Add exclusion data to enable-command
[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
21#include <assert.h>
3bd1e081
MD
22#include <poll.h>
23#include <pthread.h>
24#include <stdlib.h>
25#include <string.h>
26#include <sys/mman.h>
27#include <sys/socket.h>
28#include <sys/types.h>
29#include <unistd.h>
77c7c900 30#include <inttypes.h>
331744e3 31#include <signal.h>
3bd1e081 32
51a9e1c7 33#include <bin/lttng-consumerd/health-consumerd.h>
990570ed 34#include <common/common.h>
fb3a43a9
DG
35#include <common/utils.h>
36#include <common/compat/poll.h>
309167d2 37#include <common/index/index.h>
10a8a223 38#include <common/kernel-ctl/kernel-ctl.h>
00e2e675 39#include <common/sessiond-comm/relayd.h>
10a8a223
DG
40#include <common/sessiond-comm/sessiond-comm.h>
41#include <common/kernel-consumer/kernel-consumer.h>
00e2e675 42#include <common/relayd/relayd.h>
10a8a223 43#include <common/ust-consumer/ust-consumer.h>
d3e2ba59 44#include <common/consumer-timer.h>
10a8a223
DG
45
46#include "consumer.h"
1d1a276c 47#include "consumer-stream.h"
3bd1e081
MD
48
49struct lttng_consumer_global_data consumer_data = {
3bd1e081
MD
50 .stream_count = 0,
51 .need_update = 1,
52 .type = LTTNG_CONSUMER_UNKNOWN,
53};
54
d8ef542d
MD
55enum consumer_channel_action {
56 CONSUMER_CHANNEL_ADD,
a0cbdd2e 57 CONSUMER_CHANNEL_DEL,
d8ef542d
MD
58 CONSUMER_CHANNEL_QUIT,
59};
60
61struct consumer_channel_msg {
62 enum consumer_channel_action action;
a0cbdd2e
MD
63 struct lttng_consumer_channel *chan; /* add */
64 uint64_t key; /* del */
d8ef542d
MD
65};
66
3bd1e081
MD
67/*
68 * Flag to inform the polling thread to quit when all fd hung up. Updated by
69 * the consumer_thread_receive_fds when it notices that all fds has hung up.
70 * Also updated by the signal handler (consumer_should_exit()). Read by the
71 * polling threads.
72 */
a98dae5f 73volatile int consumer_quit;
3bd1e081 74
43c34bc3 75/*
43c34bc3
DG
76 * Global hash table containing respectively metadata and data streams. The
77 * stream element in this ht should only be updated by the metadata poll thread
78 * for the metadata and the data poll thread for the data.
79 */
40dc48e0
DG
80static struct lttng_ht *metadata_ht;
81static struct lttng_ht *data_ht;
43c34bc3 82
acdb9057
DG
83/*
84 * Notify a thread lttng pipe to poll back again. This usually means that some
85 * global state has changed so we just send back the thread in a poll wait
86 * call.
87 */
88static void notify_thread_lttng_pipe(struct lttng_pipe *pipe)
89{
90 struct lttng_consumer_stream *null_stream = NULL;
91
92 assert(pipe);
93
94 (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream));
95}
96
5c635c72
MD
97static void notify_health_quit_pipe(int *pipe)
98{
6cd525e8 99 ssize_t ret;
5c635c72 100
6cd525e8
MD
101 ret = lttng_write(pipe[1], "4", 1);
102 if (ret < 1) {
5c635c72
MD
103 PERROR("write consumer health quit");
104 }
105}
106
d8ef542d
MD
107static void notify_channel_pipe(struct lttng_consumer_local_data *ctx,
108 struct lttng_consumer_channel *chan,
a0cbdd2e 109 uint64_t key,
d8ef542d
MD
110 enum consumer_channel_action action)
111{
112 struct consumer_channel_msg msg;
6cd525e8 113 ssize_t ret;
d8ef542d 114
e56251fc
DG
115 memset(&msg, 0, sizeof(msg));
116
d8ef542d
MD
117 msg.action = action;
118 msg.chan = chan;
f21dae48 119 msg.key = key;
6cd525e8
MD
120 ret = lttng_write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg));
121 if (ret < sizeof(msg)) {
122 PERROR("notify_channel_pipe write error");
123 }
d8ef542d
MD
124}
125
a0cbdd2e
MD
126void notify_thread_del_channel(struct lttng_consumer_local_data *ctx,
127 uint64_t key)
128{
129 notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL);
130}
131
d8ef542d
MD
132static int read_channel_pipe(struct lttng_consumer_local_data *ctx,
133 struct lttng_consumer_channel **chan,
a0cbdd2e 134 uint64_t *key,
d8ef542d
MD
135 enum consumer_channel_action *action)
136{
137 struct consumer_channel_msg msg;
6cd525e8 138 ssize_t ret;
d8ef542d 139
6cd525e8
MD
140 ret = lttng_read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg));
141 if (ret < sizeof(msg)) {
142 ret = -1;
143 goto error;
d8ef542d 144 }
6cd525e8
MD
145 *action = msg.action;
146 *chan = msg.chan;
147 *key = msg.key;
148error:
149 return (int) ret;
d8ef542d
MD
150}
151
3bd1e081
MD
152/*
153 * Find a stream. The consumer_data.lock must be locked during this
154 * call.
155 */
d88aee68 156static struct lttng_consumer_stream *find_stream(uint64_t key,
8389e4f8 157 struct lttng_ht *ht)
3bd1e081 158{
e4421fec 159 struct lttng_ht_iter iter;
d88aee68 160 struct lttng_ht_node_u64 *node;
e4421fec 161 struct lttng_consumer_stream *stream = NULL;
3bd1e081 162
8389e4f8
DG
163 assert(ht);
164
d88aee68
DG
165 /* -1ULL keys are lookup failures */
166 if (key == (uint64_t) -1ULL) {
7ad0a0cb 167 return NULL;
7a57cf92 168 }
e4421fec 169
6065ceec
DG
170 rcu_read_lock();
171
d88aee68
DG
172 lttng_ht_lookup(ht, &key, &iter);
173 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
174 if (node != NULL) {
175 stream = caa_container_of(node, struct lttng_consumer_stream, node);
3bd1e081 176 }
e4421fec 177
6065ceec
DG
178 rcu_read_unlock();
179
e4421fec 180 return stream;
3bd1e081
MD
181}
182
da009f2c 183static void steal_stream_key(uint64_t key, struct lttng_ht *ht)
7ad0a0cb
MD
184{
185 struct lttng_consumer_stream *stream;
186
04253271 187 rcu_read_lock();
ffe60014 188 stream = find_stream(key, ht);
04253271 189 if (stream) {
da009f2c 190 stream->key = (uint64_t) -1ULL;
04253271
MD
191 /*
192 * We don't want the lookup to match, but we still need
193 * to iterate on this stream when iterating over the hash table. Just
194 * change the node key.
195 */
da009f2c 196 stream->node.key = (uint64_t) -1ULL;
04253271
MD
197 }
198 rcu_read_unlock();
7ad0a0cb
MD
199}
200
d56db448
DG
201/*
202 * Return a channel object for the given key.
203 *
204 * RCU read side lock MUST be acquired before calling this function and
205 * protects the channel ptr.
206 */
d88aee68 207struct lttng_consumer_channel *consumer_find_channel(uint64_t key)
3bd1e081 208{
e4421fec 209 struct lttng_ht_iter iter;
d88aee68 210 struct lttng_ht_node_u64 *node;
e4421fec 211 struct lttng_consumer_channel *channel = NULL;
3bd1e081 212
d88aee68
DG
213 /* -1ULL keys are lookup failures */
214 if (key == (uint64_t) -1ULL) {
7ad0a0cb 215 return NULL;
7a57cf92 216 }
e4421fec 217
d88aee68
DG
218 lttng_ht_lookup(consumer_data.channel_ht, &key, &iter);
219 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
220 if (node != NULL) {
221 channel = caa_container_of(node, struct lttng_consumer_channel, node);
3bd1e081 222 }
e4421fec
DG
223
224 return channel;
3bd1e081
MD
225}
226
ffe60014 227static void free_stream_rcu(struct rcu_head *head)
7ad0a0cb 228{
d88aee68
DG
229 struct lttng_ht_node_u64 *node =
230 caa_container_of(head, struct lttng_ht_node_u64, head);
ffe60014
DG
231 struct lttng_consumer_stream *stream =
232 caa_container_of(node, struct lttng_consumer_stream, node);
7ad0a0cb 233
ffe60014 234 free(stream);
7ad0a0cb
MD
235}
236
ffe60014 237static void free_channel_rcu(struct rcu_head *head)
702b1ea4 238{
d88aee68
DG
239 struct lttng_ht_node_u64 *node =
240 caa_container_of(head, struct lttng_ht_node_u64, head);
ffe60014
DG
241 struct lttng_consumer_channel *channel =
242 caa_container_of(node, struct lttng_consumer_channel, node);
702b1ea4 243
ffe60014 244 free(channel);
702b1ea4
MD
245}
246
00e2e675
DG
247/*
248 * RCU protected relayd socket pair free.
249 */
ffe60014 250static void free_relayd_rcu(struct rcu_head *head)
00e2e675 251{
d88aee68
DG
252 struct lttng_ht_node_u64 *node =
253 caa_container_of(head, struct lttng_ht_node_u64, head);
00e2e675
DG
254 struct consumer_relayd_sock_pair *relayd =
255 caa_container_of(node, struct consumer_relayd_sock_pair, node);
256
8994307f
DG
257 /*
258 * Close all sockets. This is done in the call RCU since we don't want the
259 * socket fds to be reassigned thus potentially creating bad state of the
260 * relayd object.
261 *
262 * We do not have to lock the control socket mutex here since at this stage
263 * there is no one referencing to this relayd object.
264 */
265 (void) relayd_close(&relayd->control_sock);
266 (void) relayd_close(&relayd->data_sock);
267
00e2e675
DG
268 free(relayd);
269}
270
271/*
272 * Destroy and free relayd socket pair object.
00e2e675 273 */
51230d70 274void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
275{
276 int ret;
277 struct lttng_ht_iter iter;
278
173af62f
DG
279 if (relayd == NULL) {
280 return;
281 }
282
00e2e675
DG
283 DBG("Consumer destroy and close relayd socket pair");
284
285 iter.iter.node = &relayd->node.node;
286 ret = lttng_ht_del(consumer_data.relayd_ht, &iter);
173af62f 287 if (ret != 0) {
8994307f 288 /* We assume the relayd is being or is destroyed */
173af62f
DG
289 return;
290 }
00e2e675 291
00e2e675 292 /* RCU free() call */
ffe60014
DG
293 call_rcu(&relayd->node.head, free_relayd_rcu);
294}
295
296/*
297 * Remove a channel from the global list protected by a mutex. This function is
298 * also responsible for freeing its data structures.
299 */
300void consumer_del_channel(struct lttng_consumer_channel *channel)
301{
302 int ret;
303 struct lttng_ht_iter iter;
f2a444f1 304 struct lttng_consumer_stream *stream, *stmp;
ffe60014 305
d88aee68 306 DBG("Consumer delete channel key %" PRIu64, channel->key);
ffe60014
DG
307
308 pthread_mutex_lock(&consumer_data.lock);
a9838785 309 pthread_mutex_lock(&channel->lock);
ffe60014 310
51e762e5
JD
311 /* Delete streams that might have been left in the stream list. */
312 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
313 send_node) {
314 cds_list_del(&stream->send_node);
315 /*
316 * Once a stream is added to this list, the buffers were created so
317 * we have a guarantee that this call will succeed.
318 */
319 consumer_stream_destroy(stream, NULL);
320 }
321
d3e2ba59
JD
322 if (channel->live_timer_enabled == 1) {
323 consumer_timer_live_stop(channel);
324 }
325
ffe60014
DG
326 switch (consumer_data.type) {
327 case LTTNG_CONSUMER_KERNEL:
328 break;
329 case LTTNG_CONSUMER32_UST:
330 case LTTNG_CONSUMER64_UST:
331 lttng_ustconsumer_del_channel(channel);
332 break;
333 default:
334 ERR("Unknown consumer_data type");
335 assert(0);
336 goto end;
337 }
338
339 rcu_read_lock();
340 iter.iter.node = &channel->node.node;
341 ret = lttng_ht_del(consumer_data.channel_ht, &iter);
342 assert(!ret);
343 rcu_read_unlock();
344
345 call_rcu(&channel->node.head, free_channel_rcu);
346end:
a9838785 347 pthread_mutex_unlock(&channel->lock);
ffe60014 348 pthread_mutex_unlock(&consumer_data.lock);
00e2e675
DG
349}
350
228b5bf7
DG
351/*
352 * Iterate over the relayd hash table and destroy each element. Finally,
353 * destroy the whole hash table.
354 */
355static void cleanup_relayd_ht(void)
356{
357 struct lttng_ht_iter iter;
358 struct consumer_relayd_sock_pair *relayd;
359
360 rcu_read_lock();
361
362 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
363 node.node) {
51230d70 364 consumer_destroy_relayd(relayd);
228b5bf7
DG
365 }
366
228b5bf7 367 rcu_read_unlock();
36b588ed
MD
368
369 lttng_ht_destroy(consumer_data.relayd_ht);
228b5bf7
DG
370}
371
8994307f
DG
372/*
373 * Update the end point status of all streams having the given network sequence
374 * index (relayd index).
375 *
376 * It's atomically set without having the stream mutex locked which is fine
377 * because we handle the write/read race with a pipe wakeup for each thread.
378 */
da009f2c 379static void update_endpoint_status_by_netidx(uint64_t net_seq_idx,
8994307f
DG
380 enum consumer_endpoint_status status)
381{
382 struct lttng_ht_iter iter;
383 struct lttng_consumer_stream *stream;
384
da009f2c 385 DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx);
8994307f
DG
386
387 rcu_read_lock();
388
389 /* Let's begin with metadata */
390 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
391 if (stream->net_seq_idx == net_seq_idx) {
392 uatomic_set(&stream->endpoint_status, status);
393 DBG("Delete flag set to metadata stream %d", stream->wait_fd);
394 }
395 }
396
397 /* Follow up by the data streams */
398 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
399 if (stream->net_seq_idx == net_seq_idx) {
400 uatomic_set(&stream->endpoint_status, status);
401 DBG("Delete flag set to data stream %d", stream->wait_fd);
402 }
403 }
404 rcu_read_unlock();
405}
406
407/*
408 * Cleanup a relayd object by flagging every associated streams for deletion,
409 * destroying the object meaning removing it from the relayd hash table,
410 * closing the sockets and freeing the memory in a RCU call.
411 *
412 * If a local data context is available, notify the threads that the streams'
413 * state have changed.
414 */
415static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd,
416 struct lttng_consumer_local_data *ctx)
417{
da009f2c 418 uint64_t netidx;
8994307f
DG
419
420 assert(relayd);
421
9617607b
DG
422 DBG("Cleaning up relayd sockets");
423
8994307f
DG
424 /* Save the net sequence index before destroying the object */
425 netidx = relayd->net_seq_idx;
426
427 /*
428 * Delete the relayd from the relayd hash table, close the sockets and free
429 * the object in a RCU call.
430 */
51230d70 431 consumer_destroy_relayd(relayd);
8994307f
DG
432
433 /* Set inactive endpoint to all streams */
434 update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE);
435
436 /*
437 * With a local data context, notify the threads that the streams' state
438 * have changed. The write() action on the pipe acts as an "implicit"
439 * memory barrier ordering the updates of the end point status from the
440 * read of this status which happens AFTER receiving this notify.
441 */
442 if (ctx) {
acdb9057 443 notify_thread_lttng_pipe(ctx->consumer_data_pipe);
13886d2d 444 notify_thread_lttng_pipe(ctx->consumer_metadata_pipe);
8994307f
DG
445 }
446}
447
a6ba4fe1
DG
448/*
449 * Flag a relayd socket pair for destruction. Destroy it if the refcount
450 * reaches zero.
451 *
452 * RCU read side lock MUST be aquired before calling this function.
453 */
454void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd)
455{
456 assert(relayd);
457
458 /* Set destroy flag for this object */
459 uatomic_set(&relayd->destroy_flag, 1);
460
461 /* Destroy the relayd if refcount is 0 */
462 if (uatomic_read(&relayd->refcount) == 0) {
51230d70 463 consumer_destroy_relayd(relayd);
a6ba4fe1
DG
464 }
465}
466
3bd1e081 467/*
1d1a276c
DG
468 * Completly destroy stream from every visiable data structure and the given
469 * hash table if one.
470 *
471 * One this call returns, the stream object is not longer usable nor visible.
3bd1e081 472 */
e316aad5
DG
473void consumer_del_stream(struct lttng_consumer_stream *stream,
474 struct lttng_ht *ht)
3bd1e081 475{
1d1a276c 476 consumer_stream_destroy(stream, ht);
3bd1e081
MD
477}
478
5ab66908
MD
479/*
480 * XXX naming of del vs destroy is all mixed up.
