Clean-up: consumer: fix -Wshadow error in lttng_consumer_rotate_channel
[lttng-tools.git] / src / common / consumer / consumer.c
CommitLineData
3bd1e081 1/*
ab5be9fa
MJ
2 * Copyright (C) 2011 Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * Copyright (C) 2012 David Goulet <dgoulet@efficios.com>
3bd1e081 5 *
ab5be9fa 6 * SPDX-License-Identifier: GPL-2.0-only
3bd1e081 7 *
3bd1e081
MD
8 */
9
6f9449c2 10#include "common/index/ctf-index.h"
6c1c0768 11#define _LGPL_SOURCE
3bd1e081 12#include <assert.h>
3bd1e081
MD
13#include <poll.h>
14#include <pthread.h>
15#include <stdlib.h>
16#include <string.h>
17#include <sys/mman.h>
18#include <sys/socket.h>
19#include <sys/types.h>
20#include <unistd.h>
77c7c900 21#include <inttypes.h>
331744e3 22#include <signal.h>
3bd1e081 23
51a9e1c7 24#include <bin/lttng-consumerd/health-consumerd.h>
990570ed 25#include <common/common.h>
fb3a43a9 26#include <common/utils.h>
d2956687 27#include <common/time.h>
fb3a43a9 28#include <common/compat/poll.h>
f263b7fd 29#include <common/compat/endian.h>
309167d2 30#include <common/index/index.h>
10a8a223 31#include <common/kernel-ctl/kernel-ctl.h>
00e2e675 32#include <common/sessiond-comm/relayd.h>
10a8a223
DG
33#include <common/sessiond-comm/sessiond-comm.h>
34#include <common/kernel-consumer/kernel-consumer.h>
00e2e675 35#include <common/relayd/relayd.h>
10a8a223 36#include <common/ust-consumer/ust-consumer.h>
c8fea79c
JR
37#include <common/consumer/consumer-timer.h>
38#include <common/consumer/consumer.h>
39#include <common/consumer/consumer-stream.h>
40#include <common/consumer/consumer-testpoint.h>
41#include <common/align.h>
5feafd41 42#include <common/consumer/consumer-metadata-cache.h>
d2956687
JG
43#include <common/trace-chunk.h>
44#include <common/trace-chunk-registry.h>
45#include <common/string-utils/format.h>
c35f9726 46#include <common/dynamic-array.h>
3bd1e081
MD
47
48struct lttng_consumer_global_data consumer_data = {
3bd1e081
MD
49 .stream_count = 0,
50 .need_update = 1,
51 .type = LTTNG_CONSUMER_UNKNOWN,
52};
53
d8ef542d
MD
54enum consumer_channel_action {
55 CONSUMER_CHANNEL_ADD,
a0cbdd2e 56 CONSUMER_CHANNEL_DEL,
d8ef542d
MD
57 CONSUMER_CHANNEL_QUIT,
58};
59
60struct consumer_channel_msg {
61 enum consumer_channel_action action;
a0cbdd2e
MD
62 struct lttng_consumer_channel *chan; /* add */
63 uint64_t key; /* del */
d8ef542d
MD
64};
65
80957876 66/* Flag used to temporarily pause data consumption from testpoints. */
cf0bcb51
JG
67int data_consumption_paused;
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 */
10211f5c 75int 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
5da88b0f
MD
85static const char *get_consumer_domain(void)
86{
87 switch (consumer_data.type) {
88 case LTTNG_CONSUMER_KERNEL:
89 return DEFAULT_KERNEL_TRACE_DIR;
90 case LTTNG_CONSUMER64_UST:
91 /* Fall-through. */
92 case LTTNG_CONSUMER32_UST:
93 return DEFAULT_UST_TRACE_DIR;
94 default:
95 abort();
96 }
97}
98
acdb9057
DG
99/*
100 * Notify a thread lttng pipe to poll back again. This usually means that some
101 * global state has changed so we just send back the thread in a poll wait
102 * call.
103 */
104static void notify_thread_lttng_pipe(struct lttng_pipe *pipe)
105{
106 struct lttng_consumer_stream *null_stream = NULL;
107
108 assert(pipe);
109
110 (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream));
111}
112
5c635c72
MD
113static void notify_health_quit_pipe(int *pipe)
114{
6cd525e8 115 ssize_t ret;
5c635c72 116
6cd525e8
MD
117 ret = lttng_write(pipe[1], "4", 1);
118 if (ret < 1) {
5c635c72
MD
119 PERROR("write consumer health quit");
120 }
121}
122
d8ef542d
MD
123static void notify_channel_pipe(struct lttng_consumer_local_data *ctx,
124 struct lttng_consumer_channel *chan,
a0cbdd2e 125 uint64_t key,
d8ef542d
MD
126 enum consumer_channel_action action)
127{
128 struct consumer_channel_msg msg;
6cd525e8 129 ssize_t ret;
d8ef542d 130
e56251fc
DG
131 memset(&msg, 0, sizeof(msg));
132
d8ef542d
MD
133 msg.action = action;
134 msg.chan = chan;
f21dae48 135 msg.key = key;
6cd525e8
MD
136 ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg));
137 if (ret < sizeof(msg)) {
138 PERROR("notify_channel_pipe write error");
139 }
d8ef542d
MD
140}
141
a0cbdd2e
MD
142void notify_thread_del_channel(struct lttng_consumer_local_data *ctx,
143 uint64_t key)
144{
145 notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL);
146}
147
d8ef542d
MD
148static int read_channel_pipe(struct lttng_consumer_local_data *ctx,
149 struct lttng_consumer_channel **chan,
a0cbdd2e 150 uint64_t *key,
d8ef542d
MD
151 enum consumer_channel_action *action)
152{
153 struct consumer_channel_msg msg;
6cd525e8 154 ssize_t ret;
d8ef542d 155
6cd525e8
MD
156 ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg));
157 if (ret < sizeof(msg)) {
158 ret = -1;
159 goto error;
d8ef542d 160 }
6cd525e8
MD
161 *action = msg.action;
162 *chan = msg.chan;
163 *key = msg.key;
164error:
165 return (int) ret;
d8ef542d
MD
166}
167
212d67a2
DG
168/*
169 * Cleanup the stream list of a channel. Those streams are not yet globally
170 * visible
171 */
172static void clean_channel_stream_list(struct lttng_consumer_channel *channel)
173{
174 struct lttng_consumer_stream *stream, *stmp;
175
176 assert(channel);
177
178 /* Delete streams that might have been left in the stream list. */
179 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
180 send_node) {
181 cds_list_del(&stream->send_node);
182 /*
183 * Once a stream is added to this list, the buffers were created so we
184 * have a guarantee that this call will succeed. Setting the monitor
185 * mode to 0 so we don't lock nor try to delete the stream from the
186 * global hash table.
187 */
188 stream->monitor = 0;
189 consumer_stream_destroy(stream, NULL);
190 }
191}
192
3bd1e081
MD
193/*
194 * Find a stream. The consumer_data.lock must be locked during this
195 * call.
196 */
d88aee68 197static struct lttng_consumer_stream *find_stream(uint64_t key,
8389e4f8 198 struct lttng_ht *ht)
3bd1e081 199{
e4421fec 200 struct lttng_ht_iter iter;
d88aee68 201 struct lttng_ht_node_u64 *node;
e4421fec 202 struct lttng_consumer_stream *stream = NULL;
3bd1e081 203
8389e4f8
DG
204 assert(ht);
205
d88aee68
DG
206 /* -1ULL keys are lookup failures */
207 if (key == (uint64_t) -1ULL) {
7ad0a0cb 208 return NULL;
7a57cf92 209 }
e4421fec 210
6065ceec
DG
211 rcu_read_lock();
212
d88aee68
DG
213 lttng_ht_lookup(ht, &key, &iter);
214 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
215 if (node != NULL) {
216 stream = caa_container_of(node, struct lttng_consumer_stream, node);
3bd1e081 217 }
e4421fec 218
6065ceec
DG
219 rcu_read_unlock();
220
e4421fec 221 return stream;
3bd1e081
MD
222}
223
da009f2c 224static void steal_stream_key(uint64_t key, struct lttng_ht *ht)
7ad0a0cb
MD
225{
226 struct lttng_consumer_stream *stream;
227
04253271 228 rcu_read_lock();
ffe60014 229 stream = find_stream(key, ht);
04253271 230 if (stream) {
da009f2c 231 stream->key = (uint64_t) -1ULL;
04253271
MD
232 /*
233 * We don't want the lookup to match, but we still need
234 * to iterate on this stream when iterating over the hash table. Just
235 * change the node key.
236 */
da009f2c 237 stream->node.key = (uint64_t) -1ULL;
04253271
MD
238 }
239 rcu_read_unlock();
7ad0a0cb
MD
240}
241
d56db448
DG
242/*
243 * Return a channel object for the given key.
244 *
245 * RCU read side lock MUST be acquired before calling this function and
246 * protects the channel ptr.
247 */
d88aee68 248struct lttng_consumer_channel *consumer_find_channel(uint64_t key)
3bd1e081 249{
e4421fec 250 struct lttng_ht_iter iter;
d88aee68 251 struct lttng_ht_node_u64 *node;
e4421fec 252 struct lttng_consumer_channel *channel = NULL;
3bd1e081 253
d88aee68
DG
254 /* -1ULL keys are lookup failures */
255 if (key == (uint64_t) -1ULL) {
7ad0a0cb 256 return NULL;
7a57cf92 257 }
e4421fec 258
d88aee68
DG
259 lttng_ht_lookup(consumer_data.channel_ht, &key, &iter);
260 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
261 if (node != NULL) {
262 channel = caa_container_of(node, struct lttng_consumer_channel, node);
3bd1e081 263 }
e4421fec
DG
264
265 return channel;
3bd1e081
MD
266}
267
b5a6470f
DG
268/*
269 * There is a possibility that the consumer does not have enough time between
270 * the close of the channel on the session daemon and the cleanup in here thus
271 * once we have a channel add with an existing key, we know for sure that this
272 * channel will eventually get cleaned up by all streams being closed.
273 *
274 * This function just nullifies the already existing channel key.
275 */
276static void steal_channel_key(uint64_t key)
277{
278 struct lttng_consumer_channel *channel;
279
280 rcu_read_lock();
281 channel = consumer_find_channel(key);
282 if (channel) {
283 channel->key = (uint64_t) -1ULL;
284 /*
285 * We don't want the lookup to match, but we still need to iterate on
286 * this channel when iterating over the hash table. Just change the
287 * node key.
288 */
289 channel->node.key = (uint64_t) -1ULL;
290 }
291 rcu_read_unlock();
292}
293
ffe60014 294static void free_channel_rcu(struct rcu_head *head)
702b1ea4 295{
d88aee68
DG
296 struct lttng_ht_node_u64 *node =
297 caa_container_of(head, struct lttng_ht_node_u64, head);
ffe60014
DG
298 struct lttng_consumer_channel *channel =
299 caa_container_of(node, struct lttng_consumer_channel, node);
702b1ea4 300
b83e03c4
MD
301 switch (consumer_data.type) {
302 case LTTNG_CONSUMER_KERNEL:
303 break;
304 case LTTNG_CONSUMER32_UST:
305 case LTTNG_CONSUMER64_UST:
306 lttng_ustconsumer_free_channel(channel);
307 break;
308 default:
309 ERR("Unknown consumer_data type");
310 abort();
311 }
ffe60014 312 free(channel);
702b1ea4
MD
313}
314
00e2e675
DG
315/*
316 * RCU protected relayd socket pair free.
317 */
ffe60014 318static void free_relayd_rcu(struct rcu_head *head)
00e2e675 319{
d88aee68
DG
320 struct lttng_ht_node_u64 *node =
321 caa_container_of(head, struct lttng_ht_node_u64, head);
00e2e675
DG
322 struct consumer_relayd_sock_pair *relayd =
323 caa_container_of(node, struct consumer_relayd_sock_pair, node);
324
8994307f
DG
325 /*
326 * Close all sockets. This is done in the call RCU since we don't want the
327 * socket fds to be reassigned thus potentially creating bad state of the
328 * relayd object.
329 *
330 * We do not have to lock the control socket mutex here since at this stage
331 * there is no one referencing to this relayd object.
332 */
333 (void) relayd_close(&relayd->control_sock);
334 (void) relayd_close(&relayd->data_sock);
335
3a84e2f3 336 pthread_mutex_destroy(&relayd->ctrl_sock_mutex);
00e2e675
DG
337 free(relayd);
338}
339
340/*
341 * Destroy and free relayd socket pair object.
00e2e675 342 */
51230d70 343void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
344{
345 int ret;
346 struct lttng_ht_iter iter;
347
173af62f
DG
348 if (relayd == NULL) {
349 return;
350 }
351
00e2e675
DG
352 DBG("Consumer destroy and close relayd socket pair");
353
354 iter.iter.node = &relayd->node.node;
355 ret = lttng_ht_del(consumer_data.relayd_ht, &iter);
173af62f 356 if (ret != 0) {
8994307f 357 /* We assume the relayd is being or is destroyed */
173af62f
DG
358 return;
359 }
00e2e675 360
00e2e675 361 /* RCU free() call */
ffe60014
DG
362 call_rcu(&relayd->node.head, free_relayd_rcu);
363}
364
365/*
366 * Remove a channel from the global list protected by a mutex. This function is
367 * also responsible for freeing its data structures.
368 */
369void consumer_del_channel(struct lttng_consumer_channel *channel)
370{
ffe60014
DG
371 struct lttng_ht_iter iter;
372
d88aee68 373 DBG("Consumer delete channel key %" PRIu64, channel->key);
ffe60014
DG
374
375 pthread_mutex_lock(&consumer_data.lock);
a9838785 376 pthread_mutex_lock(&channel->lock);
ffe60014 377
212d67a2
DG
378 /* Destroy streams that might have been left in the stream list. */
379 clean_channel_stream_list(channel);
51e762e5 380
d3e2ba59
JD
381 if (channel->live_timer_enabled == 1) {
382 consumer_timer_live_stop(channel);
383 }
e9404c27
JG
384 if (channel->monitor_timer_enabled == 1) {
385 consumer_timer_monitor_stop(channel);
386 }
d3e2ba59 387
ffe60014
DG
388 switch (consumer_data.type) {
389 case LTTNG_CONSUMER_KERNEL:
390 break;
391 case LTTNG_CONSUMER32_UST:
392 case LTTNG_CONSUMER64_UST:
393 lttng_ustconsumer_del_channel(channel);
394 break;
395 default:
396 ERR("Unknown consumer_data type");
397 assert(0);
398 goto end;
399 }
400
d2956687
JG
401 lttng_trace_chunk_put(channel->trace_chunk);
402 channel->trace_chunk = NULL;
5c3892a6 403
d2956687
JG
404 if (channel->is_published) {
405 int ret;
406
407 rcu_read_lock();
408 iter.iter.node = &channel->node.node;
409 ret = lttng_ht_del(consumer_data.channel_ht, &iter);
410 assert(!ret);
ffe60014 411
d2956687
JG
412 iter.iter.node = &channel->channels_by_session_id_ht_node.node;
413 ret = lttng_ht_del(consumer_data.channels_by_session_id_ht,
414 &iter);
415 assert(!ret);
416 rcu_read_unlock();
417 }
418
b6921a17
JG
419 channel->is_deleted = true;
420 call_rcu(&channel->node.head, free_channel_rcu);
ffe60014 421end:
a9838785 422 pthread_mutex_unlock(&channel->lock);
ffe60014 423 pthread_mutex_unlock(&consumer_data.lock);
00e2e675
DG
424}
425
228b5bf7
DG
426/*
427 * Iterate over the relayd hash table and destroy each element. Finally,
428 * destroy the whole hash table.
