Fix: add missing semicolons after MSG, DBG, ERR print macros
[lttng-tools.git] / src / common / consumer / consumer.c
CommitLineData
3bd1e081
MD
1/*
2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
00e2e675 4 * 2012 - David Goulet <dgoulet@efficios.com>
3bd1e081 5 *
d14d33bf
AM
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
3bd1e081 9 *
d14d33bf
AM
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
3bd1e081 14 *
d14d33bf
AM
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
3bd1e081
MD
18 */
19
6c1c0768 20#define _LGPL_SOURCE
3bd1e081 21#include <assert.h>
3bd1e081
MD
22#include <poll.h>
23#include <pthread.h>
24#include <stdlib.h>
25#include <string.h>
26#include <sys/mman.h>
27#include <sys/socket.h>
28#include <sys/types.h>
29#include <unistd.h>
77c7c900 30#include <inttypes.h>
331744e3 31#include <signal.h>
3bd1e081 32
51a9e1c7 33#include <bin/lttng-consumerd/health-consumerd.h>
990570ed 34#include <common/common.h>
fb3a43a9
DG
35#include <common/utils.h>
36#include <common/compat/poll.h>
f263b7fd 37#include <common/compat/endian.h>
309167d2 38#include <common/index/index.h>
10a8a223 39#include <common/kernel-ctl/kernel-ctl.h>
00e2e675 40#include <common/sessiond-comm/relayd.h>
10a8a223
DG
41#include <common/sessiond-comm/sessiond-comm.h>
42#include <common/kernel-consumer/kernel-consumer.h>
00e2e675 43#include <common/relayd/relayd.h>
10a8a223 44#include <common/ust-consumer/ust-consumer.h>
c8fea79c
JR
45#include <common/consumer/consumer-timer.h>
46#include <common/consumer/consumer.h>
47#include <common/consumer/consumer-stream.h>
48#include <common/consumer/consumer-testpoint.h>
49#include <common/align.h>
3bd1e081
MD
50
51struct lttng_consumer_global_data consumer_data = {
3bd1e081
MD
52 .stream_count = 0,
53 .need_update = 1,
54 .type = LTTNG_CONSUMER_UNKNOWN,
55};
56
d8ef542d
MD
57enum consumer_channel_action {
58 CONSUMER_CHANNEL_ADD,
a0cbdd2e 59 CONSUMER_CHANNEL_DEL,
d8ef542d
MD
60 CONSUMER_CHANNEL_QUIT,
61};
62
63struct consumer_channel_msg {
64 enum consumer_channel_action action;
a0cbdd2e
MD
65 struct lttng_consumer_channel *chan; /* add */
66 uint64_t key; /* del */
d8ef542d
MD
67};
68
3bd1e081
MD
69/*
70 * Flag to inform the polling thread to quit when all fd hung up. Updated by
71 * the consumer_thread_receive_fds when it notices that all fds has hung up.
72 * Also updated by the signal handler (consumer_should_exit()). Read by the
73 * polling threads.
74 */
a98dae5f 75volatile int consumer_quit;
3bd1e081 76
43c34bc3 77/*
43c34bc3
DG
78 * Global hash table containing respectively metadata and data streams. The
79 * stream element in this ht should only be updated by the metadata poll thread
80 * for the metadata and the data poll thread for the data.
81 */
40dc48e0
DG
82static struct lttng_ht *metadata_ht;
83static struct lttng_ht *data_ht;
43c34bc3 84
acdb9057
DG
85/*
86 * Notify a thread lttng pipe to poll back again. This usually means that some
87 * global state has changed so we just send back the thread in a poll wait
88 * call.
89 */
90static void notify_thread_lttng_pipe(struct lttng_pipe *pipe)
91{
92 struct lttng_consumer_stream *null_stream = NULL;
93
94 assert(pipe);
95
96 (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream));
97}
98
5c635c72
MD
99static void notify_health_quit_pipe(int *pipe)
100{
6cd525e8 101 ssize_t ret;
5c635c72 102
6cd525e8
MD
103 ret = lttng_write(pipe[1], "4", 1);
104 if (ret < 1) {
5c635c72
MD
105 PERROR("write consumer health quit");
106 }
107}
108
d8ef542d
MD
109static void notify_channel_pipe(struct lttng_consumer_local_data *ctx,
110 struct lttng_consumer_channel *chan,
a0cbdd2e 111 uint64_t key,
d8ef542d
MD
112 enum consumer_channel_action action)
113{
114 struct consumer_channel_msg msg;
6cd525e8 115 ssize_t ret;
d8ef542d 116
e56251fc
DG
117 memset(&msg, 0, sizeof(msg));
118
d8ef542d
MD
119 msg.action = action;
120 msg.chan = chan;
f21dae48 121 msg.key = key;
6cd525e8
MD
122 ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg));
123 if (ret < sizeof(msg)) {
124 PERROR("notify_channel_pipe write error");
125 }
d8ef542d
MD
126}
127
a0cbdd2e
MD
128void notify_thread_del_channel(struct lttng_consumer_local_data *ctx,
129 uint64_t key)
130{
131 notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL);
132}
133
d8ef542d
MD
134static int read_channel_pipe(struct lttng_consumer_local_data *ctx,
135 struct lttng_consumer_channel **chan,
a0cbdd2e 136 uint64_t *key,
d8ef542d
MD
137 enum consumer_channel_action *action)
138{
139 struct consumer_channel_msg msg;
6cd525e8 140 ssize_t ret;
d8ef542d 141
6cd525e8
MD
142 ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg));
143 if (ret < sizeof(msg)) {
144 ret = -1;
145 goto error;
d8ef542d 146 }
6cd525e8
MD
147 *action = msg.action;
148 *chan = msg.chan;
149 *key = msg.key;
150error:
151 return (int) ret;
d8ef542d
MD
152}
153
212d67a2
DG
154/*
155 * Cleanup the stream list of a channel. Those streams are not yet globally
156 * visible
157 */
158static void clean_channel_stream_list(struct lttng_consumer_channel *channel)
159{
160 struct lttng_consumer_stream *stream, *stmp;
161
162 assert(channel);
163
164 /* Delete streams that might have been left in the stream list. */
165 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
166 send_node) {
167 cds_list_del(&stream->send_node);
168 /*
169 * Once a stream is added to this list, the buffers were created so we
170 * have a guarantee that this call will succeed. Setting the monitor
171 * mode to 0 so we don't lock nor try to delete the stream from the
172 * global hash table.
173 */
174 stream->monitor = 0;
175 consumer_stream_destroy(stream, NULL);
176 }
177}
178
3bd1e081
MD
179/*
180 * Find a stream. The consumer_data.lock must be locked during this
181 * call.
182 */
d88aee68 183static struct lttng_consumer_stream *find_stream(uint64_t key,
8389e4f8 184 struct lttng_ht *ht)
3bd1e081 185{
e4421fec 186 struct lttng_ht_iter iter;
d88aee68 187 struct lttng_ht_node_u64 *node;
e4421fec 188 struct lttng_consumer_stream *stream = NULL;
3bd1e081 189
8389e4f8
DG
190 assert(ht);
191
d88aee68
DG
192 /* -1ULL keys are lookup failures */
193 if (key == (uint64_t) -1ULL) {
7ad0a0cb 194 return NULL;
7a57cf92 195 }
e4421fec 196
6065ceec
DG
197 rcu_read_lock();
198
d88aee68
DG
199 lttng_ht_lookup(ht, &key, &iter);
200 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
201 if (node != NULL) {
202 stream = caa_container_of(node, struct lttng_consumer_stream, node);
3bd1e081 203 }
e4421fec 204
6065ceec
DG
205 rcu_read_unlock();
206
e4421fec 207 return stream;
3bd1e081
MD
208}
209
da009f2c 210static void steal_stream_key(uint64_t key, struct lttng_ht *ht)
7ad0a0cb
MD
211{
212 struct lttng_consumer_stream *stream;
213
04253271 214 rcu_read_lock();
ffe60014 215 stream = find_stream(key, ht);
04253271 216 if (stream) {
da009f2c 217 stream->key = (uint64_t) -1ULL;
04253271
MD
218 /*
219 * We don't want the lookup to match, but we still need
220 * to iterate on this stream when iterating over the hash table. Just
221 * change the node key.
222 */
da009f2c 223 stream->node.key = (uint64_t) -1ULL;
04253271
MD
224 }
225 rcu_read_unlock();
7ad0a0cb
MD
226}
227
d56db448
DG
228/*
229 * Return a channel object for the given key.
230 *
231 * RCU read side lock MUST be acquired before calling this function and
232 * protects the channel ptr.
233 */
d88aee68 234struct lttng_consumer_channel *consumer_find_channel(uint64_t key)
3bd1e081 235{
e4421fec 236 struct lttng_ht_iter iter;
d88aee68 237 struct lttng_ht_node_u64 *node;
e4421fec 238 struct lttng_consumer_channel *channel = NULL;
3bd1e081 239
d88aee68
DG
240 /* -1ULL keys are lookup failures */
241 if (key == (uint64_t) -1ULL) {
7ad0a0cb 242 return NULL;
7a57cf92 243 }
e4421fec 244
d88aee68
DG
245 lttng_ht_lookup(consumer_data.channel_ht, &key, &iter);
246 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
247 if (node != NULL) {
248 channel = caa_container_of(node, struct lttng_consumer_channel, node);
3bd1e081 249 }
e4421fec
DG
250
251 return channel;
3bd1e081
MD
252}
253
b5a6470f
DG
254/*
255 * There is a possibility that the consumer does not have enough time between
256 * the close of the channel on the session daemon and the cleanup in here thus
257 * once we have a channel add with an existing key, we know for sure that this
258 * channel will eventually get cleaned up by all streams being closed.
259 *
260 * This function just nullifies the already existing channel key.
261 */
262static void steal_channel_key(uint64_t key)
263{
264 struct lttng_consumer_channel *channel;
265
266 rcu_read_lock();
267 channel = consumer_find_channel(key);
268 if (channel) {
269 channel->key = (uint64_t) -1ULL;
270 /*
271 * We don't want the lookup to match, but we still need to iterate on
272 * this channel when iterating over the hash table. Just change the
273 * node key.
274 */
275 channel->node.key = (uint64_t) -1ULL;
276 }
277 rcu_read_unlock();
278}
279
ffe60014 280static void free_channel_rcu(struct rcu_head *head)
702b1ea4 281{
d88aee68
DG
282 struct lttng_ht_node_u64 *node =
283 caa_container_of(head, struct lttng_ht_node_u64, head);
ffe60014
DG
284 struct lttng_consumer_channel *channel =
285 caa_container_of(node, struct lttng_consumer_channel, node);
702b1ea4 286
b83e03c4
MD
287 switch (consumer_data.type) {
288 case LTTNG_CONSUMER_KERNEL:
289 break;
290 case LTTNG_CONSUMER32_UST:
291 case LTTNG_CONSUMER64_UST:
292 lttng_ustconsumer_free_channel(channel);
293 break;
294 default:
295 ERR("Unknown consumer_data type");
296 abort();
297 }
ffe60014 298 free(channel);
702b1ea4
MD
299}
300
00e2e675
DG
301/*
302 * RCU protected relayd socket pair free.
303 */
ffe60014 304static void free_relayd_rcu(struct rcu_head *head)
00e2e675 305{
d88aee68
DG
306 struct lttng_ht_node_u64 *node =
307 caa_container_of(head, struct lttng_ht_node_u64, head);
00e2e675
DG
308 struct consumer_relayd_sock_pair *relayd =
309 caa_container_of(node, struct consumer_relayd_sock_pair, node);
310
8994307f
DG
311 /*
312 * Close all sockets. This is done in the call RCU since we don't want the
313 * socket fds to be reassigned thus potentially creating bad state of the
314 * relayd object.
315 *
316 * We do not have to lock the control socket mutex here since at this stage
317 * there is no one referencing to this relayd object.
318 */
319 (void) relayd_close(&relayd->control_sock);
320 (void) relayd_close(&relayd->data_sock);
321
00e2e675
DG
322 free(relayd);
323}
324
325/*
326 * Destroy and free relayd socket pair object.
00e2e675 327 */
51230d70 328void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
329{
330 int ret;
331 struct lttng_ht_iter iter;
332
173af62f
DG
333 if (relayd == NULL) {
334 return;
335 }
336
00e2e675
DG
337 DBG("Consumer destroy and close relayd socket pair");
338
339 iter.iter.node = &relayd->node.node;
340 ret = lttng_ht_del(consumer_data.relayd_ht, &iter);
173af62f 341 if (ret != 0) {
8994307f 342 /* We assume the relayd is being or is destroyed */
173af62f
DG
343 return;
344 }
00e2e675 345
00e2e675 346 /* RCU free() call */
ffe60014
DG
347 call_rcu(&relayd->node.head, free_relayd_rcu);
348}
349
350/*
351 * Remove a channel from the global list protected by a mutex. This function is
352 * also responsible for freeing its data structures.
353 */
354void consumer_del_channel(struct lttng_consumer_channel *channel)
355{
356 int ret;
357 struct lttng_ht_iter iter;
358
d88aee68 359 DBG("Consumer delete channel key %" PRIu64, channel->key);
ffe60014
DG
360
361 pthread_mutex_lock(&consumer_data.lock);
a9838785 362 pthread_mutex_lock(&channel->lock);
ffe60014 363
212d67a2
DG
364 /* Destroy streams that might have been left in the stream list. */
365 clean_channel_stream_list(channel);
51e762e5 366
d3e2ba59
JD
367 if (channel->live_timer_enabled == 1) {
368 consumer_timer_live_stop(channel);
369 }
370
ffe60014
DG
371 switch (consumer_data.type) {
372 case LTTNG_CONSUMER_KERNEL:
373 break;
374 case LTTNG_CONSUMER32_UST:
375 case LTTNG_CONSUMER64_UST:
376 lttng_ustconsumer_del_channel(channel);
377 break;
378 default:
379 ERR("Unknown consumer_data type");
380 assert(0);
381 goto end;
382 }
383
384 rcu_read_lock();
385 iter.iter.node = &channel->node.node;
386 ret = lttng_ht_del(consumer_data.channel_ht, &iter);
387 assert(!ret);
388 rcu_read_unlock();
389
390 call_rcu(&channel->node.head, free_channel_rcu);
391end:
a9838785 392 pthread_mutex_unlock(&channel->lock);
ffe60014 393 pthread_mutex_unlock(&consumer_data.lock);
00e2e675
DG
394}
395
228b5bf7
DG
396/*
397 * Iterate over the relayd hash table and destroy each element. Finally,
398 * destroy the whole hash table.
399 */
400static void cleanup_relayd_ht(void)
401{
402 struct lttng_ht_iter iter;
403 struct consumer_relayd_sock_pair *relayd;
404
405 rcu_read_lock();
406
407 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
408 node.node) {
51230d70 409 consumer_destroy_relayd(relayd);
228b5bf7
DG
410 }
411
228b5bf7 412 rcu_read_unlock();
36b588ed
MD
413
414 lttng_ht_destroy(consumer_data.relayd_ht);
228b5bf7
DG
415}
416
8994307f
DG
417/*
418 * Update the end point status of all streams having the given network sequence
419 * index (relayd index).
420 *
421 * It's atomically set without having the stream mutex locked which is fine
422 * because we handle the write/read race with a pipe wakeup for each thread.
423 */
da009f2c 424static void update_endpoint_status_by_netidx(uint64_t net_seq_idx,
8994307f
DG
425 enum consumer_endpoint_status status)
426{
427 struct lttng_ht_iter iter;
428 struct lttng_consumer_stream *stream;
429
da009f2c 430 DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx);
8994307f
DG
431
432 rcu_read_lock();
433
434 /* Let's begin with metadata */
435 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
436 if (stream->net_seq_idx == net_seq_idx) {
437 uatomic_set(&stream->endpoint_status, status);
438 DBG("Delete flag set to metadata stream %d", stream->wait_fd);
439 }
440 }
441
442 /* Follow up by the data streams */
443 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
444 if (stream->net_seq_idx == net_seq_idx) {
445 uatomic_set(&stream->endpoint_status, status);
446 DBG("Delete flag set to data stream %d", stream->wait_fd);
447 }
448 }
449 rcu_read_unlock();
450}
451
452/*
453 * Cleanup a relayd object by flagging every associated streams for deletion,
454 * destroying the object meaning removing it from the relayd hash table,
455 * closing the sockets and freeing the memory in a RCU call.
456 *
457 * If a local data context is available, notify the threads that the streams'
458 * state have changed.
459 */
460static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd,
461 struct lttng_consumer_local_data *ctx)
462{
da009f2c 463 uint64_t netidx;
8994307f
DG
464
465 assert(relayd);
466
9617607b
DG
467 DBG("Cleaning up relayd sockets");
468
8994307f
DG
469 /* Save the net sequence index before destroying the object */
470 netidx = relayd->net_seq_idx;
471
472 /*
473 * Delete the relayd from the relayd hash table, close the sockets and free
474 * the object in a RCU call.
475 */
51230d70 476 consumer_destroy_relayd(relayd);
8994307f
DG
477
478 /* Set inactive endpoint to all streams */
479 update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE);
480
481 /*
482 * With a local data context, notify the threads that the streams' state
483 * have changed. The write() action on the pipe acts as an "implicit"
484 * memory barrier ordering the updates of the end point status from the
485 * read of this status which happens AFTER receiving this notify.
