3fcb528d1c0ca87899d6d6142d57829c84cacee4
[lttng-tools.git] / src / common / consumer.c
1 /*
2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License
7 * as published by the Free Software Foundation; only version 2
8 * of the License.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
18 */
19
20 #define _GNU_SOURCE
21 #include <assert.h>
22 #include <fcntl.h>
23 #include <poll.h>
24 #include <pthread.h>
25 #include <stdlib.h>
26 #include <string.h>
27 #include <sys/mman.h>
28 #include <sys/socket.h>
29 #include <sys/types.h>
30 #include <unistd.h>
31
32 #include <common/common.h>
33 #include <common/kernel-ctl/kernel-ctl.h>
34 #include <common/sessiond-comm/sessiond-comm.h>
35 #include <common/kernel-consumer/kernel-consumer.h>
36 #include <common/ust-consumer/ust-consumer.h>
37
38 #include "consumer.h"
39
40 struct lttng_consumer_global_data consumer_data = {
41 .stream_count = 0,
42 .need_update = 1,
43 .type = LTTNG_CONSUMER_UNKNOWN,
44 };
45
46 /* timeout parameter, to control the polling thread grace period. */
47 int consumer_poll_timeout = -1;
48
49 /*
50 * Flag to inform the polling thread to quit when all fd hung up. Updated by
51 * the consumer_thread_receive_fds when it notices that all fds has hung up.
52 * Also updated by the signal handler (consumer_should_exit()). Read by the
53 * polling threads.
54 */
55 volatile int consumer_quit = 0;
56
57 /*
58 * Find a stream. The consumer_data.lock must be locked during this
59 * call.
60 */
61 static struct lttng_consumer_stream *consumer_find_stream(int key)
62 {
63 struct lttng_ht_iter iter;
64 struct lttng_ht_node_ulong *node;
65 struct lttng_consumer_stream *stream = NULL;
66
67 /* Negative keys are lookup failures */
68 if (key < 0)
69 return NULL;
70
71 rcu_read_lock();
72
73 lttng_ht_lookup(consumer_data.stream_ht, (void *)((unsigned long) key),
74 &iter);
75 node = lttng_ht_iter_get_node_ulong(&iter);
76 if (node != NULL) {
77 stream = caa_container_of(node, struct lttng_consumer_stream, node);
78 }
79
80 rcu_read_unlock();
81
82 return stream;
83 }
84
85 static void consumer_steal_stream_key(int key)
86 {
87 struct lttng_consumer_stream *stream;
88
89 stream = consumer_find_stream(key);
90 if (stream)
91 stream->key = -1;
92 }
93
94 static struct lttng_consumer_channel *consumer_find_channel(int key)
95 {
96 struct lttng_ht_iter iter;
97 struct lttng_ht_node_ulong *node;
98 struct lttng_consumer_channel *channel = NULL;
99
100 /* Negative keys are lookup failures */
101 if (key < 0)
102 return NULL;
103
104 rcu_read_lock();
105
106 lttng_ht_lookup(consumer_data.channel_ht, (void *)((unsigned long) key),
107 &iter);
108 node = lttng_ht_iter_get_node_ulong(&iter);
109 if (node != NULL) {
110 channel = caa_container_of(node, struct lttng_consumer_channel, node);
111 }
112
113 rcu_read_unlock();
114
115 return channel;
116 }
117
118 static void consumer_steal_channel_key(int key)
119 {
120 struct lttng_consumer_channel *channel;
121
122 channel = consumer_find_channel(key);
123 if (channel)
124 channel->key = -1;
125 }
126
127 static
128 void consumer_free_stream(struct rcu_head *head)
129 {
130 struct lttng_ht_node_ulong *node =
131 caa_container_of(head, struct lttng_ht_node_ulong, head);
132 struct lttng_consumer_stream *stream =
133 caa_container_of(node, struct lttng_consumer_stream, node);
134
135 free(stream);
136 }
137
138 /*
139 * Remove a stream from the global list protected by a mutex. This
140 * function is also responsible for freeing its data structures.
141 */
142 void consumer_del_stream(struct lttng_consumer_stream *stream)
143 {
144 int ret;
145 struct lttng_ht_iter iter;
146 struct lttng_consumer_channel *free_chan = NULL;
147
148 pthread_mutex_lock(&consumer_data.lock);
149
150 switch (consumer_data.type) {
151 case LTTNG_CONSUMER_KERNEL:
152 if (stream->mmap_base != NULL) {
153 ret = munmap(stream->mmap_base, stream->mmap_len);
154 if (ret != 0) {
155 perror("munmap");
156 }
157 }
158 break;
159 case LTTNG_CONSUMER32_UST:
160 case LTTNG_CONSUMER64_UST:
161 lttng_ustconsumer_del_stream(stream);
162 break;
163 default:
164 ERR("Unknown consumer_data type");
165 assert(0);
166 goto end;
167 }
168
169 rcu_read_lock();
170
171 /* Get stream node from hash table */
172 lttng_ht_lookup(consumer_data.stream_ht,
173 (void *)((unsigned long) stream->key), &iter);
174 /*
175 * Remove stream node from hash table. It can fail if it's been
176 * replaced due to key reuse.
