Fix: data pending race
[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) {
1337 goto end;
1338 }
1339 }
1340
1341 /* get the offset inside the fd to mmap */
3bd1e081
MD
1342 switch (consumer_data.type) {
1343 case LTTNG_CONSUMER_KERNEL:
ffe60014 1344 mmap_base = stream->mmap_base;
f02e1e8a
DG
1345 ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset);
1346 break;
7753dea8
MD
1347 case LTTNG_CONSUMER32_UST:
1348 case LTTNG_CONSUMER64_UST:
ffe60014
DG
1349 mmap_base = lttng_ustctl_get_mmap_base(stream);
1350 if (!mmap_base) {
1351 ERR("read mmap get mmap base for stream %s", stream->name);
1352 written = -1;
1353 goto end;
1354 }
1355 ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset);
331744e3 1356
f02e1e8a 1357 break;
3bd1e081
MD
1358 default:
1359 ERR("Unknown consumer_data type");
1360 assert(0);
1361 }
f02e1e8a
DG
1362 if (ret != 0) {
1363 errno = -ret;
1364 PERROR("tracer ctl get_mmap_read_offset");
1365 written = ret;
1366 goto end;
1367 }
b9182dd9 1368
f02e1e8a
DG
1369 /* Handle stream on the relayd if the output is on the network */
1370 if (relayd) {
1371 unsigned long netlen = len;
1372
1373 /*
1374 * Lock the control socket for the complete duration of the function
1375 * since from this point on we will use the socket.
1376 */
1377 if (stream->metadata_flag) {
1378 /* Metadata requires the control socket. */
1379 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1d4dfdef 1380 netlen += sizeof(struct lttcomm_relayd_metadata_payload);
f02e1e8a
DG
1381 }
1382
1d4dfdef 1383 ret = write_relayd_stream_header(stream, netlen, padding, relayd);
f02e1e8a
DG
1384 if (ret >= 0) {
1385 /* Use the returned socket. */
1386 outfd = ret;
1387
1388 /* Write metadata stream id before payload */
1389 if (stream->metadata_flag) {
1d4dfdef 1390 ret = write_relayd_metadata_id(outfd, stream, relayd, padding);
f02e1e8a 1391 if (ret < 0) {
f02e1e8a 1392 written = ret;
8994307f
DG
1393 /* Socket operation failed. We consider the relayd dead */
1394 if (ret == -EPIPE || ret == -EINVAL) {
1395 relayd_hang_up = 1;
1396 goto write_error;
1397 }
f02e1e8a
DG
1398 goto end;
1399 }
f02e1e8a 1400 }
8994307f
DG
1401 } else {
1402 /* Socket operation failed. We consider the relayd dead */
1403 if (ret == -EPIPE || ret == -EINVAL) {
1404 relayd_hang_up = 1;
1405 goto write_error;
1406 }
1407 /* Else, use the default set before which is the filesystem. */
f02e1e8a 1408 }
1d4dfdef
DG
1409 } else {
1410 /* No streaming, we have to set the len with the full padding */
1411 len += padding;
1624d5b7
JD
1412
1413 /*
1414 * Check if we need to change the tracefile before writing the packet.
1415 */
1416 if (stream->chan->tracefile_size > 0 &&
1417 (stream->tracefile_size_current + len) >
1418 stream->chan->tracefile_size) {
fe4477ee
JD
1419 ret = utils_rotate_stream_file(stream->chan->pathname,
1420 stream->name, stream->chan->tracefile_size,
1421 stream->chan->tracefile_count, stream->uid, stream->gid,
1422 stream->out_fd, &(stream->tracefile_count_current));
1624d5b7
JD
1423 if (ret < 0) {
1424 ERR("Rotating output file");
1425 goto end;
1426 }
fe4477ee 1427 outfd = stream->out_fd = ret;
a6976990
DG
1428 /* Reset current size because we just perform a rotation. */
1429 stream->tracefile_size_current = 0;
1624d5b7
JD
1430 }
1431 stream->tracefile_size_current += len;
f02e1e8a
DG
1432 }
1433
1434 while (len > 0) {
1435 do {
ffe60014 1436 ret = write(outfd, mmap_base + mmap_offset, len);
f02e1e8a 1437 } while (ret < 0 && errno == EINTR);
1d4dfdef 1438 DBG("Consumer mmap write() ret %zd (len %lu)", ret, len);
f02e1e8a 1439 if (ret < 0) {
c5c45efa
DG
1440 /*
1441 * This is possible if the fd is closed on the other side (outfd)
1442 * or any write problem. It can be verbose a bit for a normal
1443 * execution if for instance the relayd is stopped abruptly. This
1444 * can happen so set this to a DBG statement.
1445 */
1446 DBG("Error in file write mmap");
f02e1e8a
DG
1447 if (written == 0) {
1448 written = ret;
1449 }
8994307f
DG
1450 /* Socket operation failed. We consider the relayd dead */
1451 if (errno == EPIPE || errno == EINVAL) {
1452 relayd_hang_up = 1;
1453 goto write_error;
1454 }
f02e1e8a
DG
1455 goto end;
1456 } else if (ret > len) {
77c7c900 1457 PERROR("Error in file write (ret %zd > len %lu)", ret, len);
f02e1e8a
DG
1458 written += ret;
1459 goto end;
1460 } else {
1461 len -= ret;
1462 mmap_offset += ret;
1463 }
f02e1e8a
DG
1464
1465 /* This call is useless on a socket so better save a syscall. */
1466 if (!relayd) {
1467 /* This won't block, but will start writeout asynchronously */
1468 lttng_sync_file_range(outfd, stream->out_fd_offset, ret,
1469 SYNC_FILE_RANGE_WRITE);
1470 stream->out_fd_offset += ret;
1471 }
1472 written += ret;
1473 }
1474 lttng_consumer_sync_trace_file(stream, orig_offset);
1475
8994307f
DG
1476write_error:
1477 /*
1478 * This is a special case that the relayd has closed its socket. Let's
1479 * cleanup the relayd object and all associated streams.
1480 */
1481 if (relayd && relayd_hang_up) {
1482 cleanup_relayd(relayd, ctx);
1483 }
1484
f02e1e8a
DG
1485end:
1486 /* Unlock only if ctrl socket used */
1487 if (relayd && stream->metadata_flag) {
1488 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1489 }
1490
1491 rcu_read_unlock();
1492 return written;
3bd1e081
MD
1493}
1494
1495/*
1496 * Splice the data from the ring buffer to the tracefile.
1497 *
79d4ffb7
DG
1498 * It must be called with the stream lock held.
1499 *
3bd1e081
MD
1500 * Returns the number of bytes spliced.
1501 */
4078b776 1502ssize_t lttng_consumer_on_read_subbuffer_splice(
3bd1e081 1503 struct lttng_consumer_local_data *ctx,
1d4dfdef
DG
1504 struct lttng_consumer_stream *stream, unsigned long len,
1505 unsigned long padding)
3bd1e081 1506{
f02e1e8a
DG
1507 ssize_t ret = 0, written = 0, ret_splice = 0;
1508 loff_t offset = 0;
1509 off_t orig_offset = stream->out_fd_offset;
1510 int fd = stream->wait_fd;
1511 /* Default is on the disk */
1512 int outfd = stream->out_fd;
f02e1e8a 1513 struct consumer_relayd_sock_pair *relayd = NULL;
fb3a43a9 1514 int *splice_pipe;
8994307f 1515 unsigned int relayd_hang_up = 0;
f02e1e8a 1516
3bd1e081
MD
1517 switch (consumer_data.type) {
1518 case LTTNG_CONSUMER_KERNEL:
f02e1e8a 1519 break;
7753dea8
MD
1520 case LTTNG_CONSUMER32_UST:
1521 case LTTNG_CONSUMER64_UST:
f02e1e8a 1522 /* Not supported for user space tracing */
3bd1e081
MD
1523 return -ENOSYS;
1524 default:
1525 ERR("Unknown consumer_data type");
1526 assert(0);
3bd1e081
MD
1527 }
1528
f02e1e8a
DG
1529 /* RCU lock for the relayd pointer */
1530 rcu_read_lock();
1531
1532 /* Flag that the current stream if set for network streaming. */
da009f2c 1533 if (stream->net_seq_idx != (uint64_t) -1ULL) {
f02e1e8a
DG
1534 relayd = consumer_find_relayd(stream->net_seq_idx);
1535 if (relayd == NULL) {
1536 goto end;
1537 }
1538 }
1539
fb3a43a9
DG
1540 /*
1541 * Choose right pipe for splice. Metadata and trace data are handled by
1542 * different threads hence the use of two pipes in order not to race or
1543 * corrupt the written data.
