Fix: double RCU unlock on event_agent_disable_all
[lttng-tools.git] / src / common / 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
20#define _GNU_SOURCE
6c1c0768 21#define _LGPL_SOURCE
3bd1e081 22#include <assert.h>
3bd1e081
MD
23#include <poll.h>
24#include <pthread.h>
25#include <stdlib.h>
26#include <string.h>
27#include <sys/mman.h>
28#include <sys/socket.h>
29#include <sys/types.h>
30#include <unistd.h>
77c7c900 31#include <inttypes.h>
331744e3 32#include <signal.h>
3bd1e081 33
51a9e1c7 34#include <bin/lttng-consumerd/health-consumerd.h>
990570ed 35#include <common/common.h>
fb3a43a9
DG
36#include <common/utils.h>
37#include <common/compat/poll.h>
f263b7fd 38#include <common/compat/endian.h>
309167d2 39#include <common/index/index.h>
10a8a223 40#include <common/kernel-ctl/kernel-ctl.h>
00e2e675 41#include <common/sessiond-comm/relayd.h>
10a8a223
DG
42#include <common/sessiond-comm/sessiond-comm.h>
43#include <common/kernel-consumer/kernel-consumer.h>
00e2e675 44#include <common/relayd/relayd.h>
10a8a223 45#include <common/ust-consumer/ust-consumer.h>
d3e2ba59 46#include <common/consumer-timer.h>
10a8a223
DG
47
48#include "consumer.h"
1d1a276c 49#include "consumer-stream.h"
2d57de81 50#include "consumer-testpoint.h"
d07ceecd 51#include "align.h"
3bd1e081
MD
52
53struct lttng_consumer_global_data consumer_data = {
3bd1e081
MD
54 .stream_count = 0,
55 .need_update = 1,
56 .type = LTTNG_CONSUMER_UNKNOWN,
57};
58
d8ef542d
MD
59enum consumer_channel_action {
60 CONSUMER_CHANNEL_ADD,
a0cbdd2e 61 CONSUMER_CHANNEL_DEL,
d8ef542d
MD
62 CONSUMER_CHANNEL_QUIT,
63};
64
65struct consumer_channel_msg {
66 enum consumer_channel_action action;
a0cbdd2e
MD
67 struct lttng_consumer_channel *chan; /* add */
68 uint64_t key; /* del */
d8ef542d
MD
69};
70
3bd1e081
MD
71/*
72 * Flag to inform the polling thread to quit when all fd hung up. Updated by
73 * the consumer_thread_receive_fds when it notices that all fds has hung up.
74 * Also updated by the signal handler (consumer_should_exit()). Read by the
75 * polling threads.
76 */
a98dae5f 77volatile int consumer_quit;
3bd1e081 78
43c34bc3 79/*
43c34bc3
DG
80 * Global hash table containing respectively metadata and data streams. The
81 * stream element in this ht should only be updated by the metadata poll thread
82 * for the metadata and the data poll thread for the data.
83 */
40dc48e0
DG
84static struct lttng_ht *metadata_ht;
85static struct lttng_ht *data_ht;
43c34bc3 86
acdb9057
DG
87/*
88 * Notify a thread lttng pipe to poll back again. This usually means that some
89 * global state has changed so we just send back the thread in a poll wait
90 * call.
91 */
92static void notify_thread_lttng_pipe(struct lttng_pipe *pipe)
93{
94 struct lttng_consumer_stream *null_stream = NULL;
95
96 assert(pipe);
97
98 (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream));
99}
100
5c635c72
MD
101static void notify_health_quit_pipe(int *pipe)
102{
6cd525e8 103 ssize_t ret;
5c635c72 104
6cd525e8
MD
105 ret = lttng_write(pipe[1], "4", 1);
106 if (ret < 1) {
5c635c72
MD
107 PERROR("write consumer health quit");
108 }
109}
110
d8ef542d
MD
111static void notify_channel_pipe(struct lttng_consumer_local_data *ctx,
112 struct lttng_consumer_channel *chan,
a0cbdd2e 113 uint64_t key,
d8ef542d
MD
114 enum consumer_channel_action action)
115{
116 struct consumer_channel_msg msg;
6cd525e8 117 ssize_t ret;
d8ef542d 118
e56251fc
DG
119 memset(&msg, 0, sizeof(msg));
120
d8ef542d
MD
121 msg.action = action;
122 msg.chan = chan;
f21dae48 123 msg.key = key;
6cd525e8
MD
124 ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg));
125 if (ret < sizeof(msg)) {
126 PERROR("notify_channel_pipe write error");
127 }
d8ef542d
MD
128}
129
a0cbdd2e
MD
130void notify_thread_del_channel(struct lttng_consumer_local_data *ctx,
131 uint64_t key)
132{
133 notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL);
134}
135
d8ef542d
MD
136static int read_channel_pipe(struct lttng_consumer_local_data *ctx,
137 struct lttng_consumer_channel **chan,
a0cbdd2e 138 uint64_t *key,
d8ef542d
MD
139 enum consumer_channel_action *action)
140{
141 struct consumer_channel_msg msg;
6cd525e8 142 ssize_t ret;
d8ef542d 143
6cd525e8
MD
144 ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg));
145 if (ret < sizeof(msg)) {
146 ret = -1;
147 goto error;
d8ef542d 148 }
6cd525e8
MD
149 *action = msg.action;
150 *chan = msg.chan;
151 *key = msg.key;
152error:
153 return (int) ret;
d8ef542d
MD
154}
155
212d67a2
DG
156/*
157 * Cleanup the stream list of a channel. Those streams are not yet globally
158 * visible
159 */
160static void clean_channel_stream_list(struct lttng_consumer_channel *channel)
161{
162 struct lttng_consumer_stream *stream, *stmp;
163
164 assert(channel);
165
166 /* Delete streams that might have been left in the stream list. */
167 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
168 send_node) {
169 cds_list_del(&stream->send_node);
170 /*
171 * Once a stream is added to this list, the buffers were created so we
172 * have a guarantee that this call will succeed. Setting the monitor
173 * mode to 0 so we don't lock nor try to delete the stream from the
174 * global hash table.
175 */
176 stream->monitor = 0;
177 consumer_stream_destroy(stream, NULL);
178 }
179}
180
3bd1e081
MD
181/*
182 * Find a stream. The consumer_data.lock must be locked during this
183 * call.
184 */
d88aee68 185static struct lttng_consumer_stream *find_stream(uint64_t key,
8389e4f8 186 struct lttng_ht *ht)
3bd1e081 187{
e4421fec 188 struct lttng_ht_iter iter;
d88aee68 189 struct lttng_ht_node_u64 *node;
e4421fec 190 struct lttng_consumer_stream *stream = NULL;
3bd1e081 191
8389e4f8
DG
192 assert(ht);
193
d88aee68
DG
194 /* -1ULL keys are lookup failures */
195 if (key == (uint64_t) -1ULL) {
7ad0a0cb 196 return NULL;
7a57cf92 197 }
e4421fec 198
6065ceec
DG
199 rcu_read_lock();
200
d88aee68
DG
201 lttng_ht_lookup(ht, &key, &iter);
202 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
203 if (node != NULL) {
204 stream = caa_container_of(node, struct lttng_consumer_stream, node);
3bd1e081 205 }
e4421fec 206
6065ceec
DG
207 rcu_read_unlock();
208
e4421fec 209 return stream;
3bd1e081
MD
210}
211
da009f2c 212static void steal_stream_key(uint64_t key, struct lttng_ht *ht)
7ad0a0cb
MD
213{
214 struct lttng_consumer_stream *stream;
215
04253271 216 rcu_read_lock();
ffe60014 217 stream = find_stream(key, ht);
04253271 218 if (stream) {
da009f2c 219 stream->key = (uint64_t) -1ULL;
04253271
MD
220 /*
221 * We don't want the lookup to match, but we still need
222 * to iterate on this stream when iterating over the hash table. Just
223 * change the node key.
224 */
da009f2c 225 stream->node.key = (uint64_t) -1ULL;
04253271
MD
226 }
227 rcu_read_unlock();
7ad0a0cb
MD
228}
229
d56db448
DG
230/*
231 * Return a channel object for the given key.
232 *
233 * RCU read side lock MUST be acquired before calling this function and
234 * protects the channel ptr.
235 */
d88aee68 236struct lttng_consumer_channel *consumer_find_channel(uint64_t key)
3bd1e081 237{
e4421fec 238 struct lttng_ht_iter iter;
d88aee68 239 struct lttng_ht_node_u64 *node;
e4421fec 240 struct lttng_consumer_channel *channel = NULL;
3bd1e081 241
d88aee68
DG
242 /* -1ULL keys are lookup failures */
243 if (key == (uint64_t) -1ULL) {
7ad0a0cb 244 return NULL;
7a57cf92 245 }
e4421fec 246
d88aee68
DG
247 lttng_ht_lookup(consumer_data.channel_ht, &key, &iter);
248 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
249 if (node != NULL) {
250 channel = caa_container_of(node, struct lttng_consumer_channel, node);
3bd1e081 251 }
e4421fec
DG
252
253 return channel;
3bd1e081
MD
254}
255
b5a6470f
DG
256/*
257 * There is a possibility that the consumer does not have enough time between
258 * the close of the channel on the session daemon and the cleanup in here thus
259 * once we have a channel add with an existing key, we know for sure that this
260 * channel will eventually get cleaned up by all streams being closed.
261 *
262 * This function just nullifies the already existing channel key.
263 */
264static void steal_channel_key(uint64_t key)
265{
266 struct lttng_consumer_channel *channel;
267
268 rcu_read_lock();
269 channel = consumer_find_channel(key);
270 if (channel) {
271 channel->key = (uint64_t) -1ULL;
272 /*
273 * We don't want the lookup to match, but we still need to iterate on
274 * this channel when iterating over the hash table. Just change the
275 * node key.
276 */
277 channel->node.key = (uint64_t) -1ULL;
278 }
279 rcu_read_unlock();
280}
281
ffe60014 282static void free_channel_rcu(struct rcu_head *head)
702b1ea4 283{
d88aee68
DG
284 struct lttng_ht_node_u64 *node =
285 caa_container_of(head, struct lttng_ht_node_u64, head);
ffe60014
DG
286 struct lttng_consumer_channel *channel =
287 caa_container_of(node, struct lttng_consumer_channel, node);
702b1ea4 288
b83e03c4
MD
289 switch (consumer_data.type) {
290 case LTTNG_CONSUMER_KERNEL:
291 break;
292 case LTTNG_CONSUMER32_UST:
293 case LTTNG_CONSUMER64_UST:
294 lttng_ustconsumer_free_channel(channel);
295 break;
296 default:
297 ERR("Unknown consumer_data type");
298 abort();
299 }
ffe60014 300 free(channel);
702b1ea4
MD
301}
302
00e2e675
DG
303/*
304 * RCU protected relayd socket pair free.
305 */
ffe60014 306static void free_relayd_rcu(struct rcu_head *head)
00e2e675 307{
d88aee68
DG
308 struct lttng_ht_node_u64 *node =
309 caa_container_of(head, struct lttng_ht_node_u64, head);
00e2e675
DG
310 struct consumer_relayd_sock_pair *relayd =
311 caa_container_of(node, struct consumer_relayd_sock_pair, node);
312
8994307f
DG
313 /*
314 * Close all sockets. This is done in the call RCU since we don't want the
315 * socket fds to be reassigned thus potentially creating bad state of the
316 * relayd object.
317 *
318 * We do not have to lock the control socket mutex here since at this stage
319 * there is no one referencing to this relayd object.
320 */
321 (void) relayd_close(&relayd->control_sock);
322 (void) relayd_close(&relayd->data_sock);
323
00e2e675
DG
324 free(relayd);
325}
326
327/*
328 * Destroy and free relayd socket pair object.
00e2e675 329 */
51230d70 330void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
331{
332 int ret;
333 struct lttng_ht_iter iter;
334
173af62f
DG
335 if (relayd == NULL) {
336 return;
337 }
338
00e2e675
DG
339 DBG("Consumer destroy and close relayd socket pair");
340
341 iter.iter.node = &relayd->node.node;
342 ret = lttng_ht_del(consumer_data.relayd_ht, &iter);
173af62f 343 if (ret != 0) {
8994307f 344 /* We assume the relayd is being or is destroyed */
173af62f
DG
345 return;
346 }
00e2e675 347
00e2e675 348 /* RCU free() call */
ffe60014
DG
349 call_rcu(&relayd->node.head, free_relayd_rcu);
350}
351
352/*
353 * Remove a channel from the global list protected by a mutex. This function is
354 * also responsible for freeing its data structures.
355 */
356void consumer_del_channel(struct lttng_consumer_channel *channel)
357{
358 int ret;
359 struct lttng_ht_iter iter;
360
d88aee68 361 DBG("Consumer delete channel key %" PRIu64, channel->key);
ffe60014
DG
362
363 pthread_mutex_lock(&consumer_data.lock);
a9838785 364 pthread_mutex_lock(&channel->lock);
ffe60014 365
212d67a2
DG
366 /* Destroy streams that might have been left in the stream list. */
367 clean_channel_stream_list(channel);
51e762e5 368
d3e2ba59
JD
369 if (channel->live_timer_enabled == 1) {
370 consumer_timer_live_stop(channel);
371 }
372
ffe60014
DG
373 switch (consumer_data.type) {
374 case LTTNG_CONSUMER_KERNEL:
375 break;
376 case LTTNG_CONSUMER32_UST:
377 case LTTNG_CONSUMER64_UST:
378 lttng_ustconsumer_del_channel(channel);
379 break;
380 default:
381 ERR("Unknown consumer_data type");
382 assert(0);
383 goto end;
384 }
385
386 rcu_read_lock();
387 iter.iter.node = &channel->node.node;
388 ret = lttng_ht_del(consumer_data.channel_ht, &iter);
389 assert(!ret);
390 rcu_read_unlock();
391
392 call_rcu(&channel->node.head, free_channel_rcu);
393end:
a9838785 394 pthread_mutex_unlock(&channel->lock);
ffe60014 395 pthread_mutex_unlock(&consumer_data.lock);
00e2e675
DG
396}
397
228b5bf7
DG
398/*
399 * Iterate over the relayd hash table and destroy each element. Finally,
400 * destroy the whole hash table.
401 */
402static void cleanup_relayd_ht(void)
403{
404 struct lttng_ht_iter iter;
405 struct consumer_relayd_sock_pair *relayd;
406
407 rcu_read_lock();
408
409 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
410 node.node) {
51230d70 411 consumer_destroy_relayd(relayd);
228b5bf7
DG
412 }
413
228b5bf7 414 rcu_read_unlock();
36b588ed
MD
415
416 lttng_ht_destroy(consumer_data.relayd_ht);
228b5bf7
DG
417}
418
8994307f
DG
419/*
420 * Update the end point status of all streams having the given network sequence
421 * index (relayd index).
422 *
423 * It's atomically set without having the stream mutex locked which is fine
424 * because we handle the write/read race with a pipe wakeup for each thread.
