lttng-ctl: separate support of named/unnamed trigger registration
[lttng-tools.git] / src / common / consumer / consumer.c
CommitLineData
3bd1e081 1/*
ab5be9fa
MJ
2 * Copyright (C) 2011 Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * Copyright (C) 2012 David Goulet <dgoulet@efficios.com>
3bd1e081 5 *
ab5be9fa 6 * SPDX-License-Identifier: GPL-2.0-only
3bd1e081 7 *
3bd1e081
MD
8 */
9
6f9449c2 10#include "common/index/ctf-index.h"
6c1c0768 11#define _LGPL_SOURCE
3bd1e081 12#include <assert.h>
3bd1e081
MD
13#include <poll.h>
14#include <pthread.h>
15#include <stdlib.h>
16#include <string.h>
17#include <sys/mman.h>
18#include <sys/socket.h>
19#include <sys/types.h>
20#include <unistd.h>
77c7c900 21#include <inttypes.h>
331744e3 22#include <signal.h>
3bd1e081 23
51a9e1c7 24#include <bin/lttng-consumerd/health-consumerd.h>
990570ed 25#include <common/common.h>
fb3a43a9 26#include <common/utils.h>
d2956687 27#include <common/time.h>
fb3a43a9 28#include <common/compat/poll.h>
f263b7fd 29#include <common/compat/endian.h>
309167d2 30#include <common/index/index.h>
10a8a223 31#include <common/kernel-ctl/kernel-ctl.h>
00e2e675 32#include <common/sessiond-comm/relayd.h>
10a8a223
DG
33#include <common/sessiond-comm/sessiond-comm.h>
34#include <common/kernel-consumer/kernel-consumer.h>
00e2e675 35#include <common/relayd/relayd.h>
10a8a223 36#include <common/ust-consumer/ust-consumer.h>
c8fea79c
JR
37#include <common/consumer/consumer-timer.h>
38#include <common/consumer/consumer.h>
39#include <common/consumer/consumer-stream.h>
40#include <common/consumer/consumer-testpoint.h>
41#include <common/align.h>
5feafd41 42#include <common/consumer/consumer-metadata-cache.h>
d2956687
JG
43#include <common/trace-chunk.h>
44#include <common/trace-chunk-registry.h>
45#include <common/string-utils/format.h>
c35f9726 46#include <common/dynamic-array.h>
3bd1e081 47
fa29bfbf
SM
48struct lttng_consumer_global_data the_consumer_data = {
49 .stream_count = 0,
50 .need_update = 1,
51 .type = LTTNG_CONSUMER_UNKNOWN,
3bd1e081
MD
52};
53
d8ef542d
MD
54enum consumer_channel_action {
55 CONSUMER_CHANNEL_ADD,
a0cbdd2e 56 CONSUMER_CHANNEL_DEL,
d8ef542d
MD
57 CONSUMER_CHANNEL_QUIT,
58};
59
60struct consumer_channel_msg {
61 enum consumer_channel_action action;
a0cbdd2e
MD
62 struct lttng_consumer_channel *chan; /* add */
63 uint64_t key; /* del */
d8ef542d
MD
64};
65
80957876 66/* Flag used to temporarily pause data consumption from testpoints. */
cf0bcb51
JG
67int data_consumption_paused;
68
3bd1e081
MD
69/*
70 * Flag to inform the polling thread to quit when all fd hung up. Updated by
71 * the consumer_thread_receive_fds when it notices that all fds has hung up.
72 * Also updated by the signal handler (consumer_should_exit()). Read by the
73 * polling threads.
74 */
10211f5c 75int consumer_quit;
3bd1e081 76
43c34bc3 77/*
43c34bc3
DG
78 * Global hash table containing respectively metadata and data streams. The
79 * stream element in this ht should only be updated by the metadata poll thread
80 * for the metadata and the data poll thread for the data.
81 */
40dc48e0
DG
82static struct lttng_ht *metadata_ht;
83static struct lttng_ht *data_ht;
43c34bc3 84
5da88b0f
MD
85static const char *get_consumer_domain(void)
86{
fa29bfbf 87 switch (the_consumer_data.type) {
5da88b0f
MD
88 case LTTNG_CONSUMER_KERNEL:
89 return DEFAULT_KERNEL_TRACE_DIR;
90 case LTTNG_CONSUMER64_UST:
91 /* Fall-through. */
92 case LTTNG_CONSUMER32_UST:
93 return DEFAULT_UST_TRACE_DIR;
94 default:
95 abort();
96 }
97}
98
acdb9057
DG
99/*
100 * Notify a thread lttng pipe to poll back again. This usually means that some
101 * global state has changed so we just send back the thread in a poll wait
102 * call.
103 */
104static void notify_thread_lttng_pipe(struct lttng_pipe *pipe)
105{
106 struct lttng_consumer_stream *null_stream = NULL;
107
108 assert(pipe);
109
110 (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream));
111}
112
5c635c72
MD
113static void notify_health_quit_pipe(int *pipe)
114{
6cd525e8 115 ssize_t ret;
5c635c72 116
6cd525e8
MD
117 ret = lttng_write(pipe[1], "4", 1);
118 if (ret < 1) {
5c635c72
MD
119 PERROR("write consumer health quit");
120 }
121}
122
d8ef542d
MD
123static void notify_channel_pipe(struct lttng_consumer_local_data *ctx,
124 struct lttng_consumer_channel *chan,
a0cbdd2e 125 uint64_t key,
d8ef542d
MD
126 enum consumer_channel_action action)
127{
128 struct consumer_channel_msg msg;
6cd525e8 129 ssize_t ret;
d8ef542d 130
e56251fc
DG
131 memset(&msg, 0, sizeof(msg));
132
d8ef542d
MD
133 msg.action = action;
134 msg.chan = chan;
f21dae48 135 msg.key = key;
6cd525e8
MD
136 ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg));
137 if (ret < sizeof(msg)) {
138 PERROR("notify_channel_pipe write error");
139 }
d8ef542d
MD
140}
141
a0cbdd2e
MD
142void notify_thread_del_channel(struct lttng_consumer_local_data *ctx,
143 uint64_t key)
144{
145 notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL);
146}
147
d8ef542d
MD
148static int read_channel_pipe(struct lttng_consumer_local_data *ctx,
149 struct lttng_consumer_channel **chan,
a0cbdd2e 150 uint64_t *key,
d8ef542d
MD
151 enum consumer_channel_action *action)
152{
153 struct consumer_channel_msg msg;
6cd525e8 154 ssize_t ret;
d8ef542d 155
6cd525e8
MD
156 ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg));
157 if (ret < sizeof(msg)) {
158 ret = -1;
159 goto error;
d8ef542d 160 }
6cd525e8
MD
161 *action = msg.action;
162 *chan = msg.chan;
163 *key = msg.key;
164error:
165 return (int) ret;
d8ef542d
MD
166}
167
212d67a2
DG
168/*
169 * Cleanup the stream list of a channel. Those streams are not yet globally
170 * visible
171 */
172static void clean_channel_stream_list(struct lttng_consumer_channel *channel)
173{
174 struct lttng_consumer_stream *stream, *stmp;
175
176 assert(channel);
177
178 /* Delete streams that might have been left in the stream list. */
179 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
180 send_node) {
181 cds_list_del(&stream->send_node);
182 /*
183 * Once a stream is added to this list, the buffers were created so we
184 * have a guarantee that this call will succeed. Setting the monitor
185 * mode to 0 so we don't lock nor try to delete the stream from the
186 * global hash table.
187 */
188 stream->monitor = 0;
189 consumer_stream_destroy(stream, NULL);
190 }
191}
192
3bd1e081
MD
193/*
194 * Find a stream. The consumer_data.lock must be locked during this
195 * call.
196 */
d88aee68 197static struct lttng_consumer_stream *find_stream(uint64_t key,
8389e4f8 198 struct lttng_ht *ht)
3bd1e081 199{
e4421fec 200 struct lttng_ht_iter iter;
d88aee68 201 struct lttng_ht_node_u64 *node;
e4421fec 202 struct lttng_consumer_stream *stream = NULL;
3bd1e081 203
8389e4f8
DG
204 assert(ht);
205
d88aee68
DG
206 /* -1ULL keys are lookup failures */
207 if (key == (uint64_t) -1ULL) {
7ad0a0cb 208 return NULL;
7a57cf92 209 }
e4421fec 210
6065ceec
DG
211 rcu_read_lock();
212
d88aee68
DG
213 lttng_ht_lookup(ht, &key, &iter);
214 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
215 if (node != NULL) {
216 stream = caa_container_of(node, struct lttng_consumer_stream, node);
3bd1e081 217 }
e4421fec 218
6065ceec
DG
219 rcu_read_unlock();
220
e4421fec 221 return stream;
3bd1e081
MD
222}
223
da009f2c 224static void steal_stream_key(uint64_t key, struct lttng_ht *ht)
7ad0a0cb
MD
225{
226 struct lttng_consumer_stream *stream;
227
04253271 228 rcu_read_lock();
ffe60014 229 stream = find_stream(key, ht);
04253271 230 if (stream) {
da009f2c 231 stream->key = (uint64_t) -1ULL;
04253271
MD
232 /*
233 * We don't want the lookup to match, but we still need
234 * to iterate on this stream when iterating over the hash table. Just
235 * change the node key.
236 */
da009f2c 237 stream->node.key = (uint64_t) -1ULL;
04253271
MD
238 }
239 rcu_read_unlock();
7ad0a0cb
MD
240}
241
d56db448
DG
242/*
243 * Return a channel object for the given key.
244 *
245 * RCU read side lock MUST be acquired before calling this function and
246 * protects the channel ptr.
247 */
d88aee68 248struct lttng_consumer_channel *consumer_find_channel(uint64_t key)
3bd1e081 249{
e4421fec 250 struct lttng_ht_iter iter;
d88aee68 251 struct lttng_ht_node_u64 *node;
e4421fec 252 struct lttng_consumer_channel *channel = NULL;
3bd1e081 253
d88aee68
DG
254 /* -1ULL keys are lookup failures */
255 if (key == (uint64_t) -1ULL) {
7ad0a0cb 256 return NULL;
7a57cf92 257 }
e4421fec 258
fa29bfbf 259 lttng_ht_lookup(the_consumer_data.channel_ht, &key, &iter);
d88aee68 260 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
261 if (node != NULL) {
262 channel = caa_container_of(node, struct lttng_consumer_channel, node);
3bd1e081 263 }
e4421fec
DG
264
265 return channel;
3bd1e081
MD
266}
267
b5a6470f
DG
268/*
269 * There is a possibility that the consumer does not have enough time between
270 * the close of the channel on the session daemon and the cleanup in here thus
271 * once we have a channel add with an existing key, we know for sure that this
272 * channel will eventually get cleaned up by all streams being closed.
273 *
274 * This function just nullifies the already existing channel key.
275 */
276static void steal_channel_key(uint64_t key)
277{
278 struct lttng_consumer_channel *channel;
279
280 rcu_read_lock();
281 channel = consumer_find_channel(key);
282 if (channel) {
283 channel->key = (uint64_t) -1ULL;
284 /*
285 * We don't want the lookup to match, but we still need to iterate on
286 * this channel when iterating over the hash table. Just change the
287 * node key.
288 */
289 channel->node.key = (uint64_t) -1ULL;
290 }
291 rcu_read_unlock();
292}
293
ffe60014 294static void free_channel_rcu(struct rcu_head *head)
702b1ea4 295{
d88aee68
DG
296 struct lttng_ht_node_u64 *node =
297 caa_container_of(head, struct lttng_ht_node_u64, head);
ffe60014
DG
298 struct lttng_consumer_channel *channel =
299 caa_container_of(node, struct lttng_consumer_channel, node);
702b1ea4 300
fa29bfbf 301 switch (the_consumer_data.type) {
b83e03c4
MD
302 case LTTNG_CONSUMER_KERNEL:
303 break;
304 case LTTNG_CONSUMER32_UST:
305 case LTTNG_CONSUMER64_UST:
306 lttng_ustconsumer_free_channel(channel);
307 break;
308 default:
309 ERR("Unknown consumer_data type");
310 abort();
311 }
ffe60014 312 free(channel);
702b1ea4
MD
313}
314
00e2e675
DG
315/*
316 * RCU protected relayd socket pair free.
317 */
ffe60014 318static void free_relayd_rcu(struct rcu_head *head)
00e2e675 319{
d88aee68
DG
320 struct lttng_ht_node_u64 *node =
321 caa_container_of(head, struct lttng_ht_node_u64, head);
00e2e675
DG
322 struct consumer_relayd_sock_pair *relayd =
323 caa_container_of(node, struct consumer_relayd_sock_pair, node);
324
8994307f
DG
325 /*
326 * Close all sockets. This is done in the call RCU since we don't want the
327 * socket fds to be reassigned thus potentially creating bad state of the
328 * relayd object.
329 *
330 * We do not have to lock the control socket mutex here since at this stage
331 * there is no one referencing to this relayd object.
332 */
333 (void) relayd_close(&relayd->control_sock);
334 (void) relayd_close(&relayd->data_sock);
335
3a84e2f3 336 pthread_mutex_destroy(&relayd->ctrl_sock_mutex);
00e2e675
DG
337 free(relayd);
338}
339
340/*
341 * Destroy and free relayd socket pair object.
00e2e675 342 */
51230d70 343void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
344{
345 int ret;
346 struct lttng_ht_iter iter;
347
173af62f
DG
348 if (relayd == NULL) {
349 return;
350 }
351
00e2e675
DG
352 DBG("Consumer destroy and close relayd socket pair");
353
354 iter.iter.node = &relayd->node.node;
fa29bfbf 355 ret = lttng_ht_del(the_consumer_data.relayd_ht, &iter);
173af62f 356 if (ret != 0) {
8994307f 357 /* We assume the relayd is being or is destroyed */
173af62f
DG
358 return;
359 }
00e2e675 360
00e2e675 361 /* RCU free() call */
ffe60014
DG
362 call_rcu(&relayd->node.head, free_relayd_rcu);
363}
364
365/*
366 * Remove a channel from the global list protected by a mutex. This function is
367 * also responsible for freeing its data structures.
368 */
369void consumer_del_channel(struct lttng_consumer_channel *channel)
370{
ffe60014
DG
371 struct lttng_ht_iter iter;
372
d88aee68 373 DBG("Consumer delete channel key %" PRIu64, channel->key);
ffe60014 374
fa29bfbf 375 pthread_mutex_lock(&the_consumer_data.lock);
a9838785 376 pthread_mutex_lock(&channel->lock);
ffe60014 377
212d67a2
DG
378 /* Destroy streams that might have been left in the stream list. */
379 clean_channel_stream_list(channel);
51e762e5 380
d3e2ba59
JD
381 if (channel->live_timer_enabled == 1) {
382 consumer_timer_live_stop(channel);
383 }
e9404c27
JG
384 if (channel->monitor_timer_enabled == 1) {
385 consumer_timer_monitor_stop(channel);
386 }
d3e2ba59 387
fa29bfbf 388 switch (the_consumer_data.type) {
ffe60014
DG
389 case LTTNG_CONSUMER_KERNEL:
390 break;
391 case LTTNG_CONSUMER32_UST:
392 case LTTNG_CONSUMER64_UST:
393 lttng_ustconsumer_del_channel(channel);
394 break;
395 default:
396 ERR("Unknown consumer_data type");
397 assert(0);
398 goto end;
399 }
400
d2956687
JG
401 lttng_trace_chunk_put(channel->trace_chunk);
402 channel->trace_chunk = NULL;
5c3892a6 403
d2956687
JG
404 if (channel->is_published) {
405 int ret;
406
407 rcu_read_lock();
408 iter.iter.node = &channel->node.node;
fa29bfbf 409 ret = lttng_ht_del(the_consumer_data.channel_ht, &iter);
d2956687 410 assert(!ret);
ffe60014 411
d2956687 412 iter.iter.node = &channel->channels_by_session_id_ht_node.node;
fa29bfbf 413 ret = lttng_ht_del(the_consumer_data.channels_by_session_id_ht,
d2956687
JG
414 &iter);
415 assert(!ret);
416 rcu_read_unlock();
417 }
418
b6921a17
JG
419 channel->is_deleted = true;
420 call_rcu(&channel->node.head, free_channel_rcu);
ffe60014 421end:
a9838785 422 pthread_mutex_unlock(&channel->lock);
fa29bfbf 423 pthread_mutex_unlock(&the_consumer_data.lock);
00e2e675
DG
424}
425
228b5bf7
DG
426/*
427 * Iterate over the relayd hash table and destroy each element. Finally,
428 * destroy the whole hash table.
429 */
430static void cleanup_relayd_ht(void)
431{
432 struct lttng_ht_iter iter;
433 struct consumer_relayd_sock_pair *relayd;
434
435 rcu_read_lock();
436
fa29bfbf
SM
437 cds_lfht_for_each_entry(the_consumer_data.relayd_ht->ht, &iter.iter,
438 relayd, node.node) {
51230d70 439 consumer_destroy_relayd(relayd);
228b5bf7
DG
440 }
441
228b5bf7 442 rcu_read_unlock();
36b588ed 443
fa29bfbf 444 lttng_ht_destroy(the_consumer_data.relayd_ht);
228b5bf7
DG
445}
446
8994307f
DG
447/*
448 * Update the end point status of all streams having the given network sequence
449 * index (relayd index).
450 *
451 * It's atomically set without having the stream mutex locked which is fine
452 * because we handle the write/read race with a pipe wakeup for each thread.
453 */
da009f2c 454static void update_endpoint_status_by_netidx(uint64_t net_seq_idx,
8994307f
DG
455 enum consumer_endpoint_status status)
456{
457 struct lttng_ht_iter iter;
458 struct lttng_consumer_stream *stream;
459
da009f2c 460 DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx);
8994307f
DG
461
462 rcu_read_lock();
463
464 /* Let's begin with metadata */
465 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
466 if (stream->net_seq_idx == net_seq_idx) {
467 uatomic_set(&stream->endpoint_status, status);
468 DBG("Delete flag set to metadata stream %d", stream->wait_fd);
469 }
470 }
471
472 /* Follow up by the data streams */
473 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
474 if (stream->net_seq_idx == net_seq_idx) {
475 uatomic_set(&stream->endpoint_status, status);
476 DBG("Delete flag set to data stream %d", stream->wait_fd);
477 }
478 }
479 rcu_read_unlock();
480}
481
482/*
483 * Cleanup a relayd object by flagging every associated streams for deletion,
484 * destroying the object meaning removing it from the relayd hash table,
485 * closing the sockets and freeing the memory in a RCU call.
486 *
487 * If a local data context is available, notify the threads that the streams'
488 * state have changed.
489 */
9276e5c8 490void lttng_consumer_cleanup_relayd(struct consumer_relayd_sock_pair *relayd)
8994307f 491{
da009f2c 492 uint64_t netidx;
8994307f
DG
493
494 assert(relayd);
495
9276e5c8 496 DBG("Cleaning up relayd object ID %"PRIu64, relayd->net_seq_idx);
9617607b 497
8994307f
DG
498 /* Save the net sequence index before destroying the object */
499 netidx = relayd->net_seq_idx;
500
501 /*
502 * Delete the relayd from the relayd hash table, close the sockets and free
503 * the object in a RCU call.
504 */
51230d70 505 consumer_destroy_relayd(relayd);
8994307f
DG
506
507 /* Set inactive endpoint to all streams */
508 update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE);
509
510 /*
511 * With a local data context, notify the threads that the streams' state
512 * have changed. The write() action on the pipe acts as an "implicit"
513 * memory barrier ordering the updates of the end point status from the
514 * read of this status which happens AFTER receiving this notify.
515 */
9276e5c8
JR
516 notify_thread_lttng_pipe(relayd->ctx->consumer_data_pipe);
517 notify_thread_lttng_pipe(relayd->ctx->consumer_metadata_pipe);
8994307f
DG
518}
519
a6ba4fe1
DG
520/*
521 * Flag a relayd socket pair for destruction. Destroy it if the refcount
522 * reaches zero.
523 *
524 * RCU read side lock MUST be aquired before calling this function.
525 */
526void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd)
527{
528 assert(relayd);
529
530 /* Set destroy flag for this object */
531 uatomic_set(&relayd->destroy_flag, 1);
532
533 /* Destroy the relayd if refcount is 0 */
534 if (uatomic_read(&relayd->refcount) == 0) {
51230d70 535 consumer_destroy_relayd(relayd);
a6ba4fe1
DG
536 }
537}
538
3bd1e081 539/*
1d1a276c
DG
540 * Completly destroy stream from every visiable data structure and the given
541 * hash table if one.
542 *
543 * One this call returns, the stream object is not longer usable nor visible.
3bd1e081 544 */
e316aad5
DG
545void consumer_del_stream(struct lttng_consumer_stream *stream,
546 struct lttng_ht *ht)
3bd1e081 547{
1d1a276c 548 consumer_stream_destroy(stream, ht);
3bd1e081
MD
549}
550
5ab66908
MD
551/*
552 * XXX naming of del vs destroy is all mixed up.
553 */
554void consumer_del_stream_for_data(struct lttng_consumer_stream *stream)
555{
556 consumer_stream_destroy(stream, data_ht);
557}
558
559void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream)
560{
561 consumer_stream_destroy(stream, metadata_ht);
562}
563
d9a2e16e
JD
564void consumer_stream_update_channel_attributes(
565 struct lttng_consumer_stream *stream,
566 struct lttng_consumer_channel *channel)
567{
568 stream->channel_read_only_attributes.tracefile_size =
569 channel->tracefile_size;
d9a2e16e
JD
570}
571
3bd1e081
MD
572/*
573 * Add a stream to the global list protected by a mutex.
574 */
66d583dc 575void consumer_add_data_stream(struct lttng_consumer_stream *stream)
3bd1e081 576{
5ab66908 577 struct lttng_ht *ht = data_ht;
3bd1e081 578
e316aad5 579 assert(stream);
43c34bc3 580 assert(ht);
c77fc10a 581
d88aee68 582 DBG3("Adding consumer stream %" PRIu64, stream->key);
e316aad5 583
fa29bfbf 584 pthread_mutex_lock(&the_consumer_data.lock);
a9838785 585 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 586 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 587 pthread_mutex_lock(&stream->lock);
b0b335c8 588 rcu_read_lock();
e316aad5 589
43c34bc3 590 /* Steal stream identifier to avoid having streams with the same key */
ffe60014 591 steal_stream_key(stream->key, ht);
43c34bc3 592
d88aee68 593 lttng_ht_add_unique_u64(ht, &stream->node);
00e2e675 594
fa29bfbf 595 lttng_ht_add_u64(the_consumer_data.stream_per_chan_id_ht,
d8ef542d
MD
596 &stream->node_channel_id);
597
ca22feea
DG
598 /*
599 * Add stream to the stream_list_ht of the consumer data. No need to steal
600 * the key since the HT does not use it and we allow to add redundant keys
601 * into this table.
602 */
fa29bfbf
SM
603 lttng_ht_add_u64(the_consumer_data.stream_list_ht,
604 &stream->node_session_id);
ca22feea 605
e316aad5 606 /*
ffe60014
DG
607 * When nb_init_stream_left reaches 0, we don't need to trigger any action
608 * in terms of destroying the associated channel, because the action that
e316aad5
DG
609 * causes the count to become 0 also causes a stream to be added. The
610 * channel deletion will thus be triggered by the following removal of this
611 * stream.
612 */
ffe60014 613 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
614 /* Increment refcount before decrementing nb_init_stream_left */
615 cmm_smp_wmb();
ffe60014 616 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
617 }
618
619 /* Update consumer data once the node is inserted. */
fa29bfbf
SM
620 the_consumer_data.stream_count++;
621 the_consumer_data.need_update = 1;
3bd1e081 622
e316aad5 623 rcu_read_unlock();
2e818a6a 624 pthread_mutex_unlock(&stream->lock);
ec6ea7d0 625 pthread_mutex_unlock(&stream->chan->timer_lock);
a9838785 626 pthread_mutex_unlock(&stream->chan->lock);
fa29bfbf 627 pthread_mutex_unlock(&the_consumer_data.lock);
3bd1e081
MD
628}
629
00e2e675 630/*
3f8e211f
DG
631 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
632 * be acquired before calling this.
00e2e675 633 */
d09e1200 634static int add_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
635{
636 int ret = 0;
d88aee68 637 struct lttng_ht_node_u64 *node;
00e2e675
DG
638 struct lttng_ht_iter iter;
639
ffe60014 640 assert(relayd);
00e2e675 641
fa29bfbf
SM
642 lttng_ht_lookup(the_consumer_data.relayd_ht, &relayd->net_seq_idx,
643 &iter);
d88aee68 644 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675 645 if (node != NULL) {
00e2e675
DG
646 goto end;
647 }
fa29bfbf 648 lttng_ht_add_unique_u64(the_consumer_data.relayd_ht, &relayd->node);
00e2e675 649
00e2e675
DG
650end:
651 return ret;
652}
653
654/*
655 * Allocate and return a consumer relayd socket.
656 */
027a694f 657static struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair(
da009f2c 658 uint64_t net_seq_idx)
00e2e675
DG
659{
660 struct consumer_relayd_sock_pair *obj = NULL;
661
da009f2c
MD
662 /* net sequence index of -1 is a failure */
663 if (net_seq_idx == (uint64_t) -1ULL) {
00e2e675
DG
664 goto error;
665 }
666
667 obj = zmalloc(sizeof(struct consumer_relayd_sock_pair));
668 if (obj == NULL) {
669 PERROR("zmalloc relayd sock");
670 goto error;
671 }
672
673 obj->net_seq_idx = net_seq_idx;
674 obj->refcount = 0;
173af62f 675 obj->destroy_flag = 0;
f96e4545
MD
676 obj->control_sock.sock.fd = -1;
677 obj->data_sock.sock.fd = -1;
d88aee68 678 lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx);
00e2e675
DG
679 pthread_mutex_init(&obj->ctrl_sock_mutex, NULL);
680
681error:
682 return obj;
683}
684
685/*
686 * Find a relayd socket pair in the global consumer data.