1544 */
1545 if (stream->metadata_flag) {
1546 splice_pipe = ctx->consumer_splice_metadata_pipe;
1547 } else {
1548 splice_pipe = ctx->consumer_thread_pipe;
1549 }
1550
f02e1e8a 1551 /* Write metadata stream id before payload */
1d4dfdef
DG
1552 if (relayd) {
1553 int total_len = len;
f02e1e8a 1554
1d4dfdef
DG
1555 if (stream->metadata_flag) {
1556 /*
1557 * Lock the control socket for the complete duration of the function
1558 * since from this point on we will use the socket.
1559 */
1560 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1561
1562 ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd,
1563 padding);
1564 if (ret < 0) {
1565 written = ret;
8994307f
DG
1566 /* Socket operation failed. We consider the relayd dead */
1567 if (ret == -EBADF) {
1568 WARN("Remote relayd disconnected. Stopping");
1569 relayd_hang_up = 1;
1570 goto write_error;
1571 }
1d4dfdef
DG
1572 goto end;
1573 }
1574
1575 total_len += sizeof(struct lttcomm_relayd_metadata_payload);
1576 }
1577
1578 ret = write_relayd_stream_header(stream, total_len, padding, relayd);
1579 if (ret >= 0) {
1580 /* Use the returned socket. */
1581 outfd = ret;
1582 } else {
8994307f
DG
1583 /* Socket operation failed. We consider the relayd dead */
1584 if (ret == -EBADF) {
1585 WARN("Remote relayd disconnected. Stopping");
1586 relayd_hang_up = 1;
1587 goto write_error;
1588 }
f02e1e8a
DG
1589 goto end;
1590 }
1d4dfdef
DG
1591 } else {
1592 /* No streaming, we have to set the len with the full padding */
1593 len += padding;
1624d5b7
JD
1594
1595 /*
1596 * Check if we need to change the tracefile before writing the packet.
1597 */
1598 if (stream->chan->tracefile_size > 0 &&
1599 (stream->tracefile_size_current + len) >
1600 stream->chan->tracefile_size) {
fe4477ee
JD
1601 ret = utils_rotate_stream_file(stream->chan->pathname,
1602 stream->name, stream->chan->tracefile_size,
1603 stream->chan->tracefile_count, stream->uid, stream->gid,
1604 stream->out_fd, &(stream->tracefile_count_current));
1624d5b7
JD
1605 if (ret < 0) {
1606 ERR("Rotating output file");
1607 goto end;
1608 }
fe4477ee 1609 outfd = stream->out_fd = ret;
a6976990
DG
1610 /* Reset current size because we just perform a rotation. */
1611 stream->tracefile_size_current = 0;
1624d5b7
JD
1612 }
1613 stream->tracefile_size_current += len;
f02e1e8a
DG
1614 }
1615
1616 while (len > 0) {
1d4dfdef
DG
1617 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1618 (unsigned long)offset, len, fd, splice_pipe[1]);
fb3a43a9 1619 ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len,
f02e1e8a
DG
1620 SPLICE_F_MOVE | SPLICE_F_MORE);
1621 DBG("splice chan to pipe, ret %zd", ret_splice);
1622 if (ret_splice < 0) {
1623 PERROR("Error in relay splice");
1624 if (written == 0) {
1625 written = ret_splice;
1626 }
1627 ret = errno;
1628 goto splice_error;
1629 }
1630
1631 /* Handle stream on the relayd if the output is on the network */
1632 if (relayd) {
1633 if (stream->metadata_flag) {
1d4dfdef
DG
1634 size_t metadata_payload_size =
1635 sizeof(struct lttcomm_relayd_metadata_payload);
1636
f02e1e8a 1637 /* Update counter to fit the spliced data */
1d4dfdef
DG
1638 ret_splice += metadata_payload_size;
1639 len += metadata_payload_size;
f02e1e8a
DG
1640 /*
1641 * We do this so the return value can match the len passed as
1642 * argument to this function.
1643 */
1d4dfdef 1644 written -= metadata_payload_size;
f02e1e8a
DG
1645 }
1646 }
1647
1648 /* Splice data out */
fb3a43a9 1649 ret_splice = splice(splice_pipe[0], NULL, outfd, NULL,
f02e1e8a 1650 ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE);
1d4dfdef 1651 DBG("Consumer splice pipe to file, ret %zd", ret_splice);
f02e1e8a
DG
1652 if (ret_splice < 0) {
1653 PERROR("Error in file splice");
1654 if (written == 0) {
1655 written = ret_splice;
1656 }
8994307f 1657 /* Socket operation failed. We consider the relayd dead */
00c8752b 1658 if (errno == EBADF || errno == EPIPE) {
8994307f
DG
1659 WARN("Remote relayd disconnected. Stopping");
1660 relayd_hang_up = 1;
1661 goto write_error;
1662 }
f02e1e8a
DG
1663 ret = errno;
1664 goto splice_error;
1665 } else if (ret_splice > len) {
1666 errno = EINVAL;
1667 PERROR("Wrote more data than requested %zd (len: %lu)",
1668 ret_splice, len);
1669 written += ret_splice;
1670 ret = errno;
1671 goto splice_error;
1672 }
1673 len -= ret_splice;
1674
1675 /* This call is useless on a socket so better save a syscall. */
1676 if (!relayd) {
1677 /* This won't block, but will start writeout asynchronously */
1678 lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice,
1679 SYNC_FILE_RANGE_WRITE);
1680 stream->out_fd_offset += ret_splice;
1681 }
1682 written += ret_splice;
1683 }
1684 lttng_consumer_sync_trace_file(stream, orig_offset);
1685
1686 ret = ret_splice;
1687
1688 goto end;
1689
8994307f
DG
1690write_error:
1691 /*
1692 * This is a special case that the relayd has closed its socket. Let's
1693 * cleanup the relayd object and all associated streams.
1694 */
1695 if (relayd && relayd_hang_up) {
1696 cleanup_relayd(relayd, ctx);
1697 /* Skip splice error so the consumer does not fail */
1698 goto end;
1699 }
1700
f02e1e8a
DG
1701splice_error:
1702 /* send the appropriate error description to sessiond */
1703 switch (ret) {
f02e1e8a 1704 case EINVAL:
f73fabfd 1705 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL);
f02e1e8a
DG
1706 break;
1707 case ENOMEM:
f73fabfd 1708 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM);
f02e1e8a
DG
1709 break;
1710 case ESPIPE:
f73fabfd 1711 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE);
f02e1e8a
DG
1712 break;
1713 }
1714
1715end:
1716 if (relayd && stream->metadata_flag) {
1717 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1718 }
1719
1720 rcu_read_unlock();
1721 return written;
3bd1e081
MD
1722}
1723
1724/*
1725 * Take a snapshot for a specific fd
1726 *
1727 * Returns 0 on success, < 0 on error
1728 */
ffe60014 1729int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream)
3bd1e081
MD
1730{
1731 switch (consumer_data.type) {
1732 case LTTNG_CONSUMER_KERNEL:
ffe60014 1733 return lttng_kconsumer_take_snapshot(stream);
7753dea8
MD
1734 case LTTNG_CONSUMER32_UST:
1735 case LTTNG_CONSUMER64_UST:
ffe60014 1736 return lttng_ustconsumer_take_snapshot(stream);
3bd1e081
MD
1737 default:
1738 ERR("Unknown consumer_data type");
1739 assert(0);
1740 return -ENOSYS;
1741 }
3bd1e081
MD
1742}
1743
1744/*
1745 * Get the produced position
1746 *
1747 * Returns 0 on success, < 0 on error
1748 */
ffe60014 1749int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
3bd1e081
MD
1750 unsigned long *pos)
1751{
1752 switch (consumer_data.type) {
1753 case LTTNG_CONSUMER_KERNEL:
ffe60014 1754 return lttng_kconsumer_get_produced_snapshot(stream, pos);
7753dea8
MD
1755 case LTTNG_CONSUMER32_UST:
1756 case LTTNG_CONSUMER64_UST:
ffe60014 1757 return lttng_ustconsumer_get_produced_snapshot(stream, pos);
3bd1e081
MD
1758 default:
1759 ERR("Unknown consumer_data type");
1760 assert(0);
1761 return -ENOSYS;
1762 }
1763}
1764
1765int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1766 int sock, struct pollfd *consumer_sockpoll)
1767{
1768 switch (consumer_data.type) {
1769 case LTTNG_CONSUMER_KERNEL:
1770 return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
7753dea8
MD
1771 case LTTNG_CONSUMER32_UST:
1772 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
1773 return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
1774 default:
1775 ERR("Unknown consumer_data type");
1776 assert(0);
1777 return -ENOSYS;
1778 }
1779}
1780
43c34bc3
DG
1781/*
1782 * Iterate over all streams of the hashtable and free them properly.