486 */
487 if (ctx) {
acdb9057 488 notify_thread_lttng_pipe(ctx->consumer_data_pipe);
13886d2d 489 notify_thread_lttng_pipe(ctx->consumer_metadata_pipe);
8994307f
DG
490 }
491}
492
a6ba4fe1
DG
493/*
494 * Flag a relayd socket pair for destruction. Destroy it if the refcount
495 * reaches zero.
496 *
497 * RCU read side lock MUST be aquired before calling this function.
498 */
499void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd)
500{
501 assert(relayd);
502
503 /* Set destroy flag for this object */
504 uatomic_set(&relayd->destroy_flag, 1);
505
506 /* Destroy the relayd if refcount is 0 */
507 if (uatomic_read(&relayd->refcount) == 0) {
51230d70 508 consumer_destroy_relayd(relayd);
a6ba4fe1
DG
509 }
510}
511
3bd1e081 512/*
1d1a276c
DG
513 * Completly destroy stream from every visiable data structure and the given
514 * hash table if one.
515 *
516 * One this call returns, the stream object is not longer usable nor visible.
3bd1e081 517 */
e316aad5
DG
518void consumer_del_stream(struct lttng_consumer_stream *stream,
519 struct lttng_ht *ht)
3bd1e081 520{
1d1a276c 521 consumer_stream_destroy(stream, ht);
3bd1e081
MD
522}
523
5ab66908
MD
524/*
525 * XXX naming of del vs destroy is all mixed up.
526 */
527void consumer_del_stream_for_data(struct lttng_consumer_stream *stream)
528{
529 consumer_stream_destroy(stream, data_ht);
530}
531
532void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream)
533{
534 consumer_stream_destroy(stream, metadata_ht);
535}
536
d88aee68
DG
537struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key,
538 uint64_t stream_key,
3bd1e081 539 enum lttng_consumer_stream_state state,
ffe60014 540 const char *channel_name,
6df2e2c9 541 uid_t uid,
00e2e675 542 gid_t gid,
57a269f2 543 uint64_t relayd_id,
53632229 544 uint64_t session_id,
ffe60014
DG
545 int cpu,
546 int *alloc_ret,
4891ece8
DG
547 enum consumer_channel_type type,
548 unsigned int monitor)
3bd1e081 549{
ffe60014 550 int ret;
3bd1e081 551 struct lttng_consumer_stream *stream;
3bd1e081 552
effcf122 553 stream = zmalloc(sizeof(*stream));
3bd1e081 554 if (stream == NULL) {
7a57cf92 555 PERROR("malloc struct lttng_consumer_stream");
ffe60014 556 ret = -ENOMEM;
7a57cf92 557 goto end;
3bd1e081 558 }
7a57cf92 559
d56db448
DG
560 rcu_read_lock();
561
3bd1e081 562 stream->key = stream_key;
3bd1e081
MD
563 stream->out_fd = -1;
564 stream->out_fd_offset = 0;
e5d1a9b3 565 stream->output_written = 0;
3bd1e081 566 stream->state = state;
6df2e2c9
MD
567 stream->uid = uid;
568 stream->gid = gid;
ffe60014 569 stream->net_seq_idx = relayd_id;
53632229 570 stream->session_id = session_id;
4891ece8 571 stream->monitor = monitor;
774d490c 572 stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE;
309167d2 573 stream->index_fd = -1;
fb83fe64 574 stream->last_sequence_number = -1ULL;
53632229 575 pthread_mutex_init(&stream->lock, NULL);
c585821b 576 pthread_mutex_init(&stream->metadata_timer_lock, NULL);
58b1f425 577
ffe60014
DG
578 /* If channel is the metadata, flag this stream as metadata. */
579 if (type == CONSUMER_CHANNEL_TYPE_METADATA) {
580 stream->metadata_flag = 1;
581 /* Metadata is flat out. */
582 strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name));
94d49140
JD
583 /* Live rendez-vous point. */
584 pthread_cond_init(&stream->metadata_rdv, NULL);
585 pthread_mutex_init(&stream->metadata_rdv_lock, NULL);
58b1f425 586 } else {
ffe60014
DG
587 /* Format stream name to <channel_name>_<cpu_number> */
588 ret = snprintf(stream->name, sizeof(stream->name), "%s_%d",
589 channel_name, cpu);
590 if (ret < 0) {
591 PERROR("snprintf stream name");
592 goto error;
593 }
58b1f425 594 }
c30aaa51 595
ffe60014 596 /* Key is always the wait_fd for streams. */
d88aee68 597 lttng_ht_node_init_u64(&stream->node, stream->key);
ffe60014 598
d8ef542d
MD
599 /* Init node per channel id key */
600 lttng_ht_node_init_u64(&stream->node_channel_id, channel_key);
601
53632229 602 /* Init session id node with the stream session id */
d88aee68 603 lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id);
53632229 604
07b86b52
JD
605 DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64
606 " relayd_id %" PRIu64 ", session_id %" PRIu64,
607 stream->name, stream->key, channel_key,
608 stream->net_seq_idx, stream->session_id);
d56db448
DG
609
610 rcu_read_unlock();
3bd1e081 611 return stream;
c80048c6
MD
612
613error:
d56db448 614 rcu_read_unlock();
c80048c6 615 free(stream);
7a57cf92 616end:
ffe60014
DG
617 if (alloc_ret) {
618 *alloc_ret = ret;
619 }
c80048c6 620 return NULL;
3bd1e081
MD
621}
622
623/*
624 * Add a stream to the global list protected by a mutex.
625 */
5ab66908 626int consumer_add_data_stream(struct lttng_consumer_stream *stream)
3bd1e081 627{
5ab66908 628 struct lttng_ht *ht = data_ht;
3bd1e081
MD
629 int ret = 0;
630
e316aad5 631 assert(stream);
43c34bc3 632 assert(ht);
c77fc10a 633
d88aee68 634 DBG3("Adding consumer stream %" PRIu64, stream->key);
e316aad5
DG
635
636 pthread_mutex_lock(&consumer_data.lock);
a9838785 637 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 638 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 639 pthread_mutex_lock(&stream->lock);
b0b335c8 640 rcu_read_lock();
e316aad5 641
43c34bc3 642 /* Steal stream identifier to avoid having streams with the same key */
ffe60014 643 steal_stream_key(stream->key, ht);
43c34bc3 644
d88aee68 645 lttng_ht_add_unique_u64(ht, &stream->node);
00e2e675 646
d8ef542d
MD
647 lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht,
648 &stream->node_channel_id);
649
ca22feea
DG
650 /*
651 * Add stream to the stream_list_ht of the consumer data. No need to steal
652 * the key since the HT does not use it and we allow to add redundant keys
653 * into this table.
654 */
d88aee68 655 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 656
e316aad5 657 /*
ffe60014
DG
658 * When nb_init_stream_left reaches 0, we don't need to trigger any action
659 * in terms of destroying the associated channel, because the action that
e316aad5
DG
660 * causes the count to become 0 also causes a stream to be added. The
661 * channel deletion will thus be triggered by the following removal of this
662 * stream.
663 */
ffe60014 664 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
665 /* Increment refcount before decrementing nb_init_stream_left */
666 cmm_smp_wmb();
ffe60014 667 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
668 }
669
670 /* Update consumer data once the node is inserted. */
3bd1e081
MD
671 consumer_data.stream_count++;
672 consumer_data.need_update = 1;
673
e316aad5 674 rcu_read_unlock();
2e818a6a 675 pthread_mutex_unlock(&stream->lock);
ec6ea7d0 676 pthread_mutex_unlock(&stream->chan->timer_lock);
a9838785 677 pthread_mutex_unlock(&stream->chan->lock);
3bd1e081 678 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 679
3bd1e081
MD
680 return ret;
681}
682
5ab66908
MD
683void consumer_del_data_stream(struct lttng_consumer_stream *stream)
684{
685 consumer_del_stream(stream, data_ht);
686}
687
00e2e675 688/*
3f8e211f
DG
689 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
690 * be acquired before calling this.
00e2e675 691 */
d09e1200 692static int add_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
693{
694 int ret = 0;
d88aee68 695 struct lttng_ht_node_u64 *node;
00e2e675
DG
696 struct lttng_ht_iter iter;
697
ffe60014 698 assert(relayd);
00e2e675 699
00e2e675 700 lttng_ht_lookup(consumer_data.relayd_ht,
d88aee68
DG
701 &relayd->net_seq_idx, &iter);
702 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675 703 if (node != NULL) {
00e2e675
DG
704 goto end;
705 }
d88aee68 706 lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node);
00e2e675 707
00e2e675
DG
708end:
709 return ret;
710}
711
712/*
713 * Allocate and return a consumer relayd socket.
714 */
027a694f 715static struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair(
da009f2c 716 uint64_t net_seq_idx)
00e2e675
DG
717{
718 struct consumer_relayd_sock_pair *obj = NULL;
719
da009f2c
MD
720 /* net sequence index of -1 is a failure */
721 if (net_seq_idx == (uint64_t) -1ULL) {
00e2e675
DG
722 goto error;
723 }
724
725 obj = zmalloc(sizeof(struct consumer_relayd_sock_pair));
726 if (obj == NULL) {
727 PERROR("zmalloc relayd sock");
728 goto error;
729 }
730
731 obj->net_seq_idx = net_seq_idx;
732 obj->refcount = 0;
173af62f 733 obj->destroy_flag = 0;
f96e4545
MD
734 obj->control_sock.sock.fd = -1;
735 obj->data_sock.sock.fd = -1;
d88aee68 736 lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx);
00e2e675
DG
737 pthread_mutex_init(&obj->ctrl_sock_mutex, NULL);
738
739error:
740 return obj;
741}
742
743/*
744 * Find a relayd socket pair in the global consumer data.
745 *
746 * Return the object if found else NULL.
b0b335c8
MD
747 * RCU read-side lock must be held across this call and while using the
748 * returned object.
00e2e675 749 */
d88aee68 750struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key)
00e2e675
DG
751{
752 struct lttng_ht_iter iter;
d88aee68 753 struct lttng_ht_node_u64 *node;
00e2e675
DG
754 struct consumer_relayd_sock_pair *relayd = NULL;
755
756 /* Negative keys are lookup failures */
d88aee68 757 if (key == (uint64_t) -1ULL) {
00e2e675
DG
758 goto error;
759 }
760
d88aee68 761 lttng_ht_lookup(consumer_data.relayd_ht, &key,
00e2e675 762 &iter);
d88aee68 763 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675
DG
764 if (node != NULL) {
765 relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node);
766 }
767
00e2e675
DG
768error:
769 return relayd;
770}
771
10a50311
JD
772/*
773 * Find a relayd and send the stream
774 *
775 * Returns 0 on success, < 0 on error
776 */
777int consumer_send_relayd_stream(struct lttng_consumer_stream *stream,
778 char *path)
779{
780 int ret = 0;
781 struct consumer_relayd_sock_pair *relayd;
782
783 assert(stream);
784 assert(stream->net_seq_idx != -1ULL);
785 assert(path);
786
787 /* The stream is not metadata. Get relayd reference if exists. */
788 rcu_read_lock();
789 relayd = consumer_find_relayd(stream->net_seq_idx);
790 if (relayd != NULL) {
791 /* Add stream on the relayd */
792 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
793 ret = relayd_add_stream(&relayd->control_sock, stream->name,
794 path, &stream->relayd_stream_id,
795 stream->chan->tracefile_size, stream->chan->tracefile_count);
796 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
797 if (ret < 0) {
798 goto end;
799 }
1c20f0e2 800
10a50311 801 uatomic_inc(&relayd->refcount);
d01178b6 802 stream->sent_to_relayd = 1;
10a50311
JD
803 } else {
804 ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.",
805 stream->key, stream->net_seq_idx);
806 ret = -1;
807 goto end;
808 }
809
810 DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64,
811 stream->name, stream->key, stream->net_seq_idx);
812
813end:
814 rcu_read_unlock();
815 return ret;
816}
817
a4baae1b
JD
818/*
819 * Find a relayd and send the streams sent message
820 *
821 * Returns 0 on success, < 0 on error
822 */
823int consumer_send_relayd_streams_sent(uint64_t net_seq_idx)
824{
825 int ret = 0;
826 struct consumer_relayd_sock_pair *relayd;
827
828 assert(net_seq_idx != -1ULL);
829
830 /* The stream is not metadata. Get relayd reference if exists. */
831 rcu_read_lock();
832 relayd = consumer_find_relayd(net_seq_idx);
833 if (relayd != NULL) {
834 /* Add stream on the relayd */
835 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
836 ret = relayd_streams_sent(&relayd->control_sock);
837 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
838 if (ret < 0) {
839 goto end;
840 }
841 } else {
842 ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.",
843 net_seq_idx);
844 ret = -1;
845 goto end;
846 }
847
848 ret = 0;
849 DBG("All streams sent relayd id %" PRIu64, net_seq_idx);
850
851end:
852 rcu_read_unlock();
853 return ret;
854}
855
10a50311
JD
856/*
857 * Find a relayd and close the stream
858 */
859void close_relayd_stream(struct lttng_consumer_stream *stream)
860{
861 struct consumer_relayd_sock_pair *relayd;
862
863 /* The stream is not metadata. Get relayd reference if exists. */
864 rcu_read_lock();
865 relayd = consumer_find_relayd(stream->net_seq_idx);
866 if (relayd) {
867 consumer_stream_relayd_close(stream, relayd);
868 }
869 rcu_read_unlock();
870}
871
00e2e675
DG
872/*
873 * Handle stream for relayd transmission if the stream applies for network
874 * streaming where the net sequence index is set.
875 *
876 * Return destination file descriptor or negative value on error.
877 */
6197aea7 878static int write_relayd_stream_header(struct lttng_consumer_stream *stream,
1d4dfdef
DG
879 size_t data_size, unsigned long padding,
880 struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
881{
882 int outfd = -1, ret;
00e2e675
DG
883 struct lttcomm_relayd_data_hdr data_hdr;
884
885 /* Safety net */
886 assert(stream);
6197aea7 887 assert(relayd);
00e2e675
DG
888
889 /* Reset data header */
890 memset(&data_hdr, 0, sizeof(data_hdr));
891
00e2e675
DG
892 if (stream->metadata_flag) {
893 /* Caller MUST acquire the relayd control socket lock */
894 ret = relayd_send_metadata(&relayd->control_sock, data_size);
895 if (ret < 0) {
896 goto error;
897 }
898
899 /* Metadata are always sent on the control socket. */
6151a90f 900 outfd = relayd->control_sock.sock.fd;
00e2e675
DG
901 } else {
902 /* Set header with stream information */
903 data_hdr.stream_id = htobe64(stream->relayd_stream_id);
904 data_hdr.data_size = htobe32(data_size);
1d4dfdef 905 data_hdr.padding_size = htobe32(padding);
39df6d9f
DG
906 /*
907 * Note that net_seq_num below is assigned with the *current* value of
908 * next_net_seq_num and only after that the next_net_seq_num will be
909 * increment. This is why when issuing a command on the relayd using
910 * this next value, 1 should always be substracted in order to compare
911 * the last seen sequence number on the relayd side to the last sent.
912 */
3604f373 913 data_hdr.net_seq_num = htobe64(stream->next_net_seq_num);
00e2e675
DG
914 /* Other fields are zeroed previously */
915
916 ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr,
917 sizeof(data_hdr));
918 if (ret < 0) {
919 goto error;
920 }
921
3604f373
DG
922 ++stream->next_net_seq_num;
923
00e2e675 924 /* Set to go on data socket */
6151a90f 925 outfd = relayd->data_sock.sock.fd;
00e2e675
DG
926 }
927
928error:
929 return outfd;
930}
931
3bd1e081 932/*
ffe60014
DG
933 * Allocate and return a new lttng_consumer_channel object using the given key
934 * to initialize the hash table node.
935 *
936 * On error, return NULL.
3bd1e081 937 */
886224ff 938struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key,
ffe60014
DG
939 uint64_t session_id,
940 const char *pathname,
941 const char *name,
942 uid_t uid,
943 gid_t gid,
57a269f2 944 uint64_t relayd_id,
1624d5b7
JD
945 enum lttng_event_output output,
946 uint64_t tracefile_size,
2bba9e53 947 uint64_t tracefile_count,
1950109e 948 uint64_t session_id_per_pid,
ecc48a90 949 unsigned int monitor,
d7ba1388 950 unsigned int live_timer_interval,
3d071855 951 const char *root_shm_path,
d7ba1388 952 const char *shm_path)
3bd1e081
MD
953{
954 struct lttng_consumer_channel *channel;
3bd1e081 955
276b26d1 956 channel = zmalloc(sizeof(*channel));
3bd1e081 957 if (channel == NULL) {
7a57cf92 958 PERROR("malloc struct lttng_consumer_channel");
3bd1e081
MD
959 goto end;
960 }
ffe60014
DG
961
962 channel->key = key;
3bd1e081 963 channel->refcount = 0;
ffe60014 964 channel->session_id = session_id;
1950109e 965 channel->session_id_per_pid = session_id_per_pid;
ffe60014
DG
966 channel->uid = uid;
967 channel->gid = gid;
968 channel->relayd_id = relayd_id;
1624d5b7
JD
969 channel->tracefile_size = tracefile_size;
970 channel->tracefile_count = tracefile_count;
2bba9e53 971 channel->monitor = monitor;
ecc48a90 972 channel->live_timer_interval = live_timer_interval;
a9838785 973 pthread_mutex_init(&channel->lock, NULL);
ec6ea7d0 974 pthread_mutex_init(&channel->timer_lock, NULL);
ffe60014 975
0c759fc9
DG
976 switch (output) {
977 case LTTNG_EVENT_SPLICE:
978 channel->output = CONSUMER_CHANNEL_SPLICE;
979 break;
980 case LTTNG_EVENT_MMAP:
981 channel->output = CONSUMER_CHANNEL_MMAP;
982 break;
983 default:
984 assert(0);
985 free(channel);
986 channel = NULL;
987 goto end;
988 }
989
07b86b52
JD
990 /*
991 * In monitor mode, the streams associated with the channel will be put in
992 * a special list ONLY owned by this channel. So, the refcount is set to 1
993 * here meaning that the channel itself has streams that are referenced.
