relayd: create stream files relative to a session's trace chunk
[lttng-tools.git] / src / bin / lttng-sessiond / consumer.c
1 /*
2 * Copyright (C) 2012 - David Goulet <dgoulet@efficios.com>
3 * 2018 - Jérémie Galarneau <jeremie.galarneau@efficios.com>
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License, version 2 only, as
7 * 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 with
15 * this program; if not, write to the Free Software Foundation, Inc., 51
16 * Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
17 */
18
19 #define _LGPL_SOURCE
20 #include <assert.h>
21 #include <stdio.h>
22 #include <stdlib.h>
23 #include <string.h>
24 #include <sys/stat.h>
25 #include <sys/types.h>
26 #include <unistd.h>
27 #include <inttypes.h>
28
29 #include <common/common.h>
30 #include <common/defaults.h>
31 #include <common/uri.h>
32 #include <common/relayd/relayd.h>
33 #include <common/string-utils/format.h>
34
35 #include "consumer.h"
36 #include "health-sessiond.h"
37 #include "ust-app.h"
38 #include "utils.h"
39 #include "lttng-sessiond.h"
40
41 /*
42 * Send a data payload using a given consumer socket of size len.
43 *
44 * The consumer socket lock MUST be acquired before calling this since this
45 * function can change the fd value.
46 *
47 * Return 0 on success else a negative value on error.
48 */
49 int consumer_socket_send(struct consumer_socket *socket, void *msg, size_t len)
50 {
51 int fd;
52 ssize_t size;
53
54 assert(socket);
55 assert(socket->fd_ptr);
56 assert(msg);
57
58 /* Consumer socket is invalid. Stopping. */
59 fd = *socket->fd_ptr;
60 if (fd < 0) {
61 goto error;
62 }
63
64 size = lttcomm_send_unix_sock(fd, msg, len);
65 if (size < 0) {
66 /* The above call will print a PERROR on error. */
67 DBG("Error when sending data to consumer on sock %d", fd);
68 /*
69 * At this point, the socket is not usable anymore thus closing it and
70 * setting the file descriptor to -1 so it is not reused.
71 */
72
73 /* This call will PERROR on error. */
74 (void) lttcomm_close_unix_sock(fd);
75 *socket->fd_ptr = -1;
76 goto error;
77 }
78
79 return 0;
80
81 error:
82 return -1;
83 }
84
85 /*
86 * Receive a data payload using a given consumer socket of size len.
87 *
88 * The consumer socket lock MUST be acquired before calling this since this
89 * function can change the fd value.
90 *
91 * Return 0 on success else a negative value on error.
92 */
93 int consumer_socket_recv(struct consumer_socket *socket, void *msg, size_t len)
94 {
95 int fd;
96 ssize_t size;
97
98 assert(socket);
99 assert(socket->fd_ptr);
100 assert(msg);
101
102 /* Consumer socket is invalid. Stopping. */
103 fd = *socket->fd_ptr;
104 if (fd < 0) {
105 goto error;
106 }
107
108 size = lttcomm_recv_unix_sock(fd, msg, len);
109 if (size <= 0) {
110 /* The above call will print a PERROR on error. */
111 DBG("Error when receiving data from the consumer socket %d", fd);
112 /*
113 * At this point, the socket is not usable anymore thus closing it and
114 * setting the file descriptor to -1 so it is not reused.
115 */
116
117 /* This call will PERROR on error. */
118 (void) lttcomm_close_unix_sock(fd);
119 *socket->fd_ptr = -1;
120 goto error;
121 }
122
123 return 0;
124
125 error:
126 return -1;
127 }
128
129 /*
130 * Receive a reply command status message from the consumer. Consumer socket
131 * lock MUST be acquired before calling this function.
132 *
133 * Return 0 on success, -1 on recv error or a negative lttng error code which
134 * was possibly returned by the consumer.
135 */
136 int consumer_recv_status_reply(struct consumer_socket *sock)
137 {
138 int ret;
139 struct lttcomm_consumer_status_msg reply;
140
141 assert(sock);
142
143 ret = consumer_socket_recv(sock, &reply, sizeof(reply));
144 if (ret < 0) {
145 goto end;
146 }
147
148 if (reply.ret_code == LTTCOMM_CONSUMERD_SUCCESS) {
149 /* All good. */
150 ret = 0;
151 } else {
152 ret = -reply.ret_code;
153 DBG("Consumer ret code %d", ret);
154 }
155
156 end:
157 return ret;
158 }
159
160 /*
161 * Once the ASK_CHANNEL command is sent to the consumer, the channel
162 * information are sent back. This call receives that data and populates key
163 * and stream_count.
164 *
165 * On success return 0 and both key and stream_count are set. On error, a
166 * negative value is sent back and both parameters are untouched.
167 */
168 int consumer_recv_status_channel(struct consumer_socket *sock,
169 uint64_t *key, unsigned int *stream_count)
170 {
171 int ret;
172 struct lttcomm_consumer_status_channel reply;
173
174 assert(sock);
175 assert(stream_count);
176 assert(key);
177
178 ret = consumer_socket_recv(sock, &reply, sizeof(reply));
179 if (ret < 0) {
180 goto end;
181 }
182
183 /* An error is possible so don't touch the key and stream_count. */
184 if (reply.ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
185 ret = -1;
186 goto end;
187 }
188
189 *key = reply.key;
190 *stream_count = reply.stream_count;
191 ret = 0;
192
193 end:
194 return ret;
195 }
196
197 /*
198 * Send destroy relayd command to consumer.
199 *
200 * On success return positive value. On error, negative value.
201 */
202 int consumer_send_destroy_relayd(struct consumer_socket *sock,
203 struct consumer_output *consumer)
204 {
205 int ret;
206 struct lttcomm_consumer_msg msg;
207
208 assert(consumer);
209 assert(sock);
210
211 DBG2("Sending destroy relayd command to consumer sock %d", *sock->fd_ptr);
212
213 memset(&msg, 0, sizeof(msg));
214 msg.cmd_type = LTTNG_CONSUMER_DESTROY_RELAYD;
215 msg.u.destroy_relayd.net_seq_idx = consumer->net_seq_index;
216
217 pthread_mutex_lock(sock->lock);
218 ret = consumer_socket_send(sock, &msg, sizeof(msg));
219 if (ret < 0) {
220 goto error;
221 }
222
223 /* Don't check the return value. The caller will do it. */
224 ret = consumer_recv_status_reply(sock);
225
226 DBG2("Consumer send destroy relayd command done");
227
228 error:
229 pthread_mutex_unlock(sock->lock);
230 return ret;
231 }
232
233 /*
234 * For each consumer socket in the consumer output object, send a destroy
235 * relayd command.
236 */
237 void consumer_output_send_destroy_relayd(struct consumer_output *consumer)
238 {
239 struct lttng_ht_iter iter;
240 struct consumer_socket *socket;
241
242 assert(consumer);
243
244 /* Destroy any relayd connection */
245 if (consumer->type == CONSUMER_DST_NET) {
246 rcu_read_lock();
247 cds_lfht_for_each_entry(consumer->socks->ht, &iter.iter, socket,
248 node.node) {
249 int ret;
250
251 /* Send destroy relayd command */
252 ret = consumer_send_destroy_relayd(socket, consumer);
253 if (ret < 0) {
254 DBG("Unable to send destroy relayd command to consumer");
255 /* Continue since we MUST delete everything at this point. */
256 }
257 }
258 rcu_read_unlock();
259 }
260 }
261
262 /*
263 * From a consumer_data structure, allocate and add a consumer socket to the
264 * consumer output.
265 *
266 * Return 0 on success, else negative value on error
267 */
268 int consumer_create_socket(struct consumer_data *data,
269 struct consumer_output *output)
270 {
271 int ret = 0;
272 struct consumer_socket *socket;
273
274 assert(data);
275
276 if (output == NULL || data->cmd_sock < 0) {
277 /*
278 * Not an error. Possible there is simply not spawned consumer or it's
279 * disabled for the tracing session asking the socket.
280 */
281 goto error;
282 }
283
284 rcu_read_lock();
285 socket = consumer_find_socket(data->cmd_sock, output);
286 rcu_read_unlock();
287 if (socket == NULL) {
288 socket = consumer_allocate_socket(&data->cmd_sock);
289 if (socket == NULL) {
290 ret = -1;
291 goto error;
292 }
293
294 socket->registered = 0;
295 socket->lock = &data->lock;
296 rcu_read_lock();
297 consumer_add_socket(socket, output);
298 rcu_read_unlock();
299 }
300
301 socket->type = data->type;
302
303 DBG3("Consumer socket created (fd: %d) and added to output",
304 data->cmd_sock);
305
306 error:
307 return ret;
308 }
309
310 /*
311 * Return the consumer socket from the given consumer output with the right
312 * bitness. On error, returns NULL.
