
457 void *thread_listener(void *data)
458 {
459 int i, ret, pollfd, err = -1;
460 uint32_t revents, nb_fd;
461 struct lttng_poll_event events;
462 struct lttcomm_sock *live_control_sock;
463
464 DBG("[thread] Relay live listener started");
465
466 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_LISTENER);
467
468 health_code_update();
469
470 live_control_sock = init_socket(live_uri);
471 if (!live_control_sock) {
472 goto error_sock_control;
473 }
474
475 /* Pass 2 as size here for the thread quit pipe and control sockets. */
476 ret = create_thread_poll_set(&events, 2);
477 if (ret < 0) {
478 goto error_create_poll;
479 }
480
481 /* Add the control socket */
482 ret = lttng_poll_add(&events, live_control_sock->fd, LPOLLIN | LPOLLRDHUP);
483 if (ret < 0) {
484 goto error_poll_add;
485 }
486
487 lttng_relay_notify_ready();
488
489 if (testpoint(relayd_thread_live_listener)) {
490 goto error_testpoint;
491 }
492
493 while (1) {
494 health_code_update();
495
496 DBG("Listener accepting live viewers connections");
497
498 restart:
499 health_poll_entry();
500 ret = lttng_poll_wait(&events, -1);
501 health_poll_exit();
502 if (ret < 0) {
503 /*
504 * Restart interrupted system call.
505 */
506 if (errno == EINTR) {
507 goto restart;
508 }
509 goto error;
510 }
511 nb_fd = ret;
512
513 DBG("Relay new viewer connection received");
514 for (i = 0; i < nb_fd; i++) {
515 health_code_update();
516
517 /* Fetch once the poll data */
518 revents = LTTNG_POLL_GETEV(&events, i);
519 pollfd = LTTNG_POLL_GETFD(&events, i);
520
521 /* Thread quit pipe has been closed. Killing thread. */
522 ret = check_thread_quit_pipe(pollfd, revents);
523 if (ret) {
524 err = 0;
525 goto exit;
526 }
527
528 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
529 ERR("socket poll error");
530 goto error;
531 } else if (revents & LPOLLIN) {
532 /*
533 * Get allocated in this thread, enqueued to a global queue,
534 * dequeued and freed in the worker thread.
535 */
536 int val = 1;
537 struct relay_connection *new_conn;
538 struct lttcomm_sock *newsock;
539
540 new_conn = connection_create();
541 if (!new_conn) {
542 goto error;
543 }
544
545 newsock = live_control_sock->ops->accept(live_control_sock);
546 if (!newsock) {
547 PERROR("accepting control sock");
548 connection_free(new_conn);
549 goto error;
550 }
551 DBG("Relay viewer connection accepted socket %d", newsock->fd);
552
553 ret = setsockopt(newsock->fd, SOL_SOCKET, SO_REUSEADDR, &val,
554 sizeof(val));
555 if (ret < 0) {
556 PERROR("setsockopt inet");
557 lttcomm_destroy_sock(newsock);
558 connection_free(new_conn);
559 goto error;
560 }
561 new_conn->sock = newsock;
562
563 /* Enqueue request for the dispatcher thread. */
564 cds_wfcq_enqueue(&viewer_conn_queue.head, &viewer_conn_queue.tail,
565 &new_conn->qnode);
566
567 /*
568 * Wake the dispatch queue futex. Implicit memory barrier with
569 * the exchange in cds_wfcq_enqueue.
570 */
571 futex_nto1_wake(&viewer_conn_queue.futex);
572 }
573 }
574 }
575
576 exit:
577 error:
578 error_poll_add:
579 error_testpoint:
580 lttng_poll_clean(&events);
581 error_create_poll:
582 if (live_control_sock->fd >= 0) {
583 ret = live_control_sock->ops->close(live_control_sock);
584 if (ret) {
585 PERROR("close");
586 }
587 }
588 lttcomm_destroy_sock(live_control_sock);
589 error_sock_control:
590 if (err) {
591 health_error();
592 DBG("Live viewer listener thread exited with error");
593 }
594 health_unregister(health_relayd);
595 DBG("Live viewer listener thread cleanup complete");
596 if (stop_threads()) {
597 ERR("Error stopping live threads");
598 }
599 return NULL;
600 }
601
602 /*
603 * This thread manages the dispatching of the requests to worker threads
604 */
605 static
606 void *thread_dispatcher(void *data)
607 {
608 int err = -1;
609 ssize_t ret;
610 struct cds_wfcq_node *node;
611 struct relay_connection *conn = NULL;
612
613 DBG("[thread] Live viewer relay dispatcher started");
614
615 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_DISPATCHER);
616
617 if (testpoint(relayd_thread_live_dispatcher)) {
618 goto error_testpoint;
619 }
620
621 health_code_update();
622
623 while (!CMM_LOAD_SHARED(live_dispatch_thread_exit)) {
624 health_code_update();
625
626 /* Atomically prepare the queue futex */
627 futex_nto1_prepare(&viewer_conn_queue.futex);
628
629 do {
630 health_code_update();
631
632 /* Dequeue commands */
633 node = cds_wfcq_dequeue_blocking(&viewer_conn_queue.head,
634 &viewer_conn_queue.tail);
635 if (node == NULL) {
636 DBG("Woken up but nothing in the live-viewer "
637 "relay command queue");
638 /* Continue thread execution */
639 break;
640 }
641 conn = caa_container_of(node, struct relay_connection, qnode);
642 DBG("Dispatching viewer request waiting on sock %d",
643 conn->sock->fd);
644
645 /*
646 * Inform worker thread of the new request. This call is blocking
647 * so we can be assured that the data will be read at some point in
648 * time or wait to the end of the world :)
649 */
650 ret = lttng_write(live_conn_pipe[1], &conn, sizeof(conn));
651 if (ret < 0) {
652 PERROR("write conn pipe");
653 connection_destroy(conn);
654 goto error;
655 }
656 } while (node != NULL);
657
658 /* Futex wait on queue. Blocking call on futex() */
659 health_poll_entry();
660 futex_nto1_wait(&viewer_conn_queue.futex);
661 health_poll_exit();
662 }
663
664 /* Normal exit, no error */
665 err = 0;
666
667 error:
668 error_testpoint:
669 if (err) {
670 health_error();
671 ERR("Health error occurred in %s", __func__);
672 }
673 health_unregister(health_relayd);
674 DBG("Live viewer dispatch thread dying");
675 if (stop_threads()) {
676 ERR("Error stopping live threads");
677 }
678 return NULL;
679 }
680
681 /*
682 * Establish connection with the viewer and check the versions.
683 *
684 * Return 0 on success or else negative value.
