consumerd: cleanup: use buffer view interface for mmap read subbuf
[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
6c1c0768 10#define _LGPL_SOURCE
3bd1e081 11#include <assert.h>
3bd1e081
MD
12#include <poll.h>
13#include <pthread.h>
14#include <stdlib.h>
15#include <string.h>
16#include <sys/mman.h>
17#include <sys/socket.h>
18#include <sys/types.h>
19#include <unistd.h>
77c7c900 20#include <inttypes.h>
331744e3 21#include <signal.h>
3bd1e081 22
51a9e1c7 23#include <bin/lttng-consumerd/health-consumerd.h>
990570ed 24#include <common/common.h>
fb3a43a9 25#include <common/utils.h>
d2956687 26#include <common/time.h>
fb3a43a9 27#include <common/compat/poll.h>
f263b7fd 28#include <common/compat/endian.h>
309167d2 29#include <common/index/index.h>
10a8a223 30#include <common/kernel-ctl/kernel-ctl.h>
00e2e675 31#include <common/sessiond-comm/relayd.h>
10a8a223
DG
32#include <common/sessiond-comm/sessiond-comm.h>
33#include <common/kernel-consumer/kernel-consumer.h>
00e2e675 34#include <common/relayd/relayd.h>
10a8a223 35#include <common/ust-consumer/ust-consumer.h>
c8fea79c
JR
36#include <common/consumer/consumer-timer.h>
37#include <common/consumer/consumer.h>
38#include <common/consumer/consumer-stream.h>
39#include <common/consumer/consumer-testpoint.h>
40#include <common/align.h>
5feafd41 41#include <common/consumer/consumer-metadata-cache.h>
d2956687
JG
42#include <common/trace-chunk.h>
43#include <common/trace-chunk-registry.h>
44#include <common/string-utils/format.h>
c35f9726 45#include <common/dynamic-array.h>
3bd1e081
MD
46
47struct lttng_consumer_global_data consumer_data = {
3bd1e081
MD
48 .stream_count = 0,
49 .need_update = 1,
50 .type = LTTNG_CONSUMER_UNKNOWN,
51};
52
d8ef542d
MD
53enum consumer_channel_action {
54 CONSUMER_CHANNEL_ADD,
a0cbdd2e 55 CONSUMER_CHANNEL_DEL,
d8ef542d
MD
56 CONSUMER_CHANNEL_QUIT,
57};
58
59struct consumer_channel_msg {
60 enum consumer_channel_action action;
a0cbdd2e
MD
61 struct lttng_consumer_channel *chan; /* add */
62 uint64_t key; /* del */
d8ef542d
MD
63};
64
80957876 65/* Flag used to temporarily pause data consumption from testpoints. */
cf0bcb51
JG
66int data_consumption_paused;
67
3bd1e081
MD
68/*
69 * Flag to inform the polling thread to quit when all fd hung up. Updated by
70 * the consumer_thread_receive_fds when it notices that all fds has hung up.
71 * Also updated by the signal handler (consumer_should_exit()). Read by the
72 * polling threads.
73 */
10211f5c 74int consumer_quit;
3bd1e081 75
43c34bc3 76/*
43c34bc3
DG
77 * Global hash table containing respectively metadata and data streams. The
78 * stream element in this ht should only be updated by the metadata poll thread
79 * for the metadata and the data poll thread for the data.
80 */
40dc48e0
DG
81static struct lttng_ht *metadata_ht;
82static struct lttng_ht *data_ht;
43c34bc3 83
5da88b0f
MD
84static const char *get_consumer_domain(void)
85{
86 switch (consumer_data.type) {
87 case LTTNG_CONSUMER_KERNEL:
88 return DEFAULT_KERNEL_TRACE_DIR;
89 case LTTNG_CONSUMER64_UST:
90 /* Fall-through. */
91 case LTTNG_CONSUMER32_UST:
92 return DEFAULT_UST_TRACE_DIR;
93 default:
94 abort();
95 }
96}
97
acdb9057
DG
98/*
99 * Notify a thread lttng pipe to poll back again. This usually means that some
100 * global state has changed so we just send back the thread in a poll wait
101 * call.
102 */
103static void notify_thread_lttng_pipe(struct lttng_pipe *pipe)
104{
105 struct lttng_consumer_stream *null_stream = NULL;
106
107 assert(pipe);
108
109 (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream));
110}
111
5c635c72
MD
112static void notify_health_quit_pipe(int *pipe)
113{
6cd525e8 114 ssize_t ret;
5c635c72 115
6cd525e8
MD
116 ret = lttng_write(pipe[1], "4", 1);
117 if (ret < 1) {
5c635c72
MD
118 PERROR("write consumer health quit");
119 }
120}
121
d8ef542d
MD
122static void notify_channel_pipe(struct lttng_consumer_local_data *ctx,
123 struct lttng_consumer_channel *chan,
a0cbdd2e 124 uint64_t key,
d8ef542d
MD
125 enum consumer_channel_action action)
126{
127 struct consumer_channel_msg msg;
6cd525e8 128 ssize_t ret;
d8ef542d 129
e56251fc
DG
130 memset(&msg, 0, sizeof(msg));
131
d8ef542d
MD
132 msg.action = action;
133 msg.chan = chan;
f21dae48 134 msg.key = key;
6cd525e8
MD
135 ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg));
136 if (ret < sizeof(msg)) {
137 PERROR("notify_channel_pipe write error");
138 }
d8ef542d
MD
139}
140
a0cbdd2e
MD
141void notify_thread_del_channel(struct lttng_consumer_local_data *ctx,
142 uint64_t key)
143{
144 notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL);
145}
146
d8ef542d
MD
147static int read_channel_pipe(struct lttng_consumer_local_data *ctx,
148 struct lttng_consumer_channel **chan,
a0cbdd2e 149 uint64_t *key,
d8ef542d
MD
150 enum consumer_channel_action *action)
151{
152 struct consumer_channel_msg msg;
6cd525e8 153 ssize_t ret;
d8ef542d 154
6cd525e8
MD
155 ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg));
156 if (ret < sizeof(msg)) {
157 ret = -1;
158 goto error;
d8ef542d 159 }
6cd525e8
MD
160 *action = msg.action;
161 *chan = msg.chan;
162 *key = msg.key;
163error:
164 return (int) ret;
d8ef542d
MD
165}
166
212d67a2
DG
167/*
168 * Cleanup the stream list of a channel. Those streams are not yet globally
169 * visible
170 */
171static void clean_channel_stream_list(struct lttng_consumer_channel *channel)
172{
173 struct lttng_consumer_stream *stream, *stmp;
174
175 assert(channel);
176
177 /* Delete streams that might have been left in the stream list. */
178 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
179 send_node) {
180 cds_list_del(&stream->send_node);
181 /*
182 * Once a stream is added to this list, the buffers were created so we
183 * have a guarantee that this call will succeed. Setting the monitor
184 * mode to 0 so we don't lock nor try to delete the stream from the
185 * global hash table.
186 */
187 stream->monitor = 0;
188 consumer_stream_destroy(stream, NULL);
189 }
190}
191
3bd1e081
MD
192/*
193 * Find a stream. The consumer_data.lock must be locked during this
194 * call.
195 */
d88aee68 196static struct lttng_consumer_stream *find_stream(uint64_t key,
8389e4f8 197 struct lttng_ht *ht)
3bd1e081 198{
e4421fec 199 struct lttng_ht_iter iter;
d88aee68 200 struct lttng_ht_node_u64 *node;
e4421fec 201 struct lttng_consumer_stream *stream = NULL;
3bd1e081 202
8389e4f8
DG
203 assert(ht);
204
d88aee68
DG
205 /* -1ULL keys are lookup failures */
206 if (key == (uint64_t) -1ULL) {
7ad0a0cb 207 return NULL;
7a57cf92 208 }
e4421fec 209
6065ceec
DG
210 rcu_read_lock();
211
d88aee68
DG
212 lttng_ht_lookup(ht, &key, &iter);
213 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
214 if (node != NULL) {
215 stream = caa_container_of(node, struct lttng_consumer_stream, node);
3bd1e081 216 }
e4421fec 217
6065ceec
DG
218 rcu_read_unlock();
219
e4421fec 220 return stream;
3bd1e081
MD
221}
222
da009f2c 223static void steal_stream_key(uint64_t key, struct lttng_ht *ht)
7ad0a0cb
MD
224{
225 struct lttng_consumer_stream *stream;
226
04253271 227 rcu_read_lock();
ffe60014 228 stream = find_stream(key, ht);
04253271 229 if (stream) {
da009f2c 230 stream->key = (uint64_t) -1ULL;
04253271
MD
231 /*
232 * We don't want the lookup to match, but we still need
233 * to iterate on this stream when iterating over the hash table. Just
234 * change the node key.
235 */
da009f2c 236 stream->node.key = (uint64_t) -1ULL;
04253271
MD
237 }
238 rcu_read_unlock();
7ad0a0cb
MD
239}
240
d56db448
DG
241/*
242 * Return a channel object for the given key.
243 *
244 * RCU read side lock MUST be acquired before calling this function and
245 * protects the channel ptr.
246 */
d88aee68 247struct lttng_consumer_channel *consumer_find_channel(uint64_t key)
3bd1e081 248{
e4421fec 249 struct lttng_ht_iter iter;
d88aee68 250 struct lttng_ht_node_u64 *node;
e4421fec 251 struct lttng_consumer_channel *channel = NULL;
3bd1e081 252
d88aee68
DG
253 /* -1ULL keys are lookup failures */
254 if (key == (uint64_t) -1ULL) {
7ad0a0cb 255 return NULL;
7a57cf92 256 }
e4421fec 257
d88aee68
DG
258 lttng_ht_lookup(consumer_data.channel_ht, &key, &iter);
259 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
260 if (node != NULL) {
261 channel = caa_container_of(node, struct lttng_consumer_channel, node);
3bd1e081 262 }
e4421fec
DG
263
264 return channel;
3bd1e081
MD
265}
266
b5a6470f
DG
267/*
268 * There is a possibility that the consumer does not have enough time between
269 * the close of the channel on the session daemon and the cleanup in here thus
270 * once we have a channel add with an existing key, we know for sure that this
271 * channel will eventually get cleaned up by all streams being closed.
272 *
273 * This function just nullifies the already existing channel key.
274 */
275static void steal_channel_key(uint64_t key)
276{
277 struct lttng_consumer_channel *channel;
278
279 rcu_read_lock();
280 channel = consumer_find_channel(key);
281 if (channel) {
282 channel->key = (uint64_t) -1ULL;
283 /*
284 * We don't want the lookup to match, but we still need to iterate on
285 * this channel when iterating over the hash table. Just change the
286 * node key.
287 */
288 channel->node.key = (uint64_t) -1ULL;
289 }
290 rcu_read_unlock();
291}
292
ffe60014 293static void free_channel_rcu(struct rcu_head *head)
702b1ea4 294{
d88aee68
DG
295 struct lttng_ht_node_u64 *node =
296 caa_container_of(head, struct lttng_ht_node_u64, head);
ffe60014
DG
297 struct lttng_consumer_channel *channel =
298 caa_container_of(node, struct lttng_consumer_channel, node);
702b1ea4 299
b83e03c4
MD
300 switch (consumer_data.type) {
301 case LTTNG_CONSUMER_KERNEL:
302 break;
303 case LTTNG_CONSUMER32_UST:
304 case LTTNG_CONSUMER64_UST:
305 lttng_ustconsumer_free_channel(channel);
306 break;
307 default:
308 ERR("Unknown consumer_data type");
309 abort();
310 }
ffe60014 311 free(channel);
702b1ea4
MD
312}
313
00e2e675
DG
314/*
315 * RCU protected relayd socket pair free.
316 */
ffe60014 317static void free_relayd_rcu(struct rcu_head *head)
00e2e675 318{
d88aee68
DG
319 struct lttng_ht_node_u64 *node =
320 caa_container_of(head, struct lttng_ht_node_u64, head);
00e2e675
DG
321 struct consumer_relayd_sock_pair *relayd =
322 caa_container_of(node, struct consumer_relayd_sock_pair, node);
323
8994307f
DG
324 /*
325 * Close all sockets. This is done in the call RCU since we don't want the
326 * socket fds to be reassigned thus potentially creating bad state of the
327 * relayd object.
328 *
329 * We do not have to lock the control socket mutex here since at this stage
330 * there is no one referencing to this relayd object.
331 */
332 (void) relayd_close(&relayd->control_sock);
333 (void) relayd_close(&relayd->data_sock);
334
3a84e2f3 335 pthread_mutex_destroy(&relayd->ctrl_sock_mutex);
00e2e675
DG
336 free(relayd);
337}
338
339/*
340 * Destroy and free relayd socket pair object.
00e2e675 341 */
51230d70 342void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
343{
344 int ret;
345 struct lttng_ht_iter iter;
346
173af62f
DG
347 if (relayd == NULL) {
348 return;
349 }
350
00e2e675
DG
351 DBG("Consumer destroy and close relayd socket pair");
352
353 iter.iter.node = &relayd->node.node;
354 ret = lttng_ht_del(consumer_data.relayd_ht, &iter);
173af62f 355 if (ret != 0) {
8994307f 356 /* We assume the relayd is being or is destroyed */
173af62f
DG
357 return;
358 }
00e2e675 359
00e2e675 360 /* RCU free() call */
ffe60014
DG
361 call_rcu(&relayd->node.head, free_relayd_rcu);
362}
363
364/*
365 * Remove a channel from the global list protected by a mutex. This function is
366 * also responsible for freeing its data structures.
367 */
368void consumer_del_channel(struct lttng_consumer_channel *channel)
369{
ffe60014
DG
370 struct lttng_ht_iter iter;
371
d88aee68 372 DBG("Consumer delete channel key %" PRIu64, channel->key);
ffe60014
DG
373
374 pthread_mutex_lock(&consumer_data.lock);
a9838785 375 pthread_mutex_lock(&channel->lock);
ffe60014 376
212d67a2
DG
377 /* Destroy streams that might have been left in the stream list. */
378 clean_channel_stream_list(channel);
51e762e5 379
d3e2ba59
JD
380 if (channel->live_timer_enabled == 1) {
381 consumer_timer_live_stop(channel);
382 }
e9404c27
JG
383 if (channel->monitor_timer_enabled == 1) {
384 consumer_timer_monitor_stop(channel);
385 }
d3e2ba59 386
ffe60014
DG
387 switch (consumer_data.type) {
388 case LTTNG_CONSUMER_KERNEL:
389 break;
390 case LTTNG_CONSUMER32_UST:
391 case LTTNG_CONSUMER64_UST:
392 lttng_ustconsumer_del_channel(channel);
393 break;
394 default:
395 ERR("Unknown consumer_data type");
396 assert(0);
397 goto end;
398 }
399
d2956687
JG
400 lttng_trace_chunk_put(channel->trace_chunk);
401 channel->trace_chunk = NULL;
5c3892a6 402
d2956687
JG
403 if (channel->is_published) {
404 int ret;
405
406 rcu_read_lock();
407 iter.iter.node = &channel->node.node;
408 ret = lttng_ht_del(consumer_data.channel_ht, &iter);
409 assert(!ret);
ffe60014 410
d2956687
JG
411 iter.iter.node = &channel->channels_by_session_id_ht_node.node;
412 ret = lttng_ht_del(consumer_data.channels_by_session_id_ht,
413 &iter);
414 assert(!ret);
415 rcu_read_unlock();
416 }
417
b6921a17
JG
418 channel->is_deleted = true;
419 call_rcu(&channel->node.head, free_channel_rcu);
ffe60014 420end:
a9838785 421 pthread_mutex_unlock(&channel->lock);
ffe60014 422 pthread_mutex_unlock(&consumer_data.lock);
00e2e675
DG
423}
424
228b5bf7
DG
425/*
426 * Iterate over the relayd hash table and destroy each element. Finally,
427 * destroy the whole hash table.
