Store the relay protocol version in the consumer_output
[lttng-tools.git] / src / common / consumer / consumer.c
CommitLineData
3bd1e081
MD
1/*
2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
00e2e675 4 * 2012 - David Goulet <dgoulet@efficios.com>
3bd1e081 5 *
d14d33bf
AM
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
3bd1e081 9 *
d14d33bf
AM
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
3bd1e081 14 *
d14d33bf
AM
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
3bd1e081
MD
18 */
19
6c1c0768 20#define _LGPL_SOURCE
3bd1e081 21#include <assert.h>
3bd1e081
MD
22#include <poll.h>
23#include <pthread.h>
24#include <stdlib.h>
25#include <string.h>
26#include <sys/mman.h>
27#include <sys/socket.h>
28#include <sys/types.h>
29#include <unistd.h>
77c7c900 30#include <inttypes.h>
331744e3 31#include <signal.h>
3bd1e081 32
51a9e1c7 33#include <bin/lttng-consumerd/health-consumerd.h>
990570ed 34#include <common/common.h>
fb3a43a9
DG
35#include <common/utils.h>
36#include <common/compat/poll.h>
f263b7fd 37#include <common/compat/endian.h>
309167d2 38#include <common/index/index.h>
10a8a223 39#include <common/kernel-ctl/kernel-ctl.h>
00e2e675 40#include <common/sessiond-comm/relayd.h>
10a8a223
DG
41#include <common/sessiond-comm/sessiond-comm.h>
42#include <common/kernel-consumer/kernel-consumer.h>
00e2e675 43#include <common/relayd/relayd.h>
10a8a223 44#include <common/ust-consumer/ust-consumer.h>
c8fea79c
JR
45#include <common/consumer/consumer-timer.h>
46#include <common/consumer/consumer.h>
47#include <common/consumer/consumer-stream.h>
48#include <common/consumer/consumer-testpoint.h>
49#include <common/align.h>
3bd1e081
MD
50
51struct lttng_consumer_global_data consumer_data = {
3bd1e081
MD
52 .stream_count = 0,
53 .need_update = 1,
54 .type = LTTNG_CONSUMER_UNKNOWN,
55};
56
d8ef542d
MD
57enum consumer_channel_action {
58 CONSUMER_CHANNEL_ADD,
a0cbdd2e 59 CONSUMER_CHANNEL_DEL,
d8ef542d
MD
60 CONSUMER_CHANNEL_QUIT,
61};
62
63struct consumer_channel_msg {
64 enum consumer_channel_action action;
a0cbdd2e
MD
65 struct lttng_consumer_channel *chan; /* add */
66 uint64_t key; /* del */
d8ef542d
MD
67};
68
3bd1e081
MD
69/*
70 * Flag to inform the polling thread to quit when all fd hung up. Updated by
71 * the consumer_thread_receive_fds when it notices that all fds has hung up.
72 * Also updated by the signal handler (consumer_should_exit()). Read by the
73 * polling threads.
74 */
a98dae5f 75volatile int consumer_quit;
3bd1e081 76
43c34bc3 77/*
43c34bc3
DG
78 * Global hash table containing respectively metadata and data streams. The
79 * stream element in this ht should only be updated by the metadata poll thread
80 * for the metadata and the data poll thread for the data.
81 */
40dc48e0
DG
82static struct lttng_ht *metadata_ht;
83static struct lttng_ht *data_ht;
43c34bc3 84
acdb9057
DG
85/*
86 * Notify a thread lttng pipe to poll back again. This usually means that some
87 * global state has changed so we just send back the thread in a poll wait
88 * call.
89 */
90static void notify_thread_lttng_pipe(struct lttng_pipe *pipe)
91{
92 struct lttng_consumer_stream *null_stream = NULL;
93
94 assert(pipe);
95
96 (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream));
97}
98
5c635c72
MD
99static void notify_health_quit_pipe(int *pipe)
100{
6cd525e8 101 ssize_t ret;
5c635c72 102
6cd525e8
MD
103 ret = lttng_write(pipe[1], "4", 1);
104 if (ret < 1) {
5c635c72
MD
105 PERROR("write consumer health quit");
106 }
107}
108
d8ef542d
MD
109static void notify_channel_pipe(struct lttng_consumer_local_data *ctx,
110 struct lttng_consumer_channel *chan,
a0cbdd2e 111 uint64_t key,
d8ef542d
MD
112 enum consumer_channel_action action)
113{
114 struct consumer_channel_msg msg;
6cd525e8 115 ssize_t ret;
d8ef542d 116
e56251fc
DG
117 memset(&msg, 0, sizeof(msg));
118
d8ef542d
MD
119 msg.action = action;
120 msg.chan = chan;
f21dae48 121 msg.key = key;
6cd525e8
MD
122 ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg));
123 if (ret < sizeof(msg)) {
124 PERROR("notify_channel_pipe write error");
125 }
d8ef542d
MD
126}
127
a0cbdd2e
MD
128void notify_thread_del_channel(struct lttng_consumer_local_data *ctx,
129 uint64_t key)
130{
131 notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL);
132}
133
d8ef542d
MD
134static int read_channel_pipe(struct lttng_consumer_local_data *ctx,
135 struct lttng_consumer_channel **chan,
a0cbdd2e 136 uint64_t *key,
d8ef542d
MD
137 enum consumer_channel_action *action)
138{
139 struct consumer_channel_msg msg;
6cd525e8 140 ssize_t ret;
d8ef542d 141
6cd525e8
MD
142 ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg));
143 if (ret < sizeof(msg)) {
144 ret = -1;
145 goto error;
d8ef542d 146 }
6cd525e8
MD
147 *action = msg.action;
148 *chan = msg.chan;
149 *key = msg.key;
150error:
151 return (int) ret;
d8ef542d
MD
152}
153
212d67a2
DG
154/*
155 * Cleanup the stream list of a channel. Those streams are not yet globally
156 * visible
157 */
158static void clean_channel_stream_list(struct lttng_consumer_channel *channel)
159{
160 struct lttng_consumer_stream *stream, *stmp;
161
162 assert(channel);
163
164 /* Delete streams that might have been left in the stream list. */
165 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
166 send_node) {
167 cds_list_del(&stream->send_node);
168 /*
169 * Once a stream is added to this list, the buffers were created so we
170 * have a guarantee that this call will succeed. Setting the monitor
171 * mode to 0 so we don't lock nor try to delete the stream from the
172 * global hash table.
173 */
174 stream->monitor = 0;
175 consumer_stream_destroy(stream, NULL);
176 }
177}
178
3bd1e081
MD
179/*
180 * Find a stream. The consumer_data.lock must be locked during this
181 * call.
182 */
d88aee68 183static struct lttng_consumer_stream *find_stream(uint64_t key,
8389e4f8 184 struct lttng_ht *ht)
3bd1e081 185{
e4421fec 186 struct lttng_ht_iter iter;
d88aee68 187 struct lttng_ht_node_u64 *node;
e4421fec 188 struct lttng_consumer_stream *stream = NULL;
3bd1e081 189
8389e4f8
DG
190 assert(ht);
191
d88aee68
DG
192 /* -1ULL keys are lookup failures */
193 if (key == (uint64_t) -1ULL) {
7ad0a0cb 194 return NULL;
7a57cf92 195 }
e4421fec 196
6065ceec
DG
197 rcu_read_lock();
198
d88aee68
DG
199 lttng_ht_lookup(ht, &key, &iter);
200 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
201 if (node != NULL) {
202 stream = caa_container_of(node, struct lttng_consumer_stream, node);
3bd1e081 203 }
e4421fec 204
6065ceec
DG
205 rcu_read_unlock();
206
e4421fec 207 return stream;
3bd1e081
MD
208}
209
da009f2c 210static void steal_stream_key(uint64_t key, struct lttng_ht *ht)
7ad0a0cb
MD
211{
212 struct lttng_consumer_stream *stream;
213
04253271 214 rcu_read_lock();
ffe60014 215 stream = find_stream(key, ht);
04253271 216 if (stream) {
da009f2c 217 stream->key = (uint64_t) -1ULL;
04253271
MD
218 /*
219 * We don't want the lookup to match, but we still need
220 * to iterate on this stream when iterating over the hash table. Just
221 * change the node key.
222 */
da009f2c 223 stream->node.key = (uint64_t) -1ULL;
04253271
MD
224 }
225 rcu_read_unlock();
7ad0a0cb
MD
226}
227
d56db448
DG
228/*
229 * Return a channel object for the given key.
230 *
231 * RCU read side lock MUST be acquired before calling this function and
232 * protects the channel ptr.
233 */
d88aee68 234struct lttng_consumer_channel *consumer_find_channel(uint64_t key)
3bd1e081 235{
e4421fec 236 struct lttng_ht_iter iter;
d88aee68 237 struct lttng_ht_node_u64 *node;
e4421fec 238 struct lttng_consumer_channel *channel = NULL;
3bd1e081 239
d88aee68
DG
240 /* -1ULL keys are lookup failures */
241 if (key == (uint64_t) -1ULL) {
7ad0a0cb 242 return NULL;
7a57cf92 243 }
e4421fec 244
d88aee68
DG
245 lttng_ht_lookup(consumer_data.channel_ht, &key, &iter);
246 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
247 if (node != NULL) {
248 channel = caa_container_of(node, struct lttng_consumer_channel, node);
3bd1e081 249 }
e4421fec
DG
250
251 return channel;
3bd1e081
MD
252}
253
b5a6470f
DG
254/*
255 * There is a possibility that the consumer does not have enough time between
256 * the close of the channel on the session daemon and the cleanup in here thus
257 * once we have a channel add with an existing key, we know for sure that this
258 * channel will eventually get cleaned up by all streams being closed.
259 *
260 * This function just nullifies the already existing channel key.
261 */
262static void steal_channel_key(uint64_t key)
263{
264 struct lttng_consumer_channel *channel;
265
266 rcu_read_lock();
267 channel = consumer_find_channel(key);
268 if (channel) {
269 channel->key = (uint64_t) -1ULL;
270 /*
271 * We don't want the lookup to match, but we still need to iterate on
272 * this channel when iterating over the hash table. Just change the
273 * node key.
274 */
275 channel->node.key = (uint64_t) -1ULL;
276 }
277 rcu_read_unlock();
278}
279
ffe60014 280static void free_channel_rcu(struct rcu_head *head)
702b1ea4 281{
d88aee68
DG
282 struct lttng_ht_node_u64 *node =
283 caa_container_of(head, struct lttng_ht_node_u64, head);
ffe60014
DG
284 struct lttng_consumer_channel *channel =
285 caa_container_of(node, struct lttng_consumer_channel, node);
702b1ea4 286
b83e03c4
MD
287 switch (consumer_data.type) {
288 case LTTNG_CONSUMER_KERNEL:
289 break;
290 case LTTNG_CONSUMER32_UST:
291 case LTTNG_CONSUMER64_UST:
292 lttng_ustconsumer_free_channel(channel);
293 break;
294 default:
295 ERR("Unknown consumer_data type");
296 abort();
297 }
ffe60014 298 free(channel);
702b1ea4
MD
299}
300
00e2e675
DG
301/*
302 * RCU protected relayd socket pair free.
303 */
ffe60014 304static void free_relayd_rcu(struct rcu_head *head)
00e2e675 305{
d88aee68
DG
306 struct lttng_ht_node_u64 *node =
307 caa_container_of(head, struct lttng_ht_node_u64, head);
00e2e675
DG
308 struct consumer_relayd_sock_pair *relayd =
309 caa_container_of(node, struct consumer_relayd_sock_pair, node);
310
8994307f
DG
311 /*
312 * Close all sockets. This is done in the call RCU since we don't want the
313 * socket fds to be reassigned thus potentially creating bad state of the
314 * relayd object.
315 *
316 * We do not have to lock the control socket mutex here since at this stage
317 * there is no one referencing to this relayd object.
318 */
319 (void) relayd_close(&relayd->control_sock);
320 (void) relayd_close(&relayd->data_sock);
321
00e2e675
DG
322 free(relayd);
323}
324
325/*
326 * Destroy and free relayd socket pair object.
00e2e675 327 */
51230d70 328void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
329{
330 int ret;
331 struct lttng_ht_iter iter;
332
173af62f
DG
333 if (relayd == NULL) {
334 return;
335 }
336
00e2e675
DG
337 DBG("Consumer destroy and close relayd socket pair");
338
339 iter.iter.node = &relayd->node.node;
340 ret = lttng_ht_del(consumer_data.relayd_ht, &iter);
173af62f 341 if (ret != 0) {
8994307f 342 /* We assume the relayd is being or is destroyed */
173af62f
DG
343 return;
344 }
00e2e675 345
00e2e675 346 /* RCU free() call */
ffe60014
DG
347 call_rcu(&relayd->node.head, free_relayd_rcu);
348}
349
350/*
351 * Remove a channel from the global list protected by a mutex. This function is
352 * also responsible for freeing its data structures.
353 */
354void consumer_del_channel(struct lttng_consumer_channel *channel)
355{
356 int ret;
357 struct lttng_ht_iter iter;
358
d88aee68 359 DBG("Consumer delete channel key %" PRIu64, channel->key);
ffe60014
DG
360
361 pthread_mutex_lock(&consumer_data.lock);
a9838785 362 pthread_mutex_lock(&channel->lock);
ffe60014 363
212d67a2
DG
364 /* Destroy streams that might have been left in the stream list. */
365 clean_channel_stream_list(channel);
51e762e5 366
d3e2ba59
JD
367 if (channel->live_timer_enabled == 1) {
368 consumer_timer_live_stop(channel);
369 }
370
ffe60014
DG
371 switch (consumer_data.type) {
372 case LTTNG_CONSUMER_KERNEL:
373 break;
374 case LTTNG_CONSUMER32_UST:
375 case LTTNG_CONSUMER64_UST:
376 lttng_ustconsumer_del_channel(channel);
377 break;
378 default:
379 ERR("Unknown consumer_data type");
380 assert(0);
381 goto end;
382 }
383
384 rcu_read_lock();
385 iter.iter.node = &channel->node.node;
386 ret = lttng_ht_del(consumer_data.channel_ht, &iter);
387 assert(!ret);
388 rcu_read_unlock();
389
390 call_rcu(&channel->node.head, free_channel_rcu);
391end:
a9838785 392 pthread_mutex_unlock(&channel->lock);
ffe60014 393 pthread_mutex_unlock(&consumer_data.lock);
00e2e675
DG
394}
395
228b5bf7
DG
396/*
397 * Iterate over the relayd hash table and destroy each element. Finally,
398 * destroy the whole hash table.
399 */
400static void cleanup_relayd_ht(void)
401{
402 struct lttng_ht_iter iter;
403 struct consumer_relayd_sock_pair *relayd;
404
405 rcu_read_lock();
406
407 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
408 node.node) {
51230d70 409 consumer_destroy_relayd(relayd);
228b5bf7
DG
410 }
411
228b5bf7 412 rcu_read_unlock();
36b588ed
MD
413
414 lttng_ht_destroy(consumer_data.relayd_ht);
228b5bf7
DG
415}
416
8994307f
DG
417/*
418 * Update the end point status of all streams having the given network sequence
419 * index (relayd index).
420 *
421 * It's atomically set without having the stream mutex locked which is fine
422 * because we handle the write/read race with a pipe wakeup for each thread.
