Cleanup: use LTTNG_HIDDEN in src/common
[lttng-tools.git] / src / common / consumer.c
CommitLineData
3bd1e081
MD
1/*
2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
00e2e675 4 * 2012 - David Goulet <dgoulet@efficios.com>
3bd1e081 5 *
d14d33bf
AM
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
3bd1e081 9 *
d14d33bf
AM
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
3bd1e081 14 *
d14d33bf
AM
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
3bd1e081
MD
18 */
19
20#define _GNU_SOURCE
21#include <assert.h>
3bd1e081
MD
22#include <poll.h>
23#include <pthread.h>
24#include <stdlib.h>
25#include <string.h>
26#include <sys/mman.h>
27#include <sys/socket.h>
28#include <sys/types.h>
29#include <unistd.h>
77c7c900 30#include <inttypes.h>
3bd1e081 31
990570ed 32#include <common/common.h>
fb3a43a9
DG
33#include <common/utils.h>
34#include <common/compat/poll.h>
10a8a223 35#include <common/kernel-ctl/kernel-ctl.h>
00e2e675 36#include <common/sessiond-comm/relayd.h>
10a8a223
DG
37#include <common/sessiond-comm/sessiond-comm.h>
38#include <common/kernel-consumer/kernel-consumer.h>
00e2e675 39#include <common/relayd/relayd.h>
10a8a223
DG
40#include <common/ust-consumer/ust-consumer.h>
41
42#include "consumer.h"
3bd1e081
MD
43
44struct lttng_consumer_global_data consumer_data = {
3bd1e081
MD
45 .stream_count = 0,
46 .need_update = 1,
47 .type = LTTNG_CONSUMER_UNKNOWN,
48};
49
3bd1e081
MD
50/*
51 * Flag to inform the polling thread to quit when all fd hung up. Updated by
52 * the consumer_thread_receive_fds when it notices that all fds has hung up.
53 * Also updated by the signal handler (consumer_should_exit()). Read by the
54 * polling threads.
55 */
a98dae5f 56volatile int consumer_quit;
3bd1e081 57
43c34bc3 58/*
43c34bc3
DG
59 * Global hash table containing respectively metadata and data streams. The
60 * stream element in this ht should only be updated by the metadata poll thread
61 * for the metadata and the data poll thread for the data.
62 */
40dc48e0
DG
63static struct lttng_ht *metadata_ht;
64static struct lttng_ht *data_ht;
43c34bc3 65
8994307f
DG
66/*
67 * Notify a thread pipe to poll back again. This usually means that some global
68 * state has changed so we just send back the thread in a poll wait call.
69 */
70static void notify_thread_pipe(int wpipe)
71{
72 int ret;
73
74 do {
75 struct lttng_consumer_stream *null_stream = NULL;
76
77 ret = write(wpipe, &null_stream, sizeof(null_stream));
78 } while (ret < 0 && errno == EINTR);
79}
80
3bd1e081
MD
81/*
82 * Find a stream. The consumer_data.lock must be locked during this
83 * call.
84 */
8389e4f8
DG
85static struct lttng_consumer_stream *consumer_find_stream(int key,
86 struct lttng_ht *ht)
3bd1e081 87{
e4421fec
DG
88 struct lttng_ht_iter iter;
89 struct lttng_ht_node_ulong *node;
90 struct lttng_consumer_stream *stream = NULL;
3bd1e081 91
8389e4f8
DG
92 assert(ht);
93
7ad0a0cb 94 /* Negative keys are lookup failures */
7a57cf92 95 if (key < 0) {
7ad0a0cb 96 return NULL;
7a57cf92 97 }
e4421fec 98
6065ceec
DG
99 rcu_read_lock();
100
8389e4f8 101 lttng_ht_lookup(ht, (void *)((unsigned long) key), &iter);
e4421fec
DG
102 node = lttng_ht_iter_get_node_ulong(&iter);
103 if (node != NULL) {
104 stream = caa_container_of(node, struct lttng_consumer_stream, node);
3bd1e081 105 }
e4421fec 106
6065ceec
DG
107 rcu_read_unlock();
108
e4421fec 109 return stream;
3bd1e081
MD
110}
111
c869f647 112void consumer_steal_stream_key(int key, struct lttng_ht *ht)
7ad0a0cb
MD
113{
114 struct lttng_consumer_stream *stream;
115
04253271 116 rcu_read_lock();
8389e4f8 117 stream = consumer_find_stream(key, ht);
04253271 118 if (stream) {
7ad0a0cb 119 stream->key = -1;
04253271
MD
120 /*
121 * We don't want the lookup to match, but we still need
122 * to iterate on this stream when iterating over the hash table. Just
123 * change the node key.
124 */
125 stream->node.key = -1;
126 }
127 rcu_read_unlock();
7ad0a0cb
MD
128}
129
d56db448
DG
130/*
131 * Return a channel object for the given key.
132 *
133 * RCU read side lock MUST be acquired before calling this function and
134 * protects the channel ptr.
135 */
3bd1e081
MD
136static struct lttng_consumer_channel *consumer_find_channel(int key)
137{
e4421fec
DG
138 struct lttng_ht_iter iter;
139 struct lttng_ht_node_ulong *node;
140 struct lttng_consumer_channel *channel = NULL;
3bd1e081 141
7ad0a0cb 142 /* Negative keys are lookup failures */
7a57cf92 143 if (key < 0) {
7ad0a0cb 144 return NULL;
7a57cf92 145 }
e4421fec
DG
146
147 lttng_ht_lookup(consumer_data.channel_ht, (void *)((unsigned long) key),
148 &iter);
149 node = lttng_ht_iter_get_node_ulong(&iter);
150 if (node != NULL) {
151 channel = caa_container_of(node, struct lttng_consumer_channel, node);
3bd1e081 152 }
e4421fec
DG
153
154 return channel;
3bd1e081
MD
155}
156
7ad0a0cb
MD
157static void consumer_steal_channel_key(int key)
158{
159 struct lttng_consumer_channel *channel;
160
04253271 161 rcu_read_lock();
7ad0a0cb 162 channel = consumer_find_channel(key);
04253271 163 if (channel) {
7ad0a0cb 164 channel->key = -1;
04253271
MD
165 /*
166 * We don't want the lookup to match, but we still need
167 * to iterate on this channel when iterating over the hash table. Just
168 * change the node key.
169 */
170 channel->node.key = -1;
171 }
172 rcu_read_unlock();
7ad0a0cb
MD
173}
174
702b1ea4
MD
175static
176void consumer_free_stream(struct rcu_head *head)
177{
178 struct lttng_ht_node_ulong *node =
179 caa_container_of(head, struct lttng_ht_node_ulong, head);
180 struct lttng_consumer_stream *stream =
181 caa_container_of(node, struct lttng_consumer_stream, node);
182
183 free(stream);
184}
185
00e2e675
DG
186/*
187 * RCU protected relayd socket pair free.
188 */
189static void consumer_rcu_free_relayd(struct rcu_head *head)
190{
191 struct lttng_ht_node_ulong *node =
192 caa_container_of(head, struct lttng_ht_node_ulong, head);
193 struct consumer_relayd_sock_pair *relayd =
194 caa_container_of(node, struct consumer_relayd_sock_pair, node);
195
8994307f
DG
196 /*
197 * Close all sockets. This is done in the call RCU since we don't want the
198 * socket fds to be reassigned thus potentially creating bad state of the
199 * relayd object.
200 *
201 * We do not have to lock the control socket mutex here since at this stage
202 * there is no one referencing to this relayd object.
203 */
204 (void) relayd_close(&relayd->control_sock);
205 (void) relayd_close(&relayd->data_sock);
206
00e2e675
DG
207 free(relayd);
208}
209
210/*
211 * Destroy and free relayd socket pair object.
212 *
213 * This function MUST be called with the consumer_data lock acquired.
214 */
d09e1200 215static void destroy_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
216{
217 int ret;
218 struct lttng_ht_iter iter;
219
173af62f
DG
220 if (relayd == NULL) {
221 return;
222 }
223
00e2e675
DG
224 DBG("Consumer destroy and close relayd socket pair");
225
226 iter.iter.node = &relayd->node.node;
227 ret = lttng_ht_del(consumer_data.relayd_ht, &iter);
173af62f 228 if (ret != 0) {
8994307f 229 /* We assume the relayd is being or is destroyed */
173af62f
DG
230 return;
231 }
00e2e675 232
00e2e675
DG
233 /* RCU free() call */
234 call_rcu(&relayd->node.head, consumer_rcu_free_relayd);
235}
236
228b5bf7
DG
237/*
238 * Iterate over the relayd hash table and destroy each element. Finally,
239 * destroy the whole hash table.
240 */
241static void cleanup_relayd_ht(void)
242{
243 struct lttng_ht_iter iter;
244 struct consumer_relayd_sock_pair *relayd;
245
246 rcu_read_lock();
247
248 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
249 node.node) {
250 destroy_relayd(relayd);
251 }
252
253 lttng_ht_destroy(consumer_data.relayd_ht);
254
255 rcu_read_unlock();
256}
257
8994307f
DG
258/*
259 * Update the end point status of all streams having the given network sequence
260 * index (relayd index).
261 *
262 * It's atomically set without having the stream mutex locked which is fine
263 * because we handle the write/read race with a pipe wakeup for each thread.
264 */
265static void update_endpoint_status_by_netidx(int net_seq_idx,
266 enum consumer_endpoint_status status)
267{
268 struct lttng_ht_iter iter;
269 struct lttng_consumer_stream *stream;
270
271 DBG("Consumer set delete flag on stream by idx %d", net_seq_idx);
272
273 rcu_read_lock();
274
275 /* Let's begin with metadata */
276 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
277 if (stream->net_seq_idx == net_seq_idx) {
278 uatomic_set(&stream->endpoint_status, status);
279 DBG("Delete flag set to metadata stream %d", stream->wait_fd);
280 }
281 }
282
283 /* Follow up by the data streams */
284 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
285 if (stream->net_seq_idx == net_seq_idx) {
286 uatomic_set(&stream->endpoint_status, status);
287 DBG("Delete flag set to data stream %d", stream->wait_fd);
288 }
289 }
290 rcu_read_unlock();
291}
292
293/*
294 * Cleanup a relayd object by flagging every associated streams for deletion,
295 * destroying the object meaning removing it from the relayd hash table,
296 * closing the sockets and freeing the memory in a RCU call.
297 *
298 * If a local data context is available, notify the threads that the streams'
299 * state have changed.
300 */
301static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd,
302 struct lttng_consumer_local_data *ctx)
303{
304 int netidx;
305
306 assert(relayd);
307
9617607b
DG
308 DBG("Cleaning up relayd sockets");
309
8994307f
DG
310 /* Save the net sequence index before destroying the object */
311 netidx = relayd->net_seq_idx;
312
313 /*
314 * Delete the relayd from the relayd hash table, close the sockets and free
315 * the object in a RCU call.
316 */
317 destroy_relayd(relayd);
318
319 /* Set inactive endpoint to all streams */
320 update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE);
321
322 /*
323 * With a local data context, notify the threads that the streams' state
324 * have changed. The write() action on the pipe acts as an "implicit"
325 * memory barrier ordering the updates of the end point status from the
326 * read of this status which happens AFTER receiving this notify.
327 */
328 if (ctx) {
329 notify_thread_pipe(ctx->consumer_data_pipe[1]);
330 notify_thread_pipe(ctx->consumer_metadata_pipe[1]);
331 }
332}
333
a6ba4fe1
DG
334/*
335 * Flag a relayd socket pair for destruction. Destroy it if the refcount
336 * reaches zero.
337 *
338 * RCU read side lock MUST be aquired before calling this function.
339 */
340void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd)
341{
342 assert(relayd);
343
344 /* Set destroy flag for this object */
345 uatomic_set(&relayd->destroy_flag, 1);
346
347 /* Destroy the relayd if refcount is 0 */
348 if (uatomic_read(&relayd->refcount) == 0) {
d09e1200 349 destroy_relayd(relayd);
a6ba4fe1
DG
350 }
351}
352
3bd1e081
MD
353/*
354 * Remove a stream from the global list protected by a mutex. This
355 * function is also responsible for freeing its data structures.
356 */
e316aad5
DG
357void consumer_del_stream(struct lttng_consumer_stream *stream,
358 struct lttng_ht *ht)
3bd1e081
MD
359{
360 int ret;
e4421fec 361 struct lttng_ht_iter iter;
3bd1e081 362 struct lttng_consumer_channel *free_chan = NULL;
00e2e675
DG
363 struct consumer_relayd_sock_pair *relayd;
364
365 assert(stream);
3bd1e081 366
8994307f
DG
367 DBG("Consumer del stream %d", stream->wait_fd);
368
e316aad5
DG
369 if (ht == NULL) {
370 /* Means the stream was allocated but not successfully added */
371 goto free_stream;
372 }
373
3bd1e081 374 pthread_mutex_lock(&consumer_data.lock);
74251bb8 375 pthread_mutex_lock(&stream->lock);
3bd1e081
MD
376
377 switch (consumer_data.type) {
378 case LTTNG_CONSUMER_KERNEL:
379 if (stream->mmap_base != NULL) {
380 ret = munmap(stream->mmap_base, stream->mmap_len);
381 if (ret != 0) {
7a57cf92 382 PERROR("munmap");
3bd1e081
MD
383 }
384 }
385 break;
7753dea8
MD
386 case LTTNG_CONSUMER32_UST:
387 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
388 lttng_ustconsumer_del_stream(stream);
389 break;
390 default:
391 ERR("Unknown consumer_data type");
392 assert(0);
393 goto end;
394 }
395
6065ceec 396 rcu_read_lock();
04253271 397 iter.iter.node = &stream->node.node;
e316aad5 398 ret = lttng_ht_del(ht, &iter);
04253271 399 assert(!ret);
ca22feea
DG
400
401 /* Remove node session id from the consumer_data stream ht */
402 iter.iter.node = &stream->node_session_id.node;
403 ret = lttng_ht_del(consumer_data.stream_list_ht, &iter);
404 assert(!ret);
6065ceec
DG
405 rcu_read_unlock();
406
50f8ae69 407 assert(consumer_data.stream_count > 0);
3bd1e081 408 consumer_data.stream_count--;
50f8ae69 409
3bd1e081 410 if (stream->out_fd >= 0) {
4c462e79
MD
411 ret = close(stream->out_fd);
412 if (ret) {
413 PERROR("close");
414 }
3bd1e081 415 }
b5c5fc29 416 if (stream->wait_fd >= 0 && !stream->wait_fd_is_copy) {
4c462e79
MD
417 ret = close(stream->wait_fd);
418 if (ret) {
419 PERROR("close");
420 }
3bd1e081 421 }
2c1dd183 422 if (stream->shm_fd >= 0 && stream->wait_fd != stream->shm_fd) {
4c462e79
MD
423 ret = close(stream->shm_fd);
424 if (ret) {
425 PERROR("close");
426 }
3bd1e081 427 }
00e2e675
DG
428
429 /* Check and cleanup relayd */
b0b335c8 430 rcu_read_lock();
00e2e675
DG
431 relayd = consumer_find_relayd(stream->net_seq_idx);
432 if (relayd != NULL) {
b0b335c8
MD
433 uatomic_dec(&relayd->refcount);
434 assert(uatomic_read(&relayd->refcount) >= 0);
173af62f 435
3f8e211f
DG
436 /* Closing streams requires to lock the control socket. */
437 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
173af62f
DG
438 ret = relayd_send_close_stream(&relayd->control_sock,
439 stream->relayd_stream_id,
440 stream->next_net_seq_num - 1);
3f8e211f 441 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
173af62f 442 if (ret < 0) {
a4b92340
DG
443 DBG("Unable to close stream on the relayd. Continuing");
444 /*
445 * Continue here. There is nothing we can do for the relayd.