425 */
da009f2c 426static void update_endpoint_status_by_netidx(uint64_t net_seq_idx,
8994307f
DG
427 enum consumer_endpoint_status status)
428{
429 struct lttng_ht_iter iter;
430 struct lttng_consumer_stream *stream;
431
da009f2c 432 DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx);
8994307f
DG
433
434 rcu_read_lock();
435
436 /* Let's begin with metadata */
437 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
438 if (stream->net_seq_idx == net_seq_idx) {
439 uatomic_set(&stream->endpoint_status, status);
440 DBG("Delete flag set to metadata stream %d", stream->wait_fd);
441 }
442 }
443
444 /* Follow up by the data streams */
445 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
446 if (stream->net_seq_idx == net_seq_idx) {
447 uatomic_set(&stream->endpoint_status, status);
448 DBG("Delete flag set to data stream %d", stream->wait_fd);
449 }
450 }
451 rcu_read_unlock();
452}
453
454/*
455 * Cleanup a relayd object by flagging every associated streams for deletion,
456 * destroying the object meaning removing it from the relayd hash table,
457 * closing the sockets and freeing the memory in a RCU call.
458 *
459 * If a local data context is available, notify the threads that the streams'
460 * state have changed.
461 */
462static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd,
463 struct lttng_consumer_local_data *ctx)
464{
da009f2c 465 uint64_t netidx;
8994307f
DG
466
467 assert(relayd);
468
9617607b
DG
469 DBG("Cleaning up relayd sockets");
470
8994307f
DG
471 /* Save the net sequence index before destroying the object */
472 netidx = relayd->net_seq_idx;
473
474 /*
475 * Delete the relayd from the relayd hash table, close the sockets and free
476 * the object in a RCU call.
477 */
51230d70 478 consumer_destroy_relayd(relayd);
8994307f
DG
479
480 /* Set inactive endpoint to all streams */
481 update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE);
482
483 /*
484 * With a local data context, notify the threads that the streams' state
485 * have changed. The write() action on the pipe acts as an "implicit"
486 * memory barrier ordering the updates of the end point status from the
487 * read of this status which happens AFTER receiving this notify.
488 */
489 if (ctx) {
acdb9057 490 notify_thread_lttng_pipe(ctx->consumer_data_pipe);
13886d2d 491 notify_thread_lttng_pipe(ctx->consumer_metadata_pipe);
8994307f
DG
492 }
493}
494
a6ba4fe1
DG
495/*
496 * Flag a relayd socket pair for destruction. Destroy it if the refcount
497 * reaches zero.
498 *
499 * RCU read side lock MUST be aquired before calling this function.
500 */
501void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd)
502{
503 assert(relayd);
504
505 /* Set destroy flag for this object */
506 uatomic_set(&relayd->destroy_flag, 1);
507
508 /* Destroy the relayd if refcount is 0 */
509 if (uatomic_read(&relayd->refcount) == 0) {
51230d70 510 consumer_destroy_relayd(relayd);
a6ba4fe1
DG
511 }
512}
513
3bd1e081 514/*
1d1a276c
DG
515 * Completly destroy stream from every visiable data structure and the given
516 * hash table if one.
517 *
518 * One this call returns, the stream object is not longer usable nor visible.
3bd1e081 519 */
e316aad5
DG
520void consumer_del_stream(struct lttng_consumer_stream *stream,
521 struct lttng_ht *ht)
3bd1e081 522{
1d1a276c 523 consumer_stream_destroy(stream, ht);
3bd1e081
MD
524}
525
5ab66908
MD
526/*
527 * XXX naming of del vs destroy is all mixed up.
528 */
529void consumer_del_stream_for_data(struct lttng_consumer_stream *stream)
530{
531 consumer_stream_destroy(stream, data_ht);
532}
533
534void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream)
535{
536 consumer_stream_destroy(stream, metadata_ht);
537}
538
d88aee68
DG
539struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key,
540 uint64_t stream_key,
3bd1e081 541 enum lttng_consumer_stream_state state,
ffe60014 542 const char *channel_name,
6df2e2c9 543 uid_t uid,
00e2e675 544 gid_t gid,
57a269f2 545 uint64_t relayd_id,
53632229 546 uint64_t session_id,
ffe60014
DG
547 int cpu,
548 int *alloc_ret,
4891ece8
DG
549 enum consumer_channel_type type,
550 unsigned int monitor)
3bd1e081 551{
ffe60014 552 int ret;
3bd1e081 553 struct lttng_consumer_stream *stream;
3bd1e081 554
effcf122 555 stream = zmalloc(sizeof(*stream));
3bd1e081 556 if (stream == NULL) {
7a57cf92 557 PERROR("malloc struct lttng_consumer_stream");
ffe60014 558 ret = -ENOMEM;
7a57cf92 559 goto end;
3bd1e081 560 }
7a57cf92 561
d56db448
DG
562 rcu_read_lock();
563
3bd1e081 564 stream->key = stream_key;
3bd1e081
MD
565 stream->out_fd = -1;
566 stream->out_fd_offset = 0;
e5d1a9b3 567 stream->output_written = 0;
3bd1e081 568 stream->state = state;
6df2e2c9
MD
569 stream->uid = uid;
570 stream->gid = gid;
ffe60014 571 stream->net_seq_idx = relayd_id;
53632229 572 stream->session_id = session_id;
4891ece8 573 stream->monitor = monitor;
774d490c 574 stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE;
309167d2 575 stream->index_fd = -1;
53632229 576 pthread_mutex_init(&stream->lock, NULL);
c585821b 577 pthread_mutex_init(&stream->metadata_timer_lock, NULL);
58b1f425 578
ffe60014
DG
579 /* If channel is the metadata, flag this stream as metadata. */
580 if (type == CONSUMER_CHANNEL_TYPE_METADATA) {
581 stream->metadata_flag = 1;
582 /* Metadata is flat out. */
583 strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name));
94d49140
JD
584 /* Live rendez-vous point. */
585 pthread_cond_init(&stream->metadata_rdv, NULL);
586 pthread_mutex_init(&stream->metadata_rdv_lock, NULL);
58b1f425 587 } else {
ffe60014
DG
588 /* Format stream name to <channel_name>_<cpu_number> */
589 ret = snprintf(stream->name, sizeof(stream->name), "%s_%d",
590 channel_name, cpu);
591 if (ret < 0) {
592 PERROR("snprintf stream name");
593 goto error;
594 }
58b1f425 595 }
c30aaa51 596
ffe60014 597 /* Key is always the wait_fd for streams. */
d88aee68 598 lttng_ht_node_init_u64(&stream->node, stream->key);
ffe60014 599
d8ef542d
MD
600 /* Init node per channel id key */
601 lttng_ht_node_init_u64(&stream->node_channel_id, channel_key);
602
53632229 603 /* Init session id node with the stream session id */
d88aee68 604 lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id);
53632229 605
07b86b52
JD
606 DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64
607 " relayd_id %" PRIu64 ", session_id %" PRIu64,
608 stream->name, stream->key, channel_key,
609 stream->net_seq_idx, stream->session_id);
d56db448
DG
610
611 rcu_read_unlock();
3bd1e081 612 return stream;
c80048c6
MD
613
614error:
d56db448 615 rcu_read_unlock();
c80048c6 616 free(stream);
7a57cf92 617end:
ffe60014
DG
618 if (alloc_ret) {
619 *alloc_ret = ret;
620 }
c80048c6 621 return NULL;
3bd1e081
MD
622}
623
624/*
625 * Add a stream to the global list protected by a mutex.
626 */
5ab66908 627int consumer_add_data_stream(struct lttng_consumer_stream *stream)
3bd1e081 628{
5ab66908 629 struct lttng_ht *ht = data_ht;
3bd1e081
MD
630 int ret = 0;
631
e316aad5 632 assert(stream);
43c34bc3 633 assert(ht);
c77fc10a 634
d88aee68 635 DBG3("Adding consumer stream %" PRIu64, stream->key);
e316aad5
DG
636
637 pthread_mutex_lock(&consumer_data.lock);
a9838785 638 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 639 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 640 pthread_mutex_lock(&stream->lock);
b0b335c8 641 rcu_read_lock();
e316aad5 642
43c34bc3 643 /* Steal stream identifier to avoid having streams with the same key */
ffe60014 644 steal_stream_key(stream->key, ht);
43c34bc3 645
d88aee68 646 lttng_ht_add_unique_u64(ht, &stream->node);
00e2e675 647
d8ef542d
MD
648 lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht,
649 &stream->node_channel_id);
650
ca22feea
DG
651 /*
652 * Add stream to the stream_list_ht of the consumer data. No need to steal
653 * the key since the HT does not use it and we allow to add redundant keys
654 * into this table.
655 */
d88aee68 656 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 657
e316aad5 658 /*
ffe60014
DG
659 * When nb_init_stream_left reaches 0, we don't need to trigger any action
660 * in terms of destroying the associated channel, because the action that
e316aad5
DG
661 * causes the count to become 0 also causes a stream to be added. The
662 * channel deletion will thus be triggered by the following removal of this
663 * stream.
664 */
ffe60014 665 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
666 /* Increment refcount before decrementing nb_init_stream_left */
667 cmm_smp_wmb();
ffe60014 668 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
669 }
670
671 /* Update consumer data once the node is inserted. */
3bd1e081
MD
672 consumer_data.stream_count++;
673 consumer_data.need_update = 1;
674
e316aad5 675 rcu_read_unlock();
2e818a6a 676 pthread_mutex_unlock(&stream->lock);
ec6ea7d0 677 pthread_mutex_unlock(&stream->chan->timer_lock);
a9838785 678 pthread_mutex_unlock(&stream->chan->lock);
3bd1e081 679 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 680
3bd1e081
MD
681 return ret;
682}
683
5ab66908
MD
684void consumer_del_data_stream(struct lttng_consumer_stream *stream)
685{
686 consumer_del_stream(stream, data_ht);
687}
688
00e2e675 689/*
3f8e211f
DG
690 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
691 * be acquired before calling this.
00e2e675 692 */
d09e1200 693static int add_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
694{
695 int ret = 0;
d88aee68 696 struct lttng_ht_node_u64 *node;
00e2e675
DG
697 struct lttng_ht_iter iter;
698
ffe60014 699 assert(relayd);
00e2e675 700
00e2e675 701 lttng_ht_lookup(consumer_data.relayd_ht,
d88aee68
DG
702 &relayd->net_seq_idx, &iter);
703 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675 704 if (node != NULL) {
00e2e675
DG
705 goto end;
706 }
d88aee68 707 lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node);
00e2e675 708
00e2e675
DG
709end:
710 return ret;
711}
712
713/*
714 * Allocate and return a consumer relayd socket.
715 */
027a694f 716static struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair(
da009f2c 717 uint64_t net_seq_idx)
00e2e675
DG
718{
719 struct consumer_relayd_sock_pair *obj = NULL;
720
da009f2c
MD
721 /* net sequence index of -1 is a failure */
722 if (net_seq_idx == (uint64_t) -1ULL) {
00e2e675
DG
723 goto error;
724 }
725
726 obj = zmalloc(sizeof(struct consumer_relayd_sock_pair));
727 if (obj == NULL) {
728 PERROR("zmalloc relayd sock");
729 goto error;
730 }
731
732 obj->net_seq_idx = net_seq_idx;
733 obj->refcount = 0;
173af62f 734 obj->destroy_flag = 0;
f96e4545
MD
735 obj->control_sock.sock.fd = -1;
736 obj->data_sock.sock.fd = -1;
d88aee68 737 lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx);
00e2e675
DG
738 pthread_mutex_init(&obj->ctrl_sock_mutex, NULL);
739
740error:
741 return obj;
742}
743
744/*
745 * Find a relayd socket pair in the global consumer data.
746 *
747 * Return the object if found else NULL.
b0b335c8
MD
748 * RCU read-side lock must be held across this call and while using the
749 * returned object.
00e2e675 750 */
d88aee68 751struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key)
00e2e675
DG
752{
753 struct lttng_ht_iter iter;
d88aee68 754 struct lttng_ht_node_u64 *node;
00e2e675
DG
755 struct consumer_relayd_sock_pair *relayd = NULL;
756
757 /* Negative keys are lookup failures */
d88aee68 758 if (key == (uint64_t) -1ULL) {
00e2e675
DG
759 goto error;
760 }
761
d88aee68 762 lttng_ht_lookup(consumer_data.relayd_ht, &key,
00e2e675 763 &iter);
d88aee68 764 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675
DG
765 if (node != NULL) {
766 relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node);
767 }
768
00e2e675
DG
769error:
770 return relayd;
771}
772
10a50311
JD
773/*
774 * Find a relayd and send the stream
775 *
776 * Returns 0 on success, < 0 on error
777 */
778int consumer_send_relayd_stream(struct lttng_consumer_stream *stream,
779 char *path)
780{
781 int ret = 0;
782 struct consumer_relayd_sock_pair *relayd;
783
784 assert(stream);
785 assert(stream->net_seq_idx != -1ULL);
786 assert(path);
787
788 /* The stream is not metadata. Get relayd reference if exists. */
789 rcu_read_lock();
790 relayd = consumer_find_relayd(stream->net_seq_idx);
791 if (relayd != NULL) {
792 /* Add stream on the relayd */
793 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
794 ret = relayd_add_stream(&relayd->control_sock, stream->name,
795 path, &stream->relayd_stream_id,
796 stream->chan->tracefile_size, stream->chan->tracefile_count);
797 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
798 if (ret < 0) {
799 goto end;
800 }
1c20f0e2 801
10a50311 802 uatomic_inc(&relayd->refcount);
d01178b6 803 stream->sent_to_relayd = 1;
10a50311
JD
804 } else {
805 ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.",
806 stream->key, stream->net_seq_idx);
807 ret = -1;
808 goto end;
809 }
810
811 DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64,
812 stream->name, stream->key, stream->net_seq_idx);
813
814end:
815 rcu_read_unlock();
816 return ret;
817}
818
a4baae1b
JD
819/*
820 * Find a relayd and send the streams sent message
821 *
822 * Returns 0 on success, < 0 on error
823 */
824int consumer_send_relayd_streams_sent(uint64_t net_seq_idx)
825{
826 int ret = 0;
827 struct consumer_relayd_sock_pair *relayd;
828
829 assert(net_seq_idx != -1ULL);
830
831 /* The stream is not metadata. Get relayd reference if exists. */
832 rcu_read_lock();
833 relayd = consumer_find_relayd(net_seq_idx);
834 if (relayd != NULL) {
835 /* Add stream on the relayd */
836 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
837 ret = relayd_streams_sent(&relayd->control_sock);
838 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
839 if (ret < 0) {
840 goto end;
841 }
842 } else {
843 ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.",
844 net_seq_idx);
845 ret = -1;
846 goto end;
847 }
848
849 ret = 0;
850 DBG("All streams sent relayd id %" PRIu64, net_seq_idx);
851
852end:
853 rcu_read_unlock();
854 return ret;
855}
856
10a50311
JD
857/*
858 * Find a relayd and close the stream
859 */
860void close_relayd_stream(struct lttng_consumer_stream *stream)
861{
862 struct consumer_relayd_sock_pair *relayd;
863
864 /* The stream is not metadata. Get relayd reference if exists. */
865 rcu_read_lock();
866 relayd = consumer_find_relayd(stream->net_seq_idx);
867 if (relayd) {
868 consumer_stream_relayd_close(stream, relayd);
869 }
870 rcu_read_unlock();
871}
872
00e2e675
DG
873/*
874 * Handle stream for relayd transmission if the stream applies for network
875 * streaming where the net sequence index is set.
876 *
877 * Return destination file descriptor or negative value on error.
