tests: Implement tests for lttng-ust LTTNG_UST_BLOCKING_RETRY_TIMEOUT
[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
5199ffc4
JG
1232
1233/*
1234 * Flush pending writes to trace output disk file.
1235 */
1236static
00e2e675
DG
1237void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream,
1238 off_t orig_offset)
3bd1e081 1239{
c7a78aab 1240 int ret;
3bd1e081
MD
1241 int outfd = stream->out_fd;
1242
1243 /*
1244 * This does a blocking write-and-wait on any page that belongs to the
1245 * subbuffer prior to the one we just wrote.
1246 * Don't care about error values, as these are just hints and ways to
1247 * limit the amount of page cache used.
1248 */
ffe60014 1249 if (orig_offset < stream->max_sb_size) {
3bd1e081
MD
1250 return;
1251 }
ffe60014
DG
1252 lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size,
1253 stream->max_sb_size,
3bd1e081
MD
1254 SYNC_FILE_RANGE_WAIT_BEFORE
1255 | SYNC_FILE_RANGE_WRITE
1256 | SYNC_FILE_RANGE_WAIT_AFTER);
1257 /*
1258 * Give hints to the kernel about how we access the file:
1259 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1260 * we write it.
1261 *
1262 * We need to call fadvise again after the file grows because the
1263 * kernel does not seem to apply fadvise to non-existing parts of the
1264 * file.
1265 *
1266 * Call fadvise _after_ having waited for the page writeback to
1267 * complete because the dirty page writeback semantic is not well
1268 * defined. So it can be expected to lead to lower throughput in
1269 * streaming.
1270 */
c7a78aab 1271 ret = posix_fadvise(outfd, orig_offset - stream->max_sb_size,
ffe60014 1272 stream->max_sb_size, POSIX_FADV_DONTNEED);
a0d0e127 1273 if (ret && ret != -ENOSYS) {
a74a5f4a
JG
1274 errno = ret;
1275 PERROR("posix_fadvise on fd %i", outfd);
c7a78aab 1276 }
3bd1e081
MD
1277}
1278
1279/*
1280 * Initialise the necessary environnement :
1281 * - create a new context
1282 * - create the poll_pipe
1283 * - create the should_quit pipe (for signal handler)
1284 * - create the thread pipe (for splice)
1285 *
1286 * Takes a function pointer as argument, this function is called when data is
1287 * available on a buffer. This function is responsible to do the
1288 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1289 * buffer configuration and then kernctl_put_next_subbuf at the end.
1290 *
1291 * Returns a pointer to the new context or NULL on error.
1292 */
1293struct lttng_consumer_local_data *lttng_consumer_create(
1294 enum lttng_consumer_type type,
4078b776 1295 ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream,
d41f73b7 1296 struct lttng_consumer_local_data *ctx),
3bd1e081
MD
1297 int (*recv_channel)(struct lttng_consumer_channel *channel),
1298 int (*recv_stream)(struct lttng_consumer_stream *stream),
30319bcb 1299 int (*update_stream)(uint64_t stream_key, uint32_t state))
3bd1e081 1300{
d8ef542d 1301 int ret;
3bd1e081
MD
1302 struct lttng_consumer_local_data *ctx;
1303
1304 assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN ||
1305 consumer_data.type == type);
1306 consumer_data.type = type;
1307
effcf122 1308 ctx = zmalloc(sizeof(struct lttng_consumer_local_data));
3bd1e081 1309 if (ctx == NULL) {
7a57cf92 1310 PERROR("allocating context");
3bd1e081
MD
1311 goto error;
1312 }
1313
1314 ctx->consumer_error_socket = -1;
331744e3 1315 ctx->consumer_metadata_socket = -1;
75d83e50 1316 pthread_mutex_init(&ctx->metadata_socket_lock, NULL);
3bd1e081
MD
1317 /* assign the callbacks */
1318 ctx->on_buffer_ready = buffer_ready;
1319 ctx->on_recv_channel = recv_channel;
1320 ctx->on_recv_stream = recv_stream;
1321 ctx->on_update_stream = update_stream;
1322
acdb9057
DG
1323 ctx->consumer_data_pipe = lttng_pipe_open(0);
1324 if (!ctx->consumer_data_pipe) {
3bd1e081
MD
1325 goto error_poll_pipe;
1326 }
1327
02b3d176
DG
1328 ctx->consumer_wakeup_pipe = lttng_pipe_open(0);
1329 if (!ctx->consumer_wakeup_pipe) {
1330 goto error_wakeup_pipe;
1331 }
1332
3bd1e081
MD
1333 ret = pipe(ctx->consumer_should_quit);
1334 if (ret < 0) {
7a57cf92 1335 PERROR("Error creating recv pipe");
3bd1e081
MD
1336 goto error_quit_pipe;
1337 }
1338
d8ef542d
MD
1339 ret = pipe(ctx->consumer_channel_pipe);
1340 if (ret < 0) {
1341 PERROR("Error creating channel pipe");
1342 goto error_channel_pipe;
1343 }
1344
13886d2d
DG
1345 ctx->consumer_metadata_pipe = lttng_pipe_open(0);
1346 if (!ctx->consumer_metadata_pipe) {
fb3a43a9
DG
1347 goto error_metadata_pipe;
1348 }
3bd1e081 1349
fb3a43a9 1350 return ctx;
3bd1e081 1351
fb3a43a9 1352error_metadata_pipe:
d8ef542d
MD
1353 utils_close_pipe(ctx->consumer_channel_pipe);
1354error_channel_pipe:
d8ef542d 1355 utils_close_pipe(ctx->consumer_should_quit);
3bd1e081 1356error_quit_pipe:
02b3d176
DG
1357 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
1358error_wakeup_pipe:
acdb9057 1359 lttng_pipe_destroy(ctx->consumer_data_pipe);
3bd1e081
MD
1360error_poll_pipe:
1361 free(ctx);
1362error:
1363 return NULL;
1364}
1365
282dadbc
MD
1366/*
1367 * Iterate over all streams of the hashtable and free them properly.
1368 */
1369static void destroy_data_stream_ht(struct lttng_ht *ht)
1370{
1371 struct lttng_ht_iter iter;
1372 struct lttng_consumer_stream *stream;
1373
1374 if (ht == NULL) {
1375 return;
1376 }
1377
1378 rcu_read_lock();
1379 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1380 /*
1381 * Ignore return value since we are currently cleaning up so any error
1382 * can't be handled.
1383 */
1384 (void) consumer_del_stream(stream, ht);
1385 }
1386 rcu_read_unlock();
1387
1388 lttng_ht_destroy(ht);
1389}
1390
1391/*
1392 * Iterate over all streams of the metadata hashtable and free them
1393 * properly.
1394 */
1395static void destroy_metadata_stream_ht(struct lttng_ht *ht)
1396{
1397 struct lttng_ht_iter iter;
1398 struct lttng_consumer_stream *stream;
1399
1400 if (ht == NULL) {
1401 return;
1402 }
1403
1404 rcu_read_lock();
1405 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1406 /*
1407 * Ignore return value since we are currently cleaning up so any error
1408 * can't be handled.
1409 */
1410 (void) consumer_del_metadata_stream(stream, ht);
1411 }
1412 rcu_read_unlock();
1413
1414 lttng_ht_destroy(ht);
1415}
1416
3bd1e081
MD
1417/*
1418 * Close all fds associated with the instance and free the context.
1419 */
1420void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx)
1421{
4c462e79
MD
1422 int ret;
1423
ab1027f4
DG
1424 DBG("Consumer destroying it. Closing everything.");
1425
4f2e75b9
DG
1426 if (!ctx) {
1427 return;
1428 }
1429
282dadbc
MD
1430 destroy_data_stream_ht(data_ht);
1431 destroy_metadata_stream_ht(metadata_ht);
1432
4c462e79
MD
1433 ret = close(ctx->consumer_error_socket);
1434 if (ret) {
1435 PERROR("close");
1436 }
331744e3
JD
1437 ret = close(ctx->consumer_metadata_socket);
1438 if (ret) {
1439 PERROR("close");
1440 }
d8ef542d 1441 utils_close_pipe(ctx->consumer_channel_pipe);
acdb9057 1442 lttng_pipe_destroy(ctx->consumer_data_pipe);
13886d2d 1443 lttng_pipe_destroy(ctx->consumer_metadata_pipe);
02b3d176 1444 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
d8ef542d 1445 utils_close_pipe(ctx->consumer_should_quit);
fb3a43a9 1446
3bd1e081
MD
1447 unlink(ctx->consumer_command_sock_path);
1448 free(ctx);
1449}
1450
6197aea7
DG
1451/*
1452 * Write the metadata stream id on the specified file descriptor.
1453 */
1454static int write_relayd_metadata_id(int fd,
1455 struct lttng_consumer_stream *stream,
ffe60014 1456 struct consumer_relayd_sock_pair *relayd, unsigned long padding)
6197aea7 1457{
6cd525e8 1458 ssize_t ret;
1d4dfdef 1459 struct lttcomm_relayd_metadata_payload hdr;
6197aea7 1460
1d4dfdef
DG
1461 hdr.stream_id = htobe64(stream->relayd_stream_id);
1462 hdr.padding_size = htobe32(padding);
6cd525e8
MD
1463 ret = lttng_write(fd, (void *) &hdr, sizeof(hdr));
1464 if (ret < sizeof(hdr)) {
d7b75ec8 1465 /*
6f04ed72 1466 * This error means that the fd's end is closed so ignore the PERROR
d7b75ec8
DG
1467 * not to clubber the error output since this can happen in a normal
1468 * code path.
1469 */
1470 if (errno != EPIPE) {
1471 PERROR("write metadata stream id");
1472 }
1473 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno);
534d2592
DG
1474 /*
1475 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1476 * handle writting the missing part so report that as an error and
1477 * don't lie to the caller.
1478 */
1479 ret = -1;
6197aea7
DG
1480 goto end;
1481 }
1d4dfdef
DG
1482 DBG("Metadata stream id %" PRIu64 " with padding %lu written before data",
1483 stream->relayd_stream_id, padding);
6197aea7
DG
1484
1485end:
6cd525e8 1486 return (int) ret;
6197aea7
DG
1487}
1488
3bd1e081 1489/*
09e26845
DG
1490 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1491 * core function for writing trace buffers to either the local filesystem or
1492 * the network.
1493 *
79d4ffb7
DG
1494 * It must be called with the stream lock held.
1495 *
09e26845 1496 * Careful review MUST be put if any changes occur!
3bd1e081
MD
1497 *
1498 * Returns the number of bytes written
1499 */
4078b776 1500ssize_t lttng_consumer_on_read_subbuffer_mmap(
3bd1e081 1501 struct lttng_consumer_local_data *ctx,
1d4dfdef 1502 struct lttng_consumer_stream *stream, unsigned long len,
309167d2 1503 unsigned long padding,
50adc264 1504 struct ctf_packet_index *index)
3bd1e081 1505{
f02e1e8a 1506 unsigned long mmap_offset;
ffe60014 1507 void *mmap_base;
994ab360 1508 ssize_t ret = 0;
f02e1e8a
DG
1509 off_t orig_offset = stream->out_fd_offset;
1510 /* Default is on the disk */
1511 int outfd = stream->out_fd;
f02e1e8a 1512 struct consumer_relayd_sock_pair *relayd = NULL;
8994307f 1513 unsigned int relayd_hang_up = 0;
f02e1e8a
DG
1514
1515 /* RCU lock for the relayd pointer */
1516 rcu_read_lock();
1517
1518 /* Flag that the current stream if set for network streaming. */
da009f2c 1519 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1520 relayd = consumer_find_relayd(stream->net_seq_idx);
1521 if (relayd == NULL) {
56591bac 1522 ret = -EPIPE;
f02e1e8a
DG
1523 goto end;
1524 }
1525 }
1526
1527 /* get the offset inside the fd to mmap */
3bd1e081
MD
1528 switch (consumer_data.type) {
1529 case LTTNG_CONSUMER_KERNEL:
ffe60014 1530 mmap_base = stream->mmap_base;
f02e1e8a 1531 ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset);
994ab360 1532 if (ret < 0) {
56591bac 1533 PERROR("tracer ctl get_mmap_read_offset");
56591bac
MD
1534 goto end;
1535 }
f02e1e8a 1536 break;
7753dea8
MD
1537 case LTTNG_CONSUMER32_UST:
1538 case LTTNG_CONSUMER64_UST:
ffe60014
DG
1539 mmap_base = lttng_ustctl_get_mmap_base(stream);
1540 if (!mmap_base) {
1541 ERR("read mmap get mmap base for stream %s", stream->name);
994ab360 1542 ret = -EPERM;
ffe60014
DG
1543 goto end;
1544 }
1545 ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset);
56591bac
MD
1546 if (ret != 0) {
1547 PERROR("tracer ctl get_mmap_read_offset");
994ab360 1548 ret = -EINVAL;
56591bac
MD
1549 goto end;
1550 }
f02e1e8a 1551 break;
3bd1e081
MD
1552 default:
1553 ERR("Unknown consumer_data type");
1554 assert(0);
1555 }
b9182dd9 1556
f02e1e8a
DG
1557 /* Handle stream on the relayd if the output is on the network */
1558 if (relayd) {
1559 unsigned long netlen = len;
1560
1561 /*
1562 * Lock the control socket for the complete duration of the function
1563 * since from this point on we will use the socket.