446 * Chances are that the relayd has closed the socket so we just
447 * continue cleaning up.
448 */
173af62f
DG
449 }
450
451 /* Both conditions are met, we destroy the relayd. */
452 if (uatomic_read(&relayd->refcount) == 0 &&
453 uatomic_read(&relayd->destroy_flag)) {
d09e1200 454 destroy_relayd(relayd);
00e2e675 455 }
00e2e675 456 }
b0b335c8 457 rcu_read_unlock();
00e2e675 458
c30aaa51
MD
459 uatomic_dec(&stream->chan->refcount);
460 if (!uatomic_read(&stream->chan->refcount)
461 && !uatomic_read(&stream->chan->nb_init_streams)) {
3bd1e081 462 free_chan = stream->chan;
00e2e675
DG
463 }
464
3bd1e081
MD
465end:
466 consumer_data.need_update = 1;
8994307f 467 pthread_mutex_unlock(&stream->lock);
74251bb8 468 pthread_mutex_unlock(&consumer_data.lock);
3bd1e081 469
c30aaa51 470 if (free_chan) {
3bd1e081 471 consumer_del_channel(free_chan);
c30aaa51 472 }
e316aad5
DG
473
474free_stream:
475 call_rcu(&stream->node.head, consumer_free_stream);
3bd1e081
MD
476}
477
478struct lttng_consumer_stream *consumer_allocate_stream(
479 int channel_key, int stream_key,
480 int shm_fd, int wait_fd,
481 enum lttng_consumer_stream_state state,
482 uint64_t mmap_len,
483 enum lttng_event_output output,
6df2e2c9
MD
484 const char *path_name,
485 uid_t uid,
00e2e675
DG
486 gid_t gid,
487 int net_index,
c80048c6 488 int metadata_flag,
53632229 489 uint64_t session_id,
c80048c6 490 int *alloc_ret)
3bd1e081
MD
491{
492 struct lttng_consumer_stream *stream;
3bd1e081 493
effcf122 494 stream = zmalloc(sizeof(*stream));
3bd1e081 495 if (stream == NULL) {
7a57cf92 496 PERROR("malloc struct lttng_consumer_stream");
c80048c6 497 *alloc_ret = -ENOMEM;
7a57cf92 498 goto end;
3bd1e081 499 }
7a57cf92 500
d56db448
DG
501 rcu_read_lock();
502
7a57cf92
DG
503 /*
504 * Get stream's channel reference. Needed when adding the stream to the
505 * global hash table.
506 */
3bd1e081
MD
507 stream->chan = consumer_find_channel(channel_key);
508 if (!stream->chan) {
c80048c6 509 *alloc_ret = -ENOENT;
7a57cf92 510 ERR("Unable to find channel for stream %d", stream_key);
c80048c6 511 goto error;
3bd1e081 512 }
e316aad5 513
3bd1e081
MD
514 stream->key = stream_key;
515 stream->shm_fd = shm_fd;
516 stream->wait_fd = wait_fd;
517 stream->out_fd = -1;
518 stream->out_fd_offset = 0;
519 stream->state = state;
520 stream->mmap_len = mmap_len;
521 stream->mmap_base = NULL;
522 stream->output = output;
6df2e2c9
MD
523 stream->uid = uid;
524 stream->gid = gid;
00e2e675
DG
525 stream->net_seq_idx = net_index;
526 stream->metadata_flag = metadata_flag;
53632229 527 stream->session_id = session_id;
00e2e675
DG
528 strncpy(stream->path_name, path_name, sizeof(stream->path_name));
529 stream->path_name[sizeof(stream->path_name) - 1] = '\0';
53632229 530 pthread_mutex_init(&stream->lock, NULL);
58b1f425
DG
531
532 /*
533 * Index differently the metadata node because the thread is using an
534 * internal hash table to match streams in the metadata_ht to the epoll set
535 * file descriptor.
536 */
537 if (metadata_flag) {
538 lttng_ht_node_init_ulong(&stream->node, stream->wait_fd);
539 } else {
540 lttng_ht_node_init_ulong(&stream->node, stream->key);
541 }
c30aaa51 542
53632229
DG
543 /* Init session id node with the stream session id */
544 lttng_ht_node_init_ulong(&stream->node_session_id, stream->session_id);
545
c869f647
DG
546 /*
547 * The cpu number is needed before using any ustctl_* actions. Ignored for
548 * the kernel so the value does not matter.
549 */
550 pthread_mutex_lock(&consumer_data.lock);
551 stream->cpu = stream->chan->cpucount++;
552 pthread_mutex_unlock(&consumer_data.lock);
553
c30aaa51 554 DBG3("Allocated stream %s (key %d, shm_fd %d, wait_fd %d, mmap_len %llu,"
53632229
DG
555 " out_fd %d, net_seq_idx %d, session_id %" PRIu64,
556 stream->path_name, stream->key, stream->shm_fd, stream->wait_fd,
c30aaa51 557 (unsigned long long) stream->mmap_len, stream->out_fd,
53632229 558 stream->net_seq_idx, stream->session_id);
d56db448
DG
559
560 rcu_read_unlock();
3bd1e081 561 return stream;
c80048c6
MD
562
563error:
d56db448 564 rcu_read_unlock();
c80048c6 565 free(stream);
7a57cf92 566end:
c80048c6 567 return NULL;
3bd1e081
MD
568}
569
570/*
571 * Add a stream to the global list protected by a mutex.
572 */
43c34bc3
DG
573static int consumer_add_stream(struct lttng_consumer_stream *stream,
574 struct lttng_ht *ht)
3bd1e081
MD
575{
576 int ret = 0;
00e2e675 577 struct consumer_relayd_sock_pair *relayd;
3bd1e081 578
e316aad5 579 assert(stream);
43c34bc3 580 assert(ht);
c77fc10a 581
e316aad5
DG
582 DBG3("Adding consumer stream %d", stream->key);
583
584 pthread_mutex_lock(&consumer_data.lock);
2e818a6a 585 pthread_mutex_lock(&stream->lock);
b0b335c8 586 rcu_read_lock();
e316aad5 587
43c34bc3
DG
588 /* Steal stream identifier to avoid having streams with the same key */
589 consumer_steal_stream_key(stream->key, ht);
590
591 lttng_ht_add_unique_ulong(ht, &stream->node);
00e2e675 592
ca22feea
DG
593 /*
594 * Add stream to the stream_list_ht of the consumer data. No need to steal
595 * the key since the HT does not use it and we allow to add redundant keys
596 * into this table.
597 */
598 lttng_ht_add_ulong(consumer_data.stream_list_ht, &stream->node_session_id);
599
00e2e675
DG
600 /* Check and cleanup relayd */
601 relayd = consumer_find_relayd(stream->net_seq_idx);
602 if (relayd != NULL) {
b0b335c8 603 uatomic_inc(&relayd->refcount);
00e2e675
DG
604 }
605
e316aad5
DG
606 /* Update channel refcount once added without error(s). */
607 uatomic_inc(&stream->chan->refcount);
608
609 /*
610 * When nb_init_streams reaches 0, we don't need to trigger any action in
611 * terms of destroying the associated channel, because the action that
612 * causes the count to become 0 also causes a stream to be added. The
613 * channel deletion will thus be triggered by the following removal of this
614 * stream.
615 */
616 if (uatomic_read(&stream->chan->nb_init_streams) > 0) {
617 uatomic_dec(&stream->chan->nb_init_streams);
618 }
619
620 /* Update consumer data once the node is inserted. */
3bd1e081
MD
621 consumer_data.stream_count++;
622 consumer_data.need_update = 1;
623
e316aad5 624 rcu_read_unlock();
2e818a6a 625 pthread_mutex_unlock(&stream->lock);
3bd1e081 626 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 627
3bd1e081
MD
628 return ret;
629}
630
00e2e675 631/*
3f8e211f
DG
632 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
633 * be acquired before calling this.
00e2e675 634 */
d09e1200 635static int add_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
636{
637 int ret = 0;
638 struct lttng_ht_node_ulong *node;
639 struct lttng_ht_iter iter;
640
641 if (relayd == NULL) {
642 ret = -1;
643 goto end;
644 }
645
00e2e675
DG
646 lttng_ht_lookup(consumer_data.relayd_ht,
647 (void *)((unsigned long) relayd->net_seq_idx), &iter);
648 node = lttng_ht_iter_get_node_ulong(&iter);
649 if (node != NULL) {
00e2e675
DG
650 /* Relayd already exist. Ignore the insertion */
651 goto end;
652 }
653 lttng_ht_add_unique_ulong(consumer_data.relayd_ht, &relayd->node);
654
00e2e675
DG
655end:
656 return ret;
657}
658
659/*
660 * Allocate and return a consumer relayd socket.
661 */
662struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair(
663 int net_seq_idx)
664{
665 struct consumer_relayd_sock_pair *obj = NULL;
666
667 /* Negative net sequence index is a failure */
668 if (net_seq_idx < 0) {
669 goto error;
670 }
671
672 obj = zmalloc(sizeof(struct consumer_relayd_sock_pair));
673 if (obj == NULL) {
674 PERROR("zmalloc relayd sock");
675 goto error;
676 }
677
678 obj->net_seq_idx = net_seq_idx;
679 obj->refcount = 0;
173af62f 680 obj->destroy_flag = 0;
00e2e675
DG
681 lttng_ht_node_init_ulong(&obj->node, obj->net_seq_idx);
682 pthread_mutex_init(&obj->ctrl_sock_mutex, NULL);
683
684error:
685 return obj;
686}
687
688/*
689 * Find a relayd socket pair in the global consumer data.
690 *
691 * Return the object if found else NULL.
b0b335c8
MD
692 * RCU read-side lock must be held across this call and while using the
693 * returned object.
00e2e675
DG
694 */
695struct consumer_relayd_sock_pair *consumer_find_relayd(int key)
696{
697 struct lttng_ht_iter iter;
698 struct lttng_ht_node_ulong *node;
699 struct consumer_relayd_sock_pair *relayd = NULL;
700
701 /* Negative keys are lookup failures */
702 if (key < 0) {
703 goto error;
704 }
705
00e2e675
DG
706 lttng_ht_lookup(consumer_data.relayd_ht, (void *)((unsigned long) key),
707 &iter);
708 node = lttng_ht_iter_get_node_ulong(&iter);
709 if (node != NULL) {
710 relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node);
711 }
712
00e2e675
DG
713error:
714 return relayd;
715}
716
717/*
718 * Handle stream for relayd transmission if the stream applies for network
719 * streaming where the net sequence index is set.
720 *
721 * Return destination file descriptor or negative value on error.
722 */
6197aea7 723static int write_relayd_stream_header(struct lttng_consumer_stream *stream,
1d4dfdef
DG
724 size_t data_size, unsigned long padding,
725 struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
726{
727 int outfd = -1, ret;
00e2e675
DG
728 struct lttcomm_relayd_data_hdr data_hdr;
729
730 /* Safety net */
731 assert(stream);
6197aea7 732 assert(relayd);
00e2e675
DG
733
734 /* Reset data header */
735 memset(&data_hdr, 0, sizeof(data_hdr));
736
00e2e675
DG
737 if (stream->metadata_flag) {
738 /* Caller MUST acquire the relayd control socket lock */
739 ret = relayd_send_metadata(&relayd->control_sock, data_size);
740 if (ret < 0) {
741 goto error;
742 }
743
744 /* Metadata are always sent on the control socket. */
745 outfd = relayd->control_sock.fd;
746 } else {
747 /* Set header with stream information */
748 data_hdr.stream_id = htobe64(stream->relayd_stream_id);
749 data_hdr.data_size = htobe32(data_size);
1d4dfdef 750 data_hdr.padding_size = htobe32(padding);
39df6d9f
DG
751 /*
752 * Note that net_seq_num below is assigned with the *current* value of
753 * next_net_seq_num and only after that the next_net_seq_num will be
754 * increment. This is why when issuing a command on the relayd using
755 * this next value, 1 should always be substracted in order to compare
756 * the last seen sequence number on the relayd side to the last sent.
757 */
3604f373 758 data_hdr.net_seq_num = htobe64(stream->next_net_seq_num);
00e2e675
DG
759 /* Other fields are zeroed previously */
760
761 ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr,
762 sizeof(data_hdr));
763 if (ret < 0) {
764 goto error;
765 }
766
3604f373
DG
767 ++stream->next_net_seq_num;
768
00e2e675
DG
769 /* Set to go on data socket */
770 outfd = relayd->data_sock.fd;
771 }
772
773error:
774 return outfd;
775}
776
702b1ea4
MD
777static
778void consumer_free_channel(struct rcu_head *head)
779{
780 struct lttng_ht_node_ulong *node =
781 caa_container_of(head, struct lttng_ht_node_ulong, head);
782 struct lttng_consumer_channel *channel =
783 caa_container_of(node, struct lttng_consumer_channel, node);
784
785 free(channel);
786}
787
3bd1e081
MD
788/*
789 * Remove a channel from the global list protected by a mutex. This
790 * function is also responsible for freeing its data structures.