878 */
6197aea7 879static int write_relayd_stream_header(struct lttng_consumer_stream *stream,
1d4dfdef
DG
880 size_t data_size, unsigned long padding,
881 struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
882{
883 int outfd = -1, ret;
00e2e675
DG
884 struct lttcomm_relayd_data_hdr data_hdr;
885
886 /* Safety net */
887 assert(stream);
6197aea7 888 assert(relayd);
00e2e675
DG
889
890 /* Reset data header */
891 memset(&data_hdr, 0, sizeof(data_hdr));
892
00e2e675
DG
893 if (stream->metadata_flag) {
894 /* Caller MUST acquire the relayd control socket lock */
895 ret = relayd_send_metadata(&relayd->control_sock, data_size);
896 if (ret < 0) {
897 goto error;
898 }
899
900 /* Metadata are always sent on the control socket. */
6151a90f 901 outfd = relayd->control_sock.sock.fd;
00e2e675
DG
902 } else {
903 /* Set header with stream information */
904 data_hdr.stream_id = htobe64(stream->relayd_stream_id);
905 data_hdr.data_size = htobe32(data_size);
1d4dfdef 906 data_hdr.padding_size = htobe32(padding);
39df6d9f
DG
907 /*
908 * Note that net_seq_num below is assigned with the *current* value of
909 * next_net_seq_num and only after that the next_net_seq_num will be
910 * increment. This is why when issuing a command on the relayd using
911 * this next value, 1 should always be substracted in order to compare
912 * the last seen sequence number on the relayd side to the last sent.
913 */
3604f373 914 data_hdr.net_seq_num = htobe64(stream->next_net_seq_num);
00e2e675
DG
915 /* Other fields are zeroed previously */
916
917 ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr,
918 sizeof(data_hdr));
919 if (ret < 0) {
920 goto error;
921 }
922
3604f373
DG
923 ++stream->next_net_seq_num;
924
00e2e675 925 /* Set to go on data socket */
6151a90f 926 outfd = relayd->data_sock.sock.fd;
00e2e675
DG
927 }
928
929error:
930 return outfd;
931}
932
3bd1e081 933/*
ffe60014
DG
934 * Allocate and return a new lttng_consumer_channel object using the given key
935 * to initialize the hash table node.
936 *
937 * On error, return NULL.
3bd1e081 938 */
886224ff 939struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key,
ffe60014
DG
940 uint64_t session_id,
941 const char *pathname,
942 const char *name,
943 uid_t uid,
944 gid_t gid,
57a269f2 945 uint64_t relayd_id,
1624d5b7
JD
946 enum lttng_event_output output,
947 uint64_t tracefile_size,
2bba9e53 948 uint64_t tracefile_count,
1950109e 949 uint64_t session_id_per_pid,
ecc48a90 950 unsigned int monitor,
d7ba1388 951 unsigned int live_timer_interval,
3d071855 952 const char *root_shm_path,
d7ba1388 953 const char *shm_path)
3bd1e081
MD
954{
955 struct lttng_consumer_channel *channel;
3bd1e081 956
276b26d1 957 channel = zmalloc(sizeof(*channel));
3bd1e081 958 if (channel == NULL) {
7a57cf92 959 PERROR("malloc struct lttng_consumer_channel");
3bd1e081
MD
960 goto end;
961 }
ffe60014
DG
962
963 channel->key = key;
3bd1e081 964 channel->refcount = 0;
ffe60014 965 channel->session_id = session_id;
1950109e 966 channel->session_id_per_pid = session_id_per_pid;
ffe60014
DG
967 channel->uid = uid;
968 channel->gid = gid;
969 channel->relayd_id = relayd_id;
1624d5b7
JD
970 channel->tracefile_size = tracefile_size;
971 channel->tracefile_count = tracefile_count;
2bba9e53 972 channel->monitor = monitor;
ecc48a90 973 channel->live_timer_interval = live_timer_interval;
a9838785 974 pthread_mutex_init(&channel->lock, NULL);
ec6ea7d0 975 pthread_mutex_init(&channel->timer_lock, NULL);
ffe60014 976
0c759fc9
DG
977 switch (output) {
978 case LTTNG_EVENT_SPLICE:
979 channel->output = CONSUMER_CHANNEL_SPLICE;
980 break;
981 case LTTNG_EVENT_MMAP:
982 channel->output = CONSUMER_CHANNEL_MMAP;
983 break;
984 default:
985 assert(0);
986 free(channel);
987 channel = NULL;
988 goto end;
989 }
990
07b86b52
JD
991 /*
992 * In monitor mode, the streams associated with the channel will be put in
993 * a special list ONLY owned by this channel. So, the refcount is set to 1
994 * here meaning that the channel itself has streams that are referenced.
995 *
996 * On a channel deletion, once the channel is no longer visible, the
997 * refcount is decremented and checked for a zero value to delete it. With
998 * streams in no monitor mode, it will now be safe to destroy the channel.
999 */
1000 if (!channel->monitor) {
1001 channel->refcount = 1;
1002 }
1003
ffe60014
DG
1004 strncpy(channel->pathname, pathname, sizeof(channel->pathname));
1005 channel->pathname[sizeof(channel->pathname) - 1] = '\0';
1006
1007 strncpy(channel->name, name, sizeof(channel->name));
1008 channel->name[sizeof(channel->name) - 1] = '\0';
1009
3d071855
MD
1010 if (root_shm_path) {
1011 strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path));
1012 channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0';
1013 }
d7ba1388
MD
1014 if (shm_path) {
1015 strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path));
1016 channel->shm_path[sizeof(channel->shm_path) - 1] = '\0';
1017 }
1018
d88aee68 1019 lttng_ht_node_init_u64(&channel->node, channel->key);
d8ef542d
MD
1020
1021 channel->wait_fd = -1;
1022
ffe60014
DG
1023 CDS_INIT_LIST_HEAD(&channel->streams.head);
1024
d88aee68 1025 DBG("Allocated channel (key %" PRIu64 ")", channel->key)
3bd1e081 1026
3bd1e081
MD
1027end:
1028 return channel;
1029}
1030
1031/*
1032 * Add a channel to the global list protected by a mutex.
821fffb2 1033 *
b5a6470f 1034 * Always return 0 indicating success.
3bd1e081 1035 */
d8ef542d
MD
1036int consumer_add_channel(struct lttng_consumer_channel *channel,
1037 struct lttng_consumer_local_data *ctx)
3bd1e081 1038{
3bd1e081 1039 pthread_mutex_lock(&consumer_data.lock);
a9838785 1040 pthread_mutex_lock(&channel->lock);
ec6ea7d0 1041 pthread_mutex_lock(&channel->timer_lock);
c77fc10a 1042
b5a6470f
DG
1043 /*
1044 * This gives us a guarantee that the channel we are about to add to the
1045 * channel hash table will be unique. See this function comment on the why
1046 * we need to steel the channel key at this stage.
1047 */
1048 steal_channel_key(channel->key);
c77fc10a 1049
b5a6470f 1050 rcu_read_lock();
d88aee68 1051 lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node);
6065ceec 1052 rcu_read_unlock();
b5a6470f 1053
ec6ea7d0 1054 pthread_mutex_unlock(&channel->timer_lock);
a9838785 1055 pthread_mutex_unlock(&channel->lock);
3bd1e081 1056 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 1057
b5a6470f 1058 if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) {
a0cbdd2e 1059 notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD);
d8ef542d 1060 }
b5a6470f
DG
1061
1062 return 0;
3bd1e081
MD
1063}
1064
1065/*
1066 * Allocate the pollfd structure and the local view of the out fds to avoid
1067 * doing a lookup in the linked list and concurrency issues when writing is
1068 * needed. Called with consumer_data.lock held.
1069 *
1070 * Returns the number of fds in the structures.
1071 */
ffe60014
DG
1072static int update_poll_array(struct lttng_consumer_local_data *ctx,
1073 struct pollfd **pollfd, struct lttng_consumer_stream **local_stream,
1074 struct lttng_ht *ht)
3bd1e081 1075{
3bd1e081 1076 int i = 0;
e4421fec
DG
1077 struct lttng_ht_iter iter;
1078 struct lttng_consumer_stream *stream;
3bd1e081 1079
ffe60014
DG
1080 assert(ctx);
1081 assert(ht);
1082 assert(pollfd);
1083 assert(local_stream);
1084
3bd1e081 1085 DBG("Updating poll fd array");
481d6c57 1086 rcu_read_lock();
43c34bc3 1087 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
8994307f
DG
1088 /*
1089 * Only active streams with an active end point can be added to the
1090 * poll set and local stream storage of the thread.
1091 *
1092 * There is a potential race here for endpoint_status to be updated
1093 * just after the check. However, this is OK since the stream(s) will
1094 * be deleted once the thread is notified that the end point state has
1095 * changed where this function will be called back again.
1096 */
1097 if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM ||
79d4ffb7 1098 stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) {
3bd1e081
MD
1099 continue;
1100 }
7972aab2
DG
1101 /*
1102 * This clobbers way too much the debug output. Uncomment that if you
1103 * need it for debugging purposes.
1104 *
1105 * DBG("Active FD %d", stream->wait_fd);
1106 */
e4421fec 1107 (*pollfd)[i].fd = stream->wait_fd;
3bd1e081 1108 (*pollfd)[i].events = POLLIN | POLLPRI;
e4421fec 1109 local_stream[i] = stream;
3bd1e081
MD
1110 i++;
1111 }
481d6c57 1112 rcu_read_unlock();
3bd1e081
MD
1113
1114 /*
50f8ae69 1115 * Insert the consumer_data_pipe at the end of the array and don't
3bd1e081
MD
1116 * increment i so nb_fd is the number of real FD.
1117 */
acdb9057 1118 (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe);
509bb1cf 1119 (*pollfd)[i].events = POLLIN | POLLPRI;
02b3d176
DG
1120
1121 (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe);
1122 (*pollfd)[i + 1].events = POLLIN | POLLPRI;
3bd1e081
MD
1123 return i;
1124}
1125
1126/*
84382d49
MD
1127 * Poll on the should_quit pipe and the command socket return -1 on
1128 * error, 1 if should exit, 0 if data is available on the command socket
3bd1e081
MD
1129 */
1130int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll)
1131{
1132 int num_rdy;
1133
88f2b785 1134restart:
3bd1e081
MD
1135 num_rdy = poll(consumer_sockpoll, 2, -1);
1136 if (num_rdy == -1) {
88f2b785
MD
1137 /*
1138 * Restart interrupted system call.
1139 */
1140 if (errno == EINTR) {
1141 goto restart;
1142 }
7a57cf92 1143 PERROR("Poll error");
84382d49 1144 return -1;
3bd1e081 1145 }
509bb1cf 1146 if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) {
3bd1e081 1147 DBG("consumer_should_quit wake up");
84382d49 1148 return 1;
3bd1e081
MD
1149 }
1150 return 0;
3bd1e081
MD
1151}
1152
1153/*
1154 * Set the error socket.
1155 */
ffe60014
DG
1156void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx,
1157 int sock)
3bd1e081
MD
1158{
1159 ctx->consumer_error_socket = sock;
1160}
1161
1162/*
1163 * Set the command socket path.
1164 */
3bd1e081
MD
1165void lttng_consumer_set_command_sock_path(
1166 struct lttng_consumer_local_data *ctx, char *sock)
1167{
1168 ctx->consumer_command_sock_path = sock;
1169}
1170
1171/*
1172 * Send return code to the session daemon.
1173 * If the socket is not defined, we return 0, it is not a fatal error
1174 */
ffe60014 1175int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd)
3bd1e081
MD
1176{
1177 if (ctx->consumer_error_socket > 0) {
1178 return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd,
1179 sizeof(enum lttcomm_sessiond_command));
1180 }
1181
1182 return 0;
1183}
1184
1185/*
228b5bf7
DG
1186 * Close all the tracefiles and stream fds and MUST be called when all
1187 * instances are destroyed i.e. when all threads were joined and are ended.
3bd1e081
MD
1188 */
1189void lttng_consumer_cleanup(void)
1190{
e4421fec 1191 struct lttng_ht_iter iter;
ffe60014 1192 struct lttng_consumer_channel *channel;
6065ceec
DG
1193
1194 rcu_read_lock();
3bd1e081 1195
ffe60014
DG
1196 cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel,
1197 node.node) {
702b1ea4 1198 consumer_del_channel(channel);
3bd1e081 1199 }
6065ceec
DG
1200
1201 rcu_read_unlock();
d6ce1df2 1202
d6ce1df2 1203 lttng_ht_destroy(consumer_data.channel_ht);
228b5bf7
DG
1204
1205 cleanup_relayd_ht();
1206
d8ef542d
MD
1207 lttng_ht_destroy(consumer_data.stream_per_chan_id_ht);
1208
228b5bf7
DG
1209 /*
1210 * This HT contains streams that are freed by either the metadata thread or
1211 * the data thread so we do *nothing* on the hash table and simply destroy
1212 * it.
1213 */
1214 lttng_ht_destroy(consumer_data.stream_list_ht);
3bd1e081
MD
1215}
1216
1217/*
1218 * Called from signal handler.
1219 */
1220void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx)
1221{
6cd525e8
MD
1222 ssize_t ret;
1223
3bd1e081 1224 consumer_quit = 1;
6cd525e8
MD
1225 ret = lttng_write(ctx->consumer_should_quit[1], "4", 1);
1226 if (ret < 1) {
7a57cf92 1227 PERROR("write consumer quit");
3bd1e081 1228 }
ab1027f4
DG
1229
1230 DBG("Consumer flag that it should quit");
3bd1e081
MD
1231}
1232
00e2e675
DG
1233void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream,
1234 off_t orig_offset)
3bd1e081
MD
1235{
1236 int outfd = stream->out_fd;
1237
1238 /*
1239 * This does a blocking write-and-wait on any page that belongs to the
1240 * subbuffer prior to the one we just wrote.
1241 * Don't care about error values, as these are just hints and ways to
1242 * limit the amount of page cache used.
1243 */
ffe60014 1244 if (orig_offset < stream->max_sb_size) {
3bd1e081
MD
1245 return;
1246 }
ffe60014
DG
1247 lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size,
1248 stream->max_sb_size,
3bd1e081
MD
1249 SYNC_FILE_RANGE_WAIT_BEFORE
1250 | SYNC_FILE_RANGE_WRITE
1251 | SYNC_FILE_RANGE_WAIT_AFTER);
1252 /*
1253 * Give hints to the kernel about how we access the file:
1254 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1255 * we write it.
1256 *
1257 * We need to call fadvise again after the file grows because the
1258 * kernel does not seem to apply fadvise to non-existing parts of the
1259 * file.
1260 *
1261 * Call fadvise _after_ having waited for the page writeback to
1262 * complete because the dirty page writeback semantic is not well
1263 * defined. So it can be expected to lead to lower throughput in
1264 * streaming.
1265 */
ffe60014
DG
1266 posix_fadvise(outfd, orig_offset - stream->max_sb_size,
1267 stream->max_sb_size, POSIX_FADV_DONTNEED);
3bd1e081
MD
1268}
1269
1270/*
1271 * Initialise the necessary environnement :
1272 * - create a new context
1273 * - create the poll_pipe
1274 * - create the should_quit pipe (for signal handler)
1275 * - create the thread pipe (for splice)
1276 *
1277 * Takes a function pointer as argument, this function is called when data is
1278 * available on a buffer. This function is responsible to do the
1279 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1280 * buffer configuration and then kernctl_put_next_subbuf at the end.