1564 */
1565 if (stream->metadata_flag) {
1566 /* Metadata requires the control socket. */
1567 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
93ec662e
JD
1568 if (stream->reset_metadata_flag) {
1569 ret = relayd_reset_metadata(&relayd->control_sock,
1570 stream->relayd_stream_id,
1571 stream->metadata_version);
1572 if (ret < 0) {
1573 relayd_hang_up = 1;
1574 goto write_error;
1575 }
1576 stream->reset_metadata_flag = 0;
1577 }
1d4dfdef 1578 netlen += sizeof(struct lttcomm_relayd_metadata_payload);
f02e1e8a
DG
1579 }
1580
1d4dfdef 1581 ret = write_relayd_stream_header(stream, netlen, padding, relayd);
994ab360
DG
1582 if (ret < 0) {
1583 relayd_hang_up = 1;
1584 goto write_error;
1585 }
1586 /* Use the returned socket. */
1587 outfd = ret;
f02e1e8a 1588
994ab360
DG
1589 /* Write metadata stream id before payload */
1590 if (stream->metadata_flag) {
1591 ret = write_relayd_metadata_id(outfd, stream, relayd, padding);
1592 if (ret < 0) {
8994307f
DG
1593 relayd_hang_up = 1;
1594 goto write_error;
1595 }
f02e1e8a 1596 }
1d4dfdef
DG
1597 } else {
1598 /* No streaming, we have to set the len with the full padding */
1599 len += padding;
1624d5b7 1600
93ec662e
JD
1601 if (stream->metadata_flag && stream->reset_metadata_flag) {
1602 ret = utils_truncate_stream_file(stream->out_fd, 0);
1603 if (ret < 0) {
1604 ERR("Reset metadata file");
1605 goto end;
1606 }
1607 stream->reset_metadata_flag = 0;
1608 }
1609
1624d5b7
JD
1610 /*
1611 * Check if we need to change the tracefile before writing the packet.
1612 */
1613 if (stream->chan->tracefile_size > 0 &&
1614 (stream->tracefile_size_current + len) >
1615 stream->chan->tracefile_size) {
fe4477ee
JD
1616 ret = utils_rotate_stream_file(stream->chan->pathname,
1617 stream->name, stream->chan->tracefile_size,
1618 stream->chan->tracefile_count, stream->uid, stream->gid,
309167d2
JD
1619 stream->out_fd, &(stream->tracefile_count_current),
1620 &stream->out_fd);
1624d5b7
JD
1621 if (ret < 0) {
1622 ERR("Rotating output file");
1623 goto end;
1624 }
309167d2
JD
1625 outfd = stream->out_fd;
1626
1627 if (stream->index_fd >= 0) {
adf7a0a1
JG
1628 ret = close(stream->index_fd);
1629 if (ret < 0) {
1630 PERROR("Closing index");
1631 goto end;
1632 }
1633 stream->index_fd = -1;
309167d2
JD
1634 ret = index_create_file(stream->chan->pathname,
1635 stream->name, stream->uid, stream->gid,
1636 stream->chan->tracefile_size,
1637 stream->tracefile_count_current);
1638 if (ret < 0) {
1639 goto end;
1640 }
1641 stream->index_fd = ret;
1642 }
1643
a6976990
DG
1644 /* Reset current size because we just perform a rotation. */
1645 stream->tracefile_size_current = 0;
a1ae300f
JD
1646 stream->out_fd_offset = 0;
1647 orig_offset = 0;
1624d5b7
JD
1648 }
1649 stream->tracefile_size_current += len;
309167d2
JD
1650 if (index) {
1651 index->offset = htobe64(stream->out_fd_offset);
1652 }
f02e1e8a
DG
1653 }
1654
d02b8372
DG
1655 /*
1656 * This call guarantee that len or less is returned. It's impossible to
1657 * receive a ret value that is bigger than len.
1658 */
1659 ret = lttng_write(outfd, mmap_base + mmap_offset, len);
1660 DBG("Consumer mmap write() ret %zd (len %lu)", ret, len);
1661 if (ret < 0 || ((size_t) ret != len)) {
1662 /*
1663 * Report error to caller if nothing was written else at least send the
1664 * amount written.
1665 */
1666 if (ret < 0) {
994ab360 1667 ret = -errno;
f02e1e8a 1668 }
994ab360 1669 relayd_hang_up = 1;
f02e1e8a 1670
d02b8372 1671 /* Socket operation failed. We consider the relayd dead */
994ab360 1672 if (errno == EPIPE || errno == EINVAL || errno == EBADF) {
d02b8372
DG
1673 /*
1674 * This is possible if the fd is closed on the other side
1675 * (outfd) or any write problem. It can be verbose a bit for a
1676 * normal execution if for instance the relayd is stopped
1677 * abruptly. This can happen so set this to a DBG statement.
1678 */
1679 DBG("Consumer mmap write detected relayd hang up");
994ab360
DG
1680 } else {
1681 /* Unhandled error, print it and stop function right now. */
1682 PERROR("Error in write mmap (ret %zd != len %lu)", ret, len);
f02e1e8a 1683 }
994ab360 1684 goto write_error;
d02b8372
DG
1685 }
1686 stream->output_written += ret;
d02b8372
DG
1687
1688 /* This call is useless on a socket so better save a syscall. */
1689 if (!relayd) {
1690 /* This won't block, but will start writeout asynchronously */
1691 lttng_sync_file_range(outfd, stream->out_fd_offset, len,
1692 SYNC_FILE_RANGE_WRITE);
1693 stream->out_fd_offset += len;
f5dbe415 1694 lttng_consumer_sync_trace_file(stream, orig_offset);
f02e1e8a 1695 }
f02e1e8a 1696
8994307f
DG
1697write_error:
1698 /*
1699 * This is a special case that the relayd has closed its socket. Let's
1700 * cleanup the relayd object and all associated streams.
1701 */
1702 if (relayd && relayd_hang_up) {
1703 cleanup_relayd(relayd, ctx);
1704 }
1705
f02e1e8a
DG
1706end:
1707 /* Unlock only if ctrl socket used */
1708 if (relayd && stream->metadata_flag) {
1709 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1710 }
1711
1712 rcu_read_unlock();
994ab360 1713 return ret;
3bd1e081
MD
1714}
1715
1716/*
1717 * Splice the data from the ring buffer to the tracefile.
1718 *
79d4ffb7
DG
1719 * It must be called with the stream lock held.
1720 *
3bd1e081
MD
1721 * Returns the number of bytes spliced.
1722 */
4078b776 1723ssize_t lttng_consumer_on_read_subbuffer_splice(
3bd1e081 1724 struct lttng_consumer_local_data *ctx,
1d4dfdef 1725 struct lttng_consumer_stream *stream, unsigned long len,
309167d2 1726 unsigned long padding,
50adc264 1727 struct ctf_packet_index *index)
3bd1e081 1728{
f02e1e8a
DG
1729 ssize_t ret = 0, written = 0, ret_splice = 0;
1730 loff_t offset = 0;
1731 off_t orig_offset = stream->out_fd_offset;
1732 int fd = stream->wait_fd;
1733 /* Default is on the disk */
1734 int outfd = stream->out_fd;
f02e1e8a 1735 struct consumer_relayd_sock_pair *relayd = NULL;
fb3a43a9 1736 int *splice_pipe;
8994307f 1737 unsigned int relayd_hang_up = 0;
f02e1e8a 1738
3bd1e081
MD
1739 switch (consumer_data.type) {
1740 case LTTNG_CONSUMER_KERNEL:
f02e1e8a 1741 break;
7753dea8
MD
1742 case LTTNG_CONSUMER32_UST:
1743 case LTTNG_CONSUMER64_UST:
f02e1e8a 1744 /* Not supported for user space tracing */
3bd1e081
MD
1745 return -ENOSYS;
1746 default:
1747 ERR("Unknown consumer_data type");
1748 assert(0);
3bd1e081
MD
1749 }
1750
f02e1e8a
DG
1751 /* RCU lock for the relayd pointer */
1752 rcu_read_lock();
1753
1754 /* Flag that the current stream if set for network streaming. */
da009f2c 1755 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1756 relayd = consumer_find_relayd(stream->net_seq_idx);
1757 if (relayd == NULL) {
ad0b0d23 1758 written = -ret;
f02e1e8a
DG
1759 goto end;
1760 }
1761 }
a2361a61 1762 splice_pipe = stream->splice_pipe;
fb3a43a9 1763
f02e1e8a 1764 /* Write metadata stream id before payload */
1d4dfdef 1765 if (relayd) {
ad0b0d23 1766 unsigned long total_len = len;
f02e1e8a 1767
1d4dfdef
DG
1768 if (stream->metadata_flag) {
1769 /*
1770 * Lock the control socket for the complete duration of the function
1771 * since from this point on we will use the socket.
1772 */
1773 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1774
93ec662e
JD
1775 if (stream->reset_metadata_flag) {
1776 ret = relayd_reset_metadata(&relayd->control_sock,
1777 stream->relayd_stream_id,
1778 stream->metadata_version);
1779 if (ret < 0) {
1780 relayd_hang_up = 1;
1781 goto write_error;
1782 }
1783 stream->reset_metadata_flag = 0;
1784 }
1d4dfdef
DG
1785 ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd,
1786 padding);
1787 if (ret < 0) {
1788 written = ret;
ad0b0d23
DG
1789 relayd_hang_up = 1;
1790 goto write_error;
1d4dfdef
DG
1791 }
1792
1793 total_len += sizeof(struct lttcomm_relayd_metadata_payload);
1794 }
1795
1796 ret = write_relayd_stream_header(stream, total_len, padding, relayd);
ad0b0d23
DG
1797 if (ret < 0) {
1798 written = ret;
1799 relayd_hang_up = 1;
1800 goto write_error;
f02e1e8a 1801 }
ad0b0d23
DG
1802 /* Use the returned socket. */
1803 outfd = ret;
1d4dfdef
DG
1804 } else {
1805 /* No streaming, we have to set the len with the full padding */
1806 len += padding;
1624d5b7 1807
93ec662e
JD
1808 if (stream->metadata_flag && stream->reset_metadata_flag) {
1809 ret = utils_truncate_stream_file(stream->out_fd, 0);
1810 if (ret < 0) {
1811 ERR("Reset metadata file");
1812 goto end;
1813 }
1814 stream->reset_metadata_flag = 0;
1815 }
1624d5b7
JD
1816 /*
1817 * Check if we need to change the tracefile before writing the packet.