481 */
482void consumer_del_stream_for_data(struct lttng_consumer_stream *stream)
483{
484 consumer_stream_destroy(stream, data_ht);
485}
486
487void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream)
488{
489 consumer_stream_destroy(stream, metadata_ht);
490}
491
d88aee68
DG
492struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key,
493 uint64_t stream_key,
3bd1e081 494 enum lttng_consumer_stream_state state,
ffe60014 495 const char *channel_name,
6df2e2c9 496 uid_t uid,
00e2e675 497 gid_t gid,
57a269f2 498 uint64_t relayd_id,
53632229 499 uint64_t session_id,
ffe60014
DG
500 int cpu,
501 int *alloc_ret,
4891ece8
DG
502 enum consumer_channel_type type,
503 unsigned int monitor)
3bd1e081 504{
ffe60014 505 int ret;
3bd1e081 506 struct lttng_consumer_stream *stream;
3bd1e081 507
effcf122 508 stream = zmalloc(sizeof(*stream));
3bd1e081 509 if (stream == NULL) {
7a57cf92 510 PERROR("malloc struct lttng_consumer_stream");
ffe60014 511 ret = -ENOMEM;
7a57cf92 512 goto end;
3bd1e081 513 }
7a57cf92 514
d56db448
DG
515 rcu_read_lock();
516
3bd1e081 517 stream->key = stream_key;
3bd1e081
MD
518 stream->out_fd = -1;
519 stream->out_fd_offset = 0;
e5d1a9b3 520 stream->output_written = 0;
3bd1e081 521 stream->state = state;
6df2e2c9
MD
522 stream->uid = uid;
523 stream->gid = gid;
ffe60014 524 stream->net_seq_idx = relayd_id;
53632229 525 stream->session_id = session_id;
4891ece8 526 stream->monitor = monitor;
774d490c 527 stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE;
309167d2 528 stream->index_fd = -1;
53632229 529 pthread_mutex_init(&stream->lock, NULL);
58b1f425 530
ffe60014
DG
531 /* If channel is the metadata, flag this stream as metadata. */
532 if (type == CONSUMER_CHANNEL_TYPE_METADATA) {
533 stream->metadata_flag = 1;
534 /* Metadata is flat out. */
535 strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name));
94d49140
JD
536 /* Live rendez-vous point. */
537 pthread_cond_init(&stream->metadata_rdv, NULL);
538 pthread_mutex_init(&stream->metadata_rdv_lock, NULL);
58b1f425 539 } else {
ffe60014
DG
540 /* Format stream name to <channel_name>_<cpu_number> */
541 ret = snprintf(stream->name, sizeof(stream->name), "%s_%d",
542 channel_name, cpu);
543 if (ret < 0) {
544 PERROR("snprintf stream name");
545 goto error;
546 }
58b1f425 547 }
c30aaa51 548
ffe60014 549 /* Key is always the wait_fd for streams. */
d88aee68 550 lttng_ht_node_init_u64(&stream->node, stream->key);
ffe60014 551
d8ef542d
MD
552 /* Init node per channel id key */
553 lttng_ht_node_init_u64(&stream->node_channel_id, channel_key);
554
53632229 555 /* Init session id node with the stream session id */
d88aee68 556 lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id);
53632229 557
07b86b52
JD
558 DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64
559 " relayd_id %" PRIu64 ", session_id %" PRIu64,
560 stream->name, stream->key, channel_key,
561 stream->net_seq_idx, stream->session_id);
d56db448
DG
562
563 rcu_read_unlock();
3bd1e081 564 return stream;
c80048c6
MD
565
566error:
d56db448 567 rcu_read_unlock();
c80048c6 568 free(stream);
7a57cf92 569end:
ffe60014
DG
570 if (alloc_ret) {
571 *alloc_ret = ret;
572 }
c80048c6 573 return NULL;
3bd1e081
MD
574}
575
576/*
577 * Add a stream to the global list protected by a mutex.
578 */
5ab66908 579int consumer_add_data_stream(struct lttng_consumer_stream *stream)
3bd1e081 580{
5ab66908 581 struct lttng_ht *ht = data_ht;
3bd1e081
MD
582 int ret = 0;
583
e316aad5 584 assert(stream);
43c34bc3 585 assert(ht);
c77fc10a 586
d88aee68 587 DBG3("Adding consumer stream %" PRIu64, stream->key);
e316aad5
DG
588
589 pthread_mutex_lock(&consumer_data.lock);
a9838785 590 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 591 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 592 pthread_mutex_lock(&stream->lock);
b0b335c8 593 rcu_read_lock();
e316aad5 594
43c34bc3 595 /* Steal stream identifier to avoid having streams with the same key */
ffe60014 596 steal_stream_key(stream->key, ht);
43c34bc3 597
d88aee68 598 lttng_ht_add_unique_u64(ht, &stream->node);
00e2e675 599
d8ef542d
MD
600 lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht,
601 &stream->node_channel_id);
602
ca22feea
DG
603 /*
604 * Add stream to the stream_list_ht of the consumer data. No need to steal
605 * the key since the HT does not use it and we allow to add redundant keys
606 * into this table.
607 */
d88aee68 608 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 609
e316aad5 610 /*
ffe60014
DG
611 * When nb_init_stream_left reaches 0, we don't need to trigger any action
612 * in terms of destroying the associated channel, because the action that
e316aad5
DG
613 * causes the count to become 0 also causes a stream to be added. The
614 * channel deletion will thus be triggered by the following removal of this
615 * stream.
616 */
ffe60014 617 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
618 /* Increment refcount before decrementing nb_init_stream_left */
619 cmm_smp_wmb();
ffe60014 620 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
621 }
622
623 /* Update consumer data once the node is inserted. */
3bd1e081
MD
624 consumer_data.stream_count++;
625 consumer_data.need_update = 1;
626
e316aad5 627 rcu_read_unlock();
2e818a6a 628 pthread_mutex_unlock(&stream->lock);
ec6ea7d0 629 pthread_mutex_unlock(&stream->chan->timer_lock);
a9838785 630 pthread_mutex_unlock(&stream->chan->lock);
3bd1e081 631 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 632
3bd1e081
MD
633 return ret;
634}
635
5ab66908
MD
636void consumer_del_data_stream(struct lttng_consumer_stream *stream)
637{
638 consumer_del_stream(stream, data_ht);
639}
640
00e2e675 641/*
3f8e211f
DG
642 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
643 * be acquired before calling this.
00e2e675 644 */
d09e1200 645static int add_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
646{
647 int ret = 0;
d88aee68 648 struct lttng_ht_node_u64 *node;
00e2e675
DG
649 struct lttng_ht_iter iter;
650
ffe60014 651 assert(relayd);
00e2e675 652
00e2e675 653 lttng_ht_lookup(consumer_data.relayd_ht,
d88aee68
DG
654 &relayd->net_seq_idx, &iter);
655 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675 656 if (node != NULL) {
00e2e675
DG
657 goto end;
658 }
d88aee68 659 lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node);
00e2e675 660
00e2e675
DG
661end:
662 return ret;
663}
664
665/*
666 * Allocate and return a consumer relayd socket.
667 */
668struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair(
da009f2c 669 uint64_t net_seq_idx)
00e2e675
DG
670{
671 struct consumer_relayd_sock_pair *obj = NULL;
672
da009f2c
MD
673 /* net sequence index of -1 is a failure */
674 if (net_seq_idx == (uint64_t) -1ULL) {
00e2e675
DG
675 goto error;
676 }
677
678 obj = zmalloc(sizeof(struct consumer_relayd_sock_pair));
679 if (obj == NULL) {
680 PERROR("zmalloc relayd sock");
681 goto error;
682 }
683
684 obj->net_seq_idx = net_seq_idx;
685 obj->refcount = 0;
173af62f 686 obj->destroy_flag = 0;
f96e4545
MD
687 obj->control_sock.sock.fd = -1;
688 obj->data_sock.sock.fd = -1;
d88aee68 689 lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx);
00e2e675
DG
690 pthread_mutex_init(&obj->ctrl_sock_mutex, NULL);
691
692error:
693 return obj;
694}
695
696/*
697 * Find a relayd socket pair in the global consumer data.
698 *
699 * Return the object if found else NULL.
b0b335c8
MD
700 * RCU read-side lock must be held across this call and while using the
701 * returned object.
00e2e675 702 */
d88aee68 703struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key)
00e2e675
DG
704{
705 struct lttng_ht_iter iter;
d88aee68 706 struct lttng_ht_node_u64 *node;
00e2e675
DG
707 struct consumer_relayd_sock_pair *relayd = NULL;
708
709 /* Negative keys are lookup failures */
d88aee68 710 if (key == (uint64_t) -1ULL) {
00e2e675
DG
711 goto error;
712 }
713
d88aee68 714 lttng_ht_lookup(consumer_data.relayd_ht, &key,
00e2e675 715 &iter);
d88aee68 716 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675
DG
717 if (node != NULL) {
718 relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node);
719 }
720
00e2e675
DG
721error:
722 return relayd;
723}
724
10a50311
JD
725/*
726 * Find a relayd and send the stream
727 *
728 * Returns 0 on success, < 0 on error
729 */
730int consumer_send_relayd_stream(struct lttng_consumer_stream *stream,
731 char *path)
732{
733 int ret = 0;
734 struct consumer_relayd_sock_pair *relayd;
735
736 assert(stream);
737 assert(stream->net_seq_idx != -1ULL);
738 assert(path);
739
740 /* The stream is not metadata. Get relayd reference if exists. */
741 rcu_read_lock();
742 relayd = consumer_find_relayd(stream->net_seq_idx);
743 if (relayd != NULL) {
744 /* Add stream on the relayd */
745 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
746 ret = relayd_add_stream(&relayd->control_sock, stream->name,
747 path, &stream->relayd_stream_id,
748 stream->chan->tracefile_size, stream->chan->tracefile_count);
749 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
750 if (ret < 0) {
751 goto end;
752 }
1c20f0e2 753
10a50311 754 uatomic_inc(&relayd->refcount);
d01178b6 755 stream->sent_to_relayd = 1;
10a50311
JD
756 } else {
757 ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.",
758 stream->key, stream->net_seq_idx);
759 ret = -1;
760 goto end;
761 }
762
763 DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64,
764 stream->name, stream->key, stream->net_seq_idx);
765
766end:
767 rcu_read_unlock();
768 return ret;
769}
770
771/*
772 * Find a relayd and close the stream
773 */
774void close_relayd_stream(struct lttng_consumer_stream *stream)
775{
776 struct consumer_relayd_sock_pair *relayd;
777
778 /* The stream is not metadata. Get relayd reference if exists. */
779 rcu_read_lock();
780 relayd = consumer_find_relayd(stream->net_seq_idx);
781 if (relayd) {
782 consumer_stream_relayd_close(stream, relayd);
783 }
784 rcu_read_unlock();
785}
786
00e2e675
DG
787/*
788 * Handle stream for relayd transmission if the stream applies for network
789 * streaming where the net sequence index is set.
790 *
791 * Return destination file descriptor or negative value on error.
792 */
6197aea7 793static int write_relayd_stream_header(struct lttng_consumer_stream *stream,
1d4dfdef
DG
794 size_t data_size, unsigned long padding,
795 struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
796{
797 int outfd = -1, ret;
00e2e675
DG
798 struct lttcomm_relayd_data_hdr data_hdr;
799
800 /* Safety net */
801 assert(stream);
6197aea7 802 assert(relayd);
00e2e675
DG
803
804 /* Reset data header */
805 memset(&data_hdr, 0, sizeof(data_hdr));
806
00e2e675
DG
807 if (stream->metadata_flag) {
808 /* Caller MUST acquire the relayd control socket lock */
809 ret = relayd_send_metadata(&relayd->control_sock, data_size);
810 if (ret < 0) {
811 goto error;
812 }
813
814 /* Metadata are always sent on the control socket. */
6151a90f 815 outfd = relayd->control_sock.sock.fd;
00e2e675
DG
816 } else {
817 /* Set header with stream information */
818 data_hdr.stream_id = htobe64(stream->relayd_stream_id);
819 data_hdr.data_size = htobe32(data_size);
1d4dfdef 820 data_hdr.padding_size = htobe32(padding);
39df6d9f
DG
821 /*
822 * Note that net_seq_num below is assigned with the *current* value of
823 * next_net_seq_num and only after that the next_net_seq_num will be
824 * increment. This is why when issuing a command on the relayd using
825 * this next value, 1 should always be substracted in order to compare
826 * the last seen sequence number on the relayd side to the last sent.
827 */
3604f373 828 data_hdr.net_seq_num = htobe64(stream->next_net_seq_num);
00e2e675
DG
829 /* Other fields are zeroed previously */
830
831 ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr,
832 sizeof(data_hdr));
833 if (ret < 0) {
834 goto error;
835 }
836
3604f373
DG
837 ++stream->next_net_seq_num;
838
00e2e675 839 /* Set to go on data socket */
6151a90f 840 outfd = relayd->data_sock.sock.fd;
00e2e675
DG
841 }
842
843error:
844 return outfd;
845}
846
3bd1e081 847/*
ffe60014
DG
848 * Allocate and return a new lttng_consumer_channel object using the given key
849 * to initialize the hash table node.
850 *
851 * On error, return NULL.
3bd1e081 852 */
886224ff 853struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key,
ffe60014
DG
854 uint64_t session_id,
855 const char *pathname,
856 const char *name,
857 uid_t uid,
858 gid_t gid,
57a269f2 859 uint64_t relayd_id,
1624d5b7
JD
860 enum lttng_event_output output,
861 uint64_t tracefile_size,
2bba9e53 862 uint64_t tracefile_count,
1950109e 863 uint64_t session_id_per_pid,
ecc48a90
JD
864 unsigned int monitor,
865 unsigned int live_timer_interval)
3bd1e081
MD
866{
867 struct lttng_consumer_channel *channel;
3bd1e081 868
276b26d1 869 channel = zmalloc(sizeof(*channel));
3bd1e081 870 if (channel == NULL) {
7a57cf92 871 PERROR("malloc struct lttng_consumer_channel");
3bd1e081
MD
872 goto end;
873 }
ffe60014
DG
874
875 channel->key = key;
3bd1e081 876 channel->refcount = 0;
ffe60014 877 channel->session_id = session_id;
1950109e 878 channel->session_id_per_pid = session_id_per_pid;
ffe60014
DG
879 channel->uid = uid;
880 channel->gid = gid;
881 channel->relayd_id = relayd_id;
882 channel->output = output;
1624d5b7
JD
883 channel->tracefile_size = tracefile_size;
884 channel->tracefile_count = tracefile_count;
2bba9e53 885 channel->monitor = monitor;
ecc48a90 886 channel->live_timer_interval = live_timer_interval;
a9838785 887 pthread_mutex_init(&channel->lock, NULL);
ec6ea7d0 888 pthread_mutex_init(&channel->timer_lock, NULL);
ffe60014 889
07b86b52
JD
890 /*
891 * In monitor mode, the streams associated with the channel will be put in
892 * a special list ONLY owned by this channel. So, the refcount is set to 1
893 * here meaning that the channel itself has streams that are referenced.
894 *
895 * On a channel deletion, once the channel is no longer visible, the
896 * refcount is decremented and checked for a zero value to delete it. With
897 * streams in no monitor mode, it will now be safe to destroy the channel.
898 */
899 if (!channel->monitor) {
900 channel->refcount = 1;
901 }
902
ffe60014
DG
903 strncpy(channel->pathname, pathname, sizeof(channel->pathname));
904 channel->pathname[sizeof(channel->pathname) - 1] = '\0';
905
906 strncpy(channel->name, name, sizeof(channel->name));
907 channel->name[sizeof(channel->name) - 1] = '\0';
908
d88aee68 909 lttng_ht_node_init_u64(&channel->node, channel->key);
d8ef542d
MD
910
911 channel->wait_fd = -1;
912
ffe60014
DG
913 CDS_INIT_LIST_HEAD(&channel->streams.head);
914
d88aee68 915 DBG("Allocated channel (key %" PRIu64 ")", channel->key)
3bd1e081 916
3bd1e081
MD
917end:
918 return channel;
919}
920
921/*
922 * Add a channel to the global list protected by a mutex.
821fffb2
DG
923 *
924 * On success 0 is returned else a negative value.
3bd1e081 925 */
d8ef542d
MD
926int consumer_add_channel(struct lttng_consumer_channel *channel,
927 struct lttng_consumer_local_data *ctx)
3bd1e081 928{
ffe60014 929 int ret = 0;
d88aee68 930 struct lttng_ht_node_u64 *node;
c77fc10a
DG
931 struct lttng_ht_iter iter;
932
3bd1e081 933 pthread_mutex_lock(&consumer_data.lock);
a9838785 934 pthread_mutex_lock(&channel->lock);
ec6ea7d0 935 pthread_mutex_lock(&channel->timer_lock);
6065ceec 936 rcu_read_lock();
c77fc10a 937
7972aab2 938 lttng_ht_lookup(consumer_data.channel_ht, &channel->key, &iter);
d88aee68 939 node = lttng_ht_iter_get_node_u64(&iter);
c77fc10a
DG
940 if (node != NULL) {
941 /* Channel already exist. Ignore the insertion */
d88aee68
DG
942 ERR("Consumer add channel key %" PRIu64 " already exists!",
943 channel->key);
821fffb2 944 ret = -EEXIST;
c77fc10a
DG
945 goto end;
946 }
947
d88aee68 948 lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node);
c77fc10a
DG
949
950end:
6065ceec 951 rcu_read_unlock();
ec6ea7d0 952 pthread_mutex_unlock(&channel->timer_lock);
a9838785 953 pthread_mutex_unlock(&channel->lock);
3bd1e081 954 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 955
d8ef542d 956 if (!ret && channel->wait_fd != -1 &&
10a50311 957 channel->type == CONSUMER_CHANNEL_TYPE_DATA) {
a0cbdd2e 958 notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD);
d8ef542d 959 }
ffe60014 960 return ret;
3bd1e081
MD
961}
962
963/*
964 * Allocate the pollfd structure and the local view of the out fds to avoid
965 * doing a lookup in the linked list and concurrency issues when writing is
966 * needed. Called with consumer_data.lock held.
