Fix: consumer: Move sanity check within `consumer_subbuffer` functions
[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
d2956687
JG
880/*
881 * Trigger a dump of the metadata content. Following/during the succesful
882 * completion of this call, the metadata poll thread will start receiving
883 * metadata packets to consume.
884 *
885 * The caller must hold the channel and stream locks.
886 */
887static
888int consumer_metadata_stream_dump(struct lttng_consumer_stream *stream)
889{
890 int ret;
891
892 ASSERT_LOCKED(stream->chan->lock);
893 ASSERT_LOCKED(stream->lock);
894 assert(stream->metadata_flag);
895 assert(stream->chan->trace_chunk);
896
897 switch (consumer_data.type) {
898 case LTTNG_CONSUMER_KERNEL:
899 /*
900 * Reset the position of what has been read from the
901 * metadata cache to 0 so we can dump it again.
902 */
903 ret = kernctl_metadata_cache_dump(stream->wait_fd);
904 break;
905 case LTTNG_CONSUMER32_UST:
906 case LTTNG_CONSUMER64_UST:
907 /*
908 * Reset the position pushed from the metadata cache so it
909 * will write from the beginning on the next push.
910 */
911 stream->ust_metadata_pushed = 0;
912 ret = consumer_metadata_wakeup_pipe(stream->chan);
913 break;
914 default:
915 ERR("Unknown consumer_data type");
916 abort();
917 }
918 if (ret < 0) {
919 ERR("Failed to dump the metadata cache");
920 }
921 return ret;
922}
923
924static
925int lttng_consumer_channel_set_trace_chunk(
926 struct lttng_consumer_channel *channel,
927 struct lttng_trace_chunk *new_trace_chunk)
928{
d2956687 929 pthread_mutex_lock(&channel->lock);
b6921a17
JG
930 if (channel->is_deleted) {
931 /*
932 * The channel has been logically deleted and should no longer
933 * be used. It has released its reference to its current trace
934 * chunk and should not acquire a new one.
935 *
936 * Return success as there is nothing for the caller to do.
937 */
938 goto end;
939 }
d2956687
JG
940
941 /*
942 * The acquisition of the reference cannot fail (barring
943 * a severe internal error) since a reference to the published
944 * chunk is already held by the caller.
945 */
946 if (new_trace_chunk) {
947 const bool acquired_reference = lttng_trace_chunk_get(
948 new_trace_chunk);
949
950 assert(acquired_reference);
951 }
952
953 lttng_trace_chunk_put(channel->trace_chunk);
954 channel->trace_chunk = new_trace_chunk;
d2956687
JG
955end:
956 pthread_mutex_unlock(&channel->lock);
ce1aa6fe 957 return 0;
d2956687
JG
958}
959
3bd1e081 960/*
ffe60014
DG
961 * Allocate and return a new lttng_consumer_channel object using the given key
962 * to initialize the hash table node.
963 *
964 * On error, return NULL.
3bd1e081 965 */
886224ff 966struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key,
ffe60014 967 uint64_t session_id,
d2956687 968 const uint64_t *chunk_id,
ffe60014
DG
969 const char *pathname,
970 const char *name,
57a269f2 971 uint64_t relayd_id,
1624d5b7
JD
972 enum lttng_event_output output,
973 uint64_t tracefile_size,
2bba9e53 974 uint64_t tracefile_count,
1950109e 975 uint64_t session_id_per_pid,
ecc48a90 976 unsigned int monitor,
d7ba1388 977 unsigned int live_timer_interval,
a2814ea7 978 bool is_in_live_session,
3d071855 979 const char *root_shm_path,
d7ba1388 980 const char *shm_path)
3bd1e081 981{
d2956687
JG
982 struct lttng_consumer_channel *channel = NULL;
983 struct lttng_trace_chunk *trace_chunk = NULL;
984
985 if (chunk_id) {
986 trace_chunk = lttng_trace_chunk_registry_find_chunk(
987 consumer_data.chunk_registry, session_id,
988 *chunk_id);
989 if (!trace_chunk) {
990 ERR("Failed to find trace chunk reference during creation of channel");
991 goto end;
992 }
993 }
3bd1e081 994
276b26d1 995 channel = zmalloc(sizeof(*channel));
3bd1e081 996 if (channel == NULL) {
7a57cf92 997 PERROR("malloc struct lttng_consumer_channel");
3bd1e081
MD
998 goto end;
999 }
ffe60014
DG
1000
1001 channel->key = key;
3bd1e081 1002 channel->refcount = 0;
ffe60014 1003 channel->session_id = session_id;
1950109e 1004 channel->session_id_per_pid = session_id_per_pid;
ffe60014 1005 channel->relayd_id = relayd_id;
1624d5b7
JD
1006 channel->tracefile_size = tracefile_size;
1007 channel->tracefile_count = tracefile_count;
2bba9e53 1008 channel->monitor = monitor;
ecc48a90 1009 channel->live_timer_interval = live_timer_interval;
a2814ea7 1010 channel->is_live = is_in_live_session;
a9838785 1011 pthread_mutex_init(&channel->lock, NULL);
ec6ea7d0 1012 pthread_mutex_init(&channel->timer_lock, NULL);
ffe60014 1013
0c759fc9
DG
1014 switch (output) {
1015 case LTTNG_EVENT_SPLICE:
1016 channel->output = CONSUMER_CHANNEL_SPLICE;
1017 break;
1018 case LTTNG_EVENT_MMAP:
1019 channel->output = CONSUMER_CHANNEL_MMAP;
1020 break;
1021 default:
1022 assert(0);
1023 free(channel);
1024 channel = NULL;
1025 goto end;
1026 }
1027
07b86b52
JD
1028 /*
1029 * In monitor mode, the streams associated with the channel will be put in
1030 * a special list ONLY owned by this channel. So, the refcount is set to 1
1031 * here meaning that the channel itself has streams that are referenced.
1032 *
1033 * On a channel deletion, once the channel is no longer visible, the
1034 * refcount is decremented and checked for a zero value to delete it. With
1035 * streams in no monitor mode, it will now be safe to destroy the channel.
1036 */
1037 if (!channel->monitor) {
1038 channel->refcount = 1;
1039 }
1040
ffe60014
DG
1041 strncpy(channel->pathname, pathname, sizeof(channel->pathname));
1042 channel->pathname[sizeof(channel->pathname) - 1] = '\0';
1043
1044 strncpy(channel->name, name, sizeof(channel->name));
1045 channel->name[sizeof(channel->name) - 1] = '\0';
1046
3d071855
MD
1047 if (root_shm_path) {
1048 strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path));
1049 channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0';
1050 }
d7ba1388
MD
1051 if (shm_path) {
1052 strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path));
1053 channel->shm_path[sizeof(channel->shm_path) - 1] = '\0';
1054 }
1055
d88aee68 1056 lttng_ht_node_init_u64(&channel->node, channel->key);
5c3892a6
JG
1057 lttng_ht_node_init_u64(&channel->channels_by_session_id_ht_node,
1058 channel->session_id);
d8ef542d
MD
1059
1060 channel->wait_fd = -1;
ffe60014
DG
1061 CDS_INIT_LIST_HEAD(&channel->streams.head);
1062
d2956687
JG
1063 if (trace_chunk) {
1064 int ret = lttng_consumer_channel_set_trace_chunk(channel,
1065 trace_chunk);
1066 if (ret) {
1067 goto error;
1068 }
1069 }
1070
62a7b8ed 1071 DBG("Allocated channel (key %" PRIu64 ")", channel->key);
3bd1e081 1072
3bd1e081 1073end:
d2956687 1074 lttng_trace_chunk_put(trace_chunk);
3bd1e081 1075 return channel;
d2956687
JG
1076error:
1077 consumer_del_channel(channel);
1078 channel = NULL;
1079 goto end;
3bd1e081
MD
1080}
1081
1082/*
1083 * Add a channel to the global list protected by a mutex.
821fffb2 1084 *
b5a6470f 1085 * Always return 0 indicating success.
3bd1e081 1086 */
d8ef542d
MD
1087int consumer_add_channel(struct lttng_consumer_channel *channel,
1088 struct lttng_consumer_local_data *ctx)
3bd1e081 1089{
3bd1e081 1090 pthread_mutex_lock(&consumer_data.lock);
a9838785 1091 pthread_mutex_lock(&channel->lock);
ec6ea7d0 1092 pthread_mutex_lock(&channel->timer_lock);
c77fc10a 1093
b5a6470f
DG
1094 /*
1095 * This gives us a guarantee that the channel we are about to add to the
1096 * channel hash table will be unique. See this function comment on the why
1097 * we need to steel the channel key at this stage.
1098 */
1099 steal_channel_key(channel->key);
c77fc10a 1100
b5a6470f 1101 rcu_read_lock();
d88aee68 1102 lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node);
5c3892a6
JG
1103 lttng_ht_add_u64(consumer_data.channels_by_session_id_ht,
1104 &channel->channels_by_session_id_ht_node);
6065ceec 1105 rcu_read_unlock();
d2956687 1106 channel->is_published = true;
b5a6470f 1107
ec6ea7d0 1108 pthread_mutex_unlock(&channel->timer_lock);
a9838785 1109 pthread_mutex_unlock(&channel->lock);
3bd1e081 1110 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 1111
b5a6470f 1112 if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) {
a0cbdd2e 1113 notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD);
d8ef542d 1114 }
b5a6470f
DG
1115
1116 return 0;
3bd1e081
MD
1117}
1118
1119/*
1120 * Allocate the pollfd structure and the local view of the out fds to avoid
1121 * doing a lookup in the linked list and concurrency issues when writing is
1122 * needed. Called with consumer_data.lock held.
1123 *
1124 * Returns the number of fds in the structures.
1125 */
ffe60014
DG
1126static int update_poll_array(struct lttng_consumer_local_data *ctx,
1127 struct pollfd **pollfd, struct lttng_consumer_stream **local_stream,
9a2fcf78 1128 struct lttng_ht *ht, int *nb_inactive_fd)
3bd1e081 1129{
3bd1e081 1130 int i = 0;
e4421fec
DG
1131 struct lttng_ht_iter iter;
1132 struct lttng_consumer_stream *stream;
3bd1e081 1133
ffe60014
DG
1134 assert(ctx);
1135 assert(ht);
1136 assert(pollfd);
1137 assert(local_stream);
1138
3bd1e081 1139 DBG("Updating poll fd array");
9a2fcf78 1140 *nb_inactive_fd = 0;
481d6c57 1141 rcu_read_lock();
43c34bc3 1142 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
8994307f
DG
1143 /*
1144 * Only active streams with an active end point can be added to the
1145 * poll set and local stream storage of the thread.
1146 *
1147 * There is a potential race here for endpoint_status to be updated
1148 * just after the check. However, this is OK since the stream(s) will
1149 * be deleted once the thread is notified that the end point state has
1150 * changed where this function will be called back again.
9a2fcf78
JD
1151 *
1152 * We track the number of inactive FDs because they still need to be
1153 * closed by the polling thread after a wakeup on the data_pipe or
1154 * metadata_pipe.
8994307f 1155 */
d2956687 1156 if (stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) {
9a2fcf78 1157 (*nb_inactive_fd)++;
3bd1e081
MD
1158 continue;
1159 }
7972aab2
DG
1160 /*
1161 * This clobbers way too much the debug output. Uncomment that if you
1162 * need it for debugging purposes.