685 */
686 static
687 int viewer_connect(struct relay_connection *conn)
688 {
689 int ret;
690 struct lttng_viewer_connect reply, msg;
691
692 assert(conn);
693
694 conn->version_check_done = 1;
695
696 health_code_update();
697
698 DBG("Viewer is establishing a connection to the relayd.");
699
700 ret = recv_request(conn->sock, &msg, sizeof(msg));
701 if (ret < 0) {
702 goto end;
703 }
704
705 health_code_update();
706
707 memset(&reply, 0, sizeof(reply));
708 reply.major = RELAYD_VERSION_COMM_MAJOR;
709 reply.minor = RELAYD_VERSION_COMM_MINOR;
710
711 /* Major versions must be the same */
712 if (reply.major != be32toh(msg.major)) {
713 DBG("Incompatible major versions ([relayd] %u vs [client] %u)",
714 reply.major, be32toh(msg.major));
715 ret = -1;
716 goto end;
717 }
718
719 conn->major = reply.major;
720 /* We adapt to the lowest compatible version */
721 if (reply.minor <= be32toh(msg.minor)) {
722 conn->minor = reply.minor;
723 } else {
724 conn->minor = be32toh(msg.minor);
725 }
726
727 if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_COMMAND) {
728 conn->type = RELAY_VIEWER_COMMAND;
729 } else if (be32toh(msg.type) == LTTNG_VIEWER_CLIENT_NOTIFICATION) {
730 conn->type = RELAY_VIEWER_NOTIFICATION;
731 } else {
732 ERR("Unknown connection type : %u", be32toh(msg.type));
733 ret = -1;
734 goto end;
735 }
736
737 reply.major = htobe32(reply.major);
738 reply.minor = htobe32(reply.minor);
739 if (conn->type == RELAY_VIEWER_COMMAND) {
740 reply.viewer_session_id = htobe64(++last_relay_viewer_session_id);
741 }
742
743 health_code_update();
744
745 ret = send_response(conn->sock, &reply, sizeof(reply));
746 if (ret < 0) {
747 goto end;
748 }
749
750 health_code_update();
751
752 DBG("Version check done using protocol %u.%u", conn->major, conn->minor);
753 ret = 0;
754
755 end:
756 return ret;
757 }
758
759 /*
760 * Send the viewer the list of current sessions.
761 *
762 * Return 0 on success or else a negative value.
763 */
764 static
765 int viewer_list_sessions(struct relay_connection *conn)
766 {
767 int ret;
768 struct lttng_viewer_list_sessions session_list;
769 unsigned long count;
770 long approx_before, approx_after;
771 struct lttng_ht_iter iter;
772 struct lttng_viewer_session send_session;
773 struct relay_session *session;
774
775 DBG("List sessions received");
776
777 rcu_read_lock();
778 cds_lfht_count_nodes(conn->sessions_ht->ht, &approx_before, &count,
779 &approx_after);
780 session_list.sessions_count = htobe32(count);
781
782 health_code_update();
783
784 ret = send_response(conn->sock, &session_list, sizeof(session_list));
785 if (ret < 0) {
786 goto end_unlock;
787 }
788
789 health_code_update();
790
791 cds_lfht_for_each_entry(conn->sessions_ht->ht, &iter.iter, session,
792 session_n.node) {
793 health_code_update();
794
795 strncpy(send_session.session_name, session->session_name,
796 sizeof(send_session.session_name));
797 strncpy(send_session.hostname, session->hostname,
798 sizeof(send_session.hostname));
799 send_session.id = htobe64(session->id);
800 send_session.live_timer = htobe32(session->live_timer);
801 send_session.clients = htobe32(session->viewer_refcount);
802 send_session.streams = htobe32(session->stream_count);
803
804 health_code_update();
805
806 ret = send_response(conn->sock, &send_session, sizeof(send_session));
807 if (ret < 0) {
808 goto end_unlock;
809 }
810 }
811 health_code_update();
812
813 ret = 0;
814 end_unlock:
815 rcu_read_unlock();
816 return ret;
817 }
818
819 /*
820 * Check if a connection is attached to a session.
821 * Return 1 if attached, 0 if not attached, a negative value on error.
822 */
823 static
824 int session_attached(struct relay_connection *conn, uint64_t session_id)
825 {
826 struct relay_session *session;
827 int found = 0;
828
829 if (!conn->viewer_session) {
830 goto end;
831 }
832 cds_list_for_each_entry(session,
833 &conn->viewer_session->sessions_head,
834 viewer_session_list) {
835 if (session->id == session_id) {
836 found = 1;
837 goto end;
838 }
839 }
840
841 end:
842 return found;
843 }
844
845 /*
846 * Delete all streams for a specific session ID.
847 */
848 static void destroy_viewer_streams_by_session(struct relay_session *session)
849 {
850 struct relay_viewer_stream *stream;
851 struct lttng_ht_iter iter;
852
853 assert(session);
854
855 rcu_read_lock();
856 cds_lfht_for_each_entry(viewer_streams_ht->ht, &iter.iter, stream,
857 stream_n.node) {
858 struct ctf_trace *ctf_trace;
859
860 health_code_update();
861 if (stream->session_id != session->id) {
862 continue;
863 }
864
865 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
866 stream->path_name);
867 assert(ctf_trace);
868
869 viewer_stream_delete(stream);
870
871 if (stream->metadata_flag) {
872 ctf_trace->metadata_sent = 0;
873 ctf_trace->viewer_metadata_stream = NULL;
874 }
875
876 viewer_stream_destroy(ctf_trace, stream);
877 }
878 rcu_read_unlock();
879 }
880
881 static void try_destroy_streams(struct relay_session *session)
882 {
883 struct ctf_trace *ctf_trace;
884 struct lttng_ht_iter iter;
885
886 assert(session);
887
888 cds_lfht_for_each_entry(session->ctf_traces_ht->ht, &iter.iter, ctf_trace,
889 node.node) {
890 /* Attempt to destroy the ctf trace of that session. */
891 ctf_trace_try_destroy(session, ctf_trace);
892 }
893 }
894
895 /*
896 * Cleanup a session.
897 */
898 static void cleanup_session(struct relay_connection *conn,
899 struct relay_session *session)
900 {
901 /*
902 * Very important that this is done before destroying the session so we
903 * can put back every viewer stream reference from the ctf_trace.
904 */
905 destroy_viewer_streams_by_session(session);
906 try_destroy_streams(session);
907 cds_list_del(&session->viewer_session_list);
908 session_viewer_try_destroy(conn->sessions_ht, session);
909 }
910
911 /*
912 * Send the viewer the list of current sessions.