1281 *
1282 * Returns a pointer to the new context or NULL on error.
1283 */
1284struct lttng_consumer_local_data *lttng_consumer_create(
1285 enum lttng_consumer_type type,
4078b776 1286 ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream,
d41f73b7 1287 struct lttng_consumer_local_data *ctx),
3bd1e081
MD
1288 int (*recv_channel)(struct lttng_consumer_channel *channel),
1289 int (*recv_stream)(struct lttng_consumer_stream *stream),
30319bcb 1290 int (*update_stream)(uint64_t stream_key, uint32_t state))
3bd1e081 1291{
d8ef542d 1292 int ret;
3bd1e081
MD
1293 struct lttng_consumer_local_data *ctx;
1294
1295 assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN ||
1296 consumer_data.type == type);
1297 consumer_data.type = type;
1298
effcf122 1299 ctx = zmalloc(sizeof(struct lttng_consumer_local_data));
3bd1e081 1300 if (ctx == NULL) {
7a57cf92 1301 PERROR("allocating context");
3bd1e081
MD
1302 goto error;
1303 }
1304
1305 ctx->consumer_error_socket = -1;
331744e3 1306 ctx->consumer_metadata_socket = -1;
75d83e50 1307 pthread_mutex_init(&ctx->metadata_socket_lock, NULL);
3bd1e081
MD
1308 /* assign the callbacks */
1309 ctx->on_buffer_ready = buffer_ready;
1310 ctx->on_recv_channel = recv_channel;
1311 ctx->on_recv_stream = recv_stream;
1312 ctx->on_update_stream = update_stream;
1313
acdb9057
DG
1314 ctx->consumer_data_pipe = lttng_pipe_open(0);
1315 if (!ctx->consumer_data_pipe) {
3bd1e081
MD
1316 goto error_poll_pipe;
1317 }
1318
02b3d176
DG
1319 ctx->consumer_wakeup_pipe = lttng_pipe_open(0);
1320 if (!ctx->consumer_wakeup_pipe) {
1321 goto error_wakeup_pipe;
1322 }
1323
3bd1e081
MD
1324 ret = pipe(ctx->consumer_should_quit);
1325 if (ret < 0) {
7a57cf92 1326 PERROR("Error creating recv pipe");
3bd1e081
MD
1327 goto error_quit_pipe;
1328 }
1329
d8ef542d
MD
1330 ret = pipe(ctx->consumer_channel_pipe);
1331 if (ret < 0) {
1332 PERROR("Error creating channel pipe");
1333 goto error_channel_pipe;
1334 }
1335
13886d2d
DG
1336 ctx->consumer_metadata_pipe = lttng_pipe_open(0);
1337 if (!ctx->consumer_metadata_pipe) {
fb3a43a9
DG
1338 goto error_metadata_pipe;
1339 }
3bd1e081 1340
fb3a43a9 1341 return ctx;
3bd1e081 1342
fb3a43a9 1343error_metadata_pipe:
d8ef542d
MD
1344 utils_close_pipe(ctx->consumer_channel_pipe);
1345error_channel_pipe:
d8ef542d 1346 utils_close_pipe(ctx->consumer_should_quit);
3bd1e081 1347error_quit_pipe:
02b3d176
DG
1348 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
1349error_wakeup_pipe:
acdb9057 1350 lttng_pipe_destroy(ctx->consumer_data_pipe);
3bd1e081
MD
1351error_poll_pipe:
1352 free(ctx);
1353error:
1354 return NULL;
1355}
1356
282dadbc
MD
1357/*
1358 * Iterate over all streams of the hashtable and free them properly.
1359 */
1360static void destroy_data_stream_ht(struct lttng_ht *ht)
1361{
1362 struct lttng_ht_iter iter;
1363 struct lttng_consumer_stream *stream;
1364
1365 if (ht == NULL) {
1366 return;
1367 }
1368
1369 rcu_read_lock();
1370 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1371 /*
1372 * Ignore return value since we are currently cleaning up so any error
1373 * can't be handled.
1374 */
1375 (void) consumer_del_stream(stream, ht);
1376 }
1377 rcu_read_unlock();
1378
1379 lttng_ht_destroy(ht);
1380}
1381
1382/*
1383 * Iterate over all streams of the metadata hashtable and free them
1384 * properly.
1385 */
1386static void destroy_metadata_stream_ht(struct lttng_ht *ht)
1387{
1388 struct lttng_ht_iter iter;
1389 struct lttng_consumer_stream *stream;
1390
1391 if (ht == NULL) {
1392 return;
1393 }
1394
1395 rcu_read_lock();
1396 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1397 /*
1398 * Ignore return value since we are currently cleaning up so any error
1399 * can't be handled.
1400 */
1401 (void) consumer_del_metadata_stream(stream, ht);
1402 }
1403 rcu_read_unlock();
1404
1405 lttng_ht_destroy(ht);
1406}
1407
3bd1e081
MD
1408/*
1409 * Close all fds associated with the instance and free the context.
1410 */
1411void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx)
1412{
4c462e79
MD
1413 int ret;
1414
ab1027f4
DG
1415 DBG("Consumer destroying it. Closing everything.");
1416
4f2e75b9
DG
1417 if (!ctx) {
1418 return;
1419 }
1420
282dadbc
MD
1421 destroy_data_stream_ht(data_ht);
1422 destroy_metadata_stream_ht(metadata_ht);
1423
4c462e79
MD
1424 ret = close(ctx->consumer_error_socket);
1425 if (ret) {
1426 PERROR("close");
1427 }
331744e3
JD
1428 ret = close(ctx->consumer_metadata_socket);
1429 if (ret) {
1430 PERROR("close");
1431 }
d8ef542d 1432 utils_close_pipe(ctx->consumer_channel_pipe);
acdb9057 1433 lttng_pipe_destroy(ctx->consumer_data_pipe);
13886d2d 1434 lttng_pipe_destroy(ctx->consumer_metadata_pipe);
02b3d176 1435 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
d8ef542d 1436 utils_close_pipe(ctx->consumer_should_quit);
fb3a43a9 1437
3bd1e081
MD
1438 unlink(ctx->consumer_command_sock_path);
1439 free(ctx);
1440}
1441
6197aea7
DG
1442/*
1443 * Write the metadata stream id on the specified file descriptor.
1444 */
1445static int write_relayd_metadata_id(int fd,
1446 struct lttng_consumer_stream *stream,
ffe60014 1447 struct consumer_relayd_sock_pair *relayd, unsigned long padding)
6197aea7 1448{
6cd525e8 1449 ssize_t ret;
1d4dfdef 1450 struct lttcomm_relayd_metadata_payload hdr;
6197aea7 1451
1d4dfdef
DG
1452 hdr.stream_id = htobe64(stream->relayd_stream_id);
1453 hdr.padding_size = htobe32(padding);
6cd525e8
MD
1454 ret = lttng_write(fd, (void *) &hdr, sizeof(hdr));
1455 if (ret < sizeof(hdr)) {
d7b75ec8 1456 /*
6f04ed72 1457 * This error means that the fd's end is closed so ignore the PERROR
d7b75ec8
DG
1458 * not to clubber the error output since this can happen in a normal
1459 * code path.
1460 */
1461 if (errno != EPIPE) {
1462 PERROR("write metadata stream id");
1463 }
1464 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno);
534d2592
DG
1465 /*
1466 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1467 * handle writting the missing part so report that as an error and
1468 * don't lie to the caller.
1469 */
1470 ret = -1;
6197aea7
DG
1471 goto end;
1472 }
1d4dfdef
DG
1473 DBG("Metadata stream id %" PRIu64 " with padding %lu written before data",
1474 stream->relayd_stream_id, padding);
6197aea7
DG
1475
1476end:
6cd525e8 1477 return (int) ret;
6197aea7
DG
1478}
1479
3bd1e081 1480/*
09e26845
DG
1481 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1482 * core function for writing trace buffers to either the local filesystem or
1483 * the network.
1484 *
79d4ffb7
DG
1485 * It must be called with the stream lock held.
1486 *
09e26845 1487 * Careful review MUST be put if any changes occur!
3bd1e081
MD
1488 *
1489 * Returns the number of bytes written
1490 */
4078b776 1491ssize_t lttng_consumer_on_read_subbuffer_mmap(
3bd1e081 1492 struct lttng_consumer_local_data *ctx,
1d4dfdef 1493 struct lttng_consumer_stream *stream, unsigned long len,
309167d2 1494 unsigned long padding,
50adc264 1495 struct ctf_packet_index *index)
3bd1e081 1496{
f02e1e8a 1497 unsigned long mmap_offset;
ffe60014 1498 void *mmap_base;
994ab360 1499 ssize_t ret = 0;
f02e1e8a
DG
1500 off_t orig_offset = stream->out_fd_offset;
1501 /* Default is on the disk */
1502 int outfd = stream->out_fd;
f02e1e8a 1503 struct consumer_relayd_sock_pair *relayd = NULL;
8994307f 1504 unsigned int relayd_hang_up = 0;
f02e1e8a
DG
1505
1506 /* RCU lock for the relayd pointer */
1507 rcu_read_lock();
1508
1509 /* Flag that the current stream if set for network streaming. */
da009f2c 1510 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1511 relayd = consumer_find_relayd(stream->net_seq_idx);
1512 if (relayd == NULL) {
56591bac 1513 ret = -EPIPE;
f02e1e8a
DG
1514 goto end;
1515 }
1516 }
1517
1518 /* get the offset inside the fd to mmap */
3bd1e081
MD
1519 switch (consumer_data.type) {
1520 case LTTNG_CONSUMER_KERNEL:
ffe60014 1521 mmap_base = stream->mmap_base;
f02e1e8a 1522 ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset);
994ab360
DG
1523 if (ret < 0) {
1524 ret = -errno;
56591bac 1525 PERROR("tracer ctl get_mmap_read_offset");
56591bac
MD
1526 goto end;
1527 }
f02e1e8a 1528 break;
7753dea8
MD
1529 case LTTNG_CONSUMER32_UST:
1530 case LTTNG_CONSUMER64_UST:
ffe60014
DG
1531 mmap_base = lttng_ustctl_get_mmap_base(stream);
1532 if (!mmap_base) {
1533 ERR("read mmap get mmap base for stream %s", stream->name);
994ab360 1534 ret = -EPERM;
ffe60014
DG
1535 goto end;
1536 }
1537 ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset);
56591bac
MD
1538 if (ret != 0) {
1539 PERROR("tracer ctl get_mmap_read_offset");
994ab360 1540 ret = -EINVAL;
56591bac
MD
1541 goto end;
1542 }
f02e1e8a 1543 break;
3bd1e081
MD
1544 default:
1545 ERR("Unknown consumer_data type");
1546 assert(0);
1547 }
b9182dd9 1548
f02e1e8a
DG
1549 /* Handle stream on the relayd if the output is on the network */
1550 if (relayd) {
1551 unsigned long netlen = len;
1552
1553 /*
1554 * Lock the control socket for the complete duration of the function
1555 * since from this point on we will use the socket.
1556 */
1557 if (stream->metadata_flag) {
1558 /* Metadata requires the control socket. */
1559 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1d4dfdef 1560 netlen += sizeof(struct lttcomm_relayd_metadata_payload);
f02e1e8a
DG
1561 }
1562
1d4dfdef 1563 ret = write_relayd_stream_header(stream, netlen, padding, relayd);
994ab360
DG
1564 if (ret < 0) {
1565 relayd_hang_up = 1;
1566 goto write_error;
1567 }
1568 /* Use the returned socket. */
1569 outfd = ret;
f02e1e8a 1570
994ab360
DG
1571 /* Write metadata stream id before payload */
1572 if (stream->metadata_flag) {
1573 ret = write_relayd_metadata_id(outfd, stream, relayd, padding);
1574 if (ret < 0) {
8994307f
DG
1575 relayd_hang_up = 1;
1576 goto write_error;
1577 }
f02e1e8a 1578 }
1d4dfdef
DG
1579 } else {
1580 /* No streaming, we have to set the len with the full padding */
1581 len += padding;
1624d5b7
JD
1582
1583 /*
1584 * Check if we need to change the tracefile before writing the packet.
1585 */
1586 if (stream->chan->tracefile_size > 0 &&
1587 (stream->tracefile_size_current + len) >
1588 stream->chan->tracefile_size) {
fe4477ee
JD
1589 ret = utils_rotate_stream_file(stream->chan->pathname,
1590 stream->name, stream->chan->tracefile_size,
1591 stream->chan->tracefile_count, stream->uid, stream->gid,
309167d2
JD
1592 stream->out_fd, &(stream->tracefile_count_current),
1593 &stream->out_fd);
1624d5b7
JD
1594 if (ret < 0) {
1595 ERR("Rotating output file");
1596 goto end;
1597 }
309167d2
JD
1598 outfd = stream->out_fd;
1599
1600 if (stream->index_fd >= 0) {
1601 ret = index_create_file(stream->chan->pathname,
1602 stream->name, stream->uid, stream->gid,
1603 stream->chan->tracefile_size,
1604 stream->tracefile_count_current);
1605 if (ret < 0) {
1606 goto end;
1607 }
1608 stream->index_fd = ret;
1609 }
1610
a6976990
DG
1611 /* Reset current size because we just perform a rotation. */
1612 stream->tracefile_size_current = 0;
a1ae300f
JD
1613 stream->out_fd_offset = 0;
1614 orig_offset = 0;
1624d5b7
JD
1615 }
1616 stream->tracefile_size_current += len;
309167d2
JD
1617 if (index) {
1618 index->offset = htobe64(stream->out_fd_offset);
1619 }
f02e1e8a
DG
1620 }
1621
d02b8372
DG
1622 /*
1623 * This call guarantee that len or less is returned. It's impossible to
1624 * receive a ret value that is bigger than len.
1625 */
1626 ret = lttng_write(outfd, mmap_base + mmap_offset, len);
1627 DBG("Consumer mmap write() ret %zd (len %lu)", ret, len);
1628 if (ret < 0 || ((size_t) ret != len)) {
1629 /*
1630 * Report error to caller if nothing was written else at least send the
1631 * amount written.
1632 */
1633 if (ret < 0) {
994ab360 1634 ret = -errno;
f02e1e8a 1635 }
994ab360 1636 relayd_hang_up = 1;
f02e1e8a 1637
d02b8372 1638 /* Socket operation failed. We consider the relayd dead */
994ab360 1639 if (errno == EPIPE || errno == EINVAL || errno == EBADF) {
d02b8372
DG
1640 /*
1641 * This is possible if the fd is closed on the other side
1642 * (outfd) or any write problem. It can be verbose a bit for a
1643 * normal execution if for instance the relayd is stopped
1644 * abruptly. This can happen so set this to a DBG statement.
1645 */
1646 DBG("Consumer mmap write detected relayd hang up");
994ab360
DG
1647 } else {
1648 /* Unhandled error, print it and stop function right now. */
1649 PERROR("Error in write mmap (ret %zd != len %lu)", ret, len);
f02e1e8a 1650 }
994ab360 1651 goto write_error;
d02b8372
DG
1652 }
1653 stream->output_written += ret;
d02b8372
DG
1654
1655 /* This call is useless on a socket so better save a syscall. */
1656 if (!relayd) {
1657 /* This won't block, but will start writeout asynchronously */
1658 lttng_sync_file_range(outfd, stream->out_fd_offset, len,
1659 SYNC_FILE_RANGE_WRITE);
1660 stream->out_fd_offset += len;
f02e1e8a
DG
1661 }
1662 lttng_consumer_sync_trace_file(stream, orig_offset);
1663
8994307f
DG
1664write_error:
1665 /*
1666 * This is a special case that the relayd has closed its socket. Let's
1667 * cleanup the relayd object and all associated streams.