423 */
da009f2c 424static void update_endpoint_status_by_netidx(uint64_t net_seq_idx,
8994307f
DG
425 enum consumer_endpoint_status status)
426{
427 struct lttng_ht_iter iter;
428 struct lttng_consumer_stream *stream;
429
da009f2c 430 DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx);
8994307f
DG
431
432 rcu_read_lock();
433
434 /* Let's begin with metadata */
435 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
436 if (stream->net_seq_idx == net_seq_idx) {
437 uatomic_set(&stream->endpoint_status, status);
438 DBG("Delete flag set to metadata stream %d", stream->wait_fd);
439 }
440 }
441
442 /* Follow up by the data streams */
443 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
444 if (stream->net_seq_idx == net_seq_idx) {
445 uatomic_set(&stream->endpoint_status, status);
446 DBG("Delete flag set to data stream %d", stream->wait_fd);
447 }
448 }
449 rcu_read_unlock();
450}
451
452/*
453 * Cleanup a relayd object by flagging every associated streams for deletion,
454 * destroying the object meaning removing it from the relayd hash table,
455 * closing the sockets and freeing the memory in a RCU call.
456 *
457 * If a local data context is available, notify the threads that the streams'
458 * state have changed.
459 */
460static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd,
461 struct lttng_consumer_local_data *ctx)
462{
da009f2c 463 uint64_t netidx;
8994307f
DG
464
465 assert(relayd);
466
9617607b
DG
467 DBG("Cleaning up relayd sockets");
468
8994307f
DG
469 /* Save the net sequence index before destroying the object */
470 netidx = relayd->net_seq_idx;
471
472 /*
473 * Delete the relayd from the relayd hash table, close the sockets and free
474 * the object in a RCU call.
475 */
51230d70 476 consumer_destroy_relayd(relayd);
8994307f
DG
477
478 /* Set inactive endpoint to all streams */
479 update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE);
480
481 /*
482 * With a local data context, notify the threads that the streams' state
483 * have changed. The write() action on the pipe acts as an "implicit"
484 * memory barrier ordering the updates of the end point status from the
485 * read of this status which happens AFTER receiving this notify.
486 */
487 if (ctx) {
acdb9057 488 notify_thread_lttng_pipe(ctx->consumer_data_pipe);
13886d2d 489 notify_thread_lttng_pipe(ctx->consumer_metadata_pipe);
8994307f
DG
490 }
491}
492
a6ba4fe1
DG
493/*
494 * Flag a relayd socket pair for destruction. Destroy it if the refcount
495 * reaches zero.
496 *
497 * RCU read side lock MUST be aquired before calling this function.
498 */
499void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd)
500{
501 assert(relayd);
502
503 /* Set destroy flag for this object */
504 uatomic_set(&relayd->destroy_flag, 1);
505
506 /* Destroy the relayd if refcount is 0 */
507 if (uatomic_read(&relayd->refcount) == 0) {
51230d70 508 consumer_destroy_relayd(relayd);
a6ba4fe1
DG
509 }
510}
511
3bd1e081 512/*
1d1a276c
DG
513 * Completly destroy stream from every visiable data structure and the given
514 * hash table if one.
515 *
516 * One this call returns, the stream object is not longer usable nor visible.
3bd1e081 517 */
e316aad5
DG
518void consumer_del_stream(struct lttng_consumer_stream *stream,
519 struct lttng_ht *ht)
3bd1e081 520{
1d1a276c 521 consumer_stream_destroy(stream, ht);
3bd1e081
MD
522}
523
5ab66908
MD
524/*
525 * XXX naming of del vs destroy is all mixed up.
526 */
527void consumer_del_stream_for_data(struct lttng_consumer_stream *stream)
528{
529 consumer_stream_destroy(stream, data_ht);
530}
531
532void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream)
533{
534 consumer_stream_destroy(stream, metadata_ht);
535}
536
d88aee68
DG
537struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key,
538 uint64_t stream_key,
3bd1e081 539 enum lttng_consumer_stream_state state,
ffe60014 540 const char *channel_name,
6df2e2c9 541 uid_t uid,
00e2e675 542 gid_t gid,
57a269f2 543 uint64_t relayd_id,
53632229 544 uint64_t session_id,
ffe60014
DG
545 int cpu,
546 int *alloc_ret,
4891ece8
DG
547 enum consumer_channel_type type,
548 unsigned int monitor)
3bd1e081 549{
ffe60014 550 int ret;
3bd1e081 551 struct lttng_consumer_stream *stream;
3bd1e081 552
effcf122 553 stream = zmalloc(sizeof(*stream));
3bd1e081 554 if (stream == NULL) {
7a57cf92 555 PERROR("malloc struct lttng_consumer_stream");
ffe60014 556 ret = -ENOMEM;
7a57cf92 557 goto end;
3bd1e081 558 }
7a57cf92 559
d56db448
DG
560 rcu_read_lock();
561
3bd1e081 562 stream->key = stream_key;
3bd1e081
MD
563 stream->out_fd = -1;
564 stream->out_fd_offset = 0;
e5d1a9b3 565 stream->output_written = 0;
3bd1e081 566 stream->state = state;
6df2e2c9
MD
567 stream->uid = uid;
568 stream->gid = gid;
ffe60014 569 stream->net_seq_idx = relayd_id;
53632229 570 stream->session_id = session_id;
4891ece8 571 stream->monitor = monitor;
774d490c 572 stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE;
309167d2 573 stream->index_fd = -1;
fb83fe64 574 stream->last_sequence_number = -1ULL;
53632229 575 pthread_mutex_init(&stream->lock, NULL);
c585821b 576 pthread_mutex_init(&stream->metadata_timer_lock, NULL);
58b1f425 577
ffe60014
DG
578 /* If channel is the metadata, flag this stream as metadata. */
579 if (type == CONSUMER_CHANNEL_TYPE_METADATA) {
580 stream->metadata_flag = 1;
581 /* Metadata is flat out. */
582 strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name));
94d49140
JD
583 /* Live rendez-vous point. */
584 pthread_cond_init(&stream->metadata_rdv, NULL);
585 pthread_mutex_init(&stream->metadata_rdv_lock, NULL);
58b1f425 586 } else {
ffe60014
DG
587 /* Format stream name to <channel_name>_<cpu_number> */
588 ret = snprintf(stream->name, sizeof(stream->name), "%s_%d",
589 channel_name, cpu);
590 if (ret < 0) {
591 PERROR("snprintf stream name");
592 goto error;
593 }
58b1f425 594 }
c30aaa51 595
ffe60014 596 /* Key is always the wait_fd for streams. */
d88aee68 597 lttng_ht_node_init_u64(&stream->node, stream->key);
ffe60014 598
d8ef542d
MD
599 /* Init node per channel id key */
600 lttng_ht_node_init_u64(&stream->node_channel_id, channel_key);
601
53632229 602 /* Init session id node with the stream session id */
d88aee68 603 lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id);
53632229 604
07b86b52
JD
605 DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64
606 " relayd_id %" PRIu64 ", session_id %" PRIu64,
607 stream->name, stream->key, channel_key,
608 stream->net_seq_idx, stream->session_id);
d56db448
DG
609
610 rcu_read_unlock();
3bd1e081 611 return stream;
c80048c6
MD
612
613error:
d56db448 614 rcu_read_unlock();
c80048c6 615 free(stream);
7a57cf92 616end:
ffe60014
DG
617 if (alloc_ret) {
618 *alloc_ret = ret;
619 }
c80048c6 620 return NULL;
3bd1e081
MD
621}
622
623/*
624 * Add a stream to the global list protected by a mutex.
625 */
5ab66908 626int consumer_add_data_stream(struct lttng_consumer_stream *stream)
3bd1e081 627{
5ab66908 628 struct lttng_ht *ht = data_ht;
3bd1e081
MD
629 int ret = 0;
630
e316aad5 631 assert(stream);
43c34bc3 632 assert(ht);
c77fc10a 633
d88aee68 634 DBG3("Adding consumer stream %" PRIu64, stream->key);
e316aad5
DG
635
636 pthread_mutex_lock(&consumer_data.lock);
a9838785 637 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 638 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 639 pthread_mutex_lock(&stream->lock);
b0b335c8 640 rcu_read_lock();
e316aad5 641
43c34bc3 642 /* Steal stream identifier to avoid having streams with the same key */
ffe60014 643 steal_stream_key(stream->key, ht);
43c34bc3 644
d88aee68 645 lttng_ht_add_unique_u64(ht, &stream->node);
00e2e675 646
d8ef542d
MD
647 lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht,
648 &stream->node_channel_id);
649
ca22feea
DG
650 /*
651 * Add stream to the stream_list_ht of the consumer data. No need to steal
652 * the key since the HT does not use it and we allow to add redundant keys
653 * into this table.
654 */
d88aee68 655 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 656
e316aad5 657 /*
ffe60014
DG
658 * When nb_init_stream_left reaches 0, we don't need to trigger any action
659 * in terms of destroying the associated channel, because the action that
e316aad5
DG
660 * causes the count to become 0 also causes a stream to be added. The
661 * channel deletion will thus be triggered by the following removal of this
662 * stream.
663 */
ffe60014 664 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
665 /* Increment refcount before decrementing nb_init_stream_left */
666 cmm_smp_wmb();
ffe60014 667 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
668 }
669
670 /* Update consumer data once the node is inserted. */
3bd1e081
MD
671 consumer_data.stream_count++;
672 consumer_data.need_update = 1;
673
e316aad5 674 rcu_read_unlock();
2e818a6a 675 pthread_mutex_unlock(&stream->lock);
ec6ea7d0 676 pthread_mutex_unlock(&stream->chan->timer_lock);
a9838785 677 pthread_mutex_unlock(&stream->chan->lock);
3bd1e081 678 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 679
3bd1e081
MD
680 return ret;
681}
682
5ab66908
MD
683void consumer_del_data_stream(struct lttng_consumer_stream *stream)
684{
685 consumer_del_stream(stream, data_ht);
686}
687
00e2e675 688/*
3f8e211f
DG
689 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
690 * be acquired before calling this.
00e2e675 691 */
d09e1200 692static int add_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
693{
694 int ret = 0;
d88aee68 695 struct lttng_ht_node_u64 *node;
00e2e675
DG
696 struct lttng_ht_iter iter;
697
ffe60014 698 assert(relayd);
00e2e675 699
00e2e675 700 lttng_ht_lookup(consumer_data.relayd_ht,
d88aee68
DG
701 &relayd->net_seq_idx, &iter);
702 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675 703 if (node != NULL) {
00e2e675
DG
704 goto end;
705 }
d88aee68 706 lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node);
00e2e675 707
00e2e675
DG
708end:
709 return ret;
710}
711
712/*
713 * Allocate and return a consumer relayd socket.
714 */
027a694f 715static struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair(
da009f2c 716 uint64_t net_seq_idx)
00e2e675
DG
717{
718 struct consumer_relayd_sock_pair *obj = NULL;
719
da009f2c
MD
720 /* net sequence index of -1 is a failure */
721 if (net_seq_idx == (uint64_t) -1ULL) {
00e2e675
DG
722 goto error;
723 }
724
725 obj = zmalloc(sizeof(struct consumer_relayd_sock_pair));
726 if (obj == NULL) {
727 PERROR("zmalloc relayd sock");
728 goto error;
729 }
730
731 obj->net_seq_idx = net_seq_idx;
732 obj->refcount = 0;
173af62f 733 obj->destroy_flag = 0;
f96e4545
MD
734 obj->control_sock.sock.fd = -1;
735 obj->data_sock.sock.fd = -1;
d88aee68 736 lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx);
00e2e675
DG
737 pthread_mutex_init(&obj->ctrl_sock_mutex, NULL);
738
739error:
740 return obj;
741}
742
743/*
744 * Find a relayd socket pair in the global consumer data.
745 *
746 * Return the object if found else NULL.
b0b335c8
MD
747 * RCU read-side lock must be held across this call and while using the
748 * returned object.
00e2e675 749 */
d88aee68 750struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key)
00e2e675
DG
751{
752 struct lttng_ht_iter iter;
d88aee68 753 struct lttng_ht_node_u64 *node;
00e2e675
DG
754 struct consumer_relayd_sock_pair *relayd = NULL;
755
756 /* Negative keys are lookup failures */
d88aee68 757 if (key == (uint64_t) -1ULL) {
00e2e675
DG
758 goto error;
759 }
760
d88aee68 761 lttng_ht_lookup(consumer_data.relayd_ht, &key,
00e2e675 762 &iter);
d88aee68 763 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675
DG
764 if (node != NULL) {
765 relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node);
766 }
767
00e2e675
DG
768error:
769 return relayd;
770}
771
10a50311
JD
772/*
773 * Find a relayd and send the stream
774 *
775 * Returns 0 on success, < 0 on error
776 */
777int consumer_send_relayd_stream(struct lttng_consumer_stream *stream,
778 char *path)
779{
780 int ret = 0;
781 struct consumer_relayd_sock_pair *relayd;
782
783 assert(stream);
784 assert(stream->net_seq_idx != -1ULL);
785 assert(path);
786
787 /* The stream is not metadata. Get relayd reference if exists. */
788 rcu_read_lock();
789 relayd = consumer_find_relayd(stream->net_seq_idx);
790 if (relayd != NULL) {
791 /* Add stream on the relayd */
792 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
793 ret = relayd_add_stream(&relayd->control_sock, stream->name,
794 path, &stream->relayd_stream_id,
795 stream->chan->tracefile_size, stream->chan->tracefile_count);
796 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
797 if (ret < 0) {
798 goto end;
799 }
1c20f0e2 800
10a50311 801 uatomic_inc(&relayd->refcount);
d01178b6 802 stream->sent_to_relayd = 1;
10a50311
JD
803 } else {
804 ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.",
805 stream->key, stream->net_seq_idx);
806 ret = -1;
807 goto end;
808 }
809
810 DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64,
811 stream->name, stream->key, stream->net_seq_idx);
812
813end:
814 rcu_read_unlock();
815 return ret;
816}
817
a4baae1b
JD
818/*
819 * Find a relayd and send the streams sent message
820 *
821 * Returns 0 on success, < 0 on error
822 */
823int consumer_send_relayd_streams_sent(uint64_t net_seq_idx)
824{
825 int ret = 0;
826 struct consumer_relayd_sock_pair *relayd;
827
828 assert(net_seq_idx != -1ULL);
829
830 /* The stream is not metadata. Get relayd reference if exists. */
831 rcu_read_lock();
832 relayd = consumer_find_relayd(net_seq_idx);
833 if (relayd != NULL) {
834 /* Add stream on the relayd */
835 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
836 ret = relayd_streams_sent(&relayd->control_sock);
837 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
838 if (ret < 0) {
839 goto end;
840 }
841 } else {
842 ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.",
843 net_seq_idx);
844 ret = -1;
845 goto end;
846 }
847
848 ret = 0;
849 DBG("All streams sent relayd id %" PRIu64, net_seq_idx);
850
851end:
852 rcu_read_unlock();
853 return ret;
854}
855
10a50311
JD
856/*
857 * Find a relayd and close the stream
858 */
859void close_relayd_stream(struct lttng_consumer_stream *stream)
860{
861 struct consumer_relayd_sock_pair *relayd;
862
863 /* The stream is not metadata. Get relayd reference if exists. */
864 rcu_read_lock();
865 relayd = consumer_find_relayd(stream->net_seq_idx);
866 if (relayd) {
867 consumer_stream_relayd_close(stream, relayd);
868 }
869 rcu_read_unlock();
870}
871
00e2e675
DG
872/*
873 * Handle stream for relayd transmission if the stream applies for network
874 * streaming where the net sequence index is set.