429 */
430static void cleanup_relayd_ht(void)
431{
432 struct lttng_ht_iter iter;
433 struct consumer_relayd_sock_pair *relayd;
434
435 rcu_read_lock();
436
437 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
438 node.node) {
51230d70 439 consumer_destroy_relayd(relayd);
228b5bf7
DG
440 }
441
228b5bf7 442 rcu_read_unlock();
36b588ed
MD
443
444 lttng_ht_destroy(consumer_data.relayd_ht);
228b5bf7
DG
445}
446
8994307f
DG
447/*
448 * Update the end point status of all streams having the given network sequence
449 * index (relayd index).
450 *
451 * It's atomically set without having the stream mutex locked which is fine
452 * because we handle the write/read race with a pipe wakeup for each thread.
453 */
da009f2c 454static void update_endpoint_status_by_netidx(uint64_t net_seq_idx,
8994307f
DG
455 enum consumer_endpoint_status status)
456{
457 struct lttng_ht_iter iter;
458 struct lttng_consumer_stream *stream;
459
da009f2c 460 DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx);
8994307f
DG
461
462 rcu_read_lock();
463
464 /* Let's begin with metadata */
465 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
466 if (stream->net_seq_idx == net_seq_idx) {
467 uatomic_set(&stream->endpoint_status, status);
468 DBG("Delete flag set to metadata stream %d", stream->wait_fd);
469 }
470 }
471
472 /* Follow up by the data streams */
473 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
474 if (stream->net_seq_idx == net_seq_idx) {
475 uatomic_set(&stream->endpoint_status, status);
476 DBG("Delete flag set to data stream %d", stream->wait_fd);
477 }
478 }
479 rcu_read_unlock();
480}
481
482/*
483 * Cleanup a relayd object by flagging every associated streams for deletion,
484 * destroying the object meaning removing it from the relayd hash table,
485 * closing the sockets and freeing the memory in a RCU call.
486 *
487 * If a local data context is available, notify the threads that the streams'
488 * state have changed.
489 */
9276e5c8 490void lttng_consumer_cleanup_relayd(struct consumer_relayd_sock_pair *relayd)
8994307f 491{
da009f2c 492 uint64_t netidx;
8994307f
DG
493
494 assert(relayd);
495
9276e5c8 496 DBG("Cleaning up relayd object ID %"PRIu64, relayd->net_seq_idx);
9617607b 497
8994307f
DG
498 /* Save the net sequence index before destroying the object */
499 netidx = relayd->net_seq_idx;
500
501 /*
502 * Delete the relayd from the relayd hash table, close the sockets and free
503 * the object in a RCU call.
504 */
51230d70 505 consumer_destroy_relayd(relayd);
8994307f
DG
506
507 /* Set inactive endpoint to all streams */
508 update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE);
509
510 /*
511 * With a local data context, notify the threads that the streams' state
512 * have changed. The write() action on the pipe acts as an "implicit"
513 * memory barrier ordering the updates of the end point status from the
514 * read of this status which happens AFTER receiving this notify.
515 */
9276e5c8
JR
516 notify_thread_lttng_pipe(relayd->ctx->consumer_data_pipe);
517 notify_thread_lttng_pipe(relayd->ctx->consumer_metadata_pipe);
8994307f
DG
518}
519
a6ba4fe1
DG
520/*
521 * Flag a relayd socket pair for destruction. Destroy it if the refcount
522 * reaches zero.
523 *
524 * RCU read side lock MUST be aquired before calling this function.
525 */
526void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd)
527{
528 assert(relayd);
529
530 /* Set destroy flag for this object */
531 uatomic_set(&relayd->destroy_flag, 1);
532
533 /* Destroy the relayd if refcount is 0 */
534 if (uatomic_read(&relayd->refcount) == 0) {
51230d70 535 consumer_destroy_relayd(relayd);
a6ba4fe1
DG
536 }
537}
538
3bd1e081 539/*
1d1a276c
DG
540 * Completly destroy stream from every visiable data structure and the given
541 * hash table if one.
542 *
543 * One this call returns, the stream object is not longer usable nor visible.
3bd1e081 544 */
e316aad5
DG
545void consumer_del_stream(struct lttng_consumer_stream *stream,
546 struct lttng_ht *ht)
3bd1e081 547{
1d1a276c 548 consumer_stream_destroy(stream, ht);
3bd1e081
MD
549}
550
5ab66908
MD
551/*
552 * XXX naming of del vs destroy is all mixed up.
553 */
554void consumer_del_stream_for_data(struct lttng_consumer_stream *stream)
555{
556 consumer_stream_destroy(stream, data_ht);
557}
558
559void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream)
560{
561 consumer_stream_destroy(stream, metadata_ht);
562}
563
d9a2e16e
JD
564void consumer_stream_update_channel_attributes(
565 struct lttng_consumer_stream *stream,
566 struct lttng_consumer_channel *channel)
567{
568 stream->channel_read_only_attributes.tracefile_size =
569 channel->tracefile_size;
d9a2e16e
JD
570}
571
3bd1e081
MD
572/*
573 * Add a stream to the global list protected by a mutex.
574 */
66d583dc 575void consumer_add_data_stream(struct lttng_consumer_stream *stream)
3bd1e081 576{
5ab66908 577 struct lttng_ht *ht = data_ht;
3bd1e081 578
e316aad5 579 assert(stream);
43c34bc3 580 assert(ht);
c77fc10a 581
d88aee68 582 DBG3("Adding consumer stream %" PRIu64, stream->key);
e316aad5
DG
583
584 pthread_mutex_lock(&consumer_data.lock);
a9838785 585 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 586 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 587 pthread_mutex_lock(&stream->lock);
b0b335c8 588 rcu_read_lock();
e316aad5 589
43c34bc3 590 /* Steal stream identifier to avoid having streams with the same key */
ffe60014 591 steal_stream_key(stream->key, ht);
43c34bc3 592
d88aee68 593 lttng_ht_add_unique_u64(ht, &stream->node);
00e2e675 594
d8ef542d
MD
595 lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht,
596 &stream->node_channel_id);
597
ca22feea
DG
598 /*
599 * Add stream to the stream_list_ht of the consumer data. No need to steal
600 * the key since the HT does not use it and we allow to add redundant keys
601 * into this table.
602 */
d88aee68 603 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 604
e316aad5 605 /*
ffe60014
DG
606 * When nb_init_stream_left reaches 0, we don't need to trigger any action
607 * in terms of destroying the associated channel, because the action that
e316aad5
DG
608 * causes the count to become 0 also causes a stream to be added. The
609 * channel deletion will thus be triggered by the following removal of this
610 * stream.
611 */
ffe60014 612 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
613 /* Increment refcount before decrementing nb_init_stream_left */
614 cmm_smp_wmb();
ffe60014 615 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
616 }
617
618 /* Update consumer data once the node is inserted. */
3bd1e081
MD
619 consumer_data.stream_count++;
620 consumer_data.need_update = 1;
621
e316aad5 622 rcu_read_unlock();
2e818a6a 623 pthread_mutex_unlock(&stream->lock);
ec6ea7d0 624 pthread_mutex_unlock(&stream->chan->timer_lock);
a9838785 625 pthread_mutex_unlock(&stream->chan->lock);
3bd1e081 626 pthread_mutex_unlock(&consumer_data.lock);
3bd1e081
MD
627}
628
00e2e675 629/*
3f8e211f
DG
630 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
631 * be acquired before calling this.
00e2e675 632 */
d09e1200 633static int add_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
634{
635 int ret = 0;
d88aee68 636 struct lttng_ht_node_u64 *node;
00e2e675
DG
637 struct lttng_ht_iter iter;
638
ffe60014 639 assert(relayd);
00e2e675 640
00e2e675 641 lttng_ht_lookup(consumer_data.relayd_ht,
d88aee68
DG
642 &relayd->net_seq_idx, &iter);
643 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675 644 if (node != NULL) {
00e2e675
DG
645 goto end;
646 }
d88aee68 647 lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node);
00e2e675 648
00e2e675
DG
649end:
650 return ret;
651}
652
653/*
654 * Allocate and return a consumer relayd socket.
655 */
027a694f 656static struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair(
da009f2c 657 uint64_t net_seq_idx)
00e2e675
DG
658{
659 struct consumer_relayd_sock_pair *obj = NULL;
660
da009f2c
MD
661 /* net sequence index of -1 is a failure */
662 if (net_seq_idx == (uint64_t) -1ULL) {
00e2e675
DG
663 goto error;
664 }
665
666 obj = zmalloc(sizeof(struct consumer_relayd_sock_pair));
667 if (obj == NULL) {
668 PERROR("zmalloc relayd sock");
669 goto error;
670 }
671
672 obj->net_seq_idx = net_seq_idx;
673 obj->refcount = 0;
173af62f 674 obj->destroy_flag = 0;
f96e4545
MD
675 obj->control_sock.sock.fd = -1;
676 obj->data_sock.sock.fd = -1;
d88aee68 677 lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx);
00e2e675
DG
678 pthread_mutex_init(&obj->ctrl_sock_mutex, NULL);
679
680error:
681 return obj;
682}
683
684/*
685 * Find a relayd socket pair in the global consumer data.
686 *
687 * Return the object if found else NULL.
b0b335c8
MD
688 * RCU read-side lock must be held across this call and while using the
689 * returned object.
00e2e675 690 */
d88aee68 691struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key)
00e2e675
DG
692{
693 struct lttng_ht_iter iter;
d88aee68 694 struct lttng_ht_node_u64 *node;
00e2e675
DG
695 struct consumer_relayd_sock_pair *relayd = NULL;
696
697 /* Negative keys are lookup failures */
d88aee68 698 if (key == (uint64_t) -1ULL) {
00e2e675
DG
699 goto error;
700 }
701
d88aee68 702 lttng_ht_lookup(consumer_data.relayd_ht, &key,
00e2e675 703 &iter);
d88aee68 704 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675
DG
705 if (node != NULL) {
706 relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node);
707 }
708
00e2e675
DG
709error:
710 return relayd;
711}
712
10a50311
JD
713/*
714 * Find a relayd and send the stream
715 *
716 * Returns 0 on success, < 0 on error
717 */
718int consumer_send_relayd_stream(struct lttng_consumer_stream *stream,
719 char *path)
720{
721 int ret = 0;
722 struct consumer_relayd_sock_pair *relayd;
723
724 assert(stream);
725 assert(stream->net_seq_idx != -1ULL);
726 assert(path);
727
728 /* The stream is not metadata. Get relayd reference if exists. */
729 rcu_read_lock();
730 relayd = consumer_find_relayd(stream->net_seq_idx);
731 if (relayd != NULL) {
732 /* Add stream on the relayd */
733 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
734 ret = relayd_add_stream(&relayd->control_sock, stream->name,
5da88b0f 735 get_consumer_domain(), path, &stream->relayd_stream_id,
d2956687
JG
736 stream->chan->tracefile_size,
737 stream->chan->tracefile_count,
738 stream->trace_chunk);
10a50311
JD
739 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
740 if (ret < 0) {
9276e5c8
JR
741 ERR("Relayd add stream failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
742 lttng_consumer_cleanup_relayd(relayd);
10a50311
JD
743 goto end;
744 }
1c20f0e2 745
10a50311 746 uatomic_inc(&relayd->refcount);
d01178b6 747 stream->sent_to_relayd = 1;
10a50311
JD
748 } else {
749 ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.",
750 stream->key, stream->net_seq_idx);
751 ret = -1;
752 goto end;
753 }
754
755 DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64,
756 stream->name, stream->key, stream->net_seq_idx);
757
758end:
759 rcu_read_unlock();
760 return ret;
761}
762
a4baae1b
JD
763/*
764 * Find a relayd and send the streams sent message
765 *
766 * Returns 0 on success, < 0 on error
767 */
768int consumer_send_relayd_streams_sent(uint64_t net_seq_idx)
769{
770 int ret = 0;
771 struct consumer_relayd_sock_pair *relayd;
772
773 assert(net_seq_idx != -1ULL);
774
775 /* The stream is not metadata. Get relayd reference if exists. */
776 rcu_read_lock();
777 relayd = consumer_find_relayd(net_seq_idx);
778 if (relayd != NULL) {
779 /* Add stream on the relayd */
780 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
781 ret = relayd_streams_sent(&relayd->control_sock);
782 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
783 if (ret < 0) {
9276e5c8
JR
784 ERR("Relayd streams sent failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
785 lttng_consumer_cleanup_relayd(relayd);
a4baae1b
JD
786 goto end;
787 }
788 } else {
789 ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.",
790 net_seq_idx);
791 ret = -1;
792 goto end;
793 }
794
795 ret = 0;
796 DBG("All streams sent relayd id %" PRIu64, net_seq_idx);
797
798end:
799 rcu_read_unlock();
800 return ret;
801}
802
10a50311
JD
803/*
804 * Find a relayd and close the stream
805 */
806void close_relayd_stream(struct lttng_consumer_stream *stream)
807{
808 struct consumer_relayd_sock_pair *relayd;
809
810 /* The stream is not metadata. Get relayd reference if exists. */
811 rcu_read_lock();
812 relayd = consumer_find_relayd(stream->net_seq_idx);
813 if (relayd) {
814 consumer_stream_relayd_close(stream, relayd);
815 }
816 rcu_read_unlock();
817}
818
00e2e675
DG
819/*
820 * Handle stream for relayd transmission if the stream applies for network
821 * streaming where the net sequence index is set.
822 *
823 * Return destination file descriptor or negative value on error.
824 */
6197aea7 825static int write_relayd_stream_header(struct lttng_consumer_stream *stream,
1d4dfdef
DG
826 size_t data_size, unsigned long padding,
827 struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
828{
829 int outfd = -1, ret;
00e2e675
DG
830 struct lttcomm_relayd_data_hdr data_hdr;
831
832 /* Safety net */
833 assert(stream);
6197aea7 834 assert(relayd);
00e2e675
DG
835
836 /* Reset data header */
837 memset(&data_hdr, 0, sizeof(data_hdr));
838
00e2e675
DG
839 if (stream->metadata_flag) {
840 /* Caller MUST acquire the relayd control socket lock */
841 ret = relayd_send_metadata(&relayd->control_sock, data_size);
842 if (ret < 0) {
843 goto error;
844 }
845
846 /* Metadata are always sent on the control socket. */
6151a90f 847 outfd = relayd->control_sock.sock.fd;
00e2e675
DG
848 } else {
849 /* Set header with stream information */
850 data_hdr.stream_id = htobe64(stream->relayd_stream_id);
851 data_hdr.data_size = htobe32(data_size);
1d4dfdef 852 data_hdr.padding_size = htobe32(padding);
c35f9726 853
39df6d9f
DG
854 /*
855 * Note that net_seq_num below is assigned with the *current* value of
856 * next_net_seq_num and only after that the next_net_seq_num will be
857 * increment. This is why when issuing a command on the relayd using
858 * this next value, 1 should always be substracted in order to compare
859 * the last seen sequence number on the relayd side to the last sent.
