consumerd: tag metadata channel as being part of a live session
[lttng-tools.git] / src / bin / lttng-sessiond / consumer.c
1 /*
2 * Copyright (C) 2012 David Goulet <dgoulet@efficios.com>
3 * Copyright (C) 2018 Jérémie Galarneau <jeremie.galarneau@efficios.com>
4 *
5 * SPDX-License-Identifier: GPL-2.0-only
6 *
7 */
8
9 #define _LGPL_SOURCE
10 #include <assert.h>
11 #include <stdio.h>
12 #include <stdlib.h>
13 #include <string.h>
14 #include <sys/stat.h>
15 #include <sys/types.h>
16 #include <unistd.h>
17 #include <inttypes.h>
18
19 #include <common/common.h>
20 #include <common/defaults.h>
21 #include <common/uri.h>
22 #include <common/relayd/relayd.h>
23 #include <common/string-utils/format.h>
24
25 #include "consumer.h"
26 #include "health-sessiond.h"
27 #include "ust-app.h"
28 #include "utils.h"
29 #include "lttng-sessiond.h"
30
31 /*
32 * Return allocated full pathname of the session using the consumer trace path
33 * and subdir if available.
34 *
35 * The caller can safely free(3) the returned value. On error, NULL is
36 * returned.
37 */
38 char *setup_channel_trace_path(struct consumer_output *consumer,
39 const char *session_path, size_t *consumer_path_offset)
40 {
41 int ret;
42 char *pathname;
43
44 assert(consumer);
45 assert(session_path);
46
47 health_code_update();
48
49 /*
50 * Allocate the string ourself to make sure we never exceed
51 * LTTNG_PATH_MAX.
52 */
53 pathname = zmalloc(LTTNG_PATH_MAX);
54 if (!pathname) {
55 goto error;
56 }
57
58 /* Get correct path name destination */
59 if (consumer->type == CONSUMER_DST_NET &&
60 consumer->relay_major_version == 2 &&
61 consumer->relay_minor_version < 11) {
62 ret = snprintf(pathname, LTTNG_PATH_MAX, "%s%s/%s/%s",
63 consumer->dst.net.base_dir,
64 consumer->chunk_path, consumer->domain_subdir,
65 session_path);
66 *consumer_path_offset = 0;
67 } else {
68 ret = snprintf(pathname, LTTNG_PATH_MAX, "%s/%s",
69 consumer->domain_subdir, session_path);
70 *consumer_path_offset = strlen(consumer->domain_subdir) + 1;
71 }
72 DBG3("Consumer trace path relative to current trace chunk: \"%s\"",
73 pathname);
74 if (ret < 0) {
75 PERROR("Failed to format channel path");
76 goto error;
77 } else if (ret >= LTTNG_PATH_MAX) {
78 ERR("Truncation occurred while formatting channel path");
79 goto error;
80 }
81
82 return pathname;
83 error:
84 free(pathname);
85 return NULL;
86 }
87
88 /*
89 * Send a data payload using a given consumer socket of size len.
90 *
91 * The consumer socket lock MUST be acquired before calling this since this
92 * function can change the fd value.
93 *
94 * Return 0 on success else a negative value on error.
95 */
96 int consumer_socket_send(struct consumer_socket *socket, void *msg, size_t len)
97 {
98 int fd;
99 ssize_t size;
100
101 assert(socket);
102 assert(socket->fd_ptr);
103 assert(msg);
104
105 /* Consumer socket is invalid. Stopping. */
106 fd = *socket->fd_ptr;
107 if (fd < 0) {
108 goto error;
109 }
110
111 size = lttcomm_send_unix_sock(fd, msg, len);
112 if (size < 0) {
113 /* The above call will print a PERROR on error. */
114 DBG("Error when sending data to consumer on sock %d", fd);
115 /*
116 * At this point, the socket is not usable anymore thus closing it and
117 * setting the file descriptor to -1 so it is not reused.
118 */
119
120 /* This call will PERROR on error. */
121 (void) lttcomm_close_unix_sock(fd);
122 *socket->fd_ptr = -1;
123 goto error;
124 }
125
126 return 0;
127
128 error:
129 return -1;
130 }
131
132 /*
133 * Receive a data payload using a given consumer socket of size len.
134 *
135 * The consumer socket lock MUST be acquired before calling this since this
136 * function can change the fd value.
137 *
138 * Return 0 on success else a negative value on error.
139 */
140 int consumer_socket_recv(struct consumer_socket *socket, void *msg, size_t len)
141 {
142 int fd;
143 ssize_t size;
144
145 assert(socket);
146 assert(socket->fd_ptr);
147 assert(msg);
148
149 /* Consumer socket is invalid. Stopping. */
150 fd = *socket->fd_ptr;
151 if (fd < 0) {
152 goto error;
153 }
154
155 size = lttcomm_recv_unix_sock(fd, msg, len);
156 if (size <= 0) {
157 /* The above call will print a PERROR on error. */
158 DBG("Error when receiving data from the consumer socket %d", fd);
159 /*
160 * At this point, the socket is not usable anymore thus closing it and
161 * setting the file descriptor to -1 so it is not reused.
162 */
163
164 /* This call will PERROR on error. */
165 (void) lttcomm_close_unix_sock(fd);
166 *socket->fd_ptr = -1;
167 goto error;
168 }
169
170 return 0;
171
172 error:
173 return -1;
174 }
175
176 /*
177 * Receive a reply command status message from the consumer. Consumer socket
178 * lock MUST be acquired before calling this function.
179 *
180 * Return 0 on success, -1 on recv error or a negative lttng error code which
181 * was possibly returned by the consumer.
182 */
183 int consumer_recv_status_reply(struct consumer_socket *sock)
184 {
185 int ret;
186 struct lttcomm_consumer_status_msg reply;
187
188 assert(sock);
189
190 ret = consumer_socket_recv(sock, &reply, sizeof(reply));
191 if (ret < 0) {
192 goto end;
193 }
194
195 if (reply.ret_code == LTTCOMM_CONSUMERD_SUCCESS) {
196 /* All good. */
197 ret = 0;
198 } else {
199 ret = -reply.ret_code;
200 DBG("Consumer ret code %d", ret);
201 }
202
203 end:
204 return ret;
205 }
206
207 /*
208 * Once the ASK_CHANNEL command is sent to the consumer, the channel
209 * information are sent back. This call receives that data and populates key
210 * and stream_count.
211 *
212 * On success return 0 and both key and stream_count are set. On error, a
213 * negative value is sent back and both parameters are untouched.
214 */
215 int consumer_recv_status_channel(struct consumer_socket *sock,
216 uint64_t *key, unsigned int *stream_count)
217 {
218 int ret;
219 struct lttcomm_consumer_status_channel reply;
220
221 assert(sock);
222 assert(stream_count);
223 assert(key);
224
225 ret = consumer_socket_recv(sock, &reply, sizeof(reply));
226 if (ret < 0) {
227 goto end;
228 }
229
230 /* An error is possible so don't touch the key and stream_count. */
231 if (reply.ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
232 ret = -1;
233 goto end;
234 }
235
236 *key = reply.key;
237 *stream_count = reply.stream_count;
238 ret = 0;
239
240 end:
241 return ret;
242 }
243
244 /*
245 * Send destroy relayd command to consumer.
246 *
247 * On success return positive value. On error, negative value.
248 */
249 int consumer_send_destroy_relayd(struct consumer_socket *sock,
250 struct consumer_output *consumer)
251 {
252 int ret;
253 struct lttcomm_consumer_msg msg;
254
255 assert(consumer);
256 assert(sock);
257
258 DBG2("Sending destroy relayd command to consumer sock %d", *sock->fd_ptr);
259
260 memset(&msg, 0, sizeof(msg));
261 msg.cmd_type = LTTNG_CONSUMER_DESTROY_RELAYD;
262 msg.u.destroy_relayd.net_seq_idx = consumer->net_seq_index;
263
264 pthread_mutex_lock(sock->lock);
265 ret = consumer_socket_send(sock, &msg, sizeof(msg));
266 if (ret < 0) {
267 goto error;
268 }
269
270 /* Don't check the return value. The caller will do it. */
271 ret = consumer_recv_status_reply(sock);
272
273 DBG2("Consumer send destroy relayd command done");
274
275 error:
276 pthread_mutex_unlock(sock->lock);
277 return ret;
278 }
279
280 /*
281 * For each consumer socket in the consumer output object, send a destroy
282 * relayd command.
283 */
284 void consumer_output_send_destroy_relayd(struct consumer_output *consumer)
285 {
286 struct lttng_ht_iter iter;
287 struct consumer_socket *socket;
288
289 assert(consumer);
290
291 /* Destroy any relayd connection */
292 if (consumer->type == CONSUMER_DST_NET) {
293 rcu_read_lock();
294 cds_lfht_for_each_entry(consumer->socks->ht, &iter.iter, socket,
295 node.node) {
296 int ret;
297
298 /* Send destroy relayd command */
299 ret = consumer_send_destroy_relayd(socket, consumer);
300 if (ret < 0) {
301 DBG("Unable to send destroy relayd command to consumer");
302 /* Continue since we MUST delete everything at this point. */
303 }
304 }
305 rcu_read_unlock();
306 }
307 }
308
309 /*
310 * From a consumer_data structure, allocate and add a consumer socket to the
311 * consumer output.
