Fix: reset out_fd_offset when we rotate the trace file
[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
990570ed 33#include <common/common.h>
fb3a43a9
DG
34#include <common/utils.h>
35#include <common/compat/poll.h>
10a8a223 36#include <common/kernel-ctl/kernel-ctl.h>
00e2e675 37#include <common/sessiond-comm/relayd.h>
10a8a223
DG
38#include <common/sessiond-comm/sessiond-comm.h>
39#include <common/kernel-consumer/kernel-consumer.h>
00e2e675 40#include <common/relayd/relayd.h>
10a8a223
DG
41#include <common/ust-consumer/ust-consumer.h>
42
43#include "consumer.h"
1d1a276c 44#include "consumer-stream.h"
3bd1e081
MD
45
46struct lttng_consumer_global_data consumer_data = {
3bd1e081
MD
47 .stream_count = 0,
48 .need_update = 1,
49 .type = LTTNG_CONSUMER_UNKNOWN,
50};
51
d8ef542d
MD
52enum consumer_channel_action {
53 CONSUMER_CHANNEL_ADD,
a0cbdd2e 54 CONSUMER_CHANNEL_DEL,
d8ef542d
MD
55 CONSUMER_CHANNEL_QUIT,
56};
57
58struct consumer_channel_msg {
59 enum consumer_channel_action action;
a0cbdd2e
MD
60 struct lttng_consumer_channel *chan; /* add */
61 uint64_t key; /* del */
d8ef542d
MD
62};
63
3bd1e081
MD
64/*
65 * Flag to inform the polling thread to quit when all fd hung up. Updated by
66 * the consumer_thread_receive_fds when it notices that all fds has hung up.
67 * Also updated by the signal handler (consumer_should_exit()). Read by the
68 * polling threads.
69 */
a98dae5f 70volatile int consumer_quit;
3bd1e081 71
43c34bc3 72/*
43c34bc3
DG
73 * Global hash table containing respectively metadata and data streams. The
74 * stream element in this ht should only be updated by the metadata poll thread
75 * for the metadata and the data poll thread for the data.
76 */
40dc48e0
DG
77static struct lttng_ht *metadata_ht;
78static struct lttng_ht *data_ht;
43c34bc3 79
acdb9057
DG
80/*
81 * Notify a thread lttng pipe to poll back again. This usually means that some
82 * global state has changed so we just send back the thread in a poll wait
83 * call.
84 */
85static void notify_thread_lttng_pipe(struct lttng_pipe *pipe)
86{
87 struct lttng_consumer_stream *null_stream = NULL;
88
89 assert(pipe);
90
91 (void) lttng_pipe_write(pipe, &null_stream, sizeof(null_stream));
92}
93
d8ef542d
MD
94static void notify_channel_pipe(struct lttng_consumer_local_data *ctx,
95 struct lttng_consumer_channel *chan,
a0cbdd2e 96 uint64_t key,
d8ef542d
MD
97 enum consumer_channel_action action)
98{
99 struct consumer_channel_msg msg;
100 int ret;
101
e56251fc
DG
102 memset(&msg, 0, sizeof(msg));
103
d8ef542d
MD
104 msg.action = action;
105 msg.chan = chan;
f21dae48 106 msg.key = key;
d8ef542d
MD
107 do {
108 ret = write(ctx->consumer_channel_pipe[1], &msg, sizeof(msg));
109 } while (ret < 0 && errno == EINTR);
110}
111
a0cbdd2e
MD
112void notify_thread_del_channel(struct lttng_consumer_local_data *ctx,
113 uint64_t key)
114{
115 notify_channel_pipe(ctx, NULL, key, CONSUMER_CHANNEL_DEL);
116}
117
d8ef542d
MD
118static int read_channel_pipe(struct lttng_consumer_local_data *ctx,
119 struct lttng_consumer_channel **chan,
a0cbdd2e 120 uint64_t *key,
d8ef542d
MD
121 enum consumer_channel_action *action)
122{
123 struct consumer_channel_msg msg;
124 int ret;
125
126 do {
127 ret = read(ctx->consumer_channel_pipe[0], &msg, sizeof(msg));
128 } while (ret < 0 && errno == EINTR);
129 if (ret > 0) {
130 *action = msg.action;
131 *chan = msg.chan;
a0cbdd2e 132 *key = msg.key;
d8ef542d
MD
133 }
134 return ret;
135}
136
3bd1e081
MD
137/*
138 * Find a stream. The consumer_data.lock must be locked during this
139 * call.
140 */
d88aee68 141static struct lttng_consumer_stream *find_stream(uint64_t key,
8389e4f8 142 struct lttng_ht *ht)
3bd1e081 143{
e4421fec 144 struct lttng_ht_iter iter;
d88aee68 145 struct lttng_ht_node_u64 *node;
e4421fec 146 struct lttng_consumer_stream *stream = NULL;
3bd1e081 147
8389e4f8
DG
148 assert(ht);
149
d88aee68
DG
150 /* -1ULL keys are lookup failures */
151 if (key == (uint64_t) -1ULL) {
7ad0a0cb 152 return NULL;
7a57cf92 153 }
e4421fec 154
6065ceec
DG
155 rcu_read_lock();
156
d88aee68
DG
157 lttng_ht_lookup(ht, &key, &iter);
158 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
159 if (node != NULL) {
160 stream = caa_container_of(node, struct lttng_consumer_stream, node);
3bd1e081 161 }
e4421fec 162
6065ceec
DG
163 rcu_read_unlock();
164
e4421fec 165 return stream;
3bd1e081
MD
166}
167
da009f2c 168static void steal_stream_key(uint64_t key, struct lttng_ht *ht)
7ad0a0cb
MD
169{
170 struct lttng_consumer_stream *stream;
171
04253271 172 rcu_read_lock();
ffe60014 173 stream = find_stream(key, ht);
04253271 174 if (stream) {
da009f2c 175 stream->key = (uint64_t) -1ULL;
04253271
MD
176 /*
177 * We don't want the lookup to match, but we still need
178 * to iterate on this stream when iterating over the hash table. Just
179 * change the node key.
180 */
da009f2c 181 stream->node.key = (uint64_t) -1ULL;
04253271
MD
182 }
183 rcu_read_unlock();
7ad0a0cb
MD
184}
185
d56db448
DG
186/*
187 * Return a channel object for the given key.
188 *
189 * RCU read side lock MUST be acquired before calling this function and
190 * protects the channel ptr.
191 */
d88aee68 192struct lttng_consumer_channel *consumer_find_channel(uint64_t key)
3bd1e081 193{
e4421fec 194 struct lttng_ht_iter iter;
d88aee68 195 struct lttng_ht_node_u64 *node;
e4421fec 196 struct lttng_consumer_channel *channel = NULL;
3bd1e081 197
d88aee68
DG
198 /* -1ULL keys are lookup failures */
199 if (key == (uint64_t) -1ULL) {
7ad0a0cb 200 return NULL;
7a57cf92 201 }
e4421fec 202
d88aee68
DG
203 lttng_ht_lookup(consumer_data.channel_ht, &key, &iter);
204 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
205 if (node != NULL) {
206 channel = caa_container_of(node, struct lttng_consumer_channel, node);
3bd1e081 207 }
e4421fec
DG
208
209 return channel;
3bd1e081
MD
210}
211
ffe60014 212static void free_stream_rcu(struct rcu_head *head)
7ad0a0cb 213{
d88aee68
DG
214 struct lttng_ht_node_u64 *node =
215 caa_container_of(head, struct lttng_ht_node_u64, head);
ffe60014
DG
216 struct lttng_consumer_stream *stream =
217 caa_container_of(node, struct lttng_consumer_stream, node);
7ad0a0cb 218
ffe60014 219 free(stream);
7ad0a0cb
MD
220}
221
ffe60014 222static void free_channel_rcu(struct rcu_head *head)
702b1ea4 223{
d88aee68
DG
224 struct lttng_ht_node_u64 *node =
225 caa_container_of(head, struct lttng_ht_node_u64, head);
ffe60014
DG
226 struct lttng_consumer_channel *channel =
227 caa_container_of(node, struct lttng_consumer_channel, node);
702b1ea4 228
ffe60014 229 free(channel);
702b1ea4
MD
230}
231
00e2e675
DG
232/*
233 * RCU protected relayd socket pair free.
234 */
ffe60014 235static void free_relayd_rcu(struct rcu_head *head)
00e2e675 236{
d88aee68
DG
237 struct lttng_ht_node_u64 *node =
238 caa_container_of(head, struct lttng_ht_node_u64, head);
00e2e675
DG
239 struct consumer_relayd_sock_pair *relayd =
240 caa_container_of(node, struct consumer_relayd_sock_pair, node);
241
8994307f
DG
242 /*
243 * Close all sockets. This is done in the call RCU since we don't want the
244 * socket fds to be reassigned thus potentially creating bad state of the
245 * relayd object.
246 *
247 * We do not have to lock the control socket mutex here since at this stage
248 * there is no one referencing to this relayd object.
249 */
250 (void) relayd_close(&relayd->control_sock);
251 (void) relayd_close(&relayd->data_sock);
252
00e2e675
DG
253 free(relayd);
254}
255
256/*
257 * Destroy and free relayd socket pair object.
00e2e675 258 */
51230d70 259void consumer_destroy_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
260{
261 int ret;
262 struct lttng_ht_iter iter;
263
173af62f
DG
264 if (relayd == NULL) {
265 return;
266 }
267
00e2e675
DG
268 DBG("Consumer destroy and close relayd socket pair");
269
270 iter.iter.node = &relayd->node.node;
271 ret = lttng_ht_del(consumer_data.relayd_ht, &iter);
173af62f 272 if (ret != 0) {
8994307f 273 /* We assume the relayd is being or is destroyed */
173af62f
DG
274 return;
275 }
00e2e675 276
00e2e675 277 /* RCU free() call */
ffe60014
DG
278 call_rcu(&relayd->node.head, free_relayd_rcu);
279}
280
281/*
282 * Remove a channel from the global list protected by a mutex. This function is
283 * also responsible for freeing its data structures.
284 */
285void consumer_del_channel(struct lttng_consumer_channel *channel)
286{
287 int ret;
288 struct lttng_ht_iter iter;
f2a444f1 289 struct lttng_consumer_stream *stream, *stmp;
ffe60014 290
d88aee68 291 DBG("Consumer delete channel key %" PRIu64, channel->key);
ffe60014
DG
292
293 pthread_mutex_lock(&consumer_data.lock);
a9838785 294 pthread_mutex_lock(&channel->lock);
ffe60014 295
51e762e5
JD
296 /* Delete streams that might have been left in the stream list. */
297 cds_list_for_each_entry_safe(stream, stmp, &channel->streams.head,
298 send_node) {
299 cds_list_del(&stream->send_node);
300 /*
301 * Once a stream is added to this list, the buffers were created so
302 * we have a guarantee that this call will succeed.
303 */
304 consumer_stream_destroy(stream, NULL);
305 }
306
ffe60014
DG
307 switch (consumer_data.type) {
308 case LTTNG_CONSUMER_KERNEL:
309 break;
310 case LTTNG_CONSUMER32_UST:
311 case LTTNG_CONSUMER64_UST:
312 lttng_ustconsumer_del_channel(channel);
313 break;
314 default:
315 ERR("Unknown consumer_data type");
316 assert(0);
317 goto end;
318 }
319
320 rcu_read_lock();
321 iter.iter.node = &channel->node.node;
322 ret = lttng_ht_del(consumer_data.channel_ht, &iter);
323 assert(!ret);
324 rcu_read_unlock();
325
326 call_rcu(&channel->node.head, free_channel_rcu);
327end:
a9838785 328 pthread_mutex_unlock(&channel->lock);
ffe60014 329 pthread_mutex_unlock(&consumer_data.lock);
00e2e675
DG
330}
331
228b5bf7
DG
332/*
333 * Iterate over the relayd hash table and destroy each element. Finally,
334 * destroy the whole hash table.
335 */
336static void cleanup_relayd_ht(void)
337{
338 struct lttng_ht_iter iter;
339 struct consumer_relayd_sock_pair *relayd;
340
341 rcu_read_lock();
342
343 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
344 node.node) {
51230d70 345 consumer_destroy_relayd(relayd);
228b5bf7
DG
346 }
347
228b5bf7 348 rcu_read_unlock();
36b588ed
MD
349
350 lttng_ht_destroy(consumer_data.relayd_ht);
228b5bf7
DG
351}
352
8994307f
DG
353/*
354 * Update the end point status of all streams having the given network sequence
355 * index (relayd index).
356 *
357 * It's atomically set without having the stream mutex locked which is fine
358 * because we handle the write/read race with a pipe wakeup for each thread.
359 */
da009f2c 360static void update_endpoint_status_by_netidx(uint64_t net_seq_idx,
8994307f
DG
361 enum consumer_endpoint_status status)
362{
363 struct lttng_ht_iter iter;
364 struct lttng_consumer_stream *stream;
365
da009f2c 366 DBG("Consumer set delete flag on stream by idx %" PRIu64, net_seq_idx);
8994307f
DG
367
368 rcu_read_lock();
369
370 /* Let's begin with metadata */
371 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
372 if (stream->net_seq_idx == net_seq_idx) {
373 uatomic_set(&stream->endpoint_status, status);
374 DBG("Delete flag set to metadata stream %d", stream->wait_fd);
375 }
376 }
377
378 /* Follow up by the data streams */
379 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
380 if (stream->net_seq_idx == net_seq_idx) {
381 uatomic_set(&stream->endpoint_status, status);
382 DBG("Delete flag set to data stream %d", stream->wait_fd);
383 }
384 }
385 rcu_read_unlock();
386}
387
388/*
389 * Cleanup a relayd object by flagging every associated streams for deletion,
390 * destroying the object meaning removing it from the relayd hash table,
391 * closing the sockets and freeing the memory in a RCU call.
392 *
393 * If a local data context is available, notify the threads that the streams'
394 * state have changed.
395 */
396static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd,
397 struct lttng_consumer_local_data *ctx)
398{
da009f2c 399 uint64_t netidx;
8994307f
DG
400
401 assert(relayd);
402
9617607b
DG
403 DBG("Cleaning up relayd sockets");
404
8994307f
DG
405 /* Save the net sequence index before destroying the object */
406 netidx = relayd->net_seq_idx;
407
408 /*
409 * Delete the relayd from the relayd hash table, close the sockets and free
410 * the object in a RCU call.
411 */
51230d70 412 consumer_destroy_relayd(relayd);
8994307f
DG
413
414 /* Set inactive endpoint to all streams */
415 update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE);
416
417 /*
418 * With a local data context, notify the threads that the streams' state
419 * have changed. The write() action on the pipe acts as an "implicit"
420 * memory barrier ordering the updates of the end point status from the
421 * read of this status which happens AFTER receiving this notify.
422 */
423 if (ctx) {
acdb9057 424 notify_thread_lttng_pipe(ctx->consumer_data_pipe);
13886d2d 425 notify_thread_lttng_pipe(ctx->consumer_metadata_pipe);
8994307f
DG
426 }
427}
428
a6ba4fe1
DG
429/*
430 * Flag a relayd socket pair for destruction. Destroy it if the refcount
431 * reaches zero.
432 *
433 * RCU read side lock MUST be aquired before calling this function.
434 */
435void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd)
436{
437 assert(relayd);
438
439 /* Set destroy flag for this object */
440 uatomic_set(&relayd->destroy_flag, 1);
441
442 /* Destroy the relayd if refcount is 0 */
443 if (uatomic_read(&relayd->refcount) == 0) {
51230d70 444 consumer_destroy_relayd(relayd);
a6ba4fe1
DG
445 }
446}
447
3bd1e081 448/*
1d1a276c
DG
449 * Completly destroy stream from every visiable data structure and the given
450 * hash table if one.