994 *
995 * On a channel deletion, once the channel is no longer visible, the
996 * refcount is decremented and checked for a zero value to delete it. With
997 * streams in no monitor mode, it will now be safe to destroy the channel.
998 */
999 if (!channel->monitor) {
1000 channel->refcount = 1;
1001 }
1002
ffe60014
DG
1003 strncpy(channel->pathname, pathname, sizeof(channel->pathname));
1004 channel->pathname[sizeof(channel->pathname) - 1] = '\0';
1005
1006 strncpy(channel->name, name, sizeof(channel->name));
1007 channel->name[sizeof(channel->name) - 1] = '\0';
1008
3d071855
MD
1009 if (root_shm_path) {
1010 strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path));
1011 channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0';
1012 }
d7ba1388
MD
1013 if (shm_path) {
1014 strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path));
1015 channel->shm_path[sizeof(channel->shm_path) - 1] = '\0';
1016 }
1017
d88aee68 1018 lttng_ht_node_init_u64(&channel->node, channel->key);
d8ef542d
MD
1019
1020 channel->wait_fd = -1;
1021
ffe60014
DG
1022 CDS_INIT_LIST_HEAD(&channel->streams.head);
1023
62a7b8ed 1024 DBG("Allocated channel (key %" PRIu64 ")", channel->key);
3bd1e081 1025
3bd1e081
MD
1026end:
1027 return channel;
1028}
1029
1030/*
1031 * Add a channel to the global list protected by a mutex.
821fffb2 1032 *
b5a6470f 1033 * Always return 0 indicating success.
3bd1e081 1034 */
d8ef542d
MD
1035int consumer_add_channel(struct lttng_consumer_channel *channel,
1036 struct lttng_consumer_local_data *ctx)
3bd1e081 1037{
3bd1e081 1038 pthread_mutex_lock(&consumer_data.lock);
a9838785 1039 pthread_mutex_lock(&channel->lock);
ec6ea7d0 1040 pthread_mutex_lock(&channel->timer_lock);
c77fc10a 1041
b5a6470f
DG
1042 /*
1043 * This gives us a guarantee that the channel we are about to add to the
1044 * channel hash table will be unique. See this function comment on the why
1045 * we need to steel the channel key at this stage.
1046 */
1047 steal_channel_key(channel->key);
c77fc10a 1048
b5a6470f 1049 rcu_read_lock();
d88aee68 1050 lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node);
6065ceec 1051 rcu_read_unlock();
b5a6470f 1052
ec6ea7d0 1053 pthread_mutex_unlock(&channel->timer_lock);
a9838785 1054 pthread_mutex_unlock(&channel->lock);
3bd1e081 1055 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 1056
b5a6470f 1057 if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) {
a0cbdd2e 1058 notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD);
d8ef542d 1059 }
b5a6470f
DG
1060
1061 return 0;
3bd1e081
MD
1062}
1063
1064/*
1065 * Allocate the pollfd structure and the local view of the out fds to avoid
1066 * doing a lookup in the linked list and concurrency issues when writing is
1067 * needed. Called with consumer_data.lock held.
1068 *
1069 * Returns the number of fds in the structures.
1070 */
ffe60014
DG
1071static int update_poll_array(struct lttng_consumer_local_data *ctx,
1072 struct pollfd **pollfd, struct lttng_consumer_stream **local_stream,
1073 struct lttng_ht *ht)
3bd1e081 1074{
3bd1e081 1075 int i = 0;
e4421fec
DG
1076 struct lttng_ht_iter iter;
1077 struct lttng_consumer_stream *stream;
3bd1e081 1078
ffe60014
DG
1079 assert(ctx);
1080 assert(ht);
1081 assert(pollfd);
1082 assert(local_stream);
1083
3bd1e081 1084 DBG("Updating poll fd array");
481d6c57 1085 rcu_read_lock();
43c34bc3 1086 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
8994307f
DG
1087 /*
1088 * Only active streams with an active end point can be added to the
1089 * poll set and local stream storage of the thread.
1090 *
1091 * There is a potential race here for endpoint_status to be updated
1092 * just after the check. However, this is OK since the stream(s) will
1093 * be deleted once the thread is notified that the end point state has
1094 * changed where this function will be called back again.
1095 */
1096 if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM ||
79d4ffb7 1097 stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) {
3bd1e081
MD
1098 continue;
1099 }
7972aab2
DG
1100 /*
1101 * This clobbers way too much the debug output. Uncomment that if you
1102 * need it for debugging purposes.
1103 *
1104 * DBG("Active FD %d", stream->wait_fd);
1105 */
e4421fec 1106 (*pollfd)[i].fd = stream->wait_fd;
3bd1e081 1107 (*pollfd)[i].events = POLLIN | POLLPRI;
e4421fec 1108 local_stream[i] = stream;
3bd1e081
MD
1109 i++;
1110 }
481d6c57 1111 rcu_read_unlock();
3bd1e081
MD
1112
1113 /*
50f8ae69 1114 * Insert the consumer_data_pipe at the end of the array and don't
3bd1e081
MD
1115 * increment i so nb_fd is the number of real FD.
1116 */
acdb9057 1117 (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe);
509bb1cf 1118 (*pollfd)[i].events = POLLIN | POLLPRI;
02b3d176
DG
1119
1120 (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe);
1121 (*pollfd)[i + 1].events = POLLIN | POLLPRI;
3bd1e081
MD
1122 return i;
1123}
1124
1125/*
84382d49
MD
1126 * Poll on the should_quit pipe and the command socket return -1 on
1127 * error, 1 if should exit, 0 if data is available on the command socket
3bd1e081
MD
1128 */
1129int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll)
1130{
1131 int num_rdy;
1132
88f2b785 1133restart:
3bd1e081
MD
1134 num_rdy = poll(consumer_sockpoll, 2, -1);
1135 if (num_rdy == -1) {
88f2b785
MD
1136 /*
1137 * Restart interrupted system call.
1138 */
1139 if (errno == EINTR) {
1140 goto restart;
1141 }
7a57cf92 1142 PERROR("Poll error");
84382d49 1143 return -1;
3bd1e081 1144 }
509bb1cf 1145 if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) {
3bd1e081 1146 DBG("consumer_should_quit wake up");
84382d49 1147 return 1;
3bd1e081
MD
1148 }
1149 return 0;
3bd1e081
MD
1150}
1151
1152/*
1153 * Set the error socket.
1154 */
ffe60014
DG
1155void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx,
1156 int sock)
3bd1e081
MD
1157{
1158 ctx->consumer_error_socket = sock;
1159}
1160
1161/*
1162 * Set the command socket path.
1163 */
3bd1e081
MD
1164void lttng_consumer_set_command_sock_path(
1165 struct lttng_consumer_local_data *ctx, char *sock)
1166{
1167 ctx->consumer_command_sock_path = sock;
1168}
1169
1170/*
1171 * Send return code to the session daemon.
1172 * If the socket is not defined, we return 0, it is not a fatal error
1173 */
ffe60014 1174int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd)
3bd1e081
MD
1175{
1176 if (ctx->consumer_error_socket > 0) {
1177 return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd,
1178 sizeof(enum lttcomm_sessiond_command));
1179 }
1180
1181 return 0;
1182}
1183
1184/*
228b5bf7
DG
1185 * Close all the tracefiles and stream fds and MUST be called when all
1186 * instances are destroyed i.e. when all threads were joined and are ended.
3bd1e081
MD
1187 */
1188void lttng_consumer_cleanup(void)
1189{
e4421fec 1190 struct lttng_ht_iter iter;
ffe60014 1191 struct lttng_consumer_channel *channel;
6065ceec
DG
1192
1193 rcu_read_lock();
3bd1e081 1194
ffe60014
DG
1195 cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel,
1196 node.node) {
702b1ea4 1197 consumer_del_channel(channel);
3bd1e081 1198 }
6065ceec
DG
1199
1200 rcu_read_unlock();
d6ce1df2 1201
d6ce1df2 1202 lttng_ht_destroy(consumer_data.channel_ht);
228b5bf7
DG
1203
1204 cleanup_relayd_ht();
1205
d8ef542d
MD
1206 lttng_ht_destroy(consumer_data.stream_per_chan_id_ht);
1207
228b5bf7
DG
1208 /*
1209 * This HT contains streams that are freed by either the metadata thread or
1210 * the data thread so we do *nothing* on the hash table and simply destroy
1211 * it.
1212 */
1213 lttng_ht_destroy(consumer_data.stream_list_ht);
3bd1e081
MD
1214}
1215
1216/*
1217 * Called from signal handler.
1218 */
1219void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx)
1220{
6cd525e8
MD
1221 ssize_t ret;
1222
3bd1e081 1223 consumer_quit = 1;
6cd525e8
MD
1224 ret = lttng_write(ctx->consumer_should_quit[1], "4", 1);
1225 if (ret < 1) {
7a57cf92 1226 PERROR("write consumer quit");
3bd1e081 1227 }
ab1027f4
DG
1228
1229 DBG("Consumer flag that it should quit");
3bd1e081
MD
1230}
1231
00e2e675
DG
1232void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream,
1233 off_t orig_offset)
3bd1e081 1234{
c7a78aab 1235 int ret;
3bd1e081
MD
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 */
c7a78aab 1266 ret = posix_fadvise(outfd, orig_offset - stream->max_sb_size,
ffe60014 1267 stream->max_sb_size, POSIX_FADV_DONTNEED);
a0d0e127
JG
1268 if (ret && ret != -ENOSYS) {
1269 errno = -ret;
1270 PERROR("posix_fadvise");
c7a78aab 1271 }
3bd1e081
MD
1272}
1273
1274/*
1275 * Initialise the necessary environnement :
1276 * - create a new context
1277 * - create the poll_pipe
1278 * - create the should_quit pipe (for signal handler)
1279 * - create the thread pipe (for splice)
1280 *
1281 * Takes a function pointer as argument, this function is called when data is
1282 * available on a buffer. This function is responsible to do the
1283 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1284 * buffer configuration and then kernctl_put_next_subbuf at the end.
1285 *
1286 * Returns a pointer to the new context or NULL on error.
1287 */
1288struct lttng_consumer_local_data *lttng_consumer_create(
1289 enum lttng_consumer_type type,
4078b776 1290 ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream,
d41f73b7 1291 struct lttng_consumer_local_data *ctx),
3bd1e081
MD
1292 int (*recv_channel)(struct lttng_consumer_channel *channel),
1293 int (*recv_stream)(struct lttng_consumer_stream *stream),
30319bcb 1294 int (*update_stream)(uint64_t stream_key, uint32_t state))
3bd1e081 1295{
d8ef542d 1296 int ret;
3bd1e081
MD
1297 struct lttng_consumer_local_data *ctx;
1298
1299 assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN ||
1300 consumer_data.type == type);
1301 consumer_data.type = type;
1302
effcf122 1303 ctx = zmalloc(sizeof(struct lttng_consumer_local_data));
3bd1e081 1304 if (ctx == NULL) {
7a57cf92 1305 PERROR("allocating context");
3bd1e081
MD
1306 goto error;
1307 }
1308
1309 ctx->consumer_error_socket = -1;
331744e3 1310 ctx->consumer_metadata_socket = -1;
75d83e50 1311 pthread_mutex_init(&ctx->metadata_socket_lock, NULL);
3bd1e081
MD
1312 /* assign the callbacks */
1313 ctx->on_buffer_ready = buffer_ready;
1314 ctx->on_recv_channel = recv_channel;
1315 ctx->on_recv_stream = recv_stream;
1316 ctx->on_update_stream = update_stream;
1317
acdb9057
DG
1318 ctx->consumer_data_pipe = lttng_pipe_open(0);
1319 if (!ctx->consumer_data_pipe) {
3bd1e081
MD
1320 goto error_poll_pipe;
1321 }
1322
02b3d176
DG
1323 ctx->consumer_wakeup_pipe = lttng_pipe_open(0);
1324 if (!ctx->consumer_wakeup_pipe) {
1325 goto error_wakeup_pipe;
1326 }
1327
3bd1e081
MD
1328 ret = pipe(ctx->consumer_should_quit);
1329 if (ret < 0) {
7a57cf92 1330 PERROR("Error creating recv pipe");
3bd1e081
MD
1331 goto error_quit_pipe;
1332 }
1333
d8ef542d
MD
1334 ret = pipe(ctx->consumer_channel_pipe);
1335 if (ret < 0) {
1336 PERROR("Error creating channel pipe");
1337 goto error_channel_pipe;
1338 }
1339
13886d2d
DG
1340 ctx->consumer_metadata_pipe = lttng_pipe_open(0);
1341 if (!ctx->consumer_metadata_pipe) {
fb3a43a9
DG
1342 goto error_metadata_pipe;
1343 }
3bd1e081 1344
fb3a43a9 1345 return ctx;
3bd1e081 1346
fb3a43a9 1347error_metadata_pipe:
d8ef542d
MD
1348 utils_close_pipe(ctx->consumer_channel_pipe);
1349error_channel_pipe:
d8ef542d 1350 utils_close_pipe(ctx->consumer_should_quit);
3bd1e081 1351error_quit_pipe:
02b3d176
DG
1352 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
1353error_wakeup_pipe:
acdb9057 1354 lttng_pipe_destroy(ctx->consumer_data_pipe);
3bd1e081
MD
1355error_poll_pipe:
1356 free(ctx);
1357error:
1358 return NULL;
1359}
1360
282dadbc
MD
1361/*
1362 * Iterate over all streams of the hashtable and free them properly.
1363 */
1364static void destroy_data_stream_ht(struct lttng_ht *ht)
1365{
1366 struct lttng_ht_iter iter;
1367 struct lttng_consumer_stream *stream;
1368
1369 if (ht == NULL) {
1370 return;
1371 }
1372
1373 rcu_read_lock();
1374 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1375 /*
1376 * Ignore return value since we are currently cleaning up so any error
1377 * can't be handled.
1378 */
1379 (void) consumer_del_stream(stream, ht);
1380 }
1381 rcu_read_unlock();
1382
1383 lttng_ht_destroy(ht);
1384}
1385
1386/*
1387 * Iterate over all streams of the metadata hashtable and free them
1388 * properly.
1389 */
1390static void destroy_metadata_stream_ht(struct lttng_ht *ht)
1391{
1392 struct lttng_ht_iter iter;
1393 struct lttng_consumer_stream *stream;
1394
1395 if (ht == NULL) {
1396 return;
1397 }
1398
1399 rcu_read_lock();
1400 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1401 /*
1402 * Ignore return value since we are currently cleaning up so any error
1403 * can't be handled.
1404 */
1405 (void) consumer_del_metadata_stream(stream, ht);
1406 }
1407 rcu_read_unlock();
1408
1409 lttng_ht_destroy(ht);
1410}
1411
3bd1e081
MD
1412/*
1413 * Close all fds associated with the instance and free the context.
1414 */
1415void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx)
1416{
4c462e79
MD
1417 int ret;
1418
ab1027f4
DG
1419 DBG("Consumer destroying it. Closing everything.");
1420
4f2e75b9
DG
1421 if (!ctx) {
1422 return;
1423 }
1424
282dadbc
MD
1425 destroy_data_stream_ht(data_ht);
1426 destroy_metadata_stream_ht(metadata_ht);
1427
4c462e79
MD
1428 ret = close(ctx->consumer_error_socket);
1429 if (ret) {
1430 PERROR("close");
1431 }
331744e3
JD
1432 ret = close(ctx->consumer_metadata_socket);
1433 if (ret) {
1434 PERROR("close");
1435 }
d8ef542d 1436 utils_close_pipe(ctx->consumer_channel_pipe);
acdb9057 1437 lttng_pipe_destroy(ctx->consumer_data_pipe);
13886d2d 1438 lttng_pipe_destroy(ctx->consumer_metadata_pipe);
02b3d176 1439 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
d8ef542d 1440 utils_close_pipe(ctx->consumer_should_quit);
fb3a43a9 1441
3bd1e081
MD
1442 unlink(ctx->consumer_command_sock_path);
1443 free(ctx);
1444}
1445
6197aea7
DG
1446/*
1447 * Write the metadata stream id on the specified file descriptor.
1448 */
1449static int write_relayd_metadata_id(int fd,
1450 struct lttng_consumer_stream *stream,
ffe60014 1451 struct consumer_relayd_sock_pair *relayd, unsigned long padding)
6197aea7 1452{
6cd525e8 1453 ssize_t ret;
1d4dfdef 1454 struct lttcomm_relayd_metadata_payload hdr;
6197aea7 1455
1d4dfdef
DG
1456 hdr.stream_id = htobe64(stream->relayd_stream_id);
1457 hdr.padding_size = htobe32(padding);
6cd525e8
MD
1458 ret = lttng_write(fd, (void *) &hdr, sizeof(hdr));
1459 if (ret < sizeof(hdr)) {
d7b75ec8 1460 /*
6f04ed72 1461 * This error means that the fd's end is closed so ignore the PERROR
d7b75ec8
DG
1462 * not to clubber the error output since this can happen in a normal
1463 * code path.
