Fix: unchecked posix_fadvise() return value
[lttng-tools.git] / src / common / consumer / 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
6c1c0768 20#define _LGPL_SOURCE
3bd1e081 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>
331744e3 31#include <signal.h>
3bd1e081 32
51a9e1c7 33#include <bin/lttng-consumerd/health-consumerd.h>
990570ed 34#include <common/common.h>
fb3a43a9
DG
35#include <common/utils.h>
36#include <common/compat/poll.h>
f263b7fd 37#include <common/compat/endian.h>
309167d2 38#include <common/index/index.h>
10a8a223 39#include <common/kernel-ctl/kernel-ctl.h>
00e2e675 40#include <common/sessiond-comm/relayd.h>
10a8a223
DG
41#include <common/sessiond-comm/sessiond-comm.h>
42#include <common/kernel-consumer/kernel-consumer.h>
00e2e675 43#include <common/relayd/relayd.h>
10a8a223 44#include <common/ust-consumer/ust-consumer.h>
c8fea79c
JR
45#include <common/consumer/consumer-timer.h>
46#include <common/consumer/consumer.h>
47#include <common/consumer/consumer-stream.h>
48#include <common/consumer/consumer-testpoint.h>
49#include <common/align.h>
3bd1e081
MD
50
51struct lttng_consumer_global_data consumer_data = {
3bd1e081
MD
52 .stream_count = 0,
53 .need_update = 1,
54 .type = LTTNG_CONSUMER_UNKNOWN,
55};
56
d8ef542d
MD
57enum consumer_channel_action {
58 CONSUMER_CHANNEL_ADD,
a0cbdd2e 59 CONSUMER_CHANNEL_DEL,
d8ef542d
MD
60 CONSUMER_CHANNEL_QUIT,
61};
62
63struct consumer_channel_msg {
64 enum consumer_channel_action action;
a0cbdd2e
MD
65 struct lttng_consumer_channel *chan; /* add */
66 uint64_t key; /* del */
d8ef542d
MD
67};
68
3bd1e081
MD
69/*
70 * Flag to inform the polling thread to quit when all fd hung up. Updated by
71 * the consumer_thread_receive_fds when it notices that all fds has hung up.
72 * Also updated by the signal handler (consumer_should_exit()). Read by the
73 * polling threads.
74 */
a98dae5f 75volatile int consumer_quit;
3bd1e081 76
43c34bc3 77/*
43c34bc3
DG
78 * Global hash table containing respectively metadata and data streams. The
79 * stream element in this ht should only be updated by the metadata poll thread
80 * for the metadata and the data poll thread for the data.
81 */
40dc48e0
DG
82static struct lttng_ht *metadata_ht;
83static struct lttng_ht *data_ht;
43c34bc3 84
acdb9057
DG
85/*
86 * Notify a thread lttng pipe to poll back again. This usually means that some
87 * global state has changed so we just send back the thread in a poll wait
88 * call.
89 */
90static void notify_thread_lttng_pipe(struct lttng_pipe *pipe)
91{
92 struct lttng_consumer_stream *null_stream = NULL;
93
94 assert(pipe);
95
96 (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream));
97}
98
5c635c72
MD
99static void notify_health_quit_pipe(int *pipe)
100{
6cd525e8 101 ssize_t ret;
5c635c72 102
6cd525e8
MD
103 ret = lttng_write(pipe[1], "4", 1);
104 if (ret < 1) {
5c635c72
MD
105 PERROR("write consumer health quit");
106 }
107}
108
d8ef542d
MD
109static void notify_channel_pipe(struct lttng_consumer_local_data *ctx,
110 struct lttng_consumer_channel *chan,
a0cbdd2e 111 uint64_t key,
d8ef542d
MD
112 enum consumer_channel_action action)
113{
114 struct consumer_channel_msg msg;
6cd525e8 115 ssize_t ret;
d8ef542d 116
e56251fc
DG
117 memset(&msg, 0, sizeof(msg));
118
d8ef542d
MD
119 msg.action = action;
120 msg.chan = chan;
f21dae48 121 msg.key = key;
6cd525e8
MD
122 ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg));
123 if (ret < sizeof(msg)) {
124 PERROR("notify_channel_pipe write error");
125 }
d8ef542d
MD
126}
127
a0cbdd2e
MD
128void notify_thread_del_channel(struct lttng_consumer_local_data *ctx,
129 uint64_t key)
130{
131 notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL);
132}
133
d8ef542d
MD
134static int read_channel_pipe(struct lttng_consumer_local_data *ctx,
135 struct lttng_consumer_channel **chan,
a0cbdd2e 136 uint64_t *key,
d8ef542d
MD
137 enum consumer_channel_action *action)
138{
139 struct consumer_channel_msg msg;
6cd525e8 140 ssize_t ret;
d8ef542d 141
6cd525e8
MD
142 ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg));
143 if (ret < sizeof(msg)) {
144 ret = -1;
145 goto error;
d8ef542d 146 }
6cd525e8
MD
147 *action = msg.action;
148 *chan = msg.chan;
149 *key = msg.key;
150error:
151 return (int) ret;
d8ef542d
MD
152}
153
212d67a2
DG
154/*
155 * Cleanup the stream list of a channel. Those streams are not yet globally
156 * visible
157 */
158static void clean_channel_stream_list(struct lttng_consumer_channel *channel)
159{
160 struct lttng_consumer_stream *stream, *stmp;
161
162 assert(channel);
163
164 /* Delete streams that might have been left in the stream list. */
165 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
166 send_node) {
167 cds_list_del(&stream->send_node);
168 /*
169 * Once a stream is added to this list, the buffers were created so we
170 * have a guarantee that this call will succeed. Setting the monitor
171 * mode to 0 so we don't lock nor try to delete the stream from the
172 * global hash table.
173 */
174 stream->monitor = 0;
175 consumer_stream_destroy(stream, NULL);
176 }
177}
178
3bd1e081
MD
179/*
180 * Find a stream. The consumer_data.lock must be locked during this
181 * call.
182 */
d88aee68 183static struct lttng_consumer_stream *find_stream(uint64_t key,
8389e4f8 184 struct lttng_ht *ht)
3bd1e081 185{
e4421fec 186 struct lttng_ht_iter iter;
d88aee68 187 struct lttng_ht_node_u64 *node;
e4421fec 188 struct lttng_consumer_stream *stream = NULL;
3bd1e081 189
8389e4f8
DG
190 assert(ht);
191
d88aee68
DG
192 /* -1ULL keys are lookup failures */
193 if (key == (uint64_t) -1ULL) {
7ad0a0cb 194 return NULL;
7a57cf92 195 }
e4421fec 196
6065ceec
DG
197 rcu_read_lock();
198
d88aee68
DG
199 lttng_ht_lookup(ht, &key, &iter);
200 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
201 if (node != NULL) {
202 stream = caa_container_of(node, struct lttng_consumer_stream, node);
3bd1e081 203 }
e4421fec 204
6065ceec
DG
205 rcu_read_unlock();
206
e4421fec 207 return stream;
3bd1e081
MD
208}
209
da009f2c 210static void steal_stream_key(uint64_t key, struct lttng_ht *ht)
7ad0a0cb
MD
211{
212 struct lttng_consumer_stream *stream;
213
04253271 214 rcu_read_lock();
ffe60014 215 stream = find_stream(key, ht);
04253271 216 if (stream) {
da009f2c 217 stream->key = (uint64_t) -1ULL;
04253271
MD
218 /*
219 * We don't want the lookup to match, but we still need
220 * to iterate on this stream when iterating over the hash table. Just
221 * change the node key.
222 */
da009f2c 223 stream->node.key = (uint64_t) -1ULL;
04253271
MD
224 }
225 rcu_read_unlock();
7ad0a0cb
MD
226}
227
d56db448
DG
228/*
229 * Return a channel object for the given key.
230 *
231 * RCU read side lock MUST be acquired before calling this function and
232 * protects the channel ptr.
233 */
d88aee68 234struct lttng_consumer_channel *consumer_find_channel(uint64_t key)
3bd1e081 235{
e4421fec 236 struct lttng_ht_iter iter;
d88aee68 237 struct lttng_ht_node_u64 *node;
e4421fec 238 struct lttng_consumer_channel *channel = NULL;
3bd1e081 239
d88aee68
DG
240 /* -1ULL keys are lookup failures */
241 if (key == (uint64_t) -1ULL) {
7ad0a0cb 242 return NULL;
7a57cf92 243 }
e4421fec 244
d88aee68
DG
245 lttng_ht_lookup(consumer_data.channel_ht, &key, &iter);
246 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
247 if (node != NULL) {
248 channel = caa_container_of(node, struct lttng_consumer_channel, node);
3bd1e081 249 }
e4421fec
DG
250
251 return channel;
3bd1e081
MD
252}
253
b5a6470f
DG
254/*
255 * There is a possibility that the consumer does not have enough time between
256 * the close of the channel on the session daemon and the cleanup in here thus
257 * once we have a channel add with an existing key, we know for sure that this
258 * channel will eventually get cleaned up by all streams being closed.
259 *
260 * This function just nullifies the already existing channel key.
261 */
262static void steal_channel_key(uint64_t key)
263{
264 struct lttng_consumer_channel *channel;
265
266 rcu_read_lock();
267 channel = consumer_find_channel(key);
268 if (channel) {
269 channel->key = (uint64_t) -1ULL;
270 /*
271 * We don't want the lookup to match, but we still need to iterate on
272 * this channel when iterating over the hash table. Just change the
273 * node key.
274 */
275 channel->node.key = (uint64_t) -1ULL;
276 }
277 rcu_read_unlock();
278}
279
ffe60014 280static void free_channel_rcu(struct rcu_head *head)
702b1ea4 281{
d88aee68
DG
282 struct lttng_ht_node_u64 *node =
283 caa_container_of(head, struct lttng_ht_node_u64, head);
ffe60014
DG
284 struct lttng_consumer_channel *channel =
285 caa_container_of(node, struct lttng_consumer_channel, node);
702b1ea4 286
b83e03c4
MD
287 switch (consumer_data.type) {
288 case LTTNG_CONSUMER_KERNEL:
289 break;
290 case LTTNG_CONSUMER32_UST:
291 case LTTNG_CONSUMER64_UST:
292 lttng_ustconsumer_free_channel(channel);
293 break;
294 default:
295 ERR("Unknown consumer_data type");
296 abort();
297 }
ffe60014 298 free(channel);
702b1ea4
MD
299}
300
00e2e675
DG
301/*
302 * RCU protected relayd socket pair free.
303 */
ffe60014 304static void free_relayd_rcu(struct rcu_head *head)
00e2e675 305{
d88aee68
DG
306 struct lttng_ht_node_u64 *node =
307 caa_container_of(head, struct lttng_ht_node_u64, head);
00e2e675
DG
308 struct consumer_relayd_sock_pair *relayd =
309 caa_container_of(node, struct consumer_relayd_sock_pair, node);
310
8994307f
DG
311 /*
312 * Close all sockets. This is done in the call RCU since we don't want the
313 * socket fds to be reassigned thus potentially creating bad state of the
314 * relayd object.
315 *
316 * We do not have to lock the control socket mutex here since at this stage
317 * there is no one referencing to this relayd object.
318 */
319 (void) relayd_close(&relayd->control_sock);
320 (void) relayd_close(&relayd->data_sock);
321
00e2e675
DG
322 free(relayd);
323}
324
325/*
326 * Destroy and free relayd socket pair object.
00e2e675 327 */
51230d70 328void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
329{
330 int ret;
331 struct lttng_ht_iter iter;
332
173af62f
DG
333 if (relayd == NULL) {
334 return;
335 }
336
00e2e675
DG
337 DBG("Consumer destroy and close relayd socket pair");
338
339 iter.iter.node = &relayd->node.node;
340 ret = lttng_ht_del(consumer_data.relayd_ht, &iter);
173af62f 341 if (ret != 0) {
8994307f 342 /* We assume the relayd is being or is destroyed */
173af62f
DG
343 return;
344 }
00e2e675 345
00e2e675 346 /* RCU free() call */
ffe60014
DG
347 call_rcu(&relayd->node.head, free_relayd_rcu);
348}
349
350/*
351 * Remove a channel from the global list protected by a mutex. This function is
352 * also responsible for freeing its data structures.
353 */
354void consumer_del_channel(struct lttng_consumer_channel *channel)
355{
356 int ret;
357 struct lttng_ht_iter iter;
358
d88aee68 359 DBG("Consumer delete channel key %" PRIu64, channel->key);
ffe60014
DG
360
361 pthread_mutex_lock(&consumer_data.lock);
a9838785 362 pthread_mutex_lock(&channel->lock);
ffe60014 363
212d67a2
DG
364 /* Destroy streams that might have been left in the stream list. */
365 clean_channel_stream_list(channel);
51e762e5 366
d3e2ba59
JD
367 if (channel->live_timer_enabled == 1) {
368 consumer_timer_live_stop(channel);
369 }
370
ffe60014
DG
371 switch (consumer_data.type) {
372 case LTTNG_CONSUMER_KERNEL:
373 break;
374 case LTTNG_CONSUMER32_UST:
375 case LTTNG_CONSUMER64_UST:
376 lttng_ustconsumer_del_channel(channel);
377 break;
378 default:
379 ERR("Unknown consumer_data type");
380 assert(0);
381 goto end;
382 }
383
384 rcu_read_lock();
385 iter.iter.node = &channel->node.node;
386 ret = lttng_ht_del(consumer_data.channel_ht, &iter);
387 assert(!ret);
388 rcu_read_unlock();
389
390 call_rcu(&channel->node.head, free_channel_rcu);
391end:
a9838785 392 pthread_mutex_unlock(&channel->lock);
ffe60014 393 pthread_mutex_unlock(&consumer_data.lock);
00e2e675
DG
394}
395
228b5bf7
DG
396/*
397 * Iterate over the relayd hash table and destroy each element. Finally,
398 * destroy the whole hash table.
399 */
400static void cleanup_relayd_ht(void)
401{
402 struct lttng_ht_iter iter;
403 struct consumer_relayd_sock_pair *relayd;
404
405 rcu_read_lock();
406
407 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
408 node.node) {
51230d70 409 consumer_destroy_relayd(relayd);
228b5bf7
DG
410 }
411
228b5bf7 412 rcu_read_unlock();
36b588ed
MD
413
414 lttng_ht_destroy(consumer_data.relayd_ht);
228b5bf7
DG
415}
416
8994307f
DG
417/*
418 * Update the end point status of all streams having the given network sequence
419 * index (relayd index).
420 *
421 * It's atomically set without having the stream mutex locked which is fine
422 * because we handle the write/read race with a pipe wakeup for each thread.