860 */
3604f373 861 data_hdr.net_seq_num = htobe64(stream->next_net_seq_num);
00e2e675
DG
862 /* Other fields are zeroed previously */
863
864 ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr,
865 sizeof(data_hdr));
866 if (ret < 0) {
867 goto error;
868 }
869
3604f373
DG
870 ++stream->next_net_seq_num;
871
00e2e675 872 /* Set to go on data socket */
6151a90f 873 outfd = relayd->data_sock.sock.fd;
00e2e675
DG
874 }
875
876error:
877 return outfd;
878}
879
b1316da1
JG
880/*
881 * Write a character on the metadata poll pipe to wake the metadata thread.
882 * Returns 0 on success, -1 on error.
883 */
884int consumer_metadata_wakeup_pipe(const struct lttng_consumer_channel *channel)
885{
886 int ret = 0;
887
888 DBG("Waking up metadata poll thread (writing to pipe): channel name = '%s'",
889 channel->name);
890 if (channel->monitor && channel->metadata_stream) {
891 const char dummy = 'c';
892 const ssize_t write_ret = lttng_write(
893 channel->metadata_stream->ust_metadata_poll_pipe[1],
894 &dummy, 1);
895
896 if (write_ret < 1) {
897 if (errno == EWOULDBLOCK) {
898 /*
899 * This is fine, the metadata poll thread
900 * is having a hard time keeping-up, but
901 * it will eventually wake-up and consume
902 * the available data.
903 */
904 ret = 0;
905 } else {
906 PERROR("Failed to write to UST metadata pipe while attempting to wake-up the metadata poll thread");
907 ret = -1;
908 goto end;
909 }
910 }
911 }
912
913end:
914 return ret;
915}
916
d2956687
JG
917/*
918 * Trigger a dump of the metadata content. Following/during the succesful
919 * completion of this call, the metadata poll thread will start receiving
920 * metadata packets to consume.
921 *
922 * The caller must hold the channel and stream locks.
923 */
924static
925int consumer_metadata_stream_dump(struct lttng_consumer_stream *stream)
926{
927 int ret;
928
929 ASSERT_LOCKED(stream->chan->lock);
930 ASSERT_LOCKED(stream->lock);
931 assert(stream->metadata_flag);
932 assert(stream->chan->trace_chunk);
933
934 switch (consumer_data.type) {
935 case LTTNG_CONSUMER_KERNEL:
936 /*
937 * Reset the position of what has been read from the
938 * metadata cache to 0 so we can dump it again.
939 */
940 ret = kernctl_metadata_cache_dump(stream->wait_fd);
941 break;
942 case LTTNG_CONSUMER32_UST:
943 case LTTNG_CONSUMER64_UST:
944 /*
945 * Reset the position pushed from the metadata cache so it
946 * will write from the beginning on the next push.
947 */
948 stream->ust_metadata_pushed = 0;
949 ret = consumer_metadata_wakeup_pipe(stream->chan);
950 break;
951 default:
952 ERR("Unknown consumer_data type");
953 abort();
954 }
955 if (ret < 0) {
956 ERR("Failed to dump the metadata cache");
957 }
958 return ret;
959}
960
961static
962int lttng_consumer_channel_set_trace_chunk(
963 struct lttng_consumer_channel *channel,
964 struct lttng_trace_chunk *new_trace_chunk)
965{
d2956687 966 pthread_mutex_lock(&channel->lock);
b6921a17
JG
967 if (channel->is_deleted) {
968 /*
969 * The channel has been logically deleted and should no longer
970 * be used. It has released its reference to its current trace
971 * chunk and should not acquire a new one.
972 *
973 * Return success as there is nothing for the caller to do.
974 */
975 goto end;
976 }
d2956687
JG
977
978 /*
979 * The acquisition of the reference cannot fail (barring
980 * a severe internal error) since a reference to the published
981 * chunk is already held by the caller.
982 */
983 if (new_trace_chunk) {
984 const bool acquired_reference = lttng_trace_chunk_get(
985 new_trace_chunk);
986
987 assert(acquired_reference);
988 }
989
990 lttng_trace_chunk_put(channel->trace_chunk);
991 channel->trace_chunk = new_trace_chunk;
d2956687
JG
992end:
993 pthread_mutex_unlock(&channel->lock);
ce1aa6fe 994 return 0;
d2956687
JG
995}
996
3bd1e081 997/*
ffe60014
DG
998 * Allocate and return a new lttng_consumer_channel object using the given key
999 * to initialize the hash table node.
1000 *
1001 * On error, return NULL.
3bd1e081 1002 */
886224ff 1003struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key,
ffe60014 1004 uint64_t session_id,
d2956687 1005 const uint64_t *chunk_id,
ffe60014
DG
1006 const char *pathname,
1007 const char *name,
57a269f2 1008 uint64_t relayd_id,
1624d5b7
JD
1009 enum lttng_event_output output,
1010 uint64_t tracefile_size,
2bba9e53 1011 uint64_t tracefile_count,
1950109e 1012 uint64_t session_id_per_pid,
ecc48a90 1013 unsigned int monitor,
d7ba1388 1014 unsigned int live_timer_interval,
a2814ea7 1015 bool is_in_live_session,
3d071855 1016 const char *root_shm_path,
d7ba1388 1017 const char *shm_path)
3bd1e081 1018{
d2956687
JG
1019 struct lttng_consumer_channel *channel = NULL;
1020 struct lttng_trace_chunk *trace_chunk = NULL;
1021
1022 if (chunk_id) {
1023 trace_chunk = lttng_trace_chunk_registry_find_chunk(
1024 consumer_data.chunk_registry, session_id,
1025 *chunk_id);
1026 if (!trace_chunk) {
1027 ERR("Failed to find trace chunk reference during creation of channel");
1028 goto end;
1029 }
1030 }
3bd1e081 1031
276b26d1 1032 channel = zmalloc(sizeof(*channel));
3bd1e081 1033 if (channel == NULL) {
7a57cf92 1034 PERROR("malloc struct lttng_consumer_channel");
3bd1e081
MD
1035 goto end;
1036 }
ffe60014
DG
1037
1038 channel->key = key;
3bd1e081 1039 channel->refcount = 0;
ffe60014 1040 channel->session_id = session_id;
1950109e 1041 channel->session_id_per_pid = session_id_per_pid;
ffe60014 1042 channel->relayd_id = relayd_id;
1624d5b7
JD
1043 channel->tracefile_size = tracefile_size;
1044 channel->tracefile_count = tracefile_count;
2bba9e53 1045 channel->monitor = monitor;
ecc48a90 1046 channel->live_timer_interval = live_timer_interval;
a2814ea7 1047 channel->is_live = is_in_live_session;
a9838785 1048 pthread_mutex_init(&channel->lock, NULL);
ec6ea7d0 1049 pthread_mutex_init(&channel->timer_lock, NULL);
ffe60014 1050
0c759fc9
DG
1051 switch (output) {
1052 case LTTNG_EVENT_SPLICE:
1053 channel->output = CONSUMER_CHANNEL_SPLICE;
1054 break;
1055 case LTTNG_EVENT_MMAP:
1056 channel->output = CONSUMER_CHANNEL_MMAP;
1057 break;
1058 default:
1059 assert(0);
1060 free(channel);
1061 channel = NULL;
1062 goto end;
1063 }
1064
07b86b52
JD
1065 /*
1066 * In monitor mode, the streams associated with the channel will be put in
1067 * a special list ONLY owned by this channel. So, the refcount is set to 1
1068 * here meaning that the channel itself has streams that are referenced.
1069 *
1070 * On a channel deletion, once the channel is no longer visible, the
1071 * refcount is decremented and checked for a zero value to delete it. With
1072 * streams in no monitor mode, it will now be safe to destroy the channel.
1073 */
1074 if (!channel->monitor) {
1075 channel->refcount = 1;
1076 }
1077
ffe60014
DG
1078 strncpy(channel->pathname, pathname, sizeof(channel->pathname));
1079 channel->pathname[sizeof(channel->pathname) - 1] = '\0';
1080
1081 strncpy(channel->name, name, sizeof(channel->name));
1082 channel->name[sizeof(channel->name) - 1] = '\0';
1083
3d071855
MD
1084 if (root_shm_path) {
1085 strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path));
1086 channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0';
1087 }
d7ba1388
MD
1088 if (shm_path) {
1089 strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path));
1090 channel->shm_path[sizeof(channel->shm_path) - 1] = '\0';
1091 }
1092
d88aee68 1093 lttng_ht_node_init_u64(&channel->node, channel->key);
5c3892a6
JG
1094 lttng_ht_node_init_u64(&channel->channels_by_session_id_ht_node,
1095 channel->session_id);
d8ef542d
MD
1096
1097 channel->wait_fd = -1;
ffe60014
DG
1098 CDS_INIT_LIST_HEAD(&channel->streams.head);
1099
d2956687
JG
1100 if (trace_chunk) {
1101 int ret = lttng_consumer_channel_set_trace_chunk(channel,
1102 trace_chunk);
1103 if (ret) {
1104 goto error;
1105 }
1106 }
1107
62a7b8ed 1108 DBG("Allocated channel (key %" PRIu64 ")", channel->key);
3bd1e081 1109
3bd1e081 1110end:
d2956687 1111 lttng_trace_chunk_put(trace_chunk);
3bd1e081 1112 return channel;
d2956687
JG
1113error:
1114 consumer_del_channel(channel);
1115 channel = NULL;
1116 goto end;
3bd1e081
MD
1117}
1118
1119/*
1120 * Add a channel to the global list protected by a mutex.
821fffb2 1121 *
b5a6470f 1122 * Always return 0 indicating success.
3bd1e081 1123 */
d8ef542d
MD
1124int consumer_add_channel(struct lttng_consumer_channel *channel,
1125 struct lttng_consumer_local_data *ctx)
3bd1e081 1126{
3bd1e081 1127 pthread_mutex_lock(&consumer_data.lock);
a9838785 1128 pthread_mutex_lock(&channel->lock);
ec6ea7d0 1129 pthread_mutex_lock(&channel->timer_lock);
c77fc10a 1130
b5a6470f
DG
1131 /*
1132 * This gives us a guarantee that the channel we are about to add to the
1133 * channel hash table will be unique. See this function comment on the why
1134 * we need to steel the channel key at this stage.
1135 */
1136 steal_channel_key(channel->key);
c77fc10a 1137
b5a6470f 1138 rcu_read_lock();
d88aee68 1139 lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node);
5c3892a6
JG
1140 lttng_ht_add_u64(consumer_data.channels_by_session_id_ht,
1141 &channel->channels_by_session_id_ht_node);
6065ceec 1142 rcu_read_unlock();
d2956687 1143 channel->is_published = true;
b5a6470f 1144
ec6ea7d0 1145 pthread_mutex_unlock(&channel->timer_lock);
a9838785 1146 pthread_mutex_unlock(&channel->lock);
3bd1e081 1147 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 1148
b5a6470f 1149 if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) {
a0cbdd2e 1150 notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD);
d8ef542d 1151 }
b5a6470f
DG
1152
1153 return 0;
3bd1e081
MD
1154}
1155
1156/*
1157 * Allocate the pollfd structure and the local view of the out fds to avoid
1158 * doing a lookup in the linked list and concurrency issues when writing is
1159 * needed. Called with consumer_data.lock held.
1160 *
1161 * Returns the number of fds in the structures.
1162 */
ffe60014
DG
1163static int update_poll_array(struct lttng_consumer_local_data *ctx,
1164 struct pollfd **pollfd, struct lttng_consumer_stream **local_stream,
9a2fcf78 1165 struct lttng_ht *ht, int *nb_inactive_fd)
3bd1e081 1166{
3bd1e081 1167 int i = 0;
e4421fec
DG
1168 struct lttng_ht_iter iter;
1169 struct lttng_consumer_stream *stream;
3bd1e081 1170
ffe60014
DG
1171 assert(ctx);
1172 assert(ht);
1173 assert(pollfd);
1174 assert(local_stream);
1175
3bd1e081 1176 DBG("Updating poll fd array");
9a2fcf78 1177 *nb_inactive_fd = 0;
481d6c57 1178 rcu_read_lock();
43c34bc3 1179 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
8994307f
DG
1180 /*
1181 * Only active streams with an active end point can be added to the
1182 * poll set and local stream storage of the thread.
1183 *
1184 * There is a potential race here for endpoint_status to be updated
1185 * just after the check. However, this is OK since the stream(s) will
1186 * be deleted once the thread is notified that the end point state has
1187 * changed where this function will be called back again.
9a2fcf78
JD
1188 *
1189 * We track the number of inactive FDs because they still need to be
1190 * closed by the polling thread after a wakeup on the data_pipe or
1191 * metadata_pipe.
8994307f 1192 */
d2956687 1193 if (stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) {
9a2fcf78 1194 (*nb_inactive_fd)++;
3bd1e081
MD
1195 continue;
1196 }
7972aab2
DG
1197 /*
1198 * This clobbers way too much the debug output. Uncomment that if you
1199 * need it for debugging purposes.
7972aab2 1200 */
e4421fec 1201 (*pollfd)[i].fd = stream->wait_fd;
3bd1e081 1202 (*pollfd)[i].events = POLLIN | POLLPRI;
e4421fec 1203 local_stream[i] = stream;
3bd1e081
MD
1204 i++;
1205 }
481d6c57 1206 rcu_read_unlock();
3bd1e081
MD
1207
1208 /*
50f8ae69 1209 * Insert the consumer_data_pipe at the end of the array and don't
3bd1e081
MD
1210 * increment i so nb_fd is the number of real FD.
1211 */
acdb9057 1212 (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe);
509bb1cf 1213 (*pollfd)[i].events = POLLIN | POLLPRI;
02b3d176
DG
1214
1215 (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe);
1216 (*pollfd)[i + 1].events = POLLIN | POLLPRI;
3bd1e081
MD
1217 return i;
1218}
1219
1220/*
84382d49
MD
1221 * Poll on the should_quit pipe and the command socket return -1 on
1222 * error, 1 if should exit, 0 if data is available on the command socket
3bd1e081
MD
1223 */
1224int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll)
1225{
1226 int num_rdy;
1227
88f2b785 1228restart:
3bd1e081
MD
1229 num_rdy = poll(consumer_sockpoll, 2, -1);
1230 if (num_rdy == -1) {
88f2b785
MD
1231 /*
1232 * Restart interrupted system call.
1233 */
1234 if (errno == EINTR) {
1235 goto restart;
1236 }
7a57cf92 1237 PERROR("Poll error");
84382d49 1238 return -1;
3bd1e081 1239 }
509bb1cf 1240 if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) {
3bd1e081 1241 DBG("consumer_should_quit wake up");
84382d49 1242 return 1;
3bd1e081
MD
1243 }
1244 return 0;
3bd1e081
MD
1245}
1246
1247/*
1248 * Set the error socket.
1249 */
ffe60014
DG
1250void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx,
1251 int sock)
3bd1e081
MD
1252{
1253 ctx->consumer_error_socket = sock;
1254}
1255
1256/*
1257 * Set the command socket path.
1258 */
3bd1e081
MD
1259void lttng_consumer_set_command_sock_path(
1260 struct lttng_consumer_local_data *ctx, char *sock)
1261{
1262 ctx->consumer_command_sock_path = sock;
1263}
1264
1265/*
1266 * Send return code to the session daemon.