687 *
688 * Return the object if found else NULL.
b0b335c8
MD
689 * RCU read-side lock must be held across this call and while using the
690 * returned object.
00e2e675 691 */
d88aee68 692struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key)
00e2e675
DG
693{
694 struct lttng_ht_iter iter;
d88aee68 695 struct lttng_ht_node_u64 *node;
00e2e675
DG
696 struct consumer_relayd_sock_pair *relayd = NULL;
697
698 /* Negative keys are lookup failures */
d88aee68 699 if (key == (uint64_t) -1ULL) {
00e2e675
DG
700 goto error;
701 }
702
fa29bfbf 703 lttng_ht_lookup(the_consumer_data.relayd_ht, &key, &iter);
d88aee68 704 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675
DG
705 if (node != NULL) {
706 relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node);
707 }
708
00e2e675
DG
709error:
710 return relayd;
711}
712
10a50311
JD
713/*
714 * Find a relayd and send the stream
715 *
716 * Returns 0 on success, < 0 on error
717 */
718int consumer_send_relayd_stream(struct lttng_consumer_stream *stream,
719 char *path)
720{
721 int ret = 0;
722 struct consumer_relayd_sock_pair *relayd;
723
724 assert(stream);
725 assert(stream->net_seq_idx != -1ULL);
726 assert(path);
727
728 /* The stream is not metadata. Get relayd reference if exists. */
729 rcu_read_lock();
730 relayd = consumer_find_relayd(stream->net_seq_idx);
731 if (relayd != NULL) {
732 /* Add stream on the relayd */
733 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
734 ret = relayd_add_stream(&relayd->control_sock, stream->name,
5da88b0f 735 get_consumer_domain(), path, &stream->relayd_stream_id,
d2956687
JG
736 stream->chan->tracefile_size,
737 stream->chan->tracefile_count,
738 stream->trace_chunk);
10a50311
JD
739 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
740 if (ret < 0) {
9276e5c8
JR
741 ERR("Relayd add stream failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
742 lttng_consumer_cleanup_relayd(relayd);
10a50311
JD
743 goto end;
744 }
1c20f0e2 745
10a50311 746 uatomic_inc(&relayd->refcount);
d01178b6 747 stream->sent_to_relayd = 1;
10a50311
JD
748 } else {
749 ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.",
750 stream->key, stream->net_seq_idx);
751 ret = -1;
752 goto end;
753 }
754
755 DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64,
756 stream->name, stream->key, stream->net_seq_idx);
757
758end:
759 rcu_read_unlock();
760 return ret;
761}
762
a4baae1b
JD
763/*
764 * Find a relayd and send the streams sent message
765 *
766 * Returns 0 on success, < 0 on error
767 */
768int consumer_send_relayd_streams_sent(uint64_t net_seq_idx)
769{
770 int ret = 0;
771 struct consumer_relayd_sock_pair *relayd;
772
773 assert(net_seq_idx != -1ULL);
774
775 /* The stream is not metadata. Get relayd reference if exists. */
776 rcu_read_lock();
777 relayd = consumer_find_relayd(net_seq_idx);
778 if (relayd != NULL) {
779 /* Add stream on the relayd */
780 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
781 ret = relayd_streams_sent(&relayd->control_sock);
782 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
783 if (ret < 0) {
9276e5c8
JR
784 ERR("Relayd streams sent failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
785 lttng_consumer_cleanup_relayd(relayd);
a4baae1b
JD
786 goto end;
787 }
788 } else {
789 ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.",
790 net_seq_idx);
791 ret = -1;
792 goto end;
793 }
794
795 ret = 0;
796 DBG("All streams sent relayd id %" PRIu64, net_seq_idx);
797
798end:
799 rcu_read_unlock();
800 return ret;
801}
802
10a50311
JD
803/*
804 * Find a relayd and close the stream
805 */
806void close_relayd_stream(struct lttng_consumer_stream *stream)
807{
808 struct consumer_relayd_sock_pair *relayd;
809
810 /* The stream is not metadata. Get relayd reference if exists. */
811 rcu_read_lock();
812 relayd = consumer_find_relayd(stream->net_seq_idx);
813 if (relayd) {
814 consumer_stream_relayd_close(stream, relayd);
815 }
816 rcu_read_unlock();
817}
818
00e2e675
DG
819/*
820 * Handle stream for relayd transmission if the stream applies for network
821 * streaming where the net sequence index is set.
822 *
823 * Return destination file descriptor or negative value on error.
824 */
6197aea7 825static int write_relayd_stream_header(struct lttng_consumer_stream *stream,
1d4dfdef
DG
826 size_t data_size, unsigned long padding,
827 struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
828{
829 int outfd = -1, ret;
00e2e675
DG
830 struct lttcomm_relayd_data_hdr data_hdr;
831
832 /* Safety net */
833 assert(stream);
6197aea7 834 assert(relayd);
00e2e675
DG
835
836 /* Reset data header */
837 memset(&data_hdr, 0, sizeof(data_hdr));
838
00e2e675
DG
839 if (stream->metadata_flag) {
840 /* Caller MUST acquire the relayd control socket lock */
841 ret = relayd_send_metadata(&relayd->control_sock, data_size);
842 if (ret < 0) {
843 goto error;
844 }
845
846 /* Metadata are always sent on the control socket. */
6151a90f 847 outfd = relayd->control_sock.sock.fd;
00e2e675
DG
848 } else {
849 /* Set header with stream information */
850 data_hdr.stream_id = htobe64(stream->relayd_stream_id);
851 data_hdr.data_size = htobe32(data_size);
1d4dfdef 852 data_hdr.padding_size = htobe32(padding);
c35f9726 853
39df6d9f
DG
854 /*
855 * Note that net_seq_num below is assigned with the *current* value of
856 * next_net_seq_num and only after that the next_net_seq_num will be
857 * increment. This is why when issuing a command on the relayd using
858 * this next value, 1 should always be substracted in order to compare
859 * the last seen sequence number on the relayd side to the last sent.
860 */
3604f373 861 data_hdr.net_seq_num = htobe64(stream->next_net_seq_num);
00e2e675
DG
862 /* Other fields are zeroed previously */
863
864 ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr,
865 sizeof(data_hdr));
866 if (ret < 0) {
867 goto error;
868 }
869
3604f373
DG
870 ++stream->next_net_seq_num;
871
00e2e675 872 /* Set to go on data socket */
6151a90f 873 outfd = relayd->data_sock.sock.fd;
00e2e675
DG
874 }
875
876error:
877 return outfd;
878}
879
b1316da1
JG
880/*
881 * Write a character on the metadata poll pipe to wake the metadata thread.
882 * Returns 0 on success, -1 on error.
883 */
884int consumer_metadata_wakeup_pipe(const struct lttng_consumer_channel *channel)
885{
886 int ret = 0;
887
888 DBG("Waking up metadata poll thread (writing to pipe): channel name = '%s'",
889 channel->name);
890 if (channel->monitor && channel->metadata_stream) {
891 const char dummy = 'c';
892 const ssize_t write_ret = lttng_write(
893 channel->metadata_stream->ust_metadata_poll_pipe[1],
894 &dummy, 1);
895
896 if (write_ret < 1) {
897 if (errno == EWOULDBLOCK) {
898 /*
899 * This is fine, the metadata poll thread
900 * is having a hard time keeping-up, but
901 * it will eventually wake-up and consume
902 * the available data.
903 */
904 ret = 0;
905 } else {
906 PERROR("Failed to write to UST metadata pipe while attempting to wake-up the metadata poll thread");
907 ret = -1;
908 goto end;
909 }
910 }
911 }
912
913end:
914 return ret;
915}
916
d2956687
JG
917/*
918 * Trigger a dump of the metadata content. Following/during the succesful
919 * completion of this call, the metadata poll thread will start receiving
920 * metadata packets to consume.
921 *
922 * The caller must hold the channel and stream locks.
923 */
924static
925int consumer_metadata_stream_dump(struct lttng_consumer_stream *stream)
926{
927 int ret;
928
929 ASSERT_LOCKED(stream->chan->lock);
930 ASSERT_LOCKED(stream->lock);
931 assert(stream->metadata_flag);
932 assert(stream->chan->trace_chunk);
933
fa29bfbf 934 switch (the_consumer_data.type) {
d2956687
JG
935 case LTTNG_CONSUMER_KERNEL:
936 /*
937 * Reset the position of what has been read from the
938 * metadata cache to 0 so we can dump it again.
939 */
940 ret = kernctl_metadata_cache_dump(stream->wait_fd);
941 break;
942 case LTTNG_CONSUMER32_UST:
943 case LTTNG_CONSUMER64_UST:
944 /*
945 * Reset the position pushed from the metadata cache so it
946 * will write from the beginning on the next push.
947 */
948 stream->ust_metadata_pushed = 0;
949 ret = consumer_metadata_wakeup_pipe(stream->chan);
950 break;
951 default:
952 ERR("Unknown consumer_data type");
953 abort();
954 }
955 if (ret < 0) {
956 ERR("Failed to dump the metadata cache");
957 }
958 return ret;
959}
960
961static
962int lttng_consumer_channel_set_trace_chunk(
963 struct lttng_consumer_channel *channel,
964 struct lttng_trace_chunk *new_trace_chunk)
965{
d2956687 966 pthread_mutex_lock(&channel->lock);
b6921a17
JG
967 if (channel->is_deleted) {
968 /*
969 * The channel has been logically deleted and should no longer
970 * be used. It has released its reference to its current trace
971 * chunk and should not acquire a new one.
972 *
973 * Return success as there is nothing for the caller to do.
974 */
975 goto end;
976 }
d2956687
JG
977
978 /*
979 * The acquisition of the reference cannot fail (barring
980 * a severe internal error) since a reference to the published
981 * chunk is already held by the caller.
982 */
983 if (new_trace_chunk) {
984 const bool acquired_reference = lttng_trace_chunk_get(
985 new_trace_chunk);
986
987 assert(acquired_reference);
988 }
989
990 lttng_trace_chunk_put(channel->trace_chunk);
991 channel->trace_chunk = new_trace_chunk;
d2956687
JG
992end:
993 pthread_mutex_unlock(&channel->lock);
ce1aa6fe 994 return 0;
d2956687
JG
995}
996
3bd1e081 997/*
ffe60014
DG
998 * Allocate and return a new lttng_consumer_channel object using the given key
999 * to initialize the hash table node.
1000 *
1001 * On error, return NULL.
3bd1e081 1002 */
886224ff 1003struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key,
ffe60014 1004 uint64_t session_id,
d2956687 1005 const uint64_t *chunk_id,
ffe60014
DG
1006 const char *pathname,
1007 const char *name,
57a269f2 1008 uint64_t relayd_id,
1624d5b7
JD
1009 enum lttng_event_output output,
1010 uint64_t tracefile_size,
2bba9e53 1011 uint64_t tracefile_count,
1950109e 1012 uint64_t session_id_per_pid,
ecc48a90 1013 unsigned int monitor,
d7ba1388 1014 unsigned int live_timer_interval,
a2814ea7 1015 bool is_in_live_session,
3d071855 1016 const char *root_shm_path,
d7ba1388 1017 const char *shm_path)
3bd1e081 1018{
d2956687
JG
1019 struct lttng_consumer_channel *channel = NULL;
1020 struct lttng_trace_chunk *trace_chunk = NULL;
1021
1022 if (chunk_id) {
1023 trace_chunk = lttng_trace_chunk_registry_find_chunk(
fa29bfbf 1024 the_consumer_data.chunk_registry, session_id,
d2956687
JG
1025 *chunk_id);
1026 if (!trace_chunk) {
1027 ERR("Failed to find trace chunk reference during creation of channel");
1028 goto end;
1029 }
1030 }
3bd1e081 1031
276b26d1 1032 channel = zmalloc(sizeof(*channel));
3bd1e081 1033 if (channel == NULL) {
7a57cf92 1034 PERROR("malloc struct lttng_consumer_channel");
3bd1e081
MD
1035 goto end;
1036 }
ffe60014
DG
1037
1038 channel->key = key;
3bd1e081 1039 channel->refcount = 0;
ffe60014 1040 channel->session_id = session_id;
1950109e 1041 channel->session_id_per_pid = session_id_per_pid;
ffe60014 1042 channel->relayd_id = relayd_id;
1624d5b7
JD
1043 channel->tracefile_size = tracefile_size;
1044 channel->tracefile_count = tracefile_count;
2bba9e53 1045 channel->monitor = monitor;
ecc48a90 1046 channel->live_timer_interval = live_timer_interval;
a2814ea7 1047 channel->is_live = is_in_live_session;
a9838785 1048 pthread_mutex_init(&channel->lock, NULL);
ec6ea7d0 1049 pthread_mutex_init(&channel->timer_lock, NULL);
ffe60014 1050
0c759fc9
DG
1051 switch (output) {
1052 case LTTNG_EVENT_SPLICE:
1053 channel->output = CONSUMER_CHANNEL_SPLICE;
1054 break;
1055 case LTTNG_EVENT_MMAP:
1056 channel->output = CONSUMER_CHANNEL_MMAP;
1057 break;
1058 default:
1059 assert(0);
1060 free(channel);
1061 channel = NULL;
1062 goto end;
1063 }
1064
07b86b52
JD
1065 /*
1066 * In monitor mode, the streams associated with the channel will be put in
1067 * a special list ONLY owned by this channel. So, the refcount is set to 1
1068 * here meaning that the channel itself has streams that are referenced.
1069 *
1070 * On a channel deletion, once the channel is no longer visible, the
1071 * refcount is decremented and checked for a zero value to delete it. With
1072 * streams in no monitor mode, it will now be safe to destroy the channel.
1073 */
1074 if (!channel->monitor) {
1075 channel->refcount = 1;
1076 }
1077
ffe60014
DG
1078 strncpy(channel->pathname, pathname, sizeof(channel->pathname));
1079 channel->pathname[sizeof(channel->pathname) - 1] = '\0';
1080
1081 strncpy(channel->name, name, sizeof(channel->name));
1082 channel->name[sizeof(channel->name) - 1] = '\0';
1083
3d071855
MD
1084 if (root_shm_path) {
1085 strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path));
1086 channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0';
1087 }
d7ba1388
MD
1088 if (shm_path) {
1089 strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path));
1090 channel->shm_path[sizeof(channel->shm_path) - 1] = '\0';
1091 }
1092
d88aee68 1093 lttng_ht_node_init_u64(&channel->node, channel->key);
5c3892a6
JG
1094 lttng_ht_node_init_u64(&channel->channels_by_session_id_ht_node,
1095 channel->session_id);
d8ef542d
MD
1096
1097 channel->wait_fd = -1;
ffe60014
DG
1098 CDS_INIT_LIST_HEAD(&channel->streams.head);
1099
d2956687
JG
1100 if (trace_chunk) {
1101 int ret = lttng_consumer_channel_set_trace_chunk(channel,
1102 trace_chunk);
1103 if (ret) {
1104 goto error;
1105 }
1106 }
1107
62a7b8ed 1108 DBG("Allocated channel (key %" PRIu64 ")", channel->key);
3bd1e081 1109
3bd1e081 1110end:
d2956687 1111 lttng_trace_chunk_put(trace_chunk);
3bd1e081 1112 return channel;
d2956687
JG
1113error:
1114 consumer_del_channel(channel);
1115 channel = NULL;
1116 goto end;
3bd1e081
MD
1117}
1118
1119/*
1120 * Add a channel to the global list protected by a mutex.
821fffb2 1121 *
b5a6470f 1122 * Always return 0 indicating success.
3bd1e081 1123 */
d8ef542d
MD
1124int consumer_add_channel(struct lttng_consumer_channel *channel,
1125 struct lttng_consumer_local_data *ctx)
3bd1e081 1126{
fa29bfbf 1127 pthread_mutex_lock(&the_consumer_data.lock);
a9838785 1128 pthread_mutex_lock(&channel->lock);
ec6ea7d0 1129 pthread_mutex_lock(&channel->timer_lock);
c77fc10a 1130
b5a6470f
DG
1131 /*
1132 * This gives us a guarantee that the channel we are about to add to the
1133 * channel hash table will be unique. See this function comment on the why
1134 * we need to steel the channel key at this stage.
1135 */
1136 steal_channel_key(channel->key);
c77fc10a 1137
b5a6470f 1138 rcu_read_lock();
fa29bfbf
SM
1139 lttng_ht_add_unique_u64(the_consumer_data.channel_ht, &channel->node);
1140 lttng_ht_add_u64(the_consumer_data.channels_by_session_id_ht,
5c3892a6 1141 &channel->channels_by_session_id_ht_node);
6065ceec 1142 rcu_read_unlock();
d2956687 1143 channel->is_published = true;
b5a6470f 1144
ec6ea7d0 1145 pthread_mutex_unlock(&channel->timer_lock);
a9838785 1146 pthread_mutex_unlock(&channel->lock);
fa29bfbf 1147 pthread_mutex_unlock(&the_consumer_data.lock);
702b1ea4 1148
b5a6470f 1149 if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) {
a0cbdd2e 1150 notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD);
d8ef542d 1151 }
b5a6470f
DG
1152
1153 return 0;
3bd1e081
MD
1154}
1155
1156/*
1157 * Allocate the pollfd structure and the local view of the out fds to avoid
1158 * doing a lookup in the linked list and concurrency issues when writing is
1159 * needed. Called with consumer_data.lock held.
1160 *
1161 * Returns the number of fds in the structures.
1162 */
ffe60014
DG
1163static int update_poll_array(struct lttng_consumer_local_data *ctx,
1164 struct pollfd **pollfd, struct lttng_consumer_stream **local_stream,
9a2fcf78 1165 struct lttng_ht *ht, int *nb_inactive_fd)
3bd1e081 1166{
3bd1e081 1167 int i = 0;
e4421fec
DG
1168 struct lttng_ht_iter iter;
1169 struct lttng_consumer_stream *stream;
3bd1e081 1170
ffe60014
DG
1171 assert(ctx);
1172 assert(ht);
1173 assert(pollfd);
1174 assert(local_stream);
1175
3bd1e081 1176 DBG("Updating poll fd array");
9a2fcf78 1177 *nb_inactive_fd = 0;
481d6c57 1178 rcu_read_lock();
43c34bc3 1179 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
8994307f
DG
1180 /*
1181 * Only active streams with an active end point can be added to the
1182 * poll set and local stream storage of the thread.
1183 *
1184 * There is a potential race here for endpoint_status to be updated
1185 * just after the check. However, this is OK since the stream(s) will
1186 * be deleted once the thread is notified that the end point state has
1187 * changed where this function will be called back again.
9a2fcf78
JD
1188 *
1189 * We track the number of inactive FDs because they still need to be
1190 * closed by the polling thread after a wakeup on the data_pipe or
1191 * metadata_pipe.
8994307f 1192 */
d2956687 1193 if (stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) {
9a2fcf78 1194 (*nb_inactive_fd)++;
3bd1e081
MD
1195 continue;
1196 }
7972aab2
DG
1197 /*
1198 * This clobbers way too much the debug output. Uncomment that if you
1199 * need it for debugging purposes.
7972aab2 1200 */
e4421fec 1201 (*pollfd)[i].fd = stream->wait_fd;
3bd1e081 1202 (*pollfd)[i].events = POLLIN | POLLPRI;
e4421fec 1203 local_stream[i] = stream;
3bd1e081
MD
1204 i++;
1205 }
481d6c57 1206 rcu_read_unlock();
3bd1e081
MD
1207
1208 /*
50f8ae69 1209 * Insert the consumer_data_pipe at the end of the array and don't
3bd1e081
MD
1210 * increment i so nb_fd is the number of real FD.
1211 */
acdb9057 1212 (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe);
509bb1cf 1213 (*pollfd)[i].events = POLLIN | POLLPRI;
02b3d176
DG
1214
1215 (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe);
1216 (*pollfd)[i + 1].events = POLLIN | POLLPRI;
3bd1e081
MD
1217 return i;
1218}
1219
1220/*
84382d49
MD
1221 * Poll on the should_quit pipe and the command socket return -1 on
1222 * error, 1 if should exit, 0 if data is available on the command socket
3bd1e081
MD
1223 */
1224int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll)
1225{
1226 int num_rdy;
1227
88f2b785 1228restart:
3bd1e081
MD
1229 num_rdy = poll(consumer_sockpoll, 2, -1);
1230 if (num_rdy == -1) {
88f2b785
MD
1231 /*
1232 * Restart interrupted system call.
1233 */
1234 if (errno == EINTR) {
1235 goto restart;
1236 }
7a57cf92 1237 PERROR("Poll error");
84382d49 1238 return -1;
3bd1e081 1239 }
509bb1cf 1240 if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) {
3bd1e081 1241 DBG("consumer_should_quit wake up");
84382d49 1242 return 1;
3bd1e081
MD
1243 }
1244 return 0;
3bd1e081
MD
1245}
1246
1247/*
1248 * Set the error socket.
1249 */
ffe60014
DG
1250void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx,
1251 int sock)
3bd1e081
MD
1252{
1253 ctx->consumer_error_socket = sock;
1254}
1255
1256/*
1257 * Set the command socket path.
1258 */
3bd1e081
MD
1259void lttng_consumer_set_command_sock_path(
1260 struct lttng_consumer_local_data *ctx, char *sock)
1261{
1262 ctx->consumer_command_sock_path = sock;
1263}
1264
1265/*
1266 * Send return code to the session daemon.
1267 * If the socket is not defined, we return 0, it is not a fatal error
1268 */
ffe60014 1269int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd)
3bd1e081
MD
1270{
1271 if (ctx->consumer_error_socket > 0) {
1272 return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd,
1273 sizeof(enum lttcomm_sessiond_command));
1274 }
1275
1276 return 0;
1277}
1278
1279/*
228b5bf7
DG
1280 * Close all the tracefiles and stream fds and MUST be called when all
1281 * instances are destroyed i.e. when all threads were joined and are ended.
3bd1e081
MD
1282 */
1283void lttng_consumer_cleanup(void)
1284{
e4421fec 1285 struct lttng_ht_iter iter;
ffe60014 1286 struct lttng_consumer_channel *channel;
e10aec8f 1287 unsigned int trace_chunks_left;
6065ceec
DG
1288
1289 rcu_read_lock();
3bd1e081 1290
fa29bfbf
SM
1291 cds_lfht_for_each_entry(the_consumer_data.channel_ht->ht, &iter.iter,
1292 channel, node.node) {
702b1ea4 1293 consumer_del_channel(channel);
3bd1e081 1294 }
6065ceec
DG
1295
1296 rcu_read_unlock();
d6ce1df2 1297
fa29bfbf
SM
1298 lttng_ht_destroy(the_consumer_data.channel_ht);
1299 lttng_ht_destroy(the_consumer_data.channels_by_session_id_ht);
228b5bf7
DG
1300
1301 cleanup_relayd_ht();
1302
fa29bfbf 1303 lttng_ht_destroy(the_consumer_data.stream_per_chan_id_ht);
d8ef542d 1304
228b5bf7
DG
1305 /*
1306 * This HT contains streams that are freed by either the metadata thread or
1307 * the data thread so we do *nothing* on the hash table and simply destroy
1308 * it.
1309 */
fa29bfbf 1310 lttng_ht_destroy(the_consumer_data.stream_list_ht);
28cc88f3 1311
e10aec8f
MD
1312 /*
1313 * Trace chunks in the registry may still exist if the session
1314 * daemon has encountered an internal error and could not
1315 * tear down its sessions and/or trace chunks properly.
1316 *
1317 * Release the session daemon's implicit reference to any remaining
1318 * trace chunk and print an error if any trace chunk was found. Note
1319 * that there are _no_ legitimate cases for trace chunks to be left,
1320 * it is a leak. However, it can happen following a crash of the
1321 * session daemon and not emptying the registry would cause an assertion
1322 * to hit.
1323 */
1324 trace_chunks_left = lttng_trace_chunk_registry_put_each_chunk(
fa29bfbf 1325 the_consumer_data.chunk_registry);
e10aec8f
MD
1326 if (trace_chunks_left) {
1327 ERR("%u trace chunks are leaked by lttng-consumerd. "
1328 "This can be caused by an internal error of the session daemon.",
1329 trace_chunks_left);
1330 }
1331 /* Run all callbacks freeing each chunk. */
1332 rcu_barrier();
fa29bfbf 1333 lttng_trace_chunk_registry_destroy(the_consumer_data.chunk_registry);
3bd1e081
MD
1334}
1335
1336/*
1337 * Called from signal handler.
1338 */
1339void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx)
1340{
6cd525e8
MD
1341 ssize_t ret;
1342
10211f5c 1343 CMM_STORE_SHARED(consumer_quit, 1);
6cd525e8
MD
1344 ret = lttng_write(ctx->consumer_should_quit[1], "4", 1);
1345 if (ret < 1) {
7a57cf92 1346 PERROR("write consumer quit");
3bd1e081 1347 }
ab1027f4
DG
1348
1349 DBG("Consumer flag that it should quit");
3bd1e081
MD
1350}
1351
5199ffc4
JG
1352
1353/*
1354 * Flush pending writes to trace output disk file.
1355 */
1356static
00e2e675
DG
1357void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream,
1358 off_t orig_offset)
3bd1e081 1359{
c7a78aab 1360 int ret;
3bd1e081
MD
1361 int outfd = stream->out_fd;
1362
1363 /*
1364 * This does a blocking write-and-wait on any page that belongs to the
1365 * subbuffer prior to the one we just wrote.