313 *
314 * The caller MUST acquire a rcu read side lock and keep it until the socket
315 * object reference is not needed anymore.
316 */
317 struct consumer_socket *consumer_find_socket_by_bitness(int bits,
318 const struct consumer_output *consumer)
319 {
320 int consumer_fd;
321 struct consumer_socket *socket = NULL;
322
323 switch (bits) {
324 case 64:
325 consumer_fd = uatomic_read(&ust_consumerd64_fd);
326 break;
327 case 32:
328 consumer_fd = uatomic_read(&ust_consumerd32_fd);
329 break;
330 default:
331 assert(0);
332 goto end;
333 }
334
335 socket = consumer_find_socket(consumer_fd, consumer);
336 if (!socket) {
337 ERR("Consumer socket fd %d not found in consumer obj %p",
338 consumer_fd, consumer);
339 }
340
341 end:
342 return socket;
343 }
344
345 /*
346 * Find a consumer_socket in a consumer_output hashtable. Read side lock must
347 * be acquired before calling this function and across use of the
348 * returned consumer_socket.
349 */
350 struct consumer_socket *consumer_find_socket(int key,
351 const struct consumer_output *consumer)
352 {
353 struct lttng_ht_iter iter;
354 struct lttng_ht_node_ulong *node;
355 struct consumer_socket *socket = NULL;
356
357 /* Negative keys are lookup failures */
358 if (key < 0 || consumer == NULL) {
359 return NULL;
360 }
361
362 lttng_ht_lookup(consumer->socks, (void *)((unsigned long) key),
363 &iter);
364 node = lttng_ht_iter_get_node_ulong(&iter);
365 if (node != NULL) {
366 socket = caa_container_of(node, struct consumer_socket, node);
367 }
368
369 return socket;
370 }
371
372 /*
373 * Allocate a new consumer_socket and return the pointer.
374 */
375 struct consumer_socket *consumer_allocate_socket(int *fd)
376 {
377 struct consumer_socket *socket = NULL;
378
379 assert(fd);
380
381 socket = zmalloc(sizeof(struct consumer_socket));
382 if (socket == NULL) {
383 PERROR("zmalloc consumer socket");
384 goto error;
385 }
386
387 socket->fd_ptr = fd;
388 lttng_ht_node_init_ulong(&socket->node, *fd);
389
390 error:
391 return socket;
392 }
393
394 /*
395 * Add consumer socket to consumer output object. Read side lock must be
396 * acquired before calling this function.
397 */
398 void consumer_add_socket(struct consumer_socket *sock,
399 struct consumer_output *consumer)
400 {
401 assert(sock);
402 assert(consumer);
403
404 lttng_ht_add_unique_ulong(consumer->socks, &sock->node);
405 }
406
407 /*
408 * Delete consumer socket to consumer output object. Read side lock must be
409 * acquired before calling this function.
410 */
411 void consumer_del_socket(struct consumer_socket *sock,
412 struct consumer_output *consumer)
413 {
414 int ret;
415 struct lttng_ht_iter iter;
416
417 assert(sock);
418 assert(consumer);
419
420 iter.iter.node = &sock->node.node;
421 ret = lttng_ht_del(consumer->socks, &iter);
422 assert(!ret);
423 }
424
425 /*
426 * RCU destroy call function.
427 */
428 static void destroy_socket_rcu(struct rcu_head *head)
429 {
430 struct lttng_ht_node_ulong *node =
431 caa_container_of(head, struct lttng_ht_node_ulong, head);
432 struct consumer_socket *socket =
433 caa_container_of(node, struct consumer_socket, node);
434
435 free(socket);
436 }
437
438 /*
439 * Destroy and free socket pointer in a call RCU. Read side lock must be
440 * acquired before calling this function.
441 */
442 void consumer_destroy_socket(struct consumer_socket *sock)
443 {
444 assert(sock);
445
446 /*
447 * We DO NOT close the file descriptor here since it is global to the
448 * session daemon and is closed only if the consumer dies or a custom
449 * consumer was registered,
450 */
451 if (sock->registered) {
452 DBG3("Consumer socket was registered. Closing fd %d", *sock->fd_ptr);
453 lttcomm_close_unix_sock(*sock->fd_ptr);
454 }
455
456 call_rcu(&sock->node.head, destroy_socket_rcu);
457 }
458
459 /*
460 * Allocate and assign data to a consumer_output object.
461 *
462 * Return pointer to structure.
463 */
464 struct consumer_output *consumer_create_output(enum consumer_dst_type type)
465 {
466 struct consumer_output *output = NULL;
467
468 output = zmalloc(sizeof(struct consumer_output));
469 if (output == NULL) {
470 PERROR("zmalloc consumer_output");
471 goto error;
472 }
473
474 /* By default, consumer output is enabled */
475 output->enabled = 1;
476 output->type = type;
477 output->net_seq_index = (uint64_t) -1ULL;
478 urcu_ref_init(&output->ref);
479
480 output->socks = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
481
482 error:
483 return output;
484 }
485
486 /*
487 * Iterate over the consumer output socket hash table and destroy them. The
488 * socket file descriptor are only closed if the consumer output was
489 * registered meaning it's an external consumer.
490 */
491 void consumer_destroy_output_sockets(struct consumer_output *obj)
492 {
493 struct lttng_ht_iter iter;
494 struct consumer_socket *socket;
495
496 if (!obj->socks) {
497 return;
498 }
499
500 rcu_read_lock();
501 cds_lfht_for_each_entry(obj->socks->ht, &iter.iter, socket, node.node) {
502 consumer_del_socket(socket, obj);
503 consumer_destroy_socket(socket);
504 }
505 rcu_read_unlock();
506 }
507
508 /*
509 * Delete the consumer_output object from the list and free the ptr.
510 *
511 * Should *NOT* be called with RCU read-side lock held.
512 */
513 static void consumer_release_output(struct urcu_ref *ref)
514 {
515 struct consumer_output *obj =
516 caa_container_of(ref, struct consumer_output, ref);
517
518 consumer_destroy_output_sockets(obj);
519
520 if (obj->socks) {
521 /* Finally destroy HT */
522 ht_cleanup_push(obj->socks);
523 }
524
525 free(obj);
526 }
527
528 /*
529 * Get the consumer_output object.
530 */
531 void consumer_output_get(struct consumer_output *obj)
532 {
533 urcu_ref_get(&obj->ref);
534 }
535
536 /*
537 * Put the consumer_output object.
538 *
539 * Should *NOT* be called with RCU read-side lock held.
540 */
541 void consumer_output_put(struct consumer_output *obj)
542 {
543 if (!obj) {
544 return;
545 }
546 urcu_ref_put(&obj->ref, consumer_release_output);
547 }
548
549 /*
550 * Copy consumer output and returned the newly allocated copy.
551 *
552 * Should *NOT* be called with RCU read-side lock held.
553 */
554 struct consumer_output *consumer_copy_output(struct consumer_output *src)
555 {
556 int ret;
557 struct consumer_output *output;
558
559 assert(src);
560
561 output = consumer_create_output(src->type);
562 if (output == NULL) {
563 goto end;
564 }
565 output->enabled = src->enabled;
566 output->net_seq_index = src->net_seq_index;
567 memcpy(output->domain_subdir, src->domain_subdir,
568 sizeof(output->domain_subdir));
569 output->snapshot = src->snapshot;
570 output->relay_major_version = src->relay_major_version;
571 output->relay_minor_version = src->relay_minor_version;
572 memcpy(&output->dst, &src->dst, sizeof(output->dst));
573 ret = consumer_copy_sockets(output, src);
574 if (ret < 0) {
575 goto error_put;
576 }
577 end:
578 return output;
579
580 error_put:
581 consumer_output_put(output);
582 return NULL;
583 }
584
585 /*
586 * Copy consumer sockets from src to dst.
587 *
588 * Return 0 on success or else a negative value.