967 *
968 * Returns the number of fds in the structures.
969 */
ffe60014
DG
970static int update_poll_array(struct lttng_consumer_local_data *ctx,
971 struct pollfd **pollfd, struct lttng_consumer_stream **local_stream,
972 struct lttng_ht *ht)
3bd1e081 973{
3bd1e081 974 int i = 0;
e4421fec
DG
975 struct lttng_ht_iter iter;
976 struct lttng_consumer_stream *stream;
3bd1e081 977
ffe60014
DG
978 assert(ctx);
979 assert(ht);
980 assert(pollfd);
981 assert(local_stream);
982
3bd1e081 983 DBG("Updating poll fd array");
481d6c57 984 rcu_read_lock();
43c34bc3 985 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
8994307f
DG
986 /*
987 * Only active streams with an active end point can be added to the
988 * poll set and local stream storage of the thread.
989 *
990 * There is a potential race here for endpoint_status to be updated
991 * just after the check. However, this is OK since the stream(s) will
992 * be deleted once the thread is notified that the end point state has
993 * changed where this function will be called back again.
994 */
995 if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM ||
79d4ffb7 996 stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) {
3bd1e081
MD
997 continue;
998 }
7972aab2
DG
999 /*
1000 * This clobbers way too much the debug output. Uncomment that if you
1001 * need it for debugging purposes.
1002 *
1003 * DBG("Active FD %d", stream->wait_fd);
1004 */
e4421fec 1005 (*pollfd)[i].fd = stream->wait_fd;
3bd1e081 1006 (*pollfd)[i].events = POLLIN | POLLPRI;
e4421fec 1007 local_stream[i] = stream;
3bd1e081
MD
1008 i++;
1009 }
481d6c57 1010 rcu_read_unlock();
3bd1e081
MD
1011
1012 /*
50f8ae69 1013 * Insert the consumer_data_pipe at the end of the array and don't
3bd1e081
MD
1014 * increment i so nb_fd is the number of real FD.
1015 */
acdb9057 1016 (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe);
509bb1cf 1017 (*pollfd)[i].events = POLLIN | POLLPRI;
3bd1e081
MD
1018 return i;
1019}
1020
1021/*
1022 * Poll on the should_quit pipe and the command socket return -1 on error and
1023 * should exit, 0 if data is available on the command socket
1024 */
1025int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll)
1026{
1027 int num_rdy;
1028
88f2b785 1029restart:
3bd1e081
MD
1030 num_rdy = poll(consumer_sockpoll, 2, -1);
1031 if (num_rdy == -1) {
88f2b785
MD
1032 /*
1033 * Restart interrupted system call.
1034 */
1035 if (errno == EINTR) {
1036 goto restart;
1037 }
7a57cf92 1038 PERROR("Poll error");
3bd1e081
MD
1039 goto exit;
1040 }
509bb1cf 1041 if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) {
3bd1e081
MD
1042 DBG("consumer_should_quit wake up");
1043 goto exit;
1044 }
1045 return 0;
1046
1047exit:
1048 return -1;
1049}
1050
1051/*
1052 * Set the error socket.
1053 */
ffe60014
DG
1054void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx,
1055 int sock)
3bd1e081
MD
1056{
1057 ctx->consumer_error_socket = sock;
1058}
1059
1060/*
1061 * Set the command socket path.
1062 */
3bd1e081
MD
1063void lttng_consumer_set_command_sock_path(
1064 struct lttng_consumer_local_data *ctx, char *sock)
1065{
1066 ctx->consumer_command_sock_path = sock;
1067}
1068
1069/*
1070 * Send return code to the session daemon.
1071 * If the socket is not defined, we return 0, it is not a fatal error
1072 */
ffe60014 1073int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd)
3bd1e081
MD
1074{
1075 if (ctx->consumer_error_socket > 0) {
1076 return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd,
1077 sizeof(enum lttcomm_sessiond_command));
1078 }
1079
1080 return 0;
1081}
1082
1083/*
228b5bf7
DG
1084 * Close all the tracefiles and stream fds and MUST be called when all
1085 * instances are destroyed i.e. when all threads were joined and are ended.
3bd1e081
MD
1086 */
1087void lttng_consumer_cleanup(void)
1088{
e4421fec 1089 struct lttng_ht_iter iter;
ffe60014 1090 struct lttng_consumer_channel *channel;
6065ceec
DG
1091
1092 rcu_read_lock();
3bd1e081 1093
ffe60014
DG
1094 cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel,
1095 node.node) {
702b1ea4 1096 consumer_del_channel(channel);
3bd1e081 1097 }
6065ceec
DG
1098
1099 rcu_read_unlock();
d6ce1df2 1100
d6ce1df2 1101 lttng_ht_destroy(consumer_data.channel_ht);
228b5bf7
DG
1102
1103 cleanup_relayd_ht();
1104
d8ef542d
MD
1105 lttng_ht_destroy(consumer_data.stream_per_chan_id_ht);
1106
228b5bf7
DG
1107 /*
1108 * This HT contains streams that are freed by either the metadata thread or
1109 * the data thread so we do *nothing* on the hash table and simply destroy
1110 * it.
1111 */
1112 lttng_ht_destroy(consumer_data.stream_list_ht);
3bd1e081
MD
1113}
1114
1115/*
1116 * Called from signal handler.
1117 */
1118void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx)
1119{
6cd525e8
MD
1120 ssize_t ret;
1121
3bd1e081 1122 consumer_quit = 1;
6cd525e8
MD
1123 ret = lttng_write(ctx->consumer_should_quit[1], "4", 1);
1124 if (ret < 1) {
7a57cf92 1125 PERROR("write consumer quit");
3bd1e081 1126 }
ab1027f4
DG
1127
1128 DBG("Consumer flag that it should quit");
3bd1e081
MD
1129}
1130
00e2e675
DG
1131void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream,
1132 off_t orig_offset)
3bd1e081
MD
1133{
1134 int outfd = stream->out_fd;
1135
1136 /*
1137 * This does a blocking write-and-wait on any page that belongs to the
1138 * subbuffer prior to the one we just wrote.
1139 * Don't care about error values, as these are just hints and ways to
1140 * limit the amount of page cache used.
1141 */
ffe60014 1142 if (orig_offset < stream->max_sb_size) {
3bd1e081
MD
1143 return;
1144 }
ffe60014
DG
1145 lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size,
1146 stream->max_sb_size,
3bd1e081
MD
1147 SYNC_FILE_RANGE_WAIT_BEFORE
1148 | SYNC_FILE_RANGE_WRITE
1149 | SYNC_FILE_RANGE_WAIT_AFTER);
1150 /*
1151 * Give hints to the kernel about how we access the file:
1152 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1153 * we write it.
1154 *
1155 * We need to call fadvise again after the file grows because the
1156 * kernel does not seem to apply fadvise to non-existing parts of the
1157 * file.
1158 *
1159 * Call fadvise _after_ having waited for the page writeback to
1160 * complete because the dirty page writeback semantic is not well
1161 * defined. So it can be expected to lead to lower throughput in
1162 * streaming.
1163 */
ffe60014
DG
1164 posix_fadvise(outfd, orig_offset - stream->max_sb_size,
1165 stream->max_sb_size, POSIX_FADV_DONTNEED);
3bd1e081
MD
1166}
1167
1168/*
1169 * Initialise the necessary environnement :
1170 * - create a new context
1171 * - create the poll_pipe
1172 * - create the should_quit pipe (for signal handler)
1173 * - create the thread pipe (for splice)
1174 *
1175 * Takes a function pointer as argument, this function is called when data is
1176 * available on a buffer. This function is responsible to do the
1177 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1178 * buffer configuration and then kernctl_put_next_subbuf at the end.
1179 *
1180 * Returns a pointer to the new context or NULL on error.
1181 */
1182struct lttng_consumer_local_data *lttng_consumer_create(
1183 enum lttng_consumer_type type,
4078b776 1184 ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream,
d41f73b7 1185 struct lttng_consumer_local_data *ctx),
3bd1e081
MD
1186 int (*recv_channel)(struct lttng_consumer_channel *channel),
1187 int (*recv_stream)(struct lttng_consumer_stream *stream),
30319bcb 1188 int (*update_stream)(uint64_t stream_key, uint32_t state))
3bd1e081 1189{
d8ef542d 1190 int ret;
3bd1e081
MD
1191 struct lttng_consumer_local_data *ctx;
1192
1193 assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN ||
1194 consumer_data.type == type);
1195 consumer_data.type = type;
1196
effcf122 1197 ctx = zmalloc(sizeof(struct lttng_consumer_local_data));
3bd1e081 1198 if (ctx == NULL) {
7a57cf92 1199 PERROR("allocating context");
3bd1e081
MD
1200 goto error;
1201 }
1202
1203 ctx->consumer_error_socket = -1;
331744e3 1204 ctx->consumer_metadata_socket = -1;
75d83e50 1205 pthread_mutex_init(&ctx->metadata_socket_lock, NULL);
3bd1e081
MD
1206 /* assign the callbacks */
1207 ctx->on_buffer_ready = buffer_ready;
1208 ctx->on_recv_channel = recv_channel;
1209 ctx->on_recv_stream = recv_stream;
1210 ctx->on_update_stream = update_stream;
1211
acdb9057
DG
1212 ctx->consumer_data_pipe = lttng_pipe_open(0);
1213 if (!ctx->consumer_data_pipe) {
3bd1e081
MD
1214 goto error_poll_pipe;
1215 }
1216
1217 ret = pipe(ctx->consumer_should_quit);
1218 if (ret < 0) {
7a57cf92 1219 PERROR("Error creating recv pipe");
3bd1e081
MD
1220 goto error_quit_pipe;
1221 }
1222
1223 ret = pipe(ctx->consumer_thread_pipe);
1224 if (ret < 0) {
7a57cf92 1225 PERROR("Error creating thread pipe");
3bd1e081
MD
1226 goto error_thread_pipe;
1227 }
1228
d8ef542d
MD
1229 ret = pipe(ctx->consumer_channel_pipe);
1230 if (ret < 0) {
1231 PERROR("Error creating channel pipe");
1232 goto error_channel_pipe;
1233 }
1234
13886d2d
DG
1235 ctx->consumer_metadata_pipe = lttng_pipe_open(0);
1236 if (!ctx->consumer_metadata_pipe) {
fb3a43a9
DG
1237 goto error_metadata_pipe;
1238 }
3bd1e081 1239
fb3a43a9
DG
1240 ret = utils_create_pipe(ctx->consumer_splice_metadata_pipe);
1241 if (ret < 0) {
1242 goto error_splice_pipe;
1243 }
1244
1245 return ctx;
3bd1e081 1246
fb3a43a9 1247error_splice_pipe:
13886d2d 1248 lttng_pipe_destroy(ctx->consumer_metadata_pipe);
fb3a43a9 1249error_metadata_pipe:
d8ef542d
MD
1250 utils_close_pipe(ctx->consumer_channel_pipe);
1251error_channel_pipe:
fb3a43a9 1252 utils_close_pipe(ctx->consumer_thread_pipe);
3bd1e081 1253error_thread_pipe:
d8ef542d 1254 utils_close_pipe(ctx->consumer_should_quit);
3bd1e081 1255error_quit_pipe:
acdb9057 1256 lttng_pipe_destroy(ctx->consumer_data_pipe);
3bd1e081
MD
1257error_poll_pipe:
1258 free(ctx);
1259error:
1260 return NULL;
1261}
1262
1263/*
1264 * Close all fds associated with the instance and free the context.
1265 */
1266void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx)
1267{
4c462e79
MD
1268 int ret;
1269
ab1027f4
DG
1270 DBG("Consumer destroying it. Closing everything.");
1271
4c462e79
MD
1272 ret = close(ctx->consumer_error_socket);
1273 if (ret) {
1274 PERROR("close");
1275 }
331744e3
JD
1276 ret = close(ctx->consumer_metadata_socket);
1277 if (ret) {
1278 PERROR("close");
1279 }
d8ef542d
MD
1280 utils_close_pipe(ctx->consumer_thread_pipe);
1281 utils_close_pipe(ctx->consumer_channel_pipe);
acdb9057 1282 lttng_pipe_destroy(ctx->consumer_data_pipe);
13886d2d 1283 lttng_pipe_destroy(ctx->consumer_metadata_pipe);
d8ef542d 1284 utils_close_pipe(ctx->consumer_should_quit);
fb3a43a9
DG
1285 utils_close_pipe(ctx->consumer_splice_metadata_pipe);
1286
3bd1e081
MD
1287 unlink(ctx->consumer_command_sock_path);
1288 free(ctx);
1289}
1290
6197aea7
DG
1291/*
1292 * Write the metadata stream id on the specified file descriptor.
1293 */
1294static int write_relayd_metadata_id(int fd,
1295 struct lttng_consumer_stream *stream,
ffe60014 1296 struct consumer_relayd_sock_pair *relayd, unsigned long padding)
6197aea7 1297{
6cd525e8 1298 ssize_t ret;
1d4dfdef 1299 struct lttcomm_relayd_metadata_payload hdr;
6197aea7 1300
1d4dfdef
DG
1301 hdr.stream_id = htobe64(stream->relayd_stream_id);
1302 hdr.padding_size = htobe32(padding);
6cd525e8
MD
1303 ret = lttng_write(fd, (void *) &hdr, sizeof(hdr));
1304 if (ret < sizeof(hdr)) {
d7b75ec8
DG
1305 /*
1306 * This error means that the fd's end is closed so ignore the perror
1307 * not to clubber the error output since this can happen in a normal
1308 * code path.
1309 */
1310 if (errno != EPIPE) {
1311 PERROR("write metadata stream id");
1312 }
1313 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno);
534d2592
DG
1314 /*
1315 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1316 * handle writting the missing part so report that as an error and
1317 * don't lie to the caller.
1318 */
1319 ret = -1;
6197aea7
DG
1320 goto end;
1321 }
1d4dfdef
DG
1322 DBG("Metadata stream id %" PRIu64 " with padding %lu written before data",
1323 stream->relayd_stream_id, padding);
6197aea7
DG
1324
1325end:
6cd525e8 1326 return (int) ret;
6197aea7
DG
1327}
1328
3bd1e081 1329/*
09e26845
DG
1330 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1331 * core function for writing trace buffers to either the local filesystem or
1332 * the network.
1333 *
79d4ffb7
DG
1334 * It must be called with the stream lock held.
1335 *
09e26845 1336 * Careful review MUST be put if any changes occur!
3bd1e081
MD
1337 *
1338 * Returns the number of bytes written
1339 */
4078b776 1340ssize_t lttng_consumer_on_read_subbuffer_mmap(
3bd1e081 1341 struct lttng_consumer_local_data *ctx,
1d4dfdef 1342 struct lttng_consumer_stream *stream, unsigned long len,
309167d2
JD
1343 unsigned long padding,
1344 struct lttng_packet_index *index)
3bd1e081 1345{
f02e1e8a 1346 unsigned long mmap_offset;
ffe60014 1347 void *mmap_base;
f02e1e8a
DG
1348 ssize_t ret = 0, written = 0;
1349 off_t orig_offset = stream->out_fd_offset;
1350 /* Default is on the disk */
1351 int outfd = stream->out_fd;
f02e1e8a 1352 struct consumer_relayd_sock_pair *relayd = NULL;
8994307f 1353 unsigned int relayd_hang_up = 0;
f02e1e8a
DG
1354
1355 /* RCU lock for the relayd pointer */
1356 rcu_read_lock();
1357
1358 /* Flag that the current stream if set for network streaming. */
da009f2c 1359 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1360 relayd = consumer_find_relayd(stream->net_seq_idx);
1361 if (relayd == NULL) {
56591bac 1362 ret = -EPIPE;
f02e1e8a
DG
1363 goto end;
1364 }
1365 }
1366
1367 /* get the offset inside the fd to mmap */
3bd1e081
MD
1368 switch (consumer_data.type) {
1369 case LTTNG_CONSUMER_KERNEL:
ffe60014 1370 mmap_base = stream->mmap_base;
f02e1e8a 1371 ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset);
56591bac
MD
1372 if (ret != 0) {
1373 PERROR("tracer ctl get_mmap_read_offset");
1374 written = -errno;
1375 goto end;
1376 }
f02e1e8a 1377 break;
7753dea8
MD
1378 case LTTNG_CONSUMER32_UST:
1379 case LTTNG_CONSUMER64_UST:
ffe60014
DG
1380 mmap_base = lttng_ustctl_get_mmap_base(stream);
1381 if (!mmap_base) {
1382 ERR("read mmap get mmap base for stream %s", stream->name);
56591bac 1383 written = -EPERM;
ffe60014
DG
1384 goto end;
1385 }
1386 ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset);
56591bac
MD
1387 if (ret != 0) {
1388 PERROR("tracer ctl get_mmap_read_offset");
1389 written = ret;
1390 goto end;
1391 }
f02e1e8a 1392 break;
3bd1e081
MD
1393 default:
1394 ERR("Unknown consumer_data type");
1395 assert(0);
1396 }
b9182dd9 1397
f02e1e8a
DG
1398 /* Handle stream on the relayd if the output is on the network */
1399 if (relayd) {
1400 unsigned long netlen = len;
1401
1402 /*
1403 * Lock the control socket for the complete duration of the function
1404 * since from this point on we will use the socket.