1366 * Don't care about error values, as these are just hints and ways to
1367 * limit the amount of page cache used.
1368 */
ffe60014 1369 if (orig_offset < stream->max_sb_size) {
3bd1e081
MD
1370 return;
1371 }
ffe60014
DG
1372 lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size,
1373 stream->max_sb_size,
3bd1e081
MD
1374 SYNC_FILE_RANGE_WAIT_BEFORE
1375 | SYNC_FILE_RANGE_WRITE
1376 | SYNC_FILE_RANGE_WAIT_AFTER);
1377 /*
1378 * Give hints to the kernel about how we access the file:
1379 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1380 * we write it.
1381 *
1382 * We need to call fadvise again after the file grows because the
1383 * kernel does not seem to apply fadvise to non-existing parts of the
1384 * file.
1385 *
1386 * Call fadvise _after_ having waited for the page writeback to
1387 * complete because the dirty page writeback semantic is not well
1388 * defined. So it can be expected to lead to lower throughput in
1389 * streaming.
1390 */
c7a78aab 1391 ret = posix_fadvise(outfd, orig_offset - stream->max_sb_size,
ffe60014 1392 stream->max_sb_size, POSIX_FADV_DONTNEED);
a0d0e127 1393 if (ret && ret != -ENOSYS) {
a74a5f4a
JG
1394 errno = ret;
1395 PERROR("posix_fadvise on fd %i", outfd);
c7a78aab 1396 }
3bd1e081
MD
1397}
1398
1399/*
1400 * Initialise the necessary environnement :
1401 * - create a new context
1402 * - create the poll_pipe
1403 * - create the should_quit pipe (for signal handler)
1404 * - create the thread pipe (for splice)
1405 *
1406 * Takes a function pointer as argument, this function is called when data is
1407 * available on a buffer. This function is responsible to do the
1408 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1409 * buffer configuration and then kernctl_put_next_subbuf at the end.
1410 *
1411 * Returns a pointer to the new context or NULL on error.
1412 */
1413struct lttng_consumer_local_data *lttng_consumer_create(
1414 enum lttng_consumer_type type,
4078b776 1415 ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream,
6f9449c2 1416 struct lttng_consumer_local_data *ctx, bool locked_by_caller),
3bd1e081
MD
1417 int (*recv_channel)(struct lttng_consumer_channel *channel),
1418 int (*recv_stream)(struct lttng_consumer_stream *stream),
30319bcb 1419 int (*update_stream)(uint64_t stream_key, uint32_t state))
3bd1e081 1420{
d8ef542d 1421 int ret;
3bd1e081
MD
1422 struct lttng_consumer_local_data *ctx;
1423
fa29bfbf
SM
1424 assert(the_consumer_data.type == LTTNG_CONSUMER_UNKNOWN ||
1425 the_consumer_data.type == type);
1426 the_consumer_data.type = type;
3bd1e081 1427
effcf122 1428 ctx = zmalloc(sizeof(struct lttng_consumer_local_data));
3bd1e081 1429 if (ctx == NULL) {
7a57cf92 1430 PERROR("allocating context");
3bd1e081
MD
1431 goto error;
1432 }
1433
1434 ctx->consumer_error_socket = -1;
331744e3 1435 ctx->consumer_metadata_socket = -1;
75d83e50 1436 pthread_mutex_init(&ctx->metadata_socket_lock, NULL);
3bd1e081
MD
1437 /* assign the callbacks */
1438 ctx->on_buffer_ready = buffer_ready;
1439 ctx->on_recv_channel = recv_channel;
1440 ctx->on_recv_stream = recv_stream;
1441 ctx->on_update_stream = update_stream;
1442
acdb9057
DG
1443 ctx->consumer_data_pipe = lttng_pipe_open(0);
1444 if (!ctx->consumer_data_pipe) {
3bd1e081
MD
1445 goto error_poll_pipe;
1446 }
1447
02b3d176
DG
1448 ctx->consumer_wakeup_pipe = lttng_pipe_open(0);
1449 if (!ctx->consumer_wakeup_pipe) {
1450 goto error_wakeup_pipe;
1451 }
1452
3bd1e081
MD
1453 ret = pipe(ctx->consumer_should_quit);
1454 if (ret < 0) {
7a57cf92 1455 PERROR("Error creating recv pipe");
3bd1e081
MD
1456 goto error_quit_pipe;
1457 }
1458
d8ef542d
MD
1459 ret = pipe(ctx->consumer_channel_pipe);
1460 if (ret < 0) {
1461 PERROR("Error creating channel pipe");
1462 goto error_channel_pipe;
1463 }
1464
13886d2d
DG
1465 ctx->consumer_metadata_pipe = lttng_pipe_open(0);
1466 if (!ctx->consumer_metadata_pipe) {
fb3a43a9
DG
1467 goto error_metadata_pipe;
1468 }
3bd1e081 1469
e9404c27
JG
1470 ctx->channel_monitor_pipe = -1;
1471
fb3a43a9 1472 return ctx;
3bd1e081 1473
fb3a43a9 1474error_metadata_pipe:
d8ef542d
MD
1475 utils_close_pipe(ctx->consumer_channel_pipe);
1476error_channel_pipe:
d8ef542d 1477 utils_close_pipe(ctx->consumer_should_quit);
3bd1e081 1478error_quit_pipe:
02b3d176
DG
1479 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
1480error_wakeup_pipe:
acdb9057 1481 lttng_pipe_destroy(ctx->consumer_data_pipe);
3bd1e081
MD
1482error_poll_pipe:
1483 free(ctx);
1484error:
1485 return NULL;
1486}
1487
282dadbc
MD
1488/*
1489 * Iterate over all streams of the hashtable and free them properly.
1490 */
1491static void destroy_data_stream_ht(struct lttng_ht *ht)
1492{
1493 struct lttng_ht_iter iter;
1494 struct lttng_consumer_stream *stream;
1495
1496 if (ht == NULL) {
1497 return;
1498 }
1499
1500 rcu_read_lock();
1501 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1502 /*
1503 * Ignore return value since we are currently cleaning up so any error
1504 * can't be handled.
1505 */
1506 (void) consumer_del_stream(stream, ht);
1507 }
1508 rcu_read_unlock();
1509
1510 lttng_ht_destroy(ht);
1511}
1512
1513/*
1514 * Iterate over all streams of the metadata hashtable and free them
1515 * properly.
1516 */
1517static void destroy_metadata_stream_ht(struct lttng_ht *ht)
1518{
1519 struct lttng_ht_iter iter;
1520 struct lttng_consumer_stream *stream;
1521
1522 if (ht == NULL) {
1523 return;
1524 }
1525
1526 rcu_read_lock();
1527 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1528 /*
1529 * Ignore return value since we are currently cleaning up so any error
1530 * can't be handled.
1531 */
1532 (void) consumer_del_metadata_stream(stream, ht);
1533 }
1534 rcu_read_unlock();
1535
1536 lttng_ht_destroy(ht);
1537}
1538
3bd1e081
MD
1539/*
1540 * Close all fds associated with the instance and free the context.
1541 */
1542void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx)
1543{
4c462e79
MD
1544 int ret;
1545
ab1027f4
DG
1546 DBG("Consumer destroying it. Closing everything.");
1547
4f2e75b9
DG
1548 if (!ctx) {
1549 return;
1550 }
1551
282dadbc
MD
1552 destroy_data_stream_ht(data_ht);
1553 destroy_metadata_stream_ht(metadata_ht);
1554
4c462e79
MD
1555 ret = close(ctx->consumer_error_socket);
1556 if (ret) {
1557 PERROR("close");
1558 }
331744e3
JD
1559 ret = close(ctx->consumer_metadata_socket);
1560 if (ret) {
1561 PERROR("close");
1562 }
d8ef542d 1563 utils_close_pipe(ctx->consumer_channel_pipe);
acdb9057 1564 lttng_pipe_destroy(ctx->consumer_data_pipe);
13886d2d 1565 lttng_pipe_destroy(ctx->consumer_metadata_pipe);
02b3d176 1566 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
d8ef542d 1567 utils_close_pipe(ctx->consumer_should_quit);
fb3a43a9 1568
3bd1e081
MD
1569 unlink(ctx->consumer_command_sock_path);
1570 free(ctx);
1571}
1572
6197aea7
DG
1573/*
1574 * Write the metadata stream id on the specified file descriptor.
1575 */
1576static int write_relayd_metadata_id(int fd,
1577 struct lttng_consumer_stream *stream,
239f61af 1578 unsigned long padding)
6197aea7 1579{
6cd525e8 1580 ssize_t ret;
1d4dfdef 1581 struct lttcomm_relayd_metadata_payload hdr;
6197aea7 1582
1d4dfdef
DG
1583 hdr.stream_id = htobe64(stream->relayd_stream_id);
1584 hdr.padding_size = htobe32(padding);
6cd525e8
MD
1585 ret = lttng_write(fd, (void *) &hdr, sizeof(hdr));
1586 if (ret < sizeof(hdr)) {
d7b75ec8 1587 /*
6f04ed72 1588 * This error means that the fd's end is closed so ignore the PERROR
d7b75ec8
DG
1589 * not to clubber the error output since this can happen in a normal
1590 * code path.
1591 */
1592 if (errno != EPIPE) {
1593 PERROR("write metadata stream id");
1594 }
1595 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno);
534d2592
DG
1596 /*
1597 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1598 * handle writting the missing part so report that as an error and
1599 * don't lie to the caller.
1600 */
1601 ret = -1;
6197aea7
DG
1602 goto end;
1603 }
1d4dfdef
DG
1604 DBG("Metadata stream id %" PRIu64 " with padding %lu written before data",
1605 stream->relayd_stream_id, padding);
6197aea7
DG
1606
1607end:
6cd525e8 1608 return (int) ret;
6197aea7
DG
1609}
1610
3bd1e081 1611/*
09e26845
DG
1612 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1613 * core function for writing trace buffers to either the local filesystem or
1614 * the network.
1615 *
d2956687 1616 * It must be called with the stream and the channel lock held.
79d4ffb7 1617 *
09e26845 1618 * Careful review MUST be put if any changes occur!
3bd1e081
MD
1619 *
1620 * Returns the number of bytes written
1621 */
4078b776 1622ssize_t lttng_consumer_on_read_subbuffer_mmap(
128708c3 1623 struct lttng_consumer_stream *stream,
fd424d99 1624 const struct lttng_buffer_view *buffer,
6f9449c2 1625 unsigned long padding)
3bd1e081 1626{
994ab360 1627 ssize_t ret = 0;
f02e1e8a
DG
1628 off_t orig_offset = stream->out_fd_offset;
1629 /* Default is on the disk */
1630 int outfd = stream->out_fd;
f02e1e8a 1631 struct consumer_relayd_sock_pair *relayd = NULL;
8994307f 1632 unsigned int relayd_hang_up = 0;
fd424d99
JG
1633 const size_t subbuf_content_size = buffer->size - padding;
1634 size_t write_len;
f02e1e8a
DG
1635
1636 /* RCU lock for the relayd pointer */
1637 rcu_read_lock();
7fd975c5 1638 assert(stream->net_seq_idx != (uint64_t) -1ULL ||
948411cd 1639 stream->trace_chunk);
d2956687 1640
f02e1e8a 1641 /* Flag that the current stream if set for network streaming. */
da009f2c 1642 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1643 relayd = consumer_find_relayd(stream->net_seq_idx);
1644 if (relayd == NULL) {
56591bac 1645 ret = -EPIPE;
f02e1e8a
DG
1646 goto end;
1647 }
1648 }
1649
f02e1e8a
DG
1650 /* Handle stream on the relayd if the output is on the network */
1651 if (relayd) {
fd424d99 1652 unsigned long netlen = subbuf_content_size;
f02e1e8a
DG
1653
1654 /*
1655 * Lock the control socket for the complete duration of the function
1656 * since from this point on we will use the socket.
1657 */
1658 if (stream->metadata_flag) {
1659 /* Metadata requires the control socket. */
1660 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
93ec662e
JD
1661 if (stream->reset_metadata_flag) {
1662 ret = relayd_reset_metadata(&relayd->control_sock,
1663 stream->relayd_stream_id,
1664 stream->metadata_version);
1665 if (ret < 0) {
1666 relayd_hang_up = 1;
1667 goto write_error;
1668 }
1669 stream->reset_metadata_flag = 0;
1670 }
1d4dfdef 1671 netlen += sizeof(struct lttcomm_relayd_metadata_payload);
f02e1e8a
DG
1672 }
1673
1d4dfdef 1674 ret = write_relayd_stream_header(stream, netlen, padding, relayd);
994ab360
DG
1675 if (ret < 0) {
1676 relayd_hang_up = 1;
1677 goto write_error;
1678 }
1679 /* Use the returned socket. */
1680 outfd = ret;
f02e1e8a 1681
994ab360
DG
1682 /* Write metadata stream id before payload */
1683 if (stream->metadata_flag) {
239f61af 1684 ret = write_relayd_metadata_id(outfd, stream, padding);
994ab360 1685 if (ret < 0) {
8994307f
DG
1686 relayd_hang_up = 1;
1687 goto write_error;
1688 }
f02e1e8a 1689 }
1624d5b7 1690
fd424d99
JG
1691 write_len = subbuf_content_size;
1692 } else {
1693 /* No streaming; we have to write the full padding. */
93ec662e
JD
1694 if (stream->metadata_flag && stream->reset_metadata_flag) {
1695 ret = utils_truncate_stream_file(stream->out_fd, 0);
1696 if (ret < 0) {
1697 ERR("Reset metadata file");
1698 goto end;
1699 }
1700 stream->reset_metadata_flag = 0;
1701 }
1702
1624d5b7
JD
1703 /*
1704 * Check if we need to change the tracefile before writing the packet.
1705 */
1706 if (stream->chan->tracefile_size > 0 &&
fd424d99 1707 (stream->tracefile_size_current + buffer->size) >
1624d5b7 1708 stream->chan->tracefile_size) {
d2956687
JG
1709 ret = consumer_stream_rotate_output_files(stream);
1710 if (ret) {
1624d5b7
JD
1711 goto end;
1712 }
309167d2 1713 outfd = stream->out_fd;
a1ae300f 1714 orig_offset = 0;
1624d5b7 1715 }
fd424d99 1716 stream->tracefile_size_current += buffer->size;
fd424d99 1717 write_len = buffer->size;
f02e1e8a
DG
1718 }
1719
d02b8372
DG
1720 /*
1721 * This call guarantee that len or less is returned. It's impossible to
1722 * receive a ret value that is bigger than len.
1723 */
fd424d99 1724 ret = lttng_write(outfd, buffer->data, write_len);
e2d1190b 1725 DBG("Consumer mmap write() ret %zd (len %zu)", ret, write_len);
fd424d99 1726 if (ret < 0 || ((size_t) ret != write_len)) {
d02b8372
DG
1727 /*
1728 * Report error to caller if nothing was written else at least send the
1729 * amount written.
1730 */
1731 if (ret < 0) {
994ab360 1732 ret = -errno;
f02e1e8a 1733 }
994ab360 1734 relayd_hang_up = 1;
f02e1e8a 1735
d02b8372 1736 /* Socket operation failed. We consider the relayd dead */
fcf0f774 1737 if (errno == EPIPE) {
d02b8372
DG
1738 /*
1739 * This is possible if the fd is closed on the other side
1740 * (outfd) or any write problem. It can be verbose a bit for a
1741 * normal execution if for instance the relayd is stopped
1742 * abruptly. This can happen so set this to a DBG statement.
1743 */
1744 DBG("Consumer mmap write detected relayd hang up");
994ab360
DG
1745 } else {
1746 /* Unhandled error, print it and stop function right now. */
fd424d99
JG
1747 PERROR("Error in write mmap (ret %zd != write_len %zu)", ret,
1748 write_len);
f02e1e8a 1749 }
994ab360 1750 goto write_error;
d02b8372
DG
1751 }
1752 stream->output_written += ret;
d02b8372
DG
1753
1754 /* This call is useless on a socket so better save a syscall. */
1755 if (!relayd) {
1756 /* This won't block, but will start writeout asynchronously */
fd424d99 1757 lttng_sync_file_range(outfd, stream->out_fd_offset, write_len,
d02b8372 1758 SYNC_FILE_RANGE_WRITE);
fd424d99 1759 stream->out_fd_offset += write_len;
f5dbe415 1760 lttng_consumer_sync_trace_file(stream, orig_offset);
f02e1e8a 1761 }
f02e1e8a 1762
8994307f
DG
1763write_error:
1764 /*
1765 * This is a special case that the relayd has closed its socket. Let's
1766 * cleanup the relayd object and all associated streams.
1767 */
1768 if (relayd && relayd_hang_up) {
9276e5c8
JR
1769 ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
1770 lttng_consumer_cleanup_relayd(relayd);
8994307f
DG
1771 }
1772
f02e1e8a
DG
1773end:
1774 /* Unlock only if ctrl socket used */
1775 if (relayd && stream->metadata_flag) {
1776 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1777 }
1778
1779 rcu_read_unlock();
994ab360 1780 return ret;
3bd1e081
MD
1781}
1782
1783/*
1784 * Splice the data from the ring buffer to the tracefile.
1785 *
79d4ffb7
DG
1786 * It must be called with the stream lock held.
1787 *
3bd1e081
MD
1788 * Returns the number of bytes spliced.
1789 */
4078b776 1790ssize_t lttng_consumer_on_read_subbuffer_splice(
3bd1e081 1791 struct lttng_consumer_local_data *ctx,
1d4dfdef 1792 struct lttng_consumer_stream *stream, unsigned long len,
6f9449c2 1793 unsigned long padding)
3bd1e081 1794{
f02e1e8a
DG
1795 ssize_t ret = 0, written = 0, ret_splice = 0;
1796 loff_t offset = 0;
1797 off_t orig_offset = stream->out_fd_offset;
1798 int fd = stream->wait_fd;
1799 /* Default is on the disk */
1800 int outfd = stream->out_fd;
f02e1e8a 1801 struct consumer_relayd_sock_pair *relayd = NULL;
fb3a43a9 1802 int *splice_pipe;
8994307f 1803 unsigned int relayd_hang_up = 0;
f02e1e8a 1804
fa29bfbf 1805 switch (the_consumer_data.type) {
3bd1e081 1806 case LTTNG_CONSUMER_KERNEL:
f02e1e8a 1807 break;
7753dea8
MD
1808 case LTTNG_CONSUMER32_UST:
1809 case LTTNG_CONSUMER64_UST:
f02e1e8a 1810 /* Not supported for user space tracing */
3bd1e081
MD
1811 return -ENOSYS;
1812 default:
1813 ERR("Unknown consumer_data type");
1814 assert(0);
3bd1e081
MD
1815 }
1816
f02e1e8a
DG
1817 /* RCU lock for the relayd pointer */
1818 rcu_read_lock();
1819
1820 /* Flag that the current stream if set for network streaming. */
da009f2c 1821 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1822 relayd = consumer_find_relayd(stream->net_seq_idx);
1823 if (relayd == NULL) {
ad0b0d23 1824 written = -ret;
f02e1e8a
DG
1825 goto end;
1826 }
1827 }
a2361a61 1828 splice_pipe = stream->splice_pipe;
fb3a43a9 1829
f02e1e8a 1830 /* Write metadata stream id before payload */
1d4dfdef 1831 if (relayd) {
ad0b0d23 1832 unsigned long total_len = len;
f02e1e8a 1833
1d4dfdef
DG
1834 if (stream->metadata_flag) {
1835 /*
1836 * Lock the control socket for the complete duration of the function
1837 * since from this point on we will use the socket.
1838 */
1839 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1840
93ec662e
JD
1841 if (stream->reset_metadata_flag) {
1842 ret = relayd_reset_metadata(&relayd->control_sock,
1843 stream->relayd_stream_id,
1844 stream->metadata_version);
1845 if (ret < 0) {
1846 relayd_hang_up = 1;
1847 goto write_error;
1848 }
1849 stream->reset_metadata_flag = 0;
1850 }
239f61af 1851 ret = write_relayd_metadata_id(splice_pipe[1], stream,
1d4dfdef
DG
1852 padding);
1853 if (ret < 0) {
1854 written = ret;
ad0b0d23
DG
1855 relayd_hang_up = 1;
1856 goto write_error;
1d4dfdef
DG
1857 }
1858
1859 total_len += sizeof(struct lttcomm_relayd_metadata_payload);
1860 }
1861
1862 ret = write_relayd_stream_header(stream, total_len, padding, relayd);
ad0b0d23
DG
1863 if (ret < 0) {
1864 written = ret;
1865 relayd_hang_up = 1;
1866 goto write_error;
f02e1e8a 1867 }
ad0b0d23
DG
1868 /* Use the returned socket. */
1869 outfd = ret;
1d4dfdef
DG
1870 } else {
1871 /* No streaming, we have to set the len with the full padding */
1872 len += padding;
1624d5b7 1873
93ec662e
JD
1874 if (stream->metadata_flag && stream->reset_metadata_flag) {
1875 ret = utils_truncate_stream_file(stream->out_fd, 0);
1876 if (ret < 0) {
1877 ERR("Reset metadata file");
1878 goto end;
1879 }
1880 stream->reset_metadata_flag = 0;
1881 }
1624d5b7
JD
1882 /*
1883 * Check if we need to change the tracefile before writing the packet.
1884 */
1885 if (stream->chan->tracefile_size > 0 &&
1886 (stream->tracefile_size_current + len) >
1887 stream->chan->tracefile_size) {
d2956687 1888 ret = consumer_stream_rotate_output_files(stream);
1624d5b7 1889 if (ret < 0) {
ad0b0d23 1890 written = ret;
1624d5b7
JD
1891 goto end;
1892 }
309167d2 1893 outfd = stream->out_fd;
a1ae300f 1894 orig_offset = 0;
1624d5b7
JD
1895 }
1896 stream->tracefile_size_current += len;
f02e1e8a
DG
1897 }
1898
1899 while (len > 0) {
1d4dfdef
DG
1900 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1901 (unsigned long)offset, len, fd, splice_pipe[1]);
fb3a43a9 1902 ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len,
f02e1e8a
DG
1903 SPLICE_F_MOVE | SPLICE_F_MORE);
1904 DBG("splice chan to pipe, ret %zd", ret_splice);
1905 if (ret_splice < 0) {
d02b8372 1906 ret = errno;
ad0b0d23 1907 written = -ret;
d02b8372 1908 PERROR("Error in relay splice");
f02e1e8a
DG
1909 goto splice_error;
1910 }
1911
1912 /* Handle stream on the relayd if the output is on the network */
ad0b0d23
DG
1913 if (relayd && stream->metadata_flag) {
1914 size_t metadata_payload_size =
1915 sizeof(struct lttcomm_relayd_metadata_payload);
1916
1917 /* Update counter to fit the spliced data */
1918 ret_splice += metadata_payload_size;
1919 len += metadata_payload_size;
1920 /*
1921 * We do this so the return value can match the len passed as
1922 * argument to this function.
1923 */
1924 written -= metadata_payload_size;
f02e1e8a
DG
1925 }
1926
1927 /* Splice data out */
fb3a43a9 1928 ret_splice = splice(splice_pipe[0], NULL, outfd, NULL,
f02e1e8a 1929 ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE);
a2361a61
JD
1930 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1931 outfd, ret_splice);
f02e1e8a 1932 if (ret_splice < 0) {
d02b8372 1933 ret = errno;
ad0b0d23
DG
1934 written = -ret;
1935 relayd_hang_up = 1;
1936 goto write_error;
f02e1e8a 1937 } else if (ret_splice > len) {
d02b8372
DG
1938 /*
1939 * We don't expect this code path to be executed but you never know
1940 * so this is an extra protection agains a buggy splice().
1941 */
f02e1e8a 1942 ret = errno;
ad0b0d23 1943 written += ret_splice;
d02b8372
DG
1944 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice,
1945 len);
f02e1e8a 1946 goto splice_error;
d02b8372
DG
1947 } else {
1948 /* All good, update current len and continue. */
1949 len -= ret_splice;
f02e1e8a 1950 }
f02e1e8a
DG
1951
1952 /* This call is useless on a socket so better save a syscall. */
1953 if (!relayd) {
1954 /* This won't block, but will start writeout asynchronously */
1955 lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice,
1956 SYNC_FILE_RANGE_WRITE);
1957 stream->out_fd_offset += ret_splice;
1958 }
e5d1a9b3 1959 stream->output_written += ret_splice;
f02e1e8a
DG
1960 written += ret_splice;
1961 }
f5dbe415
JG
1962 if (!relayd) {
1963 lttng_consumer_sync_trace_file(stream, orig_offset);
1964 }
f02e1e8a
DG
1965 goto end;
1966
8994307f
DG
1967write_error:
1968 /*
1969 * This is a special case that the relayd has closed its socket. Let's
1970 * cleanup the relayd object and all associated streams.