1267 * If the socket is not defined, we return 0, it is not a fatal error
1268 */
ffe60014 1269int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd)
3bd1e081
MD
1270{
1271 if (ctx->consumer_error_socket > 0) {
1272 return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd,
1273 sizeof(enum lttcomm_sessiond_command));
1274 }
1275
1276 return 0;
1277}
1278
1279/*
228b5bf7
DG
1280 * Close all the tracefiles and stream fds and MUST be called when all
1281 * instances are destroyed i.e. when all threads were joined and are ended.
3bd1e081
MD
1282 */
1283void lttng_consumer_cleanup(void)
1284{
e4421fec 1285 struct lttng_ht_iter iter;
ffe60014 1286 struct lttng_consumer_channel *channel;
e10aec8f 1287 unsigned int trace_chunks_left;
6065ceec
DG
1288
1289 rcu_read_lock();
3bd1e081 1290
ffe60014
DG
1291 cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel,
1292 node.node) {
702b1ea4 1293 consumer_del_channel(channel);
3bd1e081 1294 }
6065ceec
DG
1295
1296 rcu_read_unlock();
d6ce1df2 1297
d6ce1df2 1298 lttng_ht_destroy(consumer_data.channel_ht);
5c3892a6 1299 lttng_ht_destroy(consumer_data.channels_by_session_id_ht);
228b5bf7
DG
1300
1301 cleanup_relayd_ht();
1302
d8ef542d
MD
1303 lttng_ht_destroy(consumer_data.stream_per_chan_id_ht);
1304
228b5bf7
DG
1305 /*
1306 * This HT contains streams that are freed by either the metadata thread or
1307 * the data thread so we do *nothing* on the hash table and simply destroy
1308 * it.
1309 */
1310 lttng_ht_destroy(consumer_data.stream_list_ht);
28cc88f3 1311
e10aec8f
MD
1312 /*
1313 * Trace chunks in the registry may still exist if the session
1314 * daemon has encountered an internal error and could not
1315 * tear down its sessions and/or trace chunks properly.
1316 *
1317 * Release the session daemon's implicit reference to any remaining
1318 * trace chunk and print an error if any trace chunk was found. Note
1319 * that there are _no_ legitimate cases for trace chunks to be left,
1320 * it is a leak. However, it can happen following a crash of the
1321 * session daemon and not emptying the registry would cause an assertion
1322 * to hit.
1323 */
1324 trace_chunks_left = lttng_trace_chunk_registry_put_each_chunk(
1325 consumer_data.chunk_registry);
1326 if (trace_chunks_left) {
1327 ERR("%u trace chunks are leaked by lttng-consumerd. "
1328 "This can be caused by an internal error of the session daemon.",
1329 trace_chunks_left);
1330 }
1331 /* Run all callbacks freeing each chunk. */
1332 rcu_barrier();
28cc88f3 1333 lttng_trace_chunk_registry_destroy(consumer_data.chunk_registry);
3bd1e081
MD
1334}
1335
1336/*
1337 * Called from signal handler.
1338 */
1339void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx)
1340{
6cd525e8
MD
1341 ssize_t ret;
1342
10211f5c 1343 CMM_STORE_SHARED(consumer_quit, 1);
6cd525e8
MD
1344 ret = lttng_write(ctx->consumer_should_quit[1], "4", 1);
1345 if (ret < 1) {
7a57cf92 1346 PERROR("write consumer quit");
3bd1e081 1347 }
ab1027f4
DG
1348
1349 DBG("Consumer flag that it should quit");
3bd1e081
MD
1350}
1351
5199ffc4
JG
1352
1353/*
1354 * Flush pending writes to trace output disk file.
1355 */
1356static
00e2e675
DG
1357void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream,
1358 off_t orig_offset)
3bd1e081 1359{
c7a78aab 1360 int ret;
3bd1e081
MD
1361 int outfd = stream->out_fd;
1362
1363 /*
1364 * This does a blocking write-and-wait on any page that belongs to the
1365 * subbuffer prior to the one we just wrote.
1366 * Don't care about error values, as these are just hints and ways to
1367 * limit the amount of page cache used.
1368 */
ffe60014 1369 if (orig_offset < stream->max_sb_size) {
3bd1e081
MD
1370 return;
1371 }
ffe60014
DG
1372 lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size,
1373 stream->max_sb_size,
3bd1e081
MD
1374 SYNC_FILE_RANGE_WAIT_BEFORE
1375 | SYNC_FILE_RANGE_WRITE
1376 | SYNC_FILE_RANGE_WAIT_AFTER);
1377 /*
1378 * Give hints to the kernel about how we access the file:
1379 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1380 * we write it.
1381 *
1382 * We need to call fadvise again after the file grows because the
1383 * kernel does not seem to apply fadvise to non-existing parts of the
1384 * file.
1385 *
1386 * Call fadvise _after_ having waited for the page writeback to
1387 * complete because the dirty page writeback semantic is not well
1388 * defined. So it can be expected to lead to lower throughput in
1389 * streaming.
1390 */
c7a78aab 1391 ret = posix_fadvise(outfd, orig_offset - stream->max_sb_size,
ffe60014 1392 stream->max_sb_size, POSIX_FADV_DONTNEED);
a0d0e127 1393 if (ret && ret != -ENOSYS) {
a74a5f4a
JG
1394 errno = ret;
1395 PERROR("posix_fadvise on fd %i", outfd);
c7a78aab 1396 }
3bd1e081
MD
1397}
1398
1399/*
1400 * Initialise the necessary environnement :
1401 * - create a new context
1402 * - create the poll_pipe
1403 * - create the should_quit pipe (for signal handler)
1404 * - create the thread pipe (for splice)
1405 *
1406 * Takes a function pointer as argument, this function is called when data is
1407 * available on a buffer. This function is responsible to do the
1408 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1409 * buffer configuration and then kernctl_put_next_subbuf at the end.
1410 *
1411 * Returns a pointer to the new context or NULL on error.
1412 */
1413struct lttng_consumer_local_data *lttng_consumer_create(
1414 enum lttng_consumer_type type,
4078b776 1415 ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream,
6f9449c2 1416 struct lttng_consumer_local_data *ctx, bool locked_by_caller),
3bd1e081
MD
1417 int (*recv_channel)(struct lttng_consumer_channel *channel),
1418 int (*recv_stream)(struct lttng_consumer_stream *stream),
30319bcb 1419 int (*update_stream)(uint64_t stream_key, uint32_t state))
3bd1e081 1420{
d8ef542d 1421 int ret;
3bd1e081
MD
1422 struct lttng_consumer_local_data *ctx;
1423
1424 assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN ||
1425 consumer_data.type == type);
1426 consumer_data.type = type;
1427
effcf122 1428 ctx = zmalloc(sizeof(struct lttng_consumer_local_data));
3bd1e081 1429 if (ctx == NULL) {
7a57cf92 1430 PERROR("allocating context");
3bd1e081
MD
1431 goto error;
1432 }
1433
1434 ctx->consumer_error_socket = -1;
331744e3 1435 ctx->consumer_metadata_socket = -1;
75d83e50 1436 pthread_mutex_init(&ctx->metadata_socket_lock, NULL);
3bd1e081
MD
1437 /* assign the callbacks */
1438 ctx->on_buffer_ready = buffer_ready;
1439 ctx->on_recv_channel = recv_channel;
1440 ctx->on_recv_stream = recv_stream;
1441 ctx->on_update_stream = update_stream;
1442
acdb9057
DG
1443 ctx->consumer_data_pipe = lttng_pipe_open(0);
1444 if (!ctx->consumer_data_pipe) {
3bd1e081
MD
1445 goto error_poll_pipe;
1446 }
1447
02b3d176
DG
1448 ctx->consumer_wakeup_pipe = lttng_pipe_open(0);
1449 if (!ctx->consumer_wakeup_pipe) {
1450 goto error_wakeup_pipe;
1451 }
1452
3bd1e081
MD
1453 ret = pipe(ctx->consumer_should_quit);
1454 if (ret < 0) {
7a57cf92 1455 PERROR("Error creating recv pipe");
3bd1e081
MD
1456 goto error_quit_pipe;
1457 }
1458
d8ef542d
MD
1459 ret = pipe(ctx->consumer_channel_pipe);
1460 if (ret < 0) {
1461 PERROR("Error creating channel pipe");
1462 goto error_channel_pipe;
1463 }
1464
13886d2d
DG
1465 ctx->consumer_metadata_pipe = lttng_pipe_open(0);
1466 if (!ctx->consumer_metadata_pipe) {
fb3a43a9
DG
1467 goto error_metadata_pipe;
1468 }
3bd1e081 1469
e9404c27
JG
1470 ctx->channel_monitor_pipe = -1;
1471
fb3a43a9 1472 return ctx;
3bd1e081 1473
fb3a43a9 1474error_metadata_pipe:
d8ef542d
MD
1475 utils_close_pipe(ctx->consumer_channel_pipe);
1476error_channel_pipe:
d8ef542d 1477 utils_close_pipe(ctx->consumer_should_quit);
3bd1e081 1478error_quit_pipe:
02b3d176
DG
1479 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
1480error_wakeup_pipe:
acdb9057 1481 lttng_pipe_destroy(ctx->consumer_data_pipe);
3bd1e081
MD
1482error_poll_pipe:
1483 free(ctx);
1484error:
1485 return NULL;
1486}
1487
282dadbc
MD
1488/*
1489 * Iterate over all streams of the hashtable and free them properly.
1490 */
1491static void destroy_data_stream_ht(struct lttng_ht *ht)
1492{
1493 struct lttng_ht_iter iter;
1494 struct lttng_consumer_stream *stream;
1495
1496 if (ht == NULL) {
1497 return;
1498 }
1499
1500 rcu_read_lock();
1501 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1502 /*
1503 * Ignore return value since we are currently cleaning up so any error
1504 * can't be handled.
1505 */
1506 (void) consumer_del_stream(stream, ht);
1507 }
1508 rcu_read_unlock();
1509
1510 lttng_ht_destroy(ht);
1511}
1512
1513/*
1514 * Iterate over all streams of the metadata hashtable and free them
1515 * properly.
1516 */
1517static void destroy_metadata_stream_ht(struct lttng_ht *ht)
1518{
1519 struct lttng_ht_iter iter;
1520 struct lttng_consumer_stream *stream;
1521
1522 if (ht == NULL) {
1523 return;
1524 }
1525
1526 rcu_read_lock();
1527 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1528 /*
1529 * Ignore return value since we are currently cleaning up so any error
1530 * can't be handled.
1531 */
1532 (void) consumer_del_metadata_stream(stream, ht);
1533 }
1534 rcu_read_unlock();
1535
1536 lttng_ht_destroy(ht);
1537}
1538
3bd1e081
MD
1539/*
1540 * Close all fds associated with the instance and free the context.
1541 */
1542void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx)
1543{
4c462e79
MD
1544 int ret;
1545
ab1027f4
DG
1546 DBG("Consumer destroying it. Closing everything.");
1547
4f2e75b9
DG
1548 if (!ctx) {
1549 return;
1550 }
1551
282dadbc
MD
1552 destroy_data_stream_ht(data_ht);
1553 destroy_metadata_stream_ht(metadata_ht);
1554
4c462e79
MD
1555 ret = close(ctx->consumer_error_socket);
1556 if (ret) {
1557 PERROR("close");
1558 }
331744e3
JD
1559 ret = close(ctx->consumer_metadata_socket);
1560 if (ret) {
1561 PERROR("close");
1562 }
d8ef542d 1563 utils_close_pipe(ctx->consumer_channel_pipe);
acdb9057 1564 lttng_pipe_destroy(ctx->consumer_data_pipe);
13886d2d 1565 lttng_pipe_destroy(ctx->consumer_metadata_pipe);
02b3d176 1566 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
d8ef542d 1567 utils_close_pipe(ctx->consumer_should_quit);
fb3a43a9 1568
3bd1e081
MD
1569 unlink(ctx->consumer_command_sock_path);
1570 free(ctx);
1571}
1572
6197aea7
DG
1573/*
1574 * Write the metadata stream id on the specified file descriptor.
1575 */
1576static int write_relayd_metadata_id(int fd,
1577 struct lttng_consumer_stream *stream,
239f61af 1578 unsigned long padding)
6197aea7 1579{
6cd525e8 1580 ssize_t ret;
1d4dfdef 1581 struct lttcomm_relayd_metadata_payload hdr;
6197aea7 1582
1d4dfdef
DG
1583 hdr.stream_id = htobe64(stream->relayd_stream_id);
1584 hdr.padding_size = htobe32(padding);
6cd525e8
MD
1585 ret = lttng_write(fd, (void *) &hdr, sizeof(hdr));
1586 if (ret < sizeof(hdr)) {
d7b75ec8 1587 /*
6f04ed72 1588 * This error means that the fd's end is closed so ignore the PERROR
d7b75ec8
DG
1589 * not to clubber the error output since this can happen in a normal
1590 * code path.
1591 */
1592 if (errno != EPIPE) {
1593 PERROR("write metadata stream id");
1594 }
1595 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno);
534d2592
DG
1596 /*
1597 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1598 * handle writting the missing part so report that as an error and
1599 * don't lie to the caller.
1600 */
1601 ret = -1;
6197aea7
DG
1602 goto end;
1603 }
1d4dfdef
DG
1604 DBG("Metadata stream id %" PRIu64 " with padding %lu written before data",
1605 stream->relayd_stream_id, padding);
6197aea7
DG
1606
1607end:
6cd525e8 1608 return (int) ret;
6197aea7
DG
1609}
1610
3bd1e081 1611/*
09e26845
DG
1612 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1613 * core function for writing trace buffers to either the local filesystem or
1614 * the network.
1615 *
d2956687 1616 * It must be called with the stream and the channel lock held.
79d4ffb7 1617 *
09e26845 1618 * Careful review MUST be put if any changes occur!
3bd1e081
MD
1619 *
1620 * Returns the number of bytes written
1621 */
4078b776 1622ssize_t lttng_consumer_on_read_subbuffer_mmap(
128708c3 1623 struct lttng_consumer_stream *stream,
fd424d99 1624 const struct lttng_buffer_view *buffer,
6f9449c2 1625 unsigned long padding)
3bd1e081 1626{
994ab360 1627 ssize_t ret = 0;
f02e1e8a
DG
1628 off_t orig_offset = stream->out_fd_offset;
1629 /* Default is on the disk */
1630 int outfd = stream->out_fd;
f02e1e8a 1631 struct consumer_relayd_sock_pair *relayd = NULL;
8994307f 1632 unsigned int relayd_hang_up = 0;
fd424d99
JG
1633 const size_t subbuf_content_size = buffer->size - padding;
1634 size_t write_len;
f02e1e8a
DG
1635
1636 /* RCU lock for the relayd pointer */
1637 rcu_read_lock();
7fd975c5 1638 assert(stream->net_seq_idx != (uint64_t) -1ULL ||
948411cd 1639 stream->trace_chunk);
d2956687 1640
f02e1e8a 1641 /* Flag that the current stream if set for network streaming. */
da009f2c 1642 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1643 relayd = consumer_find_relayd(stream->net_seq_idx);
1644 if (relayd == NULL) {
56591bac 1645 ret = -EPIPE;
f02e1e8a
DG
1646 goto end;
1647 }
1648 }
1649
f02e1e8a
DG
1650 /* Handle stream on the relayd if the output is on the network */
1651 if (relayd) {
fd424d99 1652 unsigned long netlen = subbuf_content_size;
f02e1e8a
DG
1653
1654 /*
1655 * Lock the control socket for the complete duration of the function
1656 * since from this point on we will use the socket.