423 */
da009f2c 424static void update_endpoint_status_by_netidx(uint64_t net_seq_idx,
8994307f
DG
425 enum consumer_endpoint_status status)
426{
427 struct lttng_ht_iter iter;
428 struct lttng_consumer_stream *stream;
429
da009f2c 430 DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx);
8994307f
DG
431
432 rcu_read_lock();
433
434 /* Let's begin with metadata */
435 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
436 if (stream->net_seq_idx == net_seq_idx) {
437 uatomic_set(&stream->endpoint_status, status);
438 DBG("Delete flag set to metadata stream %d", stream->wait_fd);
439 }
440 }
441
442 /* Follow up by the data streams */
443 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
444 if (stream->net_seq_idx == net_seq_idx) {
445 uatomic_set(&stream->endpoint_status, status);
446 DBG("Delete flag set to data stream %d", stream->wait_fd);
447 }
448 }
449 rcu_read_unlock();
450}
451
452/*
453 * Cleanup a relayd object by flagging every associated streams for deletion,
454 * destroying the object meaning removing it from the relayd hash table,
455 * closing the sockets and freeing the memory in a RCU call.
456 *
457 * If a local data context is available, notify the threads that the streams'
458 * state have changed.
459 */
460static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd,
461 struct lttng_consumer_local_data *ctx)
462{
da009f2c 463 uint64_t netidx;
8994307f
DG
464
465 assert(relayd);
466
9617607b
DG
467 DBG("Cleaning up relayd sockets");
468
8994307f
DG
469 /* Save the net sequence index before destroying the object */
470 netidx = relayd->net_seq_idx;
471
472 /*
473 * Delete the relayd from the relayd hash table, close the sockets and free
474 * the object in a RCU call.
475 */
51230d70 476 consumer_destroy_relayd(relayd);
8994307f
DG
477
478 /* Set inactive endpoint to all streams */
479 update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE);
480
481 /*
482 * With a local data context, notify the threads that the streams' state
483 * have changed. The write() action on the pipe acts as an "implicit"
484 * memory barrier ordering the updates of the end point status from the
485 * read of this status which happens AFTER receiving this notify.
486 */
487 if (ctx) {
acdb9057 488 notify_thread_lttng_pipe(ctx->consumer_data_pipe);
13886d2d 489 notify_thread_lttng_pipe(ctx->consumer_metadata_pipe);
8994307f
DG
490 }
491}
492
a6ba4fe1
DG
493/*
494 * Flag a relayd socket pair for destruction. Destroy it if the refcount
495 * reaches zero.
496 *
497 * RCU read side lock MUST be aquired before calling this function.
498 */
499void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd)
500{
501 assert(relayd);
502
503 /* Set destroy flag for this object */
504 uatomic_set(&relayd->destroy_flag, 1);
505
506 /* Destroy the relayd if refcount is 0 */
507 if (uatomic_read(&relayd->refcount) == 0) {
51230d70 508 consumer_destroy_relayd(relayd);
a6ba4fe1
DG
509 }
510}
511
3bd1e081 512/*
1d1a276c
DG
513 * Completly destroy stream from every visiable data structure and the given
514 * hash table if one.
515 *
516 * One this call returns, the stream object is not longer usable nor visible.
3bd1e081 517 */
e316aad5
DG
518void consumer_del_stream(struct lttng_consumer_stream *stream,
519 struct lttng_ht *ht)
3bd1e081 520{
1d1a276c 521 consumer_stream_destroy(stream, ht);
3bd1e081
MD
522}
523
5ab66908
MD
524/*
525 * XXX naming of del vs destroy is all mixed up.
526 */
527void consumer_del_stream_for_data(struct lttng_consumer_stream *stream)
528{
529 consumer_stream_destroy(stream, data_ht);
530}
531
532void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream)
533{
534 consumer_stream_destroy(stream, metadata_ht);
535}
536
d88aee68
DG
537struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key,
538 uint64_t stream_key,
3bd1e081 539 enum lttng_consumer_stream_state state,
ffe60014 540 const char *channel_name,
6df2e2c9 541 uid_t uid,
00e2e675 542 gid_t gid,
57a269f2 543 uint64_t relayd_id,
53632229 544 uint64_t session_id,
ffe60014
DG
545 int cpu,
546 int *alloc_ret,
4891ece8
DG
547 enum consumer_channel_type type,
548 unsigned int monitor)
3bd1e081 549{
ffe60014 550 int ret;
3bd1e081 551 struct lttng_consumer_stream *stream;
3bd1e081 552
effcf122 553 stream = zmalloc(sizeof(*stream));
3bd1e081 554 if (stream == NULL) {
7a57cf92 555 PERROR("malloc struct lttng_consumer_stream");
ffe60014 556 ret = -ENOMEM;
7a57cf92 557 goto end;
3bd1e081 558 }
7a57cf92 559
d56db448
DG
560 rcu_read_lock();
561
3bd1e081 562 stream->key = stream_key;
3bd1e081
MD
563 stream->out_fd = -1;
564 stream->out_fd_offset = 0;
e5d1a9b3 565 stream->output_written = 0;
3bd1e081 566 stream->state = state;
6df2e2c9
MD
567 stream->uid = uid;
568 stream->gid = gid;
ffe60014 569 stream->net_seq_idx = relayd_id;
53632229 570 stream->session_id = session_id;
4891ece8 571 stream->monitor = monitor;
774d490c 572 stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE;
309167d2 573 stream->index_fd = -1;
fb83fe64 574 stream->last_sequence_number = -1ULL;
53632229 575 pthread_mutex_init(&stream->lock, NULL);
c585821b 576 pthread_mutex_init(&stream->metadata_timer_lock, NULL);
58b1f425 577
ffe60014
DG
578 /* If channel is the metadata, flag this stream as metadata. */
579 if (type == CONSUMER_CHANNEL_TYPE_METADATA) {
580 stream->metadata_flag = 1;
581 /* Metadata is flat out. */
582 strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name));
94d49140
JD
583 /* Live rendez-vous point. */
584 pthread_cond_init(&stream->metadata_rdv, NULL);
585 pthread_mutex_init(&stream->metadata_rdv_lock, NULL);
58b1f425 586 } else {
ffe60014
DG
587 /* Format stream name to <channel_name>_<cpu_number> */
588 ret = snprintf(stream->name, sizeof(stream->name), "%s_%d",
589 channel_name, cpu);
590 if (ret < 0) {
591 PERROR("snprintf stream name");
592 goto error;
593 }
58b1f425 594 }
c30aaa51 595
ffe60014 596 /* Key is always the wait_fd for streams. */
d88aee68 597 lttng_ht_node_init_u64(&stream->node, stream->key);
ffe60014 598
d8ef542d
MD
599 /* Init node per channel id key */
600 lttng_ht_node_init_u64(&stream->node_channel_id, channel_key);
601
53632229 602 /* Init session id node with the stream session id */
d88aee68 603 lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id);
53632229 604
07b86b52
JD
605 DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64
606 " relayd_id %" PRIu64 ", session_id %" PRIu64,
607 stream->name, stream->key, channel_key,
608 stream->net_seq_idx, stream->session_id);
d56db448
DG
609
610 rcu_read_unlock();
3bd1e081 611 return stream;
c80048c6
MD
612
613error:
d56db448 614 rcu_read_unlock();
c80048c6 615 free(stream);
7a57cf92 616end:
ffe60014
DG
617 if (alloc_ret) {
618 *alloc_ret = ret;
619 }
c80048c6 620 return NULL;
3bd1e081
MD
621}
622
623/*
624 * Add a stream to the global list protected by a mutex.
625 */
5ab66908 626int consumer_add_data_stream(struct lttng_consumer_stream *stream)
3bd1e081 627{
5ab66908 628 struct lttng_ht *ht = data_ht;
3bd1e081
MD
629 int ret = 0;
630
e316aad5 631 assert(stream);
43c34bc3 632 assert(ht);
c77fc10a 633
d88aee68 634 DBG3("Adding consumer stream %" PRIu64, stream->key);
e316aad5
DG
635
636 pthread_mutex_lock(&consumer_data.lock);
a9838785 637 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 638 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 639 pthread_mutex_lock(&stream->lock);
b0b335c8 640 rcu_read_lock();
e316aad5 641
43c34bc3 642 /* Steal stream identifier to avoid having streams with the same key */
ffe60014 643 steal_stream_key(stream->key, ht);
43c34bc3 644
d88aee68 645 lttng_ht_add_unique_u64(ht, &stream->node);
00e2e675 646
d8ef542d
MD
647 lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht,
648 &stream->node_channel_id);
649
ca22feea
DG
650 /*
651 * Add stream to the stream_list_ht of the consumer data. No need to steal
652 * the key since the HT does not use it and we allow to add redundant keys
653 * into this table.
654 */
d88aee68 655 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 656
e316aad5 657 /*
ffe60014
DG
658 * When nb_init_stream_left reaches 0, we don't need to trigger any action
659 * in terms of destroying the associated channel, because the action that
e316aad5
DG
660 * causes the count to become 0 also causes a stream to be added. The
661 * channel deletion will thus be triggered by the following removal of this
662 * stream.
663 */
ffe60014 664 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
665 /* Increment refcount before decrementing nb_init_stream_left */
666 cmm_smp_wmb();
ffe60014 667 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
668 }
669
670 /* Update consumer data once the node is inserted. */
3bd1e081
MD
671 consumer_data.stream_count++;
672 consumer_data.need_update = 1;
673
e316aad5 674 rcu_read_unlock();
2e818a6a 675 pthread_mutex_unlock(&stream->lock);
ec6ea7d0 676 pthread_mutex_unlock(&stream->chan->timer_lock);
a9838785 677 pthread_mutex_unlock(&stream->chan->lock);
3bd1e081 678 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 679
3bd1e081
MD
680 return ret;
681}
682
5ab66908
MD
683void consumer_del_data_stream(struct lttng_consumer_stream *stream)
684{
685 consumer_del_stream(stream, data_ht);
686}
687
00e2e675 688/*
3f8e211f
DG
689 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
690 * be acquired before calling this.
00e2e675 691 */
d09e1200 692static int add_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
693{
694 int ret = 0;
d88aee68 695 struct lttng_ht_node_u64 *node;
00e2e675
DG
696 struct lttng_ht_iter iter;
697
ffe60014 698 assert(relayd);
00e2e675 699
00e2e675 700 lttng_ht_lookup(consumer_data.relayd_ht,
d88aee68
DG
701 &relayd->net_seq_idx, &iter);
702 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675 703 if (node != NULL) {
00e2e675
DG
704 goto end;
705 }
d88aee68 706 lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node);
00e2e675 707
00e2e675
DG
708end:
709 return ret;
710}
711
712/*
713 * Allocate and return a consumer relayd socket.
714 */
027a694f 715static struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair(
da009f2c 716 uint64_t net_seq_idx)
00e2e675
DG
717{
718 struct consumer_relayd_sock_pair *obj = NULL;
719
da009f2c
MD
720 /* net sequence index of -1 is a failure */
721 if (net_seq_idx == (uint64_t) -1ULL) {
00e2e675
DG
722 goto error;
723 }
724
725 obj = zmalloc(sizeof(struct consumer_relayd_sock_pair));
726 if (obj == NULL) {
727 PERROR("zmalloc relayd sock");
728 goto error;
729 }
730
731 obj->net_seq_idx = net_seq_idx;
732 obj->refcount = 0;
173af62f 733 obj->destroy_flag = 0;
f96e4545
MD
734 obj->control_sock.sock.fd = -1;
735 obj->data_sock.sock.fd = -1;
d88aee68 736 lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx);
00e2e675
DG
737 pthread_mutex_init(&obj->ctrl_sock_mutex, NULL);
738
739error:
740 return obj;
741}
742
743/*
744 * Find a relayd socket pair in the global consumer data.
745 *
746 * Return the object if found else NULL.
b0b335c8
MD
747 * RCU read-side lock must be held across this call and while using the
748 * returned object.
00e2e675 749 */
d88aee68 750struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key)
00e2e675
DG
751{
752 struct lttng_ht_iter iter;
d88aee68 753 struct lttng_ht_node_u64 *node;
00e2e675
DG
754 struct consumer_relayd_sock_pair *relayd = NULL;
755
756 /* Negative keys are lookup failures */
d88aee68 757 if (key == (uint64_t) -1ULL) {
00e2e675
DG
758 goto error;
759 }
760
d88aee68 761 lttng_ht_lookup(consumer_data.relayd_ht, &key,
00e2e675 762 &iter);
d88aee68 763 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675
DG
764 if (node != NULL) {
765 relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node);
766 }
767
00e2e675
DG
768error:
769 return relayd;
770}
771
10a50311
JD
772/*
773 * Find a relayd and send the stream
774 *
775 * Returns 0 on success, < 0 on error
776 */
777int consumer_send_relayd_stream(struct lttng_consumer_stream *stream,
778 char *path)
779{
780 int ret = 0;
781 struct consumer_relayd_sock_pair *relayd;
782
783 assert(stream);
784 assert(stream->net_seq_idx != -1ULL);
785 assert(path);
786
787 /* The stream is not metadata. Get relayd reference if exists. */
788 rcu_read_lock();
789 relayd = consumer_find_relayd(stream->net_seq_idx);
790 if (relayd != NULL) {
791 /* Add stream on the relayd */
792 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
793 ret = relayd_add_stream(&relayd->control_sock, stream->name,
794 path, &stream->relayd_stream_id,
795 stream->chan->tracefile_size, stream->chan->tracefile_count);
796 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
797 if (ret < 0) {
798 goto end;
799 }
1c20f0e2 800
10a50311 801 uatomic_inc(&relayd->refcount);
d01178b6 802 stream->sent_to_relayd = 1;
10a50311
JD
803 } else {
804 ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.",
805 stream->key, stream->net_seq_idx);
806 ret = -1;
807 goto end;
808 }
809
810 DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64,
811 stream->name, stream->key, stream->net_seq_idx);
812
813end:
814 rcu_read_unlock();
815 return ret;
816}
817
a4baae1b
JD
818/*
819 * Find a relayd and send the streams sent message
820 *
821 * Returns 0 on success, < 0 on error
822 */
823int consumer_send_relayd_streams_sent(uint64_t net_seq_idx)
824{
825 int ret = 0;
826 struct consumer_relayd_sock_pair *relayd;
827
828 assert(net_seq_idx != -1ULL);
829
830 /* The stream is not metadata. Get relayd reference if exists. */
831 rcu_read_lock();
832 relayd = consumer_find_relayd(net_seq_idx);
833 if (relayd != NULL) {
834 /* Add stream on the relayd */
835 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
836 ret = relayd_streams_sent(&relayd->control_sock);
837 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
838 if (ret < 0) {
839 goto end;
840 }
841 } else {
842 ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.",
843 net_seq_idx);
844 ret = -1;
845 goto end;
846 }
847
848 ret = 0;
849 DBG("All streams sent relayd id %" PRIu64, net_seq_idx);
850
851end:
852 rcu_read_unlock();
853 return ret;
854}
855
10a50311
JD
856/*
857 * Find a relayd and close the stream
858 */
859void close_relayd_stream(struct lttng_consumer_stream *stream)
860{
861 struct consumer_relayd_sock_pair *relayd;
862
863 /* The stream is not metadata. Get relayd reference if exists. */
864 rcu_read_lock();
865 relayd = consumer_find_relayd(stream->net_seq_idx);
866 if (relayd) {
867 consumer_stream_relayd_close(stream, relayd);
868 }
869 rcu_read_unlock();
870}
871
00e2e675
DG
872/*
873 * Handle stream for relayd transmission if the stream applies for network
874 * streaming where the net sequence index is set.