1668 */
1669 if (relayd && relayd_hang_up) {
1670 cleanup_relayd(relayd, ctx);
1671 }
1672
f02e1e8a
DG
1673end:
1674 /* Unlock only if ctrl socket used */
1675 if (relayd && stream->metadata_flag) {
1676 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1677 }
1678
1679 rcu_read_unlock();
994ab360 1680 return ret;
3bd1e081
MD
1681}
1682
1683/*
1684 * Splice the data from the ring buffer to the tracefile.
1685 *
79d4ffb7
DG
1686 * It must be called with the stream lock held.
1687 *
3bd1e081
MD
1688 * Returns the number of bytes spliced.
1689 */
4078b776 1690ssize_t lttng_consumer_on_read_subbuffer_splice(
3bd1e081 1691 struct lttng_consumer_local_data *ctx,
1d4dfdef 1692 struct lttng_consumer_stream *stream, unsigned long len,
309167d2 1693 unsigned long padding,
50adc264 1694 struct ctf_packet_index *index)
3bd1e081 1695{
f02e1e8a
DG
1696 ssize_t ret = 0, written = 0, ret_splice = 0;
1697 loff_t offset = 0;
1698 off_t orig_offset = stream->out_fd_offset;
1699 int fd = stream->wait_fd;
1700 /* Default is on the disk */
1701 int outfd = stream->out_fd;
f02e1e8a 1702 struct consumer_relayd_sock_pair *relayd = NULL;
fb3a43a9 1703 int *splice_pipe;
8994307f 1704 unsigned int relayd_hang_up = 0;
f02e1e8a 1705
3bd1e081
MD
1706 switch (consumer_data.type) {
1707 case LTTNG_CONSUMER_KERNEL:
f02e1e8a 1708 break;
7753dea8
MD
1709 case LTTNG_CONSUMER32_UST:
1710 case LTTNG_CONSUMER64_UST:
f02e1e8a 1711 /* Not supported for user space tracing */
3bd1e081
MD
1712 return -ENOSYS;
1713 default:
1714 ERR("Unknown consumer_data type");
1715 assert(0);
3bd1e081
MD
1716 }
1717
f02e1e8a
DG
1718 /* RCU lock for the relayd pointer */
1719 rcu_read_lock();
1720
1721 /* Flag that the current stream if set for network streaming. */
da009f2c 1722 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1723 relayd = consumer_find_relayd(stream->net_seq_idx);
1724 if (relayd == NULL) {
ad0b0d23 1725 written = -ret;
f02e1e8a
DG
1726 goto end;
1727 }
1728 }
a2361a61 1729 splice_pipe = stream->splice_pipe;
fb3a43a9 1730
f02e1e8a 1731 /* Write metadata stream id before payload */
1d4dfdef 1732 if (relayd) {
ad0b0d23 1733 unsigned long total_len = len;
f02e1e8a 1734
1d4dfdef
DG
1735 if (stream->metadata_flag) {
1736 /*
1737 * Lock the control socket for the complete duration of the function
1738 * since from this point on we will use the socket.
1739 */
1740 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1741
1742 ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd,
1743 padding);
1744 if (ret < 0) {
1745 written = ret;
ad0b0d23
DG
1746 relayd_hang_up = 1;
1747 goto write_error;
1d4dfdef
DG
1748 }
1749
1750 total_len += sizeof(struct lttcomm_relayd_metadata_payload);
1751 }
1752
1753 ret = write_relayd_stream_header(stream, total_len, padding, relayd);
ad0b0d23
DG
1754 if (ret < 0) {
1755 written = ret;
1756 relayd_hang_up = 1;
1757 goto write_error;
f02e1e8a 1758 }
ad0b0d23
DG
1759 /* Use the returned socket. */
1760 outfd = ret;
1d4dfdef
DG
1761 } else {
1762 /* No streaming, we have to set the len with the full padding */
1763 len += padding;
1624d5b7
JD
1764
1765 /*
1766 * Check if we need to change the tracefile before writing the packet.
1767 */
1768 if (stream->chan->tracefile_size > 0 &&
1769 (stream->tracefile_size_current + len) >
1770 stream->chan->tracefile_size) {
fe4477ee
JD
1771 ret = utils_rotate_stream_file(stream->chan->pathname,
1772 stream->name, stream->chan->tracefile_size,
1773 stream->chan->tracefile_count, stream->uid, stream->gid,
309167d2
JD
1774 stream->out_fd, &(stream->tracefile_count_current),
1775 &stream->out_fd);
1624d5b7 1776 if (ret < 0) {
ad0b0d23 1777 written = ret;
1624d5b7
JD
1778 ERR("Rotating output file");
1779 goto end;
1780 }
309167d2
JD
1781 outfd = stream->out_fd;
1782
1783 if (stream->index_fd >= 0) {
1784 ret = index_create_file(stream->chan->pathname,
1785 stream->name, stream->uid, stream->gid,
1786 stream->chan->tracefile_size,
1787 stream->tracefile_count_current);
1788 if (ret < 0) {
ad0b0d23 1789 written = ret;
309167d2
JD
1790 goto end;
1791 }
1792 stream->index_fd = ret;
1793 }
1794
a6976990
DG
1795 /* Reset current size because we just perform a rotation. */
1796 stream->tracefile_size_current = 0;
a1ae300f
JD
1797 stream->out_fd_offset = 0;
1798 orig_offset = 0;
1624d5b7
JD
1799 }
1800 stream->tracefile_size_current += len;
309167d2 1801 index->offset = htobe64(stream->out_fd_offset);
f02e1e8a
DG
1802 }
1803
1804 while (len > 0) {
1d4dfdef
DG
1805 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1806 (unsigned long)offset, len, fd, splice_pipe[1]);
fb3a43a9 1807 ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len,
f02e1e8a
DG
1808 SPLICE_F_MOVE | SPLICE_F_MORE);
1809 DBG("splice chan to pipe, ret %zd", ret_splice);
1810 if (ret_splice < 0) {
d02b8372 1811 ret = errno;
ad0b0d23 1812 written = -ret;
d02b8372 1813 PERROR("Error in relay splice");
f02e1e8a
DG
1814 goto splice_error;
1815 }
1816
1817 /* Handle stream on the relayd if the output is on the network */
ad0b0d23
DG
1818 if (relayd && stream->metadata_flag) {
1819 size_t metadata_payload_size =
1820 sizeof(struct lttcomm_relayd_metadata_payload);
1821
1822 /* Update counter to fit the spliced data */
1823 ret_splice += metadata_payload_size;
1824 len += metadata_payload_size;
1825 /*
1826 * We do this so the return value can match the len passed as
1827 * argument to this function.
1828 */
1829 written -= metadata_payload_size;
f02e1e8a
DG
1830 }
1831
1832 /* Splice data out */
fb3a43a9 1833 ret_splice = splice(splice_pipe[0], NULL, outfd, NULL,
f02e1e8a 1834 ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE);
a2361a61
JD
1835 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1836 outfd, ret_splice);
f02e1e8a 1837 if (ret_splice < 0) {
d02b8372 1838 ret = errno;
ad0b0d23
DG
1839 written = -ret;
1840 relayd_hang_up = 1;
1841 goto write_error;
f02e1e8a 1842 } else if (ret_splice > len) {
d02b8372
DG
1843 /*
1844 * We don't expect this code path to be executed but you never know
1845 * so this is an extra protection agains a buggy splice().
1846 */
f02e1e8a 1847 ret = errno;
ad0b0d23 1848 written += ret_splice;
d02b8372
DG
1849 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice,
1850 len);
f02e1e8a 1851 goto splice_error;
d02b8372
DG
1852 } else {
1853 /* All good, update current len and continue. */
1854 len -= ret_splice;
f02e1e8a 1855 }
f02e1e8a
DG
1856
1857 /* This call is useless on a socket so better save a syscall. */
1858 if (!relayd) {
1859 /* This won't block, but will start writeout asynchronously */
1860 lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice,
1861 SYNC_FILE_RANGE_WRITE);
1862 stream->out_fd_offset += ret_splice;
1863 }
e5d1a9b3 1864 stream->output_written += ret_splice;
f02e1e8a
DG
1865 written += ret_splice;
1866 }
1867 lttng_consumer_sync_trace_file(stream, orig_offset);
f02e1e8a
DG
1868 goto end;
1869
8994307f
DG
1870write_error:
1871 /*
1872 * This is a special case that the relayd has closed its socket. Let's
1873 * cleanup the relayd object and all associated streams.
1874 */
1875 if (relayd && relayd_hang_up) {
1876 cleanup_relayd(relayd, ctx);
1877 /* Skip splice error so the consumer does not fail */
1878 goto end;
1879 }
1880
f02e1e8a
DG
1881splice_error:
1882 /* send the appropriate error description to sessiond */
1883 switch (ret) {
f02e1e8a 1884 case EINVAL:
f73fabfd 1885 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL);
f02e1e8a
DG
1886 break;
1887 case ENOMEM:
f73fabfd 1888 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM);
f02e1e8a
DG
1889 break;
1890 case ESPIPE:
f73fabfd 1891 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE);
f02e1e8a
DG
1892 break;
1893 }
1894
1895end:
1896 if (relayd && stream->metadata_flag) {
1897 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1898 }
1899
1900 rcu_read_unlock();
1901 return written;
3bd1e081
MD
1902}
1903
1904/*
1905 * Take a snapshot for a specific fd
1906 *
1907 * Returns 0 on success, < 0 on error
1908 */
ffe60014 1909int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream)
3bd1e081
MD
1910{
1911 switch (consumer_data.type) {
1912 case LTTNG_CONSUMER_KERNEL:
ffe60014 1913 return lttng_kconsumer_take_snapshot(stream);
7753dea8
MD
1914 case LTTNG_CONSUMER32_UST:
1915 case LTTNG_CONSUMER64_UST:
ffe60014 1916 return lttng_ustconsumer_take_snapshot(stream);
3bd1e081
MD
1917 default:
1918 ERR("Unknown consumer_data type");
1919 assert(0);
1920 return -ENOSYS;
1921 }
3bd1e081
MD
1922}
1923
1924/*
1925 * Get the produced position
1926 *
1927 * Returns 0 on success, < 0 on error
1928 */
ffe60014 1929int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
3bd1e081
MD
1930 unsigned long *pos)
1931{
1932 switch (consumer_data.type) {
1933 case LTTNG_CONSUMER_KERNEL:
ffe60014 1934 return lttng_kconsumer_get_produced_snapshot(stream, pos);
7753dea8
MD
1935 case LTTNG_CONSUMER32_UST:
1936 case LTTNG_CONSUMER64_UST:
ffe60014 1937 return lttng_ustconsumer_get_produced_snapshot(stream, pos);
3bd1e081
MD
1938 default:
1939 ERR("Unknown consumer_data type");
1940 assert(0);
1941 return -ENOSYS;
1942 }
1943}
1944
1945int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1946 int sock, struct pollfd *consumer_sockpoll)
1947{
1948 switch (consumer_data.type) {
1949 case LTTNG_CONSUMER_KERNEL:
1950 return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
7753dea8
MD
1951 case LTTNG_CONSUMER32_UST:
1952 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
1953 return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
1954 default:
1955 ERR("Unknown consumer_data type");
1956 assert(0);
1957 return -ENOSYS;
1958 }
1959}
1960
6d574024 1961void lttng_consumer_close_all_metadata(void)
d88aee68
DG
1962{
1963 switch (consumer_data.type) {
1964 case LTTNG_CONSUMER_KERNEL:
1965 /*
1966 * The Kernel consumer has a different metadata scheme so we don't
1967 * close anything because the stream will be closed by the session
1968 * daemon.
1969 */
1970 break;
1971 case LTTNG_CONSUMER32_UST:
1972 case LTTNG_CONSUMER64_UST:
1973 /*
1974 * Close all metadata streams. The metadata hash table is passed and
1975 * this call iterates over it by closing all wakeup fd. This is safe
1976 * because at this point we are sure that the metadata producer is
1977 * either dead or blocked.
1978 */
6d574024 1979 lttng_ustconsumer_close_all_metadata(metadata_ht);
d88aee68
DG
1980 break;
1981 default:
1982 ERR("Unknown consumer_data type");
1983 assert(0);
1984 }
1985}
1986
fb3a43a9
DG
1987/*
1988 * Clean up a metadata stream and free its memory.
1989 */
e316aad5
DG
1990void consumer_del_metadata_stream(struct lttng_consumer_stream *stream,
1991 struct lttng_ht *ht)
fb3a43a9 1992{
e316aad5 1993 struct lttng_consumer_channel *free_chan = NULL;
fb3a43a9
DG
1994
1995 assert(stream);
1996 /*
1997 * This call should NEVER receive regular stream. It must always be
1998 * metadata stream and this is crucial for data structure synchronization.
1999 */
2000 assert(stream->metadata_flag);
2001
e316aad5
DG
2002 DBG3("Consumer delete metadata stream %d", stream->wait_fd);
2003
74251bb8 2004 pthread_mutex_lock(&consumer_data.lock);
a9838785 2005 pthread_mutex_lock(&stream->chan->lock);
8994307f
DG
2006 pthread_mutex_lock(&stream->lock);
2007
6d574024
DG
2008 /* Remove any reference to that stream. */
2009 consumer_stream_delete(stream, ht);
ca22feea 2010
6d574024
DG
2011 /* Close down everything including the relayd if one. */
2012 consumer_stream_close(stream);
2013 /* Destroy tracer buffers of the stream. */
2014 consumer_stream_destroy_buffers(stream);
fb3a43a9
DG
2015
2016 /* Atomically decrement channel refcount since other threads can use it. */
f2ad556d 2017 if (!uatomic_sub_return(&stream->chan->refcount, 1)
ffe60014 2018 && !uatomic_read(&stream->chan->nb_init_stream_left)) {
c30aaa51 2019 /* Go for channel deletion! */
e316aad5 2020 free_chan = stream->chan;
fb3a43a9
DG
2021 }
2022
73811ecc
DG
2023 /*
2024 * Nullify the stream reference so it is not used after deletion. The
6d574024
DG
2025 * channel lock MUST be acquired before being able to check for a NULL
2026 * pointer value.
73811ecc
DG
2027 */
2028 stream->chan->metadata_stream = NULL;
2029
8994307f 2030 pthread_mutex_unlock(&stream->lock);
a9838785 2031 pthread_mutex_unlock(&stream->chan->lock);
74251bb8 2032 pthread_mutex_unlock(&consumer_data.lock);
e316aad5
DG
2033
2034 if (free_chan) {
2035 consumer_del_channel(free_chan);
2036 }
2037
6d574024 2038 consumer_stream_free(stream);
fb3a43a9
DG
2039}
2040
2041/*
2042 * Action done with the metadata stream when adding it to the consumer internal
2043 * data structures to handle it.