791 */
792void consumer_del_channel(struct lttng_consumer_channel *channel)
793{
794 int ret;
e4421fec 795 struct lttng_ht_iter iter;
3bd1e081 796
5c540210
DG
797 DBG("Consumer delete channel key %d", channel->key);
798
3bd1e081
MD
799 pthread_mutex_lock(&consumer_data.lock);
800
801 switch (consumer_data.type) {
802 case LTTNG_CONSUMER_KERNEL:
803 break;
7753dea8
MD
804 case LTTNG_CONSUMER32_UST:
805 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
806 lttng_ustconsumer_del_channel(channel);
807 break;
808 default:
809 ERR("Unknown consumer_data type");
810 assert(0);
811 goto end;
812 }
813
6065ceec 814 rcu_read_lock();
04253271
MD
815 iter.iter.node = &channel->node.node;
816 ret = lttng_ht_del(consumer_data.channel_ht, &iter);
817 assert(!ret);
6065ceec
DG
818 rcu_read_unlock();
819
3bd1e081
MD
820 if (channel->mmap_base != NULL) {
821 ret = munmap(channel->mmap_base, channel->mmap_len);
822 if (ret != 0) {
7a57cf92 823 PERROR("munmap");
3bd1e081
MD
824 }
825 }
b5c5fc29 826 if (channel->wait_fd >= 0 && !channel->wait_fd_is_copy) {
4c462e79
MD
827 ret = close(channel->wait_fd);
828 if (ret) {
829 PERROR("close");
830 }
3bd1e081 831 }
2c1dd183 832 if (channel->shm_fd >= 0 && channel->wait_fd != channel->shm_fd) {
4c462e79
MD
833 ret = close(channel->shm_fd);
834 if (ret) {
835 PERROR("close");
836 }
3bd1e081 837 }
702b1ea4
MD
838
839 call_rcu(&channel->node.head, consumer_free_channel);
3bd1e081
MD
840end:
841 pthread_mutex_unlock(&consumer_data.lock);
842}
843
844struct lttng_consumer_channel *consumer_allocate_channel(
845 int channel_key,
846 int shm_fd, int wait_fd,
847 uint64_t mmap_len,
c30aaa51
MD
848 uint64_t max_sb_size,
849 unsigned int nb_init_streams)
3bd1e081
MD
850{
851 struct lttng_consumer_channel *channel;
852 int ret;
853
276b26d1 854 channel = zmalloc(sizeof(*channel));
3bd1e081 855 if (channel == NULL) {
7a57cf92 856 PERROR("malloc struct lttng_consumer_channel");
3bd1e081
MD
857 goto end;
858 }
859 channel->key = channel_key;
860 channel->shm_fd = shm_fd;
861 channel->wait_fd = wait_fd;
862 channel->mmap_len = mmap_len;
863 channel->max_sb_size = max_sb_size;
864 channel->refcount = 0;
c30aaa51 865 channel->nb_init_streams = nb_init_streams;
e4421fec 866 lttng_ht_node_init_ulong(&channel->node, channel->key);
3bd1e081
MD
867
868 switch (consumer_data.type) {
869 case LTTNG_CONSUMER_KERNEL:
870 channel->mmap_base = NULL;
871 channel->mmap_len = 0;
872 break;
7753dea8
MD
873 case LTTNG_CONSUMER32_UST:
874 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
875 ret = lttng_ustconsumer_allocate_channel(channel);
876 if (ret) {
877 free(channel);
878 return NULL;
879 }
880 break;
881 default:
882 ERR("Unknown consumer_data type");
883 assert(0);
884 goto end;
885 }
886 DBG("Allocated channel (key %d, shm_fd %d, wait_fd %d, mmap_len %llu, max_sb_size %llu)",
00e2e675 887 channel->key, channel->shm_fd, channel->wait_fd,
3bd1e081
MD
888 (unsigned long long) channel->mmap_len,
889 (unsigned long long) channel->max_sb_size);
890end:
891 return channel;
892}
893
894/*
895 * Add a channel to the global list protected by a mutex.
896 */
897int consumer_add_channel(struct lttng_consumer_channel *channel)
898{
c77fc10a
DG
899 struct lttng_ht_node_ulong *node;
900 struct lttng_ht_iter iter;
901
3bd1e081 902 pthread_mutex_lock(&consumer_data.lock);
7ad0a0cb
MD
903 /* Steal channel identifier, for UST */
904 consumer_steal_channel_key(channel->key);
6065ceec 905 rcu_read_lock();
c77fc10a
DG
906
907 lttng_ht_lookup(consumer_data.channel_ht,
908 (void *)((unsigned long) channel->key), &iter);
909 node = lttng_ht_iter_get_node_ulong(&iter);
910 if (node != NULL) {
911 /* Channel already exist. Ignore the insertion */
912 goto end;
913 }
914
04253271 915 lttng_ht_add_unique_ulong(consumer_data.channel_ht, &channel->node);
c77fc10a
DG
916
917end:
6065ceec 918 rcu_read_unlock();
3bd1e081 919 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 920
7ad0a0cb 921 return 0;
3bd1e081
MD
922}
923
924/*
925 * Allocate the pollfd structure and the local view of the out fds to avoid
926 * doing a lookup in the linked list and concurrency issues when writing is
927 * needed. Called with consumer_data.lock held.
928 *
929 * Returns the number of fds in the structures.
930 */
43c34bc3 931static int consumer_update_poll_array(
3bd1e081 932 struct lttng_consumer_local_data *ctx, struct pollfd **pollfd,
43c34bc3 933 struct lttng_consumer_stream **local_stream, struct lttng_ht *ht)
3bd1e081 934{
3bd1e081 935 int i = 0;
e4421fec
DG
936 struct lttng_ht_iter iter;
937 struct lttng_consumer_stream *stream;
3bd1e081
MD
938
939 DBG("Updating poll fd array");
481d6c57 940 rcu_read_lock();
43c34bc3 941 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
8994307f
DG
942 /*
943 * Only active streams with an active end point can be added to the
944 * poll set and local stream storage of the thread.
945 *
946 * There is a potential race here for endpoint_status to be updated
947 * just after the check. However, this is OK since the stream(s) will
948 * be deleted once the thread is notified that the end point state has
949 * changed where this function will be called back again.
950 */
951 if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM ||
79d4ffb7 952 stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) {
3bd1e081
MD
953 continue;
954 }
e4421fec
DG
955 DBG("Active FD %d", stream->wait_fd);
956 (*pollfd)[i].fd = stream->wait_fd;
3bd1e081 957 (*pollfd)[i].events = POLLIN | POLLPRI;
e4421fec 958 local_stream[i] = stream;
3bd1e081
MD
959 i++;
960 }
481d6c57 961 rcu_read_unlock();
3bd1e081
MD
962
963 /*
50f8ae69 964 * Insert the consumer_data_pipe at the end of the array and don't
3bd1e081
MD
965 * increment i so nb_fd is the number of real FD.
966 */
50f8ae69 967 (*pollfd)[i].fd = ctx->consumer_data_pipe[0];
509bb1cf 968 (*pollfd)[i].events = POLLIN | POLLPRI;
3bd1e081
MD
969 return i;
970}
971
972/*
973 * Poll on the should_quit pipe and the command socket return -1 on error and
974 * should exit, 0 if data is available on the command socket
975 */
976int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll)
977{
978 int num_rdy;
979
88f2b785 980restart:
3bd1e081
MD
981 num_rdy = poll(consumer_sockpoll, 2, -1);
982 if (num_rdy == -1) {
88f2b785
MD
983 /*
984 * Restart interrupted system call.
985 */
986 if (errno == EINTR) {
987 goto restart;
988 }
7a57cf92 989 PERROR("Poll error");
3bd1e081
MD
990 goto exit;
991 }
509bb1cf 992 if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) {
3bd1e081
MD
993 DBG("consumer_should_quit wake up");
994 goto exit;
995 }
996 return 0;
997
998exit:
999 return -1;
1000}
1001
1002/*
1003 * Set the error socket.
1004 */
1005void lttng_consumer_set_error_sock(
1006 struct lttng_consumer_local_data *ctx, int sock)
1007{
1008 ctx->consumer_error_socket = sock;
1009}
1010
1011/*
1012 * Set the command socket path.
1013 */
3bd1e081
MD
1014void lttng_consumer_set_command_sock_path(
1015 struct lttng_consumer_local_data *ctx, char *sock)
1016{
1017 ctx->consumer_command_sock_path = sock;
1018}
1019
1020/*
1021 * Send return code to the session daemon.
1022 * If the socket is not defined, we return 0, it is not a fatal error
1023 */
1024int lttng_consumer_send_error(
1025 struct lttng_consumer_local_data *ctx, int cmd)
1026{
1027 if (ctx->consumer_error_socket > 0) {
1028 return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd,
1029 sizeof(enum lttcomm_sessiond_command));
1030 }
1031
1032 return 0;
1033}
1034
1035/*
228b5bf7
DG
1036 * Close all the tracefiles and stream fds and MUST be called when all
1037 * instances are destroyed i.e. when all threads were joined and are ended.
3bd1e081
MD
1038 */
1039void lttng_consumer_cleanup(void)
1040{
e4421fec 1041 struct lttng_ht_iter iter;
6065ceec
DG
1042 struct lttng_ht_node_ulong *node;
1043
1044 rcu_read_lock();
3bd1e081 1045
6065ceec
DG
1046 cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, node,
1047 node) {
702b1ea4
MD
1048 struct lttng_consumer_channel *channel =
1049 caa_container_of(node, struct lttng_consumer_channel, node);
1050 consumer_del_channel(channel);
3bd1e081 1051 }
6065ceec
DG
1052
1053 rcu_read_unlock();
d6ce1df2 1054
d6ce1df2 1055 lttng_ht_destroy(consumer_data.channel_ht);
228b5bf7
DG
1056
1057 cleanup_relayd_ht();
1058
1059 /*
1060 * This HT contains streams that are freed by either the metadata thread or
1061 * the data thread so we do *nothing* on the hash table and simply destroy
1062 * it.
1063 */
1064 lttng_ht_destroy(consumer_data.stream_list_ht);
3bd1e081
MD
1065}
1066
1067/*
1068 * Called from signal handler.
1069 */
1070void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx)
1071{
1072 int ret;
1073 consumer_quit = 1;
6f94560a
MD
1074 do {
1075 ret = write(ctx->consumer_should_quit[1], "4", 1);
1076 } while (ret < 0 && errno == EINTR);
4cec016f 1077 if (ret < 0 || ret != 1) {
7a57cf92 1078 PERROR("write consumer quit");
3bd1e081 1079 }
ab1027f4
DG
1080
1081 DBG("Consumer flag that it should quit");
3bd1e081
MD
1082}
1083
00e2e675
DG
1084void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream,
1085 off_t orig_offset)
3bd1e081
MD
1086{
1087 int outfd = stream->out_fd;
1088
1089 /*
1090 * This does a blocking write-and-wait on any page that belongs to the
1091 * subbuffer prior to the one we just wrote.
1092 * Don't care about error values, as these are just hints and ways to
1093 * limit the amount of page cache used.
1094 */
1095 if (orig_offset < stream->chan->max_sb_size) {
1096 return;
1097 }
b9182dd9 1098 lttng_sync_file_range(outfd, orig_offset - stream->chan->max_sb_size,
3bd1e081
MD
1099 stream->chan->max_sb_size,
1100 SYNC_FILE_RANGE_WAIT_BEFORE
1101 | SYNC_FILE_RANGE_WRITE
1102 | SYNC_FILE_RANGE_WAIT_AFTER);
1103 /*
1104 * Give hints to the kernel about how we access the file:
1105 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1106 * we write it.
1107 *
1108 * We need to call fadvise again after the file grows because the
1109 * kernel does not seem to apply fadvise to non-existing parts of the
1110 * file.
1111 *
1112 * Call fadvise _after_ having waited for the page writeback to
1113 * complete because the dirty page writeback semantic is not well
1114 * defined. So it can be expected to lead to lower throughput in
1115 * streaming.
1116 */
1117 posix_fadvise(outfd, orig_offset - stream->chan->max_sb_size,
1118 stream->chan->max_sb_size, POSIX_FADV_DONTNEED);
1119}
1120
1121/*
1122 * Initialise the necessary environnement :
1123 * - create a new context
1124 * - create the poll_pipe
1125 * - create the should_quit pipe (for signal handler)
1126 * - create the thread pipe (for splice)
1127 *
1128 * Takes a function pointer as argument, this function is called when data is
1129 * available on a buffer. This function is responsible to do the
1130 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1131 * buffer configuration and then kernctl_put_next_subbuf at the end.
1132 *
1133 * Returns a pointer to the new context or NULL on error.