875 *
876 * Return destination file descriptor or negative value on error.
877 */
6197aea7 878static int write_relayd_stream_header(struct lttng_consumer_stream *stream,
1d4dfdef
DG
879 size_t data_size, unsigned long padding,
880 struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
881{
882 int outfd = -1, ret;
00e2e675
DG
883 struct lttcomm_relayd_data_hdr data_hdr;
884
885 /* Safety net */
886 assert(stream);
6197aea7 887 assert(relayd);
00e2e675
DG
888
889 /* Reset data header */
890 memset(&data_hdr, 0, sizeof(data_hdr));
891
00e2e675
DG
892 if (stream->metadata_flag) {
893 /* Caller MUST acquire the relayd control socket lock */
894 ret = relayd_send_metadata(&relayd->control_sock, data_size);
895 if (ret < 0) {
896 goto error;
897 }
898
899 /* Metadata are always sent on the control socket. */
6151a90f 900 outfd = relayd->control_sock.sock.fd;
00e2e675
DG
901 } else {
902 /* Set header with stream information */
903 data_hdr.stream_id = htobe64(stream->relayd_stream_id);
904 data_hdr.data_size = htobe32(data_size);
1d4dfdef 905 data_hdr.padding_size = htobe32(padding);
39df6d9f
DG
906 /*
907 * Note that net_seq_num below is assigned with the *current* value of
908 * next_net_seq_num and only after that the next_net_seq_num will be
909 * increment. This is why when issuing a command on the relayd using
910 * this next value, 1 should always be substracted in order to compare
911 * the last seen sequence number on the relayd side to the last sent.
912 */
3604f373 913 data_hdr.net_seq_num = htobe64(stream->next_net_seq_num);
00e2e675
DG
914 /* Other fields are zeroed previously */
915
916 ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr,
917 sizeof(data_hdr));
918 if (ret < 0) {
919 goto error;
920 }
921
3604f373
DG
922 ++stream->next_net_seq_num;
923
00e2e675 924 /* Set to go on data socket */
6151a90f 925 outfd = relayd->data_sock.sock.fd;
00e2e675
DG
926 }
927
928error:
929 return outfd;
930}
931
3bd1e081 932/*
ffe60014
DG
933 * Allocate and return a new lttng_consumer_channel object using the given key
934 * to initialize the hash table node.
935 *
936 * On error, return NULL.
3bd1e081 937 */
886224ff 938struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key,
ffe60014
DG
939 uint64_t session_id,
940 const char *pathname,
941 const char *name,
942 uid_t uid,
943 gid_t gid,
57a269f2 944 uint64_t relayd_id,
1624d5b7
JD
945 enum lttng_event_output output,
946 uint64_t tracefile_size,
2bba9e53 947 uint64_t tracefile_count,
1950109e 948 uint64_t session_id_per_pid,
ecc48a90 949 unsigned int monitor,
d7ba1388 950 unsigned int live_timer_interval,
3d071855 951 const char *root_shm_path,
d7ba1388 952 const char *shm_path)
3bd1e081
MD
953{
954 struct lttng_consumer_channel *channel;
3bd1e081 955
276b26d1 956 channel = zmalloc(sizeof(*channel));
3bd1e081 957 if (channel == NULL) {
7a57cf92 958 PERROR("malloc struct lttng_consumer_channel");
3bd1e081
MD
959 goto end;
960 }
ffe60014
DG
961
962 channel->key = key;
3bd1e081 963 channel->refcount = 0;
ffe60014 964 channel->session_id = session_id;
1950109e 965 channel->session_id_per_pid = session_id_per_pid;
ffe60014
DG
966 channel->uid = uid;
967 channel->gid = gid;
968 channel->relayd_id = relayd_id;
1624d5b7
JD
969 channel->tracefile_size = tracefile_size;
970 channel->tracefile_count = tracefile_count;
2bba9e53 971 channel->monitor = monitor;
ecc48a90 972 channel->live_timer_interval = live_timer_interval;
a9838785 973 pthread_mutex_init(&channel->lock, NULL);
ec6ea7d0 974 pthread_mutex_init(&channel->timer_lock, NULL);
ffe60014 975
0c759fc9
DG
976 switch (output) {
977 case LTTNG_EVENT_SPLICE:
978 channel->output = CONSUMER_CHANNEL_SPLICE;
979 break;
980 case LTTNG_EVENT_MMAP:
981 channel->output = CONSUMER_CHANNEL_MMAP;
982 break;
983 default:
984 assert(0);
985 free(channel);
986 channel = NULL;
987 goto end;
988 }
989
07b86b52
JD
990 /*
991 * In monitor mode, the streams associated with the channel will be put in
992 * a special list ONLY owned by this channel. So, the refcount is set to 1
993 * here meaning that the channel itself has streams that are referenced.
994 *
995 * On a channel deletion, once the channel is no longer visible, the
996 * refcount is decremented and checked for a zero value to delete it. With
997 * streams in no monitor mode, it will now be safe to destroy the channel.
998 */
999 if (!channel->monitor) {
1000 channel->refcount = 1;
1001 }
1002
ffe60014
DG
1003 strncpy(channel->pathname, pathname, sizeof(channel->pathname));
1004 channel->pathname[sizeof(channel->pathname) - 1] = '\0';
1005
1006 strncpy(channel->name, name, sizeof(channel->name));
1007 channel->name[sizeof(channel->name) - 1] = '\0';
1008
3d071855
MD
1009 if (root_shm_path) {
1010 strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path));
1011 channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0';
1012 }
d7ba1388
MD
1013 if (shm_path) {
1014 strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path));
1015 channel->shm_path[sizeof(channel->shm_path) - 1] = '\0';
1016 }
1017
d88aee68 1018 lttng_ht_node_init_u64(&channel->node, channel->key);
d8ef542d
MD
1019
1020 channel->wait_fd = -1;
1021
ffe60014
DG
1022 CDS_INIT_LIST_HEAD(&channel->streams.head);
1023
d88aee68 1024 DBG("Allocated channel (key %" PRIu64 ")", channel->key)
3bd1e081 1025
3bd1e081
MD
1026end:
1027 return channel;
1028}
1029
1030/*
1031 * Add a channel to the global list protected by a mutex.
821fffb2 1032 *
b5a6470f 1033 * Always return 0 indicating success.
3bd1e081 1034 */
d8ef542d
MD
1035int consumer_add_channel(struct lttng_consumer_channel *channel,
1036 struct lttng_consumer_local_data *ctx)
3bd1e081 1037{
3bd1e081 1038 pthread_mutex_lock(&consumer_data.lock);
a9838785 1039 pthread_mutex_lock(&channel->lock);
ec6ea7d0 1040 pthread_mutex_lock(&channel->timer_lock);
c77fc10a 1041
b5a6470f
DG
1042 /*
1043 * This gives us a guarantee that the channel we are about to add to the
1044 * channel hash table will be unique. See this function comment on the why
1045 * we need to steel the channel key at this stage.
1046 */
1047 steal_channel_key(channel->key);
c77fc10a 1048
b5a6470f 1049 rcu_read_lock();
d88aee68 1050 lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node);
6065ceec 1051 rcu_read_unlock();
b5a6470f 1052
ec6ea7d0 1053 pthread_mutex_unlock(&channel->timer_lock);
a9838785 1054 pthread_mutex_unlock(&channel->lock);
3bd1e081 1055 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 1056
b5a6470f 1057 if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) {
a0cbdd2e 1058 notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD);
d8ef542d 1059 }
b5a6470f
DG
1060
1061 return 0;
3bd1e081
MD
1062}
1063
1064/*
1065 * Allocate the pollfd structure and the local view of the out fds to avoid
1066 * doing a lookup in the linked list and concurrency issues when writing is
1067 * needed. Called with consumer_data.lock held.
1068 *
1069 * Returns the number of fds in the structures.
1070 */
ffe60014
DG
1071static int update_poll_array(struct lttng_consumer_local_data *ctx,
1072 struct pollfd **pollfd, struct lttng_consumer_stream **local_stream,
1073 struct lttng_ht *ht)
3bd1e081 1074{
3bd1e081 1075 int i = 0;
e4421fec
DG
1076 struct lttng_ht_iter iter;
1077 struct lttng_consumer_stream *stream;
3bd1e081 1078
ffe60014
DG
1079 assert(ctx);
1080 assert(ht);
1081 assert(pollfd);
1082 assert(local_stream);
1083
3bd1e081 1084 DBG("Updating poll fd array");
481d6c57 1085 rcu_read_lock();
43c34bc3 1086 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
8994307f
DG
1087 /*
1088 * Only active streams with an active end point can be added to the
1089 * poll set and local stream storage of the thread.
1090 *
1091 * There is a potential race here for endpoint_status to be updated
1092 * just after the check. However, this is OK since the stream(s) will
1093 * be deleted once the thread is notified that the end point state has
1094 * changed where this function will be called back again.
1095 */
1096 if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM ||
79d4ffb7 1097 stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) {
3bd1e081
MD
1098 continue;
1099 }
7972aab2
DG
1100 /*
1101 * This clobbers way too much the debug output. Uncomment that if you
1102 * need it for debugging purposes.
1103 *
1104 * DBG("Active FD %d", stream->wait_fd);
1105 */
e4421fec 1106 (*pollfd)[i].fd = stream->wait_fd;
3bd1e081 1107 (*pollfd)[i].events = POLLIN | POLLPRI;
e4421fec 1108 local_stream[i] = stream;
3bd1e081
MD
1109 i++;
1110 }
481d6c57 1111 rcu_read_unlock();
3bd1e081
MD
1112
1113 /*
50f8ae69 1114 * Insert the consumer_data_pipe at the end of the array and don't
3bd1e081
MD
1115 * increment i so nb_fd is the number of real FD.
1116 */
acdb9057 1117 (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe);
509bb1cf 1118 (*pollfd)[i].events = POLLIN | POLLPRI;
02b3d176
DG
1119
1120 (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe);
1121 (*pollfd)[i + 1].events = POLLIN | POLLPRI;
3bd1e081
MD
1122 return i;
1123}
1124
1125/*
84382d49
MD
1126 * Poll on the should_quit pipe and the command socket return -1 on
1127 * error, 1 if should exit, 0 if data is available on the command socket
3bd1e081
MD
1128 */
1129int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll)
1130{
1131 int num_rdy;
1132
88f2b785 1133restart:
3bd1e081
MD
1134 num_rdy = poll(consumer_sockpoll, 2, -1);
1135 if (num_rdy == -1) {
88f2b785
MD
1136 /*
1137 * Restart interrupted system call.
1138 */
1139 if (errno == EINTR) {
1140 goto restart;
1141 }
7a57cf92 1142 PERROR("Poll error");
84382d49 1143 return -1;
3bd1e081 1144 }
509bb1cf 1145 if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) {
3bd1e081 1146 DBG("consumer_should_quit wake up");
84382d49 1147 return 1;
3bd1e081
MD
1148 }
1149 return 0;
3bd1e081
MD
1150}
1151
1152/*
1153 * Set the error socket.
1154 */
ffe60014
DG
1155void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx,
1156 int sock)
3bd1e081
MD
1157{
1158 ctx->consumer_error_socket = sock;
1159}
1160
1161/*
1162 * Set the command socket path.
1163 */
3bd1e081
MD
1164void lttng_consumer_set_command_sock_path(
1165 struct lttng_consumer_local_data *ctx, char *sock)
1166{
1167 ctx->consumer_command_sock_path = sock;
1168}
1169
1170/*
1171 * Send return code to the session daemon.
1172 * If the socket is not defined, we return 0, it is not a fatal error
1173 */
ffe60014 1174int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd)
3bd1e081
MD
1175{
1176 if (ctx->consumer_error_socket > 0) {
1177 return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd,
1178 sizeof(enum lttcomm_sessiond_command));
1179 }
1180
1181 return 0;
1182}
1183
1184/*
228b5bf7
DG
1185 * Close all the tracefiles and stream fds and MUST be called when all
1186 * instances are destroyed i.e. when all threads were joined and are ended.
3bd1e081
MD
1187 */
1188void lttng_consumer_cleanup(void)
1189{
e4421fec 1190 struct lttng_ht_iter iter;
ffe60014 1191 struct lttng_consumer_channel *channel;
6065ceec
DG
1192
1193 rcu_read_lock();
3bd1e081 1194
ffe60014
DG
1195 cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel,
1196 node.node) {
702b1ea4 1197 consumer_del_channel(channel);
3bd1e081 1198 }
6065ceec
DG
1199
1200 rcu_read_unlock();
d6ce1df2 1201
d6ce1df2 1202 lttng_ht_destroy(consumer_data.channel_ht);
228b5bf7
DG
1203
1204 cleanup_relayd_ht();
1205
d8ef542d
MD
1206 lttng_ht_destroy(consumer_data.stream_per_chan_id_ht);
1207
228b5bf7
DG
1208 /*
1209 * This HT contains streams that are freed by either the metadata thread or
1210 * the data thread so we do *nothing* on the hash table and simply destroy
1211 * it.
1212 */
1213 lttng_ht_destroy(consumer_data.stream_list_ht);
3bd1e081
MD
1214}
1215
1216/*
1217 * Called from signal handler.
1218 */
1219void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx)
1220{
6cd525e8
MD
1221 ssize_t ret;
1222
3bd1e081 1223 consumer_quit = 1;
6cd525e8
MD
1224 ret = lttng_write(ctx->consumer_should_quit[1], "4", 1);
1225 if (ret < 1) {
7a57cf92 1226 PERROR("write consumer quit");
3bd1e081 1227 }
ab1027f4
DG
1228
1229 DBG("Consumer flag that it should quit");
3bd1e081
MD
1230}
1231
00e2e675
DG
1232void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream,
1233 off_t orig_offset)
3bd1e081 1234{
c7a78aab 1235 int ret;
3bd1e081
MD
1236 int outfd = stream->out_fd;
1237
1238 /*
1239 * This does a blocking write-and-wait on any page that belongs to the
1240 * subbuffer prior to the one we just wrote.
1241 * Don't care about error values, as these are just hints and ways to
1242 * limit the amount of page cache used.
1243 */
ffe60014 1244 if (orig_offset < stream->max_sb_size) {
3bd1e081
MD
1245 return;
1246 }
ffe60014
DG
1247 lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size,
1248 stream->max_sb_size,
3bd1e081
MD
1249 SYNC_FILE_RANGE_WAIT_BEFORE
1250 | SYNC_FILE_RANGE_WRITE
1251 | SYNC_FILE_RANGE_WAIT_AFTER);
1252 /*
1253 * Give hints to the kernel about how we access the file:
1254 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1255 * we write it.
1256 *
1257 * We need to call fadvise again after the file grows because the
1258 * kernel does not seem to apply fadvise to non-existing parts of the
1259 * file.
1260 *
1261 * Call fadvise _after_ having waited for the page writeback to
1262 * complete because the dirty page writeback semantic is not well
1263 * defined. So it can be expected to lead to lower throughput in
1264 * streaming.