875 *
876 * Return destination file descriptor or negative value on error.
877 */
6197aea7 878static int write_relayd_stream_header(struct lttng_consumer_stream *stream,
1d4dfdef
DG
879 size_t data_size, unsigned long padding,
880 struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
881{
882 int outfd = -1, ret;
00e2e675
DG
883 struct lttcomm_relayd_data_hdr data_hdr;
884
885 /* Safety net */
886 assert(stream);
6197aea7 887 assert(relayd);
00e2e675
DG
888
889 /* Reset data header */
890 memset(&data_hdr, 0, sizeof(data_hdr));
891
00e2e675
DG
892 if (stream->metadata_flag) {
893 /* Caller MUST acquire the relayd control socket lock */
894 ret = relayd_send_metadata(&relayd->control_sock, data_size);
895 if (ret < 0) {
896 goto error;
897 }
898
899 /* Metadata are always sent on the control socket. */
6151a90f 900 outfd = relayd->control_sock.sock.fd;
00e2e675
DG
901 } else {
902 /* Set header with stream information */
903 data_hdr.stream_id = htobe64(stream->relayd_stream_id);
904 data_hdr.data_size = htobe32(data_size);
1d4dfdef 905 data_hdr.padding_size = htobe32(padding);
39df6d9f
DG
906 /*
907 * Note that net_seq_num below is assigned with the *current* value of
908 * next_net_seq_num and only after that the next_net_seq_num will be
909 * increment. This is why when issuing a command on the relayd using
910 * this next value, 1 should always be substracted in order to compare
911 * the last seen sequence number on the relayd side to the last sent.
912 */
3604f373 913 data_hdr.net_seq_num = htobe64(stream->next_net_seq_num);
00e2e675
DG
914 /* Other fields are zeroed previously */
915
916 ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr,
917 sizeof(data_hdr));
918 if (ret < 0) {
919 goto error;
920 }
921
3604f373
DG
922 ++stream->next_net_seq_num;
923
00e2e675 924 /* Set to go on data socket */
6151a90f 925 outfd = relayd->data_sock.sock.fd;
00e2e675
DG
926 }
927
928error:
929 return outfd;
930}
931
3bd1e081 932/*
ffe60014
DG
933 * Allocate and return a new lttng_consumer_channel object using the given key
934 * to initialize the hash table node.
935 *
936 * On error, return NULL.
3bd1e081 937 */
886224ff 938struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key,
ffe60014
DG
939 uint64_t session_id,
940 const char *pathname,
941 const char *name,
942 uid_t uid,
943 gid_t gid,
57a269f2 944 uint64_t relayd_id,
1624d5b7
JD
945 enum lttng_event_output output,
946 uint64_t tracefile_size,
2bba9e53 947 uint64_t tracefile_count,
1950109e 948 uint64_t session_id_per_pid,
ecc48a90 949 unsigned int monitor,
d7ba1388 950 unsigned int live_timer_interval,
3d071855 951 const char *root_shm_path,
d7ba1388 952 const char *shm_path)
3bd1e081
MD
953{
954 struct lttng_consumer_channel *channel;
3bd1e081 955
276b26d1 956 channel = zmalloc(sizeof(*channel));
3bd1e081 957 if (channel == NULL) {
7a57cf92 958 PERROR("malloc struct lttng_consumer_channel");
3bd1e081
MD
959 goto end;
960 }
ffe60014
DG
961
962 channel->key = key;
3bd1e081 963 channel->refcount = 0;
ffe60014 964 channel->session_id = session_id;
1950109e 965 channel->session_id_per_pid = session_id_per_pid;
ffe60014
DG
966 channel->uid = uid;
967 channel->gid = gid;
968 channel->relayd_id = relayd_id;
1624d5b7
JD
969 channel->tracefile_size = tracefile_size;
970 channel->tracefile_count = tracefile_count;
2bba9e53 971 channel->monitor = monitor;
ecc48a90 972 channel->live_timer_interval = live_timer_interval;
a9838785 973 pthread_mutex_init(&channel->lock, NULL);
ec6ea7d0 974 pthread_mutex_init(&channel->timer_lock, NULL);
ffe60014 975
0c759fc9
DG
976 switch (output) {
977 case LTTNG_EVENT_SPLICE:
978 channel->output = CONSUMER_CHANNEL_SPLICE;
979 break;
980 case LTTNG_EVENT_MMAP:
981 channel->output = CONSUMER_CHANNEL_MMAP;
982 break;
983 default:
984 assert(0);
985 free(channel);
986 channel = NULL;
987 goto end;
988 }
989
07b86b52
JD
990 /*
991 * In monitor mode, the streams associated with the channel will be put in
992 * a special list ONLY owned by this channel. So, the refcount is set to 1
993 * here meaning that the channel itself has streams that are referenced.
994 *
995 * On a channel deletion, once the channel is no longer visible, the
996 * refcount is decremented and checked for a zero value to delete it. With
997 * streams in no monitor mode, it will now be safe to destroy the channel.
998 */
999 if (!channel->monitor) {
1000 channel->refcount = 1;
1001 }
1002
ffe60014
DG
1003 strncpy(channel->pathname, pathname, sizeof(channel->pathname));
1004 channel->pathname[sizeof(channel->pathname) - 1] = '\0';
1005
1006 strncpy(channel->name, name, sizeof(channel->name));
1007 channel->name[sizeof(channel->name) - 1] = '\0';
1008
3d071855
MD
1009 if (root_shm_path) {
1010 strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path));
1011 channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0';
1012 }
d7ba1388
MD
1013 if (shm_path) {
1014 strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path));
1015 channel->shm_path[sizeof(channel->shm_path) - 1] = '\0';
1016 }
1017
d88aee68 1018 lttng_ht_node_init_u64(&channel->node, channel->key);
d8ef542d
MD
1019
1020 channel->wait_fd = -1;
1021
ffe60014
DG
1022 CDS_INIT_LIST_HEAD(&channel->streams.head);
1023
d88aee68 1024 DBG("Allocated channel (key %" PRIu64 ")", channel->key)
3bd1e081 1025
3bd1e081
MD
1026end:
1027 return channel;
1028}
1029
1030/*
1031 * Add a channel to the global list protected by a mutex.
821fffb2 1032 *
b5a6470f 1033 * Always return 0 indicating success.
3bd1e081 1034 */
d8ef542d
MD
1035int consumer_add_channel(struct lttng_consumer_channel *channel,
1036 struct lttng_consumer_local_data *ctx)
3bd1e081 1037{
3bd1e081 1038 pthread_mutex_lock(&consumer_data.lock);
a9838785 1039 pthread_mutex_lock(&channel->lock);
ec6ea7d0 1040 pthread_mutex_lock(&channel->timer_lock);
c77fc10a 1041
b5a6470f
DG
1042 /*
1043 * This gives us a guarantee that the channel we are about to add to the
1044 * channel hash table will be unique. See this function comment on the why
1045 * we need to steel the channel key at this stage.
1046 */
1047 steal_channel_key(channel->key);
c77fc10a 1048
b5a6470f 1049 rcu_read_lock();
d88aee68 1050 lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node);
6065ceec 1051 rcu_read_unlock();
b5a6470f 1052
ec6ea7d0 1053 pthread_mutex_unlock(&channel->timer_lock);
a9838785 1054 pthread_mutex_unlock(&channel->lock);
3bd1e081 1055 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 1056
b5a6470f 1057 if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) {
a0cbdd2e 1058 notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD);
d8ef542d 1059 }
b5a6470f
DG
1060
1061 return 0;
3bd1e081
MD
1062}
1063
1064/*
1065 * Allocate the pollfd structure and the local view of the out fds to avoid
1066 * doing a lookup in the linked list and concurrency issues when writing is
1067 * needed. Called with consumer_data.lock held.
1068 *
1069 * Returns the number of fds in the structures.
1070 */
ffe60014
DG
1071static int update_poll_array(struct lttng_consumer_local_data *ctx,
1072 struct pollfd **pollfd, struct lttng_consumer_stream **local_stream,
1073 struct lttng_ht *ht)
3bd1e081 1074{
3bd1e081 1075 int i = 0;
e4421fec
DG
1076 struct lttng_ht_iter iter;
1077 struct lttng_consumer_stream *stream;
3bd1e081 1078
ffe60014
DG
1079 assert(ctx);
1080 assert(ht);
1081 assert(pollfd);
1082 assert(local_stream);
1083
3bd1e081 1084 DBG("Updating poll fd array");
481d6c57 1085 rcu_read_lock();
43c34bc3 1086 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
8994307f
DG
1087 /*
1088 * Only active streams with an active end point can be added to the
1089 * poll set and local stream storage of the thread.
1090 *
1091 * There is a potential race here for endpoint_status to be updated
1092 * just after the check. However, this is OK since the stream(s) will
1093 * be deleted once the thread is notified that the end point state has
1094 * changed where this function will be called back again.
1095 */
1096 if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM ||
79d4ffb7 1097 stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) {
3bd1e081
MD
1098 continue;
1099 }
7972aab2
DG
1100 /*
1101 * This clobbers way too much the debug output. Uncomment that if you
1102 * need it for debugging purposes.
1103 *
1104 * DBG("Active FD %d", stream->wait_fd);
1105 */
e4421fec 1106 (*pollfd)[i].fd = stream->wait_fd;
3bd1e081 1107 (*pollfd)[i].events = POLLIN | POLLPRI;
e4421fec 1108 local_stream[i] = stream;
3bd1e081
MD
1109 i++;
1110 }
481d6c57 1111 rcu_read_unlock();
3bd1e081
MD
1112
1113 /*
50f8ae69 1114 * Insert the consumer_data_pipe at the end of the array and don't
3bd1e081
MD
1115 * increment i so nb_fd is the number of real FD.
1116 */
acdb9057 1117 (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe);
509bb1cf 1118 (*pollfd)[i].events = POLLIN | POLLPRI;
02b3d176
DG
1119
1120 (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe);
1121 (*pollfd)[i + 1].events = POLLIN | POLLPRI;
3bd1e081
MD
1122 return i;
1123}
1124
1125/*
84382d49
MD
1126 * Poll on the should_quit pipe and the command socket return -1 on
1127 * error, 1 if should exit, 0 if data is available on the command socket
3bd1e081
MD
1128 */
1129int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll)
1130{
1131 int num_rdy;
1132
88f2b785 1133restart:
3bd1e081
MD
1134 num_rdy = poll(consumer_sockpoll, 2, -1);
1135 if (num_rdy == -1) {
88f2b785
MD
1136 /*
1137 * Restart interrupted system call.
1138 */
1139 if (errno == EINTR) {
1140 goto restart;
1141 }
7a57cf92 1142 PERROR("Poll error");
84382d49 1143 return -1;
3bd1e081 1144 }
509bb1cf 1145 if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) {
3bd1e081 1146 DBG("consumer_should_quit wake up");
84382d49 1147 return 1;
3bd1e081
MD
1148 }
1149 return 0;
3bd1e081
MD
1150}
1151
1152/*
1153 * Set the error socket.
1154 */
ffe60014
DG
1155void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx,
1156 int sock)
3bd1e081
MD
1157{
1158 ctx->consumer_error_socket = sock;
1159}
1160
1161/*
1162 * Set the command socket path.
1163 */
3bd1e081
MD
1164void lttng_consumer_set_command_sock_path(
1165 struct lttng_consumer_local_data *ctx, char *sock)
1166{
1167 ctx->consumer_command_sock_path = sock;
1168}
1169
1170/*
1171 * Send return code to the session daemon.
1172 * If the socket is not defined, we return 0, it is not a fatal error
1173 */
ffe60014 1174int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd)
3bd1e081
MD
1175{
1176 if (ctx->consumer_error_socket > 0) {
1177 return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd,
1178 sizeof(enum lttcomm_sessiond_command));
1179 }
1180
1181 return 0;
1182}
1183
1184/*
228b5bf7
DG
1185 * Close all the tracefiles and stream fds and MUST be called when all
1186 * instances are destroyed i.e. when all threads were joined and are ended.
3bd1e081
MD
1187 */
1188void lttng_consumer_cleanup(void)
1189{
e4421fec 1190 struct lttng_ht_iter iter;
ffe60014 1191 struct lttng_consumer_channel *channel;
6065ceec
DG
1192
1193 rcu_read_lock();
3bd1e081 1194
ffe60014
DG
1195 cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel,
1196 node.node) {
702b1ea4 1197 consumer_del_channel(channel);
3bd1e081 1198 }
6065ceec
DG
1199
1200 rcu_read_unlock();
d6ce1df2 1201
d6ce1df2 1202 lttng_ht_destroy(consumer_data.channel_ht);
228b5bf7
DG
1203
1204 cleanup_relayd_ht();
1205
d8ef542d
MD
1206 lttng_ht_destroy(consumer_data.stream_per_chan_id_ht);
1207
228b5bf7
DG
1208 /*
1209 * This HT contains streams that are freed by either the metadata thread or
1210 * the data thread so we do *nothing* on the hash table and simply destroy
1211 * it.
1212 */
1213 lttng_ht_destroy(consumer_data.stream_list_ht);
3bd1e081
MD
1214}
1215
1216/*
1217 * Called from signal handler.
1218 */
1219void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx)
1220{
6cd525e8
MD
1221 ssize_t ret;
1222
3bd1e081 1223 consumer_quit = 1;
6cd525e8
MD
1224 ret = lttng_write(ctx->consumer_should_quit[1], "4", 1);
1225 if (ret < 1) {
7a57cf92 1226 PERROR("write consumer quit");
3bd1e081 1227 }
ab1027f4
DG
1228
1229 DBG("Consumer flag that it should quit");
3bd1e081
MD
1230}
1231
00e2e675
DG
1232void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream,
1233 off_t orig_offset)
3bd1e081
MD
1234{
1235 int outfd = stream->out_fd;
1236
1237 /*
1238 * This does a blocking write-and-wait on any page that belongs to the
1239 * subbuffer prior to the one we just wrote.
1240 * Don't care about error values, as these are just hints and ways to
1241 * limit the amount of page cache used.
1242 */
ffe60014 1243 if (orig_offset < stream->max_sb_size) {
3bd1e081
MD
1244 return;
1245 }
ffe60014
DG
1246 lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size,
1247 stream->max_sb_size,
3bd1e081
MD
1248 SYNC_FILE_RANGE_WAIT_BEFORE
1249 | SYNC_FILE_RANGE_WRITE
1250 | SYNC_FILE_RANGE_WAIT_AFTER);
1251 /*
1252 * Give hints to the kernel about how we access the file:
1253 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1254 * we write it.
1255 *
1256 * We need to call fadvise again after the file grows because the
1257 * kernel does not seem to apply fadvise to non-existing parts of the
1258 * file.