913 */
914 static
915 int viewer_get_new_streams(struct relay_connection *conn)
916 {
917 int ret, send_streams = 0;
918 uint32_t nb_created = 0, nb_unsent = 0, nb_streams = 0;
919 struct lttng_viewer_new_streams_request request;
920 struct lttng_viewer_new_streams_response response;
921 struct relay_session *session;
922 uint64_t session_id;
923
924 assert(conn);
925
926 DBG("Get new streams received");
927
928 health_code_update();
929
930 /* Receive the request from the connected client. */
931 ret = recv_request(conn->sock, &request, sizeof(request));
932 if (ret < 0) {
933 goto error;
934 }
935 session_id = be64toh(request.session_id);
936
937 health_code_update();
938
939 memset(&response, 0, sizeof(response));
940
941 rcu_read_lock();
942 session = session_find_by_id(conn->sessions_ht, session_id);
943 if (!session) {
944 DBG("Relay session %" PRIu64 " not found", session_id);
945 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
946 goto send_reply;
947 }
948
949 if (!session_attached(conn, session_id)) {
950 send_streams = 0;
951 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_ERR);
952 goto send_reply;
953 }
954
955 send_streams = 1;
956 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_OK);
957
958 ret = make_viewer_streams(session, LTTNG_VIEWER_SEEK_LAST, NULL, &nb_unsent,
959 &nb_created);
960 if (ret < 0) {
961 goto end_unlock;
962 }
963 /* Only send back the newly created streams with the unsent ones. */
964 nb_streams = nb_created + nb_unsent;
965 response.streams_count = htobe32(nb_streams);
966
967 /*
968 * If the session is closed and we have no new streams to send,
969 * it means that the viewer has already received the whole trace
970 * for this session and should now close it.
971 */
972 if (nb_streams == 0 && session->close_flag) {
973 send_streams = 0;
974 response.status = htobe32(LTTNG_VIEWER_NEW_STREAMS_HUP);
975 /*
976 * Remove the session from the attached list of the connection
977 * and try to destroy it.
978 */
979 cds_list_del(&session->viewer_session_list);
980 cleanup_session(conn, session);
981 goto send_reply;
982 }
983
984 send_reply:
985 health_code_update();
986 ret = send_response(conn->sock, &response, sizeof(response));
987 if (ret < 0) {
988 goto end_unlock;
989 }
990 health_code_update();
991
992 /*
993 * Unknown or empty session, just return gracefully, the viewer knows what
994 * is happening.
995 */
996 if (!send_streams || !nb_streams) {
997 ret = 0;
998 goto end_unlock;
999 }
1000
1001 /*
1002 * Send stream and *DON'T* ignore the sent flag so every viewer streams
1003 * that were not sent from that point will be sent to the viewer.
1004 */
1005 ret = send_viewer_streams(conn->sock, session, 0);
1006 if (ret < 0) {
1007 goto end_unlock;
1008 }
1009
1010 end_unlock:
1011 rcu_read_unlock();
1012 error:
1013 return ret;
1014 }
1015
1016 /*
1017 * Send the viewer the list of current sessions.
1018 */
1019 static
1020 int viewer_attach_session(struct relay_connection *conn)
1021 {
1022 int send_streams = 0;
1023 ssize_t ret;
1024 uint32_t nb_streams = 0;
1025 enum lttng_viewer_seek seek_type;
1026 struct lttng_viewer_attach_session_request request;
1027 struct lttng_viewer_attach_session_response response;
1028 struct relay_session *session;
1029
1030 assert(conn);
1031
1032 health_code_update();
1033
1034 /* Receive the request from the connected client. */
1035 ret = recv_request(conn->sock, &request, sizeof(request));
1036 if (ret < 0) {
1037 goto error;
1038 }
1039
1040 health_code_update();
1041
1042 memset(&response, 0, sizeof(response));
1043
1044 if (!conn->viewer_session) {
1045 DBG("Client trying to attach before creating a live viewer session");
1046 response.status = htobe32(LTTNG_VIEWER_ATTACH_NO_SESSION);
1047 goto send_reply;
1048 }
1049
1050 rcu_read_lock();
1051 session = session_find_by_id(conn->sessions_ht,
1052 be64toh(request.session_id));
1053 if (!session) {
1054 DBG("Relay session %" PRIu64 " not found",
1055 be64toh(request.session_id));
1056 response.status = htobe32(LTTNG_VIEWER_ATTACH_UNK);
1057 goto send_reply;
1058 }
1059 session_viewer_attach(session);
1060 DBG("Attach session ID %" PRIu64 " received", be64toh(request.session_id));
1061
1062 if (uatomic_read(&session->viewer_refcount) > 1) {
1063 DBG("Already a viewer attached");
1064 response.status = htobe32(LTTNG_VIEWER_ATTACH_ALREADY);
1065 session_viewer_detach(session);
1066 goto send_reply;
1067 } else if (session->live_timer == 0) {
1068 DBG("Not live session");
1069 response.status = htobe32(LTTNG_VIEWER_ATTACH_NOT_LIVE);
1070 goto send_reply;
1071 } else {
1072 send_streams = 1;
1073 response.status = htobe32(LTTNG_VIEWER_ATTACH_OK);
1074 cds_list_add(&session->viewer_session_list,
1075 &conn->viewer_session->sessions_head);
1076 }
1077
1078 switch (be32toh(request.seek)) {
1079 case LTTNG_VIEWER_SEEK_BEGINNING:
1080 case LTTNG_VIEWER_SEEK_LAST:
1081 seek_type = be32toh(request.seek);
1082 break;
1083 default:
1084 ERR("Wrong seek parameter");
1085 response.status = htobe32(LTTNG_VIEWER_ATTACH_SEEK_ERR);
1086 send_streams = 0;
1087 goto send_reply;
1088 }
1089
1090 ret = make_viewer_streams(session, seek_type, &nb_streams, NULL, NULL);
1091 if (ret < 0) {
1092 goto end_unlock;
1093 }
1094 response.streams_count = htobe32(nb_streams);
1095
1096 send_reply:
1097 health_code_update();
1098 ret = send_response(conn->sock, &response, sizeof(response));
1099 if (ret < 0) {
1100 goto end_unlock;
1101 }
1102 health_code_update();
1103
1104 /*
1105 * Unknown or empty session, just return gracefully, the viewer knows what
1106 * is happening.
1107 */
1108 if (!send_streams || !nb_streams) {
1109 ret = 0;
1110 goto end_unlock;
1111 }
1112
1113 /* Send stream and ignore the sent flag. */
1114 ret = send_viewer_streams(conn->sock, session, 1);
1115 if (ret < 0) {
1116 goto end_unlock;
1117 }
1118
1119 end_unlock:
1120 rcu_read_unlock();
1121 error:
1122 return ret;
1123 }
1124
1125 /*
1126 * Open the index file if needed for the given vstream.
1127 *
1128 * If an index file is successfully opened, the index_read_fd of the stream is
1129 * set with it.
1130 *
1131 * Return 0 on success, a negative value on error (-ENOENT if not ready yet).