1265 */
c7a78aab 1266 ret = posix_fadvise(outfd, orig_offset - stream->max_sb_size,
ffe60014 1267 stream->max_sb_size, POSIX_FADV_DONTNEED);
c7a78aab
JG
1268 if (ret) {
1269 WARN("posix_fadvise() error (%i)", ret);
1270 }
3bd1e081
MD
1271}
1272
1273/*
1274 * Initialise the necessary environnement :
1275 * - create a new context
1276 * - create the poll_pipe
1277 * - create the should_quit pipe (for signal handler)
1278 * - create the thread pipe (for splice)
1279 *
1280 * Takes a function pointer as argument, this function is called when data is
1281 * available on a buffer. This function is responsible to do the
1282 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1283 * buffer configuration and then kernctl_put_next_subbuf at the end.
1284 *
1285 * Returns a pointer to the new context or NULL on error.
1286 */
1287struct lttng_consumer_local_data *lttng_consumer_create(
1288 enum lttng_consumer_type type,
4078b776 1289 ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream,
d41f73b7 1290 struct lttng_consumer_local_data *ctx),
3bd1e081
MD
1291 int (*recv_channel)(struct lttng_consumer_channel *channel),
1292 int (*recv_stream)(struct lttng_consumer_stream *stream),
30319bcb 1293 int (*update_stream)(uint64_t stream_key, uint32_t state))
3bd1e081 1294{
d8ef542d 1295 int ret;
3bd1e081
MD
1296 struct lttng_consumer_local_data *ctx;
1297
1298 assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN ||
1299 consumer_data.type == type);
1300 consumer_data.type = type;
1301
effcf122 1302 ctx = zmalloc(sizeof(struct lttng_consumer_local_data));
3bd1e081 1303 if (ctx == NULL) {
7a57cf92 1304 PERROR("allocating context");
3bd1e081
MD
1305 goto error;
1306 }
1307
1308 ctx->consumer_error_socket = -1;
331744e3 1309 ctx->consumer_metadata_socket = -1;
75d83e50 1310 pthread_mutex_init(&ctx->metadata_socket_lock, NULL);
3bd1e081
MD
1311 /* assign the callbacks */
1312 ctx->on_buffer_ready = buffer_ready;
1313 ctx->on_recv_channel = recv_channel;
1314 ctx->on_recv_stream = recv_stream;
1315 ctx->on_update_stream = update_stream;
1316
acdb9057
DG
1317 ctx->consumer_data_pipe = lttng_pipe_open(0);
1318 if (!ctx->consumer_data_pipe) {
3bd1e081
MD
1319 goto error_poll_pipe;
1320 }
1321
02b3d176
DG
1322 ctx->consumer_wakeup_pipe = lttng_pipe_open(0);
1323 if (!ctx->consumer_wakeup_pipe) {
1324 goto error_wakeup_pipe;
1325 }
1326
3bd1e081
MD
1327 ret = pipe(ctx->consumer_should_quit);
1328 if (ret < 0) {
7a57cf92 1329 PERROR("Error creating recv pipe");
3bd1e081
MD
1330 goto error_quit_pipe;
1331 }
1332
d8ef542d
MD
1333 ret = pipe(ctx->consumer_channel_pipe);
1334 if (ret < 0) {
1335 PERROR("Error creating channel pipe");
1336 goto error_channel_pipe;
1337 }
1338
13886d2d
DG
1339 ctx->consumer_metadata_pipe = lttng_pipe_open(0);
1340 if (!ctx->consumer_metadata_pipe) {
fb3a43a9
DG
1341 goto error_metadata_pipe;
1342 }
3bd1e081 1343
fb3a43a9 1344 return ctx;
3bd1e081 1345
fb3a43a9 1346error_metadata_pipe:
d8ef542d
MD
1347 utils_close_pipe(ctx->consumer_channel_pipe);
1348error_channel_pipe:
d8ef542d 1349 utils_close_pipe(ctx->consumer_should_quit);
3bd1e081 1350error_quit_pipe:
02b3d176
DG
1351 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
1352error_wakeup_pipe:
acdb9057 1353 lttng_pipe_destroy(ctx->consumer_data_pipe);
3bd1e081
MD
1354error_poll_pipe:
1355 free(ctx);
1356error:
1357 return NULL;
1358}
1359
282dadbc
MD
1360/*
1361 * Iterate over all streams of the hashtable and free them properly.
1362 */
1363static void destroy_data_stream_ht(struct lttng_ht *ht)
1364{
1365 struct lttng_ht_iter iter;
1366 struct lttng_consumer_stream *stream;
1367
1368 if (ht == NULL) {
1369 return;
1370 }
1371
1372 rcu_read_lock();
1373 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1374 /*
1375 * Ignore return value since we are currently cleaning up so any error
1376 * can't be handled.
1377 */
1378 (void) consumer_del_stream(stream, ht);
1379 }
1380 rcu_read_unlock();
1381
1382 lttng_ht_destroy(ht);
1383}
1384
1385/*
1386 * Iterate over all streams of the metadata hashtable and free them
1387 * properly.
1388 */
1389static void destroy_metadata_stream_ht(struct lttng_ht *ht)
1390{
1391 struct lttng_ht_iter iter;
1392 struct lttng_consumer_stream *stream;
1393
1394 if (ht == NULL) {
1395 return;
1396 }
1397
1398 rcu_read_lock();
1399 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1400 /*
1401 * Ignore return value since we are currently cleaning up so any error
1402 * can't be handled.
1403 */
1404 (void) consumer_del_metadata_stream(stream, ht);
1405 }
1406 rcu_read_unlock();
1407
1408 lttng_ht_destroy(ht);
1409}
1410
3bd1e081
MD
1411/*
1412 * Close all fds associated with the instance and free the context.
1413 */
1414void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx)
1415{
4c462e79
MD
1416 int ret;
1417
ab1027f4
DG
1418 DBG("Consumer destroying it. Closing everything.");
1419
4f2e75b9
DG
1420 if (!ctx) {
1421 return;
1422 }
1423
282dadbc
MD
1424 destroy_data_stream_ht(data_ht);
1425 destroy_metadata_stream_ht(metadata_ht);
1426
4c462e79
MD
1427 ret = close(ctx->consumer_error_socket);
1428 if (ret) {
1429 PERROR("close");
1430 }
331744e3
JD
1431 ret = close(ctx->consumer_metadata_socket);
1432 if (ret) {
1433 PERROR("close");
1434 }
d8ef542d 1435 utils_close_pipe(ctx->consumer_channel_pipe);
acdb9057 1436 lttng_pipe_destroy(ctx->consumer_data_pipe);
13886d2d 1437 lttng_pipe_destroy(ctx->consumer_metadata_pipe);
02b3d176 1438 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
d8ef542d 1439 utils_close_pipe(ctx->consumer_should_quit);
fb3a43a9 1440
3bd1e081
MD
1441 unlink(ctx->consumer_command_sock_path);
1442 free(ctx);
1443}
1444
6197aea7
DG
1445/*
1446 * Write the metadata stream id on the specified file descriptor.
1447 */
1448static int write_relayd_metadata_id(int fd,
1449 struct lttng_consumer_stream *stream,
ffe60014 1450 struct consumer_relayd_sock_pair *relayd, unsigned long padding)
6197aea7 1451{
6cd525e8 1452 ssize_t ret;
1d4dfdef 1453 struct lttcomm_relayd_metadata_payload hdr;
6197aea7 1454
1d4dfdef
DG
1455 hdr.stream_id = htobe64(stream->relayd_stream_id);
1456 hdr.padding_size = htobe32(padding);
6cd525e8
MD
1457 ret = lttng_write(fd, (void *) &hdr, sizeof(hdr));
1458 if (ret < sizeof(hdr)) {
d7b75ec8 1459 /*
6f04ed72 1460 * This error means that the fd's end is closed so ignore the PERROR
d7b75ec8
DG
1461 * not to clubber the error output since this can happen in a normal
1462 * code path.
1463 */
1464 if (errno != EPIPE) {
1465 PERROR("write metadata stream id");
1466 }
1467 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno);
534d2592
DG
1468 /*
1469 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1470 * handle writting the missing part so report that as an error and
1471 * don't lie to the caller.
1472 */
1473 ret = -1;
6197aea7
DG
1474 goto end;
1475 }
1d4dfdef
DG
1476 DBG("Metadata stream id %" PRIu64 " with padding %lu written before data",
1477 stream->relayd_stream_id, padding);
6197aea7
DG
1478
1479end:
6cd525e8 1480 return (int) ret;
6197aea7
DG
1481}
1482
3bd1e081 1483/*
09e26845
DG
1484 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1485 * core function for writing trace buffers to either the local filesystem or
1486 * the network.
1487 *
79d4ffb7
DG
1488 * It must be called with the stream lock held.
1489 *
09e26845 1490 * Careful review MUST be put if any changes occur!
3bd1e081
MD
1491 *
1492 * Returns the number of bytes written
1493 */
4078b776 1494ssize_t lttng_consumer_on_read_subbuffer_mmap(
3bd1e081 1495 struct lttng_consumer_local_data *ctx,
1d4dfdef 1496 struct lttng_consumer_stream *stream, unsigned long len,
309167d2 1497 unsigned long padding,
50adc264 1498 struct ctf_packet_index *index)
3bd1e081 1499{
f02e1e8a 1500 unsigned long mmap_offset;
ffe60014 1501 void *mmap_base;
994ab360 1502 ssize_t ret = 0;
f02e1e8a
DG
1503 off_t orig_offset = stream->out_fd_offset;
1504 /* Default is on the disk */
1505 int outfd = stream->out_fd;
f02e1e8a 1506 struct consumer_relayd_sock_pair *relayd = NULL;
8994307f 1507 unsigned int relayd_hang_up = 0;
f02e1e8a
DG
1508
1509 /* RCU lock for the relayd pointer */
1510 rcu_read_lock();
1511
1512 /* Flag that the current stream if set for network streaming. */
da009f2c 1513 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1514 relayd = consumer_find_relayd(stream->net_seq_idx);
1515 if (relayd == NULL) {
56591bac 1516 ret = -EPIPE;
f02e1e8a
DG
1517 goto end;
1518 }
1519 }
1520
1521 /* get the offset inside the fd to mmap */
3bd1e081
MD
1522 switch (consumer_data.type) {
1523 case LTTNG_CONSUMER_KERNEL:
ffe60014 1524 mmap_base = stream->mmap_base;
f02e1e8a 1525 ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset);
994ab360
DG
1526 if (ret < 0) {
1527 ret = -errno;
56591bac 1528 PERROR("tracer ctl get_mmap_read_offset");
56591bac
MD
1529 goto end;
1530 }
f02e1e8a 1531 break;
7753dea8
MD
1532 case LTTNG_CONSUMER32_UST:
1533 case LTTNG_CONSUMER64_UST:
ffe60014
DG
1534 mmap_base = lttng_ustctl_get_mmap_base(stream);
1535 if (!mmap_base) {
1536 ERR("read mmap get mmap base for stream %s", stream->name);
994ab360 1537 ret = -EPERM;
ffe60014
DG
1538 goto end;
1539 }
1540 ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset);
56591bac
MD
1541 if (ret != 0) {
1542 PERROR("tracer ctl get_mmap_read_offset");
994ab360 1543 ret = -EINVAL;
56591bac
MD
1544 goto end;
1545 }
f02e1e8a 1546 break;
3bd1e081
MD
1547 default:
1548 ERR("Unknown consumer_data type");
1549 assert(0);
1550 }
b9182dd9 1551
f02e1e8a
DG
1552 /* Handle stream on the relayd if the output is on the network */
1553 if (relayd) {
1554 unsigned long netlen = len;
1555
1556 /*
1557 * Lock the control socket for the complete duration of the function
1558 * since from this point on we will use the socket.
1559 */
1560 if (stream->metadata_flag) {
1561 /* Metadata requires the control socket. */
1562 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
93ec662e
JD
1563 if (stream->reset_metadata_flag) {
1564 ret = relayd_reset_metadata(&relayd->control_sock,
1565 stream->relayd_stream_id,
1566 stream->metadata_version);
1567 if (ret < 0) {
1568 relayd_hang_up = 1;
1569 goto write_error;
1570 }
1571 stream->reset_metadata_flag = 0;
1572 }
1d4dfdef 1573 netlen += sizeof(struct lttcomm_relayd_metadata_payload);
f02e1e8a
DG
1574 }
1575
1d4dfdef 1576 ret = write_relayd_stream_header(stream, netlen, padding, relayd);
994ab360
DG
1577 if (ret < 0) {
1578 relayd_hang_up = 1;
1579 goto write_error;
1580 }
1581 /* Use the returned socket. */
1582 outfd = ret;
f02e1e8a 1583
994ab360
DG
1584 /* Write metadata stream id before payload */
1585 if (stream->metadata_flag) {
1586 ret = write_relayd_metadata_id(outfd, stream, relayd, padding);
1587 if (ret < 0) {
8994307f
DG
1588 relayd_hang_up = 1;
1589 goto write_error;
1590 }
f02e1e8a 1591 }
1d4dfdef
DG
1592 } else {
1593 /* No streaming, we have to set the len with the full padding */
1594 len += padding;
1624d5b7 1595
93ec662e
JD
1596 if (stream->metadata_flag && stream->reset_metadata_flag) {
1597 ret = utils_truncate_stream_file(stream->out_fd, 0);
1598 if (ret < 0) {
1599 ERR("Reset metadata file");
1600 goto end;
1601 }
1602 stream->reset_metadata_flag = 0;
1603 }
1604
1624d5b7
JD
1605 /*
1606 * Check if we need to change the tracefile before writing the packet.
1607 */
1608 if (stream->chan->tracefile_size > 0 &&
1609 (stream->tracefile_size_current + len) >
1610 stream->chan->tracefile_size) {
fe4477ee
JD
1611 ret = utils_rotate_stream_file(stream->chan->pathname,
1612 stream->name, stream->chan->tracefile_size,
1613 stream->chan->tracefile_count, stream->uid, stream->gid,
309167d2
JD
1614 stream->out_fd, &(stream->tracefile_count_current),
1615 &stream->out_fd);
1624d5b7
JD
1616 if (ret < 0) {
1617 ERR("Rotating output file");
1618 goto end;
1619 }
309167d2
JD
1620 outfd = stream->out_fd;
1621
1622 if (stream->index_fd >= 0) {
adf7a0a1
JG
1623 ret = close(stream->index_fd);
1624 if (ret < 0) {
1625 PERROR("Closing index");
1626 goto end;
1627 }
1628 stream->index_fd = -1;
309167d2
JD
1629 ret = index_create_file(stream->chan->pathname,
1630 stream->name, stream->uid, stream->gid,
1631 stream->chan->tracefile_size,
1632 stream->tracefile_count_current);
1633 if (ret < 0) {
1634 goto end;
1635 }
1636 stream->index_fd = ret;
1637 }
1638
a6976990
DG
1639 /* Reset current size because we just perform a rotation. */
1640 stream->tracefile_size_current = 0;
a1ae300f
JD
1641 stream->out_fd_offset = 0;
1642 orig_offset = 0;
1624d5b7
JD
1643 }
1644 stream->tracefile_size_current += len;
309167d2
JD
1645 if (index) {
1646 index->offset = htobe64(stream->out_fd_offset);
1647 }
f02e1e8a
DG
1648 }
1649
d02b8372
DG
1650 /*
1651 * This call guarantee that len or less is returned. It's impossible to
1652 * receive a ret value that is bigger than len.