451 *
452 * One this call returns, the stream object is not longer usable nor visible.
3bd1e081 453 */
e316aad5
DG
454void consumer_del_stream(struct lttng_consumer_stream *stream,
455 struct lttng_ht *ht)
3bd1e081 456{
1d1a276c 457 consumer_stream_destroy(stream, ht);
3bd1e081
MD
458}
459
5ab66908
MD
460/*
461 * XXX naming of del vs destroy is all mixed up.
462 */
463void consumer_del_stream_for_data(struct lttng_consumer_stream *stream)
464{
465 consumer_stream_destroy(stream, data_ht);
466}
467
468void consumer_del_stream_for_metadata(struct lttng_consumer_stream *stream)
469{
470 consumer_stream_destroy(stream, metadata_ht);
471}
472
d88aee68
DG
473struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key,
474 uint64_t stream_key,
3bd1e081 475 enum lttng_consumer_stream_state state,
ffe60014 476 const char *channel_name,
6df2e2c9 477 uid_t uid,
00e2e675 478 gid_t gid,
57a269f2 479 uint64_t relayd_id,
53632229 480 uint64_t session_id,
ffe60014
DG
481 int cpu,
482 int *alloc_ret,
4891ece8
DG
483 enum consumer_channel_type type,
484 unsigned int monitor)
3bd1e081 485{
ffe60014 486 int ret;
3bd1e081 487 struct lttng_consumer_stream *stream;
3bd1e081 488
effcf122 489 stream = zmalloc(sizeof(*stream));
3bd1e081 490 if (stream == NULL) {
7a57cf92 491 PERROR("malloc struct lttng_consumer_stream");
ffe60014 492 ret = -ENOMEM;
7a57cf92 493 goto end;
3bd1e081 494 }
7a57cf92 495
d56db448
DG
496 rcu_read_lock();
497
3bd1e081 498 stream->key = stream_key;
3bd1e081
MD
499 stream->out_fd = -1;
500 stream->out_fd_offset = 0;
501 stream->state = state;
6df2e2c9
MD
502 stream->uid = uid;
503 stream->gid = gid;
ffe60014 504 stream->net_seq_idx = relayd_id;
53632229 505 stream->session_id = session_id;
4891ece8 506 stream->monitor = monitor;
774d490c 507 stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE;
53632229 508 pthread_mutex_init(&stream->lock, NULL);
58b1f425 509
ffe60014
DG
510 /* If channel is the metadata, flag this stream as metadata. */
511 if (type == CONSUMER_CHANNEL_TYPE_METADATA) {
512 stream->metadata_flag = 1;
513 /* Metadata is flat out. */
514 strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name));
58b1f425 515 } else {
ffe60014
DG
516 /* Format stream name to <channel_name>_<cpu_number> */
517 ret = snprintf(stream->name, sizeof(stream->name), "%s_%d",
518 channel_name, cpu);
519 if (ret < 0) {
520 PERROR("snprintf stream name");
521 goto error;
522 }
58b1f425 523 }
c30aaa51 524
ffe60014 525 /* Key is always the wait_fd for streams. */
d88aee68 526 lttng_ht_node_init_u64(&stream->node, stream->key);
ffe60014 527
d8ef542d
MD
528 /* Init node per channel id key */
529 lttng_ht_node_init_u64(&stream->node_channel_id, channel_key);
530
53632229 531 /* Init session id node with the stream session id */
d88aee68 532 lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id);
53632229 533
07b86b52
JD
534 DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64
535 " relayd_id %" PRIu64 ", session_id %" PRIu64,
536 stream->name, stream->key, channel_key,
537 stream->net_seq_idx, stream->session_id);
d56db448
DG
538
539 rcu_read_unlock();
3bd1e081 540 return stream;
c80048c6
MD
541
542error:
d56db448 543 rcu_read_unlock();
c80048c6 544 free(stream);
7a57cf92 545end:
ffe60014
DG
546 if (alloc_ret) {
547 *alloc_ret = ret;
548 }
c80048c6 549 return NULL;
3bd1e081
MD
550}
551
552/*
553 * Add a stream to the global list protected by a mutex.
554 */
5ab66908 555int consumer_add_data_stream(struct lttng_consumer_stream *stream)
3bd1e081 556{
5ab66908 557 struct lttng_ht *ht = data_ht;
3bd1e081
MD
558 int ret = 0;
559
e316aad5 560 assert(stream);
43c34bc3 561 assert(ht);
c77fc10a 562
d88aee68 563 DBG3("Adding consumer stream %" PRIu64, stream->key);
e316aad5
DG
564
565 pthread_mutex_lock(&consumer_data.lock);
a9838785 566 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 567 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 568 pthread_mutex_lock(&stream->lock);
b0b335c8 569 rcu_read_lock();
e316aad5 570
43c34bc3 571 /* Steal stream identifier to avoid having streams with the same key */
ffe60014 572 steal_stream_key(stream->key, ht);
43c34bc3 573
d88aee68 574 lttng_ht_add_unique_u64(ht, &stream->node);
00e2e675 575
d8ef542d
MD
576 lttng_ht_add_u64(consumer_data.stream_per_chan_id_ht,
577 &stream->node_channel_id);
578
ca22feea
DG
579 /*
580 * Add stream to the stream_list_ht of the consumer data. No need to steal
581 * the key since the HT does not use it and we allow to add redundant keys
582 * into this table.
583 */
d88aee68 584 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 585
e316aad5 586 /*
ffe60014
DG
587 * When nb_init_stream_left reaches 0, we don't need to trigger any action
588 * in terms of destroying the associated channel, because the action that
e316aad5
DG
589 * causes the count to become 0 also causes a stream to be added. The
590 * channel deletion will thus be triggered by the following removal of this
591 * stream.
592 */
ffe60014 593 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
594 /* Increment refcount before decrementing nb_init_stream_left */
595 cmm_smp_wmb();
ffe60014 596 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
597 }
598
599 /* Update consumer data once the node is inserted. */
3bd1e081
MD
600 consumer_data.stream_count++;
601 consumer_data.need_update = 1;
602
e316aad5 603 rcu_read_unlock();
2e818a6a 604 pthread_mutex_unlock(&stream->lock);
ec6ea7d0 605 pthread_mutex_unlock(&stream->chan->timer_lock);
a9838785 606 pthread_mutex_unlock(&stream->chan->lock);
3bd1e081 607 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 608
3bd1e081
MD
609 return ret;
610}
611
5ab66908
MD
612void consumer_del_data_stream(struct lttng_consumer_stream *stream)
613{
614 consumer_del_stream(stream, data_ht);
615}
616
00e2e675 617/*
3f8e211f
DG
618 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
619 * be acquired before calling this.
00e2e675 620 */
d09e1200 621static int add_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
622{
623 int ret = 0;
d88aee68 624 struct lttng_ht_node_u64 *node;
00e2e675
DG
625 struct lttng_ht_iter iter;
626
ffe60014 627 assert(relayd);
00e2e675 628
00e2e675 629 lttng_ht_lookup(consumer_data.relayd_ht,
d88aee68
DG
630 &relayd->net_seq_idx, &iter);
631 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675 632 if (node != NULL) {
00e2e675
DG
633 goto end;
634 }
d88aee68 635 lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node);
00e2e675 636
00e2e675
DG
637end:
638 return ret;
639}
640
641/*
642 * Allocate and return a consumer relayd socket.
643 */
644struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair(
da009f2c 645 uint64_t net_seq_idx)
00e2e675
DG
646{
647 struct consumer_relayd_sock_pair *obj = NULL;
648
da009f2c
MD
649 /* net sequence index of -1 is a failure */
650 if (net_seq_idx == (uint64_t) -1ULL) {
00e2e675
DG
651 goto error;
652 }
653
654 obj = zmalloc(sizeof(struct consumer_relayd_sock_pair));
655 if (obj == NULL) {
656 PERROR("zmalloc relayd sock");
657 goto error;
658 }
659
660 obj->net_seq_idx = net_seq_idx;
661 obj->refcount = 0;
173af62f 662 obj->destroy_flag = 0;
f96e4545
MD
663 obj->control_sock.sock.fd = -1;
664 obj->data_sock.sock.fd = -1;
d88aee68 665 lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx);
00e2e675
DG
666 pthread_mutex_init(&obj->ctrl_sock_mutex, NULL);
667
668error:
669 return obj;
670}
671
672/*
673 * Find a relayd socket pair in the global consumer data.
674 *
675 * Return the object if found else NULL.
b0b335c8
MD
676 * RCU read-side lock must be held across this call and while using the
677 * returned object.
00e2e675 678 */
d88aee68 679struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key)
00e2e675
DG
680{
681 struct lttng_ht_iter iter;
d88aee68 682 struct lttng_ht_node_u64 *node;
00e2e675
DG
683 struct consumer_relayd_sock_pair *relayd = NULL;
684
685 /* Negative keys are lookup failures */
d88aee68 686 if (key == (uint64_t) -1ULL) {
00e2e675
DG
687 goto error;
688 }
689
d88aee68 690 lttng_ht_lookup(consumer_data.relayd_ht, &key,
00e2e675 691 &iter);
d88aee68 692 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675
DG
693 if (node != NULL) {
694 relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node);
695 }
696
00e2e675
DG
697error:
698 return relayd;
699}
700
10a50311
JD
701/*
702 * Find a relayd and send the stream
703 *
704 * Returns 0 on success, < 0 on error
705 */
706int consumer_send_relayd_stream(struct lttng_consumer_stream *stream,
707 char *path)
708{
709 int ret = 0;
710 struct consumer_relayd_sock_pair *relayd;
711
712 assert(stream);
713 assert(stream->net_seq_idx != -1ULL);
714 assert(path);
715
716 /* The stream is not metadata. Get relayd reference if exists. */
717 rcu_read_lock();
718 relayd = consumer_find_relayd(stream->net_seq_idx);
719 if (relayd != NULL) {
720 /* Add stream on the relayd */
721 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
722 ret = relayd_add_stream(&relayd->control_sock, stream->name,
723 path, &stream->relayd_stream_id,
724 stream->chan->tracefile_size, stream->chan->tracefile_count);
725 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
726 if (ret < 0) {
727 goto end;
728 }
729 uatomic_inc(&relayd->refcount);
d01178b6 730 stream->sent_to_relayd = 1;
10a50311
JD
731 } else {
732 ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't send it.",
733 stream->key, stream->net_seq_idx);
734 ret = -1;
735 goto end;
736 }
737
738 DBG("Stream %s with key %" PRIu64 " sent to relayd id %" PRIu64,
739 stream->name, stream->key, stream->net_seq_idx);
740
741end:
742 rcu_read_unlock();
743 return ret;
744}
745
746/*
747 * Find a relayd and close the stream
748 */
749void close_relayd_stream(struct lttng_consumer_stream *stream)
750{
751 struct consumer_relayd_sock_pair *relayd;
752
753 /* The stream is not metadata. Get relayd reference if exists. */
754 rcu_read_lock();
755 relayd = consumer_find_relayd(stream->net_seq_idx);
756 if (relayd) {
757 consumer_stream_relayd_close(stream, relayd);
758 }
759 rcu_read_unlock();
760}
761
00e2e675
DG
762/*
763 * Handle stream for relayd transmission if the stream applies for network
764 * streaming where the net sequence index is set.
765 *
766 * Return destination file descriptor or negative value on error.
767 */
6197aea7 768static int write_relayd_stream_header(struct lttng_consumer_stream *stream,
1d4dfdef
DG
769 size_t data_size, unsigned long padding,
770 struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
771{
772 int outfd = -1, ret;
00e2e675
DG
773 struct lttcomm_relayd_data_hdr data_hdr;
774
775 /* Safety net */
776 assert(stream);
6197aea7 777 assert(relayd);
00e2e675
DG
778
779 /* Reset data header */
780 memset(&data_hdr, 0, sizeof(data_hdr));
781
00e2e675
DG
782 if (stream->metadata_flag) {
783 /* Caller MUST acquire the relayd control socket lock */
784 ret = relayd_send_metadata(&relayd->control_sock, data_size);
785 if (ret < 0) {
786 goto error;
787 }
788
789 /* Metadata are always sent on the control socket. */
6151a90f 790 outfd = relayd->control_sock.sock.fd;
00e2e675
DG
791 } else {
792 /* Set header with stream information */
793 data_hdr.stream_id = htobe64(stream->relayd_stream_id);
794 data_hdr.data_size = htobe32(data_size);
1d4dfdef 795 data_hdr.padding_size = htobe32(padding);
39df6d9f
DG
796 /*
797 * Note that net_seq_num below is assigned with the *current* value of
798 * next_net_seq_num and only after that the next_net_seq_num will be
799 * increment. This is why when issuing a command on the relayd using
800 * this next value, 1 should always be substracted in order to compare
801 * the last seen sequence number on the relayd side to the last sent.
802 */
3604f373 803 data_hdr.net_seq_num = htobe64(stream->next_net_seq_num);
00e2e675
DG
804 /* Other fields are zeroed previously */
805
806 ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr,
807 sizeof(data_hdr));
808 if (ret < 0) {
809 goto error;
810 }
811
3604f373
DG
812 ++stream->next_net_seq_num;
813
00e2e675 814 /* Set to go on data socket */
6151a90f 815 outfd = relayd->data_sock.sock.fd;
00e2e675
DG
816 }
817
818error:
819 return outfd;
820}
821
3bd1e081 822/*
ffe60014
DG
823 * Allocate and return a new lttng_consumer_channel object using the given key
824 * to initialize the hash table node.
825 *
826 * On error, return NULL.
3bd1e081 827 */
886224ff 828struct lttng_consumer_channel *consumer_allocate_channel(uint64_t key,
ffe60014
DG
829 uint64_t session_id,
830 const char *pathname,
831 const char *name,
832 uid_t uid,
833 gid_t gid,
57a269f2 834 uint64_t relayd_id,
1624d5b7
JD
835 enum lttng_event_output output,
836 uint64_t tracefile_size,
2bba9e53 837 uint64_t tracefile_count,
1950109e 838 uint64_t session_id_per_pid,
2bba9e53 839 unsigned int monitor)
3bd1e081
MD
840{
841 struct lttng_consumer_channel *channel;
3bd1e081 842
276b26d1 843 channel = zmalloc(sizeof(*channel));
3bd1e081 844 if (channel == NULL) {
7a57cf92 845 PERROR("malloc struct lttng_consumer_channel");
3bd1e081
MD
846 goto end;
847 }
ffe60014
DG
848
849 channel->key = key;
3bd1e081 850 channel->refcount = 0;
ffe60014 851 channel->session_id = session_id;
1950109e 852 channel->session_id_per_pid = session_id_per_pid;
ffe60014
DG
853 channel->uid = uid;
854 channel->gid = gid;
855 channel->relayd_id = relayd_id;
856 channel->output = output;
1624d5b7
JD
857 channel->tracefile_size = tracefile_size;
858 channel->tracefile_count = tracefile_count;
2bba9e53 859 channel->monitor = monitor;
a9838785 860 pthread_mutex_init(&channel->lock, NULL);
ec6ea7d0 861 pthread_mutex_init(&channel->timer_lock, NULL);
ffe60014 862
07b86b52
JD
863 /*
864 * In monitor mode, the streams associated with the channel will be put in
865 * a special list ONLY owned by this channel. So, the refcount is set to 1
866 * here meaning that the channel itself has streams that are referenced.
867 *
868 * On a channel deletion, once the channel is no longer visible, the
869 * refcount is decremented and checked for a zero value to delete it. With
870 * streams in no monitor mode, it will now be safe to destroy the channel.