1783 *
1784 * WARNING: *MUST* be used with data stream only.
1785 */
1786static void destroy_data_stream_ht(struct lttng_ht *ht)
1787{
43c34bc3
DG
1788 struct lttng_ht_iter iter;
1789 struct lttng_consumer_stream *stream;
1790
1791 if (ht == NULL) {
1792 return;
1793 }
1794
1795 rcu_read_lock();
1796 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
5c540210
DG
1797 /*
1798 * Ignore return value since we are currently cleaning up so any error
1799 * can't be handled.
1800 */
1801 (void) consumer_del_stream(stream, ht);
43c34bc3
DG
1802 }
1803 rcu_read_unlock();
1804
1805 lttng_ht_destroy(ht);
1806}
1807
fb3a43a9 1808/*
f724d81e 1809 * Iterate over all streams of the hashtable and free them properly.
e316aad5
DG
1810 *
1811 * XXX: Should not be only for metadata stream or else use an other name.
fb3a43a9
DG
1812 */
1813static void destroy_stream_ht(struct lttng_ht *ht)
1814{
fb3a43a9
DG
1815 struct lttng_ht_iter iter;
1816 struct lttng_consumer_stream *stream;
1817
1818 if (ht == NULL) {
1819 return;
1820 }
1821
d09e1200 1822 rcu_read_lock();
58b1f425 1823 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
5c540210
DG
1824 /*
1825 * Ignore return value since we are currently cleaning up so any error
1826 * can't be handled.
1827 */
1828 (void) consumer_del_metadata_stream(stream, ht);
fb3a43a9 1829 }
d09e1200 1830 rcu_read_unlock();
fb3a43a9
DG
1831
1832 lttng_ht_destroy(ht);
1833}
1834
d88aee68
DG
1835void lttng_consumer_close_metadata(void)
1836{
1837 switch (consumer_data.type) {
1838 case LTTNG_CONSUMER_KERNEL:
1839 /*
1840 * The Kernel consumer has a different metadata scheme so we don't
1841 * close anything because the stream will be closed by the session
1842 * daemon.
1843 */
1844 break;
1845 case LTTNG_CONSUMER32_UST:
1846 case LTTNG_CONSUMER64_UST:
1847 /*
1848 * Close all metadata streams. The metadata hash table is passed and
1849 * this call iterates over it by closing all wakeup fd. This is safe
1850 * because at this point we are sure that the metadata producer is
1851 * either dead or blocked.
1852 */
1853 lttng_ustconsumer_close_metadata(metadata_ht);
1854 break;
1855 default:
1856 ERR("Unknown consumer_data type");
1857 assert(0);
1858 }
1859}
1860
fb3a43a9
DG
1861/*
1862 * Clean up a metadata stream and free its memory.
1863 */
e316aad5
DG
1864void consumer_del_metadata_stream(struct lttng_consumer_stream *stream,
1865 struct lttng_ht *ht)
fb3a43a9
DG
1866{
1867 int ret;
e316aad5
DG
1868 struct lttng_ht_iter iter;
1869 struct lttng_consumer_channel *free_chan = NULL;
fb3a43a9
DG
1870 struct consumer_relayd_sock_pair *relayd;
1871
1872 assert(stream);
1873 /*
1874 * This call should NEVER receive regular stream. It must always be
1875 * metadata stream and this is crucial for data structure synchronization.
1876 */
1877 assert(stream->metadata_flag);
1878
e316aad5
DG
1879 DBG3("Consumer delete metadata stream %d", stream->wait_fd);
1880
1881 if (ht == NULL) {
1882 /* Means the stream was allocated but not successfully added */
ffe60014 1883 goto free_stream_rcu;
e316aad5
DG
1884 }
1885
74251bb8 1886 pthread_mutex_lock(&consumer_data.lock);
a9838785 1887 pthread_mutex_lock(&stream->chan->lock);
8994307f
DG
1888 pthread_mutex_lock(&stream->lock);
1889
fb3a43a9
DG
1890 switch (consumer_data.type) {
1891 case LTTNG_CONSUMER_KERNEL:
1892 if (stream->mmap_base != NULL) {
1893 ret = munmap(stream->mmap_base, stream->mmap_len);
1894 if (ret != 0) {
1895 PERROR("munmap metadata stream");
1896 }
1897 }
4c95e622
JD
1898 if (stream->wait_fd >= 0) {
1899 ret = close(stream->wait_fd);
1900 if (ret < 0) {
1901 PERROR("close kernel metadata wait_fd");
1902 }
1903 }
fb3a43a9
DG
1904 break;
1905 case LTTNG_CONSUMER32_UST:
1906 case LTTNG_CONSUMER64_UST:
04ef1097
MD
1907 if (stream->monitor) {
1908 /* close the write-side in close_metadata */
1909 ret = close(stream->ust_metadata_poll_pipe[0]);
1910 if (ret < 0) {
1911 PERROR("Close UST metadata read-side poll pipe");
1912 }
1913 }
fb3a43a9
DG
1914 lttng_ustconsumer_del_stream(stream);
1915 break;
1916 default:
1917 ERR("Unknown consumer_data type");
1918 assert(0);
e316aad5 1919 goto end;
fb3a43a9 1920 }
fb3a43a9 1921
c869f647 1922 rcu_read_lock();
58b1f425 1923 iter.iter.node = &stream->node.node;
c869f647
DG
1924 ret = lttng_ht_del(ht, &iter);
1925 assert(!ret);
ca22feea 1926
d8ef542d
MD
1927 iter.iter.node = &stream->node_channel_id.node;
1928 ret = lttng_ht_del(consumer_data.stream_per_chan_id_ht, &iter);
1929 assert(!ret);
1930
ca22feea
DG
1931 iter.iter.node = &stream->node_session_id.node;
1932 ret = lttng_ht_del(consumer_data.stream_list_ht, &iter);
1933 assert(!ret);
c869f647
DG
1934 rcu_read_unlock();
1935
fb3a43a9
DG
1936 if (stream->out_fd >= 0) {
1937 ret = close(stream->out_fd);
1938 if (ret) {
1939 PERROR("close");
1940 }
1941 }
1942
fb3a43a9
DG
1943 /* Check and cleanup relayd */
1944 rcu_read_lock();
1945 relayd = consumer_find_relayd(stream->net_seq_idx);
1946 if (relayd != NULL) {
1947 uatomic_dec(&relayd->refcount);
1948 assert(uatomic_read(&relayd->refcount) >= 0);
1949
1950 /* Closing streams requires to lock the control socket. */
1951 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1952 ret = relayd_send_close_stream(&relayd->control_sock,
1953 stream->relayd_stream_id, stream->next_net_seq_num - 1);
1954 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1955 if (ret < 0) {
1956 DBG("Unable to close stream on the relayd. Continuing");
1957 /*
1958 * Continue here. There is nothing we can do for the relayd.
1959 * Chances are that the relayd has closed the socket so we just
1960 * continue cleaning up.