1259 *
1260 * Call fadvise _after_ having waited for the page writeback to
1261 * complete because the dirty page writeback semantic is not well
1262 * defined. So it can be expected to lead to lower throughput in
1263 * streaming.
1264 */
ffe60014
DG
1265 posix_fadvise(outfd, orig_offset - stream->max_sb_size,
1266 stream->max_sb_size, POSIX_FADV_DONTNEED);
3bd1e081
MD
1267}
1268
1269/*
1270 * Initialise the necessary environnement :
1271 * - create a new context
1272 * - create the poll_pipe
1273 * - create the should_quit pipe (for signal handler)
1274 * - create the thread pipe (for splice)
1275 *
1276 * Takes a function pointer as argument, this function is called when data is
1277 * available on a buffer. This function is responsible to do the
1278 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1279 * buffer configuration and then kernctl_put_next_subbuf at the end.
1280 *
1281 * Returns a pointer to the new context or NULL on error.
1282 */
1283struct lttng_consumer_local_data *lttng_consumer_create(
1284 enum lttng_consumer_type type,
4078b776 1285 ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream,
d41f73b7 1286 struct lttng_consumer_local_data *ctx),
3bd1e081
MD
1287 int (*recv_channel)(struct lttng_consumer_channel *channel),
1288 int (*recv_stream)(struct lttng_consumer_stream *stream),
30319bcb 1289 int (*update_stream)(uint64_t stream_key, uint32_t state))
3bd1e081 1290{
d8ef542d 1291 int ret;
3bd1e081
MD
1292 struct lttng_consumer_local_data *ctx;
1293
1294 assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN ||
1295 consumer_data.type == type);
1296 consumer_data.type = type;
1297
effcf122 1298 ctx = zmalloc(sizeof(struct lttng_consumer_local_data));
3bd1e081 1299 if (ctx == NULL) {
7a57cf92 1300 PERROR("allocating context");
3bd1e081
MD
1301 goto error;
1302 }
1303
1304 ctx->consumer_error_socket = -1;
331744e3 1305 ctx->consumer_metadata_socket = -1;
75d83e50 1306 pthread_mutex_init(&ctx->metadata_socket_lock, NULL);
3bd1e081
MD
1307 /* assign the callbacks */
1308 ctx->on_buffer_ready = buffer_ready;
1309 ctx->on_recv_channel = recv_channel;
1310 ctx->on_recv_stream = recv_stream;
1311 ctx->on_update_stream = update_stream;
1312
acdb9057
DG
1313 ctx->consumer_data_pipe = lttng_pipe_open(0);
1314 if (!ctx->consumer_data_pipe) {
3bd1e081
MD
1315 goto error_poll_pipe;
1316 }
1317
02b3d176
DG
1318 ctx->consumer_wakeup_pipe = lttng_pipe_open(0);
1319 if (!ctx->consumer_wakeup_pipe) {
1320 goto error_wakeup_pipe;
1321 }
1322
3bd1e081
MD
1323 ret = pipe(ctx->consumer_should_quit);
1324 if (ret < 0) {
7a57cf92 1325 PERROR("Error creating recv pipe");
3bd1e081
MD
1326 goto error_quit_pipe;
1327 }
1328
d8ef542d
MD
1329 ret = pipe(ctx->consumer_channel_pipe);
1330 if (ret < 0) {
1331 PERROR("Error creating channel pipe");
1332 goto error_channel_pipe;
1333 }
1334
13886d2d
DG
1335 ctx->consumer_metadata_pipe = lttng_pipe_open(0);
1336 if (!ctx->consumer_metadata_pipe) {
fb3a43a9
DG
1337 goto error_metadata_pipe;
1338 }
3bd1e081 1339
fb3a43a9 1340 return ctx;
3bd1e081 1341
fb3a43a9 1342error_metadata_pipe:
d8ef542d
MD
1343 utils_close_pipe(ctx->consumer_channel_pipe);
1344error_channel_pipe:
d8ef542d 1345 utils_close_pipe(ctx->consumer_should_quit);
3bd1e081 1346error_quit_pipe:
02b3d176
DG
1347 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
1348error_wakeup_pipe:
acdb9057 1349 lttng_pipe_destroy(ctx->consumer_data_pipe);
3bd1e081
MD
1350error_poll_pipe:
1351 free(ctx);
1352error:
1353 return NULL;
1354}
1355
282dadbc
MD
1356/*
1357 * Iterate over all streams of the hashtable and free them properly.
1358 */
1359static void destroy_data_stream_ht(struct lttng_ht *ht)
1360{
1361 struct lttng_ht_iter iter;
1362 struct lttng_consumer_stream *stream;
1363
1364 if (ht == NULL) {
1365 return;
1366 }
1367
1368 rcu_read_lock();
1369 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1370 /*
1371 * Ignore return value since we are currently cleaning up so any error
1372 * can't be handled.
1373 */
1374 (void) consumer_del_stream(stream, ht);
1375 }
1376 rcu_read_unlock();
1377
1378 lttng_ht_destroy(ht);
1379}
1380
1381/*
1382 * Iterate over all streams of the metadata hashtable and free them
1383 * properly.
1384 */
1385static void destroy_metadata_stream_ht(struct lttng_ht *ht)
1386{
1387 struct lttng_ht_iter iter;
1388 struct lttng_consumer_stream *stream;
1389
1390 if (ht == NULL) {
1391 return;
1392 }
1393
1394 rcu_read_lock();
1395 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1396 /*
1397 * Ignore return value since we are currently cleaning up so any error
1398 * can't be handled.
1399 */
1400 (void) consumer_del_metadata_stream(stream, ht);
1401 }
1402 rcu_read_unlock();
1403
1404 lttng_ht_destroy(ht);
1405}
1406
3bd1e081
MD
1407/*
1408 * Close all fds associated with the instance and free the context.
1409 */
1410void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx)
1411{
4c462e79
MD
1412 int ret;
1413
ab1027f4
DG
1414 DBG("Consumer destroying it. Closing everything.");
1415
4f2e75b9
DG
1416 if (!ctx) {
1417 return;
1418 }
1419
282dadbc
MD
1420 destroy_data_stream_ht(data_ht);
1421 destroy_metadata_stream_ht(metadata_ht);
1422
4c462e79
MD
1423 ret = close(ctx->consumer_error_socket);
1424 if (ret) {
1425 PERROR("close");
1426 }
331744e3
JD
1427 ret = close(ctx->consumer_metadata_socket);
1428 if (ret) {
1429 PERROR("close");
1430 }
d8ef542d 1431 utils_close_pipe(ctx->consumer_channel_pipe);
acdb9057 1432 lttng_pipe_destroy(ctx->consumer_data_pipe);
13886d2d 1433 lttng_pipe_destroy(ctx->consumer_metadata_pipe);
02b3d176 1434 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
d8ef542d 1435 utils_close_pipe(ctx->consumer_should_quit);
fb3a43a9 1436
3bd1e081
MD
1437 unlink(ctx->consumer_command_sock_path);
1438 free(ctx);
1439}
1440
6197aea7
DG
1441/*
1442 * Write the metadata stream id on the specified file descriptor.
1443 */
1444static int write_relayd_metadata_id(int fd,
1445 struct lttng_consumer_stream *stream,
ffe60014 1446 struct consumer_relayd_sock_pair *relayd, unsigned long padding)
6197aea7 1447{
6cd525e8 1448 ssize_t ret;
1d4dfdef 1449 struct lttcomm_relayd_metadata_payload hdr;
6197aea7 1450
1d4dfdef
DG
1451 hdr.stream_id = htobe64(stream->relayd_stream_id);
1452 hdr.padding_size = htobe32(padding);
6cd525e8
MD
1453 ret = lttng_write(fd, (void *) &hdr, sizeof(hdr));
1454 if (ret < sizeof(hdr)) {
d7b75ec8 1455 /*
6f04ed72 1456 * This error means that the fd's end is closed so ignore the PERROR
d7b75ec8
DG
1457 * not to clubber the error output since this can happen in a normal
1458 * code path.
1459 */
1460 if (errno != EPIPE) {
1461 PERROR("write metadata stream id");
1462 }
1463 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno);
534d2592
DG
1464 /*
1465 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1466 * handle writting the missing part so report that as an error and
1467 * don't lie to the caller.
1468 */
1469 ret = -1;
6197aea7
DG
1470 goto end;
1471 }
1d4dfdef
DG
1472 DBG("Metadata stream id %" PRIu64 " with padding %lu written before data",
1473 stream->relayd_stream_id, padding);
6197aea7
DG
1474
1475end:
6cd525e8 1476 return (int) ret;
6197aea7
DG
1477}
1478
3bd1e081 1479/*
09e26845
DG
1480 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1481 * core function for writing trace buffers to either the local filesystem or
1482 * the network.
1483 *
79d4ffb7
DG
1484 * It must be called with the stream lock held.
1485 *
09e26845 1486 * Careful review MUST be put if any changes occur!
3bd1e081
MD
1487 *
1488 * Returns the number of bytes written
1489 */
4078b776 1490ssize_t lttng_consumer_on_read_subbuffer_mmap(
3bd1e081 1491 struct lttng_consumer_local_data *ctx,
1d4dfdef 1492 struct lttng_consumer_stream *stream, unsigned long len,
309167d2 1493 unsigned long padding,
50adc264 1494 struct ctf_packet_index *index)
3bd1e081 1495{
f02e1e8a 1496 unsigned long mmap_offset;
ffe60014 1497 void *mmap_base;
994ab360 1498 ssize_t ret = 0;
f02e1e8a
DG
1499 off_t orig_offset = stream->out_fd_offset;
1500 /* Default is on the disk */
1501 int outfd = stream->out_fd;
f02e1e8a 1502 struct consumer_relayd_sock_pair *relayd = NULL;
8994307f 1503 unsigned int relayd_hang_up = 0;
f02e1e8a
DG
1504
1505 /* RCU lock for the relayd pointer */
1506 rcu_read_lock();
1507
1508 /* Flag that the current stream if set for network streaming. */
da009f2c 1509 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1510 relayd = consumer_find_relayd(stream->net_seq_idx);
1511 if (relayd == NULL) {
56591bac 1512 ret = -EPIPE;
f02e1e8a
DG
1513 goto end;
1514 }
1515 }
1516
1517 /* get the offset inside the fd to mmap */
3bd1e081
MD
1518 switch (consumer_data.type) {
1519 case LTTNG_CONSUMER_KERNEL:
ffe60014 1520 mmap_base = stream->mmap_base;
f02e1e8a 1521 ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset);
994ab360
DG
1522 if (ret < 0) {
1523 ret = -errno;
56591bac 1524 PERROR("tracer ctl get_mmap_read_offset");
56591bac
MD
1525 goto end;
1526 }
f02e1e8a 1527 break;
7753dea8
MD
1528 case LTTNG_CONSUMER32_UST:
1529 case LTTNG_CONSUMER64_UST:
ffe60014
DG
1530 mmap_base = lttng_ustctl_get_mmap_base(stream);
1531 if (!mmap_base) {
1532 ERR("read mmap get mmap base for stream %s", stream->name);
994ab360 1533 ret = -EPERM;
ffe60014
DG
1534 goto end;
1535 }
1536 ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset);
56591bac
MD
1537 if (ret != 0) {
1538 PERROR("tracer ctl get_mmap_read_offset");
994ab360 1539 ret = -EINVAL;
56591bac
MD
1540 goto end;
1541 }
f02e1e8a 1542 break;
3bd1e081
MD
1543 default:
1544 ERR("Unknown consumer_data type");
1545 assert(0);
1546 }
b9182dd9 1547
f02e1e8a
DG
1548 /* Handle stream on the relayd if the output is on the network */
1549 if (relayd) {
1550 unsigned long netlen = len;
1551
1552 /*
1553 * Lock the control socket for the complete duration of the function
1554 * since from this point on we will use the socket.
1555 */
1556 if (stream->metadata_flag) {
1557 /* Metadata requires the control socket. */
1558 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1d4dfdef 1559 netlen += sizeof(struct lttcomm_relayd_metadata_payload);
f02e1e8a
DG
1560 }
1561
1d4dfdef 1562 ret = write_relayd_stream_header(stream, netlen, padding, relayd);
994ab360
DG
1563 if (ret < 0) {
1564 relayd_hang_up = 1;
1565 goto write_error;
1566 }
1567 /* Use the returned socket. */
1568 outfd = ret;
f02e1e8a 1569
994ab360
DG
1570 /* Write metadata stream id before payload */
1571 if (stream->metadata_flag) {
1572 ret = write_relayd_metadata_id(outfd, stream, relayd, padding);
1573 if (ret < 0) {
8994307f
DG
1574 relayd_hang_up = 1;
1575 goto write_error;
1576 }
f02e1e8a 1577 }
1d4dfdef
DG
1578 } else {
1579 /* No streaming, we have to set the len with the full padding */
1580 len += padding;
1624d5b7
JD
1581
1582 /*
1583 * Check if we need to change the tracefile before writing the packet.
1584 */
1585 if (stream->chan->tracefile_size > 0 &&
1586 (stream->tracefile_size_current + len) >
1587 stream->chan->tracefile_size) {
fe4477ee
JD
1588 ret = utils_rotate_stream_file(stream->chan->pathname,
1589 stream->name, stream->chan->tracefile_size,
1590 stream->chan->tracefile_count, stream->uid, stream->gid,
309167d2
JD
1591 stream->out_fd, &(stream->tracefile_count_current),
1592 &stream->out_fd);
1624d5b7
JD
1593 if (ret < 0) {
1594 ERR("Rotating output file");
1595 goto end;
1596 }
309167d2
JD
1597 outfd = stream->out_fd;
1598
1599 if (stream->index_fd >= 0) {
adf7a0a1
JG
1600 ret = close(stream->index_fd);
1601 if (ret < 0) {
1602 PERROR("Closing index");
1603 goto end;
1604 }
1605 stream->index_fd = -1;
309167d2
JD
1606 ret = index_create_file(stream->chan->pathname,
1607 stream->name, stream->uid, stream->gid,
1608 stream->chan->tracefile_size,
1609 stream->tracefile_count_current);
1610 if (ret < 0) {
1611 goto end;
1612 }
1613 stream->index_fd = ret;
1614 }
1615
a6976990
DG
1616 /* Reset current size because we just perform a rotation. */
1617 stream->tracefile_size_current = 0;
a1ae300f
JD
1618 stream->out_fd_offset = 0;
1619 orig_offset = 0;
1624d5b7
JD
1620 }
1621 stream->tracefile_size_current += len;
309167d2
JD
1622 if (index) {
1623 index->offset = htobe64(stream->out_fd_offset);
1624 }
f02e1e8a
DG
1625 }
1626
d02b8372
DG
1627 /*
1628 * This call guarantee that len or less is returned. It's impossible to
1629 * receive a ret value that is bigger than len.
1630 */
1631 ret = lttng_write(outfd, mmap_base + mmap_offset, len);
1632 DBG("Consumer mmap write() ret %zd (len %lu)", ret, len);
1633 if (ret < 0 || ((size_t) ret != len)) {
1634 /*
1635 * Report error to caller if nothing was written else at least send the
1636 * amount written.