871 */
872 if (!channel->monitor) {
873 channel->refcount = 1;
874 }
875
ffe60014
DG
876 strncpy(channel->pathname, pathname, sizeof(channel->pathname));
877 channel->pathname[sizeof(channel->pathname) - 1] = '\0';
878
879 strncpy(channel->name, name, sizeof(channel->name));
880 channel->name[sizeof(channel->name) - 1] = '\0';
881
d88aee68 882 lttng_ht_node_init_u64(&channel->node, channel->key);
d8ef542d
MD
883
884 channel->wait_fd = -1;
885
ffe60014
DG
886 CDS_INIT_LIST_HEAD(&channel->streams.head);
887
d88aee68 888 DBG("Allocated channel (key %" PRIu64 ")", channel->key)
3bd1e081 889
3bd1e081
MD
890end:
891 return channel;
892}
893
894/*
895 * Add a channel to the global list protected by a mutex.
821fffb2
DG
896 *
897 * On success 0 is returned else a negative value.
3bd1e081 898 */
d8ef542d
MD
899int consumer_add_channel(struct lttng_consumer_channel *channel,
900 struct lttng_consumer_local_data *ctx)
3bd1e081 901{
ffe60014 902 int ret = 0;
d88aee68 903 struct lttng_ht_node_u64 *node;
c77fc10a
DG
904 struct lttng_ht_iter iter;
905
3bd1e081 906 pthread_mutex_lock(&consumer_data.lock);
a9838785 907 pthread_mutex_lock(&channel->lock);
ec6ea7d0 908 pthread_mutex_lock(&channel->timer_lock);
6065ceec 909 rcu_read_lock();
c77fc10a 910
7972aab2 911 lttng_ht_lookup(consumer_data.channel_ht, &channel->key, &iter);
d88aee68 912 node = lttng_ht_iter_get_node_u64(&iter);
c77fc10a
DG
913 if (node != NULL) {
914 /* Channel already exist. Ignore the insertion */
d88aee68
DG
915 ERR("Consumer add channel key %" PRIu64 " already exists!",
916 channel->key);
821fffb2 917 ret = -EEXIST;
c77fc10a
DG
918 goto end;
919 }
920
d88aee68 921 lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node);
c77fc10a
DG
922
923end:
6065ceec 924 rcu_read_unlock();
ec6ea7d0 925 pthread_mutex_unlock(&channel->timer_lock);
a9838785 926 pthread_mutex_unlock(&channel->lock);
3bd1e081 927 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 928
d8ef542d 929 if (!ret && channel->wait_fd != -1 &&
10a50311 930 channel->type == CONSUMER_CHANNEL_TYPE_DATA) {
a0cbdd2e 931 notify_channel_pipe(ctx, channel, -1, CONSUMER_CHANNEL_ADD);
d8ef542d 932 }
ffe60014 933 return ret;
3bd1e081
MD
934}
935
936/*
937 * Allocate the pollfd structure and the local view of the out fds to avoid
938 * doing a lookup in the linked list and concurrency issues when writing is
939 * needed. Called with consumer_data.lock held.
940 *
941 * Returns the number of fds in the structures.
942 */
ffe60014
DG
943static int update_poll_array(struct lttng_consumer_local_data *ctx,
944 struct pollfd **pollfd, struct lttng_consumer_stream **local_stream,
945 struct lttng_ht *ht)
3bd1e081 946{
3bd1e081 947 int i = 0;
e4421fec
DG
948 struct lttng_ht_iter iter;
949 struct lttng_consumer_stream *stream;
3bd1e081 950
ffe60014
DG
951 assert(ctx);
952 assert(ht);
953 assert(pollfd);
954 assert(local_stream);
955
3bd1e081 956 DBG("Updating poll fd array");
481d6c57 957 rcu_read_lock();
43c34bc3 958 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
8994307f
DG
959 /*
960 * Only active streams with an active end point can be added to the
961 * poll set and local stream storage of the thread.
962 *
963 * There is a potential race here for endpoint_status to be updated
964 * just after the check. However, this is OK since the stream(s) will
965 * be deleted once the thread is notified that the end point state has
966 * changed where this function will be called back again.
967 */
968 if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM ||
79d4ffb7 969 stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) {
3bd1e081
MD
970 continue;
971 }
7972aab2
DG
972 /*
973 * This clobbers way too much the debug output. Uncomment that if you
974 * need it for debugging purposes.
975 *
976 * DBG("Active FD %d", stream->wait_fd);
977 */
e4421fec 978 (*pollfd)[i].fd = stream->wait_fd;
3bd1e081 979 (*pollfd)[i].events = POLLIN | POLLPRI;
e4421fec 980 local_stream[i] = stream;
3bd1e081
MD
981 i++;
982 }
481d6c57 983 rcu_read_unlock();
3bd1e081
MD
984
985 /*
50f8ae69 986 * Insert the consumer_data_pipe at the end of the array and don't
3bd1e081
MD
987 * increment i so nb_fd is the number of real FD.
988 */
acdb9057 989 (*pollfd)[i].fd = lttng_pipe_get_readfd(ctx->consumer_data_pipe);
509bb1cf 990 (*pollfd)[i].events = POLLIN | POLLPRI;
3bd1e081
MD
991 return i;
992}
993
994/*
995 * Poll on the should_quit pipe and the command socket return -1 on error and
996 * should exit, 0 if data is available on the command socket
997 */
998int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll)
999{
1000 int num_rdy;
1001
88f2b785 1002restart:
3bd1e081
MD
1003 num_rdy = poll(consumer_sockpoll, 2, -1);
1004 if (num_rdy == -1) {
88f2b785
MD
1005 /*
1006 * Restart interrupted system call.
1007 */
1008 if (errno == EINTR) {
1009 goto restart;
1010 }
7a57cf92 1011 PERROR("Poll error");
3bd1e081
MD
1012 goto exit;
1013 }
509bb1cf 1014 if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) {
3bd1e081
MD
1015 DBG("consumer_should_quit wake up");
1016 goto exit;
1017 }
1018 return 0;
1019
1020exit:
1021 return -1;
1022}
1023
1024/*
1025 * Set the error socket.
1026 */
ffe60014
DG
1027void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx,
1028 int sock)
3bd1e081
MD
1029{
1030 ctx->consumer_error_socket = sock;
1031}
1032
1033/*
1034 * Set the command socket path.
1035 */
3bd1e081
MD
1036void lttng_consumer_set_command_sock_path(
1037 struct lttng_consumer_local_data *ctx, char *sock)
1038{
1039 ctx->consumer_command_sock_path = sock;
1040}
1041
1042/*
1043 * Send return code to the session daemon.
1044 * If the socket is not defined, we return 0, it is not a fatal error
1045 */
ffe60014 1046int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd)
3bd1e081
MD
1047{
1048 if (ctx->consumer_error_socket > 0) {
1049 return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd,
1050 sizeof(enum lttcomm_sessiond_command));
1051 }
1052
1053 return 0;
1054}
1055
1056/*
228b5bf7
DG
1057 * Close all the tracefiles and stream fds and MUST be called when all
1058 * instances are destroyed i.e. when all threads were joined and are ended.
3bd1e081
MD
1059 */
1060void lttng_consumer_cleanup(void)
1061{
e4421fec 1062 struct lttng_ht_iter iter;
ffe60014 1063 struct lttng_consumer_channel *channel;
6065ceec
DG
1064
1065 rcu_read_lock();
3bd1e081 1066
ffe60014
DG
1067 cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel,
1068 node.node) {
702b1ea4 1069 consumer_del_channel(channel);
3bd1e081 1070 }
6065ceec
DG
1071
1072 rcu_read_unlock();
d6ce1df2 1073
d6ce1df2 1074 lttng_ht_destroy(consumer_data.channel_ht);
228b5bf7
DG
1075
1076 cleanup_relayd_ht();
1077
d8ef542d
MD
1078 lttng_ht_destroy(consumer_data.stream_per_chan_id_ht);
1079
228b5bf7
DG
1080 /*
1081 * This HT contains streams that are freed by either the metadata thread or
1082 * the data thread so we do *nothing* on the hash table and simply destroy
1083 * it.
1084 */
1085 lttng_ht_destroy(consumer_data.stream_list_ht);
3bd1e081
MD
1086}
1087
1088/*
1089 * Called from signal handler.
1090 */
1091void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx)
1092{
1093 int ret;
1094 consumer_quit = 1;
6f94560a
MD
1095 do {
1096 ret = write(ctx->consumer_should_quit[1], "4", 1);
1097 } while (ret < 0 && errno == EINTR);
4cec016f 1098 if (ret < 0 || ret != 1) {
7a57cf92 1099 PERROR("write consumer quit");
3bd1e081 1100 }
ab1027f4
DG
1101
1102 DBG("Consumer flag that it should quit");
3bd1e081
MD
1103}
1104
00e2e675
DG
1105void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream,
1106 off_t orig_offset)
3bd1e081
MD
1107{
1108 int outfd = stream->out_fd;
1109
1110 /*
1111 * This does a blocking write-and-wait on any page that belongs to the
1112 * subbuffer prior to the one we just wrote.
1113 * Don't care about error values, as these are just hints and ways to
1114 * limit the amount of page cache used.
1115 */
ffe60014 1116 if (orig_offset < stream->max_sb_size) {
3bd1e081
MD
1117 return;
1118 }
ffe60014
DG
1119 lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size,
1120 stream->max_sb_size,
3bd1e081
MD
1121 SYNC_FILE_RANGE_WAIT_BEFORE
1122 | SYNC_FILE_RANGE_WRITE
1123 | SYNC_FILE_RANGE_WAIT_AFTER);
1124 /*
1125 * Give hints to the kernel about how we access the file:
1126 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1127 * we write it.
1128 *
1129 * We need to call fadvise again after the file grows because the
1130 * kernel does not seem to apply fadvise to non-existing parts of the
1131 * file.
1132 *
1133 * Call fadvise _after_ having waited for the page writeback to
1134 * complete because the dirty page writeback semantic is not well
1135 * defined. So it can be expected to lead to lower throughput in
1136 * streaming.
1137 */
ffe60014
DG
1138 posix_fadvise(outfd, orig_offset - stream->max_sb_size,
1139 stream->max_sb_size, POSIX_FADV_DONTNEED);
3bd1e081
MD
1140}
1141
1142/*
1143 * Initialise the necessary environnement :
1144 * - create a new context
1145 * - create the poll_pipe
1146 * - create the should_quit pipe (for signal handler)
1147 * - create the thread pipe (for splice)
1148 *
1149 * Takes a function pointer as argument, this function is called when data is
1150 * available on a buffer. This function is responsible to do the
1151 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1152 * buffer configuration and then kernctl_put_next_subbuf at the end.
1153 *
1154 * Returns a pointer to the new context or NULL on error.
1155 */
1156struct lttng_consumer_local_data *lttng_consumer_create(
1157 enum lttng_consumer_type type,
4078b776 1158 ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream,
d41f73b7 1159 struct lttng_consumer_local_data *ctx),
3bd1e081
MD
1160 int (*recv_channel)(struct lttng_consumer_channel *channel),
1161 int (*recv_stream)(struct lttng_consumer_stream *stream),
30319bcb 1162 int (*update_stream)(uint64_t stream_key, uint32_t state))
3bd1e081 1163{
d8ef542d 1164 int ret;
3bd1e081
MD
1165 struct lttng_consumer_local_data *ctx;
1166
1167 assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN ||
1168 consumer_data.type == type);
1169 consumer_data.type = type;
1170
effcf122 1171 ctx = zmalloc(sizeof(struct lttng_consumer_local_data));
3bd1e081 1172 if (ctx == NULL) {
7a57cf92 1173 PERROR("allocating context");
3bd1e081
MD
1174 goto error;
1175 }
1176
1177 ctx->consumer_error_socket = -1;
331744e3 1178 ctx->consumer_metadata_socket = -1;
75d83e50 1179 pthread_mutex_init(&ctx->metadata_socket_lock, NULL);
3bd1e081
MD
1180 /* assign the callbacks */
1181 ctx->on_buffer_ready = buffer_ready;
1182 ctx->on_recv_channel = recv_channel;
1183 ctx->on_recv_stream = recv_stream;
1184 ctx->on_update_stream = update_stream;
1185
acdb9057
DG
1186 ctx->consumer_data_pipe = lttng_pipe_open(0);
1187 if (!ctx->consumer_data_pipe) {
3bd1e081
MD
1188 goto error_poll_pipe;
1189 }
1190
1191 ret = pipe(ctx->consumer_should_quit);
1192 if (ret < 0) {
7a57cf92 1193 PERROR("Error creating recv pipe");
3bd1e081
MD
1194 goto error_quit_pipe;
1195 }
1196
1197 ret = pipe(ctx->consumer_thread_pipe);
1198 if (ret < 0) {
7a57cf92 1199 PERROR("Error creating thread pipe");
3bd1e081
MD
1200 goto error_thread_pipe;
1201 }
1202
d8ef542d
MD
1203 ret = pipe(ctx->consumer_channel_pipe);
1204 if (ret < 0) {
1205 PERROR("Error creating channel pipe");
1206 goto error_channel_pipe;
1207 }
1208
13886d2d
DG
1209 ctx->consumer_metadata_pipe = lttng_pipe_open(0);
1210 if (!ctx->consumer_metadata_pipe) {
fb3a43a9
DG
1211 goto error_metadata_pipe;
1212 }
3bd1e081 1213
fb3a43a9
DG
1214 ret = utils_create_pipe(ctx->consumer_splice_metadata_pipe);
1215 if (ret < 0) {
1216 goto error_splice_pipe;
1217 }
1218
1219 return ctx;
3bd1e081 1220
fb3a43a9 1221error_splice_pipe:
13886d2d 1222 lttng_pipe_destroy(ctx->consumer_metadata_pipe);
fb3a43a9 1223error_metadata_pipe:
d8ef542d
MD
1224 utils_close_pipe(ctx->consumer_channel_pipe);
1225error_channel_pipe:
fb3a43a9 1226 utils_close_pipe(ctx->consumer_thread_pipe);
3bd1e081 1227error_thread_pipe:
d8ef542d 1228 utils_close_pipe(ctx->consumer_should_quit);
3bd1e081 1229error_quit_pipe:
acdb9057 1230 lttng_pipe_destroy(ctx->consumer_data_pipe);
3bd1e081
MD
1231error_poll_pipe:
1232 free(ctx);
1233error:
1234 return NULL;
1235}
1236
1237/*
1238 * Close all fds associated with the instance and free the context.
1239 */
1240void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx)
1241{
4c462e79
MD
1242 int ret;
1243
ab1027f4
DG
1244 DBG("Consumer destroying it. Closing everything.");
1245
4c462e79
MD
1246 ret = close(ctx->consumer_error_socket);
1247 if (ret) {
1248 PERROR("close");
1249 }
331744e3
JD
1250 ret = close(ctx->consumer_metadata_socket);
1251 if (ret) {
1252 PERROR("close");
1253 }
d8ef542d
MD
1254 utils_close_pipe(ctx->consumer_thread_pipe);
1255 utils_close_pipe(ctx->consumer_channel_pipe);
acdb9057 1256 lttng_pipe_destroy(ctx->consumer_data_pipe);
13886d2d 1257 lttng_pipe_destroy(ctx->consumer_metadata_pipe);
d8ef542d 1258 utils_close_pipe(ctx->consumer_should_quit);
fb3a43a9
DG
1259 utils_close_pipe(ctx->consumer_splice_metadata_pipe);
1260
3bd1e081
MD
1261 unlink(ctx->consumer_command_sock_path);
1262 free(ctx);
1263}
1264
6197aea7
DG
1265/*
1266 * Write the metadata stream id on the specified file descriptor.