312 *
313 * Return 0 on success, else negative value on error
314 */
315 int consumer_create_socket(struct consumer_data *data,
316 struct consumer_output *output)
317 {
318 int ret = 0;
319 struct consumer_socket *socket;
320
321 assert(data);
322
323 if (output == NULL || data->cmd_sock < 0) {
324 /*
325 * Not an error. Possible there is simply not spawned consumer or it's
326 * disabled for the tracing session asking the socket.
327 */
328 goto error;
329 }
330
331 rcu_read_lock();
332 socket = consumer_find_socket(data->cmd_sock, output);
333 rcu_read_unlock();
334 if (socket == NULL) {
335 socket = consumer_allocate_socket(&data->cmd_sock);
336 if (socket == NULL) {
337 ret = -1;
338 goto error;
339 }
340
341 socket->registered = 0;
342 socket->lock = &data->lock;
343 rcu_read_lock();
344 consumer_add_socket(socket, output);
345 rcu_read_unlock();
346 }
347
348 socket->type = data->type;
349
350 DBG3("Consumer socket created (fd: %d) and added to output",
351 data->cmd_sock);
352
353 error:
354 return ret;
355 }
356
357 /*
358 * Return the consumer socket from the given consumer output with the right
359 * bitness. On error, returns NULL.
360 *
361 * The caller MUST acquire a rcu read side lock and keep it until the socket
362 * object reference is not needed anymore.
363 */
364 struct consumer_socket *consumer_find_socket_by_bitness(int bits,
365 const struct consumer_output *consumer)
366 {
367 int consumer_fd;
368 struct consumer_socket *socket = NULL;
369
370 switch (bits) {
371 case 64:
372 consumer_fd = uatomic_read(&ust_consumerd64_fd);
373 break;
374 case 32:
375 consumer_fd = uatomic_read(&ust_consumerd32_fd);
376 break;
377 default:
378 assert(0);
379 goto end;
380 }
381
382 socket = consumer_find_socket(consumer_fd, consumer);
383 if (!socket) {
384 ERR("Consumer socket fd %d not found in consumer obj %p",
385 consumer_fd, consumer);
386 }
387
388 end:
389 return socket;
390 }
391
392 /*
393 * Find a consumer_socket in a consumer_output hashtable. Read side lock must
394 * be acquired before calling this function and across use of the
395 * returned consumer_socket.
396 */
397 struct consumer_socket *consumer_find_socket(int key,
398 const struct consumer_output *consumer)
399 {
400 struct lttng_ht_iter iter;
401 struct lttng_ht_node_ulong *node;
402 struct consumer_socket *socket = NULL;
403
404 /* Negative keys are lookup failures */
405 if (key < 0 || consumer == NULL) {
406 return NULL;
407 }
408
409 lttng_ht_lookup(consumer->socks, (void *)((unsigned long) key),
410 &iter);
411 node = lttng_ht_iter_get_node_ulong(&iter);
412 if (node != NULL) {
413 socket = caa_container_of(node, struct consumer_socket, node);
414 }
415
416 return socket;
417 }
418
419 /*
420 * Allocate a new consumer_socket and return the pointer.
421 */
422 struct consumer_socket *consumer_allocate_socket(int *fd)
423 {
424 struct consumer_socket *socket = NULL;
425
426 assert(fd);
427
428 socket = zmalloc(sizeof(struct consumer_socket));
429 if (socket == NULL) {
430 PERROR("zmalloc consumer socket");
431 goto error;
432 }
433
434 socket->fd_ptr = fd;
435 lttng_ht_node_init_ulong(&socket->node, *fd);
436
437 error:
438 return socket;
439 }
440
441 /*
442 * Add consumer socket to consumer output object. Read side lock must be
443 * acquired before calling this function.
444 */
445 void consumer_add_socket(struct consumer_socket *sock,
446 struct consumer_output *consumer)
447 {
448 assert(sock);
449 assert(consumer);
450
451 lttng_ht_add_unique_ulong(consumer->socks, &sock->node);
452 }
453
454 /*
455 * Delete consumer socket to consumer output object. Read side lock must be
456 * acquired before calling this function.
457 */
458 void consumer_del_socket(struct consumer_socket *sock,
459 struct consumer_output *consumer)
460 {
461 int ret;
462 struct lttng_ht_iter iter;
463
464 assert(sock);
465 assert(consumer);
466
467 iter.iter.node = &sock->node.node;
468 ret = lttng_ht_del(consumer->socks, &iter);
469 assert(!ret);
470 }
471
472 /*
473 * RCU destroy call function.
474 */
475 static void destroy_socket_rcu(struct rcu_head *head)
476 {
477 struct lttng_ht_node_ulong *node =
478 caa_container_of(head, struct lttng_ht_node_ulong, head);
479 struct consumer_socket *socket =
480 caa_container_of(node, struct consumer_socket, node);
481
482 free(socket);
483 }
484
485 /*
486 * Destroy and free socket pointer in a call RCU. Read side lock must be
487 * acquired before calling this function.
488 */
489 void consumer_destroy_socket(struct consumer_socket *sock)
490 {
491 assert(sock);
492
493 /*
494 * We DO NOT close the file descriptor here since it is global to the
495 * session daemon and is closed only if the consumer dies or a custom
496 * consumer was registered,
497 */
498 if (sock->registered) {
499 DBG3("Consumer socket was registered. Closing fd %d", *sock->fd_ptr);
500 lttcomm_close_unix_sock(*sock->fd_ptr);
501 }
502
503 call_rcu(&sock->node.head, destroy_socket_rcu);
504 }
505
506 /*
507 * Allocate and assign data to a consumer_output object.
508 *
509 * Return pointer to structure.
510 */
511 struct consumer_output *consumer_create_output(enum consumer_dst_type type)
512 {
513 struct consumer_output *output = NULL;
514
515 output = zmalloc(sizeof(struct consumer_output));
516 if (output == NULL) {
517 PERROR("zmalloc consumer_output");
518 goto error;
519 }
520
521 /* By default, consumer output is enabled */
522 output->enabled = 1;
523 output->type = type;
524 output->net_seq_index = (uint64_t) -1ULL;
525 urcu_ref_init(&output->ref);
526
527 output->socks = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
528
529 error:
530 return output;
531 }
532
533 /*
534 * Iterate over the consumer output socket hash table and destroy them. The
535 * socket file descriptor are only closed if the consumer output was
536 * registered meaning it's an external consumer.
537 */
538 void consumer_destroy_output_sockets(struct consumer_output *obj)
539 {
540 struct lttng_ht_iter iter;
541 struct consumer_socket *socket;
542
543 if (!obj->socks) {
544 return;
545 }
546
547 rcu_read_lock();
548 cds_lfht_for_each_entry(obj->socks->ht, &iter.iter, socket, node.node) {
549 consumer_del_socket(socket, obj);
550 consumer_destroy_socket(socket);
551 }
552 rcu_read_unlock();
553 }
554
555 /*
556 * Delete the consumer_output object from the list and free the ptr.
557 *
558 * Should *NOT* be called with RCU read-side lock held.
559 */
560 static void consumer_release_output(struct urcu_ref *ref)
561 {
562 struct consumer_output *obj =
563 caa_container_of(ref, struct consumer_output, ref);
564
565 consumer_destroy_output_sockets(obj);
566
567 if (obj->socks) {
568 /* Finally destroy HT */
569 ht_cleanup_push(obj->socks);
570 }
571
572 free(obj);
573 }
574
575 /*
576 * Get the consumer_output object.
577 */
578 void consumer_output_get(struct consumer_output *obj)
579 {
580 urcu_ref_get(&obj->ref);
581 }
582
583 /*
584 * Put the consumer_output object.
585 *
586 * Should *NOT* be called with RCU read-side lock held.
587 */
588 void consumer_output_put(struct consumer_output *obj)
589 {
590 if (!obj) {
591 return;
592 }
593 urcu_ref_put(&obj->ref, consumer_release_output);
594 }
595
596 /*
597 * Copy consumer output and returned the newly allocated copy.
598 *
599 * Should *NOT* be called with RCU read-side lock held.
600 */
601 struct consumer_output *consumer_copy_output(struct consumer_output *src)
602 {
603 int ret;
604 struct consumer_output *output;
605
606 assert(src);
607
608 output = consumer_create_output(src->type);
609 if (output == NULL) {
610 goto end;
611 }
612 output->enabled = src->enabled;
613 output->net_seq_index = src->net_seq_index;
614 memcpy(output->domain_subdir, src->domain_subdir,
615 sizeof(output->domain_subdir));
616 output->snapshot = src->snapshot;
617 output->relay_major_version = src->relay_major_version;
618 output->relay_minor_version = src->relay_minor_version;
619 output->relay_allows_clear = src->relay_allows_clear;
620 memcpy(&output->dst, &src->dst, sizeof(output->dst));
621 ret = consumer_copy_sockets(output, src);
622 if (ret < 0) {
623 goto error_put;
624 }
625 end:
626 return output;
627
628 error_put:
629 consumer_output_put(output);
630 return NULL;
631 }
632
633 /*
634 * Copy consumer sockets from src to dst.
635 *
636 * Return 0 on success or else a negative value.