177 */
178 (void) lttng_ht_del(consumer_data.stream_ht, &iter);
179
180 rcu_read_unlock();
181
182 if (consumer_data.stream_count <= 0) {
183 goto end;
184 }
185 consumer_data.stream_count--;
186 if (!stream) {
187 goto end;
188 }
189 if (stream->out_fd >= 0) {
190 ret = close(stream->out_fd);
191 if (ret) {
192 PERROR("close");
193 }
194 }
195 if (stream->wait_fd >= 0 && !stream->wait_fd_is_copy) {
196 ret = close(stream->wait_fd);
197 if (ret) {
198 PERROR("close");
199 }
200 }
201 if (stream->shm_fd >= 0 && stream->wait_fd != stream->shm_fd) {
202 ret = close(stream->shm_fd);
203 if (ret) {
204 PERROR("close");
205 }
206 }
207 if (!--stream->chan->refcount)
208 free_chan = stream->chan;
209
210 call_rcu(&stream->node.head, consumer_free_stream);
211 end:
212 consumer_data.need_update = 1;
213 pthread_mutex_unlock(&consumer_data.lock);
214
215 if (free_chan)
216 consumer_del_channel(free_chan);
217 }
218
219 struct lttng_consumer_stream *consumer_allocate_stream(
220 int channel_key, int stream_key,
221 int shm_fd, int wait_fd,
222 enum lttng_consumer_stream_state state,
223 uint64_t mmap_len,
224 enum lttng_event_output output,
225 const char *path_name,
226 uid_t uid,
227 gid_t gid)
228 {
229 struct lttng_consumer_stream *stream;
230 int ret;
231
232 stream = zmalloc(sizeof(*stream));
233 if (stream == NULL) {
234 perror("malloc struct lttng_consumer_stream");
235 goto end;
236 }
237 stream->chan = consumer_find_channel(channel_key);
238 if (!stream->chan) {
239 perror("Unable to find channel key");
240 goto end;
241 }
242 stream->chan->refcount++;
243 stream->key = stream_key;
244 stream->shm_fd = shm_fd;
245 stream->wait_fd = wait_fd;
246 stream->out_fd = -1;
247 stream->out_fd_offset = 0;
248 stream->state = state;
249 stream->mmap_len = mmap_len;
250 stream->mmap_base = NULL;
251 stream->output = output;
252 stream->uid = uid;
253 stream->gid = gid;
254 strncpy(stream->path_name, path_name, PATH_MAX - 1);
255 stream->path_name[PATH_MAX - 1] = '\0';
256 lttng_ht_node_init_ulong(&stream->node, stream->key);
257
258 switch (consumer_data.type) {
259 case LTTNG_CONSUMER_KERNEL:
260 break;
261 case LTTNG_CONSUMER32_UST:
262 case LTTNG_CONSUMER64_UST:
263 stream->cpu = stream->chan->cpucount++;
264 ret = lttng_ustconsumer_allocate_stream(stream);
265 if (ret) {
266 free(stream);
267 return NULL;
268 }
269 break;
270 default:
271 ERR("Unknown consumer_data type");
272 assert(0);
273 goto end;
274 }
275 DBG("Allocated stream %s (key %d, shm_fd %d, wait_fd %d, mmap_len %llu, out_fd %d)",
276 stream->path_name, stream->key,
277 stream->shm_fd,
278 stream->wait_fd,
279 (unsigned long long) stream->mmap_len,
280 stream->out_fd);
281 end:
282 return stream;
283 }
284
285 /*
286 * Add a stream to the global list protected by a mutex.
287 */
288 int consumer_add_stream(struct lttng_consumer_stream *stream)
289 {
290 int ret = 0;
291
292 pthread_mutex_lock(&consumer_data.lock);
293 /* Steal stream identifier, for UST */
294 consumer_steal_stream_key(stream->key);
295 rcu_read_lock();
296 /*
297 * We simply remove the old channel from the hash table. It's
298 * ok, since we know for sure the sessiond wants to replace it
299 * with the new version, because the key has been reused.
300 */
301 (void) lttng_ht_add_replace_ulong(consumer_data.stream_ht, &stream->node);
302 rcu_read_unlock();
303 consumer_data.stream_count++;
304 consumer_data.need_update = 1;
305
306 switch (consumer_data.type) {
307 case LTTNG_CONSUMER_KERNEL:
308 break;
309 case LTTNG_CONSUMER32_UST:
310 case LTTNG_CONSUMER64_UST:
311 /* Streams are in CPU number order (we rely on this) */
312 stream->cpu = stream->chan->nr_streams++;
313 break;
314 default:
315 ERR("Unknown consumer_data type");
316 assert(0);
317 goto end;
318 }
319
320 end:
321 pthread_mutex_unlock(&consumer_data.lock);
322
323 return ret;
324 }
325
326 /*
327 * Update a stream according to what we just received.