1961 */
1962 }
1963
1964 /* Both conditions are met, we destroy the relayd. */
1965 if (uatomic_read(&relayd->refcount) == 0 &&
1966 uatomic_read(&relayd->destroy_flag)) {
51230d70 1967 consumer_destroy_relayd(relayd);
fb3a43a9
DG
1968 }
1969 }
1970 rcu_read_unlock();
1971
1972 /* Atomically decrement channel refcount since other threads can use it. */
f2ad556d 1973 if (!uatomic_sub_return(&stream->chan->refcount, 1)
ffe60014 1974 && !uatomic_read(&stream->chan->nb_init_stream_left)) {
c30aaa51 1975 /* Go for channel deletion! */
e316aad5 1976 free_chan = stream->chan;
fb3a43a9
DG
1977 }
1978
e316aad5 1979end:
73811ecc
DG
1980 /*
1981 * Nullify the stream reference so it is not used after deletion. The
5e41ebe1
MD
1982 * channel lock MUST be acquired before being able to check for
1983 * a NULL pointer value.
73811ecc
DG
1984 */
1985 stream->chan->metadata_stream = NULL;
1986
8994307f 1987 pthread_mutex_unlock(&stream->lock);
a9838785 1988 pthread_mutex_unlock(&stream->chan->lock);
74251bb8 1989 pthread_mutex_unlock(&consumer_data.lock);
e316aad5
DG
1990
1991 if (free_chan) {
1992 consumer_del_channel(free_chan);
1993 }
1994
ffe60014
DG
1995free_stream_rcu:
1996 call_rcu(&stream->node.head, free_stream_rcu);
fb3a43a9
DG
1997}
1998
1999/*
2000 * Action done with the metadata stream when adding it to the consumer internal
2001 * data structures to handle it.
2002 */
5ab66908 2003int consumer_add_metadata_stream(struct lttng_consumer_stream *stream)
fb3a43a9 2004{
5ab66908 2005 struct lttng_ht *ht = metadata_ht;
e316aad5 2006 int ret = 0;
76082088 2007 struct lttng_ht_iter iter;
d88aee68 2008 struct lttng_ht_node_u64 *node;
fb3a43a9 2009
e316aad5
DG
2010 assert(stream);
2011 assert(ht);
2012
d88aee68 2013 DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key);
e316aad5
DG
2014
2015 pthread_mutex_lock(&consumer_data.lock);
a9838785 2016 pthread_mutex_lock(&stream->chan->lock);
ec6ea7d0 2017 pthread_mutex_lock(&stream->chan->timer_lock);
2e818a6a 2018 pthread_mutex_lock(&stream->lock);
e316aad5 2019
e316aad5
DG
2020 /*
2021 * From here, refcounts are updated so be _careful_ when returning an error
2022 * after this point.
2023 */
2024
fb3a43a9 2025 rcu_read_lock();
76082088
DG
2026
2027 /*
2028 * Lookup the stream just to make sure it does not exist in our internal
2029 * state. This should NEVER happen.
2030 */
d88aee68
DG
2031 lttng_ht_lookup(ht, &stream->key, &iter);
2032 node = lttng_ht_iter_get_node_u64(&iter);
76082088
DG
2033 assert(!node);
2034
e316aad5 2035 /*
ffe60014
DG
2036 * When nb_init_stream_left reaches 0, we don't need to trigger any action
2037 * in terms of destroying the associated channel, because the action that
e316aad5
DG
2038 * causes the count to become 0 also causes a stream to be added. The
2039 * channel deletion will thus be triggered by the following removal of this
2040 * stream.
2041 */
ffe60014 2042 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
f2ad556d
MD
2043 /* Increment refcount before decrementing nb_init_stream_left */
2044 cmm_smp_wmb();
ffe60014 2045 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
2046 }
2047
d88aee68 2048 lttng_ht_add_unique_u64(ht, &stream->node);
ca22feea 2049
d8ef542d
MD
2050 lttng_ht_add_unique_u64(consumer_data.stream_per_chan_id_ht,
2051 &stream->node_channel_id);
2052
ca22feea
DG
2053 /*
2054 * Add stream to the stream_list_ht of the consumer data. No need to steal
2055 * the key since the HT does not use it and we allow to add redundant keys
2056 * into this table.
2057 */
d88aee68 2058 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 2059
fb3a43a9 2060 rcu_read_unlock();
e316aad5 2061
2e818a6a 2062 pthread_mutex_unlock(&stream->lock);
a9838785 2063 pthread_mutex_unlock(&stream->chan->lock);
ec6ea7d0 2064 pthread_mutex_unlock(&stream->chan->timer_lock);
e316aad5
DG
2065 pthread_mutex_unlock(&consumer_data.lock);
2066 return ret;
fb3a43a9
DG
2067}
2068
8994307f
DG
2069/*
2070 * Delete data stream that are flagged for deletion (endpoint_status).
2071 */
2072static void validate_endpoint_status_data_stream(void)
2073{
2074 struct lttng_ht_iter iter;
2075 struct lttng_consumer_stream *stream;
2076
2077 DBG("Consumer delete flagged data stream");
2078
2079 rcu_read_lock();
2080 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
2081 /* Validate delete flag of the stream */
79d4ffb7 2082 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2083 continue;
2084 }
2085 /* Delete it right now */
2086 consumer_del_stream(stream, data_ht);
2087 }
2088 rcu_read_unlock();
2089}
2090
2091/*
2092 * Delete metadata stream that are flagged for deletion (endpoint_status).
2093 */
2094static void validate_endpoint_status_metadata_stream(
2095 struct lttng_poll_event *pollset)
2096{
2097 struct lttng_ht_iter iter;
2098 struct lttng_consumer_stream *stream;
2099
2100 DBG("Consumer delete flagged metadata stream");
2101
2102 assert(pollset);
2103
2104 rcu_read_lock();
2105 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
2106 /* Validate delete flag of the stream */
79d4ffb7 2107 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
2108 continue;
2109 }
2110 /*
2111 * Remove from pollset so the metadata thread can continue without
2112 * blocking on a deleted stream.
2113 */
2114 lttng_poll_del(pollset, stream->wait_fd);
2115
2116 /* Delete it right now */
2117 consumer_del_metadata_stream(stream, metadata_ht);
2118 }
2119 rcu_read_unlock();
2120}
2121
fb3a43a9
DG
2122/*
2123 * Thread polls on metadata file descriptor and write them on disk or on the
2124 * network.
2125 */
7d980def 2126void *consumer_thread_metadata_poll(void *data)
fb3a43a9
DG
2127{
2128 int ret, i, pollfd;
2129 uint32_t revents, nb_fd;
e316aad5 2130 struct lttng_consumer_stream *stream = NULL;
fb3a43a9 2131 struct lttng_ht_iter iter;
d88aee68 2132 struct lttng_ht_node_u64 *node;
fb3a43a9
DG
2133 struct lttng_poll_event events;
2134 struct lttng_consumer_local_data *ctx = data;
2135 ssize_t len;
2136
2137 rcu_register_thread();
2138
d88aee68 2139 metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
04bb2b64
DG
2140 if (!metadata_ht) {
2141 /* ENOMEM at this point. Better to bail out. */
d8ef542d 2142 goto end_ht;
04bb2b64
DG
2143 }
2144
fb3a43a9
DG
2145 DBG("Thread metadata poll started");
2146
fb3a43a9
DG
2147 /* Size is set to 1 for the consumer_metadata pipe */
2148 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2149 if (ret < 0) {
2150 ERR("Poll set creation failed");
d8ef542d 2151 goto end_poll;
fb3a43a9
DG
2152 }
2153
13886d2d
DG
2154 ret = lttng_poll_add(&events,
2155 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe), LPOLLIN);
fb3a43a9
DG
2156 if (ret < 0) {
2157 goto end;
2158 }
2159
2160 /* Main loop */
2161 DBG("Metadata main loop started");
2162
2163 while (1) {
fb3a43a9 2164 /* Only the metadata pipe is set */
d21b0d71 2165 if (LTTNG_POLL_GETNB(&events) == 0 && consumer_quit == 1) {
fb3a43a9
DG
2166 goto end;
2167 }
2168
2169restart:
d21b0d71 2170 DBG("Metadata poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events));
fb3a43a9
DG
2171 ret = lttng_poll_wait(&events, -1);
2172 DBG("Metadata event catched in thread");
2173 if (ret < 0) {
2174 if (errno == EINTR) {
e316aad5 2175 ERR("Poll EINTR catched");
fb3a43a9
DG
2176 goto restart;
2177 }
2178 goto error;
2179 }
2180
0d9c5d77
DG
2181 nb_fd = ret;
2182
e316aad5 2183 /* From here, the event is a metadata wait fd */
fb3a43a9
DG
2184 for (i = 0; i < nb_fd; i++) {
2185 revents = LTTNG_POLL_GETEV(&events, i);
2186 pollfd = LTTNG_POLL_GETFD(&events, i);
2187
e316aad5
DG
2188 /* Just don't waste time if no returned events for the fd */
2189 if (!revents) {
2190 continue;
2191 }
2192
13886d2d 2193 if (pollfd == lttng_pipe_get_readfd(ctx->consumer_metadata_pipe)) {
4adabd61 2194 if (revents & (LPOLLERR | LPOLLHUP )) {
fb3a43a9
DG
2195 DBG("Metadata thread pipe hung up");
2196 /*
2197 * Remove the pipe from the poll set and continue the loop
2198 * since their might be data to consume.