637 */
638 int consumer_copy_sockets(struct consumer_output *dst,
639 struct consumer_output *src)
640 {
641 int ret = 0;
642 struct lttng_ht_iter iter;
643 struct consumer_socket *socket, *copy_sock;
644
645 assert(dst);
646 assert(src);
647
648 rcu_read_lock();
649 cds_lfht_for_each_entry(src->socks->ht, &iter.iter, socket, node.node) {
650 /* Ignore socket that are already there. */
651 copy_sock = consumer_find_socket(*socket->fd_ptr, dst);
652 if (copy_sock) {
653 continue;
654 }
655
656 /* Create new socket object. */
657 copy_sock = consumer_allocate_socket(socket->fd_ptr);
658 if (copy_sock == NULL) {
659 rcu_read_unlock();
660 ret = -ENOMEM;
661 goto error;
662 }
663
664 copy_sock->registered = socket->registered;
665 /*
666 * This is valid because this lock is shared accross all consumer
667 * object being the global lock of the consumer data structure of the
668 * session daemon.
669 */
670 copy_sock->lock = socket->lock;
671 consumer_add_socket(copy_sock, dst);
672 }
673 rcu_read_unlock();
674
675 error:
676 return ret;
677 }
678
679 /*
680 * Set network URI to the consumer output.
681 *
682 * Return 0 on success. Return 1 if the URI were equal. Else, negative value on
683 * error.
684 */
685 int consumer_set_network_uri(const struct ltt_session *session,
686 struct consumer_output *output,
687 struct lttng_uri *uri)
688 {
689 int ret;
690 struct lttng_uri *dst_uri = NULL;
691
692 /* Code flow error safety net. */
693 assert(output);
694 assert(uri);
695
696 switch (uri->stype) {
697 case LTTNG_STREAM_CONTROL:
698 dst_uri = &output->dst.net.control;
699 output->dst.net.control_isset = 1;
700 if (uri->port == 0) {
701 /* Assign default port. */
702 uri->port = DEFAULT_NETWORK_CONTROL_PORT;
703 } else {
704 if (output->dst.net.data_isset && uri->port ==
705 output->dst.net.data.port) {
706 ret = -LTTNG_ERR_INVALID;
707 goto error;
708 }
709 }
710 DBG3("Consumer control URI set with port %d", uri->port);
711 break;
712 case LTTNG_STREAM_DATA:
713 dst_uri = &output->dst.net.data;
714 output->dst.net.data_isset = 1;
715 if (uri->port == 0) {
716 /* Assign default port. */
717 uri->port = DEFAULT_NETWORK_DATA_PORT;
718 } else {
719 if (output->dst.net.control_isset && uri->port ==
720 output->dst.net.control.port) {
721 ret = -LTTNG_ERR_INVALID;
722 goto error;
723 }
724 }
725 DBG3("Consumer data URI set with port %d", uri->port);
726 break;
727 default:
728 ERR("Set network uri type unknown %d", uri->stype);
729 ret = -LTTNG_ERR_INVALID;
730 goto error;
731 }
732
733 ret = uri_compare(dst_uri, uri);
734 if (!ret) {
735 /* Same URI, don't touch it and return success. */
736 DBG3("URI network compare are the same");
737 goto equal;
738 }
739
740 /* URIs were not equal, replacing it. */
741 memcpy(dst_uri, uri, sizeof(struct lttng_uri));
742 output->type = CONSUMER_DST_NET;
743 if (dst_uri->stype != LTTNG_STREAM_CONTROL) {
744 /* Only the control uri needs to contain the path. */
745 goto end;
746 }
747
748 /*
749 * If the user has specified a subdir as part of the control
750 * URL, the session's base output directory is:
751 * /RELAYD_OUTPUT_PATH/HOSTNAME/USER_SPECIFIED_DIR
752 *
753 * Hence, the "base_dir" from which all stream files and
754 * session rotation chunks are created takes the form
755 * /HOSTNAME/USER_SPECIFIED_DIR
756 *
757 * If the user has not specified an output directory as part of
758 * the control URL, the base output directory has the form:
759 * /RELAYD_OUTPUT_PATH/HOSTNAME/SESSION_NAME-CREATION_TIME
760 *
761 * Hence, the "base_dir" from which all stream files and
762 * session rotation chunks are created takes the form
763 * /HOSTNAME/SESSION_NAME-CREATION_TIME
764 *
765 * Note that automatically generated session names already
766 * contain the session's creation time. In that case, the
767 * creation time is omitted to prevent it from being duplicated
768 * in the final directory hierarchy.
769 */
770 if (*uri->subdir) {
771 if (strstr(uri->subdir, "../")) {
772 ERR("Network URI subdirs are not allowed to walk up the path hierarchy");
773 ret = -LTTNG_ERR_INVALID;
774 goto error;
775 }
776 ret = snprintf(output->dst.net.base_dir,
777 sizeof(output->dst.net.base_dir),
778 "/%s/%s/", session->hostname, uri->subdir);
779 } else {
780 if (session->has_auto_generated_name) {
781 ret = snprintf(output->dst.net.base_dir,
782 sizeof(output->dst.net.base_dir),
783 "/%s/%s/", session->hostname,
784 session->name);
785 } else {
786 char session_creation_datetime[16];
787 size_t strftime_ret;
788 struct tm *timeinfo;
789
790 timeinfo = localtime(&session->creation_time);
791 if (!timeinfo) {
792 ret = -LTTNG_ERR_FATAL;
793 goto error;
794 }
795 strftime_ret = strftime(session_creation_datetime,
796 sizeof(session_creation_datetime),
797 "%Y%m%d-%H%M%S", timeinfo);
798 if (strftime_ret == 0) {
799 ERR("Failed to format session creation timestamp while setting network URI");
800 ret = -LTTNG_ERR_FATAL;
801 goto error;
802 }
803 ret = snprintf(output->dst.net.base_dir,
804 sizeof(output->dst.net.base_dir),
805 "/%s/%s-%s/", session->hostname,
806 session->name,
807 session_creation_datetime);
808 }
809 }
810 if (ret >= sizeof(output->dst.net.base_dir)) {
811 ret = -LTTNG_ERR_INVALID;
812 ERR("Truncation occurred while setting network output base directory");
813 goto error;
814 } else if (ret == -1) {
815 ret = -LTTNG_ERR_INVALID;
816 PERROR("Error occurred while setting network output base directory");
817 goto error;
818 }
819
820 DBG3("Consumer set network uri base_dir path %s",
821 output->dst.net.base_dir);
822
823 end:
824 return 0;
825 equal:
826 return 1;
827 error:
828 return ret;
829 }
830
831 /*
832 * Send file descriptor to consumer via sock.
833 *
834 * The consumer socket lock must be held by the caller.
835 */
836 int consumer_send_fds(struct consumer_socket *sock, const int *fds,
837 size_t nb_fd)
838 {
839 int ret;
840
841 assert(fds);
842 assert(sock);
843 assert(nb_fd > 0);
844 assert(pthread_mutex_trylock(sock->lock) == EBUSY);
845
846 ret = lttcomm_send_fds_unix_sock(*sock->fd_ptr, fds, nb_fd);
847 if (ret < 0) {
848 /* The above call will print a PERROR on error. */
849 DBG("Error when sending consumer fds on sock %d", *sock->fd_ptr);
850 goto error;
851 }
852
853 ret = consumer_recv_status_reply(sock);
854 error:
855 return ret;
856 }
857
858 /*
859 * Consumer send communication message structure to consumer.
860 *
861 * The consumer socket lock must be held by the caller.
862 */
863 int consumer_send_msg(struct consumer_socket *sock,
864 struct lttcomm_consumer_msg *msg)
865 {
866 int ret;
867
868 assert(msg);
869 assert(sock);
870 assert(pthread_mutex_trylock(sock->lock) == EBUSY);
871
872 ret = consumer_socket_send(sock, msg, sizeof(struct lttcomm_consumer_msg));
873 if (ret < 0) {
874 goto error;
875 }
876
877 ret = consumer_recv_status_reply(sock);
878
879 error:
880 return ret;
881 }
882
883 /*
884 * Consumer send channel communication message structure to consumer.
885 *
886 * The consumer socket lock must be held by the caller.
887 */
888 int consumer_send_channel(struct consumer_socket *sock,
889 struct lttcomm_consumer_msg *msg)
890 {
891 int ret;
892
893 assert(msg);
894 assert(sock);
895
896 ret = consumer_send_msg(sock, msg);
897 if (ret < 0) {
898 goto error;
899 }
900
901 error:
902 return ret;
903 }
904
905 /*
906 * Populate the given consumer msg structure with the ask_channel command
907 * information.
