syscall tracing: update tests
[lttng-tools.git] / src / common / consumer.c
CommitLineData
3bd1e081
MD
1/*
2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
00e2e675 4 * 2012 - David Goulet <dgoulet@efficios.com>
3bd1e081 5 *
d14d33bf
AM
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License, version 2 only,
8 * as published by the Free Software Foundation.
3bd1e081 9 *
d14d33bf
AM
10 * This program is distributed in the hope that it will be useful, but WITHOUT
11 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
12 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
13 * more details.
3bd1e081 14 *
d14d33bf
AM
15 * You should have received a copy of the GNU General Public License along
16 * with this program; if not, write to the Free Software Foundation, Inc.,
17 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
3bd1e081
MD
18 */
19
20#define _GNU_SOURCE
6c1c0768 21#define _LGPL_SOURCE
3bd1e081 22#include <assert.h>
3bd1e081
MD
23#include <poll.h>
24#include <pthread.h>
25#include <stdlib.h>
26#include <string.h>
27#include <sys/mman.h>
28#include <sys/socket.h>
29#include <sys/types.h>
30#include <unistd.h>
77c7c900 31#include <inttypes.h>
331744e3 32#include <signal.h>
3bd1e081 33
51a9e1c7 34#include <bin/lttng-consumerd/health-consumerd.h>
990570ed 35#include <common/common.h>
fb3a43a9
DG
36#include <common/utils.h>
37#include <common/compat/poll.h>
f263b7fd 38#include <common/compat/endian.h>
309167d2 39#include <common/index/index.h>
10a8a223 40#include <common/kernel-ctl/kernel-ctl.h>
00e2e675 41#include <common/sessiond-comm/relayd.h>
10a8a223
DG
42#include <common/sessiond-comm/sessiond-comm.h>
43#include <common/kernel-consumer/kernel-consumer.h>
00e2e675 44#include <common/relayd/relayd.h>
10a8a223 45#include <common/ust-consumer/ust-consumer.h>
d3e2ba59 46#include <common/consumer-timer.h>
10a8a223
DG
47
48#include "consumer.h"
1d1a276c 49#include "consumer-stream.h"
2d57de81 50#include "consumer-testpoint.h"
d07ceecd 51#include "align.h"
3bd1e081
MD
52
53struct lttng_consumer_global_data consumer_data = {
3bd1e081
MD
54 .stream_count = 0,
55 .need_update = 1,
56 .type = LTTNG_CONSUMER_UNKNOWN,
57};
58
d8ef542d
MD
59enum consumer_channel_action {
60 CONSUMER_CHANNEL_ADD,
a0cbdd2e 61 CONSUMER_CHANNEL_DEL,
d8ef542d
MD
62 CONSUMER_CHANNEL_QUIT,
63};
64
65struct consumer_channel_msg {
66 enum consumer_channel_action action;
a0cbdd2e
MD
67 struct lttng_consumer_channel *chan; /* add */
68 uint64_t key; /* del */
d8ef542d
MD
69};
70
3bd1e081
MD
71/*
72 * Flag to inform the polling thread to quit when all fd hung up. Updated by
73 * the consumer_thread_receive_fds when it notices that all fds has hung up.
74 * Also updated by the signal handler (consumer_should_exit()). Read by the
75 * polling threads.
76 */
a98dae5f 77volatile int consumer_quit;
3bd1e081 78
43c34bc3 79/*
43c34bc3
DG
80 * Global hash table containing respectively metadata and data streams. The
81 * stream element in this ht should only be updated by the metadata poll thread
82 * for the metadata and the data poll thread for the data.
83 */
40dc48e0
DG
84static struct lttng_ht *metadata_ht;
85static struct lttng_ht *data_ht;
43c34bc3 86
acdb9057
DG
87/*
88 * Notify a thread lttng pipe to poll back again. This usually means that some
89 * global state has changed so we just send back the thread in a poll wait
90 * call.
91 */
92static void notify_thread_lttng_pipe(struct lttng_pipe *pipe)
93{
94 struct lttng_consumer_stream *null_stream = NULL;
95
96 assert(pipe);
97
98 (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream));
99}
100
5c635c72
MD
101static void notify_health_quit_pipe(int *pipe)
102{
6cd525e8 103 ssize_t ret;
5c635c72 104
6cd525e8
MD
105 ret = lttng_write(pipe[1], "4", 1);
106 if (ret < 1) {
5c635c72
MD
107 PERROR("write consumer health quit");
108 }
109}
110
d8ef542d
MD
111static void notify_channel_pipe(struct lttng_consumer_local_data *ctx,
112 struct lttng_consumer_channel *chan,
a0cbdd2e 113 uint64_t key,
d8ef542d
MD
114 enum consumer_channel_action action)
115{
116 struct consumer_channel_msg msg;
6cd525e8 117 ssize_t ret;
d8ef542d 118
e56251fc
DG
119 memset(&msg, 0, sizeof(msg));
120
d8ef542d
MD
121 msg.action = action;
122 msg.chan = chan;
f21dae48 123 msg.key = key;
6cd525e8
MD
124 ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg));
125 if (ret < sizeof(msg)) {
126 PERROR("notify_channel_pipe write error");
127 }
d8ef542d
MD
128}
129
a0cbdd2e
MD
130void notify_thread_del_channel(struct lttng_consumer_local_data *ctx,
131 uint64_t key)
132{
133 notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL);
134}
135
d8ef542d
MD
136static int read_channel_pipe(struct lttng_consumer_local_data *ctx,
137 struct lttng_consumer_channel **chan,
a0cbdd2e 138 uint64_t *key,
d8ef542d
MD
139 enum consumer_channel_action *action)
140{
141 struct consumer_channel_msg msg;
6cd525e8 142 ssize_t ret;
d8ef542d 143
6cd525e8
MD
144 ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg));
145 if (ret < sizeof(msg)) {
146 ret = -1;
147 goto error;
d8ef542d 148 }
6cd525e8
MD
149 *action = msg.action;
150 *chan = msg.chan;
151 *key = msg.key;
152error:
153 return (int) ret;
d8ef542d
MD
154}
155
212d67a2
DG
156/*
157 * Cleanup the stream list of a channel. Those streams are not yet globally
158 * visible
159 */
160static void clean_channel_stream_list(struct lttng_consumer_channel *channel)
161{
162 struct lttng_consumer_stream *stream, *stmp;
163
164 assert(channel);
165
166 /* Delete streams that might have been left in the stream list. */
167 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
168 send_node) {
169 cds_list_del(&stream->send_node);
170 /*
171 * Once a stream is added to this list, the buffers were created so we
172 * have a guarantee that this call will succeed. Setting the monitor
173 * mode to 0 so we don't lock nor try to delete the stream from the
174 * global hash table.
175 */
176 stream->monitor = 0;
177 consumer_stream_destroy(stream, NULL);
178 }
179}
180
3bd1e081
MD
181/*
182 * Find a stream. The consumer_data.lock must be locked during this
183 * call.
184 */
d88aee68 185static struct lttng_consumer_stream *find_stream(uint64_t key,
8389e4f8 186 struct lttng_ht *ht)
3bd1e081 187{
e4421fec 188 struct lttng_ht_iter iter;
d88aee68 189 struct lttng_ht_node_u64 *node;
e4421fec 190 struct lttng_consumer_stream *stream = NULL;
3bd1e081 191
8389e4f8
DG
192 assert(ht);
193
d88aee68
DG
194 /* -1ULL keys are lookup failures */
195 if (key == (uint64_t) -1ULL) {
7ad0a0cb 196 return NULL;
7a57cf92 197 }
e4421fec 198
6065ceec
DG
199 rcu_read_lock();
200
d88aee68
DG
201 lttng_ht_lookup(ht, &key, &iter);
202 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
203 if (node != NULL) {
204 stream = caa_container_of(node, struct lttng_consumer_stream, node);
3bd1e081 205 }
e4421fec 206
6065ceec
DG
207 rcu_read_unlock();
208
e4421fec 209 return stream;
3bd1e081
MD
210}
211
da009f2c 212static void steal_stream_key(uint64_t key, struct lttng_ht *ht)
7ad0a0cb
MD
213{
214 struct lttng_consumer_stream *stream;
215
04253271 216 rcu_read_lock();
ffe60014 217 stream = find_stream(key, ht);
04253271 218 if (stream) {
da009f2c 219 stream->key = (uint64_t) -1ULL;
04253271
MD
220 /*
221 * We don't want the lookup to match, but we still need
222 * to iterate on this stream when iterating over the hash table. Just
223 * change the node key.
224 */
da009f2c 225 stream->node.key = (uint64_t) -1ULL;
04253271
MD
226 }
227 rcu_read_unlock();
7ad0a0cb
MD
228}
229
d56db448
DG
230/*
231 * Return a channel object for the given key.
232 *
233 * RCU read side lock MUST be acquired before calling this function and
234 * protects the channel ptr.
235 */
d88aee68 236struct lttng_consumer_channel *consumer_find_channel(uint64_t key)
3bd1e081 237{
e4421fec 238 struct lttng_ht_iter iter;
d88aee68 239 struct lttng_ht_node_u64 *node;
e4421fec 240 struct lttng_consumer_channel *channel = NULL;
3bd1e081 241
d88aee68
DG
242 /* -1ULL keys are lookup failures */
243 if (key == (uint64_t) -1ULL) {
7ad0a0cb 244 return NULL;
7a57cf92 245 }
e4421fec 246
d88aee68
DG
247 lttng_ht_lookup(consumer_data.channel_ht, &key, &iter);
248 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
249 if (node != NULL) {
250 channel = caa_container_of(node, struct lttng_consumer_channel, node);
3bd1e081 251 }
e4421fec
DG
252
253 return channel;
3bd1e081
MD
254}
255
b5a6470f
DG
256/*
257 * There is a possibility that the consumer does not have enough time between
258 * the close of the channel on the session daemon and the cleanup in here thus
259 * once we have a channel add with an existing key, we know for sure that this
260 * channel will eventually get cleaned up by all streams being closed.
261 *
262 * This function just nullifies the already existing channel key.
263 */
264static void steal_channel_key(uint64_t key)
265{
266 struct lttng_consumer_channel *channel;
267
268 rcu_read_lock();
269 channel = consumer_find_channel(key);
270 if (channel) {
271 channel->key = (uint64_t) -1ULL;
272 /*
273 * We don't want the lookup to match, but we still need to iterate on
274 * this channel when iterating over the hash table. Just change the
275 * node key.
276 */
277 channel->node.key = (uint64_t) -1ULL;
278 }
279 rcu_read_unlock();
280}
281
ffe60014 282static void free_channel_rcu(struct rcu_head *head)
702b1ea4 283{
d88aee68
DG
284 struct lttng_ht_node_u64 *node =
285 caa_container_of(head, struct lttng_ht_node_u64, head);
ffe60014
DG
286 struct lttng_consumer_channel *channel =
287 caa_container_of(node, struct lttng_consumer_channel, node);
702b1ea4 288
ffe60014 289 free(channel);
702b1ea4
MD
290}
291
00e2e675
DG
292/*
293 * RCU protected relayd socket pair free.
294 */
ffe60014 295static void free_relayd_rcu(struct rcu_head *head)
00e2e675 296{
d88aee68
DG
297 struct lttng_ht_node_u64 *node =
298 caa_container_of(head, struct lttng_ht_node_u64, head);
00e2e675
DG
299 struct consumer_relayd_sock_pair *relayd =
300 caa_container_of(node, struct consumer_relayd_sock_pair, node);
301
8994307f
DG
302 /*
303 * Close all sockets. This is done in the call RCU since we don't want the
304 * socket fds to be reassigned thus potentially creating bad state of the
305 * relayd object.
306 *
307 * We do not have to lock the control socket mutex here since at this stage
308 * there is no one referencing to this relayd object.
309 */
310 (void) relayd_close(&relayd->control_sock);
311 (void) relayd_close(&relayd->data_sock);
312
00e2e675
DG
313 free(relayd);
314}
315
316/*
317 * Destroy and free relayd socket pair object.
00e2e675 318 */
51230d70 319void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
320{
321 int ret;
322 struct lttng_ht_iter iter;
323
173af62f
DG
324 if (relayd == NULL) {
325 return;
326 }
327
00e2e675
DG
328 DBG("Consumer destroy and close relayd socket pair");
329
330 iter.iter.node = &relayd->node.node;
331 ret = lttng_ht_del(consumer_data.relayd_ht, &iter);
173af62f 332 if (ret != 0) {
8994307f 333 /* We assume the relayd is being or is destroyed */
173af62f
DG
334 return;
335 }
00e2e675 336
00e2e675 337 /* RCU free() call */
ffe60014
DG
338 call_rcu(&relayd->node.head, free_relayd_rcu);
339}
340
341/*
342 * Remove a channel from the global list protected by a mutex. This function is
343 * also responsible for freeing its data structures.
344 */
345void consumer_del_channel(struct lttng_consumer_channel *channel)
346{
347 int ret;
348 struct lttng_ht_iter iter;
349
d88aee68 350 DBG("Consumer delete channel key %" PRIu64, channel->key);
ffe60014
DG
351
352 pthread_mutex_lock(&consumer_data.lock);
a9838785 353 pthread_mutex_lock(&channel->lock);
ffe60014 354
212d67a2
DG
355 /* Destroy streams that might have been left in the stream list. */
356 clean_channel_stream_list(channel);
51e762e5 357
d3e2ba59
JD
358 if (channel->live_timer_enabled == 1) {
359 consumer_timer_live_stop(channel);
360 }
361
ffe60014
DG
362 switch (consumer_data.type) {
363 case LTTNG_CONSUMER_KERNEL:
364 break;
365 case LTTNG_CONSUMER32_UST:
366 case LTTNG_CONSUMER64_UST:
367 lttng_ustconsumer_del_channel(channel);
368 break;
369 default:
370 ERR("Unknown consumer_data type");
371 assert(0);
372 goto end;
373 }
374
375 rcu_read_lock();
376 iter.iter.node = &channel->node.node;
377 ret = lttng_ht_del(consumer_data.channel_ht, &iter);
378 assert(!ret);
379 rcu_read_unlock();
380
381 call_rcu(&channel->node.head, free_channel_rcu);
382end:
a9838785 383 pthread_mutex_unlock(&channel->lock);
ffe60014 384 pthread_mutex_unlock(&consumer_data.lock);
00e2e675
DG
385}
386
228b5bf7
DG
387/*
388 * Iterate over the relayd hash table and destroy each element. Finally,
389 * destroy the whole hash table.
390 */
391static void cleanup_relayd_ht(void)
392{
393 struct lttng_ht_iter iter;
394 struct consumer_relayd_sock_pair *relayd;
395
396 rcu_read_lock();
397
398 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
399 node.node) {
51230d70 400 consumer_destroy_relayd(relayd);
228b5bf7
DG
401 }
402
228b5bf7 403 rcu_read_unlock();
36b588ed
MD
404
405 lttng_ht_destroy(consumer_data.relayd_ht);
228b5bf7
DG
406}
407
8994307f
DG
408/*
409 * Update the end point status of all streams having the given network sequence
410 * index (relayd index).
411 *
412 * It's atomically set without having the stream mutex locked which is fine
413 * because we handle the write/read race with a pipe wakeup for each thread.
414 */
da009f2c 415static void update_endpoint_status_by_netidx(uint64_t net_seq_idx,
8994307f
DG
416 enum consumer_endpoint_status status)
417{
418 struct lttng_ht_iter iter;
419 struct lttng_consumer_stream *stream;
420
da009f2c 421 DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx);
8994307f
DG
422
423 rcu_read_lock();
424
425 /* Let's begin with metadata */
426 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
427 if (stream->net_seq_idx == net_seq_idx) {
428 uatomic_set(&stream->endpoint_status, status);
429 DBG("Delete flag set to metadata stream %d", stream->wait_fd);
430 }
431 }
432
433 /* Follow up by the data streams */
434 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
435 if (stream->net_seq_idx == net_seq_idx) {
436 uatomic_set(&stream->endpoint_status, status);
437 DBG("Delete flag set to data stream %d", stream->wait_fd);
438 }
439 }
440 rcu_read_unlock();
441}
442
443/*
444 * Cleanup a relayd object by flagging every associated streams for deletion,
445 * destroying the object meaning removing it from the relayd hash table,
446 * closing the sockets and freeing the memory in a RCU call.
447 *
448 * If a local data context is available, notify the threads that the streams'
449 * state have changed.
450 */
451static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd,
452 struct lttng_consumer_local_data *ctx)
453{
da009f2c 454 uint64_t netidx;
8994307f
DG
455
456 assert(relayd);
457
9617607b
DG
458 DBG("Cleaning up relayd sockets");
459
8994307f
DG
460 /* Save the net sequence index before destroying the object */
461 netidx = relayd->net_seq_idx;
462
463 /*
464 * Delete the relayd from the relayd hash table, close the sockets and free
465 * the object in a RCU call.
466 */
51230d70 467 consumer_destroy_relayd(relayd);
8994307f
DG
468
469 /* Set inactive endpoint to all streams */
470 update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE);
471
472 /*
473 * With a local data context, notify the threads that the streams' state
474 * have changed. The write() action on the pipe acts as an "implicit"
475 * memory barrier ordering the updates of the end point status from the
476 * read of this status which happens AFTER receiving this notify.