589 */
590 int consumer_copy_sockets(struct consumer_output *dst,
591 struct consumer_output *src)
592 {
593 int ret = 0;
594 struct lttng_ht_iter iter;
595 struct consumer_socket *socket, *copy_sock;
596
597 assert(dst);
598 assert(src);
599
600 rcu_read_lock();
601 cds_lfht_for_each_entry(src->socks->ht, &iter.iter, socket, node.node) {
602 /* Ignore socket that are already there. */
603 copy_sock = consumer_find_socket(*socket->fd_ptr, dst);
604 if (copy_sock) {
605 continue;
606 }
607
608 /* Create new socket object. */
609 copy_sock = consumer_allocate_socket(socket->fd_ptr);
610 if (copy_sock == NULL) {
611 rcu_read_unlock();
612 ret = -ENOMEM;
613 goto error;
614 }
615
616 copy_sock->registered = socket->registered;
617 /*
618 * This is valid because this lock is shared accross all consumer
619 * object being the global lock of the consumer data structure of the
620 * session daemon.
621 */
622 copy_sock->lock = socket->lock;
623 consumer_add_socket(copy_sock, dst);
624 }
625 rcu_read_unlock();
626
627 error:
628 return ret;
629 }
630
631 /*
632 * Set network URI to the consumer output.
633 *
634 * Return 0 on success. Return 1 if the URI were equal. Else, negative value on
635 * error.
636 */
637 int consumer_set_network_uri(const struct ltt_session *session,
638 struct consumer_output *output,
639 struct lttng_uri *uri)
640 {
641 int ret;
642 struct lttng_uri *dst_uri = NULL;
643
644 /* Code flow error safety net. */
645 assert(output);
646 assert(uri);
647
648 switch (uri->stype) {
649 case LTTNG_STREAM_CONTROL:
650 dst_uri = &output->dst.net.control;
651 output->dst.net.control_isset = 1;
652 if (uri->port == 0) {
653 /* Assign default port. */
654 uri->port = DEFAULT_NETWORK_CONTROL_PORT;
655 } else {
656 if (output->dst.net.data_isset && uri->port ==
657 output->dst.net.data.port) {
658 ret = -LTTNG_ERR_INVALID;
659 goto error;
660 }
661 }
662 DBG3("Consumer control URI set with port %d", uri->port);
663 break;
664 case LTTNG_STREAM_DATA:
665 dst_uri = &output->dst.net.data;
666 output->dst.net.data_isset = 1;
667 if (uri->port == 0) {
668 /* Assign default port. */
669 uri->port = DEFAULT_NETWORK_DATA_PORT;
670 } else {
671 if (output->dst.net.control_isset && uri->port ==
672 output->dst.net.control.port) {
673 ret = -LTTNG_ERR_INVALID;
674 goto error;
675 }
676 }
677 DBG3("Consumer data URI set with port %d", uri->port);
678 break;
679 default:
680 ERR("Set network uri type unknown %d", uri->stype);
681 ret = -LTTNG_ERR_INVALID;
682 goto error;
683 }
684
685 ret = uri_compare(dst_uri, uri);
686 if (!ret) {
687 /* Same URI, don't touch it and return success. */
688 DBG3("URI network compare are the same");
689 goto equal;
690 }
691
692 /* URIs were not equal, replacing it. */
693 memcpy(dst_uri, uri, sizeof(struct lttng_uri));
694 output->type = CONSUMER_DST_NET;
695 if (dst_uri->stype != LTTNG_STREAM_CONTROL) {
696 /* Only the control uri needs to contain the path. */
697 goto end;
698 }
699
700 /*
701 * If the user has specified a subdir as part of the control
702 * URL, the session's base output directory is:
703 * /RELAYD_OUTPUT_PATH/HOSTNAME/USER_SPECIFIED_DIR
704 *
705 * Hence, the "base_dir" from which all stream files and
706 * session rotation chunks are created takes the form
707 * /HOSTNAME/USER_SPECIFIED_DIR
708 *
709 * If the user has not specified an output directory as part of
710 * the control URL, the base output directory has the form:
711 * /RELAYD_OUTPUT_PATH/HOSTNAME/SESSION_NAME-CREATION_TIME
712 *
713 * Hence, the "base_dir" from which all stream files and
714 * session rotation chunks are created takes the form
715 * /HOSTNAME/SESSION_NAME-CREATION_TIME
716 *
717 * Note that automatically generated session names already
718 * contain the session's creation time. In that case, the
719 * creation time is omitted to prevent it from being duplicated
720 * in the final directory hierarchy.
721 */
722 if (*uri->subdir) {
723 if (strstr(uri->subdir, "../")) {
724 ERR("Network URI subdirs are not allowed to walk up the path hierarchy");
725 ret = -LTTNG_ERR_INVALID;
726 goto error;
727 }
728 ret = snprintf(output->dst.net.base_dir,
729 sizeof(output->dst.net.base_dir),
730 "/%s/%s/", session->hostname, uri->subdir);
731 } else {
732 if (session->has_auto_generated_name) {
733 ret = snprintf(output->dst.net.base_dir,
734 sizeof(output->dst.net.base_dir),
735 "/%s/%s/", session->hostname,
736 session->name);
737 } else {
738 char session_creation_datetime[16];
739 size_t strftime_ret;
740 struct tm *timeinfo;
741
742 timeinfo = localtime(&session->creation_time);
743 if (!timeinfo) {
744 ret = -LTTNG_ERR_FATAL;
745 goto error;
746 }
747 strftime_ret = strftime(session_creation_datetime,
748 sizeof(session_creation_datetime),
749 "%Y%m%d-%H%M%S", timeinfo);
750 if (strftime_ret == 0) {
751 ERR("Failed to format session creation timestamp while setting network URI");
752 ret = -LTTNG_ERR_FATAL;
753 goto error;
754 }
755 ret = snprintf(output->dst.net.base_dir,
756 sizeof(output->dst.net.base_dir),
757 "/%s/%s-%s/", session->hostname,
758 session->name,
759 session_creation_datetime);
760 }
761 }
762 if (ret >= sizeof(output->dst.net.base_dir)) {
763 ret = -LTTNG_ERR_INVALID;
764 ERR("Truncation occurred while setting network output base directory");
765 goto error;
766 } else if (ret == -1) {
767 ret = -LTTNG_ERR_INVALID;
768 PERROR("Error occurred while setting network output base directory");
769 goto error;
770 }
771
772 DBG3("Consumer set network uri base_dir path %s",
773 output->dst.net.base_dir);
774
775 end:
776 return 0;
777 equal:
778 return 1;
779 error:
780 return ret;
781 }
782
783 /*
784 * Send file descriptor to consumer via sock.
785 *
786 * The consumer socket lock must be held by the caller.
787 */
788 int consumer_send_fds(struct consumer_socket *sock, const int *fds,
789 size_t nb_fd)
790 {
791 int ret;
792
793 assert(fds);
794 assert(sock);
795 assert(nb_fd > 0);
796 assert(pthread_mutex_trylock(sock->lock) == EBUSY);
797
798 ret = lttcomm_send_fds_unix_sock(*sock->fd_ptr, fds, nb_fd);
799 if (ret < 0) {
800 /* The above call will print a PERROR on error. */
801 DBG("Error when sending consumer fds on sock %d", *sock->fd_ptr);
802 goto error;
803 }
804
805 ret = consumer_recv_status_reply(sock);
806 error:
807 return ret;
808 }
809
810 /*
811 * Consumer send communication message structure to consumer.
812 *
813 * The consumer socket lock must be held by the caller.
814 */
815 int consumer_send_msg(struct consumer_socket *sock,
816 struct lttcomm_consumer_msg *msg)
817 {
818 int ret;
819
820 assert(msg);
821 assert(sock);
822 assert(pthread_mutex_trylock(sock->lock) == EBUSY);
823
824 ret = consumer_socket_send(sock, msg, sizeof(struct lttcomm_consumer_msg));
825 if (ret < 0) {
826 goto error;
827 }
828
829 ret = consumer_recv_status_reply(sock);
830
831 error:
832 return ret;
833 }
834
835 /*
836 * Consumer send channel communication message structure to consumer.
837 *
838 * The consumer socket lock must be held by the caller.
839 */
840 int consumer_send_channel(struct consumer_socket *sock,
841 struct lttcomm_consumer_msg *msg)
842 {
843 int ret;
844
845 assert(msg);
846 assert(sock);
847
848 ret = consumer_send_msg(sock, msg);
849 if (ret < 0) {
850 goto error;
851 }
852
853 error:
854 return ret;
855 }
856
857 /*
858 * Populate the given consumer msg structure with the ask_channel command
859 * information.