908 */
909 void consumer_init_ask_channel_comm_msg(struct lttcomm_consumer_msg *msg,
910 uint64_t subbuf_size,
911 uint64_t num_subbuf,
912 int overwrite,
913 unsigned int switch_timer_interval,
914 unsigned int read_timer_interval,
915 unsigned int live_timer_interval,
916 bool is_in_live_session,
917 unsigned int monitor_timer_interval,
918 int output,
919 int type,
920 uint64_t session_id,
921 const char *pathname,
922 const char *name,
923 uint64_t relayd_id,
924 uint64_t key,
925 unsigned char *uuid,
926 uint32_t chan_id,
927 uint64_t tracefile_size,
928 uint64_t tracefile_count,
929 uint64_t session_id_per_pid,
930 unsigned int monitor,
931 uint32_t ust_app_uid,
932 int64_t blocking_timeout,
933 const char *root_shm_path,
934 const char *shm_path,
935 struct lttng_trace_chunk *trace_chunk,
936 const struct lttng_credentials *buffer_credentials)
937 {
938 assert(msg);
939
940 /* Zeroed structure */
941 memset(msg, 0, sizeof(struct lttcomm_consumer_msg));
942 msg->u.ask_channel.buffer_credentials.uid = UINT32_MAX;
943 msg->u.ask_channel.buffer_credentials.gid = UINT32_MAX;
944
945 if (trace_chunk) {
946 uint64_t chunk_id;
947 enum lttng_trace_chunk_status chunk_status;
948
949 chunk_status = lttng_trace_chunk_get_id(trace_chunk, &chunk_id);
950 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
951 LTTNG_OPTIONAL_SET(&msg->u.ask_channel.chunk_id, chunk_id);
952 }
953 msg->u.ask_channel.buffer_credentials.uid = buffer_credentials->uid;
954 msg->u.ask_channel.buffer_credentials.gid = buffer_credentials->gid;
955
956 msg->cmd_type = LTTNG_CONSUMER_ASK_CHANNEL_CREATION;
957 msg->u.ask_channel.subbuf_size = subbuf_size;
958 msg->u.ask_channel.num_subbuf = num_subbuf ;
959 msg->u.ask_channel.overwrite = overwrite;
960 msg->u.ask_channel.switch_timer_interval = switch_timer_interval;
961 msg->u.ask_channel.read_timer_interval = read_timer_interval;
962 msg->u.ask_channel.live_timer_interval = live_timer_interval;
963 msg->u.ask_channel.is_live = is_in_live_session;
964 msg->u.ask_channel.monitor_timer_interval = monitor_timer_interval;
965 msg->u.ask_channel.output = output;
966 msg->u.ask_channel.type = type;
967 msg->u.ask_channel.session_id = session_id;
968 msg->u.ask_channel.session_id_per_pid = session_id_per_pid;
969 msg->u.ask_channel.relayd_id = relayd_id;
970 msg->u.ask_channel.key = key;
971 msg->u.ask_channel.chan_id = chan_id;
972 msg->u.ask_channel.tracefile_size = tracefile_size;
973 msg->u.ask_channel.tracefile_count = tracefile_count;
974 msg->u.ask_channel.monitor = monitor;
975 msg->u.ask_channel.ust_app_uid = ust_app_uid;
976 msg->u.ask_channel.blocking_timeout = blocking_timeout;
977
978 memcpy(msg->u.ask_channel.uuid, uuid, sizeof(msg->u.ask_channel.uuid));
979
980 if (pathname) {
981 strncpy(msg->u.ask_channel.pathname, pathname,
982 sizeof(msg->u.ask_channel.pathname));
983 msg->u.ask_channel.pathname[sizeof(msg->u.ask_channel.pathname)-1] = '\0';
984 }
985
986 strncpy(msg->u.ask_channel.name, name, sizeof(msg->u.ask_channel.name));
987 msg->u.ask_channel.name[sizeof(msg->u.ask_channel.name) - 1] = '\0';
988
989 if (root_shm_path) {
990 strncpy(msg->u.ask_channel.root_shm_path, root_shm_path,
991 sizeof(msg->u.ask_channel.root_shm_path));
992 msg->u.ask_channel.root_shm_path[sizeof(msg->u.ask_channel.root_shm_path) - 1] = '\0';
993 }
994 if (shm_path) {
995 strncpy(msg->u.ask_channel.shm_path, shm_path,
996 sizeof(msg->u.ask_channel.shm_path));
997 msg->u.ask_channel.shm_path[sizeof(msg->u.ask_channel.shm_path) - 1] = '\0';
998 }
999 }
1000
1001 /*
1002 * Init channel communication message structure.
1003 */
1004 void consumer_init_add_channel_comm_msg(struct lttcomm_consumer_msg *msg,
1005 uint64_t channel_key,
1006 uint64_t session_id,
1007 const char *pathname,
1008 uid_t uid,
1009 gid_t gid,
1010 uint64_t relayd_id,
1011 const char *name,
1012 unsigned int nb_init_streams,
1013 enum lttng_event_output output,
1014 int type,
1015 uint64_t tracefile_size,
1016 uint64_t tracefile_count,
1017 unsigned int monitor,
1018 unsigned int live_timer_interval,
1019 bool is_in_live_session,
1020 unsigned int monitor_timer_interval,
1021 struct lttng_trace_chunk *trace_chunk)
1022 {
1023 assert(msg);
1024
1025 /* Zeroed structure */
1026 memset(msg, 0, sizeof(struct lttcomm_consumer_msg));
1027
1028 if (trace_chunk) {
1029 uint64_t chunk_id;
1030 enum lttng_trace_chunk_status chunk_status;
1031
1032 chunk_status = lttng_trace_chunk_get_id(trace_chunk, &chunk_id);
1033 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
1034 LTTNG_OPTIONAL_SET(&msg->u.channel.chunk_id, chunk_id);
1035 }
1036
1037 /* Send channel */
1038 msg->cmd_type = LTTNG_CONSUMER_ADD_CHANNEL;
1039 msg->u.channel.channel_key = channel_key;
1040 msg->u.channel.session_id = session_id;
1041 msg->u.channel.relayd_id = relayd_id;
1042 msg->u.channel.nb_init_streams = nb_init_streams;
1043 msg->u.channel.output = output;
1044 msg->u.channel.type = type;
1045 msg->u.channel.tracefile_size = tracefile_size;
1046 msg->u.channel.tracefile_count = tracefile_count;
1047 msg->u.channel.monitor = monitor;
1048 msg->u.channel.live_timer_interval = live_timer_interval;
1049 msg->u.channel.is_live = is_in_live_session;
1050 msg->u.channel.monitor_timer_interval = monitor_timer_interval;
1051
1052 strncpy(msg->u.channel.pathname, pathname,
1053 sizeof(msg->u.channel.pathname));
1054 msg->u.channel.pathname[sizeof(msg->u.channel.pathname) - 1] = '\0';
1055
1056 strncpy(msg->u.channel.name, name, sizeof(msg->u.channel.name));
1057 msg->u.channel.name[sizeof(msg->u.channel.name) - 1] = '\0';
1058 }
1059
1060 /*
1061 * Init stream communication message structure.
1062 */
1063 void consumer_init_add_stream_comm_msg(struct lttcomm_consumer_msg *msg,
1064 uint64_t channel_key,
1065 uint64_t stream_key,
1066 int32_t cpu)
1067 {
1068 assert(msg);
1069
1070 memset(msg, 0, sizeof(struct lttcomm_consumer_msg));
1071
1072 msg->cmd_type = LTTNG_CONSUMER_ADD_STREAM;
1073 msg->u.stream.channel_key = channel_key;
1074 msg->u.stream.stream_key = stream_key;
1075 msg->u.stream.cpu = cpu;
1076 }
1077
1078 void consumer_init_streams_sent_comm_msg(struct lttcomm_consumer_msg *msg,
1079 enum lttng_consumer_command cmd,
1080 uint64_t channel_key, uint64_t net_seq_idx)
1081 {
1082 assert(msg);
1083
1084 memset(msg, 0, sizeof(struct lttcomm_consumer_msg));
1085
1086 msg->cmd_type = cmd;
1087 msg->u.sent_streams.channel_key = channel_key;
1088 msg->u.sent_streams.net_seq_idx = net_seq_idx;
1089 }
1090
1091 /*
1092 * Send stream communication structure to the consumer.
1093 */
1094 int consumer_send_stream(struct consumer_socket *sock,
1095 struct consumer_output *dst, struct lttcomm_consumer_msg *msg,
1096 const int *fds, size_t nb_fd)
1097 {
1098 int ret;
1099
1100 assert(msg);
1101 assert(dst);
1102 assert(sock);
1103 assert(fds);
1104
1105 ret = consumer_send_msg(sock, msg);
1106 if (ret < 0) {
1107 goto error;
1108 }
1109
1110 ret = consumer_send_fds(sock, fds, nb_fd);
1111 if (ret < 0) {
1112 goto error;
1113 }
1114
1115 error:
1116 return ret;
1117 }
1118
1119 /*
1120 * Send relayd socket to consumer associated with a session name.
1121 *
1122 * The consumer socket lock must be held by the caller.
1123 *
1124 * On success return positive value. On error, negative value.