1134 */
1135struct lttng_consumer_local_data *lttng_consumer_create(
1136 enum lttng_consumer_type type,
4078b776 1137 ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream,
d41f73b7 1138 struct lttng_consumer_local_data *ctx),
3bd1e081
MD
1139 int (*recv_channel)(struct lttng_consumer_channel *channel),
1140 int (*recv_stream)(struct lttng_consumer_stream *stream),
1141 int (*update_stream)(int stream_key, uint32_t state))
1142{
1143 int ret, i;
1144 struct lttng_consumer_local_data *ctx;
1145
1146 assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN ||
1147 consumer_data.type == type);
1148 consumer_data.type = type;
1149
effcf122 1150 ctx = zmalloc(sizeof(struct lttng_consumer_local_data));
3bd1e081 1151 if (ctx == NULL) {
7a57cf92 1152 PERROR("allocating context");
3bd1e081
MD
1153 goto error;
1154 }
1155
1156 ctx->consumer_error_socket = -1;
1157 /* assign the callbacks */
1158 ctx->on_buffer_ready = buffer_ready;
1159 ctx->on_recv_channel = recv_channel;
1160 ctx->on_recv_stream = recv_stream;
1161 ctx->on_update_stream = update_stream;
1162
50f8ae69 1163 ret = pipe(ctx->consumer_data_pipe);
3bd1e081 1164 if (ret < 0) {
7a57cf92 1165 PERROR("Error creating poll pipe");
3bd1e081
MD
1166 goto error_poll_pipe;
1167 }
1168
04fdd819 1169 /* set read end of the pipe to non-blocking */
50f8ae69 1170 ret = fcntl(ctx->consumer_data_pipe[0], F_SETFL, O_NONBLOCK);
04fdd819 1171 if (ret < 0) {
7a57cf92 1172 PERROR("fcntl O_NONBLOCK");
04fdd819
MD
1173 goto error_poll_fcntl;
1174 }
1175
1176 /* set write end of the pipe to non-blocking */
50f8ae69 1177 ret = fcntl(ctx->consumer_data_pipe[1], F_SETFL, O_NONBLOCK);
04fdd819 1178 if (ret < 0) {
7a57cf92 1179 PERROR("fcntl O_NONBLOCK");
04fdd819
MD
1180 goto error_poll_fcntl;
1181 }
1182
3bd1e081
MD
1183 ret = pipe(ctx->consumer_should_quit);
1184 if (ret < 0) {
7a57cf92 1185 PERROR("Error creating recv pipe");
3bd1e081
MD
1186 goto error_quit_pipe;
1187 }
1188
1189 ret = pipe(ctx->consumer_thread_pipe);
1190 if (ret < 0) {
7a57cf92 1191 PERROR("Error creating thread pipe");
3bd1e081
MD
1192 goto error_thread_pipe;
1193 }
1194
fb3a43a9
DG
1195 ret = utils_create_pipe(ctx->consumer_metadata_pipe);
1196 if (ret < 0) {
1197 goto error_metadata_pipe;
1198 }
3bd1e081 1199
fb3a43a9
DG
1200 ret = utils_create_pipe(ctx->consumer_splice_metadata_pipe);
1201 if (ret < 0) {
1202 goto error_splice_pipe;
1203 }
1204
1205 return ctx;
3bd1e081 1206
fb3a43a9
DG
1207error_splice_pipe:
1208 utils_close_pipe(ctx->consumer_metadata_pipe);
1209error_metadata_pipe:
1210 utils_close_pipe(ctx->consumer_thread_pipe);
3bd1e081
MD
1211error_thread_pipe:
1212 for (i = 0; i < 2; i++) {
1213 int err;
1214
1215 err = close(ctx->consumer_should_quit[i]);
4c462e79
MD
1216 if (err) {
1217 PERROR("close");
1218 }
3bd1e081 1219 }
04fdd819 1220error_poll_fcntl:
3bd1e081
MD
1221error_quit_pipe:
1222 for (i = 0; i < 2; i++) {
1223 int err;
1224
50f8ae69 1225 err = close(ctx->consumer_data_pipe[i]);
4c462e79
MD
1226 if (err) {
1227 PERROR("close");
1228 }
3bd1e081
MD
1229 }
1230error_poll_pipe:
1231 free(ctx);
1232error:
1233 return NULL;
1234}
1235
1236/*
1237 * Close all fds associated with the instance and free the context.
1238 */
1239void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx)
1240{
4c462e79
MD
1241 int ret;
1242
ab1027f4
DG
1243 DBG("Consumer destroying it. Closing everything.");
1244
4c462e79
MD
1245 ret = close(ctx->consumer_error_socket);
1246 if (ret) {
1247 PERROR("close");
1248 }
1249 ret = close(ctx->consumer_thread_pipe[0]);
1250 if (ret) {
1251 PERROR("close");
1252 }
1253 ret = close(ctx->consumer_thread_pipe[1]);
1254 if (ret) {
1255 PERROR("close");
1256 }
50f8ae69 1257 ret = close(ctx->consumer_data_pipe[0]);
4c462e79
MD
1258 if (ret) {
1259 PERROR("close");
1260 }
50f8ae69 1261 ret = close(ctx->consumer_data_pipe[1]);
4c462e79
MD
1262 if (ret) {
1263 PERROR("close");
1264 }
1265 ret = close(ctx->consumer_should_quit[0]);
1266 if (ret) {
1267 PERROR("close");
1268 }
1269 ret = close(ctx->consumer_should_quit[1]);
1270 if (ret) {
1271 PERROR("close");
1272 }
fb3a43a9
DG
1273 utils_close_pipe(ctx->consumer_splice_metadata_pipe);
1274
3bd1e081
MD
1275 unlink(ctx->consumer_command_sock_path);
1276 free(ctx);
1277}
1278
6197aea7
DG
1279/*
1280 * Write the metadata stream id on the specified file descriptor.
1281 */
1282static int write_relayd_metadata_id(int fd,
1283 struct lttng_consumer_stream *stream,
1d4dfdef
DG
1284 struct consumer_relayd_sock_pair *relayd,
1285 unsigned long padding)
6197aea7
DG
1286{
1287 int ret;
1d4dfdef 1288 struct lttcomm_relayd_metadata_payload hdr;
6197aea7 1289
1d4dfdef
DG
1290 hdr.stream_id = htobe64(stream->relayd_stream_id);
1291 hdr.padding_size = htobe32(padding);
6197aea7 1292 do {
1d4dfdef 1293 ret = write(fd, (void *) &hdr, sizeof(hdr));
6197aea7 1294 } while (ret < 0 && errno == EINTR);
4cec016f 1295 if (ret < 0 || ret != sizeof(hdr)) {
d7b75ec8
DG
1296 /*
1297 * This error means that the fd's end is closed so ignore the perror
1298 * not to clubber the error output since this can happen in a normal
1299 * code path.
1300 */
1301 if (errno != EPIPE) {
1302 PERROR("write metadata stream id");
1303 }
1304 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno);
534d2592
DG
1305 /*
1306 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1307 * handle writting the missing part so report that as an error and
1308 * don't lie to the caller.
1309 */
1310 ret = -1;
6197aea7
DG
1311 goto end;
1312 }
1d4dfdef
DG
1313 DBG("Metadata stream id %" PRIu64 " with padding %lu written before data",
1314 stream->relayd_stream_id, padding);
6197aea7
DG
1315
1316end:
1317 return ret;
1318}
1319
3bd1e081 1320/*
09e26845
DG
1321 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1322 * core function for writing trace buffers to either the local filesystem or
1323 * the network.
1324 *
79d4ffb7
DG
1325 * It must be called with the stream lock held.
1326 *
09e26845 1327 * Careful review MUST be put if any changes occur!
3bd1e081
MD
1328 *
1329 * Returns the number of bytes written
1330 */
4078b776 1331ssize_t lttng_consumer_on_read_subbuffer_mmap(
3bd1e081 1332 struct lttng_consumer_local_data *ctx,
1d4dfdef
DG
1333 struct lttng_consumer_stream *stream, unsigned long len,
1334 unsigned long padding)
3bd1e081 1335{
f02e1e8a
DG
1336 unsigned long mmap_offset;
1337 ssize_t ret = 0, written = 0;
1338 off_t orig_offset = stream->out_fd_offset;
1339 /* Default is on the disk */
1340 int outfd = stream->out_fd;
f02e1e8a 1341 struct consumer_relayd_sock_pair *relayd = NULL;
8994307f 1342 unsigned int relayd_hang_up = 0;
f02e1e8a
DG
1343
1344 /* RCU lock for the relayd pointer */
1345 rcu_read_lock();
1346
1347 /* Flag that the current stream if set for network streaming. */
1348 if (stream->net_seq_idx != -1) {
1349 relayd = consumer_find_relayd(stream->net_seq_idx);
1350 if (relayd == NULL) {
1351 goto end;
1352 }
1353 }
1354
1355 /* get the offset inside the fd to mmap */
3bd1e081
MD
1356 switch (consumer_data.type) {
1357 case LTTNG_CONSUMER_KERNEL:
f02e1e8a
DG
1358 ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset);
1359 break;
7753dea8
MD
1360 case LTTNG_CONSUMER32_UST:
1361 case LTTNG_CONSUMER64_UST:
f02e1e8a
DG
1362 ret = lttng_ustctl_get_mmap_read_offset(stream->chan->handle,
1363 stream->buf, &mmap_offset);
1364 break;
3bd1e081
MD
1365 default:
1366 ERR("Unknown consumer_data type");
1367 assert(0);
1368 }
f02e1e8a
DG
1369 if (ret != 0) {
1370 errno = -ret;
1371 PERROR("tracer ctl get_mmap_read_offset");
1372 written = ret;
1373 goto end;
1374 }
b9182dd9 1375
f02e1e8a
DG
1376 /* Handle stream on the relayd if the output is on the network */
1377 if (relayd) {
1378 unsigned long netlen = len;
1379
1380 /*
1381 * Lock the control socket for the complete duration of the function
1382 * since from this point on we will use the socket.
1383 */
1384 if (stream->metadata_flag) {
1385 /* Metadata requires the control socket. */
1386 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1d4dfdef 1387 netlen += sizeof(struct lttcomm_relayd_metadata_payload);
f02e1e8a
DG
1388 }
1389
1d4dfdef 1390 ret = write_relayd_stream_header(stream, netlen, padding, relayd);
f02e1e8a
DG
1391 if (ret >= 0) {
1392 /* Use the returned socket. */
1393 outfd = ret;
1394
1395 /* Write metadata stream id before payload */
1396 if (stream->metadata_flag) {
1d4dfdef 1397 ret = write_relayd_metadata_id(outfd, stream, relayd, padding);
f02e1e8a 1398 if (ret < 0) {
f02e1e8a 1399 written = ret;
8994307f
DG
1400 /* Socket operation failed. We consider the relayd dead */
1401 if (ret == -EPIPE || ret == -EINVAL) {
1402 relayd_hang_up = 1;
1403 goto write_error;
1404 }
f02e1e8a
DG
1405 goto end;
1406 }
f02e1e8a 1407 }
8994307f
DG
1408 } else {
1409 /* Socket operation failed. We consider the relayd dead */
1410 if (ret == -EPIPE || ret == -EINVAL) {
1411 relayd_hang_up = 1;
1412 goto write_error;
1413 }
1414 /* Else, use the default set before which is the filesystem. */
f02e1e8a 1415 }
1d4dfdef
DG
1416 } else {
1417 /* No streaming, we have to set the len with the full padding */
1418 len += padding;
f02e1e8a
DG
1419 }
1420
1421 while (len > 0) {
1422 do {
1423 ret = write(outfd, stream->mmap_base + mmap_offset, len);
1424 } while (ret < 0 && errno == EINTR);
1d4dfdef 1425 DBG("Consumer mmap write() ret %zd (len %lu)", ret, len);
f02e1e8a 1426 if (ret < 0) {
c5c45efa
DG
1427 /*
1428 * This is possible if the fd is closed on the other side (outfd)
1429 * or any write problem. It can be verbose a bit for a normal
1430 * execution if for instance the relayd is stopped abruptly. This
1431 * can happen so set this to a DBG statement.
1432 */
1433 DBG("Error in file write mmap");
f02e1e8a
DG
1434 if (written == 0) {
1435 written = ret;
1436 }
8994307f
DG
1437 /* Socket operation failed. We consider the relayd dead */
1438 if (errno == EPIPE || errno == EINVAL) {
1439 relayd_hang_up = 1;
1440 goto write_error;
1441 }
f02e1e8a
DG
1442 goto end;
1443 } else if (ret > len) {
77c7c900 1444 PERROR("Error in file write (ret %zd > len %lu)", ret, len);
f02e1e8a
DG
1445 written += ret;
1446 goto end;
1447 } else {
1448 len -= ret;
1449 mmap_offset += ret;
1450 }
f02e1e8a
DG
1451
1452 /* This call is useless on a socket so better save a syscall. */
1453 if (!relayd) {
1454 /* This won't block, but will start writeout asynchronously */
1455 lttng_sync_file_range(outfd, stream->out_fd_offset, ret,
1456 SYNC_FILE_RANGE_WRITE);
1457 stream->out_fd_offset += ret;
1458 }
1459 written += ret;
1460 }
1461 lttng_consumer_sync_trace_file(stream, orig_offset);
1462
8994307f
DG
1463write_error:
1464 /*
1465 * This is a special case that the relayd has closed its socket. Let's
1466 * cleanup the relayd object and all associated streams.
1467 */
1468 if (relayd && relayd_hang_up) {
1469 cleanup_relayd(relayd, ctx);
1470 }
1471
f02e1e8a
DG
1472end:
1473 /* Unlock only if ctrl socket used */
1474 if (relayd && stream->metadata_flag) {
1475 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1476 }
1477
1478 rcu_read_unlock();
1479 return written;
3bd1e081
MD
1480}
1481
1482/*
1483 * Splice the data from the ring buffer to the tracefile.
1484 *
79d4ffb7
DG
1485 * It must be called with the stream lock held.
1486 *
3bd1e081
MD
1487 * Returns the number of bytes spliced.
1488 */
4078b776 1489ssize_t lttng_consumer_on_read_subbuffer_splice(
3bd1e081 1490 struct lttng_consumer_local_data *ctx,
1d4dfdef
DG
1491 struct lttng_consumer_stream *stream, unsigned long len,
1492 unsigned long padding)
3bd1e081 1493{
f02e1e8a
DG
1494 ssize_t ret = 0, written = 0, ret_splice = 0;
1495 loff_t offset = 0;
1496 off_t orig_offset = stream->out_fd_offset;
1497 int fd = stream->wait_fd;
1498 /* Default is on the disk */
1499 int outfd = stream->out_fd;
f02e1e8a 1500 struct consumer_relayd_sock_pair *relayd = NULL;
fb3a43a9 1501 int *splice_pipe;
8994307f 1502 unsigned int relayd_hang_up = 0;
f02e1e8a 1503
3bd1e081
MD
1504 switch (consumer_data.type) {
1505 case LTTNG_CONSUMER_KERNEL:
f02e1e8a 1506 break;
7753dea8
MD
1507 case LTTNG_CONSUMER32_UST:
1508 case LTTNG_CONSUMER64_UST:
f02e1e8a 1509 /* Not supported for user space tracing */
3bd1e081
MD
1510 return -ENOSYS;
1511 default:
1512 ERR("Unknown consumer_data type");
1513 assert(0);
3bd1e081
MD
1514 }
1515
f02e1e8a
DG
1516 /* RCU lock for the relayd pointer */
1517 rcu_read_lock();
1518
1519 /* Flag that the current stream if set for network streaming. */
1520 if (stream->net_seq_idx != -1) {
1521 relayd = consumer_find_relayd(stream->net_seq_idx);
1522 if (relayd == NULL) {
1523 goto end;
1524 }
1525 }
1526
fb3a43a9
DG
1527 /*
1528 * Choose right pipe for splice. Metadata and trace data are handled by
1529 * different threads hence the use of two pipes in order not to race or
1530 * corrupt the written data.
1531 */
1532 if (stream->metadata_flag) {
1533 splice_pipe = ctx->consumer_splice_metadata_pipe;
1534 } else {
1535 splice_pipe = ctx->consumer_thread_pipe;
1536 }
1537
f02e1e8a 1538 /* Write metadata stream id before payload */
1d4dfdef
DG
1539 if (relayd) {
1540 int total_len = len;
f02e1e8a 1541
1d4dfdef
DG
1542 if (stream->metadata_flag) {
1543 /*
1544 * Lock the control socket for the complete duration of the function
1545 * since from this point on we will use the socket.