860 */
861 void consumer_init_ask_channel_comm_msg(struct lttcomm_consumer_msg *msg,
862 uint64_t subbuf_size,
863 uint64_t num_subbuf,
864 int overwrite,
865 unsigned int switch_timer_interval,
866 unsigned int read_timer_interval,
867 unsigned int live_timer_interval,
868 unsigned int monitor_timer_interval,
869 int output,
870 int type,
871 uint64_t session_id,
872 const char *pathname,
873 const char *name,
874 uint64_t relayd_id,
875 uint64_t key,
876 unsigned char *uuid,
877 uint32_t chan_id,
878 uint64_t tracefile_size,
879 uint64_t tracefile_count,
880 uint64_t session_id_per_pid,
881 unsigned int monitor,
882 uint32_t ust_app_uid,
883 int64_t blocking_timeout,
884 const char *root_shm_path,
885 const char *shm_path,
886 struct lttng_trace_chunk *trace_chunk,
887 const struct lttng_credentials *buffer_credentials)
888 {
889 assert(msg);
890
891 /* Zeroed structure */
892 memset(msg, 0, sizeof(struct lttcomm_consumer_msg));
893 msg->u.ask_channel.buffer_credentials.uid = UINT32_MAX;
894 msg->u.ask_channel.buffer_credentials.gid = UINT32_MAX;
895
896 if (monitor) {
897 assert(trace_chunk);
898 }
899
900 if (trace_chunk) {
901 uint64_t chunk_id;
902 enum lttng_trace_chunk_status chunk_status;
903
904 chunk_status = lttng_trace_chunk_get_id(trace_chunk, &chunk_id);
905 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
906 LTTNG_OPTIONAL_SET(&msg->u.ask_channel.chunk_id, chunk_id);
907 }
908 msg->u.ask_channel.buffer_credentials.uid = buffer_credentials->uid;
909 msg->u.ask_channel.buffer_credentials.gid = buffer_credentials->gid;
910
911 msg->cmd_type = LTTNG_CONSUMER_ASK_CHANNEL_CREATION;
912 msg->u.ask_channel.subbuf_size = subbuf_size;
913 msg->u.ask_channel.num_subbuf = num_subbuf ;
914 msg->u.ask_channel.overwrite = overwrite;
915 msg->u.ask_channel.switch_timer_interval = switch_timer_interval;
916 msg->u.ask_channel.read_timer_interval = read_timer_interval;
917 msg->u.ask_channel.live_timer_interval = live_timer_interval;
918 msg->u.ask_channel.monitor_timer_interval = monitor_timer_interval;
919 msg->u.ask_channel.output = output;
920 msg->u.ask_channel.type = type;
921 msg->u.ask_channel.session_id = session_id;
922 msg->u.ask_channel.session_id_per_pid = session_id_per_pid;
923 msg->u.ask_channel.relayd_id = relayd_id;
924 msg->u.ask_channel.key = key;
925 msg->u.ask_channel.chan_id = chan_id;
926 msg->u.ask_channel.tracefile_size = tracefile_size;
927 msg->u.ask_channel.tracefile_count = tracefile_count;
928 msg->u.ask_channel.monitor = monitor;
929 msg->u.ask_channel.ust_app_uid = ust_app_uid;
930 msg->u.ask_channel.blocking_timeout = blocking_timeout;
931
932 memcpy(msg->u.ask_channel.uuid, uuid, sizeof(msg->u.ask_channel.uuid));
933
934 if (pathname) {
935 strncpy(msg->u.ask_channel.pathname, pathname,
936 sizeof(msg->u.ask_channel.pathname));
937 msg->u.ask_channel.pathname[sizeof(msg->u.ask_channel.pathname)-1] = '\0';
938 }
939
940 strncpy(msg->u.ask_channel.name, name, sizeof(msg->u.ask_channel.name));
941 msg->u.ask_channel.name[sizeof(msg->u.ask_channel.name) - 1] = '\0';
942
943 if (root_shm_path) {
944 strncpy(msg->u.ask_channel.root_shm_path, root_shm_path,
945 sizeof(msg->u.ask_channel.root_shm_path));
946 msg->u.ask_channel.root_shm_path[sizeof(msg->u.ask_channel.root_shm_path) - 1] = '\0';
947 }
948 if (shm_path) {
949 strncpy(msg->u.ask_channel.shm_path, shm_path,
950 sizeof(msg->u.ask_channel.shm_path));
951 msg->u.ask_channel.shm_path[sizeof(msg->u.ask_channel.shm_path) - 1] = '\0';
952 }
953 }
954
955 /*
956 * Init channel communication message structure.
957 */
958 void consumer_init_add_channel_comm_msg(struct lttcomm_consumer_msg *msg,
959 uint64_t channel_key,
960 uint64_t session_id,
961 const char *pathname,
962 uid_t uid,
963 gid_t gid,
964 uint64_t relayd_id,
965 const char *name,
966 unsigned int nb_init_streams,
967 enum lttng_event_output output,
968 int type,
969 uint64_t tracefile_size,
970 uint64_t tracefile_count,
971 unsigned int monitor,
972 unsigned int live_timer_interval,
973 unsigned int monitor_timer_interval,
974 struct lttng_trace_chunk *trace_chunk)
975 {
976 assert(msg);
977
978 /* Zeroed structure */
979 memset(msg, 0, sizeof(struct lttcomm_consumer_msg));
980
981 if (trace_chunk) {
982 uint64_t chunk_id;
983 enum lttng_trace_chunk_status chunk_status;
984
985 chunk_status = lttng_trace_chunk_get_id(trace_chunk, &chunk_id);
986 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
987 LTTNG_OPTIONAL_SET(&msg->u.channel.chunk_id, chunk_id);
988 }
989
990 /* Send channel */
991 msg->cmd_type = LTTNG_CONSUMER_ADD_CHANNEL;
992 msg->u.channel.channel_key = channel_key;
993 msg->u.channel.session_id = session_id;
994 msg->u.channel.relayd_id = relayd_id;
995 msg->u.channel.nb_init_streams = nb_init_streams;
996 msg->u.channel.output = output;
997 msg->u.channel.type = type;
998 msg->u.channel.tracefile_size = tracefile_size;
999 msg->u.channel.tracefile_count = tracefile_count;
1000 msg->u.channel.monitor = monitor;
1001 msg->u.channel.live_timer_interval = live_timer_interval;
1002 msg->u.channel.monitor_timer_interval = monitor_timer_interval;
1003
1004 strncpy(msg->u.channel.pathname, pathname,
1005 sizeof(msg->u.channel.pathname));
1006 msg->u.channel.pathname[sizeof(msg->u.channel.pathname) - 1] = '\0';
1007
1008 strncpy(msg->u.channel.name, name, sizeof(msg->u.channel.name));
1009 msg->u.channel.name[sizeof(msg->u.channel.name) - 1] = '\0';
1010 }
1011
1012 /*
1013 * Init stream communication message structure.
1014 */
1015 void consumer_init_add_stream_comm_msg(struct lttcomm_consumer_msg *msg,
1016 uint64_t channel_key,
1017 uint64_t stream_key,
1018 int32_t cpu)
1019 {
1020 assert(msg);
1021
1022 memset(msg, 0, sizeof(struct lttcomm_consumer_msg));
1023
1024 msg->cmd_type = LTTNG_CONSUMER_ADD_STREAM;
1025 msg->u.stream.channel_key = channel_key;
1026 msg->u.stream.stream_key = stream_key;
1027 msg->u.stream.cpu = cpu;
1028 }
1029
1030 void consumer_init_streams_sent_comm_msg(struct lttcomm_consumer_msg *msg,
1031 enum lttng_consumer_command cmd,
1032 uint64_t channel_key, uint64_t net_seq_idx)
1033 {
1034 assert(msg);
1035
1036 memset(msg, 0, sizeof(struct lttcomm_consumer_msg));
1037
1038 msg->cmd_type = cmd;
1039 msg->u.sent_streams.channel_key = channel_key;
1040 msg->u.sent_streams.net_seq_idx = net_seq_idx;
1041 }
1042
1043 /*
1044 * Send stream communication structure to the consumer.
1045 */
1046 int consumer_send_stream(struct consumer_socket *sock,
1047 struct consumer_output *dst, struct lttcomm_consumer_msg *msg,
1048 const int *fds, size_t nb_fd)
1049 {
1050 int ret;
1051
1052 assert(msg);
1053 assert(dst);
1054 assert(sock);
1055 assert(fds);
1056
1057 ret = consumer_send_msg(sock, msg);
1058 if (ret < 0) {
1059 goto error;
1060 }
1061
1062 ret = consumer_send_fds(sock, fds, nb_fd);
1063 if (ret < 0) {
1064 goto error;
1065 }
1066
1067 error:
1068 return ret;
1069 }
1070
1071 /*
1072 * Send relayd socket to consumer associated with a session name.