1653 */
1654 ret = lttng_write(outfd, mmap_base + mmap_offset, len);
1655 DBG("Consumer mmap write() ret %zd (len %lu)", ret, len);
1656 if (ret < 0 || ((size_t) ret != len)) {
1657 /*
1658 * Report error to caller if nothing was written else at least send the
1659 * amount written.
1660 */
1661 if (ret < 0) {
994ab360 1662 ret = -errno;
f02e1e8a 1663 }
994ab360 1664 relayd_hang_up = 1;
f02e1e8a 1665
d02b8372 1666 /* Socket operation failed. We consider the relayd dead */
994ab360 1667 if (errno == EPIPE || errno == EINVAL || errno == EBADF) {
d02b8372
DG
1668 /*
1669 * This is possible if the fd is closed on the other side
1670 * (outfd) or any write problem. It can be verbose a bit for a
1671 * normal execution if for instance the relayd is stopped
1672 * abruptly. This can happen so set this to a DBG statement.
1673 */
1674 DBG("Consumer mmap write detected relayd hang up");
994ab360
DG
1675 } else {
1676 /* Unhandled error, print it and stop function right now. */
1677 PERROR("Error in write mmap (ret %zd != len %lu)", ret, len);
f02e1e8a 1678 }
994ab360 1679 goto write_error;
d02b8372
DG
1680 }
1681 stream->output_written += ret;
d02b8372
DG
1682
1683 /* This call is useless on a socket so better save a syscall. */
1684 if (!relayd) {
1685 /* This won't block, but will start writeout asynchronously */
1686 lttng_sync_file_range(outfd, stream->out_fd_offset, len,
1687 SYNC_FILE_RANGE_WRITE);
1688 stream->out_fd_offset += len;
f02e1e8a
DG
1689 }
1690 lttng_consumer_sync_trace_file(stream, orig_offset);
1691
8994307f
DG
1692write_error:
1693 /*
1694 * This is a special case that the relayd has closed its socket. Let's
1695 * cleanup the relayd object and all associated streams.
1696 */
1697 if (relayd && relayd_hang_up) {
1698 cleanup_relayd(relayd, ctx);
1699 }
1700
f02e1e8a
DG
1701end:
1702 /* Unlock only if ctrl socket used */
1703 if (relayd && stream->metadata_flag) {
1704 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1705 }
1706
1707 rcu_read_unlock();
994ab360 1708 return ret;
3bd1e081
MD
1709}
1710
1711/*
1712 * Splice the data from the ring buffer to the tracefile.
1713 *
79d4ffb7
DG
1714 * It must be called with the stream lock held.
1715 *
3bd1e081
MD
1716 * Returns the number of bytes spliced.
1717 */
4078b776 1718ssize_t lttng_consumer_on_read_subbuffer_splice(
3bd1e081 1719 struct lttng_consumer_local_data *ctx,
1d4dfdef 1720 struct lttng_consumer_stream *stream, unsigned long len,
309167d2 1721 unsigned long padding,
50adc264 1722 struct ctf_packet_index *index)
3bd1e081 1723{
f02e1e8a
DG
1724 ssize_t ret = 0, written = 0, ret_splice = 0;
1725 loff_t offset = 0;
1726 off_t orig_offset = stream->out_fd_offset;
1727 int fd = stream->wait_fd;
1728 /* Default is on the disk */
1729 int outfd = stream->out_fd;
f02e1e8a 1730 struct consumer_relayd_sock_pair *relayd = NULL;
fb3a43a9 1731 int *splice_pipe;
8994307f 1732 unsigned int relayd_hang_up = 0;
f02e1e8a 1733
3bd1e081
MD
1734 switch (consumer_data.type) {
1735 case LTTNG_CONSUMER_KERNEL:
f02e1e8a 1736 break;
7753dea8
MD
1737 case LTTNG_CONSUMER32_UST:
1738 case LTTNG_CONSUMER64_UST:
f02e1e8a 1739 /* Not supported for user space tracing */
3bd1e081
MD
1740 return -ENOSYS;
1741 default:
1742 ERR("Unknown consumer_data type");
1743 assert(0);
3bd1e081
MD
1744 }
1745
f02e1e8a
DG
1746 /* RCU lock for the relayd pointer */
1747 rcu_read_lock();
1748
1749 /* Flag that the current stream if set for network streaming. */
da009f2c 1750 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1751 relayd = consumer_find_relayd(stream->net_seq_idx);
1752 if (relayd == NULL) {
ad0b0d23 1753 written = -ret;
f02e1e8a
DG
1754 goto end;
1755 }
1756 }
a2361a61 1757 splice_pipe = stream->splice_pipe;
fb3a43a9 1758
f02e1e8a 1759 /* Write metadata stream id before payload */
1d4dfdef 1760 if (relayd) {
ad0b0d23 1761 unsigned long total_len = len;
f02e1e8a 1762
1d4dfdef
DG
1763 if (stream->metadata_flag) {
1764 /*
1765 * Lock the control socket for the complete duration of the function
1766 * since from this point on we will use the socket.
1767 */
1768 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1769
93ec662e
JD
1770 if (stream->reset_metadata_flag) {
1771 ret = relayd_reset_metadata(&relayd->control_sock,
1772 stream->relayd_stream_id,
1773 stream->metadata_version);
1774 if (ret < 0) {
1775 relayd_hang_up = 1;
1776 goto write_error;
1777 }
1778 stream->reset_metadata_flag = 0;
1779 }
1d4dfdef
DG
1780 ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd,
1781 padding);
1782 if (ret < 0) {
1783 written = ret;
ad0b0d23
DG
1784 relayd_hang_up = 1;
1785 goto write_error;
1d4dfdef
DG
1786 }
1787
1788 total_len += sizeof(struct lttcomm_relayd_metadata_payload);
1789 }
1790
1791 ret = write_relayd_stream_header(stream, total_len, padding, relayd);
ad0b0d23
DG
1792 if (ret < 0) {
1793 written = ret;
1794 relayd_hang_up = 1;
1795 goto write_error;
f02e1e8a 1796 }
ad0b0d23
DG
1797 /* Use the returned socket. */
1798 outfd = ret;
1d4dfdef
DG
1799 } else {
1800 /* No streaming, we have to set the len with the full padding */
1801 len += padding;
1624d5b7 1802
93ec662e
JD
1803 if (stream->metadata_flag && stream->reset_metadata_flag) {
1804 ret = utils_truncate_stream_file(stream->out_fd, 0);
1805 if (ret < 0) {
1806 ERR("Reset metadata file");
1807 goto end;
1808 }
1809 stream->reset_metadata_flag = 0;
1810 }
1624d5b7
JD
1811 /*
1812 * Check if we need to change the tracefile before writing the packet.
1813 */
1814 if (stream->chan->tracefile_size > 0 &&
1815 (stream->tracefile_size_current + len) >
1816 stream->chan->tracefile_size) {
fe4477ee
JD
1817 ret = utils_rotate_stream_file(stream->chan->pathname,
1818 stream->name, stream->chan->tracefile_size,
1819 stream->chan->tracefile_count, stream->uid, stream->gid,
309167d2
JD
1820 stream->out_fd, &(stream->tracefile_count_current),
1821 &stream->out_fd);
1624d5b7 1822 if (ret < 0) {
ad0b0d23 1823 written = ret;
1624d5b7
JD
1824 ERR("Rotating output file");
1825 goto end;
1826 }
309167d2
JD
1827 outfd = stream->out_fd;
1828
1829 if (stream->index_fd >= 0) {
3c1150e1
JD
1830 ret = close(stream->index_fd);
1831 if (ret < 0) {
1832 PERROR("Closing index");
1833 goto end;
1834 }
1835 stream->index_fd = -1;
309167d2
JD
1836 ret = index_create_file(stream->chan->pathname,
1837 stream->name, stream->uid, stream->gid,
1838 stream->chan->tracefile_size,
1839 stream->tracefile_count_current);
1840 if (ret < 0) {
ad0b0d23 1841 written = ret;
309167d2
JD
1842 goto end;
1843 }
1844 stream->index_fd = ret;
1845 }
1846
a6976990
DG
1847 /* Reset current size because we just perform a rotation. */
1848 stream->tracefile_size_current = 0;
a1ae300f
JD
1849 stream->out_fd_offset = 0;
1850 orig_offset = 0;
1624d5b7
JD
1851 }
1852 stream->tracefile_size_current += len;
309167d2 1853 index->offset = htobe64(stream->out_fd_offset);
f02e1e8a
DG
1854 }
1855
1856 while (len > 0) {
1d4dfdef
DG
1857 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1858 (unsigned long)offset, len, fd, splice_pipe[1]);
fb3a43a9 1859 ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len,
f02e1e8a
DG
1860 SPLICE_F_MOVE | SPLICE_F_MORE);
1861 DBG("splice chan to pipe, ret %zd", ret_splice);
1862 if (ret_splice < 0) {
d02b8372 1863 ret = errno;
ad0b0d23 1864 written = -ret;
d02b8372 1865 PERROR("Error in relay splice");
f02e1e8a
DG
1866 goto splice_error;
1867 }
1868
1869 /* Handle stream on the relayd if the output is on the network */
ad0b0d23
DG
1870 if (relayd && stream->metadata_flag) {
1871 size_t metadata_payload_size =
1872 sizeof(struct lttcomm_relayd_metadata_payload);
1873
1874 /* Update counter to fit the spliced data */
1875 ret_splice += metadata_payload_size;
1876 len += metadata_payload_size;
1877 /*
1878 * We do this so the return value can match the len passed as
1879 * argument to this function.
1880 */
1881 written -= metadata_payload_size;
f02e1e8a
DG
1882 }
1883
1884 /* Splice data out */
fb3a43a9 1885 ret_splice = splice(splice_pipe[0], NULL, outfd, NULL,
f02e1e8a 1886 ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE);
a2361a61
JD
1887 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1888 outfd, ret_splice);
f02e1e8a 1889 if (ret_splice < 0) {
d02b8372 1890 ret = errno;
ad0b0d23
DG
1891 written = -ret;
1892 relayd_hang_up = 1;
1893 goto write_error;
f02e1e8a 1894 } else if (ret_splice > len) {
d02b8372
DG
1895 /*
1896 * We don't expect this code path to be executed but you never know
1897 * so this is an extra protection agains a buggy splice().
1898 */
f02e1e8a 1899 ret = errno;
ad0b0d23 1900 written += ret_splice;
d02b8372
DG
1901 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice,
1902 len);
f02e1e8a 1903 goto splice_error;
d02b8372
DG
1904 } else {
1905 /* All good, update current len and continue. */
1906 len -= ret_splice;
f02e1e8a 1907 }
f02e1e8a
DG
1908
1909 /* This call is useless on a socket so better save a syscall. */
1910 if (!relayd) {
1911 /* This won't block, but will start writeout asynchronously */
1912 lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice,
1913 SYNC_FILE_RANGE_WRITE);
1914 stream->out_fd_offset += ret_splice;
1915 }
e5d1a9b3 1916 stream->output_written += ret_splice;
f02e1e8a
DG
1917 written += ret_splice;
1918 }
1919 lttng_consumer_sync_trace_file(stream, orig_offset);
f02e1e8a
DG
1920 goto end;
1921
8994307f
DG
1922write_error:
1923 /*
1924 * This is a special case that the relayd has closed its socket. Let's
1925 * cleanup the relayd object and all associated streams.
1926 */
1927 if (relayd && relayd_hang_up) {
1928 cleanup_relayd(relayd, ctx);
1929 /* Skip splice error so the consumer does not fail */
1930 goto end;
1931 }
1932
f02e1e8a
DG
1933splice_error:
1934 /* send the appropriate error description to sessiond */
1935 switch (ret) {
f02e1e8a 1936 case EINVAL:
f73fabfd 1937 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL);
f02e1e8a
DG
1938 break;
1939 case ENOMEM:
f73fabfd 1940 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM);
f02e1e8a
DG
1941 break;
1942 case ESPIPE:
f73fabfd 1943 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE);
f02e1e8a
DG
1944 break;
1945 }
1946
1947end:
1948 if (relayd && stream->metadata_flag) {
1949 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1950 }
1951
1952 rcu_read_unlock();
1953 return written;
3bd1e081
MD
1954}
1955
1956/*
1957 * Take a snapshot for a specific fd
1958 *
1959 * Returns 0 on success, < 0 on error
1960 */
ffe60014 1961int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream)
3bd1e081
MD
1962{
1963 switch (consumer_data.type) {
1964 case LTTNG_CONSUMER_KERNEL:
ffe60014 1965 return lttng_kconsumer_take_snapshot(stream);
7753dea8
MD
1966 case LTTNG_CONSUMER32_UST:
1967 case LTTNG_CONSUMER64_UST:
ffe60014 1968 return lttng_ustconsumer_take_snapshot(stream);
3bd1e081
MD
1969 default:
1970 ERR("Unknown consumer_data type");
1971 assert(0);
1972 return -ENOSYS;
1973 }
3bd1e081
MD
1974}
1975
1976/*
1977 * Get the produced position
1978 *
1979 * Returns 0 on success, < 0 on error
1980 */
ffe60014 1981int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
3bd1e081
MD
1982 unsigned long *pos)
1983{
1984 switch (consumer_data.type) {
1985 case LTTNG_CONSUMER_KERNEL:
ffe60014 1986 return lttng_kconsumer_get_produced_snapshot(stream, pos);
7753dea8
MD
1987 case LTTNG_CONSUMER32_UST:
1988 case LTTNG_CONSUMER64_UST:
ffe60014 1989 return lttng_ustconsumer_get_produced_snapshot(stream, pos);
3bd1e081
MD
1990 default:
1991 ERR("Unknown consumer_data type");
1992 assert(0);
1993 return -ENOSYS;
1994 }
1995}
1996
1997int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1998 int sock, struct pollfd *consumer_sockpoll)
1999{
2000 switch (consumer_data.type) {
2001 case LTTNG_CONSUMER_KERNEL:
2002 return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
7753dea8
MD
2003 case LTTNG_CONSUMER32_UST:
2004 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
2005 return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
2006 default:
2007 ERR("Unknown consumer_data type");
2008 assert(0);
2009 return -ENOSYS;
2010 }
2011}
2012
6d574024 2013void lttng_consumer_close_all_metadata(void)
d88aee68
DG
2014{
2015 switch (consumer_data.type) {
2016 case LTTNG_CONSUMER_KERNEL:
2017 /*
2018 * The Kernel consumer has a different metadata scheme so we don't
2019 * close anything because the stream will be closed by the session
2020 * daemon.
2021 */
2022 break;
2023 case LTTNG_CONSUMER32_UST:
2024 case LTTNG_CONSUMER64_UST:
2025 /*
2026 * Close all metadata streams. The metadata hash table is passed and
2027 * this call iterates over it by closing all wakeup fd. This is safe
2028 * because at this point we are sure that the metadata producer is
2029 * either dead or blocked.
2030 */
6d574024 2031 lttng_ustconsumer_close_all_metadata(metadata_ht);
d88aee68
DG
2032 break;
2033 default:
2034 ERR("Unknown consumer_data type");
2035 assert(0);
2036 }
2037}
2038
fb3a43a9
DG
2039/*
2040 * Clean up a metadata stream and free its memory.