1132 */
1133 static int try_open_index(struct relay_viewer_stream *vstream,
1134 struct relay_stream *rstream)
1135 {
1136 int ret = 0;
1137
1138 assert(vstream);
1139 assert(rstream);
1140
1141 if (vstream->index_read_fd >= 0) {
1142 goto end;
1143 }
1144
1145 /*
1146 * First time, we open the index file and at least one index is ready. The
1147 * race between the read and write of the total_index_received is
1148 * acceptable here since the client will be notified to simply come back
1149 * and get the next index.
1150 */
1151 if (rstream->total_index_received <= 0) {
1152 ret = -ENOENT;
1153 goto end;
1154 }
1155 ret = index_open(vstream->path_name, vstream->channel_name,
1156 vstream->tracefile_count, vstream->tracefile_count_current);
1157 if (ret >= 0) {
1158 vstream->index_read_fd = ret;
1159 ret = 0;
1160 goto end;
1161 }
1162
1163 end:
1164 return ret;
1165 }
1166
1167 /*
1168 * Check the status of the index for the given stream. This function updates
1169 * the index structure if needed and can destroy the vstream also for the HUP
1170 * situation.
1171 *
1172 * Return 0 means that we can proceed with the index. A value of 1 means that
1173 * the index has been updated and is ready to be send to the client. A negative
1174 * value indicates an error that can't be handled.
1175 */
1176 static int check_index_status(struct relay_viewer_stream *vstream,
1177 struct relay_stream *rstream, struct ctf_trace *trace,
1178 struct lttng_viewer_index *index)
1179 {
1180 int ret;
1181
1182 assert(vstream);
1183 assert(rstream);
1184 assert(index);
1185 assert(trace);
1186
1187 if (!rstream->close_flag) {
1188 /* Rotate on abort (overwrite). */
1189 if (vstream->abort_flag) {
1190 DBG("Viewer stream %" PRIu64 " rotate because of overwrite",
1191 vstream->stream_handle);
1192 ret = viewer_stream_rotate(vstream, rstream);
1193 if (ret < 0) {
1194 goto error;
1195 } else if (ret == 1) {
1196 /* EOF */
1197 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1198 goto hup;
1199 }
1200 /* ret == 0 means successful so we continue. */
1201 }
1202
1203 /* Check if we are in the same trace file at this point. */
1204 if (rstream->tracefile_count_current == vstream->tracefile_count_current) {
1205 if (rstream->beacon_ts_end != -1ULL &&
1206 vstream->last_sent_index == rstream->total_index_received) {
1207 /*
1208 * We've received a synchronization beacon and the last index
1209 * available has been sent, the index for now is inactive.
1210 */
1211 index->status = htobe32(LTTNG_VIEWER_INDEX_INACTIVE);
1212 index->timestamp_end = htobe64(rstream->beacon_ts_end);
1213 index->stream_id = htobe64(rstream->ctf_stream_id);
1214 goto index_ready;
1215 } else if (rstream->total_index_received <= vstream->last_sent_index
1216 && !vstream->close_write_flag) {
1217 /*
1218 * Reader and writer are working in the same tracefile, so we care
1219 * about the number of index received and sent. Otherwise, we read
1220 * up to EOF.
1221 */
1222 index->status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1223 goto index_ready;
1224 }
1225 }
1226 /* Nothing to do with the index, continue with it. */
1227 ret = 0;
1228 } else if (rstream->close_flag && vstream->close_write_flag &&
1229 vstream->total_index_received == vstream->last_sent_index) {
1230 /* Last index sent and current tracefile closed in write */
1231 index->status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1232 goto hup;
1233 } else {
1234 vstream->close_write_flag = 1;
1235 ret = 0;
1236 }
1237
1238 error:
1239 return ret;
1240
1241 hup:
1242 viewer_stream_delete(vstream);
1243 viewer_stream_destroy(trace, vstream);
1244 index_ready:
1245 return 1;
1246 }
1247
1248 /*
1249 * Send the next index for a stream.
1250 *
1251 * Return 0 on success or else a negative value.
1252 */
1253 static
1254 int viewer_get_next_index(struct relay_connection *conn)
1255 {
1256 int ret;
1257 ssize_t read_ret;
1258 struct lttng_viewer_get_next_index request_index;
1259 struct lttng_viewer_index viewer_index;
1260 struct ctf_packet_index packet_index;
1261 struct relay_viewer_stream *vstream;
1262 struct relay_stream *rstream;
1263 struct ctf_trace *ctf_trace;
1264 struct relay_session *session;
1265
1266 assert(conn);
1267
1268 DBG("Viewer get next index");
1269
1270 health_code_update();
1271
1272 ret = recv_request(conn->sock, &request_index, sizeof(request_index));
1273 if (ret < 0) {
1274 goto end;
1275 }
1276 health_code_update();
1277
1278 rcu_read_lock();
1279 vstream = viewer_stream_find_by_id(be64toh(request_index.stream_id));
1280 if (!vstream) {
1281 ret = -1;
1282 goto end_unlock;
1283 }
1284
1285 session = session_find_by_id(conn->sessions_ht, vstream->session_id);
1286 if (!session) {
1287 ret = -1;
1288 goto end_unlock;
1289 }
1290
1291 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht, vstream->path_name);
1292 assert(ctf_trace);
1293
1294 memset(&viewer_index, 0, sizeof(viewer_index));
1295
1296 /*
1297 * The viewer should not ask for index on metadata stream.
1298 */
1299 if (vstream->metadata_flag) {
1300 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1301 goto send_reply;
1302 }
1303
1304 rstream = stream_find_by_id(relay_streams_ht, vstream->stream_handle);
1305 assert(rstream);
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 packet arrives, it
1313 * might not be ready at the beginning of the session
1314 */
1315 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1316 } else {
1317 /* Unhandled error. */
1318 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1319 }
1320 goto send_reply;
1321 }
1322
1323 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1324 ret = check_index_status(vstream, rstream, ctf_trace, &viewer_index);
1325 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1326 if (ret < 0) {
1327 goto end_unlock;
1328 } else if (ret == 1) {
1329 /*
1330 * This means the viewer index data structure has been populated by the
1331 * check call thus we now send back the reply to the client.