1818 */
1819 if (stream->chan->tracefile_size > 0 &&
1820 (stream->tracefile_size_current + len) >
1821 stream->chan->tracefile_size) {
fe4477ee
JD
1822 ret = utils_rotate_stream_file(stream->chan->pathname,
1823 stream->name, stream->chan->tracefile_size,
1824 stream->chan->tracefile_count, stream->uid, stream->gid,
309167d2
JD
1825 stream->out_fd, &(stream->tracefile_count_current),
1826 &stream->out_fd);
1624d5b7 1827 if (ret < 0) {
ad0b0d23 1828 written = ret;
1624d5b7
JD
1829 ERR("Rotating output file");
1830 goto end;
1831 }
309167d2
JD
1832 outfd = stream->out_fd;
1833
1834 if (stream->index_fd >= 0) {
3c1150e1
JD
1835 ret = close(stream->index_fd);
1836 if (ret < 0) {
1837 PERROR("Closing index");
1838 goto end;
1839 }
1840 stream->index_fd = -1;
309167d2
JD
1841 ret = index_create_file(stream->chan->pathname,
1842 stream->name, stream->uid, stream->gid,
1843 stream->chan->tracefile_size,
1844 stream->tracefile_count_current);
1845 if (ret < 0) {
ad0b0d23 1846 written = ret;
309167d2
JD
1847 goto end;
1848 }
1849 stream->index_fd = ret;
1850 }
1851
a6976990
DG
1852 /* Reset current size because we just perform a rotation. */
1853 stream->tracefile_size_current = 0;
a1ae300f
JD
1854 stream->out_fd_offset = 0;
1855 orig_offset = 0;
1624d5b7
JD
1856 }
1857 stream->tracefile_size_current += len;
309167d2 1858 index->offset = htobe64(stream->out_fd_offset);
f02e1e8a
DG
1859 }
1860
1861 while (len > 0) {
1d4dfdef
DG
1862 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1863 (unsigned long)offset, len, fd, splice_pipe[1]);
fb3a43a9 1864 ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len,
f02e1e8a
DG
1865 SPLICE_F_MOVE | SPLICE_F_MORE);
1866 DBG("splice chan to pipe, ret %zd", ret_splice);
1867 if (ret_splice < 0) {
d02b8372 1868 ret = errno;
ad0b0d23 1869 written = -ret;
d02b8372 1870 PERROR("Error in relay splice");
f02e1e8a
DG
1871 goto splice_error;
1872 }
1873
1874 /* Handle stream on the relayd if the output is on the network */
ad0b0d23
DG
1875 if (relayd && stream->metadata_flag) {
1876 size_t metadata_payload_size =
1877 sizeof(struct lttcomm_relayd_metadata_payload);
1878
1879 /* Update counter to fit the spliced data */
1880 ret_splice += metadata_payload_size;
1881 len += metadata_payload_size;
1882 /*
1883 * We do this so the return value can match the len passed as
1884 * argument to this function.
1885 */
1886 written -= metadata_payload_size;
f02e1e8a
DG
1887 }
1888
1889 /* Splice data out */
fb3a43a9 1890 ret_splice = splice(splice_pipe[0], NULL, outfd, NULL,
f02e1e8a 1891 ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE);
a2361a61
JD
1892 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1893 outfd, ret_splice);
f02e1e8a 1894 if (ret_splice < 0) {
d02b8372 1895 ret = errno;
ad0b0d23
DG
1896 written = -ret;
1897 relayd_hang_up = 1;
1898 goto write_error;
f02e1e8a 1899 } else if (ret_splice > len) {
d02b8372
DG
1900 /*
1901 * We don't expect this code path to be executed but you never know
1902 * so this is an extra protection agains a buggy splice().
1903 */
f02e1e8a 1904 ret = errno;
ad0b0d23 1905 written += ret_splice;
d02b8372
DG
1906 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice,
1907 len);
f02e1e8a 1908 goto splice_error;
d02b8372
DG
1909 } else {
1910 /* All good, update current len and continue. */
1911 len -= ret_splice;
f02e1e8a 1912 }
f02e1e8a
DG
1913
1914 /* This call is useless on a socket so better save a syscall. */
1915 if (!relayd) {
1916 /* This won't block, but will start writeout asynchronously */
1917 lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice,
1918 SYNC_FILE_RANGE_WRITE);
1919 stream->out_fd_offset += ret_splice;
1920 }
e5d1a9b3 1921 stream->output_written += ret_splice;
f02e1e8a
DG
1922 written += ret_splice;
1923 }
f5dbe415
JG
1924 if (!relayd) {
1925 lttng_consumer_sync_trace_file(stream, orig_offset);
1926 }
f02e1e8a
DG
1927 goto end;
1928
8994307f
DG
1929write_error:
1930 /*
1931 * This is a special case that the relayd has closed its socket. Let's
1932 * cleanup the relayd object and all associated streams.
1933 */
1934 if (relayd && relayd_hang_up) {
1935 cleanup_relayd(relayd, ctx);
1936 /* Skip splice error so the consumer does not fail */
1937 goto end;
1938 }
1939
f02e1e8a
DG
1940splice_error:
1941 /* send the appropriate error description to sessiond */
1942 switch (ret) {
f02e1e8a 1943 case EINVAL:
f73fabfd 1944 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL);
f02e1e8a
DG
1945 break;
1946 case ENOMEM:
f73fabfd 1947 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM);
f02e1e8a
DG
1948 break;
1949 case ESPIPE:
f73fabfd 1950 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE);
f02e1e8a
DG
1951 break;
1952 }
1953
1954end:
1955 if (relayd && stream->metadata_flag) {
1956 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1957 }
1958
1959 rcu_read_unlock();
1960 return written;
3bd1e081
MD
1961}
1962
1963/*
1964 * Take a snapshot for a specific fd
1965 *
1966 * Returns 0 on success, < 0 on error
1967 */
ffe60014 1968int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream)
3bd1e081
MD
1969{
1970 switch (consumer_data.type) {
1971 case LTTNG_CONSUMER_KERNEL:
ffe60014 1972 return lttng_kconsumer_take_snapshot(stream);
7753dea8
MD
1973 case LTTNG_CONSUMER32_UST:
1974 case LTTNG_CONSUMER64_UST:
ffe60014 1975 return lttng_ustconsumer_take_snapshot(stream);
3bd1e081
MD
1976 default:
1977 ERR("Unknown consumer_data type");
1978 assert(0);
1979 return -ENOSYS;
1980 }
3bd1e081
MD
1981}
1982
1983/*
1984 * Get the produced position
1985 *
1986 * Returns 0 on success, < 0 on error
1987 */
ffe60014 1988int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
3bd1e081
MD
1989 unsigned long *pos)
1990{
1991 switch (consumer_data.type) {
1992 case LTTNG_CONSUMER_KERNEL:
ffe60014 1993 return lttng_kconsumer_get_produced_snapshot(stream, pos);
7753dea8
MD
1994 case LTTNG_CONSUMER32_UST:
1995 case LTTNG_CONSUMER64_UST:
ffe60014 1996 return lttng_ustconsumer_get_produced_snapshot(stream, pos);
3bd1e081
MD
1997 default:
1998 ERR("Unknown consumer_data type");
1999 assert(0);
2000 return -ENOSYS;
2001 }
2002}
2003
2004int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx,
2005 int sock, struct pollfd *consumer_sockpoll)
2006{
2007 switch (consumer_data.type) {
2008 case LTTNG_CONSUMER_KERNEL:
2009 return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
7753dea8
MD
2010 case LTTNG_CONSUMER32_UST:
2011 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
2012 return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
2013 default:
2014 ERR("Unknown consumer_data type");
2015 assert(0);
2016 return -ENOSYS;
2017 }
2018}
2019
6d574024 2020void lttng_consumer_close_all_metadata(void)
d88aee68
DG
2021{
2022 switch (consumer_data.type) {
2023 case LTTNG_CONSUMER_KERNEL:
2024 /*
2025 * The Kernel consumer has a different metadata scheme so we don't
2026 * close anything because the stream will be closed by the session
2027 * daemon.
2028 */
2029 break;
2030 case LTTNG_CONSUMER32_UST:
2031 case LTTNG_CONSUMER64_UST:
2032 /*
2033 * Close all metadata streams. The metadata hash table is passed and
2034 * this call iterates over it by closing all wakeup fd. This is safe
2035 * because at this point we are sure that the metadata producer is
2036 * either dead or blocked.
2037 */
6d574024 2038 lttng_ustconsumer_close_all_metadata(metadata_ht);
d88aee68
DG
2039 break;
2040 default:
2041 ERR("Unknown consumer_data type");
2042 assert(0);
2043 }
2044}
2045
fb3a43a9
DG
2046/*
2047 * Clean up a metadata stream and free its memory.
2048 */
e316aad5
DG
2049void consumer_del_metadata_stream(struct lttng_consumer_stream *stream,
2050 struct lttng_ht *ht)
fb3a43a9 2051{
e316aad5 2052 struct lttng_consumer_channel *free_chan = NULL;
fb3a43a9
DG
2053
2054 assert(stream);
2055 /*
2056 * This call should NEVER receive regular stream. It must always be
2057 * metadata stream and this is crucial for data structure synchronization.
2058 */
2059 assert(stream->metadata_flag);
2060
e316aad5
DG
2061 DBG3("Consumer delete metadata stream %d", stream->wait_fd);
2062
74251bb8 2063 pthread_mutex_lock(&consumer_data.lock);
a9838785 2064 pthread_mutex_lock(&stream->chan->lock);
8994307f
DG
2065 pthread_mutex_lock(&stream->lock);
2066
6d574024
DG
2067 /* Remove any reference to that stream. */
2068 consumer_stream_delete(stream, ht);
ca22feea 2069
6d574024
DG
2070 /* Close down everything including the relayd if one. */
2071 consumer_stream_close(stream);
2072 /* Destroy tracer buffers of the stream. */
2073 consumer_stream_destroy_buffers(stream);
fb3a43a9
DG
2074
2075 /* Atomically decrement channel refcount since other threads can use it. */
f2ad556d 2076 if (!uatomic_sub_return(&stream->chan->refcount, 1)
ffe60014 2077 && !uatomic_read(&stream->chan->nb_init_stream_left)) {
c30aaa51 2078 /* Go for channel deletion! */
e316aad5 2079 free_chan = stream->chan;
fb3a43a9
DG
2080 }
2081
73811ecc
DG
2082 /*
2083 * Nullify the stream reference so it is not used after deletion. The
6d574024
DG
2084 * channel lock MUST be acquired before being able to check for a NULL
2085 * pointer value.
73811ecc
DG
2086 */
2087 stream->chan->metadata_stream = NULL;
2088
8994307f 2089 pthread_mutex_unlock(&stream->lock);
a9838785 2090 pthread_mutex_unlock(&stream->chan->lock);
74251bb8 2091 pthread_mutex_unlock(&consumer_data.lock);
e316aad5
DG
2092
2093 if (free_chan) {
2094 consumer_del_channel(free_chan);
2095 }
2096
6d574024 2097 consumer_stream_free(stream);
fb3a43a9
DG
2098}
2099
2100/*
2101 * Action done with the metadata stream when adding it to the consumer internal
2102 * data structures to handle it.
2103 */
5ab66908 2104int consumer_add_metadata_stream(struct lttng_consumer_stream *stream)
fb3a43a9 2105{
5ab66908 2106 struct lttng_ht *ht = metadata_ht;
e316aad5 2107 int ret = 0;
76082088 2108 struct lttng_ht_iter iter;
d88aee68 2109 struct lttng_ht_node_u64 *node;
fb3a43a9 2110
e316aad5
DG
2111 assert(stream);
2112 assert(ht);
2113
d88aee68 2114 DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key);
e316aad5
DG
2115
2116 pthread_mutex_lock(&consumer_data.lock);
a9838785 2117 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 2118 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 2119 pthread_mutex_lock(&stream->lock);
e316aad5 2120
e316aad5
DG
2121 /*
2122 * From here, refcounts are updated so be _careful_ when returning an error
2123 * after this point.
2124 */
2125
fb3a43a9 2126 rcu_read_lock();
76082088
DG
2127
2128 /*
2129 * Lookup the stream just to make sure it does not exist in our internal
2130 * state. This should NEVER happen.
2131 */
d88aee68
DG
2132 lttng_ht_lookup(ht, &stream->key, &iter);
2133 node = lttng_ht_iter_get_node_u64(&iter);
76082088
DG
2134 assert(!node);
2135
e316aad5 2136 /*
ffe60014
DG
2137 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2138 * in terms of destroying the associated channel, because the action that
e316aad5
DG
2139 * causes the count to become 0 also causes a stream to be added. The
2140 * channel deletion will thus be triggered by the following removal of this
2141 * stream.