2044 */
5ab66908 2045int consumer_add_metadata_stream(struct lttng_consumer_stream *stream)
fb3a43a9 2046{
5ab66908 2047 struct lttng_ht *ht = metadata_ht;
e316aad5 2048 int ret = 0;
76082088 2049 struct lttng_ht_iter iter;
d88aee68 2050 struct lttng_ht_node_u64 *node;
fb3a43a9 2051
e316aad5
DG
2052 assert(stream);
2053 assert(ht);
2054
d88aee68 2055 DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key);
e316aad5
DG
2056
2057 pthread_mutex_lock(&consumer_data.lock);
a9838785 2058 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 2059 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 2060 pthread_mutex_lock(&stream->lock);
e316aad5 2061
e316aad5
DG
2062 /*
2063 * From here, refcounts are updated so be _careful_ when returning an error
2064 * after this point.
2065 */
2066
fb3a43a9 2067 rcu_read_lock();
76082088
DG
2068
2069 /*
2070 * Lookup the stream just to make sure it does not exist in our internal
2071 * state. This should NEVER happen.
2072 */
d88aee68
DG
2073 lttng_ht_lookup(ht, &stream->key, &iter);
2074 node = lttng_ht_iter_get_node_u64(&iter);
76082088
DG
2075 assert(!node);
2076
e316aad5 2077 /*
ffe60014
DG
2078 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2079 * in terms of destroying the associated channel, because the action that
e316aad5
DG
2080 * causes the count to become 0 also causes a stream to be added. The
2081 * channel deletion will thus be triggered by the following removal of this
2082 * stream.
2083 */
ffe60014 2084 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
2085 /* Increment refcount before decrementing nb_init_stream_left */
2086 cmm_smp_wmb();
ffe60014 2087 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
2088 }
2089
d88aee68 2090 lttng_ht_add_unique_u64(ht, &stream->node);
ca22feea 2091
d8ef542d
MD
2092 lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht,
2093 &stream->node_channel_id);
2094
ca22feea
DG
2095 /*
2096 * Add stream to the stream_list_ht of the consumer data. No need to steal
2097 * the key since the HT does not use it and we allow to add redundant keys
2098 * into this table.
2099 */
d88aee68 2100 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 2101
fb3a43a9 2102 rcu_read_unlock();
e316aad5 2103
2e818a6a 2104 pthread_mutex_unlock(&stream->lock);
a9838785 2105 pthread_mutex_unlock(&stream->chan->lock);
ec6ea7d0 2106 pthread_mutex_unlock(&stream->chan->timer_lock);
e316aad5
DG
2107 pthread_mutex_unlock(&consumer_data.lock);
2108 return ret;
fb3a43a9
DG
2109}
2110
8994307f
DG
2111/*
2112 * Delete data stream that are flagged for deletion (endpoint_status).
2113 */
2114static void validate_endpoint_status_data_stream(void)
2115{
2116 struct lttng_ht_iter iter;
2117 struct lttng_consumer_stream *stream;
2118
2119 DBG("Consumer delete flagged data stream");
2120
2121 rcu_read_lock();
2122 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
2123 /* Validate delete flag of the stream */
79d4ffb7 2124 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2125 continue;
2126 }
2127 /* Delete it right now */
2128 consumer_del_stream(stream, data_ht);
2129 }
2130 rcu_read_unlock();
2131}
2132
2133/*
2134 * Delete metadata stream that are flagged for deletion (endpoint_status).
2135 */
2136static void validate_endpoint_status_metadata_stream(
2137 struct lttng_poll_event *pollset)
2138{
2139 struct lttng_ht_iter iter;
2140 struct lttng_consumer_stream *stream;
2141
2142 DBG("Consumer delete flagged metadata stream");
2143
2144 assert(pollset);
2145
2146 rcu_read_lock();
2147 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
2148 /* Validate delete flag of the stream */
79d4ffb7 2149 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2150 continue;
2151 }
2152 /*
2153 * Remove from pollset so the metadata thread can continue without
2154 * blocking on a deleted stream.
2155 */
2156 lttng_poll_del(pollset, stream->wait_fd);
2157
2158 /* Delete it right now */
2159 consumer_del_metadata_stream(stream, metadata_ht);
2160 }
2161 rcu_read_unlock();
2162}
2163
fb3a43a9
DG
2164/*
2165 * Thread polls on metadata file descriptor and write them on disk or on the
2166 * network.
2167 */
7d980def 2168void *consumer_thread_metadata_poll(void *data)
fb3a43a9 2169{
1fc79fb4 2170 int ret, i, pollfd, err = -1;
fb3a43a9 2171 uint32_t revents, nb_fd;
e316aad5 2172 struct lttng_consumer_stream *stream = NULL;
fb3a43a9 2173 struct lttng_ht_iter iter;
d88aee68 2174 struct lttng_ht_node_u64 *node;
fb3a43a9
DG
2175 struct lttng_poll_event events;
2176 struct lttng_consumer_local_data *ctx = data;
2177 ssize_t len;
2178
2179 rcu_register_thread();
2180
1fc79fb4
MD
2181 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA);
2182
2d57de81
MD
2183 if (testpoint(consumerd_thread_metadata)) {
2184 goto error_testpoint;
2185 }
2186
9ce5646a
MD
2187 health_code_update();
2188
fb3a43a9
DG
2189 DBG("Thread metadata poll started");
2190
fb3a43a9
DG
2191 /* Size is set to 1 for the consumer_metadata pipe */
2192 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2193 if (ret < 0) {
2194 ERR("Poll set creation failed");
d8ef542d 2195 goto end_poll;
fb3a43a9
DG
2196 }
2197
13886d2d
DG
2198 ret = lttng_poll_add(&events,
2199 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN);
fb3a43a9
DG
2200 if (ret < 0) {
2201 goto end;
2202 }
2203
2204 /* Main loop */
2205 DBG("Metadata main loop started");
2206
2207 while (1) {
fb3a43a9 2208restart:
7fa2082e 2209 health_code_update();
9ce5646a 2210 health_poll_entry();
7fa2082e 2211 DBG("Metadata poll wait");
fb3a43a9 2212 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2213 DBG("Metadata poll return from wait with %d fd(s)",
2214 LTTNG_POLL_GETNB(&events));
9ce5646a 2215 health_poll_exit();
fb3a43a9
DG
2216 DBG("Metadata event catched in thread");
2217 if (ret < 0) {
2218 if (errno == EINTR) {
e316aad5 2219 ERR("Poll EINTR catched");
fb3a43a9
DG
2220 goto restart;
2221 }
d9607cd7
MD
2222 if (LTTNG_POLL_GETNB(&events) == 0) {
2223 err = 0; /* All is OK */
2224 }
2225 goto end;
fb3a43a9
DG
2226 }
2227
0d9c5d77
DG
2228 nb_fd = ret;
2229
e316aad5 2230 /* From here, the event is a metadata wait fd */
fb3a43a9 2231 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2232 health_code_update();
2233
fb3a43a9
DG
2234 revents = LTTNG_POLL_GETEV(&events, i);
2235 pollfd = LTTNG_POLL_GETFD(&events, i);
2236
fd20dac9
MD
2237 if (!revents) {
2238 /* No activity for this FD (poll implementation). */
2239 continue;
2240 }
2241
13886d2d 2242 if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) {
4adabd61 2243 if (revents & (LPOLLERR | LPOLLHUP )) {
fb3a43a9
DG
2244 DBG("Metadata thread pipe hung up");
2245 /*
2246 * Remove the pipe from the poll set and continue the loop
2247 * since their might be data to consume.
2248 */
13886d2d
DG
2249 lttng_poll_del(&events,
2250 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2251 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
fb3a43a9
DG
2252 continue;
2253 } else if (revents & LPOLLIN) {
13886d2d
DG
2254 ssize_t pipe_len;
2255
2256 pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe,
2257 &stream, sizeof(stream));
6cd525e8
MD
2258 if (pipe_len < sizeof(stream)) {
2259 PERROR("read metadata stream");
fb3a43a9 2260 /*
13886d2d 2261 * Continue here to handle the rest of the streams.
fb3a43a9
DG
2262 */
2263 continue;
2264 }
2265
8994307f
DG
2266 /* A NULL stream means that the state has changed. */
2267 if (stream == NULL) {
2268 /* Check for deleted streams. */
2269 validate_endpoint_status_metadata_stream(&events);
3714380f 2270 goto restart;
8994307f
DG
2271 }
2272
fb3a43a9
DG
2273 DBG("Adding metadata stream %d to poll set",
2274 stream->wait_fd);
2275
fb3a43a9
DG
2276 /* Add metadata stream to the global poll events list */
2277 lttng_poll_add(&events, stream->wait_fd,
6d574024 2278 LPOLLIN | LPOLLPRI | LPOLLHUP);
fb3a43a9
DG
2279 }
2280
e316aad5 2281 /* Handle other stream */
fb3a43a9
DG
2282 continue;
2283 }
2284
d09e1200 2285 rcu_read_lock();
d88aee68
DG
2286 {
2287 uint64_t tmp_id = (uint64_t) pollfd;
2288
2289 lttng_ht_lookup(metadata_ht, &tmp_id, &iter);
2290 }
2291 node = lttng_ht_iter_get_node_u64(&iter);
e316aad5 2292 assert(node);
fb3a43a9
DG
2293
2294 stream = caa_container_of(node, struct lttng_consumer_stream,
58b1f425 2295 node);
fb3a43a9 2296
e316aad5 2297 /* Check for error event */
4adabd61 2298 if (revents & (LPOLLERR | LPOLLHUP)) {
e316aad5 2299 DBG("Metadata fd %d is hup|err.", pollfd);
fb3a43a9
DG
2300 if (!stream->hangup_flush_done
2301 && (consumer_data.type == LTTNG_CONSUMER32_UST
2302 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2303 DBG("Attempting to flush and consume the UST buffers");
2304 lttng_ustconsumer_on_stream_hangup(stream);
2305
2306 /* We just flushed the stream now read it. */
4bb94b75 2307 do {
9ce5646a
MD
2308 health_code_update();
2309
4bb94b75
DG
2310 len = ctx->on_buffer_ready(stream, ctx);
2311 /*
2312 * We don't check the return value here since if we get
2313 * a negative len, it means an error occured thus we
2314 * simply remove it from the poll set and free the
2315 * stream.
2316 */
2317 } while (len > 0);
fb3a43a9
DG
2318 }
2319
fb3a43a9 2320 lttng_poll_del(&events, stream->wait_fd);
e316aad5
DG
2321 /*
2322 * This call update the channel states, closes file descriptors
2323 * and securely free the stream.
2324 */
2325 consumer_del_metadata_stream(stream, metadata_ht);
2326 } else if (revents & (LPOLLIN | LPOLLPRI)) {
2327 /* Get the data out of the metadata file descriptor */
2328 DBG("Metadata available on fd %d", pollfd);
2329 assert(stream->wait_fd == pollfd);
2330
04ef1097 2331 do {
9ce5646a
MD
2332 health_code_update();
2333
04ef1097
MD
2334 len = ctx->on_buffer_ready(stream, ctx);
2335 /*
2336 * We don't check the return value here since if we get
2337 * a negative len, it means an error occured thus we
2338 * simply remove it from the poll set and free the
2339 * stream.
2340 */
2341 } while (len > 0);
2342
e316aad5 2343 /* It's ok to have an unavailable sub-buffer */
b64403e3 2344 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2345 /* Clean up stream from consumer and free it. */
2346 lttng_poll_del(&events, stream->wait_fd);
2347 consumer_del_metadata_stream(stream, metadata_ht);
e316aad5 2348 }
fb3a43a9 2349 }
e316aad5
DG
2350
2351 /* Release RCU lock for the stream looked up */
d09e1200 2352 rcu_read_unlock();
fb3a43a9
DG
2353 }
2354 }
2355
1fc79fb4
MD
2356 /* All is OK */
2357 err = 0;
fb3a43a9
DG
2358end:
2359 DBG("Metadata poll thread exiting");
fb3a43a9 2360
d8ef542d
MD
2361 lttng_poll_clean(&events);
2362end_poll:
2d57de81 2363error_testpoint:
1fc79fb4
MD
2364 if (err) {
2365 health_error();
2366 ERR("Health error occurred in %s", __func__);
2367 }
2368 health_unregister(health_consumerd);
fb3a43a9
DG
2369 rcu_unregister_thread();
2370 return NULL;
2371}
2372
3bd1e081 2373/*
e4421fec 2374 * This thread polls the fds in the set to consume the data and write
3bd1e081
MD
2375 * it to tracefile if necessary.
2376 */
7d980def 2377void *consumer_thread_data_poll(void *data)
3bd1e081 2378{
1fc79fb4 2379 int num_rdy, num_hup, high_prio, ret, i, err = -1;
3bd1e081
MD
2380 struct pollfd *pollfd = NULL;
2381 /* local view of the streams */
c869f647 2382 struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL;
3bd1e081
MD
2383 /* local view of consumer_data.fds_count */
2384 int nb_fd = 0;
3bd1e081 2385 struct lttng_consumer_local_data *ctx = data;
00e2e675 2386 ssize_t len;
3bd1e081 2387
e7b994a3
DG
2388 rcu_register_thread();
2389
1fc79fb4
MD
2390 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA);
2391
2d57de81
MD
2392 if (testpoint(consumerd_thread_data)) {
2393 goto error_testpoint;
2394 }
2395
9ce5646a
MD
2396 health_code_update();
2397
4df6c8cb
MD
2398 local_stream = zmalloc(sizeof(struct lttng_consumer_stream *));
2399 if (local_stream == NULL) {
2400 PERROR("local_stream malloc");
2401 goto end;
2402 }
3bd1e081
MD
2403
2404 while (1) {
9ce5646a
MD
2405 health_code_update();
2406
3bd1e081
MD
2407 high_prio = 0;
2408 num_hup = 0;
2409
2410 /*
e4421fec 2411 * the fds set has been updated, we need to update our
3bd1e081
MD
2412 * local array as well
2413 */
2414 pthread_mutex_lock(&consumer_data.lock);
2415 if (consumer_data.need_update) {
0e428499
DG
2416 free(pollfd);
2417 pollfd = NULL;
2418
2419 free(local_stream);
2420 local_stream = NULL;
3bd1e081 2421
02b3d176
DG
2422 /*
2423 * Allocate for all fds +1 for the consumer_data_pipe and +1 for
2424 * wake up pipe.
2425 */
2426 pollfd = zmalloc((consumer_data.stream_count + 2) * sizeof(struct pollfd));
3bd1e081 2427 if (pollfd == NULL) {
7a57cf92 2428 PERROR("pollfd malloc");
3bd1e081
MD
2429 pthread_mutex_unlock(&consumer_data.lock);
2430 goto end;
2431 }
2432
02b3d176 2433 local_stream = zmalloc((consumer_data.stream_count + 2) *
747f8642 2434 sizeof(struct lttng_consumer_stream *));
3bd1e081 2435 if (local_stream == NULL) {
7a57cf92 2436 PERROR("local_stream malloc");
3bd1e081
MD
2437 pthread_mutex_unlock(&consumer_data.lock);
2438 goto end;
2439 }
ffe60014 2440 ret = update_poll_array(ctx, &pollfd, local_stream,
43c34bc3 2441 data_ht);
3bd1e081
MD
2442 if (ret < 0) {
2443 ERR("Error in allocating pollfd or local_outfds");
f73fabfd 2444 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2445 pthread_mutex_unlock(&consumer_data.lock);
2446 goto end;
2447 }
2448 nb_fd = ret;
2449 consumer_data.need_update = 0;
2450 }
2451 pthread_mutex_unlock(&consumer_data.lock);
2452
4078b776
MD
2453 /* No FDs and consumer_quit, consumer_cleanup the thread */
2454 if (nb_fd == 0 && consumer_quit == 1) {
1fc79fb4 2455 err = 0; /* All is OK */
4078b776
MD
2456 goto end;
2457 }
3bd1e081 2458 /* poll on the array of fds */
88f2b785 2459 restart:
02b3d176 2460 DBG("polling on %d fd", nb_fd + 2);
9ce5646a 2461 health_poll_entry();
02b3d176 2462 num_rdy = poll(pollfd, nb_fd + 2, -1);
9ce5646a 2463 health_poll_exit();
3bd1e081
MD
2464 DBG("poll num_rdy : %d", num_rdy);
2465 if (num_rdy == -1) {
88f2b785
MD
2466 /*
2467 * Restart interrupted system call.