1332 */
1333 goto send_reply;
1334 }
1335 /* At this point, ret MUST be 0 thus we continue with the get. */
1336 assert(!ret);
1337
1338 if (!ctf_trace->metadata_received ||
1339 ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1340 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1341 }
1342
1343 ret = check_new_streams(conn);
1344 if (ret < 0) {
1345 goto end_unlock;
1346 } else if (ret == 1) {
1347 viewer_index.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1348 }
1349
1350 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1351 pthread_mutex_lock(&vstream->overwrite_lock);
1352 if (vstream->abort_flag) {
1353 /* The file is being overwritten by the writer, we cannot use it. */
1354 pthread_mutex_unlock(&vstream->overwrite_lock);
1355 ret = viewer_stream_rotate(vstream, rstream);
1356 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1357 if (ret < 0) {
1358 goto end_unlock;
1359 } else if (ret == 1) {
1360 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1361 viewer_stream_delete(vstream);
1362 viewer_stream_destroy(ctf_trace, vstream);
1363 } else {
1364 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1365 }
1366 goto send_reply;
1367 }
1368
1369 read_ret = lttng_read(vstream->index_read_fd, &packet_index,
1370 sizeof(packet_index));
1371 pthread_mutex_unlock(&vstream->overwrite_lock);
1372 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1373 if (read_ret < 0) {
1374 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1375 viewer_stream_delete(vstream);
1376 viewer_stream_destroy(ctf_trace, vstream);
1377 goto send_reply;
1378 } else if (read_ret < sizeof(packet_index)) {
1379 pthread_mutex_lock(&rstream->viewer_stream_rotation_lock);
1380 if (vstream->close_write_flag) {
1381 ret = viewer_stream_rotate(vstream, rstream);
1382 if (ret < 0) {
1383 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1384 goto end_unlock;
1385 } else if (ret == 1) {
1386 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_HUP);
1387 viewer_stream_delete(vstream);
1388 viewer_stream_destroy(ctf_trace, vstream);
1389 } else {
1390 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_RETRY);
1391 }
1392 } else {
1393 ERR("Relay reading index file %d", vstream->index_read_fd);
1394 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_ERR);
1395 }
1396 pthread_mutex_unlock(&rstream->viewer_stream_rotation_lock);
1397 goto send_reply;
1398 } else {
1399 viewer_index.status = htobe32(LTTNG_VIEWER_INDEX_OK);
1400 vstream->last_sent_index++;
1401 }
1402
1403 /*
1404 * Indexes are stored in big endian, no need to switch before sending.
1405 */
1406 viewer_index.offset = packet_index.offset;
1407 viewer_index.packet_size = packet_index.packet_size;
1408 viewer_index.content_size = packet_index.content_size;
1409 viewer_index.timestamp_begin = packet_index.timestamp_begin;
1410 viewer_index.timestamp_end = packet_index.timestamp_end;
1411 viewer_index.events_discarded = packet_index.events_discarded;
1412 viewer_index.stream_id = packet_index.stream_id;
1413
1414 send_reply:
1415 viewer_index.flags = htobe32(viewer_index.flags);
1416 health_code_update();
1417
1418 ret = send_response(conn->sock, &viewer_index, sizeof(viewer_index));
1419 if (ret < 0) {
1420 goto end_unlock;
1421 }
1422 health_code_update();
1423
1424 DBG("Index %" PRIu64 " for stream %" PRIu64 " sent",
1425 vstream->last_sent_index, vstream->stream_handle);
1426
1427 end_unlock:
1428 rcu_read_unlock();
1429
1430 end:
1431 return ret;
1432 }
1433
1434 /*
1435 * Send the next index for a stream
1436 *
1437 * Return 0 on success or else a negative value.
1438 */
1439 static
1440 int viewer_get_packet(struct relay_connection *conn)
1441 {
1442 int ret, send_data = 0;
1443 char *data = NULL;
1444 uint32_t len = 0;
1445 ssize_t read_len;
1446 struct lttng_viewer_get_packet get_packet_info;
1447 struct lttng_viewer_trace_packet reply;
1448 struct relay_viewer_stream *stream;
1449 struct relay_session *session;
1450 struct ctf_trace *ctf_trace;
1451
1452 assert(conn);
1453
1454 DBG2("Relay get data packet");
1455
1456 health_code_update();
1457
1458 ret = recv_request(conn->sock, &get_packet_info, sizeof(get_packet_info));
1459 if (ret < 0) {
1460 goto end;
1461 }
1462 health_code_update();
1463
1464 /* From this point on, the error label can be reached. */
1465 memset(&reply, 0, sizeof(reply));
1466
1467 rcu_read_lock();
1468 stream = viewer_stream_find_by_id(be64toh(get_packet_info.stream_id));
1469 if (!stream) {
1470 goto error;
1471 }
1472
1473 session = session_find_by_id(conn->sessions_ht, stream->session_id);
1474 if (!session) {
1475 ret = -1;
1476 goto error;
1477 }
1478
1479 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1480 stream->path_name);
1481 assert(ctf_trace);
1482
1483 /*
1484 * First time we read this stream, we need open the tracefile, we should
1485 * only arrive here if an index has already been sent to the viewer, so the
1486 * tracefile must exist, if it does not it is a fatal error.
1487 */
1488 if (stream->read_fd < 0) {
1489 char fullpath[PATH_MAX];
1490
1491 if (stream->tracefile_count > 0) {
1492 ret = snprintf(fullpath, PATH_MAX, "%s/%s_%" PRIu64, stream->path_name,
1493 stream->channel_name,
1494 stream->tracefile_count_current);
1495 } else {
1496 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1497 stream->channel_name);
1498 }
1499 if (ret < 0) {
1500 goto error;
1501 }
1502 ret = open(fullpath, O_RDONLY);
1503 if (ret < 0) {
1504 PERROR("Relay opening trace file");
1505 goto error;
1506 }
1507 stream->read_fd = ret;
1508 }
1509
1510 if (!ctf_trace->metadata_received ||
1511 ctf_trace->metadata_received > ctf_trace->metadata_sent) {
1512 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1513 reply.flags |= LTTNG_VIEWER_FLAG_NEW_METADATA;
1514 goto send_reply;
1515 }
1516
1517 ret = check_new_streams(conn);
1518 if (ret < 0) {
1519 goto end_unlock;
1520 } else if (ret == 1) {
1521 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1522 reply.flags |= LTTNG_VIEWER_FLAG_NEW_STREAM;
1523 goto send_reply;
1524 }
1525
1526 len = be32toh(get_packet_info.len);
1527 data = zmalloc(len);
1528 if (!data) {
1529 PERROR("relay data zmalloc");
1530 goto error;
1531 }
1532
1533 ret = lseek(stream->read_fd, be64toh(get_packet_info.offset), SEEK_SET);
1534 if (ret < 0) {
1535 /*
1536 * If the read fd was closed by the streaming side, the
1537 * abort_flag will be set to 1, otherwise it is an error.
1538 */
1539 if (stream->abort_flag == 0) {
1540 PERROR("lseek");
1541 goto error;
1542 }
1543 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1544 goto send_reply;
1545 }
1546 read_len = lttng_read(stream->read_fd, data, len);
1547 if (read_len < len) {
1548 /*
1549 * If the read fd was closed by the streaming side, the
1550 * abort_flag will be set to 1, otherwise it is an error.