1971 */
1972 if (relayd && relayd_hang_up) {
9276e5c8
JR
1973 ERR("Relayd hangup. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
1974 lttng_consumer_cleanup_relayd(relayd);
8994307f
DG
1975 /* Skip splice error so the consumer does not fail */
1976 goto end;
1977 }
1978
f02e1e8a
DG
1979splice_error:
1980 /* send the appropriate error description to sessiond */
1981 switch (ret) {
f02e1e8a 1982 case EINVAL:
f73fabfd 1983 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL);
f02e1e8a
DG
1984 break;
1985 case ENOMEM:
f73fabfd 1986 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM);
f02e1e8a
DG
1987 break;
1988 case ESPIPE:
f73fabfd 1989 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE);
f02e1e8a
DG
1990 break;
1991 }
1992
1993end:
1994 if (relayd && stream->metadata_flag) {
1995 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1996 }
1997
1998 rcu_read_unlock();
1999 return written;
3bd1e081
MD
2000}
2001
15055ce5
JD
2002/*
2003 * Sample the snapshot positions for a specific fd
2004 *
2005 * Returns 0 on success, < 0 on error
2006 */
2007int lttng_consumer_sample_snapshot_positions(struct lttng_consumer_stream *stream)
2008{
fa29bfbf 2009 switch (the_consumer_data.type) {
15055ce5
JD
2010 case LTTNG_CONSUMER_KERNEL:
2011 return lttng_kconsumer_sample_snapshot_positions(stream);
2012 case LTTNG_CONSUMER32_UST:
2013 case LTTNG_CONSUMER64_UST:
2014 return lttng_ustconsumer_sample_snapshot_positions(stream);
2015 default:
2016 ERR("Unknown consumer_data type");
2017 assert(0);
2018 return -ENOSYS;
2019 }
2020}
3bd1e081
MD
2021/*
2022 * Take a snapshot for a specific fd
2023 *
2024 * Returns 0 on success, < 0 on error
2025 */
ffe60014 2026int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream)
3bd1e081 2027{
fa29bfbf 2028 switch (the_consumer_data.type) {
3bd1e081 2029 case LTTNG_CONSUMER_KERNEL:
ffe60014 2030 return lttng_kconsumer_take_snapshot(stream);
7753dea8
MD
2031 case LTTNG_CONSUMER32_UST:
2032 case LTTNG_CONSUMER64_UST:
ffe60014 2033 return lttng_ustconsumer_take_snapshot(stream);
3bd1e081
MD
2034 default:
2035 ERR("Unknown consumer_data type");
2036 assert(0);
2037 return -ENOSYS;
2038 }
3bd1e081
MD
2039}
2040
2041/*
2042 * Get the produced position
2043 *
2044 * Returns 0 on success, < 0 on error
2045 */
ffe60014 2046int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
3bd1e081
MD
2047 unsigned long *pos)
2048{
fa29bfbf 2049 switch (the_consumer_data.type) {
3bd1e081 2050 case LTTNG_CONSUMER_KERNEL:
ffe60014 2051 return lttng_kconsumer_get_produced_snapshot(stream, pos);
7753dea8
MD
2052 case LTTNG_CONSUMER32_UST:
2053 case LTTNG_CONSUMER64_UST:
ffe60014 2054 return lttng_ustconsumer_get_produced_snapshot(stream, pos);
3bd1e081
MD
2055 default:
2056 ERR("Unknown consumer_data type");
2057 assert(0);
2058 return -ENOSYS;
2059 }
2060}
2061
15055ce5
JD
2062/*
2063 * Get the consumed position (free-running counter position in bytes).
2064 *
2065 * Returns 0 on success, < 0 on error
2066 */
2067int lttng_consumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
2068 unsigned long *pos)
2069{
fa29bfbf 2070 switch (the_consumer_data.type) {
15055ce5
JD
2071 case LTTNG_CONSUMER_KERNEL:
2072 return lttng_kconsumer_get_consumed_snapshot(stream, pos);
2073 case LTTNG_CONSUMER32_UST:
2074 case LTTNG_CONSUMER64_UST:
2075 return lttng_ustconsumer_get_consumed_snapshot(stream, pos);
2076 default:
2077 ERR("Unknown consumer_data type");
2078 assert(0);
2079 return -ENOSYS;
2080 }
2081}
2082
3bd1e081
MD
2083int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx,
2084 int sock, struct pollfd *consumer_sockpoll)
2085{
fa29bfbf 2086 switch (the_consumer_data.type) {
3bd1e081
MD
2087 case LTTNG_CONSUMER_KERNEL:
2088 return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
7753dea8
MD
2089 case LTTNG_CONSUMER32_UST:
2090 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
2091 return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
2092 default:
2093 ERR("Unknown consumer_data type");
2094 assert(0);
2095 return -ENOSYS;
2096 }
2097}
2098
1f8d1c14 2099static
6d574024 2100void lttng_consumer_close_all_metadata(void)
d88aee68 2101{
fa29bfbf 2102 switch (the_consumer_data.type) {
d88aee68
DG
2103 case LTTNG_CONSUMER_KERNEL:
2104 /*
2105 * The Kernel consumer has a different metadata scheme so we don't
2106 * close anything because the stream will be closed by the session
2107 * daemon.
2108 */
2109 break;
2110 case LTTNG_CONSUMER32_UST:
2111 case LTTNG_CONSUMER64_UST:
2112 /*
2113 * Close all metadata streams. The metadata hash table is passed and
2114 * this call iterates over it by closing all wakeup fd. This is safe
2115 * because at this point we are sure that the metadata producer is
2116 * either dead or blocked.
2117 */
6d574024 2118 lttng_ustconsumer_close_all_metadata(metadata_ht);
d88aee68
DG
2119 break;
2120 default:
2121 ERR("Unknown consumer_data type");
2122 assert(0);
2123 }
2124}
2125
fb3a43a9
DG
2126/*
2127 * Clean up a metadata stream and free its memory.
2128 */
e316aad5
DG
2129void consumer_del_metadata_stream(struct lttng_consumer_stream *stream,
2130 struct lttng_ht *ht)
fb3a43a9 2131{
a6ef8ee6
JG
2132 struct lttng_consumer_channel *channel = NULL;
2133 bool free_channel = false;
fb3a43a9
DG
2134
2135 assert(stream);
2136 /*
2137 * This call should NEVER receive regular stream. It must always be
2138 * metadata stream and this is crucial for data structure synchronization.
2139 */
2140 assert(stream->metadata_flag);
2141
e316aad5
DG
2142 DBG3("Consumer delete metadata stream %d", stream->wait_fd);
2143
fa29bfbf 2144 pthread_mutex_lock(&the_consumer_data.lock);
a6ef8ee6
JG
2145 /*
2146 * Note that this assumes that a stream's channel is never changed and
2147 * that the stream's lock doesn't need to be taken to sample its
2148 * channel.
2149 */
2150 channel = stream->chan;
2151 pthread_mutex_lock(&channel->lock);
3dad2c0f 2152 pthread_mutex_lock(&stream->lock);
a6ef8ee6 2153 if (channel->metadata_cache) {
081424af 2154 /* Only applicable to userspace consumers. */
a6ef8ee6 2155 pthread_mutex_lock(&channel->metadata_cache->lock);
081424af 2156 }
8994307f 2157
6d574024
DG
2158 /* Remove any reference to that stream. */
2159 consumer_stream_delete(stream, ht);
ca22feea 2160
6d574024
DG
2161 /* Close down everything including the relayd if one. */
2162 consumer_stream_close(stream);
2163 /* Destroy tracer buffers of the stream. */
2164 consumer_stream_destroy_buffers(stream);
fb3a43a9
DG
2165
2166 /* Atomically decrement channel refcount since other threads can use it. */
a6ef8ee6
JG
2167 if (!uatomic_sub_return(&channel->refcount, 1)
2168 && !uatomic_read(&channel->nb_init_stream_left)) {
c30aaa51 2169 /* Go for channel deletion! */
a6ef8ee6 2170 free_channel = true;
fb3a43a9 2171 }
a6ef8ee6 2172 stream->chan = NULL;
fb3a43a9 2173
73811ecc
DG
2174 /*
2175 * Nullify the stream reference so it is not used after deletion. The
6d574024
DG
2176 * channel lock MUST be acquired before being able to check for a NULL
2177 * pointer value.
73811ecc 2178 */
a6ef8ee6 2179 channel->metadata_stream = NULL;
73811ecc 2180
a6ef8ee6
JG
2181 if (channel->metadata_cache) {
2182 pthread_mutex_unlock(&channel->metadata_cache->lock);
081424af 2183 }
3dad2c0f 2184 pthread_mutex_unlock(&stream->lock);
a6ef8ee6 2185 pthread_mutex_unlock(&channel->lock);
fa29bfbf 2186 pthread_mutex_unlock(&the_consumer_data.lock);
e316aad5 2187
a6ef8ee6
JG
2188 if (free_channel) {
2189 consumer_del_channel(channel);
e316aad5
DG
2190 }
2191
d2956687
JG
2192 lttng_trace_chunk_put(stream->trace_chunk);
2193 stream->trace_chunk = NULL;
6d574024 2194 consumer_stream_free(stream);
fb3a43a9
DG
2195}
2196
2197/*
2198 * Action done with the metadata stream when adding it to the consumer internal
2199 * data structures to handle it.
2200 */
66d583dc 2201void consumer_add_metadata_stream(struct lttng_consumer_stream *stream)
fb3a43a9 2202{
5ab66908 2203 struct lttng_ht *ht = metadata_ht;
76082088 2204 struct lttng_ht_iter iter;
d88aee68 2205 struct lttng_ht_node_u64 *node;
fb3a43a9 2206
e316aad5
DG
2207 assert(stream);
2208 assert(ht);
2209
d88aee68 2210 DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key);
e316aad5 2211
fa29bfbf 2212 pthread_mutex_lock(&the_consumer_data.lock);
a9838785 2213 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 2214 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 2215 pthread_mutex_lock(&stream->lock);
e316aad5 2216
e316aad5
DG
2217 /*
2218 * From here, refcounts are updated so be _careful_ when returning an error
2219 * after this point.
2220 */
2221
fb3a43a9 2222 rcu_read_lock();
76082088
DG
2223
2224 /*
2225 * Lookup the stream just to make sure it does not exist in our internal
2226 * state. This should NEVER happen.
2227 */
d88aee68
DG
2228 lttng_ht_lookup(ht, &stream->key, &iter);
2229 node = lttng_ht_iter_get_node_u64(&iter);
76082088
DG
2230 assert(!node);
2231
e316aad5 2232 /*
ffe60014
DG
2233 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2234 * in terms of destroying the associated channel, because the action that
e316aad5
DG
2235 * causes the count to become 0 also causes a stream to be added. The
2236 * channel deletion will thus be triggered by the following removal of this
2237 * stream.
2238 */
ffe60014 2239 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
2240 /* Increment refcount before decrementing nb_init_stream_left */
2241 cmm_smp_wmb();
ffe60014 2242 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
2243 }
2244
d88aee68 2245 lttng_ht_add_unique_u64(ht, &stream->node);
ca22feea 2246
fa29bfbf
SM
2247 lttng_ht_add_u64(the_consumer_data.stream_per_chan_id_ht,
2248 &stream->node_channel_id);
d8ef542d 2249
ca22feea
DG
2250 /*
2251 * Add stream to the stream_list_ht of the consumer data. No need to steal
2252 * the key since the HT does not use it and we allow to add redundant keys
2253 * into this table.
2254 */
fa29bfbf
SM
2255 lttng_ht_add_u64(the_consumer_data.stream_list_ht,
2256 &stream->node_session_id);
ca22feea 2257
fb3a43a9 2258 rcu_read_unlock();
e316aad5 2259
2e818a6a 2260 pthread_mutex_unlock(&stream->lock);
a9838785 2261 pthread_mutex_unlock(&stream->chan->lock);
ec6ea7d0 2262 pthread_mutex_unlock(&stream->chan->timer_lock);
fa29bfbf 2263 pthread_mutex_unlock(&the_consumer_data.lock);
fb3a43a9
DG
2264}
2265
8994307f
DG
2266/*
2267 * Delete data stream that are flagged for deletion (endpoint_status).
2268 */
2269static void validate_endpoint_status_data_stream(void)
2270{
2271 struct lttng_ht_iter iter;
2272 struct lttng_consumer_stream *stream;
2273
2274 DBG("Consumer delete flagged data stream");
2275
2276 rcu_read_lock();
2277 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
2278 /* Validate delete flag of the stream */
79d4ffb7 2279 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2280 continue;
2281 }
2282 /* Delete it right now */
2283 consumer_del_stream(stream, data_ht);
2284 }
2285 rcu_read_unlock();
2286}
2287
2288/*
2289 * Delete metadata stream that are flagged for deletion (endpoint_status).
2290 */
2291static void validate_endpoint_status_metadata_stream(
2292 struct lttng_poll_event *pollset)
2293{
2294 struct lttng_ht_iter iter;
2295 struct lttng_consumer_stream *stream;
2296
2297 DBG("Consumer delete flagged metadata stream");
2298
2299 assert(pollset);
2300
2301 rcu_read_lock();
2302 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
2303 /* Validate delete flag of the stream */
79d4ffb7 2304 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2305 continue;
2306 }
2307 /*
2308 * Remove from pollset so the metadata thread can continue without
2309 * blocking on a deleted stream.
2310 */
2311 lttng_poll_del(pollset, stream->wait_fd);
2312
2313 /* Delete it right now */
2314 consumer_del_metadata_stream(stream, metadata_ht);
2315 }
2316 rcu_read_unlock();
2317}
2318
fb3a43a9
DG
2319/*
2320 * Thread polls on metadata file descriptor and write them on disk or on the
2321 * network.
2322 */
7d980def 2323void *consumer_thread_metadata_poll(void *data)
fb3a43a9 2324{
1fc79fb4 2325 int ret, i, pollfd, err = -1;
fb3a43a9 2326 uint32_t revents, nb_fd;
e316aad5 2327 struct lttng_consumer_stream *stream = NULL;
fb3a43a9 2328 struct lttng_ht_iter iter;
d88aee68 2329 struct lttng_ht_node_u64 *node;
fb3a43a9
DG
2330 struct lttng_poll_event events;
2331 struct lttng_consumer_local_data *ctx = data;
2332 ssize_t len;
2333
2334 rcu_register_thread();
2335
1fc79fb4
MD
2336 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA);
2337
2d57de81
MD
2338 if (testpoint(consumerd_thread_metadata)) {
2339 goto error_testpoint;
2340 }
2341
9ce5646a
MD
2342 health_code_update();
2343
fb3a43a9
DG
2344 DBG("Thread metadata poll started");
2345
fb3a43a9
DG
2346 /* Size is set to 1 for the consumer_metadata pipe */
2347 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2348 if (ret < 0) {
2349 ERR("Poll set creation failed");
d8ef542d 2350 goto end_poll;
fb3a43a9
DG
2351 }
2352
13886d2d
DG
2353 ret = lttng_poll_add(&events,
2354 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN);
fb3a43a9
DG
2355 if (ret < 0) {
2356 goto end;
2357 }
2358
2359 /* Main loop */
2360 DBG("Metadata main loop started");
2361
2362 while (1) {
fb3a43a9 2363restart:
7fa2082e 2364 health_code_update();
9ce5646a 2365 health_poll_entry();
7fa2082e 2366 DBG("Metadata poll wait");
fb3a43a9 2367 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2368 DBG("Metadata poll return from wait with %d fd(s)",
2369 LTTNG_POLL_GETNB(&events));
9ce5646a 2370 health_poll_exit();
40063ead 2371 DBG("Metadata event caught in thread");
fb3a43a9
DG
2372 if (ret < 0) {
2373 if (errno == EINTR) {
40063ead 2374 ERR("Poll EINTR caught");
fb3a43a9
DG
2375 goto restart;
2376 }
d9607cd7
MD
2377 if (LTTNG_POLL_GETNB(&events) == 0) {
2378 err = 0; /* All is OK */
2379 }
2380 goto end;
fb3a43a9
DG
2381 }
2382
0d9c5d77
DG
2383 nb_fd = ret;
2384
e316aad5 2385 /* From here, the event is a metadata wait fd */
fb3a43a9 2386 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2387 health_code_update();
2388
fb3a43a9
DG
2389 revents = LTTNG_POLL_GETEV(&events, i);
2390 pollfd = LTTNG_POLL_GETFD(&events, i);
2391
13886d2d 2392 if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) {
03e43155 2393 if (revents & LPOLLIN) {
13886d2d
DG
2394 ssize_t pipe_len;
2395
2396 pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe,
2397 &stream, sizeof(stream));
6cd525e8 2398 if (pipe_len < sizeof(stream)) {
03e43155
MD
2399 if (pipe_len < 0) {
2400 PERROR("read metadata stream");
2401 }
fb3a43a9 2402 /*
03e43155
MD
2403 * Remove the pipe from the poll set and continue the loop
2404 * since their might be data to consume.
fb3a43a9 2405 */
03e43155
MD
2406 lttng_poll_del(&events,
2407 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2408 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
fb3a43a9
DG
2409 continue;
2410 }
2411
8994307f
DG
2412 /* A NULL stream means that the state has changed. */
2413 if (stream == NULL) {
2414 /* Check for deleted streams. */
2415 validate_endpoint_status_metadata_stream(&events);
3714380f 2416 goto restart;
8994307f
DG
2417 }
2418
fb3a43a9
DG
2419 DBG("Adding metadata stream %d to poll set",
2420 stream->wait_fd);
2421
fb3a43a9
DG
2422 /* Add metadata stream to the global poll events list */
2423 lttng_poll_add(&events, stream->wait_fd,
6d574024 2424 LPOLLIN | LPOLLPRI | LPOLLHUP);
03e43155
MD
2425 } else if (revents & (LPOLLERR | LPOLLHUP)) {
2426 DBG("Metadata thread pipe hung up");
2427 /*
2428 * Remove the pipe from the poll set and continue the loop
2429 * since their might be data to consume.
2430 */
2431 lttng_poll_del(&events,
2432 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2433 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
2434 continue;
2435 } else {
2436 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2437 goto end;
fb3a43a9
DG
2438 }
2439
e316aad5 2440 /* Handle other stream */
fb3a43a9
DG
2441 continue;
2442 }
2443
d09e1200 2444 rcu_read_lock();
d88aee68
DG
2445 {
2446 uint64_t tmp_id = (uint64_t) pollfd;
2447
2448 lttng_ht_lookup(metadata_ht, &tmp_id, &iter);
2449 }
2450 node = lttng_ht_iter_get_node_u64(&iter);
e316aad5 2451 assert(node);
fb3a43a9
DG
2452
2453 stream = caa_container_of(node, struct lttng_consumer_stream,
58b1f425 2454 node);
fb3a43a9 2455
03e43155
MD
2456 if (revents & (LPOLLIN | LPOLLPRI)) {
2457 /* Get the data out of the metadata file descriptor */
2458 DBG("Metadata available on fd %d", pollfd);
2459 assert(stream->wait_fd == pollfd);
2460
2461 do {
2462 health_code_update();
2463
6f9449c2 2464 len = ctx->on_buffer_ready(stream, ctx, false);
03e43155
MD
2465 /*
2466 * We don't check the return value here since if we get
83f4233d 2467 * a negative len, it means an error occurred thus we
03e43155
MD
2468 * simply remove it from the poll set and free the
2469 * stream.
2470 */
2471 } while (len > 0);
2472
2473 /* It's ok to have an unavailable sub-buffer */
2474 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
2475 /* Clean up stream from consumer and free it. */
2476 lttng_poll_del(&events, stream->wait_fd);
2477 consumer_del_metadata_stream(stream, metadata_ht);
2478 }
2479 } else if (revents & (LPOLLERR | LPOLLHUP)) {
e316aad5 2480 DBG("Metadata fd %d is hup|err.", pollfd);
fa29bfbf
SM
2481 if (!stream->hangup_flush_done &&
2482 (the_consumer_data.type == LTTNG_CONSUMER32_UST ||
2483 the_consumer_data.type ==
2484 LTTNG_CONSUMER64_UST)) {
fb3a43a9
DG
2485 DBG("Attempting to flush and consume the UST buffers");
2486 lttng_ustconsumer_on_stream_hangup(stream);
2487
2488 /* We just flushed the stream now read it. */
4bb94b75 2489 do {
9ce5646a
MD
2490 health_code_update();
2491
6f9449c2 2492 len = ctx->on_buffer_ready(stream, ctx, false);
4bb94b75
DG
2493 /*
2494 * We don't check the return value here since if we get
83f4233d 2495 * a negative len, it means an error occurred thus we
4bb94b75
DG
2496 * simply remove it from the poll set and free the
2497 * stream.
2498 */
2499 } while (len > 0);
fb3a43a9
DG
2500 }
2501
fb3a43a9 2502 lttng_poll_del(&events, stream->wait_fd);
e316aad5
DG
2503 /*
2504 * This call update the channel states, closes file descriptors
2505 * and securely free the stream.
2506 */
2507 consumer_del_metadata_stream(stream, metadata_ht);
03e43155
MD
2508 } else {
2509 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
6f2f1a70 2510 rcu_read_unlock();
03e43155 2511 goto end;
fb3a43a9 2512 }
e316aad5 2513 /* Release RCU lock for the stream looked up */
d09e1200 2514 rcu_read_unlock();
fb3a43a9
DG
2515 }
2516 }
2517
1fc79fb4
MD
2518 /* All is OK */
2519 err = 0;
fb3a43a9
DG
2520end:
2521 DBG("Metadata poll thread exiting");
fb3a43a9 2522
d8ef542d
MD
2523 lttng_poll_clean(&events);
2524end_poll:
2d57de81 2525error_testpoint:
1fc79fb4
MD
2526 if (err) {
2527 health_error();
2528 ERR("Health error occurred in %s", __func__);
2529 }
2530 health_unregister(health_consumerd);
fb3a43a9
DG
2531 rcu_unregister_thread();
2532 return NULL;
2533}
2534
3bd1e081 2535/*
e4421fec 2536 * This thread polls the fds in the set to consume the data and write
3bd1e081
MD
2537 * it to tracefile if necessary.
2538 */
7d980def 2539void *consumer_thread_data_poll(void *data)
3bd1e081 2540{
1fc79fb4 2541 int num_rdy, num_hup, high_prio, ret, i, err = -1;
3bd1e081
MD
2542 struct pollfd *pollfd = NULL;
2543 /* local view of the streams */
c869f647 2544 struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL;
3bd1e081 2545 /* local view of consumer_data.fds_count */
8bdcc002
JG
2546 int nb_fd = 0;
2547 /* 2 for the consumer_data_pipe and wake up pipe */
2548 const int nb_pipes_fd = 2;
9a2fcf78
JD
2549 /* Number of FDs with CONSUMER_ENDPOINT_INACTIVE but still open. */
2550 int nb_inactive_fd = 0;
3bd1e081 2551 struct lttng_consumer_local_data *ctx = data;
00e2e675 2552 ssize_t len;
3bd1e081 2553
e7b994a3
DG
2554 rcu_register_thread();
2555
1fc79fb4
MD
2556 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA);
2557
2d57de81
MD
2558 if (testpoint(consumerd_thread_data)) {
2559 goto error_testpoint;
2560 }
2561
9ce5646a
MD
2562 health_code_update();
2563
4df6c8cb
MD
2564 local_stream = zmalloc(sizeof(struct lttng_consumer_stream *));
2565 if (local_stream == NULL) {
2566 PERROR("local_stream malloc");
2567 goto end;
2568 }
3bd1e081
MD
2569
2570 while (1) {
9ce5646a
MD
2571 health_code_update();
2572
3bd1e081
MD
2573 high_prio = 0;
2574 num_hup = 0;
2575
2576 /*
e4421fec 2577 * the fds set has been updated, we need to update our
3bd1e081
MD
2578 * local array as well
2579 */
fa29bfbf
SM
2580 pthread_mutex_lock(&the_consumer_data.lock);
2581 if (the_consumer_data.need_update) {
0e428499
DG
2582 free(pollfd);
2583 pollfd = NULL;
2584
2585 free(local_stream);
2586 local_stream = NULL;
3bd1e081 2587
8bdcc002 2588 /* Allocate for all fds */
fa29bfbf
SM
2589 pollfd = zmalloc((the_consumer_data.stream_count +
2590 nb_pipes_fd) *
2591 sizeof(struct pollfd));
3bd1e081 2592 if (pollfd == NULL) {
7a57cf92 2593 PERROR("pollfd malloc");
fa29bfbf 2594 pthread_mutex_unlock(&the_consumer_data.lock);
3bd1e081
MD
2595 goto end;
2596 }
2597
fa29bfbf
SM
2598 local_stream = zmalloc((the_consumer_data.stream_count +
2599 nb_pipes_fd) *
747f8642 2600 sizeof(struct lttng_consumer_stream *));
3bd1e081 2601 if (local_stream == NULL) {
7a57cf92 2602 PERROR("local_stream malloc");
fa29bfbf 2603 pthread_mutex_unlock(&the_consumer_data.lock);
3bd1e081
MD
2604 goto end;
2605 }
ffe60014 2606 ret = update_poll_array(ctx, &pollfd, local_stream,
9a2fcf78 2607 data_ht, &nb_inactive_fd);
3bd1e081
MD
2608 if (ret < 0) {
2609 ERR("Error in allocating pollfd or local_outfds");
f73fabfd 2610 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
fa29bfbf 2611 pthread_mutex_unlock(&the_consumer_data.lock);
3bd1e081
MD
2612 goto end;
2613 }
2614 nb_fd = ret;
fa29bfbf 2615 the_consumer_data.need_update = 0;
3bd1e081 2616 }
fa29bfbf 2617 pthread_mutex_unlock(&the_consumer_data.lock);
3bd1e081 2618
4078b776 2619 /* No FDs and consumer_quit, consumer_cleanup the thread */
9a2fcf78
JD
2620 if (nb_fd == 0 && nb_inactive_fd == 0 &&
2621 CMM_LOAD_SHARED(consumer_quit) == 1) {
1fc79fb4 2622 err = 0; /* All is OK */
4078b776
MD
2623 goto end;
2624 }
3bd1e081 2625 /* poll on the array of fds */
88f2b785 2626 restart:
261de637 2627 DBG("polling on %d fd", nb_fd + nb_pipes_fd);
cf0bcb51
JG
2628 if (testpoint(consumerd_thread_data_poll)) {
2629 goto end;
2630 }
9ce5646a 2631 health_poll_entry();
261de637 2632 num_rdy = poll(pollfd, nb_fd + nb_pipes_fd, -1);
9ce5646a 2633 health_poll_exit();
3bd1e081
MD
2634 DBG("poll num_rdy : %d", num_rdy);
2635 if (num_rdy == -1) {
88f2b785
MD
2636 /*
2637 * Restart interrupted system call.