477 */
478 if (ctx) {
acdb9057 479 notify_thread_lttng_pipe(ctx->consumer_data_pipe);
13886d2d 480 notify_thread_lttng_pipe(ctx->consumer_metadata_pipe);
8994307f
DG
481 }
482}
483
a6ba4fe1
DG
484/*
485 * Flag a relayd socket pair for destruction. Destroy it if the refcount
486 * reaches zero.
487 *
488 * RCU read side lock MUST be aquired before calling this function.
489 */
490void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd)
491{
492 assert(relayd);
493
494 /* Set destroy flag for this object */
495 uatomic_set(&relayd->destroy_flag, 1);
496
497 /* Destroy the relayd if refcount is 0 */
498 if (uatomic_read(&relayd->refcount) == 0) {
51230d70 499 consumer_destroy_relayd(relayd);
a6ba4fe1
DG
500 }
501}
502
3bd1e081 503/*
1d1a276c
DG
504 * Completly destroy stream from every visiable data structure and the given
505 * hash table if one.
506 *
507 * One this call returns, the stream object is not longer usable nor visible.
3bd1e081 508 */
e316aad5
DG
509void consumer_del_stream(struct lttng_consumer_stream *stream,
510 struct lttng_ht *ht)
3bd1e081 511{
1d1a276c 512 consumer_stream_destroy(stream, ht);
3bd1e081
MD
513}
514
5ab66908
MD
515/*
516 * XXX naming of del vs destroy is all mixed up.
517 */
518void consumer_del_stream_for_data(struct lttng_consumer_stream *stream)
519{
520 consumer_stream_destroy(stream, data_ht);
521}
522
523void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream)
524{
525 consumer_stream_destroy(stream, metadata_ht);
526}
527
d88aee68
DG
528struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key,
529 uint64_t stream_key,
3bd1e081 530 enum lttng_consumer_stream_state state,
ffe60014 531 const char *channel_name,
6df2e2c9 532 uid_t uid,
00e2e675 533 gid_t gid,
57a269f2 534 uint64_t relayd_id,
53632229 535 uint64_t session_id,
ffe60014
DG
536 int cpu,
537 int *alloc_ret,
4891ece8
DG
538 enum consumer_channel_type type,
539 unsigned int monitor)
3bd1e081 540{
ffe60014 541 int ret;
3bd1e081 542 struct lttng_consumer_stream *stream;
3bd1e081 543
effcf122 544 stream = zmalloc(sizeof(*stream));
3bd1e081 545 if (stream == NULL) {
7a57cf92 546 PERROR("malloc struct lttng_consumer_stream");
ffe60014 547 ret = -ENOMEM;
7a57cf92 548 goto end;
3bd1e081 549 }
7a57cf92 550
d56db448
DG
551 rcu_read_lock();
552
3bd1e081 553 stream->key = stream_key;
3bd1e081
MD
554 stream->out_fd = -1;
555 stream->out_fd_offset = 0;
e5d1a9b3 556 stream->output_written = 0;
3bd1e081 557 stream->state = state;
6df2e2c9
MD
558 stream->uid = uid;
559 stream->gid = gid;
ffe60014 560 stream->net_seq_idx = relayd_id;
53632229 561 stream->session_id = session_id;
4891ece8 562 stream->monitor = monitor;
774d490c 563 stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE;
309167d2 564 stream->index_fd = -1;
53632229 565 pthread_mutex_init(&stream->lock, NULL);
58b1f425 566
ffe60014
DG
567 /* If channel is the metadata, flag this stream as metadata. */
568 if (type == CONSUMER_CHANNEL_TYPE_METADATA) {
569 stream->metadata_flag = 1;
570 /* Metadata is flat out. */
571 strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name));
94d49140
JD
572 /* Live rendez-vous point. */
573 pthread_cond_init(&stream->metadata_rdv, NULL);
574 pthread_mutex_init(&stream->metadata_rdv_lock, NULL);
58b1f425 575 } else {
ffe60014
DG
576 /* Format stream name to <channel_name>_<cpu_number> */
577 ret = snprintf(stream->name, sizeof(stream->name), "%s_%d",
578 channel_name, cpu);
579 if (ret < 0) {
580 PERROR("snprintf stream name");
581 goto error;
582 }
58b1f425 583 }
c30aaa51 584
ffe60014 585 /* Key is always the wait_fd for streams. */
d88aee68 586 lttng_ht_node_init_u64(&stream->node, stream->key);
ffe60014 587
d8ef542d
MD
588 /* Init node per channel id key */
589 lttng_ht_node_init_u64(&stream->node_channel_id, channel_key);
590
53632229 591 /* Init session id node with the stream session id */
d88aee68 592 lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id);
53632229 593
07b86b52
JD
594 DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64
595 " relayd_id %" PRIu64 ", session_id %" PRIu64,
596 stream->name, stream->key, channel_key,
597 stream->net_seq_idx, stream->session_id);
d56db448
DG
598
599 rcu_read_unlock();
3bd1e081 600 return stream;
c80048c6
MD
601
602error:
d56db448 603 rcu_read_unlock();
c80048c6 604 free(stream);
7a57cf92 605end:
ffe60014
DG
606 if (alloc_ret) {
607 *alloc_ret = ret;
608 }
c80048c6 609 return NULL;
3bd1e081
MD
610}
611
612/*
613 * Add a stream to the global list protected by a mutex.
614 */
5ab66908 615int consumer_add_data_stream(struct lttng_consumer_stream *stream)
3bd1e081 616{
5ab66908 617 struct lttng_ht *ht = data_ht;
3bd1e081
MD
618 int ret = 0;
619
e316aad5 620 assert(stream);
43c34bc3 621 assert(ht);
c77fc10a 622
d88aee68 623 DBG3("Adding consumer stream %" PRIu64, stream->key);
e316aad5
DG
624
625 pthread_mutex_lock(&consumer_data.lock);
a9838785 626 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 627 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 628 pthread_mutex_lock(&stream->lock);
b0b335c8 629 rcu_read_lock();
e316aad5 630
43c34bc3 631 /* Steal stream identifier to avoid having streams with the same key */
ffe60014 632 steal_stream_key(stream->key, ht);
43c34bc3 633
d88aee68 634 lttng_ht_add_unique_u64(ht, &stream->node);
00e2e675 635
d8ef542d
MD
636 lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht,
637 &stream->node_channel_id);
638
ca22feea
DG
639 /*
640 * Add stream to the stream_list_ht of the consumer data. No need to steal
641 * the key since the HT does not use it and we allow to add redundant keys
642 * into this table.
643 */
d88aee68 644 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 645
e316aad5 646 /*
ffe60014
DG
647 * When nb_init_stream_left reaches 0, we don't need to trigger any action
648 * in terms of destroying the associated channel, because the action that
e316aad5
DG
649 * causes the count to become 0 also causes a stream to be added. The
650 * channel deletion will thus be triggered by the following removal of this
651 * stream.
652 */
ffe60014 653 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
654 /* Increment refcount before decrementing nb_init_stream_left */
655 cmm_smp_wmb();
ffe60014 656 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
657 }
658
659 /* Update consumer data once the node is inserted. */
3bd1e081
MD
660 consumer_data.stream_count++;
661 consumer_data.need_update = 1;
662
e316aad5 663 rcu_read_unlock();
2e818a6a 664 pthread_mutex_unlock(&stream->lock);
ec6ea7d0 665 pthread_mutex_unlock(&stream->chan->timer_lock);
a9838785 666 pthread_mutex_unlock(&stream->chan->lock);
3bd1e081 667 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 668
3bd1e081
MD
669 return ret;
670}
671
5ab66908
MD
672void consumer_del_data_stream(struct lttng_consumer_stream *stream)
673{
674 consumer_del_stream(stream, data_ht);
675}
676
00e2e675 677/*
3f8e211f
DG
678 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
679 * be acquired before calling this.
00e2e675 680 */
d09e1200 681static int add_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
682{
683 int ret = 0;
d88aee68 684 struct lttng_ht_node_u64 *node;
00e2e675
DG
685 struct lttng_ht_iter iter;
686
ffe60014 687 assert(relayd);
00e2e675 688
00e2e675 689 lttng_ht_lookup(consumer_data.relayd_ht,
d88aee68
DG
690 &relayd->net_seq_idx, &iter);
691 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675 692 if (node != NULL) {
00e2e675
DG
693 goto end;
694 }
d88aee68 695 lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node);
00e2e675 696
00e2e675
DG
697end:
698 return ret;
699}
700
701/*
702 * Allocate and return a consumer relayd socket.
703 */
704struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair(
da009f2c 705 uint64_t net_seq_idx)
00e2e675
DG
706{
707 struct consumer_relayd_sock_pair *obj = NULL;
708
da009f2c
MD
709 /* net sequence index of -1 is a failure */
710 if (net_seq_idx == (uint64_t) -1ULL) {
00e2e675
DG
711 goto error;
712 }
713
714 obj = zmalloc(sizeof(struct consumer_relayd_sock_pair));
715 if (obj == NULL) {
716 PERROR("zmalloc relayd sock");
717 goto error;
718 }
719
720 obj->net_seq_idx = net_seq_idx;
721 obj->refcount = 0;
173af62f 722 obj->destroy_flag = 0;
f96e4545
MD
723 obj->control_sock.sock.fd = -1;
724 obj->data_sock.sock.fd = -1;
d88aee68 725 lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx);
00e2e675
DG
726 pthread_mutex_init(&obj->ctrl_sock_mutex, NULL);
727
728error:
729 return obj;
730}
731
732/*
733 * Find a relayd socket pair in the global consumer data.
734 *
735 * Return the object if found else NULL.
b0b335c8
MD
736 * RCU read-side lock must be held across this call and while using the
737 * returned object.
00e2e675 738 */
d88aee68 739struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key)
00e2e675
DG
740{
741 struct lttng_ht_iter iter;
d88aee68 742 struct lttng_ht_node_u64 *node;
00e2e675
DG
743 struct consumer_relayd_sock_pair *relayd = NULL;
744
745 /* Negative keys are lookup failures */
d88aee68 746 if (key == (uint64_t) -1ULL) {
00e2e675
DG
747 goto error;
748 }
749
d88aee68 750 lttng_ht_lookup(consumer_data.relayd_ht, &key,
00e2e675 751 &iter);
d88aee68 752 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675
DG
753 if (node != NULL) {
754 relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node);
755 }
756
00e2e675
DG
757error:
758 return relayd;
759}
760
10a50311
JD
761/*
762 * Find a relayd and send the stream
763 *
764 * Returns 0 on success, < 0 on error
765 */
766int consumer_send_relayd_stream(struct lttng_consumer_stream *stream,
767 char *path)
768{
769 int ret = 0;
770 struct consumer_relayd_sock_pair *relayd;
771
772 assert(stream);
773 assert(stream->net_seq_idx != -1ULL);
774 assert(path);
775
776 /* The stream is not metadata. Get relayd reference if exists. */
777 rcu_read_lock();
778 relayd = consumer_find_relayd(stream->net_seq_idx);
779 if (relayd != NULL) {
780 /* Add stream on the relayd */
781 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
782 ret = relayd_add_stream(&relayd->control_sock, stream->name,
783 path, &stream->relayd_stream_id,
784 stream->chan->tracefile_size, stream->chan->tracefile_count);
785 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
786 if (ret < 0) {
787 goto end;
788 }
1c20f0e2 789
10a50311 790 uatomic_inc(&relayd->refcount);
d01178b6 791 stream->sent_to_relayd = 1;
10a50311
JD
792 } else {
793 ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.",
794 stream->key, stream->net_seq_idx);
795 ret = -1;
796 goto end;
797 }
798
799 DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64,
800 stream->name, stream->key, stream->net_seq_idx);
801
802end:
803 rcu_read_unlock();
804 return ret;
805}
806
a4baae1b
JD
807/*
808 * Find a relayd and send the streams sent message
809 *
810 * Returns 0 on success, < 0 on error
811 */
812int consumer_send_relayd_streams_sent(uint64_t net_seq_idx)
813{
814 int ret = 0;
815 struct consumer_relayd_sock_pair *relayd;
816
817 assert(net_seq_idx != -1ULL);
818
819 /* The stream is not metadata. Get relayd reference if exists. */
820 rcu_read_lock();
821 relayd = consumer_find_relayd(net_seq_idx);
822 if (relayd != NULL) {
823 /* Add stream on the relayd */
824 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
825 ret = relayd_streams_sent(&relayd->control_sock);
826 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
827 if (ret < 0) {
828 goto end;
829 }
830 } else {
831 ERR("Relayd ID %" PRIu64 " unknown. Can't send streams_sent.",
832 net_seq_idx);
833 ret = -1;
834 goto end;
835 }
836
837 ret = 0;
838 DBG("All streams sent relayd id %" PRIu64, net_seq_idx);
839
840end:
841 rcu_read_unlock();
842 return ret;
843}
844
10a50311
JD
845/*
846 * Find a relayd and close the stream
847 */
848void close_relayd_stream(struct lttng_consumer_stream *stream)
849{
850 struct consumer_relayd_sock_pair *relayd;
851
852 /* The stream is not metadata. Get relayd reference if exists. */
853 rcu_read_lock();
854 relayd = consumer_find_relayd(stream->net_seq_idx);
855 if (relayd) {
856 consumer_stream_relayd_close(stream, relayd);
857 }
858 rcu_read_unlock();
859}
860
00e2e675
DG
861/*
862 * Handle stream for relayd transmission if the stream applies for network
863 * streaming where the net sequence index is set.
864 *
865 * Return destination file descriptor or negative value on error.
866 */
6197aea7 867static int write_relayd_stream_header(struct lttng_consumer_stream *stream,
1d4dfdef
DG
868 size_t data_size, unsigned long padding,
869 struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
870{
871 int outfd = -1, ret;
00e2e675
DG
872 struct lttcomm_relayd_data_hdr data_hdr;
873
874 /* Safety net */
875 assert(stream);
6197aea7 876 assert(relayd);
00e2e675
DG
877
878 /* Reset data header */
879 memset(&data_hdr, 0, sizeof(data_hdr));
880
00e2e675
DG
881 if (stream->metadata_flag) {
882 /* Caller MUST acquire the relayd control socket lock */
883 ret = relayd_send_metadata(&relayd->control_sock, data_size);
884 if (ret < 0) {
885 goto error;
886 }
887
888 /* Metadata are always sent on the control socket. */
6151a90f 889 outfd = relayd->control_sock.sock.fd;
00e2e675
DG
890 } else {
891 /* Set header with stream information */
892 data_hdr.stream_id = htobe64(stream->relayd_stream_id);
893 data_hdr.data_size = htobe32(data_size);
1d4dfdef 894 data_hdr.padding_size = htobe32(padding);
39df6d9f
DG
895 /*
896 * Note that net_seq_num below is assigned with the *current* value of
897 * next_net_seq_num and only after that the next_net_seq_num will be
898 * increment. This is why when issuing a command on the relayd using
899 * this next value, 1 should always be substracted in order to compare
900 * the last seen sequence number on the relayd side to the last sent.
901 */
3604f373 902 data_hdr.net_seq_num = htobe64(stream->next_net_seq_num);
00e2e675
DG
903 /* Other fields are zeroed previously */
904
905 ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr,
906 sizeof(data_hdr));
907 if (ret < 0) {
908 goto error;
909 }
910
3604f373
DG
911 ++stream->next_net_seq_num;
912
00e2e675 913 /* Set to go on data socket */
6151a90f 914 outfd = relayd->data_sock.sock.fd;
00e2e675
DG
915 }
916
917error:
918 return outfd;
919}
920
3bd1e081 921/*
ffe60014
DG
922 * Allocate and return a new lttng_consumer_channel object using the given key
923 * to initialize the hash table node.
924 *
925 * On error, return NULL.
3bd1e081 926 */
886224ff 927struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key,
ffe60014
DG
928 uint64_t session_id,
929 const char *pathname,
930 const char *name,
931 uid_t uid,
932 gid_t gid,
57a269f2 933 uint64_t relayd_id,
1624d5b7
JD
934 enum lttng_event_output output,
935 uint64_t tracefile_size,
2bba9e53 936 uint64_t tracefile_count,
1950109e 937 uint64_t session_id_per_pid,
ecc48a90 938 unsigned int monitor,
d7ba1388 939 unsigned int live_timer_interval,
3d071855 940 const char *root_shm_path,
d7ba1388 941 const char *shm_path)
3bd1e081
MD
942{
943 struct lttng_consumer_channel *channel;
3bd1e081 944
276b26d1 945 channel = zmalloc(sizeof(*channel));
3bd1e081 946 if (channel == NULL) {
7a57cf92 947 PERROR("malloc struct lttng_consumer_channel");
3bd1e081
MD
948 goto end;
949 }
ffe60014
DG
950
951 channel->key = key;
3bd1e081 952 channel->refcount = 0;
ffe60014 953 channel->session_id = session_id;
1950109e 954 channel->session_id_per_pid = session_id_per_pid;
ffe60014
DG
955 channel->uid = uid;
956 channel->gid = gid;
957 channel->relayd_id = relayd_id;
1624d5b7
JD
958 channel->tracefile_size = tracefile_size;
959 channel->tracefile_count = tracefile_count;
2bba9e53 960 channel->monitor = monitor;
ecc48a90 961 channel->live_timer_interval = live_timer_interval;
a9838785 962 pthread_mutex_init(&channel->lock, NULL);
ec6ea7d0 963 pthread_mutex_init(&channel->timer_lock, NULL);
ffe60014 964
0c759fc9
DG
965 switch (output) {
966 case LTTNG_EVENT_SPLICE:
967 channel->output = CONSUMER_CHANNEL_SPLICE;
968 break;
969 case LTTNG_EVENT_MMAP:
970 channel->output = CONSUMER_CHANNEL_MMAP;
971 break;
972 default:
973 assert(0);
974 free(channel);
975 channel = NULL;
976 goto end;
977 }
978
07b86b52
JD
979 /*
980 * In monitor mode, the streams associated with the channel will be put in
981 * a special list ONLY owned by this channel. So, the refcount is set to 1
982 * here meaning that the channel itself has streams that are referenced.