2199 */
13886d2d
DG
2200 lttng_poll_del(&events,
2201 lttng_pipe_get_readfd(ctx->consumer_metadata_pipe));
2202 lttng_pipe_read_close(ctx->consumer_metadata_pipe);
fb3a43a9
DG
2203 continue;
2204 } else if (revents & LPOLLIN) {
13886d2d
DG
2205 ssize_t pipe_len;
2206
2207 pipe_len = lttng_pipe_read(ctx->consumer_metadata_pipe,
2208 &stream, sizeof(stream));
2209 if (pipe_len < 0) {
2210 ERR("read metadata stream, ret: %ld", pipe_len);
fb3a43a9 2211 /*
13886d2d 2212 * Continue here to handle the rest of the streams.
fb3a43a9
DG
2213 */
2214 continue;
2215 }
2216
8994307f
DG
2217 /* A NULL stream means that the state has changed. */
2218 if (stream == NULL) {
2219 /* Check for deleted streams. */
2220 validate_endpoint_status_metadata_stream(&events);
3714380f 2221 goto restart;
8994307f
DG
2222 }
2223
fb3a43a9
DG
2224 DBG("Adding metadata stream %d to poll set",
2225 stream->wait_fd);
2226
fb3a43a9
DG
2227 /* Add metadata stream to the global poll events list */
2228 lttng_poll_add(&events, stream->wait_fd,
2229 LPOLLIN | LPOLLPRI);
fb3a43a9
DG
2230 }
2231
e316aad5 2232 /* Handle other stream */
fb3a43a9
DG
2233 continue;
2234 }
2235
d09e1200 2236 rcu_read_lock();
d88aee68
DG
2237 {
2238 uint64_t tmp_id = (uint64_t) pollfd;
2239
2240 lttng_ht_lookup(metadata_ht, &tmp_id, &iter);
2241 }
2242 node = lttng_ht_iter_get_node_u64(&iter);
e316aad5 2243 assert(node);
fb3a43a9
DG
2244
2245 stream = caa_container_of(node, struct lttng_consumer_stream,
58b1f425 2246 node);
fb3a43a9 2247
e316aad5 2248 /* Check for error event */
4adabd61 2249 if (revents & (LPOLLERR | LPOLLHUP)) {
e316aad5 2250 DBG("Metadata fd %d is hup|err.", pollfd);
fb3a43a9
DG
2251 if (!stream->hangup_flush_done
2252 && (consumer_data.type == LTTNG_CONSUMER32_UST
2253 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2254 DBG("Attempting to flush and consume the UST buffers");
2255 lttng_ustconsumer_on_stream_hangup(stream);
2256
2257 /* We just flushed the stream now read it. */
4bb94b75
DG
2258 do {
2259 len = ctx->on_buffer_ready(stream, ctx);
2260 /*
2261 * We don't check the return value here since if we get
2262 * a negative len, it means an error occured thus we
2263 * simply remove it from the poll set and free the
2264 * stream.
2265 */
2266 } while (len > 0);
fb3a43a9
DG
2267 }
2268
fb3a43a9 2269 lttng_poll_del(&events, stream->wait_fd);
e316aad5
DG
2270 /*
2271 * This call update the channel states, closes file descriptors
2272 * and securely free the stream.
2273 */
2274 consumer_del_metadata_stream(stream, metadata_ht);
2275 } else if (revents & (LPOLLIN | LPOLLPRI)) {
2276 /* Get the data out of the metadata file descriptor */
2277 DBG("Metadata available on fd %d", pollfd);
2278 assert(stream->wait_fd == pollfd);
2279
04ef1097
MD
2280 do {
2281 len = ctx->on_buffer_ready(stream, ctx);
2282 /*
2283 * We don't check the return value here since if we get
2284 * a negative len, it means an error occured thus we
2285 * simply remove it from the poll set and free the
2286 * stream.
2287 */
2288 } while (len > 0);
2289
e316aad5 2290 /* It's ok to have an unavailable sub-buffer */
b64403e3 2291 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2292 /* Clean up stream from consumer and free it. */
2293 lttng_poll_del(&events, stream->wait_fd);
2294 consumer_del_metadata_stream(stream, metadata_ht);
e316aad5 2295 }
fb3a43a9 2296 }
e316aad5
DG
2297
2298 /* Release RCU lock for the stream looked up */
d09e1200 2299 rcu_read_unlock();
fb3a43a9
DG
2300 }
2301 }
2302
2303error:
2304end:
2305 DBG("Metadata poll thread exiting");
fb3a43a9 2306
d8ef542d
MD
2307 lttng_poll_clean(&events);
2308end_poll:
04bb2b64 2309 destroy_stream_ht(metadata_ht);
d8ef542d 2310end_ht:
fb3a43a9
DG
2311 rcu_unregister_thread();
2312 return NULL;
2313}
2314
3bd1e081 2315/*
e4421fec 2316 * This thread polls the fds in the set to consume the data and write
3bd1e081
MD
2317 * it to tracefile if necessary.
2318 */
7d980def 2319void *consumer_thread_data_poll(void *data)
3bd1e081
MD
2320{
2321 int num_rdy, num_hup, high_prio, ret, i;
2322 struct pollfd *pollfd = NULL;
2323 /* local view of the streams */
c869f647 2324 struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL;
3bd1e081
MD
2325 /* local view of consumer_data.fds_count */
2326 int nb_fd = 0;
3bd1e081 2327 struct lttng_consumer_local_data *ctx = data;
00e2e675 2328 ssize_t len;
3bd1e081 2329
e7b994a3
DG
2330 rcu_register_thread();
2331
d88aee68 2332 data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
43c34bc3 2333 if (data_ht == NULL) {
04bb2b64 2334 /* ENOMEM at this point. Better to bail out. */
43c34bc3
DG
2335 goto end;
2336 }
2337
4df6c8cb
MD
2338 local_stream = zmalloc(sizeof(struct lttng_consumer_stream *));
2339 if (local_stream == NULL) {
2340 PERROR("local_stream malloc");
2341 goto end;
2342 }
3bd1e081
MD
2343
2344 while (1) {
2345 high_prio = 0;
2346 num_hup = 0;
2347
2348 /*
e4421fec 2349 * the fds set has been updated, we need to update our
3bd1e081
MD
2350 * local array as well
2351 */
2352 pthread_mutex_lock(&consumer_data.lock);
2353 if (consumer_data.need_update) {
0e428499
DG
2354 free(pollfd);
2355 pollfd = NULL;
2356
2357 free(local_stream);
2358 local_stream = NULL;
3bd1e081 2359
50f8ae69 2360 /* allocate for all fds + 1 for the consumer_data_pipe */
effcf122 2361 pollfd = zmalloc((consumer_data.stream_count + 1) * sizeof(struct pollfd));
3bd1e081 2362 if (pollfd == NULL) {
7a57cf92 2363 PERROR("pollfd malloc");
3bd1e081
MD
2364 pthread_mutex_unlock(&consumer_data.lock);
2365 goto end;
2366 }
2367
50f8ae69 2368 /* allocate for all fds + 1 for the consumer_data_pipe */
effcf122 2369 local_stream = zmalloc((consumer_data.stream_count + 1) *
747f8642 2370 sizeof(struct lttng_consumer_stream *));
3bd1e081 2371 if (local_stream == NULL) {
7a57cf92 2372 PERROR("local_stream malloc");
3bd1e081
MD
2373 pthread_mutex_unlock(&consumer_data.lock);
2374 goto end;
2375 }
ffe60014 2376 ret = update_poll_array(ctx, &pollfd, local_stream,
43c34bc3 2377 data_ht);
3bd1e081
MD
2378 if (ret < 0) {
2379 ERR("Error in allocating pollfd or local_outfds");
f73fabfd 2380 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2381 pthread_mutex_unlock(&consumer_data.lock);
2382 goto end;
2383 }
2384 nb_fd = ret;
2385 consumer_data.need_update = 0;
2386 }
2387 pthread_mutex_unlock(&consumer_data.lock);
2388
4078b776
MD
2389 /* No FDs and consumer_quit, consumer_cleanup the thread */
2390 if (nb_fd == 0 && consumer_quit == 1) {
2391 goto end;
2392 }
3bd1e081 2393 /* poll on the array of fds */
88f2b785 2394 restart:
3bd1e081 2395 DBG("polling on %d fd", nb_fd + 1);
cb365c03 2396 num_rdy = poll(pollfd, nb_fd + 1, -1);
3bd1e081
MD
2397 DBG("poll num_rdy : %d", num_rdy);
2398 if (num_rdy == -1) {
88f2b785
MD
2399 /*
2400 * Restart interrupted system call.