2638 */
2639 if (errno == EINTR) {
2640 goto restart;
2641 }
7a57cf92 2642 PERROR("Poll error");
f73fabfd 2643 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2644 goto end;
2645 } else if (num_rdy == 0) {
2646 DBG("Polling thread timed out");
2647 goto end;
2648 }
2649
80957876
JG
2650 if (caa_unlikely(data_consumption_paused)) {
2651 DBG("Data consumption paused, sleeping...");
2652 sleep(1);
2653 goto restart;
2654 }
2655
3bd1e081 2656 /*
50f8ae69 2657 * If the consumer_data_pipe triggered poll go directly to the
00e2e675
DG
2658 * beginning of the loop to update the array. We want to prioritize
2659 * array update over low-priority reads.
3bd1e081 2660 */
509bb1cf 2661 if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) {
ab30f567 2662 ssize_t pipe_readlen;
04fdd819 2663
50f8ae69 2664 DBG("consumer_data_pipe wake up");
acdb9057
DG
2665 pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe,
2666 &new_stream, sizeof(new_stream));
6cd525e8
MD
2667 if (pipe_readlen < sizeof(new_stream)) {
2668 PERROR("Consumer data pipe");
23f5f35d
DG
2669 /* Continue so we can at least handle the current stream(s). */
2670 continue;
2671 }
c869f647
DG
2672
2673 /*
2674 * If the stream is NULL, just ignore it. It's also possible that
2675 * the sessiond poll thread changed the consumer_quit state and is
2676 * waking us up to test it.
2677 */
2678 if (new_stream == NULL) {
8994307f 2679 validate_endpoint_status_data_stream();
c869f647
DG
2680 continue;
2681 }
2682
c869f647 2683 /* Continue to update the local streams and handle prio ones */
3bd1e081
MD
2684 continue;
2685 }
2686
02b3d176
DG
2687 /* Handle wakeup pipe. */
2688 if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) {
2689 char dummy;
2690 ssize_t pipe_readlen;
2691
2692 pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy,
2693 sizeof(dummy));
2694 if (pipe_readlen < 0) {
2695 PERROR("Consumer data wakeup pipe");
2696 }
2697 /* We've been awakened to handle stream(s). */
2698 ctx->has_wakeup = 0;
2699 }
2700
3bd1e081
MD
2701 /* Take care of high priority channels first. */
2702 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2703 health_code_update();
2704
9617607b
DG
2705 if (local_stream[i] == NULL) {
2706 continue;
2707 }
fb3a43a9 2708 if (pollfd[i].revents & POLLPRI) {
d41f73b7
MD
2709 DBG("Urgent read on fd %d", pollfd[i].fd);
2710 high_prio = 1;
6f9449c2 2711 len = ctx->on_buffer_ready(local_stream[i], ctx, false);
d41f73b7 2712 /* it's ok to have an unavailable sub-buffer */
b64403e3 2713 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2714 /* Clean the stream and free it. */
2715 consumer_del_stream(local_stream[i], data_ht);
9617607b 2716 local_stream[i] = NULL;
4078b776
MD
2717 } else if (len > 0) {
2718 local_stream[i]->data_read = 1;
d41f73b7 2719 }
3bd1e081
MD
2720 }
2721 }
2722
4078b776
MD
2723 /*
2724 * If we read high prio channel in this loop, try again
2725 * for more high prio data.
2726 */
2727 if (high_prio) {
3bd1e081
MD
2728 continue;
2729 }
2730
2731 /* Take care of low priority channels. */
4078b776 2732 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2733 health_code_update();
2734
9617607b
DG
2735 if (local_stream[i] == NULL) {
2736 continue;
2737 }
4078b776 2738 if ((pollfd[i].revents & POLLIN) ||
02b3d176
DG
2739 local_stream[i]->hangup_flush_done ||
2740 local_stream[i]->has_data) {
4078b776 2741 DBG("Normal read on fd %d", pollfd[i].fd);
6f9449c2 2742 len = ctx->on_buffer_ready(local_stream[i], ctx, false);
4078b776 2743 /* it's ok to have an unavailable sub-buffer */
b64403e3 2744 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2745 /* Clean the stream and free it. */
2746 consumer_del_stream(local_stream[i], data_ht);
9617607b 2747 local_stream[i] = NULL;
4078b776
MD
2748 } else if (len > 0) {
2749 local_stream[i]->data_read = 1;
2750 }
2751 }
2752 }
2753
2754 /* Handle hangup and errors */
2755 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2756 health_code_update();
2757
9617607b
DG
2758 if (local_stream[i] == NULL) {
2759 continue;
2760 }
4078b776
MD
2761 if (!local_stream[i]->hangup_flush_done
2762 && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL))
fa29bfbf
SM
2763 && (the_consumer_data.type == LTTNG_CONSUMER32_UST
2764 || the_consumer_data.type == LTTNG_CONSUMER64_UST)) {
4078b776 2765 DBG("fd %d is hup|err|nval. Attempting flush and read.",
9617607b 2766 pollfd[i].fd);
4078b776
MD
2767 lttng_ustconsumer_on_stream_hangup(local_stream[i]);
2768 /* Attempt read again, for the data we just flushed. */
2769 local_stream[i]->data_read = 1;
2770 }
2771 /*
2772 * If the poll flag is HUP/ERR/NVAL and we have
2773 * read no data in this pass, we can remove the
2774 * stream from its hash table.
2775 */
2776 if ((pollfd[i].revents & POLLHUP)) {
2777 DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
2778 if (!local_stream[i]->data_read) {
43c34bc3 2779 consumer_del_stream(local_stream[i], data_ht);
9617607b 2780 local_stream[i] = NULL;
4078b776
MD
2781 num_hup++;
2782 }
2783 } else if (pollfd[i].revents & POLLERR) {
2784 ERR("Error returned in polling fd %d.", pollfd[i].fd);
2785 if (!local_stream[i]->data_read) {
43c34bc3 2786 consumer_del_stream(local_stream[i], data_ht);
9617607b 2787 local_stream[i] = NULL;
4078b776
MD
2788 num_hup++;
2789 }
2790 } else if (pollfd[i].revents & POLLNVAL) {
2791 ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
2792 if (!local_stream[i]->data_read) {
43c34bc3 2793 consumer_del_stream(local_stream[i], data_ht);
9617607b 2794 local_stream[i] = NULL;
4078b776 2795 num_hup++;
3bd1e081
MD
2796 }
2797 }
9617607b
DG
2798 if (local_stream[i] != NULL) {
2799 local_stream[i]->data_read = 0;
2800 }
3bd1e081
MD
2801 }
2802 }
1fc79fb4
MD
2803 /* All is OK */
2804 err = 0;
3bd1e081
MD
2805end:
2806 DBG("polling thread exiting");
0e428499
DG
2807 free(pollfd);
2808 free(local_stream);
fb3a43a9
DG
2809
2810 /*
2811 * Close the write side of the pipe so epoll_wait() in
7d980def
DG
2812 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2813 * read side of the pipe. If we close them both, epoll_wait strangely does
2814 * not return and could create a endless wait period if the pipe is the
2815 * only tracked fd in the poll set. The thread will take care of closing
2816 * the read side.
fb3a43a9 2817 */
13886d2d 2818 (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe);
fb3a43a9 2819
2d57de81 2820error_testpoint:
1fc79fb4
MD
2821 if (err) {
2822 health_error();
2823 ERR("Health error occurred in %s", __func__);
2824 }
2825 health_unregister(health_consumerd);
2826
e7b994a3 2827 rcu_unregister_thread();
3bd1e081
MD
2828 return NULL;
2829}
2830
d8ef542d
MD
2831/*
2832 * Close wake-up end of each stream belonging to the channel. This will
2833 * allow the poll() on the stream read-side to detect when the
2834 * write-side (application) finally closes them.
2835 */
2836static
2837void consumer_close_channel_streams(struct lttng_consumer_channel *channel)
2838{
2839 struct lttng_ht *ht;
2840 struct lttng_consumer_stream *stream;
2841 struct lttng_ht_iter iter;
2842
fa29bfbf 2843 ht = the_consumer_data.stream_per_chan_id_ht;
d8ef542d
MD
2844
2845 rcu_read_lock();
2846 cds_lfht_for_each_entry_duplicate(ht->ht,
2847 ht->hash_fct(&channel->key, lttng_ht_seed),
2848 ht->match_fct, &channel->key,
2849 &iter.iter, stream, node_channel_id.node) {
f2ad556d
MD
2850 /*
2851 * Protect against teardown with mutex.
2852 */
2853 pthread_mutex_lock(&stream->lock);
2854 if (cds_lfht_is_node_deleted(&stream->node.node)) {
2855 goto next;
2856 }
fa29bfbf 2857 switch (the_consumer_data.type) {
d8ef542d
MD
2858 case LTTNG_CONSUMER_KERNEL:
2859 break;
2860 case LTTNG_CONSUMER32_UST:
2861 case LTTNG_CONSUMER64_UST:
b4a650f3
DG
2862 if (stream->metadata_flag) {
2863 /* Safe and protected by the stream lock. */
2864 lttng_ustconsumer_close_metadata(stream->chan);
2865 } else {
2866 /*
2867 * Note: a mutex is taken internally within
2868 * liblttng-ust-ctl to protect timer wakeup_fd
2869 * use from concurrent close.
2870 */
2871 lttng_ustconsumer_close_stream_wakeup(stream);
2872 }
d8ef542d
MD
2873 break;
2874 default:
2875 ERR("Unknown consumer_data type");
2876 assert(0);
2877 }
f2ad556d
MD
2878 next:
2879 pthread_mutex_unlock(&stream->lock);
d8ef542d
MD
2880 }
2881 rcu_read_unlock();
2882}
2883
2884static void destroy_channel_ht(struct lttng_ht *ht)
2885{
2886 struct lttng_ht_iter iter;
2887 struct lttng_consumer_channel *channel;
2888 int ret;
2889
2890 if (ht == NULL) {
2891 return;
2892 }
2893
2894 rcu_read_lock();
2895 cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) {
2896 ret = lttng_ht_del(ht, &iter);
2897 assert(ret != 0);
2898 }
2899 rcu_read_unlock();
2900
2901 lttng_ht_destroy(ht);
2902}
2903
2904/*
2905 * This thread polls the channel fds to detect when they are being
2906 * closed. It closes all related streams if the channel is detected as
2907 * closed. It is currently only used as a shim layer for UST because the
2908 * consumerd needs to keep the per-stream wakeup end of pipes open for
2909 * periodical flush.
2910 */
2911void *consumer_thread_channel_poll(void *data)
2912{
1fc79fb4 2913 int ret, i, pollfd, err = -1;
d8ef542d
MD
2914 uint32_t revents, nb_fd;
2915 struct lttng_consumer_channel *chan = NULL;
2916 struct lttng_ht_iter iter;
2917 struct lttng_ht_node_u64 *node;
2918 struct lttng_poll_event events;
2919 struct lttng_consumer_local_data *ctx = data;
2920 struct lttng_ht *channel_ht;
2921
2922 rcu_register_thread();
2923
1fc79fb4
MD
2924 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL);
2925
2d57de81
MD
2926 if (testpoint(consumerd_thread_channel)) {
2927 goto error_testpoint;
2928 }
2929
9ce5646a
MD
2930 health_code_update();
2931
d8ef542d
MD
2932 channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2933 if (!channel_ht) {
2934 /* ENOMEM at this point. Better to bail out. */
2935 goto end_ht;
2936 }
2937
2938 DBG("Thread channel poll started");
2939
2940 /* Size is set to 1 for the consumer_channel pipe */
2941 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2942 if (ret < 0) {
2943 ERR("Poll set creation failed");
2944 goto end_poll;
2945 }
2946
2947 ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN);
2948 if (ret < 0) {
2949 goto end;
2950 }
2951
2952 /* Main loop */
2953 DBG("Channel main loop started");
2954
2955 while (1) {
d8ef542d 2956restart:
7fa2082e
MD
2957 health_code_update();
2958 DBG("Channel poll wait");
9ce5646a 2959 health_poll_entry();
d8ef542d 2960 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2961 DBG("Channel poll return from wait with %d fd(s)",
2962 LTTNG_POLL_GETNB(&events));
9ce5646a 2963 health_poll_exit();
40063ead 2964 DBG("Channel event caught in thread");
d8ef542d
MD
2965 if (ret < 0) {
2966 if (errno == EINTR) {
40063ead 2967 ERR("Poll EINTR caught");
d8ef542d
MD
2968 goto restart;
2969 }
d9607cd7
MD
2970 if (LTTNG_POLL_GETNB(&events) == 0) {
2971 err = 0; /* All is OK */
2972 }
d8ef542d
MD
2973 goto end;
2974 }
2975
2976 nb_fd = ret;
2977
2978 /* From here, the event is a channel wait fd */
2979 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2980 health_code_update();
2981
d8ef542d
MD
2982 revents = LTTNG_POLL_GETEV(&events, i);
2983 pollfd = LTTNG_POLL_GETFD(&events, i);
2984
d8ef542d 2985 if (pollfd == ctx->consumer_channel_pipe[0]) {
03e43155 2986 if (revents & LPOLLIN) {
d8ef542d 2987 enum consumer_channel_action action;
a0cbdd2e 2988 uint64_t key;
d8ef542d 2989
a0cbdd2e 2990 ret = read_channel_pipe(ctx, &chan, &key, &action);
d8ef542d 2991 if (ret <= 0) {
03e43155
MD
2992 if (ret < 0) {
2993 ERR("Error reading channel pipe");
2994 }
2995 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
d8ef542d
MD
2996 continue;
2997 }
2998
2999 switch (action) {
3000 case CONSUMER_CHANNEL_ADD:
3001 DBG("Adding channel %d to poll set",
3002 chan->wait_fd);
3003
3004 lttng_ht_node_init_u64(&chan->wait_fd_node,
3005 chan->wait_fd);
c7260a81 3006 rcu_read_lock();
d8ef542d
MD
3007 lttng_ht_add_unique_u64(channel_ht,
3008 &chan->wait_fd_node);
c7260a81 3009 rcu_read_unlock();
d8ef542d
MD
3010 /* Add channel to the global poll events list */
3011 lttng_poll_add(&events, chan->wait_fd,
03e43155 3012 LPOLLERR | LPOLLHUP);
d8ef542d 3013 break;
a0cbdd2e
MD
3014 case CONSUMER_CHANNEL_DEL:
3015 {
b4a650f3
DG
3016 /*
3017 * This command should never be called if the channel
3018 * has streams monitored by either the data or metadata
3019 * thread. The consumer only notify this thread with a
3020 * channel del. command if it receives a destroy
3021 * channel command from the session daemon that send it
3022 * if a command prior to the GET_CHANNEL failed.
3023 */
3024
c7260a81 3025 rcu_read_lock();
a0cbdd2e
MD
3026 chan = consumer_find_channel(key);
3027 if (!chan) {
c7260a81 3028 rcu_read_unlock();
a0cbdd2e
MD
3029 ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key);
3030 break;
3031 }
3032 lttng_poll_del(&events, chan->wait_fd);
f623cc0b 3033 iter.iter.node = &chan->wait_fd_node.node;
a0cbdd2e
MD
3034 ret = lttng_ht_del(channel_ht, &iter);
3035 assert(ret == 0);
a0cbdd2e 3036
fa29bfbf 3037 switch (the_consumer_data.type) {
f2a444f1
DG
3038 case LTTNG_CONSUMER_KERNEL:
3039 break;
3040 case LTTNG_CONSUMER32_UST:
3041 case LTTNG_CONSUMER64_UST:
212d67a2
DG
3042 health_code_update();
3043 /* Destroy streams that might have been left in the stream list. */
3044 clean_channel_stream_list(chan);
f2a444f1
DG
3045 break;
3046 default:
3047 ERR("Unknown consumer_data type");
3048 assert(0);
3049 }
3050
a0cbdd2e
MD
3051 /*
3052 * Release our own refcount. Force channel deletion even if
3053 * streams were not initialized.
3054 */
3055 if (!uatomic_sub_return(&chan->refcount, 1)) {
3056 consumer_del_channel(chan);
3057 }
c7260a81 3058 rcu_read_unlock();
a0cbdd2e
MD
3059 goto restart;
3060 }
d8ef542d
MD
3061 case CONSUMER_CHANNEL_QUIT:
3062 /*
3063 * Remove the pipe from the poll set and continue the loop
3064 * since their might be data to consume.
3065 */
3066 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
3067 continue;
3068 default:
3069 ERR("Unknown action");
3070 break;
3071 }
03e43155
MD
3072 } else if (revents & (LPOLLERR | LPOLLHUP)) {
3073 DBG("Channel thread pipe hung up");
3074 /*
3075 * Remove the pipe from the poll set and continue the loop
3076 * since their might be data to consume.
3077 */
3078 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
3079 continue;
3080 } else {
3081 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
3082 goto end;
d8ef542d
MD
3083 }
3084
3085 /* Handle other stream */
3086 continue;
3087 }
3088
3089 rcu_read_lock();
3090 {
3091 uint64_t tmp_id = (uint64_t) pollfd;
3092
3093 lttng_ht_lookup(channel_ht, &tmp_id, &iter);
3094 }
3095 node = lttng_ht_iter_get_node_u64(&iter);
3096 assert(node);
3097
3098 chan = caa_container_of(node, struct lttng_consumer_channel,
3099 wait_fd_node);
3100
3101 /* Check for error event */
3102 if (revents & (LPOLLERR | LPOLLHUP)) {
3103 DBG("Channel fd %d is hup|err.", pollfd);
3104
3105 lttng_poll_del(&events, chan->wait_fd);
3106 ret = lttng_ht_del(channel_ht, &iter);
3107 assert(ret == 0);
b4a650f3
DG
3108
3109 /*
3110 * This will close the wait fd for each stream associated to
3111 * this channel AND monitored by the data/metadata thread thus
3112 * will be clean by the right thread.
3113 */
d8ef542d 3114 consumer_close_channel_streams(chan);
f2ad556d
MD
3115
3116 /* Release our own refcount */
3117 if (!uatomic_sub_return(&chan->refcount, 1)
3118 && !uatomic_read(&chan->nb_init_stream_left)) {
3119 consumer_del_channel(chan);
3120 }
03e43155
MD
3121 } else {
3122 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
3123 rcu_read_unlock();
3124 goto end;
d8ef542d
MD
3125 }
3126
3127 /* Release RCU lock for the channel looked up */
3128 rcu_read_unlock();
3129 }
3130 }
3131
1fc79fb4
MD
3132 /* All is OK */
3133 err = 0;
d8ef542d
MD
3134end:
3135 lttng_poll_clean(&events);
3136end_poll:
3137 destroy_channel_ht(channel_ht);
3138end_ht:
2d57de81 3139error_testpoint:
d8ef542d 3140 DBG("Channel poll thread exiting");
1fc79fb4
MD
3141 if (err) {
3142 health_error();
3143 ERR("Health error occurred in %s", __func__);
3144 }
3145 health_unregister(health_consumerd);
d8ef542d
MD
3146 rcu_unregister_thread();
3147 return NULL;
3148}
3149
331744e3
JD
3150static int set_metadata_socket(struct lttng_consumer_local_data *ctx,
3151 struct pollfd *sockpoll, int client_socket)
3152{
3153 int ret;
3154
3155 assert(ctx);
3156 assert(sockpoll);
3157
84382d49
MD
3158 ret = lttng_consumer_poll_socket(sockpoll);
3159 if (ret) {
331744e3
JD
3160 goto error;
3161 }
3162 DBG("Metadata connection on client_socket");
3163
3164 /* Blocking call, waiting for transmission */
3165 ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket);
3166 if (ctx->consumer_metadata_socket < 0) {
3167 WARN("On accept metadata");
3168 ret = -1;
3169 goto error;
3170 }
3171 ret = 0;
3172
3173error:
3174 return ret;
3175}
3176
3bd1e081
MD
3177/*
3178 * This thread listens on the consumerd socket and receives the file
3179 * descriptors from the session daemon.
3180 */
7d980def 3181void *consumer_thread_sessiond_poll(void *data)
3bd1e081 3182{
1fc79fb4 3183 int sock = -1, client_socket, ret, err = -1;
3bd1e081
MD
3184 /*
3185 * structure to poll for incoming data on communication socket avoids
3186 * making blocking sockets.
3187 */
3188 struct pollfd consumer_sockpoll[2];
3189 struct lttng_consumer_local_data *ctx = data;
3190
e7b994a3
DG
3191 rcu_register_thread();
3192
1fc79fb4
MD
3193 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND);
3194
2d57de81
MD
3195 if (testpoint(consumerd_thread_sessiond)) {
3196 goto error_testpoint;
3197 }
3198
9ce5646a
MD
3199 health_code_update();
3200
3bd1e081
MD
3201 DBG("Creating command socket %s", ctx->consumer_command_sock_path);
3202 unlink(ctx->consumer_command_sock_path);
3203 client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path);
3204 if (client_socket < 0) {
3205 ERR("Cannot create command socket");
3206 goto end;
3207 }
3208
3209 ret = lttcomm_listen_unix_sock(client_socket);
3210 if (ret < 0) {
3211 goto end;
3212 }
3213
32258573 3214 DBG("Sending ready command to lttng-sessiond");
f73fabfd 3215 ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY);
3bd1e081
MD
3216 /* return < 0 on error, but == 0 is not fatal */
3217 if (ret < 0) {
32258573 3218 ERR("Error sending ready command to lttng-sessiond");
3bd1e081
MD
3219 goto end;
3220 }
3221
3bd1e081
MD
3222 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3223 consumer_sockpoll[0].fd = ctx->consumer_should_quit[0];
3224 consumer_sockpoll[0].events = POLLIN | POLLPRI;
3225 consumer_sockpoll[1].fd = client_socket;
3226 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3227
84382d49
MD
3228 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3229 if (ret) {
3230 if (ret > 0) {
3231 /* should exit */
3232 err = 0;
3233 }
3bd1e081
MD
3234 goto end;
3235 }
3236 DBG("Connection on client_socket");
3237
3238 /* Blocking call, waiting for transmission */
3239 sock = lttcomm_accept_unix_sock(client_socket);
534d2592 3240 if (sock < 0) {
3bd1e081
MD
3241 WARN("On accept");
3242 goto end;
3243 }
3bd1e081 3244
331744e3
JD
3245 /*
3246 * Setup metadata socket which is the second socket connection on the
3247 * command unix socket.
3248 */
3249 ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket);
84382d49
MD
3250 if (ret) {
3251 if (ret > 0) {
3252 /* should exit */
3253 err = 0;
3254 }
331744e3
JD
3255 goto end;
3256 }
3257
d96f09c6
DG
3258 /* This socket is not useful anymore. */
3259 ret = close(client_socket);
3260 if (ret < 0) {
3261 PERROR("close client_socket");
3262 }
3263 client_socket = -1;
3264
3bd1e081
MD
3265 /* update the polling structure to poll on the established socket */
3266 consumer_sockpoll[1].fd = sock;
3267 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3268
3269 while (1) {
9ce5646a
MD
3270 health_code_update();
3271
3272 health_poll_entry();
3273 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3274 health_poll_exit();
84382d49
MD
3275 if (ret) {
3276 if (ret > 0) {
3277 /* should exit */
3278 err = 0;
3279 }
3bd1e081
MD
3280 goto end;
3281 }
3282 DBG("Incoming command on sock");
3283 ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll);
4cbc1a04
DG
3284 if (ret <= 0) {
3285 /*
3286 * This could simply be a session daemon quitting. Don't output
3287 * ERR() here.
3288 */
3289 DBG("Communication interrupted on command socket");
41ba6035 3290 err = 0;
3bd1e081
MD
3291 goto end;
3292 }
10211f5c 3293 if (CMM_LOAD_SHARED(consumer_quit)) {
3bd1e081 3294 DBG("consumer_thread_receive_fds received quit from signal");
1fc79fb4 3295 err = 0; /* All is OK */
3bd1e081
MD
3296 goto end;
3297 }
a1ed855a 3298 DBG("Received command on sock");
3bd1e081 3299 }
1fc79fb4
MD
3300 /* All is OK */
3301 err = 0;
3302
3bd1e081 3303end:
ffe60014 3304 DBG("Consumer thread sessiond poll exiting");
3bd1e081 3305
d88aee68
DG
3306 /*
3307 * Close metadata streams since the producer is the session daemon which
3308 * just died.
3309 *
3310 * NOTE: for now, this only applies to the UST tracer.
3311 */
6d574024 3312 lttng_consumer_close_all_metadata();
d88aee68 3313
3bd1e081
MD
3314 /*
3315 * when all fds have hung up, the polling thread
3316 * can exit cleanly
3317 */
10211f5c 3318 CMM_STORE_SHARED(consumer_quit, 1);
3bd1e081 3319
04fdd819 3320 /*
c869f647 3321 * Notify the data poll thread to poll back again and test the
8994307f 3322 * consumer_quit state that we just set so to quit gracefully.