428 */
429static void cleanup_relayd_ht(void)
430{
431 struct lttng_ht_iter iter;
432 struct consumer_relayd_sock_pair *relayd;
433
434 rcu_read_lock();
435
436 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
437 node.node) {
51230d70 438 consumer_destroy_relayd(relayd);
228b5bf7
DG
439 }
440
228b5bf7 441 rcu_read_unlock();
36b588ed
MD
442
443 lttng_ht_destroy(consumer_data.relayd_ht);
228b5bf7
DG
444}
445
8994307f
DG
446/*
447 * Update the end point status of all streams having the given network sequence
448 * index (relayd index).
449 *
450 * It's atomically set without having the stream mutex locked which is fine
451 * because we handle the write/read race with a pipe wakeup for each thread.
452 */
da009f2c 453static void update_endpoint_status_by_netidx(uint64_t net_seq_idx,
8994307f
DG
454 enum consumer_endpoint_status status)
455{
456 struct lttng_ht_iter iter;
457 struct lttng_consumer_stream *stream;
458
da009f2c 459 DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx);
8994307f
DG
460
461 rcu_read_lock();
462
463 /* Let's begin with metadata */
464 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
465 if (stream->net_seq_idx == net_seq_idx) {
466 uatomic_set(&stream->endpoint_status, status);
467 DBG("Delete flag set to metadata stream %d", stream->wait_fd);
468 }
469 }
470
471 /* Follow up by the data streams */
472 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
473 if (stream->net_seq_idx == net_seq_idx) {
474 uatomic_set(&stream->endpoint_status, status);
475 DBG("Delete flag set to data stream %d", stream->wait_fd);
476 }
477 }
478 rcu_read_unlock();
479}
480
481/*
482 * Cleanup a relayd object by flagging every associated streams for deletion,
483 * destroying the object meaning removing it from the relayd hash table,
484 * closing the sockets and freeing the memory in a RCU call.
485 *
486 * If a local data context is available, notify the threads that the streams'
487 * state have changed.
488 */
9276e5c8 489void lttng_consumer_cleanup_relayd(struct consumer_relayd_sock_pair *relayd)
8994307f 490{
da009f2c 491 uint64_t netidx;
8994307f
DG
492
493 assert(relayd);
494
9276e5c8 495 DBG("Cleaning up relayd object ID %"PRIu64, relayd->net_seq_idx);
9617607b 496
8994307f
DG
497 /* Save the net sequence index before destroying the object */
498 netidx = relayd->net_seq_idx;
499
500 /*
501 * Delete the relayd from the relayd hash table, close the sockets and free
502 * the object in a RCU call.
503 */
51230d70 504 consumer_destroy_relayd(relayd);
8994307f
DG
505
506 /* Set inactive endpoint to all streams */
507 update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE);
508
509 /*
510 * With a local data context, notify the threads that the streams' state
511 * have changed. The write() action on the pipe acts as an "implicit"
512 * memory barrier ordering the updates of the end point status from the
513 * read of this status which happens AFTER receiving this notify.
514 */
9276e5c8
JR
515 notify_thread_lttng_pipe(relayd->ctx->consumer_data_pipe);
516 notify_thread_lttng_pipe(relayd->ctx->consumer_metadata_pipe);
8994307f
DG
517}
518
a6ba4fe1
DG
519/*
520 * Flag a relayd socket pair for destruction. Destroy it if the refcount
521 * reaches zero.
522 *
523 * RCU read side lock MUST be aquired before calling this function.
524 */
525void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd)
526{
527 assert(relayd);
528
529 /* Set destroy flag for this object */
530 uatomic_set(&relayd->destroy_flag, 1);
531
532 /* Destroy the relayd if refcount is 0 */
533 if (uatomic_read(&relayd->refcount) == 0) {
51230d70 534 consumer_destroy_relayd(relayd);
a6ba4fe1
DG
535 }
536}
537
3bd1e081 538/*
1d1a276c
DG
539 * Completly destroy stream from every visiable data structure and the given
540 * hash table if one.
541 *
542 * One this call returns, the stream object is not longer usable nor visible.
3bd1e081 543 */
e316aad5
DG
544void consumer_del_stream(struct lttng_consumer_stream *stream,
545 struct lttng_ht *ht)
3bd1e081 546{
1d1a276c 547 consumer_stream_destroy(stream, ht);
3bd1e081
MD
548}
549
5ab66908
MD
550/*
551 * XXX naming of del vs destroy is all mixed up.
552 */
553void consumer_del_stream_for_data(struct lttng_consumer_stream *stream)
554{
555 consumer_stream_destroy(stream, data_ht);
556}
557
558void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream)
559{
560 consumer_stream_destroy(stream, metadata_ht);
561}
562
d9a2e16e
JD
563void consumer_stream_update_channel_attributes(
564 struct lttng_consumer_stream *stream,
565 struct lttng_consumer_channel *channel)
566{
567 stream->channel_read_only_attributes.tracefile_size =
568 channel->tracefile_size;
d9a2e16e
JD
569}
570
d88aee68
DG
571struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key,
572 uint64_t stream_key,
ffe60014 573 const char *channel_name,
57a269f2 574 uint64_t relayd_id,
53632229 575 uint64_t session_id,
d2956687 576 struct lttng_trace_chunk *trace_chunk,
ffe60014
DG
577 int cpu,
578 int *alloc_ret,
4891ece8 579 enum consumer_channel_type type,
d2956687 580 unsigned int monitor)
3bd1e081 581{
ffe60014 582 int ret;
3bd1e081 583 struct lttng_consumer_stream *stream;
3bd1e081 584
effcf122 585 stream = zmalloc(sizeof(*stream));
3bd1e081 586 if (stream == NULL) {
7a57cf92 587 PERROR("malloc struct lttng_consumer_stream");
ffe60014 588 ret = -ENOMEM;
7a57cf92 589 goto end;
3bd1e081 590 }
7a57cf92 591
d2956687
JG
592 if (trace_chunk && !lttng_trace_chunk_get(trace_chunk)) {
593 ERR("Failed to acquire trace chunk reference during the creation of a stream");
594 ret = -1;
595 goto error;
596 }
d56db448 597
d2956687 598 rcu_read_lock();
3bd1e081 599 stream->key = stream_key;
d2956687 600 stream->trace_chunk = trace_chunk;
3bd1e081
MD
601 stream->out_fd = -1;
602 stream->out_fd_offset = 0;
e5d1a9b3 603 stream->output_written = 0;
ffe60014 604 stream->net_seq_idx = relayd_id;
53632229 605 stream->session_id = session_id;
4891ece8 606 stream->monitor = monitor;
774d490c 607 stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE;
f8f3885c 608 stream->index_file = NULL;
fb83fe64 609 stream->last_sequence_number = -1ULL;
a40a503f 610 stream->rotate_position = -1ULL;
53632229 611 pthread_mutex_init(&stream->lock, NULL);
c585821b 612 pthread_mutex_init(&stream->metadata_timer_lock, NULL);
58b1f425 613
ffe60014
DG
614 /* If channel is the metadata, flag this stream as metadata. */
615 if (type == CONSUMER_CHANNEL_TYPE_METADATA) {
616 stream->metadata_flag = 1;
617 /* Metadata is flat out. */
618 strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name));
94d49140
JD
619 /* Live rendez-vous point. */
620 pthread_cond_init(&stream->metadata_rdv, NULL);
621 pthread_mutex_init(&stream->metadata_rdv_lock, NULL);
58b1f425 622 } else {
ffe60014
DG
623 /* Format stream name to <channel_name>_<cpu_number> */
624 ret = snprintf(stream->name, sizeof(stream->name), "%s_%d",
625 channel_name, cpu);
626 if (ret < 0) {
627 PERROR("snprintf stream name");
628 goto error;
629 }
58b1f425 630 }
c30aaa51 631
ffe60014 632 /* Key is always the wait_fd for streams. */
d88aee68 633 lttng_ht_node_init_u64(&stream->node, stream->key);
ffe60014 634
d8ef542d
MD
635 /* Init node per channel id key */
636 lttng_ht_node_init_u64(&stream->node_channel_id, channel_key);
637
53632229 638 /* Init session id node with the stream session id */
d88aee68 639 lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id);
53632229 640
07b86b52
JD
641 DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64
642 " relayd_id %" PRIu64 ", session_id %" PRIu64,
643 stream->name, stream->key, channel_key,
644 stream->net_seq_idx, stream->session_id);
d56db448
DG
645
646 rcu_read_unlock();
3bd1e081 647 return stream;
c80048c6
MD
648
649error:
d56db448 650 rcu_read_unlock();
d2956687 651 lttng_trace_chunk_put(stream->trace_chunk);
c80048c6 652 free(stream);
7a57cf92 653end:
ffe60014
DG
654 if (alloc_ret) {
655 *alloc_ret = ret;
656 }
c80048c6 657 return NULL;
3bd1e081
MD
658}
659
660/*
661 * Add a stream to the global list protected by a mutex.
662 */
66d583dc 663void consumer_add_data_stream(struct lttng_consumer_stream *stream)
3bd1e081 664{
5ab66908 665 struct lttng_ht *ht = data_ht;
3bd1e081 666
e316aad5 667 assert(stream);
43c34bc3 668 assert(ht);
c77fc10a 669
d88aee68 670 DBG3("Adding consumer stream %" PRIu64, stream->key);
e316aad5
DG
671
672 pthread_mutex_lock(&consumer_data.lock);
a9838785 673 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 674 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 675 pthread_mutex_lock(&stream->lock);
b0b335c8 676 rcu_read_lock();
e316aad5 677
43c34bc3 678 /* Steal stream identifier to avoid having streams with the same key */
ffe60014 679 steal_stream_key(stream->key, ht);
43c34bc3 680
d88aee68 681 lttng_ht_add_unique_u64(ht, &stream->node);
00e2e675 682
d8ef542d
MD
683 lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht,
684 &stream->node_channel_id);
685
ca22feea
DG
686 /*
687 * Add stream to the stream_list_ht of the consumer data. No need to steal
688 * the key since the HT does not use it and we allow to add redundant keys
689 * into this table.
690 */
d88aee68 691 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 692
e316aad5 693 /*
ffe60014
DG
694 * When nb_init_stream_left reaches 0, we don't need to trigger any action
695 * in terms of destroying the associated channel, because the action that
e316aad5
DG
696 * causes the count to become 0 also causes a stream to be added. The
697 * channel deletion will thus be triggered by the following removal of this
698 * stream.
699 */
ffe60014 700 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
701 /* Increment refcount before decrementing nb_init_stream_left */
702 cmm_smp_wmb();
ffe60014 703 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
704 }
705
706 /* Update consumer data once the node is inserted. */
3bd1e081
MD
707 consumer_data.stream_count++;
708 consumer_data.need_update = 1;
709
e316aad5 710 rcu_read_unlock();
2e818a6a 711 pthread_mutex_unlock(&stream->lock);
ec6ea7d0 712 pthread_mutex_unlock(&stream->chan->timer_lock);
a9838785 713 pthread_mutex_unlock(&stream->chan->lock);
3bd1e081 714 pthread_mutex_unlock(&consumer_data.lock);
3bd1e081
MD
715}
716
00e2e675 717/*
3f8e211f
DG
718 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
719 * be acquired before calling this.
00e2e675 720 */
d09e1200 721static int add_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
722{
723 int ret = 0;
d88aee68 724 struct lttng_ht_node_u64 *node;
00e2e675
DG
725 struct lttng_ht_iter iter;
726
ffe60014 727 assert(relayd);
00e2e675 728
00e2e675 729 lttng_ht_lookup(consumer_data.relayd_ht,
d88aee68
DG
730 &relayd->net_seq_idx, &iter);
731 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675 732 if (node != NULL) {
00e2e675
DG
733 goto end;
734 }
d88aee68 735 lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node);
00e2e675 736
00e2e675
DG
737end:
738 return ret;
739}
740
741/*
742 * Allocate and return a consumer relayd socket.
743 */
027a694f 744static struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair(
da009f2c 745 uint64_t net_seq_idx)
00e2e675
DG
746{
747 struct consumer_relayd_sock_pair *obj = NULL;
748
da009f2c
MD
749 /* net sequence index of -1 is a failure */
750 if (net_seq_idx == (uint64_t) -1ULL) {
00e2e675
DG
751 goto error;
752 }
753
754 obj = zmalloc(sizeof(struct consumer_relayd_sock_pair));
755 if (obj == NULL) {
756 PERROR("zmalloc relayd sock");
757 goto error;
758 }
759
760 obj->net_seq_idx = net_seq_idx;
761 obj->refcount = 0;
173af62f 762 obj->destroy_flag = 0;
f96e4545
MD
763 obj->control_sock.sock.fd = -1;
764 obj->data_sock.sock.fd = -1;
d88aee68 765 lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx);
00e2e675
DG
766 pthread_mutex_init(&obj->ctrl_sock_mutex, NULL);
767
768error:
769 return obj;
770}
771
772/*
773 * Find a relayd socket pair in the global consumer data.
774 *
775 * Return the object if found else NULL.
b0b335c8
MD
776 * RCU read-side lock must be held across this call and while using the
777 * returned object.