1073 *
1074 * The consumer socket lock must be held by the caller.
1075 *
1076 * On success return positive value. On error, negative value.
1077 */
1078 int consumer_send_relayd_socket(struct consumer_socket *consumer_sock,
1079 struct lttcomm_relayd_sock *rsock, struct consumer_output *consumer,
1080 enum lttng_stream_type type, uint64_t session_id,
1081 const char *session_name, const char *hostname,
1082 int session_live_timer, const uint64_t *current_chunk_id,
1083 time_t session_creation_time)
1084 {
1085 int ret;
1086 struct lttcomm_consumer_msg msg;
1087
1088 /* Code flow error. Safety net. */
1089 assert(rsock);
1090 assert(consumer);
1091 assert(consumer_sock);
1092
1093 memset(&msg, 0, sizeof(msg));
1094 /* Bail out if consumer is disabled */
1095 if (!consumer->enabled) {
1096 ret = LTTNG_OK;
1097 goto error;
1098 }
1099
1100 if (type == LTTNG_STREAM_CONTROL) {
1101 ret = relayd_create_session(rsock,
1102 &msg.u.relayd_sock.relayd_session_id,
1103 session_name, hostname, session_live_timer,
1104 consumer->snapshot, session_id,
1105 sessiond_uuid, current_chunk_id,
1106 session_creation_time);
1107 if (ret < 0) {
1108 /* Close the control socket. */
1109 (void) relayd_close(rsock);
1110 goto error;
1111 }
1112 }
1113
1114 msg.cmd_type = LTTNG_CONSUMER_ADD_RELAYD_SOCKET;
1115 /*
1116 * Assign network consumer output index using the temporary consumer since
1117 * this call should only be made from within a set_consumer_uri() function
1118 * call in the session daemon.
1119 */
1120 msg.u.relayd_sock.net_index = consumer->net_seq_index;
1121 msg.u.relayd_sock.type = type;
1122 msg.u.relayd_sock.session_id = session_id;
1123 memcpy(&msg.u.relayd_sock.sock, rsock, sizeof(msg.u.relayd_sock.sock));
1124
1125 DBG3("Sending relayd sock info to consumer on %d", *consumer_sock->fd_ptr);
1126 ret = consumer_send_msg(consumer_sock, &msg);
1127 if (ret < 0) {
1128 goto error;
1129 }
1130
1131 DBG3("Sending relayd socket file descriptor to consumer");
1132 ret = consumer_send_fds(consumer_sock, &rsock->sock.fd, 1);
1133 if (ret < 0) {
1134 goto error;
1135 }
1136
1137 DBG2("Consumer relayd socket sent");
1138
1139 error:
1140 return ret;
1141 }
1142
1143 static
1144 int consumer_send_pipe(struct consumer_socket *consumer_sock,
1145 enum lttng_consumer_command cmd, int pipe)
1146 {
1147 int ret;
1148 struct lttcomm_consumer_msg msg;
1149 const char *pipe_name;
1150 const char *command_name;
1151
1152 switch (cmd) {
1153 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1154 pipe_name = "channel monitor";
1155 command_name = "SET_CHANNEL_MONITOR_PIPE";
1156 break;
1157 default:
1158 ERR("Unexpected command received in %s (cmd = %d)", __func__,
1159 (int) cmd);
1160 abort();
1161 }
1162
1163 /* Code flow error. Safety net. */
1164
1165 memset(&msg, 0, sizeof(msg));
1166 msg.cmd_type = cmd;
1167
1168 pthread_mutex_lock(consumer_sock->lock);
1169 DBG3("Sending %s command to consumer", command_name);
1170 ret = consumer_send_msg(consumer_sock, &msg);
1171 if (ret < 0) {
1172 goto error;
1173 }
1174
1175 DBG3("Sending %s pipe %d to consumer on socket %d",
1176 pipe_name,
1177 pipe, *consumer_sock->fd_ptr);
1178 ret = consumer_send_fds(consumer_sock, &pipe, 1);
1179 if (ret < 0) {
1180 goto error;
1181 }
1182
1183 DBG2("%s pipe successfully sent", pipe_name);
1184 error:
1185 pthread_mutex_unlock(consumer_sock->lock);
1186 return ret;
1187 }
1188
1189 int consumer_send_channel_monitor_pipe(struct consumer_socket *consumer_sock,
1190 int pipe)
1191 {
1192 return consumer_send_pipe(consumer_sock,
1193 LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE, pipe);
1194 }
1195
1196 /*
1197 * Ask the consumer if the data is pending for the specific session id.
1198 * Returns 1 if data is pending, 0 otherwise, or < 0 on error.
1199 */
1200 int consumer_is_data_pending(uint64_t session_id,
1201 struct consumer_output *consumer)
1202 {
1203 int ret;
1204 int32_t ret_code = 0; /* Default is that the data is NOT pending */
1205 struct consumer_socket *socket;
1206 struct lttng_ht_iter iter;
1207 struct lttcomm_consumer_msg msg;
1208
1209 assert(consumer);
1210
1211 DBG3("Consumer data pending for id %" PRIu64, session_id);
1212
1213 memset(&msg, 0, sizeof(msg));
1214 msg.cmd_type = LTTNG_CONSUMER_DATA_PENDING;
1215 msg.u.data_pending.session_id = session_id;
1216
1217 /* Send command for each consumer */
1218 rcu_read_lock();
1219 cds_lfht_for_each_entry(consumer->socks->ht, &iter.iter, socket,
1220 node.node) {
1221 pthread_mutex_lock(socket->lock);
1222 ret = consumer_socket_send(socket, &msg, sizeof(msg));
1223 if (ret < 0) {
1224 pthread_mutex_unlock(socket->lock);
1225 goto error_unlock;
1226 }
1227
1228 /*
1229 * No need for a recv reply status because the answer to the command is
1230 * the reply status message.
1231 */
1232
1233 ret = consumer_socket_recv(socket, &ret_code, sizeof(ret_code));
1234 if (ret < 0) {
1235 pthread_mutex_unlock(socket->lock);
1236 goto error_unlock;
1237 }
1238 pthread_mutex_unlock(socket->lock);
1239
1240 if (ret_code == 1) {
1241 break;
1242 }
1243 }
1244 rcu_read_unlock();
1245
1246 DBG("Consumer data is %s pending for session id %" PRIu64,
1247 ret_code == 1 ? "" : "NOT", session_id);
1248 return ret_code;
1249
1250 error_unlock:
1251 rcu_read_unlock();
1252 return -1;
1253 }
1254
1255 /*
1256 * Send a flush command to consumer using the given channel key.
1257 *
1258 * Return 0 on success else a negative value.
1259 */
1260 int consumer_flush_channel(struct consumer_socket *socket, uint64_t key)
1261 {
1262 int ret;
1263 struct lttcomm_consumer_msg msg;
1264
1265 assert(socket);
1266
1267 DBG2("Consumer flush channel key %" PRIu64, key);
1268
1269 memset(&msg, 0, sizeof(msg));
1270 msg.cmd_type = LTTNG_CONSUMER_FLUSH_CHANNEL;
1271 msg.u.flush_channel.key = key;
1272
1273 pthread_mutex_lock(socket->lock);
1274 health_code_update();
1275
1276 ret = consumer_send_msg(socket, &msg);
1277 if (ret < 0) {
1278 goto end;
1279 }
1280
1281 end:
1282 health_code_update();
1283 pthread_mutex_unlock(socket->lock);
1284 return ret;
1285 }
1286
1287 /*
1288 * Send a clear quiescent command to consumer using the given channel key.
1289 *
1290 * Return 0 on success else a negative value.
1291 */
1292 int consumer_clear_quiescent_channel(struct consumer_socket *socket, uint64_t key)
1293 {
1294 int ret;
1295 struct lttcomm_consumer_msg msg;
1296
1297 assert(socket);
1298
1299 DBG2("Consumer clear quiescent channel key %" PRIu64, key);
1300
1301 memset(&msg, 0, sizeof(msg));
1302 msg.cmd_type = LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL;
1303 msg.u.clear_quiescent_channel.key = key;
1304
1305 pthread_mutex_lock(socket->lock);
1306 health_code_update();
1307
1308 ret = consumer_send_msg(socket, &msg);
1309 if (ret < 0) {
1310 goto end;
1311 }
1312
1313 end:
1314 health_code_update();
1315 pthread_mutex_unlock(socket->lock);
1316 return ret;
1317 }
1318
1319 /*
1320 * Send a close metadata command to consumer using the given channel key.