1637 */
1638 if (ret < 0) {
994ab360 1639 ret = -errno;
f02e1e8a 1640 }
994ab360 1641 relayd_hang_up = 1;
f02e1e8a 1642
d02b8372 1643 /* Socket operation failed. We consider the relayd dead */
994ab360 1644 if (errno == EPIPE || errno == EINVAL || errno == EBADF) {
d02b8372
DG
1645 /*
1646 * This is possible if the fd is closed on the other side
1647 * (outfd) or any write problem. It can be verbose a bit for a
1648 * normal execution if for instance the relayd is stopped
1649 * abruptly. This can happen so set this to a DBG statement.
1650 */
1651 DBG("Consumer mmap write detected relayd hang up");
994ab360
DG
1652 } else {
1653 /* Unhandled error, print it and stop function right now. */
1654 PERROR("Error in write mmap (ret %zd != len %lu)", ret, len);
f02e1e8a 1655 }
994ab360 1656 goto write_error;
d02b8372
DG
1657 }
1658 stream->output_written += ret;
d02b8372
DG
1659
1660 /* This call is useless on a socket so better save a syscall. */
1661 if (!relayd) {
1662 /* This won't block, but will start writeout asynchronously */
1663 lttng_sync_file_range(outfd, stream->out_fd_offset, len,
1664 SYNC_FILE_RANGE_WRITE);
1665 stream->out_fd_offset += len;
f02e1e8a
DG
1666 }
1667 lttng_consumer_sync_trace_file(stream, orig_offset);
1668
8994307f
DG
1669write_error:
1670 /*
1671 * This is a special case that the relayd has closed its socket. Let's
1672 * cleanup the relayd object and all associated streams.
1673 */
1674 if (relayd && relayd_hang_up) {
1675 cleanup_relayd(relayd, ctx);
1676 }
1677
f02e1e8a
DG
1678end:
1679 /* Unlock only if ctrl socket used */
1680 if (relayd && stream->metadata_flag) {
1681 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1682 }
1683
1684 rcu_read_unlock();
994ab360 1685 return ret;
3bd1e081
MD
1686}
1687
1688/*
1689 * Splice the data from the ring buffer to the tracefile.
1690 *
79d4ffb7
DG
1691 * It must be called with the stream lock held.
1692 *
3bd1e081
MD
1693 * Returns the number of bytes spliced.
1694 */
4078b776 1695ssize_t lttng_consumer_on_read_subbuffer_splice(
3bd1e081 1696 struct lttng_consumer_local_data *ctx,
1d4dfdef 1697 struct lttng_consumer_stream *stream, unsigned long len,
309167d2 1698 unsigned long padding,
50adc264 1699 struct ctf_packet_index *index)
3bd1e081 1700{
f02e1e8a
DG
1701 ssize_t ret = 0, written = 0, ret_splice = 0;
1702 loff_t offset = 0;
1703 off_t orig_offset = stream->out_fd_offset;
1704 int fd = stream->wait_fd;
1705 /* Default is on the disk */
1706 int outfd = stream->out_fd;
f02e1e8a 1707 struct consumer_relayd_sock_pair *relayd = NULL;
fb3a43a9 1708 int *splice_pipe;
8994307f 1709 unsigned int relayd_hang_up = 0;
f02e1e8a 1710
3bd1e081
MD
1711 switch (consumer_data.type) {
1712 case LTTNG_CONSUMER_KERNEL:
f02e1e8a 1713 break;
7753dea8
MD
1714 case LTTNG_CONSUMER32_UST:
1715 case LTTNG_CONSUMER64_UST:
f02e1e8a 1716 /* Not supported for user space tracing */
3bd1e081
MD
1717 return -ENOSYS;
1718 default:
1719 ERR("Unknown consumer_data type");
1720 assert(0);
3bd1e081
MD
1721 }
1722
f02e1e8a
DG
1723 /* RCU lock for the relayd pointer */
1724 rcu_read_lock();
1725
1726 /* Flag that the current stream if set for network streaming. */
da009f2c 1727 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1728 relayd = consumer_find_relayd(stream->net_seq_idx);
1729 if (relayd == NULL) {
ad0b0d23 1730 written = -ret;
f02e1e8a
DG
1731 goto end;
1732 }
1733 }
a2361a61 1734 splice_pipe = stream->splice_pipe;
fb3a43a9 1735
f02e1e8a 1736 /* Write metadata stream id before payload */
1d4dfdef 1737 if (relayd) {
ad0b0d23 1738 unsigned long total_len = len;
f02e1e8a 1739
1d4dfdef
DG
1740 if (stream->metadata_flag) {
1741 /*
1742 * Lock the control socket for the complete duration of the function
1743 * since from this point on we will use the socket.
1744 */
1745 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1746
1747 ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd,
1748 padding);
1749 if (ret < 0) {
1750 written = ret;
ad0b0d23
DG
1751 relayd_hang_up = 1;
1752 goto write_error;
1d4dfdef
DG
1753 }
1754
1755 total_len += sizeof(struct lttcomm_relayd_metadata_payload);
1756 }
1757
1758 ret = write_relayd_stream_header(stream, total_len, padding, relayd);
ad0b0d23
DG
1759 if (ret < 0) {
1760 written = ret;
1761 relayd_hang_up = 1;
1762 goto write_error;
f02e1e8a 1763 }
ad0b0d23
DG
1764 /* Use the returned socket. */
1765 outfd = ret;
1d4dfdef
DG
1766 } else {
1767 /* No streaming, we have to set the len with the full padding */
1768 len += padding;
1624d5b7
JD
1769
1770 /*
1771 * Check if we need to change the tracefile before writing the packet.
1772 */
1773 if (stream->chan->tracefile_size > 0 &&
1774 (stream->tracefile_size_current + len) >
1775 stream->chan->tracefile_size) {
fe4477ee
JD
1776 ret = utils_rotate_stream_file(stream->chan->pathname,
1777 stream->name, stream->chan->tracefile_size,
1778 stream->chan->tracefile_count, stream->uid, stream->gid,
309167d2
JD
1779 stream->out_fd, &(stream->tracefile_count_current),
1780 &stream->out_fd);
1624d5b7 1781 if (ret < 0) {
ad0b0d23 1782 written = ret;
1624d5b7
JD
1783 ERR("Rotating output file");
1784 goto end;
1785 }
309167d2
JD
1786 outfd = stream->out_fd;
1787
1788 if (stream->index_fd >= 0) {
3c1150e1
JD
1789 ret = close(stream->index_fd);
1790 if (ret < 0) {
1791 PERROR("Closing index");
1792 goto end;
1793 }
1794 stream->index_fd = -1;
309167d2
JD
1795 ret = index_create_file(stream->chan->pathname,
1796 stream->name, stream->uid, stream->gid,
1797 stream->chan->tracefile_size,
1798 stream->tracefile_count_current);
1799 if (ret < 0) {
ad0b0d23 1800 written = ret;
309167d2
JD
1801 goto end;
1802 }
1803 stream->index_fd = ret;
1804 }
1805
a6976990
DG
1806 /* Reset current size because we just perform a rotation. */
1807 stream->tracefile_size_current = 0;
a1ae300f
JD
1808 stream->out_fd_offset = 0;
1809 orig_offset = 0;
1624d5b7
JD
1810 }
1811 stream->tracefile_size_current += len;
309167d2 1812 index->offset = htobe64(stream->out_fd_offset);
f02e1e8a
DG
1813 }
1814
1815 while (len > 0) {
1d4dfdef
DG
1816 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1817 (unsigned long)offset, len, fd, splice_pipe[1]);
fb3a43a9 1818 ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len,
f02e1e8a
DG
1819 SPLICE_F_MOVE | SPLICE_F_MORE);
1820 DBG("splice chan to pipe, ret %zd", ret_splice);
1821 if (ret_splice < 0) {
d02b8372 1822 ret = errno;
ad0b0d23 1823 written = -ret;
d02b8372 1824 PERROR("Error in relay splice");
f02e1e8a
DG
1825 goto splice_error;
1826 }
1827
1828 /* Handle stream on the relayd if the output is on the network */
ad0b0d23
DG
1829 if (relayd && stream->metadata_flag) {
1830 size_t metadata_payload_size =
1831 sizeof(struct lttcomm_relayd_metadata_payload);
1832
1833 /* Update counter to fit the spliced data */
1834 ret_splice += metadata_payload_size;
1835 len += metadata_payload_size;
1836 /*
1837 * We do this so the return value can match the len passed as
1838 * argument to this function.
1839 */
1840 written -= metadata_payload_size;
f02e1e8a
DG
1841 }
1842
1843 /* Splice data out */
fb3a43a9 1844 ret_splice = splice(splice_pipe[0], NULL, outfd, NULL,
f02e1e8a 1845 ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE);
a2361a61
JD
1846 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1847 outfd, ret_splice);
f02e1e8a 1848 if (ret_splice < 0) {
d02b8372 1849 ret = errno;
ad0b0d23
DG
1850 written = -ret;
1851 relayd_hang_up = 1;
1852 goto write_error;
f02e1e8a 1853 } else if (ret_splice > len) {
d02b8372
DG
1854 /*
1855 * We don't expect this code path to be executed but you never know
1856 * so this is an extra protection agains a buggy splice().
1857 */
f02e1e8a 1858 ret = errno;
ad0b0d23 1859 written += ret_splice;
d02b8372
DG
1860 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice,
1861 len);
f02e1e8a 1862 goto splice_error;
d02b8372
DG
1863 } else {
1864 /* All good, update current len and continue. */
1865 len -= ret_splice;
f02e1e8a 1866 }
f02e1e8a
DG
1867
1868 /* This call is useless on a socket so better save a syscall. */
1869 if (!relayd) {
1870 /* This won't block, but will start writeout asynchronously */
1871 lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice,
1872 SYNC_FILE_RANGE_WRITE);
1873 stream->out_fd_offset += ret_splice;
1874 }
e5d1a9b3 1875 stream->output_written += ret_splice;
f02e1e8a
DG
1876 written += ret_splice;
1877 }
1878 lttng_consumer_sync_trace_file(stream, orig_offset);
f02e1e8a
DG
1879 goto end;
1880
8994307f
DG
1881write_error:
1882 /*
1883 * This is a special case that the relayd has closed its socket. Let's
1884 * cleanup the relayd object and all associated streams.
1885 */
1886 if (relayd && relayd_hang_up) {
1887 cleanup_relayd(relayd, ctx);
1888 /* Skip splice error so the consumer does not fail */
1889 goto end;
1890 }
1891
f02e1e8a
DG
1892splice_error:
1893 /* send the appropriate error description to sessiond */
1894 switch (ret) {
f02e1e8a 1895 case EINVAL:
f73fabfd 1896 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL);
f02e1e8a
DG
1897 break;
1898 case ENOMEM:
f73fabfd 1899 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM);
f02e1e8a
DG
1900 break;
1901 case ESPIPE:
f73fabfd 1902 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE);
f02e1e8a
DG
1903 break;
1904 }
1905
1906end:
1907 if (relayd && stream->metadata_flag) {
1908 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1909 }
1910
1911 rcu_read_unlock();
1912 return written;
3bd1e081
MD
1913}
1914
1915/*
1916 * Take a snapshot for a specific fd
1917 *
1918 * Returns 0 on success, < 0 on error
1919 */
ffe60014 1920int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream)
3bd1e081
MD
1921{
1922 switch (consumer_data.type) {
1923 case LTTNG_CONSUMER_KERNEL:
ffe60014 1924 return lttng_kconsumer_take_snapshot(stream);
7753dea8
MD
1925 case LTTNG_CONSUMER32_UST:
1926 case LTTNG_CONSUMER64_UST:
ffe60014 1927 return lttng_ustconsumer_take_snapshot(stream);
3bd1e081
MD
1928 default:
1929 ERR("Unknown consumer_data type");
1930 assert(0);
1931 return -ENOSYS;
1932 }
3bd1e081
MD
1933}
1934
1935/*
1936 * Get the produced position
1937 *
1938 * Returns 0 on success, < 0 on error
1939 */
ffe60014 1940int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
3bd1e081
MD
1941 unsigned long *pos)
1942{
1943 switch (consumer_data.type) {
1944 case LTTNG_CONSUMER_KERNEL:
ffe60014 1945 return lttng_kconsumer_get_produced_snapshot(stream, pos);
7753dea8
MD
1946 case LTTNG_CONSUMER32_UST:
1947 case LTTNG_CONSUMER64_UST:
ffe60014 1948 return lttng_ustconsumer_get_produced_snapshot(stream, pos);
3bd1e081
MD
1949 default:
1950 ERR("Unknown consumer_data type");
1951 assert(0);
1952 return -ENOSYS;
1953 }
1954}
1955
1956int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1957 int sock, struct pollfd *consumer_sockpoll)
1958{
1959 switch (consumer_data.type) {
1960 case LTTNG_CONSUMER_KERNEL:
1961 return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
7753dea8
MD
1962 case LTTNG_CONSUMER32_UST:
1963 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
1964 return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
1965 default:
1966 ERR("Unknown consumer_data type");
1967 assert(0);
1968 return -ENOSYS;
1969 }
1970}
1971
6d574024 1972void lttng_consumer_close_all_metadata(void)
d88aee68
DG
1973{
1974 switch (consumer_data.type) {
1975 case LTTNG_CONSUMER_KERNEL:
1976 /*
1977 * The Kernel consumer has a different metadata scheme so we don't
1978 * close anything because the stream will be closed by the session
1979 * daemon.
1980 */
1981 break;
1982 case LTTNG_CONSUMER32_UST:
1983 case LTTNG_CONSUMER64_UST:
1984 /*
1985 * Close all metadata streams. The metadata hash table is passed and
1986 * this call iterates over it by closing all wakeup fd. This is safe
1987 * because at this point we are sure that the metadata producer is
1988 * either dead or blocked.
1989 */
6d574024 1990 lttng_ustconsumer_close_all_metadata(metadata_ht);
d88aee68
DG
1991 break;
1992 default:
1993 ERR("Unknown consumer_data type");
1994 assert(0);
1995 }
1996}
1997
fb3a43a9
DG
1998/*
1999 * Clean up a metadata stream and free its memory.
2000 */
e316aad5
DG
2001void consumer_del_metadata_stream(struct lttng_consumer_stream *stream,
2002 struct lttng_ht *ht)
fb3a43a9 2003{
e316aad5 2004 struct lttng_consumer_channel *free_chan = NULL;
fb3a43a9
DG
2005
2006 assert(stream);
2007 /*
2008 * This call should NEVER receive regular stream. It must always be
2009 * metadata stream and this is crucial for data structure synchronization.
2010 */
2011 assert(stream->metadata_flag);
2012
e316aad5
DG
2013 DBG3("Consumer delete metadata stream %d", stream->wait_fd);
2014
74251bb8 2015 pthread_mutex_lock(&consumer_data.lock);
a9838785 2016 pthread_mutex_lock(&stream->chan->lock);
8994307f
DG
2017 pthread_mutex_lock(&stream->lock);
2018
6d574024
DG
2019 /* Remove any reference to that stream. */
2020 consumer_stream_delete(stream, ht);
ca22feea 2021
6d574024
DG
2022 /* Close down everything including the relayd if one. */
2023 consumer_stream_close(stream);
2024 /* Destroy tracer buffers of the stream. */
2025 consumer_stream_destroy_buffers(stream);
fb3a43a9
DG
2026
2027 /* Atomically decrement channel refcount since other threads can use it. */
f2ad556d 2028 if (!uatomic_sub_return(&stream->chan->refcount, 1)
ffe60014 2029 && !uatomic_read(&stream->chan->nb_init_stream_left)) {
c30aaa51 2030 /* Go for channel deletion! */
e316aad5 2031 free_chan = stream->chan;
fb3a43a9
DG
2032 }
2033
73811ecc
DG
2034 /*
2035 * Nullify the stream reference so it is not used after deletion. The
6d574024
DG
2036 * channel lock MUST be acquired before being able to check for a NULL
2037 * pointer value.