983 *
984 * On a channel deletion, once the channel is no longer visible, the
985 * refcount is decremented and checked for a zero value to delete it. With
986 * streams in no monitor mode, it will now be safe to destroy the channel.
987 */
988 if (!channel->monitor) {
989 channel->refcount = 1;
990 }
991
ffe60014
DG
992 strncpy(channel->pathname, pathname, sizeof(channel->pathname));
993 channel->pathname[sizeof(channel->pathname) - 1] = '\0';
994
995 strncpy(channel->name, name, sizeof(channel->name));
996 channel->name[sizeof(channel->name) - 1] = '\0';
997
3d071855
MD
998 if (root_shm_path) {
999 strncpy(channel->root_shm_path, root_shm_path, sizeof(channel->root_shm_path));
1000 channel->root_shm_path[sizeof(channel->root_shm_path) - 1] = '\0';
1001 }
d7ba1388
MD
1002 if (shm_path) {
1003 strncpy(channel->shm_path, shm_path, sizeof(channel->shm_path));
1004 channel->shm_path[sizeof(channel->shm_path) - 1] = '\0';
1005 }
1006
d88aee68 1007 lttng_ht_node_init_u64(&channel->node, channel->key);
d8ef542d
MD
1008
1009 channel->wait_fd = -1;
1010
ffe60014
DG
1011 CDS_INIT_LIST_HEAD(&channel->streams.head);
1012
d88aee68 1013 DBG("Allocated channel (key %" PRIu64 ")", channel->key)
3bd1e081 1014
3bd1e081
MD
1015end:
1016 return channel;
1017}
1018
1019/*
1020 * Add a channel to the global list protected by a mutex.
821fffb2 1021 *
b5a6470f 1022 * Always return 0 indicating success.
3bd1e081 1023 */
d8ef542d
MD
1024int consumer_add_channel(struct lttng_consumer_channel *channel,
1025 struct lttng_consumer_local_data *ctx)
3bd1e081 1026{
3bd1e081 1027 pthread_mutex_lock(&consumer_data.lock);
a9838785 1028 pthread_mutex_lock(&channel->lock);
ec6ea7d0 1029 pthread_mutex_lock(&channel->timer_lock);
c77fc10a 1030
b5a6470f
DG
1031 /*
1032 * This gives us a guarantee that the channel we are about to add to the
1033 * channel hash table will be unique. See this function comment on the why
1034 * we need to steel the channel key at this stage.
1035 */
1036 steal_channel_key(channel->key);
c77fc10a 1037
b5a6470f 1038 rcu_read_lock();
d88aee68 1039 lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node);
6065ceec 1040 rcu_read_unlock();
b5a6470f 1041
ec6ea7d0 1042 pthread_mutex_unlock(&channel->timer_lock);
a9838785 1043 pthread_mutex_unlock(&channel->lock);
3bd1e081 1044 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 1045
b5a6470f 1046 if (channel->wait_fd != -1 && channel->type == CONSUMER_CHANNEL_TYPE_DATA) {
a0cbdd2e 1047 notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD);
d8ef542d 1048 }
b5a6470f
DG
1049
1050 return 0;
3bd1e081
MD
1051}
1052
1053/*
1054 * Allocate the pollfd structure and the local view of the out fds to avoid
1055 * doing a lookup in the linked list and concurrency issues when writing is
1056 * needed. Called with consumer_data.lock held.
1057 *
1058 * Returns the number of fds in the structures.
1059 */
ffe60014
DG
1060static int update_poll_array(struct lttng_consumer_local_data *ctx,
1061 struct pollfd **pollfd, struct lttng_consumer_stream **local_stream,
1062 struct lttng_ht *ht)
3bd1e081 1063{
3bd1e081 1064 int i = 0;
e4421fec
DG
1065 struct lttng_ht_iter iter;
1066 struct lttng_consumer_stream *stream;
3bd1e081 1067
ffe60014
DG
1068 assert(ctx);
1069 assert(ht);
1070 assert(pollfd);
1071 assert(local_stream);
1072
3bd1e081 1073 DBG("Updating poll fd array");
481d6c57 1074 rcu_read_lock();
43c34bc3 1075 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
8994307f
DG
1076 /*
1077 * Only active streams with an active end point can be added to the
1078 * poll set and local stream storage of the thread.
1079 *
1080 * There is a potential race here for endpoint_status to be updated
1081 * just after the check. However, this is OK since the stream(s) will
1082 * be deleted once the thread is notified that the end point state has
1083 * changed where this function will be called back again.
1084 */
1085 if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM ||
79d4ffb7 1086 stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) {
3bd1e081
MD
1087 continue;
1088 }
7972aab2
DG
1089 /*
1090 * This clobbers way too much the debug output. Uncomment that if you
1091 * need it for debugging purposes.
1092 *
1093 * DBG("Active FD %d", stream->wait_fd);
1094 */
e4421fec 1095 (*pollfd)[i].fd = stream->wait_fd;
3bd1e081 1096 (*pollfd)[i].events = POLLIN | POLLPRI;
e4421fec 1097 local_stream[i] = stream;
3bd1e081
MD
1098 i++;
1099 }
481d6c57 1100 rcu_read_unlock();
3bd1e081
MD
1101
1102 /*
50f8ae69 1103 * Insert the consumer_data_pipe at the end of the array and don't
3bd1e081
MD
1104 * increment i so nb_fd is the number of real FD.
1105 */
acdb9057 1106 (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe);
509bb1cf 1107 (*pollfd)[i].events = POLLIN | POLLPRI;
02b3d176
DG
1108
1109 (*pollfd)[i + 1].fd = lttng_pipe_get_readfd(ctx->consumer_wakeup_pipe);
1110 (*pollfd)[i + 1].events = POLLIN | POLLPRI;
3bd1e081
MD
1111 return i;
1112}
1113
1114/*
84382d49
MD
1115 * Poll on the should_quit pipe and the command socket return -1 on
1116 * error, 1 if should exit, 0 if data is available on the command socket
3bd1e081
MD
1117 */
1118int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll)
1119{
1120 int num_rdy;
1121
88f2b785 1122restart:
3bd1e081
MD
1123 num_rdy = poll(consumer_sockpoll, 2, -1);
1124 if (num_rdy == -1) {
88f2b785
MD
1125 /*
1126 * Restart interrupted system call.
1127 */
1128 if (errno == EINTR) {
1129 goto restart;
1130 }
7a57cf92 1131 PERROR("Poll error");
84382d49 1132 return -1;
3bd1e081 1133 }
509bb1cf 1134 if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) {
3bd1e081 1135 DBG("consumer_should_quit wake up");
84382d49 1136 return 1;
3bd1e081
MD
1137 }
1138 return 0;
3bd1e081
MD
1139}
1140
1141/*
1142 * Set the error socket.
1143 */
ffe60014
DG
1144void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx,
1145 int sock)
3bd1e081
MD
1146{
1147 ctx->consumer_error_socket = sock;
1148}
1149
1150/*
1151 * Set the command socket path.
1152 */
3bd1e081
MD
1153void lttng_consumer_set_command_sock_path(
1154 struct lttng_consumer_local_data *ctx, char *sock)
1155{
1156 ctx->consumer_command_sock_path = sock;
1157}
1158
1159/*
1160 * Send return code to the session daemon.
1161 * If the socket is not defined, we return 0, it is not a fatal error
1162 */
ffe60014 1163int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd)
3bd1e081
MD
1164{
1165 if (ctx->consumer_error_socket > 0) {
1166 return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd,
1167 sizeof(enum lttcomm_sessiond_command));
1168 }
1169
1170 return 0;
1171}
1172
1173/*
228b5bf7
DG
1174 * Close all the tracefiles and stream fds and MUST be called when all
1175 * instances are destroyed i.e. when all threads were joined and are ended.
3bd1e081
MD
1176 */
1177void lttng_consumer_cleanup(void)
1178{
e4421fec 1179 struct lttng_ht_iter iter;
ffe60014 1180 struct lttng_consumer_channel *channel;
6065ceec
DG
1181
1182 rcu_read_lock();
3bd1e081 1183
ffe60014
DG
1184 cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel,
1185 node.node) {
702b1ea4 1186 consumer_del_channel(channel);
3bd1e081 1187 }
6065ceec
DG
1188
1189 rcu_read_unlock();
d6ce1df2 1190
d6ce1df2 1191 lttng_ht_destroy(consumer_data.channel_ht);
228b5bf7
DG
1192
1193 cleanup_relayd_ht();
1194
d8ef542d
MD
1195 lttng_ht_destroy(consumer_data.stream_per_chan_id_ht);
1196
228b5bf7
DG
1197 /*
1198 * This HT contains streams that are freed by either the metadata thread or
1199 * the data thread so we do *nothing* on the hash table and simply destroy
1200 * it.
1201 */
1202 lttng_ht_destroy(consumer_data.stream_list_ht);
3bd1e081
MD
1203}
1204
1205/*
1206 * Called from signal handler.
1207 */
1208void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx)
1209{
6cd525e8
MD
1210 ssize_t ret;
1211
3bd1e081 1212 consumer_quit = 1;
6cd525e8
MD
1213 ret = lttng_write(ctx->consumer_should_quit[1], "4", 1);
1214 if (ret < 1) {
7a57cf92 1215 PERROR("write consumer quit");
3bd1e081 1216 }
ab1027f4
DG
1217
1218 DBG("Consumer flag that it should quit");
3bd1e081
MD
1219}
1220
00e2e675
DG
1221void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream,
1222 off_t orig_offset)
3bd1e081
MD
1223{
1224 int outfd = stream->out_fd;
1225
1226 /*
1227 * This does a blocking write-and-wait on any page that belongs to the
1228 * subbuffer prior to the one we just wrote.
1229 * Don't care about error values, as these are just hints and ways to
1230 * limit the amount of page cache used.
1231 */
ffe60014 1232 if (orig_offset < stream->max_sb_size) {
3bd1e081
MD
1233 return;
1234 }
ffe60014
DG
1235 lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size,
1236 stream->max_sb_size,
3bd1e081
MD
1237 SYNC_FILE_RANGE_WAIT_BEFORE
1238 | SYNC_FILE_RANGE_WRITE
1239 | SYNC_FILE_RANGE_WAIT_AFTER);
1240 /*
1241 * Give hints to the kernel about how we access the file:
1242 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1243 * we write it.
1244 *
1245 * We need to call fadvise again after the file grows because the
1246 * kernel does not seem to apply fadvise to non-existing parts of the
1247 * file.
1248 *
1249 * Call fadvise _after_ having waited for the page writeback to
1250 * complete because the dirty page writeback semantic is not well
1251 * defined. So it can be expected to lead to lower throughput in
1252 * streaming.
1253 */
ffe60014
DG
1254 posix_fadvise(outfd, orig_offset - stream->max_sb_size,
1255 stream->max_sb_size, POSIX_FADV_DONTNEED);
3bd1e081
MD
1256}
1257
1258/*
1259 * Initialise the necessary environnement :
1260 * - create a new context
1261 * - create the poll_pipe
1262 * - create the should_quit pipe (for signal handler)
1263 * - create the thread pipe (for splice)
1264 *
1265 * Takes a function pointer as argument, this function is called when data is
1266 * available on a buffer. This function is responsible to do the
1267 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1268 * buffer configuration and then kernctl_put_next_subbuf at the end.
1269 *
1270 * Returns a pointer to the new context or NULL on error.
1271 */
1272struct lttng_consumer_local_data *lttng_consumer_create(
1273 enum lttng_consumer_type type,
4078b776 1274 ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream,
d41f73b7 1275 struct lttng_consumer_local_data *ctx),
3bd1e081
MD
1276 int (*recv_channel)(struct lttng_consumer_channel *channel),
1277 int (*recv_stream)(struct lttng_consumer_stream *stream),
30319bcb 1278 int (*update_stream)(uint64_t stream_key, uint32_t state))
3bd1e081 1279{
d8ef542d 1280 int ret;
3bd1e081
MD
1281 struct lttng_consumer_local_data *ctx;
1282
1283 assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN ||
1284 consumer_data.type == type);
1285 consumer_data.type = type;
1286
effcf122 1287 ctx = zmalloc(sizeof(struct lttng_consumer_local_data));
3bd1e081 1288 if (ctx == NULL) {
7a57cf92 1289 PERROR("allocating context");
3bd1e081
MD
1290 goto error;
1291 }
1292
1293 ctx->consumer_error_socket = -1;
331744e3 1294 ctx->consumer_metadata_socket = -1;
75d83e50 1295 pthread_mutex_init(&ctx->metadata_socket_lock, NULL);
3bd1e081
MD
1296 /* assign the callbacks */
1297 ctx->on_buffer_ready = buffer_ready;
1298 ctx->on_recv_channel = recv_channel;
1299 ctx->on_recv_stream = recv_stream;
1300 ctx->on_update_stream = update_stream;
1301
acdb9057
DG
1302 ctx->consumer_data_pipe = lttng_pipe_open(0);
1303 if (!ctx->consumer_data_pipe) {
3bd1e081
MD
1304 goto error_poll_pipe;
1305 }
1306
02b3d176
DG
1307 ctx->consumer_wakeup_pipe = lttng_pipe_open(0);
1308 if (!ctx->consumer_wakeup_pipe) {
1309 goto error_wakeup_pipe;
1310 }
1311
3bd1e081
MD
1312 ret = pipe(ctx->consumer_should_quit);
1313 if (ret < 0) {
7a57cf92 1314 PERROR("Error creating recv pipe");
3bd1e081
MD
1315 goto error_quit_pipe;
1316 }
1317
d8ef542d
MD
1318 ret = pipe(ctx->consumer_channel_pipe);
1319 if (ret < 0) {
1320 PERROR("Error creating channel pipe");
1321 goto error_channel_pipe;
1322 }
1323
13886d2d
DG
1324 ctx->consumer_metadata_pipe = lttng_pipe_open(0);
1325 if (!ctx->consumer_metadata_pipe) {
fb3a43a9
DG
1326 goto error_metadata_pipe;
1327 }
3bd1e081 1328
fb3a43a9 1329 return ctx;
3bd1e081 1330
fb3a43a9 1331error_metadata_pipe:
d8ef542d
MD
1332 utils_close_pipe(ctx->consumer_channel_pipe);
1333error_channel_pipe:
d8ef542d 1334 utils_close_pipe(ctx->consumer_should_quit);
3bd1e081 1335error_quit_pipe:
02b3d176
DG
1336 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
1337error_wakeup_pipe:
acdb9057 1338 lttng_pipe_destroy(ctx->consumer_data_pipe);
3bd1e081
MD
1339error_poll_pipe:
1340 free(ctx);
1341error:
1342 return NULL;
1343}
1344
282dadbc
MD
1345/*
1346 * Iterate over all streams of the hashtable and free them properly.
1347 */
1348static void destroy_data_stream_ht(struct lttng_ht *ht)
1349{
1350 struct lttng_ht_iter iter;
1351 struct lttng_consumer_stream *stream;
1352
1353 if (ht == NULL) {
1354 return;
1355 }
1356
1357 rcu_read_lock();
1358 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1359 /*
1360 * Ignore return value since we are currently cleaning up so any error
1361 * can't be handled.
1362 */
1363 (void) consumer_del_stream(stream, ht);
1364 }
1365 rcu_read_unlock();
1366
1367 lttng_ht_destroy(ht);
1368}
1369
1370/*
1371 * Iterate over all streams of the metadata hashtable and free them
1372 * properly.
1373 */
1374static void destroy_metadata_stream_ht(struct lttng_ht *ht)
1375{
1376 struct lttng_ht_iter iter;
1377 struct lttng_consumer_stream *stream;
1378
1379 if (ht == NULL) {
1380 return;
1381 }
1382
1383 rcu_read_lock();
1384 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
1385 /*
1386 * Ignore return value since we are currently cleaning up so any error
1387 * can't be handled.
1388 */
1389 (void) consumer_del_metadata_stream(stream, ht);
1390 }
1391 rcu_read_unlock();
1392
1393 lttng_ht_destroy(ht);
1394}
1395
3bd1e081
MD
1396/*
1397 * Close all fds associated with the instance and free the context.
1398 */
1399void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx)
1400{
4c462e79
MD
1401 int ret;
1402
ab1027f4
DG
1403 DBG("Consumer destroying it. Closing everything.");
1404
4f2e75b9
DG
1405 if (!ctx) {
1406 return;
1407 }
1408
282dadbc
MD
1409 destroy_data_stream_ht(data_ht);
1410 destroy_metadata_stream_ht(metadata_ht);
1411
4c462e79
MD
1412 ret = close(ctx->consumer_error_socket);
1413 if (ret) {
1414 PERROR("close");
1415 }
331744e3
JD
1416 ret = close(ctx->consumer_metadata_socket);
1417 if (ret) {
1418 PERROR("close");
1419 }
d8ef542d 1420 utils_close_pipe(ctx->consumer_channel_pipe);
acdb9057 1421 lttng_pipe_destroy(ctx->consumer_data_pipe);
13886d2d 1422 lttng_pipe_destroy(ctx->consumer_metadata_pipe);
02b3d176 1423 lttng_pipe_destroy(ctx->consumer_wakeup_pipe);
d8ef542d 1424 utils_close_pipe(ctx->consumer_should_quit);
fb3a43a9 1425
3bd1e081
MD
1426 unlink(ctx->consumer_command_sock_path);
1427 free(ctx);
1428}
1429
6197aea7
DG
1430/*
1431 * Write the metadata stream id on the specified file descriptor.