1267 */
1268static int write_relayd_metadata_id(int fd,
1269 struct lttng_consumer_stream *stream,
ffe60014 1270 struct consumer_relayd_sock_pair *relayd, unsigned long padding)
6197aea7
DG
1271{
1272 int ret;
1d4dfdef 1273 struct lttcomm_relayd_metadata_payload hdr;
6197aea7 1274
1d4dfdef
DG
1275 hdr.stream_id = htobe64(stream->relayd_stream_id);
1276 hdr.padding_size = htobe32(padding);
6197aea7 1277 do {
1d4dfdef 1278 ret = write(fd, (void *) &hdr, sizeof(hdr));
6197aea7 1279 } while (ret < 0 && errno == EINTR);
4cec016f 1280 if (ret < 0 || ret != sizeof(hdr)) {
d7b75ec8
DG
1281 /*
1282 * This error means that the fd's end is closed so ignore the perror
1283 * not to clubber the error output since this can happen in a normal
1284 * code path.
1285 */
1286 if (errno != EPIPE) {
1287 PERROR("write metadata stream id");
1288 }
1289 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno);
534d2592
DG
1290 /*
1291 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1292 * handle writting the missing part so report that as an error and
1293 * don't lie to the caller.
1294 */
1295 ret = -1;
6197aea7
DG
1296 goto end;
1297 }
1d4dfdef
DG
1298 DBG("Metadata stream id %" PRIu64 " with padding %lu written before data",
1299 stream->relayd_stream_id, padding);
6197aea7
DG
1300
1301end:
1302 return ret;
1303}
1304
3bd1e081 1305/*
09e26845
DG
1306 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1307 * core function for writing trace buffers to either the local filesystem or
1308 * the network.
1309 *
79d4ffb7
DG
1310 * It must be called with the stream lock held.
1311 *
09e26845 1312 * Careful review MUST be put if any changes occur!
3bd1e081
MD
1313 *
1314 * Returns the number of bytes written
1315 */
4078b776 1316ssize_t lttng_consumer_on_read_subbuffer_mmap(
3bd1e081 1317 struct lttng_consumer_local_data *ctx,
1d4dfdef
DG
1318 struct lttng_consumer_stream *stream, unsigned long len,
1319 unsigned long padding)
3bd1e081 1320{
f02e1e8a 1321 unsigned long mmap_offset;
ffe60014 1322 void *mmap_base;
f02e1e8a
DG
1323 ssize_t ret = 0, written = 0;
1324 off_t orig_offset = stream->out_fd_offset;
1325 /* Default is on the disk */
1326 int outfd = stream->out_fd;
f02e1e8a 1327 struct consumer_relayd_sock_pair *relayd = NULL;
8994307f 1328 unsigned int relayd_hang_up = 0;
f02e1e8a
DG
1329
1330 /* RCU lock for the relayd pointer */
1331 rcu_read_lock();
1332
1333 /* Flag that the current stream if set for network streaming. */
da009f2c 1334 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1335 relayd = consumer_find_relayd(stream->net_seq_idx);
1336 if (relayd == NULL) {
56591bac 1337 ret = -EPIPE;
f02e1e8a
DG
1338 goto end;
1339 }
1340 }
1341
1342 /* get the offset inside the fd to mmap */
3bd1e081
MD
1343 switch (consumer_data.type) {
1344 case LTTNG_CONSUMER_KERNEL:
ffe60014 1345 mmap_base = stream->mmap_base;
f02e1e8a 1346 ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset);
56591bac
MD
1347 if (ret != 0) {
1348 PERROR("tracer ctl get_mmap_read_offset");
1349 written = -errno;
1350 goto end;
1351 }
f02e1e8a 1352 break;
7753dea8
MD
1353 case LTTNG_CONSUMER32_UST:
1354 case LTTNG_CONSUMER64_UST:
ffe60014
DG
1355 mmap_base = lttng_ustctl_get_mmap_base(stream);
1356 if (!mmap_base) {
1357 ERR("read mmap get mmap base for stream %s", stream->name);
56591bac 1358 written = -EPERM;
ffe60014
DG
1359 goto end;
1360 }
1361 ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset);
56591bac
MD
1362 if (ret != 0) {
1363 PERROR("tracer ctl get_mmap_read_offset");
1364 written = ret;
1365 goto end;
1366 }
f02e1e8a 1367 break;
3bd1e081
MD
1368 default:
1369 ERR("Unknown consumer_data type");
1370 assert(0);
1371 }
b9182dd9 1372
f02e1e8a
DG
1373 /* Handle stream on the relayd if the output is on the network */
1374 if (relayd) {
1375 unsigned long netlen = len;
1376
1377 /*
1378 * Lock the control socket for the complete duration of the function
1379 * since from this point on we will use the socket.
1380 */
1381 if (stream->metadata_flag) {
1382 /* Metadata requires the control socket. */
1383 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1d4dfdef 1384 netlen += sizeof(struct lttcomm_relayd_metadata_payload);
f02e1e8a
DG
1385 }
1386
1d4dfdef 1387 ret = write_relayd_stream_header(stream, netlen, padding, relayd);
f02e1e8a
DG
1388 if (ret >= 0) {
1389 /* Use the returned socket. */
1390 outfd = ret;
1391
1392 /* Write metadata stream id before payload */
1393 if (stream->metadata_flag) {
1d4dfdef 1394 ret = write_relayd_metadata_id(outfd, stream, relayd, padding);
f02e1e8a 1395 if (ret < 0) {
f02e1e8a 1396 written = ret;
8994307f
DG
1397 /* Socket operation failed. We consider the relayd dead */
1398 if (ret == -EPIPE || ret == -EINVAL) {
1399 relayd_hang_up = 1;
1400 goto write_error;
1401 }
f02e1e8a
DG
1402 goto end;
1403 }
f02e1e8a 1404 }
8994307f
DG
1405 } else {
1406 /* Socket operation failed. We consider the relayd dead */
1407 if (ret == -EPIPE || ret == -EINVAL) {
1408 relayd_hang_up = 1;
1409 goto write_error;
1410 }
1411 /* Else, use the default set before which is the filesystem. */
f02e1e8a 1412 }
1d4dfdef
DG
1413 } else {
1414 /* No streaming, we have to set the len with the full padding */
1415 len += padding;
1624d5b7
JD
1416
1417 /*
1418 * Check if we need to change the tracefile before writing the packet.
1419 */
1420 if (stream->chan->tracefile_size > 0 &&
1421 (stream->tracefile_size_current + len) >
1422 stream->chan->tracefile_size) {
fe4477ee
JD
1423 ret = utils_rotate_stream_file(stream->chan->pathname,
1424 stream->name, stream->chan->tracefile_size,
1425 stream->chan->tracefile_count, stream->uid, stream->gid,
1426 stream->out_fd, &(stream->tracefile_count_current));
1624d5b7
JD
1427 if (ret < 0) {
1428 ERR("Rotating output file");
1429 goto end;
1430 }
fe4477ee 1431 outfd = stream->out_fd = ret;
a6976990
DG
1432 /* Reset current size because we just perform a rotation. */
1433 stream->tracefile_size_current = 0;
a1ae300f
JD
1434 stream->out_fd_offset = 0;
1435 orig_offset = 0;
1624d5b7
JD
1436 }
1437 stream->tracefile_size_current += len;
f02e1e8a
DG
1438 }
1439
1440 while (len > 0) {
1441 do {
ffe60014 1442 ret = write(outfd, mmap_base + mmap_offset, len);
f02e1e8a 1443 } while (ret < 0 && errno == EINTR);
1d4dfdef 1444 DBG("Consumer mmap write() ret %zd (len %lu)", ret, len);
f02e1e8a 1445 if (ret < 0) {
c5c45efa
DG
1446 /*
1447 * This is possible if the fd is closed on the other side (outfd)
1448 * or any write problem. It can be verbose a bit for a normal
1449 * execution if for instance the relayd is stopped abruptly. This
1450 * can happen so set this to a DBG statement.
1451 */
1452 DBG("Error in file write mmap");
f02e1e8a 1453 if (written == 0) {
56591bac 1454 written = -errno;
f02e1e8a 1455 }
8994307f
DG
1456 /* Socket operation failed. We consider the relayd dead */
1457 if (errno == EPIPE || errno == EINVAL) {
1458 relayd_hang_up = 1;
1459 goto write_error;
1460 }
f02e1e8a
DG
1461 goto end;
1462 } else if (ret > len) {
77c7c900 1463 PERROR("Error in file write (ret %zd > len %lu)", ret, len);
f02e1e8a
DG
1464 written += ret;
1465 goto end;
1466 } else {
1467 len -= ret;
1468 mmap_offset += ret;
1469 }
f02e1e8a
DG
1470
1471 /* This call is useless on a socket so better save a syscall. */
1472 if (!relayd) {
1473 /* This won't block, but will start writeout asynchronously */
1474 lttng_sync_file_range(outfd, stream->out_fd_offset, ret,
1475 SYNC_FILE_RANGE_WRITE);
1476 stream->out_fd_offset += ret;
1477 }
1478 written += ret;
1479 }
1480 lttng_consumer_sync_trace_file(stream, orig_offset);
1481
8994307f
DG
1482write_error:
1483 /*
1484 * This is a special case that the relayd has closed its socket. Let's
1485 * cleanup the relayd object and all associated streams.
1486 */
1487 if (relayd && relayd_hang_up) {
1488 cleanup_relayd(relayd, ctx);
1489 }
1490
f02e1e8a
DG
1491end:
1492 /* Unlock only if ctrl socket used */
1493 if (relayd && stream->metadata_flag) {
1494 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1495 }
1496
1497 rcu_read_unlock();
1498 return written;
3bd1e081
MD
1499}
1500
1501/*
1502 * Splice the data from the ring buffer to the tracefile.
1503 *
79d4ffb7
DG
1504 * It must be called with the stream lock held.
1505 *
3bd1e081
MD
1506 * Returns the number of bytes spliced.
1507 */
4078b776 1508ssize_t lttng_consumer_on_read_subbuffer_splice(
3bd1e081 1509 struct lttng_consumer_local_data *ctx,
1d4dfdef
DG
1510 struct lttng_consumer_stream *stream, unsigned long len,
1511 unsigned long padding)
3bd1e081 1512{
f02e1e8a
DG
1513 ssize_t ret = 0, written = 0, ret_splice = 0;
1514 loff_t offset = 0;
1515 off_t orig_offset = stream->out_fd_offset;
1516 int fd = stream->wait_fd;
1517 /* Default is on the disk */
1518 int outfd = stream->out_fd;
f02e1e8a 1519 struct consumer_relayd_sock_pair *relayd = NULL;
fb3a43a9 1520 int *splice_pipe;
8994307f 1521 unsigned int relayd_hang_up = 0;
f02e1e8a 1522
3bd1e081
MD
1523 switch (consumer_data.type) {
1524 case LTTNG_CONSUMER_KERNEL:
f02e1e8a 1525 break;
7753dea8
MD
1526 case LTTNG_CONSUMER32_UST:
1527 case LTTNG_CONSUMER64_UST:
f02e1e8a 1528 /* Not supported for user space tracing */
3bd1e081
MD
1529 return -ENOSYS;
1530 default:
1531 ERR("Unknown consumer_data type");
1532 assert(0);
3bd1e081
MD
1533 }
1534
f02e1e8a
DG
1535 /* RCU lock for the relayd pointer */
1536 rcu_read_lock();
1537
1538 /* Flag that the current stream if set for network streaming. */
da009f2c 1539 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1540 relayd = consumer_find_relayd(stream->net_seq_idx);
1541 if (relayd == NULL) {
56591bac 1542 ret = -EPIPE;
f02e1e8a
DG
1543 goto end;
1544 }
1545 }
1546
fb3a43a9
DG
1547 /*
1548 * Choose right pipe for splice. Metadata and trace data are handled by
1549 * different threads hence the use of two pipes in order not to race or
1550 * corrupt the written data.
1551 */
1552 if (stream->metadata_flag) {
1553 splice_pipe = ctx->consumer_splice_metadata_pipe;
1554 } else {
1555 splice_pipe = ctx->consumer_thread_pipe;
1556 }
1557
f02e1e8a 1558 /* Write metadata stream id before payload */
1d4dfdef
DG
1559 if (relayd) {
1560 int total_len = len;
f02e1e8a 1561
1d4dfdef
DG
1562 if (stream->metadata_flag) {
1563 /*
1564 * Lock the control socket for the complete duration of the function
1565 * since from this point on we will use the socket.
1566 */
1567 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1568
1569 ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd,
1570 padding);
1571 if (ret < 0) {
1572 written = ret;
8994307f
DG
1573 /* Socket operation failed. We consider the relayd dead */
1574 if (ret == -EBADF) {
1575 WARN("Remote relayd disconnected. Stopping");
1576 relayd_hang_up = 1;
1577 goto write_error;
1578 }
1d4dfdef
DG
1579 goto end;
1580 }
1581
1582 total_len += sizeof(struct lttcomm_relayd_metadata_payload);
1583 }
1584
1585 ret = write_relayd_stream_header(stream, total_len, padding, relayd);
1586 if (ret >= 0) {
1587 /* Use the returned socket. */
1588 outfd = ret;
1589 } else {
8994307f
DG
1590 /* Socket operation failed. We consider the relayd dead */
1591 if (ret == -EBADF) {
1592 WARN("Remote relayd disconnected. Stopping");
1593 relayd_hang_up = 1;
1594 goto write_error;
1595 }
f02e1e8a
DG
1596 goto end;
1597 }
1d4dfdef
DG
1598 } else {
1599 /* No streaming, we have to set the len with the full padding */
1600 len += padding;
1624d5b7
JD
1601
1602 /*
1603 * Check if we need to change the tracefile before writing the packet.
1604 */
1605 if (stream->chan->tracefile_size > 0 &&
1606 (stream->tracefile_size_current + len) >
1607 stream->chan->tracefile_size) {
fe4477ee
JD
1608 ret = utils_rotate_stream_file(stream->chan->pathname,
1609 stream->name, stream->chan->tracefile_size,
1610 stream->chan->tracefile_count, stream->uid, stream->gid,
1611 stream->out_fd, &(stream->tracefile_count_current));
1624d5b7
JD
1612 if (ret < 0) {
1613 ERR("Rotating output file");
1614 goto end;
1615 }
fe4477ee 1616 outfd = stream->out_fd = ret;
a6976990
DG
1617 /* Reset current size because we just perform a rotation. */
1618 stream->tracefile_size_current = 0;
a1ae300f
JD
1619 stream->out_fd_offset = 0;
1620 orig_offset = 0;
1624d5b7
JD
1621 }
1622 stream->tracefile_size_current += len;
f02e1e8a
DG
1623 }
1624
1625 while (len > 0) {
1d4dfdef
DG
1626 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1627 (unsigned long)offset, len, fd, splice_pipe[1]);
fb3a43a9 1628 ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len,
f02e1e8a
DG
1629 SPLICE_F_MOVE | SPLICE_F_MORE);
1630 DBG("splice chan to pipe, ret %zd", ret_splice);
1631 if (ret_splice < 0) {
1632 PERROR("Error in relay splice");
1633 if (written == 0) {
1634 written = ret_splice;
1635 }
1636 ret = errno;
1637 goto splice_error;
1638 }
1639
1640 /* Handle stream on the relayd if the output is on the network */
1641 if (relayd) {
1642 if (stream->metadata_flag) {
1d4dfdef
DG
1643 size_t metadata_payload_size =
1644 sizeof(struct lttcomm_relayd_metadata_payload);
1645
f02e1e8a 1646 /* Update counter to fit the spliced data */
1d4dfdef
DG
1647 ret_splice += metadata_payload_size;
1648 len += metadata_payload_size;
f02e1e8a
DG
1649 /*
1650 * We do this so the return value can match the len passed as
1651 * argument to this function.