1405 */
1406 if (stream->metadata_flag) {
1407 /* Metadata requires the control socket. */
1408 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1d4dfdef 1409 netlen += sizeof(struct lttcomm_relayd_metadata_payload);
f02e1e8a
DG
1410 }
1411
1d4dfdef 1412 ret = write_relayd_stream_header(stream, netlen, padding, relayd);
f02e1e8a
DG
1413 if (ret >= 0) {
1414 /* Use the returned socket. */
1415 outfd = ret;
1416
1417 /* Write metadata stream id before payload */
1418 if (stream->metadata_flag) {
1d4dfdef 1419 ret = write_relayd_metadata_id(outfd, stream, relayd, padding);
f02e1e8a 1420 if (ret < 0) {
f02e1e8a 1421 written = ret;
8994307f
DG
1422 /* Socket operation failed. We consider the relayd dead */
1423 if (ret == -EPIPE || ret == -EINVAL) {
1424 relayd_hang_up = 1;
1425 goto write_error;
1426 }
f02e1e8a
DG
1427 goto end;
1428 }
f02e1e8a 1429 }
8994307f
DG
1430 } else {
1431 /* Socket operation failed. We consider the relayd dead */
1432 if (ret == -EPIPE || ret == -EINVAL) {
1433 relayd_hang_up = 1;
1434 goto write_error;
1435 }
1436 /* Else, use the default set before which is the filesystem. */
f02e1e8a 1437 }
1d4dfdef
DG
1438 } else {
1439 /* No streaming, we have to set the len with the full padding */
1440 len += padding;
1624d5b7
JD
1441
1442 /*
1443 * Check if we need to change the tracefile before writing the packet.
1444 */
1445 if (stream->chan->tracefile_size > 0 &&
1446 (stream->tracefile_size_current + len) >
1447 stream->chan->tracefile_size) {
fe4477ee
JD
1448 ret = utils_rotate_stream_file(stream->chan->pathname,
1449 stream->name, stream->chan->tracefile_size,
1450 stream->chan->tracefile_count, stream->uid, stream->gid,
309167d2
JD
1451 stream->out_fd, &(stream->tracefile_count_current),
1452 &stream->out_fd);
1624d5b7
JD
1453 if (ret < 0) {
1454 ERR("Rotating output file");
1455 goto end;
1456 }
309167d2
JD
1457 outfd = stream->out_fd;
1458
1459 if (stream->index_fd >= 0) {
1460 ret = index_create_file(stream->chan->pathname,
1461 stream->name, stream->uid, stream->gid,
1462 stream->chan->tracefile_size,
1463 stream->tracefile_count_current);
1464 if (ret < 0) {
1465 goto end;
1466 }
1467 stream->index_fd = ret;
1468 }
1469
a6976990
DG
1470 /* Reset current size because we just perform a rotation. */
1471 stream->tracefile_size_current = 0;
a1ae300f
JD
1472 stream->out_fd_offset = 0;
1473 orig_offset = 0;
1624d5b7
JD
1474 }
1475 stream->tracefile_size_current += len;
309167d2
JD
1476 if (index) {
1477 index->offset = htobe64(stream->out_fd_offset);
1478 }
f02e1e8a
DG
1479 }
1480
1481 while (len > 0) {
6cd525e8 1482 ret = lttng_write(outfd, mmap_base + mmap_offset, len);
1d4dfdef 1483 DBG("Consumer mmap write() ret %zd (len %lu)", ret, len);
6cd525e8 1484 if (ret < len) {
c5c45efa
DG
1485 /*
1486 * This is possible if the fd is closed on the other side (outfd)
1487 * or any write problem. It can be verbose a bit for a normal
1488 * execution if for instance the relayd is stopped abruptly. This
1489 * can happen so set this to a DBG statement.
1490 */
1491 DBG("Error in file write mmap");
f02e1e8a 1492 if (written == 0) {
56591bac 1493 written = -errno;
f02e1e8a 1494 }
8994307f
DG
1495 /* Socket operation failed. We consider the relayd dead */
1496 if (errno == EPIPE || errno == EINVAL) {
1497 relayd_hang_up = 1;
1498 goto write_error;
1499 }
f02e1e8a
DG
1500 goto end;
1501 } else if (ret > len) {
77c7c900 1502 PERROR("Error in file write (ret %zd > len %lu)", ret, len);
f02e1e8a
DG
1503 written += ret;
1504 goto end;
1505 } else {
1506 len -= ret;
1507 mmap_offset += ret;
1508 }
f02e1e8a
DG
1509
1510 /* This call is useless on a socket so better save a syscall. */
1511 if (!relayd) {
1512 /* This won't block, but will start writeout asynchronously */
1513 lttng_sync_file_range(outfd, stream->out_fd_offset, ret,
1514 SYNC_FILE_RANGE_WRITE);
1515 stream->out_fd_offset += ret;
1516 }
e5d1a9b3 1517 stream->output_written += ret;
f02e1e8a
DG
1518 written += ret;
1519 }
1520 lttng_consumer_sync_trace_file(stream, orig_offset);
1521
8994307f
DG
1522write_error:
1523 /*
1524 * This is a special case that the relayd has closed its socket. Let's
1525 * cleanup the relayd object and all associated streams.
1526 */
1527 if (relayd && relayd_hang_up) {
1528 cleanup_relayd(relayd, ctx);
1529 }
1530
f02e1e8a
DG
1531end:
1532 /* Unlock only if ctrl socket used */
1533 if (relayd && stream->metadata_flag) {
1534 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1535 }
1536
1537 rcu_read_unlock();
1538 return written;
3bd1e081
MD
1539}
1540
1541/*
1542 * Splice the data from the ring buffer to the tracefile.
1543 *
79d4ffb7
DG
1544 * It must be called with the stream lock held.
1545 *
3bd1e081
MD
1546 * Returns the number of bytes spliced.
1547 */
4078b776 1548ssize_t lttng_consumer_on_read_subbuffer_splice(
3bd1e081 1549 struct lttng_consumer_local_data *ctx,
1d4dfdef 1550 struct lttng_consumer_stream *stream, unsigned long len,
309167d2
JD
1551 unsigned long padding,
1552 struct lttng_packet_index *index)
3bd1e081 1553{
f02e1e8a
DG
1554 ssize_t ret = 0, written = 0, ret_splice = 0;
1555 loff_t offset = 0;
1556 off_t orig_offset = stream->out_fd_offset;
1557 int fd = stream->wait_fd;
1558 /* Default is on the disk */
1559 int outfd = stream->out_fd;
f02e1e8a 1560 struct consumer_relayd_sock_pair *relayd = NULL;
fb3a43a9 1561 int *splice_pipe;
8994307f 1562 unsigned int relayd_hang_up = 0;
f02e1e8a 1563
3bd1e081
MD
1564 switch (consumer_data.type) {
1565 case LTTNG_CONSUMER_KERNEL:
f02e1e8a 1566 break;
7753dea8
MD
1567 case LTTNG_CONSUMER32_UST:
1568 case LTTNG_CONSUMER64_UST:
f02e1e8a 1569 /* Not supported for user space tracing */
3bd1e081
MD
1570 return -ENOSYS;
1571 default:
1572 ERR("Unknown consumer_data type");
1573 assert(0);
3bd1e081
MD
1574 }
1575
f02e1e8a
DG
1576 /* RCU lock for the relayd pointer */
1577 rcu_read_lock();
1578
1579 /* Flag that the current stream if set for network streaming. */
da009f2c 1580 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1581 relayd = consumer_find_relayd(stream->net_seq_idx);
1582 if (relayd == NULL) {
56591bac 1583 ret = -EPIPE;
f02e1e8a
DG
1584 goto end;
1585 }
1586 }
1587
fb3a43a9
DG
1588 /*
1589 * Choose right pipe for splice. Metadata and trace data are handled by
1590 * different threads hence the use of two pipes in order not to race or
1591 * corrupt the written data.
1592 */
1593 if (stream->metadata_flag) {
1594 splice_pipe = ctx->consumer_splice_metadata_pipe;
1595 } else {
1596 splice_pipe = ctx->consumer_thread_pipe;
1597 }
1598
f02e1e8a 1599 /* Write metadata stream id before payload */
1d4dfdef
DG
1600 if (relayd) {
1601 int total_len = len;
f02e1e8a 1602
1d4dfdef
DG
1603 if (stream->metadata_flag) {
1604 /*
1605 * Lock the control socket for the complete duration of the function
1606 * since from this point on we will use the socket.
1607 */
1608 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1609
1610 ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd,
1611 padding);
1612 if (ret < 0) {
1613 written = ret;
8994307f
DG
1614 /* Socket operation failed. We consider the relayd dead */
1615 if (ret == -EBADF) {
1616 WARN("Remote relayd disconnected. Stopping");
1617 relayd_hang_up = 1;
1618 goto write_error;
1619 }
1d4dfdef
DG
1620 goto end;
1621 }
1622
1623 total_len += sizeof(struct lttcomm_relayd_metadata_payload);
1624 }
1625
1626 ret = write_relayd_stream_header(stream, total_len, padding, relayd);
1627 if (ret >= 0) {
1628 /* Use the returned socket. */
1629 outfd = ret;
1630 } else {
8994307f
DG
1631 /* Socket operation failed. We consider the relayd dead */
1632 if (ret == -EBADF) {
1633 WARN("Remote relayd disconnected. Stopping");
1634 relayd_hang_up = 1;
1635 goto write_error;
1636 }
f02e1e8a
DG
1637 goto end;
1638 }
1d4dfdef
DG
1639 } else {
1640 /* No streaming, we have to set the len with the full padding */
1641 len += padding;
1624d5b7
JD
1642
1643 /*
1644 * Check if we need to change the tracefile before writing the packet.
1645 */
1646 if (stream->chan->tracefile_size > 0 &&
1647 (stream->tracefile_size_current + len) >
1648 stream->chan->tracefile_size) {
fe4477ee
JD
1649 ret = utils_rotate_stream_file(stream->chan->pathname,
1650 stream->name, stream->chan->tracefile_size,
1651 stream->chan->tracefile_count, stream->uid, stream->gid,
309167d2
JD
1652 stream->out_fd, &(stream->tracefile_count_current),
1653 &stream->out_fd);
1624d5b7
JD
1654 if (ret < 0) {
1655 ERR("Rotating output file");
1656 goto end;
1657 }
309167d2
JD
1658 outfd = stream->out_fd;
1659
1660 if (stream->index_fd >= 0) {
1661 ret = index_create_file(stream->chan->pathname,
1662 stream->name, stream->uid, stream->gid,
1663 stream->chan->tracefile_size,
1664 stream->tracefile_count_current);
1665 if (ret < 0) {
1666 goto end;
1667 }
1668 stream->index_fd = ret;
1669 }
1670
a6976990
DG
1671 /* Reset current size because we just perform a rotation. */
1672 stream->tracefile_size_current = 0;
a1ae300f
JD
1673 stream->out_fd_offset = 0;
1674 orig_offset = 0;
1624d5b7
JD
1675 }
1676 stream->tracefile_size_current += len;
309167d2 1677 index->offset = htobe64(stream->out_fd_offset);
f02e1e8a
DG
1678 }
1679
1680 while (len > 0) {
1d4dfdef
DG
1681 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1682 (unsigned long)offset, len, fd, splice_pipe[1]);
fb3a43a9 1683 ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len,
f02e1e8a
DG
1684 SPLICE_F_MOVE | SPLICE_F_MORE);
1685 DBG("splice chan to pipe, ret %zd", ret_splice);
1686 if (ret_splice < 0) {
1687 PERROR("Error in relay splice");
1688 if (written == 0) {
1689 written = ret_splice;
1690 }
1691 ret = errno;
1692 goto splice_error;
1693 }
1694
1695 /* Handle stream on the relayd if the output is on the network */
1696 if (relayd) {
1697 if (stream->metadata_flag) {
1d4dfdef
DG
1698 size_t metadata_payload_size =
1699 sizeof(struct lttcomm_relayd_metadata_payload);
1700
f02e1e8a 1701 /* Update counter to fit the spliced data */
1d4dfdef
DG
1702 ret_splice += metadata_payload_size;
1703 len += metadata_payload_size;
f02e1e8a
DG
1704 /*
1705 * We do this so the return value can match the len passed as
1706 * argument to this function.
1707 */
1d4dfdef 1708 written -= metadata_payload_size;
f02e1e8a
DG
1709 }
1710 }
1711
1712 /* Splice data out */
fb3a43a9 1713 ret_splice = splice(splice_pipe[0], NULL, outfd, NULL,
f02e1e8a 1714 ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE);
1d4dfdef 1715 DBG("Consumer splice pipe to file, ret %zd", ret_splice);
f02e1e8a
DG
1716 if (ret_splice < 0) {
1717 PERROR("Error in file splice");
1718 if (written == 0) {
1719 written = ret_splice;
1720 }
8994307f 1721 /* Socket operation failed. We consider the relayd dead */
00c8752b 1722 if (errno == EBADF || errno == EPIPE) {
8994307f
DG
1723 WARN("Remote relayd disconnected. Stopping");
1724 relayd_hang_up = 1;
1725 goto write_error;
1726 }
f02e1e8a
DG
1727 ret = errno;
1728 goto splice_error;
1729 } else if (ret_splice > len) {
1730 errno = EINVAL;
1731 PERROR("Wrote more data than requested %zd (len: %lu)",
1732 ret_splice, len);
1733 written += ret_splice;
1734 ret = errno;
1735 goto splice_error;
1736 }
1737 len -= ret_splice;
1738
1739 /* This call is useless on a socket so better save a syscall. */
1740 if (!relayd) {
1741 /* This won't block, but will start writeout asynchronously */
1742 lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice,
1743 SYNC_FILE_RANGE_WRITE);
1744 stream->out_fd_offset += ret_splice;
1745 }
e5d1a9b3 1746 stream->output_written += ret_splice;
f02e1e8a
DG
1747 written += ret_splice;
1748 }
1749 lttng_consumer_sync_trace_file(stream, orig_offset);
1750
1751 ret = ret_splice;
1752
1753 goto end;
1754
8994307f
DG
1755write_error:
1756 /*
1757 * This is a special case that the relayd has closed its socket. Let's
1758 * cleanup the relayd object and all associated streams.
1759 */
1760 if (relayd && relayd_hang_up) {
1761 cleanup_relayd(relayd, ctx);
1762 /* Skip splice error so the consumer does not fail */
1763 goto end;
1764 }
1765
f02e1e8a
DG
1766splice_error:
1767 /* send the appropriate error description to sessiond */
1768 switch (ret) {
f02e1e8a 1769 case EINVAL:
f73fabfd 1770 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL);
f02e1e8a
DG
1771 break;
1772 case ENOMEM:
f73fabfd 1773 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM);
f02e1e8a
DG
1774 break;
1775 case ESPIPE:
f73fabfd 1776 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE);
f02e1e8a
DG
1777 break;
1778 }
1779
1780end:
1781 if (relayd && stream->metadata_flag) {
1782 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1783 }
1784
1785 rcu_read_unlock();
1786 return written;
3bd1e081
MD
1787}
1788
1789/*
1790 * Take a snapshot for a specific fd
1791 *
1792 * Returns 0 on success, < 0 on error
1793 */
ffe60014 1794int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream)
3bd1e081
MD
1795{
1796 switch (consumer_data.type) {
1797 case LTTNG_CONSUMER_KERNEL:
ffe60014 1798 return lttng_kconsumer_take_snapshot(stream);
7753dea8
MD
1799 case LTTNG_CONSUMER32_UST:
1800 case LTTNG_CONSUMER64_UST:
ffe60014 1801 return lttng_ustconsumer_take_snapshot(stream);
3bd1e081
MD
1802 default:
1803 ERR("Unknown consumer_data type");
1804 assert(0);
1805 return -ENOSYS;
1806 }
3bd1e081
MD
1807}
1808
1809/*
1810 * Get the produced position
1811 *
1812 * Returns 0 on success, < 0 on error
1813 */
ffe60014 1814int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
3bd1e081
MD
1815 unsigned long *pos)
1816{
1817 switch (consumer_data.type) {
1818 case LTTNG_CONSUMER_KERNEL:
ffe60014 1819 return lttng_kconsumer_get_produced_snapshot(stream, pos);
7753dea8
MD
1820 case LTTNG_CONSUMER32_UST:
1821 case LTTNG_CONSUMER64_UST:
ffe60014 1822 return lttng_ustconsumer_get_produced_snapshot(stream, pos);
3bd1e081
MD
1823 default:
1824 ERR("Unknown consumer_data type");
1825 assert(0);
1826 return -ENOSYS;
1827 }
1828}
1829
1830int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1831 int sock, struct pollfd *consumer_sockpoll)
1832{
1833 switch (consumer_data.type) {
1834 case LTTNG_CONSUMER_KERNEL:
1835 return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
7753dea8
MD
1836 case LTTNG_CONSUMER32_UST:
1837 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
1838 return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
1839 default:
1840 ERR("Unknown consumer_data type");
1841 assert(0);
1842 return -ENOSYS;
1843 }
1844}
1845
43c34bc3
DG
1846/*
1847 * Iterate over all streams of the hashtable and free them properly.