00e2e675 778 */
d88aee68 779struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key)
00e2e675
DG
780{
781 struct lttng_ht_iter iter;
d88aee68 782 struct lttng_ht_node_u64 *node;
00e2e675
DG
783 struct consumer_relayd_sock_pair *relayd = NULL;
784
785 /* Negative keys are lookup failures */
d88aee68 786 if (key == (uint64_t) -1ULL) {
00e2e675
DG
787 goto error;
788 }
789
d88aee68 790 lttng_ht_lookup(consumer_data.relayd_ht, &key,
00e2e675 791 &iter);
d88aee68 792 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675
DG
793 if (node != NULL) {
794 relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node);
795 }
796
00e2e675
DG
797error:
798 return relayd;
799}
800
10a50311
JD
801/*
802 * Find a relayd and send the stream
803 *
804 * Returns 0 on success, < 0 on error
805 */
806int consumer_send_relayd_stream(struct lttng_consumer_stream *stream,
807 char *path)
808{
809 int ret = 0;
810 struct consumer_relayd_sock_pair *relayd;
811
812 assert(stream);
813 assert(stream->net_seq_idx != -1ULL);
814 assert(path);
815
816 /* The stream is not metadata. Get relayd reference if exists. */
817 rcu_read_lock();
818 relayd = consumer_find_relayd(stream->net_seq_idx);
819 if (relayd != NULL) {
820 /* Add stream on the relayd */
821 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
822 ret = relayd_add_stream(&relayd->control_sock, stream->name,
5da88b0f 823 get_consumer_domain(), path, &stream->relayd_stream_id,
d2956687
JG
824 stream->chan->tracefile_size,
825 stream->chan->tracefile_count,
826 stream->trace_chunk);
10a50311
JD
827 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
828 if (ret < 0) {
9276e5c8
JR
829 ERR("Relayd add stream failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
830 lttng_consumer_cleanup_relayd(relayd);
10a50311
JD
831 goto end;
832 }
1c20f0e2 833
10a50311 834 uatomic_inc(&relayd->refcount);
d01178b6 835 stream->sent_to_relayd = 1;
10a50311
JD
836 } else {
837 ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.",
838 stream->key, stream->net_seq_idx);
839 ret = -1;
840 goto end;
841 }
842
843 DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64,
844 stream->name, stream->key, stream->net_seq_idx);
845
846end:
847 rcu_read_unlock();
848 return ret;
849}
850
a4baae1b
JD
851/*
852 * Find a relayd and send the streams sent message
853 *
854 * Returns 0 on success, < 0 on error
855 */
856int consumer_send_relayd_streams_sent(uint64_t net_seq_idx)
857{
858 int ret = 0;
859 struct consumer_relayd_sock_pair *relayd;
860
861 assert(net_seq_idx != -1ULL);
862
863 /* The stream is not metadata. Get relayd reference if exists. */
864 rcu_read_lock();
865 relayd = consumer_find_relayd(net_seq_idx);
866 if (relayd != NULL) {
867 /* Add stream on the relayd */
868 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
869 ret = relayd_streams_sent(&relayd->control_sock);
870 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
871 if (ret < 0) {
9276e5c8
JR
872 ERR("Relayd streams sent failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
873 lttng_consumer_cleanup_relayd(relayd);
a4baae1b
JD
874 goto end;
875 }
876 } else {
877 ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.",
878 net_seq_idx);
879 ret = -1;
880 goto end;
881 }
882
883 ret = 0;
884 DBG("All streams sent relayd id %" PRIu64, net_seq_idx);
885
886end:
887 rcu_read_unlock();
888 return ret;
889}
890
10a50311
JD
891/*
892 * Find a relayd and close the stream
893 */
894void close_relayd_stream(struct lttng_consumer_stream *stream)
895{
896 struct consumer_relayd_sock_pair *relayd;
897
898 /* The stream is not metadata. Get relayd reference if exists. */
899 rcu_read_lock();
900 relayd = consumer_find_relayd(stream->net_seq_idx);
901 if (relayd) {
902 consumer_stream_relayd_close(stream, relayd);
903 }
904 rcu_read_unlock();
905}
906
00e2e675
DG
907/*
908 * Handle stream for relayd transmission if the stream applies for network
909 * streaming where the net sequence index is set.
910 *
911 * Return destination file descriptor or negative value on error.
912 */
6197aea7 913static int write_relayd_stream_header(struct lttng_consumer_stream *stream,
1d4dfdef
DG
914 size_t data_size, unsigned long padding,
915 struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
916{
917 int outfd = -1, ret;
00e2e675
DG
918 struct lttcomm_relayd_data_hdr data_hdr;
919
920 /* Safety net */
921 assert(stream);
6197aea7 922 assert(relayd);
00e2e675
DG
923
924 /* Reset data header */
925 memset(&data_hdr, 0, sizeof(data_hdr));
926
00e2e675
DG
927 if (stream->metadata_flag) {
928 /* Caller MUST acquire the relayd control socket lock */
929 ret = relayd_send_metadata(&relayd->control_sock, data_size);
930 if (ret < 0) {
931 goto error;
932 }
933
934 /* Metadata are always sent on the control socket. */
6151a90f 935 outfd = relayd->control_sock.sock.fd;
00e2e675
DG
936 } else {
937 /* Set header with stream information */
938 data_hdr.stream_id = htobe64(stream->relayd_stream_id);
939 data_hdr.data_size = htobe32(data_size);
1d4dfdef 940 data_hdr.padding_size = htobe32(padding);
c35f9726 941
39df6d9f
DG
942 /*
943 * Note that net_seq_num below is assigned with the *current* value of
944 * next_net_seq_num and only after that the next_net_seq_num will be
945 * increment. This is why when issuing a command on the relayd using
946 * this next value, 1 should always be substracted in order to compare
947 * the last seen sequence number on the relayd side to the last sent.
948 */
3604f373 949 data_hdr.net_seq_num = htobe64(stream->next_net_seq_num);
00e2e675
DG
950 /* Other fields are zeroed previously */
951
952 ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr,
953 sizeof(data_hdr));
954 if (ret < 0) {
955 goto error;
956 }
957
3604f373
DG
958 ++stream->next_net_seq_num;
959
00e2e675 960 /* Set to go on data socket */
6151a90f 961 outfd = relayd->data_sock.sock.fd;
00e2e675
DG
962 }
963
964error:
965 return outfd;
966}
967
d2956687
JG
968/*
969 * Trigger a dump of the metadata content. Following/during the succesful
970 * completion of this call, the metadata poll thread will start receiving
971 * metadata packets to consume.
972 *
973 * The caller must hold the channel and stream locks.
974 */
975static
976int consumer_metadata_stream_dump(struct lttng_consumer_stream *stream)
977{
978 int ret;
979
980 ASSERT_LOCKED(stream->chan->lock);
981 ASSERT_LOCKED(stream->lock);
982 assert(stream->metadata_flag);
983 assert(stream->chan->trace_chunk);
984
985 switch (consumer_data.type) {
986 case LTTNG_CONSUMER_KERNEL:
987 /*
988 * Reset the position of what has been read from the
989 * metadata cache to 0 so we can dump it again.
990 */
991 ret = kernctl_metadata_cache_dump(stream->wait_fd);
992 break;
993 case LTTNG_CONSUMER32_UST:
994 case LTTNG_CONSUMER64_UST:
995 /*
996 * Reset the position pushed from the metadata cache so it
997 * will write from the beginning on the next push.
998 */
999 stream->ust_metadata_pushed = 0;
1000 ret = consumer_metadata_wakeup_pipe(stream->chan);
1001 break;
1002 default:
1003 ERR("Unknown consumer_data type");
1004 abort();
1005 }
1006 if (ret < 0) {
1007 ERR("Failed to dump the metadata cache");
1008 }
1009 return ret;
1010}
1011
1012static
1013int lttng_consumer_channel_set_trace_chunk(
1014 struct lttng_consumer_channel *channel,
1015 struct lttng_trace_chunk *new_trace_chunk)
1016{
d2956687 1017 pthread_mutex_lock(&channel->lock);
b6921a17
JG
1018 if (channel->is_deleted) {
1019 /*
1020 * The channel has been logically deleted and should no longer
1021 * be used. It has released its reference to its current trace
1022 * chunk and should not acquire a new one.
1023 *
1024 * Return success as there is nothing for the caller to do.
1025 */
1026 goto end;
1027 }
d2956687
JG
1028
1029 /*
1030 * The acquisition of the reference cannot fail (barring
1031 * a severe internal error) since a reference to the published
1032 * chunk is already held by the caller.
1033 */
1034 if (new_trace_chunk) {
1035 const bool acquired_reference = lttng_trace_chunk_get(
1036 new_trace_chunk);
1037
1038 assert(acquired_reference);
1039 }
1040
1041 lttng_trace_chunk_put(channel->trace_chunk);
1042 channel->trace_chunk = new_trace_chunk;
d2956687
JG
1043end:
1044 pthread_mutex_unlock(&channel->lock);
ce1aa6fe 1045 return 0;
d2956687
JG
1046}
1047
3bd1e081 1048/*
ffe60014
DG
1049 * Allocate and return a new lttng_consumer_channel object using the given key
1050 * to initialize the hash table node.
1051 *
1052 * On error, return NULL.
3bd1e081 1053 */
886224ff 1054struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key,
ffe60014 1055 uint64_t session_id,
d2956687 1056 const uint64_t *chunk_id,
ffe60014
DG
1057 const char *pathname,
1058 const char *name,
57a269f2 1059 uint64_t relayd_id,
1624d5b7
JD
1060 enum lttng_event_output output,
1061 uint64_t tracefile_size,
2bba9e53 1062 uint64_t tracefile_count,
1950109e 1063 uint64_t session_id_per_pid,
ecc48a90 1064 unsigned int monitor,
d7ba1388 1065 unsigned int live_timer_interval,
3d071855 1066 const char *root_shm_path,
d7ba1388 1067 const char *shm_path)
3bd1e081 1068{
d2956687
JG
1069 struct lttng_consumer_channel *channel = NULL;
1070 struct lttng_trace_chunk *trace_chunk = NULL;
1071
1072 if (chunk_id) {
1073 trace_chunk = lttng_trace_chunk_registry_find_chunk(
1074 consumer_data.chunk_registry, session_id,
1075 *chunk_id);
1076 if (!trace_chunk) {
1077 ERR("Failed to find trace chunk reference during creation of channel");
1078 goto end;
1079 }
1080 }
3bd1e081 1081
276b26d1 1082 channel = zmalloc(sizeof(*channel));
3bd1e081 1083 if (channel == NULL) {
7a57cf92 1084 PERROR("malloc struct lttng_consumer_channel");
3bd1e081
MD
1085 goto end;
1086 }
ffe60014
DG
1087
1088 channel->key = key;
3bd1e081 1089 channel->refcount = 0;
ffe60014 1090 channel->session_id = session_id;
1950109e 1091 channel->session_id_per_pid = session_id_per_pid;
ffe60014 1092 channel->relayd_id = relayd_id;
1624d5b7
JD
1093 channel->tracefile_size = tracefile_size;
1094 channel->tracefile_count = tracefile_count;
2bba9e53 1095 channel->monitor = monitor;
ecc48a90 1096 channel->live_timer_interval = live_timer_interval;
a9838785 1097 pthread_mutex_init(&channel->lock, NULL);
ec6ea7d0 1098 pthread_mutex_init(&channel->timer_lock, NULL);
ffe60014 1099
0c759fc9
DG
1100 switch (output) {
1101 case LTTNG_EVENT_SPLICE:
1102 channel->output = CONSUMER_CHANNEL_SPLICE;
1103 break;
1104 case LTTNG_EVENT_MMAP:
1105 channel->output = CONSUMER_CHANNEL_MMAP;
1106 break;
1107 default:
1108 assert(0);
1109 free(channel);
1110 channel = NULL;
1111 goto end;
1112 }
1113
07b86b52
JD
1114 /*
1115 * In monitor mode, the streams associated with the channel will be put in
1116 * a special list ONLY owned by this channel. So, the refcount is set to 1
1117 * here meaning that the channel itself has streams that are referenced.
1118 *
1119 * On a channel deletion, once the channel is no longer visible, the
1120 * refcount is decremented and checked for a zero value to delete it. With
1121 * streams in no monitor mode, it will now be safe to destroy the channel.
1122 */
1123 if (!channel->monitor) {
1124 channel->refcount = 1;
1125 }
1126
ffe60014
DG
1127 strncpy(channel->pathname, pathname, sizeof(channel->pathname));
1128 channel->pathname[sizeof(channel->pathname) - 1] = '\0';
1129
1130 strncpy(channel->name, name, sizeof(channel->name));
1131 channel->name[sizeof(channel->name) - 1] = '\0';
1132
3d071855
MD
1133 if (root_shm_path) {
1134 strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path));
1135 channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0';
1136 }
d7ba1388
MD
1137 if (shm_path) {
1138 strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path));
1139 channel->shm_path[sizeof(channel->shm_path) - 1] = '\0';
1140 }
1141
d88aee68 1142 lttng_ht_node_init_u64(&channel->node, channel->key);
5c3892a6
JG
1143 lttng_ht_node_init_u64(&channel->channels_by_session_id_ht_node,
1144 channel->session_id);
d8ef542d
MD
1145
1146 channel->wait_fd = -1;
ffe60014
DG
1147 CDS_INIT_LIST_HEAD(&channel->streams.head);
1148
d2956687
JG
1149 if (trace_chunk) {
1150 int ret = lttng_consumer_channel_set_trace_chunk(channel,
1151 trace_chunk);
1152 if (ret) {
1153 goto error;
1154 }
1155 }
1156
62a7b8ed 1157 DBG("Allocated channel (key %" PRIu64 ")", channel->key);
3bd1e081 1158
3bd1e081 1159end:
d2956687 1160 lttng_trace_chunk_put(trace_chunk);
3bd1e081 1161 return channel;
d2956687
JG
1162error:
1163 consumer_del_channel(channel);
1164 channel = NULL;
1165 goto end;
3bd1e081
MD
1166}
1167
1168/*
1169 * Add a channel to the global list protected by a mutex.
821fffb2 1170 *
b5a6470f 1171 * Always return 0 indicating success.
3bd1e081 1172 */
d8ef542d
MD
1173int consumer_add_channel(struct lttng_consumer_channel *channel,
1174 struct lttng_consumer_local_data *ctx)
3bd1e081 1175{
3bd1e081 1176 pthread_mutex_lock(&consumer_data.lock);
a9838785 1177 pthread_mutex_lock(&channel->lock);
ec6ea7d0 1178 pthread_mutex_lock(&channel->timer_lock);
c77fc10a 1179
b5a6470f
DG
1180 /*
1181 * This gives us a guarantee that the channel we are about to add to the
1182 * channel hash table will be unique. See this function comment on the why
1183 * we need to steel the channel key at this stage.
1184 */
1185 steal_channel_key(channel->key);
c77fc10a 1186
b5a6470f 1187 rcu_read_lock();
d88aee68 1188 lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node);
5c3892a6
JG
1189 lttng_ht_add_u64(consumer_data.channels_by_session_id_ht,
1190 &channel->channels_by_session_id_ht_node);
6065ceec 1191 rcu_read_unlock();
d2956687 1192 channel->is_published = true;
b5a6470f 1193
ec6ea7d0 1194 pthread_mutex_unlock(&channel->timer_lock);
a9838785 1195 pthread_mutex_unlock(&channel->lock);
3bd1e081 1196 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 1197
b5a6470f 1198 if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) {
a0cbdd2e 1199 notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD);
d8ef542d 1200 }
b5a6470f
DG
1201
1202 return 0;
3bd1e081
MD
1203}
1204
1205/*
1206 * Allocate the pollfd structure and the local view of the out fds to avoid
1207 * doing a lookup in the linked list and concurrency issues when writing is
1208 * needed. Called with consumer_data.lock held.
1209 *
1210 * Returns the number of fds in the structures.
1211 */
ffe60014
DG
1212static int update_poll_array(struct lttng_consumer_local_data *ctx,
1213 struct pollfd **pollfd, struct lttng_consumer_stream **local_stream,
9a2fcf78 1214 struct lttng_ht *ht, int *nb_inactive_fd)
3bd1e081 1215{
3bd1e081 1216 int i = 0;
e4421fec
DG
1217 struct lttng_ht_iter iter;
1218 struct lttng_consumer_stream *stream;
3bd1e081 1219
ffe60014
DG
1220 assert(ctx);
1221 assert(ht);
1222 assert(pollfd);
1223 assert(local_stream);
1224
3bd1e081 1225 DBG("Updating poll fd array");
9a2fcf78 1226 *nb_inactive_fd = 0;
481d6c57 1227 rcu_read_lock();
43c34bc3 1228 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
8994307f
DG
1229 /*
1230 * Only active streams with an active end point can be added to the
1231 * poll set and local stream storage of the thread.
1232 *
1233 * There is a potential race here for endpoint_status to be updated
1234 * just after the check. However, this is OK since the stream(s) will
1235 * be deleted once the thread is notified that the end point state has
1236 * changed where this function will be called back again.
9a2fcf78
JD
1237 *
1238 * We track the number of inactive FDs because they still need to be
1239 * closed by the polling thread after a wakeup on the data_pipe or
1240 * metadata_pipe.