1432 */
1433static int write_relayd_metadata_id(int fd,
1434 struct lttng_consumer_stream *stream,
ffe60014 1435 struct consumer_relayd_sock_pair *relayd, unsigned long padding)
6197aea7 1436{
6cd525e8 1437 ssize_t ret;
1d4dfdef 1438 struct lttcomm_relayd_metadata_payload hdr;
6197aea7 1439
1d4dfdef
DG
1440 hdr.stream_id = htobe64(stream->relayd_stream_id);
1441 hdr.padding_size = htobe32(padding);
6cd525e8
MD
1442 ret = lttng_write(fd, (void *) &hdr, sizeof(hdr));
1443 if (ret < sizeof(hdr)) {
d7b75ec8 1444 /*
6f04ed72 1445 * This error means that the fd's end is closed so ignore the PERROR
d7b75ec8
DG
1446 * not to clubber the error output since this can happen in a normal
1447 * code path.
1448 */
1449 if (errno != EPIPE) {
1450 PERROR("write metadata stream id");
1451 }
1452 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno);
534d2592
DG
1453 /*
1454 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1455 * handle writting the missing part so report that as an error and
1456 * don't lie to the caller.
1457 */
1458 ret = -1;
6197aea7
DG
1459 goto end;
1460 }
1d4dfdef
DG
1461 DBG("Metadata stream id %" PRIu64 " with padding %lu written before data",
1462 stream->relayd_stream_id, padding);
6197aea7
DG
1463
1464end:
6cd525e8 1465 return (int) ret;
6197aea7
DG
1466}
1467
3bd1e081 1468/*
09e26845
DG
1469 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1470 * core function for writing trace buffers to either the local filesystem or
1471 * the network.
1472 *
79d4ffb7
DG
1473 * It must be called with the stream lock held.
1474 *
09e26845 1475 * Careful review MUST be put if any changes occur!
3bd1e081
MD
1476 *
1477 * Returns the number of bytes written
1478 */
4078b776 1479ssize_t lttng_consumer_on_read_subbuffer_mmap(
3bd1e081 1480 struct lttng_consumer_local_data *ctx,
1d4dfdef 1481 struct lttng_consumer_stream *stream, unsigned long len,
309167d2 1482 unsigned long padding,
50adc264 1483 struct ctf_packet_index *index)
3bd1e081 1484{
f02e1e8a 1485 unsigned long mmap_offset;
ffe60014 1486 void *mmap_base;
994ab360 1487 ssize_t ret = 0;
f02e1e8a
DG
1488 off_t orig_offset = stream->out_fd_offset;
1489 /* Default is on the disk */
1490 int outfd = stream->out_fd;
f02e1e8a 1491 struct consumer_relayd_sock_pair *relayd = NULL;
8994307f 1492 unsigned int relayd_hang_up = 0;
f02e1e8a
DG
1493
1494 /* RCU lock for the relayd pointer */
1495 rcu_read_lock();
1496
1497 /* Flag that the current stream if set for network streaming. */
da009f2c 1498 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1499 relayd = consumer_find_relayd(stream->net_seq_idx);
1500 if (relayd == NULL) {
56591bac 1501 ret = -EPIPE;
f02e1e8a
DG
1502 goto end;
1503 }
1504 }
1505
1506 /* get the offset inside the fd to mmap */
3bd1e081
MD
1507 switch (consumer_data.type) {
1508 case LTTNG_CONSUMER_KERNEL:
ffe60014 1509 mmap_base = stream->mmap_base;
f02e1e8a 1510 ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset);
994ab360
DG
1511 if (ret < 0) {
1512 ret = -errno;
56591bac 1513 PERROR("tracer ctl get_mmap_read_offset");
56591bac
MD
1514 goto end;
1515 }
f02e1e8a 1516 break;
7753dea8
MD
1517 case LTTNG_CONSUMER32_UST:
1518 case LTTNG_CONSUMER64_UST:
ffe60014
DG
1519 mmap_base = lttng_ustctl_get_mmap_base(stream);
1520 if (!mmap_base) {
1521 ERR("read mmap get mmap base for stream %s", stream->name);
994ab360 1522 ret = -EPERM;
ffe60014
DG
1523 goto end;
1524 }
1525 ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset);
56591bac
MD
1526 if (ret != 0) {
1527 PERROR("tracer ctl get_mmap_read_offset");
994ab360 1528 ret = -EINVAL;
56591bac
MD
1529 goto end;
1530 }
f02e1e8a 1531 break;
3bd1e081
MD
1532 default:
1533 ERR("Unknown consumer_data type");
1534 assert(0);
1535 }
b9182dd9 1536
f02e1e8a
DG
1537 /* Handle stream on the relayd if the output is on the network */
1538 if (relayd) {
1539 unsigned long netlen = len;
1540
1541 /*
1542 * Lock the control socket for the complete duration of the function
1543 * since from this point on we will use the socket.
1544 */
1545 if (stream->metadata_flag) {
1546 /* Metadata requires the control socket. */
1547 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1d4dfdef 1548 netlen += sizeof(struct lttcomm_relayd_metadata_payload);
f02e1e8a
DG
1549 }
1550
1d4dfdef 1551 ret = write_relayd_stream_header(stream, netlen, padding, relayd);
994ab360
DG
1552 if (ret < 0) {
1553 relayd_hang_up = 1;
1554 goto write_error;
1555 }
1556 /* Use the returned socket. */
1557 outfd = ret;
f02e1e8a 1558
994ab360
DG
1559 /* Write metadata stream id before payload */
1560 if (stream->metadata_flag) {
1561 ret = write_relayd_metadata_id(outfd, stream, relayd, padding);
1562 if (ret < 0) {
8994307f
DG
1563 relayd_hang_up = 1;
1564 goto write_error;
1565 }
f02e1e8a 1566 }
1d4dfdef
DG
1567 } else {
1568 /* No streaming, we have to set the len with the full padding */
1569 len += padding;
1624d5b7
JD
1570
1571 /*
1572 * Check if we need to change the tracefile before writing the packet.
1573 */
1574 if (stream->chan->tracefile_size > 0 &&
1575 (stream->tracefile_size_current + len) >
1576 stream->chan->tracefile_size) {
fe4477ee
JD
1577 ret = utils_rotate_stream_file(stream->chan->pathname,
1578 stream->name, stream->chan->tracefile_size,
1579 stream->chan->tracefile_count, stream->uid, stream->gid,
309167d2
JD
1580 stream->out_fd, &(stream->tracefile_count_current),
1581 &stream->out_fd);
1624d5b7
JD
1582 if (ret < 0) {
1583 ERR("Rotating output file");
1584 goto end;
1585 }
309167d2
JD
1586 outfd = stream->out_fd;
1587
1588 if (stream->index_fd >= 0) {
1589 ret = index_create_file(stream->chan->pathname,
1590 stream->name, stream->uid, stream->gid,
1591 stream->chan->tracefile_size,
1592 stream->tracefile_count_current);
1593 if (ret < 0) {
1594 goto end;
1595 }
1596 stream->index_fd = ret;
1597 }
1598
a6976990
DG
1599 /* Reset current size because we just perform a rotation. */
1600 stream->tracefile_size_current = 0;
a1ae300f
JD
1601 stream->out_fd_offset = 0;
1602 orig_offset = 0;
1624d5b7
JD
1603 }
1604 stream->tracefile_size_current += len;
309167d2
JD
1605 if (index) {
1606 index->offset = htobe64(stream->out_fd_offset);
1607 }
f02e1e8a
DG
1608 }
1609
d02b8372
DG
1610 /*
1611 * This call guarantee that len or less is returned. It's impossible to
1612 * receive a ret value that is bigger than len.
1613 */
1614 ret = lttng_write(outfd, mmap_base + mmap_offset, len);
1615 DBG("Consumer mmap write() ret %zd (len %lu)", ret, len);
1616 if (ret < 0 || ((size_t) ret != len)) {
1617 /*
1618 * Report error to caller if nothing was written else at least send the
1619 * amount written.
1620 */
1621 if (ret < 0) {
994ab360 1622 ret = -errno;
f02e1e8a 1623 }
994ab360 1624 relayd_hang_up = 1;
f02e1e8a 1625
d02b8372 1626 /* Socket operation failed. We consider the relayd dead */
994ab360 1627 if (errno == EPIPE || errno == EINVAL || errno == EBADF) {
d02b8372
DG
1628 /*
1629 * This is possible if the fd is closed on the other side
1630 * (outfd) or any write problem. It can be verbose a bit for a
1631 * normal execution if for instance the relayd is stopped
1632 * abruptly. This can happen so set this to a DBG statement.
1633 */
1634 DBG("Consumer mmap write detected relayd hang up");
994ab360
DG
1635 } else {
1636 /* Unhandled error, print it and stop function right now. */
1637 PERROR("Error in write mmap (ret %zd != len %lu)", ret, len);
f02e1e8a 1638 }
994ab360 1639 goto write_error;
d02b8372
DG
1640 }
1641 stream->output_written += ret;
d02b8372
DG
1642
1643 /* This call is useless on a socket so better save a syscall. */
1644 if (!relayd) {
1645 /* This won't block, but will start writeout asynchronously */
1646 lttng_sync_file_range(outfd, stream->out_fd_offset, len,
1647 SYNC_FILE_RANGE_WRITE);
1648 stream->out_fd_offset += len;
f02e1e8a
DG
1649 }
1650 lttng_consumer_sync_trace_file(stream, orig_offset);
1651
8994307f
DG
1652write_error:
1653 /*
1654 * This is a special case that the relayd has closed its socket. Let's
1655 * cleanup the relayd object and all associated streams.
1656 */
1657 if (relayd && relayd_hang_up) {
1658 cleanup_relayd(relayd, ctx);
1659 }
1660
f02e1e8a
DG
1661end:
1662 /* Unlock only if ctrl socket used */
1663 if (relayd && stream->metadata_flag) {
1664 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1665 }
1666
1667 rcu_read_unlock();
994ab360 1668 return ret;
3bd1e081
MD
1669}
1670
1671/*
1672 * Splice the data from the ring buffer to the tracefile.
1673 *
79d4ffb7
DG
1674 * It must be called with the stream lock held.
1675 *
3bd1e081
MD
1676 * Returns the number of bytes spliced.
1677 */
4078b776 1678ssize_t lttng_consumer_on_read_subbuffer_splice(
3bd1e081 1679 struct lttng_consumer_local_data *ctx,
1d4dfdef 1680 struct lttng_consumer_stream *stream, unsigned long len,
309167d2 1681 unsigned long padding,
50adc264 1682 struct ctf_packet_index *index)
3bd1e081 1683{
f02e1e8a
DG
1684 ssize_t ret = 0, written = 0, ret_splice = 0;
1685 loff_t offset = 0;
1686 off_t orig_offset = stream->out_fd_offset;
1687 int fd = stream->wait_fd;
1688 /* Default is on the disk */
1689 int outfd = stream->out_fd;
f02e1e8a 1690 struct consumer_relayd_sock_pair *relayd = NULL;
fb3a43a9 1691 int *splice_pipe;
8994307f 1692 unsigned int relayd_hang_up = 0;
f02e1e8a 1693
3bd1e081
MD
1694 switch (consumer_data.type) {
1695 case LTTNG_CONSUMER_KERNEL:
f02e1e8a 1696 break;
7753dea8
MD
1697 case LTTNG_CONSUMER32_UST:
1698 case LTTNG_CONSUMER64_UST:
f02e1e8a 1699 /* Not supported for user space tracing */
3bd1e081
MD
1700 return -ENOSYS;
1701 default:
1702 ERR("Unknown consumer_data type");
1703 assert(0);
3bd1e081
MD
1704 }
1705
f02e1e8a
DG
1706 /* RCU lock for the relayd pointer */
1707 rcu_read_lock();
1708
1709 /* Flag that the current stream if set for network streaming. */
da009f2c 1710 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1711 relayd = consumer_find_relayd(stream->net_seq_idx);
1712 if (relayd == NULL) {
ad0b0d23 1713 written = -ret;
f02e1e8a
DG
1714 goto end;
1715 }
1716 }
a2361a61 1717 splice_pipe = stream->splice_pipe;
fb3a43a9 1718
f02e1e8a 1719 /* Write metadata stream id before payload */
1d4dfdef 1720 if (relayd) {
ad0b0d23 1721 unsigned long total_len = len;
f02e1e8a 1722
1d4dfdef
DG
1723 if (stream->metadata_flag) {
1724 /*
1725 * Lock the control socket for the complete duration of the function
1726 * since from this point on we will use the socket.
1727 */
1728 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1729
1730 ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd,
1731 padding);
1732 if (ret < 0) {
1733 written = ret;
ad0b0d23
DG
1734 relayd_hang_up = 1;
1735 goto write_error;
1d4dfdef
DG
1736 }
1737
1738 total_len += sizeof(struct lttcomm_relayd_metadata_payload);
1739 }
1740
1741 ret = write_relayd_stream_header(stream, total_len, padding, relayd);
ad0b0d23
DG
1742 if (ret < 0) {
1743 written = ret;
1744 relayd_hang_up = 1;
1745 goto write_error;
f02e1e8a 1746 }
ad0b0d23
DG
1747 /* Use the returned socket. */
1748 outfd = ret;
1d4dfdef
DG
1749 } else {
1750 /* No streaming, we have to set the len with the full padding */
1751 len += padding;
1624d5b7
JD
1752
1753 /*
1754 * Check if we need to change the tracefile before writing the packet.
1755 */
1756 if (stream->chan->tracefile_size > 0 &&
1757 (stream->tracefile_size_current + len) >
1758 stream->chan->tracefile_size) {
fe4477ee
JD
1759 ret = utils_rotate_stream_file(stream->chan->pathname,
1760 stream->name, stream->chan->tracefile_size,
1761 stream->chan->tracefile_count, stream->uid, stream->gid,
309167d2
JD
1762 stream->out_fd, &(stream->tracefile_count_current),
1763 &stream->out_fd);
1624d5b7 1764 if (ret < 0) {
ad0b0d23 1765 written = ret;
1624d5b7
JD
1766 ERR("Rotating output file");
1767 goto end;
1768 }
309167d2
JD
1769 outfd = stream->out_fd;
1770
1771 if (stream->index_fd >= 0) {
1772 ret = index_create_file(stream->chan->pathname,
1773 stream->name, stream->uid, stream->gid,
1774 stream->chan->tracefile_size,
1775 stream->tracefile_count_current);
1776 if (ret < 0) {
ad0b0d23 1777 written = ret;
309167d2
JD
1778 goto end;
1779 }
1780 stream->index_fd = ret;
1781 }
1782
a6976990
DG
1783 /* Reset current size because we just perform a rotation. */
1784 stream->tracefile_size_current = 0;
a1ae300f
JD
1785 stream->out_fd_offset = 0;
1786 orig_offset = 0;
1624d5b7
JD
1787 }
1788 stream->tracefile_size_current += len;
309167d2 1789 index->offset = htobe64(stream->out_fd_offset);
f02e1e8a
DG
1790 }
1791
1792 while (len > 0) {
1d4dfdef
DG
1793 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1794 (unsigned long)offset, len, fd, splice_pipe[1]);
fb3a43a9 1795 ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len,
f02e1e8a
DG
1796 SPLICE_F_MOVE | SPLICE_F_MORE);
1797 DBG("splice chan to pipe, ret %zd", ret_splice);
1798 if (ret_splice < 0) {
d02b8372 1799 ret = errno;
ad0b0d23 1800 written = -ret;
d02b8372 1801 PERROR("Error in relay splice");
f02e1e8a
DG
1802 goto splice_error;
1803 }
1804
1805 /* Handle stream on the relayd if the output is on the network */
ad0b0d23
DG
1806 if (relayd && stream->metadata_flag) {
1807 size_t metadata_payload_size =
1808 sizeof(struct lttcomm_relayd_metadata_payload);
1809
1810 /* Update counter to fit the spliced data */
1811 ret_splice += metadata_payload_size;
1812 len += metadata_payload_size;
1813 /*
1814 * We do this so the return value can match the len passed as
1815 * argument to this function.
1816 */
1817 written -= metadata_payload_size;
f02e1e8a
DG
1818 }
1819
1820 /* Splice data out */
fb3a43a9 1821 ret_splice = splice(splice_pipe[0], NULL, outfd, NULL,
f02e1e8a 1822 ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE);
a2361a61
JD
1823 DBG("Consumer splice pipe to file (out_fd: %d), ret %zd",
1824 outfd, ret_splice);
f02e1e8a 1825 if (ret_splice < 0) {
d02b8372 1826 ret = errno;
ad0b0d23
DG
1827 written = -ret;
1828 relayd_hang_up = 1;
1829 goto write_error;
f02e1e8a 1830 } else if (ret_splice > len) {
d02b8372
DG
1831 /*
1832 * We don't expect this code path to be executed but you never know
1833 * so this is an extra protection agains a buggy splice().