1551 */
1552 if (stream->abort_flag == 0) {
1553 PERROR("Relay reading trace file, fd: %d, offset: %" PRIu64,
1554 stream->read_fd,
1555 be64toh(get_packet_info.offset));
1556 goto error;
1557 } else {
1558 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_EOF);
1559 goto send_reply;
1560 }
1561 }
1562 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_OK);
1563 reply.len = htobe32(len);
1564 send_data = 1;
1565 goto send_reply;
1566
1567 error:
1568 reply.status = htobe32(LTTNG_VIEWER_GET_PACKET_ERR);
1569
1570 send_reply:
1571 reply.flags = htobe32(reply.flags);
1572
1573 health_code_update();
1574
1575 ret = send_response(conn->sock, &reply, sizeof(reply));
1576 if (ret < 0) {
1577 goto end_unlock;
1578 }
1579 health_code_update();
1580
1581 if (send_data) {
1582 health_code_update();
1583 ret = send_response(conn->sock, data, len);
1584 if (ret < 0) {
1585 goto end_unlock;
1586 }
1587 health_code_update();
1588 }
1589
1590 DBG("Sent %u bytes for stream %" PRIu64, len,
1591 be64toh(get_packet_info.stream_id));
1592
1593 end_unlock:
1594 free(data);
1595 rcu_read_unlock();
1596
1597 end:
1598 return ret;
1599 }
1600
1601 /*
1602 * Send the session's metadata
1603 *
1604 * Return 0 on success else a negative value.
1605 */
1606 static
1607 int viewer_get_metadata(struct relay_connection *conn)
1608 {
1609 int ret = 0;
1610 ssize_t read_len;
1611 uint64_t len = 0;
1612 char *data = NULL;
1613 struct lttng_viewer_get_metadata request;
1614 struct lttng_viewer_metadata_packet reply;
1615 struct relay_viewer_stream *stream;
1616 struct ctf_trace *ctf_trace;
1617 struct relay_session *session;
1618
1619 assert(conn);
1620
1621 DBG("Relay get metadata");
1622
1623 health_code_update();
1624
1625 ret = recv_request(conn->sock, &request, sizeof(request));
1626 if (ret < 0) {
1627 goto end;
1628 }
1629 health_code_update();
1630
1631 memset(&reply, 0, sizeof(reply));
1632
1633 rcu_read_lock();
1634 stream = viewer_stream_find_by_id(be64toh(request.stream_id));
1635 if (!stream || !stream->metadata_flag) {
1636 ERR("Invalid metadata stream");
1637 goto error;
1638 }
1639
1640 session = session_find_by_id(conn->sessions_ht, stream->session_id);
1641 if (!session) {
1642 ret = -1;
1643 goto error;
1644 }
1645
1646 ctf_trace = ctf_trace_find_by_path(session->ctf_traces_ht,
1647 stream->path_name);
1648 assert(ctf_trace);
1649 assert(ctf_trace->metadata_sent <= ctf_trace->metadata_received);
1650
1651 len = ctf_trace->metadata_received - ctf_trace->metadata_sent;
1652 if (len == 0) {
1653 reply.status = htobe32(LTTNG_VIEWER_NO_NEW_METADATA);
1654 goto send_reply;
1655 }
1656
1657 /* first time, we open the metadata file */
1658 if (stream->read_fd < 0) {
1659 char fullpath[PATH_MAX];
1660
1661 ret = snprintf(fullpath, PATH_MAX, "%s/%s", stream->path_name,
1662 stream->channel_name);
1663 if (ret < 0) {
1664 goto error;
1665 }
1666 ret = open(fullpath, O_RDONLY);
1667 if (ret < 0) {
1668 PERROR("Relay opening metadata file");
1669 goto error;
1670 }
1671 stream->read_fd = ret;
1672 }
1673
1674 reply.len = htobe64(len);
1675 data = zmalloc(len);
1676 if (!data) {
1677 PERROR("viewer metadata zmalloc");
1678 goto error;
1679 }
1680
1681 read_len = lttng_read(stream->read_fd, data, len);
1682 if (read_len < len) {
1683 PERROR("Relay reading metadata file");
1684 goto error;
1685 }
1686 ctf_trace->metadata_sent += read_len;
1687 reply.status = htobe32(LTTNG_VIEWER_METADATA_OK);
1688 goto send_reply;
1689
1690 error:
1691 reply.status = htobe32(LTTNG_VIEWER_METADATA_ERR);
1692
1693 send_reply:
1694 health_code_update();
1695 ret = send_response(conn->sock, &reply, sizeof(reply));
1696 if (ret < 0) {
1697 goto end_unlock;
1698 }
1699 health_code_update();
1700
1701 if (len > 0) {
1702 ret = send_response(conn->sock, data, len);
1703 if (ret < 0) {
1704 goto end_unlock;
1705 }
1706 }
1707
1708 DBG("Sent %" PRIu64 " bytes of metadata for stream %" PRIu64, len,
1709 be64toh(request.stream_id));
1710
1711 DBG("Metadata sent");
1712
1713 end_unlock:
1714 free(data);
1715 rcu_read_unlock();
1716 end:
1717 return ret;
1718 }
1719
1720 /*
1721 * Create a viewer session.
1722 *
1723 * Return 0 on success or else a negative value.
1724 */
1725 static
1726 int viewer_create_session(struct relay_connection *conn)
1727 {
1728 int ret;
1729 struct lttng_viewer_create_session_response resp;
1730
1731 DBG("Viewer create session received");
1732
1733 memset(&resp, 0, sizeof(resp));
1734 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_OK);
1735 conn->viewer_session = zmalloc(sizeof(*conn->viewer_session));
1736 if (!conn->viewer_session) {
1737 ERR("Allocation viewer session");
1738 resp.status = htobe32(LTTNG_VIEWER_CREATE_SESSION_ERR);
1739 goto send_reply;
1740 }
1741 CDS_INIT_LIST_HEAD(&conn->viewer_session->sessions_head);
1742
1743 send_reply:
1744 health_code_update();
1745 ret = send_response(conn->sock, &resp, sizeof(resp));
1746 if (ret < 0) {
1747 goto end;
1748 }
1749 health_code_update();
1750 ret = 0;
1751
1752 end:
1753 return ret;
1754 }
1755
1756
1757 /*
1758 * live_relay_unknown_command: send -1 if received unknown command
1759 */
1760 static
1761 void live_relay_unknown_command(struct relay_connection *conn)
1762 {
1763 struct lttcomm_relayd_generic_reply reply;
1764
1765 memset(&reply, 0, sizeof(reply));
1766 reply.ret_code = htobe32(LTTNG_ERR_UNK);
1767 (void) send_response(conn->sock, &reply, sizeof(reply));
1768 }
1769
1770 /*
1771 * Process the commands received on the control socket
1772 */
1773 static
1774 int process_control(struct lttng_viewer_cmd *recv_hdr,
1775 struct relay_connection *conn)
1776 {
1777 int ret = 0;
1778 uint32_t msg_value;
1779
1780 assert(recv_hdr);
1781 assert(conn);
1782
1783 msg_value = be32toh(recv_hdr->cmd);
1784
1785 /*
1786 * Make sure we've done the version check before any command other then a
1787 * new client connection.