73811ecc
DG
2038 */
2039 stream->chan->metadata_stream = NULL;
2040
8994307f 2041 pthread_mutex_unlock(&stream->lock);
a9838785 2042 pthread_mutex_unlock(&stream->chan->lock);
74251bb8 2043 pthread_mutex_unlock(&consumer_data.lock);
e316aad5
DG
2044
2045 if (free_chan) {
2046 consumer_del_channel(free_chan);
2047 }
2048
6d574024 2049 consumer_stream_free(stream);
fb3a43a9
DG
2050}
2051
2052/*
2053 * Action done with the metadata stream when adding it to the consumer internal
2054 * data structures to handle it.
2055 */
5ab66908 2056int consumer_add_metadata_stream(struct lttng_consumer_stream *stream)
fb3a43a9 2057{
5ab66908 2058 struct lttng_ht *ht = metadata_ht;
e316aad5 2059 int ret = 0;
76082088 2060 struct lttng_ht_iter iter;
d88aee68 2061 struct lttng_ht_node_u64 *node;
fb3a43a9 2062
e316aad5
DG
2063 assert(stream);
2064 assert(ht);
2065
d88aee68 2066 DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key);
e316aad5
DG
2067
2068 pthread_mutex_lock(&consumer_data.lock);
a9838785 2069 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 2070 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 2071 pthread_mutex_lock(&stream->lock);
e316aad5 2072
e316aad5
DG
2073 /*
2074 * From here, refcounts are updated so be _careful_ when returning an error
2075 * after this point.
2076 */
2077
fb3a43a9 2078 rcu_read_lock();
76082088
DG
2079
2080 /*
2081 * Lookup the stream just to make sure it does not exist in our internal
2082 * state. This should NEVER happen.
2083 */
d88aee68
DG
2084 lttng_ht_lookup(ht, &stream->key, &iter);
2085 node = lttng_ht_iter_get_node_u64(&iter);
76082088
DG
2086 assert(!node);
2087
e316aad5 2088 /*
ffe60014
DG
2089 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2090 * in terms of destroying the associated channel, because the action that
e316aad5
DG
2091 * causes the count to become 0 also causes a stream to be added. The
2092 * channel deletion will thus be triggered by the following removal of this
2093 * stream.
2094 */
ffe60014 2095 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
2096 /* Increment refcount before decrementing nb_init_stream_left */
2097 cmm_smp_wmb();
ffe60014 2098 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
2099 }
2100
d88aee68 2101 lttng_ht_add_unique_u64(ht, &stream->node);
ca22feea 2102
d8ef542d
MD
2103 lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht,
2104 &stream->node_channel_id);
2105
ca22feea
DG
2106 /*
2107 * Add stream to the stream_list_ht of the consumer data. No need to steal
2108 * the key since the HT does not use it and we allow to add redundant keys
2109 * into this table.
2110 */
d88aee68 2111 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 2112
fb3a43a9 2113 rcu_read_unlock();
e316aad5 2114
2e818a6a 2115 pthread_mutex_unlock(&stream->lock);
a9838785 2116 pthread_mutex_unlock(&stream->chan->lock);
ec6ea7d0 2117 pthread_mutex_unlock(&stream->chan->timer_lock);
e316aad5
DG
2118 pthread_mutex_unlock(&consumer_data.lock);
2119 return ret;
fb3a43a9
DG
2120}
2121
8994307f
DG
2122/*
2123 * Delete data stream that are flagged for deletion (endpoint_status).
2124 */
2125static void validate_endpoint_status_data_stream(void)
2126{
2127 struct lttng_ht_iter iter;
2128 struct lttng_consumer_stream *stream;
2129
2130 DBG("Consumer delete flagged data stream");
2131
2132 rcu_read_lock();
2133 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
2134 /* Validate delete flag of the stream */
79d4ffb7 2135 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2136 continue;
2137 }
2138 /* Delete it right now */
2139 consumer_del_stream(stream, data_ht);
2140 }
2141 rcu_read_unlock();
2142}
2143
2144/*
2145 * Delete metadata stream that are flagged for deletion (endpoint_status).
2146 */
2147static void validate_endpoint_status_metadata_stream(
2148 struct lttng_poll_event *pollset)
2149{
2150 struct lttng_ht_iter iter;
2151 struct lttng_consumer_stream *stream;
2152
2153 DBG("Consumer delete flagged metadata stream");
2154
2155 assert(pollset);
2156
2157 rcu_read_lock();
2158 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
2159 /* Validate delete flag of the stream */
79d4ffb7 2160 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2161 continue;
2162 }
2163 /*
2164 * Remove from pollset so the metadata thread can continue without
2165 * blocking on a deleted stream.
2166 */
2167 lttng_poll_del(pollset, stream->wait_fd);
2168
2169 /* Delete it right now */
2170 consumer_del_metadata_stream(stream, metadata_ht);
2171 }
2172 rcu_read_unlock();
2173}
2174
fb3a43a9
DG
2175/*
2176 * Thread polls on metadata file descriptor and write them on disk or on the
2177 * network.
2178 */
7d980def 2179void *consumer_thread_metadata_poll(void *data)
fb3a43a9 2180{
1fc79fb4 2181 int ret, i, pollfd, err = -1;
fb3a43a9 2182 uint32_t revents, nb_fd;
e316aad5 2183 struct lttng_consumer_stream *stream = NULL;
fb3a43a9 2184 struct lttng_ht_iter iter;
d88aee68 2185 struct lttng_ht_node_u64 *node;
fb3a43a9
DG
2186 struct lttng_poll_event events;
2187 struct lttng_consumer_local_data *ctx = data;
2188 ssize_t len;
2189
2190 rcu_register_thread();
2191
1fc79fb4
MD
2192 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA);
2193
2d57de81
MD
2194 if (testpoint(consumerd_thread_metadata)) {
2195 goto error_testpoint;
2196 }
2197
9ce5646a
MD
2198 health_code_update();
2199
fb3a43a9
DG
2200 DBG("Thread metadata poll started");
2201
fb3a43a9
DG
2202 /* Size is set to 1 for the consumer_metadata pipe */
2203 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2204 if (ret < 0) {
2205 ERR("Poll set creation failed");
d8ef542d 2206 goto end_poll;
fb3a43a9
DG
2207 }
2208
13886d2d
DG
2209 ret = lttng_poll_add(&events,
2210 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN);
fb3a43a9
DG
2211 if (ret < 0) {
2212 goto end;
2213 }
2214
2215 /* Main loop */
2216 DBG("Metadata main loop started");
2217
2218 while (1) {
fb3a43a9 2219restart:
7fa2082e 2220 health_code_update();
9ce5646a 2221 health_poll_entry();
7fa2082e 2222 DBG("Metadata poll wait");
fb3a43a9 2223 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2224 DBG("Metadata poll return from wait with %d fd(s)",
2225 LTTNG_POLL_GETNB(&events));
9ce5646a 2226 health_poll_exit();
fb3a43a9
DG
2227 DBG("Metadata event catched in thread");
2228 if (ret < 0) {
2229 if (errno == EINTR) {
e316aad5 2230 ERR("Poll EINTR catched");
fb3a43a9
DG
2231 goto restart;
2232 }
d9607cd7
MD
2233 if (LTTNG_POLL_GETNB(&events) == 0) {
2234 err = 0; /* All is OK */
2235 }
2236 goto end;
fb3a43a9
DG
2237 }
2238
0d9c5d77
DG
2239 nb_fd = ret;
2240
e316aad5 2241 /* From here, the event is a metadata wait fd */
fb3a43a9 2242 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2243 health_code_update();
2244
fb3a43a9
DG
2245 revents = LTTNG_POLL_GETEV(&events, i);
2246 pollfd = LTTNG_POLL_GETFD(&events, i);
2247
fd20dac9
MD
2248 if (!revents) {
2249 /* No activity for this FD (poll implementation). */
2250 continue;
2251 }
2252
13886d2d 2253 if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) {
03e43155 2254 if (revents & LPOLLIN) {
13886d2d
DG
2255 ssize_t pipe_len;
2256
2257 pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe,
2258 &stream, sizeof(stream));
6cd525e8 2259 if (pipe_len < sizeof(stream)) {
03e43155
MD
2260 if (pipe_len < 0) {
2261 PERROR("read metadata stream");
2262 }
fb3a43a9 2263 /*
03e43155
MD
2264 * Remove the pipe from the poll set and continue the loop
2265 * since their might be data to consume.
fb3a43a9 2266 */
03e43155
MD
2267 lttng_poll_del(&events,
2268 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2269 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
fb3a43a9
DG
2270 continue;
2271 }
2272
8994307f
DG
2273 /* A NULL stream means that the state has changed. */
2274 if (stream == NULL) {
2275 /* Check for deleted streams. */
2276 validate_endpoint_status_metadata_stream(&events);
3714380f 2277 goto restart;
8994307f
DG
2278 }
2279
fb3a43a9
DG
2280 DBG("Adding metadata stream %d to poll set",
2281 stream->wait_fd);
2282
fb3a43a9
DG
2283 /* Add metadata stream to the global poll events list */
2284 lttng_poll_add(&events, stream->wait_fd,
6d574024 2285 LPOLLIN | LPOLLPRI | LPOLLHUP);
03e43155
MD
2286 } else if (revents & (LPOLLERR | LPOLLHUP)) {
2287 DBG("Metadata thread pipe hung up");
2288 /*
2289 * Remove the pipe from the poll set and continue the loop
2290 * since their might be data to consume.
2291 */
2292 lttng_poll_del(&events,
2293 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2294 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
2295 continue;
2296 } else {
2297 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2298 goto end;
fb3a43a9
DG
2299 }
2300
e316aad5 2301 /* Handle other stream */
fb3a43a9
DG
2302 continue;
2303 }
2304
d09e1200 2305 rcu_read_lock();
d88aee68
DG
2306 {
2307 uint64_t tmp_id = (uint64_t) pollfd;
2308
2309 lttng_ht_lookup(metadata_ht, &tmp_id, &iter);
2310 }
2311 node = lttng_ht_iter_get_node_u64(&iter);
e316aad5 2312 assert(node);
fb3a43a9
DG
2313
2314 stream = caa_container_of(node, struct lttng_consumer_stream,
58b1f425 2315 node);
fb3a43a9 2316
03e43155
MD
2317 if (revents & (LPOLLIN | LPOLLPRI)) {
2318 /* Get the data out of the metadata file descriptor */
2319 DBG("Metadata available on fd %d", pollfd);
2320 assert(stream->wait_fd == pollfd);
2321
2322 do {
2323 health_code_update();
2324
2325 len = ctx->on_buffer_ready(stream, ctx);
2326 /*
2327 * We don't check the return value here since if we get
2328 * a negative len, it means an error occured thus we
2329 * simply remove it from the poll set and free the
2330 * stream.
2331 */
2332 } while (len > 0);
2333
2334 /* It's ok to have an unavailable sub-buffer */
2335 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
2336 /* Clean up stream from consumer and free it. */
2337 lttng_poll_del(&events, stream->wait_fd);
2338 consumer_del_metadata_stream(stream, metadata_ht);
2339 }
2340 } else if (revents & (LPOLLERR | LPOLLHUP)) {
e316aad5 2341 DBG("Metadata fd %d is hup|err.", pollfd);
fb3a43a9
DG
2342 if (!stream->hangup_flush_done
2343 && (consumer_data.type == LTTNG_CONSUMER32_UST
2344 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2345 DBG("Attempting to flush and consume the UST buffers");
2346 lttng_ustconsumer_on_stream_hangup(stream);
2347
2348 /* We just flushed the stream now read it. */
4bb94b75 2349 do {
9ce5646a
MD
2350 health_code_update();
2351
4bb94b75
DG
2352 len = ctx->on_buffer_ready(stream, ctx);
2353 /*
2354 * We don't check the return value here since if we get
2355 * a negative len, it means an error occured thus we
2356 * simply remove it from the poll set and free the
2357 * stream.
2358 */
2359 } while (len > 0);
fb3a43a9
DG
2360 }
2361
fb3a43a9 2362 lttng_poll_del(&events, stream->wait_fd);
e316aad5
DG
2363 /*
2364 * This call update the channel states, closes file descriptors
2365 * and securely free the stream.
2366 */
2367 consumer_del_metadata_stream(stream, metadata_ht);
03e43155
MD
2368 } else {
2369 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
6f2f1a70 2370 rcu_read_unlock();
03e43155 2371 goto end;
fb3a43a9 2372 }
e316aad5 2373 /* Release RCU lock for the stream looked up */
d09e1200 2374 rcu_read_unlock();
fb3a43a9
DG
2375 }
2376 }
2377
1fc79fb4
MD
2378 /* All is OK */
2379 err = 0;
fb3a43a9
DG
2380end:
2381 DBG("Metadata poll thread exiting");
fb3a43a9 2382
d8ef542d
MD
2383 lttng_poll_clean(&events);
2384end_poll:
2d57de81 2385error_testpoint:
1fc79fb4
MD
2386 if (err) {
2387 health_error();
2388 ERR("Health error occurred in %s", __func__);
2389 }
2390 health_unregister(health_consumerd);
fb3a43a9
DG
2391 rcu_unregister_thread();
2392 return NULL;
2393}
2394
3bd1e081 2395/*
e4421fec 2396 * This thread polls the fds in the set to consume the data and write
3bd1e081
MD
2397 * it to tracefile if necessary.
2398 */
7d980def 2399void *consumer_thread_data_poll(void *data)
3bd1e081 2400{
1fc79fb4 2401 int num_rdy, num_hup, high_prio, ret, i, err = -1;
3bd1e081
MD
2402 struct pollfd *pollfd = NULL;
2403 /* local view of the streams */
c869f647 2404 struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL;
3bd1e081
MD
2405 /* local view of consumer_data.fds_count */
2406 int nb_fd = 0;
3bd1e081 2407 struct lttng_consumer_local_data *ctx = data;
00e2e675 2408 ssize_t len;
3bd1e081 2409
e7b994a3
DG
2410 rcu_register_thread();
2411
1fc79fb4
MD
2412 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA);
2413
2d57de81
MD
2414 if (testpoint(consumerd_thread_data)) {
2415 goto error_testpoint;
2416 }
2417
9ce5646a
MD
2418 health_code_update();
2419
4df6c8cb
MD
2420 local_stream = zmalloc(sizeof(struct lttng_consumer_stream *));
2421 if (local_stream == NULL) {
2422 PERROR("local_stream malloc");
2423 goto end;
2424 }
3bd1e081
MD
2425
2426 while (1) {
9ce5646a
MD
2427 health_code_update();
2428
3bd1e081
MD
2429 high_prio = 0;
2430 num_hup = 0;
2431
2432 /*
e4421fec 2433 * the fds set has been updated, we need to update our
3bd1e081
MD
2434 * local array as well
2435 */
2436 pthread_mutex_lock(&consumer_data.lock);
2437 if (consumer_data.need_update) {
0e428499
DG
2438 free(pollfd);
2439 pollfd = NULL;
2440
2441 free(local_stream);
2442 local_stream = NULL;
3bd1e081 2443
02b3d176
DG
2444 /*
2445 * Allocate for all fds +1 for the consumer_data_pipe and +1 for
2446 * wake up pipe.