1848 *
1849 * WARNING: *MUST* be used with data stream only.
1850 */
1851static void destroy_data_stream_ht(struct lttng_ht *ht)
1852{
43c34bc3
DG
1853 struct lttng_ht_iter iter;
1854 struct lttng_consumer_stream *stream;
1855
1856 if (ht == NULL) {
1857 return;
1858 }
1859
1860 rcu_read_lock();
1861 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
5c540210
DG
1862 /*
1863 * Ignore return value since we are currently cleaning up so any error
1864 * can't be handled.
1865 */
1866 (void) consumer_del_stream(stream, ht);
43c34bc3
DG
1867 }
1868 rcu_read_unlock();
1869
1870 lttng_ht_destroy(ht);
1871}
1872
fb3a43a9 1873/*
f724d81e 1874 * Iterate over all streams of the hashtable and free them properly.
e316aad5
DG
1875 *
1876 * XXX: Should not be only for metadata stream or else use an other name.
fb3a43a9
DG
1877 */
1878static void destroy_stream_ht(struct lttng_ht *ht)
1879{
fb3a43a9
DG
1880 struct lttng_ht_iter iter;
1881 struct lttng_consumer_stream *stream;
1882
1883 if (ht == NULL) {
1884 return;
1885 }
1886
d09e1200 1887 rcu_read_lock();
58b1f425 1888 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
5c540210
DG
1889 /*
1890 * Ignore return value since we are currently cleaning up so any error
1891 * can't be handled.
1892 */
1893 (void) consumer_del_metadata_stream(stream, ht);
fb3a43a9 1894 }
d09e1200 1895 rcu_read_unlock();
fb3a43a9
DG
1896
1897 lttng_ht_destroy(ht);
1898}
1899
d88aee68
DG
1900void lttng_consumer_close_metadata(void)
1901{
1902 switch (consumer_data.type) {
1903 case LTTNG_CONSUMER_KERNEL:
1904 /*
1905 * The Kernel consumer has a different metadata scheme so we don't
1906 * close anything because the stream will be closed by the session
1907 * daemon.
1908 */
1909 break;
1910 case LTTNG_CONSUMER32_UST:
1911 case LTTNG_CONSUMER64_UST:
1912 /*
1913 * Close all metadata streams. The metadata hash table is passed and
1914 * this call iterates over it by closing all wakeup fd. This is safe
1915 * because at this point we are sure that the metadata producer is
1916 * either dead or blocked.
1917 */
1918 lttng_ustconsumer_close_metadata(metadata_ht);
1919 break;
1920 default:
1921 ERR("Unknown consumer_data type");
1922 assert(0);
1923 }
1924}
1925
fb3a43a9
DG
1926/*
1927 * Clean up a metadata stream and free its memory.
1928 */
e316aad5
DG
1929void consumer_del_metadata_stream(struct lttng_consumer_stream *stream,
1930 struct lttng_ht *ht)
fb3a43a9
DG
1931{
1932 int ret;
e316aad5
DG
1933 struct lttng_ht_iter iter;
1934 struct lttng_consumer_channel *free_chan = NULL;
fb3a43a9
DG
1935 struct consumer_relayd_sock_pair *relayd;
1936
1937 assert(stream);
1938 /*
1939 * This call should NEVER receive regular stream. It must always be
1940 * metadata stream and this is crucial for data structure synchronization.
1941 */
1942 assert(stream->metadata_flag);
1943
e316aad5
DG
1944 DBG3("Consumer delete metadata stream %d", stream->wait_fd);
1945
1946 if (ht == NULL) {
1947 /* Means the stream was allocated but not successfully added */
ffe60014 1948 goto free_stream_rcu;
e316aad5
DG
1949 }
1950
74251bb8 1951 pthread_mutex_lock(&consumer_data.lock);
a9838785 1952 pthread_mutex_lock(&stream->chan->lock);
8994307f
DG
1953 pthread_mutex_lock(&stream->lock);
1954
fb3a43a9
DG
1955 switch (consumer_data.type) {
1956 case LTTNG_CONSUMER_KERNEL:
1957 if (stream->mmap_base != NULL) {
1958 ret = munmap(stream->mmap_base, stream->mmap_len);
1959 if (ret != 0) {
1960 PERROR("munmap metadata stream");
1961 }
1962 }
4c95e622
JD
1963 if (stream->wait_fd >= 0) {
1964 ret = close(stream->wait_fd);
1965 if (ret < 0) {
1966 PERROR("close kernel metadata wait_fd");
1967 }
1968 }
fb3a43a9
DG
1969 break;
1970 case LTTNG_CONSUMER32_UST:
1971 case LTTNG_CONSUMER64_UST:
04ef1097
MD
1972 if (stream->monitor) {
1973 /* close the write-side in close_metadata */
1974 ret = close(stream->ust_metadata_poll_pipe[0]);
1975 if (ret < 0) {
1976 PERROR("Close UST metadata read-side poll pipe");
1977 }
1978 }
fb3a43a9
DG
1979 lttng_ustconsumer_del_stream(stream);
1980 break;
1981 default:
1982 ERR("Unknown consumer_data type");
1983 assert(0);
e316aad5 1984 goto end;
fb3a43a9 1985 }
fb3a43a9 1986
c869f647 1987 rcu_read_lock();
58b1f425 1988 iter.iter.node = &stream->node.node;
c869f647
DG
1989 ret = lttng_ht_del(ht, &iter);
1990 assert(!ret);
ca22feea 1991
d8ef542d
MD
1992 iter.iter.node = &stream->node_channel_id.node;
1993 ret = lttng_ht_del(consumer_data.stream_per_chan_id_ht, &iter);
1994 assert(!ret);
1995
ca22feea
DG
1996 iter.iter.node = &stream->node_session_id.node;
1997 ret = lttng_ht_del(consumer_data.stream_list_ht, &iter);
1998 assert(!ret);
c869f647
DG
1999 rcu_read_unlock();
2000
fb3a43a9
DG
2001 if (stream->out_fd >= 0) {
2002 ret = close(stream->out_fd);
2003 if (ret) {
2004 PERROR("close");
2005 }
2006 }
2007
fb3a43a9
DG
2008 /* Check and cleanup relayd */
2009 rcu_read_lock();
2010 relayd = consumer_find_relayd(stream->net_seq_idx);
2011 if (relayd != NULL) {
2012 uatomic_dec(&relayd->refcount);
2013 assert(uatomic_read(&relayd->refcount) >= 0);
2014
2015 /* Closing streams requires to lock the control socket. */
2016 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
2017 ret = relayd_send_close_stream(&relayd->control_sock,
2018 stream->relayd_stream_id, stream->next_net_seq_num - 1);
2019 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
2020 if (ret < 0) {
2021 DBG("Unable to close stream on the relayd. Continuing");
2022 /*
2023 * Continue here. There is nothing we can do for the relayd.
2024 * Chances are that the relayd has closed the socket so we just
2025 * continue cleaning up.
2026 */
2027 }
2028
2029 /* Both conditions are met, we destroy the relayd. */
2030 if (uatomic_read(&relayd->refcount) == 0 &&
2031 uatomic_read(&relayd->destroy_flag)) {
51230d70 2032 consumer_destroy_relayd(relayd);
fb3a43a9
DG
2033 }
2034 }
2035 rcu_read_unlock();
2036
2037 /* Atomically decrement channel refcount since other threads can use it. */
f2ad556d 2038 if (!uatomic_sub_return(&stream->chan->refcount, 1)
ffe60014 2039 && !uatomic_read(&stream->chan->nb_init_stream_left)) {
c30aaa51 2040 /* Go for channel deletion! */
e316aad5 2041 free_chan = stream->chan;
fb3a43a9
DG
2042 }
2043
e316aad5 2044end:
73811ecc
DG
2045 /*
2046 * Nullify the stream reference so it is not used after deletion. The
5e41ebe1
MD
2047 * channel lock MUST be acquired before being able to check for
2048 * a NULL pointer value.
73811ecc
DG
2049 */
2050 stream->chan->metadata_stream = NULL;
2051
8994307f 2052 pthread_mutex_unlock(&stream->lock);
a9838785 2053 pthread_mutex_unlock(&stream->chan->lock);
74251bb8 2054 pthread_mutex_unlock(&consumer_data.lock);
e316aad5
DG
2055
2056 if (free_chan) {
2057 consumer_del_channel(free_chan);
2058 }
2059
ffe60014
DG
2060free_stream_rcu:
2061 call_rcu(&stream->node.head, free_stream_rcu);
fb3a43a9
DG
2062}
2063
2064/*
2065 * Action done with the metadata stream when adding it to the consumer internal
2066 * data structures to handle it.
2067 */
5ab66908 2068int consumer_add_metadata_stream(struct lttng_consumer_stream *stream)
fb3a43a9 2069{
5ab66908 2070 struct lttng_ht *ht = metadata_ht;
e316aad5 2071 int ret = 0;
76082088 2072 struct lttng_ht_iter iter;
d88aee68 2073 struct lttng_ht_node_u64 *node;
fb3a43a9 2074
e316aad5
DG
2075 assert(stream);
2076 assert(ht);
2077
d88aee68 2078 DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key);
e316aad5
DG
2079
2080 pthread_mutex_lock(&consumer_data.lock);
a9838785 2081 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 2082 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 2083 pthread_mutex_lock(&stream->lock);
e316aad5 2084
e316aad5
DG
2085 /*
2086 * From here, refcounts are updated so be _careful_ when returning an error
2087 * after this point.
2088 */
2089
fb3a43a9 2090 rcu_read_lock();
76082088
DG
2091
2092 /*
2093 * Lookup the stream just to make sure it does not exist in our internal
2094 * state. This should NEVER happen.
2095 */
d88aee68
DG
2096 lttng_ht_lookup(ht, &stream->key, &iter);
2097 node = lttng_ht_iter_get_node_u64(&iter);
76082088
DG
2098 assert(!node);
2099
e316aad5 2100 /*
ffe60014
DG
2101 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2102 * in terms of destroying the associated channel, because the action that
e316aad5
DG
2103 * causes the count to become 0 also causes a stream to be added. The
2104 * channel deletion will thus be triggered by the following removal of this
2105 * stream.
2106 */
ffe60014 2107 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
2108 /* Increment refcount before decrementing nb_init_stream_left */
2109 cmm_smp_wmb();
ffe60014 2110 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
2111 }
2112
d88aee68 2113 lttng_ht_add_unique_u64(ht, &stream->node);
ca22feea 2114
d8ef542d
MD
2115 lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht,
2116 &stream->node_channel_id);
2117
ca22feea
DG
2118 /*
2119 * Add stream to the stream_list_ht of the consumer data. No need to steal
2120 * the key since the HT does not use it and we allow to add redundant keys
2121 * into this table.
2122 */
d88aee68 2123 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 2124
fb3a43a9 2125 rcu_read_unlock();
e316aad5 2126
2e818a6a 2127 pthread_mutex_unlock(&stream->lock);
a9838785 2128 pthread_mutex_unlock(&stream->chan->lock);
ec6ea7d0 2129 pthread_mutex_unlock(&stream->chan->timer_lock);
e316aad5
DG
2130 pthread_mutex_unlock(&consumer_data.lock);
2131 return ret;
fb3a43a9
DG
2132}
2133
8994307f
DG
2134/*
2135 * Delete data stream that are flagged for deletion (endpoint_status).
2136 */
2137static void validate_endpoint_status_data_stream(void)
2138{
2139 struct lttng_ht_iter iter;
2140 struct lttng_consumer_stream *stream;
2141
2142 DBG("Consumer delete flagged data stream");
2143
2144 rcu_read_lock();
2145 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
2146 /* Validate delete flag of the stream */
79d4ffb7 2147 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2148 continue;
2149 }
2150 /* Delete it right now */
2151 consumer_del_stream(stream, data_ht);
2152 }
2153 rcu_read_unlock();
2154}
2155
2156/*
2157 * Delete metadata stream that are flagged for deletion (endpoint_status).
2158 */
2159static void validate_endpoint_status_metadata_stream(
2160 struct lttng_poll_event *pollset)
2161{
2162 struct lttng_ht_iter iter;
2163 struct lttng_consumer_stream *stream;
2164
2165 DBG("Consumer delete flagged metadata stream");
2166
2167 assert(pollset);
2168
2169 rcu_read_lock();
2170 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
2171 /* Validate delete flag of the stream */
79d4ffb7 2172 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2173 continue;
2174 }
2175 /*
2176 * Remove from pollset so the metadata thread can continue without
2177 * blocking on a deleted stream.
2178 */
2179 lttng_poll_del(pollset, stream->wait_fd);
2180
2181 /* Delete it right now */
2182 consumer_del_metadata_stream(stream, metadata_ht);
2183 }
2184 rcu_read_unlock();
2185}
2186
fb3a43a9
DG
2187/*
2188 * Thread polls on metadata file descriptor and write them on disk or on the
2189 * network.
2190 */
7d980def 2191void *consumer_thread_metadata_poll(void *data)
fb3a43a9 2192{
1fc79fb4 2193 int ret, i, pollfd, err = -1;
fb3a43a9 2194 uint32_t revents, nb_fd;
e316aad5 2195 struct lttng_consumer_stream *stream = NULL;
fb3a43a9 2196 struct lttng_ht_iter iter;
d88aee68 2197 struct lttng_ht_node_u64 *node;
fb3a43a9
DG
2198 struct lttng_poll_event events;
2199 struct lttng_consumer_local_data *ctx = data;
2200 ssize_t len;
2201
2202 rcu_register_thread();
2203
1fc79fb4
MD
2204 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_METADATA);
2205
9ce5646a
MD
2206 health_code_update();
2207
d88aee68 2208 metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
04bb2b64
DG
2209 if (!metadata_ht) {
2210 /* ENOMEM at this point. Better to bail out. */
d8ef542d 2211 goto end_ht;
04bb2b64
DG
2212 }
2213
fb3a43a9
DG
2214 DBG("Thread metadata poll started");
2215
fb3a43a9
DG
2216 /* Size is set to 1 for the consumer_metadata pipe */
2217 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2218 if (ret < 0) {
2219 ERR("Poll set creation failed");
d8ef542d 2220 goto end_poll;
fb3a43a9
DG
2221 }
2222
13886d2d
DG
2223 ret = lttng_poll_add(&events,
2224 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN);
fb3a43a9
DG
2225 if (ret < 0) {
2226 goto end;
2227 }
2228
2229 /* Main loop */
2230 DBG("Metadata main loop started");
2231
2232 while (1) {
9ce5646a
MD
2233 health_code_update();
2234
fb3a43a9 2235 /* Only the metadata pipe is set */
d21b0d71 2236 if (LTTNG_POLL_GETNB(&events) == 0 && consumer_quit == 1) {
1fc79fb4 2237 err = 0; /* All is OK */
fb3a43a9
DG
2238 goto end;
2239 }
2240
2241restart:
d21b0d71 2242 DBG("Metadata poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events));
9ce5646a 2243 health_poll_entry();
fb3a43a9 2244 ret = lttng_poll_wait(&events, -1);
9ce5646a 2245 health_poll_exit();
fb3a43a9
DG
2246 DBG("Metadata event catched in thread");
2247 if (ret < 0) {
2248 if (errno == EINTR) {
e316aad5 2249 ERR("Poll EINTR catched");
fb3a43a9
DG
2250 goto restart;
2251 }
2252 goto error;
2253 }
2254
0d9c5d77
DG
2255 nb_fd = ret;
2256
e316aad5 2257 /* From here, the event is a metadata wait fd */
fb3a43a9 2258 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2259 health_code_update();
2260
fb3a43a9
DG
2261 revents = LTTNG_POLL_GETEV(&events, i);
2262 pollfd = LTTNG_POLL_GETFD(&events, i);
2263
13886d2d 2264 if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) {
4adabd61 2265 if (revents & (LPOLLERR | LPOLLHUP )) {
fb3a43a9
DG
2266 DBG("Metadata thread pipe hung up");
2267 /*
2268 * Remove the pipe from the poll set and continue the loop
2269 * since their might be data to consume.