7972aab2 1163 */
e4421fec 1164 (*pollfd)[i].fd = stream->wait_fd;
3bd1e081 1165 (*pollfd)[i].events = POLLIN | POLLPRI;
e4421fec 1166 local_stream[i] = stream;
3bd1e081
MD
1167 i++;
1168 }
481d6c57 1169 rcu_read_unlock();
3bd1e081
MD
1170
1171 /*
50f8ae69 1172 * Insert the consumer_data_pipe at the end of the array and don't
3bd1e081
MD
1173 * increment i so nb_fd is the number of real FD.
1174 */
acdb9057 1175 (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe);
509bb1cf 1176 (*pollfd)[i].events = POLLIN | POLLPRI;
02b3d176
DG
1177
1178 (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe);
1179 (*pollfd)[i + 1].events = POLLIN | POLLPRI;
3bd1e081
MD
1180 return i;
1181}
1182
1183/*
84382d49
MD
1184 * Poll on the should_quit pipe and the command socket return -1 on
1185 * error, 1 if should exit, 0 if data is available on the command socket
3bd1e081
MD
1186 */
1187int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll)
1188{
1189 int num_rdy;
1190
88f2b785 1191restart:
3bd1e081
MD
1192 num_rdy = poll(consumer_sockpoll, 2, -1);
1193 if (num_rdy == -1) {
88f2b785
MD
1194 /*
1195 * Restart interrupted system call.
1196 */
1197 if (errno == EINTR) {
1198 goto restart;
1199 }
7a57cf92 1200 PERROR("Poll error");
84382d49 1201 return -1;
3bd1e081 1202 }
509bb1cf 1203 if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) {
3bd1e081 1204 DBG("consumer_should_quit wake up");
84382d49 1205 return 1;
3bd1e081
MD
1206 }
1207 return 0;
3bd1e081
MD
1208}
1209
1210/*
1211 * Set the error socket.
1212 */
ffe60014
DG
1213void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx,
1214 int sock)
3bd1e081
MD
1215{
1216 ctx->consumer_error_socket = sock;
1217}
1218
1219/*
1220 * Set the command socket path.
1221 */
3bd1e081
MD
1222void lttng_consumer_set_command_sock_path(
1223 struct lttng_consumer_local_data *ctx, char *sock)
1224{
1225 ctx->consumer_command_sock_path = sock;
1226}
1227
1228/*
1229 * Send return code to the session daemon.
1230 * If the socket is not defined, we return 0, it is not a fatal error
1231 */
ffe60014 1232int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd)
3bd1e081
MD
1233{
1234 if (ctx->consumer_error_socket > 0) {
1235 return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd,
1236 sizeof(enum lttcomm_sessiond_command));
1237 }
1238
1239 return 0;
1240}
1241
1242/*
228b5bf7
DG
1243 * Close all the tracefiles and stream fds and MUST be called when all
1244 * instances are destroyed i.e. when all threads were joined and are ended.
3bd1e081
MD
1245 */
1246void lttng_consumer_cleanup(void)
1247{
e4421fec 1248 struct lttng_ht_iter iter;
ffe60014 1249 struct lttng_consumer_channel *channel;
e10aec8f 1250 unsigned int trace_chunks_left;
6065ceec
DG
1251
1252 rcu_read_lock();
3bd1e081 1253
ffe60014
DG
1254 cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel,
1255 node.node) {
702b1ea4 1256 consumer_del_channel(channel);
3bd1e081 1257 }
6065ceec
DG
1258
1259 rcu_read_unlock();
d6ce1df2 1260
d6ce1df2 1261 lttng_ht_destroy(consumer_data.channel_ht);
5c3892a6 1262 lttng_ht_destroy(consumer_data.channels_by_session_id_ht);
228b5bf7
DG
1263
1264 cleanup_relayd_ht();
1265
d8ef542d
MD
1266 lttng_ht_destroy(consumer_data.stream_per_chan_id_ht);
1267
228b5bf7
DG
1268 /*
1269 * This HT contains streams that are freed by either the metadata thread or
1270 * the data thread so we do *nothing* on the hash table and simply destroy
1271 * it.
1272 */
1273 lttng_ht_destroy(consumer_data.stream_list_ht);
28cc88f3 1274
e10aec8f
MD
1275 /*
1276 * Trace chunks in the registry may still exist if the session
1277 * daemon has encountered an internal error and could not
1278 * tear down its sessions and/or trace chunks properly.
1279 *
1280 * Release the session daemon's implicit reference to any remaining
1281 * trace chunk and print an error if any trace chunk was found. Note
1282 * that there are _no_ legitimate cases for trace chunks to be left,
1283 * it is a leak. However, it can happen following a crash of the
1284 * session daemon and not emptying the registry would cause an assertion
1285 * to hit.
1286 */
1287 trace_chunks_left = lttng_trace_chunk_registry_put_each_chunk(
1288 consumer_data.chunk_registry);
1289 if (trace_chunks_left) {
1290 ERR("%u trace chunks are leaked by lttng-consumerd. "
1291 "This can be caused by an internal error of the session daemon.",
1292 trace_chunks_left);
1293 }
1294 /* Run all callbacks freeing each chunk. */
1295 rcu_barrier();
28cc88f3 1296 lttng_trace_chunk_registry_destroy(consumer_data.chunk_registry);
3bd1e081
MD
1297}
1298
1299/*
1300 * Called from signal handler.
1301 */
1302void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx)
1303{
6cd525e8
MD
1304 ssize_t ret;
1305
10211f5c 1306 CMM_STORE_SHARED(consumer_quit, 1);
6cd525e8
MD
1307 ret = lttng_write(ctx->consumer_should_quit[1], "4", 1);
1308 if (ret < 1) {
7a57cf92 1309 PERROR("write consumer quit");
3bd1e081 1310 }
ab1027f4
DG
1311
1312 DBG("Consumer flag that it should quit");
3bd1e081
MD
1313}
1314
5199ffc4
JG
1315
1316/*
1317 * Flush pending writes to trace output disk file.
1318 */
1319static
00e2e675
DG
1320void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream,
1321 off_t orig_offset)
3bd1e081 1322{
c7a78aab 1323 int ret;
3bd1e081
MD
1324 int outfd = stream->out_fd;
1325
1326 /*
1327 * This does a blocking write-and-wait on any page that belongs to the
1328 * subbuffer prior to the one we just wrote.
1329 * Don't care about error values, as these are just hints and ways to
1330 * limit the amount of page cache used.
1331 */
ffe60014 1332 if (orig_offset < stream->max_sb_size) {
3bd1e081
MD
1333 return;
1334 }
ffe60014
DG
1335 lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size,
1336 stream->max_sb_size,
3bd1e081
MD
1337 SYNC_FILE_RANGE_WAIT_BEFORE
1338 | SYNC_FILE_RANGE_WRITE
1339 | SYNC_FILE_RANGE_WAIT_AFTER);
1340 /*
1341 * Give hints to the kernel about how we access the file:
1342 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1343 * we write it.
1344 *
1345 * We need to call fadvise again after the file grows because the
1346 * kernel does not seem to apply fadvise to non-existing parts of the
1347 * file.
1348 *
1349 * Call fadvise _after_ having waited for the page writeback to
1350 * complete because the dirty page writeback semantic is not well
1351 * defined. So it can be expected to lead to lower throughput in
1352 * streaming.
1353 */
c7a78aab 1354 ret = posix_fadvise(outfd, orig_offset - stream->max_sb_size,
ffe60014 1355 stream->max_sb_size, POSIX_FADV_DONTNEED);
a0d0e127 1356 if (ret && ret != -ENOSYS) {
a74a5f4a
JG
1357 errno = ret;
1358 PERROR("posix_fadvise on fd %i", outfd);
c7a78aab 1359 }
3bd1e081
MD
1360}
1361
1362/*
1363 * Initialise the necessary environnement :
1364 * - create a new context
1365 * - create the poll_pipe
1366 * - create the should_quit pipe (for signal handler)
1367 * - create the thread pipe (for splice)
1368 *
1369 * Takes a function pointer as argument, this function is called when data is
1370 * available on a buffer. This function is responsible to do the
1371 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1372 * buffer configuration and then kernctl_put_next_subbuf at the end.
1373 *
1374 * Returns a pointer to the new context or NULL on error.
1375 */
1376struct lttng_consumer_local_data *lttng_consumer_create(
1377 enum lttng_consumer_type type,
4078b776 1378 ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream,
6f9449c2 1379 struct lttng_consumer_local_data *ctx, bool locked_by_caller),
3bd1e081
MD
1380 int (*recv_channel)(struct lttng_consumer_channel *channel),
1381 int (*recv_stream)(struct lttng_consumer_stream *stream),
30319bcb 1382 int (*update_stream)(uint64_t stream_key, uint32_t state))
3bd1e081 1383{
d8ef542d 1384 int ret;
3bd1e081
MD
1385 struct lttng_consumer_local_data *ctx;
1386
1387 assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN ||
1388 consumer_data.type == type);
1389 consumer_data.type = type;
1390
effcf122 1391 ctx = zmalloc(sizeof(struct lttng_consumer_local_data));
3bd1e081 1392 if (ctx == NULL) {
7a57cf92 1393 PERROR("allocating context");
3bd1e081
MD
1394 goto error;
1395 }
1396
1397 ctx->consumer_error_socket = -1;
331744e3 1398 ctx->consumer_metadata_socket = -1;
75d83e50 1399 pthread_mutex_init(&ctx->metadata_socket_lock, NULL);
3bd1e081
MD
1400 /* assign the callbacks */
1401 ctx->on_buffer_ready = buffer_ready;
1402 ctx->on_recv_channel = recv_channel;
1403 ctx->on_recv_stream = recv_stream;
1404 ctx->on_update_stream = update_stream;
1405
acdb9057
DG
1406 ctx->consumer_data_pipe = lttng_pipe_open(0);
1407 if (!ctx->consumer_data_pipe) {
3bd1e081
MD
1408 goto error_poll_pipe;
1409 }
1410
02b3d176
DG
1411 ctx->consumer_wakeup_pipe = lttng_pipe_open(0);
1412 if (!ctx->consumer_wakeup_pipe) {
1413 goto error_wakeup_pipe;
1414 }
1415
3bd1e081
MD
1416 ret = pipe(ctx->consumer_should_quit);
1417 if (ret < 0) {
7a57cf92 1418 PERROR("Error creating recv pipe");
3bd1e081
MD
1419 goto error_quit_pipe;
1420 }
1421
d8ef542d
MD
1422 ret = pipe(ctx->consumer_channel_pipe);
1423 if (ret < 0) {
1424 PERROR("Error creating channel pipe");
1425 goto error_channel_pipe;
1426 }
1427
13886d2d
DG
1428 ctx->consumer_metadata_pipe = lttng_pipe_open(0);
1429 if (!ctx->consumer_metadata_pipe) {
fb3a43a9
DG
1430 goto error_metadata_pipe;
1431 }
3bd1e081 1432
e9404c27
JG
1433 ctx->channel_monitor_pipe = -1;
1434
fb3a43a9 1435 return ctx;
3bd1e081 1436
fb3a43a9 1437error_metadata_pipe:
d8ef542d
MD
1438 utils_close_pipe(ctx->consumer_channel_pipe);
1439error_channel_pipe:
d8ef542d 1440 utils_close_pipe(ctx->consumer_should_quit);
3bd1e081 1441error_quit_pipe:
02b3d176
DG
1442 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
1443error_wakeup_pipe:
acdb9057 1444 lttng_pipe_destroy(ctx->consumer_data_pipe);
3bd1e081
MD
1445error_poll_pipe:
1446 free(ctx);
1447error:
1448 return NULL;
1449}
1450
282dadbc
MD
1451/*
1452 * Iterate over all streams of the hashtable and free them properly.