2142 */
ffe60014 2143 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
2144 /* Increment refcount before decrementing nb_init_stream_left */
2145 cmm_smp_wmb();
ffe60014 2146 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
2147 }
2148
d88aee68 2149 lttng_ht_add_unique_u64(ht, &stream->node);
ca22feea 2150
d8ef542d
MD
2151 lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht,
2152 &stream->node_channel_id);
2153
ca22feea
DG
2154 /*
2155 * Add stream to the stream_list_ht of the consumer data. No need to steal
2156 * the key since the HT does not use it and we allow to add redundant keys
2157 * into this table.
2158 */
d88aee68 2159 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 2160
fb3a43a9 2161 rcu_read_unlock();
e316aad5 2162
2e818a6a 2163 pthread_mutex_unlock(&stream->lock);
a9838785 2164 pthread_mutex_unlock(&stream->chan->lock);
ec6ea7d0 2165 pthread_mutex_unlock(&stream->chan->timer_lock);
e316aad5
DG
2166 pthread_mutex_unlock(&consumer_data.lock);
2167 return ret;
fb3a43a9
DG
2168}
2169
8994307f
DG
2170/*
2171 * Delete data stream that are flagged for deletion (endpoint_status).
2172 */
2173static void validate_endpoint_status_data_stream(void)
2174{
2175 struct lttng_ht_iter iter;
2176 struct lttng_consumer_stream *stream;
2177
2178 DBG("Consumer delete flagged data stream");
2179
2180 rcu_read_lock();
2181 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
2182 /* Validate delete flag of the stream */
79d4ffb7 2183 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2184 continue;
2185 }
2186 /* Delete it right now */
2187 consumer_del_stream(stream, data_ht);
2188 }
2189 rcu_read_unlock();
2190}
2191
2192/*
2193 * Delete metadata stream that are flagged for deletion (endpoint_status).
2194 */
2195static void validate_endpoint_status_metadata_stream(
2196 struct lttng_poll_event *pollset)
2197{
2198 struct lttng_ht_iter iter;
2199 struct lttng_consumer_stream *stream;
2200
2201 DBG("Consumer delete flagged metadata stream");
2202
2203 assert(pollset);
2204
2205 rcu_read_lock();
2206 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
2207 /* Validate delete flag of the stream */
79d4ffb7 2208 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2209 continue;
2210 }
2211 /*
2212 * Remove from pollset so the metadata thread can continue without
2213 * blocking on a deleted stream.
2214 */
2215 lttng_poll_del(pollset, stream->wait_fd);
2216
2217 /* Delete it right now */
2218 consumer_del_metadata_stream(stream, metadata_ht);
2219 }
2220 rcu_read_unlock();
2221}
2222
fb3a43a9
DG
2223/*
2224 * Thread polls on metadata file descriptor and write them on disk or on the
2225 * network.
2226 */
7d980def 2227void *consumer_thread_metadata_poll(void *data)
fb3a43a9 2228{
1fc79fb4 2229 int ret, i, pollfd, err = -1;
fb3a43a9 2230 uint32_t revents, nb_fd;
e316aad5 2231 struct lttng_consumer_stream *stream = NULL;
fb3a43a9 2232 struct lttng_ht_iter iter;
d88aee68 2233 struct lttng_ht_node_u64 *node;
fb3a43a9
DG
2234 struct lttng_poll_event events;
2235 struct lttng_consumer_local_data *ctx = data;
2236 ssize_t len;
2237
2238 rcu_register_thread();
2239
1fc79fb4
MD
2240 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA);
2241
2d57de81
MD
2242 if (testpoint(consumerd_thread_metadata)) {
2243 goto error_testpoint;
2244 }
2245
9ce5646a
MD
2246 health_code_update();
2247
fb3a43a9
DG
2248 DBG("Thread metadata poll started");
2249
fb3a43a9
DG
2250 /* Size is set to 1 for the consumer_metadata pipe */
2251 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2252 if (ret < 0) {
2253 ERR("Poll set creation failed");
d8ef542d 2254 goto end_poll;
fb3a43a9
DG
2255 }
2256
13886d2d
DG
2257 ret = lttng_poll_add(&events,
2258 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN);
fb3a43a9
DG
2259 if (ret < 0) {
2260 goto end;
2261 }
2262
2263 /* Main loop */
2264 DBG("Metadata main loop started");
2265
2266 while (1) {
fb3a43a9 2267restart:
7fa2082e 2268 health_code_update();
9ce5646a 2269 health_poll_entry();
7fa2082e 2270 DBG("Metadata poll wait");
fb3a43a9 2271 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2272 DBG("Metadata poll return from wait with %d fd(s)",
2273 LTTNG_POLL_GETNB(&events));
9ce5646a 2274 health_poll_exit();
40063ead 2275 DBG("Metadata event caught in thread");
fb3a43a9
DG
2276 if (ret < 0) {
2277 if (errno == EINTR) {
40063ead 2278 ERR("Poll EINTR caught");
fb3a43a9
DG
2279 goto restart;
2280 }
d9607cd7
MD
2281 if (LTTNG_POLL_GETNB(&events) == 0) {
2282 err = 0; /* All is OK */
2283 }
2284 goto end;
fb3a43a9
DG
2285 }
2286
0d9c5d77
DG
2287 nb_fd = ret;
2288
e316aad5 2289 /* From here, the event is a metadata wait fd */
fb3a43a9 2290 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2291 health_code_update();
2292
fb3a43a9
DG
2293 revents = LTTNG_POLL_GETEV(&events, i);
2294 pollfd = LTTNG_POLL_GETFD(&events, i);
2295
fd20dac9
MD
2296 if (!revents) {
2297 /* No activity for this FD (poll implementation). */
2298 continue;
2299 }
2300
13886d2d 2301 if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) {
03e43155 2302 if (revents & LPOLLIN) {
13886d2d
DG
2303 ssize_t pipe_len;
2304
2305 pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe,
2306 &stream, sizeof(stream));
6cd525e8 2307 if (pipe_len < sizeof(stream)) {
03e43155
MD
2308 if (pipe_len < 0) {
2309 PERROR("read metadata stream");
2310 }
fb3a43a9 2311 /*
03e43155
MD
2312 * Remove the pipe from the poll set and continue the loop
2313 * since their might be data to consume.
fb3a43a9 2314 */
03e43155
MD
2315 lttng_poll_del(&events,
2316 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2317 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
fb3a43a9
DG
2318 continue;
2319 }
2320
8994307f
DG
2321 /* A NULL stream means that the state has changed. */
2322 if (stream == NULL) {
2323 /* Check for deleted streams. */
2324 validate_endpoint_status_metadata_stream(&events);
3714380f 2325 goto restart;
8994307f
DG
2326 }
2327
fb3a43a9
DG
2328 DBG("Adding metadata stream %d to poll set",
2329 stream->wait_fd);
2330
fb3a43a9
DG
2331 /* Add metadata stream to the global poll events list */
2332 lttng_poll_add(&events, stream->wait_fd,
6d574024 2333 LPOLLIN | LPOLLPRI | LPOLLHUP);
03e43155
MD
2334 } else if (revents & (LPOLLERR | LPOLLHUP)) {
2335 DBG("Metadata thread pipe hung up");
2336 /*
2337 * Remove the pipe from the poll set and continue the loop
2338 * since their might be data to consume.
2339 */
2340 lttng_poll_del(&events,
2341 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2342 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
2343 continue;
2344 } else {
2345 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2346 goto end;
fb3a43a9
DG
2347 }
2348
e316aad5 2349 /* Handle other stream */
fb3a43a9
DG
2350 continue;
2351 }
2352
d09e1200 2353 rcu_read_lock();
d88aee68
DG
2354 {
2355 uint64_t tmp_id = (uint64_t) pollfd;
2356
2357 lttng_ht_lookup(metadata_ht, &tmp_id, &iter);
2358 }
2359 node = lttng_ht_iter_get_node_u64(&iter);
e316aad5 2360 assert(node);
fb3a43a9
DG
2361
2362 stream = caa_container_of(node, struct lttng_consumer_stream,
58b1f425 2363 node);
fb3a43a9 2364
03e43155
MD
2365 if (revents & (LPOLLIN | LPOLLPRI)) {
2366 /* Get the data out of the metadata file descriptor */
2367 DBG("Metadata available on fd %d", pollfd);
2368 assert(stream->wait_fd == pollfd);
2369
2370 do {
2371 health_code_update();
2372
2373 len = ctx->on_buffer_ready(stream, ctx);
2374 /*
2375 * We don't check the return value here since if we get
83f4233d 2376 * a negative len, it means an error occurred thus we
03e43155
MD
2377 * simply remove it from the poll set and free the
2378 * stream.
2379 */
2380 } while (len > 0);
2381
2382 /* It's ok to have an unavailable sub-buffer */
2383 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
2384 /* Clean up stream from consumer and free it. */
2385 lttng_poll_del(&events, stream->wait_fd);
2386 consumer_del_metadata_stream(stream, metadata_ht);
2387 }
2388 } else if (revents & (LPOLLERR | LPOLLHUP)) {
e316aad5 2389 DBG("Metadata fd %d is hup|err.", pollfd);
fb3a43a9
DG
2390 if (!stream->hangup_flush_done
2391 && (consumer_data.type == LTTNG_CONSUMER32_UST
2392 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2393 DBG("Attempting to flush and consume the UST buffers");
2394 lttng_ustconsumer_on_stream_hangup(stream);
2395
2396 /* We just flushed the stream now read it. */
4bb94b75 2397 do {
9ce5646a
MD
2398 health_code_update();
2399
4bb94b75
DG
2400 len = ctx->on_buffer_ready(stream, ctx);
2401 /*
2402 * We don't check the return value here since if we get
83f4233d 2403 * a negative len, it means an error occurred thus we
4bb94b75
DG
2404 * simply remove it from the poll set and free the
2405 * stream.
2406 */
2407 } while (len > 0);
fb3a43a9
DG
2408 }
2409
fb3a43a9 2410 lttng_poll_del(&events, stream->wait_fd);
e316aad5
DG
2411 /*
2412 * This call update the channel states, closes file descriptors
2413 * and securely free the stream.
2414 */
2415 consumer_del_metadata_stream(stream, metadata_ht);
03e43155
MD
2416 } else {
2417 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
6f2f1a70 2418 rcu_read_unlock();
03e43155 2419 goto end;
fb3a43a9 2420 }
e316aad5 2421 /* Release RCU lock for the stream looked up */
d09e1200 2422 rcu_read_unlock();
fb3a43a9
DG
2423 }
2424 }
2425
1fc79fb4
MD
2426 /* All is OK */
2427 err = 0;
fb3a43a9
DG
2428end:
2429 DBG("Metadata poll thread exiting");
fb3a43a9 2430
d8ef542d
MD
2431 lttng_poll_clean(&events);
2432end_poll:
2d57de81 2433error_testpoint:
1fc79fb4
MD
2434 if (err) {
2435 health_error();
2436 ERR("Health error occurred in %s", __func__);
2437 }
2438 health_unregister(health_consumerd);
fb3a43a9
DG
2439 rcu_unregister_thread();
2440 return NULL;
2441}
2442
3bd1e081 2443/*
e4421fec 2444 * This thread polls the fds in the set to consume the data and write
3bd1e081
MD
2445 * it to tracefile if necessary.
2446 */
7d980def 2447void *consumer_thread_data_poll(void *data)
3bd1e081 2448{
1fc79fb4 2449 int num_rdy, num_hup, high_prio, ret, i, err = -1;
3bd1e081
MD
2450 struct pollfd *pollfd = NULL;
2451 /* local view of the streams */
c869f647 2452 struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL;
3bd1e081
MD
2453 /* local view of consumer_data.fds_count */
2454 int nb_fd = 0;
3bd1e081 2455 struct lttng_consumer_local_data *ctx = data;
00e2e675 2456 ssize_t len;
3bd1e081 2457
e7b994a3
DG
2458 rcu_register_thread();
2459
1fc79fb4
MD
2460 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA);
2461
2d57de81
MD
2462 if (testpoint(consumerd_thread_data)) {
2463 goto error_testpoint;
2464 }
2465
9ce5646a
MD
2466 health_code_update();
2467
4df6c8cb
MD
2468 local_stream = zmalloc(sizeof(struct lttng_consumer_stream *));
2469 if (local_stream == NULL) {
2470 PERROR("local_stream malloc");
2471 goto end;
2472 }
3bd1e081
MD
2473
2474 while (1) {
9ce5646a
MD
2475 health_code_update();
2476
3bd1e081
MD
2477 high_prio = 0;
2478 num_hup = 0;
2479
2480 /*
e4421fec 2481 * the fds set has been updated, we need to update our
3bd1e081
MD
2482 * local array as well
2483 */
2484 pthread_mutex_lock(&consumer_data.lock);
2485 if (consumer_data.need_update) {
0e428499
DG
2486 free(pollfd);
2487 pollfd = NULL;
2488
2489 free(local_stream);
2490 local_stream = NULL;
3bd1e081 2491
02b3d176
DG
2492 /*
2493 * Allocate for all fds +1 for the consumer_data_pipe and +1 for
2494 * wake up pipe.