2270 */
13886d2d
DG
2271 lttng_poll_del(&events,
2272 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2273 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
fb3a43a9
DG
2274 continue;
2275 } else if (revents & LPOLLIN) {
13886d2d
DG
2276 ssize_t pipe_len;
2277
2278 pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe,
2279 &stream, sizeof(stream));
6cd525e8
MD
2280 if (pipe_len < sizeof(stream)) {
2281 PERROR("read metadata stream");
fb3a43a9 2282 /*
13886d2d 2283 * Continue here to handle the rest of the streams.
fb3a43a9
DG
2284 */
2285 continue;
2286 }
2287
8994307f
DG
2288 /* A NULL stream means that the state has changed. */
2289 if (stream == NULL) {
2290 /* Check for deleted streams. */
2291 validate_endpoint_status_metadata_stream(&events);
3714380f 2292 goto restart;
8994307f
DG
2293 }
2294
fb3a43a9
DG
2295 DBG("Adding metadata stream %d to poll set",
2296 stream->wait_fd);
2297
fb3a43a9
DG
2298 /* Add metadata stream to the global poll events list */
2299 lttng_poll_add(&events, stream->wait_fd,
2300 LPOLLIN | LPOLLPRI);
fb3a43a9
DG
2301 }
2302
e316aad5 2303 /* Handle other stream */
fb3a43a9
DG
2304 continue;
2305 }
2306
d09e1200 2307 rcu_read_lock();
d88aee68
DG
2308 {
2309 uint64_t tmp_id = (uint64_t) pollfd;
2310
2311 lttng_ht_lookup(metadata_ht, &tmp_id, &iter);
2312 }
2313 node = lttng_ht_iter_get_node_u64(&iter);
e316aad5 2314 assert(node);
fb3a43a9
DG
2315
2316 stream = caa_container_of(node, struct lttng_consumer_stream,
58b1f425 2317 node);
fb3a43a9 2318
e316aad5 2319 /* Check for error event */
4adabd61 2320 if (revents & (LPOLLERR | LPOLLHUP)) {
e316aad5 2321 DBG("Metadata fd %d is hup|err.", pollfd);
fb3a43a9
DG
2322 if (!stream->hangup_flush_done
2323 && (consumer_data.type == LTTNG_CONSUMER32_UST
2324 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2325 DBG("Attempting to flush and consume the UST buffers");
2326 lttng_ustconsumer_on_stream_hangup(stream);
2327
2328 /* We just flushed the stream now read it. */
4bb94b75 2329 do {
9ce5646a
MD
2330 health_code_update();
2331
4bb94b75
DG
2332 len = ctx->on_buffer_ready(stream, ctx);
2333 /*
2334 * We don't check the return value here since if we get
2335 * a negative len, it means an error occured thus we
2336 * simply remove it from the poll set and free the
2337 * stream.
2338 */
2339 } while (len > 0);
fb3a43a9
DG
2340 }
2341
fb3a43a9 2342 lttng_poll_del(&events, stream->wait_fd);
e316aad5
DG
2343 /*
2344 * This call update the channel states, closes file descriptors
2345 * and securely free the stream.
2346 */
2347 consumer_del_metadata_stream(stream, metadata_ht);
2348 } else if (revents & (LPOLLIN | LPOLLPRI)) {
2349 /* Get the data out of the metadata file descriptor */
2350 DBG("Metadata available on fd %d", pollfd);
2351 assert(stream->wait_fd == pollfd);
2352
04ef1097 2353 do {
9ce5646a
MD
2354 health_code_update();
2355
04ef1097
MD
2356 len = ctx->on_buffer_ready(stream, ctx);
2357 /*
2358 * We don't check the return value here since if we get
2359 * a negative len, it means an error occured thus we
2360 * simply remove it from the poll set and free the
2361 * stream.
2362 */
2363 } while (len > 0);
2364
e316aad5 2365 /* It's ok to have an unavailable sub-buffer */
b64403e3 2366 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2367 /* Clean up stream from consumer and free it. */
2368 lttng_poll_del(&events, stream->wait_fd);
2369 consumer_del_metadata_stream(stream, metadata_ht);
e316aad5 2370 }
fb3a43a9 2371 }
e316aad5
DG
2372
2373 /* Release RCU lock for the stream looked up */
d09e1200 2374 rcu_read_unlock();
fb3a43a9
DG
2375 }
2376 }
2377
1fc79fb4
MD
2378 /* All is OK */
2379 err = 0;
fb3a43a9
DG
2380error:
2381end:
2382 DBG("Metadata poll thread exiting");
fb3a43a9 2383
d8ef542d
MD
2384 lttng_poll_clean(&events);
2385end_poll:
04bb2b64 2386 destroy_stream_ht(metadata_ht);
d8ef542d 2387end_ht:
1fc79fb4
MD
2388 if (err) {
2389 health_error();
2390 ERR("Health error occurred in %s", __func__);
2391 }
2392 health_unregister(health_consumerd);
fb3a43a9
DG
2393 rcu_unregister_thread();
2394 return NULL;
2395}
2396
3bd1e081 2397/*
e4421fec 2398 * This thread polls the fds in the set to consume the data and write
3bd1e081
MD
2399 * it to tracefile if necessary.
2400 */
7d980def 2401void *consumer_thread_data_poll(void *data)
3bd1e081 2402{
1fc79fb4 2403 int num_rdy, num_hup, high_prio, ret, i, err = -1;
3bd1e081
MD
2404 struct pollfd *pollfd = NULL;
2405 /* local view of the streams */
c869f647 2406 struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL;
3bd1e081
MD
2407 /* local view of consumer_data.fds_count */
2408 int nb_fd = 0;
3bd1e081 2409 struct lttng_consumer_local_data *ctx = data;
00e2e675 2410 ssize_t len;
3bd1e081 2411
e7b994a3
DG
2412 rcu_register_thread();
2413
1fc79fb4
MD
2414 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_DATA);
2415
9ce5646a
MD
2416 health_code_update();
2417
d88aee68 2418 data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
43c34bc3 2419 if (data_ht == NULL) {
04bb2b64 2420 /* ENOMEM at this point. Better to bail out. */
43c34bc3
DG
2421 goto end;
2422 }
2423
4df6c8cb
MD
2424 local_stream = zmalloc(sizeof(struct lttng_consumer_stream *));
2425 if (local_stream == NULL) {
2426 PERROR("local_stream malloc");
2427 goto end;
2428 }
3bd1e081
MD
2429
2430 while (1) {
9ce5646a
MD
2431 health_code_update();
2432
3bd1e081
MD
2433 high_prio = 0;
2434 num_hup = 0;
2435
2436 /*
e4421fec 2437 * the fds set has been updated, we need to update our
3bd1e081
MD
2438 * local array as well
2439 */
2440 pthread_mutex_lock(&consumer_data.lock);
2441 if (consumer_data.need_update) {
0e428499
DG
2442 free(pollfd);
2443 pollfd = NULL;
2444
2445 free(local_stream);
2446 local_stream = NULL;
3bd1e081 2447
50f8ae69 2448 /* allocate for all fds + 1 for the consumer_data_pipe */
effcf122 2449 pollfd = zmalloc((consumer_data.stream_count + 1) * sizeof(struct pollfd));
3bd1e081 2450 if (pollfd == NULL) {
7a57cf92 2451 PERROR("pollfd malloc");
3bd1e081
MD
2452 pthread_mutex_unlock(&consumer_data.lock);
2453 goto end;
2454 }
2455
50f8ae69 2456 /* allocate for all fds + 1 for the consumer_data_pipe */
effcf122 2457 local_stream = zmalloc((consumer_data.stream_count + 1) *
747f8642 2458 sizeof(struct lttng_consumer_stream *));
3bd1e081 2459 if (local_stream == NULL) {
7a57cf92 2460 PERROR("local_stream malloc");
3bd1e081
MD
2461 pthread_mutex_unlock(&consumer_data.lock);
2462 goto end;
2463 }
ffe60014 2464 ret = update_poll_array(ctx, &pollfd, local_stream,
43c34bc3 2465 data_ht);
3bd1e081
MD
2466 if (ret < 0) {
2467 ERR("Error in allocating pollfd or local_outfds");
f73fabfd 2468 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2469 pthread_mutex_unlock(&consumer_data.lock);
2470 goto end;
2471 }
2472 nb_fd = ret;
2473 consumer_data.need_update = 0;
2474 }
2475 pthread_mutex_unlock(&consumer_data.lock);
2476
4078b776
MD
2477 /* No FDs and consumer_quit, consumer_cleanup the thread */
2478 if (nb_fd == 0 && consumer_quit == 1) {
1fc79fb4 2479 err = 0; /* All is OK */
4078b776
MD
2480 goto end;
2481 }
3bd1e081 2482 /* poll on the array of fds */
88f2b785 2483 restart:
3bd1e081 2484 DBG("polling on %d fd", nb_fd + 1);
9ce5646a 2485 health_poll_entry();
cb365c03 2486 num_rdy = poll(pollfd, nb_fd + 1, -1);
9ce5646a 2487 health_poll_exit();
3bd1e081
MD
2488 DBG("poll num_rdy : %d", num_rdy);
2489 if (num_rdy == -1) {
88f2b785
MD
2490 /*
2491 * Restart interrupted system call.
2492 */
2493 if (errno == EINTR) {
2494 goto restart;
2495 }
7a57cf92 2496 PERROR("Poll error");
f73fabfd 2497 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2498 goto end;
2499 } else if (num_rdy == 0) {
2500 DBG("Polling thread timed out");
2501 goto end;
2502 }
2503
3bd1e081 2504 /*
50f8ae69 2505 * If the consumer_data_pipe triggered poll go directly to the
00e2e675
DG
2506 * beginning of the loop to update the array. We want to prioritize
2507 * array update over low-priority reads.
3bd1e081 2508 */
509bb1cf 2509 if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) {
ab30f567 2510 ssize_t pipe_readlen;
04fdd819 2511
50f8ae69 2512 DBG("consumer_data_pipe wake up");
acdb9057
DG
2513 pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe,
2514 &new_stream, sizeof(new_stream));
6cd525e8
MD
2515 if (pipe_readlen < sizeof(new_stream)) {
2516 PERROR("Consumer data pipe");
23f5f35d
DG
2517 /* Continue so we can at least handle the current stream(s). */
2518 continue;
2519 }
c869f647
DG
2520
2521 /*
2522 * If the stream is NULL, just ignore it. It's also possible that
2523 * the sessiond poll thread changed the consumer_quit state and is
2524 * waking us up to test it.
2525 */
2526 if (new_stream == NULL) {
8994307f 2527 validate_endpoint_status_data_stream();
c869f647
DG
2528 continue;
2529 }
2530
c869f647 2531 /* Continue to update the local streams and handle prio ones */
3bd1e081
MD
2532 continue;
2533 }
2534
2535 /* Take care of high priority channels first. */
2536 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2537 health_code_update();
2538
9617607b
DG
2539 if (local_stream[i] == NULL) {
2540 continue;
2541 }
fb3a43a9 2542 if (pollfd[i].revents & POLLPRI) {
d41f73b7
MD
2543 DBG("Urgent read on fd %d", pollfd[i].fd);
2544 high_prio = 1;
4078b776 2545 len = ctx->on_buffer_ready(local_stream[i], ctx);
d41f73b7 2546 /* it's ok to have an unavailable sub-buffer */
b64403e3 2547 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2548 /* Clean the stream and free it. */
2549 consumer_del_stream(local_stream[i], data_ht);
9617607b 2550 local_stream[i] = NULL;
4078b776
MD
2551 } else if (len > 0) {
2552 local_stream[i]->data_read = 1;
d41f73b7 2553 }
3bd1e081
MD
2554 }
2555 }
2556
4078b776
MD
2557 /*
2558 * If we read high prio channel in this loop, try again
2559 * for more high prio data.
2560 */
2561 if (high_prio) {
3bd1e081
MD
2562 continue;
2563 }
2564
2565 /* Take care of low priority channels. */
4078b776 2566 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2567 health_code_update();
2568
9617607b
DG
2569 if (local_stream[i] == NULL) {
2570 continue;
2571 }
4078b776
MD
2572 if ((pollfd[i].revents & POLLIN) ||
2573 local_stream[i]->hangup_flush_done) {
4078b776
MD
2574 DBG("Normal read on fd %d", pollfd[i].fd);
2575 len = ctx->on_buffer_ready(local_stream[i], ctx);
2576 /* it's ok to have an unavailable sub-buffer */
b64403e3 2577 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2578 /* Clean the stream and free it. */
2579 consumer_del_stream(local_stream[i], data_ht);
9617607b 2580 local_stream[i] = NULL;
4078b776
MD
2581 } else if (len > 0) {
2582 local_stream[i]->data_read = 1;
2583 }
2584 }
2585 }
2586
2587 /* Handle hangup and errors */
2588 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2589 health_code_update();
2590
9617607b
DG
2591 if (local_stream[i] == NULL) {
2592 continue;
2593 }
4078b776
MD
2594 if (!local_stream[i]->hangup_flush_done
2595 && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL))
2596 && (consumer_data.type == LTTNG_CONSUMER32_UST
2597 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2598 DBG("fd %d is hup|err|nval. Attempting flush and read.",
9617607b 2599 pollfd[i].fd);
4078b776
MD
2600 lttng_ustconsumer_on_stream_hangup(local_stream[i]);
2601 /* Attempt read again, for the data we just flushed. */
2602 local_stream[i]->data_read = 1;
2603 }
2604 /*
2605 * If the poll flag is HUP/ERR/NVAL and we have
2606 * read no data in this pass, we can remove the
2607 * stream from its hash table.
2608 */
2609 if ((pollfd[i].revents & POLLHUP)) {
2610 DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
2611 if (!local_stream[i]->data_read) {
43c34bc3 2612 consumer_del_stream(local_stream[i], data_ht);
9617607b 2613 local_stream[i] = NULL;
4078b776
MD
2614 num_hup++;
2615 }
2616 } else if (pollfd[i].revents & POLLERR) {
2617 ERR("Error returned in polling fd %d.", pollfd[i].fd);
2618 if (!local_stream[i]->data_read) {
43c34bc3 2619 consumer_del_stream(local_stream[i], data_ht);
9617607b 2620 local_stream[i] = NULL;
4078b776
MD
2621 num_hup++;
2622 }
2623 } else if (pollfd[i].revents & POLLNVAL) {
2624 ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
2625 if (!local_stream[i]->data_read) {
43c34bc3 2626 consumer_del_stream(local_stream[i], data_ht);
9617607b 2627 local_stream[i] = NULL;
4078b776 2628 num_hup++;
3bd1e081
MD
2629 }
2630 }
9617607b
DG
2631 if (local_stream[i] != NULL) {
2632 local_stream[i]->data_read = 0;
2633 }
3bd1e081
MD
2634 }
2635 }
1fc79fb4
MD
2636 /* All is OK */
2637 err = 0;
3bd1e081
MD
2638end:
2639 DBG("polling thread exiting");
0e428499
DG
2640 free(pollfd);
2641 free(local_stream);
fb3a43a9
DG
2642
2643 /*
2644 * Close the write side of the pipe so epoll_wait() in
7d980def
DG
2645 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2646 * read side of the pipe. If we close them both, epoll_wait strangely does
2647 * not return and could create a endless wait period if the pipe is the
2648 * only tracked fd in the poll set. The thread will take care of closing
2649 * the read side.
fb3a43a9 2650 */
13886d2d 2651 (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe);
fb3a43a9 2652
04bb2b64 2653 destroy_data_stream_ht(data_ht);
43c34bc3 2654
1fc79fb4
MD
2655 if (err) {
2656 health_error();
2657 ERR("Health error occurred in %s", __func__);
2658 }
2659 health_unregister(health_consumerd);
2660
e7b994a3 2661 rcu_unregister_thread();
3bd1e081
MD
2662 return NULL;
2663}
2664
d8ef542d
MD
2665/*
2666 * Close wake-up end of each stream belonging to the channel. This will
2667 * allow the poll() on the stream read-side to detect when the
2668 * write-side (application) finally closes them.
2669 */
2670static
2671void consumer_close_channel_streams(struct lttng_consumer_channel *channel)
2672{
2673 struct lttng_ht *ht;
2674 struct lttng_consumer_stream *stream;
2675 struct lttng_ht_iter iter;
2676
2677 ht = consumer_data.stream_per_chan_id_ht;
2678
2679 rcu_read_lock();
2680 cds_lfht_for_each_entry_duplicate(ht->ht,
2681 ht->hash_fct(&channel->key, lttng_ht_seed),
2682 ht->match_fct, &channel->key,
2683 &iter.iter, stream, node_channel_id.node) {
f2ad556d
MD
2684 /*
2685 * Protect against teardown with mutex.
2686 */
2687 pthread_mutex_lock(&stream->lock);
2688 if (cds_lfht_is_node_deleted(&stream->node.node)) {
2689 goto next;
2690 }
d8ef542d
MD
2691 switch (consumer_data.type) {
2692 case LTTNG_CONSUMER_KERNEL:
2693 break;
2694 case LTTNG_CONSUMER32_UST:
2695 case LTTNG_CONSUMER64_UST:
2696 /*
2697 * Note: a mutex is taken internally within
2698 * liblttng-ust-ctl to protect timer wakeup_fd
2699 * use from concurrent close.