1453 */
1454static void destroy_data_stream_ht(struct lttng_ht *ht)
1455{
1456 struct lttng_ht_iter iter;
1457 struct lttng_consumer_stream *stream;
1458
1459 if (ht == NULL) {
1460 return;
1461 }
1462
1463 rcu_read_lock();
1464 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1465 /*
1466 * Ignore return value since we are currently cleaning up so any error
1467 * can't be handled.
1468 */
1469 (void) consumer_del_stream(stream, ht);
1470 }
1471 rcu_read_unlock();
1472
1473 lttng_ht_destroy(ht);
1474}
1475
1476/*
1477 * Iterate over all streams of the metadata hashtable and free them
1478 * properly.
1479 */
1480static void destroy_metadata_stream_ht(struct lttng_ht *ht)
1481{
1482 struct lttng_ht_iter iter;
1483 struct lttng_consumer_stream *stream;
1484
1485 if (ht == NULL) {
1486 return;
1487 }
1488
1489 rcu_read_lock();
1490 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1491 /*
1492 * Ignore return value since we are currently cleaning up so any error
1493 * can't be handled.
1494 */
1495 (void) consumer_del_metadata_stream(stream, ht);
1496 }
1497 rcu_read_unlock();
1498
1499 lttng_ht_destroy(ht);
1500}
1501
3bd1e081
MD
1502/*
1503 * Close all fds associated with the instance and free the context.
1504 */
1505void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx)
1506{
4c462e79
MD
1507 int ret;
1508
ab1027f4
DG
1509 DBG("Consumer destroying it. Closing everything.");
1510
4f2e75b9
DG
1511 if (!ctx) {
1512 return;
1513 }
1514
282dadbc
MD
1515 destroy_data_stream_ht(data_ht);
1516 destroy_metadata_stream_ht(metadata_ht);
1517
4c462e79
MD
1518 ret = close(ctx->consumer_error_socket);
1519 if (ret) {
1520 PERROR("close");
1521 }
331744e3
JD
1522 ret = close(ctx->consumer_metadata_socket);
1523 if (ret) {
1524 PERROR("close");
1525 }
d8ef542d 1526 utils_close_pipe(ctx->consumer_channel_pipe);
acdb9057 1527 lttng_pipe_destroy(ctx->consumer_data_pipe);
13886d2d 1528 lttng_pipe_destroy(ctx->consumer_metadata_pipe);
02b3d176 1529 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
d8ef542d 1530 utils_close_pipe(ctx->consumer_should_quit);
fb3a43a9 1531
3bd1e081
MD
1532 unlink(ctx->consumer_command_sock_path);
1533 free(ctx);
1534}
1535
6197aea7
DG
1536/*
1537 * Write the metadata stream id on the specified file descriptor.
1538 */
1539static int write_relayd_metadata_id(int fd,
1540 struct lttng_consumer_stream *stream,
239f61af 1541 unsigned long padding)
6197aea7 1542{
6cd525e8 1543 ssize_t ret;
1d4dfdef 1544 struct lttcomm_relayd_metadata_payload hdr;
6197aea7 1545
1d4dfdef
DG
1546 hdr.stream_id = htobe64(stream->relayd_stream_id);
1547 hdr.padding_size = htobe32(padding);
6cd525e8
MD
1548 ret = lttng_write(fd, (void *) &hdr, sizeof(hdr));
1549 if (ret < sizeof(hdr)) {
d7b75ec8 1550 /*
6f04ed72 1551 * This error means that the fd's end is closed so ignore the PERROR
d7b75ec8
DG
1552 * not to clubber the error output since this can happen in a normal
1553 * code path.
1554 */
1555 if (errno != EPIPE) {
1556 PERROR("write metadata stream id");
1557 }
1558 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno);
534d2592
DG
1559 /*
1560 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1561 * handle writting the missing part so report that as an error and
1562 * don't lie to the caller.
1563 */
1564 ret = -1;
6197aea7
DG
1565 goto end;
1566 }
1d4dfdef
DG
1567 DBG("Metadata stream id %" PRIu64 " with padding %lu written before data",
1568 stream->relayd_stream_id, padding);
6197aea7
DG
1569
1570end:
6cd525e8 1571 return (int) ret;
6197aea7
DG
1572}
1573
3bd1e081 1574/*
09e26845
DG
1575 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1576 * core function for writing trace buffers to either the local filesystem or
1577 * the network.
1578 *
d2956687 1579 * It must be called with the stream and the channel lock held.
79d4ffb7 1580 *
09e26845 1581 * Careful review MUST be put if any changes occur!
3bd1e081
MD
1582 *
1583 * Returns the number of bytes written
1584 */
4078b776 1585ssize_t lttng_consumer_on_read_subbuffer_mmap(
128708c3 1586 struct lttng_consumer_stream *stream,
fd424d99 1587 const struct lttng_buffer_view *buffer,
6f9449c2 1588 unsigned long padding)
3bd1e081 1589{
994ab360 1590 ssize_t ret = 0;
f02e1e8a
DG
1591 off_t orig_offset = stream->out_fd_offset;
1592 /* Default is on the disk */
1593 int outfd = stream->out_fd;
f02e1e8a 1594 struct consumer_relayd_sock_pair *relayd = NULL;
8994307f 1595 unsigned int relayd_hang_up = 0;
fd424d99
JG
1596 const size_t subbuf_content_size = buffer->size - padding;
1597 size_t write_len;
f02e1e8a
DG
1598
1599 /* RCU lock for the relayd pointer */
1600 rcu_read_lock();
7fd975c5 1601 assert(stream->net_seq_idx != (uint64_t) -1ULL ||
948411cd 1602 stream->trace_chunk);
d2956687 1603
f02e1e8a 1604 /* Flag that the current stream if set for network streaming. */
da009f2c 1605 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1606 relayd = consumer_find_relayd(stream->net_seq_idx);
1607 if (relayd == NULL) {
56591bac 1608 ret = -EPIPE;
f02e1e8a
DG
1609 goto end;
1610 }
1611 }
1612
f02e1e8a
DG
1613 /* Handle stream on the relayd if the output is on the network */
1614 if (relayd) {
fd424d99 1615 unsigned long netlen = subbuf_content_size;
f02e1e8a
DG
1616
1617 /*
1618 * Lock the control socket for the complete duration of the function
1619 * since from this point on we will use the socket.
1620 */
1621 if (stream->metadata_flag) {
1622 /* Metadata requires the control socket. */
1623 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
93ec662e
JD
1624 if (stream->reset_metadata_flag) {
1625 ret = relayd_reset_metadata(&relayd->control_sock,
1626 stream->relayd_stream_id,
1627 stream->metadata_version);
1628 if (ret < 0) {
1629 relayd_hang_up = 1;
1630 goto write_error;
1631 }
1632 stream->reset_metadata_flag = 0;
1633 }
1d4dfdef 1634 netlen += sizeof(struct lttcomm_relayd_metadata_payload);
f02e1e8a
DG
1635 }
1636
1d4dfdef 1637 ret = write_relayd_stream_header(stream, netlen, padding, relayd);
994ab360
DG
1638 if (ret < 0) {
1639 relayd_hang_up = 1;
1640 goto write_error;
1641 }
1642 /* Use the returned socket. */
1643 outfd = ret;
f02e1e8a 1644
994ab360
DG
1645 /* Write metadata stream id before payload */
1646 if (stream->metadata_flag) {
239f61af 1647 ret = write_relayd_metadata_id(outfd, stream, padding);
994ab360 1648 if (ret < 0) {
8994307f
DG
1649 relayd_hang_up = 1;
1650 goto write_error;
1651 }
f02e1e8a 1652 }
1624d5b7 1653
fd424d99
JG
1654 write_len = subbuf_content_size;
1655 } else {
1656 /* No streaming; we have to write the full padding. */
93ec662e
JD
1657 if (stream->metadata_flag && stream->reset_metadata_flag) {
1658 ret = utils_truncate_stream_file(stream->out_fd, 0);
1659 if (ret < 0) {
1660 ERR("Reset metadata file");
1661 goto end;
1662 }
1663 stream->reset_metadata_flag = 0;
1664 }
1665
1624d5b7
JD
1666 /*
1667 * Check if we need to change the tracefile before writing the packet.
1668 */
1669 if (stream->chan->tracefile_size > 0 &&
fd424d99 1670 (stream->tracefile_size_current + buffer->size) >
1624d5b7 1671 stream->chan->tracefile_size) {
d2956687
JG
1672 ret = consumer_stream_rotate_output_files(stream);
1673 if (ret) {
1624d5b7
JD
1674 goto end;
1675 }
309167d2 1676 outfd = stream->out_fd;
a1ae300f 1677 orig_offset = 0;
1624d5b7 1678 }
fd424d99 1679 stream->tracefile_size_current += buffer->size;
fd424d99 1680 write_len = buffer->size;
f02e1e8a
DG
1681 }
1682
d02b8372
DG
1683 /*
1684 * This call guarantee that len or less is returned. It's impossible to
1685 * receive a ret value that is bigger than len.
1686 */
fd424d99 1687 ret = lttng_write(outfd, buffer->data, write_len);
e2d1190b 1688 DBG("Consumer mmap write() ret %zd (len %zu)", ret, write_len);
fd424d99 1689 if (ret < 0 || ((size_t) ret != write_len)) {
d02b8372
DG
1690 /*
1691 * Report error to caller if nothing was written else at least send the
1692 * amount written.
1693 */
1694 if (ret < 0) {
994ab360 1695 ret = -errno;
f02e1e8a 1696 }
994ab360 1697 relayd_hang_up = 1;
f02e1e8a 1698
d02b8372 1699 /* Socket operation failed. We consider the relayd dead */
fcf0f774 1700 if (errno == EPIPE) {
d02b8372
DG
1701 /*
1702 * This is possible if the fd is closed on the other side
1703 * (outfd) or any write problem. It can be verbose a bit for a
1704 * normal execution if for instance the relayd is stopped
1705 * abruptly. This can happen so set this to a DBG statement.
1706 */
1707 DBG("Consumer mmap write detected relayd hang up");
994ab360
DG
1708 } else {
1709 /* Unhandled error, print it and stop function right now. */
fd424d99
JG
1710 PERROR("Error in write mmap (ret %zd != write_len %zu)", ret,
1711 write_len);
f02e1e8a 1712 }
994ab360 1713 goto write_error;
d02b8372
DG
1714 }
1715 stream->output_written += ret;
d02b8372
DG
1716
1717 /* This call is useless on a socket so better save a syscall. */
1718 if (!relayd) {
1719 /* This won't block, but will start writeout asynchronously */
fd424d99 1720 lttng_sync_file_range(outfd, stream->out_fd_offset, write_len,
d02b8372 1721 SYNC_FILE_RANGE_WRITE);
fd424d99 1722 stream->out_fd_offset += write_len;
f5dbe415 1723 lttng_consumer_sync_trace_file(stream, orig_offset);
f02e1e8a 1724 }
f02e1e8a 1725
8994307f
DG
1726write_error:
1727 /*
1728 * This is a special case that the relayd has closed its socket. Let's
1729 * cleanup the relayd object and all associated streams.
1730 */
1731 if (relayd && relayd_hang_up) {
9276e5c8
JR
1732 ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
1733 lttng_consumer_cleanup_relayd(relayd);
8994307f
DG
1734 }
1735
f02e1e8a
DG
1736end:
1737 /* Unlock only if ctrl socket used */
1738 if (relayd && stream->metadata_flag) {
1739 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1740 }
1741
1742 rcu_read_unlock();
994ab360 1743 return ret;
3bd1e081
MD
1744}
1745
1746/*
1747 * Splice the data from the ring buffer to the tracefile.