1657 */
1658 if (stream->metadata_flag) {
1659 /* Metadata requires the control socket. */
1660 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
93ec662e
JD
1661 if (stream->reset_metadata_flag) {
1662 ret = relayd_reset_metadata(&relayd->control_sock,
1663 stream->relayd_stream_id,
1664 stream->metadata_version);
1665 if (ret < 0) {
1666 relayd_hang_up = 1;
1667 goto write_error;
1668 }
1669 stream->reset_metadata_flag = 0;
1670 }
1d4dfdef 1671 netlen += sizeof(struct lttcomm_relayd_metadata_payload);
f02e1e8a
DG
1672 }
1673
1d4dfdef 1674 ret = write_relayd_stream_header(stream, netlen, padding, relayd);
994ab360
DG
1675 if (ret < 0) {
1676 relayd_hang_up = 1;
1677 goto write_error;
1678 }
1679 /* Use the returned socket. */
1680 outfd = ret;
f02e1e8a 1681
994ab360
DG
1682 /* Write metadata stream id before payload */
1683 if (stream->metadata_flag) {
239f61af 1684 ret = write_relayd_metadata_id(outfd, stream, padding);
994ab360 1685 if (ret < 0) {
8994307f
DG
1686 relayd_hang_up = 1;
1687 goto write_error;
1688 }
f02e1e8a 1689 }
1624d5b7 1690
fd424d99
JG
1691 write_len = subbuf_content_size;
1692 } else {
1693 /* No streaming; we have to write the full padding. */
93ec662e
JD
1694 if (stream->metadata_flag && stream->reset_metadata_flag) {
1695 ret = utils_truncate_stream_file(stream->out_fd, 0);
1696 if (ret < 0) {
1697 ERR("Reset metadata file");
1698 goto end;
1699 }
1700 stream->reset_metadata_flag = 0;
1701 }
1702
1624d5b7
JD
1703 /*
1704 * Check if we need to change the tracefile before writing the packet.
1705 */
1706 if (stream->chan->tracefile_size > 0 &&
fd424d99 1707 (stream->tracefile_size_current + buffer->size) >
1624d5b7 1708 stream->chan->tracefile_size) {
d2956687
JG
1709 ret = consumer_stream_rotate_output_files(stream);
1710 if (ret) {
1624d5b7
JD
1711 goto end;
1712 }
309167d2 1713 outfd = stream->out_fd;
a1ae300f 1714 orig_offset = 0;
1624d5b7 1715 }
fd424d99 1716 stream->tracefile_size_current += buffer->size;
fd424d99 1717 write_len = buffer->size;
f02e1e8a
DG
1718 }
1719
d02b8372
DG
1720 /*
1721 * This call guarantee that len or less is returned. It's impossible to
1722 * receive a ret value that is bigger than len.
1723 */
fd424d99 1724 ret = lttng_write(outfd, buffer->data, write_len);
e2d1190b 1725 DBG("Consumer mmap write() ret %zd (len %zu)", ret, write_len);
fd424d99 1726 if (ret < 0 || ((size_t) ret != write_len)) {
d02b8372
DG
1727 /*
1728 * Report error to caller if nothing was written else at least send the
1729 * amount written.
1730 */
1731 if (ret < 0) {
994ab360 1732 ret = -errno;
f02e1e8a 1733 }
994ab360 1734 relayd_hang_up = 1;
f02e1e8a 1735
d02b8372 1736 /* Socket operation failed. We consider the relayd dead */
fcf0f774 1737 if (errno == EPIPE) {
d02b8372
DG
1738 /*
1739 * This is possible if the fd is closed on the other side
1740 * (outfd) or any write problem. It can be verbose a bit for a
1741 * normal execution if for instance the relayd is stopped
1742 * abruptly. This can happen so set this to a DBG statement.
1743 */
1744 DBG("Consumer mmap write detected relayd hang up");
994ab360
DG
1745 } else {
1746 /* Unhandled error, print it and stop function right now. */
fd424d99
JG
1747 PERROR("Error in write mmap (ret %zd != write_len %zu)", ret,
1748 write_len);
f02e1e8a 1749 }
994ab360 1750 goto write_error;
d02b8372
DG
1751 }
1752 stream->output_written += ret;
d02b8372
DG
1753
1754 /* This call is useless on a socket so better save a syscall. */
1755 if (!relayd) {
1756 /* This won't block, but will start writeout asynchronously */
fd424d99 1757 lttng_sync_file_range(outfd, stream->out_fd_offset, write_len,
d02b8372 1758 SYNC_FILE_RANGE_WRITE);
fd424d99 1759 stream->out_fd_offset += write_len;
f5dbe415 1760 lttng_consumer_sync_trace_file(stream, orig_offset);
f02e1e8a 1761 }
f02e1e8a 1762
8994307f
DG
1763write_error:
1764 /*
1765 * This is a special case that the relayd has closed its socket. Let's
1766 * cleanup the relayd object and all associated streams.
1767 */
1768 if (relayd && relayd_hang_up) {
9276e5c8
JR
1769 ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
1770 lttng_consumer_cleanup_relayd(relayd);
8994307f
DG
1771 }
1772
f02e1e8a
DG
1773end:
1774 /* Unlock only if ctrl socket used */
1775 if (relayd && stream->metadata_flag) {
1776 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1777 }
1778
1779 rcu_read_unlock();
994ab360 1780 return ret;
3bd1e081
MD
1781}
1782
1783/*
1784 * Splice the data from the ring buffer to the tracefile.
1785 *
79d4ffb7
DG
1786 * It must be called with the stream lock held.
1787 *
3bd1e081
MD
1788 * Returns the number of bytes spliced.
1789 */
4078b776 1790ssize_t lttng_consumer_on_read_subbuffer_splice(
3bd1e081 1791 struct lttng_consumer_local_data *ctx,
1d4dfdef 1792 struct lttng_consumer_stream *stream, unsigned long len,
6f9449c2 1793 unsigned long padding)
3bd1e081 1794{
f02e1e8a
DG
1795 ssize_t ret = 0, written = 0, ret_splice = 0;
1796 loff_t offset = 0;
1797 off_t orig_offset = stream->out_fd_offset;
1798 int fd = stream->wait_fd;
1799 /* Default is on the disk */
1800 int outfd = stream->out_fd;
f02e1e8a 1801 struct consumer_relayd_sock_pair *relayd = NULL;
fb3a43a9 1802 int *splice_pipe;
8994307f 1803 unsigned int relayd_hang_up = 0;
f02e1e8a 1804
3bd1e081
MD
1805 switch (consumer_data.type) {
1806 case LTTNG_CONSUMER_KERNEL:
f02e1e8a 1807 break;
7753dea8
MD
1808 case LTTNG_CONSUMER32_UST:
1809 case LTTNG_CONSUMER64_UST:
f02e1e8a 1810 /* Not supported for user space tracing */
3bd1e081
MD
1811 return -ENOSYS;
1812 default:
1813 ERR("Unknown consumer_data type");
1814 assert(0);
3bd1e081
MD
1815 }
1816
f02e1e8a
DG
1817 /* RCU lock for the relayd pointer */
1818 rcu_read_lock();
1819
1820 /* Flag that the current stream if set for network streaming. */
da009f2c 1821 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1822 relayd = consumer_find_relayd(stream->net_seq_idx);
1823 if (relayd == NULL) {
ad0b0d23 1824 written = -ret;
f02e1e8a
DG
1825 goto end;
1826 }
1827 }
a2361a61 1828 splice_pipe = stream->splice_pipe;
fb3a43a9 1829
f02e1e8a 1830 /* Write metadata stream id before payload */
1d4dfdef 1831 if (relayd) {
ad0b0d23 1832 unsigned long total_len = len;
f02e1e8a 1833
1d4dfdef
DG
1834 if (stream->metadata_flag) {
1835 /*
1836 * Lock the control socket for the complete duration of the function
1837 * since from this point on we will use the socket.
1838 */
1839 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1840
93ec662e
JD
1841 if (stream->reset_metadata_flag) {
1842 ret = relayd_reset_metadata(&relayd->control_sock,
1843 stream->relayd_stream_id,
1844 stream->metadata_version);
1845 if (ret < 0) {
1846 relayd_hang_up = 1;
1847 goto write_error;
1848 }
1849 stream->reset_metadata_flag = 0;
1850 }
239f61af 1851 ret = write_relayd_metadata_id(splice_pipe[1], stream,
1d4dfdef
DG
1852 padding);
1853 if (ret < 0) {
1854 written = ret;
ad0b0d23
DG
1855 relayd_hang_up = 1;
1856 goto write_error;
1d4dfdef
DG
1857 }
1858
1859 total_len += sizeof(struct lttcomm_relayd_metadata_payload);
1860 }
1861
1862 ret = write_relayd_stream_header(stream, total_len, padding, relayd);
ad0b0d23
DG
1863 if (ret < 0) {
1864 written = ret;
1865 relayd_hang_up = 1;
1866 goto write_error;
f02e1e8a 1867 }
ad0b0d23
DG
1868 /* Use the returned socket. */
1869 outfd = ret;
1d4dfdef
DG
1870 } else {
1871 /* No streaming, we have to set the len with the full padding */
1872 len += padding;
1624d5b7 1873
93ec662e
JD
1874 if (stream->metadata_flag && stream->reset_metadata_flag) {
1875 ret = utils_truncate_stream_file(stream->out_fd, 0);
1876 if (ret < 0) {
1877 ERR("Reset metadata file");
1878 goto end;
1879 }
1880 stream->reset_metadata_flag = 0;
1881 }
1624d5b7
JD
1882 /*
1883 * Check if we need to change the tracefile before writing the packet.
1884 */
1885 if (stream->chan->tracefile_size > 0 &&
1886 (stream->tracefile_size_current + len) >
1887 stream->chan->tracefile_size) {
d2956687 1888 ret = consumer_stream_rotate_output_files(stream);
1624d5b7 1889 if (ret < 0) {
ad0b0d23 1890 written = ret;
1624d5b7
JD
1891 goto end;
1892 }
309167d2 1893 outfd = stream->out_fd;
a1ae300f 1894 orig_offset = 0;
1624d5b7
JD
1895 }
1896 stream->tracefile_size_current += len;
f02e1e8a
DG
1897 }
1898
1899 while (len > 0) {
1d4dfdef
DG
1900 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1901 (unsigned long)offset, len, fd, splice_pipe[1]);
fb3a43a9 1902 ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len,
f02e1e8a
DG
1903 SPLICE_F_MOVE | SPLICE_F_MORE);
1904 DBG("splice chan to pipe, ret %zd", ret_splice);
1905 if (ret_splice < 0) {
d02b8372 1906 ret = errno;
ad0b0d23 1907 written = -ret;
d02b8372 1908 PERROR("Error in relay splice");
f02e1e8a
DG
1909 goto splice_error;
1910 }
1911
1912 /* Handle stream on the relayd if the output is on the network */
ad0b0d23
DG
1913 if (relayd && stream->metadata_flag) {
1914 size_t metadata_payload_size =
1915 sizeof(struct lttcomm_relayd_metadata_payload);
1916
1917 /* Update counter to fit the spliced data */
1918 ret_splice += metadata_payload_size;
1919 len += metadata_payload_size;
1920 /*
1921 * We do this so the return value can match the len passed as
1922 * argument to this function.
1923 */
1924 written -= metadata_payload_size;
f02e1e8a
DG
1925 }
1926
1927 /* Splice data out */
fb3a43a9 1928 ret_splice = splice(splice_pipe[0], NULL, outfd, NULL,
f02e1e8a 1929 ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE);
a2361a61
JD
1930 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1931 outfd, ret_splice);
f02e1e8a 1932 if (ret_splice < 0) {
d02b8372 1933 ret = errno;
ad0b0d23
DG
1934 written = -ret;
1935 relayd_hang_up = 1;
1936 goto write_error;
f02e1e8a 1937 } else if (ret_splice > len) {
d02b8372
DG
1938 /*
1939 * We don't expect this code path to be executed but you never know
1940 * so this is an extra protection agains a buggy splice().
1941 */
f02e1e8a 1942 ret = errno;
ad0b0d23 1943 written += ret_splice;
d02b8372
DG
1944 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice,
1945 len);
f02e1e8a 1946 goto splice_error;
d02b8372
DG
1947 } else {
1948 /* All good, update current len and continue. */
1949 len -= ret_splice;
f02e1e8a 1950 }
f02e1e8a
DG
1951
1952 /* This call is useless on a socket so better save a syscall. */
1953 if (!relayd) {
1954 /* This won't block, but will start writeout asynchronously */
1955 lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice,
1956 SYNC_FILE_RANGE_WRITE);
1957 stream->out_fd_offset += ret_splice;
1958 }
e5d1a9b3 1959 stream->output_written += ret_splice;
f02e1e8a
DG
1960 written += ret_splice;
1961 }
f5dbe415
JG
1962 if (!relayd) {
1963 lttng_consumer_sync_trace_file(stream, orig_offset);
1964 }
f02e1e8a
DG
1965 goto end;
1966
8994307f
DG
1967write_error:
1968 /*
1969 * This is a special case that the relayd has closed its socket. Let's
1970 * cleanup the relayd object and all associated streams.
1971 */
1972 if (relayd && relayd_hang_up) {
9276e5c8
JR
1973 ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
1974 lttng_consumer_cleanup_relayd(relayd);
8994307f
DG
1975 /* Skip splice error so the consumer does not fail */
1976 goto end;
1977 }
1978
f02e1e8a
DG
1979splice_error:
1980 /* send the appropriate error description to sessiond */
1981 switch (ret) {
f02e1e8a 1982 case EINVAL:
f73fabfd 1983 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL);
f02e1e8a
DG
1984 break;
1985 case ENOMEM:
f73fabfd 1986 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM);
f02e1e8a
DG
1987 break;
1988 case ESPIPE:
f73fabfd 1989 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE);
f02e1e8a
DG
1990 break;
1991 }
1992
1993end:
1994 if (relayd && stream->metadata_flag) {
1995 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1996 }
1997
1998 rcu_read_unlock();
1999 return written;
3bd1e081
MD
2000}
2001
15055ce5
JD
2002/*
2003 * Sample the snapshot positions for a specific fd
2004 *
2005 * Returns 0 on success, < 0 on error
2006 */
2007int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream *stream)
2008{
2009 switch (consumer_data.type) {
2010 case LTTNG_CONSUMER_KERNEL:
2011 return lttng_kconsumer_sample_snapshot_positions(stream);
2012 case LTTNG_CONSUMER32_UST:
2013 case LTTNG_CONSUMER64_UST:
2014 return lttng_ustconsumer_sample_snapshot_positions(stream);
2015 default:
2016 ERR("Unknown consumer_data type");
2017 assert(0);
2018 return -ENOSYS;
2019 }
2020}
3bd1e081
MD
2021/*
2022 * Take a snapshot for a specific fd
2023 *
2024 * Returns 0 on success, < 0 on error
2025 */
ffe60014 2026int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream)
3bd1e081
MD
2027{
2028 switch (consumer_data.type) {
2029 case LTTNG_CONSUMER_KERNEL:
ffe60014 2030 return lttng_kconsumer_take_snapshot(stream);
7753dea8
MD
2031 case LTTNG_CONSUMER32_UST:
2032 case LTTNG_CONSUMER64_UST:
ffe60014 2033 return lttng_ustconsumer_take_snapshot(stream);
3bd1e081
MD
2034 default:
2035 ERR("Unknown consumer_data type");
2036 assert(0);
2037 return -ENOSYS;
2038 }
3bd1e081
MD
2039}
2040
2041/*
2042 * Get the produced position
2043 *
2044 * Returns 0 on success, < 0 on error
2045 */
ffe60014 2046int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
3bd1e081
MD
2047 unsigned long *pos)
2048{
2049 switch (consumer_data.type) {
2050 case LTTNG_CONSUMER_KERNEL:
ffe60014 2051 return lttng_kconsumer_get_produced_snapshot(stream, pos);
7753dea8
MD
2052 case LTTNG_CONSUMER32_UST:
2053 case LTTNG_CONSUMER64_UST:
ffe60014 2054 return lttng_ustconsumer_get_produced_snapshot(stream, pos);
3bd1e081
MD
2055 default:
2056 ERR("Unknown consumer_data type");
2057 assert(0);
2058 return -ENOSYS;
2059 }
2060}
2061
15055ce5
JD
2062/*
2063 * Get the consumed position (free-running counter position in bytes).