328 */
329 void consumer_change_stream_state(int stream_key,
330 enum lttng_consumer_stream_state state)
331 {
332 struct lttng_consumer_stream *stream;
333
334 pthread_mutex_lock(&consumer_data.lock);
335 stream = consumer_find_stream(stream_key);
336 if (stream) {
337 stream->state = state;
338 }
339 consumer_data.need_update = 1;
340 pthread_mutex_unlock(&consumer_data.lock);
341 }
342
343 static
344 void consumer_free_channel(struct rcu_head *head)
345 {
346 struct lttng_ht_node_ulong *node =
347 caa_container_of(head, struct lttng_ht_node_ulong, head);
348 struct lttng_consumer_channel *channel =
349 caa_container_of(node, struct lttng_consumer_channel, node);
350
351 free(channel);
352 }
353
354 /*
355 * Remove a channel from the global list protected by a mutex. This
356 * function is also responsible for freeing its data structures.
357 */
358 void consumer_del_channel(struct lttng_consumer_channel *channel)
359 {
360 int ret;
361 struct lttng_ht_iter iter;
362
363 pthread_mutex_lock(&consumer_data.lock);
364
365 switch (consumer_data.type) {
366 case LTTNG_CONSUMER_KERNEL:
367 break;
368 case LTTNG_CONSUMER32_UST:
369 case LTTNG_CONSUMER64_UST:
370 lttng_ustconsumer_del_channel(channel);
371 break;
372 default:
373 ERR("Unknown consumer_data type");
374 assert(0);
375 goto end;
376 }
377
378 rcu_read_lock();
379
380 lttng_ht_lookup(consumer_data.channel_ht,
381 (void *)((unsigned long) channel->key), &iter);
382
383 /*
384 * Remove channel node from hash table. It can fail if it's been
385 * replaced due to key reuse.
386 */
387 (void) lttng_ht_del(consumer_data.channel_ht, &iter);
388
389 rcu_read_unlock();
390
391 if (channel->mmap_base != NULL) {
392 ret = munmap(channel->mmap_base, channel->mmap_len);
393 if (ret != 0) {
394 perror("munmap");
395 }
396 }
397 if (channel->wait_fd >= 0 && !channel->wait_fd_is_copy) {
398 ret = close(channel->wait_fd);
399 if (ret) {
400 PERROR("close");
401 }
402 }
403 if (channel->shm_fd >= 0 && channel->wait_fd != channel->shm_fd) {
404 ret = close(channel->shm_fd);
405 if (ret) {
406 PERROR("close");
407 }
408 }
409
410 call_rcu(&channel->node.head, consumer_free_channel);
411 end:
412 pthread_mutex_unlock(&consumer_data.lock);
413 }
414
415 struct lttng_consumer_channel *consumer_allocate_channel(
416 int channel_key,
417 int shm_fd, int wait_fd,
418 uint64_t mmap_len,
419 uint64_t max_sb_size)
420 {
421 struct lttng_consumer_channel *channel;
422 int ret;
423
424 channel = zmalloc(sizeof(*channel));
425 if (channel == NULL) {
426 perror("malloc struct lttng_consumer_channel");
427 goto end;
428 }
429 channel->key = channel_key;
430 channel->shm_fd = shm_fd;
431 channel->wait_fd = wait_fd;
432 channel->mmap_len = mmap_len;
433 channel->max_sb_size = max_sb_size;
434 channel->refcount = 0;
435 channel->nr_streams = 0;
436 lttng_ht_node_init_ulong(&channel->node, channel->key);
437
438 switch (consumer_data.type) {
439 case LTTNG_CONSUMER_KERNEL:
440 channel->mmap_base = NULL;
441 channel->mmap_len = 0;
442 break;
443 case LTTNG_CONSUMER32_UST:
444 case LTTNG_CONSUMER64_UST:
445 ret = lttng_ustconsumer_allocate_channel(channel);
446 if (ret) {
447 free(channel);
448 return NULL;
449 }
450 break;
451 default:
452 ERR("Unknown consumer_data type");
453 assert(0);
454 goto end;
455 }
456 DBG("Allocated channel (key %d, shm_fd %d, wait_fd %d, mmap_len %llu, max_sb_size %llu)",
457 channel->key,
458 channel->shm_fd,
459 channel->wait_fd,
460 (unsigned long long) channel->mmap_len,
461 (unsigned long long) channel->max_sb_size);
462 end:
463 return channel;
464 }
465
466 /*
467 * Add a channel to the global list protected by a mutex.
468 */
469 int consumer_add_channel(struct lttng_consumer_channel *channel)
470 {
471 pthread_mutex_lock(&consumer_data.lock);
472 /* Steal channel identifier, for UST */
473 consumer_steal_channel_key(channel->key);
474 rcu_read_lock();
475 /*
476 * We simply remove the old channel from the hash table. It's
477 * ok, since we know for sure the sessiond wants to replace it
478 * with the new version, because the key has been reused.
479 */
480 (void) lttng_ht_add_replace_ulong(consumer_data.channel_ht, &channel->node);
481 rcu_read_unlock();
482 pthread_mutex_unlock(&consumer_data.lock);
483
484 return 0;
485 }
486
487 /*
488 * Allocate the pollfd structure and the local view of the out fds to avoid
489 * doing a lookup in the linked list and concurrency issues when writing is
490 * needed. Called with consumer_data.lock held.