1748 *
79d4ffb7
DG
1749 * It must be called with the stream lock held.
1750 *
3bd1e081
MD
1751 * Returns the number of bytes spliced.
1752 */
4078b776 1753ssize_t lttng_consumer_on_read_subbuffer_splice(
3bd1e081 1754 struct lttng_consumer_local_data *ctx,
1d4dfdef 1755 struct lttng_consumer_stream *stream, unsigned long len,
6f9449c2 1756 unsigned long padding)
3bd1e081 1757{
f02e1e8a
DG
1758 ssize_t ret = 0, written = 0, ret_splice = 0;
1759 loff_t offset = 0;
1760 off_t orig_offset = stream->out_fd_offset;
1761 int fd = stream->wait_fd;
1762 /* Default is on the disk */
1763 int outfd = stream->out_fd;
f02e1e8a 1764 struct consumer_relayd_sock_pair *relayd = NULL;
fb3a43a9 1765 int *splice_pipe;
8994307f 1766 unsigned int relayd_hang_up = 0;
f02e1e8a 1767
3bd1e081
MD
1768 switch (consumer_data.type) {
1769 case LTTNG_CONSUMER_KERNEL:
f02e1e8a 1770 break;
7753dea8
MD
1771 case LTTNG_CONSUMER32_UST:
1772 case LTTNG_CONSUMER64_UST:
f02e1e8a 1773 /* Not supported for user space tracing */
3bd1e081
MD
1774 return -ENOSYS;
1775 default:
1776 ERR("Unknown consumer_data type");
1777 assert(0);
3bd1e081
MD
1778 }
1779
f02e1e8a
DG
1780 /* RCU lock for the relayd pointer */
1781 rcu_read_lock();
1782
1783 /* Flag that the current stream if set for network streaming. */
da009f2c 1784 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1785 relayd = consumer_find_relayd(stream->net_seq_idx);
1786 if (relayd == NULL) {
ad0b0d23 1787 written = -ret;
f02e1e8a
DG
1788 goto end;
1789 }
1790 }
a2361a61 1791 splice_pipe = stream->splice_pipe;
fb3a43a9 1792
f02e1e8a 1793 /* Write metadata stream id before payload */
1d4dfdef 1794 if (relayd) {
ad0b0d23 1795 unsigned long total_len = len;
f02e1e8a 1796
1d4dfdef
DG
1797 if (stream->metadata_flag) {
1798 /*
1799 * Lock the control socket for the complete duration of the function
1800 * since from this point on we will use the socket.
1801 */
1802 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1803
93ec662e
JD
1804 if (stream->reset_metadata_flag) {
1805 ret = relayd_reset_metadata(&relayd->control_sock,
1806 stream->relayd_stream_id,
1807 stream->metadata_version);
1808 if (ret < 0) {
1809 relayd_hang_up = 1;
1810 goto write_error;
1811 }
1812 stream->reset_metadata_flag = 0;
1813 }
239f61af 1814 ret = write_relayd_metadata_id(splice_pipe[1], stream,
1d4dfdef
DG
1815 padding);
1816 if (ret < 0) {
1817 written = ret;
ad0b0d23
DG
1818 relayd_hang_up = 1;
1819 goto write_error;
1d4dfdef
DG
1820 }
1821
1822 total_len += sizeof(struct lttcomm_relayd_metadata_payload);
1823 }
1824
1825 ret = write_relayd_stream_header(stream, total_len, padding, relayd);
ad0b0d23
DG
1826 if (ret < 0) {
1827 written = ret;
1828 relayd_hang_up = 1;
1829 goto write_error;
f02e1e8a 1830 }
ad0b0d23
DG
1831 /* Use the returned socket. */
1832 outfd = ret;
1d4dfdef
DG
1833 } else {
1834 /* No streaming, we have to set the len with the full padding */
1835 len += padding;
1624d5b7 1836
93ec662e
JD
1837 if (stream->metadata_flag && stream->reset_metadata_flag) {
1838 ret = utils_truncate_stream_file(stream->out_fd, 0);
1839 if (ret < 0) {
1840 ERR("Reset metadata file");
1841 goto end;
1842 }
1843 stream->reset_metadata_flag = 0;
1844 }
1624d5b7
JD
1845 /*
1846 * Check if we need to change the tracefile before writing the packet.
1847 */
1848 if (stream->chan->tracefile_size > 0 &&
1849 (stream->tracefile_size_current + len) >
1850 stream->chan->tracefile_size) {
d2956687 1851 ret = consumer_stream_rotate_output_files(stream);
1624d5b7 1852 if (ret < 0) {
ad0b0d23 1853 written = ret;
1624d5b7
JD
1854 goto end;
1855 }
309167d2 1856 outfd = stream->out_fd;
a1ae300f 1857 orig_offset = 0;
1624d5b7
JD
1858 }
1859 stream->tracefile_size_current += len;
f02e1e8a
DG
1860 }
1861
1862 while (len > 0) {
1d4dfdef
DG
1863 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1864 (unsigned long)offset, len, fd, splice_pipe[1]);
fb3a43a9 1865 ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len,
f02e1e8a
DG
1866 SPLICE_F_MOVE | SPLICE_F_MORE);
1867 DBG("splice chan to pipe, ret %zd", ret_splice);
1868 if (ret_splice < 0) {
d02b8372 1869 ret = errno;
ad0b0d23 1870 written = -ret;
d02b8372 1871 PERROR("Error in relay splice");
f02e1e8a
DG
1872 goto splice_error;
1873 }
1874
1875 /* Handle stream on the relayd if the output is on the network */
ad0b0d23
DG
1876 if (relayd && stream->metadata_flag) {
1877 size_t metadata_payload_size =
1878 sizeof(struct lttcomm_relayd_metadata_payload);
1879
1880 /* Update counter to fit the spliced data */
1881 ret_splice += metadata_payload_size;
1882 len += metadata_payload_size;
1883 /*
1884 * We do this so the return value can match the len passed as
1885 * argument to this function.
1886 */
1887 written -= metadata_payload_size;
f02e1e8a
DG
1888 }
1889
1890 /* Splice data out */
fb3a43a9 1891 ret_splice = splice(splice_pipe[0], NULL, outfd, NULL,
f02e1e8a 1892 ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE);
a2361a61
JD
1893 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1894 outfd, ret_splice);
f02e1e8a 1895 if (ret_splice < 0) {
d02b8372 1896 ret = errno;
ad0b0d23
DG
1897 written = -ret;
1898 relayd_hang_up = 1;
1899 goto write_error;
f02e1e8a 1900 } else if (ret_splice > len) {
d02b8372
DG
1901 /*
1902 * We don't expect this code path to be executed but you never know
1903 * so this is an extra protection agains a buggy splice().
1904 */
f02e1e8a 1905 ret = errno;
ad0b0d23 1906 written += ret_splice;
d02b8372
DG
1907 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice,
1908 len);
f02e1e8a 1909 goto splice_error;
d02b8372
DG
1910 } else {
1911 /* All good, update current len and continue. */
1912 len -= ret_splice;
f02e1e8a 1913 }
f02e1e8a
DG
1914
1915 /* This call is useless on a socket so better save a syscall. */
1916 if (!relayd) {
1917 /* This won't block, but will start writeout asynchronously */
1918 lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice,
1919 SYNC_FILE_RANGE_WRITE);
1920 stream->out_fd_offset += ret_splice;
1921 }
e5d1a9b3 1922 stream->output_written += ret_splice;
f02e1e8a
DG
1923 written += ret_splice;
1924 }
f5dbe415
JG
1925 if (!relayd) {
1926 lttng_consumer_sync_trace_file(stream, orig_offset);
1927 }
f02e1e8a
DG
1928 goto end;
1929
8994307f
DG
1930write_error:
1931 /*
1932 * This is a special case that the relayd has closed its socket. Let's
1933 * cleanup the relayd object and all associated streams.
1934 */
1935 if (relayd && relayd_hang_up) {
9276e5c8
JR
1936 ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
1937 lttng_consumer_cleanup_relayd(relayd);
8994307f
DG
1938 /* Skip splice error so the consumer does not fail */
1939 goto end;
1940 }
1941
f02e1e8a
DG
1942splice_error:
1943 /* send the appropriate error description to sessiond */
1944 switch (ret) {
f02e1e8a 1945 case EINVAL:
f73fabfd 1946 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL);
f02e1e8a
DG
1947 break;
1948 case ENOMEM:
f73fabfd 1949 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM);
f02e1e8a
DG
1950 break;
1951 case ESPIPE:
f73fabfd 1952 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE);
f02e1e8a
DG
1953 break;
1954 }
1955
1956end:
1957 if (relayd && stream->metadata_flag) {
1958 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1959 }
1960
1961 rcu_read_unlock();
1962 return written;
3bd1e081
MD
1963}
1964
15055ce5
JD
1965/*
1966 * Sample the snapshot positions for a specific fd
1967 *
1968 * Returns 0 on success, < 0 on error
1969 */
1970int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream *stream)
1971{
1972 switch (consumer_data.type) {
1973 case LTTNG_CONSUMER_KERNEL:
1974 return lttng_kconsumer_sample_snapshot_positions(stream);
1975 case LTTNG_CONSUMER32_UST:
1976 case LTTNG_CONSUMER64_UST:
1977 return lttng_ustconsumer_sample_snapshot_positions(stream);
1978 default:
1979 ERR("Unknown consumer_data type");
1980 assert(0);
1981 return -ENOSYS;
1982 }
1983}
3bd1e081
MD
1984/*
1985 * Take a snapshot for a specific fd
1986 *
1987 * Returns 0 on success, < 0 on error
1988 */
ffe60014 1989int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream)
3bd1e081
MD
1990{
1991 switch (consumer_data.type) {
1992 case LTTNG_CONSUMER_KERNEL:
ffe60014 1993 return lttng_kconsumer_take_snapshot(stream);
7753dea8
MD
1994 case LTTNG_CONSUMER32_UST:
1995 case LTTNG_CONSUMER64_UST:
ffe60014 1996 return lttng_ustconsumer_take_snapshot(stream);
3bd1e081
MD
1997 default:
1998 ERR("Unknown consumer_data type");
1999 assert(0);
2000 return -ENOSYS;
2001 }
3bd1e081
MD
2002}
2003
2004/*
2005 * Get the produced position
2006 *
2007 * Returns 0 on success, < 0 on error
2008 */
ffe60014 2009int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
3bd1e081
MD
2010 unsigned long *pos)
2011{
2012 switch (consumer_data.type) {
2013 case LTTNG_CONSUMER_KERNEL:
ffe60014 2014 return lttng_kconsumer_get_produced_snapshot(stream, pos);
7753dea8
MD
2015 case LTTNG_CONSUMER32_UST:
2016 case LTTNG_CONSUMER64_UST:
ffe60014 2017 return lttng_ustconsumer_get_produced_snapshot(stream, pos);
3bd1e081
MD
2018 default:
2019 ERR("Unknown consumer_data type");
2020 assert(0);
2021 return -ENOSYS;
2022 }
2023}
2024
15055ce5
JD
2025/*
2026 * Get the consumed position (free-running counter position in bytes).