8994307f 1241 */
d2956687 1242 if (stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) {
9a2fcf78 1243 (*nb_inactive_fd)++;
3bd1e081
MD
1244 continue;
1245 }
7972aab2
DG
1246 /*
1247 * This clobbers way too much the debug output. Uncomment that if you
1248 * need it for debugging purposes.
7972aab2 1249 */
e4421fec 1250 (*pollfd)[i].fd = stream->wait_fd;
3bd1e081 1251 (*pollfd)[i].events = POLLIN | POLLPRI;
e4421fec 1252 local_stream[i] = stream;
3bd1e081
MD
1253 i++;
1254 }
481d6c57 1255 rcu_read_unlock();
3bd1e081
MD
1256
1257 /*
50f8ae69 1258 * Insert the consumer_data_pipe at the end of the array and don't
3bd1e081
MD
1259 * increment i so nb_fd is the number of real FD.
1260 */
acdb9057 1261 (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe);
509bb1cf 1262 (*pollfd)[i].events = POLLIN | POLLPRI;
02b3d176
DG
1263
1264 (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe);
1265 (*pollfd)[i + 1].events = POLLIN | POLLPRI;
3bd1e081
MD
1266 return i;
1267}
1268
1269/*
84382d49
MD
1270 * Poll on the should_quit pipe and the command socket return -1 on
1271 * error, 1 if should exit, 0 if data is available on the command socket
3bd1e081
MD
1272 */
1273int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll)
1274{
1275 int num_rdy;
1276
88f2b785 1277restart:
3bd1e081
MD
1278 num_rdy = poll(consumer_sockpoll, 2, -1);
1279 if (num_rdy == -1) {
88f2b785
MD
1280 /*
1281 * Restart interrupted system call.
1282 */
1283 if (errno == EINTR) {
1284 goto restart;
1285 }
7a57cf92 1286 PERROR("Poll error");
84382d49 1287 return -1;
3bd1e081 1288 }
509bb1cf 1289 if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) {
3bd1e081 1290 DBG("consumer_should_quit wake up");
84382d49 1291 return 1;
3bd1e081
MD
1292 }
1293 return 0;
3bd1e081
MD
1294}
1295
1296/*
1297 * Set the error socket.
1298 */
ffe60014
DG
1299void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx,
1300 int sock)
3bd1e081
MD
1301{
1302 ctx->consumer_error_socket = sock;
1303}
1304
1305/*
1306 * Set the command socket path.
1307 */
3bd1e081
MD
1308void lttng_consumer_set_command_sock_path(
1309 struct lttng_consumer_local_data *ctx, char *sock)
1310{
1311 ctx->consumer_command_sock_path = sock;
1312}
1313
1314/*
1315 * Send return code to the session daemon.
1316 * If the socket is not defined, we return 0, it is not a fatal error
1317 */
ffe60014 1318int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd)
3bd1e081
MD
1319{
1320 if (ctx->consumer_error_socket > 0) {
1321 return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd,
1322 sizeof(enum lttcomm_sessiond_command));
1323 }
1324
1325 return 0;
1326}
1327
1328/*
228b5bf7
DG
1329 * Close all the tracefiles and stream fds and MUST be called when all
1330 * instances are destroyed i.e. when all threads were joined and are ended.
3bd1e081
MD
1331 */
1332void lttng_consumer_cleanup(void)
1333{
e4421fec 1334 struct lttng_ht_iter iter;
ffe60014 1335 struct lttng_consumer_channel *channel;
e10aec8f 1336 unsigned int trace_chunks_left;
6065ceec
DG
1337
1338 rcu_read_lock();
3bd1e081 1339
ffe60014
DG
1340 cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel,
1341 node.node) {
702b1ea4 1342 consumer_del_channel(channel);
3bd1e081 1343 }
6065ceec
DG
1344
1345 rcu_read_unlock();
d6ce1df2 1346
d6ce1df2 1347 lttng_ht_destroy(consumer_data.channel_ht);
5c3892a6 1348 lttng_ht_destroy(consumer_data.channels_by_session_id_ht);
228b5bf7
DG
1349
1350 cleanup_relayd_ht();
1351
d8ef542d
MD
1352 lttng_ht_destroy(consumer_data.stream_per_chan_id_ht);
1353
228b5bf7
DG
1354 /*
1355 * This HT contains streams that are freed by either the metadata thread or
1356 * the data thread so we do *nothing* on the hash table and simply destroy
1357 * it.
1358 */
1359 lttng_ht_destroy(consumer_data.stream_list_ht);
28cc88f3 1360
e10aec8f
MD
1361 /*
1362 * Trace chunks in the registry may still exist if the session
1363 * daemon has encountered an internal error and could not
1364 * tear down its sessions and/or trace chunks properly.
1365 *
1366 * Release the session daemon's implicit reference to any remaining
1367 * trace chunk and print an error if any trace chunk was found. Note
1368 * that there are _no_ legitimate cases for trace chunks to be left,
1369 * it is a leak. However, it can happen following a crash of the
1370 * session daemon and not emptying the registry would cause an assertion
1371 * to hit.
1372 */
1373 trace_chunks_left = lttng_trace_chunk_registry_put_each_chunk(
1374 consumer_data.chunk_registry);
1375 if (trace_chunks_left) {
1376 ERR("%u trace chunks are leaked by lttng-consumerd. "
1377 "This can be caused by an internal error of the session daemon.",
1378 trace_chunks_left);
1379 }
1380 /* Run all callbacks freeing each chunk. */
1381 rcu_barrier();
28cc88f3 1382 lttng_trace_chunk_registry_destroy(consumer_data.chunk_registry);
3bd1e081
MD
1383}
1384
1385/*
1386 * Called from signal handler.
1387 */
1388void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx)
1389{
6cd525e8
MD
1390 ssize_t ret;
1391
10211f5c 1392 CMM_STORE_SHARED(consumer_quit, 1);
6cd525e8
MD
1393 ret = lttng_write(ctx->consumer_should_quit[1], "4", 1);
1394 if (ret < 1) {
7a57cf92 1395 PERROR("write consumer quit");
3bd1e081 1396 }
ab1027f4
DG
1397
1398 DBG("Consumer flag that it should quit");
3bd1e081
MD
1399}
1400
5199ffc4
JG
1401
1402/*
1403 * Flush pending writes to trace output disk file.
1404 */
1405static
00e2e675
DG
1406void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream,
1407 off_t orig_offset)
3bd1e081 1408{
c7a78aab 1409 int ret;
3bd1e081
MD
1410 int outfd = stream->out_fd;
1411
1412 /*
1413 * This does a blocking write-and-wait on any page that belongs to the
1414 * subbuffer prior to the one we just wrote.
1415 * Don't care about error values, as these are just hints and ways to
1416 * limit the amount of page cache used.
1417 */
ffe60014 1418 if (orig_offset < stream->max_sb_size) {
3bd1e081
MD
1419 return;
1420 }
ffe60014
DG
1421 lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size,
1422 stream->max_sb_size,
3bd1e081
MD
1423 SYNC_FILE_RANGE_WAIT_BEFORE
1424 | SYNC_FILE_RANGE_WRITE
1425 | SYNC_FILE_RANGE_WAIT_AFTER);
1426 /*
1427 * Give hints to the kernel about how we access the file:
1428 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1429 * we write it.
1430 *
1431 * We need to call fadvise again after the file grows because the
1432 * kernel does not seem to apply fadvise to non-existing parts of the
1433 * file.
1434 *
1435 * Call fadvise _after_ having waited for the page writeback to
1436 * complete because the dirty page writeback semantic is not well
1437 * defined. So it can be expected to lead to lower throughput in
1438 * streaming.
1439 */
c7a78aab 1440 ret = posix_fadvise(outfd, orig_offset - stream->max_sb_size,
ffe60014 1441 stream->max_sb_size, POSIX_FADV_DONTNEED);
a0d0e127 1442 if (ret && ret != -ENOSYS) {
a74a5f4a
JG
1443 errno = ret;
1444 PERROR("posix_fadvise on fd %i", outfd);
c7a78aab 1445 }
3bd1e081
MD
1446}
1447
1448/*
1449 * Initialise the necessary environnement :
1450 * - create a new context
1451 * - create the poll_pipe
1452 * - create the should_quit pipe (for signal handler)
1453 * - create the thread pipe (for splice)
1454 *
1455 * Takes a function pointer as argument, this function is called when data is
1456 * available on a buffer. This function is responsible to do the
1457 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1458 * buffer configuration and then kernctl_put_next_subbuf at the end.
1459 *
1460 * Returns a pointer to the new context or NULL on error.
1461 */
1462struct lttng_consumer_local_data *lttng_consumer_create(
1463 enum lttng_consumer_type type,
4078b776 1464 ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream,
d41f73b7 1465 struct lttng_consumer_local_data *ctx),
3bd1e081
MD
1466 int (*recv_channel)(struct lttng_consumer_channel *channel),
1467 int (*recv_stream)(struct lttng_consumer_stream *stream),
30319bcb 1468 int (*update_stream)(uint64_t stream_key, uint32_t state))
3bd1e081 1469{
d8ef542d 1470 int ret;
3bd1e081
MD
1471 struct lttng_consumer_local_data *ctx;
1472
1473 assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN ||
1474 consumer_data.type == type);
1475 consumer_data.type = type;
1476
effcf122 1477 ctx = zmalloc(sizeof(struct lttng_consumer_local_data));
3bd1e081 1478 if (ctx == NULL) {
7a57cf92 1479 PERROR("allocating context");
3bd1e081
MD
1480 goto error;
1481 }
1482
1483 ctx->consumer_error_socket = -1;
331744e3 1484 ctx->consumer_metadata_socket = -1;
75d83e50 1485 pthread_mutex_init(&ctx->metadata_socket_lock, NULL);
3bd1e081
MD
1486 /* assign the callbacks */
1487 ctx->on_buffer_ready = buffer_ready;
1488 ctx->on_recv_channel = recv_channel;
1489 ctx->on_recv_stream = recv_stream;
1490 ctx->on_update_stream = update_stream;
1491
acdb9057
DG
1492 ctx->consumer_data_pipe = lttng_pipe_open(0);
1493 if (!ctx->consumer_data_pipe) {
3bd1e081
MD
1494 goto error_poll_pipe;
1495 }
1496
02b3d176
DG
1497 ctx->consumer_wakeup_pipe = lttng_pipe_open(0);
1498 if (!ctx->consumer_wakeup_pipe) {
1499 goto error_wakeup_pipe;
1500 }
1501
3bd1e081
MD
1502 ret = pipe(ctx->consumer_should_quit);
1503 if (ret < 0) {
7a57cf92 1504 PERROR("Error creating recv pipe");
3bd1e081
MD
1505 goto error_quit_pipe;
1506 }
1507
d8ef542d
MD
1508 ret = pipe(ctx->consumer_channel_pipe);
1509 if (ret < 0) {
1510 PERROR("Error creating channel pipe");
1511 goto error_channel_pipe;
1512 }
1513
13886d2d
DG
1514 ctx->consumer_metadata_pipe = lttng_pipe_open(0);
1515 if (!ctx->consumer_metadata_pipe) {
fb3a43a9
DG
1516 goto error_metadata_pipe;
1517 }
3bd1e081 1518
e9404c27
JG
1519 ctx->channel_monitor_pipe = -1;
1520
fb3a43a9 1521 return ctx;
3bd1e081 1522
fb3a43a9 1523error_metadata_pipe:
d8ef542d
MD
1524 utils_close_pipe(ctx->consumer_channel_pipe);
1525error_channel_pipe:
d8ef542d 1526 utils_close_pipe(ctx->consumer_should_quit);
3bd1e081 1527error_quit_pipe:
02b3d176
DG
1528 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
1529error_wakeup_pipe:
acdb9057 1530 lttng_pipe_destroy(ctx->consumer_data_pipe);
3bd1e081
MD
1531error_poll_pipe:
1532 free(ctx);
1533error:
1534 return NULL;
1535}
1536
282dadbc
MD
1537/*
1538 * Iterate over all streams of the hashtable and free them properly.
1539 */
1540static void destroy_data_stream_ht(struct lttng_ht *ht)
1541{
1542 struct lttng_ht_iter iter;
1543 struct lttng_consumer_stream *stream;
1544
1545 if (ht == NULL) {
1546 return;
1547 }
1548
1549 rcu_read_lock();
1550 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1551 /*
1552 * Ignore return value since we are currently cleaning up so any error
1553 * can't be handled.
1554 */
1555 (void) consumer_del_stream(stream, ht);
1556 }
1557 rcu_read_unlock();
1558
1559 lttng_ht_destroy(ht);
1560}
1561
1562/*
1563 * Iterate over all streams of the metadata hashtable and free them
1564 * properly.
1565 */
1566static void destroy_metadata_stream_ht(struct lttng_ht *ht)
1567{
1568 struct lttng_ht_iter iter;
1569 struct lttng_consumer_stream *stream;
1570
1571 if (ht == NULL) {
1572 return;
1573 }
1574
1575 rcu_read_lock();
1576 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1577 /*
1578 * Ignore return value since we are currently cleaning up so any error
1579 * can't be handled.
1580 */
1581 (void) consumer_del_metadata_stream(stream, ht);
1582 }
1583 rcu_read_unlock();
1584
1585 lttng_ht_destroy(ht);
1586}
1587
3bd1e081
MD
1588/*
1589 * Close all fds associated with the instance and free the context.
1590 */
1591void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx)
1592{
4c462e79
MD
1593 int ret;
1594
ab1027f4
DG
1595 DBG("Consumer destroying it. Closing everything.");
1596
4f2e75b9
DG
1597 if (!ctx) {
1598 return;
1599 }
1600
282dadbc
MD
1601 destroy_data_stream_ht(data_ht);
1602 destroy_metadata_stream_ht(metadata_ht);
1603
4c462e79
MD
1604 ret = close(ctx->consumer_error_socket);
1605 if (ret) {
1606 PERROR("close");
1607 }
331744e3
JD
1608 ret = close(ctx->consumer_metadata_socket);
1609 if (ret) {
1610 PERROR("close");
1611 }
d8ef542d 1612 utils_close_pipe(ctx->consumer_channel_pipe);
acdb9057 1613 lttng_pipe_destroy(ctx->consumer_data_pipe);
13886d2d 1614 lttng_pipe_destroy(ctx->consumer_metadata_pipe);
02b3d176 1615 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
d8ef542d 1616 utils_close_pipe(ctx->consumer_should_quit);
fb3a43a9 1617
3bd1e081
MD
1618 unlink(ctx->consumer_command_sock_path);
1619 free(ctx);
1620}
1621
6197aea7
DG
1622/*
1623 * Write the metadata stream id on the specified file descriptor.
1624 */
1625static int write_relayd_metadata_id(int fd,
1626 struct lttng_consumer_stream *stream,
239f61af 1627 unsigned long padding)
6197aea7 1628{
6cd525e8 1629 ssize_t ret;
1d4dfdef 1630 struct lttcomm_relayd_metadata_payload hdr;
6197aea7 1631
1d4dfdef
DG
1632 hdr.stream_id = htobe64(stream->relayd_stream_id);
1633 hdr.padding_size = htobe32(padding);
6cd525e8
MD
1634 ret = lttng_write(fd, (void *) &hdr, sizeof(hdr));
1635 if (ret < sizeof(hdr)) {
d7b75ec8 1636 /*
6f04ed72 1637 * This error means that the fd's end is closed so ignore the PERROR
d7b75ec8
DG
1638 * not to clubber the error output since this can happen in a normal
1639 * code path.