1834 */
f02e1e8a 1835 ret = errno;
ad0b0d23 1836 written += ret_splice;
d02b8372
DG
1837 PERROR("Wrote more data than requested %zd (len: %lu)", ret_splice,
1838 len);
f02e1e8a 1839 goto splice_error;
d02b8372
DG
1840 } else {
1841 /* All good, update current len and continue. */
1842 len -= ret_splice;
f02e1e8a 1843 }
f02e1e8a
DG
1844
1845 /* This call is useless on a socket so better save a syscall. */
1846 if (!relayd) {
1847 /* This won't block, but will start writeout asynchronously */
1848 lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice,
1849 SYNC_FILE_RANGE_WRITE);
1850 stream->out_fd_offset += ret_splice;
1851 }
e5d1a9b3 1852 stream->output_written += ret_splice;
f02e1e8a
DG
1853 written += ret_splice;
1854 }
1855 lttng_consumer_sync_trace_file(stream, orig_offset);
f02e1e8a
DG
1856 goto end;
1857
8994307f
DG
1858write_error:
1859 /*
1860 * This is a special case that the relayd has closed its socket. Let's
1861 * cleanup the relayd object and all associated streams.
1862 */
1863 if (relayd && relayd_hang_up) {
1864 cleanup_relayd(relayd, ctx);
1865 /* Skip splice error so the consumer does not fail */
1866 goto end;
1867 }
1868
f02e1e8a
DG
1869splice_error:
1870 /* send the appropriate error description to sessiond */
1871 switch (ret) {
f02e1e8a 1872 case EINVAL:
f73fabfd 1873 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL);
f02e1e8a
DG
1874 break;
1875 case ENOMEM:
f73fabfd 1876 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM);
f02e1e8a
DG
1877 break;
1878 case ESPIPE:
f73fabfd 1879 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE);
f02e1e8a
DG
1880 break;
1881 }
1882
1883end:
1884 if (relayd && stream->metadata_flag) {
1885 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1886 }
1887
1888 rcu_read_unlock();
1889 return written;
3bd1e081
MD
1890}
1891
1892/*
1893 * Take a snapshot for a specific fd
1894 *
1895 * Returns 0 on success, < 0 on error
1896 */
ffe60014 1897int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream)
3bd1e081
MD
1898{
1899 switch (consumer_data.type) {
1900 case LTTNG_CONSUMER_KERNEL:
ffe60014 1901 return lttng_kconsumer_take_snapshot(stream);
7753dea8
MD
1902 case LTTNG_CONSUMER32_UST:
1903 case LTTNG_CONSUMER64_UST:
ffe60014 1904 return lttng_ustconsumer_take_snapshot(stream);
3bd1e081
MD
1905 default:
1906 ERR("Unknown consumer_data type");
1907 assert(0);
1908 return -ENOSYS;
1909 }
3bd1e081
MD
1910}
1911
1912/*
1913 * Get the produced position
1914 *
1915 * Returns 0 on success, < 0 on error
1916 */
ffe60014 1917int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
3bd1e081
MD
1918 unsigned long *pos)
1919{
1920 switch (consumer_data.type) {
1921 case LTTNG_CONSUMER_KERNEL:
ffe60014 1922 return lttng_kconsumer_get_produced_snapshot(stream, pos);
7753dea8
MD
1923 case LTTNG_CONSUMER32_UST:
1924 case LTTNG_CONSUMER64_UST:
ffe60014 1925 return lttng_ustconsumer_get_produced_snapshot(stream, pos);
3bd1e081
MD
1926 default:
1927 ERR("Unknown consumer_data type");
1928 assert(0);
1929 return -ENOSYS;
1930 }
1931}
1932
1933int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1934 int sock, struct pollfd *consumer_sockpoll)
1935{
1936 switch (consumer_data.type) {
1937 case LTTNG_CONSUMER_KERNEL:
1938 return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
7753dea8
MD
1939 case LTTNG_CONSUMER32_UST:
1940 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
1941 return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
1942 default:
1943 ERR("Unknown consumer_data type");
1944 assert(0);
1945 return -ENOSYS;
1946 }
1947}
1948
6d574024 1949void lttng_consumer_close_all_metadata(void)
d88aee68
DG
1950{
1951 switch (consumer_data.type) {
1952 case LTTNG_CONSUMER_KERNEL:
1953 /*
1954 * The Kernel consumer has a different metadata scheme so we don't
1955 * close anything because the stream will be closed by the session
1956 * daemon.
1957 */
1958 break;
1959 case LTTNG_CONSUMER32_UST:
1960 case LTTNG_CONSUMER64_UST:
1961 /*
1962 * Close all metadata streams. The metadata hash table is passed and
1963 * this call iterates over it by closing all wakeup fd. This is safe
1964 * because at this point we are sure that the metadata producer is
1965 * either dead or blocked.
1966 */
6d574024 1967 lttng_ustconsumer_close_all_metadata(metadata_ht);
d88aee68
DG
1968 break;
1969 default:
1970 ERR("Unknown consumer_data type");
1971 assert(0);
1972 }
1973}
1974
fb3a43a9
DG
1975/*
1976 * Clean up a metadata stream and free its memory.
1977 */
e316aad5
DG
1978void consumer_del_metadata_stream(struct lttng_consumer_stream *stream,
1979 struct lttng_ht *ht)
fb3a43a9 1980{
e316aad5 1981 struct lttng_consumer_channel *free_chan = NULL;
fb3a43a9
DG
1982
1983 assert(stream);
1984 /*
1985 * This call should NEVER receive regular stream. It must always be
1986 * metadata stream and this is crucial for data structure synchronization.
1987 */
1988 assert(stream->metadata_flag);
1989
e316aad5
DG
1990 DBG3("Consumer delete metadata stream %d", stream->wait_fd);
1991
74251bb8 1992 pthread_mutex_lock(&consumer_data.lock);
a9838785 1993 pthread_mutex_lock(&stream->chan->lock);
8994307f
DG
1994 pthread_mutex_lock(&stream->lock);
1995
6d574024
DG
1996 /* Remove any reference to that stream. */
1997 consumer_stream_delete(stream, ht);
ca22feea 1998
6d574024
DG
1999 /* Close down everything including the relayd if one. */
2000 consumer_stream_close(stream);
2001 /* Destroy tracer buffers of the stream. */
2002 consumer_stream_destroy_buffers(stream);
fb3a43a9
DG
2003
2004 /* Atomically decrement channel refcount since other threads can use it. */
f2ad556d 2005 if (!uatomic_sub_return(&stream->chan->refcount, 1)
ffe60014 2006 && !uatomic_read(&stream->chan->nb_init_stream_left)) {
c30aaa51 2007 /* Go for channel deletion! */
e316aad5 2008 free_chan = stream->chan;
fb3a43a9
DG
2009 }
2010
73811ecc
DG
2011 /*
2012 * Nullify the stream reference so it is not used after deletion. The
6d574024
DG
2013 * channel lock MUST be acquired before being able to check for a NULL
2014 * pointer value.
73811ecc
DG
2015 */
2016 stream->chan->metadata_stream = NULL;
2017
8994307f 2018 pthread_mutex_unlock(&stream->lock);
a9838785 2019 pthread_mutex_unlock(&stream->chan->lock);
74251bb8 2020 pthread_mutex_unlock(&consumer_data.lock);
e316aad5
DG
2021
2022 if (free_chan) {
2023 consumer_del_channel(free_chan);
2024 }
2025
6d574024 2026 consumer_stream_free(stream);
fb3a43a9
DG
2027}
2028
2029/*
2030 * Action done with the metadata stream when adding it to the consumer internal
2031 * data structures to handle it.
2032 */
5ab66908 2033int consumer_add_metadata_stream(struct lttng_consumer_stream *stream)
fb3a43a9 2034{
5ab66908 2035 struct lttng_ht *ht = metadata_ht;
e316aad5 2036 int ret = 0;
76082088 2037 struct lttng_ht_iter iter;
d88aee68 2038 struct lttng_ht_node_u64 *node;
fb3a43a9 2039
e316aad5
DG
2040 assert(stream);
2041 assert(ht);
2042
d88aee68 2043 DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key);
e316aad5
DG
2044
2045 pthread_mutex_lock(&consumer_data.lock);
a9838785 2046 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 2047 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 2048 pthread_mutex_lock(&stream->lock);
e316aad5 2049
e316aad5
DG
2050 /*
2051 * From here, refcounts are updated so be _careful_ when returning an error
2052 * after this point.
2053 */
2054
fb3a43a9 2055 rcu_read_lock();
76082088
DG
2056
2057 /*
2058 * Lookup the stream just to make sure it does not exist in our internal
2059 * state. This should NEVER happen.
2060 */
d88aee68
DG
2061 lttng_ht_lookup(ht, &stream->key, &iter);
2062 node = lttng_ht_iter_get_node_u64(&iter);
76082088
DG
2063 assert(!node);
2064
e316aad5 2065 /*
ffe60014
DG
2066 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2067 * in terms of destroying the associated channel, because the action that
e316aad5
DG
2068 * causes the count to become 0 also causes a stream to be added. The
2069 * channel deletion will thus be triggered by the following removal of this
2070 * stream.
2071 */
ffe60014 2072 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
2073 /* Increment refcount before decrementing nb_init_stream_left */
2074 cmm_smp_wmb();
ffe60014 2075 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
2076 }
2077
d88aee68 2078 lttng_ht_add_unique_u64(ht, &stream->node);
ca22feea 2079
d8ef542d
MD
2080 lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht,
2081 &stream->node_channel_id);
2082
ca22feea
DG
2083 /*
2084 * Add stream to the stream_list_ht of the consumer data. No need to steal
2085 * the key since the HT does not use it and we allow to add redundant keys
2086 * into this table.
2087 */
d88aee68 2088 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 2089
fb3a43a9 2090 rcu_read_unlock();
e316aad5 2091
2e818a6a 2092 pthread_mutex_unlock(&stream->lock);
a9838785 2093 pthread_mutex_unlock(&stream->chan->lock);
ec6ea7d0 2094 pthread_mutex_unlock(&stream->chan->timer_lock);
e316aad5
DG
2095 pthread_mutex_unlock(&consumer_data.lock);
2096 return ret;
fb3a43a9
DG
2097}
2098
8994307f
DG
2099/*
2100 * Delete data stream that are flagged for deletion (endpoint_status).
2101 */
2102static void validate_endpoint_status_data_stream(void)
2103{
2104 struct lttng_ht_iter iter;
2105 struct lttng_consumer_stream *stream;
2106
2107 DBG("Consumer delete flagged data stream");
2108
2109 rcu_read_lock();
2110 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
2111 /* Validate delete flag of the stream */
79d4ffb7 2112 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2113 continue;
2114 }
2115 /* Delete it right now */
2116 consumer_del_stream(stream, data_ht);
2117 }
2118 rcu_read_unlock();
2119}
2120
2121/*
2122 * Delete metadata stream that are flagged for deletion (endpoint_status).
2123 */
2124static void validate_endpoint_status_metadata_stream(
2125 struct lttng_poll_event *pollset)
2126{
2127 struct lttng_ht_iter iter;
2128 struct lttng_consumer_stream *stream;
2129
2130 DBG("Consumer delete flagged metadata stream");
2131
2132 assert(pollset);
2133
2134 rcu_read_lock();
2135 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
2136 /* Validate delete flag of the stream */
79d4ffb7 2137 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2138 continue;
2139 }
2140 /*
2141 * Remove from pollset so the metadata thread can continue without
2142 * blocking on a deleted stream.
2143 */
2144 lttng_poll_del(pollset, stream->wait_fd);
2145
2146 /* Delete it right now */
2147 consumer_del_metadata_stream(stream, metadata_ht);
2148 }
2149 rcu_read_unlock();
2150}
2151
fb3a43a9
DG
2152/*
2153 * Thread polls on metadata file descriptor and write them on disk or on the
2154 * network.
2155 */
7d980def 2156void *consumer_thread_metadata_poll(void *data)
fb3a43a9 2157{
1fc79fb4 2158 int ret, i, pollfd, err = -1;
fb3a43a9 2159 uint32_t revents, nb_fd;
e316aad5 2160 struct lttng_consumer_stream *stream = NULL;
fb3a43a9 2161 struct lttng_ht_iter iter;
d88aee68 2162 struct lttng_ht_node_u64 *node;
fb3a43a9
DG
2163 struct lttng_poll_event events;
2164 struct lttng_consumer_local_data *ctx = data;
2165 ssize_t len;
2166
2167 rcu_register_thread();
2168
1fc79fb4
MD
2169 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA);
2170
2d57de81
MD
2171 if (testpoint(consumerd_thread_metadata)) {
2172 goto error_testpoint;
2173 }
2174
9ce5646a
MD
2175 health_code_update();
2176
fb3a43a9
DG
2177 DBG("Thread metadata poll started");
2178
fb3a43a9
DG
2179 /* Size is set to 1 for the consumer_metadata pipe */
2180 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2181 if (ret < 0) {
2182 ERR("Poll set creation failed");
d8ef542d 2183 goto end_poll;
fb3a43a9
DG
2184 }
2185
13886d2d
DG
2186 ret = lttng_poll_add(&events,
2187 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN);
fb3a43a9
DG
2188 if (ret < 0) {
2189 goto end;
2190 }
2191
2192 /* Main loop */
2193 DBG("Metadata main loop started");
2194
2195 while (1) {
fb3a43a9 2196restart:
7fa2082e 2197 health_code_update();
9ce5646a 2198 health_poll_entry();
7fa2082e 2199 DBG("Metadata poll wait");
fb3a43a9 2200 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2201 DBG("Metadata poll return from wait with %d fd(s)",
2202 LTTNG_POLL_GETNB(&events));
9ce5646a 2203 health_poll_exit();
fb3a43a9
DG
2204 DBG("Metadata event catched in thread");
2205 if (ret < 0) {
2206 if (errno == EINTR) {
e316aad5 2207 ERR("Poll EINTR catched");
fb3a43a9
DG
2208 goto restart;
2209 }
d9607cd7
MD
2210 if (LTTNG_POLL_GETNB(&events) == 0) {
2211 err = 0; /* All is OK */
2212 }
2213 goto end;
fb3a43a9
DG
2214 }
2215
0d9c5d77
DG
2216 nb_fd = ret;
2217
e316aad5 2218 /* From here, the event is a metadata wait fd */
fb3a43a9 2219 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2220 health_code_update();
2221
fb3a43a9
DG
2222 revents = LTTNG_POLL_GETEV(&events, i);
2223 pollfd = LTTNG_POLL_GETFD(&events, i);
2224
fd20dac9
MD
2225 if (!revents) {
2226 /* No activity for this FD (poll implementation). */
2227 continue;
2228 }
2229
13886d2d 2230 if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) {
4adabd61 2231 if (revents & (LPOLLERR | LPOLLHUP )) {
fb3a43a9
DG
2232 DBG("Metadata thread pipe hung up");
2233 /*
2234 * Remove the pipe from the poll set and continue the loop
2235 * since their might be data to consume.
2236 */
13886d2d
DG
2237 lttng_poll_del(&events,
2238 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2239 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
fb3a43a9
DG
2240 continue;
2241 } else if (revents & LPOLLIN) {
13886d2d
DG
2242 ssize_t pipe_len;
2243
2244 pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe,
2245 &stream, sizeof(stream));
6cd525e8
MD
2246 if (pipe_len < sizeof(stream)) {
2247 PERROR("read metadata stream");
fb3a43a9 2248 /*
13886d2d 2249 * Continue here to handle the rest of the streams.
fb3a43a9
DG
2250 */
2251 continue;
2252 }
2253
8994307f
DG
2254 /* A NULL stream means that the state has changed. */
2255 if (stream == NULL) {
2256 /* Check for deleted streams. */
2257 validate_endpoint_status_metadata_stream(&events);
3714380f 2258 goto restart;
8994307f
DG
2259 }
2260
fb3a43a9
DG
2261 DBG("Adding metadata stream %d to poll set",
2262 stream->wait_fd);
2263
fb3a43a9
DG
2264 /* Add metadata stream to the global poll events list */
2265 lttng_poll_add(&events, stream->wait_fd,
6d574024 2266 LPOLLIN | LPOLLPRI | LPOLLHUP);
fb3a43a9
DG
2267 }
2268
e316aad5 2269 /* Handle other stream */
fb3a43a9
DG
2270 continue;
2271 }
2272
d09e1200 2273 rcu_read_lock();
d88aee68
DG
2274 {
2275 uint64_t tmp_id = (uint64_t) pollfd;
2276
2277 lttng_ht_lookup(metadata_ht, &tmp_id, &iter);
2278 }
2279 node = lttng_ht_iter_get_node_u64(&iter);
e316aad5 2280 assert(node);
fb3a43a9
DG
2281
2282 stream = caa_container_of(node, struct lttng_consumer_stream,
58b1f425 2283 node);
fb3a43a9 2284
e316aad5 2285 /* Check for error event */
4adabd61 2286 if (revents & (LPOLLERR | LPOLLHUP)) {
e316aad5 2287 DBG("Metadata fd %d is hup|err.", pollfd);
fb3a43a9
DG
2288 if (!stream->hangup_flush_done
2289 && (consumer_data.type == LTTNG_CONSUMER32_UST
2290 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2291 DBG("Attempting to flush and consume the UST buffers");
2292 lttng_ustconsumer_on_stream_hangup(stream);
2293
2294 /* We just flushed the stream now read it. */
4bb94b75 2295 do {
9ce5646a
MD
2296 health_code_update();
2297
4bb94b75
DG
2298 len = ctx->on_buffer_ready(stream, ctx);
2299 /*
2300 * We don't check the return value here since if we get
2301 * a negative len, it means an error occured thus we
2302 * simply remove it from the poll set and free the
2303 * stream.