2027 *
2028 * Returns 0 on success, < 0 on error
2029 */
2030int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
2031 unsigned long *pos)
2032{
2033 switch (consumer_data.type) {
2034 case LTTNG_CONSUMER_KERNEL:
2035 return lttng_kconsumer_get_consumed_snapshot(stream, pos);
2036 case LTTNG_CONSUMER32_UST:
2037 case LTTNG_CONSUMER64_UST:
2038 return lttng_ustconsumer_get_consumed_snapshot(stream, pos);
2039 default:
2040 ERR("Unknown consumer_data type");
2041 assert(0);
2042 return -ENOSYS;
2043 }
2044}
2045
3bd1e081
MD
2046int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx,
2047 int sock, struct pollfd *consumer_sockpoll)
2048{
2049 switch (consumer_data.type) {
2050 case LTTNG_CONSUMER_KERNEL:
2051 return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
7753dea8
MD
2052 case LTTNG_CONSUMER32_UST:
2053 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
2054 return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
2055 default:
2056 ERR("Unknown consumer_data type");
2057 assert(0);
2058 return -ENOSYS;
2059 }
2060}
2061
1f8d1c14 2062static
6d574024 2063void lttng_consumer_close_all_metadata(void)
d88aee68
DG
2064{
2065 switch (consumer_data.type) {
2066 case LTTNG_CONSUMER_KERNEL:
2067 /*
2068 * The Kernel consumer has a different metadata scheme so we don't
2069 * close anything because the stream will be closed by the session
2070 * daemon.
2071 */
2072 break;
2073 case LTTNG_CONSUMER32_UST:
2074 case LTTNG_CONSUMER64_UST:
2075 /*
2076 * Close all metadata streams. The metadata hash table is passed and
2077 * this call iterates over it by closing all wakeup fd. This is safe
2078 * because at this point we are sure that the metadata producer is
2079 * either dead or blocked.
2080 */
6d574024 2081 lttng_ustconsumer_close_all_metadata(metadata_ht);
d88aee68
DG
2082 break;
2083 default:
2084 ERR("Unknown consumer_data type");
2085 assert(0);
2086 }
2087}
2088
fb3a43a9
DG
2089/*
2090 * Clean up a metadata stream and free its memory.
2091 */
e316aad5
DG
2092void consumer_del_metadata_stream(struct lttng_consumer_stream *stream,
2093 struct lttng_ht *ht)
fb3a43a9 2094{
a6ef8ee6
JG
2095 struct lttng_consumer_channel *channel = NULL;
2096 bool free_channel = false;
fb3a43a9
DG
2097
2098 assert(stream);
2099 /*
2100 * This call should NEVER receive regular stream. It must always be
2101 * metadata stream and this is crucial for data structure synchronization.
2102 */
2103 assert(stream->metadata_flag);
2104
e316aad5
DG
2105 DBG3("Consumer delete metadata stream %d", stream->wait_fd);
2106
74251bb8 2107 pthread_mutex_lock(&consumer_data.lock);
a6ef8ee6
JG
2108 /*
2109 * Note that this assumes that a stream's channel is never changed and
2110 * that the stream's lock doesn't need to be taken to sample its
2111 * channel.
2112 */
2113 channel = stream->chan;
2114 pthread_mutex_lock(&channel->lock);
3dad2c0f 2115 pthread_mutex_lock(&stream->lock);
a6ef8ee6 2116 if (channel->metadata_cache) {
081424af 2117 /* Only applicable to userspace consumers. */
a6ef8ee6 2118 pthread_mutex_lock(&channel->metadata_cache->lock);
081424af 2119 }
8994307f 2120
6d574024
DG
2121 /* Remove any reference to that stream. */
2122 consumer_stream_delete(stream, ht);
ca22feea 2123
6d574024
DG
2124 /* Close down everything including the relayd if one. */
2125 consumer_stream_close(stream);
2126 /* Destroy tracer buffers of the stream. */
2127 consumer_stream_destroy_buffers(stream);
fb3a43a9
DG
2128
2129 /* Atomically decrement channel refcount since other threads can use it. */
a6ef8ee6
JG
2130 if (!uatomic_sub_return(&channel->refcount, 1)
2131 && !uatomic_read(&channel->nb_init_stream_left)) {
c30aaa51 2132 /* Go for channel deletion! */
a6ef8ee6 2133 free_channel = true;
fb3a43a9 2134 }
a6ef8ee6 2135 stream->chan = NULL;
fb3a43a9 2136
73811ecc
DG
2137 /*
2138 * Nullify the stream reference so it is not used after deletion. The
6d574024
DG
2139 * channel lock MUST be acquired before being able to check for a NULL
2140 * pointer value.
73811ecc 2141 */
a6ef8ee6 2142 channel->metadata_stream = NULL;
73811ecc 2143
a6ef8ee6
JG
2144 if (channel->metadata_cache) {
2145 pthread_mutex_unlock(&channel->metadata_cache->lock);
081424af 2146 }
3dad2c0f 2147 pthread_mutex_unlock(&stream->lock);
a6ef8ee6 2148 pthread_mutex_unlock(&channel->lock);
74251bb8 2149 pthread_mutex_unlock(&consumer_data.lock);
e316aad5 2150
a6ef8ee6
JG
2151 if (free_channel) {
2152 consumer_del_channel(channel);
e316aad5
DG
2153 }
2154
d2956687
JG
2155 lttng_trace_chunk_put(stream->trace_chunk);
2156 stream->trace_chunk = NULL;
6d574024 2157 consumer_stream_free(stream);
fb3a43a9
DG
2158}
2159
2160/*
2161 * Action done with the metadata stream when adding it to the consumer internal
2162 * data structures to handle it.
2163 */
66d583dc 2164void consumer_add_metadata_stream(struct lttng_consumer_stream *stream)
fb3a43a9 2165{
5ab66908 2166 struct lttng_ht *ht = metadata_ht;
76082088 2167 struct lttng_ht_iter iter;
d88aee68 2168 struct lttng_ht_node_u64 *node;
fb3a43a9 2169
e316aad5
DG
2170 assert(stream);
2171 assert(ht);
2172
d88aee68 2173 DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key);
e316aad5
DG
2174
2175 pthread_mutex_lock(&consumer_data.lock);
a9838785 2176 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 2177 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 2178 pthread_mutex_lock(&stream->lock);
e316aad5 2179
e316aad5
DG
2180 /*
2181 * From here, refcounts are updated so be _careful_ when returning an error
2182 * after this point.
2183 */
2184
fb3a43a9 2185 rcu_read_lock();
76082088
DG
2186
2187 /*
2188 * Lookup the stream just to make sure it does not exist in our internal
2189 * state. This should NEVER happen.
2190 */
d88aee68
DG
2191 lttng_ht_lookup(ht, &stream->key, &iter);
2192 node = lttng_ht_iter_get_node_u64(&iter);
76082088
DG
2193 assert(!node);
2194
e316aad5 2195 /*
ffe60014
DG
2196 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2197 * in terms of destroying the associated channel, because the action that
e316aad5
DG
2198 * causes the count to become 0 also causes a stream to be added. The
2199 * channel deletion will thus be triggered by the following removal of this
2200 * stream.
2201 */
ffe60014 2202 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
2203 /* Increment refcount before decrementing nb_init_stream_left */
2204 cmm_smp_wmb();
ffe60014 2205 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
2206 }
2207
d88aee68 2208 lttng_ht_add_unique_u64(ht, &stream->node);
ca22feea 2209
446156b4 2210 lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht,
d8ef542d
MD
2211 &stream->node_channel_id);
2212
ca22feea
DG
2213 /*
2214 * Add stream to the stream_list_ht of the consumer data. No need to steal
2215 * the key since the HT does not use it and we allow to add redundant keys
2216 * into this table.
2217 */
d88aee68 2218 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 2219
fb3a43a9 2220 rcu_read_unlock();
e316aad5 2221
2e818a6a 2222 pthread_mutex_unlock(&stream->lock);
a9838785 2223 pthread_mutex_unlock(&stream->chan->lock);
ec6ea7d0 2224 pthread_mutex_unlock(&stream->chan->timer_lock);
e316aad5 2225 pthread_mutex_unlock(&consumer_data.lock);
fb3a43a9
DG
2226}
2227
8994307f
DG
2228/*
2229 * Delete data stream that are flagged for deletion (endpoint_status).
2230 */
2231static void validate_endpoint_status_data_stream(void)
2232{
2233 struct lttng_ht_iter iter;
2234 struct lttng_consumer_stream *stream;
2235
2236 DBG("Consumer delete flagged data stream");
2237
2238 rcu_read_lock();
2239 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
2240 /* Validate delete flag of the stream */
79d4ffb7 2241 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2242 continue;
2243 }
2244 /* Delete it right now */
2245 consumer_del_stream(stream, data_ht);
2246 }
2247 rcu_read_unlock();
2248}
2249
2250/*
2251 * Delete metadata stream that are flagged for deletion (endpoint_status).
2252 */
2253static void validate_endpoint_status_metadata_stream(
2254 struct lttng_poll_event *pollset)
2255{
2256 struct lttng_ht_iter iter;
2257 struct lttng_consumer_stream *stream;
2258
2259 DBG("Consumer delete flagged metadata stream");
2260
2261 assert(pollset);
2262
2263 rcu_read_lock();
2264 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
2265 /* Validate delete flag of the stream */
79d4ffb7 2266 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2267 continue;
2268 }
2269 /*
2270 * Remove from pollset so the metadata thread can continue without
2271 * blocking on a deleted stream.
2272 */
2273 lttng_poll_del(pollset, stream->wait_fd);
2274
2275 /* Delete it right now */
2276 consumer_del_metadata_stream(stream, metadata_ht);
2277 }
2278 rcu_read_unlock();
2279}
2280
fb3a43a9
DG
2281/*
2282 * Thread polls on metadata file descriptor and write them on disk or on the
2283 * network.
2284 */
7d980def 2285void *consumer_thread_metadata_poll(void *data)
fb3a43a9 2286{
1fc79fb4 2287 int ret, i, pollfd, err = -1;
fb3a43a9 2288 uint32_t revents, nb_fd;
e316aad5 2289 struct lttng_consumer_stream *stream = NULL;
fb3a43a9 2290 struct lttng_ht_iter iter;
d88aee68 2291 struct lttng_ht_node_u64 *node;
fb3a43a9
DG
2292 struct lttng_poll_event events;
2293 struct lttng_consumer_local_data *ctx = data;
2294 ssize_t len;
2295
2296 rcu_register_thread();
2297
1fc79fb4
MD
2298 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA);
2299
2d57de81
MD
2300 if (testpoint(consumerd_thread_metadata)) {
2301 goto error_testpoint;
2302 }
2303
9ce5646a
MD
2304 health_code_update();
2305
fb3a43a9
DG
2306 DBG("Thread metadata poll started");
2307
fb3a43a9
DG
2308 /* Size is set to 1 for the consumer_metadata pipe */
2309 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2310 if (ret < 0) {
2311 ERR("Poll set creation failed");
d8ef542d 2312 goto end_poll;
fb3a43a9
DG
2313 }
2314
13886d2d
DG
2315 ret = lttng_poll_add(&events,
2316 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN);
fb3a43a9
DG
2317 if (ret < 0) {
2318 goto end;
2319 }
2320
2321 /* Main loop */
2322 DBG("Metadata main loop started");
2323
2324 while (1) {
fb3a43a9 2325restart:
7fa2082e 2326 health_code_update();
9ce5646a 2327 health_poll_entry();
7fa2082e 2328 DBG("Metadata poll wait");
fb3a43a9 2329 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2330 DBG("Metadata poll return from wait with %d fd(s)",
2331 LTTNG_POLL_GETNB(&events));
9ce5646a 2332 health_poll_exit();
40063ead 2333 DBG("Metadata event caught in thread");
fb3a43a9
DG
2334 if (ret < 0) {
2335 if (errno == EINTR) {
40063ead 2336 ERR("Poll EINTR caught");
fb3a43a9
DG
2337 goto restart;
2338 }
d9607cd7
MD
2339 if (LTTNG_POLL_GETNB(&events) == 0) {
2340 err = 0; /* All is OK */
2341 }
2342 goto end;
fb3a43a9
DG
2343 }
2344
0d9c5d77
DG
2345 nb_fd = ret;
2346
e316aad5 2347 /* From here, the event is a metadata wait fd */
fb3a43a9 2348 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2349 health_code_update();
2350
fb3a43a9
DG
2351 revents = LTTNG_POLL_GETEV(&events, i);
2352 pollfd = LTTNG_POLL_GETFD(&events, i);
2353
13886d2d 2354 if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) {
03e43155 2355 if (revents & LPOLLIN) {
13886d2d
DG
2356 ssize_t pipe_len;
2357
2358 pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe,
2359 &stream, sizeof(stream));
6cd525e8 2360 if (pipe_len < sizeof(stream)) {
03e43155
MD
2361 if (pipe_len < 0) {
2362 PERROR("read metadata stream");
2363 }
fb3a43a9 2364 /*
03e43155
MD
2365 * Remove the pipe from the poll set and continue the loop
2366 * since their might be data to consume.