1652 */
1d4dfdef 1653 written -= metadata_payload_size;
f02e1e8a
DG
1654 }
1655 }
1656
1657 /* Splice data out */
fb3a43a9 1658 ret_splice = splice(splice_pipe[0], NULL, outfd, NULL,
f02e1e8a 1659 ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE);
1d4dfdef 1660 DBG("Consumer splice pipe to file, ret %zd", ret_splice);
f02e1e8a
DG
1661 if (ret_splice < 0) {
1662 PERROR("Error in file splice");
1663 if (written == 0) {
1664 written = ret_splice;
1665 }
8994307f 1666 /* Socket operation failed. We consider the relayd dead */
00c8752b 1667 if (errno == EBADF || errno == EPIPE) {
8994307f
DG
1668 WARN("Remote relayd disconnected. Stopping");
1669 relayd_hang_up = 1;
1670 goto write_error;
1671 }
f02e1e8a
DG
1672 ret = errno;
1673 goto splice_error;
1674 } else if (ret_splice > len) {
1675 errno = EINVAL;
1676 PERROR("Wrote more data than requested %zd (len: %lu)",
1677 ret_splice, len);
1678 written += ret_splice;
1679 ret = errno;
1680 goto splice_error;
1681 }
1682 len -= ret_splice;
1683
1684 /* This call is useless on a socket so better save a syscall. */
1685 if (!relayd) {
1686 /* This won't block, but will start writeout asynchronously */
1687 lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice,
1688 SYNC_FILE_RANGE_WRITE);
1689 stream->out_fd_offset += ret_splice;
1690 }
1691 written += ret_splice;
1692 }
1693 lttng_consumer_sync_trace_file(stream, orig_offset);
1694
1695 ret = ret_splice;
1696
1697 goto end;
1698
8994307f
DG
1699write_error:
1700 /*
1701 * This is a special case that the relayd has closed its socket. Let's
1702 * cleanup the relayd object and all associated streams.
1703 */
1704 if (relayd && relayd_hang_up) {
1705 cleanup_relayd(relayd, ctx);
1706 /* Skip splice error so the consumer does not fail */
1707 goto end;
1708 }
1709
f02e1e8a
DG
1710splice_error:
1711 /* send the appropriate error description to sessiond */
1712 switch (ret) {
f02e1e8a 1713 case EINVAL:
f73fabfd 1714 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL);
f02e1e8a
DG
1715 break;
1716 case ENOMEM:
f73fabfd 1717 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM);
f02e1e8a
DG
1718 break;
1719 case ESPIPE:
f73fabfd 1720 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE);
f02e1e8a
DG
1721 break;
1722 }
1723
1724end:
1725 if (relayd && stream->metadata_flag) {
1726 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1727 }
1728
1729 rcu_read_unlock();
1730 return written;
3bd1e081
MD
1731}
1732
1733/*
1734 * Take a snapshot for a specific fd
1735 *
1736 * Returns 0 on success, < 0 on error
1737 */
ffe60014 1738int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream)
3bd1e081
MD
1739{
1740 switch (consumer_data.type) {
1741 case LTTNG_CONSUMER_KERNEL:
ffe60014 1742 return lttng_kconsumer_take_snapshot(stream);
7753dea8
MD
1743 case LTTNG_CONSUMER32_UST:
1744 case LTTNG_CONSUMER64_UST:
ffe60014 1745 return lttng_ustconsumer_take_snapshot(stream);
3bd1e081
MD
1746 default:
1747 ERR("Unknown consumer_data type");
1748 assert(0);
1749 return -ENOSYS;
1750 }
3bd1e081
MD
1751}
1752
1753/*
1754 * Get the produced position
1755 *
1756 * Returns 0 on success, < 0 on error
1757 */
ffe60014 1758int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
3bd1e081
MD
1759 unsigned long *pos)
1760{
1761 switch (consumer_data.type) {
1762 case LTTNG_CONSUMER_KERNEL:
ffe60014 1763 return lttng_kconsumer_get_produced_snapshot(stream, pos);
7753dea8
MD
1764 case LTTNG_CONSUMER32_UST:
1765 case LTTNG_CONSUMER64_UST:
ffe60014 1766 return lttng_ustconsumer_get_produced_snapshot(stream, pos);
3bd1e081
MD
1767 default:
1768 ERR("Unknown consumer_data type");
1769 assert(0);
1770 return -ENOSYS;
1771 }
1772}
1773
1774int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1775 int sock, struct pollfd *consumer_sockpoll)
1776{
1777 switch (consumer_data.type) {
1778 case LTTNG_CONSUMER_KERNEL:
1779 return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
7753dea8
MD
1780 case LTTNG_CONSUMER32_UST:
1781 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
1782 return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
1783 default:
1784 ERR("Unknown consumer_data type");
1785 assert(0);
1786 return -ENOSYS;
1787 }
1788}
1789
43c34bc3
DG
1790/*
1791 * Iterate over all streams of the hashtable and free them properly.
1792 *
1793 * WARNING: *MUST* be used with data stream only.
1794 */
1795static void destroy_data_stream_ht(struct lttng_ht *ht)
1796{
43c34bc3
DG
1797 struct lttng_ht_iter iter;
1798 struct lttng_consumer_stream *stream;
1799
1800 if (ht == NULL) {
1801 return;
1802 }
1803
1804 rcu_read_lock();
1805 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
5c540210
DG
1806 /*
1807 * Ignore return value since we are currently cleaning up so any error
1808 * can't be handled.
1809 */
1810 (void) consumer_del_stream(stream, ht);
43c34bc3
DG
1811 }
1812 rcu_read_unlock();
1813
1814 lttng_ht_destroy(ht);
1815}
1816
fb3a43a9 1817/*
f724d81e 1818 * Iterate over all streams of the hashtable and free them properly.
e316aad5
DG
1819 *
1820 * XXX: Should not be only for metadata stream or else use an other name.
fb3a43a9
DG
1821 */
1822static void destroy_stream_ht(struct lttng_ht *ht)
1823{
fb3a43a9
DG
1824 struct lttng_ht_iter iter;
1825 struct lttng_consumer_stream *stream;
1826
1827 if (ht == NULL) {
1828 return;
1829 }
1830
d09e1200 1831 rcu_read_lock();
58b1f425 1832 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
5c540210
DG
1833 /*
1834 * Ignore return value since we are currently cleaning up so any error
1835 * can't be handled.
1836 */
1837 (void) consumer_del_metadata_stream(stream, ht);
fb3a43a9 1838 }
d09e1200 1839 rcu_read_unlock();
fb3a43a9
DG
1840
1841 lttng_ht_destroy(ht);
1842}
1843
d88aee68
DG
1844void lttng_consumer_close_metadata(void)
1845{
1846 switch (consumer_data.type) {
1847 case LTTNG_CONSUMER_KERNEL:
1848 /*
1849 * The Kernel consumer has a different metadata scheme so we don't
1850 * close anything because the stream will be closed by the session
1851 * daemon.
1852 */
1853 break;
1854 case LTTNG_CONSUMER32_UST:
1855 case LTTNG_CONSUMER64_UST:
1856 /*
1857 * Close all metadata streams. The metadata hash table is passed and
1858 * this call iterates over it by closing all wakeup fd. This is safe
1859 * because at this point we are sure that the metadata producer is
1860 * either dead or blocked.
1861 */
1862 lttng_ustconsumer_close_metadata(metadata_ht);
1863 break;
1864 default:
1865 ERR("Unknown consumer_data type");
1866 assert(0);
1867 }
1868}
1869
fb3a43a9
DG
1870/*
1871 * Clean up a metadata stream and free its memory.
1872 */
e316aad5
DG
1873void consumer_del_metadata_stream(struct lttng_consumer_stream *stream,
1874 struct lttng_ht *ht)
fb3a43a9
DG
1875{
1876 int ret;
e316aad5
DG
1877 struct lttng_ht_iter iter;
1878 struct lttng_consumer_channel *free_chan = NULL;
fb3a43a9
DG
1879 struct consumer_relayd_sock_pair *relayd;
1880
1881 assert(stream);
1882 /*
1883 * This call should NEVER receive regular stream. It must always be
1884 * metadata stream and this is crucial for data structure synchronization.
1885 */
1886 assert(stream->metadata_flag);
1887
e316aad5
DG
1888 DBG3("Consumer delete metadata stream %d", stream->wait_fd);
1889
1890 if (ht == NULL) {
1891 /* Means the stream was allocated but not successfully added */
ffe60014 1892 goto free_stream_rcu;
e316aad5
DG
1893 }
1894
74251bb8 1895 pthread_mutex_lock(&consumer_data.lock);
a9838785 1896 pthread_mutex_lock(&stream->chan->lock);
8994307f
DG
1897 pthread_mutex_lock(&stream->lock);
1898
fb3a43a9
DG
1899 switch (consumer_data.type) {
1900 case LTTNG_CONSUMER_KERNEL:
1901 if (stream->mmap_base != NULL) {
1902 ret = munmap(stream->mmap_base, stream->mmap_len);
1903 if (ret != 0) {
1904 PERROR("munmap metadata stream");
1905 }
1906 }
4c95e622
JD
1907 if (stream->wait_fd >= 0) {
1908 ret = close(stream->wait_fd);
1909 if (ret < 0) {
1910 PERROR("close kernel metadata wait_fd");
1911 }
1912 }
fb3a43a9
DG
1913 break;
1914 case LTTNG_CONSUMER32_UST:
1915 case LTTNG_CONSUMER64_UST:
04ef1097
MD
1916 if (stream->monitor) {
1917 /* close the write-side in close_metadata */
1918 ret = close(stream->ust_metadata_poll_pipe[0]);
1919 if (ret < 0) {
1920 PERROR("Close UST metadata read-side poll pipe");
1921 }
1922 }
fb3a43a9
DG
1923 lttng_ustconsumer_del_stream(stream);
1924 break;
1925 default:
1926 ERR("Unknown consumer_data type");
1927 assert(0);
e316aad5 1928 goto end;
fb3a43a9 1929 }
fb3a43a9 1930
c869f647 1931 rcu_read_lock();
58b1f425 1932 iter.iter.node = &stream->node.node;
c869f647
DG
1933 ret = lttng_ht_del(ht, &iter);
1934 assert(!ret);
ca22feea 1935
d8ef542d
MD
1936 iter.iter.node = &stream->node_channel_id.node;
1937 ret = lttng_ht_del(consumer_data.stream_per_chan_id_ht, &iter);
1938 assert(!ret);
1939
ca22feea
DG
1940 iter.iter.node = &stream->node_session_id.node;
1941 ret = lttng_ht_del(consumer_data.stream_list_ht, &iter);
1942 assert(!ret);
c869f647
DG
1943 rcu_read_unlock();
1944
fb3a43a9
DG
1945 if (stream->out_fd >= 0) {
1946 ret = close(stream->out_fd);
1947 if (ret) {
1948 PERROR("close");
1949 }
1950 }
1951
fb3a43a9
DG
1952 /* Check and cleanup relayd */
1953 rcu_read_lock();
1954 relayd = consumer_find_relayd(stream->net_seq_idx);
1955 if (relayd != NULL) {
1956 uatomic_dec(&relayd->refcount);
1957 assert(uatomic_read(&relayd->refcount) >= 0);
1958
1959 /* Closing streams requires to lock the control socket. */
1960 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1961 ret = relayd_send_close_stream(&relayd->control_sock,
1962 stream->relayd_stream_id, stream->next_net_seq_num - 1);
1963 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1964 if (ret < 0) {
1965 DBG("Unable to close stream on the relayd. Continuing");
1966 /*
1967 * Continue here. There is nothing we can do for the relayd.
1968 * Chances are that the relayd has closed the socket so we just
1969 * continue cleaning up.
1970 */
1971 }
1972
1973 /* Both conditions are met, we destroy the relayd. */
1974 if (uatomic_read(&relayd->refcount) == 0 &&
1975 uatomic_read(&relayd->destroy_flag)) {
51230d70 1976 consumer_destroy_relayd(relayd);
fb3a43a9
DG
1977 }
1978 }
1979 rcu_read_unlock();
1980
1981 /* Atomically decrement channel refcount since other threads can use it. */
f2ad556d 1982 if (!uatomic_sub_return(&stream->chan->refcount, 1)
ffe60014 1983 && !uatomic_read(&stream->chan->nb_init_stream_left)) {
c30aaa51 1984 /* Go for channel deletion! */
e316aad5 1985 free_chan = stream->chan;
fb3a43a9
DG
1986 }
1987
e316aad5 1988end:
73811ecc
DG
1989 /*
1990 * Nullify the stream reference so it is not used after deletion. The
5e41ebe1
MD
1991 * channel lock MUST be acquired before being able to check for
1992 * a NULL pointer value.
73811ecc
DG
1993 */
1994 stream->chan->metadata_stream = NULL;
1995
8994307f 1996 pthread_mutex_unlock(&stream->lock);
a9838785 1997 pthread_mutex_unlock(&stream->chan->lock);
74251bb8 1998 pthread_mutex_unlock(&consumer_data.lock);
e316aad5
DG
1999
2000 if (free_chan) {
2001 consumer_del_channel(free_chan);
2002 }
2003
ffe60014
DG
2004free_stream_rcu:
2005 call_rcu(&stream->node.head, free_stream_rcu);
fb3a43a9
DG
2006}
2007
2008/*
2009 * Action done with the metadata stream when adding it to the consumer internal
2010 * data structures to handle it.
2011 */
5ab66908 2012int consumer_add_metadata_stream(struct lttng_consumer_stream *stream)
fb3a43a9 2013{
5ab66908 2014 struct lttng_ht *ht = metadata_ht;
e316aad5 2015 int ret = 0;
76082088 2016 struct lttng_ht_iter iter;
d88aee68 2017 struct lttng_ht_node_u64 *node;
fb3a43a9 2018
e316aad5
DG
2019 assert(stream);
2020 assert(ht);
2021
d88aee68 2022 DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key);
e316aad5
DG
2023
2024 pthread_mutex_lock(&consumer_data.lock);
a9838785 2025 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 2026 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 2027 pthread_mutex_lock(&stream->lock);
e316aad5 2028
e316aad5
DG
2029 /*
2030 * From here, refcounts are updated so be _careful_ when returning an error
2031 * after this point.
2032 */
2033
fb3a43a9 2034 rcu_read_lock();
76082088
DG
2035
2036 /*
2037 * Lookup the stream just to make sure it does not exist in our internal
2038 * state. This should NEVER happen.
2039 */
d88aee68
DG
2040 lttng_ht_lookup(ht, &stream->key, &iter);
2041 node = lttng_ht_iter_get_node_u64(&iter);
76082088
DG
2042 assert(!node);
2043
e316aad5 2044 /*
ffe60014
DG
2045 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2046 * in terms of destroying the associated channel, because the action that
e316aad5
DG
2047 * causes the count to become 0 also causes a stream to be added. The
2048 * channel deletion will thus be triggered by the following removal of this
2049 * stream.
2050 */
ffe60014 2051 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
2052 /* Increment refcount before decrementing nb_init_stream_left */
2053 cmm_smp_wmb();
ffe60014 2054 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
2055 }
2056
d88aee68 2057 lttng_ht_add_unique_u64(ht, &stream->node);
ca22feea 2058
d8ef542d
MD
2059 lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht,
2060 &stream->node_channel_id);
2061
ca22feea
DG
2062 /*
2063 * Add stream to the stream_list_ht of the consumer data. No need to steal
2064 * the key since the HT does not use it and we allow to add redundant keys
2065 * into this table.