1464 */
1465 if (errno != EPIPE) {
1466 PERROR("write metadata stream id");
1467 }
1468 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno);
534d2592
DG
1469 /*
1470 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1471 * handle writting the missing part so report that as an error and
1472 * don't lie to the caller.
1473 */
1474 ret = -1;
6197aea7
DG
1475 goto end;
1476 }
1d4dfdef
DG
1477 DBG("Metadata stream id %" PRIu64 " with padding %lu written before data",
1478 stream->relayd_stream_id, padding);
6197aea7
DG
1479
1480end:
6cd525e8 1481 return (int) ret;
6197aea7
DG
1482}
1483
3bd1e081 1484/*
09e26845
DG
1485 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1486 * core function for writing trace buffers to either the local filesystem or
1487 * the network.
1488 *
79d4ffb7
DG
1489 * It must be called with the stream lock held.
1490 *
09e26845 1491 * Careful review MUST be put if any changes occur!
3bd1e081
MD
1492 *
1493 * Returns the number of bytes written
1494 */
4078b776 1495ssize_t lttng_consumer_on_read_subbuffer_mmap(
3bd1e081 1496 struct lttng_consumer_local_data *ctx,
1d4dfdef 1497 struct lttng_consumer_stream *stream, unsigned long len,
309167d2 1498 unsigned long padding,
50adc264 1499 struct ctf_packet_index *index)
3bd1e081 1500{
f02e1e8a 1501 unsigned long mmap_offset;
ffe60014 1502 void *mmap_base;
994ab360 1503 ssize_t ret = 0;
f02e1e8a
DG
1504 off_t orig_offset = stream->out_fd_offset;
1505 /* Default is on the disk */
1506 int outfd = stream->out_fd;
f02e1e8a 1507 struct consumer_relayd_sock_pair *relayd = NULL;
8994307f 1508 unsigned int relayd_hang_up = 0;
f02e1e8a
DG
1509
1510 /* RCU lock for the relayd pointer */
1511 rcu_read_lock();
1512
1513 /* Flag that the current stream if set for network streaming. */
da009f2c 1514 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1515 relayd = consumer_find_relayd(stream->net_seq_idx);
1516 if (relayd == NULL) {
56591bac 1517 ret = -EPIPE;
f02e1e8a
DG
1518 goto end;
1519 }
1520 }
1521
1522 /* get the offset inside the fd to mmap */
3bd1e081
MD
1523 switch (consumer_data.type) {
1524 case LTTNG_CONSUMER_KERNEL:
ffe60014 1525 mmap_base = stream->mmap_base;
f02e1e8a 1526 ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset);
994ab360
DG
1527 if (ret < 0) {
1528 ret = -errno;
56591bac 1529 PERROR("tracer ctl get_mmap_read_offset");
56591bac
MD
1530 goto end;
1531 }
f02e1e8a 1532 break;
7753dea8
MD
1533 case LTTNG_CONSUMER32_UST:
1534 case LTTNG_CONSUMER64_UST:
ffe60014
DG
1535 mmap_base = lttng_ustctl_get_mmap_base(stream);
1536 if (!mmap_base) {
1537 ERR("read mmap get mmap base for stream %s", stream->name);
994ab360 1538 ret = -EPERM;
ffe60014
DG
1539 goto end;
1540 }
1541 ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset);
56591bac
MD
1542 if (ret != 0) {
1543 PERROR("tracer ctl get_mmap_read_offset");
994ab360 1544 ret = -EINVAL;
56591bac
MD
1545 goto end;
1546 }
f02e1e8a 1547 break;
3bd1e081
MD
1548 default:
1549 ERR("Unknown consumer_data type");
1550 assert(0);
1551 }
b9182dd9 1552
f02e1e8a
DG
1553 /* Handle stream on the relayd if the output is on the network */
1554 if (relayd) {
1555 unsigned long netlen = len;
1556
1557 /*
1558 * Lock the control socket for the complete duration of the function
1559 * since from this point on we will use the socket.
1560 */
1561 if (stream->metadata_flag) {
1562 /* Metadata requires the control socket. */
1563 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
93ec662e
JD
1564 if (stream->reset_metadata_flag) {
1565 ret = relayd_reset_metadata(&relayd->control_sock,
1566 stream->relayd_stream_id,
1567 stream->metadata_version);
1568 if (ret < 0) {
1569 relayd_hang_up = 1;
1570 goto write_error;
1571 }
1572 stream->reset_metadata_flag = 0;
1573 }
1d4dfdef 1574 netlen += sizeof(struct lttcomm_relayd_metadata_payload);
f02e1e8a
DG
1575 }
1576
1d4dfdef 1577 ret = write_relayd_stream_header(stream, netlen, padding, relayd);
994ab360
DG
1578 if (ret < 0) {
1579 relayd_hang_up = 1;
1580 goto write_error;
1581 }
1582 /* Use the returned socket. */
1583 outfd = ret;
f02e1e8a 1584
994ab360
DG
1585 /* Write metadata stream id before payload */
1586 if (stream->metadata_flag) {
1587 ret = write_relayd_metadata_id(outfd, stream, relayd, padding);
1588 if (ret < 0) {
8994307f
DG
1589 relayd_hang_up = 1;
1590 goto write_error;
1591 }
f02e1e8a 1592 }
1d4dfdef
DG
1593 } else {
1594 /* No streaming, we have to set the len with the full padding */
1595 len += padding;
1624d5b7 1596
93ec662e
JD
1597 if (stream->metadata_flag && stream->reset_metadata_flag) {
1598 ret = utils_truncate_stream_file(stream->out_fd, 0);
1599 if (ret < 0) {
1600 ERR("Reset metadata file");
1601 goto end;
1602 }
1603 stream->reset_metadata_flag = 0;
1604 }
1605
1624d5b7
JD
1606 /*
1607 * Check if we need to change the tracefile before writing the packet.
1608 */
1609 if (stream->chan->tracefile_size > 0 &&
1610 (stream->tracefile_size_current + len) >
1611 stream->chan->tracefile_size) {
fe4477ee
JD
1612 ret = utils_rotate_stream_file(stream->chan->pathname,
1613 stream->name, stream->chan->tracefile_size,
1614 stream->chan->tracefile_count, stream->uid, stream->gid,
309167d2
JD
1615 stream->out_fd, &(stream->tracefile_count_current),
1616 &stream->out_fd);
1624d5b7
JD
1617 if (ret < 0) {
1618 ERR("Rotating output file");
1619 goto end;
1620 }
309167d2
JD
1621 outfd = stream->out_fd;
1622
1623 if (stream->index_fd >= 0) {
adf7a0a1
JG
1624 ret = close(stream->index_fd);
1625 if (ret < 0) {
1626 PERROR("Closing index");
1627 goto end;
1628 }
1629 stream->index_fd = -1;
309167d2
JD
1630 ret = index_create_file(stream->chan->pathname,
1631 stream->name, stream->uid, stream->gid,
1632 stream->chan->tracefile_size,
1633 stream->tracefile_count_current);
1634 if (ret < 0) {
1635 goto end;
1636 }
1637 stream->index_fd = ret;
1638 }
1639
a6976990
DG
1640 /* Reset current size because we just perform a rotation. */
1641 stream->tracefile_size_current = 0;
a1ae300f
JD
1642 stream->out_fd_offset = 0;
1643 orig_offset = 0;
1624d5b7
JD
1644 }
1645 stream->tracefile_size_current += len;
309167d2
JD
1646 if (index) {
1647 index->offset = htobe64(stream->out_fd_offset);
1648 }
f02e1e8a
DG
1649 }
1650
d02b8372
DG
1651 /*
1652 * This call guarantee that len or less is returned. It's impossible to
1653 * receive a ret value that is bigger than len.
1654 */
1655 ret = lttng_write(outfd, mmap_base + mmap_offset, len);
1656 DBG("Consumer mmap write() ret %zd (len %lu)", ret, len);
1657 if (ret < 0 || ((size_t) ret != len)) {
1658 /*
1659 * Report error to caller if nothing was written else at least send the
1660 * amount written.
1661 */
1662 if (ret < 0) {
994ab360 1663 ret = -errno;
f02e1e8a 1664 }
994ab360 1665 relayd_hang_up = 1;
f02e1e8a 1666
d02b8372 1667 /* Socket operation failed. We consider the relayd dead */
994ab360 1668 if (errno == EPIPE || errno == EINVAL || errno == EBADF) {
d02b8372
DG
1669 /*
1670 * This is possible if the fd is closed on the other side
1671 * (outfd) or any write problem. It can be verbose a bit for a
1672 * normal execution if for instance the relayd is stopped
1673 * abruptly. This can happen so set this to a DBG statement.
1674 */
1675 DBG("Consumer mmap write detected relayd hang up");
994ab360
DG
1676 } else {
1677 /* Unhandled error, print it and stop function right now. */
1678 PERROR("Error in write mmap (ret %zd != len %lu)", ret, len);
f02e1e8a 1679 }
994ab360 1680 goto write_error;
d02b8372
DG
1681 }
1682 stream->output_written += ret;
d02b8372
DG
1683
1684 /* This call is useless on a socket so better save a syscall. */
1685 if (!relayd) {
1686 /* This won't block, but will start writeout asynchronously */
1687 lttng_sync_file_range(outfd, stream->out_fd_offset, len,
1688 SYNC_FILE_RANGE_WRITE);
1689 stream->out_fd_offset += len;
f02e1e8a
DG
1690 }
1691 lttng_consumer_sync_trace_file(stream, orig_offset);
1692
8994307f
DG
1693write_error:
1694 /*
1695 * This is a special case that the relayd has closed its socket. Let's
1696 * cleanup the relayd object and all associated streams.
1697 */
1698 if (relayd && relayd_hang_up) {
1699 cleanup_relayd(relayd, ctx);
1700 }
1701
f02e1e8a
DG
1702end:
1703 /* Unlock only if ctrl socket used */
1704 if (relayd && stream->metadata_flag) {
1705 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1706 }
1707
1708 rcu_read_unlock();
994ab360 1709 return ret;
3bd1e081
MD
1710}
1711
1712/*
1713 * Splice the data from the ring buffer to the tracefile.
1714 *
79d4ffb7
DG
1715 * It must be called with the stream lock held.
1716 *
3bd1e081
MD
1717 * Returns the number of bytes spliced.
1718 */
4078b776 1719ssize_t lttng_consumer_on_read_subbuffer_splice(
3bd1e081 1720 struct lttng_consumer_local_data *ctx,
1d4dfdef 1721 struct lttng_consumer_stream *stream, unsigned long len,
309167d2 1722 unsigned long padding,
50adc264 1723 struct ctf_packet_index *index)
3bd1e081 1724{
f02e1e8a
DG
1725 ssize_t ret = 0, written = 0, ret_splice = 0;
1726 loff_t offset = 0;
1727 off_t orig_offset = stream->out_fd_offset;
1728 int fd = stream->wait_fd;
1729 /* Default is on the disk */
1730 int outfd = stream->out_fd;
f02e1e8a 1731 struct consumer_relayd_sock_pair *relayd = NULL;
fb3a43a9 1732 int *splice_pipe;
8994307f 1733 unsigned int relayd_hang_up = 0;
f02e1e8a 1734
3bd1e081
MD
1735 switch (consumer_data.type) {
1736 case LTTNG_CONSUMER_KERNEL:
f02e1e8a 1737 break;
7753dea8
MD
1738 case LTTNG_CONSUMER32_UST:
1739 case LTTNG_CONSUMER64_UST:
f02e1e8a 1740 /* Not supported for user space tracing */
3bd1e081
MD
1741 return -ENOSYS;
1742 default:
1743 ERR("Unknown consumer_data type");
1744 assert(0);
3bd1e081
MD
1745 }
1746
f02e1e8a
DG
1747 /* RCU lock for the relayd pointer */
1748 rcu_read_lock();
1749
1750 /* Flag that the current stream if set for network streaming. */
da009f2c 1751 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1752 relayd = consumer_find_relayd(stream->net_seq_idx);
1753 if (relayd == NULL) {
ad0b0d23 1754 written = -ret;
f02e1e8a
DG
1755 goto end;
1756 }
1757 }
a2361a61 1758 splice_pipe = stream->splice_pipe;
fb3a43a9 1759
f02e1e8a 1760 /* Write metadata stream id before payload */
1d4dfdef 1761 if (relayd) {
ad0b0d23 1762 unsigned long total_len = len;
f02e1e8a 1763
1d4dfdef
DG
1764 if (stream->metadata_flag) {
1765 /*
1766 * Lock the control socket for the complete duration of the function
1767 * since from this point on we will use the socket.
1768 */
1769 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1770
93ec662e
JD
1771 if (stream->reset_metadata_flag) {
1772 ret = relayd_reset_metadata(&relayd->control_sock,
1773 stream->relayd_stream_id,
1774 stream->metadata_version);
1775 if (ret < 0) {
1776 relayd_hang_up = 1;
1777 goto write_error;
1778 }
1779 stream->reset_metadata_flag = 0;
1780 }
1d4dfdef
DG
1781 ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd,
1782 padding);
1783 if (ret < 0) {
1784 written = ret;
ad0b0d23
DG
1785 relayd_hang_up = 1;
1786 goto write_error;
1d4dfdef
DG
1787 }
1788
1789 total_len += sizeof(struct lttcomm_relayd_metadata_payload);
1790 }
1791
1792 ret = write_relayd_stream_header(stream, total_len, padding, relayd);
ad0b0d23
DG
1793 if (ret < 0) {
1794 written = ret;
1795 relayd_hang_up = 1;
1796 goto write_error;
f02e1e8a 1797 }
ad0b0d23
DG
1798 /* Use the returned socket. */
1799 outfd = ret;
1d4dfdef
DG
1800 } else {
1801 /* No streaming, we have to set the len with the full padding */
1802 len += padding;
1624d5b7 1803
93ec662e
JD
1804 if (stream->metadata_flag && stream->reset_metadata_flag) {
1805 ret = utils_truncate_stream_file(stream->out_fd, 0);
1806 if (ret < 0) {
1807 ERR("Reset metadata file");
1808 goto end;
1809 }
1810 stream->reset_metadata_flag = 0;
1811 }
1624d5b7
JD
1812 /*
1813 * Check if we need to change the tracefile before writing the packet.
1814 */
1815 if (stream->chan->tracefile_size > 0 &&
1816 (stream->tracefile_size_current + len) >
1817 stream->chan->tracefile_size) {
fe4477ee
JD
1818 ret = utils_rotate_stream_file(stream->chan->pathname,
1819 stream->name, stream->chan->tracefile_size,
1820 stream->chan->tracefile_count, stream->uid, stream->gid,
309167d2
JD
1821 stream->out_fd, &(stream->tracefile_count_current),
1822 &stream->out_fd);
1624d5b7 1823 if (ret < 0) {
ad0b0d23 1824 written = ret;
1624d5b7
JD
1825 ERR("Rotating output file");
1826 goto end;
1827 }
309167d2
JD
1828 outfd = stream->out_fd;
1829
1830 if (stream->index_fd >= 0) {
3c1150e1
JD
1831 ret = close(stream->index_fd);
1832 if (ret < 0) {
1833 PERROR("Closing index");
1834 goto end;
1835 }
1836 stream->index_fd = -1;
309167d2
JD
1837 ret = index_create_file(stream->chan->pathname,
1838 stream->name, stream->uid, stream->gid,
1839 stream->chan->tracefile_size,
1840 stream->tracefile_count_current);
1841 if (ret < 0) {
ad0b0d23 1842 written = ret;
309167d2
JD
1843 goto end;
1844 }
1845 stream->index_fd = ret;
1846 }
1847
a6976990
DG
1848 /* Reset current size because we just perform a rotation. */
1849 stream->tracefile_size_current = 0;
a1ae300f
JD
1850 stream->out_fd_offset = 0;
1851 orig_offset = 0;
1624d5b7
JD
1852 }
1853 stream->tracefile_size_current += len;
309167d2 1854 index->offset = htobe64(stream->out_fd_offset);
f02e1e8a
DG
1855 }
1856
1857 while (len > 0) {
1d4dfdef
DG
1858 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1859 (unsigned long)offset, len, fd, splice_pipe[1]);
fb3a43a9 1860 ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len,
f02e1e8a
DG
1861 SPLICE_F_MOVE | SPLICE_F_MORE);
1862 DBG("splice chan to pipe, ret %zd", ret_splice);
1863 if (ret_splice < 0) {
d02b8372 1864 ret = errno;
ad0b0d23 1865 written = -ret;
d02b8372 1866 PERROR("Error in relay splice");
f02e1e8a
DG
1867 goto splice_error;
1868 }
1869
1870 /* Handle stream on the relayd if the output is on the network */
ad0b0d23
DG
1871 if (relayd && stream->metadata_flag) {
1872 size_t metadata_payload_size =
1873 sizeof(struct lttcomm_relayd_metadata_payload);
1874
1875 /* Update counter to fit the spliced data */
1876 ret_splice += metadata_payload_size;
1877 len += metadata_payload_size;
1878 /*
1879 * We do this so the return value can match the len passed as
1880 * argument to this function.
1881 */
1882 written -= metadata_payload_size;
f02e1e8a
DG
1883 }
1884
1885 /* Splice data out */
fb3a43a9 1886 ret_splice = splice(splice_pipe[0], NULL, outfd, NULL,
f02e1e8a 1887 ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE);
a2361a61
JD
1888 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1889 outfd, ret_splice);
f02e1e8a 1890 if (ret_splice < 0) {
d02b8372 1891 ret = errno;
ad0b0d23
DG
1892 written = -ret;
1893 relayd_hang_up = 1;
1894 goto write_error;
f02e1e8a 1895 } else if (ret_splice > len) {
d02b8372
DG
1896 /*
1897 * We don't expect this code path to be executed but you never know
1898 * so this is an extra protection agains a buggy splice().