491 *
492 * Returns the number of fds in the structures.
493 */
494 int consumer_update_poll_array(
495 struct lttng_consumer_local_data *ctx, struct pollfd **pollfd,
496 struct lttng_consumer_stream **local_stream)
497 {
498 int i = 0;
499 struct lttng_ht_iter iter;
500 struct lttng_consumer_stream *stream;
501
502 DBG("Updating poll fd array");
503 rcu_read_lock();
504 cds_lfht_for_each_entry(consumer_data.stream_ht->ht, &iter.iter, stream,
505 node.node) {
506 if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM) {
507 continue;
508 }
509 DBG("Active FD %d", stream->wait_fd);
510 (*pollfd)[i].fd = stream->wait_fd;
511 (*pollfd)[i].events = POLLIN | POLLPRI;
512 local_stream[i] = stream;
513 i++;
514 }
515 rcu_read_unlock();
516
517 /*
518 * Insert the consumer_poll_pipe at the end of the array and don't
519 * increment i so nb_fd is the number of real FD.
520 */
521 (*pollfd)[i].fd = ctx->consumer_poll_pipe[0];
522 (*pollfd)[i].events = POLLIN;
523 return i;
524 }
525
526 /*
527 * Poll on the should_quit pipe and the command socket return -1 on error and
528 * should exit, 0 if data is available on the command socket
529 */
530 int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll)
531 {
532 int num_rdy;
533
534 restart:
535 num_rdy = poll(consumer_sockpoll, 2, -1);
536 if (num_rdy == -1) {
537 /*
538 * Restart interrupted system call.
539 */
540 if (errno == EINTR) {
541 goto restart;
542 }
543 perror("Poll error");
544 goto exit;
545 }
546 if (consumer_sockpoll[0].revents == POLLIN) {
547 DBG("consumer_should_quit wake up");
548 goto exit;
549 }
550 return 0;
551
552 exit:
553 return -1;
554 }
555
556 /*
557 * Set the error socket.
558 */
559 void lttng_consumer_set_error_sock(
560 struct lttng_consumer_local_data *ctx, int sock)
561 {
562 ctx->consumer_error_socket = sock;
563 }
564
565 /*
566 * Set the command socket path.
567 */
568
569 void lttng_consumer_set_command_sock_path(
570 struct lttng_consumer_local_data *ctx, char *sock)
571 {
572 ctx->consumer_command_sock_path = sock;
573 }
574
575 /*
576 * Send return code to the session daemon.
577 * If the socket is not defined, we return 0, it is not a fatal error
578 */
579 int lttng_consumer_send_error(
580 struct lttng_consumer_local_data *ctx, int cmd)
581 {
582 if (ctx->consumer_error_socket > 0) {
583 return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd,
584 sizeof(enum lttcomm_sessiond_command));
585 }
586
587 return 0;
588 }
589
590 /*
591 * Close all the tracefiles and stream fds, should be called when all instances
592 * are destroyed.
593 */
594 void lttng_consumer_cleanup(void)
595 {
596 struct lttng_ht_iter iter;
597 struct lttng_ht_node_ulong *node;
598
599 rcu_read_lock();
600
601 /*
602 * close all outfd. Called when there are no more threads running (after
603 * joining on the threads), no need to protect list iteration with mutex.
604 */
605 cds_lfht_for_each_entry(consumer_data.stream_ht->ht, &iter.iter, node,
606 node) {
607 struct lttng_consumer_stream *stream =
608 caa_container_of(node, struct lttng_consumer_stream, node);
609 consumer_del_stream(stream);
610 }
611
612 cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, node,
613 node) {
614 struct lttng_consumer_channel *channel =
615 caa_container_of(node, struct lttng_consumer_channel, node);
616 consumer_del_channel(channel);
617 }
618
619 rcu_read_unlock();
620 }
621
622 /*
623 * Called from signal handler.
624 */
625 void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx)
626 {
627 int ret;
628 consumer_quit = 1;
629 ret = write(ctx->consumer_should_quit[1], "4", 1);
630 if (ret < 0) {
631 perror("write consumer quit");
632 }
633 }
634
635 void lttng_consumer_sync_trace_file(
636 struct lttng_consumer_stream *stream, off_t orig_offset)
637 {
638 int outfd = stream->out_fd;
639
640 /*
641 * This does a blocking write-and-wait on any page that belongs to the
642 * subbuffer prior to the one we just wrote.
643 * Don't care about error values, as these are just hints and ways to
644 * limit the amount of page cache used.
645 */
646 if (orig_offset < stream->chan->max_sb_size) {
647 return;
648 }
649 sync_file_range(outfd, orig_offset - stream->chan->max_sb_size,
650 stream->chan->max_sb_size,
651 SYNC_FILE_RANGE_WAIT_BEFORE
652 | SYNC_FILE_RANGE_WRITE
653 | SYNC_FILE_RANGE_WAIT_AFTER);
654 /*
655 * Give hints to the kernel about how we access the file:
656 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
657 * we write it.