1546 */
1547 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1548
1549 ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd,
1550 padding);
1551 if (ret < 0) {
1552 written = ret;
8994307f
DG
1553 /* Socket operation failed. We consider the relayd dead */
1554 if (ret == -EBADF) {
1555 WARN("Remote relayd disconnected. Stopping");
1556 relayd_hang_up = 1;
1557 goto write_error;
1558 }
1d4dfdef
DG
1559 goto end;
1560 }
1561
1562 total_len += sizeof(struct lttcomm_relayd_metadata_payload);
1563 }
1564
1565 ret = write_relayd_stream_header(stream, total_len, padding, relayd);
1566 if (ret >= 0) {
1567 /* Use the returned socket. */
1568 outfd = ret;
1569 } else {
8994307f
DG
1570 /* Socket operation failed. We consider the relayd dead */
1571 if (ret == -EBADF) {
1572 WARN("Remote relayd disconnected. Stopping");
1573 relayd_hang_up = 1;
1574 goto write_error;
1575 }
f02e1e8a
DG
1576 goto end;
1577 }
1d4dfdef
DG
1578 } else {
1579 /* No streaming, we have to set the len with the full padding */
1580 len += padding;
f02e1e8a
DG
1581 }
1582
1583 while (len > 0) {
1d4dfdef
DG
1584 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1585 (unsigned long)offset, len, fd, splice_pipe[1]);
fb3a43a9 1586 ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len,
f02e1e8a
DG
1587 SPLICE_F_MOVE | SPLICE_F_MORE);
1588 DBG("splice chan to pipe, ret %zd", ret_splice);
1589 if (ret_splice < 0) {
1590 PERROR("Error in relay splice");
1591 if (written == 0) {
1592 written = ret_splice;
1593 }
1594 ret = errno;
1595 goto splice_error;
1596 }
1597
1598 /* Handle stream on the relayd if the output is on the network */
1599 if (relayd) {
1600 if (stream->metadata_flag) {
1d4dfdef
DG
1601 size_t metadata_payload_size =
1602 sizeof(struct lttcomm_relayd_metadata_payload);
1603
f02e1e8a 1604 /* Update counter to fit the spliced data */
1d4dfdef
DG
1605 ret_splice += metadata_payload_size;
1606 len += metadata_payload_size;
f02e1e8a
DG
1607 /*
1608 * We do this so the return value can match the len passed as
1609 * argument to this function.
1610 */
1d4dfdef 1611 written -= metadata_payload_size;
f02e1e8a
DG
1612 }
1613 }
1614
1615 /* Splice data out */
fb3a43a9 1616 ret_splice = splice(splice_pipe[0], NULL, outfd, NULL,
f02e1e8a 1617 ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE);
1d4dfdef 1618 DBG("Consumer splice pipe to file, ret %zd", ret_splice);
f02e1e8a
DG
1619 if (ret_splice < 0) {
1620 PERROR("Error in file splice");
1621 if (written == 0) {
1622 written = ret_splice;
1623 }
8994307f 1624 /* Socket operation failed. We consider the relayd dead */
00c8752b 1625 if (errno == EBADF || errno == EPIPE) {
8994307f
DG
1626 WARN("Remote relayd disconnected. Stopping");
1627 relayd_hang_up = 1;
1628 goto write_error;
1629 }
f02e1e8a
DG
1630 ret = errno;
1631 goto splice_error;
1632 } else if (ret_splice > len) {
1633 errno = EINVAL;
1634 PERROR("Wrote more data than requested %zd (len: %lu)",
1635 ret_splice, len);
1636 written += ret_splice;
1637 ret = errno;
1638 goto splice_error;
1639 }
1640 len -= ret_splice;
1641
1642 /* This call is useless on a socket so better save a syscall. */
1643 if (!relayd) {
1644 /* This won't block, but will start writeout asynchronously */
1645 lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice,
1646 SYNC_FILE_RANGE_WRITE);
1647 stream->out_fd_offset += ret_splice;
1648 }
1649 written += ret_splice;
1650 }
1651 lttng_consumer_sync_trace_file(stream, orig_offset);
1652
1653 ret = ret_splice;
1654
1655 goto end;
1656
8994307f
DG
1657write_error:
1658 /*
1659 * This is a special case that the relayd has closed its socket. Let's
1660 * cleanup the relayd object and all associated streams.
1661 */
1662 if (relayd && relayd_hang_up) {
1663 cleanup_relayd(relayd, ctx);
1664 /* Skip splice error so the consumer does not fail */
1665 goto end;
1666 }
1667
f02e1e8a
DG
1668splice_error:
1669 /* send the appropriate error description to sessiond */
1670 switch (ret) {
f02e1e8a 1671 case EINVAL:
f73fabfd 1672 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL);
f02e1e8a
DG
1673 break;
1674 case ENOMEM:
f73fabfd 1675 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM);
f02e1e8a
DG
1676 break;
1677 case ESPIPE:
f73fabfd 1678 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE);
f02e1e8a
DG
1679 break;
1680 }
1681
1682end:
1683 if (relayd && stream->metadata_flag) {
1684 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1685 }
1686
1687 rcu_read_unlock();
1688 return written;
3bd1e081
MD
1689}
1690
1691/*
1692 * Take a snapshot for a specific fd
1693 *
1694 * Returns 0 on success, < 0 on error
1695 */
1696int lttng_consumer_take_snapshot(struct lttng_consumer_local_data *ctx,
1697 struct lttng_consumer_stream *stream)
1698{
1699 switch (consumer_data.type) {
1700 case LTTNG_CONSUMER_KERNEL:
1701 return lttng_kconsumer_take_snapshot(ctx, stream);
7753dea8
MD
1702 case LTTNG_CONSUMER32_UST:
1703 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
1704 return lttng_ustconsumer_take_snapshot(ctx, stream);
1705 default:
1706 ERR("Unknown consumer_data type");
1707 assert(0);
1708 return -ENOSYS;
1709 }
1710
1711}
1712
1713/*
1714 * Get the produced position
1715 *
1716 * Returns 0 on success, < 0 on error
1717 */
1718int lttng_consumer_get_produced_snapshot(
1719 struct lttng_consumer_local_data *ctx,
1720 struct lttng_consumer_stream *stream,
1721 unsigned long *pos)
1722{
1723 switch (consumer_data.type) {
1724 case LTTNG_CONSUMER_KERNEL:
1725 return lttng_kconsumer_get_produced_snapshot(ctx, stream, pos);
7753dea8
MD
1726 case LTTNG_CONSUMER32_UST:
1727 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
1728 return lttng_ustconsumer_get_produced_snapshot(ctx, stream, pos);
1729 default:
1730 ERR("Unknown consumer_data type");
1731 assert(0);
1732 return -ENOSYS;
1733 }
1734}
1735
1736int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1737 int sock, struct pollfd *consumer_sockpoll)
1738{
1739 switch (consumer_data.type) {
1740 case LTTNG_CONSUMER_KERNEL:
1741 return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
7753dea8
MD
1742 case LTTNG_CONSUMER32_UST:
1743 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
1744 return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
1745 default:
1746 ERR("Unknown consumer_data type");
1747 assert(0);
1748 return -ENOSYS;
1749 }
1750}
1751
43c34bc3
DG
1752/*
1753 * Iterate over all streams of the hashtable and free them properly.
1754 *
1755 * WARNING: *MUST* be used with data stream only.
1756 */
1757static void destroy_data_stream_ht(struct lttng_ht *ht)
1758{
43c34bc3
DG
1759 struct lttng_ht_iter iter;
1760 struct lttng_consumer_stream *stream;
1761
1762 if (ht == NULL) {
1763 return;
1764 }
1765
1766 rcu_read_lock();
1767 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
5c540210
DG
1768 /*
1769 * Ignore return value since we are currently cleaning up so any error
1770 * can't be handled.
1771 */
1772 (void) consumer_del_stream(stream, ht);
43c34bc3
DG
1773 }
1774 rcu_read_unlock();
1775
1776 lttng_ht_destroy(ht);
1777}
1778
fb3a43a9 1779/*
f724d81e 1780 * Iterate over all streams of the hashtable and free them properly.
e316aad5
DG
1781 *
1782 * XXX: Should not be only for metadata stream or else use an other name.
fb3a43a9
DG
1783 */
1784static void destroy_stream_ht(struct lttng_ht *ht)
1785{
fb3a43a9
DG
1786 struct lttng_ht_iter iter;
1787 struct lttng_consumer_stream *stream;
1788
1789 if (ht == NULL) {
1790 return;
1791 }
1792
d09e1200 1793 rcu_read_lock();
58b1f425 1794 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
5c540210
DG
1795 /*
1796 * Ignore return value since we are currently cleaning up so any error
1797 * can't be handled.
1798 */
1799 (void) consumer_del_metadata_stream(stream, ht);
fb3a43a9 1800 }
d09e1200 1801 rcu_read_unlock();
fb3a43a9
DG
1802
1803 lttng_ht_destroy(ht);
1804}
1805
1806/*
1807 * Clean up a metadata stream and free its memory.
1808 */
e316aad5
DG
1809void consumer_del_metadata_stream(struct lttng_consumer_stream *stream,
1810 struct lttng_ht *ht)
fb3a43a9
DG
1811{
1812 int ret;
e316aad5
DG
1813 struct lttng_ht_iter iter;
1814 struct lttng_consumer_channel *free_chan = NULL;
fb3a43a9
DG
1815 struct consumer_relayd_sock_pair *relayd;
1816
1817 assert(stream);
1818 /*
1819 * This call should NEVER receive regular stream. It must always be
1820 * metadata stream and this is crucial for data structure synchronization.
1821 */
1822 assert(stream->metadata_flag);
1823
e316aad5
DG
1824 DBG3("Consumer delete metadata stream %d", stream->wait_fd);
1825
1826 if (ht == NULL) {
1827 /* Means the stream was allocated but not successfully added */
1828 goto free_stream;
1829 }
1830
74251bb8 1831 pthread_mutex_lock(&consumer_data.lock);
8994307f
DG
1832 pthread_mutex_lock(&stream->lock);
1833
fb3a43a9
DG
1834 switch (consumer_data.type) {
1835 case LTTNG_CONSUMER_KERNEL:
1836 if (stream->mmap_base != NULL) {
1837 ret = munmap(stream->mmap_base, stream->mmap_len);
1838 if (ret != 0) {
1839 PERROR("munmap metadata stream");
1840 }
1841 }
1842 break;
1843 case LTTNG_CONSUMER32_UST:
1844 case LTTNG_CONSUMER64_UST:
1845 lttng_ustconsumer_del_stream(stream);
1846 break;
1847 default:
1848 ERR("Unknown consumer_data type");
1849 assert(0);
e316aad5 1850 goto end;
fb3a43a9 1851 }
fb3a43a9 1852
c869f647 1853 rcu_read_lock();
58b1f425 1854 iter.iter.node = &stream->node.node;
c869f647
DG
1855 ret = lttng_ht_del(ht, &iter);
1856 assert(!ret);
ca22feea
DG
1857
1858 /* Remove node session id from the consumer_data stream ht */
1859 iter.iter.node = &stream->node_session_id.node;
1860 ret = lttng_ht_del(consumer_data.stream_list_ht, &iter);
1861 assert(!ret);
c869f647
DG
1862 rcu_read_unlock();
1863
fb3a43a9
DG
1864 if (stream->out_fd >= 0) {
1865 ret = close(stream->out_fd);
1866 if (ret) {
1867 PERROR("close");
1868 }
1869 }
1870
1871 if (stream->wait_fd >= 0 && !stream->wait_fd_is_copy) {
1872 ret = close(stream->wait_fd);
1873 if (ret) {
1874 PERROR("close");
1875 }
1876 }
1877
1878 if (stream->shm_fd >= 0 && stream->wait_fd != stream->shm_fd) {
1879 ret = close(stream->shm_fd);
1880 if (ret) {
1881 PERROR("close");
1882 }
1883 }
1884
1885 /* Check and cleanup relayd */
1886 rcu_read_lock();
1887 relayd = consumer_find_relayd(stream->net_seq_idx);
1888 if (relayd != NULL) {
1889 uatomic_dec(&relayd->refcount);
1890 assert(uatomic_read(&relayd->refcount) >= 0);
1891
1892 /* Closing streams requires to lock the control socket. */
1893 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1894 ret = relayd_send_close_stream(&relayd->control_sock,
1895 stream->relayd_stream_id, stream->next_net_seq_num - 1);
1896 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1897 if (ret < 0) {
1898 DBG("Unable to close stream on the relayd. Continuing");
1899 /*
1900 * Continue here. There is nothing we can do for the relayd.
1901 * Chances are that the relayd has closed the socket so we just
1902 * continue cleaning up.
1903 */
1904 }
1905
1906 /* Both conditions are met, we destroy the relayd. */
1907 if (uatomic_read(&relayd->refcount) == 0 &&
1908 uatomic_read(&relayd->destroy_flag)) {
d09e1200 1909 destroy_relayd(relayd);
fb3a43a9
DG
1910 }
1911 }
1912 rcu_read_unlock();
1913
1914 /* Atomically decrement channel refcount since other threads can use it. */
1915 uatomic_dec(&stream->chan->refcount);
c30aaa51
MD
1916 if (!uatomic_read(&stream->chan->refcount)
1917 && !uatomic_read(&stream->chan->nb_init_streams)) {
1918 /* Go for channel deletion! */
e316aad5 1919 free_chan = stream->chan;
fb3a43a9
DG
1920 }
1921
e316aad5 1922end:
8994307f 1923 pthread_mutex_unlock(&stream->lock);
74251bb8 1924 pthread_mutex_unlock(&consumer_data.lock);
e316aad5
DG
1925
1926 if (free_chan) {
1927 consumer_del_channel(free_chan);
1928 }
1929
1930free_stream:
58b1f425 1931 call_rcu(&stream->node.head, consumer_free_stream);
fb3a43a9
DG
1932}
1933
1934/*
1935 * Action done with the metadata stream when adding it to the consumer internal
1936 * data structures to handle it.