1125 */
1126 int consumer_send_relayd_socket(struct consumer_socket *consumer_sock,
1127 struct lttcomm_relayd_sock *rsock, struct consumer_output *consumer,
1128 enum lttng_stream_type type, uint64_t session_id,
1129 const char *session_name, const char *hostname,
1130 const char *base_path, int session_live_timer,
1131 const uint64_t *current_chunk_id, time_t session_creation_time,
1132 bool session_name_contains_creation_time)
1133 {
1134 int ret;
1135 struct lttcomm_consumer_msg msg;
1136
1137 /* Code flow error. Safety net. */
1138 assert(rsock);
1139 assert(consumer);
1140 assert(consumer_sock);
1141
1142 memset(&msg, 0, sizeof(msg));
1143 /* Bail out if consumer is disabled */
1144 if (!consumer->enabled) {
1145 ret = LTTNG_OK;
1146 goto error;
1147 }
1148
1149 if (type == LTTNG_STREAM_CONTROL) {
1150 char output_path[LTTNG_PATH_MAX] = {};
1151 uint64_t relayd_session_id;
1152
1153 ret = relayd_create_session(rsock,
1154 &relayd_session_id,
1155 session_name, hostname, base_path,
1156 session_live_timer,
1157 consumer->snapshot, session_id,
1158 sessiond_uuid, current_chunk_id,
1159 session_creation_time,
1160 session_name_contains_creation_time,
1161 output_path);
1162 if (ret < 0) {
1163 /* Close the control socket. */
1164 (void) relayd_close(rsock);
1165 goto error;
1166 }
1167 msg.u.relayd_sock.relayd_session_id = relayd_session_id;
1168 DBG("Created session on relay, output path reply: %s",
1169 output_path);
1170 }
1171
1172 msg.cmd_type = LTTNG_CONSUMER_ADD_RELAYD_SOCKET;
1173 /*
1174 * Assign network consumer output index using the temporary consumer since
1175 * this call should only be made from within a set_consumer_uri() function
1176 * call in the session daemon.
1177 */
1178 msg.u.relayd_sock.net_index = consumer->net_seq_index;
1179 msg.u.relayd_sock.type = type;
1180 msg.u.relayd_sock.session_id = session_id;
1181 memcpy(&msg.u.relayd_sock.sock, rsock, sizeof(msg.u.relayd_sock.sock));
1182
1183 DBG3("Sending relayd sock info to consumer on %d", *consumer_sock->fd_ptr);
1184 ret = consumer_send_msg(consumer_sock, &msg);
1185 if (ret < 0) {
1186 goto error;
1187 }
1188
1189 DBG3("Sending relayd socket file descriptor to consumer");
1190 ret = consumer_send_fds(consumer_sock, ALIGNED_CONST_PTR(rsock->sock.fd), 1);
1191 if (ret < 0) {
1192 goto error;
1193 }
1194
1195 DBG2("Consumer relayd socket sent");
1196
1197 error:
1198 return ret;
1199 }
1200
1201 static
1202 int consumer_send_pipe(struct consumer_socket *consumer_sock,
1203 enum lttng_consumer_command cmd, int pipe)
1204 {
1205 int ret;
1206 struct lttcomm_consumer_msg msg;
1207 const char *pipe_name;
1208 const char *command_name;
1209
1210 switch (cmd) {
1211 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1212 pipe_name = "channel monitor";
1213 command_name = "SET_CHANNEL_MONITOR_PIPE";
1214 break;
1215 default:
1216 ERR("Unexpected command received in %s (cmd = %d)", __func__,
1217 (int) cmd);
1218 abort();
1219 }
1220
1221 /* Code flow error. Safety net. */
1222
1223 memset(&msg, 0, sizeof(msg));
1224 msg.cmd_type = cmd;
1225
1226 pthread_mutex_lock(consumer_sock->lock);
1227 DBG3("Sending %s command to consumer", command_name);
1228 ret = consumer_send_msg(consumer_sock, &msg);
1229 if (ret < 0) {
1230 goto error;
1231 }
1232
1233 DBG3("Sending %s pipe %d to consumer on socket %d",
1234 pipe_name,
1235 pipe, *consumer_sock->fd_ptr);
1236 ret = consumer_send_fds(consumer_sock, &pipe, 1);
1237 if (ret < 0) {
1238 goto error;
1239 }
1240
1241 DBG2("%s pipe successfully sent", pipe_name);
1242 error:
1243 pthread_mutex_unlock(consumer_sock->lock);
1244 return ret;
1245 }
1246
1247 int consumer_send_channel_monitor_pipe(struct consumer_socket *consumer_sock,
1248 int pipe)
1249 {
1250 return consumer_send_pipe(consumer_sock,
1251 LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE, pipe);
1252 }
1253
1254 /*
1255 * Ask the consumer if the data is pending for the specific session id.
1256 * Returns 1 if data is pending, 0 otherwise, or < 0 on error.
1257 */
1258 int consumer_is_data_pending(uint64_t session_id,
1259 struct consumer_output *consumer)
1260 {
1261 int ret;
1262 int32_t ret_code = 0; /* Default is that the data is NOT pending */
1263 struct consumer_socket *socket;
1264 struct lttng_ht_iter iter;
1265 struct lttcomm_consumer_msg msg;
1266
1267 assert(consumer);
1268
1269 DBG3("Consumer data pending for id %" PRIu64, session_id);
1270
1271 memset(&msg, 0, sizeof(msg));
1272 msg.cmd_type = LTTNG_CONSUMER_DATA_PENDING;
1273 msg.u.data_pending.session_id = session_id;
1274
1275 /* Send command for each consumer */
1276 rcu_read_lock();
1277 cds_lfht_for_each_entry(consumer->socks->ht, &iter.iter, socket,
1278 node.node) {
1279 pthread_mutex_lock(socket->lock);
1280 ret = consumer_socket_send(socket, &msg, sizeof(msg));
1281 if (ret < 0) {
1282 pthread_mutex_unlock(socket->lock);
1283 goto error_unlock;
1284 }
1285
1286 /*
1287 * No need for a recv reply status because the answer to the command is
1288 * the reply status message.
1289 */
1290
1291 ret = consumer_socket_recv(socket, &ret_code, sizeof(ret_code));
1292 if (ret < 0) {
1293 pthread_mutex_unlock(socket->lock);
1294 goto error_unlock;
1295 }
1296 pthread_mutex_unlock(socket->lock);
1297
1298 if (ret_code == 1) {
1299 break;
1300 }
1301 }
1302 rcu_read_unlock();
1303
1304 DBG("Consumer data is %s pending for session id %" PRIu64,
1305 ret_code == 1 ? "" : "NOT", session_id);
1306 return ret_code;
1307
1308 error_unlock:
1309 rcu_read_unlock();
1310 return -1;
1311 }
1312
1313 /*
1314 * Send a flush command to consumer using the given channel key.
1315 *
1316 * Return 0 on success else a negative value.
1317 */
1318 int consumer_flush_channel(struct consumer_socket *socket, uint64_t key)
1319 {
1320 int ret;
1321 struct lttcomm_consumer_msg msg;
1322
1323 assert(socket);
1324
1325 DBG2("Consumer flush channel key %" PRIu64, key);
1326
1327 memset(&msg, 0, sizeof(msg));
1328 msg.cmd_type = LTTNG_CONSUMER_FLUSH_CHANNEL;
1329 msg.u.flush_channel.key = key;
1330
1331 pthread_mutex_lock(socket->lock);
1332 health_code_update();
1333
1334 ret = consumer_send_msg(socket, &msg);
1335 if (ret < 0) {
1336 goto end;
1337 }
1338
1339 end:
1340 health_code_update();
1341 pthread_mutex_unlock(socket->lock);
1342 return ret;
1343 }
1344
1345 /*
1346 * Send a clear quiescent command to consumer using the given channel key.
1347 *
1348 * Return 0 on success else a negative value.
1349 */
1350 int consumer_clear_quiescent_channel(struct consumer_socket *socket, uint64_t key)
1351 {
1352 int ret;
1353 struct lttcomm_consumer_msg msg;
1354
1355 assert(socket);
1356
1357 DBG2("Consumer clear quiescent channel key %" PRIu64, key);
1358
1359 memset(&msg, 0, sizeof(msg));
1360 msg.cmd_type = LTTNG_CONSUMER_CLEAR_QUIESCENT_CHANNEL;
1361 msg.u.clear_quiescent_channel.key = key;
1362
1363 pthread_mutex_lock(socket->lock);
1364 health_code_update();
1365
1366 ret = consumer_send_msg(socket, &msg);
1367 if (ret < 0) {
1368 goto end;
1369 }
1370
1371 end:
1372 health_code_update();
1373 pthread_mutex_unlock(socket->lock);
1374 return ret;
1375 }
1376
1377 /*
1378 * Send a close metadata command to consumer using the given channel key.
1379 * Called with registry lock held.
1380 *
1381 * Return 0 on success else a negative value.
1382 */
1383 int consumer_close_metadata(struct consumer_socket *socket,
1384 uint64_t metadata_key)
1385 {
1386 int ret;
1387 struct lttcomm_consumer_msg msg;
1388
1389 assert(socket);
1390
1391 DBG2("Consumer close metadata channel key %" PRIu64, metadata_key);
1392
1393 memset(&msg, 0, sizeof(msg));
1394 msg.cmd_type = LTTNG_CONSUMER_CLOSE_METADATA;
1395 msg.u.close_metadata.key = metadata_key;
1396
1397 pthread_mutex_lock(socket->lock);
1398 health_code_update();
1399
1400 ret = consumer_send_msg(socket, &msg);
1401 if (ret < 0) {
1402 goto end;
1403 }
1404
1405 end:
1406 health_code_update();
1407 pthread_mutex_unlock(socket->lock);
1408 return ret;
1409 }
1410
1411 /*
1412 * Send a setup metdata command to consumer using the given channel key.
1413 *
1414 * Return 0 on success else a negative value.