2495 */
2496 pollfd = zmalloc((consumer_data.stream_count + 2) * sizeof(struct pollfd));
3bd1e081 2497 if (pollfd == NULL) {
7a57cf92 2498 PERROR("pollfd malloc");
3bd1e081
MD
2499 pthread_mutex_unlock(&consumer_data.lock);
2500 goto end;
2501 }
2502
02b3d176 2503 local_stream = zmalloc((consumer_data.stream_count + 2) *
747f8642 2504 sizeof(struct lttng_consumer_stream *));
3bd1e081 2505 if (local_stream == NULL) {
7a57cf92 2506 PERROR("local_stream malloc");
3bd1e081
MD
2507 pthread_mutex_unlock(&consumer_data.lock);
2508 goto end;
2509 }
ffe60014 2510 ret = update_poll_array(ctx, &pollfd, local_stream,
43c34bc3 2511 data_ht);
3bd1e081
MD
2512 if (ret < 0) {
2513 ERR("Error in allocating pollfd or local_outfds");
f73fabfd 2514 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2515 pthread_mutex_unlock(&consumer_data.lock);
2516 goto end;
2517 }
2518 nb_fd = ret;
2519 consumer_data.need_update = 0;
2520 }
2521 pthread_mutex_unlock(&consumer_data.lock);
2522
4078b776
MD
2523 /* No FDs and consumer_quit, consumer_cleanup the thread */
2524 if (nb_fd == 0 && consumer_quit == 1) {
1fc79fb4 2525 err = 0; /* All is OK */
4078b776
MD
2526 goto end;
2527 }
3bd1e081 2528 /* poll on the array of fds */
88f2b785 2529 restart:
02b3d176 2530 DBG("polling on %d fd", nb_fd + 2);
9ce5646a 2531 health_poll_entry();
02b3d176 2532 num_rdy = poll(pollfd, nb_fd + 2, -1);
9ce5646a 2533 health_poll_exit();
3bd1e081
MD
2534 DBG("poll num_rdy : %d", num_rdy);
2535 if (num_rdy == -1) {
88f2b785
MD
2536 /*
2537 * Restart interrupted system call.
2538 */
2539 if (errno == EINTR) {
2540 goto restart;
2541 }
7a57cf92 2542 PERROR("Poll error");
f73fabfd 2543 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2544 goto end;
2545 } else if (num_rdy == 0) {
2546 DBG("Polling thread timed out");
2547 goto end;
2548 }
2549
3bd1e081 2550 /*
50f8ae69 2551 * If the consumer_data_pipe triggered poll go directly to the
00e2e675
DG
2552 * beginning of the loop to update the array. We want to prioritize
2553 * array update over low-priority reads.
3bd1e081 2554 */
509bb1cf 2555 if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) {
ab30f567 2556 ssize_t pipe_readlen;
04fdd819 2557
50f8ae69 2558 DBG("consumer_data_pipe wake up");
acdb9057
DG
2559 pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe,
2560 &new_stream, sizeof(new_stream));
6cd525e8
MD
2561 if (pipe_readlen < sizeof(new_stream)) {
2562 PERROR("Consumer data pipe");
23f5f35d
DG
2563 /* Continue so we can at least handle the current stream(s). */
2564 continue;
2565 }
c869f647
DG
2566
2567 /*
2568 * If the stream is NULL, just ignore it. It's also possible that
2569 * the sessiond poll thread changed the consumer_quit state and is
2570 * waking us up to test it.
2571 */
2572 if (new_stream == NULL) {
8994307f 2573 validate_endpoint_status_data_stream();
c869f647
DG
2574 continue;
2575 }
2576
c869f647 2577 /* Continue to update the local streams and handle prio ones */
3bd1e081
MD
2578 continue;
2579 }
2580
02b3d176
DG
2581 /* Handle wakeup pipe. */
2582 if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) {
2583 char dummy;
2584 ssize_t pipe_readlen;
2585
2586 pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy,
2587 sizeof(dummy));
2588 if (pipe_readlen < 0) {
2589 PERROR("Consumer data wakeup pipe");
2590 }
2591 /* We've been awakened to handle stream(s). */
2592 ctx->has_wakeup = 0;
2593 }
2594
3bd1e081
MD
2595 /* Take care of high priority channels first. */
2596 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2597 health_code_update();
2598
9617607b
DG
2599 if (local_stream[i] == NULL) {
2600 continue;
2601 }
fb3a43a9 2602 if (pollfd[i].revents & POLLPRI) {
d41f73b7
MD
2603 DBG("Urgent read on fd %d", pollfd[i].fd);
2604 high_prio = 1;
4078b776 2605 len = ctx->on_buffer_ready(local_stream[i], ctx);
d41f73b7 2606 /* it's ok to have an unavailable sub-buffer */
b64403e3 2607 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2608 /* Clean the stream and free it. */
2609 consumer_del_stream(local_stream[i], data_ht);
9617607b 2610 local_stream[i] = NULL;
4078b776
MD
2611 } else if (len > 0) {
2612 local_stream[i]->data_read = 1;
d41f73b7 2613 }
3bd1e081
MD
2614 }
2615 }
2616
4078b776
MD
2617 /*
2618 * If we read high prio channel in this loop, try again
2619 * for more high prio data.
2620 */
2621 if (high_prio) {
3bd1e081
MD
2622 continue;
2623 }
2624
2625 /* Take care of low priority channels. */
4078b776 2626 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2627 health_code_update();
2628
9617607b
DG
2629 if (local_stream[i] == NULL) {
2630 continue;
2631 }
4078b776 2632 if ((pollfd[i].revents & POLLIN) ||
02b3d176
DG
2633 local_stream[i]->hangup_flush_done ||
2634 local_stream[i]->has_data) {
4078b776
MD
2635 DBG("Normal read on fd %d", pollfd[i].fd);
2636 len = ctx->on_buffer_ready(local_stream[i], ctx);
2637 /* it's ok to have an unavailable sub-buffer */
b64403e3 2638 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2639 /* Clean the stream and free it. */
2640 consumer_del_stream(local_stream[i], data_ht);
9617607b 2641 local_stream[i] = NULL;
4078b776
MD
2642 } else if (len > 0) {
2643 local_stream[i]->data_read = 1;
2644 }
2645 }
2646 }
2647
2648 /* Handle hangup and errors */
2649 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2650 health_code_update();
2651
9617607b
DG
2652 if (local_stream[i] == NULL) {
2653 continue;
2654 }
4078b776
MD
2655 if (!local_stream[i]->hangup_flush_done
2656 && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL))
2657 && (consumer_data.type == LTTNG_CONSUMER32_UST
2658 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2659 DBG("fd %d is hup|err|nval. Attempting flush and read.",
9617607b 2660 pollfd[i].fd);
4078b776
MD
2661 lttng_ustconsumer_on_stream_hangup(local_stream[i]);
2662 /* Attempt read again, for the data we just flushed. */
2663 local_stream[i]->data_read = 1;
2664 }
2665 /*
2666 * If the poll flag is HUP/ERR/NVAL and we have
2667 * read no data in this pass, we can remove the
2668 * stream from its hash table.
2669 */
2670 if ((pollfd[i].revents & POLLHUP)) {
2671 DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
2672 if (!local_stream[i]->data_read) {
43c34bc3 2673 consumer_del_stream(local_stream[i], data_ht);
9617607b 2674 local_stream[i] = NULL;
4078b776
MD
2675 num_hup++;
2676 }
2677 } else if (pollfd[i].revents & POLLERR) {
2678 ERR("Error returned in polling fd %d.", pollfd[i].fd);
2679 if (!local_stream[i]->data_read) {
43c34bc3 2680 consumer_del_stream(local_stream[i], data_ht);
9617607b 2681 local_stream[i] = NULL;
4078b776
MD
2682 num_hup++;
2683 }
2684 } else if (pollfd[i].revents & POLLNVAL) {
2685 ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
2686 if (!local_stream[i]->data_read) {
43c34bc3 2687 consumer_del_stream(local_stream[i], data_ht);
9617607b 2688 local_stream[i] = NULL;
4078b776 2689 num_hup++;
3bd1e081
MD
2690 }
2691 }
9617607b
DG
2692 if (local_stream[i] != NULL) {
2693 local_stream[i]->data_read = 0;
2694 }
3bd1e081
MD
2695 }
2696 }
1fc79fb4
MD
2697 /* All is OK */
2698 err = 0;
3bd1e081
MD
2699end:
2700 DBG("polling thread exiting");
0e428499
DG
2701 free(pollfd);
2702 free(local_stream);
fb3a43a9
DG
2703
2704 /*
2705 * Close the write side of the pipe so epoll_wait() in
7d980def
DG
2706 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2707 * read side of the pipe. If we close them both, epoll_wait strangely does
2708 * not return and could create a endless wait period if the pipe is the
2709 * only tracked fd in the poll set. The thread will take care of closing
2710 * the read side.
fb3a43a9 2711 */
13886d2d 2712 (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe);
fb3a43a9 2713
2d57de81 2714error_testpoint:
1fc79fb4
MD
2715 if (err) {
2716 health_error();
2717 ERR("Health error occurred in %s", __func__);
2718 }
2719 health_unregister(health_consumerd);
2720
e7b994a3 2721 rcu_unregister_thread();
3bd1e081
MD
2722 return NULL;
2723}
2724
d8ef542d
MD
2725/*
2726 * Close wake-up end of each stream belonging to the channel. This will
2727 * allow the poll() on the stream read-side to detect when the
2728 * write-side (application) finally closes them.
2729 */
2730static
2731void consumer_close_channel_streams(struct lttng_consumer_channel *channel)
2732{
2733 struct lttng_ht *ht;
2734 struct lttng_consumer_stream *stream;
2735 struct lttng_ht_iter iter;
2736
2737 ht = consumer_data.stream_per_chan_id_ht;
2738
2739 rcu_read_lock();
2740 cds_lfht_for_each_entry_duplicate(ht->ht,
2741 ht->hash_fct(&channel->key, lttng_ht_seed),
2742 ht->match_fct, &channel->key,
2743 &iter.iter, stream, node_channel_id.node) {
f2ad556d
MD
2744 /*
2745 * Protect against teardown with mutex.
2746 */
2747 pthread_mutex_lock(&stream->lock);
2748 if (cds_lfht_is_node_deleted(&stream->node.node)) {
2749 goto next;
2750 }
d8ef542d
MD
2751 switch (consumer_data.type) {
2752 case LTTNG_CONSUMER_KERNEL:
2753 break;
2754 case LTTNG_CONSUMER32_UST:
2755 case LTTNG_CONSUMER64_UST:
b4a650f3
DG
2756 if (stream->metadata_flag) {
2757 /* Safe and protected by the stream lock. */
2758 lttng_ustconsumer_close_metadata(stream->chan);
2759 } else {
2760 /*
2761 * Note: a mutex is taken internally within
2762 * liblttng-ust-ctl to protect timer wakeup_fd
2763 * use from concurrent close.