1321 * Called with registry lock held.
1322 *
1323 * Return 0 on success else a negative value.
1324 */
1325 int consumer_close_metadata(struct consumer_socket *socket,
1326 uint64_t metadata_key)
1327 {
1328 int ret;
1329 struct lttcomm_consumer_msg msg;
1330
1331 assert(socket);
1332
1333 DBG2("Consumer close metadata channel key %" PRIu64, metadata_key);
1334
1335 memset(&msg, 0, sizeof(msg));
1336 msg.cmd_type = LTTNG_CONSUMER_CLOSE_METADATA;
1337 msg.u.close_metadata.key = metadata_key;
1338
1339 pthread_mutex_lock(socket->lock);
1340 health_code_update();
1341
1342 ret = consumer_send_msg(socket, &msg);
1343 if (ret < 0) {
1344 goto end;
1345 }
1346
1347 end:
1348 health_code_update();
1349 pthread_mutex_unlock(socket->lock);
1350 return ret;
1351 }
1352
1353 /*
1354 * Send a setup metdata command to consumer using the given channel key.
1355 *
1356 * Return 0 on success else a negative value.
1357 */
1358 int consumer_setup_metadata(struct consumer_socket *socket,
1359 uint64_t metadata_key)
1360 {
1361 int ret;
1362 struct lttcomm_consumer_msg msg;
1363
1364 assert(socket);
1365
1366 DBG2("Consumer setup metadata channel key %" PRIu64, metadata_key);
1367
1368 memset(&msg, 0, sizeof(msg));
1369 msg.cmd_type = LTTNG_CONSUMER_SETUP_METADATA;
1370 msg.u.setup_metadata.key = metadata_key;
1371
1372 pthread_mutex_lock(socket->lock);
1373 health_code_update();
1374
1375 ret = consumer_send_msg(socket, &msg);
1376 if (ret < 0) {
1377 goto end;
1378 }
1379
1380 end:
1381 health_code_update();
1382 pthread_mutex_unlock(socket->lock);
1383 return ret;
1384 }
1385
1386 /*
1387 * Send metadata string to consumer.
1388 * RCU read-side lock must be held to guarantee existence of socket.
1389 *
1390 * Return 0 on success else a negative value.
1391 */
1392 int consumer_push_metadata(struct consumer_socket *socket,
1393 uint64_t metadata_key, char *metadata_str, size_t len,
1394 size_t target_offset, uint64_t version)
1395 {
1396 int ret;
1397 struct lttcomm_consumer_msg msg;
1398
1399 assert(socket);
1400
1401 DBG2("Consumer push metadata to consumer socket %d", *socket->fd_ptr);
1402
1403 pthread_mutex_lock(socket->lock);
1404
1405 memset(&msg, 0, sizeof(msg));
1406 msg.cmd_type = LTTNG_CONSUMER_PUSH_METADATA;
1407 msg.u.push_metadata.key = metadata_key;
1408 msg.u.push_metadata.target_offset = target_offset;
1409 msg.u.push_metadata.len = len;
1410 msg.u.push_metadata.version = version;
1411
1412 health_code_update();
1413 ret = consumer_send_msg(socket, &msg);
1414 if (ret < 0 || len == 0) {
1415 goto end;
1416 }
1417
1418 DBG3("Consumer pushing metadata on sock %d of len %zu", *socket->fd_ptr,
1419 len);
1420
1421 ret = consumer_socket_send(socket, metadata_str, len);
1422 if (ret < 0) {
1423 goto end;
1424 }
1425
1426 health_code_update();
1427 ret = consumer_recv_status_reply(socket);
1428 if (ret < 0) {
1429 goto end;
1430 }
1431
1432 end:
1433 pthread_mutex_unlock(socket->lock);
1434 health_code_update();
1435 return ret;
1436 }
1437
1438 /*
1439 * Ask the consumer to snapshot a specific channel using the key.
1440 *
1441 * Returns LTTNG_OK on success or else an LTTng error code.
1442 */
1443 enum lttng_error_code consumer_snapshot_channel(struct consumer_socket *socket,
1444 uint64_t key, const struct consumer_output *output, int metadata,
1445 uid_t uid, gid_t gid, const char *channel_path, int wait,
1446 uint64_t nb_packets_per_stream)
1447 {
1448 int ret;
1449 enum lttng_error_code status = LTTNG_OK;
1450 struct lttcomm_consumer_msg msg;
1451
1452 assert(socket);
1453 assert(output);
1454
1455 DBG("Consumer snapshot channel key %" PRIu64, key);
1456
1457 memset(&msg, 0, sizeof(msg));
1458 msg.cmd_type = LTTNG_CONSUMER_SNAPSHOT_CHANNEL;
1459 msg.u.snapshot_channel.key = key;
1460 msg.u.snapshot_channel.nb_packets_per_stream = nb_packets_per_stream;
1461 msg.u.snapshot_channel.metadata = metadata;
1462
1463 if (output->type == CONSUMER_DST_NET) {
1464 msg.u.snapshot_channel.relayd_id =
1465 output->net_seq_index;
1466 msg.u.snapshot_channel.use_relayd = 1;
1467 } else {
1468 msg.u.snapshot_channel.relayd_id = (uint64_t) -1ULL;
1469 }
1470 ret = lttng_strncpy(msg.u.snapshot_channel.pathname,
1471 channel_path,
1472 sizeof(msg.u.snapshot_channel.pathname));
1473 if (ret < 0) {
1474 ERR("Snapshot path exceeds the maximal allowed length of %zu bytes (%zu bytes required) with path \"%s\"",
1475 sizeof(msg.u.snapshot_channel.pathname),
1476 strlen(channel_path),
1477 channel_path);
1478 status = LTTNG_ERR_SNAPSHOT_FAIL;
1479 goto error;
1480 }
1481
1482 health_code_update();
1483 pthread_mutex_lock(socket->lock);
1484 ret = consumer_send_msg(socket, &msg);
1485 pthread_mutex_unlock(socket->lock);
1486 if (ret < 0) {
1487 switch (-ret) {
1488 case LTTCOMM_CONSUMERD_CHAN_NOT_FOUND:
1489 status = LTTNG_ERR_CHAN_NOT_FOUND;
1490 break;
1491 default:
1492 status = LTTNG_ERR_SNAPSHOT_FAIL;
1493 break;
1494 }
1495 goto error;
1496 }
1497
1498 error:
1499 health_code_update();
1500 return status;
1501 }
1502
1503 /*
1504 * Ask the consumer the number of discarded events for a channel.
1505 */
1506 int consumer_get_discarded_events(uint64_t session_id, uint64_t channel_key,
1507 struct consumer_output *consumer, uint64_t *discarded)
1508 {
1509 int ret;
1510 struct consumer_socket *socket;
1511 struct lttng_ht_iter iter;
1512 struct lttcomm_consumer_msg msg;
1513
1514 assert(consumer);
1515
1516 DBG3("Consumer discarded events id %" PRIu64, session_id);
1517
1518 memset(&msg, 0, sizeof(msg));
1519 msg.cmd_type = LTTNG_CONSUMER_DISCARDED_EVENTS;
1520 msg.u.discarded_events.session_id = session_id;
1521 msg.u.discarded_events.channel_key = channel_key;
1522
1523 *discarded = 0;
1524
1525 /* Send command for each consumer */
1526 rcu_read_lock();
1527 cds_lfht_for_each_entry(consumer->socks->ht, &iter.iter, socket,
1528 node.node) {
1529 uint64_t consumer_discarded = 0;
1530 pthread_mutex_lock(socket->lock);
1531 ret = consumer_socket_send(socket, &msg, sizeof(msg));
1532 if (ret < 0) {
1533 pthread_mutex_unlock(socket->lock);
1534 goto end;
1535 }
1536
1537 /*
1538 * No need for a recv reply status because the answer to the
1539 * command is the reply status message.