658 *
659 * We need to call fadvise again after the file grows because the
660 * kernel does not seem to apply fadvise to non-existing parts of the
661 * file.
662 *
663 * Call fadvise _after_ having waited for the page writeback to
664 * complete because the dirty page writeback semantic is not well
665 * defined. So it can be expected to lead to lower throughput in
666 * streaming.
667 */
668 posix_fadvise(outfd, orig_offset - stream->chan->max_sb_size,
669 stream->chan->max_sb_size, POSIX_FADV_DONTNEED);
670 }
671
672 /*
673 * Initialise the necessary environnement :
674 * - create a new context
675 * - create the poll_pipe
676 * - create the should_quit pipe (for signal handler)
677 * - create the thread pipe (for splice)
678 *
679 * Takes a function pointer as argument, this function is called when data is
680 * available on a buffer. This function is responsible to do the
681 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
682 * buffer configuration and then kernctl_put_next_subbuf at the end.
683 *
684 * Returns a pointer to the new context or NULL on error.
685 */
686 struct lttng_consumer_local_data *lttng_consumer_create(
687 enum lttng_consumer_type type,
688 ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream,
689 struct lttng_consumer_local_data *ctx),
690 int (*recv_channel)(struct lttng_consumer_channel *channel),
691 int (*recv_stream)(struct lttng_consumer_stream *stream),
692 int (*update_stream)(int stream_key, uint32_t state))
693 {
694 int ret, i;
695 struct lttng_consumer_local_data *ctx;
696
697 assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN ||
698 consumer_data.type == type);
699 consumer_data.type = type;
700
701 ctx = zmalloc(sizeof(struct lttng_consumer_local_data));
702 if (ctx == NULL) {
703 perror("allocating context");
704 goto error;
705 }
706
707 ctx->consumer_error_socket = -1;
708 /* assign the callbacks */
709 ctx->on_buffer_ready = buffer_ready;
710 ctx->on_recv_channel = recv_channel;
711 ctx->on_recv_stream = recv_stream;
712 ctx->on_update_stream = update_stream;
713
714 ret = pipe(ctx->consumer_poll_pipe);
715 if (ret < 0) {
716 perror("Error creating poll pipe");
717 goto error_poll_pipe;
718 }
719
720 ret = pipe(ctx->consumer_should_quit);
721 if (ret < 0) {
722 perror("Error creating recv pipe");
723 goto error_quit_pipe;
724 }
725
726 ret = pipe(ctx->consumer_thread_pipe);
727 if (ret < 0) {
728 perror("Error creating thread pipe");
729 goto error_thread_pipe;
730 }
731
732 return ctx;
733
734
735 error_thread_pipe:
736 for (i = 0; i < 2; i++) {
737 int err;
738
739 err = close(ctx->consumer_should_quit[i]);
740 if (err) {
741 PERROR("close");
742 }
743 }
744 error_quit_pipe:
745 for (i = 0; i < 2; i++) {
746 int err;
747
748 err = close(ctx->consumer_poll_pipe[i]);
749 if (err) {
750 PERROR("close");
751 }
752 }
753 error_poll_pipe:
754 free(ctx);
755 error:
756 return NULL;
757 }
758
759 /*
760 * Close all fds associated with the instance and free the context.
761 */
762 void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx)
763 {
764 int ret;
765
766 ret = close(ctx->consumer_error_socket);
767 if (ret) {
768 PERROR("close");
769 }
770 ret = close(ctx->consumer_thread_pipe[0]);
771 if (ret) {
772 PERROR("close");
773 }
774 ret = close(ctx->consumer_thread_pipe[1]);
775 if (ret) {
776 PERROR("close");
777 }
778 ret = close(ctx->consumer_poll_pipe[0]);
779 if (ret) {
780 PERROR("close");
781 }
782 ret = close(ctx->consumer_poll_pipe[1]);
783 if (ret) {
784 PERROR("close");
785 }
786 ret = close(ctx->consumer_should_quit[0]);
787 if (ret) {
788 PERROR("close");
789 }
790 ret = close(ctx->consumer_should_quit[1]);
791 if (ret) {
792 PERROR("close");
793 }
794 unlink(ctx->consumer_command_sock_path);
795 free(ctx);
796 }
797
798 /*
799 * Mmap the ring buffer, read it and write the data to the tracefile.
800 *
801 * Returns the number of bytes written
802 */
803 ssize_t lttng_consumer_on_read_subbuffer_mmap(
804 struct lttng_consumer_local_data *ctx,
805 struct lttng_consumer_stream *stream, unsigned long len)
806 {
807 switch (consumer_data.type) {
808 case LTTNG_CONSUMER_KERNEL:
809 return lttng_kconsumer_on_read_subbuffer_mmap(ctx, stream, len);
810 case LTTNG_CONSUMER32_UST:
811 case LTTNG_CONSUMER64_UST:
812 return lttng_ustconsumer_on_read_subbuffer_mmap(ctx, stream, len);
813 default:
814 ERR("Unknown consumer_data type");
815 assert(0);
816 }
817 }
818
819 /*
820 * Splice the data from the ring buffer to the tracefile.