fb3a43a9 2367 */
03e43155
MD
2368 lttng_poll_del(&events,
2369 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2370 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
fb3a43a9
DG
2371 continue;
2372 }
2373
8994307f
DG
2374 /* A NULL stream means that the state has changed. */
2375 if (stream == NULL) {
2376 /* Check for deleted streams. */
2377 validate_endpoint_status_metadata_stream(&events);
3714380f 2378 goto restart;
8994307f
DG
2379 }
2380
fb3a43a9
DG
2381 DBG("Adding metadata stream %d to poll set",
2382 stream->wait_fd);
2383
fb3a43a9
DG
2384 /* Add metadata stream to the global poll events list */
2385 lttng_poll_add(&events, stream->wait_fd,
6d574024 2386 LPOLLIN | LPOLLPRI | LPOLLHUP);
03e43155
MD
2387 } else if (revents & (LPOLLERR | LPOLLHUP)) {
2388 DBG("Metadata thread pipe hung up");
2389 /*
2390 * Remove the pipe from the poll set and continue the loop
2391 * since their might be data to consume.
2392 */
2393 lttng_poll_del(&events,
2394 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2395 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
2396 continue;
2397 } else {
2398 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2399 goto end;
fb3a43a9
DG
2400 }
2401
e316aad5 2402 /* Handle other stream */
fb3a43a9
DG
2403 continue;
2404 }
2405
d09e1200 2406 rcu_read_lock();
d88aee68
DG
2407 {
2408 uint64_t tmp_id = (uint64_t) pollfd;
2409
2410 lttng_ht_lookup(metadata_ht, &tmp_id, &iter);
2411 }
2412 node = lttng_ht_iter_get_node_u64(&iter);
e316aad5 2413 assert(node);
fb3a43a9
DG
2414
2415 stream = caa_container_of(node, struct lttng_consumer_stream,
58b1f425 2416 node);
fb3a43a9 2417
03e43155
MD
2418 if (revents & (LPOLLIN | LPOLLPRI)) {
2419 /* Get the data out of the metadata file descriptor */
2420 DBG("Metadata available on fd %d", pollfd);
2421 assert(stream->wait_fd == pollfd);
2422
2423 do {
2424 health_code_update();
2425
6f9449c2 2426 len = ctx->on_buffer_ready(stream, ctx, false);
03e43155
MD
2427 /*
2428 * We don't check the return value here since if we get
83f4233d 2429 * a negative len, it means an error occurred thus we
03e43155
MD
2430 * simply remove it from the poll set and free the
2431 * stream.
2432 */
2433 } while (len > 0);
2434
2435 /* It's ok to have an unavailable sub-buffer */
2436 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
2437 /* Clean up stream from consumer and free it. */
2438 lttng_poll_del(&events, stream->wait_fd);
2439 consumer_del_metadata_stream(stream, metadata_ht);
2440 }
2441 } else if (revents & (LPOLLERR | LPOLLHUP)) {
e316aad5 2442 DBG("Metadata fd %d is hup|err.", pollfd);
fb3a43a9
DG
2443 if (!stream->hangup_flush_done
2444 && (consumer_data.type == LTTNG_CONSUMER32_UST
2445 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2446 DBG("Attempting to flush and consume the UST buffers");
2447 lttng_ustconsumer_on_stream_hangup(stream);
2448
2449 /* We just flushed the stream now read it. */
4bb94b75 2450 do {
9ce5646a
MD
2451 health_code_update();
2452
6f9449c2 2453 len = ctx->on_buffer_ready(stream, ctx, false);
4bb94b75
DG
2454 /*
2455 * We don't check the return value here since if we get
83f4233d 2456 * a negative len, it means an error occurred thus we
4bb94b75
DG
2457 * simply remove it from the poll set and free the
2458 * stream.
2459 */
2460 } while (len > 0);
fb3a43a9
DG
2461 }
2462
fb3a43a9 2463 lttng_poll_del(&events, stream->wait_fd);
e316aad5
DG
2464 /*
2465 * This call update the channel states, closes file descriptors
2466 * and securely free the stream.
2467 */
2468 consumer_del_metadata_stream(stream, metadata_ht);
03e43155
MD
2469 } else {
2470 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
6f2f1a70 2471 rcu_read_unlock();
03e43155 2472 goto end;
fb3a43a9 2473 }
e316aad5 2474 /* Release RCU lock for the stream looked up */
d09e1200 2475 rcu_read_unlock();
fb3a43a9
DG
2476 }
2477 }
2478
1fc79fb4
MD
2479 /* All is OK */
2480 err = 0;
fb3a43a9
DG
2481end:
2482 DBG("Metadata poll thread exiting");
fb3a43a9 2483
d8ef542d
MD
2484 lttng_poll_clean(&events);
2485end_poll:
2d57de81 2486error_testpoint:
1fc79fb4
MD
2487 if (err) {
2488 health_error();
2489 ERR("Health error occurred in %s", __func__);
2490 }
2491 health_unregister(health_consumerd);
fb3a43a9
DG
2492 rcu_unregister_thread();
2493 return NULL;
2494}
2495
3bd1e081 2496/*
e4421fec 2497 * This thread polls the fds in the set to consume the data and write
3bd1e081
MD
2498 * it to tracefile if necessary.
2499 */
7d980def 2500void *consumer_thread_data_poll(void *data)
3bd1e081 2501{
1fc79fb4 2502 int num_rdy, num_hup, high_prio, ret, i, err = -1;
3bd1e081
MD
2503 struct pollfd *pollfd = NULL;
2504 /* local view of the streams */
c869f647 2505 struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL;
3bd1e081 2506 /* local view of consumer_data.fds_count */
8bdcc002
JG
2507 int nb_fd = 0;
2508 /* 2 for the consumer_data_pipe and wake up pipe */
2509 const int nb_pipes_fd = 2;
9a2fcf78
JD
2510 /* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */
2511 int nb_inactive_fd = 0;
3bd1e081 2512 struct lttng_consumer_local_data *ctx = data;
00e2e675 2513 ssize_t len;
3bd1e081 2514
e7b994a3
DG
2515 rcu_register_thread();
2516
1fc79fb4
MD
2517 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA);
2518
2d57de81
MD
2519 if (testpoint(consumerd_thread_data)) {
2520 goto error_testpoint;
2521 }
2522
9ce5646a
MD
2523 health_code_update();
2524
4df6c8cb
MD
2525 local_stream = zmalloc(sizeof(struct lttng_consumer_stream *));
2526 if (local_stream == NULL) {
2527 PERROR("local_stream malloc");
2528 goto end;
2529 }
3bd1e081
MD
2530
2531 while (1) {
9ce5646a
MD
2532 health_code_update();
2533
3bd1e081
MD
2534 high_prio = 0;
2535 num_hup = 0;
2536
2537 /*
e4421fec 2538 * the fds set has been updated, we need to update our
3bd1e081
MD
2539 * local array as well
2540 */
2541 pthread_mutex_lock(&consumer_data.lock);
2542 if (consumer_data.need_update) {
0e428499
DG
2543 free(pollfd);
2544 pollfd = NULL;
2545
2546 free(local_stream);
2547 local_stream = NULL;
3bd1e081 2548
8bdcc002 2549 /* Allocate for all fds */
261de637 2550 pollfd = zmalloc((consumer_data.stream_count + nb_pipes_fd) * sizeof(struct pollfd));
3bd1e081 2551 if (pollfd == NULL) {
7a57cf92 2552 PERROR("pollfd malloc");
3bd1e081
MD
2553 pthread_mutex_unlock(&consumer_data.lock);
2554 goto end;
2555 }
2556
261de637 2557 local_stream = zmalloc((consumer_data.stream_count + nb_pipes_fd) *
747f8642 2558 sizeof(struct lttng_consumer_stream *));
3bd1e081 2559 if (local_stream == NULL) {
7a57cf92 2560 PERROR("local_stream malloc");
3bd1e081
MD
2561 pthread_mutex_unlock(&consumer_data.lock);
2562 goto end;
2563 }
ffe60014 2564 ret = update_poll_array(ctx, &pollfd, local_stream,
9a2fcf78 2565 data_ht, &nb_inactive_fd);
3bd1e081
MD
2566 if (ret < 0) {
2567 ERR("Error in allocating pollfd or local_outfds");
f73fabfd 2568 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2569 pthread_mutex_unlock(&consumer_data.lock);
2570 goto end;
2571 }
2572 nb_fd = ret;
2573 consumer_data.need_update = 0;
2574 }
2575 pthread_mutex_unlock(&consumer_data.lock);
2576
4078b776 2577 /* No FDs and consumer_quit, consumer_cleanup the thread */
9a2fcf78
JD
2578 if (nb_fd == 0 && nb_inactive_fd == 0 &&
2579 CMM_LOAD_SHARED(consumer_quit) == 1) {
1fc79fb4 2580 err = 0; /* All is OK */
4078b776
MD
2581 goto end;
2582 }
3bd1e081 2583 /* poll on the array of fds */
88f2b785 2584 restart:
261de637 2585 DBG("polling on %d fd", nb_fd + nb_pipes_fd);
cf0bcb51
JG
2586 if (testpoint(consumerd_thread_data_poll)) {
2587 goto end;
2588 }
9ce5646a 2589 health_poll_entry();
261de637 2590 num_rdy = poll(pollfd, nb_fd + nb_pipes_fd, -1);
9ce5646a 2591 health_poll_exit();
3bd1e081
MD
2592 DBG("poll num_rdy : %d", num_rdy);
2593 if (num_rdy == -1) {
88f2b785
MD
2594 /*
2595 * Restart interrupted system call.
2596 */
2597 if (errno == EINTR) {
2598 goto restart;
2599 }
7a57cf92 2600 PERROR("Poll error");
f73fabfd 2601 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2602 goto end;
2603 } else if (num_rdy == 0) {
2604 DBG("Polling thread timed out");
2605 goto end;
2606 }
2607
80957876
JG
2608 if (caa_unlikely(data_consumption_paused)) {
2609 DBG("Data consumption paused, sleeping...");
2610 sleep(1);
2611 goto restart;
2612 }
2613
3bd1e081 2614 /*
50f8ae69 2615 * If the consumer_data_pipe triggered poll go directly to the
00e2e675
DG
2616 * beginning of the loop to update the array. We want to prioritize
2617 * array update over low-priority reads.