1640 */
1641 if (errno != EPIPE) {
1642 PERROR("write metadata stream id");
1643 }
1644 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno);
534d2592
DG
1645 /*
1646 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1647 * handle writting the missing part so report that as an error and
1648 * don't lie to the caller.
1649 */
1650 ret = -1;
6197aea7
DG
1651 goto end;
1652 }
1d4dfdef
DG
1653 DBG("Metadata stream id %" PRIu64 " with padding %lu written before data",
1654 stream->relayd_stream_id, padding);
6197aea7
DG
1655
1656end:
6cd525e8 1657 return (int) ret;
6197aea7
DG
1658}
1659
3bd1e081 1660/*
09e26845
DG
1661 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1662 * core function for writing trace buffers to either the local filesystem or
1663 * the network.
1664 *
d2956687 1665 * It must be called with the stream and the channel lock held.
79d4ffb7 1666 *
09e26845 1667 * Careful review MUST be put if any changes occur!
3bd1e081
MD
1668 *
1669 * Returns the number of bytes written
1670 */
4078b776 1671ssize_t lttng_consumer_on_read_subbuffer_mmap(
3bd1e081 1672 struct lttng_consumer_local_data *ctx,
128708c3 1673 struct lttng_consumer_stream *stream,
fd424d99 1674 const struct lttng_buffer_view *buffer,
309167d2 1675 unsigned long padding,
50adc264 1676 struct ctf_packet_index *index)
3bd1e081 1677{
994ab360 1678 ssize_t ret = 0;
f02e1e8a
DG
1679 off_t orig_offset = stream->out_fd_offset;
1680 /* Default is on the disk */
1681 int outfd = stream->out_fd;
f02e1e8a 1682 struct consumer_relayd_sock_pair *relayd = NULL;
8994307f 1683 unsigned int relayd_hang_up = 0;
fd424d99
JG
1684 const size_t subbuf_content_size = buffer->size - padding;
1685 size_t write_len;
f02e1e8a
DG
1686
1687 /* RCU lock for the relayd pointer */
1688 rcu_read_lock();
7fd975c5 1689 assert(stream->net_seq_idx != (uint64_t) -1ULL ||
948411cd 1690 stream->trace_chunk);
d2956687 1691
f02e1e8a 1692 /* Flag that the current stream if set for network streaming. */
da009f2c 1693 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1694 relayd = consumer_find_relayd(stream->net_seq_idx);
1695 if (relayd == NULL) {
56591bac 1696 ret = -EPIPE;
f02e1e8a
DG
1697 goto end;
1698 }
1699 }
1700
f02e1e8a
DG
1701 /* Handle stream on the relayd if the output is on the network */
1702 if (relayd) {
fd424d99 1703 unsigned long netlen = subbuf_content_size;
f02e1e8a
DG
1704
1705 /*
1706 * Lock the control socket for the complete duration of the function
1707 * since from this point on we will use the socket.
1708 */
1709 if (stream->metadata_flag) {
1710 /* Metadata requires the control socket. */
1711 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
93ec662e
JD
1712 if (stream->reset_metadata_flag) {
1713 ret = relayd_reset_metadata(&relayd->control_sock,
1714 stream->relayd_stream_id,
1715 stream->metadata_version);
1716 if (ret < 0) {
1717 relayd_hang_up = 1;
1718 goto write_error;
1719 }
1720 stream->reset_metadata_flag = 0;
1721 }
1d4dfdef 1722 netlen += sizeof(struct lttcomm_relayd_metadata_payload);
f02e1e8a
DG
1723 }
1724
1d4dfdef 1725 ret = write_relayd_stream_header(stream, netlen, padding, relayd);
994ab360
DG
1726 if (ret < 0) {
1727 relayd_hang_up = 1;
1728 goto write_error;
1729 }
1730 /* Use the returned socket. */
1731 outfd = ret;
f02e1e8a 1732
994ab360
DG
1733 /* Write metadata stream id before payload */
1734 if (stream->metadata_flag) {
239f61af 1735 ret = write_relayd_metadata_id(outfd, stream, padding);
994ab360 1736 if (ret < 0) {
8994307f
DG
1737 relayd_hang_up = 1;
1738 goto write_error;
1739 }
f02e1e8a 1740 }
1624d5b7 1741
fd424d99
JG
1742 write_len = subbuf_content_size;
1743 } else {
1744 /* No streaming; we have to write the full padding. */
93ec662e
JD
1745 if (stream->metadata_flag && stream->reset_metadata_flag) {
1746 ret = utils_truncate_stream_file(stream->out_fd, 0);
1747 if (ret < 0) {
1748 ERR("Reset metadata file");
1749 goto end;
1750 }
1751 stream->reset_metadata_flag = 0;
1752 }
1753
1624d5b7
JD
1754 /*
1755 * Check if we need to change the tracefile before writing the packet.
1756 */
1757 if (stream->chan->tracefile_size > 0 &&
fd424d99 1758 (stream->tracefile_size_current + buffer->size) >
1624d5b7 1759 stream->chan->tracefile_size) {
d2956687
JG
1760 ret = consumer_stream_rotate_output_files(stream);
1761 if (ret) {
1624d5b7
JD
1762 goto end;
1763 }
309167d2 1764 outfd = stream->out_fd;
a1ae300f 1765 orig_offset = 0;
1624d5b7 1766 }
fd424d99 1767 stream->tracefile_size_current += buffer->size;
309167d2
JD
1768 if (index) {
1769 index->offset = htobe64(stream->out_fd_offset);
1770 }
fd424d99
JG
1771
1772 write_len = buffer->size;
f02e1e8a
DG
1773 }
1774
d02b8372
DG
1775 /*
1776 * This call guarantee that len or less is returned. It's impossible to
1777 * receive a ret value that is bigger than len.
1778 */
fd424d99
JG
1779 ret = lttng_write(outfd, buffer->data, write_len);
1780 DBG("Consumer mmap write() ret %zd (len %lu)", ret, write_len);
1781 if (ret < 0 || ((size_t) ret != write_len)) {
d02b8372
DG
1782 /*
1783 * Report error to caller if nothing was written else at least send the
1784 * amount written.
1785 */
1786 if (ret < 0) {
994ab360 1787 ret = -errno;
f02e1e8a 1788 }
994ab360 1789 relayd_hang_up = 1;
f02e1e8a 1790
d02b8372 1791 /* Socket operation failed. We consider the relayd dead */
fcf0f774 1792 if (errno == EPIPE) {
d02b8372
DG
1793 /*
1794 * This is possible if the fd is closed on the other side
1795 * (outfd) or any write problem. It can be verbose a bit for a
1796 * normal execution if for instance the relayd is stopped
1797 * abruptly. This can happen so set this to a DBG statement.
1798 */
1799 DBG("Consumer mmap write detected relayd hang up");
994ab360
DG
1800 } else {
1801 /* Unhandled error, print it and stop function right now. */
fd424d99
JG
1802 PERROR("Error in write mmap (ret %zd != write_len %zu)", ret,
1803 write_len);
f02e1e8a 1804 }
994ab360 1805 goto write_error;
d02b8372
DG
1806 }
1807 stream->output_written += ret;
d02b8372
DG
1808
1809 /* This call is useless on a socket so better save a syscall. */
1810 if (!relayd) {
1811 /* This won't block, but will start writeout asynchronously */
fd424d99 1812 lttng_sync_file_range(outfd, stream->out_fd_offset, write_len,
d02b8372 1813 SYNC_FILE_RANGE_WRITE);
fd424d99 1814 stream->out_fd_offset += write_len;
f5dbe415 1815 lttng_consumer_sync_trace_file(stream, orig_offset);
f02e1e8a 1816 }
f02e1e8a 1817
8994307f
DG
1818write_error:
1819 /*
1820 * This is a special case that the relayd has closed its socket. Let's
1821 * cleanup the relayd object and all associated streams.
1822 */
1823 if (relayd && relayd_hang_up) {
9276e5c8
JR
1824 ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
1825 lttng_consumer_cleanup_relayd(relayd);
8994307f
DG
1826 }
1827
f02e1e8a
DG
1828end:
1829 /* Unlock only if ctrl socket used */
1830 if (relayd && stream->metadata_flag) {
1831 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1832 }
1833
1834 rcu_read_unlock();
994ab360 1835 return ret;
3bd1e081
MD
1836}
1837
1838/*
1839 * Splice the data from the ring buffer to the tracefile.
1840 *
79d4ffb7
DG
1841 * It must be called with the stream lock held.
1842 *
3bd1e081
MD
1843 * Returns the number of bytes spliced.
1844 */
4078b776 1845ssize_t lttng_consumer_on_read_subbuffer_splice(
3bd1e081 1846 struct lttng_consumer_local_data *ctx,
1d4dfdef 1847 struct lttng_consumer_stream *stream, unsigned long len,
309167d2 1848 unsigned long padding,
50adc264 1849 struct ctf_packet_index *index)
3bd1e081 1850{
f02e1e8a
DG
1851 ssize_t ret = 0, written = 0, ret_splice = 0;
1852 loff_t offset = 0;
1853 off_t orig_offset = stream->out_fd_offset;
1854 int fd = stream->wait_fd;
1855 /* Default is on the disk */
1856 int outfd = stream->out_fd;
f02e1e8a 1857 struct consumer_relayd_sock_pair *relayd = NULL;
fb3a43a9 1858 int *splice_pipe;
8994307f 1859 unsigned int relayd_hang_up = 0;
f02e1e8a 1860
3bd1e081
MD
1861 switch (consumer_data.type) {
1862 case LTTNG_CONSUMER_KERNEL:
f02e1e8a 1863 break;
7753dea8
MD
1864 case LTTNG_CONSUMER32_UST:
1865 case LTTNG_CONSUMER64_UST:
f02e1e8a 1866 /* Not supported for user space tracing */
3bd1e081
MD
1867 return -ENOSYS;
1868 default:
1869 ERR("Unknown consumer_data type");
1870 assert(0);
3bd1e081
MD
1871 }
1872
f02e1e8a
DG
1873 /* RCU lock for the relayd pointer */
1874 rcu_read_lock();
1875
1876 /* Flag that the current stream if set for network streaming. */
da009f2c 1877 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1878 relayd = consumer_find_relayd(stream->net_seq_idx);
1879 if (relayd == NULL) {
ad0b0d23 1880 written = -ret;
f02e1e8a
DG
1881 goto end;
1882 }
1883 }
a2361a61 1884 splice_pipe = stream->splice_pipe;
fb3a43a9 1885
f02e1e8a 1886 /* Write metadata stream id before payload */
1d4dfdef 1887 if (relayd) {
ad0b0d23 1888 unsigned long total_len = len;
f02e1e8a 1889
1d4dfdef
DG
1890 if (stream->metadata_flag) {
1891 /*
1892 * Lock the control socket for the complete duration of the function
1893 * since from this point on we will use the socket.
1894 */
1895 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1896
93ec662e
JD
1897 if (stream->reset_metadata_flag) {
1898 ret = relayd_reset_metadata(&relayd->control_sock,
1899 stream->relayd_stream_id,
1900 stream->metadata_version);
1901 if (ret < 0) {
1902 relayd_hang_up = 1;
1903 goto write_error;
1904 }
1905 stream->reset_metadata_flag = 0;
1906 }
239f61af 1907 ret = write_relayd_metadata_id(splice_pipe[1], stream,
1d4dfdef
DG
1908 padding);
1909 if (ret < 0) {
1910 written = ret;
ad0b0d23
DG
1911 relayd_hang_up = 1;
1912 goto write_error;
1d4dfdef
DG
1913 }
1914
1915 total_len += sizeof(struct lttcomm_relayd_metadata_payload);
1916 }
1917
1918 ret = write_relayd_stream_header(stream, total_len, padding, relayd);
ad0b0d23
DG
1919 if (ret < 0) {
1920 written = ret;
1921 relayd_hang_up = 1;
1922 goto write_error;
f02e1e8a 1923 }
ad0b0d23
DG
1924 /* Use the returned socket. */
1925 outfd = ret;
1d4dfdef
DG
1926 } else {
1927 /* No streaming, we have to set the len with the full padding */
1928 len += padding;
1624d5b7 1929
93ec662e
JD
1930 if (stream->metadata_flag && stream->reset_metadata_flag) {
1931 ret = utils_truncate_stream_file(stream->out_fd, 0);
1932 if (ret < 0) {
1933 ERR("Reset metadata file");
1934 goto end;
1935 }
1936 stream->reset_metadata_flag = 0;
1937 }
1624d5b7
JD
1938 /*
1939 * Check if we need to change the tracefile before writing the packet.
1940 */
1941 if (stream->chan->tracefile_size > 0 &&
1942 (stream->tracefile_size_current + len) >
1943 stream->chan->tracefile_size) {
d2956687 1944 ret = consumer_stream_rotate_output_files(stream);
1624d5b7 1945 if (ret < 0) {
ad0b0d23 1946 written = ret;
1624d5b7
JD
1947 goto end;
1948 }
309167d2 1949 outfd = stream->out_fd;
a1ae300f 1950 orig_offset = 0;
1624d5b7
JD
1951 }
1952 stream->tracefile_size_current += len;
309167d2 1953 index->offset = htobe64(stream->out_fd_offset);
f02e1e8a
DG
1954 }
1955
1956 while (len > 0) {
1d4dfdef
DG
1957 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1958 (unsigned long)offset, len, fd, splice_pipe[1]);
fb3a43a9 1959 ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len,
f02e1e8a
DG
1960 SPLICE_F_MOVE | SPLICE_F_MORE);
1961 DBG("splice chan to pipe, ret %zd", ret_splice);
1962 if (ret_splice < 0) {
d02b8372 1963 ret = errno;
ad0b0d23 1964 written = -ret;
d02b8372 1965 PERROR("Error in relay splice");
f02e1e8a
DG
1966 goto splice_error;
1967 }
1968
1969 /* Handle stream on the relayd if the output is on the network */
ad0b0d23
DG
1970 if (relayd && stream->metadata_flag) {
1971 size_t metadata_payload_size =
1972 sizeof(struct lttcomm_relayd_metadata_payload);
1973
1974 /* Update counter to fit the spliced data */
1975 ret_splice += metadata_payload_size;
1976 len += metadata_payload_size;
1977 /*
1978 * We do this so the return value can match the len passed as
1979 * argument to this function.
1980 */
1981 written -= metadata_payload_size;
f02e1e8a
DG
1982 }
1983
1984 /* Splice data out */
fb3a43a9 1985 ret_splice = splice(splice_pipe[0], NULL, outfd, NULL,
f02e1e8a 1986 ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE);
a2361a61
JD
1987 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1988 outfd, ret_splice);
f02e1e8a 1989 if (ret_splice < 0) {
d02b8372 1990 ret = errno;
ad0b0d23
DG
1991 written = -ret;
1992 relayd_hang_up = 1;
1993 goto write_error;
f02e1e8a 1994 } else if (ret_splice > len) {
d02b8372
DG
1995 /*
1996 * We don't expect this code path to be executed but you never know
1997 * so this is an extra protection agains a buggy splice().