821 *
822 * Returns the number of bytes spliced.
823 */
824 ssize_t lttng_consumer_on_read_subbuffer_splice(
825 struct lttng_consumer_local_data *ctx,
826 struct lttng_consumer_stream *stream, unsigned long len)
827 {
828 switch (consumer_data.type) {
829 case LTTNG_CONSUMER_KERNEL:
830 return lttng_kconsumer_on_read_subbuffer_splice(ctx, stream, len);
831 case LTTNG_CONSUMER32_UST:
832 case LTTNG_CONSUMER64_UST:
833 return -ENOSYS;
834 default:
835 ERR("Unknown consumer_data type");
836 assert(0);
837 return -ENOSYS;
838 }
839
840 }
841
842 /*
843 * Take a snapshot for a specific fd
844 *
845 * Returns 0 on success, < 0 on error
846 */
847 int lttng_consumer_take_snapshot(struct lttng_consumer_local_data *ctx,
848 struct lttng_consumer_stream *stream)
849 {
850 switch (consumer_data.type) {
851 case LTTNG_CONSUMER_KERNEL:
852 return lttng_kconsumer_take_snapshot(ctx, stream);
853 case LTTNG_CONSUMER32_UST:
854 case LTTNG_CONSUMER64_UST:
855 return lttng_ustconsumer_take_snapshot(ctx, stream);
856 default:
857 ERR("Unknown consumer_data type");
858 assert(0);
859 return -ENOSYS;
860 }
861
862 }
863
864 /*
865 * Get the produced position
866 *
867 * Returns 0 on success, < 0 on error
868 */
869 int lttng_consumer_get_produced_snapshot(
870 struct lttng_consumer_local_data *ctx,
871 struct lttng_consumer_stream *stream,
872 unsigned long *pos)
873 {
874 switch (consumer_data.type) {
875 case LTTNG_CONSUMER_KERNEL:
876 return lttng_kconsumer_get_produced_snapshot(ctx, stream, pos);
877 case LTTNG_CONSUMER32_UST:
878 case LTTNG_CONSUMER64_UST:
879 return lttng_ustconsumer_get_produced_snapshot(ctx, stream, pos);
880 default:
881 ERR("Unknown consumer_data type");
882 assert(0);
883 return -ENOSYS;
884 }
885 }
886
887 int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx,
888 int sock, struct pollfd *consumer_sockpoll)
889 {
890 switch (consumer_data.type) {
891 case LTTNG_CONSUMER_KERNEL:
892 return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
893 case LTTNG_CONSUMER32_UST:
894 case LTTNG_CONSUMER64_UST:
895 return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
896 default:
897 ERR("Unknown consumer_data type");
898 assert(0);
899 return -ENOSYS;
900 }
901 }
902
903 /*
904 * This thread polls the fds in the set to consume the data and write
905 * it to tracefile if necessary.
906 */
907 void *lttng_consumer_thread_poll_fds(void *data)
908 {
909 int num_rdy, num_hup, high_prio, ret, i;
910 struct pollfd *pollfd = NULL;
911 /* local view of the streams */
912 struct lttng_consumer_stream **local_stream = NULL;
913 /* local view of consumer_data.fds_count */
914 int nb_fd = 0;
915 char tmp;
916 int tmp2;
917 struct lttng_consumer_local_data *ctx = data;
918
919 rcu_register_thread();
920
921 local_stream = zmalloc(sizeof(struct lttng_consumer_stream));
922
923 while (1) {
924 high_prio = 0;
925 num_hup = 0;
926
927 /*
928 * the fds set has been updated, we need to update our
929 * local array as well
930 */
931 pthread_mutex_lock(&consumer_data.lock);
932 if (consumer_data.need_update) {
933 if (pollfd != NULL) {
934 free(pollfd);
935 pollfd = NULL;
936 }
937 if (local_stream != NULL) {
938 free(local_stream);
939 local_stream = NULL;
940 }
941
942 /* allocate for all fds + 1 for the consumer_poll_pipe */
943 pollfd = zmalloc((consumer_data.stream_count + 1) * sizeof(struct pollfd));
944 if (pollfd == NULL) {
945 perror("pollfd malloc");
946 pthread_mutex_unlock(&consumer_data.lock);
947 goto end;
948 }
949
950 /* allocate for all fds + 1 for the consumer_poll_pipe */
951 local_stream = zmalloc((consumer_data.stream_count + 1) *
952 sizeof(struct lttng_consumer_stream));
953 if (local_stream == NULL) {
954 perror("local_stream malloc");
955 pthread_mutex_unlock(&consumer_data.lock);
956 goto end;
957 }
958 ret = consumer_update_poll_array(ctx, &pollfd, local_stream);
959 if (ret < 0) {
960 ERR("Error in allocating pollfd or local_outfds");
961 lttng_consumer_send_error(ctx, CONSUMERD_POLL_ERROR);
962 pthread_mutex_unlock(&consumer_data.lock);
963 goto end;
964 }
965 nb_fd = ret;
966 consumer_data.need_update = 0;
967 }
968 pthread_mutex_unlock(&consumer_data.lock);
969
970 /* No FDs and consumer_quit, consumer_cleanup the thread */
971 if (nb_fd == 0 && consumer_quit == 1) {
972 goto end;
973 }
974 /* poll on the array of fds */
975 restart:
976 DBG("polling on %d fd", nb_fd + 1);
977 num_rdy = poll(pollfd, nb_fd + 1, consumer_poll_timeout);
978 DBG("poll num_rdy : %d", num_rdy);
979 if (num_rdy == -1) {
980 /*
981 * Restart interrupted system call.