2764 */
2765 lttng_ustconsumer_close_stream_wakeup(stream);
2766 }
d8ef542d
MD
2767 break;
2768 default:
2769 ERR("Unknown consumer_data type");
2770 assert(0);
2771 }
f2ad556d
MD
2772 next:
2773 pthread_mutex_unlock(&stream->lock);
d8ef542d
MD
2774 }
2775 rcu_read_unlock();
2776}
2777
2778static void destroy_channel_ht(struct lttng_ht *ht)
2779{
2780 struct lttng_ht_iter iter;
2781 struct lttng_consumer_channel *channel;
2782 int ret;
2783
2784 if (ht == NULL) {
2785 return;
2786 }
2787
2788 rcu_read_lock();
2789 cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) {
2790 ret = lttng_ht_del(ht, &iter);
2791 assert(ret != 0);
2792 }
2793 rcu_read_unlock();
2794
2795 lttng_ht_destroy(ht);
2796}
2797
2798/*
2799 * This thread polls the channel fds to detect when they are being
2800 * closed. It closes all related streams if the channel is detected as
2801 * closed. It is currently only used as a shim layer for UST because the
2802 * consumerd needs to keep the per-stream wakeup end of pipes open for
2803 * periodical flush.
2804 */
2805void *consumer_thread_channel_poll(void *data)
2806{
1fc79fb4 2807 int ret, i, pollfd, err = -1;
d8ef542d
MD
2808 uint32_t revents, nb_fd;
2809 struct lttng_consumer_channel *chan = NULL;
2810 struct lttng_ht_iter iter;
2811 struct lttng_ht_node_u64 *node;
2812 struct lttng_poll_event events;
2813 struct lttng_consumer_local_data *ctx = data;
2814 struct lttng_ht *channel_ht;
2815
2816 rcu_register_thread();
2817
1fc79fb4
MD
2818 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL);
2819
2d57de81
MD
2820 if (testpoint(consumerd_thread_channel)) {
2821 goto error_testpoint;
2822 }
2823
9ce5646a
MD
2824 health_code_update();
2825
d8ef542d
MD
2826 channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2827 if (!channel_ht) {
2828 /* ENOMEM at this point. Better to bail out. */
2829 goto end_ht;
2830 }
2831
2832 DBG("Thread channel poll started");
2833
2834 /* Size is set to 1 for the consumer_channel pipe */
2835 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2836 if (ret < 0) {
2837 ERR("Poll set creation failed");
2838 goto end_poll;
2839 }
2840
2841 ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN);
2842 if (ret < 0) {
2843 goto end;
2844 }
2845
2846 /* Main loop */
2847 DBG("Channel main loop started");
2848
2849 while (1) {
d8ef542d 2850restart:
7fa2082e
MD
2851 health_code_update();
2852 DBG("Channel poll wait");
9ce5646a 2853 health_poll_entry();
d8ef542d 2854 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2855 DBG("Channel poll return from wait with %d fd(s)",
2856 LTTNG_POLL_GETNB(&events));
9ce5646a 2857 health_poll_exit();
40063ead 2858 DBG("Channel event caught in thread");
d8ef542d
MD
2859 if (ret < 0) {
2860 if (errno == EINTR) {
40063ead 2861 ERR("Poll EINTR caught");
d8ef542d
MD
2862 goto restart;
2863 }
d9607cd7
MD
2864 if (LTTNG_POLL_GETNB(&events) == 0) {
2865 err = 0; /* All is OK */
2866 }
d8ef542d
MD
2867 goto end;
2868 }
2869
2870 nb_fd = ret;
2871
2872 /* From here, the event is a channel wait fd */
2873 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2874 health_code_update();
2875
d8ef542d
MD
2876 revents = LTTNG_POLL_GETEV(&events, i);
2877 pollfd = LTTNG_POLL_GETFD(&events, i);
2878
d8ef542d 2879 if (!revents) {
fd20dac9 2880 /* No activity for this FD (poll implementation). */
d8ef542d
MD
2881 continue;
2882 }
fd20dac9 2883
d8ef542d 2884 if (pollfd == ctx->consumer_channel_pipe[0]) {
03e43155 2885 if (revents & LPOLLIN) {
d8ef542d 2886 enum consumer_channel_action action;
a0cbdd2e 2887 uint64_t key;
d8ef542d 2888
a0cbdd2e 2889 ret = read_channel_pipe(ctx, &chan, &key, &action);
d8ef542d 2890 if (ret <= 0) {
03e43155
MD
2891 if (ret < 0) {
2892 ERR("Error reading channel pipe");
2893 }
2894 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
d8ef542d
MD
2895 continue;
2896 }
2897
2898 switch (action) {
2899 case CONSUMER_CHANNEL_ADD:
2900 DBG("Adding channel %d to poll set",
2901 chan->wait_fd);
2902
2903 lttng_ht_node_init_u64(&chan->wait_fd_node,
2904 chan->wait_fd);
c7260a81 2905 rcu_read_lock();
d8ef542d
MD
2906 lttng_ht_add_unique_u64(channel_ht,
2907 &chan->wait_fd_node);
c7260a81 2908 rcu_read_unlock();
d8ef542d
MD
2909 /* Add channel to the global poll events list */
2910 lttng_poll_add(&events, chan->wait_fd,
03e43155 2911 LPOLLERR | LPOLLHUP);
d8ef542d 2912 break;
a0cbdd2e
MD
2913 case CONSUMER_CHANNEL_DEL:
2914 {
b4a650f3
DG
2915 /*
2916 * This command should never be called if the channel
2917 * has streams monitored by either the data or metadata
2918 * thread. The consumer only notify this thread with a
2919 * channel del. command if it receives a destroy
2920 * channel command from the session daemon that send it
2921 * if a command prior to the GET_CHANNEL failed.
2922 */
2923
c7260a81 2924 rcu_read_lock();
a0cbdd2e
MD
2925 chan = consumer_find_channel(key);
2926 if (!chan) {
c7260a81 2927 rcu_read_unlock();
a0cbdd2e
MD
2928 ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key);
2929 break;
2930 }
2931 lttng_poll_del(&events, chan->wait_fd);
f623cc0b 2932 iter.iter.node = &chan->wait_fd_node.node;
a0cbdd2e
MD
2933 ret = lttng_ht_del(channel_ht, &iter);
2934 assert(ret == 0);
a0cbdd2e 2935
f2a444f1
DG
2936 switch (consumer_data.type) {
2937 case LTTNG_CONSUMER_KERNEL:
2938 break;
2939 case LTTNG_CONSUMER32_UST:
2940 case LTTNG_CONSUMER64_UST:
212d67a2
DG
2941 health_code_update();
2942 /* Destroy streams that might have been left in the stream list. */
2943 clean_channel_stream_list(chan);
f2a444f1
DG
2944 break;
2945 default:
2946 ERR("Unknown consumer_data type");
2947 assert(0);
2948 }
2949
a0cbdd2e
MD
2950 /*
2951 * Release our own refcount. Force channel deletion even if
2952 * streams were not initialized.
2953 */
2954 if (!uatomic_sub_return(&chan->refcount, 1)) {
2955 consumer_del_channel(chan);
2956 }
c7260a81 2957 rcu_read_unlock();
a0cbdd2e
MD
2958 goto restart;
2959 }
d8ef542d
MD
2960 case CONSUMER_CHANNEL_QUIT:
2961 /*
2962 * Remove the pipe from the poll set and continue the loop
2963 * since their might be data to consume.
2964 */
2965 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2966 continue;
2967 default:
2968 ERR("Unknown action");
2969 break;
2970 }
03e43155
MD
2971 } else if (revents & (LPOLLERR | LPOLLHUP)) {
2972 DBG("Channel thread pipe hung up");
2973 /*
2974 * Remove the pipe from the poll set and continue the loop
2975 * since their might be data to consume.
2976 */
2977 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2978 continue;
2979 } else {
2980 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
2981 goto end;
d8ef542d
MD
2982 }
2983
2984 /* Handle other stream */
2985 continue;
2986 }
2987
2988 rcu_read_lock();
2989 {
2990 uint64_t tmp_id = (uint64_t) pollfd;
2991
2992 lttng_ht_lookup(channel_ht, &tmp_id, &iter);
2993 }
2994 node = lttng_ht_iter_get_node_u64(&iter);
2995 assert(node);
2996
2997 chan = caa_container_of(node, struct lttng_consumer_channel,
2998 wait_fd_node);
2999
3000 /* Check for error event */
3001 if (revents & (LPOLLERR | LPOLLHUP)) {
3002 DBG("Channel fd %d is hup|err.", pollfd);
3003
3004 lttng_poll_del(&events, chan->wait_fd);
3005 ret = lttng_ht_del(channel_ht, &iter);
3006 assert(ret == 0);
b4a650f3
DG
3007
3008 /*
3009 * This will close the wait fd for each stream associated to
3010 * this channel AND monitored by the data/metadata thread thus
3011 * will be clean by the right thread.
3012 */
d8ef542d 3013 consumer_close_channel_streams(chan);
f2ad556d
MD
3014
3015 /* Release our own refcount */
3016 if (!uatomic_sub_return(&chan->refcount, 1)
3017 && !uatomic_read(&chan->nb_init_stream_left)) {
3018 consumer_del_channel(chan);
3019 }
03e43155
MD
3020 } else {
3021 ERR("Unexpected poll events %u for sock %d", revents, pollfd);
3022 rcu_read_unlock();
3023 goto end;
d8ef542d
MD
3024 }
3025
3026 /* Release RCU lock for the channel looked up */
3027 rcu_read_unlock();
3028 }
3029 }
3030
1fc79fb4
MD
3031 /* All is OK */
3032 err = 0;
d8ef542d
MD
3033end:
3034 lttng_poll_clean(&events);
3035end_poll:
3036 destroy_channel_ht(channel_ht);
3037end_ht:
2d57de81 3038error_testpoint:
d8ef542d 3039 DBG("Channel poll thread exiting");
1fc79fb4
MD
3040 if (err) {
3041 health_error();
3042 ERR("Health error occurred in %s", __func__);
3043 }
3044 health_unregister(health_consumerd);
d8ef542d
MD
3045 rcu_unregister_thread();
3046 return NULL;
3047}
3048
331744e3
JD
3049static int set_metadata_socket(struct lttng_consumer_local_data *ctx,
3050 struct pollfd *sockpoll, int client_socket)
3051{
3052 int ret;
3053
3054 assert(ctx);
3055 assert(sockpoll);
3056
84382d49
MD
3057 ret = lttng_consumer_poll_socket(sockpoll);
3058 if (ret) {
331744e3
JD
3059 goto error;
3060 }
3061 DBG("Metadata connection on client_socket");
3062
3063 /* Blocking call, waiting for transmission */
3064 ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket);
3065 if (ctx->consumer_metadata_socket < 0) {
3066 WARN("On accept metadata");
3067 ret = -1;
3068 goto error;
3069 }
3070 ret = 0;
3071
3072error:
3073 return ret;
3074}
3075
3bd1e081
MD
3076/*
3077 * This thread listens on the consumerd socket and receives the file
3078 * descriptors from the session daemon.
3079 */
7d980def 3080void *consumer_thread_sessiond_poll(void *data)
3bd1e081 3081{
1fc79fb4 3082 int sock = -1, client_socket, ret, err = -1;
3bd1e081
MD
3083 /*
3084 * structure to poll for incoming data on communication socket avoids
3085 * making blocking sockets.
3086 */
3087 struct pollfd consumer_sockpoll[2];
3088 struct lttng_consumer_local_data *ctx = data;
3089
e7b994a3
DG
3090 rcu_register_thread();
3091
1fc79fb4
MD
3092 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND);
3093
2d57de81
MD
3094 if (testpoint(consumerd_thread_sessiond)) {
3095 goto error_testpoint;
3096 }
3097
9ce5646a
MD
3098 health_code_update();
3099
3bd1e081
MD
3100 DBG("Creating command socket %s", ctx->consumer_command_sock_path);
3101 unlink(ctx->consumer_command_sock_path);
3102 client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path);
3103 if (client_socket < 0) {
3104 ERR("Cannot create command socket");
3105 goto end;
3106 }
3107
3108 ret = lttcomm_listen_unix_sock(client_socket);
3109 if (ret < 0) {
3110 goto end;
3111 }
3112
32258573 3113 DBG("Sending ready command to lttng-sessiond");
f73fabfd 3114 ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY);
3bd1e081
MD
3115 /* return < 0 on error, but == 0 is not fatal */
3116 if (ret < 0) {
32258573 3117 ERR("Error sending ready command to lttng-sessiond");
3bd1e081
MD
3118 goto end;
3119 }
3120
3bd1e081
MD
3121 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3122 consumer_sockpoll[0].fd = ctx->consumer_should_quit[0];
3123 consumer_sockpoll[0].events = POLLIN | POLLPRI;
3124 consumer_sockpoll[1].fd = client_socket;
3125 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3126
84382d49
MD
3127 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3128 if (ret) {
3129 if (ret > 0) {
3130 /* should exit */
3131 err = 0;
3132 }
3bd1e081
MD
3133 goto end;
3134 }
3135 DBG("Connection on client_socket");
3136
3137 /* Blocking call, waiting for transmission */
3138 sock = lttcomm_accept_unix_sock(client_socket);
534d2592 3139 if (sock < 0) {
3bd1e081
MD
3140 WARN("On accept");
3141 goto end;
3142 }
3bd1e081 3143
331744e3
JD
3144 /*
3145 * Setup metadata socket which is the second socket connection on the
3146 * command unix socket.