1540 */
1541 ret = consumer_socket_recv(socket, &consumer_discarded,
1542 sizeof(consumer_discarded));
1543 if (ret < 0) {
1544 ERR("get discarded events");
1545 pthread_mutex_unlock(socket->lock);
1546 goto end;
1547 }
1548 pthread_mutex_unlock(socket->lock);
1549 *discarded += consumer_discarded;
1550 }
1551 ret = 0;
1552 DBG("Consumer discarded %" PRIu64 " events in session id %" PRIu64,
1553 *discarded, session_id);
1554
1555 end:
1556 rcu_read_unlock();
1557 return ret;
1558 }
1559
1560 /*
1561 * Ask the consumer the number of lost packets for a channel.
1562 */
1563 int consumer_get_lost_packets(uint64_t session_id, uint64_t channel_key,
1564 struct consumer_output *consumer, uint64_t *lost)
1565 {
1566 int ret;
1567 struct consumer_socket *socket;
1568 struct lttng_ht_iter iter;
1569 struct lttcomm_consumer_msg msg;
1570
1571 assert(consumer);
1572
1573 DBG3("Consumer lost packets id %" PRIu64, session_id);
1574
1575 memset(&msg, 0, sizeof(msg));
1576 msg.cmd_type = LTTNG_CONSUMER_LOST_PACKETS;
1577 msg.u.lost_packets.session_id = session_id;
1578 msg.u.lost_packets.channel_key = channel_key;
1579
1580 *lost = 0;
1581
1582 /* Send command for each consumer */
1583 rcu_read_lock();
1584 cds_lfht_for_each_entry(consumer->socks->ht, &iter.iter, socket,
1585 node.node) {
1586 uint64_t consumer_lost = 0;
1587 pthread_mutex_lock(socket->lock);
1588 ret = consumer_socket_send(socket, &msg, sizeof(msg));
1589 if (ret < 0) {
1590 pthread_mutex_unlock(socket->lock);
1591 goto end;
1592 }
1593
1594 /*
1595 * No need for a recv reply status because the answer to the
1596 * command is the reply status message.
1597 */
1598 ret = consumer_socket_recv(socket, &consumer_lost,
1599 sizeof(consumer_lost));
1600 if (ret < 0) {
1601 ERR("get lost packets");
1602 pthread_mutex_unlock(socket->lock);
1603 goto end;
1604 }
1605 pthread_mutex_unlock(socket->lock);
1606 *lost += consumer_lost;
1607 }
1608 ret = 0;
1609 DBG("Consumer lost %" PRIu64 " packets in session id %" PRIu64,
1610 *lost, session_id);
1611
1612 end:
1613 rcu_read_unlock();
1614 return ret;
1615 }
1616
1617 /*
1618 * Ask the consumer to rotate a channel.
1619 *
1620 * The new_chunk_id is the session->rotate_count that has been incremented
1621 * when the rotation started. On the relay, this allows to keep track in which
1622 * chunk each stream is currently writing to (for the rotate_pending operation).
1623 */
1624 int consumer_rotate_channel(struct consumer_socket *socket, uint64_t key,
1625 uid_t uid, gid_t gid, struct consumer_output *output,
1626 bool is_metadata_channel)
1627 {
1628 int ret;
1629 struct lttcomm_consumer_msg msg;
1630
1631 assert(socket);
1632
1633 DBG("Consumer rotate channel key %" PRIu64, key);
1634
1635 pthread_mutex_lock(socket->lock);
1636 memset(&msg, 0, sizeof(msg));
1637 msg.cmd_type = LTTNG_CONSUMER_ROTATE_CHANNEL;
1638 msg.u.rotate_channel.key = key;
1639 msg.u.rotate_channel.metadata = !!is_metadata_channel;
1640
1641 if (output->type == CONSUMER_DST_NET) {
1642 msg.u.rotate_channel.relayd_id = output->net_seq_index;
1643 } else {
1644 msg.u.rotate_channel.relayd_id = (uint64_t) -1ULL;
1645 }
1646
1647 health_code_update();
1648 ret = consumer_send_msg(socket, &msg);
1649 if (ret < 0) {
1650 switch (-ret) {
1651 case LTTCOMM_CONSUMERD_CHAN_NOT_FOUND:
1652 ret = -LTTNG_ERR_CHAN_NOT_FOUND;
1653 break;
1654 default:
1655 ret = -LTTNG_ERR_ROTATION_FAIL_CONSUMER;
1656 break;
1657 }
1658 goto error;
1659 }
1660 error:
1661 pthread_mutex_unlock(socket->lock);
1662 health_code_update();
1663 return ret;
1664 }
1665
1666 int consumer_init(struct consumer_socket *socket,
1667 const lttng_uuid sessiond_uuid)
1668 {
1669 int ret;
1670 struct lttcomm_consumer_msg msg = {
1671 .cmd_type = LTTNG_CONSUMER_INIT,
1672 };
1673
1674 assert(socket);
1675
1676 DBG("Sending consumer initialization command");
1677 lttng_uuid_copy(msg.u.init.sessiond_uuid, sessiond_uuid);
1678
1679 health_code_update();
1680 ret = consumer_send_msg(socket, &msg);
1681 if (ret < 0) {
1682 goto error;
1683 }
1684
1685 error:
1686 health_code_update();
1687 return ret;
1688 }
1689
1690 /*
1691 * Ask the consumer to create a new chunk for a given session.
1692 *
1693 * Called with the consumer socket lock held.
1694 */
1695 int consumer_create_trace_chunk(struct consumer_socket *socket,
1696 uint64_t relayd_id, uint64_t session_id,
1697 struct lttng_trace_chunk *chunk)
1698 {
1699 int ret;
1700 enum lttng_trace_chunk_status chunk_status;
1701 struct lttng_credentials chunk_credentials;
1702 const struct lttng_directory_handle *chunk_directory_handle;
1703 int chunk_dirfd;
1704 const char *chunk_name;
1705 bool chunk_name_overriden;
1706 uint64_t chunk_id;
1707 time_t creation_timestamp;
1708 char creation_timestamp_buffer[ISO8601_STR_LEN];
1709 const char *creation_timestamp_str = "(none)";
1710 const bool chunk_has_local_output = relayd_id == -1ULL;
1711 struct lttcomm_consumer_msg msg = {
1712 .cmd_type = LTTNG_CONSUMER_CREATE_TRACE_CHUNK,
1713 .u.create_trace_chunk.session_id = session_id,
1714 };
1715
1716 assert(socket);
1717 assert(chunk);
1718
1719 if (relayd_id != -1ULL) {
1720 LTTNG_OPTIONAL_SET(&msg.u.create_trace_chunk.relayd_id,
1721 relayd_id);
1722 }
1723
1724 chunk_status = lttng_trace_chunk_get_name(chunk, &chunk_name,
1725 &chunk_name_overriden);
1726 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK &&
1727 chunk_status != LTTNG_TRACE_CHUNK_STATUS_NONE) {
1728 ERR("Failed to get name of trace chunk");
1729 ret = -LTTNG_ERR_FATAL;
1730 goto error;
1731 }
1732 if (chunk_name_overriden) {
1733 ret = lttng_strncpy(msg.u.create_trace_chunk.override_name,
1734 chunk_name,
1735 sizeof(msg.u.create_trace_chunk.override_name));
1736 if (ret) {
1737 ERR("Trace chunk name \"%s\" exceeds the maximal length allowed by the consumer protocol",
1738 chunk_name);
1739 ret = -LTTNG_ERR_FATAL;
1740 goto error;
1741 }
1742 }
1743
1744 chunk_status = lttng_trace_chunk_get_creation_timestamp(chunk,
1745 &creation_timestamp);
1746 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1747 ret = -LTTNG_ERR_FATAL;
1748 goto error;
1749 }
1750 msg.u.create_trace_chunk.creation_timestamp =
1751 (uint64_t) creation_timestamp;
1752 /* Only used for logging purposes. */
1753 ret = time_to_iso8601_str(creation_timestamp,
1754 creation_timestamp_buffer,
1755 sizeof(creation_timestamp_buffer));
1756 creation_timestamp_str = !ret ? creation_timestamp_buffer :
1757 "(formatting error)";
1758
1759 chunk_status = lttng_trace_chunk_get_id(chunk, &chunk_id);
1760 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1761 /*
1762 * Anonymous trace chunks should never be transmitted
1763 * to remote peers (consumerd and relayd). They are used
1764 * internally for backward-compatibility purposes.
1765 */
1766 ret = -LTTNG_ERR_FATAL;
1767 goto error;
1768 }
1769 msg.u.create_trace_chunk.chunk_id = chunk_id;
1770
1771 if (chunk_has_local_output) {
1772 chunk_status = lttng_trace_chunk_get_chunk_directory_handle(
1773 chunk, &chunk_directory_handle);
1774 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1775 ret = -LTTNG_ERR_FATAL;
1776 goto error;
1777 }
1778
1779 /*
1780 * This will only compile on platforms that support
1781 * dirfd (POSIX.2008). This is fine as the session daemon
1782 * is only built for such platforms.