1415 */
1416 int consumer_setup_metadata(struct consumer_socket *socket,
1417 uint64_t metadata_key)
1418 {
1419 int ret;
1420 struct lttcomm_consumer_msg msg;
1421
1422 assert(socket);
1423
1424 DBG2("Consumer setup metadata channel key %" PRIu64, metadata_key);
1425
1426 memset(&msg, 0, sizeof(msg));
1427 msg.cmd_type = LTTNG_CONSUMER_SETUP_METADATA;
1428 msg.u.setup_metadata.key = metadata_key;
1429
1430 pthread_mutex_lock(socket->lock);
1431 health_code_update();
1432
1433 ret = consumer_send_msg(socket, &msg);
1434 if (ret < 0) {
1435 goto end;
1436 }
1437
1438 end:
1439 health_code_update();
1440 pthread_mutex_unlock(socket->lock);
1441 return ret;
1442 }
1443
1444 /*
1445 * Send metadata string to consumer.
1446 * RCU read-side lock must be held to guarantee existence of socket.
1447 *
1448 * Return 0 on success else a negative value.
1449 */
1450 int consumer_push_metadata(struct consumer_socket *socket,
1451 uint64_t metadata_key, char *metadata_str, size_t len,
1452 size_t target_offset, uint64_t version)
1453 {
1454 int ret;
1455 struct lttcomm_consumer_msg msg;
1456
1457 assert(socket);
1458
1459 DBG2("Consumer push metadata to consumer socket %d", *socket->fd_ptr);
1460
1461 pthread_mutex_lock(socket->lock);
1462
1463 memset(&msg, 0, sizeof(msg));
1464 msg.cmd_type = LTTNG_CONSUMER_PUSH_METADATA;
1465 msg.u.push_metadata.key = metadata_key;
1466 msg.u.push_metadata.target_offset = target_offset;
1467 msg.u.push_metadata.len = len;
1468 msg.u.push_metadata.version = version;
1469
1470 health_code_update();
1471 ret = consumer_send_msg(socket, &msg);
1472 if (ret < 0 || len == 0) {
1473 goto end;
1474 }
1475
1476 DBG3("Consumer pushing metadata on sock %d of len %zu", *socket->fd_ptr,
1477 len);
1478
1479 ret = consumer_socket_send(socket, metadata_str, len);
1480 if (ret < 0) {
1481 goto end;
1482 }
1483
1484 health_code_update();
1485 ret = consumer_recv_status_reply(socket);
1486 if (ret < 0) {
1487 goto end;
1488 }
1489
1490 end:
1491 pthread_mutex_unlock(socket->lock);
1492 health_code_update();
1493 return ret;
1494 }
1495
1496 /*
1497 * Ask the consumer to snapshot a specific channel using the key.
1498 *
1499 * Returns LTTNG_OK on success or else an LTTng error code.
1500 */
1501 enum lttng_error_code consumer_snapshot_channel(struct consumer_socket *socket,
1502 uint64_t key, const struct consumer_output *output, int metadata,
1503 uid_t uid, gid_t gid, const char *channel_path, int wait,
1504 uint64_t nb_packets_per_stream)
1505 {
1506 int ret;
1507 enum lttng_error_code status = LTTNG_OK;
1508 struct lttcomm_consumer_msg msg;
1509
1510 assert(socket);
1511 assert(output);
1512
1513 DBG("Consumer snapshot channel key %" PRIu64, key);
1514
1515 memset(&msg, 0, sizeof(msg));
1516 msg.cmd_type = LTTNG_CONSUMER_SNAPSHOT_CHANNEL;
1517 msg.u.snapshot_channel.key = key;
1518 msg.u.snapshot_channel.nb_packets_per_stream = nb_packets_per_stream;
1519 msg.u.snapshot_channel.metadata = metadata;
1520
1521 if (output->type == CONSUMER_DST_NET) {
1522 msg.u.snapshot_channel.relayd_id =
1523 output->net_seq_index;
1524 msg.u.snapshot_channel.use_relayd = 1;
1525 } else {
1526 msg.u.snapshot_channel.relayd_id = (uint64_t) -1ULL;
1527 }
1528 ret = lttng_strncpy(msg.u.snapshot_channel.pathname,
1529 channel_path,
1530 sizeof(msg.u.snapshot_channel.pathname));
1531 if (ret < 0) {
1532 ERR("Snapshot path exceeds the maximal allowed length of %zu bytes (%zu bytes required) with path \"%s\"",
1533 sizeof(msg.u.snapshot_channel.pathname),
1534 strlen(channel_path),
1535 channel_path);
1536 status = LTTNG_ERR_SNAPSHOT_FAIL;
1537 goto error;
1538 }
1539
1540 health_code_update();
1541 pthread_mutex_lock(socket->lock);
1542 ret = consumer_send_msg(socket, &msg);
1543 pthread_mutex_unlock(socket->lock);
1544 if (ret < 0) {
1545 switch (-ret) {
1546 case LTTCOMM_CONSUMERD_CHAN_NOT_FOUND:
1547 status = LTTNG_ERR_CHAN_NOT_FOUND;
1548 break;
1549 default:
1550 status = LTTNG_ERR_SNAPSHOT_FAIL;
1551 break;
1552 }
1553 goto error;
1554 }
1555
1556 error:
1557 health_code_update();
1558 return status;
1559 }
1560
1561 /*
1562 * Ask the consumer the number of discarded events for a channel.
1563 */
1564 int consumer_get_discarded_events(uint64_t session_id, uint64_t channel_key,
1565 struct consumer_output *consumer, uint64_t *discarded)
1566 {
1567 int ret;
1568 struct consumer_socket *socket;
1569 struct lttng_ht_iter iter;
1570 struct lttcomm_consumer_msg msg;
1571
1572 assert(consumer);
1573
1574 DBG3("Consumer discarded events id %" PRIu64, session_id);
1575
1576 memset(&msg, 0, sizeof(msg));
1577 msg.cmd_type = LTTNG_CONSUMER_DISCARDED_EVENTS;
1578 msg.u.discarded_events.session_id = session_id;
1579 msg.u.discarded_events.channel_key = channel_key;
1580
1581 *discarded = 0;
1582
1583 /* Send command for each consumer */
1584 rcu_read_lock();
1585 cds_lfht_for_each_entry(consumer->socks->ht, &iter.iter, socket,
1586 node.node) {
1587 uint64_t consumer_discarded = 0;
1588 pthread_mutex_lock(socket->lock);
1589 ret = consumer_socket_send(socket, &msg, sizeof(msg));
1590 if (ret < 0) {
1591 pthread_mutex_unlock(socket->lock);
1592 goto end;
1593 }
1594
1595 /*
1596 * No need for a recv reply status because the answer to the
1597 * command is the reply status message.
1598 */
1599 ret = consumer_socket_recv(socket, &consumer_discarded,
1600 sizeof(consumer_discarded));
1601 if (ret < 0) {
1602 ERR("get discarded events");
1603 pthread_mutex_unlock(socket->lock);
1604 goto end;
1605 }
1606 pthread_mutex_unlock(socket->lock);
1607 *discarded += consumer_discarded;
1608 }
1609 ret = 0;
1610 DBG("Consumer discarded %" PRIu64 " events in session id %" PRIu64,
1611 *discarded, session_id);
1612
1613 end:
1614 rcu_read_unlock();
1615 return ret;
1616 }
1617
1618 /*
1619 * Ask the consumer the number of lost packets for a channel.
1620 */
1621 int consumer_get_lost_packets(uint64_t session_id, uint64_t channel_key,
1622 struct consumer_output *consumer, uint64_t *lost)
1623 {
1624 int ret;
1625 struct consumer_socket *socket;
1626 struct lttng_ht_iter iter;
1627 struct lttcomm_consumer_msg msg;
1628
1629 assert(consumer);
1630
1631 DBG3("Consumer lost packets id %" PRIu64, session_id);
1632
1633 memset(&msg, 0, sizeof(msg));
1634 msg.cmd_type = LTTNG_CONSUMER_LOST_PACKETS;
1635 msg.u.lost_packets.session_id = session_id;
1636 msg.u.lost_packets.channel_key = channel_key;
1637
1638 *lost = 0;
1639
1640 /* Send command for each consumer */
1641 rcu_read_lock();
1642 cds_lfht_for_each_entry(consumer->socks->ht, &iter.iter, socket,
1643 node.node) {
1644 uint64_t consumer_lost = 0;
1645 pthread_mutex_lock(socket->lock);
1646 ret = consumer_socket_send(socket, &msg, sizeof(msg));
1647 if (ret < 0) {
1648 pthread_mutex_unlock(socket->lock);
1649 goto end;
1650 }
1651
1652 /*
1653 * No need for a recv reply status because the answer to the
1654 * command is the reply status message.
1655 */
1656 ret = consumer_socket_recv(socket, &consumer_lost,
1657 sizeof(consumer_lost));
1658 if (ret < 0) {
1659 ERR("get lost packets");
1660 pthread_mutex_unlock(socket->lock);
1661 goto end;
1662 }
1663 pthread_mutex_unlock(socket->lock);
1664 *lost += consumer_lost;
1665 }
1666 ret = 0;
1667 DBG("Consumer lost %" PRIu64 " packets in session id %" PRIu64,
1668 *lost, session_id);
1669
1670 end:
1671 rcu_read_unlock();
1672 return ret;
1673 }
1674
1675 /*
1676 * Ask the consumer to rotate a channel.