982 */
983 if (errno == EINTR) {
984 goto restart;
985 }
986 perror("Poll error");
987 lttng_consumer_send_error(ctx, CONSUMERD_POLL_ERROR);
988 goto end;
989 } else if (num_rdy == 0) {
990 DBG("Polling thread timed out");
991 goto end;
992 }
993
994 /*
995 * If the consumer_poll_pipe triggered poll go
996 * directly to the beginning of the loop to update the
997 * array. We want to prioritize array update over
998 * low-priority reads.
999 */
1000 if (pollfd[nb_fd].revents & POLLIN) {
1001 DBG("consumer_poll_pipe wake up");
1002 tmp2 = read(ctx->consumer_poll_pipe[0], &tmp, 1);
1003 if (tmp2 < 0) {
1004 perror("read consumer poll");
1005 }
1006 continue;
1007 }
1008
1009 /* Take care of high priority channels first. */
1010 for (i = 0; i < nb_fd; i++) {
1011 if (pollfd[i].revents & POLLPRI) {
1012 ssize_t len;
1013
1014 DBG("Urgent read on fd %d", pollfd[i].fd);
1015 high_prio = 1;
1016 len = ctx->on_buffer_ready(local_stream[i], ctx);
1017 /* it's ok to have an unavailable sub-buffer */
1018 if (len < 0 && len != -EAGAIN) {
1019 goto end;
1020 } else if (len > 0) {
1021 local_stream[i]->data_read = 1;
1022 }
1023 }
1024 }
1025
1026 /*
1027 * If we read high prio channel in this loop, try again
1028 * for more high prio data.
1029 */
1030 if (high_prio) {
1031 continue;
1032 }
1033
1034 /* Take care of low priority channels. */
1035 for (i = 0; i < nb_fd; i++) {
1036 if ((pollfd[i].revents & POLLIN) ||
1037 local_stream[i]->hangup_flush_done) {
1038 ssize_t len;
1039
1040 assert(!(pollfd[i].revents & POLLERR));
1041 assert(!(pollfd[i].revents & POLLNVAL));
1042 DBG("Normal read on fd %d", pollfd[i].fd);
1043 len = ctx->on_buffer_ready(local_stream[i], ctx);
1044 /* it's ok to have an unavailable sub-buffer */
1045 if (len < 0 && len != -EAGAIN) {
1046 goto end;
1047 } else if (len > 0) {
1048 local_stream[i]->data_read = 1;
1049 }
1050 }
1051 }
1052
1053 /* Handle hangup and errors */
1054 for (i = 0; i < nb_fd; i++) {
1055 if (!local_stream[i]->hangup_flush_done
1056 && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL))
1057 && (consumer_data.type == LTTNG_CONSUMER32_UST
1058 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
1059 DBG("fd %d is hup|err|nval. Attempting flush and read.",
1060 pollfd[i].fd);
1061 lttng_ustconsumer_on_stream_hangup(local_stream[i]);
1062 /* Attempt read again, for the data we just flushed. */
1063 local_stream[i]->data_read = 1;
1064 }
1065 /*
1066 * If the poll flag is HUP/ERR/NVAL and we have
1067 * read no data in this pass, we can remove the
1068 * stream from its hash table.
1069 */
1070 if ((pollfd[i].revents & POLLHUP)) {
1071 DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
1072 if (!local_stream[i]->data_read) {
1073 consumer_del_stream(local_stream[i]);
1074 num_hup++;
1075 }
1076 } else if (pollfd[i].revents & POLLERR) {
1077 ERR("Error returned in polling fd %d.", pollfd[i].fd);
1078 if (!local_stream[i]->data_read) {
1079 consumer_del_stream(local_stream[i]);
1080 num_hup++;
1081 }
1082 } else if (pollfd[i].revents & POLLNVAL) {
1083 ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
1084 if (!local_stream[i]->data_read) {
1085 consumer_del_stream(local_stream[i]);
1086 num_hup++;
1087 }
1088 }
1089 local_stream[i]->data_read = 0;
1090 }
1091 }
1092 end:
1093 DBG("polling thread exiting");
1094 if (pollfd != NULL) {
1095 free(pollfd);
1096 pollfd = NULL;
1097 }
1098 if (local_stream != NULL) {
1099 free(local_stream);
1100 local_stream = NULL;
1101 }
1102 rcu_unregister_thread();
1103 return NULL;
1104 }
1105
1106 /*
1107 * This thread listens on the consumerd socket and receives the file
1108 * descriptors from the session daemon.
1109 */
1110 void *lttng_consumer_thread_receive_fds(void *data)
1111 {
1112 int sock, client_socket, ret;
1113 /*
1114 * structure to poll for incoming data on communication socket avoids
1115 * making blocking sockets.