3bd1e081 2618 */
509bb1cf 2619 if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) {
ab30f567 2620 ssize_t pipe_readlen;
04fdd819 2621
50f8ae69 2622 DBG("consumer_data_pipe wake up");
acdb9057
DG
2623 pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe,
2624 &new_stream, sizeof(new_stream));
6cd525e8
MD
2625 if (pipe_readlen < sizeof(new_stream)) {
2626 PERROR("Consumer data pipe");
23f5f35d
DG
2627 /* Continue so we can at least handle the current stream(s). */
2628 continue;
2629 }
c869f647
DG
2630
2631 /*
2632 * If the stream is NULL, just ignore it. It's also possible that
2633 * the sessiond poll thread changed the consumer_quit state and is
2634 * waking us up to test it.
2635 */
2636 if (new_stream == NULL) {
8994307f 2637 validate_endpoint_status_data_stream();
c869f647
DG
2638 continue;
2639 }
2640
c869f647 2641 /* Continue to update the local streams and handle prio ones */
3bd1e081
MD
2642 continue;
2643 }
2644
02b3d176
DG
2645 /* Handle wakeup pipe. */
2646 if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) {
2647 char dummy;
2648 ssize_t pipe_readlen;
2649
2650 pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy,
2651 sizeof(dummy));
2652 if (pipe_readlen < 0) {
2653 PERROR("Consumer data wakeup pipe");
2654 }
2655 /* We've been awakened to handle stream(s). */
2656 ctx->has_wakeup = 0;
2657 }
2658
3bd1e081
MD
2659 /* Take care of high priority channels first. */
2660 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2661 health_code_update();
2662
9617607b
DG
2663 if (local_stream[i] == NULL) {
2664 continue;
2665 }
fb3a43a9 2666 if (pollfd[i].revents & POLLPRI) {
d41f73b7
MD
2667 DBG("Urgent read on fd %d", pollfd[i].fd);
2668 high_prio = 1;
6f9449c2 2669 len = ctx->on_buffer_ready(local_stream[i], ctx, false);
d41f73b7 2670 /* it's ok to have an unavailable sub-buffer */
b64403e3 2671 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2672 /* Clean the stream and free it. */
2673 consumer_del_stream(local_stream[i], data_ht);
9617607b 2674 local_stream[i] = NULL;
4078b776
MD
2675 } else if (len > 0) {
2676 local_stream[i]->data_read = 1;
d41f73b7 2677 }
3bd1e081
MD
2678 }
2679 }
2680
4078b776
MD
2681 /*
2682 * If we read high prio channel in this loop, try again
2683 * for more high prio data.
2684 */
2685 if (high_prio) {
3bd1e081
MD
2686 continue;
2687 }
2688
2689 /* Take care of low priority channels. */
4078b776 2690 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2691 health_code_update();
2692
9617607b
DG
2693 if (local_stream[i] == NULL) {
2694 continue;
2695 }
4078b776 2696 if ((pollfd[i].revents & POLLIN) ||
02b3d176
DG
2697 local_stream[i]->hangup_flush_done ||
2698 local_stream[i]->has_data) {
4078b776 2699 DBG("Normal read on fd %d", pollfd[i].fd);
6f9449c2 2700 len = ctx->on_buffer_ready(local_stream[i], ctx, false);
4078b776 2701 /* it's ok to have an unavailable sub-buffer */
b64403e3 2702 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2703 /* Clean the stream and free it. */
2704 consumer_del_stream(local_stream[i], data_ht);
9617607b 2705 local_stream[i] = NULL;
4078b776
MD
2706 } else if (len > 0) {
2707 local_stream[i]->data_read = 1;
2708 }
2709 }
2710 }
2711
2712 /* Handle hangup and errors */
2713 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2714 health_code_update();
2715
9617607b
DG
2716 if (local_stream[i] == NULL) {
2717 continue;
2718 }
4078b776
MD
2719 if (!local_stream[i]->hangup_flush_done
2720 && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL))
2721 && (consumer_data.type == LTTNG_CONSUMER32_UST
2722 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2723 DBG("fd %d is hup|err|nval. Attempting flush and read.",
9617607b 2724 pollfd[i].fd);
4078b776
MD
2725 lttng_ustconsumer_on_stream_hangup(local_stream[i]);
2726 /* Attempt read again, for the data we just flushed. */
2727 local_stream[i]->data_read = 1;
2728 }
2729 /*
2730 * If the poll flag is HUP/ERR/NVAL and we have
2731 * read no data in this pass, we can remove the
2732 * stream from its hash table.
2733 */
2734 if ((pollfd[i].revents & POLLHUP)) {
2735 DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
2736 if (!local_stream[i]->data_read) {
43c34bc3 2737 consumer_del_stream(local_stream[i], data_ht);
9617607b 2738 local_stream[i] = NULL;
4078b776
MD
2739 num_hup++;
2740 }
2741 } else if (pollfd[i].revents & POLLERR) {
2742 ERR("Error returned in polling fd %d.", pollfd[i].fd);
2743 if (!local_stream[i]->data_read) {
43c34bc3 2744 consumer_del_stream(local_stream[i], data_ht);
9617607b 2745 local_stream[i] = NULL;
4078b776
MD
2746 num_hup++;
2747 }
2748 } else if (pollfd[i].revents & POLLNVAL) {
2749 ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
2750 if (!local_stream[i]->data_read) {
43c34bc3 2751 consumer_del_stream(local_stream[i], data_ht);
9617607b 2752 local_stream[i] = NULL;
4078b776 2753 num_hup++;
3bd1e081
MD
2754 }
2755 }
9617607b
DG
2756 if (local_stream[i] != NULL) {
2757 local_stream[i]->data_read = 0;
2758 }
3bd1e081
MD
2759 }
2760 }
1fc79fb4
MD
2761 /* All is OK */
2762 err = 0;
3bd1e081
MD
2763end:
2764 DBG("polling thread exiting");
0e428499
DG
2765 free(pollfd);
2766 free(local_stream);
fb3a43a9
DG
2767
2768 /*
2769 * Close the write side of the pipe so epoll_wait() in
7d980def
DG
2770 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2771 * read side of the pipe. If we close them both, epoll_wait strangely does
2772 * not return and could create a endless wait period if the pipe is the
2773 * only tracked fd in the poll set. The thread will take care of closing
2774 * the read side.
fb3a43a9 2775 */
13886d2d 2776 (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe);
fb3a43a9 2777
2d57de81 2778error_testpoint:
1fc79fb4
MD
2779 if (err) {
2780 health_error();
2781 ERR("Health error occurred in %s", __func__);
2782 }
2783 health_unregister(health_consumerd);
2784
e7b994a3 2785 rcu_unregister_thread();
3bd1e081
MD
2786 return NULL;
2787}
2788
d8ef542d
MD
2789/*
2790 * Close wake-up end of each stream belonging to the channel. This will
2791 * allow the poll() on the stream read-side to detect when the
2792 * write-side (application) finally closes them.
2793 */
2794static
2795void consumer_close_channel_streams(struct lttng_consumer_channel *channel)
2796{
2797 struct lttng_ht *ht;
2798 struct lttng_consumer_stream *stream;
2799 struct lttng_ht_iter iter;
2800
2801 ht = consumer_data.stream_per_chan_id_ht;
2802
2803 rcu_read_lock();
2804 cds_lfht_for_each_entry_duplicate(ht->ht,
2805 ht->hash_fct(&channel->key, lttng_ht_seed),
2806 ht->match_fct, &channel->key,
2807 &iter.iter, stream, node_channel_id.node) {
f2ad556d
MD
2808 /*
2809 * Protect against teardown with mutex.
2810 */
2811 pthread_mutex_lock(&stream->lock);
2812 if (cds_lfht_is_node_deleted(&stream->node.node)) {
2813 goto next;
2814 }
d8ef542d
MD
2815 switch (consumer_data.type) {
2816 case LTTNG_CONSUMER_KERNEL:
2817 break;
2818 case LTTNG_CONSUMER32_UST:
2819 case LTTNG_CONSUMER64_UST:
b4a650f3
DG
2820 if (stream->metadata_flag) {
2821 /* Safe and protected by the stream lock. */
2822 lttng_ustconsumer_close_metadata(stream->chan);
2823 } else {
2824 /*
2825 * Note: a mutex is taken internally within
2826 * liblttng-ust-ctl to protect timer wakeup_fd
2827 * use from concurrent close.
2828 */
2829 lttng_ustconsumer_close_stream_wakeup(stream);
2830 }
d8ef542d
MD
2831 break;
2832 default:
2833 ERR("Unknown consumer_data type");
2834 assert(0);
2835 }
f2ad556d
MD
2836 next:
2837 pthread_mutex_unlock(&stream->lock);
d8ef542d
MD
2838 }
2839 rcu_read_unlock();
2840}
2841
2842static void destroy_channel_ht(struct lttng_ht *ht)
2843{
2844 struct lttng_ht_iter iter;
2845 struct lttng_consumer_channel *channel;
2846 int ret;
2847
2848 if (ht == NULL) {
2849 return;
2850 }
2851
2852 rcu_read_lock();
2853 cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) {
2854 ret = lttng_ht_del(ht, &iter);
2855 assert(ret != 0);
2856 }
2857 rcu_read_unlock();
2858
2859 lttng_ht_destroy(ht);
2860}
2861
2862/*
2863 * This thread polls the channel fds to detect when they are being
2864 * closed. It closes all related streams if the channel is detected as
2865 * closed. It is currently only used as a shim layer for UST because the
2866 * consumerd needs to keep the per-stream wakeup end of pipes open for
2867 * periodical flush.