1899 */
f02e1e8a 1900 ret = errno;
ad0b0d23 1901 written += ret_splice;
d02b8372
DG
1902 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice,
1903 len);
f02e1e8a 1904 goto splice_error;
d02b8372
DG
1905 } else {
1906 /* All good, update current len and continue. */
1907 len -= ret_splice;
f02e1e8a 1908 }
f02e1e8a
DG
1909
1910 /* This call is useless on a socket so better save a syscall. */
1911 if (!relayd) {
1912 /* This won't block, but will start writeout asynchronously */
1913 lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice,
1914 SYNC_FILE_RANGE_WRITE);
1915 stream->out_fd_offset += ret_splice;
1916 }
e5d1a9b3 1917 stream->output_written += ret_splice;
f02e1e8a
DG
1918 written += ret_splice;
1919 }
1920 lttng_consumer_sync_trace_file(stream, orig_offset);
f02e1e8a
DG
1921 goto end;
1922
8994307f
DG
1923write_error:
1924 /*
1925 * This is a special case that the relayd has closed its socket. Let's
1926 * cleanup the relayd object and all associated streams.
1927 */
1928 if (relayd && relayd_hang_up) {
1929 cleanup_relayd(relayd, ctx);
1930 /* Skip splice error so the consumer does not fail */
1931 goto end;
1932 }
1933
f02e1e8a
DG
1934splice_error:
1935 /* send the appropriate error description to sessiond */
1936 switch (ret) {
f02e1e8a 1937 case EINVAL:
f73fabfd 1938 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL);
f02e1e8a
DG
1939 break;
1940 case ENOMEM:
f73fabfd 1941 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM);
f02e1e8a
DG
1942 break;
1943 case ESPIPE:
f73fabfd 1944 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE);
f02e1e8a
DG
1945 break;
1946 }
1947
1948end:
1949 if (relayd && stream->metadata_flag) {
1950 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1951 }
1952
1953 rcu_read_unlock();
1954 return written;
3bd1e081
MD
1955}
1956
1957/*
1958 * Take a snapshot for a specific fd
1959 *
1960 * Returns 0 on success, < 0 on error
1961 */
ffe60014 1962int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream)
3bd1e081
MD
1963{
1964 switch (consumer_data.type) {
1965 case LTTNG_CONSUMER_KERNEL:
ffe60014 1966 return lttng_kconsumer_take_snapshot(stream);
7753dea8
MD
1967 case LTTNG_CONSUMER32_UST:
1968 case LTTNG_CONSUMER64_UST:
ffe60014 1969 return lttng_ustconsumer_take_snapshot(stream);
3bd1e081
MD
1970 default:
1971 ERR("Unknown consumer_data type");
1972 assert(0);
1973 return -ENOSYS;
1974 }
3bd1e081
MD
1975}
1976
1977/*
1978 * Get the produced position
1979 *
1980 * Returns 0 on success, < 0 on error
1981 */
ffe60014 1982int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
3bd1e081
MD
1983 unsigned long *pos)
1984{
1985 switch (consumer_data.type) {
1986 case LTTNG_CONSUMER_KERNEL:
ffe60014 1987 return lttng_kconsumer_get_produced_snapshot(stream, pos);
7753dea8
MD
1988 case LTTNG_CONSUMER32_UST:
1989 case LTTNG_CONSUMER64_UST:
ffe60014 1990 return lttng_ustconsumer_get_produced_snapshot(stream, pos);
3bd1e081
MD
1991 default:
1992 ERR("Unknown consumer_data type");
1993 assert(0);
1994 return -ENOSYS;
1995 }
1996}
1997
1998int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1999 int sock, struct pollfd *consumer_sockpoll)
2000{
2001 switch (consumer_data.type) {
2002 case LTTNG_CONSUMER_KERNEL:
2003 return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
7753dea8
MD
2004 case LTTNG_CONSUMER32_UST:
2005 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
2006 return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
2007 default:
2008 ERR("Unknown consumer_data type");
2009 assert(0);
2010 return -ENOSYS;
2011 }
2012}
2013
6d574024 2014void lttng_consumer_close_all_metadata(void)
d88aee68
DG
2015{
2016 switch (consumer_data.type) {
2017 case LTTNG_CONSUMER_KERNEL:
2018 /*
2019 * The Kernel consumer has a different metadata scheme so we don't
2020 * close anything because the stream will be closed by the session
2021 * daemon.
2022 */
2023 break;
2024 case LTTNG_CONSUMER32_UST:
2025 case LTTNG_CONSUMER64_UST:
2026 /*
2027 * Close all metadata streams. The metadata hash table is passed and
2028 * this call iterates over it by closing all wakeup fd. This is safe
2029 * because at this point we are sure that the metadata producer is
2030 * either dead or blocked.
2031 */
6d574024 2032 lttng_ustconsumer_close_all_metadata(metadata_ht);
d88aee68
DG
2033 break;
2034 default:
2035 ERR("Unknown consumer_data type");
2036 assert(0);
2037 }
2038}
2039
fb3a43a9
DG
2040/*
2041 * Clean up a metadata stream and free its memory.
2042 */
e316aad5
DG
2043void consumer_del_metadata_stream(struct lttng_consumer_stream *stream,
2044 struct lttng_ht *ht)
fb3a43a9 2045{
e316aad5 2046 struct lttng_consumer_channel *free_chan = NULL;
fb3a43a9
DG
2047
2048 assert(stream);
2049 /*
2050 * This call should NEVER receive regular stream. It must always be
2051 * metadata stream and this is crucial for data structure synchronization.
2052 */
2053 assert(stream->metadata_flag);
2054
e316aad5
DG
2055 DBG3("Consumer delete metadata stream %d", stream->wait_fd);
2056
74251bb8 2057 pthread_mutex_lock(&consumer_data.lock);
a9838785 2058 pthread_mutex_lock(&stream->chan->lock);
8994307f
DG
2059 pthread_mutex_lock(&stream->lock);
2060
6d574024
DG
2061 /* Remove any reference to that stream. */
2062 consumer_stream_delete(stream, ht);
ca22feea 2063
6d574024
DG
2064 /* Close down everything including the relayd if one. */
2065 consumer_stream_close(stream);
2066 /* Destroy tracer buffers of the stream. */
2067 consumer_stream_destroy_buffers(stream);
fb3a43a9
DG
2068
2069 /* Atomically decrement channel refcount since other threads can use it. */
f2ad556d 2070 if (!uatomic_sub_return(&stream->chan->refcount, 1)
ffe60014 2071 && !uatomic_read(&stream->chan->nb_init_stream_left)) {
c30aaa51 2072 /* Go for channel deletion! */
e316aad5 2073 free_chan = stream->chan;
fb3a43a9
DG
2074 }
2075
73811ecc
DG
2076 /*
2077 * Nullify the stream reference so it is not used after deletion. The
6d574024
DG
2078 * channel lock MUST be acquired before being able to check for a NULL
2079 * pointer value.
73811ecc
DG
2080 */
2081 stream->chan->metadata_stream = NULL;
2082
8994307f 2083 pthread_mutex_unlock(&stream->lock);
a9838785 2084 pthread_mutex_unlock(&stream->chan->lock);
74251bb8 2085 pthread_mutex_unlock(&consumer_data.lock);
e316aad5
DG
2086
2087 if (free_chan) {
2088 consumer_del_channel(free_chan);
2089 }
2090
6d574024 2091 consumer_stream_free(stream);
fb3a43a9
DG
2092}
2093
2094/*
2095 * Action done with the metadata stream when adding it to the consumer internal
2096 * data structures to handle it.
2097 */
5ab66908 2098int consumer_add_metadata_stream(struct lttng_consumer_stream *stream)
fb3a43a9 2099{
5ab66908 2100 struct lttng_ht *ht = metadata_ht;
e316aad5 2101 int ret = 0;
76082088 2102 struct lttng_ht_iter iter;
d88aee68 2103 struct lttng_ht_node_u64 *node;
fb3a43a9 2104
e316aad5
DG
2105 assert(stream);
2106 assert(ht);
2107
d88aee68 2108 DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key);
e316aad5
DG
2109
2110 pthread_mutex_lock(&consumer_data.lock);
a9838785 2111 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 2112 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 2113 pthread_mutex_lock(&stream->lock);
e316aad5 2114
e316aad5
DG
2115 /*
2116 * From here, refcounts are updated so be _careful_ when returning an error
2117 * after this point.
2118 */
2119
fb3a43a9 2120 rcu_read_lock();
76082088
DG
2121
2122 /*
2123 * Lookup the stream just to make sure it does not exist in our internal
2124 * state. This should NEVER happen.
2125 */
d88aee68
DG
2126 lttng_ht_lookup(ht, &stream->key, &iter);
2127 node = lttng_ht_iter_get_node_u64(&iter);
76082088
DG
2128 assert(!node);
2129
e316aad5 2130 /*
ffe60014
DG
2131 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2132 * in terms of destroying the associated channel, because the action that
e316aad5
DG
2133 * causes the count to become 0 also causes a stream to be added. The
2134 * channel deletion will thus be triggered by the following removal of this
2135 * stream.
2136 */
ffe60014 2137 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
2138 /* Increment refcount before decrementing nb_init_stream_left */
2139 cmm_smp_wmb();
ffe60014 2140 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
2141 }
2142
d88aee68 2143 lttng_ht_add_unique_u64(ht, &stream->node);
ca22feea 2144
d8ef542d
MD
2145 lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht,
2146 &stream->node_channel_id);
2147
ca22feea
DG
2148 /*
2149 * Add stream to the stream_list_ht of the consumer data. No need to steal
2150 * the key since the HT does not use it and we allow to add redundant keys
2151 * into this table.
2152 */
d88aee68 2153 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 2154
fb3a43a9 2155 rcu_read_unlock();
e316aad5 2156
2e818a6a 2157 pthread_mutex_unlock(&stream->lock);
a9838785 2158 pthread_mutex_unlock(&stream->chan->lock);
ec6ea7d0 2159 pthread_mutex_unlock(&stream->chan->timer_lock);
e316aad5
DG
2160 pthread_mutex_unlock(&consumer_data.lock);
2161 return ret;
fb3a43a9
DG
2162}
2163
8994307f
DG
2164/*
2165 * Delete data stream that are flagged for deletion (endpoint_status).
2166 */
2167static void validate_endpoint_status_data_stream(void)
2168{
2169 struct lttng_ht_iter iter;
2170 struct lttng_consumer_stream *stream;
2171
2172 DBG("Consumer delete flagged data stream");
2173
2174 rcu_read_lock();
2175 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
2176 /* Validate delete flag of the stream */
79d4ffb7 2177 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2178 continue;
2179 }
2180 /* Delete it right now */
2181 consumer_del_stream(stream, data_ht);
2182 }
2183 rcu_read_unlock();
2184}
2185
2186/*
2187 * Delete metadata stream that are flagged for deletion (endpoint_status).
2188 */
2189static void validate_endpoint_status_metadata_stream(
2190 struct lttng_poll_event *pollset)
2191{
2192 struct lttng_ht_iter iter;
2193 struct lttng_consumer_stream *stream;
2194
2195 DBG("Consumer delete flagged metadata stream");
2196
2197 assert(pollset);
2198
2199 rcu_read_lock();
2200 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
2201 /* Validate delete flag of the stream */
79d4ffb7 2202 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2203 continue;
2204 }
2205 /*
2206 * Remove from pollset so the metadata thread can continue without
2207 * blocking on a deleted stream.
2208 */
2209 lttng_poll_del(pollset, stream->wait_fd);
2210
2211 /* Delete it right now */
2212 consumer_del_metadata_stream(stream, metadata_ht);
2213 }
2214 rcu_read_unlock();
2215}
2216
fb3a43a9
DG
2217/*
2218 * Thread polls on metadata file descriptor and write them on disk or on the
2219 * network.
2220 */
7d980def 2221void *consumer_thread_metadata_poll(void *data)
fb3a43a9 2222{
1fc79fb4 2223 int ret, i, pollfd, err = -1;
fb3a43a9 2224 uint32_t revents, nb_fd;
e316aad5 2225 struct lttng_consumer_stream *stream = NULL;
fb3a43a9 2226 struct lttng_ht_iter iter;
d88aee68 2227 struct lttng_ht_node_u64 *node;
fb3a43a9
DG
2228 struct lttng_poll_event events;
2229 struct lttng_consumer_local_data *ctx = data;
2230 ssize_t len;
2231
2232 rcu_register_thread();
2233
1fc79fb4
MD
2234 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA);
2235
2d57de81
MD
2236 if (testpoint(consumerd_thread_metadata)) {
2237 goto error_testpoint;
2238 }
2239
9ce5646a
MD
2240 health_code_update();
2241
fb3a43a9
DG
2242 DBG("Thread metadata poll started");
2243
fb3a43a9
DG
2244 /* Size is set to 1 for the consumer_metadata pipe */
2245 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2246 if (ret < 0) {
2247 ERR("Poll set creation failed");
d8ef542d 2248 goto end_poll;
fb3a43a9
DG
2249 }
2250
13886d2d
DG
2251 ret = lttng_poll_add(&events,
2252 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN);
fb3a43a9
DG
2253 if (ret < 0) {
2254 goto end;
2255 }
2256
2257 /* Main loop */
2258 DBG("Metadata main loop started");
2259
2260 while (1) {
fb3a43a9 2261restart:
7fa2082e 2262 health_code_update();
9ce5646a 2263 health_poll_entry();
7fa2082e 2264 DBG("Metadata poll wait");
fb3a43a9 2265 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2266 DBG("Metadata poll return from wait with %d fd(s)",
2267 LTTNG_POLL_GETNB(&events));
9ce5646a 2268 health_poll_exit();
40063ead 2269 DBG("Metadata event caught in thread");
fb3a43a9
DG
2270 if (ret < 0) {
2271 if (errno == EINTR) {
40063ead 2272 ERR("Poll EINTR caught");
fb3a43a9
DG
2273 goto restart;
2274 }
d9607cd7
MD
2275 if (LTTNG_POLL_GETNB(&events) == 0) {
2276 err = 0; /* All is OK */
2277 }
2278 goto end;
fb3a43a9
DG
2279 }
2280
0d9c5d77
DG
2281 nb_fd = ret;
2282
e316aad5 2283 /* From here, the event is a metadata wait fd */
fb3a43a9 2284 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2285 health_code_update();
2286
fb3a43a9
DG
2287 revents = LTTNG_POLL_GETEV(&events, i);
2288 pollfd = LTTNG_POLL_GETFD(&events, i);
2289
fd20dac9
MD
2290 if (!revents) {
2291 /* No activity for this FD (poll implementation). */
2292 continue;
2293 }
2294
13886d2d 2295 if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) {
03e43155 2296 if (revents & LPOLLIN) {
13886d2d
DG
2297 ssize_t pipe_len;
2298
2299 pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe,
2300 &stream, sizeof(stream));
6cd525e8 2301 if (pipe_len < sizeof(stream)) {
03e43155
MD
2302 if (pipe_len < 0) {
2303 PERROR("read metadata stream");
2304 }
fb3a43a9 2305 /*
03e43155
MD
2306 * Remove the pipe from the poll set and continue the loop
2307 * since their might be data to consume.
fb3a43a9 2308 */
03e43155
MD
2309 lttng_poll_del(&events,
2310 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2311 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
fb3a43a9
DG
2312 continue;
2313 }
2314
8994307f
DG
2315 /* A NULL stream means that the state has changed. */
2316 if (stream == NULL) {
2317 /* Check for deleted streams. */
2318 validate_endpoint_status_metadata_stream(&events);
3714380f 2319 goto restart;
8994307f
DG
2320 }
2321
fb3a43a9
DG
2322 DBG("Adding metadata stream %d to poll set",
2323 stream->wait_fd);
2324
fb3a43a9
DG
2325 /* Add metadata stream to the global poll events list */
2326 lttng_poll_add(&events, stream->wait_fd,
6d574024 2327 LPOLLIN | LPOLLPRI | LPOLLHUP);
03e43155
MD
2328 } else if (revents & (LPOLLERR | LPOLLHUP)) {
2329 DBG("Metadata thread pipe hung up");
2330 /*
2331 * Remove the pipe from the poll set and continue the loop
2332 * since their might be data to consume.
2333 */
2334 lttng_poll_del(&events,
2335 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2336 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
2337 continue;
2338 } else {
2339 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2340 goto end;
fb3a43a9
DG
2341 }
2342
e316aad5 2343 /* Handle other stream */
fb3a43a9
DG
2344 continue;
2345 }
2346
d09e1200 2347 rcu_read_lock();
d88aee68
DG
2348 {
2349 uint64_t tmp_id = (uint64_t) pollfd;
2350
2351 lttng_ht_lookup(metadata_ht, &tmp_id, &iter);
2352 }
2353 node = lttng_ht_iter_get_node_u64(&iter);
e316aad5 2354 assert(node);
fb3a43a9
DG
2355
2356 stream = caa_container_of(node, struct lttng_consumer_stream,
58b1f425 2357 node);
fb3a43a9 2358
03e43155
MD
2359 if (revents & (LPOLLIN | LPOLLPRI)) {
2360 /* Get the data out of the metadata file descriptor */
2361 DBG("Metadata available on fd %d", pollfd);
2362 assert(stream->wait_fd == pollfd);
2363
2364 do {
2365 health_code_update();
2366
2367 len = ctx->on_buffer_ready(stream, ctx);
2368 /*
2369 * We don't check the return value here since if we get
2370 * a negative len, it means an error occured thus we
2371 * simply remove it from the poll set and free the
2372 * stream.