2700 */
2701 lttng_ustconsumer_close_stream_wakeup(stream);
2702 break;
2703 default:
2704 ERR("Unknown consumer_data type");
2705 assert(0);
2706 }
f2ad556d
MD
2707 next:
2708 pthread_mutex_unlock(&stream->lock);
d8ef542d
MD
2709 }
2710 rcu_read_unlock();
2711}
2712
2713static void destroy_channel_ht(struct lttng_ht *ht)
2714{
2715 struct lttng_ht_iter iter;
2716 struct lttng_consumer_channel *channel;
2717 int ret;
2718
2719 if (ht == NULL) {
2720 return;
2721 }
2722
2723 rcu_read_lock();
2724 cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) {
2725 ret = lttng_ht_del(ht, &iter);
2726 assert(ret != 0);
2727 }
2728 rcu_read_unlock();
2729
2730 lttng_ht_destroy(ht);
2731}
2732
2733/*
2734 * This thread polls the channel fds to detect when they are being
2735 * closed. It closes all related streams if the channel is detected as
2736 * closed. It is currently only used as a shim layer for UST because the
2737 * consumerd needs to keep the per-stream wakeup end of pipes open for
2738 * periodical flush.
2739 */
2740void *consumer_thread_channel_poll(void *data)
2741{
1fc79fb4 2742 int ret, i, pollfd, err = -1;
d8ef542d
MD
2743 uint32_t revents, nb_fd;
2744 struct lttng_consumer_channel *chan = NULL;
2745 struct lttng_ht_iter iter;
2746 struct lttng_ht_node_u64 *node;
2747 struct lttng_poll_event events;
2748 struct lttng_consumer_local_data *ctx = data;
2749 struct lttng_ht *channel_ht;
2750
2751 rcu_register_thread();
2752
1fc79fb4
MD
2753 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_CHANNEL);
2754
9ce5646a
MD
2755 health_code_update();
2756
d8ef542d
MD
2757 channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2758 if (!channel_ht) {
2759 /* ENOMEM at this point. Better to bail out. */
2760 goto end_ht;
2761 }
2762
2763 DBG("Thread channel poll started");
2764
2765 /* Size is set to 1 for the consumer_channel pipe */
2766 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2767 if (ret < 0) {
2768 ERR("Poll set creation failed");
2769 goto end_poll;
2770 }
2771
2772 ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN);
2773 if (ret < 0) {
2774 goto end;
2775 }
2776
2777 /* Main loop */
2778 DBG("Channel main loop started");
2779
2780 while (1) {
9ce5646a
MD
2781 health_code_update();
2782
d8ef542d
MD
2783 /* Only the channel pipe is set */
2784 if (LTTNG_POLL_GETNB(&events) == 0 && consumer_quit == 1) {
1fc79fb4 2785 err = 0; /* All is OK */
d8ef542d
MD
2786 goto end;
2787 }
2788
2789restart:
2790 DBG("Channel poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events));
9ce5646a 2791 health_poll_entry();
d8ef542d 2792 ret = lttng_poll_wait(&events, -1);
9ce5646a 2793 health_poll_exit();
d8ef542d
MD
2794 DBG("Channel event catched in thread");
2795 if (ret < 0) {
2796 if (errno == EINTR) {
2797 ERR("Poll EINTR catched");
2798 goto restart;
2799 }
2800 goto end;
2801 }
2802
2803 nb_fd = ret;
2804
2805 /* From here, the event is a channel wait fd */
2806 for (i = 0; i < nb_fd; i++) {
9ce5646a
MD
2807 health_code_update();
2808
d8ef542d
MD
2809 revents = LTTNG_POLL_GETEV(&events, i);
2810 pollfd = LTTNG_POLL_GETFD(&events, i);
2811
2812 /* Just don't waste time if no returned events for the fd */
2813 if (!revents) {
2814 continue;
2815 }
2816 if (pollfd == ctx->consumer_channel_pipe[0]) {
2817 if (revents & (LPOLLERR | LPOLLHUP)) {
2818 DBG("Channel thread pipe hung up");
2819 /*
2820 * Remove the pipe from the poll set and continue the loop
2821 * since their might be data to consume.
2822 */
2823 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2824 continue;
2825 } else if (revents & LPOLLIN) {
2826 enum consumer_channel_action action;
a0cbdd2e 2827 uint64_t key;
d8ef542d 2828
a0cbdd2e 2829 ret = read_channel_pipe(ctx, &chan, &key, &action);
d8ef542d
MD
2830 if (ret <= 0) {
2831 ERR("Error reading channel pipe");
2832 continue;
2833 }
2834
2835 switch (action) {
2836 case CONSUMER_CHANNEL_ADD:
2837 DBG("Adding channel %d to poll set",
2838 chan->wait_fd);
2839
2840 lttng_ht_node_init_u64(&chan->wait_fd_node,
2841 chan->wait_fd);
c7260a81 2842 rcu_read_lock();
d8ef542d
MD
2843 lttng_ht_add_unique_u64(channel_ht,
2844 &chan->wait_fd_node);
c7260a81 2845 rcu_read_unlock();
d8ef542d
MD
2846 /* Add channel to the global poll events list */
2847 lttng_poll_add(&events, chan->wait_fd,
2848 LPOLLIN | LPOLLPRI);
2849 break;
a0cbdd2e
MD
2850 case CONSUMER_CHANNEL_DEL:
2851 {
f2a444f1
DG
2852 struct lttng_consumer_stream *stream, *stmp;
2853
c7260a81 2854 rcu_read_lock();
a0cbdd2e
MD
2855 chan = consumer_find_channel(key);
2856 if (!chan) {
c7260a81 2857 rcu_read_unlock();
a0cbdd2e
MD
2858 ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key);
2859 break;
2860 }
2861 lttng_poll_del(&events, chan->wait_fd);
f623cc0b 2862 iter.iter.node = &chan->wait_fd_node.node;
a0cbdd2e
MD
2863 ret = lttng_ht_del(channel_ht, &iter);
2864 assert(ret == 0);
2865 consumer_close_channel_streams(chan);
2866
f2a444f1
DG
2867 switch (consumer_data.type) {
2868 case LTTNG_CONSUMER_KERNEL:
2869 break;
2870 case LTTNG_CONSUMER32_UST:
2871 case LTTNG_CONSUMER64_UST:
2872 /* Delete streams that might have been left in the stream list. */
2873 cds_list_for_each_entry_safe(stream, stmp, &chan->streams.head,
2874 send_node) {
9ce5646a
MD
2875 health_code_update();
2876
f2a444f1
DG
2877 cds_list_del(&stream->send_node);
2878 lttng_ustconsumer_del_stream(stream);
2879 uatomic_sub(&stream->chan->refcount, 1);
2880 assert(&chan->refcount);
2881 free(stream);
2882 }
2883 break;
2884 default:
2885 ERR("Unknown consumer_data type");
2886 assert(0);
2887 }
2888
a0cbdd2e
MD
2889 /*
2890 * Release our own refcount. Force channel deletion even if
2891 * streams were not initialized.
2892 */
2893 if (!uatomic_sub_return(&chan->refcount, 1)) {
2894 consumer_del_channel(chan);
2895 }
c7260a81 2896 rcu_read_unlock();
a0cbdd2e
MD
2897 goto restart;
2898 }
d8ef542d
MD
2899 case CONSUMER_CHANNEL_QUIT:
2900 /*
2901 * Remove the pipe from the poll set and continue the loop
2902 * since their might be data to consume.
2903 */
2904 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2905 continue;
2906 default:
2907 ERR("Unknown action");
2908 break;
2909 }
2910 }
2911
2912 /* Handle other stream */
2913 continue;
2914 }
2915
2916 rcu_read_lock();
2917 {
2918 uint64_t tmp_id = (uint64_t) pollfd;
2919
2920 lttng_ht_lookup(channel_ht, &tmp_id, &iter);
2921 }
2922 node = lttng_ht_iter_get_node_u64(&iter);
2923 assert(node);
2924
2925 chan = caa_container_of(node, struct lttng_consumer_channel,
2926 wait_fd_node);
2927
2928 /* Check for error event */
2929 if (revents & (LPOLLERR | LPOLLHUP)) {
2930 DBG("Channel fd %d is hup|err.", pollfd);
2931
2932 lttng_poll_del(&events, chan->wait_fd);
2933 ret = lttng_ht_del(channel_ht, &iter);
2934 assert(ret == 0);
2935 consumer_close_channel_streams(chan);
f2ad556d
MD
2936
2937 /* Release our own refcount */
2938 if (!uatomic_sub_return(&chan->refcount, 1)
2939 && !uatomic_read(&chan->nb_init_stream_left)) {
2940 consumer_del_channel(chan);
2941 }
d8ef542d
MD
2942 }
2943
2944 /* Release RCU lock for the channel looked up */
2945 rcu_read_unlock();
2946 }
2947 }
2948
1fc79fb4
MD
2949 /* All is OK */
2950 err = 0;
d8ef542d
MD
2951end:
2952 lttng_poll_clean(&events);
2953end_poll:
2954 destroy_channel_ht(channel_ht);
2955end_ht:
2956 DBG("Channel poll thread exiting");
1fc79fb4
MD
2957 if (err) {
2958 health_error();
2959 ERR("Health error occurred in %s", __func__);
2960 }
2961 health_unregister(health_consumerd);
d8ef542d
MD
2962 rcu_unregister_thread();
2963 return NULL;
2964}
2965
331744e3
JD
2966static int set_metadata_socket(struct lttng_consumer_local_data *ctx,
2967 struct pollfd *sockpoll, int client_socket)
2968{
2969 int ret;
2970
2971 assert(ctx);
2972 assert(sockpoll);
2973
2974 if (lttng_consumer_poll_socket(sockpoll) < 0) {
2975 ret = -1;
2976 goto error;
2977 }
2978 DBG("Metadata connection on client_socket");
2979
2980 /* Blocking call, waiting for transmission */
2981 ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket);
2982 if (ctx->consumer_metadata_socket < 0) {
2983 WARN("On accept metadata");
2984 ret = -1;
2985 goto error;
2986 }
2987 ret = 0;
2988
2989error:
2990 return ret;
2991}
2992
3bd1e081
MD
2993/*
2994 * This thread listens on the consumerd socket and receives the file
2995 * descriptors from the session daemon.
2996 */
7d980def 2997void *consumer_thread_sessiond_poll(void *data)
3bd1e081 2998{
1fc79fb4 2999 int sock = -1, client_socket, ret, err = -1;
3bd1e081
MD
3000 /*
3001 * structure to poll for incoming data on communication socket avoids
3002 * making blocking sockets.
3003 */
3004 struct pollfd consumer_sockpoll[2];
3005 struct lttng_consumer_local_data *ctx = data;
3006
e7b994a3
DG
3007 rcu_register_thread();
3008
1fc79fb4
MD
3009 health_register(health_consumerd, HEALTH_CONSUMERD_TYPE_SESSIOND);
3010
9ce5646a
MD
3011 health_code_update();
3012
3bd1e081
MD
3013 DBG("Creating command socket %s", ctx->consumer_command_sock_path);
3014 unlink(ctx->consumer_command_sock_path);
3015 client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path);
3016 if (client_socket < 0) {
3017 ERR("Cannot create command socket");
3018 goto end;
3019 }
3020
3021 ret = lttcomm_listen_unix_sock(client_socket);
3022 if (ret < 0) {
3023 goto end;
3024 }
3025
32258573 3026 DBG("Sending ready command to lttng-sessiond");
f73fabfd 3027 ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY);
3bd1e081
MD
3028 /* return < 0 on error, but == 0 is not fatal */
3029 if (ret < 0) {
32258573 3030 ERR("Error sending ready command to lttng-sessiond");
3bd1e081
MD
3031 goto end;
3032 }
3033
3bd1e081
MD
3034 /* prepare the FDs to poll : to client socket and the should_quit pipe */
3035 consumer_sockpoll[0].fd = ctx->consumer_should_quit[0];
3036 consumer_sockpoll[0].events = POLLIN | POLLPRI;
3037 consumer_sockpoll[1].fd = client_socket;
3038 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3039
3040 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
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);
3057 if (ret < 0) {
3058 goto end;
3059 }
3060
d96f09c6
DG
3061 /* This socket is not useful anymore. */
3062 ret = close(client_socket);
3063 if (ret < 0) {
3064 PERROR("close client_socket");
3065 }
3066 client_socket = -1;
3067
3bd1e081
MD
3068 /* update the polling structure to poll on the established socket */
3069 consumer_sockpoll[1].fd = sock;
3070 consumer_sockpoll[1].events = POLLIN | POLLPRI;
3071
3072 while (1) {
9ce5646a
MD
3073 health_code_update();
3074
3075 health_poll_entry();
3076 ret = lttng_consumer_poll_socket(consumer_sockpoll);
3077 health_poll_exit();
3078 if (ret < 0) {
3bd1e081
MD
3079 goto end;
3080 }
3081 DBG("Incoming command on sock");
3082 ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll);
3083 if (ret == -ENOENT) {
3084 DBG("Received STOP command");
3085 goto end;
3086 }
4cbc1a04
DG
3087 if (ret <= 0) {
3088 /*
3089 * This could simply be a session daemon quitting. Don't output
3090 * ERR() here.
3091 */
3092 DBG("Communication interrupted on command socket");
41ba6035 3093 err = 0;
3bd1e081
MD
3094 goto end;
3095 }
3096 if (consumer_quit) {
3097 DBG("consumer_thread_receive_fds received quit from signal");
1fc79fb4 3098 err = 0; /* All is OK */
3bd1e081
MD
3099 goto end;
3100 }
ffe60014 3101 DBG("received command on sock");
3bd1e081 3102 }
1fc79fb4
MD
3103 /* All is OK */
3104 err = 0;
3105
3bd1e081 3106end:
ffe60014 3107 DBG("Consumer thread sessiond poll exiting");
3bd1e081 3108
d88aee68
DG
3109 /*
3110 * Close metadata streams since the producer is the session daemon which
3111 * just died.
3112 *
3113 * NOTE: for now, this only applies to the UST tracer.
3114 */
3115 lttng_consumer_close_metadata();
3116
3bd1e081
MD
3117 /*
3118 * when all fds have hung up, the polling thread
3119 * can exit cleanly
3120 */
3121 consumer_quit = 1;
3122
04fdd819 3123 /*
c869f647 3124 * Notify the data poll thread to poll back again and test the
8994307f 3125 * consumer_quit state that we just set so to quit gracefully.
04fdd819 3126 */
acdb9057 3127 notify_thread_lttng_pipe(ctx->consumer_data_pipe);
c869f647 3128
a0cbdd2e 3129 notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT);
d8ef542d 3130
5c635c72
MD
3131 notify_health_quit_pipe(health_quit_pipe);
3132
d96f09c6
DG
3133 /* Cleaning up possibly open sockets. */
3134 if (sock >= 0) {
3135 ret = close(sock);
3136 if (ret < 0) {
3137 PERROR("close sock sessiond poll");
3138 }
3139 }
3140 if (client_socket >= 0) {
38476d24 3141 ret = close(client_socket);
d96f09c6
DG
3142 if (ret < 0) {
3143 PERROR("close client_socket sessiond poll");
3144 }
3145 }
3146
1fc79fb4
MD
3147 if (err) {
3148 health_error();
3149 ERR("Health error occurred in %s", __func__);
3150 }
3151 health_unregister(health_consumerd);
3152
e7b994a3 3153 rcu_unregister_thread();
3bd1e081
MD
3154 return NULL;
3155}
d41f73b7 3156
4078b776 3157ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream,
d41f73b7
MD
3158 struct lttng_consumer_local_data *ctx)
3159{
74251bb8
DG
3160 ssize_t ret;
3161
3162 pthread_mutex_lock(&stream->lock);
94d49140
JD
3163 if (stream->metadata_flag) {
3164 pthread_mutex_lock(&stream->metadata_rdv_lock);
3165 }
74251bb8 3166
d41f73b7
MD
3167 switch (consumer_data.type) {
3168 case LTTNG_CONSUMER_KERNEL:
74251bb8
DG
3169 ret = lttng_kconsumer_read_subbuffer(stream, ctx);
3170 break;
7753dea8
MD
3171 case LTTNG_CONSUMER32_UST:
3172 case LTTNG_CONSUMER64_UST:
74251bb8
DG
3173 ret = lttng_ustconsumer_read_subbuffer(stream, ctx);
3174 break;
d41f73b7
MD
3175 default:
3176 ERR("Unknown consumer_data type");
3177 assert(0);
74251bb8
DG
3178 ret = -ENOSYS;
3179 break;
d41f73b7 3180 }
74251bb8 3181
94d49140
JD
3182 if (stream->metadata_flag) {
3183 pthread_cond_broadcast(&stream->metadata_rdv);
3184 pthread_mutex_unlock(&stream->metadata_rdv_lock);
3185 }
74251bb8
DG
3186 pthread_mutex_unlock(&stream->lock);
3187 return ret;
d41f73b7
MD
3188}
3189
3190int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream)
3191{
3192 switch (consumer_data.type) {
3193 case LTTNG_CONSUMER_KERNEL:
3194 return lttng_kconsumer_on_recv_stream(stream);
7753dea8
MD
3195 case LTTNG_CONSUMER32_UST:
3196 case LTTNG_CONSUMER64_UST:
d41f73b7
MD
3197 return lttng_ustconsumer_on_recv_stream(stream);
3198 default:
3199 ERR("Unknown consumer_data type");
3200 assert(0);
3201 return -ENOSYS;
3202 }
3203}
e4421fec
DG
3204
3205/*
3206 * Allocate and set consumer data hash tables.