1998 */
f02e1e8a 1999 ret = errno;
ad0b0d23 2000 written += ret_splice;
d02b8372
DG
2001 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice,
2002 len);
f02e1e8a 2003 goto splice_error;
d02b8372
DG
2004 } else {
2005 /* All good, update current len and continue. */
2006 len -= ret_splice;
f02e1e8a 2007 }
f02e1e8a
DG
2008
2009 /* This call is useless on a socket so better save a syscall. */
2010 if (!relayd) {
2011 /* This won't block, but will start writeout asynchronously */
2012 lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice,
2013 SYNC_FILE_RANGE_WRITE);
2014 stream->out_fd_offset += ret_splice;
2015 }
e5d1a9b3 2016 stream->output_written += ret_splice;
f02e1e8a
DG
2017 written += ret_splice;
2018 }
f5dbe415
JG
2019 if (!relayd) {
2020 lttng_consumer_sync_trace_file(stream, orig_offset);
2021 }
f02e1e8a
DG
2022 goto end;
2023
8994307f
DG
2024write_error:
2025 /*
2026 * This is a special case that the relayd has closed its socket. Let's
2027 * cleanup the relayd object and all associated streams.
2028 */
2029 if (relayd && relayd_hang_up) {
9276e5c8
JR
2030 ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
2031 lttng_consumer_cleanup_relayd(relayd);
8994307f
DG
2032 /* Skip splice error so the consumer does not fail */
2033 goto end;
2034 }
2035
f02e1e8a
DG
2036splice_error:
2037 /* send the appropriate error description to sessiond */
2038 switch (ret) {
f02e1e8a 2039 case EINVAL:
f73fabfd 2040 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL);
f02e1e8a
DG
2041 break;
2042 case ENOMEM:
f73fabfd 2043 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM);
f02e1e8a
DG
2044 break;
2045 case ESPIPE:
f73fabfd 2046 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE);
f02e1e8a
DG
2047 break;
2048 }
2049
2050end:
2051 if (relayd && stream->metadata_flag) {
2052 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
2053 }
2054
2055 rcu_read_unlock();
2056 return written;
3bd1e081
MD
2057}
2058
15055ce5
JD
2059/*
2060 * Sample the snapshot positions for a specific fd
2061 *
2062 * Returns 0 on success, < 0 on error
2063 */
2064int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream *stream)
2065{
2066 switch (consumer_data.type) {
2067 case LTTNG_CONSUMER_KERNEL:
2068 return lttng_kconsumer_sample_snapshot_positions(stream);
2069 case LTTNG_CONSUMER32_UST:
2070 case LTTNG_CONSUMER64_UST:
2071 return lttng_ustconsumer_sample_snapshot_positions(stream);
2072 default:
2073 ERR("Unknown consumer_data type");
2074 assert(0);
2075 return -ENOSYS;
2076 }
2077}
3bd1e081
MD
2078/*
2079 * Take a snapshot for a specific fd
2080 *
2081 * Returns 0 on success, < 0 on error
2082 */
ffe60014 2083int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream)
3bd1e081
MD
2084{
2085 switch (consumer_data.type) {
2086 case LTTNG_CONSUMER_KERNEL:
ffe60014 2087 return lttng_kconsumer_take_snapshot(stream);
7753dea8
MD
2088 case LTTNG_CONSUMER32_UST:
2089 case LTTNG_CONSUMER64_UST:
ffe60014 2090 return lttng_ustconsumer_take_snapshot(stream);
3bd1e081
MD
2091 default:
2092 ERR("Unknown consumer_data type");
2093 assert(0);
2094 return -ENOSYS;
2095 }
3bd1e081
MD
2096}
2097
2098/*
2099 * Get the produced position
2100 *
2101 * Returns 0 on success, < 0 on error
2102 */
ffe60014 2103int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
3bd1e081
MD
2104 unsigned long *pos)
2105{
2106 switch (consumer_data.type) {
2107 case LTTNG_CONSUMER_KERNEL:
ffe60014 2108 return lttng_kconsumer_get_produced_snapshot(stream, pos);
7753dea8
MD
2109 case LTTNG_CONSUMER32_UST:
2110 case LTTNG_CONSUMER64_UST:
ffe60014 2111 return lttng_ustconsumer_get_produced_snapshot(stream, pos);
3bd1e081
MD
2112 default:
2113 ERR("Unknown consumer_data type");
2114 assert(0);
2115 return -ENOSYS;
2116 }
2117}
2118
15055ce5
JD
2119/*
2120 * Get the consumed position (free-running counter position in bytes).
2121 *
2122 * Returns 0 on success, < 0 on error
2123 */
2124int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
2125 unsigned long *pos)
2126{
2127 switch (consumer_data.type) {
2128 case LTTNG_CONSUMER_KERNEL:
2129 return lttng_kconsumer_get_consumed_snapshot(stream, pos);
2130 case LTTNG_CONSUMER32_UST:
2131 case LTTNG_CONSUMER64_UST:
2132 return lttng_ustconsumer_get_consumed_snapshot(stream, pos);
2133 default:
2134 ERR("Unknown consumer_data type");
2135 assert(0);
2136 return -ENOSYS;
2137 }
2138}
2139
3bd1e081
MD
2140int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx,
2141 int sock, struct pollfd *consumer_sockpoll)
2142{
2143 switch (consumer_data.type) {
2144 case LTTNG_CONSUMER_KERNEL:
2145 return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
7753dea8
MD
2146 case LTTNG_CONSUMER32_UST:
2147 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
2148 return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
2149 default:
2150 ERR("Unknown consumer_data type");
2151 assert(0);
2152 return -ENOSYS;
2153 }
2154}
2155
1f8d1c14 2156static
6d574024 2157void lttng_consumer_close_all_metadata(void)
d88aee68
DG
2158{
2159 switch (consumer_data.type) {
2160 case LTTNG_CONSUMER_KERNEL:
2161 /*
2162 * The Kernel consumer has a different metadata scheme so we don't
2163 * close anything because the stream will be closed by the session
2164 * daemon.
2165 */
2166 break;
2167 case LTTNG_CONSUMER32_UST:
2168 case LTTNG_CONSUMER64_UST:
2169 /*
2170 * Close all metadata streams. The metadata hash table is passed and
2171 * this call iterates over it by closing all wakeup fd. This is safe
2172 * because at this point we are sure that the metadata producer is
2173 * either dead or blocked.
2174 */
6d574024 2175 lttng_ustconsumer_close_all_metadata(metadata_ht);
d88aee68
DG
2176 break;
2177 default:
2178 ERR("Unknown consumer_data type");
2179 assert(0);
2180 }
2181}
2182
fb3a43a9
DG
2183/*
2184 * Clean up a metadata stream and free its memory.
2185 */
e316aad5
DG
2186void consumer_del_metadata_stream(struct lttng_consumer_stream *stream,
2187 struct lttng_ht *ht)
fb3a43a9 2188{
a6ef8ee6
JG
2189 struct lttng_consumer_channel *channel = NULL;
2190 bool free_channel = false;
fb3a43a9
DG
2191
2192 assert(stream);
2193 /*
2194 * This call should NEVER receive regular stream. It must always be
2195 * metadata stream and this is crucial for data structure synchronization.
2196 */
2197 assert(stream->metadata_flag);
2198
e316aad5
DG
2199 DBG3("Consumer delete metadata stream %d", stream->wait_fd);
2200
74251bb8 2201 pthread_mutex_lock(&consumer_data.lock);
a6ef8ee6
JG
2202 /*
2203 * Note that this assumes that a stream's channel is never changed and
2204 * that the stream's lock doesn't need to be taken to sample its
2205 * channel.
2206 */
2207 channel = stream->chan;
2208 pthread_mutex_lock(&channel->lock);
3dad2c0f 2209 pthread_mutex_lock(&stream->lock);
a6ef8ee6 2210 if (channel->metadata_cache) {
081424af 2211 /* Only applicable to userspace consumers. */
a6ef8ee6 2212 pthread_mutex_lock(&channel->metadata_cache->lock);
081424af 2213 }
8994307f 2214
6d574024
DG
2215 /* Remove any reference to that stream. */
2216 consumer_stream_delete(stream, ht);
ca22feea 2217
6d574024
DG
2218 /* Close down everything including the relayd if one. */
2219 consumer_stream_close(stream);
2220 /* Destroy tracer buffers of the stream. */
2221 consumer_stream_destroy_buffers(stream);
fb3a43a9
DG
2222
2223 /* Atomically decrement channel refcount since other threads can use it. */
a6ef8ee6
JG
2224 if (!uatomic_sub_return(&channel->refcount, 1)
2225 && !uatomic_read(&channel->nb_init_stream_left)) {
c30aaa51 2226 /* Go for channel deletion! */
a6ef8ee6 2227 free_channel = true;
fb3a43a9 2228 }
a6ef8ee6 2229 stream->chan = NULL;
fb3a43a9 2230
73811ecc
DG
2231 /*
2232 * Nullify the stream reference so it is not used after deletion. The
6d574024
DG
2233 * channel lock MUST be acquired before being able to check for a NULL
2234 * pointer value.
73811ecc 2235 */
a6ef8ee6 2236 channel->metadata_stream = NULL;
73811ecc 2237
a6ef8ee6
JG
2238 if (channel->metadata_cache) {
2239 pthread_mutex_unlock(&channel->metadata_cache->lock);
081424af 2240 }
3dad2c0f 2241 pthread_mutex_unlock(&stream->lock);
a6ef8ee6 2242 pthread_mutex_unlock(&channel->lock);
74251bb8 2243 pthread_mutex_unlock(&consumer_data.lock);
e316aad5 2244
a6ef8ee6
JG
2245 if (free_channel) {
2246 consumer_del_channel(channel);
e316aad5
DG
2247 }
2248
d2956687
JG
2249 lttng_trace_chunk_put(stream->trace_chunk);
2250 stream->trace_chunk = NULL;
6d574024 2251 consumer_stream_free(stream);
fb3a43a9
DG
2252}
2253
2254/*
2255 * Action done with the metadata stream when adding it to the consumer internal
2256 * data structures to handle it.
2257 */
66d583dc 2258void consumer_add_metadata_stream(struct lttng_consumer_stream *stream)
fb3a43a9 2259{
5ab66908 2260 struct lttng_ht *ht = metadata_ht;
76082088 2261 struct lttng_ht_iter iter;
d88aee68 2262 struct lttng_ht_node_u64 *node;
fb3a43a9 2263
e316aad5
DG
2264 assert(stream);
2265 assert(ht);
2266
d88aee68 2267 DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key);
e316aad5
DG
2268
2269 pthread_mutex_lock(&consumer_data.lock);
a9838785 2270 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 2271 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 2272 pthread_mutex_lock(&stream->lock);
e316aad5 2273
e316aad5
DG
2274 /*
2275 * From here, refcounts are updated so be _careful_ when returning an error
2276 * after this point.
2277 */
2278
fb3a43a9 2279 rcu_read_lock();
76082088
DG
2280
2281 /*
2282 * Lookup the stream just to make sure it does not exist in our internal
2283 * state. This should NEVER happen.
2284 */
d88aee68
DG
2285 lttng_ht_lookup(ht, &stream->key, &iter);
2286 node = lttng_ht_iter_get_node_u64(&iter);
76082088
DG
2287 assert(!node);
2288
e316aad5 2289 /*
ffe60014
DG
2290 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2291 * in terms of destroying the associated channel, because the action that
e316aad5
DG
2292 * causes the count to become 0 also causes a stream to be added. The
2293 * channel deletion will thus be triggered by the following removal of this
2294 * stream.
2295 */
ffe60014 2296 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
2297 /* Increment refcount before decrementing nb_init_stream_left */
2298 cmm_smp_wmb();
ffe60014 2299 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
2300 }
2301
d88aee68 2302 lttng_ht_add_unique_u64(ht, &stream->node);
ca22feea 2303
446156b4 2304 lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht,
d8ef542d
MD
2305 &stream->node_channel_id);
2306
ca22feea
DG
2307 /*
2308 * Add stream to the stream_list_ht of the consumer data. No need to steal
2309 * the key since the HT does not use it and we allow to add redundant keys
2310 * into this table.
2311 */
d88aee68 2312 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 2313
fb3a43a9 2314 rcu_read_unlock();
e316aad5 2315
2e818a6a 2316 pthread_mutex_unlock(&stream->lock);
a9838785 2317 pthread_mutex_unlock(&stream->chan->lock);
ec6ea7d0 2318 pthread_mutex_unlock(&stream->chan->timer_lock);
e316aad5 2319 pthread_mutex_unlock(&consumer_data.lock);
fb3a43a9
DG
2320}
2321
8994307f
DG
2322/*
2323 * Delete data stream that are flagged for deletion (endpoint_status).
2324 */
2325static void validate_endpoint_status_data_stream(void)
2326{
2327 struct lttng_ht_iter iter;
2328 struct lttng_consumer_stream *stream;
2329
2330 DBG("Consumer delete flagged data stream");
2331
2332 rcu_read_lock();
2333 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
2334 /* Validate delete flag of the stream */
79d4ffb7 2335 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2336 continue;
2337 }
2338 /* Delete it right now */
2339 consumer_del_stream(stream, data_ht);
2340 }
2341 rcu_read_unlock();
2342}
2343
2344/*
2345 * Delete metadata stream that are flagged for deletion (endpoint_status).
2346 */
2347static void validate_endpoint_status_metadata_stream(
2348 struct lttng_poll_event *pollset)
2349{
2350 struct lttng_ht_iter iter;
2351 struct lttng_consumer_stream *stream;
2352
2353 DBG("Consumer delete flagged metadata stream");
2354
2355 assert(pollset);
2356
2357 rcu_read_lock();
2358 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
2359 /* Validate delete flag of the stream */
79d4ffb7 2360 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2361 continue;
2362 }
2363 /*
2364 * Remove from pollset so the metadata thread can continue without
2365 * blocking on a deleted stream.
2366 */
2367 lttng_poll_del(pollset, stream->wait_fd);
2368
2369 /* Delete it right now */
2370 consumer_del_metadata_stream(stream, metadata_ht);
2371 }
2372 rcu_read_unlock();
2373}
2374
fb3a43a9
DG
2375/*
2376 * Thread polls on metadata file descriptor and write them on disk or on the
2377 * network.