2447 */
2448 pollfd = zmalloc((consumer_data.stream_count + 2) * sizeof(struct pollfd));
3bd1e081 2449 if (pollfd == NULL) {
7a57cf92 2450 PERROR("pollfd malloc");
3bd1e081
MD
2451 pthread_mutex_unlock(&consumer_data.lock);
2452 goto end;
2453 }
2454
02b3d176 2455 local_stream = zmalloc((consumer_data.stream_count + 2) *
747f8642 2456 sizeof(struct lttng_consumer_stream *));
3bd1e081 2457 if (local_stream == NULL) {
7a57cf92 2458 PERROR("local_stream malloc");
3bd1e081
MD
2459 pthread_mutex_unlock(&consumer_data.lock);
2460 goto end;
2461 }
ffe60014 2462 ret = update_poll_array(ctx, &pollfd, local_stream,
43c34bc3 2463 data_ht);
3bd1e081
MD
2464 if (ret < 0) {
2465 ERR("Error in allocating pollfd or local_outfds");
f73fabfd 2466 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2467 pthread_mutex_unlock(&consumer_data.lock);
2468 goto end;
2469 }
2470 nb_fd = ret;
2471 consumer_data.need_update = 0;
2472 }
2473 pthread_mutex_unlock(&consumer_data.lock);
2474
4078b776
MD
2475 /* No FDs and consumer_quit, consumer_cleanup the thread */
2476 if (nb_fd == 0 && consumer_quit == 1) {
1fc79fb4 2477 err = 0; /* All is OK */
4078b776
MD
2478 goto end;
2479 }
3bd1e081 2480 /* poll on the array of fds */
88f2b785 2481 restart:
02b3d176 2482 DBG("polling on %d fd", nb_fd + 2);
9ce5646a 2483 health_poll_entry();
02b3d176 2484 num_rdy = poll(pollfd, nb_fd + 2, -1);
9ce5646a 2485 health_poll_exit();
3bd1e081
MD
2486 DBG("poll num_rdy : %d", num_rdy);
2487 if (num_rdy == -1) {
88f2b785
MD
2488 /*
2489 * Restart interrupted system call.
2490 */
2491 if (errno == EINTR) {
2492 goto restart;
2493 }
7a57cf92 2494 PERROR("Poll error");
f73fabfd 2495 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2496 goto end;
2497 } else if (num_rdy == 0) {
2498 DBG("Polling thread timed out");
2499 goto end;
2500 }
2501
3bd1e081 2502 /*
50f8ae69 2503 * If the consumer_data_pipe triggered poll go directly to the
00e2e675
DG
2504 * beginning of the loop to update the array. We want to prioritize
2505 * array update over low-priority reads.
3bd1e081 2506 */
509bb1cf 2507 if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) {
ab30f567 2508 ssize_t pipe_readlen;
04fdd819 2509
50f8ae69 2510 DBG("consumer_data_pipe wake up");
acdb9057
DG
2511 pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe,
2512 &new_stream, sizeof(new_stream));
6cd525e8
MD
2513 if (pipe_readlen < sizeof(new_stream)) {
2514 PERROR("Consumer data pipe");
23f5f35d
DG
2515 /* Continue so we can at least handle the current stream(s). */
2516 continue;
2517 }
c869f647
DG
2518
2519 /*
2520 * If the stream is NULL, just ignore it. It's also possible that
2521 * the sessiond poll thread changed the consumer_quit state and is
2522 * waking us up to test it.
2523 */
2524 if (new_stream == NULL) {
8994307f 2525 validate_endpoint_status_data_stream();
c869f647
DG
2526 continue;
2527 }
2528
c869f647 2529 /* Continue to update the local streams and handle prio ones */
3bd1e081
MD
2530 continue;
2531 }
2532
02b3d176
DG
2533 /* Handle wakeup pipe. */
2534 if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) {
2535 char dummy;
2536 ssize_t pipe_readlen;
2537
2538 pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy,
2539 sizeof(dummy));
2540 if (pipe_readlen < 0) {
2541 PERROR("Consumer data wakeup pipe");
2542 }
2543 /* We've been awakened to handle stream(s). */
2544 ctx->has_wakeup = 0;
2545 }
2546
3bd1e081
MD
2547 /* Take care of high priority channels first. */
2548 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2549 health_code_update();
2550
9617607b
DG
2551 if (local_stream[i] == NULL) {
2552 continue;
2553 }
fb3a43a9 2554 if (pollfd[i].revents & POLLPRI) {
d41f73b7
MD
2555 DBG("Urgent read on fd %d", pollfd[i].fd);
2556 high_prio = 1;
4078b776 2557 len = ctx->on_buffer_ready(local_stream[i], ctx);
d41f73b7 2558 /* it's ok to have an unavailable sub-buffer */
b64403e3 2559 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2560 /* Clean the stream and free it. */
2561 consumer_del_stream(local_stream[i], data_ht);
9617607b 2562 local_stream[i] = NULL;
4078b776
MD
2563 } else if (len > 0) {
2564 local_stream[i]->data_read = 1;
d41f73b7 2565 }
3bd1e081
MD
2566 }
2567 }
2568
4078b776
MD
2569 /*
2570 * If we read high prio channel in this loop, try again
2571 * for more high prio data.
2572 */
2573 if (high_prio) {
3bd1e081
MD
2574 continue;
2575 }
2576
2577 /* Take care of low priority channels. */
4078b776 2578 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2579 health_code_update();
2580
9617607b
DG
2581 if (local_stream[i] == NULL) {
2582 continue;
2583 }
4078b776 2584 if ((pollfd[i].revents & POLLIN) ||
02b3d176
DG
2585 local_stream[i]->hangup_flush_done ||
2586 local_stream[i]->has_data) {
4078b776
MD
2587 DBG("Normal read on fd %d", pollfd[i].fd);
2588 len = ctx->on_buffer_ready(local_stream[i], ctx);
2589 /* it's ok to have an unavailable sub-buffer */
b64403e3 2590 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2591 /* Clean the stream and free it. */
2592 consumer_del_stream(local_stream[i], data_ht);
9617607b 2593 local_stream[i] = NULL;
4078b776
MD
2594 } else if (len > 0) {
2595 local_stream[i]->data_read = 1;
2596 }
2597 }
2598 }
2599
2600 /* Handle hangup and errors */
2601 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2602 health_code_update();
2603
9617607b
DG
2604 if (local_stream[i] == NULL) {
2605 continue;
2606 }
4078b776
MD
2607 if (!local_stream[i]->hangup_flush_done
2608 && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL))
2609 && (consumer_data.type == LTTNG_CONSUMER32_UST
2610 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2611 DBG("fd %d is hup|err|nval. Attempting flush and read.",
9617607b 2612 pollfd[i].fd);
4078b776
MD
2613 lttng_ustconsumer_on_stream_hangup(local_stream[i]);
2614 /* Attempt read again, for the data we just flushed. */
2615 local_stream[i]->data_read = 1;
2616 }
2617 /*
2618 * If the poll flag is HUP/ERR/NVAL and we have
2619 * read no data in this pass, we can remove the
2620 * stream from its hash table.
2621 */
2622 if ((pollfd[i].revents & POLLHUP)) {
2623 DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
2624 if (!local_stream[i]->data_read) {
43c34bc3 2625 consumer_del_stream(local_stream[i], data_ht);
9617607b 2626 local_stream[i] = NULL;
4078b776
MD
2627 num_hup++;
2628 }
2629 } else if (pollfd[i].revents & POLLERR) {
2630 ERR("Error returned in polling fd %d.", pollfd[i].fd);
2631 if (!local_stream[i]->data_read) {
43c34bc3 2632 consumer_del_stream(local_stream[i], data_ht);
9617607b 2633 local_stream[i] = NULL;
4078b776
MD
2634 num_hup++;
2635 }
2636 } else if (pollfd[i].revents & POLLNVAL) {
2637 ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
2638 if (!local_stream[i]->data_read) {
43c34bc3 2639 consumer_del_stream(local_stream[i], data_ht);
9617607b 2640 local_stream[i] = NULL;
4078b776 2641 num_hup++;
3bd1e081
MD
2642 }
2643 }
9617607b
DG
2644 if (local_stream[i] != NULL) {
2645 local_stream[i]->data_read = 0;
2646 }
3bd1e081
MD
2647 }
2648 }
1fc79fb4
MD
2649 /* All is OK */
2650 err = 0;
3bd1e081
MD
2651end:
2652 DBG("polling thread exiting");
0e428499
DG
2653 free(pollfd);
2654 free(local_stream);
fb3a43a9
DG
2655
2656 /*
2657 * Close the write side of the pipe so epoll_wait() in
7d980def
DG
2658 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2659 * read side of the pipe. If we close them both, epoll_wait strangely does
2660 * not return and could create a endless wait period if the pipe is the
2661 * only tracked fd in the poll set. The thread will take care of closing
2662 * the read side.
fb3a43a9 2663 */
13886d2d 2664 (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe);
fb3a43a9 2665
2d57de81 2666error_testpoint:
1fc79fb4
MD
2667 if (err) {
2668 health_error();
2669 ERR("Health error occurred in %s", __func__);
2670 }
2671 health_unregister(health_consumerd);
2672
e7b994a3 2673 rcu_unregister_thread();
3bd1e081
MD
2674 return NULL;
2675}
2676
d8ef542d
MD
2677/*
2678 * Close wake-up end of each stream belonging to the channel. This will
2679 * allow the poll() on the stream read-side to detect when the
2680 * write-side (application) finally closes them.
2681 */
2682static
2683void consumer_close_channel_streams(struct lttng_consumer_channel *channel)
2684{
2685 struct lttng_ht *ht;
2686 struct lttng_consumer_stream *stream;
2687 struct lttng_ht_iter iter;
2688
2689 ht = consumer_data.stream_per_chan_id_ht;
2690
2691 rcu_read_lock();
2692 cds_lfht_for_each_entry_duplicate(ht->ht,
2693 ht->hash_fct(&channel->key, lttng_ht_seed),
2694 ht->match_fct, &channel->key,
2695 &iter.iter, stream, node_channel_id.node) {
f2ad556d
MD
2696 /*
2697 * Protect against teardown with mutex.
2698 */
2699 pthread_mutex_lock(&stream->lock);
2700 if (cds_lfht_is_node_deleted(&stream->node.node)) {
2701 goto next;
2702 }
d8ef542d
MD
2703 switch (consumer_data.type) {
2704 case LTTNG_CONSUMER_KERNEL:
2705 break;
2706 case LTTNG_CONSUMER32_UST:
2707 case LTTNG_CONSUMER64_UST:
b4a650f3
DG
2708 if (stream->metadata_flag) {
2709 /* Safe and protected by the stream lock. */
2710 lttng_ustconsumer_close_metadata(stream->chan);
2711 } else {
2712 /*
2713 * Note: a mutex is taken internally within
2714 * liblttng-ust-ctl to protect timer wakeup_fd
2715 * use from concurrent close.
2716 */
2717 lttng_ustconsumer_close_stream_wakeup(stream);
2718 }
d8ef542d
MD
2719 break;
2720 default:
2721 ERR("Unknown consumer_data type");
2722 assert(0);
2723 }
f2ad556d
MD
2724 next:
2725 pthread_mutex_unlock(&stream->lock);
d8ef542d
MD
2726 }
2727 rcu_read_unlock();
2728}
2729
2730static void destroy_channel_ht(struct lttng_ht *ht)
2731{
2732 struct lttng_ht_iter iter;
2733 struct lttng_consumer_channel *channel;
2734 int ret;
2735
2736 if (ht == NULL) {
2737 return;
2738 }
2739
2740 rcu_read_lock();
2741 cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) {
2742 ret = lttng_ht_del(ht, &iter);
2743 assert(ret != 0);
2744 }
2745 rcu_read_unlock();
2746
2747 lttng_ht_destroy(ht);
2748}
2749
2750/*
2751 * This thread polls the channel fds to detect when they are being
2752 * closed. It closes all related streams if the channel is detected as
2753 * closed. It is currently only used as a shim layer for UST because the
2754 * consumerd needs to keep the per-stream wakeup end of pipes open for
2755 * periodical flush.
2756 */
2757void *consumer_thread_channel_poll(void *data)
2758{
1fc79fb4 2759 int ret, i, pollfd, err = -1;
d8ef542d
MD
2760 uint32_t revents, nb_fd;
2761 struct lttng_consumer_channel *chan = NULL;
2762 struct lttng_ht_iter iter;
2763 struct lttng_ht_node_u64 *node;
2764 struct lttng_poll_event events;
2765 struct lttng_consumer_local_data *ctx = data;
2766 struct lttng_ht *channel_ht;
2767
2768 rcu_register_thread();
2769
1fc79fb4
MD
2770 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL);
2771
2d57de81
MD
2772 if (testpoint(consumerd_thread_channel)) {
2773 goto error_testpoint;
2774 }
2775
9ce5646a
MD
2776 health_code_update();
2777
d8ef542d
MD
2778 channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2779 if (!channel_ht) {
2780 /* ENOMEM at this point. Better to bail out. */
2781 goto end_ht;
2782 }
2783
2784 DBG("Thread channel poll started");
2785
2786 /* Size is set to 1 for the consumer_channel pipe */
2787 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2788 if (ret < 0) {
2789 ERR("Poll set creation failed");
2790 goto end_poll;
2791 }
2792
2793 ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN);
2794 if (ret < 0) {
2795 goto end;
2796 }
2797
2798 /* Main loop */
2799 DBG("Channel main loop started");
2800
2801 while (1) {
d8ef542d 2802restart:
7fa2082e
MD
2803 health_code_update();
2804 DBG("Channel poll wait");
9ce5646a 2805 health_poll_entry();
d8ef542d 2806 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2807 DBG("Channel poll return from wait with %d fd(s)",
2808 LTTNG_POLL_GETNB(&events));
9ce5646a 2809 health_poll_exit();
d8ef542d
MD
2810 DBG("Channel event catched in thread");
2811 if (ret < 0) {
2812 if (errno == EINTR) {
2813 ERR("Poll EINTR catched");
2814 goto restart;
2815 }
d9607cd7
MD
2816 if (LTTNG_POLL_GETNB(&events) == 0) {
2817 err = 0; /* All is OK */
2818 }
d8ef542d
MD
2819 goto end;
2820 }
2821
2822 nb_fd = ret;
2823
2824 /* From here, the event is a channel wait fd */
2825 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2826 health_code_update();
2827
d8ef542d
MD
2828 revents = LTTNG_POLL_GETEV(&events, i);
2829 pollfd = LTTNG_POLL_GETFD(&events, i);
2830
d8ef542d 2831 if (!revents) {
fd20dac9 2832 /* No activity for this FD (poll implementation). */
d8ef542d
MD
2833 continue;
2834 }
fd20dac9 2835
d8ef542d 2836 if (pollfd == ctx->consumer_channel_pipe[0]) {
03e43155 2837 if (revents & LPOLLIN) {
d8ef542d 2838 enum consumer_channel_action action;
a0cbdd2e 2839 uint64_t key;
d8ef542d 2840
a0cbdd2e 2841 ret = read_channel_pipe(ctx, &chan, &key, &action);
d8ef542d 2842 if (ret <= 0) {
03e43155
MD
2843 if (ret < 0) {
2844 ERR("Error reading channel pipe");
2845 }
2846 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
d8ef542d
MD
2847 continue;
2848 }
2849
2850 switch (action) {
2851 case CONSUMER_CHANNEL_ADD:
2852 DBG("Adding channel %d to poll set",
2853 chan->wait_fd);
2854
2855 lttng_ht_node_init_u64(&chan->wait_fd_node,
2856 chan->wait_fd);
c7260a81 2857 rcu_read_lock();
d8ef542d
MD
2858 lttng_ht_add_unique_u64(channel_ht,
2859 &chan->wait_fd_node);
c7260a81 2860 rcu_read_unlock();
d8ef542d
MD
2861 /* Add channel to the global poll events list */
2862 lttng_poll_add(&events, chan->wait_fd,
03e43155 2863 LPOLLERR | LPOLLHUP);
d8ef542d 2864 break;
a0cbdd2e
MD
2865 case CONSUMER_CHANNEL_DEL:
2866 {
b4a650f3
DG
2867 /*
2868 * This command should never be called if the channel
2869 * has streams monitored by either the data or metadata
2870 * thread. The consumer only notify this thread with a
2871 * channel del. command if it receives a destroy
2872 * channel command from the session daemon that send it
2873 * if a command prior to the GET_CHANNEL failed.