3147 */
3148 ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket);
84382d49
MD
3149 if (ret) {
3150 if (ret > 0) {
3151 /* should exit */
3152 err = 0;
3153 }
331744e3
JD
3154 goto end;
3155 }
3156
d96f09c6
DG
3157 /* This socket is not useful anymore. */
3158 ret = close(client_socket);
3159 if (ret < 0) {
3160 PERROR("close client_socket");
3161 }
3162 client_socket = -1;
3163
3bd1e081
MD
3164 /* update the polling structure to poll on the established socket */
3165 consumer_sockpoll[1].fd = sock;
3166 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3167
3168 while (1) {
9ce5646a
MD
3169 health_code_update();
3170
3171 health_poll_entry();
3172 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3173 health_poll_exit();
84382d49
MD
3174 if (ret) {
3175 if (ret > 0) {
3176 /* should exit */
3177 err = 0;
3178 }
3bd1e081
MD
3179 goto end;
3180 }
3181 DBG("Incoming command on sock");
3182 ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll);
4cbc1a04
DG
3183 if (ret <= 0) {
3184 /*
3185 * This could simply be a session daemon quitting. Don't output
3186 * ERR() here.
3187 */
3188 DBG("Communication interrupted on command socket");
41ba6035 3189 err = 0;
3bd1e081
MD
3190 goto end;
3191 }
3192 if (consumer_quit) {
3193 DBG("consumer_thread_receive_fds received quit from signal");
1fc79fb4 3194 err = 0; /* All is OK */
3bd1e081
MD
3195 goto end;
3196 }
ffe60014 3197 DBG("received command on sock");
3bd1e081 3198 }
1fc79fb4
MD
3199 /* All is OK */
3200 err = 0;
3201
3bd1e081 3202end:
ffe60014 3203 DBG("Consumer thread sessiond poll exiting");
3bd1e081 3204
d88aee68
DG
3205 /*
3206 * Close metadata streams since the producer is the session daemon which
3207 * just died.
3208 *
3209 * NOTE: for now, this only applies to the UST tracer.
3210 */
6d574024 3211 lttng_consumer_close_all_metadata();
d88aee68 3212
3bd1e081
MD
3213 /*
3214 * when all fds have hung up, the polling thread
3215 * can exit cleanly
3216 */
3217 consumer_quit = 1;
3218
04fdd819 3219 /*
c869f647 3220 * Notify the data poll thread to poll back again and test the
8994307f 3221 * consumer_quit state that we just set so to quit gracefully.
04fdd819 3222 */
acdb9057 3223 notify_thread_lttng_pipe(ctx->consumer_data_pipe);
c869f647 3224
a0cbdd2e 3225 notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT);
d8ef542d 3226
5c635c72
MD
3227 notify_health_quit_pipe(health_quit_pipe);
3228
d96f09c6
DG
3229 /* Cleaning up possibly open sockets. */
3230 if (sock >= 0) {
3231 ret = close(sock);
3232 if (ret < 0) {
3233 PERROR("close sock sessiond poll");
3234 }
3235 }
3236 if (client_socket >= 0) {
38476d24 3237 ret = close(client_socket);
d96f09c6
DG
3238 if (ret < 0) {
3239 PERROR("close client_socket sessiond poll");
3240 }
3241 }
3242
2d57de81 3243error_testpoint:
1fc79fb4
MD
3244 if (err) {
3245 health_error();
3246 ERR("Health error occurred in %s", __func__);
3247 }
3248 health_unregister(health_consumerd);
3249
e7b994a3 3250 rcu_unregister_thread();
3bd1e081
MD
3251 return NULL;
3252}
d41f73b7 3253
4078b776 3254ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream,
d41f73b7
MD
3255 struct lttng_consumer_local_data *ctx)
3256{
74251bb8
DG
3257 ssize_t ret;
3258
3259 pthread_mutex_lock(&stream->lock);
94d49140
JD
3260 if (stream->metadata_flag) {
3261 pthread_mutex_lock(&stream->metadata_rdv_lock);
3262 }
74251bb8 3263
d41f73b7
MD
3264 switch (consumer_data.type) {
3265 case LTTNG_CONSUMER_KERNEL:
74251bb8
DG
3266 ret = lttng_kconsumer_read_subbuffer(stream, ctx);
3267 break;
7753dea8
MD
3268 case LTTNG_CONSUMER32_UST:
3269 case LTTNG_CONSUMER64_UST:
74251bb8
DG
3270 ret = lttng_ustconsumer_read_subbuffer(stream, ctx);
3271 break;
d41f73b7
MD
3272 default:
3273 ERR("Unknown consumer_data type");
3274 assert(0);
74251bb8
DG
3275 ret = -ENOSYS;
3276 break;
d41f73b7 3277 }
74251bb8 3278
94d49140
JD
3279 if (stream->metadata_flag) {
3280 pthread_cond_broadcast(&stream->metadata_rdv);
3281 pthread_mutex_unlock(&stream->metadata_rdv_lock);
3282 }
74251bb8
DG
3283 pthread_mutex_unlock(&stream->lock);
3284 return ret;
d41f73b7
MD
3285}
3286
3287int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream)
3288{
3289 switch (consumer_data.type) {
3290 case LTTNG_CONSUMER_KERNEL:
3291 return lttng_kconsumer_on_recv_stream(stream);
7753dea8
MD
3292 case LTTNG_CONSUMER32_UST:
3293 case LTTNG_CONSUMER64_UST:
d41f73b7
MD
3294 return lttng_ustconsumer_on_recv_stream(stream);
3295 default:
3296 ERR("Unknown consumer_data type");
3297 assert(0);
3298 return -ENOSYS;
3299 }
3300}
e4421fec
DG
3301
3302/*
3303 * Allocate and set consumer data hash tables.
3304 */
282dadbc 3305int lttng_consumer_init(void)
e4421fec 3306{
d88aee68 3307 consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
282dadbc
MD
3308 if (!consumer_data.channel_ht) {
3309 goto error;
3310 }
3311
d88aee68 3312 consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
282dadbc
MD
3313 if (!consumer_data.relayd_ht) {
3314 goto error;
3315 }
3316
d88aee68 3317 consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
282dadbc
MD
3318 if (!consumer_data.stream_list_ht) {
3319 goto error;
3320 }
3321
d8ef542d 3322 consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
282dadbc
MD
3323 if (!consumer_data.stream_per_chan_id_ht) {
3324 goto error;
3325 }
3326
3327 data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3328 if (!data_ht) {
3329 goto error;
3330 }
3331
3332 metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3333 if (!metadata_ht) {
3334 goto error;
3335 }
3336
3337 return 0;
3338
3339error:
3340 return -1;
e4421fec 3341}
7735ef9e
DG
3342
3343/*
3344 * Process the ADD_RELAYD command receive by a consumer.
3345 *
3346 * This will create a relayd socket pair and add it to the relayd hash table.
3347 * The caller MUST acquire a RCU read side lock before calling it.
3348 */
da009f2c 3349int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type,
7735ef9e 3350 struct lttng_consumer_local_data *ctx, int sock,
6151a90f 3351 struct pollfd *consumer_sockpoll,
d3e2ba59
JD
3352 struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id,
3353 uint64_t relayd_session_id)
7735ef9e 3354{
cd2b09ed 3355 int fd = -1, ret = -1, relayd_created = 0;
0c759fc9 3356 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
d4298c99 3357 struct consumer_relayd_sock_pair *relayd = NULL;
7735ef9e 3358
6151a90f
JD
3359 assert(ctx);
3360 assert(relayd_sock);
3361
da009f2c 3362 DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx);
7735ef9e
DG
3363
3364 /* Get relayd reference if exists. */
3365 relayd = consumer_find_relayd(net_seq_idx);
3366 if (relayd == NULL) {
da009f2c 3367 assert(sock_type == LTTNG_STREAM_CONTROL);
7735ef9e
DG
3368 /* Not found. Allocate one. */
3369 relayd = consumer_allocate_relayd_sock_pair(net_seq_idx);
3370 if (relayd == NULL) {
0d08d75e 3371 ret = -ENOMEM;
618a6a28
MD
3372 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
3373 goto error;
0d08d75e 3374 } else {
30319bcb 3375 relayd->sessiond_session_id = sessiond_id;
0d08d75e 3376 relayd_created = 1;
7735ef9e 3377 }
0d08d75e
DG
3378
3379 /*
3380 * This code path MUST continue to the consumer send status message to
3381 * we can notify the session daemon and continue our work without
3382 * killing everything.
3383 */
da009f2c
MD
3384 } else {
3385 /*
3386 * relayd key should never be found for control socket.
3387 */
3388 assert(sock_type != LTTNG_STREAM_CONTROL);
0d08d75e
DG
3389 }
3390
3391 /* First send a status message before receiving the fds. */
0c759fc9 3392 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
618a6a28 3393 if (ret < 0) {
0d08d75e 3394 /* Somehow, the session daemon is not responding anymore. */
618a6a28
MD
3395 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3396 goto error_nosignal;
7735ef9e
DG
3397 }
3398
3399 /* Poll on consumer socket. */
84382d49
MD
3400 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3401 if (ret) {
3402 /* Needing to exit in the middle of a command: error. */
0d08d75e 3403 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
7735ef9e 3404 ret = -EINTR;
618a6a28 3405 goto error_nosignal;
7735ef9e
DG
3406 }
3407
3408 /* Get relayd socket from session daemon */
3409 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
3410 if (ret != sizeof(fd)) {
7735ef9e 3411 ret = -1;
4028eeb9 3412 fd = -1; /* Just in case it gets set with an invalid value. */
0d08d75e
DG
3413
3414 /*
3415 * Failing to receive FDs might indicate a major problem such as
3416 * reaching a fd limit during the receive where the kernel returns a
3417 * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we
3418 * don't take any chances and stop everything.
3419 *
3420 * XXX: Feature request #558 will fix that and avoid this possible
3421 * issue when reaching the fd limit.
3422 */
3423 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
618a6a28 3424 ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD;
f50f23d9
DG
3425 goto error;
3426 }
3427
7735ef9e
DG
3428 /* Copy socket information and received FD */
3429 switch (sock_type) {
3430 case LTTNG_STREAM_CONTROL:
3431 /* Copy received lttcomm socket */
6151a90f
JD
3432 lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock);
3433 ret = lttcomm_create_sock(&relayd->control_sock.sock);
4028eeb9 3434 /* Handle create_sock error. */
f66c074c 3435 if (ret < 0) {
618a6a28 3436 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
4028eeb9 3437 goto error;
f66c074c 3438 }
da009f2c
MD
3439 /*
3440 * Close the socket created internally by
3441 * lttcomm_create_sock, so we can replace it by the one
3442 * received from sessiond.