2304 */
2305 } while (len > 0);
fb3a43a9
DG
2306 }
2307
fb3a43a9 2308 lttng_poll_del(&events, stream->wait_fd);
e316aad5
DG
2309 /*
2310 * This call update the channel states, closes file descriptors
2311 * and securely free the stream.
2312 */
2313 consumer_del_metadata_stream(stream, metadata_ht);
2314 } else if (revents & (LPOLLIN | LPOLLPRI)) {
2315 /* Get the data out of the metadata file descriptor */
2316 DBG("Metadata available on fd %d", pollfd);
2317 assert(stream->wait_fd == pollfd);
2318
04ef1097 2319 do {
9ce5646a
MD
2320 health_code_update();
2321
04ef1097
MD
2322 len = ctx->on_buffer_ready(stream, ctx);
2323 /*
2324 * We don't check the return value here since if we get
2325 * a negative len, it means an error occured thus we
2326 * simply remove it from the poll set and free the
2327 * stream.
2328 */
2329 } while (len > 0);
2330
e316aad5 2331 /* It's ok to have an unavailable sub-buffer */
b64403e3 2332 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2333 /* Clean up stream from consumer and free it. */
2334 lttng_poll_del(&events, stream->wait_fd);
2335 consumer_del_metadata_stream(stream, metadata_ht);
e316aad5 2336 }
fb3a43a9 2337 }
e316aad5
DG
2338
2339 /* Release RCU lock for the stream looked up */
d09e1200 2340 rcu_read_unlock();
fb3a43a9
DG
2341 }
2342 }
2343
1fc79fb4
MD
2344 /* All is OK */
2345 err = 0;
fb3a43a9
DG
2346end:
2347 DBG("Metadata poll thread exiting");
fb3a43a9 2348
d8ef542d
MD
2349 lttng_poll_clean(&events);
2350end_poll:
2d57de81 2351error_testpoint:
1fc79fb4
MD
2352 if (err) {
2353 health_error();
2354 ERR("Health error occurred in %s", __func__);
2355 }
2356 health_unregister(health_consumerd);
fb3a43a9
DG
2357 rcu_unregister_thread();
2358 return NULL;
2359}
2360
3bd1e081 2361/*
e4421fec 2362 * This thread polls the fds in the set to consume the data and write
3bd1e081
MD
2363 * it to tracefile if necessary.
2364 */
7d980def 2365void *consumer_thread_data_poll(void *data)
3bd1e081 2366{
1fc79fb4 2367 int num_rdy, num_hup, high_prio, ret, i, err = -1;
3bd1e081
MD
2368 struct pollfd *pollfd = NULL;
2369 /* local view of the streams */
c869f647 2370 struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL;
3bd1e081
MD
2371 /* local view of consumer_data.fds_count */
2372 int nb_fd = 0;
3bd1e081 2373 struct lttng_consumer_local_data *ctx = data;
00e2e675 2374 ssize_t len;
3bd1e081 2375
e7b994a3
DG
2376 rcu_register_thread();
2377
1fc79fb4
MD
2378 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA);
2379
2d57de81
MD
2380 if (testpoint(consumerd_thread_data)) {
2381 goto error_testpoint;
2382 }
2383
9ce5646a
MD
2384 health_code_update();
2385
4df6c8cb
MD
2386 local_stream = zmalloc(sizeof(struct lttng_consumer_stream *));
2387 if (local_stream == NULL) {
2388 PERROR("local_stream malloc");
2389 goto end;
2390 }
3bd1e081
MD
2391
2392 while (1) {
9ce5646a
MD
2393 health_code_update();
2394
3bd1e081
MD
2395 high_prio = 0;
2396 num_hup = 0;
2397
2398 /*
e4421fec 2399 * the fds set has been updated, we need to update our
3bd1e081
MD
2400 * local array as well
2401 */
2402 pthread_mutex_lock(&consumer_data.lock);
2403 if (consumer_data.need_update) {
0e428499
DG
2404 free(pollfd);
2405 pollfd = NULL;
2406
2407 free(local_stream);
2408 local_stream = NULL;
3bd1e081 2409
02b3d176
DG
2410 /*
2411 * Allocate for all fds +1 for the consumer_data_pipe and +1 for
2412 * wake up pipe.
2413 */
2414 pollfd = zmalloc((consumer_data.stream_count + 2) * sizeof(struct pollfd));
3bd1e081 2415 if (pollfd == NULL) {
7a57cf92 2416 PERROR("pollfd malloc");
3bd1e081
MD
2417 pthread_mutex_unlock(&consumer_data.lock);
2418 goto end;
2419 }
2420
02b3d176 2421 local_stream = zmalloc((consumer_data.stream_count + 2) *
747f8642 2422 sizeof(struct lttng_consumer_stream *));
3bd1e081 2423 if (local_stream == NULL) {
7a57cf92 2424 PERROR("local_stream malloc");
3bd1e081
MD
2425 pthread_mutex_unlock(&consumer_data.lock);
2426 goto end;
2427 }
ffe60014 2428 ret = update_poll_array(ctx, &pollfd, local_stream,
43c34bc3 2429 data_ht);
3bd1e081
MD
2430 if (ret < 0) {
2431 ERR("Error in allocating pollfd or local_outfds");
f73fabfd 2432 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2433 pthread_mutex_unlock(&consumer_data.lock);
2434 goto end;
2435 }
2436 nb_fd = ret;
2437 consumer_data.need_update = 0;
2438 }
2439 pthread_mutex_unlock(&consumer_data.lock);
2440
4078b776
MD
2441 /* No FDs and consumer_quit, consumer_cleanup the thread */
2442 if (nb_fd == 0 && consumer_quit == 1) {
1fc79fb4 2443 err = 0; /* All is OK */
4078b776
MD
2444 goto end;
2445 }
3bd1e081 2446 /* poll on the array of fds */
88f2b785 2447 restart:
02b3d176 2448 DBG("polling on %d fd", nb_fd + 2);
9ce5646a 2449 health_poll_entry();
02b3d176 2450 num_rdy = poll(pollfd, nb_fd + 2, -1);
9ce5646a 2451 health_poll_exit();
3bd1e081
MD
2452 DBG("poll num_rdy : %d", num_rdy);
2453 if (num_rdy == -1) {
88f2b785
MD
2454 /*
2455 * Restart interrupted system call.
2456 */
2457 if (errno == EINTR) {
2458 goto restart;
2459 }
7a57cf92 2460 PERROR("Poll error");
f73fabfd 2461 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2462 goto end;
2463 } else if (num_rdy == 0) {
2464 DBG("Polling thread timed out");
2465 goto end;
2466 }
2467
3bd1e081 2468 /*
50f8ae69 2469 * If the consumer_data_pipe triggered poll go directly to the
00e2e675
DG
2470 * beginning of the loop to update the array. We want to prioritize
2471 * array update over low-priority reads.
3bd1e081 2472 */
509bb1cf 2473 if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) {
ab30f567 2474 ssize_t pipe_readlen;
04fdd819 2475
50f8ae69 2476 DBG("consumer_data_pipe wake up");
acdb9057
DG
2477 pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe,
2478 &new_stream, sizeof(new_stream));
6cd525e8
MD
2479 if (pipe_readlen < sizeof(new_stream)) {
2480 PERROR("Consumer data pipe");
23f5f35d
DG
2481 /* Continue so we can at least handle the current stream(s). */
2482 continue;
2483 }
c869f647
DG
2484
2485 /*
2486 * If the stream is NULL, just ignore it. It's also possible that
2487 * the sessiond poll thread changed the consumer_quit state and is
2488 * waking us up to test it.
2489 */
2490 if (new_stream == NULL) {
8994307f 2491 validate_endpoint_status_data_stream();
c869f647
DG
2492 continue;
2493 }
2494
c869f647 2495 /* Continue to update the local streams and handle prio ones */
3bd1e081
MD
2496 continue;
2497 }
2498
02b3d176
DG
2499 /* Handle wakeup pipe. */
2500 if (pollfd[nb_fd + 1].revents & (POLLIN | POLLPRI)) {
2501 char dummy;
2502 ssize_t pipe_readlen;
2503
2504 pipe_readlen = lttng_pipe_read(ctx->consumer_wakeup_pipe, &dummy,
2505 sizeof(dummy));
2506 if (pipe_readlen < 0) {
2507 PERROR("Consumer data wakeup pipe");
2508 }
2509 /* We've been awakened to handle stream(s). */
2510 ctx->has_wakeup = 0;
2511 }
2512
3bd1e081
MD
2513 /* Take care of high priority channels first. */
2514 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2515 health_code_update();
2516
9617607b
DG
2517 if (local_stream[i] == NULL) {
2518 continue;
2519 }
fb3a43a9 2520 if (pollfd[i].revents & POLLPRI) {
d41f73b7
MD
2521 DBG("Urgent read on fd %d", pollfd[i].fd);
2522 high_prio = 1;
4078b776 2523 len = ctx->on_buffer_ready(local_stream[i], ctx);
d41f73b7 2524 /* it's ok to have an unavailable sub-buffer */
b64403e3 2525 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2526 /* Clean the stream and free it. */
2527 consumer_del_stream(local_stream[i], data_ht);
9617607b 2528 local_stream[i] = NULL;
4078b776
MD
2529 } else if (len > 0) {
2530 local_stream[i]->data_read = 1;
d41f73b7 2531 }
3bd1e081
MD
2532 }
2533 }
2534
4078b776
MD
2535 /*
2536 * If we read high prio channel in this loop, try again
2537 * for more high prio data.
2538 */
2539 if (high_prio) {
3bd1e081
MD
2540 continue;
2541 }
2542
2543 /* Take care of low priority channels. */
4078b776 2544 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2545 health_code_update();
2546
9617607b
DG
2547 if (local_stream[i] == NULL) {
2548 continue;
2549 }
4078b776 2550 if ((pollfd[i].revents & POLLIN) ||
02b3d176
DG
2551 local_stream[i]->hangup_flush_done ||
2552 local_stream[i]->has_data) {
4078b776
MD
2553 DBG("Normal read on fd %d", pollfd[i].fd);
2554 len = ctx->on_buffer_ready(local_stream[i], ctx);
2555 /* it's ok to have an unavailable sub-buffer */
b64403e3 2556 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2557 /* Clean the stream and free it. */
2558 consumer_del_stream(local_stream[i], data_ht);
9617607b 2559 local_stream[i] = NULL;
4078b776
MD
2560 } else if (len > 0) {
2561 local_stream[i]->data_read = 1;
2562 }
2563 }
2564 }
2565
2566 /* Handle hangup and errors */
2567 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2568 health_code_update();
2569
9617607b
DG
2570 if (local_stream[i] == NULL) {
2571 continue;
2572 }
4078b776
MD
2573 if (!local_stream[i]->hangup_flush_done
2574 && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL))
2575 && (consumer_data.type == LTTNG_CONSUMER32_UST
2576 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2577 DBG("fd %d is hup|err|nval. Attempting flush and read.",
9617607b 2578 pollfd[i].fd);
4078b776
MD
2579 lttng_ustconsumer_on_stream_hangup(local_stream[i]);
2580 /* Attempt read again, for the data we just flushed. */
2581 local_stream[i]->data_read = 1;
2582 }
2583 /*
2584 * If the poll flag is HUP/ERR/NVAL and we have
2585 * read no data in this pass, we can remove the
2586 * stream from its hash table.
2587 */
2588 if ((pollfd[i].revents & POLLHUP)) {
2589 DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
2590 if (!local_stream[i]->data_read) {
43c34bc3 2591 consumer_del_stream(local_stream[i], data_ht);
9617607b 2592 local_stream[i] = NULL;
4078b776
MD
2593 num_hup++;
2594 }
2595 } else if (pollfd[i].revents & POLLERR) {
2596 ERR("Error returned in polling fd %d.", pollfd[i].fd);
2597 if (!local_stream[i]->data_read) {
43c34bc3 2598 consumer_del_stream(local_stream[i], data_ht);
9617607b 2599 local_stream[i] = NULL;
4078b776
MD
2600 num_hup++;
2601 }
2602 } else if (pollfd[i].revents & POLLNVAL) {
2603 ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
2604 if (!local_stream[i]->data_read) {
43c34bc3 2605 consumer_del_stream(local_stream[i], data_ht);
9617607b 2606 local_stream[i] = NULL;
4078b776 2607 num_hup++;
3bd1e081
MD
2608 }
2609 }
9617607b
DG
2610 if (local_stream[i] != NULL) {
2611 local_stream[i]->data_read = 0;
2612 }
3bd1e081
MD
2613 }
2614 }
1fc79fb4
MD
2615 /* All is OK */
2616 err = 0;
3bd1e081
MD
2617end:
2618 DBG("polling thread exiting");
0e428499
DG
2619 free(pollfd);
2620 free(local_stream);
fb3a43a9
DG
2621
2622 /*
2623 * Close the write side of the pipe so epoll_wait() in
7d980def
DG
2624 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2625 * read side of the pipe. If we close them both, epoll_wait strangely does
2626 * not return and could create a endless wait period if the pipe is the
2627 * only tracked fd in the poll set. The thread will take care of closing
2628 * the read side.
fb3a43a9 2629 */
13886d2d 2630 (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe);
fb3a43a9 2631
2d57de81 2632error_testpoint:
1fc79fb4
MD
2633 if (err) {
2634 health_error();
2635 ERR("Health error occurred in %s", __func__);
2636 }
2637 health_unregister(health_consumerd);
2638
e7b994a3 2639 rcu_unregister_thread();
3bd1e081
MD
2640 return NULL;
2641}
2642
d8ef542d
MD
2643/*
2644 * Close wake-up end of each stream belonging to the channel. This will
2645 * allow the poll() on the stream read-side to detect when the
2646 * write-side (application) finally closes them.
2647 */
2648static
2649void consumer_close_channel_streams(struct lttng_consumer_channel *channel)
2650{
2651 struct lttng_ht *ht;
2652 struct lttng_consumer_stream *stream;
2653 struct lttng_ht_iter iter;
2654
2655 ht = consumer_data.stream_per_chan_id_ht;
2656
2657 rcu_read_lock();
2658 cds_lfht_for_each_entry_duplicate(ht->ht,
2659 ht->hash_fct(&channel->key, lttng_ht_seed),
2660 ht->match_fct, &channel->key,
2661 &iter.iter, stream, node_channel_id.node) {
f2ad556d
MD
2662 /*
2663 * Protect against teardown with mutex.
2664 */
2665 pthread_mutex_lock(&stream->lock);
2666 if (cds_lfht_is_node_deleted(&stream->node.node)) {
2667 goto next;
2668 }
d8ef542d
MD
2669 switch (consumer_data.type) {
2670 case LTTNG_CONSUMER_KERNEL:
2671 break;
2672 case LTTNG_CONSUMER32_UST:
2673 case LTTNG_CONSUMER64_UST:
b4a650f3
DG
2674 if (stream->metadata_flag) {
2675 /* Safe and protected by the stream lock. */
2676 lttng_ustconsumer_close_metadata(stream->chan);
2677 } else {
2678 /*
2679 * Note: a mutex is taken internally within
2680 * liblttng-ust-ctl to protect timer wakeup_fd
2681 * use from concurrent close.
2682 */
2683 lttng_ustconsumer_close_stream_wakeup(stream);
2684 }
d8ef542d
MD
2685 break;
2686 default:
2687 ERR("Unknown consumer_data type");
2688 assert(0);
2689 }
f2ad556d
MD
2690 next:
2691 pthread_mutex_unlock(&stream->lock);
d8ef542d
MD
2692 }
2693 rcu_read_unlock();
2694}
2695
2696static void destroy_channel_ht(struct lttng_ht *ht)
2697{
2698 struct lttng_ht_iter iter;
2699 struct lttng_consumer_channel *channel;
2700 int ret;
2701
2702 if (ht == NULL) {
2703 return;
2704 }
2705
2706 rcu_read_lock();
2707 cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) {
2708 ret = lttng_ht_del(ht, &iter);
2709 assert(ret != 0);
2710 }
2711 rcu_read_unlock();
2712
2713 lttng_ht_destroy(ht);
2714}
2715
2716/*
2717 * This thread polls the channel fds to detect when they are being
2718 * closed. It closes all related streams if the channel is detected as
2719 * closed. It is currently only used as a shim layer for UST because the
2720 * consumerd needs to keep the per-stream wakeup end of pipes open for
2721 * periodical flush.