2401 */
2402 if (errno == EINTR) {
2403 goto restart;
2404 }
7a57cf92 2405 PERROR("Poll error");
f73fabfd 2406 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2407 goto end;
2408 } else if (num_rdy == 0) {
2409 DBG("Polling thread timed out");
2410 goto end;
2411 }
2412
3bd1e081 2413 /*
50f8ae69 2414 * If the consumer_data_pipe triggered poll go directly to the
00e2e675
DG
2415 * beginning of the loop to update the array. We want to prioritize
2416 * array update over low-priority reads.
3bd1e081 2417 */
509bb1cf 2418 if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) {
ab30f567 2419 ssize_t pipe_readlen;
04fdd819 2420
50f8ae69 2421 DBG("consumer_data_pipe wake up");
acdb9057
DG
2422 pipe_readlen = lttng_pipe_read(ctx->consumer_data_pipe,
2423 &new_stream, sizeof(new_stream));
23f5f35d 2424 if (pipe_readlen < 0) {
acdb9057 2425 ERR("Consumer data pipe ret %ld", pipe_readlen);
23f5f35d
DG
2426 /* Continue so we can at least handle the current stream(s). */
2427 continue;
2428 }
c869f647
DG
2429
2430 /*
2431 * If the stream is NULL, just ignore it. It's also possible that
2432 * the sessiond poll thread changed the consumer_quit state and is
2433 * waking us up to test it.
2434 */
2435 if (new_stream == NULL) {
8994307f 2436 validate_endpoint_status_data_stream();
c869f647
DG
2437 continue;
2438 }
2439
c869f647 2440 /* Continue to update the local streams and handle prio ones */
3bd1e081
MD
2441 continue;
2442 }
2443
2444 /* Take care of high priority channels first. */
2445 for (i = 0; i < nb_fd; i++) {
9617607b
DG
2446 if (local_stream[i] == NULL) {
2447 continue;
2448 }
fb3a43a9 2449 if (pollfd[i].revents & POLLPRI) {
d41f73b7
MD
2450 DBG("Urgent read on fd %d", pollfd[i].fd);
2451 high_prio = 1;
4078b776 2452 len = ctx->on_buffer_ready(local_stream[i], ctx);
d41f73b7 2453 /* it's ok to have an unavailable sub-buffer */
b64403e3 2454 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2455 /* Clean the stream and free it. */
2456 consumer_del_stream(local_stream[i], data_ht);
9617607b 2457 local_stream[i] = NULL;
4078b776
MD
2458 } else if (len > 0) {
2459 local_stream[i]->data_read = 1;
d41f73b7 2460 }
3bd1e081
MD
2461 }
2462 }
2463
4078b776
MD
2464 /*
2465 * If we read high prio channel in this loop, try again
2466 * for more high prio data.
2467 */
2468 if (high_prio) {
3bd1e081
MD
2469 continue;
2470 }
2471
2472 /* Take care of low priority channels. */
4078b776 2473 for (i = 0; i < nb_fd; i++) {
9617607b
DG
2474 if (local_stream[i] == NULL) {
2475 continue;
2476 }
4078b776
MD
2477 if ((pollfd[i].revents & POLLIN) ||
2478 local_stream[i]->hangup_flush_done) {
4078b776
MD
2479 DBG("Normal read on fd %d", pollfd[i].fd);
2480 len = ctx->on_buffer_ready(local_stream[i], ctx);
2481 /* it's ok to have an unavailable sub-buffer */
b64403e3 2482 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2483 /* Clean the stream and free it. */
2484 consumer_del_stream(local_stream[i], data_ht);
9617607b 2485 local_stream[i] = NULL;
4078b776
MD
2486 } else if (len > 0) {
2487 local_stream[i]->data_read = 1;
2488 }
2489 }
2490 }
2491
2492 /* Handle hangup and errors */
2493 for (i = 0; i < nb_fd; i++) {
9617607b
DG
2494 if (local_stream[i] == NULL) {
2495 continue;
2496 }
4078b776
MD
2497 if (!local_stream[i]->hangup_flush_done
2498 && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL))
2499 && (consumer_data.type == LTTNG_CONSUMER32_UST
2500 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2501 DBG("fd %d is hup|err|nval. Attempting flush and read.",
9617607b 2502 pollfd[i].fd);
4078b776
MD
2503 lttng_ustconsumer_on_stream_hangup(local_stream[i]);
2504 /* Attempt read again, for the data we just flushed. */
2505 local_stream[i]->data_read = 1;
2506 }
2507 /*
2508 * If the poll flag is HUP/ERR/NVAL and we have
2509 * read no data in this pass, we can remove the
2510 * stream from its hash table.
2511 */
2512 if ((pollfd[i].revents & POLLHUP)) {
2513 DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
2514 if (!local_stream[i]->data_read) {
43c34bc3 2515 consumer_del_stream(local_stream[i], data_ht);
9617607b 2516 local_stream[i] = NULL;
4078b776
MD
2517 num_hup++;
2518 }
2519 } else if (pollfd[i].revents & POLLERR) {
2520 ERR("Error returned in polling fd %d.", pollfd[i].fd);
2521 if (!local_stream[i]->data_read) {
43c34bc3 2522 consumer_del_stream(local_stream[i], data_ht);
9617607b 2523 local_stream[i] = NULL;
4078b776
MD
2524 num_hup++;
2525 }
2526 } else if (pollfd[i].revents & POLLNVAL) {
2527 ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
2528 if (!local_stream[i]->data_read) {
43c34bc3 2529 consumer_del_stream(local_stream[i], data_ht);
9617607b 2530 local_stream[i] = NULL;
4078b776 2531 num_hup++;
3bd1e081
MD
2532 }
2533 }
9617607b
DG
2534 if (local_stream[i] != NULL) {
2535 local_stream[i]->data_read = 0;
2536 }
3bd1e081
MD
2537 }
2538 }
2539end:
2540 DBG("polling thread exiting");
0e428499
DG
2541 free(pollfd);
2542 free(local_stream);
fb3a43a9
DG
2543
2544 /*
2545 * Close the write side of the pipe so epoll_wait() in
7d980def
DG
2546 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2547 * read side of the pipe. If we close them both, epoll_wait strangely does
2548 * not return and could create a endless wait period if the pipe is the
2549 * only tracked fd in the poll set. The thread will take care of closing
2550 * the read side.
fb3a43a9 2551 */
13886d2d 2552 (void) lttng_pipe_write_close(ctx->consumer_metadata_pipe);
fb3a43a9 2553
04bb2b64 2554 destroy_data_stream_ht(data_ht);
43c34bc3 2555
e7b994a3 2556 rcu_unregister_thread();
3bd1e081
MD
2557 return NULL;
2558}
2559
d8ef542d
MD
2560/*
2561 * Close wake-up end of each stream belonging to the channel. This will
2562 * allow the poll() on the stream read-side to detect when the
2563 * write-side (application) finally closes them.