2874 */
2875
c7260a81 2876 rcu_read_lock();
a0cbdd2e
MD
2877 chan = consumer_find_channel(key);
2878 if (!chan) {
c7260a81 2879 rcu_read_unlock();
a0cbdd2e
MD
2880 ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key);
2881 break;
2882 }
2883 lttng_poll_del(&events, chan->wait_fd);
f623cc0b 2884 iter.iter.node = &chan->wait_fd_node.node;
a0cbdd2e
MD
2885 ret = lttng_ht_del(channel_ht, &iter);
2886 assert(ret == 0);
a0cbdd2e 2887
f2a444f1
DG
2888 switch (consumer_data.type) {
2889 case LTTNG_CONSUMER_KERNEL:
2890 break;
2891 case LTTNG_CONSUMER32_UST:
2892 case LTTNG_CONSUMER64_UST:
212d67a2
DG
2893 health_code_update();
2894 /* Destroy streams that might have been left in the stream list. */
2895 clean_channel_stream_list(chan);
f2a444f1
DG
2896 break;
2897 default:
2898 ERR("Unknown consumer_data type");
2899 assert(0);
2900 }
2901
a0cbdd2e
MD
2902 /*
2903 * Release our own refcount. Force channel deletion even if
2904 * streams were not initialized.
2905 */
2906 if (!uatomic_sub_return(&chan->refcount, 1)) {
2907 consumer_del_channel(chan);
2908 }
c7260a81 2909 rcu_read_unlock();
a0cbdd2e
MD
2910 goto restart;
2911 }
d8ef542d
MD
2912 case CONSUMER_CHANNEL_QUIT:
2913 /*
2914 * Remove the pipe from the poll set and continue the loop
2915 * since their might be data to consume.
2916 */
2917 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2918 continue;
2919 default:
2920 ERR("Unknown action");
2921 break;
2922 }
03e43155
MD
2923 } else if (revents & (LPOLLERR | LPOLLHUP)) {
2924 DBG("Channel thread pipe hung up");
2925 /*
2926 * Remove the pipe from the poll set and continue the loop
2927 * since their might be data to consume.
2928 */
2929 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2930 continue;
2931 } else {
2932 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2933 goto end;
d8ef542d
MD
2934 }
2935
2936 /* Handle other stream */
2937 continue;
2938 }
2939
2940 rcu_read_lock();
2941 {
2942 uint64_t tmp_id = (uint64_t) pollfd;
2943
2944 lttng_ht_lookup(channel_ht, &tmp_id, &iter);
2945 }
2946 node = lttng_ht_iter_get_node_u64(&iter);
2947 assert(node);
2948
2949 chan = caa_container_of(node, struct lttng_consumer_channel,
2950 wait_fd_node);
2951
2952 /* Check for error event */
2953 if (revents & (LPOLLERR | LPOLLHUP)) {
2954 DBG("Channel fd %d is hup|err.", pollfd);
2955
2956 lttng_poll_del(&events, chan->wait_fd);
2957 ret = lttng_ht_del(channel_ht, &iter);
2958 assert(ret == 0);
b4a650f3
DG
2959
2960 /*
2961 * This will close the wait fd for each stream associated to
2962 * this channel AND monitored by the data/metadata thread thus
2963 * will be clean by the right thread.
2964 */
d8ef542d 2965 consumer_close_channel_streams(chan);
f2ad556d
MD
2966
2967 /* Release our own refcount */
2968 if (!uatomic_sub_return(&chan->refcount, 1)
2969 && !uatomic_read(&chan->nb_init_stream_left)) {
2970 consumer_del_channel(chan);
2971 }
03e43155
MD
2972 } else {
2973 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2974 rcu_read_unlock();
2975 goto end;
d8ef542d
MD
2976 }
2977
2978 /* Release RCU lock for the channel looked up */
2979 rcu_read_unlock();
2980 }
2981 }
2982
1fc79fb4
MD
2983 /* All is OK */
2984 err = 0;
d8ef542d
MD
2985end:
2986 lttng_poll_clean(&events);
2987end_poll:
2988 destroy_channel_ht(channel_ht);
2989end_ht:
2d57de81 2990error_testpoint:
d8ef542d 2991 DBG("Channel poll thread exiting");
1fc79fb4
MD
2992 if (err) {
2993 health_error();
2994 ERR("Health error occurred in %s", __func__);
2995 }
2996 health_unregister(health_consumerd);
d8ef542d
MD
2997 rcu_unregister_thread();
2998 return NULL;
2999}
3000
331744e3
JD
3001static int set_metadata_socket(struct lttng_consumer_local_data *ctx,
3002 struct pollfd *sockpoll, int client_socket)
3003{
3004 int ret;
3005
3006 assert(ctx);
3007 assert(sockpoll);
3008
84382d49
MD
3009 ret = lttng_consumer_poll_socket(sockpoll);
3010 if (ret) {
331744e3
JD
3011 goto error;
3012 }
3013 DBG("Metadata connection on client_socket");
3014
3015 /* Blocking call, waiting for transmission */
3016 ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket);
3017 if (ctx->consumer_metadata_socket < 0) {
3018 WARN("On accept metadata");
3019 ret = -1;
3020 goto error;
3021 }
3022 ret = 0;
3023
3024error:
3025 return ret;
3026}
3027
3bd1e081
MD
3028/*
3029 * This thread listens on the consumerd socket and receives the file
3030 * descriptors from the session daemon.
3031 */
7d980def 3032void *consumer_thread_sessiond_poll(void *data)
3bd1e081 3033{
1fc79fb4 3034 int sock = -1, client_socket, ret, err = -1;
3bd1e081
MD
3035 /*
3036 * structure to poll for incoming data on communication socket avoids
3037 * making blocking sockets.
3038 */
3039 struct pollfd consumer_sockpoll[2];
3040 struct lttng_consumer_local_data *ctx = data;
3041
e7b994a3
DG
3042 rcu_register_thread();
3043
1fc79fb4
MD
3044 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND);
3045
2d57de81
MD
3046 if (testpoint(consumerd_thread_sessiond)) {
3047 goto error_testpoint;
3048 }
3049
9ce5646a
MD
3050 health_code_update();
3051
3bd1e081
MD
3052 DBG("Creating command socket %s", ctx->consumer_command_sock_path);
3053 unlink(ctx->consumer_command_sock_path);
3054 client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path);
3055 if (client_socket < 0) {
3056 ERR("Cannot create command socket");
3057 goto end;
3058 }
3059
3060 ret = lttcomm_listen_unix_sock(client_socket);
3061 if (ret < 0) {
3062 goto end;
3063 }
3064
32258573 3065 DBG("Sending ready command to lttng-sessiond");
f73fabfd 3066 ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY);
3bd1e081
MD
3067 /* return < 0 on error, but == 0 is not fatal */
3068 if (ret < 0) {
32258573 3069 ERR("Error sending ready command to lttng-sessiond");
3bd1e081
MD
3070 goto end;
3071 }
3072
3bd1e081
MD
3073 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3074 consumer_sockpoll[0].fd = ctx->consumer_should_quit[0];
3075 consumer_sockpoll[0].events = POLLIN | POLLPRI;
3076 consumer_sockpoll[1].fd = client_socket;
3077 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3078
84382d49
MD
3079 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3080 if (ret) {
3081 if (ret > 0) {
3082 /* should exit */
3083 err = 0;
3084 }
3bd1e081
MD
3085 goto end;
3086 }
3087 DBG("Connection on client_socket");
3088
3089 /* Blocking call, waiting for transmission */
3090 sock = lttcomm_accept_unix_sock(client_socket);
534d2592 3091 if (sock < 0) {
3bd1e081
MD
3092 WARN("On accept");
3093 goto end;
3094 }
3bd1e081 3095
331744e3
JD
3096 /*
3097 * Setup metadata socket which is the second socket connection on the
3098 * command unix socket.
3099 */
3100 ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket);
84382d49
MD
3101 if (ret) {
3102 if (ret > 0) {
3103 /* should exit */
3104 err = 0;
3105 }
331744e3
JD
3106 goto end;
3107 }
3108
d96f09c6
DG
3109 /* This socket is not useful anymore. */
3110 ret = close(client_socket);
3111 if (ret < 0) {
3112 PERROR("close client_socket");
3113 }
3114 client_socket = -1;
3115
3bd1e081
MD
3116 /* update the polling structure to poll on the established socket */
3117 consumer_sockpoll[1].fd = sock;
3118 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3119
3120 while (1) {
9ce5646a
MD
3121 health_code_update();
3122
3123 health_poll_entry();
3124 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3125 health_poll_exit();
84382d49
MD
3126 if (ret) {
3127 if (ret > 0) {
3128 /* should exit */
3129 err = 0;
3130 }
3bd1e081
MD
3131 goto end;
3132 }
3133 DBG("Incoming command on sock");
3134 ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll);
4cbc1a04
DG
3135 if (ret <= 0) {
3136 /*
3137 * This could simply be a session daemon quitting. Don't output
3138 * ERR() here.
3139 */
3140 DBG("Communication interrupted on command socket");
41ba6035 3141 err = 0;
3bd1e081
MD
3142 goto end;
3143 }
3144 if (consumer_quit) {
3145 DBG("consumer_thread_receive_fds received quit from signal");
1fc79fb4 3146 err = 0; /* All is OK */
3bd1e081
MD
3147 goto end;
3148 }
ffe60014 3149 DBG("received command on sock");
3bd1e081 3150 }
1fc79fb4
MD
3151 /* All is OK */
3152 err = 0;
3153
3bd1e081 3154end:
ffe60014 3155 DBG("Consumer thread sessiond poll exiting");
3bd1e081 3156
d88aee68
DG
3157 /*
3158 * Close metadata streams since the producer is the session daemon which
3159 * just died.
3160 *
3161 * NOTE: for now, this only applies to the UST tracer.
3162 */
6d574024 3163 lttng_consumer_close_all_metadata();
d88aee68 3164
3bd1e081
MD
3165 /*
3166 * when all fds have hung up, the polling thread
3167 * can exit cleanly
3168 */
3169 consumer_quit = 1;
3170
04fdd819 3171 /*
c869f647 3172 * Notify the data poll thread to poll back again and test the
8994307f 3173 * consumer_quit state that we just set so to quit gracefully.
04fdd819 3174 */
acdb9057 3175 notify_thread_lttng_pipe(ctx->consumer_data_pipe);
c869f647 3176
a0cbdd2e 3177 notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT);
d8ef542d 3178
5c635c72
MD
3179 notify_health_quit_pipe(health_quit_pipe);
3180
d96f09c6
DG
3181 /* Cleaning up possibly open sockets. */
3182 if (sock >= 0) {
3183 ret = close(sock);
3184 if (ret < 0) {
3185 PERROR("close sock sessiond poll");
3186 }
3187 }
3188 if (client_socket >= 0) {
38476d24 3189 ret = close(client_socket);
d96f09c6
DG
3190 if (ret < 0) {
3191 PERROR("close client_socket sessiond poll");
3192 }
3193 }
3194
2d57de81 3195error_testpoint:
1fc79fb4
MD
3196 if (err) {
3197 health_error();
3198 ERR("Health error occurred in %s", __func__);
3199 }
3200 health_unregister(health_consumerd);
3201
e7b994a3 3202 rcu_unregister_thread();
3bd1e081
MD
3203 return NULL;
3204}
d41f73b7 3205
4078b776 3206ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream,
d41f73b7
MD
3207 struct lttng_consumer_local_data *ctx)
3208{
74251bb8
DG
3209 ssize_t ret;
3210
3211 pthread_mutex_lock(&stream->lock);
94d49140
JD
3212 if (stream->metadata_flag) {
3213 pthread_mutex_lock(&stream->metadata_rdv_lock);
3214 }
74251bb8 3215
d41f73b7
MD
3216 switch (consumer_data.type) {
3217 case LTTNG_CONSUMER_KERNEL:
74251bb8
DG
3218 ret = lttng_kconsumer_read_subbuffer(stream, ctx);
3219 break;
7753dea8
MD
3220 case LTTNG_CONSUMER32_UST:
3221 case LTTNG_CONSUMER64_UST:
74251bb8
DG
3222 ret = lttng_ustconsumer_read_subbuffer(stream, ctx);
3223 break;
d41f73b7
MD
3224 default:
3225 ERR("Unknown consumer_data type");
3226 assert(0);
74251bb8
DG
3227 ret = -ENOSYS;
3228 break;
d41f73b7 3229 }
74251bb8 3230
94d49140
JD
3231 if (stream->metadata_flag) {
3232 pthread_cond_broadcast(&stream->metadata_rdv);
3233 pthread_mutex_unlock(&stream->metadata_rdv_lock);
3234 }
74251bb8
DG
3235 pthread_mutex_unlock(&stream->lock);
3236 return ret;
d41f73b7
MD
3237}
3238
3239int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream)
3240{
3241 switch (consumer_data.type) {
3242 case LTTNG_CONSUMER_KERNEL:
3243 return lttng_kconsumer_on_recv_stream(stream);
7753dea8
MD
3244 case LTTNG_CONSUMER32_UST:
3245 case LTTNG_CONSUMER64_UST:
d41f73b7
MD
3246 return lttng_ustconsumer_on_recv_stream(stream);
3247 default:
3248 ERR("Unknown consumer_data type");
3249 assert(0);
3250 return -ENOSYS;
3251 }
3252}
e4421fec
DG
3253
3254/*
3255 * Allocate and set consumer data hash tables.
3256 */
282dadbc 3257int l