1788 */
1789 if (msg_value != LTTNG_VIEWER_CONNECT && !conn->version_check_done) {
1790 ERR("Viewer conn value %" PRIu32 " before version check", msg_value);
1791 ret = -1;
1792 goto end;
1793 }
1794
1795 switch (msg_value) {
1796 case LTTNG_VIEWER_CONNECT:
1797 ret = viewer_connect(conn);
1798 break;
1799 case LTTNG_VIEWER_LIST_SESSIONS:
1800 ret = viewer_list_sessions(conn);
1801 break;
1802 case LTTNG_VIEWER_ATTACH_SESSION:
1803 ret = viewer_attach_session(conn);
1804 break;
1805 case LTTNG_VIEWER_GET_NEXT_INDEX:
1806 ret = viewer_get_next_index(conn);
1807 break;
1808 case LTTNG_VIEWER_GET_PACKET:
1809 ret = viewer_get_packet(conn);
1810 break;
1811 case LTTNG_VIEWER_GET_METADATA:
1812 ret = viewer_get_metadata(conn);
1813 break;
1814 case LTTNG_VIEWER_GET_NEW_STREAMS:
1815 ret = viewer_get_new_streams(conn);
1816 break;
1817 case LTTNG_VIEWER_CREATE_SESSION:
1818 ret = viewer_create_session(conn);
1819 break;
1820 default:
1821 ERR("Received unknown viewer command (%u)", be32toh(recv_hdr->cmd));
1822 live_relay_unknown_command(conn);
1823 ret = -1;
1824 goto end;
1825 }
1826
1827 end:
1828 return ret;
1829 }
1830
1831 static
1832 void cleanup_connection_pollfd(struct lttng_poll_event *events, int pollfd)
1833 {
1834 int ret;
1835
1836 assert(events);
1837
1838 (void) lttng_poll_del(events, pollfd);
1839
1840 ret = close(pollfd);
1841 if (ret < 0) {
1842 ERR("Closing pollfd %d", pollfd);
1843 }
1844 }
1845
1846 /*
1847 * Delete and destroy a connection.
1848 *
1849 * RCU read side lock MUST be acquired.
1850 */
1851 static void destroy_connection(struct lttng_ht *relay_connections_ht,
1852 struct relay_connection *conn)
1853 {
1854 struct relay_session *session, *tmp_session;
1855
1856 assert(relay_connections_ht);
1857 assert(conn);
1858
1859 connection_delete(relay_connections_ht, conn);
1860
1861 if (!conn->viewer_session) {
1862 goto end;
1863 }
1864
1865 rcu_read_lock();
1866 cds_list_for_each_entry_safe(session, tmp_session,
1867 &conn->viewer_session->sessions_head,
1868 viewer_session_list) {
1869 DBG("Cleaning connection of session ID %" PRIu64, session->id);
1870 cleanup_session(conn, session);
1871 }
1872 rcu_read_unlock();
1873
1874 end:
1875 connection_destroy(conn);
1876 }
1877
1878 /*
1879 * This thread does the actual work
1880 */
1881 static
1882 void *thread_worker(void *data)
1883 {
1884 int ret, err = -1;
1885 uint32_t nb_fd;
1886 struct relay_connection *conn;
1887 struct lttng_poll_event events;
1888 struct lttng_ht *relay_connections_ht;
1889 struct lttng_ht_iter iter;
1890 struct lttng_viewer_cmd recv_hdr;
1891 struct relay_local_data *relay_ctx = (struct relay_local_data *) data;
1892 struct lttng_ht *sessions_ht = relay_ctx->sessions_ht;
1893
1894 DBG("[thread] Live viewer relay worker started");
1895
1896 rcu_register_thread();
1897
1898 health_register(health_relayd, HEALTH_RELAYD_TYPE_LIVE_WORKER);
1899
1900 if (testpoint(relayd_thread_live_worker)) {
1901 goto error_testpoint;
1902 }
1903
1904 /* table of connections indexed on socket */
1905 relay_connections_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1906 if (!relay_connections_ht) {
1907 goto relay_connections_ht_error;
1908 }
1909
1910 ret = create_thread_poll_set(&events, 2);
1911 if (ret < 0) {
1912 goto error_poll_create;
1913 }
1914
1915 ret = lttng_poll_add(&events, live_conn_pipe[0], LPOLLIN | LPOLLRDHUP);
1916 if (ret < 0) {
1917 goto error;
1918 }
1919
1920 restart:
1921 while (1) {
1922 int i;
1923
1924 health_code_update();
1925
1926 /* Infinite blocking call, waiting for transmission */
1927 DBG3("Relayd live viewer worker thread polling...");
1928 health_poll_entry();
1929 ret = lttng_poll_wait(&events, -1);
1930 health_poll_exit();
1931 if (ret < 0) {
1932 /*
1933 * Restart interrupted system call.
1934 */
1935 if (errno == EINTR) {
1936 goto restart;
1937 }
1938 goto error;
1939 }
1940
1941 nb_fd = ret;
1942
1943 /*
1944 * Process control. The control connection is prioritised so we don't
1945 * starve it with high throughput tracing data on the data
1946 * connection.