2564 */
2565static
2566void consumer_close_channel_streams(struct lttng_consumer_channel *channel)
2567{
2568 struct lttng_ht *ht;
2569 struct lttng_consumer_stream *stream;
2570 struct lttng_ht_iter iter;
2571
2572 ht = consumer_data.stream_per_chan_id_ht;
2573
2574 rcu_read_lock();
2575 cds_lfht_for_each_entry_duplicate(ht->ht,
2576 ht->hash_fct(&channel->key, lttng_ht_seed),
2577 ht->match_fct, &channel->key,
2578 &iter.iter, stream, node_channel_id.node) {
f2ad556d
MD
2579 /*
2580 * Protect against teardown with mutex.
2581 */
2582 pthread_mutex_lock(&stream->lock);
2583 if (cds_lfht_is_node_deleted(&stream->node.node)) {
2584 goto next;
2585 }
d8ef542d
MD
2586 switch (consumer_data.type) {
2587 case LTTNG_CONSUMER_KERNEL:
2588 break;
2589 case LTTNG_CONSUMER32_UST:
2590 case LTTNG_CONSUMER64_UST:
2591 /*
2592 * Note: a mutex is taken internally within
2593 * liblttng-ust-ctl to protect timer wakeup_fd
2594 * use from concurrent close.
2595 */
2596 lttng_ustconsumer_close_stream_wakeup(stream);
2597 break;
2598 default:
2599 ERR("Unknown consumer_data type");
2600 assert(0);
2601 }
f2ad556d
MD
2602 next:
2603 pthread_mutex_unlock(&stream->lock);
d8ef542d
MD
2604 }
2605 rcu_read_unlock();
2606}
2607
2608static void destroy_channel_ht(struct lttng_ht *ht)
2609{
2610 struct lttng_ht_iter iter;
2611 struct lttng_consumer_channel *channel;
2612 int ret;
2613
2614 if (ht == NULL) {
2615 return;
2616 }
2617
2618 rcu_read_lock();
2619 cds_lfht_for_each_entry(ht->ht, &iter.iter, channel, wait_fd_node.node) {
2620 ret = lttng_ht_del(ht, &iter);
2621 assert(ret != 0);
2622 }
2623 rcu_read_unlock();
2624
2625 lttng_ht_destroy(ht);
2626}
2627
2628/*
2629 * This thread polls the channel fds to detect when they are being
2630 * closed. It closes all related streams if the channel is detected as
2631 * closed. It is currently only used as a shim layer for UST because the
2632 * consumerd needs to keep the per-stream wakeup end of pipes open for
2633 * periodical flush.
2634 */
2635void *consumer_thread_channel_poll(void *data)
2636{
2637 int ret, i, pollfd;
2638 uint32_t revents, nb_fd;
2639 struct lttng_consumer_channel *chan = NULL;
2640 struct lttng_ht_iter iter;
2641 struct lttng_ht_node_u64 *node;
2642 struct lttng_poll_event events;
2643 struct lttng_consumer_local_data *ctx = data;
2644 struct lttng_ht *channel_ht;
2645
2646 rcu_register_thread();
2647
2648 channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2649 if (!channel_ht) {
2650 /* ENOMEM at this point. Better to bail out. */
2651 goto end_ht;
2652 }
2653
2654 DBG("Thread channel poll started");
2655
2656 /* Size is set to 1 for the consumer_channel pipe */
2657 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2658 if (ret < 0) {
2659 ERR("Poll set creation failed");
2660 goto end_poll;
2661 }
2662
2663 ret = lttng_poll_add(&events, ctx->consumer_channel_pipe[0], LPOLLIN);
2664 if (ret < 0) {
2665 goto end;
2666 }
2667
2668 /* Main loop */
2669 DBG("Channel main loop started");
2670
2671 while (1) {
2672 /* Only the channel pipe is set */
2673 if (LTTNG_POLL_GETNB(&events) == 0 && consumer_quit == 1) {
2674 goto end;
2675 }
2676
2677restart:
2678 DBG("Channel poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events));
2679 ret = lttng_poll_wait(&events, -1);
2680 DBG("Channel event catched in thread");
2681 if (ret < 0) {
2682 if (errno == EINTR) {
2683 ERR("Poll EINTR catched");
2684 goto restart;
2685 }
2686 goto end;
2687 }
2688
2689 nb_fd = ret;
2690
2691 /* From here, the event is a channel wait fd */
2692 for (i = 0; i < nb_fd; i++) {
2693 revents = LTTNG_POLL_GETEV(&events, i);
2694 pollfd = LTTNG_POLL_GETFD(&events, i);
2695
2696 /* Just don't waste time if no returned events for the fd */
2697 if (!revents) {
2698 continue;
2699 }
2700 if (pollfd == ctx->consumer_channel_pipe[0]) {
2701 if (revents & (LPOLLERR | LPOLLHUP)) {
2702 DBG("Channel thread pipe hung up");
2703 /*
2704 * Remove the pipe from the poll set and continue the loop
2705 * since their might be data to consume.
2706 */
2707 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2708 continue;
2709 } else if (revents & LPOLLIN) {
2710 enum consumer_channel_action action;
a0cbdd2e 2711 uint64_t key;
d8ef542d 2712
a0cbdd2e 2713 ret = read_channel_pipe(ctx, &chan, &key, &action);
d8ef542d
MD
2714 if (ret <= 0) {
2715 ERR("Error reading channel pipe");
2716 continue;
2717 }
2718
2719 switch (action) {
2720 case CONSUMER_CHANNEL_ADD:
2721 DBG("Adding channel %d to poll set",
2722 chan->wait_fd);
2723
2724 lttng_ht_node_init_u64(&chan->wait_fd_node,
2725 chan->wait_fd);
c7260a81 2726 rcu_read_lock();
d8ef542d
MD
2727 lttng_ht_add_unique_u64(channel_ht,
2728 &chan->wait_fd_node);
c7260a81 2729 rcu_read_unlock();
d8ef542d
MD
2730 /* Add channel to the global poll events list */
2731 lttng_poll_add(&events, chan->wait_fd,
2732 LPOLLIN | LPOLLPRI);
2733 break;
a0cbdd2e
MD
2734 case CONSUMER_CHANNEL_DEL:
2735 {
f2a444f1
DG
2736 struct lttng_consumer_stream *stream, *stmp;
2737
c7260a81 2738 rcu_read_lock();
a0cbdd2e
MD
2739 chan = consumer_find_channel(key);
2740 if (!chan) {
c7260a81 2741 rcu_read_unlock();
a0cbdd2e
MD
2742 ERR("UST consumer get channel key %" PRIu64 " not found for del channel", key);
2743 break;
2744 }
2745 lttng_poll_del(&events, chan->wait_fd);
f623cc0b 2746 iter.iter.node = &chan->wait_fd_node.node;
a0cbdd2e
MD
2747 ret = lttng_ht_del(channel_ht, &iter);
2748 assert(ret == 0);
2749 consumer_close_channel_streams(chan);
2750
f2a444f1
DG
2751 switch (consumer_data.type) {
2752 case LTTNG_CONSUMER_KERNEL:
2753 break;
2754 case LTTNG_CONSUMER32_UST:
2755 case LTTNG_CONSUMER64_UST:
2756 /* Delete streams that might have been left in the stream list. */
2757 cds_list_for_each_entry_safe(stream, stmp, &chan->streams.head,
2758 send_node) {
2759 cds_list_del(&stream->send_node);
2760 lttng_ustconsumer_del_stream(stream);
2761 uatomic_sub(&stream->chan->refcount, 1);
2762 assert(&chan->refcount);
2763 free(stream);
2764 }
2765 break;
2766 default:
2767 ERR("Unknown consumer_data type");
2768 assert(0);
2769 }
2770
a0cbdd2e
MD
2771 /*
2772 * Release our own refcount. Force channel deletion even if
2773 * streams were not initialized.
2774 */
2775 if (!uatomic_sub_return(&chan->refcount, 1)) {
2776 consumer_del_channel(chan);
2777 }
c7260a81 2778 rcu_read_unlock();
a0cbdd2e
MD
2779 goto restart;
2780 }
d8ef542d
MD
2781 case CONSUMER_CHANNEL_QUIT:
2782 /*
2783 * Remove the pipe from the poll set and continue the loop
2784 * since their might be data to consume.