2868 */
2869void *consumer_thread_channel_poll(void *data)
2870{
1fc79fb4 2871 int ret, i, pollfd, err = -1;
d8ef542d
MD
2872 uint32_t revents, nb_fd;
2873 struct lttng_consumer_channel *chan = NULL;
2874 struct lttng_ht_iter iter;
2875 struct lttng_ht_node_u64 *node;
2876 struct lttng_poll_event events;
2877 struct lttng_consumer_local_data *ctx = data;
2878 struct lttng_ht *channel_ht;
2879
2880 rcu_register_thread();
2881
1fc79fb4
MD
2882 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL);
2883
2d57de81
MD
2884 if (testpoint(consumerd_thread_channel)) {
2885 goto error_testpoint;
2886 }
2887
9ce5646a
MD
2888 health_code_update();
2889
d8ef542d
MD
2890 channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2891 if (!channel_ht) {
2892 /* ENOMEM at this point. Better to bail out. */
2893 goto end_ht;
2894 }
2895
2896 DBG("Thread channel poll started");
2897
2898 /* Size is set to 1 for the consumer_channel pipe */
2899 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2900 if (ret < 0) {
2901 ERR("Poll set creation failed");
2902 goto end_poll;
2903 }
2904
2905 ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN);
2906 if (ret < 0) {
2907 goto end;
2908 }
2909
2910 /* Main loop */
2911 DBG("Channel main loop started");
2912
2913 while (1) {
d8ef542d 2914restart:
7fa2082e
MD
2915 health_code_update();
2916 DBG("Channel poll wait");
9ce5646a 2917 health_poll_entry();
d8ef542d 2918 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2919 DBG("Channel poll return from wait with %d fd(s)",
2920 LTTNG_POLL_GETNB(&events));
9ce5646a 2921 health_poll_exit();
40063ead 2922 DBG("Channel event caught in thread");
d8ef542d
MD
2923 if (ret < 0) {
2924 if (errno == EINTR) {
40063ead 2925 ERR("Poll EINTR caught");
d8ef542d
MD
2926 goto restart;
2927 }
d9607cd7
MD
2928 if (LTTNG_POLL_GETNB(&events) == 0) {
2929 err = 0; /* All is OK */
2930 }
d8ef542d
MD
2931 goto end;
2932 }
2933
2934 nb_fd = ret;
2935
2936 /* From here, the event is a channel wait fd */
2937 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2938 health_code_update();
2939
d8ef542d
MD
2940 revents = LTTNG_POLL_GETEV(&events, i);
2941 pollfd = LTTNG_POLL_GETFD(&events, i);
2942
d8ef542d 2943 if (pollfd == ctx->consumer_channel_pipe[0]) {
03e43155 2944 if (revents & LPOLLIN) {
d8ef542d 2945 enum consumer_channel_action action;
a0cbdd2e 2946 uint64_t key;
d8ef542d 2947
a0cbdd2e 2948 ret = read_channel_pipe(ctx, &chan, &key, &action);
d8ef542d 2949 if (ret <= 0) {
03e43155
MD
2950 if (ret < 0) {
2951 ERR("Error reading channel pipe");
2952 }
2953 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
d8ef542d
MD
2954 continue;
2955 }
2956
2957 switch (action) {
2958 case CONSUMER_CHANNEL_ADD:
2959 DBG("Adding channel %d to poll set",
2960 chan->wait_fd);
2961
2962 lttng_ht_node_init_u64(&chan->wait_fd_node,
2963 chan->wait_fd);
c7260a81 2964 rcu_read_lock();
d8ef542d
MD
2965 lttng_ht_add_unique_u64(channel_ht,
2966 &chan->wait_fd_node);
c7260a81 2967 rcu_read_unlock();
d8ef542d
MD
2968 /* Add channel to the global poll events list */
2969 lttng_poll_add(&events, chan->wait_fd,
03e43155 2970 LPOLLERR | LPOLLHUP);
d8ef542d 2971 break;
a0cbdd2e
MD
2972 case CONSUMER_CHANNEL_DEL:
2973 {
b4a650f3
DG
2974 /*
2975 * This command should never be called if the channel
2976 * has streams monitored by either the data or metadata
2977 * thread. The consumer only notify this thread with a
2978 * channel del. command if it receives a destroy
2979 * channel command from the session daemon that send it
2980 * if a command prior to the GET_CHANNEL failed.
2981 */
2982
c7260a81 2983 rcu_read_lock();
a0cbdd2e
MD
2984 chan = consumer_find_channel(key);
2985 if (!chan) {
c7260a81 2986 rcu_read_unlock();
a0cbdd2e
MD
2987 ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key);
2988 break;
2989 }
2990 lttng_poll_del(&events, chan->wait_fd);
f623cc0b 2991 iter.iter.node = &chan->wait_fd_node.node;
a0cbdd2e
MD
2992 ret = lttng_ht_del(channel_ht, &iter);
2993 assert(ret == 0);
a0cbdd2e 2994
f2a444f1
DG
2995 switch (consumer_data.type) {
2996 case LTTNG_CONSUMER_KERNEL:
2997 break;
2998 case LTTNG_CONSUMER32_UST:
2999 case LTTNG_CONSUMER64_UST:
212d67a2
DG
3000 health_code_update();
3001 /* Destroy streams that might have been left in the stream list. */
3002 clean_channel_stream_list(chan);
f2a444f1
DG
3003 break;
3004 default:
3005 ERR("Unknown consumer_data type");
3006 assert(0);
3007 }
3008
a0cbdd2e
MD
3009 /*
3010 * Release our own refcount. Force channel deletion even if
3011 * streams were not initialized.
3012 */
3013 if (!uatomic_sub_return(&chan->refcount, 1)) {
3014 consumer_del_channel(chan);
3015 }
c7260a81 3016 rcu_read_unlock();
a0cbdd2e
MD
3017 goto restart;
3018 }
d8ef542d
MD
3019 case CONSUMER_CHANNEL_QUIT:
3020 /*
3021 * Remove the pipe from the poll set and continue the loop
3022 * since their might be data to consume.
3023 */
3024 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
3025 continue;
3026 default:
3027 ERR("Unknown action");
3028 break;
3029 }
03e43155
MD
3030 } else if (revents & (LPOLLERR | LPOLLHUP)) {
3031 DBG("Channel thread pipe hung up");
3032 /*
3033 * Remove the pipe from the poll set and continue the loop
3034 * since their might be data to consume.
3035 */
3036 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
3037 continue;
3038 } else {
3039 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
3040 goto end;
d8ef542d
MD
3041 }
3042
3043 /* Handle other stream */
3044 continue;
3045 }
3046
3047 rcu_read_lock();
3048 {
3049 uint64_t tmp_id = (uint64_t) pollfd;
3050
3051 lttng_ht_lookup(channel_ht, &tmp_id, &iter);
3052 }
3053 node = lttng_ht_iter_get_node_u64(&iter);
3054 assert(node);
3055
3056 chan = caa_container_of(node, struct lttng_consumer_channel,
3057 wait_fd_node);
3058
3059 /* Check for error event */
3060 if (revents & (LPOLLERR | LPOLLHUP)) {
3061 DBG("Channel fd %d is hup|err.", pollfd);
3062
3063 lttng_poll_del(&events, chan->wait_fd);
3064 ret = lttng_ht_del(channel_ht, &iter);
3065 assert(ret == 0);
b4a650f3
DG
3066
3067 /*
3068 * This will close the wait fd for each stream associated to
3069 * this channel AND monitored by the data/metadata thread thus
3070 * will be clean by the right thread.
3071 */
d8ef542d 3072 consumer_close_channel_streams(chan);
f2ad556d
MD
3073
3074 /* Release our own refcount */
3075 if (!uatomic_sub_return(&chan->refcount, 1)
3076 && !uatomic_read(&chan->nb_init_stream_left)) {
3077 consumer_del_channel(chan);
3078 }
03e43155
MD
3079 } else {
3080 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
3081 rcu_read_unlock();
3082 goto end;
d8ef542d
MD
3083 }
3084
3085 /* Release RCU lock for the channel looked up */
3086 rcu_read_unlock();
3087 }
3088 }
3089
1fc79fb4
MD
3090 /* All is OK */
3091 err = 0;
d8ef542d
MD
3092end:
3093 lttng_poll_clean(&events);
3094end_poll:
3095 destroy_channel_ht(channel_ht);
3096end_ht:
2d57de81 3097error_testpoint:
d8ef542d 3098 DBG("Channel poll thread exiting");
1fc79fb4
MD
3099 if (err) {
3100 health_error();
3101 ERR("Health error occurred in %s", __func__);
3102 }
3103 health_unregister(health_consumerd);
d8ef542d
MD
3104 rcu_unregister_thread();
3105 return NULL;
3106}
3107
331744e3
JD
3108static int set_metadata_socket(struct lttng_consumer_local_data *ctx,
3109 struct pollfd *sockpoll, int client_socket)
3110{
3111 int ret;
3112
3113 assert(ctx);
3114 assert(sockpoll);
3115
84382d49
MD
3116 ret = lttng_consumer_poll_socket(sockpoll);
3117 if (ret) {
331744e3
JD
3118 goto error;
3119 }
3120 DBG("Metadata connection on client_socket");
3121
3122 /* Blocking call, waiting for transmission */
3123 ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket);
3124 if (ctx->consumer_metadata_socket < 0) {
3125 WARN("On accept metadata");
3126 ret = -1;
3127 goto error;
3128 }
3129 ret = 0;
3130
3131error:
3132 return ret;
3133}
3134
3bd1e081
MD
3135/*
3136 * This thread listens on the consumerd socket and receives the file
3137 * descriptors from the session daemon.
3138 */
7d980def 3139void *consumer_thread_sessiond_poll(void *data)
3bd1e081 3140{
1fc79fb4 3141 int sock = -1, client_socket, ret, err = -1;
3bd1e081
MD
3142 /*
3143 * structure to poll for incoming data on communication socket avoids
3144 * making blocking sockets.
3145 */
3146 struct pollfd consumer_sockpoll[2];
3147 struct lttng_consumer_local_data *ctx = data;
3148
e7b994a3
DG
3149 rcu_register_thread();
3150
1fc79fb4
MD
3151 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND);
3152
2d57de81
MD
3153 if (testpoint(consumerd_thread_sessiond)) {
3154 goto error_testpoint;
3155 }
3156
9ce5646a
MD
3157 health_code_update();
3158
3bd1e081
MD
3159 DBG("Creating command socket %s", ctx->consumer_command_sock_path);
3160 unlink(ctx->consumer_command_sock_path);
3161 client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path);
3162 if (client_socket < 0) {
3163 ERR("Cannot create command socket");
3164 goto end;
3165 }
3166
3167 ret = lttcomm_listen_unix_sock(client_socket);
3168 if (ret < 0) {
3169 goto end;
3170 }
3171
32258573 3172 DBG("Sending ready command to lttng-sessiond");
f73fabfd 3173 ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY);
3bd1e081
MD
3174 /* return < 0 on error, but == 0 is not fatal */
3175 if (ret < 0) {
32258573 3176 ERR("Error sending ready command to lttng-sessiond");
3bd1e081
MD
3177 goto end;
3178 }
3179
3bd1e081
MD
3180 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3181 consumer_sockpoll[0].fd = ctx->consumer_should_quit[0];
3182 consumer_sockpoll[0].events = POLLIN | POLLPRI;
3183 consumer_sockpoll[1].fd = client_socket;
3184 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3185
84382d49
MD
3186 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3187 if (ret) {
3188 if (ret > 0) {
3189 /* should exit */
3190 err = 0;
3191 }
3bd1e081
MD
3192 goto end;
3193 }
3194 DBG("Connection on client_socket");
3195
3196 /* Blocking call, waiting for transmission */
3197 sock = lttcomm_accept_unix_sock(client_socket);
534d2592 3198 if (sock < 0) {
3bd1e081
MD
3199 WARN("On accept");
3200 goto end;
3201 }
3bd1e081 3202
331744e3
JD
3203 /*
3204 * Setup metadata socket which is the second socket connection on the
3205 * command unix socket.
3206 */
3207 ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket);
84382d49
MD
3208 if (ret) {
3209 if (ret > 0) {
3210 /* should exit */
3211 err = 0;
3212 }
331744e3
JD
3213 goto end;
3214 }
3215
d96f09c6
DG
3216 /* This socket is not useful anymore. */
3217 ret = close(client_socket);
3218 if (ret < 0) {
3219 PERROR("close client_socket");
3220 }
3221 client_socket = -1;
3222
3bd1e081
MD
3223 /* update the polling structure to poll on the established socket */
3224 consumer_sockpoll[1].fd = sock;
3225 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3226
3227 while (1) {
9ce5646a
MD
3228 health_code_update();
3229
3230 health_poll_entry();
3231 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3232 health_poll_exit();
84382d49
MD
3233 if (ret) {
3234 if (ret > 0) {
3235 /* should exit */
3236 err = 0;
3237 }
3bd1e081
MD
3238 goto end;
3239 }
3240 DBG("Incoming command on sock");
3241 ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll);
4cbc1a04
DG
3242 if (ret <= 0) {
3243 /*
3244 * This could simply be a session daemon quitting. Don't output
3245 * ERR() here.
3246 */
3247 DBG("Communication interrupted on command socket");
41ba6035