2066 */
d88aee68 2067 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 2068
fb3a43a9 2069 rcu_read_unlock();
e316aad5 2070
2e818a6a 2071 pthread_mutex_unlock(&stream->lock);
a9838785 2072 pthread_mutex_unlock(&stream->chan->lock);
ec6ea7d0 2073 pthread_mutex_unlock(&stream->chan->timer_lock);
e316aad5
DG
2074 pthread_mutex_unlock(&consumer_data.lock);
2075 return ret;
fb3a43a9
DG
2076}
2077
8994307f
DG
2078/*
2079 * Delete data stream that are flagged for deletion (endpoint_status).
2080 */
2081static void validate_endpoint_status_data_stream(void)
2082{
2083 struct lttng_ht_iter iter;
2084 struct lttng_consumer_stream *stream;
2085
2086 DBG("Consumer delete flagged data stream");
2087
2088 rcu_read_lock();
2089 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
2090 /* Validate delete flag of the stream */
79d4ffb7 2091 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2092 continue;
2093 }
2094 /* Delete it right now */
2095 consumer_del_stream(stream, data_ht);
2096 }
2097 rcu_read_unlock();
2098}
2099
2100/*
2101 * Delete metadata stream that are flagged for deletion (endpoint_status).
2102 */
2103static void validate_endpoint_status_metadata_stream(
2104 struct lttng_poll_event *pollset)
2105{
2106 struct lttng_ht_iter iter;
2107 struct lttng_consumer_stream *stream;
2108
2109 DBG("Consumer delete flagged metadata stream");
2110
2111 assert(pollset);
2112
2113 rcu_read_lock();
2114 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
2115 /* Validate delete flag of the stream */
79d4ffb7 2116 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2117 continue;
2118 }
2119 /*
2120 * Remove from pollset so the metadata thread can continue without
2121 * blocking on a deleted stream.
2122 */
2123 lttng_poll_del(pollset, stream->wait_fd);
2124
2125 /* Delete it right now */
2126 consumer_del_metadata_stream(stream, metadata_ht);
2127 }
2128 rcu_read_unlock();
2129}
2130
fb3a43a9
DG
2131/*
2132 * Thread polls on metadata file descriptor and write them on disk or on the
2133 * network.
2134 */
7d980def 2135void *consumer_thread_metadata_poll(void *data)
fb3a43a9
DG
2136{
2137 int ret, i, pollfd;
2138 uint32_t revents, nb_fd;
e316aad5 2139 struct lttng_consumer_stream *stream = NULL;
fb3a43a9 2140 struct lttng_ht_iter iter;
d88aee68 2141 struct lttng_ht_node_u64 *node;
fb3a43a9
DG
2142 struct lttng_poll_event events;
2143 struct lttng_consumer_local_data *ctx = data;
2144 ssize_t len;
2145
2146 rcu_register_thread();
2147
d88aee68 2148 metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
04bb2b64
DG
2149 if (!metadata_ht) {
2150 /* ENOMEM at this point. Better to bail out. */
d8ef542d 2151 goto end_ht;
04bb2b64
DG
2152 }
2153
fb3a43a9
DG
2154 DBG("Thread metadata poll started");
2155
fb3a43a9
DG
2156 /* Size is set to 1 for the consumer_metadata pipe */
2157 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2158 if (ret < 0) {
2159 ERR("Poll set creation failed");
d8ef542d 2160 goto end_poll;
fb3a43a9
DG
2161 }
2162
13886d2d
DG
2163 ret = lttng_poll_add(&events,
2164 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN);
fb3a43a9
DG
2165 if (ret < 0) {
2166 goto end;
2167 }
2168
2169 /* Main loop */
2170 DBG("Metadata main loop started");
2171
2172 while (1) {
fb3a43a9 2173 /* Only the metadata pipe is set */
d21b0d71 2174 if (LTTNG_POLL_GETNB(&events) == 0 && consumer_quit == 1) {
fb3a43a9
DG
2175 goto end;
2176 }
2177
2178restart:
d21b0d71 2179 DBG("Metadata poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events));
fb3a43a9
DG
2180 ret = lttng_poll_wait(&events, -1);
2181 DBG("Metadata event catched in thread");
2182 if (ret < 0) {
2183 if (errno == EINTR) {
e316aad5 2184 ERR("Poll EINTR catched");
fb3a43a9
DG
2185 goto restart;
2186 }
2187 goto error;
2188 }
2189
0d9c5d77
DG
2190 nb_fd = ret;
2191
e316aad5 2192 /* From here, the event is a metadata wait fd */
fb3a43a9
DG
2193 for (i = 0; i < nb_fd; i++) {
2194 revents = LTTNG_POLL_GETEV(&events, i);
2195 pollfd = LTTNG_POLL_GETFD(&events, i);
2196
e316aad5
DG
2197 /* Just don't waste time if no returned events for the fd */
2198 if (!revents) {
2199 continue;
2200 }
2201
13886d2d 2202 if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) {
4adabd61 2203 if (revents & (LPOLLERR | LPOLLHUP )) {
fb3a43a9
DG
2204 DBG("Metadata thread pipe hung up");
2205 /*
2206 * Remove the pipe from the poll set and continue the loop
2207 * since their might be data to consume.
2208 */
13886d2d
DG
2209 lttng_poll_del(&events,
2210 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2211 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
fb3a43a9
DG
2212 continue;
2213 } else if (revents & LPOLLIN) {
13886d2d
DG
2214 ssize_t pipe_len;
2215
2216 pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe,
2217 &stream, sizeof(stream));
2218 if (pipe_len < 0) {
6a00837f 2219 ERR("read metadata stream, ret: %zd", pipe_len);
fb3a43a9 2220 /*
13886d2d 2221 * Continue here to handle the rest of the streams.
fb3a43a9
DG
2222 */
2223 continue;
2224 }
2225
8994307f
DG
2226 /* A NULL stream means that the state has changed. */
2227 if (stream == NULL) {
2228 /* Check for deleted streams. */
2229 validate_endpoint_status_metadata_stream(&events);
3714380f 2230 goto restart;
8994307f
DG
2231 }
2232
fb3a43a9
DG
2233 DBG("Adding metadata stream %d to poll set",
2234 stream->wait_fd);
2235
fb3a43a9
DG
2236 /* Add metadata stream to the global poll events list */
2237 lttng_poll_add(&events, stream->wait_fd,
2238 LPOLLIN | LPOLLPRI);
fb3a43a9
DG
2239 }
2240
e316aad5 2241 /* Handle other stream */
fb3a43a9
DG
2242 continue;
2243 }
2244
d09e1200 2245 rcu_read_lock();
d88aee68
DG
2246 {
2247 uint64_t tmp_id = (uint64_t) pollfd;
2248
2249 lttng_ht_lookup(metadata_ht, &tmp_id, &iter);
2250 }
2251 node = lttng_ht_iter_get_node_u64(&iter);
e316aad5 2252 assert(node);
fb3a43a9
DG
2253
2254 stream = caa_container_of(node, struct lttng_consumer_stream,
58b1f425 2255 node);
fb3a43a9 2256
e316aad5 2257 /* Check for error event */
4adabd61 2258 if (revents & (LPOLLERR | LPOLLHUP)) {
e316aad5 2259 DBG("Metadata fd %d is hup|err.", pollfd);
fb3a43a9
DG
2260 if (!stream->hangup_flush_done
2261 && (consumer_data.type == LTTNG_CONSUMER32_UST
2262 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2263 DBG("Attempting to flush and consume the UST buffers");
2264 lttng_ustconsumer_on_stream_hangup(stream);
2265
2266 /* We just flushed the stream now read it. */
4bb94b75
DG
2267 do {
2268 len = ctx->on_buffer_ready(stream, ctx);
2269 /*
2270 * We don't check the return value here since if we get
2271 * a negative len, it means an error occured thus we
2272 * simply remove it from the poll set and free the
2273 * stream.
2274 */
2275 } while (len > 0);
fb3a43a9
DG
2276 }
2277
fb3a43a9 2278 lttng_poll_del(&events, stream->wait_fd);
e316aad5
DG
2279 /*
2280 * This call update the channel states, closes file descriptors
2281 * and securely free the stream.
2282 */
2283 consumer_del_metadata_stream(stream, metadata_ht);
2284 } else if (revents & (LPOLLIN | LPOLLPRI)) {
2285 /* Get the data out of the metadata file descriptor */
2286 DBG("Metadata available on fd %d", pollfd);
2287 assert(stream->wait_fd == pollfd);
2288
04ef1097
MD
2289 do {
2290 len = ctx->on_buffer_ready(stream, ctx);
2291 /*
2292 * We don't check the return value here since if we get
2293 * a negative len, it means an error occured thus we
2294 * simply remove it from the poll set and free the
2295 * stream.
2296 */
2297 } while (len > 0);
2298
e316aad5 2299 /* It's ok to have an unavailable sub-buffer */
b64403e3 2300 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2301 /* Clean up stream from consumer and free it. */
2302 lttng_poll_del(&events, stream->wait_fd);
2303 consumer_del_metadata_stream(stream, metadata_ht);
e316aad5 2304 }
fb3a43a9 2305 }
e316aad5
DG
2306
2307 /* Release RCU lock for the stream looked up */
d09e1200 2308 rcu_read_unlock();
fb3a43a9
DG
2309 }
2310 }
2311
2312error:
2313end:
2314 DBG("Metadata poll thread exiting");
fb3a43a9 2315
d8ef542d
MD
2316 lttng_poll_clean(&events);
2317end_poll:
04bb2b64 2318 destroy_stream_ht(metadata_ht);
d8ef542d 2319end_ht:
fb3a43a9
DG
2320 rcu_unregister_thread();
2321 return NULL;
2322}
2323
3bd1e081 2324/*
e4421fec 2325 * This thread polls the fds in the set to consume the data and write
3bd1e081
MD
2326 * it to tracefile if necessary.
2327 */
7d980def 2328void *consumer_thread_data_poll(void *data)
3bd1e081
MD
2329{
2330 int num_rdy, num_hup, high_prio, ret, i;
2331 struct pollfd *pollfd = NULL;
2332 /* local view of the streams */
c869f647 2333 struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL;
3bd1e081
MD
2334 /* local view of consumer_data.fds_count */
2335 int nb_fd = 0;
3bd1e081 2336 struct lttng_consumer_local_data *ctx = data;
00e2e675 2337 ssize_t len;
3bd1e081 2338
e7b994a3
DG
2339 rcu_register_thread();
2340
d88aee68 2341 data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
43c34bc3 2342 if (data_ht == NULL) {
04bb2b64 2343 /* ENOMEM at this point. Better to bail out. */
43c34bc3
DG
2344 goto end;
2345 }
2346
4df6c8cb
MD
2347 local_stream = zmalloc(sizeof(struct lttng_consumer_stream *));
2348 if (local_stream == NULL) {
2349 PERROR("local_stream malloc");
2350 goto end;
2351 }
3bd1e081
MD
2352
2353 while (1) {
2354 high_prio = 0;
2355 num_hup = 0;
2356
2357 /*
e4421fec 2358 * the fds set has been updated, we need to update our
3bd1e081
MD
2359 * local array as well
2360 */
2361 pthread_mutex_lock(&consumer_data.lock);
2362 if (consumer_data.need_update) {
0e428499
DG
2363 free(pollfd);
2364 pollfd = NULL;
2365
2366 free(local_stream);
2367 local_stream = NULL;
3bd1e081 2368
50f8ae69 2369 /* allocate for all fds + 1 for the consumer_data_pipe */
effcf122 2370 pollfd = zmalloc((consumer_data.stream_count + 1) * sizeof(struct pollfd));
3bd1e081 2371 if (pollfd == NULL) {
7a57cf92 2372 PERROR("pollfd malloc");
3bd1e081
MD
2373 pthread_mutex_unlock(&consumer_data.lock);
2374 goto end;
2375 }
2376
50f8ae69 2377 /* allocate for all fds + 1 for the consumer_data_pipe */
effcf122 2378 local_stream = zmalloc((consumer_data.stream_count + 1) *
747f8642 2379 sizeof(struct lttng_consumer_stream *));
3bd1e081 2380 if (local_stream == NULL) {
7a57cf92 2381 PERROR("local_stream malloc");
3bd1e081
MD
2382 pthread_mutex_unlock(&consumer_data.lock);
2383 goto end;
2384 }
ffe60014 2385 ret = update_poll_array(ctx, &pollfd, local_stream,
43c34bc3 2386 data_ht);
3bd1e081
MD
2387 if (ret < 0) {
2388 ERR("Error in allocating pollfd or local_outfds");
f73fabfd 2389 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2390 pthread_mutex_unlock(&consumer_data.lock);
2391 goto end;
2392 }
2393 nb_fd = ret;
2394 consumer_data.need_update = 0;
2395 }
2396 pthread_mutex_unlock(&consumer_data.lock);
2397
4078b776
MD
2398 /* No FDs and consumer_quit, consumer_cleanup the thread */
2399 if (nb_fd == 0 && consumer_quit == 1) {
2400 goto end;
2401 }
3bd1e081 2402 /* poll on the array of fds */
88f2b785 2403 restart:
3bd1e081 2404 DBG("polling on %d fd", nb_fd + 1);
cb365c03 2405 num_rdy = poll(pollfd, nb_fd + 1, -1);
3bd1e081
MD
2406 DBG("poll num_rdy : %d", num_rdy);
2407 if (num_rdy == -1) {
88f2b785
MD
2408 /*
2409 * Restart interrupted system call.
2410 */
2411 if (errno == EINTR) {
2412 goto restart;
2413 }
7a57cf92 2414 PERROR("Poll error");
f73fabfd 2415 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2416 goto end;
2417 } else if (num_rdy == 0) {
2418 DBG("Polling thread timed out");
2419 goto end;
2420 }
2421
3bd1e081 2422 /*
50f8ae69 2423 * If the consumer_data_pipe triggered poll go directly to the
00e2e675
DG
2424 * beginning of the loop to update the array. We want to prioritize
2425 * array update over low-priority reads.
3bd1e081 2426 */
509bb1cf 2427 if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) {
ab30f567 2428 ssize_t pipe_readlen;
04fdd819 2429
50f8ae69 2430 DBG("consumer_data_pipe wake up");
acdb9057
DG
2431 pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe,
2432 &new_stream, sizeof(new_stream));
23f5f35d 2433 if (pipe_readlen < 0) {
6a00837f 2434 ERR("Consumer data pipe ret %zd", pipe_readlen);
23f5f35d
DG
2435 /* Continue so we can at least handle the current stream(s). */
2436 continue;
2437 }
c869f647
DG
2438
2439 /*
2440 * If the stream is NULL, just ignore it. It's also possible that
2441 * the sessiond poll thread changed the consumer_quit state and is
2442 * waking us up to test it.
2443 */
2444 if (new_stream == NULL) {
8994307f 2445 validate_endpoint_status_data_stream();
c869f647
DG
2446 continue;
2447 }
2448
c869f647 2449 /* Continue to update the local streams and handle prio ones */
3bd1e081
MD
2450 continue;
2451 }
2452
2453 /* Take care of high priority channels first. */
2454 for (i = 0; i < nb_fd; i++) {
9617607b
DG
2455 if (local_stream[i] == NULL) {
2456 continue;
2457 }
fb3a43a9 2458 if (pollfd[i].revents & POLLPRI) {
d41f73b7
MD
2459 DBG("Urgent read on fd %d", pollfd[i].fd);
2460 high_prio = 1;
4078b776 2461 len = ctx->on_buffer_ready(local_stream[i], ctx);
d41f73b7 2462 /* it's ok to have an unavailable sub-buffer */
b64403e3 2463 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2464 /* Clean the stream and free it. */
2465 consumer_del_stream(local_stream[i], data_ht);
9617607b 2466 local_stream[i] = NULL;
4078b776
MD
2467 } else if (len > 0) {
2468 local_stream[i]->data_read = 1;
d41f73b7 2469 }
3bd1e081
MD
2470 }
2471 }
2472
4078b776
MD
2473 /*
2474 * If we read high prio channel in this loop, try again
2475 * for more high prio data.