04fdd819 3323 */
acdb9057 3324 notify_thread_lttng_pipe(ctx->consumer_data_pipe);
c869f647 3325
a0cbdd2e 3326 notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT);
d8ef542d 3327
5c635c72
MD
3328 notify_health_quit_pipe(health_quit_pipe);
3329
d96f09c6
DG
3330 /* Cleaning up possibly open sockets. */
3331 if (sock >= 0) {
3332 ret = close(sock);
3333 if (ret < 0) {
3334 PERROR("close sock sessiond poll");
3335 }
3336 }
3337 if (client_socket >= 0) {
38476d24 3338 ret = close(client_socket);
d96f09c6
DG
3339 if (ret < 0) {
3340 PERROR("close client_socket sessiond poll");
3341 }
3342 }
3343
2d57de81 3344error_testpoint:
1fc79fb4
MD
3345 if (err) {
3346 health_error();
3347 ERR("Health error occurred in %s", __func__);
3348 }
3349 health_unregister(health_consumerd);
3350
e7b994a3 3351 rcu_unregister_thread();
3bd1e081
MD
3352 return NULL;
3353}
d41f73b7 3354
503fefca
JG
3355static int post_consume(struct lttng_consumer_stream *stream,
3356 const struct stream_subbuffer *subbuffer,
3357 struct lttng_consumer_local_data *ctx)
f96af312 3358{
503fefca 3359 size_t i;
f96af312 3360 int ret = 0;
503fefca
JG
3361 const size_t count = lttng_dynamic_array_get_count(
3362 &stream->read_subbuffer_ops.post_consume_cbs);
f96af312 3363
503fefca
JG
3364 for (i = 0; i < count; i++) {
3365 const post_consume_cb op = *(post_consume_cb *) lttng_dynamic_array_get_element(
3366 &stream->read_subbuffer_ops.post_consume_cbs,
3367 i);
3368
3369 ret = op(stream, subbuffer, ctx);
3370 if (ret) {
3371 goto end;
f96af312 3372 }
f96af312 3373 }
f96af312
JG
3374end:
3375 return ret;
3376}
3377
4078b776 3378ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream,
6f9449c2
JG
3379 struct lttng_consumer_local_data *ctx,
3380 bool locked_by_caller)
d41f73b7 3381{
12bddd1d 3382 ssize_t ret, written_bytes = 0;
23d56598 3383 int rotation_ret;
6f9449c2 3384 struct stream_subbuffer subbuffer = {};
74251bb8 3385
6f9449c2
JG
3386 if (!locked_by_caller) {
3387 stream->read_subbuffer_ops.lock(stream);
3388 }
3389
3390 if (stream->read_subbuffer_ops.on_wake_up) {
3391 ret = stream->read_subbuffer_ops.on_wake_up(stream);
3392 if (ret) {
3393 goto end;
3394 }
94d49140 3395 }
74251bb8 3396
23d56598
JG
3397 /*
3398 * If the stream was flagged to be ready for rotation before we extract
3399 * the next packet, rotate it now.
3400 */
3401 if (stream->rotate_ready) {
3402 DBG("Rotate stream before consuming data");
3403 ret = lttng_consumer_rotate_stream(ctx, stream);
3404 if (ret < 0) {
3405 ERR("Stream rotation error before consuming data");
3406 goto end;
3407 }
3408 }
3409
6f9449c2
JG
3410 ret = stream->read_subbuffer_ops.get_next_subbuffer(stream, &subbuffer);
3411 if (ret) {
3412 if (ret == -ENODATA) {
3413 /* Not an error. */
3414 ret = 0;
e66d26f5 3415 goto sleep_stream;
6f9449c2
JG
3416 }
3417 goto end;
d41f73b7 3418 }
74251bb8 3419
6f9449c2
JG
3420 ret = stream->read_subbuffer_ops.pre_consume_subbuffer(
3421 stream, &subbuffer);
3422 if (ret) {
3423 goto error_put_subbuf;
3424 }
3425
3426 written_bytes = stream->read_subbuffer_ops.consume_subbuffer(
3427 ctx, stream, &subbuffer);
514775d9
FD
3428 if (written_bytes <= 0) {
3429 ERR("Error consuming subbuffer: (%zd)", written_bytes);
3430 ret = (int) written_bytes;
3431 goto error_put_subbuf;
6f9449c2
JG
3432 }
3433
3434 ret = stream->read_subbuffer_ops.put_next_subbuffer(stream, &subbuffer);
3435 if (ret) {
23d56598
JG
3436 goto end;
3437 }
3438
503fefca
JG
3439 ret = post_consume(stream, &subbuffer, ctx);
3440 if (ret) {
3441 goto end;
6f9449c2
JG
3442 }
3443
23d56598
JG
3444 /*
3445 * After extracting the packet, we check if the stream is now ready to
3446 * be rotated and perform the action immediately.
3447 *
3448 * Don't overwrite `ret` as callers expect the number of bytes
3449 * consumed to be returned on success.
3450 */
3451 rotation_ret = lttng_consumer_stream_is_rotate_ready(stream);
3452 if (rotation_ret == 1) {
3453 rotation_ret = lttng_consumer_rotate_stream(ctx, stream);
3454 if (rotation_ret < 0) {
3455 ret = rotation_ret;
3456 ERR("Stream rotation error after consuming data");
3457 goto end;
3458 }
503fefca 3459
23d56598
JG
3460 } else if (rotation_ret < 0) {
3461 ret = rotation_ret;
3462 ERR("Failed to check if stream was ready to rotate after consuming data");
3463 goto end;
3464 }
3465
82e72193 3466sleep_stream:
6f9449c2
JG
3467 if (stream->read_subbuffer_ops.on_sleep) {
3468 stream->read_subbuffer_ops.on_sleep(stream, ctx);
3469 }
3470
3471 ret = written_bytes;
23d56598 3472end:
6f9449c2
JG
3473 if (!locked_by_caller) {
3474 stream->read_subbuffer_ops.unlock(stream);
94d49140 3475 }
6f9449c2 3476
74251bb8 3477 return ret;
6f9449c2
JG
3478error_put_subbuf:
3479 (void) stream->read_subbuffer_ops.put_next_subbuffer(stream, &subbuffer);
3480 goto end;
d41f73b7
MD
3481}
3482
3483int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream)
3484{
fa29bfbf 3485 switch (the_consumer_data.type) {
d41f73b7
MD
3486 case LTTNG_CONSUMER_KERNEL:
3487 return lttng_kconsumer_on_recv_stream(stream);
7753dea8
MD
3488 case LTTNG_CONSUMER32_UST:
3489 case LTTNG_CONSUMER64_UST:
d41f73b7
MD
3490 return lttng_ustconsumer_on_recv_stream(stream);
3491 default:
3492 ERR("Unknown consumer_data type");
3493 assert(0);
3494 return -ENOSYS;
3495 }
3496}
e4421fec
DG
3497
3498/*
3499 * Allocate and set consumer data hash tables.
3500 */
282dadbc 3501int lttng_consumer_init(void)
e4421fec 3502{
fa29bfbf
SM
3503 the_consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3504 if (!the_consumer_data.channel_ht) {
282dadbc
MD
3505 goto error;
3506 }
3507
fa29bfbf 3508 the_consumer_data.channels_by_session_id_ht =
5c3892a6 3509 lttng_ht_new(0, LTTNG_HT_TYPE_U64);
fa29bfbf 3510 if (!the_consumer_data.channels_by_session_id_ht) {
5c3892a6
JG
3511 goto error;
3512 }
3513
fa29bfbf
SM
3514 the_consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3515 if (!the_consumer_data.relayd_ht) {
282dadbc
MD
3516 goto error;
3517 }
3518
fa29bfbf
SM
3519 the_consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3520 if (!the_consumer_data.stream_list_ht) {
282dadbc
MD
3521 goto error;
3522 }
3523
fa29bfbf
SM
3524 the_consumer_data.stream_per_chan_id_ht =
3525 lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3526 if (!the_consumer_data.stream_per_chan_id_ht) {
282dadbc
MD
3527 goto error;
3528 }
3529
3530 data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3531 if (!data_ht) {
3532 goto error;
3533 }
3534
3535 metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3536 if (!metadata_ht) {
3537 goto error;
3538 }
3539
fa29bfbf
SM
3540 the_consumer_data.chunk_registry = lttng_trace_chunk_registry_create();
3541 if (!the_consumer_data.chunk_registry) {
28cc88f3
JG
3542 goto error;
3543 }
3544
282dadbc
MD
3545 return 0;
3546
3547error:
3548 return -1;
e4421fec 3549}
7735ef9e
DG
3550
3551/*
3552 * Process the ADD_RELAYD command receive by a consumer.
3553 *
3554 * This will create a relayd socket pair and add it to the relayd hash table.
3555 * The caller MUST acquire a RCU read side lock before calling it.
3556 */
2527bf85 3557 void consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type,
7735ef9e 3558 struct lttng_consumer_local_data *ctx, int sock,
6151a90f 3559 struct pollfd *consumer_sockpoll,
d3e2ba59
JD
3560 struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id,
3561 uint64_t relayd_session_id)
7735ef9e 3562{
cd2b09ed 3563 int fd = -1, ret = -1, relayd_created = 0;
0c759fc9 3564 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
d4298c99 3565 struct consumer_relayd_sock_pair *relayd = NULL;
7735ef9e 3566
6151a90f
JD
3567 assert(ctx);
3568 assert(relayd_sock);
3569
da009f2c 3570 DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx);
7735ef9e
DG
3571
3572 /* Get relayd reference if exists. */
3573 relayd = consumer_find_relayd(net_seq_idx);
3574 if (relayd == NULL) {
da009f2c 3575 assert(sock_type == LTTNG_STREAM_CONTROL);
7735ef9e
DG
3576 /* Not found. Allocate one. */
3577 relayd = consumer_allocate_relayd_sock_pair(net_seq_idx);
3578 if (relayd == NULL) {
618a6a28
MD
3579 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
3580 goto error;
0d08d75e 3581 } else {
30319bcb 3582 relayd->sessiond_session_id = sessiond_id;
0d08d75e 3583 relayd_created = 1;
7735ef9e 3584 }
0d08d75e
DG
3585
3586 /*
3587 * This code path MUST continue to the consumer send status message to
3588 * we can notify the session daemon and continue our work without
3589 * killing everything.
3590 */
da009f2c
MD
3591 } else {
3592 /*
3593 * relayd key should never be found for control socket.
3594 */
3595 assert(sock_type != LTTNG_STREAM_CONTROL);
0d08d75e
DG
3596 }
3597
3598 /* First send a status message before receiving the fds. */
0c759fc9 3599 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
618a6a28 3600 if (ret < 0) {
0d08d75e 3601 /* Somehow, the session daemon is not responding anymore. */
618a6a28
MD
3602 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3603 goto error_nosignal;
7735ef9e
DG
3604 }
3605
3606 /* Poll on consumer socket. */
84382d49
MD
3607 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3608 if (ret) {
3609 /* Needing to exit in the middle of a command: error. */
0d08d75e 3610 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
618a6a28 3611 goto error_nosignal;
7735ef9e
DG
3612 }
3613
3614 /* Get relayd socket from session daemon */
3615 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
3616 if (ret != sizeof(fd)) {
4028eeb9 3617 fd = -1; /* Just in case it gets set with an invalid value. */
0d08d75e
DG
3618
3619 /*
3620 * Failing to receive FDs might indicate a major problem such as
3621 * reaching a fd limit during the receive where the kernel returns a
3622 * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we
3623 * don't take any chances and stop everything.
3624 *
3625 * XXX: Feature request #558 will fix that and avoid this possible
3626 * issue when reaching the fd limit.
3627 */
3628 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
618a6a28 3629 ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD;
f50f23d9
DG
3630 goto error;
3631 }
3632
7735ef9e
DG
3633 /* Copy socket information and received FD */
3634 switch (sock_type) {
3635 case LTTNG_STREAM_CONTROL:
3636 /* Copy received lttcomm socket */
6151a90f
JD
3637 lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock);
3638 ret = lttcomm_create_sock(&relayd->control_sock.sock);
4028eeb9 3639 /* Handle create_sock error. */
f66c074c 3640 if (ret < 0) {
618a6a28 3641 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
4028eeb9 3642 goto error;
f66c074c 3643 }
da009f2c
MD
3644 /*
3645 * Close the socket created internally by
3646 * lttcomm_create_sock, so we can replace it by the one
3647 * received from sessiond.
3648 */
3649 if (close(relayd->control_sock.sock.fd)) {
3650 PERROR("close");
3651 }
7735ef9e
DG
3652
3653 /* Assign new file descriptor */
6151a90f
JD
3654 relayd->control_sock.sock.fd = fd;
3655 /* Assign version values. */
3656 relayd->control_sock.major = relayd_sock->major;
3657 relayd->control_sock.minor = relayd_sock->minor;
c5b6f4f0 3658
d3e2ba59 3659 relayd->relayd_session_id = relayd_session_id;
c5b6f4f0 3660
7735ef9e
DG
3661 break;
3662 case LTTNG_STREAM_DATA:
3663 /* Copy received lttcomm socket */
6151a90f
JD
3664 lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock);
3665 ret = lttcomm_create_sock(&relayd->data_sock.sock);
4028eeb9 3666 /* Handle create_sock error. */
f66c074c 3667 if (ret < 0) {
618a6a28 3668 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
4028eeb9 3669 goto error;
f66c074c 3670 }
da009f2c
MD
3671 /*
3672 * Close the socket created internally by
3673 * lttcomm_create_sock, so we can replace it by the one
3674 * received from sessiond.
3675 */
3676 if (close(relayd->data_sock.sock.fd)) {
3677 PERROR("close");
3678 }
7735ef9e
DG
3679
3680 /* Assign new file descriptor */
6151a90f
JD
3681 relayd->data_sock.sock.fd = fd;
3682 /* Assign version values. */
3683 relayd->data_sock.major = relayd_sock->major;
3684 relayd->data_sock.minor = relayd_sock->minor;
7735ef9e
DG
3685 break;
3686 default:
3687 ERR("Unknown relayd socket type (%d)", sock_type);
618a6a28 3688 ret_code = LTTCOMM_CONSUMERD_FATAL;
7735ef9e
DG
3689 goto error;
3690 }
3691
d88aee68 3692 DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)",
7735ef9e
DG
3693 sock_type == LTTNG_STREAM_CONTROL ? "control" : "data",
3694 relayd->net_seq_idx, fd);
39d9954c
FD
3695 /*
3696 * We gave the ownership of the fd to the relayd structure. Set the
3697 * fd to -1 so we don't call close() on it in the error path below.
3698 */
3699 fd = -1;
7735ef9e 3700
618a6a28
MD
3701 /* We successfully added the socket. Send status back. */
3702 ret = consumer_send_status_msg(sock, ret_code);
3703 if (ret < 0) {
3704 /* Somehow, the session daemon is not responding anymore. */
3705 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3706 goto error_nosignal;
3707 }
3708
7735ef9e
DG
3709 /*
3710 * Add relayd socket pair to consumer data hashtable. If object already
3711 * exists or on error, the function gracefully returns.
3712 */
9276e5c8 3713 relayd->ctx = ctx;
d09e1200 3714 add_relayd(relayd);
7735ef9e
DG
3715
3716 /* All good! */
2527bf85 3717 return;
7735ef9e
DG
3718
3719error:
618a6a28
MD
3720 if (consumer_send_status_msg(sock, ret_code) < 0) {
3721 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3722 }
3723
3724error_nosignal:
4028eeb9
DG
3725 /* Close received socket if valid. */
3726 if (fd >= 0) {
3727 if (close(fd)) {
3728 PERROR("close received socket");
3729 }
3730 }
cd2b09ed
DG
3731
3732 if (relayd_created) {
cd2b09ed
DG
3733 free(relayd);
3734 }
7735ef9e 3735}
ca22feea 3736
f7079f67
DG
3737/*
3738 * Search for a relayd associated to the session id and return the reference.
3739 *
3740 * A rcu read side lock MUST be acquire before calling this function and locked
3741 * until the relayd object is no longer necessary.
3742 */
3743static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id)
3744{
3745 struct lttng_ht_iter iter;
f7079f67 3746 struct consumer_relayd_sock_pair *relayd = NULL;
f7079f67
DG
3747
3748 /* Iterate over all relayd since they are indexed by net_seq_idx. */
fa29bfbf
SM
3749 cds_lfht_for_each_entry(the_consumer_data.relayd_ht->ht, &iter.iter,
3750 relayd, node.node) {
18261bd1
DG
3751 /*
3752 * Check by sessiond id which is unique here where the relayd session
3753 * id might not be when having multiple relayd.
3754 */
3755 if (relayd->sessiond_session_id == id) {
f7079f67 3756 /* Found the relayd. There can be only one per id. */
18261bd1 3757 goto found;
f7079f67
DG
3758 }
3759 }
3760
18261bd1
DG
3761 return NULL;
3762
3763found:
f7079f67
DG
3764 return relayd;
3765}
3766
ca22feea
DG
3767/*
3768 * Check if for a given session id there is still data needed to be extract
3769 * from the buffers.
3770 *
6d805429 3771 * Return 1 if data is pending or else 0 meaning ready to be read.
ca22feea 3772 */
6d805429 3773int consumer_data_pending(uint64_t id)
ca22feea
DG
3774{
3775 int ret;
3776 struct lttng_ht_iter iter;
3777 struct lttng_ht *ht;
3778 struct lttng_consumer_stream *stream;
f7079f67 3779 struct consumer_relayd_sock_pair *relayd = NULL;
6d805429 3780 int (*data_pending)(struct lttng_consumer_stream *);
ca22feea 3781
6d805429 3782 DBG("Consumer data pending command on session id %" PRIu64, id);
ca22feea 3783
6f6eda74 3784 rcu_read_lock();
fa29bfbf 3785 pthread_mutex_lock(&the_consumer_data.lock);
ca22feea 3786
fa29bfbf 3787 switch (the_consumer_data.type) {
ca22feea 3788 case LTTNG_CONSUMER_KERNEL:
6d805429 3789 data_pending = lttng_kconsumer_data_pending;
ca22feea
DG
3790 break;
3791 case LTTNG_CONSUMER32_UST:
3792 case LTTNG_CONSUMER64_UST:
6d805429 3793 data_pending = lttng_ustconsumer_data_pending;
ca22feea
DG
3794 break;
3795 default:
3796 ERR("Unknown consumer data type");
3797 assert(0);
3798 }
3799
3800 /* Ease our life a bit */
fa29bfbf 3801 ht = the_consumer_data.stream_list_ht;
ca22feea 3802
c8f59ee5 3803 cds_lfht_for_each_entry_duplicate(ht->ht,
d88aee68
DG
3804 ht->hash_fct(&id, lttng_ht_seed),
3805 ht->match_fct, &id,
ca22feea 3806 &iter.iter, stream, node_session_id.node) {
bb586a6e 3807 pthread_mutex_lock(&stream->lock);
ca22feea 3808
4e9a4686
DG
3809 /*
3810 * A removed node from the hash table indicates that the stream has
3811 * been deleted thus having a guarantee that the buffers are closed
3812 * on the consumer side. However, data can still be transmitted
3813 * over the network so don't skip the relayd check.
3814 */
3815 ret = cds_lfht_is_node_deleted(&stream->node.node);
3816 if (!ret) {
3817 /* Check the stream if there is data in the buffers. */
6d805429
DG
3818 ret = data_pending(stream);
3819 if (ret == 1) {
4e9a4686 3820 pthread_mutex_unlock(&stream->lock);
f7079f67 3821 goto data_pending;
4e9a4686
DG
3822 }
3823 }
3824
d9f0c7c7
JR
3825 pthread_mutex_unlock(&stream->lock);
3826 }
3827
3828 relayd = find_relayd_by_session_id(id);
3829 if (relayd) {
3830 unsigned int is_data_inflight = 0;
3831
3832 /* Send init command for data pending. */
3833 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3834 ret = relayd_begin_data_pending(&relayd->control_sock,
3835 relayd->relayd_session_id);
3836 if (ret < 0) {
3837 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
3838 /* Communication error thus the relayd so no data pending. */
3839 goto data_not_pending;
3840 }
3841
3842 cds_lfht_for_each_entry_duplicate(ht->ht,
3843 ht->hash_fct(&id, lttng_ht_seed),
3844 ht->match_fct, &id,
3845 &iter.iter, stream, node_session_id.node) {
c8f59ee5 3846 if (stream->metadata_flag) {
ad7051c0
DG
3847 ret = relayd_quiescent_control(&relayd->control_sock,
3848 stream->relayd_stream_id);
c8f59ee5 3849 } else {
6d805429 3850 ret = relayd_data_pending(&relayd->control_sock,
39df6d9f
DG
3851 stream->relayd_stream_id,
3852 stream->next_net_seq_num - 1);
c8f59ee5 3853 }
d9f0c7c7
JR
3854
3855 if (ret == 1) {
3856 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
3857 goto data_pending;
3858 } else if (ret < 0) {
9276e5c8
JR
3859 ERR("Relayd data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
3860 lttng_consumer_cleanup_relayd(relayd);
3861 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
9276e5c8
JR
3862 goto data_not_pending;
3863 }
c8f59ee5 3864 }
f7079f67 3865
d9f0c7c7 3866 /* Send end command for data pending. */
f7079f67
DG
3867 ret = relayd_end_data_pending(&relayd->control_sock,
3868 relayd->relayd_session_id, &is_data_inflight);
3869 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
bdd88757 3870 if (ret < 0) {
9276e5c8
JR
3871 ERR("Relayd end data pending failed. Cleaning up relayd %" PRIu64".", relayd->net_seq_idx);
3872 lttng_consumer_cleanup_relayd(relayd);
f7079f67
DG
3873 goto data_not_pending;
3874 }
bdd88757
DG
3875 if (is_data_inflight) {
3876 goto data_pending;
3877 }
f7079f67
DG
3878 }
3879
ca22feea 3880 /*
f7079f67
DG
3881 * Finding _no_ node in the hash table and no inflight data means that the
3882 * stream(s) have been removed thus data is guaranteed to be available for
3883 * analysis from the trace files.
ca22feea
DG
3884 */
3885
f7079f67 3886data_not_pending:
ca22feea 3887 /* Data is available to be read by a viewer. */
fa29bfbf 3888 pthread_mutex_unlock(&the_consumer_data.lock);
c8f59ee5 3889 rcu_read_unlock();
6d805429 3890 return 0;
ca22feea 3891
f7079f67 3892data_pending:
ca22feea 3893 /* Data is still being extracted from buffers. */
fa29bfbf 3894 pthread_mutex_unlock(&the_consumer_data.lock);
c8f59ee5 3895 rcu_read_unlock();
6d805429 3896 return 1;
ca22feea 3897}
f50f23d9
DG
3898
3899/*
3900 * Send a ret code status message to the sessiond daemon.
3901 *
3902 * Return the sendmsg() return value.
3903 */
3904int consumer_send_status_msg(int sock, int ret_code)
3905{
3906 struct lttcomm_consumer_status_msg msg;
3907
53efb85a 3908 memset(&msg, 0, sizeof(msg));
f50f23d9
DG
3909 msg.ret_code = ret_code;
3910
3911 return lttcomm_send_unix_sock(sock, &msg, sizeof(msg));
3912}
ffe60014
DG
3913
3914/*
3915 * Send a channel status message to the sessiond daemon.
3916 *
3917 * Return the sendmsg() return value.
3918 */
3919int consumer_send_status_channel(int sock,
3920 struct lttng_consumer_channel *channel)
3921{
3922 struct lttcomm_consumer_status_channel msg;
3923
3924 assert(sock >= 0);
3925
53efb85a 3926 memset(&msg, 0, sizeof(msg));
ffe60014 3927 if (!channel) {
0c759fc9 3928 msg.ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
ffe60014 3929 } else {
0c759fc9 3930 msg.ret_code = LTTCOMM_CONSUMERD_SUCCESS;
ffe60014
DG
3931 msg.key = channel->key;
3932 msg.stream_count = channel->streams.count;
3933 }
3934
3935 return lttcomm_send_unix_sock(sock, &msg, sizeof(msg));
3936}
5c786ded 3937
d07ceecd
MD
3938unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos,
3939 unsigned long produced_pos, uint64_t nb_packets_per_stream,
3940 uint64_t max_sb_size)
5c786ded 3941{
d07ceecd 3942 unsigned long start_pos;
5c786ded 3943
d07ceecd
MD
3944 if (!nb_packets_per_stream) {
3945 return consumed_pos; /* Grab everything */
3946 }
3947 start_pos = produced_pos - offset_align_floor(produced_pos, max_sb_size);
3948 start_pos -= max_sb_size * nb_packets_per_stream;
3949 if ((long) (start_pos - consumed_pos) < 0) {
3950 return consumed_pos; /* Grab everything */
3951 }
3952 return start_pos;
5c786ded 3953}
a1ae2ea5 3954
c1dcb8bb
JG
3955/* Stream lock must be held by the caller. */
3956static int sample_stream_positions(struct lttng_consumer_stream *stream,
3957 unsigned long *produced, unsigned long *consumed)
3958{
3959 int ret;
3960
3961 ASSERT_LOCKED(stream->lock);
3962
3963 ret = lttng_consumer_sample_snapshot_positions(stream);
3964 if (ret < 0) {
3965 ERR("Failed to sample snapshot positions");
3966 goto end;
3967 }
3968
3969 ret = lttng_consumer_get_produced_snapshot(stream, produced);
3970 if (ret < 0) {
3971 ERR("Failed to sample produced position");
3972 goto end;
3973 }
3974
3975 ret = lttng_consumer_get_consumed_snapshot(stream, consumed);
3976 if (ret < 0) {
3977 ERR("Failed to sample consumed position");
3978 goto end;
3979 }
3980
3981end:
3982 return ret;
3983}
3984
b99a8d42
JD
3985/*
3986 * Sample the rotate position for all the streams of a channel. If a stream
3987 * is already at the rotate position (produced == consumed), we flag it as
3988 * ready for rotation. The rotation of ready streams occurs after we have
3989 * replied to the session daemon that we have finished sampling the positions.