2378 */
7d980def 2379void *consumer_thread_metadata_poll(void *data)
fb3a43a9 2380{
1fc79fb4 2381 int ret, i, pollfd, err = -1;
fb3a43a9 2382 uint32_t revents, nb_fd;
e316aad5 2383 struct lttng_consumer_stream *stream = NULL;
fb3a43a9 2384 struct lttng_ht_iter iter;
d88aee68 2385 struct lttng_ht_node_u64 *node;
fb3a43a9
DG
2386 struct lttng_poll_event events;
2387 struct lttng_consumer_local_data *ctx = data;
2388 ssize_t len;
2389
2390 rcu_register_thread();
2391
1fc79fb4
MD
2392 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA);
2393
2d57de81
MD
2394 if (testpoint(consumerd_thread_metadata)) {
2395 goto error_testpoint;
2396 }
2397
9ce5646a
MD
2398 health_code_update();
2399
fb3a43a9
DG
2400 DBG("Thread metadata poll started");
2401
fb3a43a9
DG
2402 /* Size is set to 1 for the consumer_metadata pipe */
2403 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2404 if (ret < 0) {
2405 ERR("Poll set creation failed");
d8ef542d 2406 goto end_poll;
fb3a43a9
DG
2407 }
2408
13886d2d
DG
2409 ret = lttng_poll_add(&events,
2410 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN);
fb3a43a9
DG
2411 if (ret < 0) {
2412 goto end;
2413 }
2414
2415 /* Main loop */
2416 DBG("Metadata main loop started");
2417
2418 while (1) {
fb3a43a9 2419restart:
7fa2082e 2420 health_code_update();
9ce5646a 2421 health_poll_entry();
7fa2082e 2422 DBG("Metadata poll wait");
fb3a43a9 2423 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2424 DBG("Metadata poll return from wait with %d fd(s)",
2425 LTTNG_POLL_GETNB(&events));
9ce5646a 2426 health_poll_exit();
40063ead 2427 DBG("Metadata event caught in thread");
fb3a43a9
DG
2428 if (ret < 0) {
2429 if (errno == EINTR) {
40063ead 2430 ERR("Poll EINTR caught");
fb3a43a9
DG
2431 goto restart;
2432 }
d9607cd7
MD
2433 if (LTTNG_POLL_GETNB(&events) == 0) {
2434 err = 0; /* All is OK */
2435 }
2436 goto end;
fb3a43a9
DG
2437 }
2438
0d9c5d77
DG
2439 nb_fd = ret;
2440
e316aad5 2441 /* From here, the event is a metadata wait fd */
fb3a43a9 2442 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2443 health_code_update();
2444
fb3a43a9
DG
2445 revents = LTTNG_POLL_GETEV(&events, i);
2446 pollfd = LTTNG_POLL_GETFD(&events, i);
2447
13886d2d 2448 if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) {
03e43155 2449 if (revents & LPOLLIN) {
13886d2d
DG
2450 ssize_t pipe_len;
2451
2452 pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe,
2453 &stream, sizeof(stream));
6cd525e8 2454 if (pipe_len < sizeof(stream)) {
03e43155
MD
2455 if (pipe_len < 0) {
2456 PERROR("read metadata stream");
2457 }
fb3a43a9 2458 /*
03e43155
MD
2459 * Remove the pipe from the poll set and continue the loop
2460 * since their might be data to consume.
fb3a43a9 2461 */
03e43155
MD
2462 lttng_poll_del(&events,
2463 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2464 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
fb3a43a9
DG
2465 continue;
2466 }
2467
8994307f
DG
2468 /* A NULL stream means that the state has changed. */
2469 if (stream == NULL) {
2470 /* Check for deleted streams. */
2471 validate_endpoint_status_metadata_stream(&events);
3714380f 2472 goto restart;
8994307f
DG
2473 }
2474
fb3a43a9
DG
2475 DBG("Adding metadata stream %d to poll set",
2476 stream->wait_fd);
2477
fb3a43a9
DG
2478 /* Add metadata stream to the global poll events list */
2479 lttng_poll_add(&events, stream->wait_fd,
6d574024 2480 LPOLLIN | LPOLLPRI | LPOLLHUP);
03e43155
MD
2481 } else if (revents & (LPOLLERR | LPOLLHUP)) {
2482 DBG("Metadata thread pipe hung up");
2483 /*
2484 * Remove the pipe from the poll set and continue the loop
2485 * since their might be data to consume.
2486 */
2487 lttng_poll_del(&events,
2488 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2489 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
2490 continue;
2491 } else {
2492 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2493 goto end;
fb3a43a9
DG
2494 }
2495
e316aad5 2496 /* Handle other stream */
fb3a43a9
DG
2497 continue;
2498 }
2499
d09e1200 2500 rcu_read_lock();
d88aee68
DG
2501 {
2502 uint64_t tmp_id = (uint64_t) pollfd;
2503
2504 lttng_ht_lookup(metadata_ht, &tmp_id, &iter);
2505 }
2506 node = lttng_ht_iter_get_node_u64(&iter);
e316aad5 2507 assert(node);
fb3a43a9
DG
2508
2509 stream = caa_container_of(node, struct lttng_consumer_stream,
58b1f425 2510 node);
fb3a43a9 2511
03e43155
MD
2512 if (revents & (LPOLLIN | LPOLLPRI)) {
2513 /* Get the data out of the metadata file descriptor */
2514 DBG("Metadata available on fd %d", pollfd);
2515 assert(stream->wait_fd == pollfd);
2516
2517 do {
2518 health_code_update();
2519
2520 len = ctx->on_buffer_ready(stream, ctx);
2521 /*
2522 * We don't check the return value here since if we get
83f4233d 2523 * a negative len, it means an error occurred thus we
03e43155
MD
2524 * simply remove it from the poll set and free the
2525 * stream.
2526 */
2527 } while (len > 0);
2528
2529 /* It's ok to have an unavailable sub-buffer */
2530 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
2531 /* Clean up stream from consumer and free it. */
2532 lttng_poll_del(&events, stream->wait_fd);
2533 consumer_del_metadata_stream(stream, metadata_ht);
2534 }
2535 } else if (revents & (LPOLLERR | LPOLLHUP)) {
e316aad5 2536 DBG("Metadata fd %d is hup|err.", pollfd);
fb3a43a9
DG
2537 if (!stream->hangup_flush_done
2538 && (consumer_data.type == LTTNG_CONSUMER32_UST
2539 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2540 DBG("Attempting to flush and consume the UST buffers");
2541 lttng_ustconsumer_on_stream_hangup(stream);
2542
2543 /* We just flushed the stream now read it. */
4bb94b75 2544 do {
9ce5646a
MD
2545 health_code_update();
2546
4bb94b75
DG
2547 len = ctx->on_buffer_ready(stream, ctx);
2548 /*
2549 * We don't check the return value here since if we get
83f4233d 2550 * a negative len, it means an error occurred thus we
4bb94b75
DG
2551 * simply remove it from the poll set and free the
2552 * stream.
2553 */
2554 } while (len > 0);
fb3a43a9
DG
2555 }
2556
fb3a43a9 2557 lttng_poll_del(&events, stream->wait_fd);
e316aad5
DG
2558 /*
2559 * This call update the channel states, closes file descriptors
2560 * and securely free the stream.
2561 */
2562 consumer_del_metadata_stream(stream, metadata_ht);
03e43155
MD
2563 } else {
2564 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
6f2f1a70 2565 rcu_read_unlock();
03e43155 2566 goto end;
fb3a43a9 2567 }
e316aad5 2568 /* Release RCU lock for the stream looked up */
d09e1200 2569 rcu_read_unlock();
fb3a43a9
DG
2570 }
2571 }
2572
1fc79fb4
MD
2573 /* All is OK */
2574 err = 0;
fb3a43a9
DG
2575end:
2576 DBG("Metadata poll thread exiting");
fb3a43a9 2577
d8ef542d
MD
2578 lttng_poll_clean(&events);
2579end_poll:
2d57de81 2580error_testpoint:
1fc79fb4
MD
2581 if (err) {
2582 health_error();
2583 ERR("Health error occurred in %s", __func__);
2584 }
2585 health_unregister(health_consumerd);
fb3a43a9
DG
2586 rcu_unregister_thread();
2587 return NULL;
2588}
2589
3bd1e081 2590/*
e4421fec 2591 * This thread polls the fds in the set to consume the data and write
3bd1e081
MD
2592 * it to tracefile if necessary.
2593 */
7d980def 2594void *consumer_thread_data_poll(void *data)
3bd1e081 2595{
1fc79fb4 2596 int num_rdy, num_hup, high_prio, ret, i, err = -1;
3bd1e081
MD
2597 struct pollfd *pollfd = NULL;
2598 /* local view of the streams */
c869f647 2599 struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL;
3bd1e081 2600 /* local view of consumer_data.fds_count */
8bdcc002
JG
2601 int nb_fd = 0;
2602 /* 2 for the consumer_data_pipe and wake up pipe */
2603 const int nb_pipes_fd = 2;
9a2fcf78
JD
2604 /* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */
2605 int nb_inactive_fd = 0;
3bd1e081 2606 struct lttng_consumer_local_data *ctx = data;
00e2e675 2607 ssize_t len;
3bd1e081 2608
e7b994a3
DG
2609 rcu_register_thread();
2610
1fc79fb4
MD
2611 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA);
2612
2d57de81
MD
2613 if (testpoint(consumerd_thread_data)) {
2614 goto error_testpoint;
2615 }
2616
9ce5646a
MD
2617 health_code_update();
2618
4df6c8cb
MD
2619 local_stream = zmalloc(sizeof(struct lttng_consumer_stream *));
2620 if (local_stream == NULL) {
2621 PERROR("local_stream malloc");
2622 goto end;
2623 }
3bd1e081
MD
2624
2625 while (1) {
9ce5646a
MD
2626 health_code_update();
2627
3bd1e081
MD
2628 high_prio = 0;
2629 num_hup = 0;
2630
2631 /*
e4421fec 2632 * the fds set has been updated, we need to update our
3bd1e081
MD
2633 * local array as well
2634 */
2635 pthread_mutex_lock(&consumer_data.lock);
2636 if (consumer_data.need_update) {
0e428499
DG
2637 free(pollfd);
2638 pollfd = NULL;
2639
2640 free(local_stream);
2641 local_stream = NULL;
3bd1e081 2642
8bdcc002 2643 /* Allocate for all fds */
261de637 2644 pollfd = zmalloc((consumer_data.stream_count + nb_pipes_fd) * sizeof(struct pollfd));
3bd1e081 2645 if (pollfd == NULL) {
7a57cf92 2646 PERROR("pollfd malloc");
3bd1e081
MD
2647 pthread_mutex_unlock(&consumer_data.lock);
2648 goto end;
2649 }
2650
261de637 2651 local_stream = zmalloc((consumer_data.stream_count + nb_pipes_fd) *
747f8642 2652 sizeof(struct lttng_consumer_stream *));
3bd1e081 2653 if (local_stream == NULL) {
7a57cf92 2654 PERROR("local_stream malloc");
3bd1e081
MD
2655 pthread_mutex_unlock(&consumer_data.lock);
2656 goto end;
2657 }
ffe60014 2658 ret = update_poll_array(ctx, &pollfd, local_stream,
9a2fcf78 2659 data_ht, &nb_inactive_fd);
3bd1e081
MD
2660 if (ret < 0) {
2661 ERR("Error in allocating pollfd or local_outfds");
f73fabfd 2662 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2663 pthread_mutex_unlock(&consumer_data.lock);
2664 goto end;
2665 }
2666 nb_fd = ret;
2667 consumer_data.need_update = 0;
2668 }
2669 pthread_mutex_unlock(&consumer_data.lock);
2670
4078b776 2671 /* No FDs and consumer_quit, consumer_cleanup the thread */
9a2fcf78
JD
2672 if (nb_fd == 0 && nb_inactive_fd == 0 &&
2673 CMM_LOAD_SHARED(consumer_quit) == 1) {
1fc79fb4 2674 err = 0; /* All is OK */
4078b776
MD
2675 goto end;
2676 }
3bd1e081 2677 /* poll on the array of fds */
88f2b785 2678 restart:
261de637 2679 DBG("polling on %d fd", nb_fd + nb_pipes_fd);
cf0bcb51
JG
2680 if (testpoint(consumerd_thread_data_poll)) {
2681 goto end;
2682 }
9ce5646a 2683 health_poll_entry();
261de637 2684 num_rdy = poll(pollfd, nb_fd + nb_pipes_fd, -1);
9ce5646a 2685 health_poll_exit();
3bd1e081
MD
2686 DBG("poll num_rdy : %d", num_rdy);
2687 if (num_rdy == -1) {
88f2b785
MD
2688 /*
2689 * Restart interrupted system call.
2690 */
2691 if (errno == EINTR) {
2692 goto restart;
2693 }
7a57cf92 2694 PERROR("Poll error");
f73fabfd 2695 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2696 goto end;
2697 } else if (num_rdy == 0) {
2698 DBG("Polling thread timed out");
2699 goto end;
2700 }
2701
80957876
JG
2702 if (caa_unlikely(data_consumption_paused)) {
2703 DBG("Data consumption paused, sleeping...");
2704 sleep(1);
2705 goto restart;
2706 }
2707
3bd1e081 2708 /*
50f8ae69 2709 * If the consumer_data_pipe triggered poll go directly to the
00e2e675
DG
2710 * beginning of the loop to update the array. We want to prioritize
2711 * array update over low-priority reads.
3bd1e081 2712 */
509bb1cf 2713 if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) {
ab30f567 2714 ssize_t pipe_readlen;
04fdd819 2715
50f8ae69 2716 DBG("consumer_data_pipe wake up");
acdb9057
DG
2717 pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe,
2718 &new_stream, sizeof(new_stream));
6cd525e8
MD
2719 if (pipe_readlen < sizeof(new_stream)) {
2720 PERROR("Consumer data pipe");
23f5f35d
DG
2721 /* Continue so we can at least handle the current stream(s). */
2722 continue;
2723 }
c869f647
DG
2724
2725 /*
2726 * If the stream is NULL, just ignore it. It's also possible that
2727 * the sessiond poll thread changed the consumer_quit state and is
2728 * waking us up to test it.
2729 */
2730 if (new_stream == NULL) {
8994307f 2731 validate_endpoint_status_data_stream();
c869f647
DG
2732 continue;
2733 }
2734
c869f647 2735 /* Continue to update the local streams and handle prio ones */
3bd1e081
MD
2736 continue;
2737 }
2738
02b3d176
DG
2739 /* Handle wakeup pipe. */
2740 if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) {
2741 char dummy;
2742 ssize_t pipe_readlen;
2743
2744 pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy,
2745 sizeof(dummy));
2746 if (pipe_readlen < 0) {
2747 PERROR("Consumer data wakeup pipe");
2748 }
2749 /* We've been awakened to handle stream(s). */
2750 ctx->has_wakeup = 0;
2751 }
2752
3bd1e081
MD
2753 /* Take care of high priority channels first. */
2754 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2755 health_code_update();
2756
9617607b
DG
2757 if (local_stream[i] == NULL) {
2758 continue;
2759 }
fb3a43a9 2760 if (pollfd[i].revents & POLLPRI) {
d41f73b7
MD
2761 DBG("Urgent read on fd %d", pollfd[i].fd);
2762 high_prio = 1;
4078b776 2763 len = ctx->on_buffer_ready(local_stream[i], ctx);
d41f73b7 2764 /* it's ok to have an unavailable sub-buffer */
b64403e3 2765 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2766 /* Clean the stream and free it. */
2767 consumer_del_stream(local_stream[i], data_ht);
9617607b 2768 local_stream[i] = NULL;
4078b776
MD
2769 } else if (len > 0) {
2770 local_stream[i]->data_read = 1;
d41f73b7 2771 }
3bd1e081
MD
2772 }
2773 }
2774
4078b776
MD
2775 /*
2776 * If we read high prio channel in this loop, try again
2777 * for more high prio data.
2778 */
2779 if (high_prio) {
3bd1e081
MD
2780 continue;
2781 }
2782
2783 /* Take care of low priority channels. */
4078b776 2784 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2785 health_code_update();
2786
9617607b
DG
2787 if (local_stream[i] == NULL) {
2788 continue;
2789 }
4078b776 2790 if ((pollfd[i].revents & POLLIN) ||
02b3d176
DG
2791 local_stream[i]->hangup_flush_done ||
2792 local_stream[i]->has_data) {
4078b776
MD
2793 DBG("Normal read on fd %d", pollfd[i].fd);
2794 len = ctx->on_buffer_ready(local_stream[i], ctx);
2795 /* it's ok to have an unavailable sub-buffer */
b64403e3 2796 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2797 /* Clean the stream and free it. */
2798 consumer_del_stream(local_stream[i], data_ht);
9617607b 2799 local_stream[i] = NULL;
4078b776
MD
2800 } else if (len > 0) {
2801 local_stream[i]->data_read = 1;
2802 }
2803 }
2804 }
2805
2806 /* Handle hangup and errors */
2807 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2808 health_code_update();
2809
9617607b
DG
2810 if (local_stream[i] == NULL) {
2811 continue;
2812 }
4078b776
MD
2813 if (!local_stream[i]->hangup_flush_done
2814 && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL))
2815 && (consumer_data.type == LTTNG_CONSUMER32_UST
2816 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2817 DBG("fd %d is hup|err|nval. Attempting flush and read.",
9617607b 2818 pollfd[i].fd);
4078b776
MD
2819 lttng_ustconsumer_on_stream_hangup(local_stream[i]);
2820 /* Attempt read again, for the data we just flushed. */
2821 local_stream[i]->data_read = 1;
2822 }
2823 /*
2824 * If the poll flag is HUP/ERR/NVAL and we have
2825 * read no data in this pass, we can remove the
2826 * stream from its hash table.