2064 *
2065 * Returns 0 on success, < 0 on error
2066 */
2067int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
2068 unsigned long *pos)
2069{
2070 switch (consumer_data.type) {
2071 case LTTNG_CONSUMER_KERNEL:
2072 return lttng_kconsumer_get_consumed_snapshot(stream, pos);
2073 case LTTNG_CONSUMER32_UST:
2074 case LTTNG_CONSUMER64_UST:
2075 return lttng_ustconsumer_get_consumed_snapshot(stream, pos);
2076 default:
2077 ERR("Unknown consumer_data type");
2078 assert(0);
2079 return -ENOSYS;
2080 }
2081}
2082
3bd1e081
MD
2083int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx,
2084 int sock, struct pollfd *consumer_sockpoll)
2085{
2086 switch (consumer_data.type) {
2087 case LTTNG_CONSUMER_KERNEL:
2088 return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
7753dea8
MD
2089 case LTTNG_CONSUMER32_UST:
2090 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
2091 return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
2092 default:
2093 ERR("Unknown consumer_data type");
2094 assert(0);
2095 return -ENOSYS;
2096 }
2097}
2098
1f8d1c14 2099static
6d574024 2100void lttng_consumer_close_all_metadata(void)
d88aee68
DG
2101{
2102 switch (consumer_data.type) {
2103 case LTTNG_CONSUMER_KERNEL:
2104 /*
2105 * The Kernel consumer has a different metadata scheme so we don't
2106 * close anything because the stream will be closed by the session
2107 * daemon.
2108 */
2109 break;
2110 case LTTNG_CONSUMER32_UST:
2111 case LTTNG_CONSUMER64_UST:
2112 /*
2113 * Close all metadata streams. The metadata hash table is passed and
2114 * this call iterates over it by closing all wakeup fd. This is safe
2115 * because at this point we are sure that the metadata producer is
2116 * either dead or blocked.
2117 */
6d574024 2118 lttng_ustconsumer_close_all_metadata(metadata_ht);
d88aee68
DG
2119 break;
2120 default:
2121 ERR("Unknown consumer_data type");
2122 assert(0);
2123 }
2124}
2125
fb3a43a9
DG
2126/*
2127 * Clean up a metadata stream and free its memory.
2128 */
e316aad5
DG
2129void consumer_del_metadata_stream(struct lttng_consumer_stream *stream,
2130 struct lttng_ht *ht)
fb3a43a9 2131{
a6ef8ee6
JG
2132 struct lttng_consumer_channel *channel = NULL;
2133 bool free_channel = false;
fb3a43a9
DG
2134
2135 assert(stream);
2136 /*
2137 * This call should NEVER receive regular stream. It must always be
2138 * metadata stream and this is crucial for data structure synchronization.
2139 */
2140 assert(stream->metadata_flag);
2141
e316aad5
DG
2142 DBG3("Consumer delete metadata stream %d", stream->wait_fd);
2143
74251bb8 2144 pthread_mutex_lock(&consumer_data.lock);
a6ef8ee6
JG
2145 /*
2146 * Note that this assumes that a stream's channel is never changed and
2147 * that the stream's lock doesn't need to be taken to sample its
2148 * channel.
2149 */
2150 channel = stream->chan;
2151 pthread_mutex_lock(&channel->lock);
3dad2c0f 2152 pthread_mutex_lock(&stream->lock);
a6ef8ee6 2153 if (channel->metadata_cache) {
081424af 2154 /* Only applicable to userspace consumers. */
a6ef8ee6 2155 pthread_mutex_lock(&channel->metadata_cache->lock);
081424af 2156 }
8994307f 2157
6d574024
DG
2158 /* Remove any reference to that stream. */
2159 consumer_stream_delete(stream, ht);
ca22feea 2160
6d574024
DG
2161 /* Close down everything including the relayd if one. */
2162 consumer_stream_close(stream);
2163 /* Destroy tracer buffers of the stream. */
2164 consumer_stream_destroy_buffers(stream);
fb3a43a9
DG
2165
2166 /* Atomically decrement channel refcount since other threads can use it. */
a6ef8ee6
JG
2167 if (!uatomic_sub_return(&channel->refcount, 1)
2168 && !uatomic_read(&channel->nb_init_stream_left)) {
c30aaa51 2169 /* Go for channel deletion! */
a6ef8ee6 2170 free_channel = true;
fb3a43a9 2171 }
a6ef8ee6 2172 stream->chan = NULL;
fb3a43a9 2173
73811ecc
DG
2174 /*
2175 * Nullify the stream reference so it is not used after deletion. The
6d574024
DG
2176 * channel lock MUST be acquired before being able to check for a NULL
2177 * pointer value.
73811ecc 2178 */
a6ef8ee6 2179 channel->metadata_stream = NULL;
73811ecc 2180
a6ef8ee6
JG
2181 if (channel->metadata_cache) {
2182 pthread_mutex_unlock(&channel->metadata_cache->lock);
081424af 2183 }
3dad2c0f 2184 pthread_mutex_unlock(&stream->lock);
a6ef8ee6 2185 pthread_mutex_unlock(&channel->lock);
74251bb8 2186 pthread_mutex_unlock(&consumer_data.lock);
e316aad5 2187
a6ef8ee6
JG
2188 if (free_channel) {
2189 consumer_del_channel(channel);
e316aad5
DG
2190 }
2191
d2956687
JG
2192 lttng_trace_chunk_put(stream->trace_chunk);
2193 stream->trace_chunk = NULL;
6d574024 2194 consumer_stream_free(stream);
fb3a43a9
DG
2195}
2196
2197/*
2198 * Action done with the metadata stream when adding it to the consumer internal
2199 * data structures to handle it.
2200 */
66d583dc 2201void consumer_add_metadata_stream(struct lttng_consumer_stream *stream)
fb3a43a9 2202{
5ab66908 2203 struct lttng_ht *ht = metadata_ht;
76082088 2204 struct lttng_ht_iter iter;
d88aee68 2205 struct lttng_ht_node_u64 *node;
fb3a43a9 2206
e316aad5
DG
2207 assert(stream);
2208 assert(ht);
2209
d88aee68 2210 DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key);
e316aad5
DG
2211
2212 pthread_mutex_lock(&consumer_data.lock);
a9838785 2213 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 2214 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 2215 pthread_mutex_lock(&stream->lock);
e316aad5 2216
e316aad5
DG
2217 /*
2218 * From here, refcounts are updated so be _careful_ when returning an error
2219 * after this point.
2220 */
2221
fb3a43a9 2222 rcu_read_lock();
76082088
DG
2223
2224 /*
2225 * Lookup the stream just to make sure it does not exist in our internal
2226 * state. This should NEVER happen.
2227 */
d88aee68
DG
2228 lttng_ht_lookup(ht, &stream->key, &iter);
2229 node = lttng_ht_iter_get_node_u64(&iter);
76082088
DG
2230 assert(!node);
2231
e316aad5 2232 /*
ffe60014
DG
2233 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2234 * in terms of destroying the associated channel, because the action that
e316aad5
DG
2235 * causes the count to become 0 also causes a stream to be added. The
2236 * channel deletion will thus be triggered by the following removal of this
2237 * stream.
2238 */
ffe60014 2239 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
2240 /* Increment refcount before decrementing nb_init_stream_left */
2241 cmm_smp_wmb();
ffe60014 2242 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
2243 }
2244
d88aee68 2245 lttng_ht_add_unique_u64(ht, &stream->node);
ca22feea 2246
446156b4 2247 lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht,
d8ef542d
MD
2248 &stream->node_channel_id);
2249
ca22feea
DG
2250 /*
2251 * Add stream to the stream_list_ht of the consumer data. No need to steal
2252 * the key since the HT does not use it and we allow to add redundant keys
2253 * into this table.
2254 */
d88aee68 2255 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 2256
fb3a43a9 2257 rcu_read_unlock();
e316aad5 2258
2e818a6a 2259 pthread_mutex_unlock(&stream->lock);
a9838785 2260 pthread_mutex_unlock(&stream->chan->lock);
ec6ea7d0 2261 pthread_mutex_unlock(&stream->chan->timer_lock);
e316aad5 2262 pthread_mutex_unlock(&consumer_data.lock);
fb3a43a9
DG
2263}
2264
8994307f
DG
2265/*
2266 * Delete data stream that are flagged for deletion (endpoint_status).
2267 */
2268static void validate_endpoint_status_data_stream(void)
2269{
2270 struct lttng_ht_iter iter;
2271 struct lttng_consumer_stream *stream;
2272
2273 DBG("Consumer delete flagged data stream");
2274
2275 rcu_read_lock();
2276 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
2277 /* Validate delete flag of the stream */
79d4ffb7 2278 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2279 continue;
2280 }
2281 /* Delete it right now */
2282 consumer_del_stream(stream, data_ht);
2283 }
2284 rcu_read_unlock();
2285}
2286
2287/*
2288 * Delete metadata stream that are flagged for deletion (endpoint_status).
2289 */
2290static void validate_endpoint_status_metadata_stream(
2291 struct lttng_poll_event *pollset)
2292{
2293 struct lttng_ht_iter iter;
2294 struct lttng_consumer_stream *stream;
2295
2296 DBG("Consumer delete flagged metadata stream");
2297
2298 assert(pollset);
2299
2300 rcu_read_lock();
2301 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
2302 /* Validate delete flag of the stream */
79d4ffb7 2303 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2304 continue;
2305 }
2306 /*
2307 * Remove from pollset so the metadata thread can continue without
2308 * blocking on a deleted stream.
2309 */
2310 lttng_poll_del(pollset, stream->wait_fd);
2311
2312 /* Delete it right now */
2313 consumer_del_metadata_stream(stream, metadata_ht);
2314 }
2315 rcu_read_unlock();
2316}
2317
fb3a43a9
DG
2318/*
2319 * Thread polls on metadata file descriptor and write them on disk or on the
2320 * network.
2321 */
7d980def 2322void *consumer_thread_metadata_poll(void *data)
fb3a43a9 2323{
1fc79fb4 2324 int ret, i, pollfd, err = -1;
fb3a43a9 2325 uint32_t revents, nb_fd;
e316aad5 2326 struct lttng_consumer_stream *stream = NULL;
fb3a43a9 2327 struct lttng_ht_iter iter;
d88aee68 2328 struct lttng_ht_node_u64 *node;
fb3a43a9
DG
2329 struct lttng_poll_event events;
2330 struct lttng_consumer_local_data *ctx = data;
2331 ssize_t len;
2332
2333 rcu_register_thread();
2334
1fc79fb4
MD
2335 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA);
2336
2d57de81
MD
2337 if (testpoint(consumerd_thread_metadata)) {
2338 goto error_testpoint;
2339 }
2340
9ce5646a
MD
2341 health_code_update();
2342
fb3a43a9
DG
2343 DBG("Thread metadata poll started");
2344
fb3a43a9
DG
2345 /* Size is set to 1 for the consumer_metadata pipe */
2346 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2347 if (ret < 0) {
2348 ERR("Poll set creation failed");
d8ef542d 2349 goto end_poll;
fb3a43a9
DG
2350 }
2351
13886d2d
DG
2352 ret = lttng_poll_add(&events,
2353 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN);
fb3a43a9
DG
2354 if (ret < 0) {
2355 goto end;
2356 }
2357
2358 /* Main loop */
2359 DBG("Metadata main loop started");
2360
2361 while (1) {
fb3a43a9 2362restart:
7fa2082e 2363 health_code_update();
9ce5646a 2364 health_poll_entry();
7fa2082e 2365 DBG("Metadata poll wait");
fb3a43a9 2366 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2367 DBG("Metadata poll return from wait with %d fd(s)",
2368 LTTNG_POLL_GETNB(&events));
9ce5646a 2369 health_poll_exit();
40063ead 2370 DBG("Metadata event caught in thread");
fb3a43a9
DG
2371 if (ret < 0) {
2372 if (errno == EINTR) {
40063ead 2373 ERR("Poll EINTR caught");
fb3a43a9
DG
2374 goto restart;
2375 }
d9607cd7
MD
2376 if (LTTNG_POLL_GETNB(&events) == 0) {
2377 err = 0; /* All is OK */
2378 }
2379 goto end;
fb3a43a9
DG
2380 }
2381
0d9c5d77
DG
2382 nb_fd = ret;
2383
e316aad5 2384 /* From here, the event is a metadata wait fd */
fb3a43a9 2385 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2386 health_code_update();
2387
fb3a43a9
DG
2388 revents = LTTNG_POLL_GETEV(&events, i);
2389 pollfd = LTTNG_POLL_GETFD(&events, i);
2390
13886d2d 2391 if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) {
03e43155 2392 if (revents & LPOLLIN) {
13886d2d
DG
2393 ssize_t pipe_len;
2394
2395 pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe,
2396 &stream, sizeof(stream));
6cd525e8 2397 if (pipe_len < sizeof(stream)) {
03e43155
MD
2398 if (pipe_len < 0) {
2399 PERROR("read metadata stream");
2400 }
fb3a43a9 2401 /*
03e43155
MD
2402 * Remove the pipe from the poll set and continue the loop
2403 * since their might be data to consume.
fb3a43a9 2404 */
03e43155
MD
2405 lttng_poll_del(&events,
2406 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2407 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
fb3a43a9
DG
2408 continue;
2409 }
2410
8994307f
DG
2411 /* A NULL stream means that the state has changed. */
2412 if (stream == NULL) {
2413 /* Check for deleted streams. */
2414 validate_endpoint_status_metadata_stream(&events);
3714380f 2415 goto restart;
8994307f
DG
2416 }
2417
fb3a43a9
DG
2418 DBG("Adding metadata stream %d to poll set",
2419 stream->wait_fd);
2420
fb3a43a9
DG
2421 /* Add metadata stream to the global poll events list */
2422 lttng_poll_add(&events, stream->wait_fd,
6d574024 2423 LPOLLIN | LPOLLPRI | LPOLLHUP);
03e43155
MD
2424 } else if (revents & (LPOLLERR | LPOLLHUP)) {
2425 DBG("Metadata thread pipe hung up");
2426 /*
2427 * Remove the pipe from the poll set and continue the loop
2428 * since their might be data to consume.