2468 */
2469 if (errno == EINTR) {
2470 goto restart;
2471 }
7a57cf92 2472 PERROR("Poll error");
f73fabfd 2473 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2474 goto end;
2475 } else if (num_rdy == 0) {
2476 DBG("Polling thread timed out");
2477 goto end;
2478 }
2479
3bd1e081 2480 /*
50f8ae69 2481 * If the consumer_data_pipe triggered poll go directly to the
00e2e675
DG
2482 * beginning of the loop to update the array. We want to prioritize
2483 * array update over low-priority reads.
3bd1e081 2484 */
509bb1cf 2485 if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) {
ab30f567 2486 ssize_t pipe_readlen;
04fdd819 2487
50f8ae69 2488 DBG("consumer_data_pipe wake up");
acdb9057
DG
2489 pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe,
2490 &new_stream, sizeof(new_stream));
6cd525e8
MD
2491 if (pipe_readlen < sizeof(new_stream)) {
2492 PERROR("Consumer data pipe");
23f5f35d
DG
2493 /* Continue so we can at least handle the current stream(s). */
2494 continue;
2495 }
c869f647
DG
2496
2497 /*
2498 * If the stream is NULL, just ignore it. It's also possible that
2499 * the sessiond poll thread changed the consumer_quit state and is
2500 * waking us up to test it.
2501 */
2502 if (new_stream == NULL) {
8994307f 2503 validate_endpoint_status_data_stream();
c869f647
DG
2504 continue;
2505 }
2506
c869f647 2507 /* Continue to update the local streams and handle prio ones */
3bd1e081
MD
2508 continue;
2509 }
2510
02b3d176
DG
2511 /* Handle wakeup pipe. */
2512 if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) {
2513 char dummy;
2514 ssize_t pipe_readlen;
2515
2516 pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy,
2517 sizeof(dummy));
2518 if (pipe_readlen < 0) {
2519 PERROR("Consumer data wakeup pipe");
2520 }
2521 /* We've been awakened to handle stream(s). */
2522 ctx->has_wakeup = 0;
2523 }
2524
3bd1e081
MD
2525 /* Take care of high priority channels first. */
2526 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2527 health_code_update();
2528
9617607b
DG
2529 if (local_stream[i] == NULL) {
2530 continue;
2531 }
fb3a43a9 2532 if (pollfd[i].revents & POLLPRI) {
d41f73b7
MD
2533 DBG("Urgent read on fd %d", pollfd[i].fd);
2534 high_prio = 1;
4078b776 2535 len = ctx->on_buffer_ready(local_stream[i], ctx);
d41f73b7 2536 /* it's ok to have an unavailable sub-buffer */
b64403e3 2537 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2538 /* Clean the stream and free it. */
2539 consumer_del_stream(local_stream[i], data_ht);
9617607b 2540 local_stream[i] = NULL;
4078b776
MD
2541 } else if (len > 0) {
2542 local_stream[i]->data_read = 1;
d41f73b7 2543 }
3bd1e081
MD
2544 }
2545 }
2546
4078b776
MD
2547 /*
2548 * If we read high prio channel in this loop, try again
2549 * for more high prio data.
2550 */
2551 if (high_prio) {
3bd1e081
MD
2552 continue;
2553 }
2554
2555 /* Take care of low priority channels. */
4078b776 2556 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2557 health_code_update();
2558
9617607b
DG
2559 if (local_stream[i] == NULL) {
2560 continue;
2561 }
4078b776 2562 if ((pollfd[i].revents & POLLIN) ||
02b3d176
DG
2563 local_stream[i]->hangup_flush_done ||
2564 local_stream[i]->has_data) {
4078b776
MD
2565 DBG("Normal read on fd %d", pollfd[i].fd);
2566 len = ctx->on_buffer_ready(local_stream[i], ctx);
2567 /* it's ok to have an unavailable sub-buffer */
b64403e3 2568 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2569 /* Clean the stream and free it. */
2570 consumer_del_stream(local_stream[i], data_ht);
9617607b 2571 local_stream[i] = NULL;
4078b776
MD
2572 } else if (len > 0) {
2573 local_stream[i]->data_read = 1;
2574 }
2575 }
2576 }
2577
2578 /* Handle hangup and errors */
2579 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2580 health_code_update();
2581
9617607b
DG
2582 if (local_stream[i] == NULL) {
2583 continue;
2584 }
4078b776
MD
2585 if (!local_stream[i]->hangup_flush_done
2586 && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL))
2587 && (consumer_data.type == LTTNG_CONSUMER32_UST
2588 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2589 DBG("fd %d is hup|err|nval. Attempting flush and read.",
9617607b 2590 pollfd[i].fd);
4078b776
MD
2591 lttng_ustconsumer_on_stream_hangup(local_stream[i]);
2592 /* Attempt read again, for the data we just flushed. */
2593 local_stream[i]->data_read = 1;
2594 }
2595 /*
2596 * If the poll flag is HUP/ERR/NVAL and we have
2597 * read no data in this pass, we can remove the
2598 * stream from its hash table.
2599 */
2600 if ((pollfd[i].revents & POLLHUP)) {
2601 DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
2602 if (!local_stream[i]->data_read) {
43c34bc3 2603 consumer_del_stream(local_stream[i], data_ht);
9617607b 2604 local_stream[i] = NULL;
4078b776
MD
2605 num_hup++;
2606 }
2607 } else if (pollfd[i].revents & POLLERR) {
2608 ERR("Error returned in polling fd %d.", pollfd[i].fd);
2609 if (!local_stream[i]->data_read) {
43c34bc3 2610 consumer_del_stream(local_stream[i], data_ht);
9617607b 2611 local_stream[i] = NULL;
4078b776
MD
2612 num_hup++;
2613 }
2614 } else if (pollfd[i].revents & POLLNVAL) {
2615 ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
2616 if (!local_stream[i]->data_read) {
43c34bc3 2617 consumer_del_stream(local_stream[i], data_ht);
9617607b 2618 local_stream[i] = NULL;
4078b776 2619 num_hup++;
3bd1e081
MD
2620 }
2621 }
9617607b
DG
2622 if (local_stream[i] != NULL) {
2623 local_stream[i]->data_read = 0;
2624 }
3bd1e081
MD
2625 }
2626 }
1fc79fb4
MD
2627 /* All is OK */
2628 err = 0;
3bd1e081
MD
2629end:
2630 DBG("polling thread exiting");
0e428499
DG
2631 free(pollfd);
2632 free(local_stream);
fb3a43a9
DG
2633
2634 /*
2635 * Close the write side of the pipe so epoll_wait() in
7d980def
DG
2636 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2637 * read side of the pipe. If we close them both, epoll_wait strangely does
2638 * not return and could create a endless wait period if the pipe is the
2639 * only tracked fd in the poll set. The thread will take care of closing
2640 * the read side.
fb3a43a9 2641 */
13886d2d 2642 (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe);
fb3a43a9 2643
2d57de81 2644error_testpoint:
1fc79fb4
MD
2645 if (err) {
2646 health_error();
2647 ERR("Health error occurred in %s", __func__);
2648 }
2649 health_unregister(health_consumerd);
2650
e7b994a3 2651 rcu_unregister_thread();
3bd1e081
MD
2652 return NULL;
2653}
2654
d8ef542d
MD
2655/*
2656 * Close wake-up end of each stream belonging to the channel. This will
2657 * allow the poll() on the stream read-side to detect when the
2658 * write-side (application) finally closes them.
2659 */
2660static
2661void consumer_close_channel_streams(struct lttng_consumer_channel *channel)
2662{
2663 struct lttng_ht *ht;
2664 struct lttng_consumer_stream *stream;
2665 struct lttng_ht_iter iter;
2666
2667 ht = consumer_data.stream_per_chan_id_ht;
2668
2669 rcu_read_lock();
2670 cds_lfht_for_each_entry_duplicate(ht->ht,
2671 ht->hash_fct(&channel->key, lttng_ht_seed),
2672 ht->match_fct, &channel->key,
2673 &iter.iter, stream, node_channel_id.node) {
f2ad556d
MD
2674 /*
2675 * Protect against teardown with mutex.
2676 */
2677 pthread_mutex_lock(&stream->lock);
2678 if (cds_lfht_is_node_deleted(&stream->node.node)) {
2679 goto next;
2680 }
d8ef542d
MD
2681 switch (consumer_data.type) {
2682 case LTTNG_CONSUMER_KERNEL:
2683 break;
2684 case LTTNG_CONSUMER32_UST:
2685 case LTTNG_CONSUMER64_UST:
b4a650f3
DG
2686 if (stream->metadata_flag) {
2687 /* Safe and protected by the stream lock. */
2688 lttng_ustconsumer_close_metadata(stream->chan);
2689 } else {
2690 /*
2691 * Note: a mutex is taken internally within
2692 * liblttng-ust-ctl to protect timer wakeup_fd
2693 * use from concurrent close.
2694 */
2695 lttng_ustconsumer_close_stream_wakeup(stream);
2696 }
d8ef542d
MD
2697 break;
2698 default:
2699 ERR("Unknown consumer_data type");
2700 assert(0);
2701 }
f2ad556d
MD
2702 next:
2703 pthread_mutex_unlock(&stream->lock);
d8ef542d
MD
2704 }
2705 rcu_read_unlock();
2706}
2707
2708static void destroy_channel_ht(struct lttng_ht *ht)
2709{
2710 struct lttng_ht_iter iter;
2711 struct lttng_consumer_channel *channel;
2712 int ret;
2713
2714 if (ht == NULL) {
2715 return;
2716 }
2717
2718 rcu_read_lock();
2719 cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) {
2720 ret = lttng_ht_del(ht, &iter);
2721 assert(ret != 0);
2722 }
2723 rcu_read_unlock();
2724
2725 lttng_ht_destroy(ht);
2726}
2727
2728/*
2729 * This thread polls the channel fds to detect when they are being
2730 * closed. It closes all related streams if the channel is detected as
2731 * closed. It is currently only used as a shim layer for UST because the
2732 * consumerd needs to keep the per-stream wakeup end of pipes open for
2733 * periodical flush.
2734 */
2735void *consumer_thread_channel_poll(void *data)
2736{
1fc79fb4 2737 int ret, i, pollfd, err = -1;
d8ef542d
MD
2738 uint32_t revents, nb_fd;
2739 struct lttng_consumer_channel *chan = NULL;
2740 struct lttng_ht_iter iter;
2741 struct lttng_ht_node_u64 *node;
2742 struct lttng_poll_event events;
2743 struct lttng_consumer_local_data *ctx = data;
2744 struct lttng_ht *channel_ht;
2745
2746 rcu_register_thread();
2747
1fc79fb4
MD
2748 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL);
2749
2d57de81
MD
2750 if (testpoint(consumerd_thread_channel)) {
2751 goto error_testpoint;
2752 }
2753
9ce5646a
MD
2754 health_code_update();
2755
d8ef542d
MD
2756 channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2757 if (!channel_ht) {
2758 /* ENOMEM at this point. Better to bail out. */
2759 goto end_ht;
2760 }
2761
2762 DBG("Thread channel poll started");
2763
2764 /* Size is set to 1 for the consumer_channel pipe */
2765 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2766 if (ret < 0) {
2767 ERR("Poll set creation failed");
2768 goto end_poll;
2769 }
2770
2771 ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN);
2772 if (ret < 0) {
2773 goto end;
2774 }
2775
2776 /* Main loop */
2777 DBG("Channel main loop started");
2778
2779 while (1) {
d8ef542d 2780restart:
7fa2082e
MD
2781 health_code_update();
2782 DBG("Channel poll wait");
9ce5646a 2783 health_poll_entry();
d8ef542d 2784 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2785 DBG("Channel poll return from wait with %d fd(s)",
2786 LTTNG_POLL_GETNB(&events));
9ce5646a 2787 health_poll_exit();
d8ef542d
MD
2788 DBG("Channel event catched in thread");
2789 if (ret < 0) {
2790 if (errno == EINTR) {
2791 ERR("Poll EINTR catched");
2792 goto restart;
2793 }
d9607cd7
MD
2794 if (LTTNG_POLL_GETNB(&events) == 0) {
2795 err = 0; /* All is OK */
2796 }
d8ef542d
MD
2797 goto end;
2798 }
2799
2800 nb_fd = ret;
2801
2802 /* From here, the event is a channel wait fd */
2803 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2804 health_code_update();
2805
d8ef542d
MD
2806 revents = LTTNG_POLL_GETEV(&events, i);
2807 pollfd = LTTNG_POLL_GETFD(&events, i);
2808
d8ef542d 2809 if (!revents) {
fd20dac9 2810 /* No activity for this FD (poll implementation). */
d8ef542d
MD
2811 continue;
2812 }
fd20dac9 2813
d8ef542d
MD
2814 if (pollfd == ctx->consumer_channel_pipe[0]) {
2815 if (revents & (LPOLLERR | LPOLLHUP)) {
2816 DBG("Channel thread pipe hung up");
2817 /*
2818 * Remove the pipe from the poll set and continue the loop
2819 * since their might be data to consume.
2820 */
2821 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2822 continue;
2823 } else if (revents & LPOLLIN) {
2824 enum consumer_channel_action action;
a0cbdd2e 2825 uint64_t key;
d8ef542d 2826
a0cbdd2e 2827 ret = read_channel_pipe(ctx, &chan, &key, &action);
d8ef542d
MD
2828 if (ret <= 0) {
2829 ERR("Error reading channel pipe");
2830 continue;
2831 }
2832
2833 switch (action) {
2834 case CONSUMER_CHANNEL_ADD:
2835 DBG("Adding channel %d to poll set",
2836 chan->wait_fd);
2837
2838 lttng_ht_node_init_u64(&chan->wait_fd_node,
2839 chan->wait_fd);
c7260a81 2840 rcu_read_lock();
d8ef542d
MD
2841 lttng_ht_add_unique_u64(channel_ht,
2842 &chan->wait_fd_node);
c7260a81 2843 rcu_read_unlock();
d8ef542d
MD
2844 /* Add channel to the global poll events list */
2845 lttng_poll_add(&events, chan->wait_fd,
2846 LPOLLIN | LPOLLPRI);
2847 break;
a0cbdd2e
MD
2848 case CONSUMER_CHANNEL_DEL:
2849 {
b4a650f3
DG
2850 /*
2851 * This command should never be called if the channel
2852 * has streams monitored by either the data or metadata
2853 * thread. The consumer only notify this thread with a
2854 * channel del. command if it receives a destroy
2855 * channel command from the session daemon that send it
2856 * if a command prior to the GET_CHANNEL failed.