2041 */
e316aad5
DG
2042void consumer_del_metadata_stream(struct lttng_consumer_stream *stream,
2043 struct lttng_ht *ht)
fb3a43a9 2044{
e316aad5 2045 struct lttng_consumer_channel *free_chan = NULL;
fb3a43a9
DG
2046
2047 assert(stream);
2048 /*
2049 * This call should NEVER receive regular stream. It must always be
2050 * metadata stream and this is crucial for data structure synchronization.
2051 */
2052 assert(stream->metadata_flag);
2053
e316aad5
DG
2054 DBG3("Consumer delete metadata stream %d", stream->wait_fd);
2055
74251bb8 2056 pthread_mutex_lock(&consumer_data.lock);
a9838785 2057 pthread_mutex_lock(&stream->chan->lock);
8994307f
DG
2058 pthread_mutex_lock(&stream->lock);
2059
6d574024
DG
2060 /* Remove any reference to that stream. */
2061 consumer_stream_delete(stream, ht);
ca22feea 2062
6d574024
DG
2063 /* Close down everything including the relayd if one. */
2064 consumer_stream_close(stream);
2065 /* Destroy tracer buffers of the stream. */
2066 consumer_stream_destroy_buffers(stream);
fb3a43a9
DG
2067
2068 /* Atomically decrement channel refcount since other threads can use it. */
f2ad556d 2069 if (!uatomic_sub_return(&stream->chan->refcount, 1)
ffe60014 2070 && !uatomic_read(&stream->chan->nb_init_stream_left)) {
c30aaa51 2071 /* Go for channel deletion! */
e316aad5 2072 free_chan = stream->chan;
fb3a43a9
DG
2073 }
2074
73811ecc
DG
2075 /*
2076 * Nullify the stream reference so it is not used after deletion. The
6d574024
DG
2077 * channel lock MUST be acquired before being able to check for a NULL
2078 * pointer value.
73811ecc
DG
2079 */
2080 stream->chan->metadata_stream = NULL;
2081
8994307f 2082 pthread_mutex_unlock(&stream->lock);
a9838785 2083 pthread_mutex_unlock(&stream->chan->lock);
74251bb8 2084 pthread_mutex_unlock(&consumer_data.lock);
e316aad5
DG
2085
2086 if (free_chan) {
2087 consumer_del_channel(free_chan);
2088 }
2089
6d574024 2090 consumer_stream_free(stream);
fb3a43a9
DG
2091}
2092
2093/*
2094 * Action done with the metadata stream when adding it to the consumer internal
2095 * data structures to handle it.
2096 */
5ab66908 2097int consumer_add_metadata_stream(struct lttng_consumer_stream *stream)
fb3a43a9 2098{
5ab66908 2099 struct lttng_ht *ht = metadata_ht;
e316aad5 2100 int ret = 0;
76082088 2101 struct lttng_ht_iter iter;
d88aee68 2102 struct lttng_ht_node_u64 *node;
fb3a43a9 2103
e316aad5
DG
2104 assert(stream);
2105 assert(ht);
2106
d88aee68 2107 DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key);
e316aad5
DG
2108
2109 pthread_mutex_lock(&consumer_data.lock);
a9838785 2110 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 2111 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 2112 pthread_mutex_lock(&stream->lock);
e316aad5 2113
e316aad5
DG
2114 /*
2115 * From here, refcounts are updated so be _careful_ when returning an error
2116 * after this point.
2117 */
2118
fb3a43a9 2119 rcu_read_lock();
76082088
DG
2120
2121 /*
2122 * Lookup the stream just to make sure it does not exist in our internal
2123 * state. This should NEVER happen.
2124 */
d88aee68
DG
2125 lttng_ht_lookup(ht, &stream->key, &iter);
2126 node = lttng_ht_iter_get_node_u64(&iter);
76082088
DG
2127 assert(!node);
2128
e316aad5 2129 /*
ffe60014
DG
2130 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2131 * in terms of destroying the associated channel, because the action that
e316aad5
DG
2132 * causes the count to become 0 also causes a stream to be added. The
2133 * channel deletion will thus be triggered by the following removal of this
2134 * stream.
2135 */
ffe60014 2136 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
2137 /* Increment refcount before decrementing nb_init_stream_left */
2138 cmm_smp_wmb();
ffe60014 2139 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
2140 }
2141
d88aee68 2142 lttng_ht_add_unique_u64(ht, &stream->node);
ca22feea 2143
d8ef542d
MD
2144 lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht,
2145 &stream->node_channel_id);
2146
ca22feea
DG
2147 /*
2148 * Add stream to the stream_list_ht of the consumer data. No need to steal
2149 * the key since the HT does not use it and we allow to add redundant keys
2150 * into this table.
2151 */
d88aee68 2152 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 2153
fb3a43a9 2154 rcu_read_unlock();
e316aad5 2155
2e818a6a 2156 pthread_mutex_unlock(&stream->lock);
a9838785 2157 pthread_mutex_unlock(&stream->chan->lock);
ec6ea7d0 2158 pthread_mutex_unlock(&stream->chan->timer_lock);
e316aad5
DG
2159 pthread_mutex_unlock(&consumer_data.lock);
2160 return ret;
fb3a43a9
DG
2161}
2162
8994307f
DG
2163/*
2164 * Delete data stream that are flagged for deletion (endpoint_status).
2165 */
2166static void validate_endpoint_status_data_stream(void)
2167{
2168 struct lttng_ht_iter iter;
2169 struct lttng_consumer_stream *stream;
2170
2171 DBG("Consumer delete flagged data stream");
2172
2173 rcu_read_lock();
2174 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
2175 /* Validate delete flag of the stream */
79d4ffb7 2176 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2177 continue;
2178 }
2179 /* Delete it right now */
2180 consumer_del_stream(stream, data_ht);
2181 }
2182 rcu_read_unlock();
2183}
2184
2185/*
2186 * Delete metadata stream that are flagged for deletion (endpoint_status).
2187 */
2188static void validate_endpoint_status_metadata_stream(
2189 struct lttng_poll_event *pollset)
2190{
2191 struct lttng_ht_iter iter;
2192 struct lttng_consumer_stream *stream;
2193
2194 DBG("Consumer delete flagged metadata stream");
2195
2196 assert(pollset);
2197
2198 rcu_read_lock();
2199 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
2200 /* Validate delete flag of the stream */
79d4ffb7 2201 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2202 continue;
2203 }
2204 /*
2205 * Remove from pollset so the metadata thread can continue without
2206 * blocking on a deleted stream.
2207 */
2208 lttng_poll_del(pollset, stream->wait_fd);
2209
2210 /* Delete it right now */
2211 consumer_del_metadata_stream(stream, metadata_ht);
2212 }
2213 rcu_read_unlock();
2214}
2215
fb3a43a9
DG
2216/*
2217 * Thread polls on metadata file descriptor and write them on disk or on the
2218 * network.
2219 */
7d980def 2220void *consumer_thread_metadata_poll(void *data)
fb3a43a9 2221{
1fc79fb4 2222 int ret, i, pollfd, err = -1;
fb3a43a9 2223 uint32_t revents, nb_fd;
e316aad5 2224 struct lttng_consumer_stream *stream = NULL;
fb3a43a9 2225 struct lttng_ht_iter iter;
d88aee68 2226 struct lttng_ht_node_u64 *node;
fb3a43a9
DG
2227 struct lttng_poll_event events;
2228 struct lttng_consumer_local_data *ctx = data;
2229 ssize_t len;
2230
2231 rcu_register_thread();
2232
1fc79fb4
MD
2233 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA);
2234
2d57de81
MD
2235 if (testpoint(consumerd_thread_metadata)) {
2236 goto error_testpoint;
2237 }
2238
9ce5646a
MD
2239 health_code_update();
2240
fb3a43a9
DG
2241 DBG("Thread metadata poll started");
2242
fb3a43a9
DG
2243 /* Size is set to 1 for the consumer_metadata pipe */
2244 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2245 if (ret < 0) {
2246 ERR("Poll set creation failed");
d8ef542d 2247 goto end_poll;
fb3a43a9
DG
2248 }
2249
13886d2d
DG
2250 ret = lttng_poll_add(&events,
2251 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN);
fb3a43a9
DG
2252 if (ret < 0) {
2253 goto end;
2254 }
2255
2256 /* Main loop */
2257 DBG("Metadata main loop started");
2258
2259 while (1) {
fb3a43a9 2260restart:
7fa2082e 2261 health_code_update();
9ce5646a 2262 health_poll_entry();
7fa2082e 2263 DBG("Metadata poll wait");
fb3a43a9 2264 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2265 DBG("Metadata poll return from wait with %d fd(s)",
2266 LTTNG_POLL_GETNB(&events));
9ce5646a 2267 health_poll_exit();
40063ead 2268 DBG("Metadata event caught in thread");
fb3a43a9
DG
2269 if (ret < 0) {
2270 if (errno == EINTR) {
40063ead 2271 ERR("Poll EINTR caught");
fb3a43a9
DG
2272 goto restart;
2273 }
d9607cd7
MD
2274 if (LTTNG_POLL_GETNB(&events) == 0) {
2275 err = 0; /* All is OK */
2276 }
2277 goto end;
fb3a43a9
DG
2278 }
2279
0d9c5d77
DG
2280 nb_fd = ret;
2281
e316aad5 2282 /* From here, the event is a metadata wait fd */
fb3a43a9 2283 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2284 health_code_update();
2285
fb3a43a9
DG
2286 revents = LTTNG_POLL_GETEV(&events, i);
2287 pollfd = LTTNG_POLL_GETFD(&events, i);
2288
fd20dac9
MD
2289 if (!revents) {
2290 /* No activity for this FD (poll implementation). */
2291 continue;
2292 }
2293
13886d2d 2294 if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) {
03e43155 2295 if (revents & LPOLLIN) {
13886d2d
DG
2296 ssize_t pipe_len;
2297
2298 pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe,
2299 &stream, sizeof(stream));
6cd525e8 2300 if (pipe_len < sizeof(stream)) {
03e43155
MD
2301 if (pipe_len < 0) {
2302 PERROR("read metadata stream");
2303 }
fb3a43a9 2304 /*
03e43155
MD
2305 * Remove the pipe from the poll set and continue the loop
2306 * since their might be data to consume.
fb3a43a9 2307 */
03e43155
MD
2308 lttng_poll_del(&events,
2309 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2310 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
fb3a43a9
DG
2311 continue;
2312 }
2313
8994307f
DG
2314 /* A NULL stream means that the state has changed. */
2315 if (stream == NULL) {
2316 /* Check for deleted streams. */
2317 validate_endpoint_status_metadata_stream(&events);
3714380f 2318 goto restart;
8994307f
DG
2319 }
2320
fb3a43a9
DG
2321 DBG("Adding metadata stream %d to poll set",
2322 stream->wait_fd);
2323
fb3a43a9
DG
2324 /* Add metadata stream to the global poll events list */
2325 lttng_poll_add(&events, stream->wait_fd,
6d574024 2326 LPOLLIN | LPOLLPRI | LPOLLHUP);
03e43155
MD
2327 } else if (revents & (LPOLLERR | LPOLLHUP)) {
2328 DBG("Metadata thread pipe hung up");
2329 /*
2330 * Remove the pipe from the poll set and continue the loop
2331 * since their might be data to consume.
2332 */
2333 lttng_poll_del(&events,
2334 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2335 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
2336 continue;
2337 } else {
2338 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2339 goto end;
fb3a43a9
DG
2340 }
2341
e316aad5 2342 /* Handle other stream */
fb3a43a9
DG
2343 continue;
2344 }
2345
d09e1200 2346 rcu_read_lock();
d88aee68
DG
2347 {
2348 uint64_t tmp_id = (uint64_t) pollfd;
2349
2350 lttng_ht_lookup(metadata_ht, &tmp_id, &iter);
2351 }
2352 node = lttng_ht_iter_get_node_u64(&iter);
e316aad5 2353 assert(node);
fb3a43a9
DG
2354
2355 stream = caa_container_of(node, struct lttng_consumer_stream,
58b1f425 2356 node);
fb3a43a9 2357
03e43155
MD
2358 if (revents & (LPOLLIN | LPOLLPRI)) {
2359 /* Get the data out of the metadata file descriptor */
2360 DBG("Metadata available on fd %d", pollfd);
2361 assert(stream->wait_fd == pollfd);
2362
2363 do {
2364 health_code_update();
2365
2366 len = ctx->on_buffer_ready(stream, ctx);
2367 /*
2368 * We don't check the return value here since if we get
2369 * a negative len, it means an error occured thus we
2370 * simply remove it from the poll set and free the
2371 * stream.
2372 */
2373 } while (len > 0);
2374
2375 /* It's ok to have an unavailable sub-buffer */
2376 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
2377 /* Clean up stream from consumer and free it. */
2378 lttng_poll_del(&events, stream->wait_fd);
2379 consumer_del_metadata_stream(stream, metadata_ht);
2380 }
2381 } else if (revents & (LPOLLERR | LPOLLHUP)) {
e316aad5 2382 DBG("Metadata fd %d is hup|err.", pollfd);
fb3a43a9
DG
2383 if (!stream->hangup_flush_done
2384 && (consumer_data.type == LTTNG_CONSUMER32_UST
2385 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2386 DBG("Attempting to flush and consume the UST buffers");
2387 lttng_ustconsumer_on_stream_hangup(stream);
2388
2389 /* We just flushed the stream now read it. */
4bb94b75 2390 do {
9ce5646a
MD
2391 health_code_update();
2392
4bb94b75
DG
2393 len = ctx->on_buffer_ready(stream, ctx);
2394 /*
2395 * We don't check the return value here since if we get
2396 * a negative len, it means an error occured thus we
2397 * simply remove it from the poll set and free the
2398 * stream.
2399 */
2400 } while (len > 0);
fb3a43a9
DG
2401 }
2402
fb3a43a9 2403 lttng_poll_del(&events, stream->wait_fd);
e316aad5
DG
2404 /*
2405 * This call update the channel states, closes file descriptors
2406 * and securely free the stream.
2407 */
2408 consumer_del_metadata_stream(stream, metadata_ht);
03e43155
MD
2409 } else {
2410 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
6f2f1a70 2411 rcu_read_unlock();
03e43155 2412 goto end;
fb3a43a9 2413 }
e316aad5 2414 /* Release RCU lock for the stream looked up */
d09e1200 2415 rcu_read_unlock();
fb3a43a9
DG
2416 }
2417 }
2418
1fc79fb4
MD
2419 /* All is OK */
2420 err = 0;
fb3a43a9
DG
2421end:
2422 DBG("Metadata poll thread exiting");
fb3a43a9 2423
d8ef542d
MD
2424 lttng_poll_clean(&events);
2425end_poll:
2d57de81 2426error_testpoint:
1fc79fb4
MD
2427 if (err) {
2428 health_error();
2429 ERR("Health error occurred in %s", __func__);
2430 }
2431 health_unregister(health_consumerd);
fb3a43a9
DG
2432 rcu_unregister_thread();
2433 return NULL;
2434}
2435
3bd1e081 2436/*
e4421fec 2437 * This thread polls the fds in the set to consume the data and write
3bd1e081
MD
2438 * it to tracefile if necessary.