92b7a7f8 3990 * Must be called with RCU read-side lock held to ensure existence of channel.
b99a8d42
JD
3991 *
3992 * Returns 0 on success, < 0 on error
3993 */
92b7a7f8 3994int lttng_consumer_rotate_channel(struct lttng_consumer_channel *channel,
d2956687 3995 uint64_t key, uint64_t relayd_id, uint32_t metadata,
b99a8d42
JD
3996 struct lttng_consumer_local_data *ctx)
3997{
3998 int ret;
b99a8d42
JD
3999 struct lttng_consumer_stream *stream;
4000 struct lttng_ht_iter iter;
fa29bfbf 4001 struct lttng_ht *ht = the_consumer_data.stream_per_chan_id_ht;
c35f9726
JG
4002 struct lttng_dynamic_array stream_rotation_positions;
4003 uint64_t next_chunk_id, stream_count = 0;
4004 enum lttng_trace_chunk_status chunk_status;
4005 const bool is_local_trace = relayd_id == -1ULL;
4006 struct consumer_relayd_sock_pair *relayd = NULL;
4007 bool rotating_to_new_chunk = true;
b32703d6
JG
4008 /* Array of `struct lttng_consumer_stream *` */
4009 struct lttng_dynamic_pointer_array streams_packet_to_open;
4010 size_t stream_idx;
b99a8d42
JD
4011
4012 DBG("Consumer sample rotate position for channel %" PRIu64, key);
4013
c35f9726
JG
4014 lttng_dynamic_array_init(&stream_rotation_positions,
4015 sizeof(struct relayd_stream_rotation_position), NULL);
b32703d6 4016 lttng_dynamic_pointer_array_init(&streams_packet_to_open, NULL);
c35f9726 4017
b99a8d42
JD
4018 rcu_read_lock();
4019
b99a8d42 4020 pthread_mutex_lock(&channel->lock);
c35f9726
JG
4021 assert(channel->trace_chunk);
4022 chunk_status = lttng_trace_chunk_get_id(channel->trace_chunk,
4023 &next_chunk_id);
4024 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
4025 ret = -1;
4026 goto end_unlock_channel;
4027 }
b99a8d42
JD
4028
4029 cds_lfht_for_each_entry_duplicate(ht->ht,
4030 ht->hash_fct(&channel->key, lttng_ht_seed),
4031 ht->match_fct, &channel->key, &iter.iter,
4032 stream, node_channel_id.node) {
a40a503f 4033 unsigned long produced_pos = 0, consumed_pos = 0;
b99a8d42
JD
4034
4035 health_code_update();
4036
4037 /*
4038 * Lock stream because we are about to change its state.
4039 */
4040 pthread_mutex_lock(&stream->lock);
4041
c35f9726
JG
4042 if (stream->trace_chunk == stream->chan->trace_chunk) {
4043 rotating_to_new_chunk = false;
4044 }
4045
a40a503f 4046 /*
c1dcb8bb 4047 * Do not flush a packet when rotating from a NULL trace
a9dde553 4048 * chunk. The stream has no means to output data, and the prior
c1dcb8bb
JG
4049 * rotation which rotated to NULL performed that side-effect
4050 * already. No new data can be produced when a stream has no
4051 * associated trace chunk (e.g. a stop followed by a rotate).
a40a503f 4052 */
a9dde553 4053 if (stream->trace_chunk) {
c1dcb8bb
JG
4054 bool flush_active;
4055
4056 if (stream->metadata_flag) {
4057 /*
4058 * Don't produce an empty metadata packet,
4059 * simply close the current one.
4060 *
4061 * Metadata is regenerated on every trace chunk
4062 * switch; there is no concern that no data was
4063 * produced.
4064 */
4065 flush_active = true;
4066 } else {
4067 /*
4068 * Only flush an empty packet if the "packet
4069 * open" could not be performed on transition
4070 * to a new trace chunk and no packets were
4071 * consumed within the chunk's lifetime.
4072 */
4073 if (stream->opened_packet_in_current_trace_chunk) {
4074 flush_active = true;
4075 } else {
4076 /*
4077 * Stream could have been full at the
4078 * time of rotation, but then have had
4079 * no activity at all.
4080 *
4081 * It is important to flush a packet
4082 * to prevent 0-length files from being
4083 * produced as most viewers choke on
4084 * them.
4085 *
4086 * Unfortunately viewers will not be
4087 * able to know that tracing was active
4088 * for this stream during this trace
4089 * chunk's lifetime.
4090 */
4091 ret = sample_stream_positions(stream, &produced_pos, &consumed_pos);
4092 if (ret) {
4093 goto end_unlock_stream;
4094 }
4095
4096 /*
4097 * Don't flush an empty packet if data
4098 * was produced; it will be consumed
4099 * before the rotation completes.
4100 */
4101 flush_active = produced_pos != consumed_pos;
4102 if (!flush_active) {
c1dcb8bb
JG
4103 const char *trace_chunk_name;
4104 uint64_t trace_chunk_id;
4105
4106 chunk_status = lttng_trace_chunk_get_name(
4107 stream->trace_chunk,
4108 &trace_chunk_name,
4109 NULL);
4110 if (chunk_status == LTTNG_TRACE_CHUNK_STATUS_NONE) {
4111 trace_chunk_name = "none";
4112 }
4113
4114 /*
4115 * Consumer trace chunks are
4116 * never anonymous.
4117 */
4118 chunk_status = lttng_trace_chunk_get_id(
4119 stream->trace_chunk,
4120 &trace_chunk_id);
4121 assert(chunk_status ==
4122 LTTNG_TRACE_CHUNK_STATUS_OK);
4123
4124 DBG("Unable to open packet for stream during trace chunk's lifetime. "
4125 "Flushing an empty packet to prevent an empty file from being created: "
4126 "stream id = %" PRIu64 ", trace chunk name = `%s`, trace chunk id = %" PRIu64,
4127 stream->key, trace_chunk_name, trace_chunk_id);
4128 }
4129 }
4130 }
4131
a9dde553 4132 /*
c1dcb8bb
JG
4133 * Close the current packet before sampling the
4134 * ring buffer positions.
a9dde553 4135 */
c1dcb8bb 4136 ret = consumer_stream_flush_buffer(stream, flush_active);
a9dde553
MD
4137 if (ret < 0) {
4138 ERR("Failed to flush stream %" PRIu64 " during channel rotation",
4139 stream->key);
4140 goto end_unlock_stream;
4141 }
b99a8d42
JD
4142 }
4143
a40a503f
MD
4144 ret = lttng_consumer_take_snapshot(stream);
4145 if (ret < 0 && ret != -ENODATA && ret != -EAGAIN) {
4146 ERR("Failed to sample snapshot position during channel rotation");
b99a8d42
JD
4147 goto end_unlock_stream;
4148 }
a40a503f
MD
4149 if (!ret) {
4150 ret = lttng_consumer_get_produced_snapshot(stream,
4151 &produced_pos);
4152 if (ret < 0) {
4153 ERR("Failed to sample produced position during channel rotation");
4154 goto end_unlock_stream;
4155 }
b99a8d42 4156
a40a503f
MD
4157 ret = lttng_consumer_get_consumed_snapshot(stream,
4158 &consumed_pos);
4159 if (ret < 0) {
4160 ERR("Failed to sample consumed position during channel rotation");
4161 goto end_unlock_stream;
4162 }
4163 }
4164 /*
4165 * Align produced position on the start-of-packet boundary of the first
4166 * packet going into the next trace chunk.
4167 */
4168 produced_pos = ALIGN_FLOOR(produced_pos, stream->max_sb_size);
4169 if (consumed_pos == produced_pos) {
f8528c7a
MD
4170 DBG("Set rotate ready for stream %" PRIu64 " produced = %lu consumed = %lu",
4171 stream->key, produced_pos, consumed_pos);
b99a8d42 4172 stream->rotate_ready = true;
f8528c7a
MD
4173 } else {
4174 DBG("Different consumed and produced positions "
4175 "for stream %" PRIu64 " produced = %lu consumed = %lu",
4176 stream->key, produced_pos, consumed_pos);
b99a8d42 4177 }
633d0182 4178 /*
a40a503f
MD
4179 * The rotation position is based on the packet_seq_num of the
4180 * packet following the last packet that was consumed for this
4181 * stream, incremented by the offset between produced and
4182 * consumed positions. This rotation position is a lower bound
4183 * (inclusive) at which the next trace chunk starts. Since it
4184 * is a lower bound, it is OK if the packet_seq_num does not
4185 * correspond exactly to the same packet identified by the
4186 * consumed_pos, which can happen in overwrite mode.
633d0182 4187 */
a40a503f
MD
4188 if (stream->sequence_number_unavailable) {
4189 /*
4190 * Rotation should never be performed on a session which
4191 * interacts with a pre-2.8 lttng-modules, which does
4192 * not implement packet sequence number.
4193 */
4194 ERR("Failure to rotate stream %" PRIu64 ": sequence number unavailable",
b99a8d42 4195 stream->key);
a40a503f 4196 ret = -1;
b99a8d42
JD
4197 goto end_unlock_stream;
4198 }
a40a503f
MD
4199 stream->rotate_position = stream->last_sequence_number + 1 +
4200 ((produced_pos - consumed_pos) / stream->max_sb_size);
f8528c7a
MD
4201 DBG("Set rotation position for stream %" PRIu64 " at position %" PRIu64,
4202 stream->key, stream->rotate_position);
b99a8d42 4203
c35f9726 4204 if (!is_local_trace) {
633d0182
JG
4205 /*
4206 * The relay daemon control protocol expects a rotation
4207 * position as "the sequence number of the first packet
a40a503f 4208 * _after_ the current trace chunk".
633d0182 4209 */
c35f9726
JG
4210 const struct relayd_stream_rotation_position position = {
4211 .stream_id = stream->relayd_stream_id,
a40a503f 4212 .rotate_at_seq_num = stream->rotate_position,
c35f9726
JG
4213 };
4214
4215 ret = lttng_dynamic_array_add_element(
4216 &stream_rotation_positions,
4217 &position);
4218 if (ret) {
4219 ERR("Failed to allocate stream rotation position");
4220 goto end_unlock_stream;
4221 }
4222 stream_count++;
4223 }
f96af312
JG
4224
4225 stream->opened_packet_in_current_trace_chunk = false;
4226
4227 if (rotating_to_new_chunk && !stream->metadata_flag) {
4228 /*
4229 * Attempt to flush an empty packet as close to the
4230 * rotation point as possible. In the event where a
4231 * stream remains inactive after the rotation point,
4232 * this ensures that the new trace chunk has a
4233 * beginning timestamp set at the begining of the
4234 * trace chunk instead of only creating an empty
4235 * packet when the trace chunk is stopped.
4236 *
4237 * This indicates to the viewers that the stream
4238 * was being recorded, but more importantly it
4239 * allows viewers to determine a useable trace
4240 * intersection.
4241 *
4242 * This presents a problem in the case where the
4243 * ring-buffer is completely full.
4244 *
4245 * Consider the following scenario:
4246 * - The consumption of data is slow (slow network,
4247 * for instance),
4248 * - The ring buffer is full,
4249 * - A rotation is initiated,
4250 * - The flush below does nothing (no space left to
4251 * open a new packet),
4252 * - The other streams rotate very soon, and new
4253 * data is produced in the new chunk,
4254 * - This stream completes its rotation long after the
4255 * rotation was initiated
4256 * - The session is stopped before any event can be
4257 * produced in this stream's buffers.
4258 *
4259 * The resulting trace chunk will have a single packet
4260 * temporaly at the end of the trace chunk for this
4261 * stream making the stream intersection more narrow
4262 * than it should be.
4263 *
4264 * To work-around this, an empty flush is performed
4265 * after the first consumption of a packet during a
4266 * rotation if open_packet fails. The idea is that
4267 * consuming a packet frees enough space to switch
4268 * packets in this scenario and allows the tracer to
4269 * "stamp" the beginning of the new trace chunk at the
4270 * earliest possible point.
b32703d6
JG
4271 *
4272 * The packet open is performed after the channel
4273 * rotation to ensure that no attempt to open a packet
4274 * is performed in a stream that has no active trace
4275 * chunk.
f96af312 4276 */
b32703d6
JG
4277 ret = lttng_dynamic_pointer_array_add_pointer(
4278 &streams_packet_to_open, stream);
4279 if (ret) {
4280 PERROR("Failed to add a stream pointer to array of streams in which to open a packet");
f96af312
JG
4281 ret = -1;
4282 goto end_unlock_stream;
f96af312
JG
4283 }
4284 }
4285
b99a8d42
JD
4286 pthread_mutex_unlock(&stream->lock);
4287 }
c35f9726 4288 stream = NULL;
b99a8d42 4289
b32703d6
JG
4290 if (!is_local_trace) {
4291 relayd = consumer_find_relayd(relayd_id);
4292 if (!relayd) {
4293 ERR("Failed to find relayd %" PRIu64, relayd_id);
4294 ret = -1;
4295 goto end_unlock_channel;
4296 }
c35f9726 4297
b32703d6
JG
4298 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
4299 ret = relayd_rotate_streams(&relayd->control_sock, stream_count,
4300 rotating_to_new_chunk ? &next_chunk_id : NULL,
4301 (const struct relayd_stream_rotation_position *)
4302 stream_rotation_positions.buffer
4303 .data);
4304 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
4305 if (ret < 0) {
4306 ERR("Relayd rotate stream failed. Cleaning up relayd %" PRIu64,
4307 relayd->net_seq_idx);
4308 lttng_consumer_cleanup_relayd(relayd);
4309 goto end_unlock_channel;
4310 }
c35f9726
JG
4311 }
4312
b32703d6
JG
4313 for (stream_idx = 0;
4314 stream_idx < lttng_dynamic_pointer_array_get_count(
4315 &streams_packet_to_open);
4316 stream_idx++) {
4317 enum consumer_stream_open_packet_status status;
4318
4319 stream = lttng_dynamic_pointer_array_get_pointer(
4320 &streams_packet_to_open, stream_idx);
4321
4322 pthread_mutex_lock(&stream->lock);
4323 status = consumer_stream_open_packet(stream);
4324 pthread_mutex_unlock(&stream->lock);
4325 switch (status) {
4326 case CONSUMER_STREAM_OPEN_PACKET_STATUS_OPENED:
4327 DBG("Opened a packet after a rotation: stream id = %" PRIu64
4328 ", channel name = %s, session id = %" PRIu64,
4329 stream->key, stream->chan->name,
4330 stream->chan->session_id);
4331 break;
4332 case CONSUMER_STREAM_OPEN_PACKET_STATUS_NO_SPACE:
4333 /*
4334 * Can't open a packet as there is no space left
4335 * in the buffer. A new packet will be opened
4336 * once one has been consumed.
4337 */
4338 DBG("No space left to open a packet after a rotation: stream id = %" PRIu64
4339 ", channel name = %s, session id = %" PRIu64,
4340 stream->key, stream->chan->name,
4341 stream->chan->session_id);
4342 break;
4343 case CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR:
4344 /* Logged by callee. */
4345 ret = -1;
7a86c13d 4346 goto end_unlock_channel;
b32703d6
JG
4347 default:
4348 abort();
4349 }
c35f9726
JG
4350 }
4351
b32703d6 4352 pthread_mutex_unlock(&channel->lock);
b99a8d42
JD
4353 ret = 0;
4354 goto end;
4355
4356end_unlock_stream:
4357 pthread_mutex_unlock(&stream->lock);
c35f9726 4358end_unlock_channel:
b99a8d42
JD
4359 pthread_mutex_unlock(&channel->lock);
4360end:
4361 rcu_read_unlock();
c35f9726 4362 lttng_dynamic_array_reset(&stream_rotation_positions);
b32703d6 4363 lttng_dynamic_pointer_array_reset(&streams_packet_to_open);
b99a8d42
JD
4364 return ret;
4365}
4366
5f3aff8b
MD
4367static
4368int consumer_clear_buffer(struct lttng_consumer_stream *stream)
4369{
4370 int ret = 0;
4371 unsigned long consumed_pos_before, consumed_pos_after;
4372
4373 ret = lttng_consumer_sample_snapshot_positions(stream);
4374 if (ret < 0) {
4375 ERR("Taking snapshot positions");
4376 goto end;
4377 }
4378
4379 ret = lttng_consumer_get_consumed_snapshot(stream, &consumed_pos_before);
4380 if (ret < 0) {
4381 ERR("Consumed snapshot position");
4382 goto end;
4383 }
4384
fa29bfbf 4385 switch (the_consumer_data.type) {
5f3aff8b
MD
4386 case LTTNG_CONSUMER_KERNEL:
4387 ret = kernctl_buffer_clear(stream->wait_fd);
4388 if (ret < 0) {
96393977 4389 ERR("Failed to clear kernel stream (ret = %d)", ret);
5f3aff8b
MD
4390 goto end;
4391 }
4392 break;
4393 case LTTNG_CONSUMER32_UST:
4394 case LTTNG_CONSUMER64_UST:
4395 lttng_ustconsumer_clear_buffer(stream);
4396 break;
4397 default:
4398 ERR("Unknown consumer_data type");
4399 abort();
4400 }
4401
4402 ret = lttng_consumer_sample_snapshot_positions(stream);
4403 if (ret < 0) {
4404 ERR("Taking snapshot positions");
4405 goto end;
4406 }
4407 ret = lttng_consumer_get_consumed_snapshot(stream, &consumed_pos_after);
4408 if (ret < 0) {
4409 ERR("Consumed snapshot position");
4410 goto end;
4411 }
4412 DBG("clear: before: %lu after: %lu", consumed_pos_before, consumed_pos_after);
4413end:
4414 return ret;
4415}
4416
4417static
4418int consumer_clear_stream(struct lttng_consumer_stream *stream)
4419{
4420 int ret;
4421
503fefca 4422 ret = consumer_stream_flush_buffer(stream, 1);
5f3aff8b
MD
4423 if (ret < 0) {
4424 ERR("Failed to flush stream %" PRIu64 " during channel clear",
4425 stream->key);
4426 ret = LTTCOMM_CONSUMERD_FATAL;
4427 goto error;
4428 }
4429
4430 ret = consumer_clear_buffer(stream);
4431 if (ret < 0) {
4432 ERR("Failed to clear stream %" PRIu64 " during channel clear",
4433 stream->key);
4434 ret = LTTCOMM_CONSUMERD_FATAL;
4435 goto error;
4436 }
4437
4438 ret = LTTCOMM_CONSUMERD_SUCCESS;
4439error:
4440 return ret;
4441}
4442
4443static
4444int consumer_clear_unmonitored_channel(struct lttng_consumer_channel *channel)
4445{
4446 int ret;
4447 struct lttng_consumer_stream *stream;
4448
4449 rcu_read_lock();
4450 pthread_mutex_lock(&channel->lock);
4451 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
4452 health_code_update();
4453 pthread_mutex_lock(&stream->lock);
4454 ret = consumer_clear_stream(stream);
4455 if (ret) {
4456 goto error_unlock;
4457 }
4458 pthread_mutex_unlock(&stream->lock);
4459 }
4460 pthread_mutex_unlock(&channel->lock);
4461 rcu_read_unlock();
4462 return 0;
4463
4464error_unlock:
4465 pthread_mutex_unlock(&stream->lock);
4466 pthread_mutex_unlock(&channel->lock);
4467 rcu_read_unlock();
5f3aff8b
MD
4468 return ret;
4469}
4470
02d02e31
JD
4471/*
4472 * Check if a stream is ready to be rotated after extracting it.
4473 *
4474 * Return 1 if it is ready for rotation, 0 if it is not, a negative value on
4475 * error. Stream lock must be held.
4476 */
4477int lttng_consumer_stream_is_rotate_ready(struct lttng_consumer_stream *stream)
4478{
f8528c7a
MD
4479 DBG("Check is rotate ready for stream %" PRIu64
4480 " ready %u rotate_position %" PRIu64
4481 " last_sequence_number %" PRIu64,
4482 stream->key, stream->rotate_ready,
4483 stream->rotate_position, stream->last_sequence_number);
02d02e31 4484 if (stream->rotate_ready) {
a40a503f 4485 return 1;
02d02e31
JD
4486 }
4487
4488 /*
a40a503f
MD
4489 * If packet seq num is unavailable, it means we are interacting
4490 * with a pre-2.8 lttng-modules which does not implement the
4491 * sequence number. Rotation should never be used by sessiond in this
4492 * scenario.
02d02e31 4493 */
a40a503f
MD
4494 if (stream->sequence_number_unavailable) {
4495 ERR("Internal error: rotation used on stream %" PRIu64
4496 " with unavailable sequence number",
4497 stream->key);
4498 return -1;
02d02e31
JD
4499 }
4500
a40a503f
MD
4501 if (stream->rotate_position == -1ULL ||
4502 stream->last_sequence_number == -1ULL) {
4503 return 0;
02d02e31
JD
4504 }
4505
a40a503f
MD
4506 /*
4507 * Rotate position not reached yet. The stream rotate position is
4508 * the position of the next packet belonging to the next trace chunk,
4509 * but consumerd considers rotation ready when reaching the last
4510 * packet of the current chunk, hence the "rotate_position - 1".
4511 */
f8528c7a
MD
4512
4513 DBG("Check is rotate ready for stream %" PRIu64
4514 " last_sequence_number %" PRIu64
4515 " rotate_position %" PRIu64,
4516 stream->key, stream->last_sequence_number,
4517 stream->rotate_position);
a40a503f
MD
4518 if (stream->last_sequence_number >= stream->rotate_position - 1) {
4519 return 1;
02d02e31 4520 }
02d02e31 4521
a40a503f 4522 return 0;
02d02e31
JD
4523}
4524
d73bf3d7
JD
4525/*
4526 * Reset the state for a stream after a rotation occurred.
4527 */
4528void lttng_consumer_reset_stream_rotate_state(struct lttng_consumer_stream *stream)
4529{
f8528c7a
MD
4530 DBG("lttng_consumer_reset_stream_rotate_state for stream %" PRIu64,
4531 stream->key);
a40a503f 4532 stream->rotate_position = -1ULL;
d73bf3d7
JD
4533 stream->rotate_ready = false;
4534}
4535
4536/*
4537 * Perform the rotation a local stream file.
4538 */
d2956687 4539static
d73bf3d7
JD
4540int rotate_local_stream(struct lttng_consumer_local_data *ctx,
4541 struct lttng_consumer_stream *stream)
4542{
d2956687 4543 int ret = 0;
d73bf3d7 4544
d2956687 4545 DBG("Rotate local stream: stream key %" PRIu64 ", channel key %" PRIu64,
d73bf3d7 4546 stream->key,
d2956687 4547 stream->chan->key);
d73bf3d7 4548 stream->tracefile_size_current = 0;
d2956687 4549 stream->tracefile_count_current = 0;
d73bf3d7 4550
d2956687
JG
4551 if (stream->out_fd >= 0) {
4552 ret = close(stream->out_fd);
4553 if (ret) {
4554 PERROR("Failed to close stream out_fd of channel \"%s\"",
4555 stream->chan->name);
4556 }
4557 stream->out_fd = -1;
4558 }
d73bf3d7 4559
d2956687 4560 if (stream->index_file) {
d73bf3d7 4561 lttng_index_file_put(stream->index_file);
d2956687 4562 stream->index_file = NULL;
d73bf3d7
JD
4563 }
4564
d2956687
JG
4565 if (!stream->trace_chunk) {
4566 goto end;
4567 }
d73bf3d7 4568
d2956687 4569 ret = consumer_stream_create_output_files(stream, true);
d73bf3d7
JD
4570end:
4571 return ret;
d73bf3d7
JD
4572}
4573
d73bf3d7
JD
4574/*
4575 * Performs the stream rotation for the rotate session feature if needed.
d2956687 4576 * It must be called with the channel and stream locks held.
d73bf3d7
JD
4577 *
4578 * Return 0 on success, a negative number of error.
4579 */
4580int lttng_consumer_rotate_stream(struct lttng_consumer_local_data *ctx,
d2956687 4581 struct lttng_consumer_stream *stream)
d73bf3d7
JD
4582{
4583 int ret;
4584
4585 DBG("Consumer rotate stream %" PRIu64, stream->key);
4586
d2956687
JG
4587 /*
4588 * Update the stream's 'current' chunk to the session's (channel)
4589 * now-current chunk.
4590 */
4591 lttng_trace_chunk_put(stream->trace_chunk);
4592 if (stream->chan->trace_chunk == stream->trace_chunk) {
4593 /*
4594 * A channel can be rotated and not have a "next" chunk
4595 * to transition to. In that case, the channel's "current chunk"
4596 * has not been closed yet, but it has not been updated to
4597 * a "next" trace chunk either. Hence, the stream, like its
4598 * parent channel, becomes part of no chunk and can't output
4599 * anything until a new trace chunk is created.
4600 */
4601 stream->trace_chunk = NULL;
4602 } else if (stream->chan->trace_chunk &&
4603 !lttng_trace_chunk_get(stream->chan->trace_chunk)) {
4604 ERR("Failed to acquire a reference to channel's trace chunk during stream rotation");
4605 ret = -1;
4606 goto error;
4607 } else {
4608 /*
4609 * Update the stream's trace chunk to its parent channel's
4610 * current trace chunk.
4611 */
4612 stream->trace_chunk = stream->chan->trace_chunk;
4613 }
4614
c35f9726 4615 if (stream->net_seq_idx == (uint64_t) -1ULL) {
d73bf3d7 4616 ret = rotate_local_stream(ctx, stream);
c35f9726
JG
4617 if (ret < 0) {
4618 ERR("Failed to rotate stream, ret = %i", ret);
4619 goto error;
4620 }
d73bf3d7
JD
4621 }
4622
d2956687
JG
4623 if (stream->metadata_flag && stream->trace_chunk) {
4624 /*
4625 * If the stream has transitioned to a new trace
4626 * chunk, the metadata should be re-dumped to the
4627 * newest chunk.
4628 *
4629 * However, it is possible for a stream to transition to
4630 * a "no-chunk" state. This can happen if a rotation
4631 * occurs on an inactive session. In such cases, the metadata
4632 * regeneration will happen when the next trace chunk is
4633 * created.