2857 */
2858
c7260a81 2859 rcu_read_lock();
a0cbdd2e
MD
2860 chan = consumer_find_channel(key);
2861 if (!chan) {
c7260a81 2862 rcu_read_unlock();
a0cbdd2e
MD
2863 ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key);
2864 break;
2865 }
2866 lttng_poll_del(&events, chan->wait_fd);
f623cc0b 2867 iter.iter.node = &chan->wait_fd_node.node;
a0cbdd2e
MD
2868 ret = lttng_ht_del(channel_ht, &iter);
2869 assert(ret == 0);
a0cbdd2e 2870
f2a444f1
DG
2871 switch (consumer_data.type) {
2872 case LTTNG_CONSUMER_KERNEL:
2873 break;
2874 case LTTNG_CONSUMER32_UST:
2875 case LTTNG_CONSUMER64_UST:
212d67a2
DG
2876 health_code_update();
2877 /* Destroy streams that might have been left in the stream list. */
2878 clean_channel_stream_list(chan);
f2a444f1
DG
2879 break;
2880 default:
2881 ERR("Unknown consumer_data type");
2882 assert(0);
2883 }
2884
a0cbdd2e
MD
2885 /*
2886 * Release our own refcount. Force channel deletion even if
2887 * streams were not initialized.
2888 */
2889 if (!uatomic_sub_return(&chan->refcount, 1)) {
2890 consumer_del_channel(chan);
2891 }
c7260a81 2892 rcu_read_unlock();
a0cbdd2e
MD
2893 goto restart;
2894 }
d8ef542d
MD
2895 case CONSUMER_CHANNEL_QUIT:
2896 /*
2897 * Remove the pipe from the poll set and continue the loop
2898 * since their might be data to consume.
2899 */
2900 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2901 continue;
2902 default:
2903 ERR("Unknown action");
2904 break;
2905 }
2906 }
2907
2908 /* Handle other stream */
2909 continue;
2910 }
2911
2912 rcu_read_lock();
2913 {
2914 uint64_t tmp_id = (uint64_t) pollfd;
2915
2916 lttng_ht_lookup(channel_ht, &tmp_id, &iter);
2917 }
2918 node = lttng_ht_iter_get_node_u64(&iter);
2919 assert(node);
2920
2921 chan = caa_container_of(node, struct lttng_consumer_channel,
2922 wait_fd_node);
2923
2924 /* Check for error event */
2925 if (revents & (LPOLLERR | LPOLLHUP)) {
2926 DBG("Channel fd %d is hup|err.", pollfd);
2927
2928 lttng_poll_del(&events, chan->wait_fd);
2929 ret = lttng_ht_del(channel_ht, &iter);
2930 assert(ret == 0);
b4a650f3
DG
2931
2932 /*
2933 * This will close the wait fd for each stream associated to
2934 * this channel AND monitored by the data/metadata thread thus
2935 * will be clean by the right thread.
2936 */
d8ef542d 2937 consumer_close_channel_streams(chan);
f2ad556d
MD
2938
2939 /* Release our own refcount */
2940 if (!uatomic_sub_return(&chan->refcount, 1)
2941 && !uatomic_read(&chan->nb_init_stream_left)) {
2942 consumer_del_channel(chan);
2943 }
d8ef542d
MD
2944 }
2945
2946 /* Release RCU lock for the channel looked up */
2947 rcu_read_unlock();
2948 }
2949 }
2950
1fc79fb4
MD
2951 /* All is OK */
2952 err = 0;
d8ef542d
MD
2953end:
2954 lttng_poll_clean(&events);
2955end_poll:
2956 destroy_channel_ht(channel_ht);
2957end_ht:
2d57de81 2958error_testpoint:
d8ef542d 2959 DBG("Channel poll thread exiting");
1fc79fb4
MD
2960 if (err) {
2961 health_error();
2962 ERR("Health error occurred in %s", __func__);
2963 }
2964 health_unregister(health_consumerd);
d8ef542d
MD
2965 rcu_unregister_thread();
2966 return NULL;
2967}
2968
331744e3
JD
2969static int set_metadata_socket(struct lttng_consumer_local_data *ctx,
2970 struct pollfd *sockpoll, int client_socket)
2971{
2972 int ret;
2973
2974 assert(ctx);
2975 assert(sockpoll);
2976
84382d49
MD
2977 ret = lttng_consumer_poll_socket(sockpoll);
2978 if (ret) {
331744e3
JD
2979 goto error;
2980 }
2981 DBG("Metadata connection on client_socket");
2982
2983 /* Blocking call, waiting for transmission */
2984 ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket);
2985 if (ctx->consumer_metadata_socket < 0) {
2986 WARN("On accept metadata");
2987 ret = -1;
2988 goto error;
2989 }
2990 ret = 0;
2991
2992error:
2993 return ret;
2994}
2995
3bd1e081
MD
2996/*
2997 * This thread listens on the consumerd socket and receives the file
2998 * descriptors from the session daemon.
2999 */
7d980def 3000void *consumer_thread_sessiond_poll(void *data)
3bd1e081 3001{
1fc79fb4 3002 int sock = -1, client_socket, ret, err = -1;
3bd1e081
MD
3003 /*
3004 * structure to poll for incoming data on communication socket avoids
3005 * making blocking sockets.
3006 */
3007 struct pollfd consumer_sockpoll[2];
3008 struct lttng_consumer_local_data *ctx = data;
3009
e7b994a3
DG
3010 rcu_register_thread();
3011
1fc79fb4
MD
3012 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND);
3013
2d57de81
MD
3014 if (testpoint(consumerd_thread_sessiond)) {
3015 goto error_testpoint;
3016 }
3017
9ce5646a
MD
3018 health_code_update();
3019
3bd1e081
MD
3020 DBG("Creating command socket %s", ctx->consumer_command_sock_path);
3021 unlink(ctx->consumer_command_sock_path);
3022 client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path);
3023 if (client_socket < 0) {
3024 ERR("Cannot create command socket");
3025 goto end;
3026 }
3027
3028 ret = lttcomm_listen_unix_sock(client_socket);
3029 if (ret < 0) {
3030 goto end;
3031 }
3032
32258573 3033 DBG("Sending ready command to lttng-sessiond");
f73fabfd 3034 ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY);
3bd1e081
MD
3035 /* return < 0 on error, but == 0 is not fatal */
3036 if (ret < 0) {
32258573 3037 ERR("Error sending ready command to lttng-sessiond");
3bd1e081
MD
3038 goto end;
3039 }
3040
3bd1e081
MD
3041 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3042 consumer_sockpoll[0].fd = ctx->consumer_should_quit[0];
3043 consumer_sockpoll[0].events = POLLIN | POLLPRI;
3044 consumer_sockpoll[1].fd = client_socket;
3045 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3046
84382d49
MD
3047 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3048 if (ret) {
3049 if (ret > 0) {
3050 /* should exit */
3051 err = 0;
3052 }
3bd1e081
MD
3053 goto end;
3054 }
3055 DBG("Connection on client_socket");
3056
3057 /* Blocking call, waiting for transmission */
3058 sock = lttcomm_accept_unix_sock(client_socket);
534d2592 3059 if (sock < 0) {
3bd1e081
MD
3060 WARN("On accept");
3061 goto end;
3062 }
3bd1e081 3063
331744e3
JD
3064 /*
3065 * Setup metadata socket which is the second socket connection on the
3066 * command unix socket.
3067 */
3068 ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket);
84382d49
MD
3069 if (ret) {
3070 if (ret > 0) {
3071 /* should exit */
3072 err = 0;
3073 }
331744e3
JD
3074 goto end;
3075 }
3076
d96f09c6
DG
3077 /* This socket is not useful anymore. */
3078 ret = close(client_socket);
3079 if (ret < 0) {
3080 PERROR("close client_socket");
3081 }
3082 client_socket = -1;
3083
3bd1e081
MD
3084 /* update the polling structure to poll on the established socket */
3085 consumer_sockpoll[1].fd = sock;
3086 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3087
3088 while (1) {
9ce5646a
MD
3089 health_code_update();
3090
3091 health_poll_entry();
3092 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3093 health_poll_exit();
84382d49
MD
3094 if (ret) {
3095 if (ret > 0) {
3096 /* should exit */
3097 err = 0;
3098 }
3bd1e081
MD
3099 goto end;
3100 }
3101 DBG("Incoming command on sock");
3102 ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll);
4cbc1a04
DG
3103 if (ret <= 0) {
3104 /*
3105 * This could simply be a session daemon quitting. Don't output
3106 * ERR() here.
3107 */
3108 DBG("Communication interrupted on command socket");
41ba6035 3109 err = 0;
3bd1e081
MD
3110 goto end;
3111 }
3112 if (consumer_quit) {
3113 DBG("consumer_thread_receive_fds received quit from signal");
1fc79fb4 3114 err = 0; /* All is OK */
3bd1e081
MD
3115 goto end;
3116 }
ffe60014 3117 DBG("received command on sock");
3bd1e081 3118 }
1fc79fb4
MD
3119 /* All is OK */
3120 err = 0;
3121
3bd1e081 3122end:
ffe60014 3123 DBG("Consumer thread sessiond poll exiting");
3bd1e081 3124
d88aee68
DG
3125 /*
3126 * Close metadata streams since the producer is the session daemon which
3127 * just died.
3128 *
3129 * NOTE: for now, this only applies to the UST tracer.
3130 */
6d574024 3131 lttng_consumer_close_all_metadata();
d88aee68 3132
3bd1e081
MD
3133 /*
3134 * when all fds have hung up, the polling thread
3135 * can exit cleanly
3136 */
3137 consumer_quit = 1;
3138
04fdd819 3139 /*
c869f647 3140 * Notify the data poll thread to poll back again and test the
8994307f 3141 * consumer_quit state that we just set so to quit gracefully.
04fdd819 3142 */
acdb9057 3143 notify_thread_lttng_pipe(ctx->consumer_data_pipe);
c869f647 3144
a0cbdd2e 3145 notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT);
d8ef542d 3146
5c635c72
MD
3147 notify_health_quit_pipe(health_quit_pipe);
3148
d96f09c6
DG
3149 /* Cleaning up possibly open sockets. */
3150 if (sock >= 0) {
3151 ret = close(sock);
3152 if (ret < 0) {
3153 PERROR("close sock sessiond poll");
3154 }
3155 }
3156 if (client_socket >= 0) {
38476d24 3157 ret = close(client_socket);
d96f09c6
DG
3158 if (ret < 0) {
3159 PERROR("close client_socket sessiond poll");
3160 }
3161 }
3162
2d57de81 3163error_testpoint:
1fc79fb4
MD
3164 if (err) {
3165 health_error();
3166 ERR("Health error occurred in %s", __func__);
3167 }
3168 health_unregister(health_consumerd);
3169
e7b994a3 3170 rcu_unregister_thread();
3bd1e081
MD
3171 return NULL;
3172}
d41f73b7 3173
4078b776 3174ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream,
d41f73b7
MD
3175 struct lttng_consumer_local_data *ctx)
3176{
74251bb8
DG
3177 ssize_t ret;
3178
3179 pthread_mutex_lock(&stream->lock);
94d49140
JD
3180 if (stream->metadata_flag) {
3181 pthread_mutex_lock(&stream->metadata_rdv_lock);
3182 }
74251bb8 3183
d41f73b7
MD
3184 switch (consumer_data.type) {
3185 case LTTNG_CONSUMER_KERNEL:
74251bb8
DG
3186 ret = lttng_kconsumer_read_subbuffer(stream, ctx);
3187 break;
7753dea8
MD
3188 case LTTNG_CONSUMER32_UST:
3189 case LTTNG_CONSUMER64_UST:
74251bb8
DG
3190 ret = lttng_ustconsumer_read_subbuffer(stream, ctx);
3191 break;
d41f73b7
MD
3192 default:
3193 ERR("Unknown consumer_data type");
3194 assert(0);
74251bb8
DG
3195 ret = -ENOSYS;
3196 break;
d41f73b7 3197 }
74251bb8 3198
94d49140
JD
3199 if (stream->metadata_flag) {
3200 pthread_cond_broadcast(&stream->metadata_rdv);
3201 pthread_mutex_unlock(&stream->metadata_rdv_lock);
3202 }
74251bb8
DG
3203 pthread_mutex_unlock(&stream->lock);
3204 return ret;
d41f73b7
MD
3205}
3206
3207int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream)
3208{
3209 switch (consumer_data.type) {
3210 case LTTNG_CONSUMER_KERNEL:
3211 return lttng_kconsumer_on_recv_stream(stream);
7753dea8
MD
3212 case LTTNG_CONSUMER32_UST:
3213 case LTTNG_CONSUMER64_UST:
d41f73b7
MD
3214 return lttng_ustconsumer_on_recv_stream(stream);
3215 default:
3216 ERR("Unknown consumer_data type");
3217 assert(0);
3218 return -ENOSYS;
3219 }
3220}
e4421fec
DG
3221
3222/*
3223 * Allocate and set consumer data hash tables.
3224 */
282dadbc 3225int lttng_consumer_init(void)
e4421fec 3226{
d88aee68 3227 consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
282dadbc
MD
3228 if (!consumer_data.channel_ht) {
3229 goto error;
3230 }
3231
d88aee68 3232 consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
282dadbc
MD
3233 if (!consumer_data.relayd_ht) {
3234 goto error;
3235 }
3236
d88aee68 3237 consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
282dadbc
MD
3238 if (!consumer_data.stream_list_ht) {
3239 goto error;
3240 }
3241
d8ef542d 3242 consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
282dadbc
MD
3243 if (!consumer_data.stream_per_chan_id_ht) {
3244 goto error;
3245 }
3246
3247 data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3248 if (!data_ht) {
3249 goto error;
3250 }
3251
3252 metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3253 if (!metadata_ht) {
3254 goto error;
3255 }
3256
3257 return 0;
3258
3259error:
3260 return -1;
e4421fec 3261}
7735ef9e
DG
3262
3263/*
3264 * Process the ADD_RELAYD command receive by a consumer.
3265 *
3266 * This will create a relayd socket pair and add it to the relayd hash table.
3267 * The caller MUST acquire a RCU read side lock before calling it.
3268 */
da009f2c 3269int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type,
7735ef9e 3270 struct lttng_consumer_local_data *ctx, int sock,
6151a90f 3271 struct pollfd *consumer_sockpoll,
d3e2ba59
JD
3272 struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id,
3273 uint64_t relayd_session_id)
7735ef9e 3274{
cd2b09ed 3275 int fd = -1, ret = -1, relayd_created = 0;
0c759fc9 3276 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
d4298c99 3277 struct consumer_relayd_sock_pair *relayd = NULL;
7735ef9e 3278
6151a90f
JD
3279 assert(ctx);
3280 assert(relayd_sock);
3281
da009f2c 3282 DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx);
7735ef9e
DG
3283
3284 /* Get relayd reference if exists. */
3285 relayd = consumer_find_relayd(net_seq_idx);
3286 if (relayd == NULL) {
da009f2c 3287 assert(sock_type == LTTNG_STREAM_CONTROL);
7735ef9e
DG
3288 /* Not found. A