1677 *
1678 * The new_chunk_id is the session->rotate_count that has been incremented
1679 * when the rotation started. On the relay, this allows to keep track in which
1680 * chunk each stream is currently writing to (for the rotate_pending operation).
1681 */
1682 int consumer_rotate_channel(struct consumer_socket *socket, uint64_t key,
1683 uid_t uid, gid_t gid, struct consumer_output *output,
1684 bool is_metadata_channel)
1685 {
1686 int ret;
1687 struct lttcomm_consumer_msg msg;
1688
1689 assert(socket);
1690
1691 DBG("Consumer rotate channel key %" PRIu64, key);
1692
1693 pthread_mutex_lock(socket->lock);
1694 memset(&msg, 0, sizeof(msg));
1695 msg.cmd_type = LTTNG_CONSUMER_ROTATE_CHANNEL;
1696 msg.u.rotate_channel.key = key;
1697 msg.u.rotate_channel.metadata = !!is_metadata_channel;
1698
1699 if (output->type == CONSUMER_DST_NET) {
1700 msg.u.rotate_channel.relayd_id = output->net_seq_index;
1701 } else {
1702 msg.u.rotate_channel.relayd_id = (uint64_t) -1ULL;
1703 }
1704
1705 health_code_update();
1706 ret = consumer_send_msg(socket, &msg);
1707 if (ret < 0) {
1708 switch (-ret) {
1709 case LTTCOMM_CONSUMERD_CHAN_NOT_FOUND:
1710 ret = -LTTNG_ERR_CHAN_NOT_FOUND;
1711 break;
1712 default:
1713 ret = -LTTNG_ERR_ROTATION_FAIL_CONSUMER;
1714 break;
1715 }
1716 goto error;
1717 }
1718 error:
1719 pthread_mutex_unlock(socket->lock);
1720 health_code_update();
1721 return ret;
1722 }
1723
1724 int consumer_clear_channel(struct consumer_socket *socket, uint64_t key)
1725 {
1726 int ret;
1727 struct lttcomm_consumer_msg msg;
1728
1729 assert(socket);
1730
1731 DBG("Consumer clear channel %" PRIu64, key);
1732
1733 memset(&msg, 0, sizeof(msg));
1734 msg.cmd_type = LTTNG_CONSUMER_CLEAR_CHANNEL;
1735 msg.u.clear_channel.key = key;
1736
1737 health_code_update();
1738
1739 pthread_mutex_lock(socket->lock);
1740 ret = consumer_send_msg(socket, &msg);
1741 if (ret < 0) {
1742 goto error_socket;
1743 }
1744
1745 error_socket:
1746 pthread_mutex_unlock(socket->lock);
1747
1748 health_code_update();
1749 return ret;
1750 }
1751
1752 int consumer_init(struct consumer_socket *socket,
1753 const lttng_uuid sessiond_uuid)
1754 {
1755 int ret;
1756 struct lttcomm_consumer_msg msg = {
1757 .cmd_type = LTTNG_CONSUMER_INIT,
1758 };
1759
1760 assert(socket);
1761
1762 DBG("Sending consumer initialization command");
1763 lttng_uuid_copy(msg.u.init.sessiond_uuid, sessiond_uuid);
1764
1765 health_code_update();
1766 ret = consumer_send_msg(socket, &msg);
1767 if (ret < 0) {
1768 goto error;
1769 }
1770
1771 error:
1772 health_code_update();
1773 return ret;
1774 }
1775
1776 /*
1777 * Ask the consumer to create a new chunk for a given session.
1778 *
1779 * Called with the consumer socket lock held.
1780 */
1781 int consumer_create_trace_chunk(struct consumer_socket *socket,
1782 uint64_t relayd_id, uint64_t session_id,
1783 struct lttng_trace_chunk *chunk,
1784 const char *domain_subdir)
1785 {
1786 int ret;
1787 enum lttng_trace_chunk_status chunk_status;
1788 struct lttng_credentials chunk_credentials;
1789 const struct lttng_directory_handle *chunk_directory_handle = NULL;
1790 struct lttng_directory_handle *domain_handle = NULL;
1791 int domain_dirfd;
1792 const char *chunk_name;
1793 bool chunk_name_overridden;
1794 uint64_t chunk_id;
1795 time_t creation_timestamp;
1796 char creation_timestamp_buffer[ISO8601_STR_LEN];
1797 const char *creation_timestamp_str = "(none)";
1798 const bool chunk_has_local_output = relayd_id == -1ULL;
1799 enum lttng_trace_chunk_status tc_status;
1800 struct lttcomm_consumer_msg msg = {
1801 .cmd_type = LTTNG_CONSUMER_CREATE_TRACE_CHUNK,
1802 .u.create_trace_chunk.session_id = session_id,
1803 };
1804
1805 assert(socket);
1806 assert(chunk);
1807
1808 if (relayd_id != -1ULL) {
1809 LTTNG_OPTIONAL_SET(&msg.u.create_trace_chunk.relayd_id,
1810 relayd_id);
1811 }
1812
1813 chunk_status = lttng_trace_chunk_get_name(chunk, &chunk_name,
1814 &chunk_name_overridden);
1815 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK &&
1816 chunk_status != LTTNG_TRACE_CHUNK_STATUS_NONE) {
1817 ERR("Failed to get name of trace chunk");
1818 ret = -LTTNG_ERR_FATAL;
1819 goto error;
1820 }
1821 if (chunk_name_overridden) {
1822 ret = lttng_strncpy(msg.u.create_trace_chunk.override_name,
1823 chunk_name,
1824 sizeof(msg.u.create_trace_chunk.override_name));
1825 if (ret) {
1826 ERR("Trace chunk name \"%s\" exceeds the maximal length allowed by the consumer protocol",
1827 chunk_name);
1828 ret = -LTTNG_ERR_FATAL;
1829 goto error;
1830 }
1831 }
1832
1833 chunk_status = lttng_trace_chunk_get_creation_timestamp(chunk,
1834 &creation_timestamp);
1835 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1836 ret = -LTTNG_ERR_FATAL;
1837 goto error;
1838 }
1839 msg.u.create_trace_chunk.creation_timestamp =
1840 (uint64_t) creation_timestamp;
1841 /* Only used for logging purposes. */
1842 ret = time_to_iso8601_str(creation_timestamp,
1843 creation_timestamp_buffer,
1844 sizeof(creation_timestamp_buffer));
1845 creation_timestamp_str = !ret ? creation_timestamp_buffer :
1846 "(formatting error)";
1847
1848 chunk_status = lttng_trace_chunk_get_id(chunk, &chunk_id);
1849 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1850 /*
1851 * Anonymous trace chunks should never be transmitted
1852 * to remote peers (consumerd and relayd). They are used
1853 * internally for backward-compatibility purposes.
1854 */
1855 ret = -LTTNG_ERR_FATAL;
1856 goto error;
1857 }
1858 msg.u.create_trace_chunk.chunk_id = chunk_id;
1859
1860 if (chunk_has_local_output) {
1861 chunk_status = lttng_trace_chunk_borrow_chunk_directory_handle(
1862 chunk, &chunk_directory_handle);
1863 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1864 ret = -LTTNG_ERR_FATAL;
1865 goto error;
1866 }
1867 chunk_status = lttng_trace_chunk_get_credentials(
1868 chunk, &chunk_credentials);
1869 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1870 /*
1871 * Not associating credentials to a sessiond chunk is a
1872 * fatal internal error.
1873 */
1874 ret = -LTTNG_ERR_FATAL;
1875 goto error;
1876 }
1877 tc_status = lttng_trace_chunk_create_subdirectory(
1878 chunk, domain_subdir);
1879 if (tc_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1880 PERROR("Failed to create chunk domain output directory \"%s\"",
1881 domain_subdir);
1882 ret = -LTTNG_ERR_FATAL;
1883 goto error;
1884 }
1885 domain_handle = lttng_directory_handle_create_from_handle(
1886 domain_subdir,
1887 chunk_directory_handle);
1888 if (!domain_handle) {
1889 ret = -LTTNG_ERR_FATAL;
1890 goto error;
1891 }
1892
1893 /*
1894 * This will only compile on platforms that support
1895 * dirfd (POSIX.2008). This is fine as the session daemon
1896 * is only built for such platforms.
1897 *
1898 * The ownership of the chunk directory handle's is maintained
1899 * by the trace chunk.
1900 */
1901 domain_dirfd = lttng_directory_handle_get_dirfd(
1902 domain_handle);
1903 assert(domain_dirfd >= 0);
1904
1905 msg.u.create_trace_chunk.credentials.value.uid =
1906 chunk_credentials.uid;
1907 msg.u.create_trace_chunk.credentials.value.gid =
1908 chunk_credentials.gid;
1909 msg.u.create_trace_chunk.credentials.is_set = 1;
1910 }
1911
1912 DBG("Sending consumer create trace chunk command: relayd_id = %" PRId64
1913 ", session_id = %" PRIu64 ", chunk_id = %" PRIu64
1914 ", creation_timestamp = %s",
1915 relayd_id, session_id, chunk_id,
1916 creation_timestamp_str);
1917 health_code_update();
1918 ret = consumer_send_msg(socket, &msg);
1919 health_code_update();
1920 if (ret < 0) {
1921 ERR("Trace chunk creation error on consumer");
1922 ret = -LTTNG_ERR_CREATE_TRACE_CHUNK_FAIL_CONSUMER;
1923 goto error;
1924 }
1925
1926 if (chunk_has_local_output) {
1927 DBG("Sending trace chunk domain directory fd to consumer");
1928 health_code_update();
1929 ret = consumer_send_fds(socket, &domain_dirfd, 1);
1930 health_code_update();
1931 if (ret < 0) {
1932 ERR("Trace chunk creation error on consumer");
1933 ret = -LTTNG_ERR_CREATE_TRACE_CHUNK_FAIL_CONSUMER;
1934 goto error;
1935 }
1936 }
1937 error:
1938 lttng_directory_handle_put(domain_handle);
1939 return ret;
1940 }
1941
1942 /*
1943 * Ask the consumer to close a trace chunk for a given session.