4634 */
4635 ret = consumer_metadata_stream_dump(stream);
4636 if (ret) {
4637 goto error;
d73bf3d7
JD
4638 }
4639 }
4640 lttng_consumer_reset_stream_rotate_state(stream);
4641
4642 ret = 0;
4643
4644error:
4645 return ret;
4646}
4647
b99a8d42
JD
4648/*
4649 * Rotate all the ready streams now.
4650 *
4651 * This is especially important for low throughput streams that have already
4652 * been consumed, we cannot wait for their next packet to perform the
4653 * rotation.
92b7a7f8
MD
4654 * Need to be called with RCU read-side lock held to ensure existence of
4655 * channel.
b99a8d42
JD
4656 *
4657 * Returns 0 on success, < 0 on error
4658 */
92b7a7f8
MD
4659int lttng_consumer_rotate_ready_streams(struct lttng_consumer_channel *channel,
4660 uint64_t key, struct lttng_consumer_local_data *ctx)
b99a8d42
JD
4661{
4662 int ret;
b99a8d42
JD
4663 struct lttng_consumer_stream *stream;
4664 struct lttng_ht_iter iter;
fa29bfbf 4665 struct lttng_ht *ht = the_consumer_data.stream_per_chan_id_ht;
b99a8d42
JD
4666
4667 rcu_read_lock();
4668
4669 DBG("Consumer rotate ready streams in channel %" PRIu64, key);
4670
b99a8d42
JD
4671 cds_lfht_for_each_entry_duplicate(ht->ht,
4672 ht->hash_fct(&channel->key, lttng_ht_seed),
4673 ht->match_fct, &channel->key, &iter.iter,
4674 stream, node_channel_id.node) {
4675 health_code_update();
4676
d2956687 4677 pthread_mutex_lock(&stream->chan->lock);
b99a8d42
JD
4678 pthread_mutex_lock(&stream->lock);
4679
4680 if (!stream->rotate_ready) {
4681 pthread_mutex_unlock(&stream->lock);
d2956687 4682 pthread_mutex_unlock(&stream->chan->lock);
b99a8d42
JD
4683 continue;
4684 }
4685 DBG("Consumer rotate ready stream %" PRIu64, stream->key);
4686
d2956687 4687 ret = lttng_consumer_rotate_stream(ctx, stream);
b99a8d42 4688 pthread_mutex_unlock(&stream->lock);
d2956687 4689 pthread_mutex_unlock(&stream->chan->lock);
b99a8d42
JD
4690 if (ret) {
4691 goto end;
4692 }
4693 }
4694
4695 ret = 0;
4696
4697end:
4698 rcu_read_unlock();
4699 return ret;
4700}
4701
d2956687
JG
4702enum lttcomm_return_code lttng_consumer_init_command(
4703 struct lttng_consumer_local_data *ctx,
4704 const lttng_uuid sessiond_uuid)
00fb02ac 4705{
d2956687 4706 enum lttcomm_return_code ret;
c70636a7 4707 char uuid_str[LTTNG_UUID_STR_LEN];
00fb02ac 4708
d2956687
JG
4709 if (ctx->sessiond_uuid.is_set) {
4710 ret = LTTCOMM_CONSUMERD_ALREADY_SET;
00fb02ac
JD
4711 goto end;
4712 }
4713
d2956687
JG
4714 ctx->sessiond_uuid.is_set = true;
4715 memcpy(ctx->sessiond_uuid.value, sessiond_uuid, sizeof(lttng_uuid));
4716 ret = LTTCOMM_CONSUMERD_SUCCESS;
4717 lttng_uuid_to_str(sessiond_uuid, uuid_str);
4718 DBG("Received session daemon UUID: %s", uuid_str);
00fb02ac
JD
4719end:
4720 return ret;
4721}
4722
d2956687
JG
4723enum lttcomm_return_code lttng_consumer_create_trace_chunk(
4724 const uint64_t *relayd_id, uint64_t session_id,
4725 uint64_t chunk_id,
4726 time_t chunk_creation_timestamp,
4727 const char *chunk_override_name,
4728 const struct lttng_credentials *credentials,
4729 struct lttng_directory_handle *chunk_directory_handle)
00fb02ac
JD
4730{
4731 int ret;
d2956687 4732 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
7ea24db3 4733 struct lttng_trace_chunk *created_chunk = NULL, *published_chunk = NULL;
d2956687
JG
4734 enum lttng_trace_chunk_status chunk_status;
4735 char relayd_id_buffer[MAX_INT_DEC_LEN(*relayd_id)];
4736 char creation_timestamp_buffer[ISO8601_STR_LEN];
4737 const char *relayd_id_str = "(none)";
4738 const char *creation_timestamp_str;
4739 struct lttng_ht_iter iter;
4740 struct lttng_consumer_channel *channel;
92816cc3 4741
d2956687
JG
4742 if (relayd_id) {
4743 /* Only used for logging purposes. */
4744 ret = snprintf(relayd_id_buffer, sizeof(relayd_id_buffer),
4745 "%" PRIu64, *relayd_id);
4746 if (ret > 0 && ret < sizeof(relayd_id_buffer)) {
4747 relayd_id_str = relayd_id_buffer;
4748 } else {
4749 relayd_id_str = "(formatting error)";
4750 }
d01ef216 4751 }
d2956687 4752
d01ef216 4753 /* Local protocol error. */
d2956687
JG
4754 assert(chunk_creation_timestamp);
4755 ret = time_to_iso8601_str(chunk_creation_timestamp,
4756 creation_timestamp_buffer,
4757 sizeof(creation_timestamp_buffer));
4758 creation_timestamp_str = !ret ? creation_timestamp_buffer :
4759 "(formatting error)";
4760
4761 DBG("Consumer create trace chunk command: relay_id = %s"
4762 ", session_id = %" PRIu64 ", chunk_id = %" PRIu64
4763 ", chunk_override_name = %s"
4764 ", chunk_creation_timestamp = %s",
4765 relayd_id_str, session_id, chunk_id,
4766 chunk_override_name ? : "(none)",
4767 creation_timestamp_str);
92816cc3
JG
4768
4769 /*
d2956687
JG
4770 * The trace chunk registry, as used by the consumer daemon, implicitly
4771 * owns the trace chunks. This is only needed in the consumer since
4772 * the consumer has no notion of a session beyond session IDs being
4773 * used to identify other objects.
4774 *
4775 * The lttng_trace_chunk_registry_publish() call below provides a
4776 * reference which is not released; it implicitly becomes the session
4777 * daemon's reference to the chunk in the consumer daemon.
4778 *
4779 * The lifetime of trace chunks in the consumer daemon is managed by
4780 * the session daemon through the LTTNG_CONSUMER_CREATE_TRACE_CHUNK
4781 * and LTTNG_CONSUMER_DESTROY_TRACE_CHUNK commands.
92816cc3 4782 */
d2956687 4783 created_chunk = lttng_trace_chunk_create(chunk_id,
a7ceb342 4784 chunk_creation_timestamp, NULL);
d2956687
JG
4785 if (!created_chunk) {
4786 ERR("Failed to create trace chunk");
4787 ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED;
7ea24db3 4788 goto error;
d2956687 4789 }
92816cc3 4790
d2956687
JG
4791 if (chunk_override_name) {
4792 chunk_status = lttng_trace_chunk_override_name(created_chunk,
4793 chunk_override_name);
4794 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
4795 ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED;
7ea24db3 4796 goto error;
92816cc3
JG
4797 }
4798 }
4799
d2956687
JG
4800 if (chunk_directory_handle) {
4801 chunk_status = lttng_trace_chunk_set_credentials(created_chunk,
4802 credentials);
4803 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
4804 ERR("Failed to set trace chunk credentials");
4805 ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED;
7ea24db3 4806 goto error;
d2956687
JG
4807 }
4808 /*
4809 * The consumer daemon has no ownership of the chunk output
4810 * directory.
4811 */
4812 chunk_status = lttng_trace_chunk_set_as_user(created_chunk,
4813 chunk_directory_handle);
cbf53d23 4814 chunk_directory_handle = NULL;
d2956687
JG
4815 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
4816 ERR("Failed to set trace chunk's directory handle");
4817 ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED;
7ea24db3 4818 goto error;
92816cc3
JG
4819 }
4820 }
4821
d2956687 4822 published_chunk = lttng_trace_chunk_registry_publish_chunk(
fa29bfbf 4823 the_consumer_data.chunk_registry, session_id,
d2956687
JG
4824 created_chunk);
4825 lttng_trace_chunk_put(created_chunk);
4826 created_chunk = NULL;
4827 if (!published_chunk) {
4828 ERR("Failed to publish trace chunk");
4829 ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED;
7ea24db3 4830 goto error;
d88744a4
JD
4831 }
4832
d2956687 4833 rcu_read_lock();
fa29bfbf
SM
4834 cds_lfht_for_each_entry_duplicate(
4835 the_consumer_data.channels_by_session_id_ht->ht,
4836 the_consumer_data.channels_by_session_id_ht->hash_fct(
d2956687 4837 &session_id, lttng_ht_seed),
fa29bfbf 4838 the_consumer_data.channels_by_session_id_ht->match_fct,
d2956687
JG
4839 &session_id, &iter.iter, channel,
4840 channels_by_session_id_ht_node.node) {
4841 ret = lttng_consumer_channel_set_trace_chunk(channel,
4842 published_chunk);
4843 if (ret) {
4844 /*
4845 * Roll-back the creation of this chunk.
4846 *
4847 * This is important since the session daemon will
4848 * assume that the creation of this chunk failed and
4849 * will never ask for it to be closed, resulting
4850 * in a leak and an inconsistent state for some
4851 * channels.
4852 */
4853 enum lttcomm_return_code close_ret;
ecd1a12f 4854 char path[LTTNG_PATH_MAX];
d2956687
JG
4855
4856 DBG("Failed to set new trace chunk on existing channels, rolling back");
4857 close_ret = lttng_consumer_close_trace_chunk(relayd_id,
4858 session_id, chunk_id,
ecd1a12f
MD
4859 chunk_creation_timestamp, NULL,
4860 path);
d2956687
JG
4861 if (close_ret != LTTCOMM_CONSUMERD_SUCCESS) {
4862 ERR("Failed to roll-back the creation of new chunk: session_id = %" PRIu64 ", chunk_id = %" PRIu64,
4863 session_id, chunk_id);
4864 }
a1ae2ea5 4865
d2956687
JG
4866 ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED;
4867 break;
4868 }
a1ae2ea5
JD
4869 }
4870
e5add6d0
JG
4871 if (relayd_id) {
4872 struct consumer_relayd_sock_pair *relayd;
4873
4874 relayd = consumer_find_relayd(*relayd_id);
4875 if (relayd) {
4876 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
4877 ret = relayd_create_trace_chunk(
4878 &relayd->control_sock, published_chunk);
4879 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
4880 } else {
4881 ERR("Failed to find relay daemon socket: relayd_id = %" PRIu64, *relayd_id);
4882 }
4883
4884 if (!relayd || ret) {
4885 enum lttcomm_return_code close_ret;
ecd1a12f 4886 char path[LTTNG_PATH_MAX];
e5add6d0
JG
4887
4888 close_ret = lttng_consumer_close_trace_chunk(relayd_id,
4889 session_id,
4890 chunk_id,
bbc4768c 4891 chunk_creation_timestamp,
ecd1a12f 4892 NULL, path);
e5add6d0
JG
4893 if (close_ret != LTTCOMM_CONSUMERD_SUCCESS) {
4894 ERR("Failed to roll-back the creation of new chunk: session_id = %" PRIu64 ", chunk_id = %" PRIu64,
4895 session_id,
4896 chunk_id);
4897 }
4898
4899 ret_code = LTTCOMM_CONSUMERD_CREATE_TRACE_CHUNK_FAILED;
7ea24db3 4900 goto error_unlock;
e5add6d0
JG
4901 }
4902 }
7ea24db3 4903error_unlock:
e5add6d0 4904 rcu_read_unlock();
7ea24db3 4905error:
d2956687
JG
4906 /* Release the reference returned by the "publish" operation. */
4907 lttng_trace_chunk_put(published_chunk);
9bb5f1f8 4908 lttng_trace_chunk_put(created_chunk);
d2956687 4909 return ret_code;
a1ae2ea5
JD
4910}
4911
d2956687
JG
4912enum lttcomm_return_code lttng_consumer_close_trace_chunk(
4913 const uint64_t *relayd_id, uint64_t session_id,
bbc4768c 4914 uint64_t chunk_id, time_t chunk_close_timestamp,
ecd1a12f
MD
4915 const enum lttng_trace_chunk_command_type *close_command,
4916 char *path)
a1ae2ea5 4917{
d2956687
JG
4918 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
4919 struct lttng_trace_chunk *chunk;
4920 char relayd_id_buffer[MAX_INT_DEC_LEN(*relayd_id)];
4921 const char *relayd_id_str = "(none)";
bbc4768c 4922 const char *close_command_name = "none";
d2956687
JG
4923 struct lttng_ht_iter iter;
4924 struct lttng_consumer_channel *channel;
4925 enum lttng_trace_chunk_status chunk_status;
a1ae2ea5 4926
d2956687
JG
4927 if (relayd_id) {
4928 int ret;
4929
4930 /* Only used for logging purposes. */
4931 ret = snprintf(relayd_id_buffer, sizeof(relayd_id_buffer),
4932 "%" PRIu64, *relayd_id);
4933 if (ret > 0 && ret < sizeof(relayd_id_buffer)) {
4934 relayd_id_str = relayd_id_buffer;
4935 } else {
4936 relayd_id_str = "(formatting error)";
4937 }
bbc4768c
JG
4938 }
4939 if (close_command) {
4940 close_command_name = lttng_trace_chunk_command_type_get_name(
4941 *close_command);
4942 }
d2956687
JG
4943
4944 DBG("Consumer close trace chunk command: relayd_id = %s"
bbc4768c
JG
4945 ", session_id = %" PRIu64 ", chunk_id = %" PRIu64
4946 ", close command = %s",
4947 relayd_id_str, session_id, chunk_id,
4948 close_command_name);
4949
d2956687 4950 chunk = lttng_trace_chunk_registry_find_chunk(
fa29bfbf 4951 the_consumer_data.chunk_registry, session_id, chunk_id);
bbc4768c 4952 if (!chunk) {
d2956687
JG
4953 ERR("Failed to find chunk: session_id = %" PRIu64
4954 ", chunk_id = %" PRIu64,
4955 session_id, chunk_id);
4956 ret_code = LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK;
a1ae2ea5
JD
4957 goto end;
4958 }
4959
d2956687
JG
4960 chunk_status = lttng_trace_chunk_set_close_timestamp(chunk,
4961 chunk_close_timestamp);
4962 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
4963 ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED;
4964 goto end;
45f1d9a1 4965 }
bbc4768c
JG
4966
4967 if (close_command) {
4968 chunk_status = lttng_trace_chunk_set_close_command(
4969 chunk, *close_command);
4970 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
4971 ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED;
4972 goto end;
4973 }
4974 }
a1ae2ea5 4975
d2956687
JG
4976 /*
4977 * chunk is now invalid to access as we no longer hold a reference to
4978 * it; it is only kept around to compare it (by address) to the
4979 * current chunk found in the session's channels.
4980 */
4981 rcu_read_lock();
fa29bfbf 4982 cds_lfht_for_each_entry(the_consumer_data.channel_ht->ht, &iter.iter,
d2956687
JG
4983 channel, node.node) {
4984 int ret;
a1ae2ea5 4985
d2956687
JG
4986 /*
4987 * Only change the channel's chunk to NULL if it still
4988 * references the chunk being closed. The channel may
4989 * reference a newer channel in the case of a session
4990 * rotation. When a session rotation occurs, the "next"
4991 * chunk is created before the "current" chunk is closed.
4992 */
4993 if (channel->trace_chunk != chunk) {
4994 continue;
4995 }
4996 ret = lttng_consumer_channel_set_trace_chunk(channel, NULL);
4997 if (ret) {
4998 /*
4999 * Attempt to close the chunk on as many channels as
5000 * possible.
5001 */
5002 ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED;
5003 }
a1ae2ea5 5004 }
bbc4768c
JG
5005
5006 if (relayd_id) {
5007 int ret;
5008 struct consumer_relayd_sock_pair *relayd;
5009
5010 relayd = consumer_find_relayd(*relayd_id);
5011 if (relayd) {
5012 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
5013 ret = relayd_close_trace_chunk(
ecd1a12f
MD
5014 &relayd->control_sock, chunk,
5015 path);
bbc4768c
JG
5016 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
5017 } else {
5018 ERR("Failed to find relay daemon socket: relayd_id = %" PRIu64,
5019 *relayd_id);
5020 }
5021
5022 if (!relayd || ret) {
5023 ret_code = LTTCOMM_CONSUMERD_CLOSE_TRACE_CHUNK_FAILED;
5024 goto error_unlock;
5025 }
5026 }
5027error_unlock:
d2956687
JG
5028 rcu_read_unlock();
5029end:
bbc4768c
JG
5030 /*
5031 * Release the reference returned by the "find" operation and
5032 * the session daemon's implicit reference to the chunk.
5033 */
5034 lttng_trace_chunk_put(chunk);
5035 lttng_trace_chunk_put(chunk);
5036
d2956687 5037 return ret_code;
a1ae2ea5 5038}
3654ed19 5039
d2956687
JG
5040enum lttcomm_return_code lttng_consumer_trace_chunk_exists(
5041 const uint64_t *relayd_id, uint64_t session_id,
5042 uint64_t chunk_id)
3654ed19 5043{
c35f9726 5044 int ret;
d2956687 5045 enum lttcomm_return_code ret_code;
d2956687
JG
5046 char relayd_id_buffer[MAX_INT_DEC_LEN(*relayd_id)];
5047 const char *relayd_id_str = "(none)";
c35f9726
JG
5048 const bool is_local_trace = !relayd_id;
5049 struct consumer_relayd_sock_pair *relayd = NULL;
6b584c2e 5050 bool chunk_exists_local, chunk_exists_remote;
d2956687
JG
5051
5052 if (relayd_id) {
d2956687
JG
5053 /* Only used for logging purposes. */
5054 ret = snprintf(relayd_id_buffer, sizeof(relayd_id_buffer),
5055 "%" PRIu64, *relayd_id);
5056 if (ret > 0 && ret < sizeof(relayd_id_buffer)) {
5057 relayd_id_str = relayd_id_buffer;
5058 } else {
5059 relayd_id_str = "(formatting error)";
5060 }
d01ef216 5061 }
d2956687
JG
5062
5063 DBG("Consumer trace chunk exists command: relayd_id = %s"
d2956687 5064 ", chunk_id = %" PRIu64, relayd_id_str,
c35f9726 5065 chunk_id);
6b584c2e 5066 ret = lttng_trace_chunk_registry_chunk_exists(
fa29bfbf
SM
5067 the_consumer_data.chunk_registry, session_id, chunk_id,
5068 &chunk_exists_local);
6b584c2e
JG
5069 if (ret) {
5070 /* Internal error. */
5071 ERR("Failed to query the existence of a trace chunk");
5072 ret_code = LTTCOMM_CONSUMERD_FATAL;
13e3b280 5073 goto end;
6b584c2e
JG
5074 }
5075 DBG("Trace chunk %s locally",
5076 chunk_exists_local ? "exists" : "does not exist");
5077 if (chunk_exists_local) {
c35f9726 5078 ret_code = LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_LOCAL;
c35f9726
JG
5079 goto end;
5080 } else if (is_local_trace) {
5081 ret_code = LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK;
5082 goto end;
5083 }
5084
5085 rcu_read_lock();
5086 relayd = consumer_find_relayd(*relayd_id);
5087 if (!relayd) {
5088 ERR("Failed to find relayd %" PRIu64, *relayd_id);
5089 ret_code = LTTCOMM_CONSUMERD_INVALID_PARAMETERS;
5090 goto end_rcu_unlock;
5091 }
5092 DBG("Looking up existence of trace chunk on relay daemon");
5093 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
5094 ret = relayd_trace_chunk_exists(&relayd->control_sock, chunk_id,
5095 &chunk_exists_remote);
5096 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
5097 if (ret < 0) {
5098 ERR("Failed to look-up the existence of trace chunk on relay daemon");
5099 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
5100 goto end_rcu_unlock;
5101 }
5102
5103 ret_code = chunk_exists_remote ?
5104 LTTCOMM_CONSUMERD_TRACE_CHUNK_EXISTS_REMOTE :
d2956687 5105 LTTCOMM_CONSUMERD_UNKNOWN_TRACE_CHUNK;
c35f9726
JG
5106 DBG("Trace chunk %s on relay daemon",
5107 chunk_exists_remote ? "exists" : "does not exist");
d2956687 5108
c35f9726
JG
5109end_rcu_unlock:
5110 rcu_read_unlock();
5111end:
d2956687 5112 return ret_code;
3654ed19 5113}
5f3aff8b
MD
5114
5115static
5116int consumer_clear_monitored_channel(struct lttng_consumer_channel *channel)
5117{
5118 struct lttng_ht *ht;
5119 struct lttng_consumer_stream *stream;
5120 struct lttng_ht_iter iter;
5121 int ret;
5122
fa29bfbf 5123 ht = the_consumer_data.stream_per_chan_id_ht;
5f3aff8b
MD
5124
5125 rcu_read_lock();
5126 cds_lfht_for_each_entry_duplicate(ht->ht,
5127 ht->hash_fct(&channel->key, lttng_ht_seed),
5128 ht->match_fct, &channel->key,
5129 &iter.iter, stream, node_channel_id.node) {
5130 /*
5131 * Protect against teardown with mutex.
5132 */
5133 pthread_mutex_lock(&stream->lock);
5134 if (cds_lfht_is_node_deleted(&stream->node.node)) {
5135 goto next;
5136 }
5137 ret = consumer_clear_stream(stream);
5138 if (ret) {
5139 goto error_unlock;
5140 }
5141 next:
5142 pthread_mutex_unlock(&stream->lock);
5143 }
5144 rcu_read_unlock();
5145 return LTTCOMM_CONSUMERD_SUCCESS;
5146
5147error_unlock:
5148 pthread_mutex_unlock(&stream->lock);
5149 rcu_read_unlock();
5150 return ret;
5151}
5152
5153int lttng_consumer_clear_channel(struct lttng_consumer_channel *channel)
5154{
5155 int ret;
5156
5157 DBG("Consumer clear channel %" PRIu64, channel->key);
5158
5159 if (channel->type == CONSUMER_CHANNEL_TYPE_METADATA) {
5160 /*
5161 * Nothing to do for the metadata channel/stream.
5162 * Snapshot mechanism already take care of the metadata
5163 * handling/generation, and monitored channels only need to
5164 * have their data stream cleared..
5165 */
5166 ret = LTTCOMM_CONSUMERD_SUCCESS;
5167 goto end;
5168 }
5169
5170 if (!channel->monitor) {
5171 ret = consumer_clear_unmonitored_channel(channel);
5172 } else {
5173 ret = consumer_clear_monitored_channel(channel);
5174 }
5175end:
5176 return ret;
5177}
04ed9e10
JG
5178
5179enum lttcomm_return_code lttng_consumer_open_channel_packets(
5180 struct lttng_consumer_channel *channel)
5181{
5182 struct lttng_consumer_stream *stream;
5183 enum lttcomm_return_code ret = LTTCOMM_CONSUMERD_SUCCESS;
5184
5185 if (channel->metadata_stream) {
5186 ERR("Open channel packets command attempted on a metadata channel");
5187 ret = LTTCOMM_CONSUMERD_INVALID_PARAMETERS;
5188 goto end;
5189 }
5190
5191 rcu_read_lock();
5192 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
503fefca 5193 enum consumer_stream_open_packet_status status;
04ed9e10
JG
5194
5195 pthread_mutex_lock(&stream->lock);
5196 if (cds_lfht_is_node_deleted(&stream->node.node)) {
5197 goto next;
5198 }
5199
503fefca 5200 status = consumer_stream_open_packet(stream);
04ed9e10 5201 switch (status) {
503fefca 5202 case CONSUMER_STREAM_OPEN_PACKET_STATUS_OPENED:
04ed9e10
JG
5203 DBG("Opened a packet in \"open channel packets\" command: stream id = %" PRIu64
5204 ", channel name = %s, session id = %" PRIu64,
5205 stream->key, stream->chan->name,
5206 stream->chan->session_id);
5207 stream->opened_packet_in_current_trace_chunk = true;
5208 break;
503fefca 5209 case CONSUMER_STREAM_OPEN_PACKET_STATUS_NO_SPACE:
04ed9e10
JG
5210 DBG("No space left to open a packet in \"open channel packets\" command: stream id = %" PRIu64
5211 ", channel name = %s, session id = %" PRIu64,
5212 stream->key, stream->chan->name,
5213 stream->chan->session_id);
5214 break;
503fefca 5215 case CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR:
04ed9e10
JG
5216 /*
5217 * Only unexpected internal errors can lead to this
5218 * failing. Report an unknown error.
5219 */
5220 ERR("Failed to flush empty buffer in \"open channel packets\" command: stream id = %" PRIu64
5221 ", channel id = %" PRIu64
5222 ", channel name = %s"
5223 ", session id = %" PRIu64,
5224 stream->key, channel->key,
5225 channel->name, channel->session_id);
5226 ret = LTTCOMM_CONSUMERD_UNKNOWN_ERROR;
5227 goto error_unlock;
5228 default:
5229 abort();
5230 }
5231
5232 next:
5233 pthread_mutex_unlock(&stream->lock);
5234 }
5235
5236end_rcu_unlock:
5237 rcu_read_unlock();
5238end:
5239 return ret;
5240
5241error_unlock:
5242 pthread_mutex_unlock(&stream->lock);
5243 goto end_rcu_unlock;
5244}
This page took 0.41302 seconds and 4 git commands to generate.