1783 *
1784 * The ownership of the chunk directory handle's is maintained
1785 * by the trace chunk.
1786 */
1787 chunk_dirfd = lttng_directory_handle_get_dirfd(
1788 chunk_directory_handle);
1789 assert(chunk_dirfd >= 0);
1790
1791 chunk_status = lttng_trace_chunk_get_credentials(
1792 chunk, &chunk_credentials);
1793 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1794 /*
1795 * Not associating credentials to a sessiond chunk is a
1796 * fatal internal error.
1797 */
1798 ret = -LTTNG_ERR_FATAL;
1799 goto error;
1800 }
1801 msg.u.create_trace_chunk.credentials.value.uid =
1802 chunk_credentials.uid;
1803 msg.u.create_trace_chunk.credentials.value.gid =
1804 chunk_credentials.gid;
1805 msg.u.create_trace_chunk.credentials.is_set = 1;
1806 }
1807
1808 DBG("Sending consumer create trace chunk command: relayd_id = %" PRId64
1809 ", session_id = %" PRIu64 ", chunk_id = %" PRIu64
1810 ", creation_timestamp = %s",
1811 relayd_id, session_id, chunk_id,
1812 creation_timestamp_str);
1813 health_code_update();
1814 ret = consumer_send_msg(socket, &msg);
1815 health_code_update();
1816 if (ret < 0) {
1817 ERR("Trace chunk creation error on consumer");
1818 ret = -LTTNG_ERR_CREATE_TRACE_CHUNK_FAIL_CONSUMER;
1819 goto error;
1820 }
1821
1822 if (chunk_has_local_output) {
1823 DBG("Sending trace chunk directory fd to consumer");
1824 health_code_update();
1825 ret = consumer_send_fds(socket, &chunk_dirfd, 1);
1826 health_code_update();
1827 if (ret < 0) {
1828 ERR("Trace chunk creation error on consumer");
1829 ret = -LTTNG_ERR_CREATE_TRACE_CHUNK_FAIL_CONSUMER;
1830 goto error;
1831 }
1832 }
1833 error:
1834 return ret;
1835 }
1836
1837 /*
1838 * Ask the consumer to close a trace chunk for a given session.
1839 *
1840 * Called with the consumer socket lock held.
1841 */
1842 int consumer_close_trace_chunk(struct consumer_socket *socket,
1843 uint64_t relayd_id, uint64_t session_id,
1844 struct lttng_trace_chunk *chunk)
1845 {
1846 int ret;
1847 enum lttng_trace_chunk_status chunk_status;
1848 struct lttcomm_consumer_msg msg = {
1849 .cmd_type = LTTNG_CONSUMER_CLOSE_TRACE_CHUNK,
1850 .u.close_trace_chunk.session_id = session_id,
1851 };
1852 uint64_t chunk_id;
1853 time_t close_timestamp;
1854 enum lttng_trace_chunk_command_type close_command;
1855 const char *close_command_name = "none";
1856
1857 assert(socket);
1858
1859 if (relayd_id != -1ULL) {
1860 LTTNG_OPTIONAL_SET(
1861 &msg.u.close_trace_chunk.relayd_id, relayd_id);
1862 }
1863
1864 chunk_status = lttng_trace_chunk_get_close_command(
1865 chunk, &close_command);
1866 switch (chunk_status) {
1867 case LTTNG_TRACE_CHUNK_STATUS_OK:
1868 LTTNG_OPTIONAL_SET(&msg.u.close_trace_chunk.close_command,
1869 (uint32_t) close_command);
1870 break;
1871 case LTTNG_TRACE_CHUNK_STATUS_NONE:
1872 break;
1873 default:
1874 ERR("Failed to get trace chunk close command");
1875 ret = -1;
1876 goto error;
1877 }
1878
1879 chunk_status = lttng_trace_chunk_get_id(chunk, &chunk_id);
1880 /*
1881 * Anonymous trace chunks should never be transmitted to remote peers
1882 * (consumerd and relayd). They are used internally for
1883 * backward-compatibility purposes.
1884 */
1885 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
1886 msg.u.close_trace_chunk.chunk_id = chunk_id;
1887
1888 chunk_status = lttng_trace_chunk_get_close_timestamp(chunk,
1889 &close_timestamp);
1890 /*
1891 * A trace chunk should be closed locally before being closed remotely.
1892 * Otherwise, the close timestamp would never be transmitted to the
1893 * peers.
1894 */
1895 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
1896 msg.u.close_trace_chunk.close_timestamp = (uint64_t) close_timestamp;
1897
1898 if (msg.u.close_trace_chunk.close_command.is_set) {
1899 close_command_name = lttng_trace_chunk_command_type_get_name(
1900 close_command);
1901 }
1902 DBG("Sending consumer close trace chunk command: relayd_id = %" PRId64
1903 ", session_id = %" PRIu64 ", chunk_id = %" PRIu64
1904 ", close command = \"%s\"",
1905 relayd_id, session_id, chunk_id, close_command_name);
1906
1907 health_code_update();
1908 ret = consumer_send_msg(socket, &msg);
1909 if (ret < 0) {
1910 ret = -LTTNG_ERR_CLOSE_TRACE_CHUNK_FAIL_CONSUMER;
1911 goto error;
1912 }
1913
1914 error:
1915 health_code_update();
1916 return ret;
1917 }
1918
1919 /*
1920 * Ask the consumer if a trace chunk exists.
1921 *
1922 * Called with the consumer socket lock held.
1923 * Returns 0 on success, or a negative value on error.
1924 */
1925 int consumer_trace_chunk_exists(struct consumer_socket *socket,
1926 uint64_t relayd_id, uint64_t session_id,
1927 struct lttng_trace_chunk *chunk,
1928 enum consumer_trace_chunk_exists_status *result)
1929 {
1930 int ret;
1931 enum lttng_trace_chunk_status chunk_status;
1932 struct lttcomm_consumer_msg msg = {
1933 .cmd_type = LTTNG_CONSUMER_TRACE_CHUNK_EXISTS,
1934 .u.trace_chunk_exists.session_id = session_id,
1935 };
1936 uint64_t chunk_id;
1937 const char *consumer_reply_str;
1938
1939 assert(socket);
1940
1941 if (relayd_id != -1ULL) {
1942 LTTNG_OPTIONAL_SET(&msg.u.trace_chunk_exists.relayd_id,
1943 relayd_id);
1944 }
1945
1946 chunk_status = lttng_trace_chunk_get_id(chunk, &chunk_id);
1947 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1948 /*
1949 * Anonymous trace chunks should never be transmitted
1950 * to remote peers (consumerd and relayd). They are used
1951 * internally for backward-compatibility purposes.
1952 */
1953 ret = -LTTNG_ERR_FATAL;
1954 goto error;
1955 }
1956 msg.u.trace_chunk_exists.chunk_id = chunk_id;
1957
1958 DBG("Sending consumer trace chunk exists command: relayd_id = %" PRId64
1959 ", session_id = %" PRIu64
1960 ", chunk_id = %" PRIu64, relayd_id, session_id, chunk_id);
1961
1962 health_code_update();
1963 ret = consumer_send_msg(socket, &msg);
1964 switch (-ret) {
1965 case LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK:
1966 consumer_reply_str = "unknown trace chunk";
1967 *result = CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK;
1968 break;
1969 case LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_LOCAL:
1970 consumer_reply_str = "trace chunk exists locally";
1971 *result = CONSUMER_TRACE_CHUNK_EXISTS_STATUS_EXISTS_LOCAL;
1972 break;
1973 case LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_REMOTE:
1974 consumer_reply_str = "trace chunk exists on remote peer";
1975 *result = CONSUMER_TRACE_CHUNK_EXISTS_STATUS_EXISTS_REMOTE;
1976 break;
1977 default:
1978 ERR("Consumer returned an error from TRACE_CHUNK_EXISTS command");
1979 ret = -1;
1980 goto error;
1981 }
1982 DBG("Consumer reply to TRACE_CHUNK_EXISTS command: %s",
1983 consumer_reply_str);
1984 ret = 0;
1985 error:
1986 health_code_update();
1987 return ret;
1988 }
This page took 0.101599 seconds and 6 git commands to generate.