1944 *
1945 * Called with the consumer socket lock held.
1946 */
1947 int consumer_close_trace_chunk(struct consumer_socket *socket,
1948 uint64_t relayd_id, uint64_t session_id,
1949 struct lttng_trace_chunk *chunk,
1950 char *closed_trace_chunk_path)
1951 {
1952 int ret;
1953 enum lttng_trace_chunk_status chunk_status;
1954 struct lttcomm_consumer_msg msg = {
1955 .cmd_type = LTTNG_CONSUMER_CLOSE_TRACE_CHUNK,
1956 .u.close_trace_chunk.session_id = session_id,
1957 };
1958 struct lttcomm_consumer_close_trace_chunk_reply reply;
1959 uint64_t chunk_id;
1960 time_t close_timestamp;
1961 enum lttng_trace_chunk_command_type close_command;
1962 const char *close_command_name = "none";
1963 struct lttng_dynamic_buffer path_reception_buffer;
1964
1965 assert(socket);
1966 lttng_dynamic_buffer_init(&path_reception_buffer);
1967
1968 if (relayd_id != -1ULL) {
1969 LTTNG_OPTIONAL_SET(
1970 &msg.u.close_trace_chunk.relayd_id, relayd_id);
1971 }
1972
1973 chunk_status = lttng_trace_chunk_get_close_command(
1974 chunk, &close_command);
1975 switch (chunk_status) {
1976 case LTTNG_TRACE_CHUNK_STATUS_OK:
1977 LTTNG_OPTIONAL_SET(&msg.u.close_trace_chunk.close_command,
1978 (uint32_t) close_command);
1979 break;
1980 case LTTNG_TRACE_CHUNK_STATUS_NONE:
1981 break;
1982 default:
1983 ERR("Failed to get trace chunk close command");
1984 ret = -1;
1985 goto error;
1986 }
1987
1988 chunk_status = lttng_trace_chunk_get_id(chunk, &chunk_id);
1989 /*
1990 * Anonymous trace chunks should never be transmitted to remote peers
1991 * (consumerd and relayd). They are used internally for
1992 * backward-compatibility purposes.
1993 */
1994 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
1995 msg.u.close_trace_chunk.chunk_id = chunk_id;
1996
1997 chunk_status = lttng_trace_chunk_get_close_timestamp(chunk,
1998 &close_timestamp);
1999 /*
2000 * A trace chunk should be closed locally before being closed remotely.
2001 * Otherwise, the close timestamp would never be transmitted to the
2002 * peers.
2003 */
2004 assert(chunk_status == LTTNG_TRACE_CHUNK_STATUS_OK);
2005 msg.u.close_trace_chunk.close_timestamp = (uint64_t) close_timestamp;
2006
2007 if (msg.u.close_trace_chunk.close_command.is_set) {
2008 close_command_name = lttng_trace_chunk_command_type_get_name(
2009 close_command);
2010 }
2011 DBG("Sending consumer close trace chunk command: relayd_id = %" PRId64
2012 ", session_id = %" PRIu64 ", chunk_id = %" PRIu64
2013 ", close command = \"%s\"",
2014 relayd_id, session_id, chunk_id, close_command_name);
2015
2016 health_code_update();
2017 ret = consumer_socket_send(socket, &msg, sizeof(struct lttcomm_consumer_msg));
2018 if (ret < 0) {
2019 ret = -LTTNG_ERR_CLOSE_TRACE_CHUNK_FAIL_CONSUMER;
2020 goto error;
2021 }
2022 ret = consumer_socket_recv(socket, &reply, sizeof(reply));
2023 if (ret < 0) {
2024 ret = -LTTNG_ERR_CLOSE_TRACE_CHUNK_FAIL_CONSUMER;
2025 goto error;
2026 }
2027 if (reply.path_length >= LTTNG_PATH_MAX) {
2028 ERR("Invalid path returned by relay daemon: %" PRIu32 "bytes exceeds maximal allowed length of %d bytes",
2029 reply.path_length, LTTNG_PATH_MAX);
2030 ret = -LTTNG_ERR_INVALID_PROTOCOL;
2031 goto error;
2032 }
2033 ret = lttng_dynamic_buffer_set_size(&path_reception_buffer,
2034 reply.path_length);
2035 if (ret) {
2036 ERR("Failed to allocate reception buffer of path returned by the \"close trace chunk\" command");
2037 ret = -LTTNG_ERR_NOMEM;
2038 goto error;
2039 }
2040 ret = consumer_socket_recv(socket, path_reception_buffer.data,
2041 path_reception_buffer.size);
2042 if (ret < 0) {
2043 ERR("Communication error while receiving path of closed trace chunk");
2044 ret = -LTTNG_ERR_CLOSE_TRACE_CHUNK_FAIL_CONSUMER;
2045 goto error;
2046 }
2047 if (path_reception_buffer.data[path_reception_buffer.size - 1] != '\0') {
2048 ERR("Invalid path returned by relay daemon: not null-terminated");
2049 ret = -LTTNG_ERR_INVALID_PROTOCOL;
2050 goto error;
2051 }
2052 if (closed_trace_chunk_path) {
2053 /*
2054 * closed_trace_chunk_path is assumed to have a length >=
2055 * LTTNG_PATH_MAX
2056 */
2057 memcpy(closed_trace_chunk_path, path_reception_buffer.data,
2058 path_reception_buffer.size);
2059 }
2060 error:
2061 lttng_dynamic_buffer_reset(&path_reception_buffer);
2062 health_code_update();
2063 return ret;
2064 }
2065
2066 /*
2067 * Ask the consumer if a trace chunk exists.
2068 *
2069 * Called with the consumer socket lock held.
2070 * Returns 0 on success, or a negative value on error.
2071 */
2072 int consumer_trace_chunk_exists(struct consumer_socket *socket,
2073 uint64_t relayd_id, uint64_t session_id,
2074 struct lttng_trace_chunk *chunk,
2075 enum consumer_trace_chunk_exists_status *result)
2076 {
2077 int ret;
2078 enum lttng_trace_chunk_status chunk_status;
2079 struct lttcomm_consumer_msg msg = {
2080 .cmd_type = LTTNG_CONSUMER_TRACE_CHUNK_EXISTS,
2081 .u.trace_chunk_exists.session_id = session_id,
2082 };
2083 uint64_t chunk_id;
2084 const char *consumer_reply_str;
2085
2086 assert(socket);
2087
2088 if (relayd_id != -1ULL) {
2089 LTTNG_OPTIONAL_SET(&msg.u.trace_chunk_exists.relayd_id,
2090 relayd_id);
2091 }
2092
2093 chunk_status = lttng_trace_chunk_get_id(chunk, &chunk_id);
2094 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
2095 /*
2096 * Anonymous trace chunks should never be transmitted
2097 * to remote peers (consumerd and relayd). They are used
2098 * internally for backward-compatibility purposes.
2099 */
2100 ret = -LTTNG_ERR_FATAL;
2101 goto error;
2102 }
2103 msg.u.trace_chunk_exists.chunk_id = chunk_id;
2104
2105 DBG("Sending consumer trace chunk exists command: relayd_id = %" PRId64
2106 ", session_id = %" PRIu64
2107 ", chunk_id = %" PRIu64, relayd_id, session_id, chunk_id);
2108
2109 health_code_update();
2110 ret = consumer_send_msg(socket, &msg);
2111 switch (-ret) {
2112 case LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK:
2113 consumer_reply_str = "unknown trace chunk";
2114 *result = CONSUMER_TRACE_CHUNK_EXISTS_STATUS_UNKNOWN_CHUNK;
2115 break;
2116 case LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_LOCAL:
2117 consumer_reply_str = "trace chunk exists locally";
2118 *result = CONSUMER_TRACE_CHUNK_EXISTS_STATUS_EXISTS_LOCAL;
2119 break;
2120 case LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_REMOTE:
2121 consumer_reply_str = "trace chunk exists on remote peer";
2122 *result = CONSUMER_TRACE_CHUNK_EXISTS_STATUS_EXISTS_REMOTE;
2123 break;
2124 default:
2125 ERR("Consumer returned an error from TRACE_CHUNK_EXISTS command");
2126 ret = -1;
2127 goto error;
2128 }
2129 DBG("Consumer reply to TRACE_CHUNK_EXISTS command: %s",
2130 consumer_reply_str);
2131 ret = 0;
2132 error:
2133 health_code_update();
2134 return ret;
2135 }
This page took 0.108765 seconds and 6 git commands to generate.