2429 */
2430 lttng_poll_del(&events,
2431 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2432 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
2433 continue;
2434 } else {
2435 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2436 goto end;
fb3a43a9
DG
2437 }
2438
e316aad5 2439 /* Handle other stream */
fb3a43a9
DG
2440 continue;
2441 }
2442
d09e1200 2443 rcu_read_lock();
d88aee68
DG
2444 {
2445 uint64_t tmp_id = (uint64_t) pollfd;
2446
2447 lttng_ht_lookup(metadata_ht, &tmp_id, &iter);
2448 }
2449 node = lttng_ht_iter_get_node_u64(&iter);
e316aad5 2450 assert(node);
fb3a43a9
DG
2451
2452 stream = caa_container_of(node, struct lttng_consumer_stream,
58b1f425 2453 node);
fb3a43a9 2454
03e43155
MD
2455 if (revents & (LPOLLIN | LPOLLPRI)) {
2456 /* Get the data out of the metadata file descriptor */
2457 DBG("Metadata available on fd %d", pollfd);
2458 assert(stream->wait_fd == pollfd);
2459
2460 do {
2461 health_code_update();
2462
6f9449c2 2463 len = ctx->on_buffer_ready(stream, ctx, false);
03e43155
MD
2464 /*
2465 * We don't check the return value here since if we get
83f4233d 2466 * a negative len, it means an error occurred thus we
03e43155
MD
2467 * simply remove it from the poll set and free the
2468 * stream.
2469 */
2470 } while (len > 0);
2471
2472 /* It's ok to have an unavailable sub-buffer */
2473 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
2474 /* Clean up stream from consumer and free it. */
2475 lttng_poll_del(&events, stream->wait_fd);
2476 consumer_del_metadata_stream(stream, metadata_ht);
2477 }
2478 } else if (revents & (LPOLLERR | LPOLLHUP)) {
e316aad5 2479 DBG("Metadata fd %d is hup|err.", pollfd);
fb3a43a9
DG
2480 if (!stream->hangup_flush_done
2481 && (consumer_data.type == LTTNG_CONSUMER32_UST
2482 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2483 DBG("Attempting to flush and consume the UST buffers");
2484 lttng_ustconsumer_on_stream_hangup(stream);
2485
2486 /* We just flushed the stream now read it. */
4bb94b75 2487 do {
9ce5646a
MD
2488 health_code_update();
2489
6f9449c2 2490 len = ctx->on_buffer_ready(stream, ctx, false);
4bb94b75
DG
2491 /*
2492 * We don't check the return value here since if we get
83f4233d 2493 * a negative len, it means an error occurred thus we
4bb94b75
DG
2494 * simply remove it from the poll set and free the
2495 * stream.
2496 */
2497 } while (len > 0);
fb3a43a9
DG
2498 }
2499
fb3a43a9 2500 lttng_poll_del(&events, stream->wait_fd);
e316aad5
DG
2501 /*
2502 * This call update the channel states, closes file descriptors
2503 * and securely free the stream.
2504 */
2505 consumer_del_metadata_stream(stream, metadata_ht);
03e43155
MD
2506 } else {
2507 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
6f2f1a70 2508 rcu_read_unlock();
03e43155 2509 goto end;
fb3a43a9 2510 }
e316aad5 2511 /* Release RCU lock for the stream looked up */
d09e1200 2512 rcu_read_unlock();
fb3a43a9
DG
2513 }
2514 }
2515
1fc79fb4
MD
2516 /* All is OK */
2517 err = 0;
fb3a43a9
DG
2518end:
2519 DBG("Metadata poll thread exiting");
fb3a43a9 2520
d8ef542d
MD
2521 lttng_poll_clean(&events);
2522end_poll:
2d57de81 2523error_testpoint:
1fc79fb4
MD
2524 if (err) {
2525 health_error();
2526 ERR("Health error occurred in %s", __func__);
2527 }
2528 health_unregister(health_consumerd);
fb3a43a9
DG
2529 rcu_unregister_thread();
2530 return NULL;
2531}
2532
3bd1e081 2533/*
e4421fec 2534 * This thread polls the fds in the set to consume the data and write
3bd1e081
MD
2535 * it to tracefile if necessary.
2536 */
7d980def 2537void *consumer_thread_data_poll(void *data)
3bd1e081 2538{
1fc79fb4 2539 int num_rdy, num_hup, high_prio, ret, i, err = -1;
3bd1e081
MD
2540 struct pollfd *pollfd = NULL;
2541 /* local view of the streams */
c869f647 2542 struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL;
3bd1e081 2543 /* local view of consumer_data.fds_count */
8bdcc002
JG
2544 int nb_fd = 0;
2545 /* 2 for the consumer_data_pipe and wake up pipe */
2546 const int nb_pipes_fd = 2;
9a2fcf78
JD
2547 /* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */
2548 int nb_inactive_fd = 0;
3bd1e081 2549 struct lttng_consumer_local_data *ctx = data;
00e2e675 2550 ssize_t len;
3bd1e081 2551
e7b994a3
DG
2552 rcu_register_thread();
2553
1fc79fb4
MD
2554 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA);
2555
2d57de81
MD
2556 if (testpoint(consumerd_thread_data)) {
2557 goto error_testpoint;
2558 }
2559
9ce5646a
MD
2560 health_code_update();
2561
4df6c8cb
MD
2562 local_stream = zmalloc(sizeof(struct lttng_consumer_stream *));
2563 if (local_stream == NULL) {
2564 PERROR("local_stream malloc");
2565 goto end;
2566 }
3bd1e081
MD
2567
2568 while (1) {
9ce5646a
MD
2569 health_code_update();
2570
3bd1e081
MD
2571 high_prio = 0;
2572 num_hup = 0;
2573
2574 /*
e4421fec 2575 * the fds set has been updated, we need to update our
3bd1e081
MD
2576 * local array as well
2577 */
2578 pthread_mutex_lock(&consumer_data.lock);
2579 if (consumer_data.need_update) {
0e428499
DG
2580 free(pollfd);
2581 pollfd = NULL;
2582
2583 free(local_stream);
2584 local_stream = NULL;
3bd1e081 2585
8bdcc002 2586 /* Allocate for all fds */
261de637 2587 pollfd = zmalloc((consumer_data.stream_count + nb_pipes_fd) * sizeof(struct pollfd));
3bd1e081 2588 if (pollfd == NULL) {
7a57cf92 2589 PERROR("pollfd malloc");
3bd1e081
MD
2590 pthread_mutex_unlock(&consumer_data.lock);
2591 goto end;
2592 }
2593
261de637 2594 local_stream = zmalloc((consumer_data.stream_count + nb_pipes_fd) *
747f8642 2595 sizeof(struct lttng_consumer_stream *));
3bd1e081 2596 if (local_stream == NULL) {
7a57cf92 2597 PERROR("local_stream malloc");
3bd1e081
MD
2598 pthread_mutex_unlock(&consumer_data.lock);
2599 goto end;
2600 }
ffe60014 2601 ret = update_poll_array(ctx, &pollfd, local_stream,
9a2fcf78 2602 data_ht, &nb_inactive_fd);
3bd1e081
MD
2603 if (ret < 0) {
2604 ERR("Error in allocating pollfd or local_outfds");
f73fabfd 2605 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2606 pthread_mutex_unlock(&consumer_data.lock);
2607 goto end;
2608 }
2609 nb_fd = ret;
2610 consumer_data.need_update = 0;
2611 }
2612 pthread_mutex_unlock(&consumer_data.lock);
2613
4078b776 2614 /* No FDs and consumer_quit, consumer_cleanup the thread */
9a2fcf78
JD
2615 if (nb_fd == 0 && nb_inactive_fd == 0 &&
2616 CMM_LOAD_SHARED(consumer_quit) == 1) {
1fc79fb4 2617 err = 0; /* All is OK */
4078b776
MD
2618 goto end;
2619 }
3bd1e081 2620 /* poll on the array of fds */
88f2b785 2621 restart:
261de637 2622 DBG("polling on %d fd", nb_fd + nb_pipes_fd);
cf0bcb51
JG
2623 if (testpoint(consumerd_thread_data_poll)) {
2624 goto end;
2625 }
9ce5646a 2626 health_poll_entry();
261de637 2627 num_rdy = poll(pollfd, nb_fd + nb_pipes_fd, -1);
9ce5646a 2628 health_poll_exit();
3bd1e081
MD
2629 DBG("poll num_rdy : %d", num_rdy);
2630 if (num_rdy == -1) {
88f2b785
MD
2631 /*
2632 * Restart interrupted system call.
2633 */
2634 if (errno == EINTR) {
2635 goto restart;
2636 }
7a57cf92 2637 PERROR("Poll error");
f73fabfd 2638 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2639 goto end;
2640 } else if (num_rdy == 0) {
2641 DBG("Polling thread timed out");
2642 goto end;
2643 }
2644
80957876
JG
2645 if (caa_unlikely(data_consumption_paused)) {
2646 DBG("Data consumption paused, sleeping...");
2647 sleep(1);
2648 goto restart;
2649 }
2650
3bd1e081 2651 /*
50f8ae69 2652 * If the consumer_data_pipe triggered poll go directly to the
00e2e675
DG
2653 * beginning of the loop to update the array. We want to prioritize
2654 * array update over low-priority reads.
3bd1e081 2655 */
509bb1cf 2656 if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) {
ab30f567 2657 ssize_t pipe_readlen;
04fdd819 2658
50f8ae69 2659 DBG("consumer_data_pipe wake up");
acdb9057
DG
2660 pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe,
2661 &new_stream, sizeof(new_stream));
6cd525e8
MD
2662 if (pipe_readlen < sizeof(new_stream)) {
2663 PERROR("Consumer data pipe");
23f5f35d
DG
2664 /* Continue so we can at least handle the current stream(s). */
2665 continue;
2666 }
c869f647
DG
2667
2668 /*
2669 * If the stream is NULL, just ignore it. It's also possible that
2670 * the sessiond poll thread changed the consumer_quit state and is
2671 * waking us up to test it.
2672 */
2673 if (new_stream == NULL) {
8994307f 2674 validate_endpoint_status_data_stream();
c869f647
DG
2675 continue;
2676 }
2677
c869f647 2678 /* Continue to update the local streams and handle prio ones */
3bd1e081
MD
2679 continue;
2680 }
2681
02b3d176
DG
2682 /* Handle wakeup pipe. */
2683 if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) {
2684 char dummy;
2685 ssize_t pipe_readlen;
2686
2687 pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy,
2688 sizeof(dummy));
2689 if (pipe_readlen < 0) {
2690 PERROR("Consumer data wakeup pipe");
2691 }
2692 /* We've been awakened to handle stream(s). */
2693 ctx->has_wakeup = 0;
2694 }
2695
3bd1e081
MD
2696 /* Take care of high priority channels first. */
2697 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2698 health_code_update();
2699
9617607b
DG
2700 if (local_stream[i] == NULL) {
2701 continue;
2702 }
fb3a43a9 2703 if (pollfd[i].revents & POLLPRI) {
d41f73b7
MD
2704 DBG("Urgent read on fd %d", pollfd[i].fd);
2705 high_prio = 1;
6f9449c2 2706 len = ctx->on_buffer_ready(local_stream[i], ctx, false);
d41f73b7 2707 /* it's ok to have an unavailable sub-buffer */
b64403e3 2708 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2709 /* Clean the stream and free it. */
2710 consumer_del_stream(local_stream[i], data_ht);
9617607b 2711 local_stream[i] = NULL;
4078b776
MD
2712 } else if (len > 0) {
2713 local_stream[i]->data_read = 1;
d41f73b7 2714 }
3bd1e081
MD
2715 }
2716 }
2717
4078b776
MD
2718 /*
2719 * If we read high prio channel in this loop, try again
2720 * for more high prio data.
2721 */
2722 if (high_prio) {
3bd1e081
MD
2723 continue;
2724 }
2725
2726 /* Take care of low priority channels. */
4078b776 2727 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2728 health_code_update();
2729
9617607b
DG
2730 if (local_stream[i] == NULL) {
2731 continue;
2732 }
4078b776 2733 if ((pollfd[i].revents & POLLIN) ||
02b3d176
DG
2734 local_stream[i]->hangup_flush_done ||
2735 local_stream[i]->has_data) {
4078b776 2736 DBG("Normal read on fd %d", pollfd[i].fd);
6f9449c2 2737 len = ctx->on_buffer_ready(local_stream[i], ctx, false);
4078b776 2738 /* it's ok to have an unavailable sub-buffer */
b64403e3 2739 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2740 /* Clean the stream and free it. */
2741 consumer_del_stream(local_stream[i], data_ht);
9617607b 2742 local_stream[i] = NULL;
4078b776
MD
2743 } else if (len > 0) {
2744 local_stream[i]->data_read = 1;
2745 }
2746 }
2747 }
2748
2749 /* Handle hangup and errors */
2750 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2751 health_code_update();
2752
9617607b
DG
2753 if (local_stream[i] == NULL) {
2754 continue;
2755 }
4078b776
MD
2756 if (!local_stream[i]->hangup_flush_done
2757 && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL))
2758 && (consumer_data.type == LTTNG_CONSUMER32_UST
2759 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2760 DBG("fd %d is hup|err|nval. Attempting flush and read.",
9617607b 2761 pollfd[i].fd);
4078b776
MD
2762 lttng_ustconsumer_on_stream_hangup(local_stream[i]);
2763 /* Attempt read again, for the data we just flushed. */
2764 local_stream[i]->data_read = 1;
2765 }
2766 /*
2767 * If the poll flag is HUP/ERR/NVAL and we have
2768 * read no data in this pass, we can remove the
2769 * stream from its hash table.
2770 */
2771 if ((pollfd[i].revents & POLLHUP)) {
2772 DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
2773 if (!local_stream[i]->data_read) {
43c34bc3 2774 consumer_del_stream(local_stream[i], data_ht);
9617607b 2775 local_stream[i] = NULL;
4078b776
MD
2776 num_hup++;
2777 }
2778 } else if (pollfd[i].revents & POLLERR) {
2779 ERR("Error returned in polling fd %d.", pollfd[i].fd);
2780 if (!local_stream[i]->data_read) {
43c34bc3 2781 consumer_del_stream(local_stream[i], data_ht);
9617607b 2782 local_stream[i] = NULL;
4078b776
MD
2783 num_hup++;
2784 }
2785 } else if (pollfd[i].revents & POLLNVAL) {
2786 ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
2787 if (!local_stream[i]->data_read) {
43c34bc3 2788 consumer_del_stream(local_stream[i], data_ht);
9617607b 2789 local_stream[i] = NULL;
4078b776 2790 num_hup++;
3bd1e081
MD
2791 }
2792 }
9617607b
DG
2793 if (local_stream[i] != NULL) {
2794 local_stream[i]->data_read = 0;
2795 }
3bd1e081
MD
2796 }
2797 }
1fc79fb4
MD
2798 /* All is OK */
2799 err = 0;
3bd1e081
MD
2800end:
2801 DBG("polling thread exiting");
0e428499
DG
2802 free(pollfd);
2803 free(local_stream);
fb3a43a9
DG
2804
2805 /*
2806 * Close the write side of the pipe so epoll_wait() in
7d980def
DG
2807 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2808 * read side of the pipe. If we close them both, epoll_wait strangely does
2809 * not return and could create a endless wait period if the pipe is the
2810 * only tracked fd in the poll set. The thread will take care of closing
2811 * the read side.
fb3a43a9 2812 */
13886d2d 2813 (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe);
fb3a43a9 2814
2d57de81 2815error_testpoint:
1fc79fb4
MD
2816 if (err) {
2817 health_error();
2818 ERR("Health error occurred in %s", __func__);
2819 }
2820 health_unregister(health_consumerd);
2821
e7b994a3 2822 rcu_unregister_thread();
3bd1e081
MD
2823 return NULL;
2824}
2825
d8ef542d
MD
2826/*
2827 * Close wake-up end of each stream belonging to the channel. This will
2828 * allow the poll() on the stream read-side to detect when the
2829 * write-side (application) finally closes them.