2373 */
2374 } while (len > 0);
2375
2376 /* It's ok to have an unavailable sub-buffer */
2377 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
2378 /* Clean up stream from consumer and free it. */
2379 lttng_poll_del(&events, stream->wait_fd);
2380 consumer_del_metadata_stream(stream, metadata_ht);
2381 }
2382 } else if (revents & (LPOLLERR | LPOLLHUP)) {
e316aad5 2383 DBG("Metadata fd %d is hup|err.", pollfd);
fb3a43a9
DG
2384 if (!stream->hangup_flush_done
2385 && (consumer_data.type == LTTNG_CONSUMER32_UST
2386 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2387 DBG("Attempting to flush and consume the UST buffers");
2388 lttng_ustconsumer_on_stream_hangup(stream);
2389
2390 /* We just flushed the stream now read it. */
4bb94b75 2391 do {
9ce5646a
MD
2392 health_code_update();
2393
4bb94b75
DG
2394 len = ctx->on_buffer_ready(stream, ctx);
2395 /*
2396 * We don't check the return value here since if we get
2397 * a negative len, it means an error occured thus we
2398 * simply remove it from the poll set and free the
2399 * stream.
2400 */
2401 } while (len > 0);
fb3a43a9
DG
2402 }
2403
fb3a43a9 2404 lttng_poll_del(&events, stream->wait_fd);
e316aad5
DG
2405 /*
2406 * This call update the channel states, closes file descriptors
2407 * and securely free the stream.
2408 */
2409 consumer_del_metadata_stream(stream, metadata_ht);
03e43155
MD
2410 } else {
2411 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
6f2f1a70 2412 rcu_read_unlock();
03e43155 2413 goto end;
fb3a43a9 2414 }
e316aad5 2415 /* Release RCU lock for the stream looked up */
d09e1200 2416 rcu_read_unlock();
fb3a43a9
DG
2417 }
2418 }
2419
1fc79fb4
MD
2420 /* All is OK */
2421 err = 0;
fb3a43a9
DG
2422end:
2423 DBG("Metadata poll thread exiting");
fb3a43a9 2424
d8ef542d
MD
2425 lttng_poll_clean(&events);
2426end_poll:
2d57de81 2427error_testpoint:
1fc79fb4
MD
2428 if (err) {
2429 health_error();
2430 ERR("Health error occurred in %s", __func__);
2431 }
2432 health_unregister(health_consumerd);
fb3a43a9
DG
2433 rcu_unregister_thread();
2434 return NULL;
2435}
2436
3bd1e081 2437/*
e4421fec 2438 * This thread polls the fds in the set to consume the data and write
3bd1e081
MD
2439 * it to tracefile if necessary.
2440 */
7d980def 2441void *consumer_thread_data_poll(void *data)
3bd1e081 2442{
1fc79fb4 2443 int num_rdy, num_hup, high_prio, ret, i, err = -1;
3bd1e081
MD
2444 struct pollfd *pollfd = NULL;
2445 /* local view of the streams */
c869f647 2446 struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL;
3bd1e081
MD
2447 /* local view of consumer_data.fds_count */
2448 int nb_fd = 0;
3bd1e081 2449 struct lttng_consumer_local_data *ctx = data;
00e2e675 2450 ssize_t len;
3bd1e081 2451
e7b994a3
DG
2452 rcu_register_thread();
2453
1fc79fb4
MD
2454 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA);
2455
2d57de81
MD
2456 if (testpoint(consumerd_thread_data)) {
2457 goto error_testpoint;
2458 }
2459
9ce5646a
MD
2460 health_code_update();
2461
4df6c8cb
MD
2462 local_stream = zmalloc(sizeof(struct lttng_consumer_stream *));
2463 if (local_stream == NULL) {
2464 PERROR("local_stream malloc");
2465 goto end;
2466 }
3bd1e081
MD
2467
2468 while (1) {
9ce5646a
MD
2469 health_code_update();
2470
3bd1e081
MD
2471 high_prio = 0;
2472 num_hup = 0;
2473
2474 /*
e4421fec 2475 * the fds set has been updated, we need to update our
3bd1e081
MD
2476 * local array as well
2477 */
2478 pthread_mutex_lock(&consumer_data.lock);
2479 if (consumer_data.need_update) {
0e428499
DG
2480 free(pollfd);
2481 pollfd = NULL;
2482
2483 free(local_stream);
2484 local_stream = NULL;
3bd1e081 2485
02b3d176
DG
2486 /*
2487 * Allocate for all fds +1 for the consumer_data_pipe and +1 for
2488 * wake up pipe.
2489 */
2490 pollfd = zmalloc((consumer_data.stream_count + 2) * sizeof(struct pollfd));
3bd1e081 2491 if (pollfd == NULL) {
7a57cf92 2492 PERROR("pollfd malloc");
3bd1e081
MD
2493 pthread_mutex_unlock(&consumer_data.lock);
2494 goto end;
2495 }
2496
02b3d176 2497 local_stream = zmalloc((consumer_data.stream_count + 2) *
747f8642 2498 sizeof(struct lttng_consumer_stream *));
3bd1e081 2499 if (local_stream == NULL) {
7a57cf92 2500 PERROR("local_stream malloc");
3bd1e081
MD
2501 pthread_mutex_unlock(&consumer_data.lock);
2502 goto end;
2503 }
ffe60014 2504 ret = update_poll_array(ctx, &pollfd, local_stream,
43c34bc3 2505 data_ht);
3bd1e081
MD
2506 if (ret < 0) {
2507 ERR("Error in allocating pollfd or local_outfds");
f73fabfd 2508 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2509 pthread_mutex_unlock(&consumer_data.lock);
2510 goto end;
2511 }
2512 nb_fd = ret;
2513 consumer_data.need_update = 0;
2514 }
2515 pthread_mutex_unlock(&consumer_data.lock);
2516
4078b776
MD
2517 /* No FDs and consumer_quit, consumer_cleanup the thread */
2518 if (nb_fd == 0 && consumer_quit == 1) {
1fc79fb4 2519 err = 0; /* All is OK */
4078b776
MD
2520 goto end;
2521 }
3bd1e081 2522 /* poll on the array of fds */
88f2b785 2523 restart:
02b3d176 2524 DBG("polling on %d fd", nb_fd + 2);
9ce5646a 2525 health_poll_entry();
02b3d176 2526 num_rdy = poll(pollfd, nb_fd + 2, -1);
9ce5646a 2527 health_poll_exit();
3bd1e081
MD
2528 DBG("poll num_rdy : %d", num_rdy);
2529 if (num_rdy == -1) {
88f2b785
MD
2530 /*
2531 * Restart interrupted system call.
2532 */
2533 if (errno == EINTR) {
2534 goto restart;
2535 }
7a57cf92 2536 PERROR("Poll error");
f73fabfd 2537 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2538 goto end;
2539 } else if (num_rdy == 0) {
2540 DBG("Polling thread timed out");
2541 goto end;
2542 }
2543
3bd1e081 2544 /*
50f8ae69 2545 * If the consumer_data_pipe triggered poll go directly to the
00e2e675
DG
2546 * beginning of the loop to update the array. We want to prioritize
2547 * array update over low-priority reads.
3bd1e081 2548 */
509bb1cf 2549 if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) {
ab30f567 2550 ssize_t pipe_readlen;
04fdd819 2551
50f8ae69 2552 DBG("consumer_data_pipe wake up");
acdb9057
DG
2553 pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe,
2554 &new_stream, sizeof(new_stream));
6cd525e8
MD
2555 if (pipe_readlen < sizeof(new_stream)) {
2556 PERROR("Consumer data pipe");
23f5f35d
DG
2557 /* Continue so we can at least handle the current stream(s). */
2558 continue;
2559 }
c869f647
DG
2560
2561 /*
2562 * If the stream is NULL, just ignore it. It's also possible that
2563 * the sessiond poll thread changed the consumer_quit state and is
2564 * waking us up to test it.
2565 */
2566 if (new_stream == NULL) {
8994307f 2567 validate_endpoint_status_data_stream();
c869f647
DG
2568 continue;
2569 }
2570
c869f647 2571 /* Continue to update the local streams and handle prio ones */
3bd1e081
MD
2572 continue;
2573 }
2574
02b3d176
DG
2575 /* Handle wakeup pipe. */
2576 if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) {
2577 char dummy;
2578 ssize_t pipe_readlen;
2579
2580 pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy,
2581 sizeof(dummy));
2582 if (pipe_readlen < 0) {
2583 PERROR("Consumer data wakeup pipe");
2584 }
2585 /* We've been awakened to handle stream(s). */
2586 ctx->has_wakeup = 0;
2587 }
2588
3bd1e081
MD
2589 /* Take care of high priority channels first. */
2590 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2591 health_code_update();
2592
9617607b
DG
2593 if (local_stream[i] == NULL) {
2594 continue;
2595 }
fb3a43a9 2596 if (pollfd[i].revents & POLLPRI) {
d41f73b7
MD
2597 DBG("Urgent read on fd %d", pollfd[i].fd);
2598 high_prio = 1;
4078b776 2599 len = ctx->on_buffer_ready(local_stream[i], ctx);
d41f73b7 2600 /* it's ok to have an unavailable sub-buffer */
b64403e3 2601 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2602 /* Clean the stream and free it. */
2603 consumer_del_stream(local_stream[i], data_ht);
9617607b 2604 local_stream[i] = NULL;
4078b776
MD
2605 } else if (len > 0) {
2606 local_stream[i]->data_read = 1;
d41f73b7 2607 }
3bd1e081
MD
2608 }
2609 }
2610
4078b776
MD
2611 /*
2612 * If we read high prio channel in this loop, try again
2613 * for more high prio data.
2614 */
2615 if (high_prio) {
3bd1e081
MD
2616 continue;
2617 }
2618
2619 /* Take care of low priority channels. */
4078b776 2620 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2621 health_code_update();
2622
9617607b
DG
2623 if (local_stream[i] == NULL) {
2624 continue;
2625 }
4078b776 2626 if ((pollfd[i].revents & POLLIN) ||
02b3d176
DG
2627 local_stream[i]->hangup_flush_done ||
2628 local_stream[i]->has_data) {
4078b776
MD
2629 DBG("Normal read on fd %d", pollfd[i].fd);
2630 len = ctx->on_buffer_ready(local_stream[i], ctx);
2631 /* it's ok to have an unavailable sub-buffer */
b64403e3 2632 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2633 /* Clean the stream and free it. */
2634 consumer_del_stream(local_stream[i], data_ht);
9617607b 2635 local_stream[i] = NULL;
4078b776
MD
2636 } else if (len > 0) {
2637 local_stream[i]->data_read = 1;
2638 }
2639 }
2640 }
2641
2642 /* Handle hangup and errors */
2643 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2644 health_code_update();
2645
9617607b
DG
2646 if (local_stream[i] == NULL) {
2647 continue;
2648 }
4078b776
MD
2649 if (!local_stream[i]->hangup_flush_done
2650 && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL))
2651 && (consumer_data.type == LTTNG_CONSUMER32_UST
2652 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2653 DBG("fd %d is hup|err|nval. Attempting flush and read.",
9617607b 2654 pollfd[i].fd);
4078b776
MD
2655 lttng_ustconsumer_on_stream_hangup(local_stream[i]);
2656 /* Attempt read again, for the data we just flushed. */
2657 local_stream[i]->data_read = 1;
2658 }
2659 /*
2660 * If the poll flag is HUP/ERR/NVAL and we have
2661 * read no data in this pass, we can remove the
2662 * stream from its hash table.
2663 */
2664 if ((pollfd[i].revents & POLLHUP)) {
2665 DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
2666 if (!local_stream[i]->data_read) {
43c34bc3 2667 consumer_del_stream(local_stream[i], data_ht);
9617607b 2668 local_stream[i] = NULL;
4078b776
MD
2669 num_hup++;
2670 }
2671 } else if (pollfd[i].revents & POLLERR) {
2672 ERR("Error returned in polling fd %d.", pollfd[i].fd);
2673 if (!local_stream[i]->data_read) {
43c34bc3 2674 consumer_del_stream(local_stream[i], data_ht);
9617607b 2675 local_stream[i] = NULL;
4078b776
MD
2676 num_hup++;
2677 }
2678 } else if (pollfd[i].revents & POLLNVAL) {
2679 ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
2680 if (!local_stream[i]->data_read) {
43c34bc3 2681 consumer_del_stream(local_stream[i], data_ht);
9617607b 2682 local_stream[i] = NULL;
4078b776 2683 num_hup++;
3bd1e081
MD
2684 }
2685 }
9617607b
DG
2686 if (local_stream[i] != NULL) {
2687 local_stream[i]->data_read = 0;
2688 }
3bd1e081
MD
2689 }
2690 }
1fc79fb4
MD
2691 /* All is OK */
2692 err = 0;
3bd1e081
MD
2693end:
2694 DBG("polling thread exiting");
0e428499
DG
2695 free(pollfd);
2696 free(local_stream);
fb3a43a9
DG
2697
2698 /*
2699 * Close the write side of the pipe so epoll_wait() in
7d980def
DG
2700 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2701 * read side of the pipe. If we close them both, epoll_wait strangely does
2702 * not return and could create a endless wait period if the pipe is the
2703 * only tracked fd in the poll set. The thread will take care of closing
2704 * the read side.
fb3a43a9 2705 */
13886d2d 2706 (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe);
fb3a43a9 2707
2d57de81 2708error_testpoint:
1fc79fb4
MD
2709 if (err) {
2710 health_error();
2711 ERR("Health error occurred in %s", __func__);
2712 }
2713 health_unregister(health_consumerd);
2714
e7b994a3 2715 rcu_unregister_thread();
3bd1e081
MD
2716 return NULL;
2717}
2718
d8ef542d
MD
2719/*
2720 * Close wake-up end of each stream belonging to the channel. This will
2721 * allow the poll() on the stream read-side to detect when the
2722 * write-side (application) finally closes them.
2723 */
2724static
2725void consumer_close_channel_streams(struct lttng_consumer_channel *channel)
2726{
2727 struct lttng_ht *ht;
2728 struct lttng_consumer_stream *stream;
2729 struct lttng_ht_iter iter;
2730
2731 ht = consumer_data.stream_per_chan_id_ht;
2732
2733 rcu_read_lock();
2734 cds_lfht_for_each_entry_duplicate(ht->ht,
2735 ht->hash_fct(&channel->key, lttng_ht_seed),
2736 ht->match_fct, &channel->key,
2737 &iter.iter, stream, node_channel_id.node) {
f2ad556d
MD
2738 /*
2739 * Protect against teardown with mutex.
2740 */
2741 pthread_mutex_lock(&stream->lock);
2742 if (cds_lfht_is_node_deleted(&stream->node.node)) {
2743 goto next;
2744 }
d8ef542d
MD
2745 switch (consumer_data.type) {
2746 case LTTNG_CONSUMER_KERNEL:
2747 break;
2748 case LTTNG_CONSUMER32_UST:
2749 case LTTNG_CONSUMER64_UST:
b4a650f3
DG
2750 if (stream->metadata_flag) {
2751 /* Safe and protected by the stream lock. */
2752 lttng_ustconsumer_close_metadata(stream->chan);
2753 } else {
2754 /*
2755 * Note: a mutex is taken internally within
2756 * liblttng-ust-ctl to protect timer wakeup_fd
2757 * use from concurrent close.
2758 */
2759 lttng_ustconsumer_close_stream_wakeup(stream);
2760 }
d8ef542d
MD
2761 break;
2762 default:
2763 ERR("Unknown consumer_data type");
2764 assert(0);
2765 }
f2ad556d
MD
2766 next:
2767 pthread_mutex_unlock(&stream->lock);
d8ef542d
MD
2768 }
2769 rcu_read_unlock();
2770}
2771
2772static void destroy_channel_ht(struct lttng_ht *ht)
2773{
2774 struct lttng_ht_iter iter;
2775 struct lttng_consumer_channel *channel;
2776 int ret;
2777
2778 if (ht == NULL) {
2779 return;
2780 }
2781
2782 rcu_read_lock();
2783 cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) {
2784 ret = lttng_ht_del(ht, &iter);
2785 assert(ret != 0);
2786 }
2787 rcu_read_unlock();
2788
2789 lttng_ht_destroy(ht);
2790}
2791
2792/*
2793 * This thread polls the channel fds to detect when they are being
2794 * closed. It closes all related streams if the channel is detected as
2795 * closed. It is currently only used as a shim layer for UST because the
2796 * consumerd needs to keep the per-stream wakeup end of pipes open for
2797 * periodical flush.