1937 */
e316aad5
DG
1938static int consumer_add_metadata_stream(struct lttng_consumer_stream *stream,
1939 struct lttng_ht *ht)
fb3a43a9 1940{
e316aad5 1941 int ret = 0;
fb3a43a9 1942 struct consumer_relayd_sock_pair *relayd;
76082088
DG
1943 struct lttng_ht_iter iter;
1944 struct lttng_ht_node_ulong *node;
fb3a43a9 1945
e316aad5
DG
1946 assert(stream);
1947 assert(ht);
1948
1949 DBG3("Adding metadata stream %d to hash table", stream->wait_fd);
1950
1951 pthread_mutex_lock(&consumer_data.lock);
2e818a6a 1952 pthread_mutex_lock(&stream->lock);
e316aad5 1953
e316aad5
DG
1954 /*
1955 * From here, refcounts are updated so be _careful_ when returning an error
1956 * after this point.
1957 */
1958
fb3a43a9 1959 rcu_read_lock();
76082088
DG
1960
1961 /*
1962 * Lookup the stream just to make sure it does not exist in our internal
1963 * state. This should NEVER happen.
1964 */
1965 lttng_ht_lookup(ht, (void *)((unsigned long) stream->wait_fd), &iter);
1966 node = lttng_ht_iter_get_node_ulong(&iter);
1967 assert(!node);
1968
e316aad5 1969 /* Find relayd and, if one is found, increment refcount. */
fb3a43a9
DG
1970 relayd = consumer_find_relayd(stream->net_seq_idx);
1971 if (relayd != NULL) {
1972 uatomic_inc(&relayd->refcount);
1973 }
e316aad5
DG
1974
1975 /* Update channel refcount once added without error(s). */
1976 uatomic_inc(&stream->chan->refcount);
1977
1978 /*
1979 * When nb_init_streams reaches 0, we don't need to trigger any action in
1980 * terms of destroying the associated channel, because the action that
1981 * causes the count to become 0 also causes a stream to be added. The
1982 * channel deletion will thus be triggered by the following removal of this
1983 * stream.
1984 */
1985 if (uatomic_read(&stream->chan->nb_init_streams) > 0) {
1986 uatomic_dec(&stream->chan->nb_init_streams);
1987 }
1988
58b1f425 1989 lttng_ht_add_unique_ulong(ht, &stream->node);
ca22feea
DG
1990
1991 /*
1992 * Add stream to the stream_list_ht of the consumer data. No need to steal
1993 * the key since the HT does not use it and we allow to add redundant keys
1994 * into this table.
1995 */
1996 lttng_ht_add_ulong(consumer_data.stream_list_ht, &stream->node_session_id);
1997
fb3a43a9 1998 rcu_read_unlock();
e316aad5 1999
2e818a6a 2000 pthread_mutex_unlock(&stream->lock);
e316aad5
DG
2001 pthread_mutex_unlock(&consumer_data.lock);
2002 return ret;
fb3a43a9
DG
2003}
2004
8994307f
DG
2005/*
2006 * Delete data stream that are flagged for deletion (endpoint_status).
2007 */
2008static void validate_endpoint_status_data_stream(void)
2009{
2010 struct lttng_ht_iter iter;
2011 struct lttng_consumer_stream *stream;
2012
2013 DBG("Consumer delete flagged data stream");
2014
2015 rcu_read_lock();
2016 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
2017 /* Validate delete flag of the stream */
79d4ffb7 2018 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2019 continue;
2020 }
2021 /* Delete it right now */
2022 consumer_del_stream(stream, data_ht);
2023 }
2024 rcu_read_unlock();
2025}
2026
2027/*
2028 * Delete metadata stream that are flagged for deletion (endpoint_status).
2029 */
2030static void validate_endpoint_status_metadata_stream(
2031 struct lttng_poll_event *pollset)
2032{
2033 struct lttng_ht_iter iter;
2034 struct lttng_consumer_stream *stream;
2035
2036 DBG("Consumer delete flagged metadata stream");
2037
2038 assert(pollset);
2039
2040 rcu_read_lock();
2041 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
2042 /* Validate delete flag of the stream */
79d4ffb7 2043 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2044 continue;
2045 }
2046 /*
2047 * Remove from pollset so the metadata thread can continue without
2048 * blocking on a deleted stream.
2049 */
2050 lttng_poll_del(pollset, stream->wait_fd);
2051
2052 /* Delete it right now */
2053 consumer_del_metadata_stream(stream, metadata_ht);
2054 }
2055 rcu_read_unlock();
2056}
2057
fb3a43a9
DG
2058/*
2059 * Thread polls on metadata file descriptor and write them on disk or on the
2060 * network.
2061 */
7d980def 2062void *consumer_thread_metadata_poll(void *data)
fb3a43a9
DG
2063{
2064 int ret, i, pollfd;
2065 uint32_t revents, nb_fd;
e316aad5 2066 struct lttng_consumer_stream *stream = NULL;
fb3a43a9
DG
2067 struct lttng_ht_iter iter;
2068 struct lttng_ht_node_ulong *node;
fb3a43a9
DG
2069 struct lttng_poll_event events;
2070 struct lttng_consumer_local_data *ctx = data;
2071 ssize_t len;
2072
2073 rcu_register_thread();
2074
04bb2b64
DG
2075 metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2076 if (!metadata_ht) {
2077 /* ENOMEM at this point. Better to bail out. */
2078 goto error;
2079 }
2080
fb3a43a9
DG
2081 DBG("Thread metadata poll started");
2082
fb3a43a9
DG
2083 /* Size is set to 1 for the consumer_metadata pipe */
2084 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2085 if (ret < 0) {
2086 ERR("Poll set creation failed");
2087 goto end;
2088 }
2089
2090 ret = lttng_poll_add(&events, ctx->consumer_metadata_pipe[0], LPOLLIN);
2091 if (ret < 0) {
2092 goto end;
2093 }
2094
2095 /* Main loop */
2096 DBG("Metadata main loop started");
2097
2098 while (1) {
fb3a43a9 2099 /* Only the metadata pipe is set */
d21b0d71 2100 if (LTTNG_POLL_GETNB(&events) == 0 && consumer_quit == 1) {
fb3a43a9
DG
2101 goto end;
2102 }
2103
2104restart:
d21b0d71 2105 DBG("Metadata poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events));
fb3a43a9
DG
2106 ret = lttng_poll_wait(&events, -1);
2107 DBG("Metadata event catched in thread");
2108 if (ret < 0) {
2109 if (errno == EINTR) {
e316aad5 2110 ERR("Poll EINTR catched");
fb3a43a9
DG
2111 goto restart;
2112 }
2113 goto error;
2114 }
2115
0d9c5d77
DG
2116 nb_fd = ret;
2117
e316aad5 2118 /* From here, the event is a metadata wait fd */
fb3a43a9
DG
2119 for (i = 0; i < nb_fd; i++) {
2120 revents = LTTNG_POLL_GETEV(&events, i);
2121 pollfd = LTTNG_POLL_GETFD(&events, i);
2122
e316aad5
DG
2123 /* Just don't waste time if no returned events for the fd */
2124 if (!revents) {
2125 continue;
2126 }
2127
fb3a43a9 2128 if (pollfd == ctx->consumer_metadata_pipe[0]) {
4adabd61 2129 if (revents & (LPOLLERR | LPOLLHUP )) {
fb3a43a9
DG
2130 DBG("Metadata thread pipe hung up");
2131 /*
2132 * Remove the pipe from the poll set and continue the loop
2133 * since their might be data to consume.
2134 */
2135 lttng_poll_del(&events, ctx->consumer_metadata_pipe[0]);
f66c074c
DG
2136 ret = close(ctx->consumer_metadata_pipe[0]);
2137 if (ret < 0) {
2138 PERROR("close metadata pipe");
2139 }
fb3a43a9
DG
2140 continue;
2141 } else if (revents & LPOLLIN) {
fb3a43a9 2142 do {
633d0084
DG
2143 /* Get the stream pointer received */
2144 ret = read(pollfd, &stream, sizeof(stream));
fb3a43a9 2145 } while (ret < 0 && errno == EINTR);
633d0084
DG
2146 if (ret < 0 ||
2147 ret < sizeof(struct lttng_consumer_stream *)) {
fb3a43a9 2148 PERROR("read metadata stream");
fb3a43a9
DG
2149 /*
2150 * Let's continue here and hope we can still work
2151 * without stopping the consumer. XXX: Should we?
2152 */
2153 continue;
2154 }
2155
8994307f
DG
2156 /* A NULL stream means that the state has changed. */
2157 if (stream == NULL) {
2158 /* Check for deleted streams. */
2159 validate_endpoint_status_metadata_stream(&events);
3714380f 2160 goto restart;
8994307f
DG
2161 }
2162
fb3a43a9
DG
2163 DBG("Adding metadata stream %d to poll set",
2164 stream->wait_fd);
2165
e316aad5
DG
2166 ret = consumer_add_metadata_stream(stream, metadata_ht);
2167 if (ret) {
2168 ERR("Unable to add metadata stream");
2169 /* Stream was not setup properly. Continuing. */
2170 consumer_del_metadata_stream(stream, NULL);
2171 continue;
2172 }
fb3a43a9
DG
2173
2174 /* Add metadata stream to the global poll events list */
2175 lttng_poll_add(&events, stream->wait_fd,
2176 LPOLLIN | LPOLLPRI);
fb3a43a9
DG
2177 }
2178
e316aad5 2179 /* Handle other stream */
fb3a43a9
DG
2180 continue;
2181 }
2182
d09e1200 2183 rcu_read_lock();
fb3a43a9
DG
2184 lttng_ht_lookup(metadata_ht, (void *)((unsigned long) pollfd),
2185 &iter);
2186 node = lttng_ht_iter_get_node_ulong(&iter);
e316aad5 2187 assert(node);
fb3a43a9
DG
2188
2189 stream = caa_container_of(node, struct lttng_consumer_stream,
58b1f425 2190 node);
fb3a43a9 2191
e316aad5 2192 /* Check for error event */
4adabd61 2193 if (revents & (LPOLLERR | LPOLLHUP)) {
e316aad5 2194 DBG("Metadata fd %d is hup|err.", pollfd);
fb3a43a9
DG
2195 if (!stream->hangup_flush_done
2196 && (consumer_data.type == LTTNG_CONSUMER32_UST
2197 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2198 DBG("Attempting to flush and consume the UST buffers");
2199 lttng_ustconsumer_on_stream_hangup(stream);
2200
2201 /* We just flushed the stream now read it. */
4bb94b75
DG
2202 do {
2203 len = ctx->on_buffer_ready(stream, ctx);
2204 /*
2205 * We don't check the return value here since if we get
2206 * a negative len, it means an error occured thus we
2207 * simply remove it from the poll set and free the
2208 * stream.
2209 */
2210 } while (len > 0);
fb3a43a9
DG
2211 }
2212
fb3a43a9 2213 lttng_poll_del(&events, stream->wait_fd);
e316aad5
DG
2214 /*
2215 * This call update the channel states, closes file descriptors
2216 * and securely free the stream.
2217 */
2218 consumer_del_metadata_stream(stream, metadata_ht);
2219 } else if (revents & (LPOLLIN | LPOLLPRI)) {
2220 /* Get the data out of the metadata file descriptor */
2221 DBG("Metadata available on fd %d", pollfd);
2222 assert(stream->wait_fd == pollfd);
2223
2224 len = ctx->on_buffer_ready(stream, ctx);
2225 /* It's ok to have an unavailable sub-buffer */
b64403e3 2226 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2227 /* Clean up stream from consumer and free it. */
2228 lttng_poll_del(&events, stream->wait_fd);
2229 consumer_del_metadata_stream(stream, metadata_ht);
e316aad5
DG
2230 } else if (len > 0) {
2231 stream->data_read = 1;
2232 }
fb3a43a9 2233 }
e316aad5
DG
2234
2235 /* Release RCU lock for the stream looked up */
d09e1200 2236 rcu_read_unlock();
fb3a43a9
DG
2237 }
2238 }
2239
2240error:
2241end:
2242 DBG("Metadata poll thread exiting");
2243 lttng_poll_clean(&events);
2244
04bb2b64 2245 destroy_stream_ht(metadata_ht);
fb3a43a9
DG
2246
2247 rcu_unregister_thread();
2248 return NULL;
2249}
2250
3bd1e081 2251/*
e4421fec 2252 * This thread polls the fds in the set to consume the data and write
3bd1e081
MD
2253 * it to tracefile if necessary.
2254 */
7d980def 2255void *consumer_thread_data_poll(void *data)
3bd1e081
MD
2256{
2257 int num_rdy, num_hup, high_prio, ret, i;
2258 struct pollfd *pollfd = NULL;
2259 /* local view of the streams */
c869f647 2260 struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL;
3bd1e081
MD
2261 /* local view of consumer_data.fds_count */
2262 int nb_fd = 0;
3bd1e081 2263 struct lttng_consumer_local_data *ctx = data;
00e2e675 2264 ssize_t len;
3bd1e081 2265
e7b994a3
DG
2266 rcu_register_thread();
2267
43c34bc3
DG
2268 data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
2269 if (data_ht == NULL) {
04bb2b64 2270 /* ENOMEM at this point. Better to bail out. */
43c34bc3
DG
2271 goto end;
2272 }
2273
effcf122 2274 local_stream = zmalloc(sizeof(struct lttng_consumer_stream));
3bd1e081
MD
2275
2276 while (1) {
2277 high_prio = 0;
2278 num_hup = 0;
2279
2280 /*
e4421fec 2281 * the fds set has been updated, we need to update our
3bd1e081
MD
2282 * local array as well
2283 */
2284 pthread_mutex_lock(&consumer_data.lock);
2285 if (consumer_data.need_update) {
0e428499
DG
2286 free(pollfd);
2287 pollfd = NULL;
2288
2289 free(local_stream);
2290 local_stream = NULL;
3bd1e081 2291
50f8ae69 2292 /* allocate for all fds + 1 for the consumer_data_pipe */
effcf122 2293 pollfd = zmalloc((consumer_data.stream_count + 1) * sizeof(struct pollfd));
3bd1e081 2294 if (pollfd == NULL) {
7a57cf92 2295 PERROR("pollfd malloc");
3bd1e081
MD
2296 pthread_mutex_unlock(&consumer_data.lock);
2297 goto end;
2298 }
2299
50f8ae69 2300 /* allocate for all fds + 1 for the consumer_data_pipe */
effcf122 2301 local_stream = zmalloc((consumer_data.stream_count + 1) *
3bd1e081
MD
2302 sizeof(struct lttng_consumer_stream));
2303 if (local_stream == NULL) {
7a57cf92 2304 PERROR("local_stream malloc");
3bd1e081
MD
2305 pthread_mutex_unlock(&consumer_data.lock);
2306 goto end;
2307 }
43c34bc3
DG
2308 ret = consumer_update_poll_array(ctx, &pollfd, local_stream,
2309 data_ht);
3bd1e081
MD
2310 if (ret < 0) {
2311 ERR("Error in allocating pollfd or local_outfds");
f73fabfd 2312 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2313 pthread_mutex_unlock(&consumer_data.lock);
2314 goto end;
2315 }
2316 nb_fd = ret;
2317 consumer_data.need_update = 0;
2318 }
2319 pthread_mutex_unlock(&consumer_data.lock);
2320
4078b776
MD
2321 /* No FDs and consumer_quit, consumer_cleanup the thread */
2322 if (nb_fd == 0 && consumer_quit == 1) {
2323 goto end;
2324 }
3bd1e081 2325 /* poll on the array of fds */
88f2b785 2326 restart:
3bd1e081 2327 DBG("polling on %d fd", nb_fd + 1);
cb365c03 2328 num_rdy = poll(pollfd, nb_fd + 1, -1);
3bd1e081
MD
2329 DBG("poll num_rdy : %d", num_rdy);
2330 if (num_rdy == -1) {
88f2b785
MD
2331 /*
2332 * Restart interrupted system call.