2827 */
2828 if ((pollfd[i].revents & POLLHUP)) {
2829 DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
2830 if (!local_stream[i]->data_read) {
43c34bc3 2831 consumer_del_stream(local_stream[i], data_ht);
9617607b 2832 local_stream[i] = NULL;
4078b776
MD
2833 num_hup++;
2834 }
2835 } else if (pollfd[i].revents & POLLERR) {
2836 ERR("Error returned in polling fd %d.", pollfd[i].fd);
2837 if (!local_stream[i]->data_read) {
43c34bc3 2838 consumer_del_stream(local_stream[i], data_ht);
9617607b 2839 local_stream[i] = NULL;
4078b776
MD
2840 num_hup++;
2841 }
2842 } else if (pollfd[i].revents & POLLNVAL) {
2843 ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
2844 if (!local_stream[i]->data_read) {
43c34bc3 2845 consumer_del_stream(local_stream[i], data_ht);
9617607b 2846 local_stream[i] = NULL;
4078b776 2847 num_hup++;
3bd1e081
MD
2848 }
2849 }
9617607b
DG
2850 if (local_stream[i] != NULL) {
2851 local_stream[i]->data_read = 0;
2852 }
3bd1e081
MD
2853 }
2854 }
1fc79fb4
MD
2855 /* All is OK */
2856 err = 0;
3bd1e081
MD
2857end:
2858 DBG("polling thread exiting");
0e428499
DG
2859 free(pollfd);
2860 free(local_stream);
fb3a43a9
DG
2861
2862 /*
2863 * Close the write side of the pipe so epoll_wait() in
7d980def
DG
2864 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2865 * read side of the pipe. If we close them both, epoll_wait strangely does
2866 * not return and could create a endless wait period if the pipe is the
2867 * only tracked fd in the poll set. The thread will take care of closing
2868 * the read side.
fb3a43a9 2869 */
13886d2d 2870 (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe);
fb3a43a9 2871
2d57de81 2872error_testpoint:
1fc79fb4
MD
2873 if (err) {
2874 health_error();
2875 ERR("Health error occurred in %s", __func__);
2876 }
2877 health_unregister(health_consumerd);
2878
e7b994a3 2879 rcu_unregister_thread();
3bd1e081
MD
2880 return NULL;
2881}
2882
d8ef542d
MD
2883/*
2884 * Close wake-up end of each stream belonging to the channel. This will
2885 * allow the poll() on the stream read-side to detect when the
2886 * write-side (application) finally closes them.
2887 */
2888static
2889void consumer_close_channel_streams(struct lttng_consumer_channel *channel)
2890{
2891 struct lttng_ht *ht;
2892 struct lttng_consumer_stream *stream;
2893 struct lttng_ht_iter iter;
2894
2895 ht = consumer_data.stream_per_chan_id_ht;
2896
2897 rcu_read_lock();
2898 cds_lfht_for_each_entry_duplicate(ht->ht,
2899 ht->hash_fct(&channel->key, lttng_ht_seed),
2900 ht->match_fct, &channel->key,
2901 &iter.iter, stream, node_channel_id.node) {
f2ad556d
MD
2902 /*
2903 * Protect against teardown with mutex.
2904 */
2905 pthread_mutex_lock(&stream->lock);
2906 if (cds_lfht_is_node_deleted(&stream->node.node)) {
2907 goto next;
2908 }
d8ef542d
MD
2909 switch (consumer_data.type) {
2910 case LTTNG_CONSUMER_KERNEL:
2911 break;
2912 case LTTNG_CONSUMER32_UST:
2913 case LTTNG_CONSUMER64_UST:
b4a650f3
DG
2914 if (stream->metadata_flag) {
2915 /* Safe and protected by the stream lock. */
2916 lttng_ustconsumer_close_metadata(stream->chan);
2917 } else {
2918 /*
2919 * Note: a mutex is taken internally within
2920 * liblttng-ust-ctl to protect timer wakeup_fd
2921 * use from concurrent close.
2922 */
2923 lttng_ustconsumer_close_stream_wakeup(stream);
2924 }
d8ef542d
MD
2925 break;
2926 default:
2927 ERR("Unknown consumer_data type");
2928 assert(0);
2929 }
f2ad556d
MD
2930 next:
2931 pthread_mutex_unlock(&stream->lock);
d8ef542d
MD
2932 }
2933 rcu_read_unlock();
2934}
2935
2936static void destroy_channel_ht(struct lttng_ht *ht)
2937{
2938 struct lttng_ht_iter iter;
2939 struct lttng_consumer_channel *channel;
2940 int ret;
2941
2942 if (ht == NULL) {
2943 return;
2944 }
2945
2946 rcu_read_lock();
2947 cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) {
2948 ret = lttng_ht_del(ht, &iter);
2949 assert(ret != 0);
2950 }
2951 rcu_read_unlock();
2952
2953 lttng_ht_destroy(ht);
2954}
2955
2956/*
2957 * This thread polls the channel fds to detect when they are being
2958 * closed. It closes all related streams if the channel is detected as
2959 * closed. It is currently only used as a shim layer for UST because the
2960 * consumerd needs to keep the per-stream wakeup end of pipes open for
2961 * periodical flush.
2962 */
2963void *consumer_thread_channel_poll(void *data)
2964{
1fc79fb4 2965 int ret, i, pollfd, err = -1;
d8ef542d
MD
2966 uint32_t revents, nb_fd;
2967 struct lttng_consumer_channel *chan = NULL;
2968 struct lttng_ht_iter iter;
2969 struct lttng_ht_node_u64 *node;
2970 struct lttng_poll_event events;
2971 struct lttng_consumer_local_data *ctx = data;
2972 struct lttng_ht *channel_ht;
2973
2974 rcu_register_thread();
2975
1fc79fb4
MD
2976 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL);
2977
2d57de81
MD
2978 if (testpoint(consumerd_thread_channel)) {
2979 goto error_testpoint;
2980 }
2981
9ce5646a
MD
2982 health_code_update();
2983
d8ef542d
MD
2984 channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2985 if (!channel_ht) {
2986 /* ENOMEM at this point. Better to bail out. */
2987 goto end_ht;
2988 }
2989
2990 DBG("Thread channel poll started");
2991
2992 /* Size is set to 1 for the consumer_channel pipe */
2993 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2994 if (ret < 0) {
2995 ERR("Poll set creation failed");
2996 goto end_poll;
2997 }
2998
2999 ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN);
3000 if (ret < 0) {
3001 goto end;
3002 }
3003
3004 /* Main loop */
3005 DBG("Channel main loop started");
3006
3007 while (1) {
d8ef542d 3008restart:
7fa2082e
MD
3009 health_code_update();
3010 DBG("Channel poll wait");
9ce5646a 3011 health_poll_entry();
d8ef542d 3012 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
3013 DBG("Channel poll return from wait with %d fd(s)",
3014 LTTNG_POLL_GETNB(&events));
9ce5646a 3015 health_poll_exit();
40063ead 3016 DBG("Channel event caught in thread");
d8ef542d
MD
3017 if (ret < 0) {
3018 if (errno == EINTR) {
40063ead 3019 ERR("Poll EINTR caught");
d8ef542d
MD
3020 goto restart;
3021 }
d9607cd7
MD
3022 if (LTTNG_POLL_GETNB(&events) == 0) {
3023 err = 0; /* All is OK */
3024 }
d8ef542d
MD
3025 goto end;
3026 }
3027
3028 nb_fd = ret;
3029
3030 /* From here, the event is a channel wait fd */
3031 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
3032 health_code_update();
3033
d8ef542d
MD
3034 revents = LTTNG_POLL_GETEV(&events, i);
3035 pollfd = LTTNG_POLL_GETFD(&events, i);
3036
d8ef542d 3037 if (pollfd == ctx->consumer_channel_pipe[0]) {
03e43155 3038 if (revents & LPOLLIN) {
d8ef542d 3039 enum consumer_channel_action action;
a0cbdd2e 3040 uint64_t key;
d8ef542d 3041
a0cbdd2e 3042 ret = read_channel_pipe(ctx, &chan, &key, &action);
d8ef542d 3043 if (ret <= 0) {
03e43155
MD
3044 if (ret < 0) {
3045 ERR("Error reading channel pipe");
3046 }
3047 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
d8ef542d
MD
3048 continue;
3049 }
3050
3051 switch (action) {
3052 case CONSUMER_CHANNEL_ADD:
3053 DBG("Adding channel %d to poll set",
3054 chan->wait_fd);
3055
3056 lttng_ht_node_init_u64(&chan->wait_fd_node,
3057 chan->wait_fd);
c7260a81 3058 rcu_read_lock();
d8ef542d
MD
3059 lttng_ht_add_unique_u64(channel_ht,
3060 &chan->wait_fd_node);
c7260a81 3061 rcu_read_unlock();
d8ef542d
MD
3062 /* Add channel to the global poll events list */
3063 lttng_poll_add(&events, chan->wait_fd,
03e43155 3064 LPOLLERR | LPOLLHUP);
d8ef542d 3065 break;
a0cbdd2e
MD
3066 case CONSUMER_CHANNEL_DEL:
3067 {
b4a650f3
DG
3068 /*
3069 * This command should never be called if the channel
3070 * has streams monitored by either the data or metadata
3071 * thread. The consumer only notify this thread with a
3072 * channel del. command if it receives a destroy
3073 * channel command from the session daemon that send it
3074 * if a command prior to the GET_CHANNEL failed.
3075 */
3076
c7260a81 3077 rcu_read_lock();
a0cbdd2e
MD
3078 chan = consumer_find_channel(key);
3079 if (!chan) {
c7260a81 3080 rcu_read_unlock();
a0cbdd2e
MD
3081 ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key);
3082 break;
3083 }
3084 lttng_poll_del(&events, chan->wait_fd);
f623cc0b 3085 iter.iter.node = &chan->wait_fd_node.node;
a0cbdd2e
MD
3086 ret = lttng_ht_del(channel_ht, &iter);
3087 assert(ret == 0);
a0cbdd2e 3088
f2a444f1
DG
3089 switch (consumer_data.type) {
3090 case LTTNG_CONSUMER_KERNEL:
3091 break;
3092 case LTTNG_CONSUMER32_UST:
3093 case LTTNG_CONSUMER64_UST:
212d67a2
DG
3094 health_code_update();
3095 /* Destroy streams that might have been left in the stream list. */
3096 clean_channel_stream_list(chan);
f2a444f1
DG
3097 break;
3098 default:
3099 ERR("Unknown consumer_data type");
3100 assert(0);
3101 }
3102
a0cbdd2e
MD
3103 /*
3104 * Release our own refcount. Force channel deletion even if
3105 * streams were not initialized.
3106 */
3107 if (!uatomic_sub_return(&chan->refcount, 1)) {
3108 consumer_del_channel(chan);
3109 }
c7260a81 3110 rcu_read_unlock();
a0cbdd2e
MD
3111 goto restart;
3112 }
d8ef542d
MD
3113 case CONSUMER_CHANNEL_QUIT:
3114 /*
3115 * Remove the pipe from the poll set and continue the loop
3116 * since their might be data to consume.
3117 */
3118 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
3119 continue;
3120 default:
3121 ERR("Unknown action");
3122 break;
3123 }
03e43155
MD
3124 } else if (revents & (LPOLLERR | LPOLLHUP)) {
3125 DBG("Channel thread pipe hung up");
3126 /*
3127 * Remove the pipe from the poll set and continue the loop
3128 * since their might be data to consume.
3129 */
3130 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
3131 continue;
3132 } else {
3133 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
3134 goto end;
d8ef542d
MD
3135 }
3136
3137 /* Handle other stream */
3138 continue;
3139 }
3140
3141 rcu_read_lock();
3142 {
3143 uint64_t tmp_id = (uint64_t) pollfd;
3144
3145 lttng_ht_lookup(channel_ht, &tmp_id, &iter);
3146 }
3147 node = lttng_ht_iter_get_node_u64(&iter);
3148 assert(node);
3149
3150 chan = caa_container_of(node, struct lttng_consumer_channel,
3151 wait_fd_node);
3152
3153 /* Check for error event */
3154 if (revents & (LPOLLERR | LPOLLHUP)) {
3155 DBG("Channel fd %d is hup|err.", pollfd);
3156
3157 lttng_poll_del(&events, chan->wait_fd);
3158 ret = lttng_ht_del(channel_ht, &iter);
3159 assert(ret == 0);
b4a650f3
DG
3160
3161 /*
3162 * This will close the wait fd for each stream associated to
3163 * this channel AND monitored by the data/metadata thread thus
3164 * will be clean by the right thread.
3165 */
d8ef542d 3166 consumer_close_channel_streams(chan);
f2ad556d
MD
3167
3168 /* Release our own refcount */
3169 if (!uatomic_sub_return(&chan->refcount, 1)
3170 && !uatomic_read(&chan->nb_init_stream_left)) {
3171 consumer_del_channel(chan);
3172 }
03e43155
MD
3173 } else {
3174 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
3175 rcu_read_unlock();
3176 goto end;
d8ef542d
MD
3177 }
3178
3179 /* Release RCU lock for the channel looked up */
3180 rcu_read_unlock();
3181 }
3182 }
3183
1fc79fb4
MD
3184 /* All is OK */
3185 err = 0;
d8ef542d
MD
3186end:
3187 lttng_poll_clean(&events);
3188end_poll:
3189 destroy_channel_ht(channel_ht);
3190end_ht:
2d57de81 3191error_testpoint:
d8ef542d 3192 DBG("Channel poll thread exiting");
1fc79fb4
MD
3193 if (err) {
3194 health_error();
3195 ERR("Health error occurred in %s", __func__);
3196 }
3197 health_unregister(health_consumerd);
d8ef542d
MD
3198 rcu_unregister_thread();
3199 return NULL;
3200}
3201
331744e3
JD
3202static int set_metadata_socket(struct lttng_consumer_local_data *ctx,
3203 struct pollfd *sockpoll, int client_socket)
3204{
3205 int ret;
3206
3207 assert(ctx);
3208 assert(sockpoll);
3209
84382d49
MD
3210 ret = lttng_consumer_poll_socket(sockpoll);
3211 if (ret) {
331744e3
JD
3212 goto error;
3213 }
3214 DBG("Metadata connection on client_socket");
3215
3216 /* Blocking call, waiting for transmission */
3217 ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket);
3218 if (ctx->consumer_metadata_socket < 0) {
3219 WARN("On accept metadata");
3220 ret = -1;
3221 goto error;
3222 }
3223 ret = 0;
3224
3225error:
3226 return ret;
3227}
3228
3bd1e081
MD
3229/*
3230 * This thread listens on the consumerd socket and receives the file
3231 * descriptors from the session daemon.