2722 */
2723void *consumer_thread_channel_poll(void *data)
2724{
1fc79fb4 2725 int ret, i, pollfd, err = -1;
d8ef542d
MD
2726 uint32_t revents, nb_fd;
2727 struct lttng_consumer_channel *chan = NULL;
2728 struct lttng_ht_iter iter;
2729 struct lttng_ht_node_u64 *node;
2730 struct lttng_poll_event events;
2731 struct lttng_consumer_local_data *ctx = data;
2732 struct lttng_ht *channel_ht;
2733
2734 rcu_register_thread();
2735
1fc79fb4
MD
2736 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL);
2737
2d57de81
MD
2738 if (testpoint(consumerd_thread_channel)) {
2739 goto error_testpoint;
2740 }
2741
9ce5646a
MD
2742 health_code_update();
2743
d8ef542d
MD
2744 channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2745 if (!channel_ht) {
2746 /* ENOMEM at this point. Better to bail out. */
2747 goto end_ht;
2748 }
2749
2750 DBG("Thread channel poll started");
2751
2752 /* Size is set to 1 for the consumer_channel pipe */
2753 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2754 if (ret < 0) {
2755 ERR("Poll set creation failed");
2756 goto end_poll;
2757 }
2758
2759 ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN);
2760 if (ret < 0) {
2761 goto end;
2762 }
2763
2764 /* Main loop */
2765 DBG("Channel main loop started");
2766
2767 while (1) {
d8ef542d 2768restart:
7fa2082e
MD
2769 health_code_update();
2770 DBG("Channel poll wait");
9ce5646a 2771 health_poll_entry();
d8ef542d 2772 ret = lttng_poll_wait(&events, -1);
7fa2082e
MD
2773 DBG("Channel poll return from wait with %d fd(s)",
2774 LTTNG_POLL_GETNB(&events));
9ce5646a 2775 health_poll_exit();
d8ef542d
MD
2776 DBG("Channel event catched in thread");
2777 if (ret < 0) {
2778 if (errno == EINTR) {
2779 ERR("Poll EINTR catched");
2780 goto restart;
2781 }
d9607cd7
MD
2782 if (LTTNG_POLL_GETNB(&events) == 0) {
2783 err = 0; /* All is OK */
2784 }
d8ef542d
MD
2785 goto end;
2786 }
2787
2788 nb_fd = ret;
2789
2790 /* From here, the event is a channel wait fd */
2791 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2792 health_code_update();
2793
d8ef542d
MD
2794 revents = LTTNG_POLL_GETEV(&events, i);
2795 pollfd = LTTNG_POLL_GETFD(&events, i);
2796
d8ef542d 2797 if (!revents) {
fd20dac9 2798 /* No activity for this FD (poll implementation). */
d8ef542d
MD
2799 continue;
2800 }
fd20dac9 2801
d8ef542d
MD
2802 if (pollfd == ctx->consumer_channel_pipe[0]) {
2803 if (revents & (LPOLLERR | LPOLLHUP)) {
2804 DBG("Channel thread pipe hung up");
2805 /*
2806 * Remove the pipe from the poll set and continue the loop
2807 * since their might be data to consume.
2808 */
2809 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2810 continue;
2811 } else if (revents & LPOLLIN) {
2812 enum consumer_channel_action action;
a0cbdd2e 2813 uint64_t key;
d8ef542d 2814
a0cbdd2e 2815 ret = read_channel_pipe(ctx, &chan, &key, &action);
d8ef542d
MD
2816 if (ret <= 0) {
2817 ERR("Error reading channel pipe");
2818 continue;
2819 }
2820
2821 switch (action) {
2822 case CONSUMER_CHANNEL_ADD:
2823 DBG("Adding channel %d to poll set",
2824 chan->wait_fd);
2825
2826 lttng_ht_node_init_u64(&chan->wait_fd_node,
2827 chan->wait_fd);
c7260a81 2828 rcu_read_lock();
d8ef542d
MD
2829 lttng_ht_add_unique_u64(channel_ht,
2830 &chan->wait_fd_node);
c7260a81 2831 rcu_read_unlock();
d8ef542d
MD
2832 /* Add channel to the global poll events list */
2833 lttng_poll_add(&events, chan->wait_fd,
2834 LPOLLIN | LPOLLPRI);
2835 break;
a0cbdd2e
MD
2836 case CONSUMER_CHANNEL_DEL:
2837 {
b4a650f3
DG
2838 /*
2839 * This command should never be called if the channel
2840 * has streams monitored by either the data or metadata
2841 * thread. The consumer only notify this thread with a
2842 * channel del. command if it receives a destroy
2843 * channel command from the session daemon that send it
2844 * if a command prior to the GET_CHANNEL failed.
2845 */
2846
c7260a81 2847 rcu_read_lock();
a0cbdd2e
MD
2848 chan = consumer_find_channel(key);
2849 if (!chan) {
c7260a81 2850 rcu_read_unlock();
a0cbdd2e
MD
2851 ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key);
2852 break;
2853 }
2854 lttng_poll_del(&events, chan->wait_fd);
f623cc0b 2855 iter.iter.node = &chan->wait_fd_node.node;
a0cbdd2e
MD
2856 ret = lttng_ht_del(channel_ht, &iter);
2857 assert(ret == 0);
a0cbdd2e 2858
f2a444f1
DG
2859 switch (consumer_data.type) {
2860 case LTTNG_CONSUMER_KERNEL:
2861 break;
2862 case LTTNG_CONSUMER32_UST:
2863 case LTTNG_CONSUMER64_UST:
212d67a2
DG
2864 health_code_update();
2865 /* Destroy streams that might have been left in the stream list. */
2866 clean_channel_stream_list(chan);
f2a444f1
DG
2867 break;
2868 default:
2869 ERR("Unknown consumer_data type");
2870 assert(0);
2871 }
2872
a0cbdd2e
MD
2873 /*
2874 * Release our own refcount. Force channel deletion even if
2875 * streams were not initialized.
2876 */
2877 if (!uatomic_sub_return(&chan->refcount, 1)) {
2878 consumer_del_channel(chan);
2879 }
c7260a81 2880 rcu_read_unlock();
a0cbdd2e
MD
2881 goto restart;
2882 }
d8ef542d
MD
2883 case CONSUMER_CHANNEL_QUIT:
2884 /*
2885 * Remove the pipe from the poll set and continue the loop
2886 * since their might be data to consume.
2887 */
2888 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2889 continue;
2890 default:
2891 ERR("Unknown action");
2892 break;
2893 }
2894 }
2895
2896 /* Handle other stream */
2897 continue;
2898 }
2899
2900 rcu_read_lock();
2901 {
2902 uint64_t tmp_id = (uint64_t) pollfd;
2903
2904 lttng_ht_lookup(channel_ht, &tmp_id, &iter);
2905 }
2906 node = lttng_ht_iter_get_node_u64(&iter);
2907 assert(node);
2908
2909 chan = caa_container_of(node, struct lttng_consumer_channel,
2910 wait_fd_node);
2911
2912 /* Check for error event */
2913 if (revents & (LPOLLERR | LPOLLHUP)) {
2914 DBG("Channel fd %d is hup|err.", pollfd);
2915
2916 lttng_poll_del(&events, chan->wait_fd);
2917 ret = lttng_ht_del(channel_ht, &iter);
2918 assert(ret == 0);
b4a650f3
DG
2919
2920 /*
2921 * This will close the wait fd for each stream associated to
2922 * this channel AND monitored by the data/metadata thread thus
2923 * will be clean by the right thread.
2924 */
d8ef542d 2925 consumer_close_channel_streams(chan);
f2ad556d
MD
2926
2927 /* Release our own refcount */
2928 if (!uatomic_sub_return(&chan->refcount, 1)
2929 && !uatomic_read(&chan->nb_init_stream_left)) {
2930 consumer_del_channel(chan);
2931 }
d8ef542d
MD
2932 }
2933
2934 /* Release RCU lock for the channel looked up */
2935 rcu_read_unlock();
2936 }
2937 }
2938
1fc79fb4
MD
2939 /* All is OK */
2940 err = 0;
d8ef542d
MD
2941end:
2942 lttng_poll_clean(&events);
2943end_poll:
2944 destroy_channel_ht(channel_ht);
2945end_ht:
2d57de81 2946error_testpoint:
d8ef542d 2947 DBG("Channel poll thread exiting");
1fc79fb4
MD
2948 if (err) {
2949 health_error();
2950 ERR("Health error occurred in %s", __func__);
2951 }
2952 health_unregister(health_consumerd);
d8ef542d
MD
2953 rcu_unregister_thread();
2954 return NULL;
2955}
2956
331744e3
JD
2957static int set_metadata_socket(struct lttng_consumer_local_data *ctx,
2958 struct pollfd *sockpoll, int client_socket)
2959{
2960 int ret;
2961
2962 assert(ctx);
2963 assert(sockpoll);
2964
84382d49
MD
2965 ret = lttng_consumer_poll_socket(sockpoll);
2966 if (ret) {
331744e3
JD
2967 goto error;
2968 }
2969 DBG("Metadata connection on client_socket");
2970
2971 /* Blocking call, waiting for transmission */
2972 ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket);
2973 if (ctx->consumer_metadata_socket < 0) {
2974 WARN("On accept metadata");
2975 ret = -1;
2976 goto error;
2977 }
2978 ret = 0;
2979
2980error:
2981 return ret;
2982}
2983
3bd1e081
MD
2984/*
2985 * This thread listens on the consumerd socket and receives the file
2986 * descriptors from the session daemon.
2987 */
7d980def 2988void *consumer_thread_sessiond_poll(void *data)
3bd1e081 2989{
1fc79fb4 2990 int sock = -1, client_socket, ret, err = -1;
3bd1e081
MD
2991 /*
2992 * structure to poll for incoming data on communication socket avoids
2993 * making blocking sockets.
2994 */
2995 struct pollfd consumer_sockpoll[2];
2996 struct lttng_consumer_local_data *ctx = data;
2997
e7b994a3
DG
2998 rcu_register_thread();
2999
1fc79fb4
MD
3000 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND);
3001
2d57de81
MD
3002 if (testpoint(consumerd_thread_sessiond)) {
3003 goto error_testpoint;
3004 }
3005
9ce5646a
MD
3006 health_code_update();
3007
3bd1e081
MD
3008 DBG("Creating command socket %s", ctx->consumer_command_sock_path);
3009 unlink(ctx->consumer_command_sock_path);
3010 client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path);
3011 if (client_socket < 0) {
3012 ERR("Cannot create command socket");
3013 goto end;
3014 }
3015
3016 ret = lttcomm_listen_unix_sock(client_socket);
3017 if (ret < 0) {
3018 goto end;
3019 }
3020
32258573 3021 DBG("Sending ready command to lttng-sessiond");
f73fabfd 3022 ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY);
3bd1e081
MD
3023 /* return < 0 on error, but == 0 is not fatal */
3024 if (ret < 0) {
32258573 3025 ERR("Error sending ready command to lttng-sessiond");
3bd1e081
MD
3026 goto end;
3027 }
3028
3bd1e081
MD
3029 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3030 consumer_sockpoll[0].fd = ctx->consumer_should_quit[0];
3031 consumer_sockpoll[0].events = POLLIN | POLLPRI;
3032 consumer_sockpoll[1].fd = client_socket;
3033 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3034
84382d49
MD
3035 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3036 if (ret) {
3037 if (ret > 0) {
3038 /* should exit */
3039 err = 0;
3040 }
3bd1e081
MD
3041 goto end;
3042 }
3043 DBG("Connection on client_socket");
3044
3045 /* Blocking call, waiting for transmission */
3046 sock = lttcomm_accept_unix_sock(client_socket);
534d2592 3047 if (sock < 0) {
3bd1e081
MD
3048 WARN("On accept");
3049 goto end;
3050 }
3bd1e081 3051
331744e3
JD
3052 /*
3053 * Setup metadata socket which is the second socket connection on the
3054 * command unix socket.
3055 */
3056 ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket);
84382d49
MD
3057 if (ret) {
3058 if (ret > 0) {
3059 /* should exit */
3060 err = 0;
3061 }
331744e3
JD
3062 goto end;
3063 }
3064
d96f09c6
DG
3065 /* This socket is not useful anymore. */
3066 ret = close(client_socket);
3067 if (ret < 0) {
3068 PERROR("close client_socket");
3069 }
3070 client_socket = -1;
3071
3bd1e081
MD
3072 /* update the polling structure to poll on the established socket */
3073 consumer_sockpoll[1].fd = sock;
3074 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3075
3076 while (1) {
9ce5646a
MD
3077 health_code_update();
3078
3079 health_poll_entry();
3080 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3081 health_poll_exit();
84382d49
MD
3082 if (ret) {
3083 if (ret > 0) {
3084 /* should exit */
3085 err = 0;
3086 }
3bd1e081
MD
3087 goto end;
3088 }
3089 DBG("Incoming command on sock");
3090 ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll);
4cbc1a04
DG
3091 if (ret <= 0) {
3092 /*
3093 * This could simply be a session daemon quitting. Don't output
3094 * ERR() here.
3095 */
3096 DBG("Communication interrupted on command socket");
41ba6035 3097 err = 0;
3bd1e081
MD
3098 goto end;
3099 }
3100 if (consumer_quit) {
3101 DBG("consumer_thread_receive_fds received quit from signal");
1fc79fb4 3102 err = 0; /* All is OK */
3bd1e081
MD
3103 goto end;
3104 }
ffe60014 3105 DBG("received command on sock");
3bd1e081 3106 }
1fc79fb4
MD
3107 /* All is OK */
3108 err = 0;
3109
3bd1e081 3110end:
ffe60014 3111 DBG("Consumer thread sessiond poll exiting");
3bd1e081 3112
d88aee68
DG
3113 /*
3114 * Close metadata streams since the producer is the session daemon which
3115 * just died.
3116 *
3117 * NOTE: for now, this only applies to the UST tracer.
3118 */
6d574024 3119 lttng_consumer_close_all_metadata();
d88aee68 3120
3bd1e081
MD
3121 /*
3122 * when all fds have hung up, the polling thread
3123 * can exit cleanly
3124 */
3125 consumer_quit = 1;
3126
04fdd819 3127 /*
c869f647 3128 * Notify the data poll thread to poll back again and test the
8994307f 3129 * consumer_quit state that we just set so to quit gracefully.
04fdd819 3130 */
acdb9057 3131 notify_thread_lttng_pipe(ctx->consumer_data_pipe);
c869f647 3132
a0cbdd2e 3133 notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT);
d8ef542d 3134
5c635c72
MD
3135 notify_health_quit_pipe(health_quit_pipe);
3136
d96f09c6
DG
3137 /* Cleaning up possibly open sockets. */
3138 if (sock >= 0) {
3139 ret = close(sock);
3140 if (ret < 0) {
3141 PERROR("close sock sessiond poll");
3142 }
3143 }
3144 if (client_socket >= 0) {
38476d24 3145 ret = close(client_socket);
d96f09c6
DG
3146 if (ret < 0) {
3147 PERROR("close client_socket sessiond poll");
3148 }
3149 }
3150
2d57de81 3151error_testpoint:
1fc79fb4
MD
3152 if (err) {
3153 health_error();
3154 ERR("Health error occurred in %s", __func__);
3155 }
3156 health_unregister(health_consumerd);
3157
e7b994a3 3158 rcu_unregister_thread();
3bd1e081
MD
3159 return NULL;
3160}
d41f73b7 3161
4078b776 3162ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream,
d41f73b7
MD
3163 struct lttng_consumer_local_data *ctx)
3164{
74251bb8
DG
3165 ssize_t ret;
3166
3167 pthread_mutex_lock(&stream->lock);
94d49140
JD
3168 if (stream->metadata_flag) {
3169 pthread_mutex_lock(&stream->metadata_rdv_lock);
3170 }
74251bb8 3171
d41f73b7
MD
3172 switch (consumer_data.type) {
3173 case LTTNG_CONSUMER_KERNEL:
74251bb8
DG
3174 ret = lttng_kconsumer_read_subbuffer(stream, ctx);
3175 break;
7753dea8
MD
3176 case LTTNG_CONSUMER32_UST:
3177 case LTTNG_CONSUMER64_UST:
74251bb8
DG
3178 ret = lttng_ustconsumer_read_subbuffer(stream, ctx);
3179 break;
d41f73b7
MD
3180 default:
3181 ERR("Unknown consumer_data type");
3182 assert(0);
74251bb8
DG
3183 ret = -ENOSYS;
3184 break;
d41f73b7 3185 }
74251bb8 3186
94d49140
JD
3187 if (stream->metadata_flag) {
3188 pthread_cond_broadcast(&stream->metadata_rdv);
3189 pthread_mutex_unlock(&stream->metadata_rdv_lock);
3190 }
74251bb8
DG
3191 pthread_mutex_unlock(&stream->lock);
3192 return ret;
d41f73b7
MD
3193}
3194
3195int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream)
3196{
3197 switch (consumer_data.type) {
3198 case LTTNG_CONSUMER_KERNEL:
3199 return lttng_kconsumer_on_recv_stream(stream);
7753dea8
MD
3200 case LTTNG_CONSUMER32_UST:
3201 case LTTNG_CONSUMER64_UST:
d41f73b7
MD
3202 return lttng_ustconsumer_on_recv_stream(stream);
3203 default:
3204 ERR("Unknown consumer_data type");
3205 assert(0);
3206 return -ENOSYS;
3207 }
3208}
e4421fec
DG
3209
3210/*
3211 * Allocate and set consumer data hash tables.
3212 */
282dadbc 3213int lttng_consumer_init(void)
e4421fec 3214{
d88aee68 3215 consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
282dadbc
MD
3216 if (!consumer_data.channel_ht) {
3217 goto error;
3218 }
3219
d88aee68 3220 consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
282dadbc
MD
3221 if (!consumer_data.relayd_ht) {
3222 goto error;
3223 }
3224
d88aee68 3225 consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
282dadbc
MD
3226 if (!consumer_data.stream_list_ht) {
3227 goto error;
3228 }
3229
d8ef542d 3230 consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
282dadbc
MD
3231 if (!consumer_data.stream_per_chan_id_ht) {
3232 goto error;
3233 }
3234
3235 data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3236 if (!data_ht) {
3237 goto error;
3238 }
3239
3240 metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3241 if (!metadata_ht) {
3242 goto error;
3243 }
3244
3245 return 0;
3246
3247error:
3248 return -1;
e4421fec 3249}
7735ef9e
DG
3250
3251/*
3252 * Process the ADD_RELAYD command receive by a consumer.