2476 */
2477 if (high_prio) {
3bd1e081
MD
2478 continue;
2479 }
2480
2481 /* Take care of low priority channels. */
4078b776 2482 for (i = 0; i < nb_fd; i++) {
9617607b
DG
2483 if (local_stream[i] == NULL) {
2484 continue;
2485 }
4078b776
MD
2486 if ((pollfd[i].revents & POLLIN) ||
2487 local_stream[i]->hangup_flush_done) {
4078b776
MD
2488 DBG("Normal read on fd %d", pollfd[i].fd);
2489 len = ctx->on_buffer_ready(local_stream[i], ctx);
2490 /* it's ok to have an unavailable sub-buffer */
b64403e3 2491 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2492 /* Clean the stream and free it. */
2493 consumer_del_stream(local_stream[i], data_ht);
9617607b 2494 local_stream[i] = NULL;
4078b776
MD
2495 } else if (len > 0) {
2496 local_stream[i]->data_read = 1;
2497 }
2498 }
2499 }
2500
2501 /* Handle hangup and errors */
2502 for (i = 0; i < nb_fd; i++) {
9617607b
DG
2503 if (local_stream[i] == NULL) {
2504 continue;
2505 }
4078b776
MD
2506 if (!local_stream[i]->hangup_flush_done
2507 && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL))
2508 && (consumer_data.type == LTTNG_CONSUMER32_UST
2509 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2510 DBG("fd %d is hup|err|nval. Attempting flush and read.",
9617607b 2511 pollfd[i].fd);
4078b776
MD
2512 lttng_ustconsumer_on_stream_hangup(local_stream[i]);
2513 /* Attempt read again, for the data we just flushed. */
2514 local_stream[i]->data_read = 1;
2515 }
2516 /*
2517 * If the poll flag is HUP/ERR/NVAL and we have
2518 * read no data in this pass, we can remove the
2519 * stream from its hash table.
2520 */
2521 if ((pollfd[i].revents & POLLHUP)) {
2522 DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
2523 if (!local_stream[i]->data_read) {
43c34bc3 2524 consumer_del_stream(local_stream[i], data_ht);
9617607b 2525 local_stream[i] = NULL;
4078b776
MD
2526 num_hup++;
2527 }
2528 } else if (pollfd[i].revents & POLLERR) {
2529 ERR("Error returned in polling fd %d.", pollfd[i].fd);
2530 if (!local_stream[i]->data_read) {
43c34bc3 2531 consumer_del_stream(local_stream[i], data_ht);
9617607b 2532 local_stream[i] = NULL;
4078b776
MD
2533 num_hup++;
2534 }
2535 } else if (pollfd[i].revents & POLLNVAL) {
2536 ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
2537 if (!local_stream[i]->data_read) {
43c34bc3 2538 consumer_del_stream(local_stream[i], data_ht);
9617607b 2539 local_stream[i] = NULL;
4078b776 2540 num_hup++;
3bd1e081
MD
2541 }
2542 }
9617607b
DG
2543 if (local_stream[i] != NULL) {
2544 local_stream[i]->data_read = 0;
2545 }
3bd1e081
MD
2546 }
2547 }
2548end:
2549 DBG("polling thread exiting");
0e428499
DG
2550 free(pollfd);
2551 free(local_stream);
fb3a43a9
DG
2552
2553 /*
2554 * Close the write side of the pipe so epoll_wait() in
7d980def
DG
2555 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2556 * read side of the pipe. If we close them both, epoll_wait strangely does
2557 * not return and could create a endless wait period if the pipe is the
2558 * only tracked fd in the poll set. The thread will take care of closing
2559 * the read side.
fb3a43a9 2560 */
13886d2d 2561 (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe);
fb3a43a9 2562
04bb2b64 2563 destroy_data_stream_ht(data_ht);
43c34bc3 2564
e7b994a3 2565 rcu_unregister_thread();
3bd1e081
MD
2566 return NULL;
2567}
2568
d8ef542d
MD
2569/*
2570 * Close wake-up end of each stream belonging to the channel. This will
2571 * allow the poll() on the stream read-side to detect when the
2572 * write-side (application) finally closes them.
2573 */
2574static
2575void consumer_close_channel_streams(struct lttng_consumer_channel *channel)
2576{
2577 struct lttng_ht *ht;
2578 struct lttng_consumer_stream *stream;
2579 struct lttng_ht_iter iter;
2580
2581 ht = consumer_data.stream_per_chan_id_ht;
2582
2583 rcu_read_lock();
2584 cds_lfht_for_each_entry_duplicate(ht->ht,
2585 ht->hash_fct(&channel->key, lttng_ht_seed),
2586 ht->match_fct, &channel->key,
2587 &iter.iter, stream, node_channel_id.node) {
f2ad556d
MD
2588 /*
2589 * Protect against teardown with mutex.
2590 */
2591 pthread_mutex_lock(&stream->lock);
2592 if (cds_lfht_is_node_deleted(&stream->node.node)) {
2593 goto next;
2594 }
d8ef542d
MD
2595 switch (consumer_data.type) {
2596 case LTTNG_CONSUMER_KERNEL:
2597 break;
2598 case LTTNG_CONSUMER32_UST:
2599 case LTTNG_CONSUMER64_UST:
2600 /*
2601 * Note: a mutex is taken internally within
2602 * liblttng-ust-ctl to protect timer wakeup_fd
2603 * use from concurrent close.
2604 */
2605 lttng_ustconsumer_close_stream_wakeup(stream);
2606 break;
2607 default:
2608 ERR("Unknown consumer_data type");
2609 assert(0);
2610 }
f2ad556d
MD
2611 next:
2612 pthread_mutex_unlock(&stream->lock);
d8ef542d
MD
2613 }
2614 rcu_read_unlock();
2615}
2616
2617static void destroy_channel_ht(struct lttng_ht *ht)
2618{
2619 struct lttng_ht_iter iter;
2620 struct lttng_consumer_channel *channel;
2621 int ret;
2622
2623 if (ht == NULL) {
2624 return;
2625 }
2626
2627 rcu_read_lock();
2628 cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) {
2629 ret = lttng_ht_del(ht, &iter);
2630 assert(ret != 0);
2631 }
2632 rcu_read_unlock();
2633
2634 lttng_ht_destroy(ht);
2635}
2636
2637/*
2638 * This thread polls the channel fds to detect when they are being
2639 * closed. It closes all related streams if the channel is detected as
2640 * closed. It is currently only used as a shim layer for UST because the
2641 * consumerd needs to keep the per-stream wakeup end of pipes open for
2642 * periodical flush.
2643 */
2644void *consumer_thread_channel_poll(void *data)
2645{
2646 int ret, i, pollfd;
2647 uint32_t revents, nb_fd;
2648 struct lttng_consumer_channel *chan = NULL;
2649 struct lttng_ht_iter iter;
2650 struct lttng_ht_node_u64 *node;
2651 struct lttng_poll_event events;
2652 struct lttng_consumer_local_data *ctx = data;
2653 struct lttng_ht *channel_ht;
2654
2655 rcu_register_thread();
2656
2657 channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2658 if (!channel_ht) {
2659 /* ENOMEM at this point. Better to bail out. */
2660 goto end_ht;
2661 }
2662
2663 DBG("Thread channel poll started");
2664
2665 /* Size is set to 1 for the consumer_channel pipe */
2666 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2667 if (ret < 0) {
2668 ERR("Poll set creation failed");
2669 goto end_poll;
2670 }
2671
2672 ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN);
2673 if (ret < 0) {
2674 goto end;
2675 }
2676
2677 /* Main loop */
2678 DBG("Channel main loop started");
2679
2680 while (1) {
2681 /* Only the channel pipe is set */
2682 if (LTTNG_POLL_GETNB(&events) == 0 && consumer_quit == 1) {
2683 goto end;
2684 }
2685
2686restart:
2687 DBG("Channel poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events));
2688 ret = lttng_poll_wait(&events, -1);
2689 DBG("Channel event catched in thread");
2690 if (ret < 0) {
2691 if (errno == EINTR) {
2692 ERR("Poll EINTR catched");
2693 goto restart;
2694 }
2695 goto end;
2696 }
2697
2698 nb_fd = ret;
2699
2700 /* From here, the event is a channel wait fd */
2701 for (i = 0; i < nb_fd; i++) {
2702 revents = LTTNG_POLL_GETEV(&events, i);
2703 pollfd = LTTNG_POLL_GETFD(&events, i);
2704
2705 /* Just don't waste time if no returned events for the fd */
2706 if (!revents) {
2707 continue;
2708 }
2709 if (pollfd == ctx->consumer_channel_pipe[0]) {
2710 if (revents & (LPOLLERR | LPOLLHUP)) {
2711 DBG("Channel thread pipe hung up");
2712 /*
2713 * Remove the pipe from the poll set and continue the loop
2714 * since their might be data to consume.
2715 */
2716 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2717 continue;
2718 } else if (revents & LPOLLIN) {
2719 enum consumer_channel_action action;
a0cbdd2e 2720 uint64_t key;
d8ef542d 2721
a0cbdd2e 2722 ret = read_channel_pipe(ctx, &chan, &key, &action);
d8ef542d
MD
2723 if (ret <= 0) {
2724 ERR("Error reading channel pipe");
2725 continue;
2726 }
2727
2728 switch (action) {
2729 case CONSUMER_CHANNEL_ADD:
2730 DBG("Adding channel %d to poll set",
2731 chan->wait_fd);
2732
2733 lttng_ht_node_init_u64(&chan->wait_fd_node,
2734 chan->wait_fd);
c7260a81 2735 rcu_read_lock();
d8ef542d
MD
2736 lttng_ht_add_unique_u64(channel_ht,
2737 &chan->wait_fd_node);
c7260a81 2738 rcu_read_unlock();
d8ef542d
MD
2739 /* Add channel to the global poll events list */
2740 lttng_poll_add(&events, chan->wait_fd,
2741 LPOLLIN | LPOLLPRI);
2742 break;
a0cbdd2e
MD
2743 case CONSUMER_CHANNEL_DEL:
2744 {
f2a444f1
DG
2745 struct lttng_consumer_stream *stream, *stmp;
2746
c7260a81 2747 rcu_read_lock();
a0cbdd2e
MD
2748 chan = consumer_find_channel(key);
2749 if (!chan) {
c7260a81 2750 rcu_read_unlock();
a0cbdd2e
MD
2751 ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key);
2752 break;
2753 }
2754 lttng_poll_del(&events, chan->wait_fd);
f623cc0b 2755 iter.iter.node = &chan->wait_fd_node.node;
a0cbdd2e
MD
2756 ret = lttng_ht_del(channel_ht, &iter);
2757 assert(ret == 0);
2758 consumer_close_channel_streams(chan);
2759
f2a444f1
DG
2760 switch (consumer_data.type) {
2761 case LTTNG_CONSUMER_KERNEL:
2762 break;
2763 case LTTNG_CONSUMER32_UST:
2764 case LTTNG_CONSUMER64_UST:
2765 /* Delete streams that might have been left in the stream list. */
2766 cds_list_for_each_entry_safe(stream, stmp, &chan->streams.head,
2767 send_node) {
2768 cds_list_del(&stream->send_node);
2769 lttng_ustconsumer_del_stream(stream);
2770 uatomic_sub(&stream->chan->refcount, 1);
2771 assert(&chan->refcount);
2772 free(stream);
2773 }
2774 break;
2775 default:
2776 ERR("Unknown consumer_data type");
2777 assert(0);
2778 }
2779
a0cbdd2e
MD
2780 /*
2781 * Release our own refcount. Force channel deletion even if
2782 * streams were not initialized.
2783 */
2784 if (!uatomic_sub_return(&chan->refcount, 1)) {
2785 consumer_del_channel(chan);
2786 }
c7260a81 2787 rcu_read_unlock();
a0cbdd2e
MD
2788 goto restart;
2789 }
d8ef542d
MD
2790 case CONSUMER_CHANNEL_QUIT:
2791 /*
2792 * Remove the pipe from the poll set and continue the loop
2793 * since their might be data to consume.
2794 */
2795 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2796 continue;
2797 default:
2798 ERR("Unknown action");
2799 break;
2800 }
2801 }
2802
2803 /* Handle other stream */
2804 continue;
2805 }
2806
2807 rcu_read_lock();
2808 {
2809 uint64_t tmp_id = (uint64_t) pollfd;
2810
2811 lttng_ht_lookup(channel_ht, &tmp_id, &iter);
2812 }
2813 node = lttng_ht_iter_get_node_u64(&iter);
2814 assert(node);
2815
2816 chan = caa_container_of(node, struct lttng_consumer_channel,
2817 wait_fd_node);
2818
2819 /* Check for error event */
2820 if (revents & (LPOLLERR | LPOLLHUP)) {
2821 DBG("Channel fd %d is hup|err.", pollfd);
2822
2823 lttng_poll_del(&events, chan->wait_fd);
2824 ret = lttng_ht_del(channel_ht, &iter);
2825 assert(ret == 0);
2826 consumer_close_channel_streams(chan);
f2ad556d
MD
2827
2828 /* Release our own refcount */
2829 if (!uatomic_sub_return(&chan->refcount, 1)
2830 && !uatomic_read(&chan->nb_init_stream_left)) {
2831 consumer_del_channel(chan);
2832 }
d8ef542d
MD
2833 }
2834
2835 /* Release RCU lock for the channel looked up */
2836 rcu_read_unlock();
2837 }
2838 }
2839
2840end:
2841 lttng_poll_clean(&events);
2842end_poll:
2843 destroy_channel_ht(channel_ht);
2844end_ht:
2845 DBG("Channel poll thread exiting");
2846 rcu_unregister_thread();
2847 return NULL;
2848}
2849
331744e3
JD
2850static int set_metadata_socket(struct lttng_consumer_local_data *ctx,
2851 struct pollfd *sockpoll, int client_socket)
2852{
2853 int ret;
2854
2855 assert(ctx);
2856 assert(sockpoll);
2857
2858 if (lttng_consumer_poll_socket(sockpoll) < 0) {
2859 ret = -1;
2860 goto error;
2861 }
2862 DBG("Metadata connection on client_socket");
2863
2864 /* Blocking call, waiting for transmission */
2865 ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket);
2866 if (ctx->consumer_metadata_socket < 0) {
2867 WARN("On accept metadata");
2868 ret = -1;
2869 goto error;
2870 }
2871 ret = 0;
2872
2873error:
2874 return ret;
2875}
2876
3bd1e081
MD
2877/*
2878 * This thread listens on the consumerd socket and receives the file
2879 * descriptors from the session daemon.
2880 */
7d980def 2881void *consumer_thread_sessiond_poll(void *data)
3bd1e081 2882{
d96f09c6 2883 int sock = -1, client_socket, ret;
3bd1e081
MD
2884 /*
2885 * structure to poll for incoming data on communication socket avoids
2886 * making blocking sockets.