1947 */
1948 for (i = 0; i < nb_fd; i++) {
1949 /* Fetch once the poll data */
1950 uint32_t revents = LTTNG_POLL_GETEV(&events, i);
1951 int pollfd = LTTNG_POLL_GETFD(&events, i);
1952
1953 health_code_update();
1954
1955 /* Thread quit pipe has been closed. Killing thread. */
1956 ret = check_thread_quit_pipe(pollfd, revents);
1957 if (ret) {
1958 err = 0;
1959 goto exit;
1960 }
1961
1962 /* Inspect the relay conn pipe for new connection */
1963 if (pollfd == live_conn_pipe[0]) {
1964 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1965 ERR("Relay live pipe error");
1966 goto error;
1967 } else if (revents & LPOLLIN) {
1968 ret = lttng_read(live_conn_pipe[0], &conn, sizeof(conn));
1969 if (ret < 0) {
1970 goto error;
1971 }
1972 conn->sessions_ht = sessions_ht;
1973 connection_init(conn);
1974 lttng_poll_add(&events, conn->sock->fd,
1975 LPOLLIN | LPOLLRDHUP);
1976 rcu_read_lock();
1977 lttng_ht_add_unique_ulong(relay_connections_ht,
1978 &conn->sock_n);
1979 rcu_read_unlock();
1980 DBG("Connection socket %d added", conn->sock->fd);
1981 }
1982 } else {
1983 rcu_read_lock();
1984 conn = connection_find_by_sock(relay_connections_ht, pollfd);
1985 /* If not found, there is a synchronization issue. */
1986 assert(conn);
1987
1988 if (revents & (LPOLLERR | LPOLLHUP | LPOLLRDHUP)) {
1989 cleanup_connection_pollfd(&events, pollfd);
1990 destroy_connection(relay_connections_ht, conn);
1991 } else if (revents & LPOLLIN) {
1992 ret = conn->sock->ops->recvmsg(conn->sock, &recv_hdr,
1993 sizeof(recv_hdr), 0);
1994 if (ret <= 0) {
1995 /* Connection closed */
1996 cleanup_connection_pollfd(&events, pollfd);
1997 destroy_connection(relay_connections_ht, conn);
1998 DBG("Viewer control conn closed with %d", pollfd);
1999 } else {
2000 ret = process_control(&recv_hdr, conn);
2001 if (ret < 0) {
2002 /* Clear the session on error. */
2003 cleanup_connection_pollfd(&events, pollfd);
2004 destroy_connection(relay_connections_ht, conn);
2005 DBG("Viewer connection closed with %d", pollfd);
2006 }
2007 }
2008 }
2009 rcu_read_unlock();
2010 }
2011 }
2012 }
2013
2014 exit:
2015 error:
2016 lttng_poll_clean(&events);
2017
2018 /* Cleanup reamaining connection object. */
2019 rcu_read_lock();
2020 cds_lfht_for_each_entry(relay_connections_ht->ht, &iter.iter, conn,
2021 sock_n.node) {
2022 health_code_update();
2023 destroy_connection(relay_connections_ht, conn);
2024 }
2025 rcu_read_unlock();
2026 error_poll_create:
2027 lttng_ht_destroy(relay_connections_ht);
2028 relay_connections_ht_error:
2029 /* Close relay conn pipes */
2030 utils_close_pipe(live_conn_pipe);
2031 if (err) {
2032 DBG("Viewer worker thread exited with error");
2033 }
2034 DBG("Viewer worker thread cleanup complete");
2035 error_testpoint:
2036 if (err) {
2037 health_error();
2038 ERR("Health error occurred in %s", __func__);
2039 }
2040 health_unregister(health_relayd);
2041 if (stop_threads()) {
2042 ERR("Error stopping live threads");
2043 }
2044 rcu_unregister_thread();
2045 return NULL;
2046 }
2047
2048 /*
2049 * Create the relay command pipe to wake thread_manage_apps.
2050 * Closed in cleanup().
2051 */
2052 static int create_conn_pipe(void)
2053 {
2054 return utils_create_pipe_cloexec(live_conn_pipe);
2055 }
2056
2057 int relayd_live_stop(void)
2058 {
2059 return stop_threads();
2060 }
2061
2062 int relayd_live_join(void)
2063 {
2064 int ret, retval = 0;
2065 void *status;
2066
2067 ret = pthread_join(live_listener_thread, &status);
2068 if (ret) {
2069 errno = ret;
2070 PERROR("pthread_join live listener");
2071 retval = -1;
2072 }
2073
2074 ret = pthread_join(live_worker_thread, &status);
2075 if (ret) {
2076 errno = ret;
2077 PERROR("pthread_join live worker");
2078 retval = -1;
2079 }
2080
2081 ret = pthread_join(live_dispatcher_thread, &status);
2082 if (ret) {
2083 errno = ret;
2084 PERROR("pthread_join live dispatcher");
2085 retval = -1;
2086 }
2087
2088 cleanup_relayd_live();
2089
2090 return retval;
2091 }
2092
2093 /*
2094 * main
2095 */
2096 int relayd_live_create(struct lttng_uri *uri,
2097 struct relay_local_data *relay_ctx)
2098 {
2099 int ret = 0, retval = 0;
2100 void *status;
2101 int is_root;
2102
2103 if (!uri) {
2104 retval = -1;
2105 goto exit_init_data;
2106 }
2107 live_uri = uri;
2108
2109 /* Check if daemon is UID = 0 */
2110 is_root = !getuid();
2111
2112 if (!is_root) {
2113 if (live_uri->port < 1024) {
2114 ERR("Need to be root to use ports < 1024");
2115 retval = -1;
2116 goto exit_init_data;
2117 }
2118 }
2119
2120 /* Setup the thread apps communication pipe. */
2121 if (create_conn_pipe()) {
2122 retval = -1;
2123 goto exit_init_data;
2124 }
2125
2126 /* Init relay command queue. */
2127 cds_wfcq_init(&viewer_conn_queue.head, &viewer_conn_queue.tail);
2128
2129 /* Set up max poll set size */
2130 lttng_poll_set_max_size();
2131
2132 /* Setup the dispatcher thread */
2133 ret = pthread_create(&live_dispatcher_thread, NULL,
2134 thread_dispatcher, (void *) NULL);
2135 if (ret) {
2136 errno = ret;
2137 PERROR("pthread_create viewer dispatcher");
2138 retval = -1;
2139 goto exit_dispatcher_thread;
2140 }
2141
2142 /* Setup the worker thread */
2143 ret = pthread_create(&live_worker_thread, NULL,
2144 thread_worker, relay_ctx);
2145 if (ret) {
2146 errno = ret;
2147 PERROR("pthread_create viewer worker");
2148 retval = -1;
2149 goto exit_worker_thread;
2150 }
2151
2152 /* Setup the listener thread */
2153 ret = pthread_create(&live_listener_thread, NULL,
2154 thread_listener, (void *) NULL);
2155 if (ret) {
2156 errno = ret;
2157 PERROR("pthread_create viewer listener");
2158 retval = -1;
2159 goto exit_listener_thread;
2160 }
2161
2162 /*
2163 * All OK, started all threads.
2164 */
2165 return retval;
2166
2167
2168 ret = pthread_join(live_listener_thread, &status);
2169 if (ret) {
2170 errno = ret;
2171 PERROR("pthread_join live listener");
2172 retval = -1;
2173 }
2174 exit_listener_thread:
2175
2176 ret = pthread_join(live_worker_thread, &status);
2177 if (ret) {
2178 errno = ret;
2179 PERROR("pthread_join live worker");
2180 retval = -1;
2181 }
2182 exit_worker_thread:
2183
2184 ret = pthread_join(live_dispatcher_thread, &status);
2185 if (ret) {
2186 errno = ret;
2187 PERROR("pthread_join live dispatcher");
2188 retval = -1;
2189 }
2190 exit_dispatcher_thread:
2191
2192 exit_init_data:
2193 cleanup_relayd_live();
2194
2195 return retval;
2196 }
This page took 0.110792 seconds and 5 git commands to generate.