3253 *
3254 * This will create a relayd socket pair and add it to the relayd hash table.
3255 * The caller MUST acquire a RCU read side lock before calling it.
3256 */
da009f2c 3257int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type,
7735ef9e 3258 struct lttng_consumer_local_data *ctx, int sock,
6151a90f 3259 struct pollfd *consumer_sockpoll,
d3e2ba59
JD
3260 struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id,
3261 uint64_t relayd_session_id)
7735ef9e 3262{
cd2b09ed 3263 int fd = -1, ret = -1, relayd_created = 0;
0c759fc9 3264 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
d4298c99 3265 struct consumer_relayd_sock_pair *relayd = NULL;
7735ef9e 3266
6151a90f
JD
3267 assert(ctx);
3268 assert(relayd_sock);
3269
da009f2c 3270 DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx);
7735ef9e
DG
3271
3272 /* Get relayd reference if exists. */
3273 relayd = consumer_find_relayd(net_seq_idx);
3274 if (relayd == NULL) {
da009f2c 3275 assert(sock_type == LTTNG_STREAM_CONTROL);
7735ef9e
DG
3276 /* Not found. Allocate one. */
3277 relayd = consumer_allocate_relayd_sock_pair(net_seq_idx);
3278 if (relayd == NULL) {
0d08d75e 3279 ret = -ENOMEM;
618a6a28
MD
3280 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
3281 goto error;
0d08d75e 3282 } else {
30319bcb 3283 relayd->sessiond_session_id = sessiond_id;
0d08d75e 3284 relayd_created = 1;
7735ef9e 3285 }
0d08d75e
DG
3286
3287 /*
3288 * This code path MUST continue to the consumer send status message to
3289 * we can notify the session daemon and continue our work without
3290 * killing everything.
3291 */
da009f2c
MD
3292 } else {
3293 /*
3294 * relayd key should never be found for control socket.
3295 */
3296 assert(sock_type != LTTNG_STREAM_CONTROL);
0d08d75e
DG
3297 }
3298
3299 /* First send a status message before receiving the fds. */
0c759fc9 3300 ret = consumer_send_status_msg(sock, LTTCOMM_CONSUMERD_SUCCESS);
618a6a28 3301 if (ret < 0) {
0d08d75e 3302 /* Somehow, the session daemon is not responding anymore. */
618a6a28
MD
3303 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3304 goto error_nosignal;
7735ef9e
DG
3305 }
3306
3307 /* Poll on consumer socket. */
84382d49
MD
3308 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3309 if (ret) {
3310 /* Needing to exit in the middle of a command: error. */
0d08d75e 3311 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
7735ef9e 3312 ret = -EINTR;
618a6a28 3313 goto error_nosignal;
7735ef9e
DG
3314 }
3315
3316 /* Get relayd socket from session daemon */
3317 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
3318 if (ret != sizeof(fd)) {
7735ef9e 3319 ret = -1;
4028eeb9 3320 fd = -1; /* Just in case it gets set with an invalid value. */
0d08d75e
DG
3321
3322 /*
3323 * Failing to receive FDs might indicate a major problem such as
3324 * reaching a fd limit during the receive where the kernel returns a
3325 * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we
3326 * don't take any chances and stop everything.
3327 *
3328 * XXX: Feature request #558 will fix that and avoid this possible
3329 * issue when reaching the fd limit.
3330 */
3331 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
618a6a28 3332 ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD;
f50f23d9
DG
3333 goto error;
3334 }
3335
7735ef9e
DG
3336 /* Copy socket information and received FD */
3337 switch (sock_type) {
3338 case LTTNG_STREAM_CONTROL:
3339 /* Copy received lttcomm socket */
6151a90f
JD
3340 lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock);
3341 ret = lttcomm_create_sock(&relayd->control_sock.sock);
4028eeb9 3342 /* Handle create_sock error. */
f66c074c 3343 if (ret < 0) {
618a6a28 3344 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
4028eeb9 3345 goto error;
f66c074c 3346 }
da009f2c
MD
3347 /*
3348 * Close the socket created internally by
3349 * lttcomm_create_sock, so we can replace it by the one
3350 * received from sessiond.
3351 */
3352 if (close(relayd->control_sock.sock.fd)) {
3353 PERROR("close");
3354 }
7735ef9e
DG
3355
3356 /* Assign new file descriptor */
6151a90f 3357 relayd->control_sock.sock.fd = fd;
4b29f1ce 3358 fd = -1; /* For error path */
6151a90f
JD
3359 /* Assign version values. */
3360 relayd->control_sock.major = relayd_sock->major;
3361 relayd->control_sock.minor = relayd_sock->minor;
c5b6f4f0 3362
d3e2ba59 3363 relayd->relayd_session_id = relayd_session_id;
c5b6f4f0 3364
7735ef9e
DG
3365 break;
3366 case LTTNG_STREAM_DATA:
3367 /* Copy received lttcomm socket */
6151a90f
JD
3368 lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock);
3369 ret = lttcomm_create_sock(&relayd->data_sock.sock);
4028eeb9 3370 /* Handle create_sock error. */
f66c074c 3371 if (ret < 0) {
618a6a28 3372 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
4028eeb9 3373 goto error;
f66c074c 3374 }
da009f2c
MD
3375 /*
3376 * Close the socket created internally by
3377 * lttcomm_create_sock, so we can replace it by the one
3378 * received from sessiond.
3379 */
3380 if (close(relayd->data_sock.sock.fd)) {
3381 PERROR("close");
3382 }
7735ef9e
DG
3383
3384 /* Assign new file descriptor */
6151a90f 3385 relayd->data_sock.sock.fd = fd;
4b29f1ce 3386 fd = -1; /* for eventual error paths */
6151a90f
JD
3387 /* Assign version values. */
3388 relayd->data_sock.major = relayd_sock->major;
3389 relayd->data_sock.minor = relayd_sock->minor;
7735ef9e
DG
3390 break;
3391 default:
3392 ERR("Unknown relayd socket type (%d)", sock_type);
59e71485 3393 ret = -1;
618a6a28 3394 ret_code = LTTCOMM_CONSUMERD_FATAL;
7735ef9e
DG
3395 goto error;
3396 }
3397
d88aee68 3398 DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)",
7735ef9e
DG
3399 sock_type == LTTNG_STREAM_CONTROL ? "control" : "data",
3400 relayd->net_seq_idx, fd);
3401
618a6a28
MD
3402 /* We successfully added the socket. Send status back. */
3403 ret = consumer_send_status_msg(sock, ret_code);
3404 if (ret < 0) {
3405 /* Somehow, the session daemon is not responding anymore. */
3406 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3407 goto error_nosignal;
3408 }
3409
7735ef9e
DG
3410 /*
3411 * Add relayd socket pair to consumer data hashtable. If object already
3412 * exists or on error, the function gracefully returns.
3413 */
d09e1200 3414 add_relayd(relayd);
7735ef9e
DG
3415
3416 /* All good! */
4028eeb9 3417 return 0;
7735ef9e
DG
3418
3419error:
618a6a28
MD
3420 if (consumer_send_status_msg(sock, ret_code) < 0) {
3421 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3422 }
3423
3424error_nosignal:
4028eeb9
DG
3425 /* Close received socket if valid. */
3426 if (fd >= 0) {
3427 if (close(fd)) {
3428 PERROR("close received socket");
3429 }
3430 }
cd2b09ed
DG
3431
3432 if (relayd_created) {
cd2b09ed
DG
3433 free(relayd);
3434 }
3435
7735ef9e
DG
3436 return ret;
3437}
ca22feea 3438
4e9a4686
DG
3439/*
3440 * Try to lock the stream mutex.
3441 *
3442 * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
3443 */
3444static int stream_try_lock(struct lttng_consumer_stream *stream)
3445{
3446 int ret;
3447
3448 assert(stream);
3449
3450 /*
3451 * Try to lock the stream mutex. On failure, we know that the stream is
3452 * being used else where hence there is data still being extracted.
3453 */
3454 ret = pthread_mutex_trylock(&stream->lock);
3455 if (ret) {
3456 /* For both EBUSY and EINVAL error, the mutex is NOT locked. */
3457 ret = 0;
3458 goto end;
3459 }
3460
3461 ret = 1;
3462
3463end:
3464 return ret;
3465}
3466
f7079f67
DG
3467/*
3468 * Search for a relayd associated to the session id and return the reference.
3469 *
3470 * A rcu read side lock MUST be acquire before calling this function and locked
3471 * until the relayd object is no longer necessary.
3472 */
3473static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id)
3474{
3475 struct lttng_ht_iter iter;
f7079f67 3476 struct consumer_relayd_sock_pair *relayd = NULL;
f7079f67
DG
3477
3478 /* Iterate over all relayd since they are indexed by net_seq_idx. */
3479 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
3480 node.node) {
18261bd1
DG
3481 /*
3482 * Check by sessiond id which is unique here where the relayd session
3483 * id might not be when having multiple relayd.
3484 */
3485 if (relayd->sessiond_session_id == id) {
f7079f67 3486 /* Found the relayd. There can be only one per id. */
18261bd1 3487 goto found;
f7079f67
DG
3488 }
3489 }
3490
18261bd1
DG
3491 return NULL;
3492
3493found:
f7079f67
DG
3494 return relayd;
3495}
3496
ca22feea
DG
3497/*
3498 * Check if for a given session id there is still data needed to be extract
3499 * from the buffers.
3500 *
6d805429 3501 * Return 1 if data is pending or else 0 meaning ready to be read.
ca22feea 3502 */
6d805429 3503int consumer_data_pending(uint64_t id)
ca22feea
DG
3504{
3505 int ret;
3506 struct lttng_ht_iter iter;
3507 struct lttng_ht *ht;
3508 struct lttng_consumer_stream *stream;
f7079f67 3509 struct consumer_relayd_sock_pair *relayd = NULL;
6d805429 3510 int (*data_pending)(struct lttng_consumer_stream *);
ca22feea 3511
6d805429 3512 DBG("Consumer data pending command on session id %" PRIu64, id);
ca22feea 3513
6f6eda74 3514 rcu_read_lock();
ca22feea
DG
3515 pthread_mutex_lock(&consumer_data.lock);
3516
3517 switch (consumer_data.type) {
3518 case LTTNG_CONSUMER_KERNEL:
6d805429 3519 data_pending = lttng_kconsumer_data_pending;
ca22feea
DG
3520 break;
3521 case LTTNG_CONSUMER32_UST:
3522 case LTTNG_CONSUMER64_UST:
6d805429 3523 data_pending = lttng_ustconsumer_data_pending;
ca22feea
DG
3524 break;
3525 default:
3526 ERR("Unknown consumer data type");
3527 assert(0);
3528 }
3529
3530 /* Ease our life a bit */
3531 ht = consumer_data.stream_list_ht;
3532
f7079f67
DG
3533 relayd = find_relayd_by_session_id(id);
3534 if (relayd) {
3535 /* Send init command for data pending. */
3536 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3537 ret = relayd_begin_data_pending(&relayd->control_sock,
3538 relayd->relayd_session_id);
3539 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
3540 if (ret < 0) {
3541 /* Communication error thus the relayd so no data pending. */
3542 goto data_not_pending;
3543 }
3544 }
3545
c8f59ee5 3546 cds_lfht_for_each_entry_duplicate(ht->ht,
d88aee68
DG
3547 ht->hash_fct(&id, lttng_ht_seed),
3548 ht->match_fct, &id,
ca22feea 3549 &iter.iter, stream, node_session_id.node) {
4e9a4686
DG
3550 /* If this call fails, the stream is being used hence data pending. */
3551 ret = stream_try_lock(stream);
3552 if (!ret) {
f7079f67 3553 goto data_pending;
ca22feea 3554 }
ca22feea 3555
4e9a4686
DG
3556 /*
3557 * A removed node from the hash table indicates that the stream has
3558 * been deleted thus having a guarantee that the buffers are closed
3559 * on the consumer side. However, data can still be transmitted
3560 * over the network so don't skip the relayd check.
3561 */
3562 ret = cds_lfht_is_node_deleted(&stream->node.node);
3563 if (!ret) {
3564 /* Check the stream if there is data in the buffers. */
6d805429
DG
3565 ret = data_pending(stream);
3566 if (ret == 1) {
4e9a4686 3567 pthread_mutex_unlock(&stream->lock);
f7079f67 3568 goto data_pending;
4e9a4686
DG
3569 }
3570 }
3571
3572 /* Relayd check */
f7079f67 3573 if (relayd) {
c8f59ee5
DG
3574 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3575 if (stream->metadata_flag) {
ad7051c0
DG
3576 ret = relayd_quiescent_control(&relayd->control_sock,
3577 stream->relayd_stream_id);
c8f59ee5 3578 } else {
6d805429 3579 ret = relayd_data_pending(&relayd->control_sock,
39df6d9f
DG
3580 stream->relayd_stream_id,
3581 stream->next_net_seq_num - 1);
c8f59ee5
DG
3582 }
3583 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
6d805429 3584 if (ret == 1) {
4e9a4686 3585 pthread_mutex_unlock(&stream->lock);
f7079f67 3586 goto data_pending;
c8f59ee5
DG
3587 }
3588 }
4e9a4686 3589 pthread_mutex_unlock(&stream->lock);
c8f59ee5 3590 }
ca22feea 3591
f7079f67
DG
3592 if (relayd) {
3593 unsigned int is_data_inflight = 0;
3594
3595 /* Send init command for data pending. */
3596 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3597 ret = relayd_end_data_pending(&relayd->control_sock,
3598 relayd->relayd_session_id, &is_data_inflight);
3599 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
bdd88757 3600 if (ret < 0) {
f7079f67
DG
3601 goto data_not_pending;
3602 }
bdd88757
DG
3603 if (is_data_inflight) {
3604 goto data_pending;
3605 }
f7079f67
DG
3606 }
3607
ca22feea 3608 /*
f7079f67
DG
3609 * Finding _no_ node in the hash table and no inflight data means that the
3610 * stream(s) have been removed thus data is guaranteed to be available for
3611 * analysis from the trace files.
ca22feea
DG
3612 */
3613
f7079f67 3614data_not_pending:
ca22feea
DG
3615 /* Data is available to be read by a viewer. */
3616 pthread_mutex_unlock(&consumer_data.lock);
c8f59ee5 3617 rcu_read_unlock();
6d805429 3618 return 0;
ca22feea 3619
f7079f67 3620data_pending:
ca22feea
DG
3621 /* Data is still being extracted from buffers. */
3622 pthread_mutex_unlock(&consumer_data.lock);
c8f59ee5 3623 rcu_read_unlock();
6d805429 3624 return 1;
ca22feea 3625}
f50f23d9
DG
3626
3627/*
3628 * Send a ret code status message to the sessiond daemon.
3629 *
3630 * Return the sendmsg() return value.
3631 */
3632int consumer_send_status_msg(int sock, int ret_code)
3633{
3634 struct lttcomm_consumer_status_msg msg;
3635
53efb85a 3636 memset(&msg, 0, sizeof(msg));
f50f23d9
DG
3637 msg.ret_code = ret_code;
3638
3639 return lttcomm_send_unix_sock(sock, &msg, sizeof(msg));
3640}
ffe60014
DG
3641
3642/*
3643 * Send a channel status message to the sessiond daemon.
3644 *
3645 * Return the sendmsg() return value.
3646 */
3647int consumer_send_status_channel(int sock,
3648 struct lttng_consumer_channel *channel)
3649{
3650 struct lttcomm_consumer_status_channel msg;
3651
3652 assert(sock >= 0);
3653
53efb85a 3654 memset(&msg, 0, sizeof(msg));
ffe60014 3655 if (!channel) {
0c759fc9 3656 msg.ret_code = LTTCOMM_CONSUMERD_CHANNEL_FAIL;
ffe60014 3657 } else {
0c759fc9 3658 msg.ret_code = LTTCOMM_CONSUMERD_SUCCESS;
ffe60014
DG
3659 msg.key = channel->key;
3660 msg.stream_count = channel->streams.count;
3661 }
3662
3663 return lttcomm_send_unix_sock(sock, &msg, sizeof(msg));
3664}
5c786ded 3665
d07ceecd
MD
3666unsigned long consumer_get_consume_start_pos(unsigned long consumed_pos,
3667 unsigned long produced_pos, uint64_t nb_packets_per_stream,
3668 uint64_t max_sb_size)
5c786ded 3669{
d07ceecd 3670 unsigned long start_pos;
5c786ded 3671
d07ceecd
MD
3672 if (!nb_packets_per_stream) {
3673 return consumed_pos; /* Grab everything */
3674 }
3675 start_pos = produced_pos - offset_align_floor(produced_pos, max_sb_size);
3676 start_pos -= max_sb_size * nb_packets_per_stream;
3677 if ((long) (start_pos - consumed_pos) < 0) {
3678 return consumed_pos; /* Grab everything */
3679 }
3680 return start_pos;
5c786ded 3681}
This page took 0.274955 seconds and 5 git commands to generate.