2439 */
7d980def 2440void *consumer_thread_data_poll(void *data)
3bd1e081 2441{
1fc79fb4 2442 int num_rdy, num_hup, high_prio, ret, i, err = -1;
3bd1e081
MD
2443 struct pollfd *pollfd = NULL;
2444 /* local view of the streams */
c869f647 2445 struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL;
3bd1e081
MD
2446 /* local view of consumer_data.fds_count */
2447 int nb_fd = 0;
3bd1e081 2448 struct lttng_consumer_local_data *ctx = data;
00e2e675 2449 ssize_t len;
3bd1e081 2450
e7b994a3
DG
2451 rcu_register_thread();
2452
1fc79fb4
MD
2453 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA);
2454
2d57de81
MD
2455 if (testpoint(consumerd_thread_data)) {
2456 goto error_testpoint;
2457 }
2458
9ce5646a
MD
2459 health_code_update();
2460
4df6c8cb
MD
2461 local_stream = zmalloc(sizeof(struct lttng_consumer_stream *));
2462 if (local_stream == NULL) {
2463 PERROR("local_stream malloc");
2464 goto end;
2465 }
3bd1e081
MD
2466
2467 while (1) {
9ce5646a
MD
2468 health_code_update();
2469
3bd1e081
MD
2470 high_prio = 0;
2471 num_hup = 0;
2472
2473 /*
e4421fec 2474 * the fds set has been updated, we need to update our
3bd1e081
MD
2475 * local array as well
2476 */
2477 pthread_mutex_lock(&consumer_data.lock);
2478 if (consumer_data.need_update) {
0e428499
DG
2479 free(pollfd);
2480 pollfd = NULL;
2481
2482 free(local_stream);
2483 local_stream = NULL;
3bd1e081 2484
02b3d176
DG
2485 /*
2486 * Allocate for all fds +1 for the consumer_data_pipe and +1 for
2487 * wake up pipe.
2488 */
2489 pollfd = zmalloc((consumer_data.stream_count + 2) * sizeof(struct pollfd));
3bd1e081 2490 if (pollfd == NULL) {
7a57cf92 2491 PERROR("pollfd malloc");
3bd1e081
MD
2492 pthread_mutex_unlock(&consumer_data.lock);
2493 goto end;
2494 }
2495
02b3d176 2496 local_stream = zmalloc((consumer_data.stream_count + 2) *
747f8642 2497 sizeof(struct lttng_consumer_stream *));
3bd1e081 2498 if (local_stream == NULL) {
7a57cf92 2499 PERROR("local_stream malloc");
3bd1e081
MD
2500 pthread_mutex_unlock(&consumer_data.lock);
2501 goto end;
2502 }
ffe60014 2503 ret = update_poll_array(ctx, &pollfd, local_stream,
43c34bc3 2504 data_ht);
3bd1e081
MD
2505 if (ret < 0) {
2506 ERR("Error in allocating pollfd or local_outfds");
f73fabfd 2507 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2508 pthread_mutex_unlock(&consumer_data.lock);
2509 goto end;
2510 }
2511 nb_fd = ret;
2512 consumer_data.need_update = 0;
2513 }
2514 pthread_mutex_unlock(&consumer_data.lock);
2515
4078b776
MD
2516 /* No FDs and consumer_quit, consumer_cleanup the thread */
2517 if (nb_fd == 0 && consumer_quit == 1) {
1fc79fb4 2518 err = 0; /* All is OK */
4078b776
MD
2519 goto end;
2520 }
3bd1e081 2521 /* poll on the array of fds */
88f2b785 2522 restart:
02b3d176 2523 DBG("polling on %d fd", nb_fd + 2);
9ce5646a 2524 health_poll_entry();
02b3d176 2525 num_rdy = poll(pollfd, nb_fd + 2, -1);
9ce5646a 2526 health_poll_exit();
3bd1e081
MD
2527 DBG("poll num_rdy : %d", num_rdy);
2528 if (num_rdy == -1) {
88f2b785
MD
2529 /*
2530 * Restart interrupted system call.
2531 */
2532 if (errno == EINTR) {
2533 goto restart;
2534 }
7a57cf92 2535 PERROR("Poll error");
f73fabfd 2536 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2537 goto end;
2538 } else if (num_rdy == 0) {
2539 DBG("Polling thread timed out");
2540 goto end;
2541 }
2542
3bd1e081 2543 /*
50f8ae69 2544 * If the consumer_data_pipe triggered poll go directly to the
00e2e675
DG
2545 * beginning of the loop to update the array. We want to prioritize
2546 * array update over low-priority reads.
3bd1e081 2547 */
509bb1cf 2548 if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) {
ab30f567 2549 ssize_t pipe_readlen;
04fdd819 2550
50f8ae69 2551 DBG("consumer_data_pipe wake up");
acdb9057
DG
2552 pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe,
2553 &new_stream, sizeof(new_stream));
6cd525e8
MD
2554 if (pipe_readlen < sizeof(new_stream)) {
2555 PERROR("Consumer data pipe");
23f5f35d
DG
2556 /* Continue so we can at least handle the current stream(s). */
2557 continue;
2558 }
c869f647
DG
2559
2560 /*
2561 * If the stream is NULL, just ignore it. It's also possible that
2562 * the sessiond poll thread changed the consumer_quit state and is
2563 * waking us up to test it.
2564 */
2565 if (new_stream == NULL) {
8994307f 2566 validate_endpoint_status_data_stream();
c869f647
DG
2567 continue;
2568 }
2569
c869f647 2570 /* Continue to update the local streams and handle prio ones */
3bd1e081
MD
2571 continue;
2572 }
2573
02b3d176
DG
2574 /* Handle wakeup pipe. */
2575 if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) {
2576 char dummy;
2577 ssize_t pipe_readlen;
2578
2579 pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy,
2580 sizeof(dummy));
2581 if (pipe_readlen < 0) {
2582 PERROR("Consumer data wakeup pipe");
2583 }
2584 /* We've been awakened to handle stream(s). */
2585 ctx->has_wakeup = 0;
2586 }
2587
3bd1e081
MD
2588 /* Take care of high priority channels first. */
2589 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2590 health_code_update();
2591
9617607b
DG
2592 if (local_stream[i] == NULL) {
2593 continue;
2594 }
fb3a43a9 2595 if (pollfd[i].revents & POLLPRI) {
d41f73b7
MD
2596 DBG("Urgent read on fd %d", pollfd[i].fd);
2597 high_prio = 1;
4078b776 2598 len = ctx->on_buffer_ready(local_stream[i], ctx);
d41f73b7 2599 /* it's ok to have an unavailable sub-buffer */
b64403e3 2600 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2601 /* Clean the stream and free it. */
2602 consumer_del_stream(local_stream[i], data_ht);
9617607b 2603 local_stream[i] = NULL;
4078b776
MD
2604 } else if (len > 0) {
2605 local_stream[i]->data_read = 1;
d41f73b7 2606 }
3bd1e081
MD
2607 }
2608 }
2609
4078b776
MD
2610 /*
2611 * If we read high prio channel in this loop, try again
2612 * for more high prio data.
2613 */
2614 if (high_prio) {
3bd1e081
MD
2615 continue;
2616 }
2617
2618 /* Take care of low priority channels. */
4078b776 2619 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2620 health_code_update();
2621
9617607b
DG
2622 if (local_stream[i] == NULL) {
2623 continue;
2624 }
4078b776 2625 if ((pollfd[i].revents & POLLIN) ||
02b3d176
DG
2626 local_stream[i]->hangup_flush_done ||
2627 local_stream[i]->has_data) {
4078b776
MD
2628 DBG("Normal read on fd %d", pollfd[i].fd);
2629 len = ctx->on_buffer_ready(local_stream[i], ctx);
2630 /* it's ok to have an unavailable sub-buffer */
b64403e3 2631 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2632 /* Clean the stream and free it. */
2633 consumer_del_stream(local_stream[i], data_ht);
9617607b 2634 local_stream[i] = NULL;
4078b776
MD
2635 } else if (len > 0) {
2636 local_stream[i]->data_read = 1;
2637 }
2638 }
2639 }
2640
2641 /* Handle hangup and errors */
2642 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2643 health_code_update();
2644
9617607b
DG
2645 if (local_stream[i] == NULL) {
2646 continue;
2647 }
4078b776
MD
2648 if (!local_stream[i]->hangup_flush_done
2649 && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL))
2650 && (consumer_data.type == LTTNG_CONSUMER32_UST
2651 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2652 DBG("fd %d is hup|err|nval. Attempting flush and read.",
9617607b 2653 pollfd[i].fd);
4078b776
MD
2654 lttng_ustconsumer_on_stream_hangup(local_stream[i]);
2655 /* Attempt read again, for the data we just flushed. */
2656 local_stream[i]->data_read = 1;
2657 }
2658 /*
2659 * If the poll flag is HUP/ERR/NVAL and we have
2660 * read no data in this pass, we can remove the
2661 * stream from its hash table.
2662 */
2663 if ((pollfd[i].revents & POLLHUP)) {
2664 DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
2665 if (!local_stream[i]->data_read) {
43c34bc3 2666 consumer_del_stream(local_stream[i], data_ht);
9617607b 2667 local_stream[i] = NULL;
4078b776
MD
2668 num_hup++;
2669 }
2670 } else if (pollfd[i].revents & POLLERR) {
2671 ERR("Error returned in polling fd %d.", pollfd[i].fd);
2672 if (!local_stream[i]->data_read) {
43c34bc3 2673 consumer_del_stream(local_stream[i], data_ht);
9617607b 2674 local_stream[i] = NULL;
4078b776
MD
2675 num_hup++;
2676 }
2677 } else if (pollfd[i].revents & POLLNVAL) {
2678 ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
2679 if (!local_stream[i]->data_read) {
43c34bc3 2680 consumer_del_stream(local_stream[i], data_ht);
9617607b 2681 local_stream[i] = NULL;
4078b776 2682 num_hup++;
3bd1e081
MD
2683 }
2684 }
9617607b
DG
2685 if (local_stream[i] != NULL) {
2686 local_stream[i]->data_read = 0;
2687 }
3bd1e081
MD
2688 }
2689 }
1fc79fb4
MD
2690 /* All is OK */
2691 err = 0;
3bd1e081
MD
2692end:
2693 DBG("polling thread exiting");
0e428499
DG
2694 free(pollfd);
2695 free(local_stream);
fb3a43a9
DG
2696
2697 /*
2698 * Close the write side of the pipe so epoll_wait() in
7d980def
DG
2699 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2700 * read side of the pipe. If we close them both, epoll_wait strangely does
2701 * not return and could create a endless wait period if the pipe is the
2702 * only tracked fd in the poll set. The thread will take care of closing
2703 * the read side.
fb3a43a9 2704 */
13886d2d 2705 (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe);
fb3a43a9 2706
2d57de81 2707error_testpoint:
1fc79fb4
MD
2708 if (err) {
2709 health_error();
2710 ERR("Health error occurred in %s", __func__);
2711 }
2712 health_unregister(health_consumerd);
2713
e7b994a3 2714 rcu_unregister_thread();
3bd1e081
MD
2715 return NULL;
2716}
2717
d8ef542d
MD
2718/*
2719 * Close wake-up end of each stream belonging to the channel. This will
2720 * allow the poll() on the stream read-side to detect when the
2721 * write-side (application) finally closes them.
2722 */
2723static
2724void consumer_close_channel_streams(struct lttng_consumer_channel *channel)
2725{
2726 struct lttng_ht *ht;
2727 struct lttng_consumer_stream *stream;
2728 struct lttng_ht_iter iter;
2729
2730 ht = consumer_data.stream_per_chan_id_ht;
2731
2732 rcu_read_lock();
2733 cds_lfht_for_each_entry_duplicate(ht->ht,
2734 ht->hash_fct(&channel->key, lttng_ht_seed),
2735 ht->match_fct, &channel->key,
2736 &iter.iter, stream, node_channel_id.node) {
f2ad556d
MD
2737 /*
2738 * Protect against teardown with mutex.
2739 */
2740 pthread_mutex_lock(&stream->lock);
2741 if (cds_lfht_is_node_deleted(&stream->node.node)) {
2742 goto next;
2743 }
d8ef542d
MD
2744 switch (consumer_data.type) {
2745 case LTTNG_CONSUMER_KERNEL:
2746 break;
2747 case LTTNG_CONSUMER32_UST:
2748 case LTTNG_CONSUMER64_UST:
b4a650f3
DG
2749 if (stream->metadata_flag) {
2750 /* Safe and protected by the stream lock. */
2751 lttng_ustconsumer_close_metadata(stream->chan);
2752 } else {
2753 /*
2754 * Note: a mutex is taken internally within
2755 * liblttng-ust-ctl to protect timer wakeup_fd
2756 * use from concurrent close.
2757 */
2758 lttng_ustconsumer_close_stream_wakeup(stream);
2759 }
d8ef542d
MD
2760 break;
2761 default:
2762 ERR("Unknown consumer_data type");
2763 assert(0);
2764 }
f2ad556d
MD
2765 next:
2766 pthread_mutex_unlock(&stream->lock);
d8ef542d
MD
2767 }
2768 rcu_read_unlock();
2769}
2770
2771static void destroy_channel_ht(struct lttng_ht *ht)
2772{
2773 struct lttng_ht_iter iter;
2774 struct lttng_consumer_channel *channel;
2775 int ret;
2776
2777 if (ht == NULL) {
2778 return;
2779 }
2780
2781 rcu_read_lock();
2782 cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) {
2783 ret = lttng_ht_del(ht, &iter);
2784 assert(ret != 0);
2785 }
2786 rcu_read_unlock();
2787
2788 lttng_ht_destroy(ht);
2789}
2790
2791/*
2792 * This thread polls the channel fds to detect when they are being
2793 * closed. It closes all related streams if the channel is detected as
2794 * closed. It is currently only used as a shim layer for UST because the
2795 * consumerd needs to keep the per-stream wakeup end of pipes open for
2796 * periodical flush.