2798 */
2799void *consumer_thread_channel_poll(void *data)
2800{
1fc79fb4 2801 int ret, i, pollfd, err = -1;
d8ef542d
MD
2802 uint32_t revents, nb_fd;
2803 struct lttng_consumer_channel *chan = NULL;
2804 struct lttng_ht_iter iter;
2805 struct lttng_ht_node_u64 *node;
2806 struct lttng_poll_event events;
2807 struct lttng_consumer_local_data *ctx = data;
2808 struct lttng_ht *channel_ht;
2809
2810 rcu_register_thread();
2811
1fc79fb4
MD
2812 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL);
2813
2d57de81
MD
2814 if (testpoint(consumerd_thread_channel)) {
2815 goto error_testpoint;
2816 }
2817
9ce5646a
MD
2818 health_code_update();
2819
d8ef542d
MD
2820 channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2821 if (!channel_ht) {
2822 /* ENOMEM at this point. Better to bail out. */
2823 goto end_ht;
2824 }
2825
2826 DBG("Thread channel poll started");
2827
2828 /* Size is set to 1 for the consumer_channel pipe */
2829 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2830 if (ret < 0) {
2831 ERR("Poll set creation failed");
2832 goto end_poll;
2833 }
2834
2835 ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN);
2836 if (ret < 0) {
2837 goto end;
2838 }
2839
2840 /* Main loop */
2841 DBG("Channel main loop started");
2842
2843 while (1) {
d8ef542d 2844restart:
7fa2082e
MD
2845 health_code_update();
2846 DBG("Channel poll wait");
9ce5646a 2847 health_poll_entry();
d8ef542d 2848 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2849 DBG("Channel poll return from wait with %d fd(s)",
2850 LTTNG_POLL_GETNB(&events));
9ce5646a 2851 health_poll_exit();
40063ead 2852 DBG("Channel event caught in thread");
d8ef542d
MD
2853 if (ret < 0) {
2854 if (errno == EINTR) {
40063ead 2855 ERR("Poll EINTR caught");
d8ef542d
MD
2856 goto restart;
2857 }
d9607cd7
MD
2858 if (LTTNG_POLL_GETNB(&events) == 0) {
2859 err = 0; /* All is OK */
2860 }
d8ef542d
MD
2861 goto end;
2862 }
2863
2864 nb_fd = ret;
2865
2866 /* From here, the event is a channel wait fd */
2867 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2868 health_code_update();
2869
d8ef542d
MD
2870 revents = LTTNG_POLL_GETEV(&events, i);
2871 pollfd = LTTNG_POLL_GETFD(&events, i);
2872
d8ef542d 2873 if (!revents) {
fd20dac9 2874 /* No activity for this FD (poll implementation). */
d8ef542d
MD
2875 continue;
2876 }
fd20dac9 2877
d8ef542d 2878 if (pollfd == ctx->consumer_channel_pipe[0]) {
03e43155 2879 if (revents & LPOLLIN) {
d8ef542d 2880 enum consumer_channel_action action;
a0cbdd2e 2881 uint64_t key;
d8ef542d 2882
a0cbdd2e 2883 ret = read_channel_pipe(ctx, &chan, &key, &action);
d8ef542d 2884 if (ret <= 0) {
03e43155
MD
2885 if (ret < 0) {
2886 ERR("Error reading channel pipe");
2887 }
2888 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
d8ef542d
MD
2889 continue;
2890 }
2891
2892 switch (action) {
2893 case CONSUMER_CHANNEL_ADD:
2894 DBG("Adding channel %d to poll set",
2895 chan->wait_fd);
2896
2897 lttng_ht_node_init_u64(&chan->wait_fd_node,
2898 chan->wait_fd);
c7260a81 2899 rcu_read_lock();
d8ef542d
MD
2900 lttng_ht_add_unique_u64(channel_ht,
2901 &chan->wait_fd_node);
c7260a81 2902 rcu_read_unlock();
d8ef542d
MD
2903 /* Add channel to the global poll events list */
2904 lttng_poll_add(&events, chan->wait_fd,
03e43155 2905 LPOLLERR | LPOLLHUP);
d8ef542d 2906 break;
a0cbdd2e
MD
2907 case CONSUMER_CHANNEL_DEL:
2908 {
b4a650f3
DG
2909 /*
2910 * This command should never be called if the channel
2911 * has streams monitored by either the data or metadata
2912 * thread. The consumer only notify this thread with a
2913 * channel del. command if it receives a destroy
2914 * channel command from the session daemon that send it
2915 * if a command prior to the GET_CHANNEL failed.
2916 */
2917
c7260a81 2918 rcu_read_lock();
a0cbdd2e
MD
2919 chan = consumer_find_channel(key);
2920 if (!chan) {
c7260a81 2921 rcu_read_unlock();
a0cbdd2e
MD
2922 ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key);
2923 break;
2924 }
2925 lttng_poll_del(&events, chan->wait_fd);
f623cc0b 2926 iter.iter.node = &chan->wait_fd_node.node;
a0cbdd2e
MD
2927 ret = lttng_ht_del(channel_ht, &iter);
2928 assert(ret == 0);
a0cbdd2e 2929
f2a444f1
DG
2930 switch (consumer_data.type) {
2931 case LTTNG_CONSUMER_KERNEL:
2932 break;
2933 case LTTNG_CONSUMER32_UST:
2934 case LTTNG_CONSUMER64_UST:
212d67a2
DG
2935 health_code_update();
2936 /* Destroy streams that might have been left in the stream list. */
2937 clean_channel_stream_list(chan);
f2a444f1
DG
2938 break;
2939 default:
2940 ERR("Unknown consumer_data type");
2941 assert(0);
2942 }
2943
a0cbdd2e
MD
2944 /*
2945 * Release our own refcount. Force channel deletion even if
2946 * streams were not initialized.
2947 */
2948 if (!uatomic_sub_return(&chan->refcount, 1)) {
2949 consumer_del_channel(chan);
2950 }
c7260a81 2951 rcu_read_unlock();
a0cbdd2e
MD
2952 goto restart;
2953 }
d8ef542d
MD
2954 case CONSUMER_CHANNEL_QUIT:
2955 /*
2956 * Remove the pipe from the poll set and continue the loop
2957 * since their might be data to consume.
2958 */
2959 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2960 continue;
2961 default:
2962 ERR("Unknown action");
2963 break;
2964 }
03e43155
MD
2965 } else if (revents & (LPOLLERR | LPOLLHUP)) {
2966 DBG("Channel thread pipe hung up");
2967 /*
2968 * Remove the pipe from the poll set and continue the loop
2969 * since their might be data to consume.
2970 */
2971 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2972 continue;
2973 } else {
2974 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2975 goto end;
d8ef542d
MD
2976 }
2977
2978 /* Handle other stream */
2979 continue;
2980 }
2981
2982 rcu_read_lock();
2983 {
2984 uint64_t tmp_id = (uint64_t) pollfd;
2985
2986 lttng_ht_lookup(channel_ht, &tmp_id, &iter);
2987 }
2988 node = lttng_ht_iter_get_node_u64(&iter);
2989 assert(node);
2990
2991 chan = caa_container_of(node, struct lttng_consumer_channel,
2992 wait_fd_node);
2993
2994 /* Check for error event */
2995 if (revents & (LPOLLERR | LPOLLHUP)) {
2996 DBG("Channel fd %d is hup|err.", pollfd);
2997
2998 lttng_poll_del(&events, chan->wait_fd);
2999 ret = lttng_ht_del(channel_ht, &iter);
3000 assert(ret == 0);
b4a650f3
DG
3001
3002 /*
3003 * This will close the wait fd for each stream associated to
3004 * this channel AND monitored by the data/metadata thread thus
3005 * will be clean by the right thread.
3006 */
d8ef542d 3007 consumer_close_channel_streams(chan);
f2ad556d
MD
3008
3009 /* Release our own refcount */
3010 if (!uatomic_sub_return(&chan->refcount, 1)
3011 && !uatomic_read(&chan->nb_init_stream_left)) {
3012 consumer_del_channel(chan);
3013 }
03e43155
MD
3014 } else {
3015 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
3016 rcu_read_unlock();
3017 goto end;
d8ef542d
MD
3018 }
3019
3020 /* Release RCU lock for the channel looked up */
3021 rcu_read_unlock();
3022 }
3023 }
3024
1fc79fb4
MD
3025 /* All is OK */
3026 err = 0;
d8ef542d
MD
3027end:
3028 lttng_poll_clean(&events);
3029end_poll:
3030 destroy_channel_ht(channel_ht);
3031end_ht:
2d57de81 3032error_testpoint:
d8ef542d 3033 DBG("Channel poll thread exiting");
1fc79fb4
MD
3034 if (err) {
3035 health_error();
3036 ERR("Health error occurred in %s", __func__);
3037 }
3038 health_unregister(health_consumerd);
d8ef542d
MD
3039 rcu_unregister_thread();
3040 return NULL;
3041}
3042
331744e3
JD
3043static int set_metadata_socket(struct lttng_consumer_local_data *ctx,
3044 struct pollfd *sockpoll, int client_socket)
3045{
3046 int ret;
3047
3048 assert(ctx);
3049 assert(sockpoll);
3050
84382d49
MD
3051 ret = lttng_consumer_poll_socket(sockpoll);
3052 if (ret) {
331744e3
JD
3053 goto error;
3054 }
3055 DBG("Metadata connection on client_socket");
3056
3057 /* Blocking call, waiting for transmission */
3058 ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket);
3059 if (ctx->consumer_metadata_socket < 0) {
3060 WARN("On accept metadata");
3061 ret = -1;
3062 goto error;
3063 }
3064 ret = 0;
3065
3066error:
3067 return ret;
3068}
3069
3bd1e081
MD
3070/*
3071 * This thread listens on the consumerd socket and receives the file
3072 * descriptors from the session daemon.
3073 */
7d980def 3074void *consumer_thread_sessiond_poll(void *data)
3bd1e081 3075{
1fc79fb4 3076 int sock = -1, client_socket, ret, err = -1;
3bd1e081
MD
3077 /*
3078 * structure to poll for incoming data on communication socket avoids
3079 * making blocking sockets.
3080 */
3081 struct pollfd consumer_sockpoll[2];
3082 struct lttng_consumer_local_data *ctx = data;
3083
e7b994a3
DG
3084 rcu_register_thread();
3085
1fc79fb4
MD
3086 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND);
3087
2d57de81
MD
3088 if (testpoint(consumerd_thread_sessiond)) {
3089 goto error_testpoint;
3090 }
3091
9ce5646a
MD
3092 health_code_update();
3093
3bd1e081
MD
3094 DBG("Creating command socket %s", ctx->consumer_command_sock_path);
3095 unlink(ctx->consumer_command_sock_path);
3096 client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path);
3097 if (client_socket < 0) {
3098 ERR("Cannot create command socket");
3099 goto end;
3100 }
3101
3102 ret = lttcomm_listen_unix_sock(client_socket);
3103 if (ret < 0) {
3104 goto end;
3105 }
3106
32258573 3107 DBG("Sending ready command to lttng-sessiond");
f73fabfd 3108 ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY);
3bd1e081
MD
3109 /* return < 0 on error, but == 0 is not fatal */
3110 if (ret < 0) {
32258573 3111 ERR("Error sending ready command to lttng-sessiond");
3bd1e081
MD
3112 goto end;
3113 }
3114
3bd1e081
MD
3115 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3116 consumer_sockpoll[0].fd = ctx->consumer_should_quit[0];
3117 consumer_sockpoll[0].events = POLLIN | POLLPRI;
3118 consumer_sockpoll[1].fd = client_socket;
3119 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3120
84382d49
MD
3121 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3122 if (ret) {
3123 if (ret > 0) {
3124 /* should exit */
3125 err = 0;
3126 }
3bd1e081
MD
3127 goto end;
3128 }
3129 DBG("Connection on client_socket");
3130
3131 /* Blocking call, waiting for transmission */
3132 sock = lttcomm_accept_unix_sock(client_socket);
534d2592 3133 if (sock < 0) {
3bd1e081
MD
3134 WARN("On accept");
3135 goto end;
3136 }
3bd1e081 3137
331744e3
JD
3138 /*
3139 * Setup metadata socket which is the second socket connection on the
3140 * command unix socket.
3141 */
3142 ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket);
84382d49
MD
3143 if (ret) {
3144 if (ret > 0) {
3145 /* should exit */
3146 err = 0;
3147 }
331744e3
JD
3148 goto end;
3149 }
3150
d96f09c6
DG
3151 /* This socket is not useful anymore. */
3152 ret = close(client_socket);
3153 if (ret < 0) {
3154 PERROR("close client_socket");
3155 }
3156 client_socket = -1;
3157
3bd1e081
MD
3158 /* update the polling structure to poll on the established socket */
3159 consumer_sockpoll[1].fd = sock;
3160 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3161
3162 while (1) {
9ce5646a
MD
3163 health_code_update();
3164
3165 health_poll_entry();
3166 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3167 health_poll_exit();
84382d49
MD
3168 if (ret) {
3169 if (ret > 0) {
3170 /* should exit */
3171 err = 0;
3172 }
3bd1e081
MD
3173 goto end;
3174 }
3175 DBG("Incoming command on sock");
3176 ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll);
4cbc1a04
DG
3177 if (ret <= 0) {
3178 /*
3179 * This could simply be a session daemon quitting. Don't output
3180 * ERR() here.
3181 */
3182 DBG("Communication interrupted on command socket");
41ba6035 3183 err = 0;
3bd1e081
MD
3184 goto end;
3185 }
3186 if (consumer_quit) {
3187 DBG("consumer_thread_receive_fds received quit from signal");
1fc79fb4 3188 err = 0; /* All is OK */
3bd1e081
MD
3189 goto end;
3190 }
ffe60014 3191 DBG("received command on sock");
3bd1e081 3192 }
1fc79fb4
MD
3193 /* All is OK */
3194 err = 0;
3195
3bd1e081 3196end:
ffe60014 3197 DBG("Consumer thread sessiond poll exiting");
3bd1e081 3198
d88aee68
DG
3199 /*
3200 * Close metadata streams since the producer is the session daemon which
3201 * just died.
3202 *
3203 * NOTE: for now, this only applies to the UST tracer.
3204 */
6d574024 3205 lttng_consumer_close_all_metadata();
d88aee68 3206
3bd1e081
MD
3207 /*
3208 * when all fds have hung up, the polling thread
3209 * can exit cleanly
3210 */
3211 consumer_quit = 1;
3212
04fdd819 3213 /*
c869f647 3214 * Notify the data poll thread to poll back again and test the
8994307f 3215 * consumer_quit state that we just set so to quit gracefully.
04fdd819 3216 */
acdb9057 3217 notify_thread_lttng_pipe(ctx->consumer_data_pipe);
c869f647 3218
a0cbdd2e 3219 notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT);
d8ef542d 3220
5c635c72
MD
3221 notify_health_quit_pipe(health_quit_pipe);
3222
d96f09c6
DG
3223 /* Cleaning up possibly open sockets. */
3224 if (sock >= 0) {
3225 ret = close(sock);
3226 if (ret < 0) {
3227 PERROR("close sock sessiond poll");
3228 }
3229 }
3230 if (client_socket >= 0) {
38476d24 3231 ret = close(client_socket);
d96f09c6
DG
3232 if (ret < 0) {
3233 PERROR("close client_socket sessiond poll");
3234 }
3235 }
3236
2d57de81 3237error_testpoint:
1fc79fb4
MD
3238 if (err) {
3239 health_error();
3240 ERR("Health error occurred in %s", __func__);
3241 }
3242 health_unregister(health_consumerd);
3243
e7b994a3 3244 rcu_unregister_thread();
3bd1e081
MD
3245 return NULL;
3246}
d41f73b7 3247
4078b776 3248ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream,
d41f73b7
MD
3249 struct lttng_consumer_local_data *ctx)
3250{
74251bb8
DG
3251 ssize_t ret;
3252
3253 pthread_mutex_lock(&stream->lock);
94d49140
JD
3254 if (stream->metadata_flag) {
3255 pthread_mutex_lock(&stream->metadata_rdv_lock);
3256 }
74251bb8 3257
d41f73b7
MD
3258 switch (consumer_data.type) {
3259 case LTTNG_CONSUMER_KERNEL:
74251bb8
DG
3260 ret = lttng_kconsumer_read_subbuffer(stream, ctx);
3261 break;
7753dea8
MD
3262 case LTTNG_CONSUMER32_UST:
3263 case LTTNG_CONSUMER64_UST:
74251bb8
DG
3264 ret = lttng_ustconsumer_read_subbuffer(stream, ctx);
3265 break;
d41f73b7
MD
3266 default:
3267 ERR("Unknown consumer_data type");
3268 assert(0);
74251bb8
DG
3269 ret = -ENOSYS;
3270 break;
d41f73b7 3271 }
74251bb8 3272
94d49140
JD
3273 if (stream->metadata_flag) {
3274 pthread_cond_broadcast(&stream->metadata_rdv);
3275 pthread_mutex_unlock(&stream->metadata_rdv_lock);
3276 }
74251bb8
DG
3277 pthread_mutex_unlock(&stream->lock);
3278 return ret;
d41f73b7
MD
3279}
3280
3281int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream)
3282{
3283 switch (consumer_data.type) {
3284 case LTTNG_CONSUMER_KERNEL:
3285 return lttng_kconsumer_on_recv_stream(stream);
7753dea8
MD
3286 case LTTNG_CONSUMER32_UST:
3287 case LTTNG_CONSUMER64_UST:
d41f73b7
MD
3288 return lttng_ustconsumer_on_recv_stream(stream);
3289 default:
3290 ERR("Unknown consumer_data type");
3291 assert(0);
3292 return -ENOSYS;
3293 }
3294}
e4421fec
DG
3295
3296/*
3297 * Allocate and set consumer data hash tables.