2333 */
2334 if (errno == EINTR) {
2335 goto restart;
2336 }
7a57cf92 2337 PERROR("Poll error");
f73fabfd 2338 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2339 goto end;
2340 } else if (num_rdy == 0) {
2341 DBG("Polling thread timed out");
2342 goto end;
2343 }
2344
3bd1e081 2345 /*
50f8ae69 2346 * If the consumer_data_pipe triggered poll go directly to the
00e2e675
DG
2347 * beginning of the loop to update the array. We want to prioritize
2348 * array update over low-priority reads.
3bd1e081 2349 */
509bb1cf 2350 if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) {
ab30f567 2351 ssize_t pipe_readlen;
04fdd819 2352
50f8ae69 2353 DBG("consumer_data_pipe wake up");
04fdd819
MD
2354 /* Consume 1 byte of pipe data */
2355 do {
50f8ae69 2356 pipe_readlen = read(ctx->consumer_data_pipe[0], &new_stream,
c869f647 2357 sizeof(new_stream));
04fdd819 2358 } while (pipe_readlen == -1 && errno == EINTR);
23f5f35d
DG
2359 if (pipe_readlen < 0) {
2360 PERROR("read consumer data pipe");
2361 /* Continue so we can at least handle the current stream(s). */
2362 continue;
2363 }
c869f647
DG
2364
2365 /*
2366 * If the stream is NULL, just ignore it. It's also possible that
2367 * the sessiond poll thread changed the consumer_quit state and is
2368 * waking us up to test it.
2369 */
2370 if (new_stream == NULL) {
8994307f 2371 validate_endpoint_status_data_stream();
c869f647
DG
2372 continue;
2373 }
2374
43c34bc3 2375 ret = consumer_add_stream(new_stream, data_ht);
c869f647
DG
2376 if (ret) {
2377 ERR("Consumer add stream %d failed. Continuing",
2378 new_stream->key);
2379 /*
2380 * At this point, if the add_stream fails, it is not in the
2381 * hash table thus passing the NULL value here.
2382 */
2383 consumer_del_stream(new_stream, NULL);
2384 }
2385
2386 /* Continue to update the local streams and handle prio ones */
3bd1e081
MD
2387 continue;
2388 }
2389
2390 /* Take care of high priority channels first. */
2391 for (i = 0; i < nb_fd; i++) {
9617607b
DG
2392 if (local_stream[i] == NULL) {
2393 continue;
2394 }
fb3a43a9 2395 if (pollfd[i].revents & POLLPRI) {
d41f73b7
MD
2396 DBG("Urgent read on fd %d", pollfd[i].fd);
2397 high_prio = 1;
4078b776 2398 len = ctx->on_buffer_ready(local_stream[i], ctx);
d41f73b7 2399 /* it's ok to have an unavailable sub-buffer */
b64403e3 2400 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2401 /* Clean the stream and free it. */
2402 consumer_del_stream(local_stream[i], data_ht);
9617607b 2403 local_stream[i] = NULL;
4078b776
MD
2404 } else if (len > 0) {
2405 local_stream[i]->data_read = 1;
d41f73b7 2406 }
3bd1e081
MD
2407 }
2408 }
2409
4078b776
MD
2410 /*
2411 * If we read high prio channel in this loop, try again
2412 * for more high prio data.
2413 */
2414 if (high_prio) {
3bd1e081
MD
2415 continue;
2416 }
2417
2418 /* Take care of low priority channels. */
4078b776 2419 for (i = 0; i < nb_fd; i++) {
9617607b
DG
2420 if (local_stream[i] == NULL) {
2421 continue;
2422 }
4078b776
MD
2423 if ((pollfd[i].revents & POLLIN) ||
2424 local_stream[i]->hangup_flush_done) {
4078b776
MD
2425 DBG("Normal read on fd %d", pollfd[i].fd);
2426 len = ctx->on_buffer_ready(local_stream[i], ctx);
2427 /* it's ok to have an unavailable sub-buffer */
b64403e3 2428 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2429 /* Clean the stream and free it. */
2430 consumer_del_stream(local_stream[i], data_ht);
9617607b 2431 local_stream[i] = NULL;
4078b776
MD
2432 } else if (len > 0) {
2433 local_stream[i]->data_read = 1;
2434 }
2435 }
2436 }
2437
2438 /* Handle hangup and errors */
2439 for (i = 0; i < nb_fd; i++) {
9617607b
DG
2440 if (local_stream[i] == NULL) {
2441 continue;
2442 }
4078b776
MD
2443 if (!local_stream[i]->hangup_flush_done
2444 && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL))
2445 && (consumer_data.type == LTTNG_CONSUMER32_UST
2446 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2447 DBG("fd %d is hup|err|nval. Attempting flush and read.",
9617607b 2448 pollfd[i].fd);
4078b776
MD
2449 lttng_ustconsumer_on_stream_hangup(local_stream[i]);
2450 /* Attempt read again, for the data we just flushed. */
2451 local_stream[i]->data_read = 1;
2452 }
2453 /*
2454 * If the poll flag is HUP/ERR/NVAL and we have
2455 * read no data in this pass, we can remove the
2456 * stream from its hash table.
2457 */
2458 if ((pollfd[i].revents & POLLHUP)) {
2459 DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
2460 if (!local_stream[i]->data_read) {
43c34bc3 2461 consumer_del_stream(local_stream[i], data_ht);
9617607b 2462 local_stream[i] = NULL;
4078b776
MD
2463 num_hup++;
2464 }
2465 } else if (pollfd[i].revents & POLLERR) {
2466 ERR("Error returned in polling fd %d.", pollfd[i].fd);
2467 if (!local_stream[i]->data_read) {
43c34bc3 2468 consumer_del_stream(local_stream[i], data_ht);
9617607b 2469 local_stream[i] = NULL;
4078b776
MD
2470 num_hup++;
2471 }
2472 } else if (pollfd[i].revents & POLLNVAL) {
2473 ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
2474 if (!local_stream[i]->data_read) {
43c34bc3 2475 consumer_del_stream(local_stream[i], data_ht);
9617607b 2476 local_stream[i] = NULL;
4078b776 2477 num_hup++;
3bd1e081
MD
2478 }
2479 }
9617607b
DG
2480 if (local_stream[i] != NULL) {
2481 local_stream[i]->data_read = 0;
2482 }
3bd1e081
MD
2483 }
2484 }
2485end:
2486 DBG("polling thread exiting");
0e428499
DG
2487 free(pollfd);
2488 free(local_stream);
fb3a43a9
DG
2489
2490 /*
2491 * Close the write side of the pipe so epoll_wait() in
7d980def
DG
2492 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2493 * read side of the pipe. If we close them both, epoll_wait strangely does
2494 * not return and could create a endless wait period if the pipe is the
2495 * only tracked fd in the poll set. The thread will take care of closing
2496 * the read side.
fb3a43a9 2497 */
f66c074c
DG
2498 ret = close(ctx->consumer_metadata_pipe[1]);
2499 if (ret < 0) {
2500 PERROR("close data pipe");
2501 }
fb3a43a9 2502
04bb2b64 2503 destroy_data_stream_ht(data_ht);
43c34bc3 2504
e7b994a3 2505 rcu_unregister_thread();
3bd1e081
MD
2506 return NULL;
2507}
2508
2509/*
2510 * This thread listens on the consumerd socket and receives the file
2511 * descriptors from the session daemon.
2512 */
7d980def 2513void *consumer_thread_sessiond_poll(void *data)
3bd1e081 2514{
d96f09c6 2515 int sock = -1, client_socket, ret;
3bd1e081
MD
2516 /*
2517 * structure to poll for incoming data on communication socket avoids
2518 * making blocking sockets.
2519 */
2520 struct pollfd consumer_sockpoll[2];
2521 struct lttng_consumer_local_data *ctx = data;
2522
e7b994a3
DG
2523 rcu_register_thread();
2524
3bd1e081
MD
2525 DBG("Creating command socket %s", ctx->consumer_command_sock_path);
2526 unlink(ctx->consumer_command_sock_path);
2527 client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path);
2528 if (client_socket < 0) {
2529 ERR("Cannot create command socket");
2530 goto end;
2531 }
2532
2533 ret = lttcomm_listen_unix_sock(client_socket);
2534 if (ret < 0) {
2535 goto end;
2536 }
2537
32258573 2538 DBG("Sending ready command to lttng-sessiond");
f73fabfd 2539 ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY);
3bd1e081
MD
2540 /* return < 0 on error, but == 0 is not fatal */
2541 if (ret < 0) {
32258573 2542 ERR("Error sending ready command to lttng-sessiond");
3bd1e081
MD
2543 goto end;
2544 }
2545
2546 ret = fcntl(client_socket, F_SETFL, O_NONBLOCK);
2547 if (ret < 0) {
7a57cf92 2548 PERROR("fcntl O_NONBLOCK");
3bd1e081
MD
2549 goto end;
2550 }
2551
2552 /* prepare the FDs to poll : to client socket and the should_quit pipe */
2553 consumer_sockpoll[0].fd = ctx->consumer_should_quit[0];
2554 consumer_sockpoll[0].events = POLLIN | POLLPRI;
2555 consumer_sockpoll[1].fd = client_socket;
2556 consumer_sockpoll[1].events = POLLIN | POLLPRI;
2557
2558 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
2559 goto end;
2560 }
2561 DBG("Connection on client_socket");
2562
2563 /* Blocking call, waiting for transmission */
2564 sock = lttcomm_accept_unix_sock(client_socket);
534d2592 2565 if (sock < 0) {
3bd1e081
MD
2566 WARN("On accept");
2567 goto end;
2568 }
2569 ret = fcntl(sock, F_SETFL, O_NONBLOCK);
2570 if (ret < 0) {
7a57cf92 2571 PERROR("fcntl O_NONBLOCK");
3bd1e081
MD
2572 goto end;
2573 }
2574
d96f09c6
DG
2575 /* This socket is not useful anymore. */
2576 ret = close(client_socket);
2577 if (ret < 0) {
2578 PERROR("close client_socket");
2579 }
2580 client_socket = -1;
2581
3bd1e081
MD
2582 /* update the polling structure to poll on the established socket */
2583 consumer_sockpoll[1].fd = sock;
2584 consumer_sockpoll[1].events = POLLIN | POLLPRI;
2585
2586 while (1) {
2587 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
2588 goto end;
2589 }
2590 DBG("Incoming command on sock");
2591 ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll);
2592 if (ret == -ENOENT) {
2593 DBG("Received STOP command");
2594 goto end;
2595 }
4cbc1a04
DG
2596 if (ret <= 0) {
2597 /*
2598 * This could simply be a session daemon quitting. Don't output
2599 * ERR() here.
2600 */
2601 DBG("Communication interrupted on command socket");
3bd1e081
MD
2602 goto end;
2603 }
2604 if (consumer_quit) {
2605 DBG("consumer_thread_receive_fds received quit from signal");
2606 goto end;
2607 }
2608 DBG("received fds on sock");
2609 }
2610end:
2611 DBG("consumer_thread_receive_fds exiting");
2612
2613 /*
2614 * when all fds have hung up, the polling thread
2615 * can exit cleanly
2616 */
2617 consumer_quit = 1;
2618
04fdd819 2619 /*
c869f647 2620 * Notify the data poll thread to poll back again and test the
8994307f 2621 * consumer_quit state that we just set so to quit gracefully.
04fdd819 2622 */
8994307f 2623 notify_thread_pipe(ctx->consumer_data_pipe[1]);
c869f647 2624
d96f09c6
DG
2625 /* Cleaning up possibly open sockets. */
2626 if (sock >= 0) {
2627 ret = close(sock);
2628 if (ret < 0) {
2629 PERROR("close sock sessiond poll");
2630 }
2631 }
2632 if (client_socket >= 0) {
2633 ret = close(sock);
2634 if (ret < 0) {
2635 PERROR("close client_socket sessiond poll");
2636 }
2637 }
2638
e7b994a3 2639 rcu_unregister_thread();
3bd1e081
MD
2640 return NULL;
2641}
d41f73b7 2642
4078b776 2643ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream,
d41f73b7
MD
2644 struct lttng_consumer_local_data *ctx)
2645{
74251bb8
DG
2646 ssize_t ret;
2647
2648 pthread_mutex_lock(&stream->lock);
2649
d41f73b7
MD
2650 switch (consumer_data.type) {
2651 case LTTNG_CONSUMER_KERNEL:
74251bb8
DG
2652 ret = lttng_kconsumer_read_subbuffer(stream, ctx);
2653 break;
7753dea8
MD
2654 case LTTNG_CONSUMER32_UST:
2655 case LTTNG_CONSUMER64_UST:
74251bb8
DG
2656 ret = lttng_ustconsumer_read_subbuffer(stream, ctx);
2657 break;
d41f73b7
MD
2658 default:
2659 ERR("Unknown consumer_data type");
2660 assert(0);
74251bb8
DG
2661 ret = -ENOSYS;
2662 break;
d41f73b7 2663 }
74251bb8
DG
2664
2665 pthread_mutex_unlock(&stream->lock);
2666 return ret;
d41f73b7
MD
2667}
2668
2669int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream)
2670{
2671 switch (consumer_data.type) {
2672 case LTTNG_CONSUMER_KERNEL:
2673 return lttng_kconsumer_on_recv_stream(stream);
7753dea8
MD
2674 case LTTNG_CONSUMER32_UST:
2675 case LTTNG_CONSUMER64_UST:
d41f73b7
MD
2676 return lttng_ustconsumer_on_recv_stream(stream);
2677 default:
2678 ERR("Unknown consumer_data type");
2679 assert(0);
2680 return -ENOSYS;
2681 }
2682}
e4421fec
DG
2683
2684/*
2685 * Allocate and set consumer data hash tables.