2797 */
2798void *consumer_thread_channel_poll(void *data)
2799{
1fc79fb4 2800 int ret, i, pollfd, err = -1;
d8ef542d
MD
2801 uint32_t revents, nb_fd;
2802 struct lttng_consumer_channel *chan = NULL;
2803 struct lttng_ht_iter iter;
2804 struct lttng_ht_node_u64 *node;
2805 struct lttng_poll_event events;
2806 struct lttng_consumer_local_data *ctx = data;
2807 struct lttng_ht *channel_ht;
2808
2809 rcu_register_thread();
2810
1fc79fb4
MD
2811 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL);
2812
2d57de81
MD
2813 if (testpoint(consumerd_thread_channel)) {
2814 goto error_testpoint;
2815 }
2816
9ce5646a
MD
2817 health_code_update();
2818
d8ef542d
MD
2819 channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2820 if (!channel_ht) {
2821 /* ENOMEM at this point. Better to bail out. */
2822 goto end_ht;
2823 }
2824
2825 DBG("Thread channel poll started");
2826
2827 /* Size is set to 1 for the consumer_channel pipe */
2828 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2829 if (ret < 0) {
2830 ERR("Poll set creation failed");
2831 goto end_poll;
2832 }
2833
2834 ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN);
2835 if (ret < 0) {
2836 goto end;
2837 }
2838
2839 /* Main loop */
2840 DBG("Channel main loop started");
2841
2842 while (1) {
d8ef542d 2843restart:
7fa2082e
MD
2844 health_code_update();
2845 DBG("Channel poll wait");
9ce5646a 2846 health_poll_entry();
d8ef542d 2847 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2848 DBG("Channel poll return from wait with %d fd(s)",
2849 LTTNG_POLL_GETNB(&events));
9ce5646a 2850 health_poll_exit();
40063ead 2851 DBG("Channel event caught in thread");
d8ef542d
MD
2852 if (ret < 0) {
2853 if (errno == EINTR) {
40063ead 2854 ERR("Poll EINTR caught");
d8ef542d
MD
2855 goto restart;
2856 }
d9607cd7
MD
2857 if (LTTNG_POLL_GETNB(&events) == 0) {
2858 err = 0; /* All is OK */
2859 }
d8ef542d
MD
2860 goto end;
2861 }
2862
2863 nb_fd = ret;
2864
2865 /* From here, the event is a channel wait fd */
2866 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2867 health_code_update();
2868
d8ef542d
MD
2869 revents = LTTNG_POLL_GETEV(&events, i);
2870 pollfd = LTTNG_POLL_GETFD(&events, i);
2871
d8ef542d 2872 if (!revents) {
fd20dac9 2873 /* No activity for this FD (poll implementation). */
d8ef542d
MD
2874 continue;
2875 }
fd20dac9 2876
d8ef542d 2877 if (pollfd == ctx->consumer_channel_pipe[0]) {
03e43155 2878 if (revents & LPOLLIN) {
d8ef542d 2879 enum consumer_channel_action action;
a0cbdd2e 2880 uint64_t key;
d8ef542d 2881
a0cbdd2e 2882 ret = read_channel_pipe(ctx, &chan, &key, &action);
d8ef542d 2883 if (ret <= 0) {
03e43155
MD
2884 if (ret < 0) {
2885 ERR("Error reading channel pipe");
2886 }
2887 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
d8ef542d
MD
2888 continue;
2889 }
2890
2891 switch (action) {
2892 case CONSUMER_CHANNEL_ADD:
2893 DBG("Adding channel %d to poll set",
2894 chan->wait_fd);
2895
2896 lttng_ht_node_init_u64(&chan->wait_fd_node,
2897 chan->wait_fd);
c7260a81 2898 rcu_read_lock();
d8ef542d
MD
2899 lttng_ht_add_unique_u64(channel_ht,
2900 &chan->wait_fd_node);
c7260a81 2901 rcu_read_unlock();
d8ef542d
MD
2902 /* Add channel to the global poll events list */
2903 lttng_poll_add(&events, chan->wait_fd,
03e43155 2904 LPOLLERR | LPOLLHUP);
d8ef542d 2905 break;
a0cbdd2e
MD
2906 case CONSUMER_CHANNEL_DEL:
2907 {
b4a650f3
DG
2908 /*
2909 * This command should never be called if the channel
2910 * has streams monitored by either the data or metadata
2911 * thread. The consumer only notify this thread with a
2912 * channel del. command if it receives a destroy
2913 * channel command from the session daemon that send it
2914 * if a command prior to the GET_CHANNEL failed.
2915 */
2916
c7260a81 2917 rcu_read_lock();
a0cbdd2e
MD
2918 chan = consumer_find_channel(key);
2919 if (!chan) {
c7260a81 2920 rcu_read_unlock();
a0cbdd2e
MD
2921 ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key);
2922 break;
2923 }
2924 lttng_poll_del(&events, chan->wait_fd);
f623cc0b 2925 iter.iter.node = &chan->wait_fd_node.node;
a0cbdd2e
MD
2926 ret = lttng_ht_del(channel_ht, &iter);
2927 assert(ret == 0);
a0cbdd2e 2928
f2a444f1
DG
2929 switch (consumer_data.type) {
2930 case LTTNG_CONSUMER_KERNEL:
2931 break;
2932 case LTTNG_CONSUMER32_UST:
2933 case LTTNG_CONSUMER64_UST:
212d67a2
DG
2934 health_code_update();
2935 /* Destroy streams that might have been left in the stream list. */
2936 clean_channel_stream_list(chan);
f2a444f1
DG
2937 break;
2938 default:
2939 ERR("Unknown consumer_data type");
2940 assert(0);
2941 }
2942
a0cbdd2e
MD
2943 /*
2944 * Release our own refcount. Force channel deletion even if
2945 * streams were not initialized.
2946 */
2947 if (!uatomic_sub_return(&chan->refcount, 1)) {
2948 consumer_del_channel(chan);
2949 }
c7260a81 2950 rcu_read_unlock();
a0cbdd2e
MD
2951 goto restart;
2952 }
d8ef542d
MD
2953 case CONSUMER_CHANNEL_QUIT:
2954 /*
2955 * Remove the pipe from the poll set and continue the loop
2956 * since their might be data to consume.
2957 */
2958 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2959 continue;
2960 default:
2961 ERR("Unknown action");
2962 break;
2963 }
03e43155
MD
2964 } else if (revents & (LPOLLERR | LPOLLHUP)) {
2965 DBG("Channel thread pipe hung up");
2966 /*
2967 * Remove the pipe from the poll set and continue the loop
2968 * since their might be data to consume.
2969 */
2970 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2971 continue;
2972 } else {
2973 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2974 goto end;
d8ef542d
MD
2975 }
2976
2977 /* Handle other stream */
2978 continue;
2979 }
2980
2981 rcu_read_lock();
2982 {
2983 uint64_t tmp_id = (uint64_t) pollfd;
2984
2985 lttng_ht_lookup(channel_ht, &tmp_id, &iter);
2986 }
2987 node = lttng_ht_iter_get_node_u64(&iter);
2988 assert(node);
2989
2990 chan = caa_container_of(node, struct lttng_consumer_channel,
2991 wait_fd_node);
2992
2993 /* Check for error event */
2994 if (revents & (LPOLLERR | LPOLLHUP)) {
2995 DBG("Channel fd %d is hup|err.", pollfd);
2996
2997 lttng_poll_del(&events, chan->wait_fd);
2998 ret = lttng_ht_del(channel_ht, &iter);
2999 assert(ret == 0);
b4a650f3
DG
3000
3001 /*
3002 * This will close the wait fd for each stream associated to
3003 * this channel AND monitored by the data/metadata thread thus
3004 * will be clean by the right thread.
3005 */
d8ef542d 3006 consumer_close_channel_streams(chan);
f2ad556d
MD
3007
3008 /* Release our own refcount */
3009 if (!uatomic_sub_return(&chan->refcount, 1)
3010 && !uatomic_read(&chan->nb_init_stream_left)) {
3011 consumer_del_channel(chan);
3012 }
03e43155
MD
3013 } else {
3014 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
3015 rcu_read_unlock();
3016 goto end;
d8ef542d
MD
3017 }
3018
3019 /* Release RCU lock for the channel looked up */
3020 rcu_read_unlock();
3021 }
3022 }
3023
1fc79fb4
MD
3024 /* All is OK */
3025 err = 0;
d8ef542d
MD
3026end:
3027 lttng_poll_clean(&events);
3028end_poll:
3029 destroy_channel_ht(channel_ht);
3030end_ht:
2d57de81 3031error_testpoint:
d8ef542d 3032 DBG("Channel poll thread exiting");
1fc79fb4
MD
3033 if (err) {
3034 health_error();
3035 ERR("Health error occurred in %s", __func__);
3036 }
3037 health_unregister(health_consumerd);
d8ef542d
MD
3038 rcu_unregister_thread();
3039 return NULL;
3040}
3041
331744e3
JD
3042static int set_metadata_socket(struct lttng_consumer_local_data *ctx,
3043 struct pollfd *sockpoll, int client_socket)
3044{
3045 int ret;
3046
3047 assert(ctx);
3048 assert(sockpoll);
3049
84382d49
MD
3050 ret = lttng_consumer_poll_socket(sockpoll);
3051 if (ret) {
331744e3
JD
3052 goto error;
3053 }
3054 DBG("Metadata connection on client_socket");
3055
3056 /* Blocking call, waiting for transmission */
3057 ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket);
3058 if (ctx->consumer_metadata_socket < 0) {
3059 WARN("On accept metadata");
3060 ret = -1;
3061 goto error;
3062 }
3063 ret = 0;
3064
3065error:
3066 return ret;
3067}
3068
3bd1e081
MD
3069/*
3070 * This thread listens on the consumerd socket and receives the file
3071 * descriptors from the session daemon.
3072 */
7d980def 3073void *consumer_thread_sessiond_poll(void *data)
3bd1e081 3074{
1fc79fb4 3075 int sock = -1, client_socket, ret, err = -1;
3bd1e081
MD
3076 /*
3077 * structure to poll for incoming data on communication socket avoids
3078 * making blocking sockets.
3079 */
3080 struct pollfd consumer_sockpoll[2];
3081 struct lttng_consumer_local_data *ctx = data;
3082
e7b994a3
DG
3083 rcu_register_thread();
3084
1fc79fb4
MD
3085 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND);
3086
2d57de81
MD
3087 if (testpoint(consumerd_thread_sessiond)) {
3088 goto error_testpoint;
3089 }
3090
9ce5646a
MD
3091 health_code_update();
3092
3bd1e081
MD
3093 DBG("Creating command socket %s", ctx->consumer_command_sock_path);
3094 unlink(ctx->consumer_command_sock_path);
3095 client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path);
3096 if (client_socket < 0) {
3097 ERR("Cannot create command socket");
3098 goto end;
3099 }
3100
3101 ret = lttcomm_listen_unix_sock(client_socket);
3102 if (ret < 0) {
3103 goto end;
3104 }
3105
32258573 3106 DBG("Sending ready command to lttng-sessiond");
f73fabfd 3107 ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY);
3bd1e081
MD
3108 /* return < 0 on error, but == 0 is not fatal */
3109 if (ret < 0) {
32258573 3110 ERR("Error sending ready command to lttng-sessiond");
3bd1e081
MD
3111 goto end;
3112 }
3113
3bd1e081
MD
3114 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3115 consumer_sockpoll[0].fd = ctx->consumer_should_quit[0];
3116 consumer_sockpoll[0].events = POLLIN | POLLPRI;
3117 consumer_sockpoll[1].fd = client_socket;
3118 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3119
84382d49
MD
3120 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3121 if (ret) {
3122 if (ret > 0) {
3123 /* should exit */
3124 err = 0;
3125 }
3bd1e081
MD
3126 goto end;
3127 }
3128 DBG("Connection on client_socket");
3129
3130 /* Blocking call, waiting for transmission */
3131 sock = lttcomm_accept_unix_sock(client_socket);
534d2592 3132 if (sock < 0) {
3bd1e081
MD
3133 WARN("On accept");
3134 goto end;
3135 }
3bd1e081 3136
331744e3
JD
3137 /*
3138 * Setup metadata socket which is the second socket connection on the
3139 * command unix socket.
3140 */
3141 ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket);
84382d49
MD
3142 if (ret) {
3143 if (ret > 0) {
3144 /* should exit */
3145 err = 0;
3146 }
331744e3
JD
3147 goto end;
3148 }
3149
d96f09c6
DG
3150 /* This socket is not useful anymore. */
3151 ret = close(client_socket);
3152 if (ret < 0) {
3153 PERROR("close client_socket");
3154 }
3155 client_socket = -1;
3156
3bd1e081
MD
3157 /* update the polling structure to poll on the established socket */
3158 consumer_sockpoll[1].fd = sock;
3159 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3160
3161 while (1) {
9ce5646a
MD
3162 health_code_update();
3163
3164 health_poll_entry();
3165 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3166 health_poll_exit();
84382d49
MD
3167 if (ret) {
3168 if (ret > 0) {
3169 /* should exit */
3170 err = 0;
3171 }
3bd1e081
MD
3172 goto end;
3173 }
3174 DBG("Incoming command on sock");
3175 ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll);
4cbc1a04
DG
3176 if (ret <= 0) {
3177 /*
3178 * This could simply be a session daemon quitting. Don't output
3179 * ERR() here.
3180 */
3181 DBG("Communication interrupted on command socket");
41ba6035 3182 err = 0;
3bd1e081
MD
3183 goto end;
3184 }
3185 if (consumer_quit) {
3186 DBG("consumer_thread_receive_fds received quit from signal");
1fc79fb4 3187 err = 0; /* All is OK */
3bd1e081
MD
3188 goto end;
3189 }
ffe60014 3190 DBG("received command on sock");
3bd1e081 3191 }
1fc79fb4
MD
3192 /* All is OK */
3193 err = 0;
3194
3bd1e081 3195end:
ffe60014 3196 DBG("Consumer thread sessiond poll exiting");
3bd1e081 3197
d88aee68
DG
3198 /*
3199 * Close metadata streams since the producer is the session daemon which
3200 * just died.
3201 *
3202 * NOTE: for now, this only applies to the UST tracer.
3203 */
6d574024 3204 lttng_consumer_close_all_metadata();
d88aee68 3205
3bd1e081
MD
3206 /*
3207 * when all fds have hung up, the polling thread
3208 * can exit cleanly
3209 */
3210 consumer_quit = 1;
3211
04fdd819 3212 /*
c869f647 3213 * Notify the data poll thread to poll back again and test the
8994307f 3214 * consumer_quit state that we just set so to quit gracefully.
04fdd819 3215 */
acdb9057 3216 notify_thread_lttng_pipe(ctx->consumer_data_pipe);
c869f647 3217
a0cbdd2e 3218 notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT);
d8ef542d 3219
5c635c72
MD
3220 notify_health_quit_pipe(health_quit_pipe);
3221
d96f09c6
DG
3222 /* Cleaning up possibly open sockets. */
3223 if (sock >= 0) {
3224 ret = close(sock);
3225 if (ret < 0) {
3226 PERROR("close sock sessiond poll");
3227 }
3228 }
3229 if (client_socket >= 0) {
38476d24 3230 ret = close(client_socket);
d96f09c6
DG
3231 if (ret < 0) {
3232 PERROR("close client_socket sessiond poll");
3233 }
3234 }
3235
2d57de81 3236error_testpoint:
1fc79fb4
MD
3237 if (err) {
3238 health_error();
3239 ERR("Health error occurred in %s", __func__);
3240 }
3241 health_unregister(health_consumerd);
3242
e7b994a3 3243 rcu_unregister_thread();
3bd1e081
MD
3244 return NULL;
3245}
d41f73b7 3246
4078b776 3247ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream,
d41f73b7
MD
3248 struct lttng_consumer_local_data *ctx)
3249{
74251bb8
DG
3250 ssize_t ret;
3251
3252 pthread_mutex_lock(&stream->lock);
94d49140
JD
3253 if (stream->metadata_flag) {
3254 pthread_mutex_lock(&stream->metadata_rdv_lock);
3255 }
74251bb8 3256