3298 */
282dadbc 3299int lttng_consumer_init(void)
e4421fec 3300{
d88aee68 3301 consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
282dadbc
MD
3302 if (!consumer_data.channel_ht) {
3303 goto error;
3304 }
3305
d88aee68 3306 consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
282dadbc
MD
3307 if (!consumer_data.relayd_ht) {
3308 goto error;
3309 }
3310
d88aee68 3311 consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
282dadbc
MD
3312 if (!consumer_data.stream_list_ht) {
3313 goto error;
3314 }
3315
d8ef542d 3316 consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
282dadbc
MD
3317 if (!consumer_data.stream_per_chan_id_ht) {
3318 goto error;
3319 }
3320
3321 data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3322 if (!data_ht) {
3323 goto error;
3324 }
3325
3326 metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3327 if (!metadata_ht) {
3328 goto error;
3329 }
3330
3331 return 0;
3332
3333error:
3334 return -1;
e4421fec 3335}
7735ef9e
DG
3336
3337/*
3338 * Process the ADD_RELAYD command receive by a consumer.
3339 *
3340 * This will create a relayd socket pair and add it to the relayd hash table.
3341 * The caller MUST acquire a RCU read side lock before calling it.
3342 */
da009f2c 3343int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type,
7735ef9e 3344 struct lttng_consumer_local_data *ctx, int sock,
6151a90f 3345 struct pollfd *consumer_sockpoll,
d3e2ba59
JD
3346 struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id,
3347 uint64_t relayd_session_id)
7735ef9e 3348{
cd2b09ed 3349 int fd = -1, ret = -1, relayd_created = 0;
0c759fc9 3350 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
d4298c99 3351 struct consumer_relayd_sock_pair *relayd = NULL;
7735ef9e 3352
6151a90f
JD
3353 assert(ctx);
3354 assert(relayd_sock);
3355
da009f2c 3356 DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx);
7735ef9e
DG
3357
3358 /* Get relayd reference if exists. */
3359 relayd = consumer_find_relayd(net_seq_idx);
3360 if (relayd == NULL) {
da009f2c 3361 assert(sock_type == LTTNG_STREAM_CONTROL);
7735ef9e
DG
3362 /* Not found. Allocate one. */
3363 relayd = consumer_allocate_relayd_sock_pair(net_seq_idx);
3364 if (relayd == NULL) {
0d08d75e 3365 ret = -ENOMEM;
618a6a28
MD
3366 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
3367 goto error;
0d08d75e 3368 } else {
30319bcb 3369 relayd->sessiond_session_id = sessiond_id;
0d08d75e 3370 relayd_created = 1;
7735ef9e 3371 }
0d08d75e
DG
3372
3373 /*
3374 * This code path MUST continue to the consumer send status message to
3375 * we can notify the session daemon and continue our work without
3376 * killing everything.
3377 */
da009f2c
MD
3378 } else {
3379 /*
3380 * relayd key should never be found for control socket.
3381 */
3382 assert(sock_type != LTTNG_STREAM_CONTROL);
0d08d75e
DG
3383 }
3384
3385 /* First send a status message before receiving the fds. */
0c759fc9 3386 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
618a6a28 3387 if (ret < 0) {
0d08d75e 3388 /* Somehow, the session daemon is not responding anymore. */
618a6a28
MD
3389 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3390 goto error_nosignal;
7735ef9e
DG
3391 }
3392
3393 /* Poll on consumer socket. */
84382d49
MD
3394 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3395 if (ret) {
3396 /* Needing to exit in the middle of a command: error. */
0d08d75e 3397 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
7735ef9e 3398 ret = -EINTR;
618a6a28 3399 goto error_nosignal;
7735ef9e
DG
3400 }
3401
3402 /* Get relayd socket from session daemon */
3403 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
3404 if (ret != sizeof(fd)) {
7735ef9e 3405 ret = -1;
4028eeb9 3406 fd = -1; /* Just in case it gets set with an invalid value. */
0d08d75e
DG
3407
3408 /*
3409 * Failing to receive FDs might indicate a major problem such as
3410 * reaching a fd limit during the receive where the kernel returns a
3411 * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we
3412 * don't take any chances and stop everything.
3413 *
3414 * XXX: Feature request #558 will fix that and avoid this possible
3415 * issue when reaching the fd limit.
3416 */
3417 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
618a6a28 3418 ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD;
f50f23d9
DG
3419 goto error;
3420 }
3421
7735ef9e
DG
3422 /* Copy socket information and received FD */
3423 switch (sock_type) {
3424 case LTTNG_STREAM_CONTROL:
3425 /* Copy received lttcomm socket */
6151a90f
JD
3426 lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock);
3427 ret = lttcomm_create_sock(&relayd->control_sock.sock);
4028eeb9 3428 /* Handle create_sock error. */
f66c074c 3429 if (ret < 0) {
618a6a28 3430 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
4028eeb9 3431 goto error;
f66c074c 3432 }
da009f2c
MD
3433 /*
3434 * Close the socket created internally by
3435 * lttcomm_create_sock, so we can replace it by the one
3436 * received from sessiond.
3437 */
3438 if (close(relayd->control_sock.sock.fd)) {
3439 PERROR("close");
3440 }
7735ef9e
DG
3441
3442 /* Assign new file descriptor */
6151a90f 3443 relayd->control_sock.sock.fd = fd;
4b29f1ce 3444 fd = -1; /* For error path */
6151a90f
JD
3445 /* Assign version values. */
3446 relayd->control_sock.major = relayd_sock->major;
3447 relayd->control_sock.minor = relayd_sock->minor;
c5b6f4f0 3448
d3e2ba59 3449 relayd->relayd_session_id = relayd_session_id;
c5b6f4f0 3450
7735ef9e
DG
3451 break;
3452 case LTTNG_STREAM_DATA:
3453 /* Copy received lttcomm socket */
6151a90f
JD
3454 lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock);
3455 ret = lttcomm_create_sock(&relayd->data_sock.sock);
4028eeb9 3456 /* Handle create_sock error. */
f66c074c 3457 if (ret < 0) {
618a6a28 3458 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
4028eeb9 3459 goto error;
f66c074c 3460 }
da009f2c
MD
3461 /*
3462 * Close the socket created internally by
3463 * lttcomm_create_sock, so we can replace it by the one
3464 * received from sessiond.
3465 */
3466 if (close(relayd->data_sock.sock.fd)) {
3467 PERROR("close");
3468 }
7735ef9e
DG
3469
3470 /* Assign new file descriptor */
6151a90f 3471 relayd->data_sock.sock.fd = fd;
4b29f1ce 3472 fd = -1; /* for eventual error paths */
6151a90f
JD
3473 /* Assign version values. */
3474 relayd->data_sock.major = relayd_sock->major;
3475 relayd->data_sock.minor = relayd_sock->minor;
7735ef9e
DG
3476 break;
3477 default:
3478 ERR("Unknown relayd socket type (%d)", sock_type);
59e71485 3479 ret = -1;
618a6a28 3480 ret_code = LTTCOMM_CONSUMERD_FATAL;
7735ef9e
DG
3481 goto error;
3482 }
3483
d88aee68 3484 DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)",
7735ef9e
DG
3485 sock_type == LTTNG_STREAM_CONTROL ? "control" : "data",
3486 relayd->net_seq_idx, fd);
3487
618a6a28
MD
3488 /* We successfully added the socket. Send status back. */
3489 ret = consumer_send_status_msg(sock, ret_code);
3490 if (ret < 0) {
3491 /* Somehow, the session daemon is not responding anymore. */
3492 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3493 goto error_nosignal;
3494 }
3495
7735ef9e
DG
3496 /*
3497 * Add relayd socket pair to consumer data hashtable. If object already
3498 * exists or on error, the function gracefully returns.
3499 */
d09e1200 3500 add_relayd(relayd);
7735ef9e
DG
3501
3502 /* All good! */
4028eeb9 3503 return 0;
7735ef9e
DG
3504
3505error:
618a6a28
MD
3506 if (consumer_send_status_msg(sock, ret_code) < 0) {
3507 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3508 }
3509
3510error_nosignal:
4028eeb9
DG
3511 /* Close received socket if valid. */
3512 if (fd >= 0) {
3513 if (close(fd)) {
3514 PERROR("close received socket");
3515 }
3516 }
cd2b09ed
DG
3517
3518 if (relayd_created) {
cd2b09ed
DG
3519 free(relayd);
3520 }
3521
7735ef9e
DG
3522 return ret;
3523}
ca22feea 3524
4e9a4686
DG
3525/*
3526 * Try to lock the stream mutex.
3527 *
3528 * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
3529 */
3530static int stream_try_lock(struct lttng_consumer_stream *stream)
3531{
3532 int ret;
3533
3534 assert(stream);
3535
3536 /*
3537 * Try to lock the stream mutex. On failure, we know that the stream is
3538 * being used else where hence there is data still being extracted.
3539 */
3540 ret = pthread_mutex_trylock(&stream->lock);
3541 if (ret) {
3542 /* For both EBUSY and EINVAL error, the mutex is NOT locked. */
3543 ret = 0;
3544 goto end;
3545 }
3546
3547 ret = 1;
3548
3549end:
3550 return ret;
3551}
3552
f7079f67
DG
3553/*
3554 * Search for a relayd associated to the session id and return the reference.
3555 *
3556 * A rcu read side lock MUST be acquire before calling this function and locked
3557 * until the relayd object is no longer necessary.
3558 */
3559static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id)
3560{
3561 struct lttng_ht_iter iter;
f7079f67 3562 struct consumer_relayd_sock_pair *relayd = NULL;
f7079f67
DG
3563
3564 /* Iterate over all relayd since they are indexed by net_seq_idx. */
3565 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
3566 node.node) {
18261bd1
DG
3567 /*
3568 * Check by sessiond id which is unique here where the relayd session
3569 * id might not be when having multiple relayd.
3570 */
3571 if (relayd->sessiond_session_id == id) {
f7079f67 3572 /* Found the relayd. There can be only one per id. */
18261bd1 3573 goto found;
f7079f67
DG
3574 }
3575 }
3576
18261bd1
DG
3577 return NULL;
3578
3579found:
f7079f67
DG
3580 return relayd;
3581}
3582
ca22feea
DG
3583/*
3584 * Check if for a given session id there is still data needed to be extract
3585 * from the buffers.
3586 *
6d805429 3587 * Return 1 if data is pending or else 0 meaning ready to be read.
ca22feea 3588 */
6d805429 3589int consumer_data_pending(uint64_t id)
ca22feea
DG
3590{
3591 int ret;
3592 struct lttng_ht_iter iter;
3593 struct lttng_ht *ht;
3594 struct lttng_consumer_stream *stream;
f7079f67 3595 struct consumer_relayd_sock_pair *relayd = NULL;
6d805429 3596 int (*data_pending)(struct lttng_consumer_stream *);
ca22feea 3597
6d805429 3598 DBG("Consumer data pending command on session id %" PRIu64, id);
ca22feea 3599
6f6eda74 3600 rcu_read_lock();
ca22feea
DG
3601 pthread_mutex_lock(&consumer_data.lock);
3602
3603 switch (consumer_data.type) {
3604 case LTTNG_CONSUMER_KERNEL:
6d805429 3605 data_pending = lttng_kconsumer_data_pending;
ca22feea
DG
3606 break;
3607 case LTTNG_CONSUMER32_UST:
3608 case LTTNG_CONSUMER64_UST:
6d805429 3609 data_pending = lttng_ustconsumer_data_pending;
ca22feea
DG
3610 break;
3611 default:
3612 ERR("Unknown consumer data type");
3613 assert(0);
3614 }
3615
3616 /* Ease our life a bit */
3617 ht = consumer_data.stream_list_ht;
3618
f7079f67
DG
3619 relayd = find_relayd_by_session_id(id);
3620 if (relayd) {
3621 /* Send init command for data pending. */
3622 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3623 ret = relayd_begin_data_pending(&relayd->control_sock,
3624 relayd->relayd_session_id);
3625 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
3626 if (ret < 0) {
3627 /* Communication error thus the relayd so no data pending. */
3628 goto data_not_pending;
3629 }
3630 }
3631
c8f59ee5 3632 cds_lfht_for_each_entry_duplicate(ht->ht,
d88aee68
DG
3633 ht->hash_fct(&id, lttng_ht_seed),
3634 ht->match_fct, &id,
ca22feea 3635 &iter.iter, stream, node_session_id.node) {
4e9a4686
DG
3636 /* If this call fails, the stream is being used hence data pending. */
3637 ret = stream_try_lock(stream);
3638 if (!ret) {
f7079f67 3639 goto data_pending;
ca22feea 3640 }
ca22feea 3641
4e9a4686
DG
3642 /*
3643 * A removed node from the hash table indicates that the stream has
3644 * been deleted thus having a guarantee that the buffers are closed
3645 * on the consumer side. However, data can still be transmitted
3646 * over the network so don't skip the relayd check.
3647 */
3648 ret = cds_lfht_is_node_deleted(&stream->node.node);
3649 if (!ret) {
3650 /* Check the stream if there is data in the buffers. */
6d805429
DG
3651 ret = data_pending(stream);
3652 if (ret == 1) {
4e9a4686 3653 pthread_mutex_unlock(&stream->lock);
f7079f67 3654 goto data_pending;
4e9a4686
DG
3655 }
3656 }
3657
3658 /* Relayd check */
f7079f67 3659 if (relayd) {
c8f59ee5
DG
3660 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3661 if (stream->metadata_flag) {
ad7051c0
DG
3662 ret = relayd_quiescent_control(&relayd->control_sock,
3663 stream->relayd_stream_id);
c8f59ee5 3664 } else {
6d805429 3665 ret = relayd_data_pending(&relayd->control_sock,
39df6d9f
DG
3666 stream->relayd_stream_id,
3667 stream->next_net_seq_num - 1);
c8f59ee5
DG
3668 }
3669 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
6d805429 3670 if (ret == 1) {
4e9a4686 3671 pthread_mutex_unlock(&stream->lock);
f7079f67 3672 goto data_pending;
c8f59ee5
DG
3673 }
3674 }
4e9a4686 3675 pthread_mutex_unlock(&stream->lock);
c8f59ee5 3676 }
ca22feea 3677
f7079f67
DG
3678 if (relayd) {
3679 unsigned int is_data_inflight = 0;
3680
3681 /* Send init command for data pending. */
3682 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3683 ret = relayd_end_data_pending(&relayd->control_sock,
3684 relayd->relayd_session_id, &is_data_inflight);
3685 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
bdd88757 3686 if (ret < 0) {
f7079f67
DG
3687 goto data_not_pending;
3688 }
bdd88757
DG
3689 if (is_data_inflight) {
3690 goto data_pending;
3691 }
f7079f67
DG
3692 }
3693
ca22feea 3694 /*
f7079f67
DG
3695 * Finding _no_ node in the hash table and no inflight data means that the
3696 * stream(s) have been removed thus data is guaranteed to be available for
3697 * analysis from the trace files.
ca22feea
DG
3698 */
3699
f7079f67 3700data_not_pending:
ca22feea
DG
3701 /* Data is available to be read by a viewer. */
3702 pthread_mutex_unlock(&consumer_data.lock);
c8f59ee5 3703 rcu_read_unlock();
6d805429 3704 return 0;
ca22feea 3705
f7079f67 3706data_pending:
ca22feea
DG
3707 /* Data is still being extracted from buffers. */
3708 pthread_mutex_unlock(&consumer_data.lock);
c8f59ee5 3709 rcu_read_unlock();
6d805429 3710 return 1;
ca22feea 3711}
f50f23d9
DG
3712
3713/*
3714 * Send a ret code status message to the sessiond daemon.
3715 *
3716 * Return the sendmsg() return value.
3717 */
3718int consumer_send_status_msg(int sock, int ret_code)
3719{
3720 struct lttcomm_consumer_status_msg msg;
3721
53efb85a 3722 memset(&msg, 0, sizeof(msg));
f50f23d9
DG
3723 msg.ret_code = ret_code;
3724
3725 return lttcomm_send_unix_sock(sock, &msg, sizeof(msg));
3726}
ffe60014
DG
3727
3728/*
3729 * Send a channel status message to the sessiond daemon.
3730 *
3731 * Return the sendmsg() return value.
3732 */
3733int consumer_send_status_channel(int sock,
3734 struct lttng_consumer_channel *channel)
3735{
3736 struct lttcomm_consumer_status_channel msg;
3737
3738 assert(sock >= 0);
3739
53efb85a 3740 memset(&msg, 0, sizeof(msg));
ffe60014 3741 if (!channel) {
0c759fc9 3742 msg.ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
ffe60014 3743 } else {
0c759fc9 3744 msg.ret_code = LTTCOMM_CONSUMERD_SUCCESS;
ffe60014
DG
3745 msg.key = channel->key;
3746 msg.stream_count = channel->streams.count;
3747 }
3748
3749 return lttcomm_send_unix_sock(sock, &msg, sizeof(msg));
3750}
5c786ded 3751
d07ceecd
MD
3752unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos,
3753 unsigned long produced_pos, uint64_t nb_packets_per_stream,
3754 uint64_t max_sb_size)
5c786ded 3755{
d07ceecd 3756 unsigned long start_pos;
5c786ded 3757
d07ceecd
MD
3758 if (!nb_packets_per_stream) {
3759 return consumed_pos; /* Grab everything */
3760 }
3761 start_pos = produced_pos - offset_align_floor(produced_pos, max_sb_size);
3762 start_pos -= max_sb_size * nb_packets_per_stream;
3763 if ((long) (start_pos - consumed_pos) < 0) {
3764 return consumed_pos; /* Grab everything */
3765 }
3766 return start_pos;
5c786ded 3767}
This page took 0.280571 seconds and 4 git commands to generate.