2686 */
2687void lttng_consumer_init(void)
2688{
e4421fec 2689 consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
00e2e675 2690 consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
53632229 2691 consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_ULONG);
e4421fec 2692}
7735ef9e
DG
2693
2694/*
2695 * Process the ADD_RELAYD command receive by a consumer.
2696 *
2697 * This will create a relayd socket pair and add it to the relayd hash table.
2698 * The caller MUST acquire a RCU read side lock before calling it.
2699 */
2700int consumer_add_relayd_socket(int net_seq_idx, int sock_type,
2701 struct lttng_consumer_local_data *ctx, int sock,
46e6455f
DG
2702 struct pollfd *consumer_sockpoll, struct lttcomm_sock *relayd_sock,
2703 unsigned int sessiond_id)
7735ef9e 2704{
cd2b09ed 2705 int fd = -1, ret = -1, relayd_created = 0;
f50f23d9 2706 enum lttng_error_code ret_code = LTTNG_OK;
7735ef9e
DG
2707 struct consumer_relayd_sock_pair *relayd;
2708
2709 DBG("Consumer adding relayd socket (idx: %d)", net_seq_idx);
2710
f50f23d9
DG
2711 /* First send a status message before receiving the fds. */
2712 ret = consumer_send_status_msg(sock, ret_code);
2713 if (ret < 0) {
2714 /* Somehow, the session daemon is not responding anymore. */
2715 goto error;
2716 }
2717
7735ef9e
DG
2718 /* Get relayd reference if exists. */
2719 relayd = consumer_find_relayd(net_seq_idx);
2720 if (relayd == NULL) {
2721 /* Not found. Allocate one. */
2722 relayd = consumer_allocate_relayd_sock_pair(net_seq_idx);
2723 if (relayd == NULL) {
f73fabfd 2724 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
59e71485 2725 ret = -1;
7735ef9e
DG
2726 goto error;
2727 }
f7079f67 2728 relayd->sessiond_session_id = (uint64_t) sessiond_id;
cd2b09ed 2729 relayd_created = 1;
7735ef9e
DG
2730 }
2731
2732 /* Poll on consumer socket. */
2733 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
2734 ret = -EINTR;
2735 goto error;
2736 }
2737
2738 /* Get relayd socket from session daemon */
2739 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
2740 if (ret != sizeof(fd)) {
f73fabfd 2741 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
7735ef9e 2742 ret = -1;
4028eeb9 2743 fd = -1; /* Just in case it gets set with an invalid value. */
7735ef9e
DG
2744 goto error;
2745 }
2746
f50f23d9
DG
2747 /* We have the fds without error. Send status back. */
2748 ret = consumer_send_status_msg(sock, ret_code);
2749 if (ret < 0) {
2750 /* Somehow, the session daemon is not responding anymore. */
2751 goto error;
2752 }
2753
7735ef9e
DG
2754 /* Copy socket information and received FD */
2755 switch (sock_type) {
2756 case LTTNG_STREAM_CONTROL:
2757 /* Copy received lttcomm socket */
2758 lttcomm_copy_sock(&relayd->control_sock, relayd_sock);
2759 ret = lttcomm_create_sock(&relayd->control_sock);
4028eeb9
DG
2760 /* Immediately try to close the created socket if valid. */
2761 if (relayd->control_sock.fd >= 0) {
2762 if (close(relayd->control_sock.fd)) {
2763 PERROR("close relayd control socket");
2764 }
7735ef9e 2765 }
4028eeb9 2766 /* Handle create_sock error. */
f66c074c 2767 if (ret < 0) {
4028eeb9 2768 goto error;
f66c074c 2769 }
7735ef9e
DG
2770
2771 /* Assign new file descriptor */
2772 relayd->control_sock.fd = fd;
c5b6f4f0
DG
2773
2774 /*
59e71485
DG
2775 * Create a session on the relayd and store the returned id. Lock the
2776 * control socket mutex if the relayd was NOT created before.
c5b6f4f0 2777 */
59e71485
DG
2778 if (!relayd_created) {
2779 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
2780 }
c5b6f4f0 2781 ret = relayd_create_session(&relayd->control_sock,
f7079f67 2782 &relayd->relayd_session_id);
59e71485
DG
2783 if (!relayd_created) {
2784 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
2785 }
c5b6f4f0
DG
2786 if (ret < 0) {
2787 goto error;
2788 }
2789
7735ef9e
DG
2790 break;
2791 case LTTNG_STREAM_DATA:
2792 /* Copy received lttcomm socket */
2793 lttcomm_copy_sock(&relayd->data_sock, relayd_sock);
2794 ret = lttcomm_create_sock(&relayd->data_sock);
4028eeb9
DG
2795 /* Immediately try to close the created socket if valid. */
2796 if (relayd->data_sock.fd >= 0) {
2797 if (close(relayd->data_sock.fd)) {
2798 PERROR("close relayd data socket");
2799 }
7735ef9e 2800 }
4028eeb9 2801 /* Handle create_sock error. */
f66c074c 2802 if (ret < 0) {
4028eeb9 2803 goto error;
f66c074c 2804 }
7735ef9e
DG
2805
2806 /* Assign new file descriptor */
2807 relayd->data_sock.fd = fd;
2808 break;
2809 default:
2810 ERR("Unknown relayd socket type (%d)", sock_type);
59e71485 2811 ret = -1;
7735ef9e
DG
2812 goto error;
2813 }
2814
2815 DBG("Consumer %s socket created successfully with net idx %d (fd: %d)",
2816 sock_type == LTTNG_STREAM_CONTROL ? "control" : "data",
2817 relayd->net_seq_idx, fd);
2818
2819 /*
2820 * Add relayd socket pair to consumer data hashtable. If object already
2821 * exists or on error, the function gracefully returns.
2822 */
d09e1200 2823 add_relayd(relayd);
7735ef9e
DG
2824
2825 /* All good! */
4028eeb9 2826 return 0;
7735ef9e
DG
2827
2828error:
4028eeb9
DG
2829 /* Close received socket if valid. */
2830 if (fd >= 0) {
2831 if (close(fd)) {
2832 PERROR("close received socket");
2833 }
2834 }
cd2b09ed
DG
2835
2836 if (relayd_created) {
2837 /* We just want to cleanup. Ignore ret value. */
2838 (void) relayd_close(&relayd->control_sock);
2839 (void) relayd_close(&relayd->data_sock);
2840 free(relayd);
2841 }
2842
7735ef9e
DG
2843 return ret;
2844}
ca22feea 2845
4e9a4686
DG
2846/*
2847 * Try to lock the stream mutex.
2848 *
2849 * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
2850 */
2851static int stream_try_lock(struct lttng_consumer_stream *stream)
2852{
2853 int ret;
2854
2855 assert(stream);
2856
2857 /*
2858 * Try to lock the stream mutex. On failure, we know that the stream is
2859 * being used else where hence there is data still being extracted.
2860 */
2861 ret = pthread_mutex_trylock(&stream->lock);
2862 if (ret) {
2863 /* For both EBUSY and EINVAL error, the mutex is NOT locked. */
2864 ret = 0;
2865 goto end;
2866 }
2867
2868 ret = 1;
2869
2870end:
2871 return ret;
2872}
2873
f7079f67
DG
2874/*
2875 * Search for a relayd associated to the session id and return the reference.
2876 *
2877 * A rcu read side lock MUST be acquire before calling this function and locked
2878 * until the relayd object is no longer necessary.
2879 */
2880static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id)
2881{
2882 struct lttng_ht_iter iter;
f7079f67 2883 struct consumer_relayd_sock_pair *relayd = NULL;
f7079f67
DG
2884
2885 /* Iterate over all relayd since they are indexed by net_seq_idx. */
2886 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
2887 node.node) {
18261bd1
DG
2888 /*
2889 * Check by sessiond id which is unique here where the relayd session
2890 * id might not be when having multiple relayd.
2891 */
2892 if (relayd->sessiond_session_id == id) {
f7079f67 2893 /* Found the relayd. There can be only one per id. */
18261bd1 2894 goto found;
f7079f67
DG
2895 }
2896 }
2897
18261bd1
DG
2898 return NULL;
2899
2900found:
f7079f67
DG
2901 return relayd;
2902}
2903
ca22feea
DG
2904/*
2905 * Check if for a given session id there is still data needed to be extract
2906 * from the buffers.
2907 *
6d805429 2908 * Return 1 if data is pending or else 0 meaning ready to be read.
ca22feea 2909 */
6d805429 2910int consumer_data_pending(uint64_t id)
ca22feea
DG
2911{
2912 int ret;
2913 struct lttng_ht_iter iter;
2914 struct lttng_ht *ht;
2915 struct lttng_consumer_stream *stream;
f7079f67 2916 struct consumer_relayd_sock_pair *relayd = NULL;
6d805429 2917 int (*data_pending)(struct lttng_consumer_stream *);
ca22feea 2918
6d805429 2919 DBG("Consumer data pending command on session id %" PRIu64, id);
ca22feea 2920
6f6eda74 2921 rcu_read_lock();
ca22feea
DG
2922 pthread_mutex_lock(&consumer_data.lock);
2923
2924 switch (consumer_data.type) {
2925 case LTTNG_CONSUMER_KERNEL:
6d805429 2926 data_pending = lttng_kconsumer_data_pending;
ca22feea
DG
2927 break;
2928 case LTTNG_CONSUMER32_UST:
2929 case LTTNG_CONSUMER64_UST:
6d805429 2930 data_pending = lttng_ustconsumer_data_pending;
ca22feea
DG
2931 break;
2932 default:
2933 ERR("Unknown consumer data type");
2934 assert(0);
2935 }
2936
2937 /* Ease our life a bit */
2938 ht = consumer_data.stream_list_ht;
2939
f7079f67
DG
2940 relayd = find_relayd_by_session_id(id);
2941 if (relayd) {
2942 /* Send init command for data pending. */
2943 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
2944 ret = relayd_begin_data_pending(&relayd->control_sock,
2945 relayd->relayd_session_id);
2946 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
2947 if (ret < 0) {
2948 /* Communication error thus the relayd so no data pending. */
2949 goto data_not_pending;
2950 }
2951 }
2952
c8f59ee5 2953 cds_lfht_for_each_entry_duplicate(ht->ht,
b6314938 2954 ht->hash_fct((void *)((unsigned long) id), lttng_ht_seed),
ca22feea
DG
2955 ht->match_fct, (void *)((unsigned long) id),
2956 &iter.iter, stream, node_session_id.node) {
4e9a4686
DG
2957 /* If this call fails, the stream is being used hence data pending. */
2958 ret = stream_try_lock(stream);
2959 if (!ret) {
f7079f67 2960 goto data_pending;
ca22feea 2961 }
ca22feea 2962
4e9a4686
DG
2963 /*
2964 * A removed node from the hash table indicates that the stream has
2965 * been deleted thus having a guarantee that the buffers are closed
2966 * on the consumer side. However, data can still be transmitted
2967 * over the network so don't skip the relayd check.
2968 */
2969 ret = cds_lfht_is_node_deleted(&stream->node.node);
2970 if (!ret) {
2971 /* Check the stream if there is data in the buffers. */
6d805429
DG
2972 ret = data_pending(stream);
2973 if (ret == 1) {
4e9a4686 2974 pthread_mutex_unlock(&stream->lock);
f7079f67 2975 goto data_pending;
4e9a4686
DG
2976 }
2977 }
2978
2979 /* Relayd check */
f7079f67 2980 if (relayd) {
c8f59ee5
DG
2981 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
2982 if (stream->metadata_flag) {
ad7051c0
DG
2983 ret = relayd_quiescent_control(&relayd->control_sock,
2984 stream->relayd_stream_id);
c8f59ee5 2985 } else {
6d805429 2986 ret = relayd_data_pending(&relayd->control_sock,
39df6d9f
DG
2987 stream->relayd_stream_id,
2988 stream->next_net_seq_num - 1);
c8f59ee5
DG
2989 }
2990 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
6d805429 2991 if (ret == 1) {
4e9a4686 2992 pthread_mutex_unlock(&stream->lock);
f7079f67 2993 goto data_pending;
c8f59ee5
DG
2994 }
2995 }
4e9a4686 2996 pthread_mutex_unlock(&stream->lock);
c8f59ee5 2997 }
ca22feea 2998
f7079f67
DG
2999 if (relayd) {
3000 unsigned int is_data_inflight = 0;
3001
3002 /* Send init command for data pending. */
3003 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3004 ret = relayd_end_data_pending(&relayd->control_sock,
3005 relayd->relayd_session_id, &is_data_inflight);
3006 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
bdd88757 3007 if (ret < 0) {
f7079f67
DG
3008 goto data_not_pending;
3009 }
bdd88757
DG
3010 if (is_data_inflight) {
3011 goto data_pending;
3012 }
f7079f67
DG
3013 }
3014
ca22feea 3015 /*
f7079f67
DG
3016 * Finding _no_ node in the hash table and no inflight data means that the
3017 * stream(s) have been removed thus data is guaranteed to be available for
3018 * analysis from the trace files.
ca22feea
DG
3019 */
3020
f7079f67 3021data_not_pending:
ca22feea
DG
3022 /* Data is available to be read by a viewer. */
3023 pthread_mutex_unlock(&consumer_data.lock);
c8f59ee5 3024 rcu_read_unlock();
6d805429 3025 return 0;
ca22feea 3026
f7079f67 3027data_pending:
ca22feea
DG
3028 /* Data is still being extracted from buffers. */
3029 pthread_mutex_unlock(&consumer_data.lock);
c8f59ee5 3030 rcu_read_unlock();
6d805429 3031 return 1;
ca22feea 3032}
f50f23d9
DG
3033
3034/*
3035 * Send a ret code status message to the sessiond daemon.
3036 *
3037 * Return the sendmsg() return value.
3038 */
3039int consumer_send_status_msg(int sock, int ret_code)
3040{
3041 struct lttcomm_consumer_status_msg msg;
3042
3043 msg.ret_code = ret_code;
3044
3045 return lttcomm_send_unix_sock(sock, &msg, sizeof(msg));
3046}
This page took 0.215725 seconds and 5 git commands to generate.