2830 */
2831static
2832void consumer_close_channel_streams(struct lttng_consumer_channel *channel)
2833{
2834 struct lttng_ht *ht;
2835 struct lttng_consumer_stream *stream;
2836 struct lttng_ht_iter iter;
2837
2838 ht = consumer_data.stream_per_chan_id_ht;
2839
2840 rcu_read_lock();
2841 cds_lfht_for_each_entry_duplicate(ht->ht,
2842 ht->hash_fct(&channel->key, lttng_ht_seed),
2843 ht->match_fct, &channel->key,
2844 &iter.iter, stream, node_channel_id.node) {
f2ad556d
MD
2845 /*
2846 * Protect against teardown with mutex.
2847 */
2848 pthread_mutex_lock(&stream->lock);
2849 if (cds_lfht_is_node_deleted(&stream->node.node)) {
2850 goto next;
2851 }
d8ef542d
MD
2852 switch (consumer_data.type) {
2853 case LTTNG_CONSUMER_KERNEL:
2854 break;
2855 case LTTNG_CONSUMER32_UST:
2856 case LTTNG_CONSUMER64_UST:
b4a650f3
DG
2857 if (stream->metadata_flag) {
2858 /* Safe and protected by the stream lock. */
2859 lttng_ustconsumer_close_metadata(stream->chan);
2860 } else {
2861 /*
2862 * Note: a mutex is taken internally within
2863 * liblttng-ust-ctl to protect timer wakeup_fd
2864 * use from concurrent close.
2865 */
2866 lttng_ustconsumer_close_stream_wakeup(stream);
2867 }
d8ef542d
MD
2868 break;
2869 default:
2870 ERR("Unknown consumer_data type");
2871 assert(0);
2872 }
f2ad556d
MD
2873 next:
2874 pthread_mutex_unlock(&stream->lock);
d8ef542d
MD
2875 }
2876 rcu_read_unlock();
2877}
2878
2879static void destroy_channel_ht(struct lttng_ht *ht)
2880{
2881 struct lttng_ht_iter iter;
2882 struct lttng_consumer_channel *channel;
2883 int ret;
2884
2885 if (ht == NULL) {
2886 return;
2887 }
2888
2889 rcu_read_lock();
2890 cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) {
2891 ret = lttng_ht_del(ht, &iter);
2892 assert(ret != 0);
2893 }
2894 rcu_read_unlock();
2895
2896 lttng_ht_destroy(ht);
2897}
2898
2899/*
2900 * This thread polls the channel fds to detect when they are being
2901 * closed. It closes all related streams if the channel is detected as
2902 * closed. It is currently only used as a shim layer for UST because the
2903 * consumerd needs to keep the per-stream wakeup end of pipes open for
2904 * periodical flush.
2905 */
2906void *consumer_thread_channel_poll(void *data)
2907{
1fc79fb4 2908 int ret, i, pollfd, err = -1;
d8ef542d
MD
2909 uint32_t revents, nb_fd;
2910 struct lttng_consumer_channel *chan = NULL;
2911 struct lttng_ht_iter iter;
2912 struct lttng_ht_node_u64 *node;
2913 struct lttng_poll_event events;
2914 struct lttng_consumer_local_data *ctx = data;
2915 struct lttng_ht *channel_ht;
2916
2917 rcu_register_thread();
2918
1fc79fb4
MD
2919 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL);
2920
2d57de81
MD
2921 if (testpoint(consumerd_thread_channel)) {
2922 goto error_testpoint;
2923 }
2924
9ce5646a
MD
2925 health_code_update();
2926
d8ef542d
MD
2927 channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2928 if (!channel_ht) {
2929 /* ENOMEM at this point. Better to bail out. */
2930 goto end_ht;
2931 }
2932
2933 DBG("Thread channel poll started");
2934
2935 /* Size is set to 1 for the consumer_channel pipe */
2936 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2937 if (ret < 0) {
2938 ERR("Poll set creation failed");
2939 goto end_poll;
2940 }
2941
2942 ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN);
2943 if (ret < 0) {
2944 goto end;
2945 }
2946
2947 /* Main loop */
2948 DBG("Channel main loop started");
2949
2950 while (1) {
d8ef542d 2951restart:
7fa2082e
MD
2952 health_code_update();
2953 DBG("Channel poll wait");
9ce5646a 2954 health_poll_entry();
d8ef542d 2955 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2956 DBG("Channel poll return from wait with %d fd(s)",
2957 LTTNG_POLL_GETNB(&events));
9ce5646a 2958 health_poll_exit();
40063ead 2959 DBG("Channel event caught in thread");
d8ef542d
MD
2960 if (ret < 0) {
2961 if (errno == EINTR) {
40063ead 2962 ERR("Poll EINTR caught");
d8ef542d
MD
2963 goto restart;
2964 }
d9607cd7
MD
2965 if (LTTNG_POLL_GETNB(&events) == 0) {
2966 err = 0; /* All is OK */
2967 }
d8ef542d
MD
2968 goto end;
2969 }
2970
2971 nb_fd = ret;
2972
2973 /* From here, the event is a channel wait fd */
2974 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2975 health_code_update();
2976
d8ef542d
MD
2977 revents = LTTNG_POLL_GETEV(&events, i);
2978 pollfd = LTTNG_POLL_GETFD(&events, i);
2979
d8ef542d 2980 if (pollfd == ctx->consumer_channel_pipe[0]) {
03e43155 2981 if (revents & LPOLLIN) {
d8ef542d 2982 enum consumer_channel_action action;
a0cbdd2e 2983 uint64_t key;
d8ef542d 2984
a0cbdd2e 2985 ret = read_channel_pipe(ctx, &chan, &key, &action);
d8ef542d 2986 if (ret <= 0) {
03e43155
MD
2987 if (ret < 0) {
2988 ERR("Error reading channel pipe");
2989 }
2990 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
d8ef542d
MD
2991 continue;
2992 }
2993
2994 switch (action) {
2995 case CONSUMER_CHANNEL_ADD:
2996 DBG("Adding channel %d to poll set",
2997 chan->wait_fd);
2998
2999 lttng_ht_node_init_u64(&chan->wait_fd_node,
3000 chan->wait_fd);
c7260a81 3001 rcu_read_lock();
d8ef542d
MD
3002 lttng_ht_add_unique_u64(channel_ht,
3003 &chan->wait_fd_node);
c7260a81 3004 rcu_read_unlock();
d8ef542d
MD
3005 /* Add channel to the global poll events list */
3006 lttng_poll_add(&events, chan->wait_fd,
03e43155 3007 LPOLLERR | LPOLLHUP);
d8ef542d 3008 break;
a0cbdd2e
MD
3009 case CONSUMER_CHANNEL_DEL:
3010 {
b4a650f3
DG
3011 /*
3012 * This command should never be called if the channel
3013 * has streams monitored by either the data or metadata
3014 * thread. The consumer only notify this thread with a
3015 * channel del. command if it receives a destroy
3016 * channel command from the session daemon that send it
3017 * if a command prior to the GET_CHANNEL failed.
3018 */
3019
c7260a81 3020 rcu_read_lock();
a0cbdd2e
MD
3021 chan = consumer_find_channel(key);
3022 if (!chan) {
c7260a81 3023 rcu_read_unlock();
a0cbdd2e
MD
3024 ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key);
3025 break;
3026 }
3027 lttng_poll_del(&events, chan->wait_fd);
f623cc0b 3028 iter.iter.node = &chan->wait_fd_node.node;
a0cbdd2e
MD
3029 ret = lttng_ht_del(channel_ht, &iter);
3030 assert(ret == 0);
a0cbdd2e 3031
f2a444f1
DG
3032 switch (consumer_data.type) {
3033 case LTTNG_CONSUMER_KERNEL:
3034 break;
3035 case LTTNG_CONSUMER32_UST:
3036 case LTTNG_CONSUMER64_UST:
212d67a2
DG
3037 health_code_update();
3038 /* Destroy streams that might have been left in the stream list. */
3039 clean_channel_stream_list(chan);
f2a444f1
DG
3040 break;
3041 default:
3042 ERR("Unknown consumer_data type");
3043 assert(0);
3044 }
3045
a0cbdd2e
MD
3046 /*
3047 * Release our own refcount. Force channel deletion even if
3048 * streams were not initialized.
3049 */
3050 if (!uatomic_sub_return(&chan->refcount, 1)) {
3051 consumer_del_channel(chan);
3052 }
c7260a81 3053 rcu_read_unlock();
a0cbdd2e
MD
3054 goto restart;
3055 }
d8ef542d
MD
3056 case CONSUMER_CHANNEL_QUIT:
3057 /*
3058 * Remove the pipe from the poll set and continue the loop
3059 * since their might be data to consume.
3060 */
3061 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
3062 continue;
3063 default:
3064 ERR("Unknown action");
3065 break;
3066 }
03e43155
MD
3067 } else if (revents & (LPOLLERR | LPOLLHUP)) {
3068 DBG("Channel thread pipe hung up");
3069 /*
3070 * Remove the pipe from the poll set and continue the loop
3071 * since their might be data to consume.
3072 */
3073 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
3074 continue;
3075 } else {
3076 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
3077 goto end;
d8ef542d
MD
3078 }
3079
3080 /* Handle other stream */
3081 continue;
3082 }
3083
3084 rcu_read_lock();
3085 {
3086 uint64_t tmp_id = (uint64_t) pollfd;
3087
3088 lttng_ht_lookup(channel_ht, &tmp_id, &iter);
3089 }
3090 node = lttng_ht_iter_get_node_u64(&iter);
3091 assert(node);
3092
3093 chan = caa_container_of(node, struct lttng_consumer_channel,
3094 wait_fd_node);
3095
3096 /* Check for error event */
3097 if (revents & (LPOLLERR | LPOLLHUP)) {
3098 DBG("Channel fd %d is hup|err.", pollfd);
3099
3100 lttng_poll_del(&events, chan->wait_fd);
3101 ret = lttng_ht_del(channel_ht, &iter);
3102 assert(ret == 0);
b4a650f3
DG
3103
3104 /*
3105 * This will close the wait fd for each stream associated to
3106 * this channel AND monitored by the data/metadata thread thus
3107 * will be clean by the right thread.
3108 */
d8ef542d 3109 consumer_close_channel_streams(chan);
f2ad556d
MD
3110
3111 /* Release our own refcount */
3112 if (!uatomic_sub_return(&chan->refcount, 1)
3113 && !uatomic_read(&chan->nb_init_stream_left)) {
3114 consumer_del_channel(chan);
3115 }
03e43155
MD
3116 } else {
3117 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
3118 rcu_read_unlock();
3119 goto end;
d8ef542d
MD
3120 }
3121
3122 /* Release RCU lock for the channel looked up */
3123 rcu_read_unlock();
3124 }
3125 }
3126
1fc79fb4
MD
3127 /* All is OK */
3128 err = 0;
d8ef542d
MD
3129end:
3130 lttng_poll_clean(&events);
3131end_poll:
3132 destroy_channel_ht(channel_ht);
3133end_ht:
2d57de81 3134error_testpoint:
d8ef542d 3135 DBG("Channel poll thread exiting");
1fc79fb4
MD
3136 if (err) {
3137 health_error();
3138 ERR("Health error occurred in %s", __func__);
3139 }
3140 health_unregister(health_consumerd);
d8ef542d
MD
3141 rcu_unregister_thread();
3142 return NULL;
3143}
3144
331744e3
JD
3145static int set_metadata_socket(struct lttng_consumer_local_data *ctx,
3146 struct pollfd *sockpoll, int client_socket)
3147{
3148 int ret;
3149
3150 assert(ctx);
3151 assert(sockpoll);
3152
84382d49
MD
3153 ret = lttng_consumer_poll_socket(sockpoll);
3154 if (ret) {
331744e3
JD
3155 goto error;
3156 }
3157 DBG("Metadata connection on client_socket");
3158
3159 /* Blocking call, waiting for transmission */
3160 ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket);
3161 if (ctx->consumer_metadata_socket < 0) {
3162 WARN("On accept metadata");
3163 ret = -1;
3164 goto error;
3165 }
3166 ret = 0;
3167
3168error:
3169 return ret;
3170}
3171
3bd1e081
MD
3172/*
3173 * This thread listens on the consumerd socket and receives the file
3174 * descriptors from the session daemon.
3175 */
7d980def 3176void *consumer_thread_sessiond_poll(void *data)
3bd1e081 3177{
1fc79fb4 3178 int sock = -1, client_socket, ret, err = -1;
3bd1e081
MD
3179 /*
3180 * structure to poll for incoming data on communication socket avoids
3181 * making blocking sockets.
3182 */
3183 struct pollfd consumer_sockpoll[2];
3184 struct lttng_consumer_local_data *ctx = data;
3185
e7b994a3
DG
3186 rcu_register_thread();
3187
1fc79fb4
MD
3188 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND);
3189
2d57de81
MD
3190 if (testpoint(consumerd_thread_sessiond)) {
3191 goto error_testpoint;
3192 }
3193
9ce5646a
MD
3194 health_code_update();
3195
3bd1e081
MD
3196 DBG("Creating command socket %s", ctx->consumer_command_sock_path);
3197 unlink(ctx->consumer_command_sock_path);
3198 client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path);
3199 if (client_socket < 0) {
3200 ERR("Cannot create command socket");
3201 goto end;
3202 }
3203
3204 ret = lttcomm_listen_unix_sock(client_socket);
3205 if (ret < 0) {
3206 goto end;
3207 }
3208
32258573 3209 DBG("Sending ready command to lttng-sessiond");
f73fabfd 3210 ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY);
3bd1e081
MD
3211 /* return < 0 on error, but == 0 is not fatal */
3212 if (ret < 0) {
32258573 3213 ERR("Error sending ready command to lttng-sessiond");
3bd1e081
MD
3214 goto end;
3215 }
3216
3bd1e081
MD
3217 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3218 consumer_sockpoll[0].fd = ctx->consumer_should_quit[0];
3219 consumer_sockpoll[0].events = POLLIN | POLLPRI;
3220 consumer_sockpoll[1].fd = client_socket;
3221 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3222
84382d49
MD
3223 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3224 if (ret) {
3225 if (ret > 0) {
3226 /* should exit */
3227 err = 0;
3228 }
3bd1e081
MD
3229 goto end;
3230 }
3231 DBG("Connection on client_socket");
3232
3233 /* Blocking call, waiting for transmission */
3234 sock = lttcomm_accept_unix_sock(client_socket);
534d2592 3235 if (sock < 0) {
3bd1e081
MD
3236 WARN("On accept");
3237 goto end;
3238 }
3bd1e081 3239
331744e3
JD
3240 /*
3241 * Setup metadata socket which is the second socket connection on the
3242 * command unix socket.
3243 */
3244 ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket);
84382d49
MD
3245 if (ret) {
3246 if (ret > 0) {
3247 /* should exit */
3248 err = 0;
3249 }
331744e3
JD
3250 goto end;
3251 }
3252