2887 */
2888 struct pollfd consumer_sockpoll[2];
2889 struct lttng_consumer_local_data *ctx = data;
2890
e7b994a3
DG
2891 rcu_register_thread();
2892
3bd1e081
MD
2893 DBG("Creating command socket %s", ctx->consumer_command_sock_path);
2894 unlink(ctx->consumer_command_sock_path);
2895 client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path);
2896 if (client_socket < 0) {
2897 ERR("Cannot create command socket");
2898 goto end;
2899 }
2900
2901 ret = lttcomm_listen_unix_sock(client_socket);
2902 if (ret < 0) {
2903 goto end;
2904 }
2905
32258573 2906 DBG("Sending ready command to lttng-sessiond");
f73fabfd 2907 ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY);
3bd1e081
MD
2908 /* return < 0 on error, but == 0 is not fatal */
2909 if (ret < 0) {
32258573 2910 ERR("Error sending ready command to lttng-sessiond");
3bd1e081
MD
2911 goto end;
2912 }
2913
3bd1e081
MD
2914 /* prepare the FDs to poll : to client socket and the should_quit pipe */
2915 consumer_sockpoll[0].fd = ctx->consumer_should_quit[0];
2916 consumer_sockpoll[0].events = POLLIN | POLLPRI;
2917 consumer_sockpoll[1].fd = client_socket;
2918 consumer_sockpoll[1].events = POLLIN | POLLPRI;
2919
2920 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
2921 goto end;
2922 }
2923 DBG("Connection on client_socket");
2924
2925 /* Blocking call, waiting for transmission */
2926 sock = lttcomm_accept_unix_sock(client_socket);
534d2592 2927 if (sock < 0) {
3bd1e081
MD
2928 WARN("On accept");
2929 goto end;
2930 }
3bd1e081 2931
331744e3
JD
2932 /*
2933 * Setup metadata socket which is the second socket connection on the
2934 * command unix socket.
2935 */
2936 ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket);
2937 if (ret < 0) {
2938 goto end;
2939 }
2940
d96f09c6
DG
2941 /* This socket is not useful anymore. */
2942 ret = close(client_socket);
2943 if (ret < 0) {
2944 PERROR("close client_socket");
2945 }
2946 client_socket = -1;
2947
3bd1e081
MD
2948 /* update the polling structure to poll on the established socket */
2949 consumer_sockpoll[1].fd = sock;
2950 consumer_sockpoll[1].events = POLLIN | POLLPRI;
2951
2952 while (1) {
2953 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
2954 goto end;
2955 }
2956 DBG("Incoming command on sock");
2957 ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll);
2958 if (ret == -ENOENT) {
2959 DBG("Received STOP command");
2960 goto end;
2961 }
4cbc1a04
DG
2962 if (ret <= 0) {
2963 /*
2964 * This could simply be a session daemon quitting. Don't output
2965 * ERR() here.
2966 */
2967 DBG("Communication interrupted on command socket");
3bd1e081
MD
2968 goto end;
2969 }
2970 if (consumer_quit) {
2971 DBG("consumer_thread_receive_fds received quit from signal");
2972 goto end;
2973 }
ffe60014 2974 DBG("received command on sock");
3bd1e081
MD
2975 }
2976end:
ffe60014 2977 DBG("Consumer thread sessiond poll exiting");
3bd1e081 2978
d88aee68
DG
2979 /*
2980 * Close metadata streams since the producer is the session daemon which
2981 * just died.
2982 *
2983 * NOTE: for now, this only applies to the UST tracer.
2984 */
2985 lttng_consumer_close_metadata();
2986
3bd1e081
MD
2987 /*
2988 * when all fds have hung up, the polling thread
2989 * can exit cleanly
2990 */
2991 consumer_quit = 1;
2992
04fdd819 2993 /*
c869f647 2994 * Notify the data poll thread to poll back again and test the
8994307f 2995 * consumer_quit state that we just set so to quit gracefully.
04fdd819 2996 */
acdb9057 2997 notify_thread_lttng_pipe(ctx->consumer_data_pipe);
c869f647 2998
a0cbdd2e 2999 notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT);
d8ef542d 3000
d96f09c6
DG
3001 /* Cleaning up possibly open sockets. */
3002 if (sock >= 0) {
3003 ret = close(sock);
3004 if (ret < 0) {
3005 PERROR("close sock sessiond poll");
3006 }
3007 }
3008 if (client_socket >= 0) {
38476d24 3009 ret = close(client_socket);
d96f09c6
DG
3010 if (ret < 0) {
3011 PERROR("close client_socket sessiond poll");
3012 }
3013 }
3014
e7b994a3 3015 rcu_unregister_thread();
3bd1e081
MD
3016 return NULL;
3017}
d41f73b7 3018
4078b776 3019ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream,
d41f73b7
MD
3020 struct lttng_consumer_local_data *ctx)
3021{
74251bb8
DG
3022 ssize_t ret;
3023
3024 pthread_mutex_lock(&stream->lock);
3025
d41f73b7
MD
3026 switch (consumer_data.type) {
3027 case LTTNG_CONSUMER_KERNEL:
74251bb8
DG
3028 ret = lttng_kconsumer_read_subbuffer(stream, ctx);
3029 break;
7753dea8
MD
3030 case LTTNG_CONSUMER32_UST:
3031 case LTTNG_CONSUMER64_UST:
74251bb8
DG
3032 ret = lttng_ustconsumer_read_subbuffer(stream, ctx);
3033 break;
d41f73b7
MD
3034 default:
3035 ERR("Unknown consumer_data type");
3036 assert(0);
74251bb8
DG
3037 ret = -ENOSYS;
3038 break;
d41f73b7 3039 }
74251bb8
DG
3040
3041 pthread_mutex_unlock(&stream->lock);
3042 return ret;
d41f73b7
MD
3043}
3044
3045int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream)
3046{
3047 switch (consumer_data.type) {
3048 case LTTNG_CONSUMER_KERNEL:
3049 return lttng_kconsumer_on_recv_stream(stream);
7753dea8
MD
3050 case LTTNG_CONSUMER32_UST:
3051 case LTTNG_CONSUMER64_UST:
d41f73b7
MD
3052 return lttng_ustconsumer_on_recv_stream(stream);
3053 default:
3054 ERR("Unknown consumer_data type");
3055 assert(0);
3056 return -ENOSYS;
3057 }
3058}
e4421fec
DG
3059
3060/*
3061 * Allocate and set consumer data hash tables.
3062 */
3063void lttng_consumer_init(void)
3064{
d88aee68
DG
3065 consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3066 consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3067 consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
d8ef542d 3068 consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
e4421fec 3069}
7735ef9e
DG
3070
3071/*
3072 * Process the ADD_RELAYD command receive by a consumer.
3073 *
3074 * This will create a relayd socket pair and add it to the relayd hash table.
3075 * The caller MUST acquire a RCU read side lock before calling it.
3076 */
da009f2c 3077int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type,
7735ef9e 3078 struct lttng_consumer_local_data *ctx, int sock,
6151a90f 3079 struct pollfd *consumer_sockpoll,
30319bcb 3080 struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id)
7735ef9e 3081{
cd2b09ed 3082 int fd = -1, ret = -1, relayd_created = 0;
f50f23d9 3083 enum lttng_error_code ret_code = LTTNG_OK;
d4298c99 3084 struct consumer_relayd_sock_pair *relayd = NULL;
7735ef9e 3085
6151a90f
JD
3086 assert(ctx);
3087 assert(relayd_sock);
3088
da009f2c 3089 DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx);
7735ef9e
DG
3090
3091 /* Get relayd reference if exists. */
3092 relayd = consumer_find_relayd(net_seq_idx);
3093 if (relayd == NULL) {
da009f2c 3094 assert(sock_type == LTTNG_STREAM_CONTROL);
7735ef9e
DG
3095 /* Not found. Allocate one. */
3096 relayd = consumer_allocate_relayd_sock_pair(net_seq_idx);
3097 if (relayd == NULL) {
0d08d75e 3098 ret = -ENOMEM;
618a6a28
MD
3099 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
3100 goto error;
0d08d75e 3101 } else {
30319bcb 3102 relayd->sessiond_session_id = sessiond_id;
0d08d75e 3103 relayd_created = 1;
7735ef9e 3104 }
0d08d75e
DG
3105
3106 /*
3107 * This code path MUST continue to the consumer send status message to
3108 * we can notify the session daemon and continue our work without
3109 * killing everything.
3110 */
da009f2c
MD
3111 } else {
3112 /*
3113 * relayd key should never be found for control socket.
3114 */
3115 assert(sock_type != LTTNG_STREAM_CONTROL);
0d08d75e
DG
3116 }
3117
3118 /* First send a status message before receiving the fds. */
618a6a28
MD
3119 ret = consumer_send_status_msg(sock, LTTNG_OK);
3120 if (ret < 0) {
0d08d75e 3121 /* Somehow, the session daemon is not responding anymore. */
618a6a28
MD
3122 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3123 goto error_nosignal;
7735ef9e
DG
3124 }
3125
3126 /* Poll on consumer socket. */
3127 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
0d08d75e 3128 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
7735ef9e 3129 ret = -EINTR;
618a6a28 3130 goto error_nosignal;
7735ef9e
DG
3131 }
3132
3133 /* Get relayd socket from session daemon */
3134 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
3135 if (ret != sizeof(fd)) {
7735ef9e 3136 ret = -1;
4028eeb9 3137 fd = -1; /* Just in case it gets set with an invalid value. */
0d08d75e
DG
3138
3139 /*
3140 * Failing to receive FDs might indicate a major problem such as
3141 * reaching a fd limit during the receive where the kernel returns a
3142 * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we
3143 * don't take any chances and stop everything.
3144 *
3145 * XXX: Feature request #558 will fix that and avoid this possible
3146 * issue when reaching the fd limit.
3147 */
3148 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
618a6a28 3149 ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD;
f50f23d9
DG
3150 goto error;
3151 }
3152
7735ef9e
DG
3153 /* Copy socket information and received FD */
3154 switch (sock_type) {
3155 case LTTNG_STREAM_CONTROL:
3156 /* Copy received lttcomm socket */
6151a90f
JD
3157 lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock);
3158 ret = lttcomm_create_sock(&relayd->control_sock.sock);
4028eeb9 3159 /* Handle create_sock error. */
f66c074c 3160 if (ret < 0) {
618a6a28 3161 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
4028eeb9 3162 goto error;
f66c074c 3163 }
da009f2c
MD
3164 /*
3165 * Close the socket created internally by
3166 * lttcomm_create_sock, so we can replace it by the one
3167 * received from sessiond.
3168 */
3169 if (close(relayd->control_sock.sock.fd)) {
3170 PERROR("close");
3171 }
7735ef9e
DG
3172
3173 /* Assign new file descriptor */
6151a90f 3174 relayd->control_sock.sock.fd = fd;
4b29f1ce 3175 fd = -1; /* For error path */
6151a90f
JD
3176 /* Assign version values. */
3177 relayd->control_sock.major = relayd_sock->major;
3178 relayd->control_sock.minor = relayd_sock->minor;
c5b6f4f0
DG
3179
3180 /*
59e71485
DG
3181 * Create a session on the relayd and store the returned id. Lock the
3182 * control socket mutex if the relayd was NOT created before.
c5b6f4f0 3183 */
59e71485
DG
3184 if (!relayd_created) {
3185 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3186 }
c5b6f4f0 3187 ret = relayd_create_session(&relayd->control_sock,
f7079f67 3188 &relayd->relayd_session_id);
59e71485
DG
3189 if (!relayd_created) {
3190 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
3191 }
c5b6f4f0 3192 if (ret < 0) {
ffe60014
DG
3193 /*
3194 * Close all sockets of a relayd object. It will be freed if it was
3195 * created at the error code path or else it will be garbage
3196 * collect.
3197 */
3198 (void) relayd_close(&relayd->control_sock);
3199 (void) relayd_close(&relayd->data_sock);
618a6a28 3200 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
c5b6f4f0
DG
3201 goto error;
3202 }
3203
7735ef9e
DG
3204 break;
3205 case LTTNG_STREAM_DATA:
3206 /* Copy received lttcomm socket */
6151a90f
JD
3207 lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock);
3208 ret = lttcomm_create_sock(&relayd->data_sock.sock);
4028eeb9 3209 /* Handle create_sock error. */
f66c074c 3210 if (ret < 0) {
618a6a28 3211 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
4028eeb9 3212 goto error;
f66c074c 3213 }
da009f2c
MD
3214 /*
3215 * Close the socket created internally by
3216 * lttcomm_create_sock, so we can replace it by the one
3217 * received from sessiond.
3218 */
3219 if (close(relayd->data_sock.sock.fd)) {
3220 PERROR("close");
3221 }
7735ef9e
DG
3222
3223 /* Assign new file descriptor */
6151a90f 3224 relayd->data_sock.sock.fd = fd;
4b29f1ce 3225 fd = -1; /* for eventual error paths */
6151a90f
JD
3226 /* Assign version values. */
3227 relayd->data_sock.major = relayd_sock->major;
3228 relayd->data_sock.minor = relayd_sock->minor;
7735ef9e
DG
3229 break;
3230 default:
3231 ERR("Unknown relayd socket type (%d)", sock_type);
59e71485 3232 ret = -1;
618a6a28 3233 ret_code = LTTCOMM_CONSUMERD_FATAL;
7735ef9e
DG
3234 goto error;
3235 }
3236
d88aee68 3237 DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)",
7735ef9e
DG
3238 sock_type == LTTNG_STREAM_CONTROL ? "control" : "data",
3239 relayd->net_seq_idx, fd);
3240
618a6a28
MD
3241 /* We successfully added the socket. Send status back. */
3242 ret = consumer_send_status_msg(sock, ret_code);
3243 if (ret < 0) {
3244 /* Somehow, the session daemon is not responding anymore. */
3245 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3246 goto error_nosignal;
3247 }
3248
7735ef9e
DG
3249 /*
3250 * Add relayd socket pair to consumer data hashtable. If object already
3251 * exists or on error, the function gracefully returns.
3252 */
d09e1200 3253 add_relayd(relayd);
7735ef9e
DG
3254
3255 /* All good! */
4028eeb9 3256 return 0;
7735ef9e
DG
3257
3258error:
618a6a28
MD
3259 if (consumer_send_status_msg(sock, ret_code) < 0) {
3260 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3261 }
3262
3263error_nosignal:
4028eeb9
DG
3264 /* Close received socket if valid. */
3265 if (fd >= 0) {
3266 if (close(fd)) {
3267 PERROR("close received socket");
3268 }
3269 }
cd2b09ed
DG
3270
3271 if (relayd_created) {
cd2b09ed
DG
3272 free(relayd);
3273 }
3274
7735ef9e
DG
3275 return ret;
3276}
ca22feea 3277
4e9a4686
DG
3278/*
3279 * Try to lock the stream mutex.
3280 *
3281 * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
3282 */
3283static int stream_try_lock(struct lttng_consumer_stream *stream)
3284{
3285 int ret;
3286
3287 assert(stream);
3288
3289 /*
3290 * Try to lock the stream mutex. On failure, we know that the stream is
3291 * being used else where hence there is data still being extracted.
3292 */
3293 ret = pthread_mutex_trylock(&stream->lock);
3294 if (ret) {
3295 /* For both EBUSY and EINVAL error, the mutex is NOT locked. */
3296 ret = 0;
3297 goto end;
3298 }
3299
3300 ret = 1;
3301
3302end:
3303 return ret;
3304}
3305
f7079f67
DG
3306/*
3307 * Search for a relayd associated to the session id and return the reference.
3308 *
3309 * A rcu read side lock MUST be acquire before calling this function and locked
3310 * until the relayd object is no longer necessary.
3311 */
3312static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id)
3313{
3314 struct lttng_ht_iter iter;
f7079f67 3315 struct consumer_relayd_sock_pair *relayd = NULL;
f7079f67
DG
3316
3317 /* Iterate over all relayd since they are indexed by net_seq_idx. */
3318 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
3319 node.node) {
18261bd1
DG
3320 /*
3321 * Check by sessiond id which is unique here where the relayd session
3322 * id might not be when having multiple relayd.
3323 */
3324 if (relayd->sessiond_session_id ==