3443 */
3444 if (close(relayd->control_sock.sock.fd)) {
3445 PERROR("close");
3446 }
7735ef9e
DG
3447
3448 /* Assign new file descriptor */
6151a90f 3449 relayd->control_sock.sock.fd = fd;
4b29f1ce 3450 fd = -1; /* For error path */
6151a90f
JD
3451 /* Assign version values. */
3452 relayd->control_sock.major = relayd_sock->major;
3453 relayd->control_sock.minor = relayd_sock->minor;
c5b6f4f0 3454
d3e2ba59 3455 relayd->relayd_session_id = relayd_session_id;
c5b6f4f0 3456
7735ef9e
DG
3457 break;
3458 case LTTNG_STREAM_DATA:
3459 /* Copy received lttcomm socket */
6151a90f
JD
3460 lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock);
3461 ret = lttcomm_create_sock(&relayd->data_sock.sock);
4028eeb9 3462 /* Handle create_sock error. */
f66c074c 3463 if (ret < 0) {
618a6a28 3464 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
4028eeb9 3465 goto error;
f66c074c 3466 }
da009f2c
MD
3467 /*
3468 * Close the socket created internally by
3469 * lttcomm_create_sock, so we can replace it by the one
3470 * received from sessiond.
3471 */
3472 if (close(relayd->data_sock.sock.fd)) {
3473 PERROR("close");
3474 }
7735ef9e
DG
3475
3476 /* Assign new file descriptor */
6151a90f 3477 relayd->data_sock.sock.fd = fd;
4b29f1ce 3478 fd = -1; /* for eventual error paths */
6151a90f
JD
3479 /* Assign version values. */
3480 relayd->data_sock.major = relayd_sock->major;
3481 relayd->data_sock.minor = relayd_sock->minor;
7735ef9e
DG
3482 break;
3483 default:
3484 ERR("Unknown relayd socket type (%d)", sock_type);
59e71485 3485 ret = -1;
618a6a28 3486 ret_code = LTTCOMM_CONSUMERD_FATAL;
7735ef9e
DG
3487 goto error;
3488 }
3489
d88aee68 3490 DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)",
7735ef9e
DG
3491 sock_type == LTTNG_STREAM_CONTROL ? "control" : "data",
3492 relayd->net_seq_idx, fd);
3493
618a6a28
MD
3494 /* We successfully added the socket. Send status back. */
3495 ret = consumer_send_status_msg(sock, ret_code);
3496 if (ret < 0) {
3497 /* Somehow, the session daemon is not responding anymore. */
3498 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3499 goto error_nosignal;
3500 }
3501
7735ef9e
DG
3502 /*
3503 * Add relayd socket pair to consumer data hashtable. If object already
3504 * exists or on error, the function gracefully returns.
3505 */
d09e1200 3506 add_relayd(relayd);
7735ef9e
DG
3507
3508 /* All good! */
4028eeb9 3509 return 0;
7735ef9e
DG
3510
3511error:
618a6a28
MD
3512 if (consumer_send_status_msg(sock, ret_code) < 0) {
3513 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3514 }
3515
3516error_nosignal:
4028eeb9
DG
3517 /* Close received socket if valid. */
3518 if (fd >= 0) {
3519 if (close(fd)) {
3520 PERROR("close received socket");
3521 }
3522 }
cd2b09ed
DG
3523
3524 if (relayd_created) {
cd2b09ed
DG
3525 free(relayd);
3526 }
3527
7735ef9e
DG
3528 return ret;
3529}
ca22feea 3530
4e9a4686
DG
3531/*
3532 * Try to lock the stream mutex.
3533 *
3534 * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
3535 */
3536static int stream_try_lock(struct lttng_consumer_stream *stream)
3537{
3538 int ret;
3539
3540 assert(stream);
3541
3542 /*
3543 * Try to lock the stream mutex. On failure, we know that the stream is
3544 * being used else where hence there is data still being extracted.
3545 */
3546 ret = pthread_mutex_trylock(&stream->lock);
3547 if (ret) {
3548 /* For both EBUSY and EINVAL error, the mutex is NOT locked. */
3549 ret = 0;
3550 goto end;
3551 }
3552
3553 ret = 1;
3554
3555end:
3556 return ret;
3557}
3558
f7079f67
DG
3559/*
3560 * Search for a relayd associated to the session id and return the reference.
3561 *
3562 * A rcu read side lock MUST be acquire before calling this function and locked
3563 * until the relayd object is no longer necessary.
3564 */
3565static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id)
3566{
3567 struct lttng_ht_iter iter;
f7079f67 3568 struct consumer_relayd_sock_pair *relayd = NULL;
f7079f67
DG
3569
3570 /* Iterate over all relayd since they are indexed by net_seq_idx. */
3571 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
3572 node.node) {
18261bd1
DG
3573 /*
3574 * Check by sessiond id which is unique here where the relayd session
3575 * id might not be when having multiple relayd.
3576 */
3577 if (relayd->sessiond_session_id == id) {
f7079f67 3578 /* Found the relayd. There can be only one per id. */
18261bd1 3579 goto found;
f7079f67
DG
3580 }
3581 }
3582
18261bd1
DG
3583 return NULL;
3584
3585found:
f7079f67
DG
3586 return relayd;
3587}
3588
ca22feea
DG
3589/*
3590 * Check if for a given session id there is still data needed to be extract
3591 * from the buffers.
3592 *
6d805429 3593 * Return 1 if data is pending or else 0 meaning ready to be read.
ca22feea 3594 */
6d805429 3595int consumer_data_pending(uint64_t id)
ca22feea
DG
3596{
3597 int ret;
3598 struct lttng_ht_iter iter;
3599 struct lttng_ht *ht;
3600 struct lttng_consumer_stream *stream;
f7079f67 3601 struct consumer_relayd_sock_pair *relayd = NULL;
6d805429 3602 int (*data_pending)(struct lttng_consumer_stream *);
ca22feea 3603
6d805429 3604 DBG("Consumer data pending command on session id %" PRIu64, id);
ca22feea 3605
6f6eda74 3606 rcu_read_lock();
ca22feea
DG
3607 pthread_mutex_lock(&consumer_data.lock);
3608
3609 switch (consumer_data.type) {
3610 case LTTNG_CONSUMER_KERNEL:
6d805429 3611 data_pending = lttng_kconsumer_data_pending;
ca22feea
DG
3612 break;
3613 case LTTNG_CONSUMER32_UST:
3614 case LTTNG_CONSUMER64_UST:
6d805429 3615 data_pending = lttng_ustconsumer_data_pending;
ca22feea
DG
3616 break;
3617 default:
3618 ERR("Unknown consumer data type");
3619 assert(0);
3620 }
3621
3622 /* Ease our life a bit */
3623 ht = consumer_data.stream_list_ht;
3624
f7079f67
DG
3625 relayd = find_relayd_by_session_id(id);
3626 if (relayd) {
3627 /* Send init command for data pending. */
3628 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3629 ret = relayd_begin_data_pending(&relayd->control_sock,
3630 relayd->relayd_session_id);
3631 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
3632 if (ret < 0) {
3633 /* Communication error thus the relayd so no data pending. */
3634 goto data_not_pending;
3635 }
3636 }
3637
c8f59ee5 3638 cds_lfht_for_each_entry_duplicate(ht->ht,
d88aee68
DG
3639 ht->hash_fct(&id, lttng_ht_seed),
3640 ht->match_fct, &id,
ca22feea 3641 &iter.iter, stream, node_session_id.node) {
4e9a4686
DG
3642 /* If this call fails, the stream is being used hence data pending. */
3643 ret = stream_try_lock(stream);
3644 if (!ret) {
f7079f67 3645 goto data_pending;
ca22feea 3646 }
ca22feea 3647
4e9a4686
DG
3648 /*
3649 * A removed node from the hash table indicates that the stream has
3650 * been deleted thus having a guarantee that the buffers are closed
3651 * on the consumer side. However, data can still be transmitted
3652 * over the network so don't skip the relayd check.
3653 */
3654 ret = cds_lfht_is_node_deleted(&stream->node.node);
3655 if (!ret) {
3656 /* Check the stream if there is data in the buffers. */
6d805429
DG
3657 ret = data_pending(stream);
3658 if (ret == 1) {
4e9a4686 3659 pthread_mutex_unlock(&stream->lock);
f7079f67 3660 goto data_pending;
4e9a4686
DG
3661 }
3662 }
3663
3664 /* Relayd check */
f7079f67 3665 if (relayd) {
c8f59ee5
DG
3666 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3667 if (stream->metadata_flag) {
ad7051c0
DG
3668 ret = relayd_quiescent_control(&relayd->control_sock,
3669 stream->relayd_stream_id);
c8f59ee5 3670 } else {
6d805429 3671 ret = relayd_data_pending(&relayd->control_sock,
39df6d9f
DG
3672 stream->relayd_stream_id,
3673 stream->next_net_seq_num - 1);
c8f59ee5
DG
3674 }
3675 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
6d805429 3676 if (ret == 1) {
4e9a4686 3677 pthread_mutex_unlock(&stream->lock);
f7079f67 3678 goto data_pending;
c8f59ee5
DG
3679 }
3680 }
4e9a4686 3681 pthread_mutex_unlock(&stream->lock);
c8f59ee5 3682 }
ca22feea 3683
f7079f67
DG
3684 if (relayd) {
3685 unsigned int is_data_inflight = 0;
3686
3687 /* Send init command for data pending. */
3688 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3689 ret = relayd_end_data_pending(&relayd->control_sock,
3690 relayd->relayd_session_id, &is_data_inflight);
3691 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
bdd88757 3692 if (ret < 0) {
f7079f67
DG
3693 goto data_not_pending;
3694 }
bdd88757
DG
3695 if (is_data_inflight) {
3696 goto data_pending;
3697 }
f7079f67
DG
3698 }
3699
ca22feea 3700 /*
f7079f67
DG
3701 * Finding _no_ node in the hash table and no inflight data means that the
3702 * stream(s) have been removed thus data is guaranteed to be available for
3703 * analysis from the trace files.
ca22feea
DG
3704 */
3705
f7079f67 3706data_not_pending:
ca22feea
DG
3707 /* Data is available to be read by a viewer. */
3708 pthread_mutex_unlock(&consumer_data.lock);
c8f59ee5 3709 rcu_read_unlock();
6d805429 3710 return 0;
ca22feea 3711
f7079f67 3712data_pending:
ca22feea
DG
3713 /* Data is still being extracted from buffers. */
3714 pthread_mutex_unlock(&consumer_data.lock);
c8f59ee5 3715 rcu_read_unlock();
6d805429 3716 return 1;
ca22feea 3717}
f50f23d9
DG
3718
3719/*
3720 * Send a ret code status message to the sessiond daemon.
3721 *
3722 * Return the sendmsg() return value.
3723 */
3724int consumer_send_status_msg(int sock, int ret_code)
3725{
3726 struct lttcomm_consumer_status_msg msg;
3727
53efb85a 3728 memset(&msg, 0, sizeof(msg));
f50f23d9
DG
3729 msg.ret_code = ret_code;
3730
3731 return lttcomm_send_unix_sock(sock, &msg, sizeof(msg));
3732}
ffe60014
DG
3733
3734/*
3735 * Send a channel status message to the sessiond daemon.
3736 *
3737 * Return the sendmsg() return value.
3738 */
3739int consumer_send_status_channel(int sock,
3740 struct lttng_consumer_channel *channel)
3741{
3742 struct lttcomm_consumer_status_channel msg;
3743
3744 assert(sock >= 0);
3745
53efb85a 3746 memset(&msg, 0, sizeof(msg));
ffe60014 3747 if (!channel) {
0c759fc9 3748 msg.ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
ffe60014 3749 } else {
0c759fc9 3750 msg.ret_code = LTTCOMM_CONSUMERD_SUCCESS;
ffe60014
DG
3751 msg.key = channel->key;
3752 msg.stream_count = channel->streams.count;
3753 }
3754
3755 return lttcomm_send_unix_sock(sock, &msg, sizeof(msg));
3756}
5c786ded 3757
d07ceecd
MD
3758unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos,
3759 unsigned long produced_pos, uint64_t nb_packets_per_stream,
3760 uint64_t max_sb_size)
5c786ded 3761{
d07ceecd 3762 unsigned long start_pos;
5c786ded 3763
d07ceecd
MD
3764 if (!nb_packets_per_stream) {
3765 return consumed_pos; /* Grab everything */
3766 }
3767 start_pos = produced_pos - offset_align_floor(produced_pos, max_sb_size);
3768 start_pos -= max_sb_size * nb_packets_per_stream;
3769 if ((long) (start_pos - consumed_pos) < 0) {
3770 return consumed_pos; /* Grab everything */
3771 }
3772 return start_pos;
5c786ded 3773}
This page took 0.284479 seconds and 4 git commands to generate.