3207 */
3208void lttng_consumer_init(void)
3209{
d88aee68
DG
3210 consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3211 consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3212 consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
d8ef542d 3213 consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
e4421fec 3214}
7735ef9e
DG
3215
3216/*
3217 * Process the ADD_RELAYD command receive by a consumer.
3218 *
3219 * This will create a relayd socket pair and add it to the relayd hash table.
3220 * The caller MUST acquire a RCU read side lock before calling it.
3221 */
da009f2c 3222int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type,
7735ef9e 3223 struct lttng_consumer_local_data *ctx, int sock,
6151a90f 3224 struct pollfd *consumer_sockpoll,
d3e2ba59
JD
3225 struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id,
3226 uint64_t relayd_session_id)
7735ef9e 3227{
cd2b09ed 3228 int fd = -1, ret = -1, relayd_created = 0;
f50f23d9 3229 enum lttng_error_code ret_code = LTTNG_OK;
d4298c99 3230 struct consumer_relayd_sock_pair *relayd = NULL;
7735ef9e 3231
6151a90f
JD
3232 assert(ctx);
3233 assert(relayd_sock);
3234
da009f2c 3235 DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx);
7735ef9e
DG
3236
3237 /* Get relayd reference if exists. */
3238 relayd = consumer_find_relayd(net_seq_idx);
3239 if (relayd == NULL) {
da009f2c 3240 assert(sock_type == LTTNG_STREAM_CONTROL);
7735ef9e
DG
3241 /* Not found. Allocate one. */
3242 relayd = consumer_allocate_relayd_sock_pair(net_seq_idx);
3243 if (relayd == NULL) {
0d08d75e 3244 ret = -ENOMEM;
618a6a28
MD
3245 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
3246 goto error;
0d08d75e 3247 } else {
30319bcb 3248 relayd->sessiond_session_id = sessiond_id;
0d08d75e 3249 relayd_created = 1;
7735ef9e 3250 }
0d08d75e
DG
3251
3252 /*
3253 * This code path MUST continue to the consumer send status message to
3254 * we can notify the session daemon and continue our work without
3255 * killing everything.
3256 */
da009f2c
MD
3257 } else {
3258 /*
3259 * relayd key should never be found for control socket.
3260 */
3261 assert(sock_type != LTTNG_STREAM_CONTROL);
0d08d75e
DG
3262 }
3263
3264 /* First send a status message before receiving the fds. */
618a6a28
MD
3265 ret = consumer_send_status_msg(sock, LTTNG_OK);
3266 if (ret < 0) {
0d08d75e 3267 /* Somehow, the session daemon is not responding anymore. */
618a6a28
MD
3268 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3269 goto error_nosignal;
7735ef9e
DG
3270 }
3271
3272 /* Poll on consumer socket. */
3273 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
0d08d75e 3274 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
7735ef9e 3275 ret = -EINTR;
618a6a28 3276 goto error_nosignal;
7735ef9e
DG
3277 }
3278
3279 /* Get relayd socket from session daemon */
3280 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
3281 if (ret != sizeof(fd)) {
7735ef9e 3282 ret = -1;
4028eeb9 3283 fd = -1; /* Just in case it gets set with an invalid value. */
0d08d75e
DG
3284
3285 /*
3286 * Failing to receive FDs might indicate a major problem such as
3287 * reaching a fd limit during the receive where the kernel returns a
3288 * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we
3289 * don't take any chances and stop everything.
3290 *
3291 * XXX: Feature request #558 will fix that and avoid this possible
3292 * issue when reaching the fd limit.
3293 */
3294 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
618a6a28 3295 ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD;
f50f23d9
DG
3296 goto error;
3297 }
3298
7735ef9e
DG
3299 /* Copy socket information and received FD */
3300 switch (sock_type) {
3301 case LTTNG_STREAM_CONTROL:
3302 /* Copy received lttcomm socket */
6151a90f
JD
3303 lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock);
3304 ret = lttcomm_create_sock(&relayd->control_sock.sock);
4028eeb9 3305 /* Handle create_sock error. */
f66c074c 3306 if (ret < 0) {
618a6a28 3307 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
4028eeb9 3308 goto error;
f66c074c 3309 }
da009f2c
MD
3310 /*
3311 * Close the socket created internally by
3312 * lttcomm_create_sock, so we can replace it by the one
3313 * received from sessiond.
3314 */
3315 if (close(relayd->control_sock.sock.fd)) {
3316 PERROR("close");
3317 }
7735ef9e
DG
3318
3319 /* Assign new file descriptor */
6151a90f 3320 relayd->control_sock.sock.fd = fd;
4b29f1ce 3321 fd = -1; /* For error path */
6151a90f
JD
3322 /* Assign version values. */
3323 relayd->control_sock.major = relayd_sock->major;
3324 relayd->control_sock.minor = relayd_sock->minor;
c5b6f4f0 3325
d3e2ba59 3326 relayd->relayd_session_id = relayd_session_id;
c5b6f4f0 3327
7735ef9e
DG
3328 break;
3329 case LTTNG_STREAM_DATA:
3330 /* Copy received lttcomm socket */
6151a90f
JD
3331 lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock);
3332 ret = lttcomm_create_sock(&relayd->data_sock.sock);
4028eeb9 3333 /* Handle create_sock error. */
f66c074c 3334 if (ret < 0) {
618a6a28 3335 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
4028eeb9 3336 goto error;
f66c074c 3337 }
da009f2c
MD
3338 /*
3339 * Close the socket created internally by
3340 * lttcomm_create_sock, so we can replace it by the one
3341 * received from sessiond.
3342 */
3343 if (close(relayd->data_sock.sock.fd)) {
3344 PERROR("close");
3345 }
7735ef9e
DG
3346
3347 /* Assign new file descriptor */
6151a90f 3348 relayd->data_sock.sock.fd = fd;
4b29f1ce 3349 fd = -1; /* for eventual error paths */
6151a90f
JD
3350 /* Assign version values. */
3351 relayd->data_sock.major = relayd_sock->major;
3352 relayd->data_sock.minor = relayd_sock->minor;
7735ef9e
DG
3353 break;
3354 default:
3355 ERR("Unknown relayd socket type (%d)", sock_type);
59e71485 3356 ret = -1;
618a6a28 3357 ret_code = LTTCOMM_CONSUMERD_FATAL;
7735ef9e
DG
3358 goto error;
3359 }
3360
d88aee68 3361 DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)",
7735ef9e
DG
3362 sock_type == LTTNG_STREAM_CONTROL ? "control" : "data",
3363 relayd->net_seq_idx, fd);
3364
618a6a28
MD
3365 /* We successfully added the socket. Send status back. */
3366 ret = consumer_send_status_msg(sock, ret_code);
3367 if (ret < 0) {
3368 /* Somehow, the session daemon is not responding anymore. */
3369 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3370 goto error_nosignal;
3371 }
3372
7735ef9e
DG
3373 /*
3374 * Add relayd socket pair to consumer data hashtable. If object already
3375 * exists or on error, the function gracefully returns.
3376 */
d09e1200 3377 add_relayd(relayd);
7735ef9e
DG
3378
3379 /* All good! */
4028eeb9 3380 return 0;
7735ef9e
DG
3381
3382error:
618a6a28
MD
3383 if (consumer_send_status_msg(sock, ret_code) < 0) {
3384 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3385 }
3386
3387error_nosignal:
4028eeb9
DG
3388 /* Close received socket if valid. */
3389 if (fd >= 0) {
3390 if (close(fd)) {
3391 PERROR("close received socket");
3392 }
3393 }
cd2b09ed
DG
3394
3395 if (relayd_created) {
cd2b09ed
DG
3396 free(relayd);
3397 }
3398
7735ef9e
DG
3399 return ret;
3400}
ca22feea 3401
4e9a4686
DG
3402/*
3403 * Try to lock the stream mutex.
3404 *
3405 * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
3406 */
3407static int stream_try_lock(struct lttng_consumer_stream *stream)
3408{
3409 int ret;
3410
3411 assert(stream);
3412
3413 /*
3414 * Try to lock the stream mutex. On failure, we know that the stream is
3415 * being used else where hence there is data still being extracted.
3416 */
3417 ret = pthread_mutex_trylock(&stream->lock);
3418 if (ret) {
3419 /* For both EBUSY and EINVAL error, the mutex is NOT locked. */
3420 ret = 0;
3421 goto end;
3422 }
3423
3424 ret = 1;
3425
3426end:
3427 return ret;
3428}
3429
f7079f67
DG
3430/*
3431 * Search for a relayd associated to the session id and return the reference.
3432 *
3433 * A rcu read side lock MUST be acquire before calling this function and locked
3434 * until the relayd object is no longer necessary.
3435 */
3436static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id)
3437{
3438 struct lttng_ht_iter iter;
f7079f67 3439 struct consumer_relayd_sock_pair *relayd = NULL;
f7079f67
DG
3440
3441 /* Iterate over all relayd since they are indexed by net_seq_idx. */
3442 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
3443 node.node) {
18261bd1
DG
3444 /*
3445 * Check by sessiond id which is unique here where the relayd session
3446 * id might not be when having multiple relayd.
3447 */
3448 if (relayd->sessiond_session_id == id) {
f7079f67 3449 /* Found the relayd. There can be only one per id. */
18261bd1 3450 goto found;
f7079f67
DG
3451 }
3452 }
3453
18261bd1
DG
3454 return NULL;
3455
3456found:
f7079f67
DG
3457 return relayd;
3458}
3459
ca22feea
DG
3460/*
3461 * Check if for a given session id there is still data needed to be extract
3462 * from the buffers.
3463 *
6d805429 3464 * Return 1 if data is pending or else 0 meaning ready to be read.
ca22feea 3465 */
6d805429 3466int consumer_data_pending(uint64_t id)
ca22feea
DG
3467{
3468 int ret;
3469 struct lttng_ht_iter iter;
3470 struct lttng_ht *ht;
3471 struct lttng_consumer_stream *stream;
f7079f67 3472 struct consumer_relayd_sock_pair *relayd = NULL;
6d805429 3473 int (*data_pending)(struct lttng_consumer_stream *);
ca22feea 3474
6d805429 3475 DBG("Consumer data pending command on session id %" PRIu64, id);
ca22feea 3476
6f6eda74 3477 rcu_read_lock();
ca22feea
DG
3478 pthread_mutex_lock(&consumer_data.lock);
3479
3480 switch (consumer_data.type) {
3481 case LTTNG_CONSUMER_KERNEL:
6d805429 3482 data_pending = lttng_kconsumer_data_pending;
ca22feea
DG
3483 break;
3484 case LTTNG_CONSUMER32_UST:
3485 case LTTNG_CONSUMER64_UST:
6d805429 3486 data_pending = lttng_ustconsumer_data_pending;
ca22feea
DG
3487 break;
3488 default:
3489 ERR("Unknown consumer data type");
3490 assert(0);
3491 }
3492
3493 /* Ease our life a bit */
3494 ht = consumer_data.stream_list_ht;
3495
f7079f67
DG
3496 relayd = find_relayd_by_session_id(id);
3497 if (relayd) {
3498 /* Send init command for data pending. */
3499 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3500 ret = relayd_begin_data_pending(&relayd->control_sock,
3501 relayd->relayd_session_id);
3502 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
3503 if (ret < 0) {
3504 /* Communication error thus the relayd so no data pending. */
3505 goto data_not_pending;
3506 }
3507 }
3508
c8f59ee5 3509 cds_lfht_for_each_entry_duplicate(ht->ht,
d88aee68
DG
3510 ht->hash_fct(&id, lttng_ht_seed),
3511 ht->match_fct, &id,
ca22feea 3512 &iter.iter, stream, node_session_id.node) {
4e9a4686
DG
3513 /* If this call fails, the stream is being used hence data pending. */
3514 ret = stream_try_lock(stream);
3515 if (!ret) {
f7079f67 3516 goto data_pending;
ca22feea 3517 }
ca22feea 3518
4e9a4686
DG
3519 /*
3520 * A removed node from the hash table indicates that the stream has
3521 * been deleted thus having a guarantee that the buffers are closed
3522 * on the consumer side. However, data can still be transmitted
3523 * over the network so don't skip the relayd check.
3524 */
3525 ret = cds_lfht_is_node_deleted(&stream->node.node);
3526 if (!ret) {
e5d1a9b3
MD
3527 /*
3528 * An empty output file is not valid. We need at least one packet
3529 * generated per stream, even if it contains no event, so it
3530 * contains at least one packet header.
3531 */
3532 if (stream->output_written == 0) {
3533 pthread_mutex_unlock(&stream->lock);
3534 goto data_pending;
3535 }
4e9a4686 3536 /* Check the stream if there is data in the buffers. */
6d805429
DG
3537 ret = data_pending(stream);
3538 if (ret == 1) {
4e9a4686 3539 pthread_mutex_unlock(&stream->lock);
f7079f67 3540 goto data_pending;
4e9a4686
DG
3541 }
3542 }
3543
3544 /* Relayd check */
f7079f67 3545 if (relayd) {
c8f59ee5
DG
3546 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3547 if (stream->metadata_flag) {
ad7051c0
DG
3548 ret = relayd_quiescent_control(&relayd->control_sock,
3549 stream->relayd_stream_id);
c8f59ee5 3550 } else {
6d805429 3551 ret = relayd_data_pending(&relayd->control_sock,
39df6d9f
DG
3552 stream->relayd_stream_id,
3553 stream->next_net_seq_num - 1);
c8f59ee5
DG
3554 }
3555 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
6d805429 3556 if (ret == 1) {
4e9a4686 3557 pthread_mutex_unlock(&stream->lock);
f7079f67 3558 goto data_pending;
c8f59ee5
DG
3559 }
3560 }
4e9a4686 3561 pthread_mutex_unlock(&stream->lock);
c8f59ee5 3562 }
ca22feea 3563
f7079f67
DG
3564 if (relayd) {
3565 unsigned int is_data_inflight = 0;
3566
3567 /* Send init command for data pending. */
3568 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3569 ret = relayd_end_data_pending(&relayd->control_sock,
3570 relayd->relayd_session_id, &is_data_inflight);
3571 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
bdd88757 3572 if (ret < 0) {
f7079f67
DG
3573 goto data_not_pending;
3574 }
bdd88757
DG
3575 if (is_data_inflight) {
3576 goto data_pending;
3577 }
f7079f67
DG
3578 }
3579
ca22feea 3580 /*
f7079f67
DG
3581 * Finding _no_ node in the hash table and no inflight data means that the
3582 * stream(s) have been removed thus data is guaranteed to be available for
3583 * analysis from the trace files.
ca22feea
DG
3584 */
3585
f7079f67 3586data_not_pending:
ca22feea
DG
3587 /* Data is available to be read by a viewer. */
3588 pthread_mutex_unlock(&consumer_data.lock);
c8f59ee5 3589 rcu_read_unlock();
6d805429 3590 return 0;
ca22feea 3591
f7079f67 3592data_pending:
ca22feea
DG
3593 /* Data is still being extracted from buffers. */
3594 pthread_mutex_unlock(&consumer_data.lock);
c8f59ee5 3595 rcu_read_unlock();
6d805429 3596 return 1;
ca22feea 3597}
f50f23d9
DG
3598
3599/*
3600 * Send a ret code status message to the sessiond daemon.
3601 *
3602 * Return the sendmsg() return value.
3603 */
3604int consumer_send_status_msg(int sock, int ret_code)
3605{
3606 struct lttcomm_consumer_status_msg msg;
3607
3608 msg.ret_code = ret_code;
3609
3610 return lttcomm_send_unix_sock(sock, &msg, sizeof(msg));
3611}
ffe60014
DG
3612
3613/*
3614 * Send a channel status message to the sessiond daemon.
3615 *
3616 * Return the sendmsg() return value.
3617 */
3618int consumer_send_status_channel(int sock,
3619 struct lttng_consumer_channel *channel)
3620{
3621 struct lttcomm_consumer_status_channel msg;
3622
3623 assert(sock >= 0);
3624
3625 if (!channel) {
3626 msg.ret_code = -LTTNG_ERR_UST_CHAN_FAIL;
3627 } else {
3628 msg.ret_code = LTTNG_OK;
3629 msg.key = channel->key;
3630 msg.stream_count = channel->streams.count;
3631 }
3632
3633 return lttcomm_send_unix_sock(sock, &msg, sizeof(msg));
3634}
5c786ded
JD
3635
3636/*
3637 * Using a maximum stream size with the produced and consumed position of a
3638 * stream, computes the new consumed position to be as close as possible to the
3639 * maximum possible stream size.
3640 *
3641 * If maximum stream size is lower than the possible buffer size (produced -
3642 * consumed), the consumed_pos given is returned untouched else the new value
3643 * is returned.
3644 */
3645unsigned long consumer_get_consumed_maxsize(unsigned long consumed_pos,
3646 unsigned long produced_pos, uint64_t max_stream_size)
3647{
3648 if (max_stream_size && max_stream_size < (produced_pos - consumed_pos)) {
3649 /* Offset from the produced position to get the latest buffers. */
3650 return produced_pos - max_stream_size;
3651 }
3652
3653 return consumed_pos;
3654}
This page took 0.271104 seconds and 5 git commands to generate.