1116 */
1117 struct pollfd consumer_sockpoll[2];
1118 struct lttng_consumer_local_data *ctx = data;
1119
1120 rcu_register_thread();
1121
1122 DBG("Creating command socket %s", ctx->consumer_command_sock_path);
1123 unlink(ctx->consumer_command_sock_path);
1124 client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path);
1125 if (client_socket < 0) {
1126 ERR("Cannot create command socket");
1127 goto end;
1128 }
1129
1130 ret = lttcomm_listen_unix_sock(client_socket);
1131 if (ret < 0) {
1132 goto end;
1133 }
1134
1135 DBG("Sending ready command to lttng-sessiond");
1136 ret = lttng_consumer_send_error(ctx, CONSUMERD_COMMAND_SOCK_READY);
1137 /* return < 0 on error, but == 0 is not fatal */
1138 if (ret < 0) {
1139 ERR("Error sending ready command to lttng-sessiond");
1140 goto end;
1141 }
1142
1143 ret = fcntl(client_socket, F_SETFL, O_NONBLOCK);
1144 if (ret < 0) {
1145 perror("fcntl O_NONBLOCK");
1146 goto end;
1147 }
1148
1149 /* prepare the FDs to poll : to client socket and the should_quit pipe */
1150 consumer_sockpoll[0].fd = ctx->consumer_should_quit[0];
1151 consumer_sockpoll[0].events = POLLIN | POLLPRI;
1152 consumer_sockpoll[1].fd = client_socket;
1153 consumer_sockpoll[1].events = POLLIN | POLLPRI;
1154
1155 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
1156 goto end;
1157 }
1158 DBG("Connection on client_socket");
1159
1160 /* Blocking call, waiting for transmission */
1161 sock = lttcomm_accept_unix_sock(client_socket);
1162 if (sock <= 0) {
1163 WARN("On accept");
1164 goto end;
1165 }
1166 ret = fcntl(sock, F_SETFL, O_NONBLOCK);
1167 if (ret < 0) {
1168 perror("fcntl O_NONBLOCK");
1169 goto end;
1170 }
1171
1172 /* update the polling structure to poll on the established socket */
1173 consumer_sockpoll[1].fd = sock;
1174 consumer_sockpoll[1].events = POLLIN | POLLPRI;
1175
1176 while (1) {
1177 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
1178 goto end;
1179 }
1180 DBG("Incoming command on sock");
1181 ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll);
1182 if (ret == -ENOENT) {
1183 DBG("Received STOP command");
1184 goto end;
1185 }
1186 if (ret < 0) {
1187 ERR("Communication interrupted on command socket");
1188 goto end;
1189 }
1190 if (consumer_quit) {
1191 DBG("consumer_thread_receive_fds received quit from signal");
1192 goto end;
1193 }
1194 DBG("received fds on sock");
1195 }
1196 end:
1197 DBG("consumer_thread_receive_fds exiting");
1198
1199 /*
1200 * when all fds have hung up, the polling thread
1201 * can exit cleanly
1202 */
1203 consumer_quit = 1;
1204
1205 /*
1206 * 2s of grace period, if no polling events occur during
1207 * this period, the polling thread will exit even if there
1208 * are still open FDs (should not happen, but safety mechanism).
1209 */
1210 consumer_poll_timeout = LTTNG_CONSUMER_POLL_TIMEOUT;
1211
1212 /* wake up the polling thread */
1213 ret = write(ctx->consumer_poll_pipe[1], "4", 1);
1214 if (ret < 0) {
1215 perror("poll pipe write");
1216 }
1217 rcu_unregister_thread();
1218 return NULL;
1219 }
1220
1221 ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream,
1222 struct lttng_consumer_local_data *ctx)
1223 {
1224 switch (consumer_data.type) {
1225 case LTTNG_CONSUMER_KERNEL:
1226 return lttng_kconsumer_read_subbuffer(stream, ctx);
1227 case LTTNG_CONSUMER32_UST:
1228 case LTTNG_CONSUMER64_UST:
1229 return lttng_ustconsumer_read_subbuffer(stream, ctx);
1230 default:
1231 ERR("Unknown consumer_data type");
1232 assert(0);
1233 return -ENOSYS;
1234 }
1235 }
1236
1237 int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream)
1238 {
1239 switch (consumer_data.type) {
1240 case LTTNG_CONSUMER_KERNEL:
1241 return lttng_kconsumer_on_recv_stream(stream);
1242 case LTTNG_CONSUMER32_UST:
1243 case LTTNG_CONSUMER64_UST:
1244 return lttng_ustconsumer_on_recv_stream(stream);
1245 default:
1246 ERR("Unknown consumer_data type");
1247 assert(0);
1248 return -ENOSYS;
1249 }
1250 }
1251
1252 /*
1253 * Allocate and set consumer data hash tables.
1254 */
1255 void lttng_consumer_init(void)
1256 {
1257 consumer_data.stream_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1258 consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
1259 }
1260
This page took 0.089429 seconds and 3 git commands to generate.