2785 */
2786 lttng_poll_del(&events, ctx->consumer_channel_pipe[0]);
2787 continue;
2788 default:
2789 ERR("Unknown action");
2790 break;
2791 }
2792 }
2793
2794 /* Handle other stream */
2795 continue;
2796 }
2797
2798 rcu_read_lock();
2799 {
2800 uint64_t tmp_id = (uint64_t) pollfd;
2801
2802 lttng_ht_lookup(channel_ht, &tmp_id, &iter);
2803 }
2804 node = lttng_ht_iter_get_node_u64(&iter);
2805 assert(node);
2806
2807 chan = caa_container_of(node, struct lttng_consumer_channel,
2808 wait_fd_node);
2809
2810 /* Check for error event */
2811 if (revents & (LPOLLERR | LPOLLHUP)) {
2812 DBG("Channel fd %d is hup|err.", pollfd);
2813
2814 lttng_poll_del(&events, chan->wait_fd);
2815 ret = lttng_ht_del(channel_ht, &iter);
2816 assert(ret == 0);
2817 consumer_close_channel_streams(chan);
f2ad556d
MD
2818
2819 /* Release our own refcount */
2820 if (!uatomic_sub_return(&chan->refcount, 1)
2821 && !uatomic_read(&chan->nb_init_stream_left)) {
2822 consumer_del_channel(chan);
2823 }
d8ef542d
MD
2824 }
2825
2826 /* Release RCU lock for the channel looked up */
2827 rcu_read_unlock();
2828 }
2829 }
2830
2831end:
2832 lttng_poll_clean(&events);
2833end_poll:
2834 destroy_channel_ht(channel_ht);
2835end_ht:
2836 DBG("Channel poll thread exiting");
2837 rcu_unregister_thread();
2838 return NULL;
2839}
2840
331744e3
JD
2841static int set_metadata_socket(struct lttng_consumer_local_data *ctx,
2842 struct pollfd *sockpoll, int client_socket)
2843{
2844 int ret;
2845
2846 assert(ctx);
2847 assert(sockpoll);
2848
2849 if (lttng_consumer_poll_socket(sockpoll) < 0) {
2850 ret = -1;
2851 goto error;
2852 }
2853 DBG("Metadata connection on client_socket");
2854
2855 /* Blocking call, waiting for transmission */
2856 ctx->consumer_metadata_socket = lttcomm_accept_unix_sock(client_socket);
2857 if (ctx->consumer_metadata_socket < 0) {
2858 WARN("On accept metadata");
2859 ret = -1;
2860 goto error;
2861 }
2862 ret = 0;
2863
2864error:
2865 return ret;
2866}
2867
3bd1e081
MD
2868/*
2869 * This thread listens on the consumerd socket and receives the file
2870 * descriptors from the session daemon.
2871 */
7d980def 2872void *consumer_thread_sessiond_poll(void *data)
3bd1e081 2873{
d96f09c6 2874 int sock = -1, client_socket, ret;
3bd1e081
MD
2875 /*
2876 * structure to poll for incoming data on communication socket avoids
2877 * making blocking sockets.
2878 */
2879 struct pollfd consumer_sockpoll[2];
2880 struct lttng_consumer_local_data *ctx = data;
2881
e7b994a3
DG
2882 rcu_register_thread();
2883
3bd1e081
MD
2884 DBG("Creating command socket %s", ctx->consumer_command_sock_path);
2885 unlink(ctx->consumer_command_sock_path);
2886 client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path);
2887 if (client_socket < 0) {
2888 ERR("Cannot create command socket");
2889 goto end;
2890 }
2891
2892 ret = lttcomm_listen_unix_sock(client_socket);
2893 if (ret < 0) {
2894 goto end;
2895 }
2896
32258573 2897 DBG("Sending ready command to lttng-sessiond");
f73fabfd 2898 ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY);
3bd1e081
MD
2899 /* return < 0 on error, but == 0 is not fatal */
2900 if (ret < 0) {
32258573 2901 ERR("Error sending ready command to lttng-sessiond");
3bd1e081
MD
2902 goto end;
2903 }
2904
3bd1e081
MD
2905 /* prepare the FDs to poll : to client socket and the should_quit pipe */
2906 consumer_sockpoll[0].fd = ctx->consumer_should_quit[0];
2907 consumer_sockpoll[0].events = POLLIN | POLLPRI;
2908 consumer_sockpoll[1].fd = client_socket;
2909 consumer_sockpoll[1].events = POLLIN | POLLPRI;
2910
2911 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
2912 goto end;
2913 }
2914 DBG("Connection on client_socket");
2915
2916 /* Blocking call, waiting for transmission */
2917 sock = lttcomm_accept_unix_sock(client_socket);
534d2592 2918 if (sock < 0) {
3bd1e081
MD
2919 WARN("On accept");
2920 goto end;
2921 }
3bd1e081 2922
331744e3
JD
2923 /*
2924 * Setup metadata socket which is the second socket connection on the
2925 * command unix socket.
2926 */
2927 ret = set_metadata_socket(ctx, consumer_sockpoll, client_socket);
2928 if (ret < 0) {
2929 goto end;
2930 }
2931
d96f09c6
DG
2932 /* This socket is not useful anymore. */
2933 ret = close(client_socket);
2934 if (ret < 0) {
2935 PERROR("close client_socket");
2936 }
2937 client_socket = -1;
2938
3bd1e081
MD
2939 /* update the polling structure to poll on the established socket */
2940 consumer_sockpoll[1].fd = sock;
2941 consumer_sockpoll[1].events = POLLIN | POLLPRI;
2942
2943 while (1) {
2944 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
2945 goto end;
2946 }
2947 DBG("Incoming command on sock");
2948 ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll);
2949 if (ret == -ENOENT) {
2950 DBG("Received STOP command");
2951 goto end;
2952 }
4cbc1a04
DG
2953 if (ret <= 0) {
2954 /*
2955 * This could simply be a session daemon quitting. Don't output
2956 * ERR() here.
2957 */
2958 DBG("Communication interrupted on command socket");
3bd1e081
MD
2959 goto end;
2960 }
2961 if (consumer_quit) {
2962 DBG("consumer_thread_receive_fds received quit from signal");
2963 goto end;
2964 }
ffe60014 2965 DBG("received command on sock");
3bd1e081
MD
2966 }
2967end:
ffe60014 2968 DBG("Consumer thread sessiond poll exiting");
3bd1e081 2969
d88aee68
DG
2970 /*
2971 * Close metadata streams since the producer is the session daemon which
2972 * just died.
2973 *
2974 * NOTE: for now, this only applies to the UST tracer.
2975 */
2976 lttng_consumer_close_metadata();
2977
3bd1e081
MD
2978 /*
2979 * when all fds have hung up, the polling thread
2980 * can exit cleanly
2981 */
2982 consumer_quit = 1;
2983
04fdd819 2984 /*
c869f647 2985 * Notify the data poll thread to poll back again and test the
8994307f 2986 * consumer_quit state that we just set so to quit gracefully.
04fdd819 2987 */
acdb9057 2988 notify_thread_lttng_pipe(ctx->consumer_data_pipe);
c869f647 2989
a0cbdd2e 2990 notify_channel_pipe(ctx, NULL, -1, CONSUMER_CHANNEL_QUIT);
d8ef542d 2991
d96f09c6
DG
2992 /* Cleaning up possibly open sockets. */
2993 if (sock >= 0) {
2994 ret = close(sock);
2995 if (ret < 0) {
2996 PERROR("close sock sessiond poll");
2997 }
2998 }
2999 if (client_socket >= 0) {
38476d24 3000 ret = close(client_socket);
d96f09c6
DG
3001 if (ret < 0) {
3002 PERROR("close client_socket sessiond poll");
3003 }
3004 }
3005
e7b994a3 3006 rcu_unregister_thread();
3bd1e081
MD
3007 return NULL;
3008}
d41f73b7 3009
4078b776 3010ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream,
d41f73b7
MD
3011 struct lttng_consumer_local_data *ctx)
3012{
74251bb8
DG
3013 ssize_t ret;
3014
3015 pthread_mutex_lock(&stream->lock);
3016
d41f73b7
MD
3017 switch (consumer_data.type) {
3018 case LTTNG_CONSUMER_KERNEL:
74251bb8
DG
3019 ret = lttng_kconsumer_read_subbuffer(stream, ctx);
3020 break;
7753dea8
MD
3021 case LTTNG_CONSUMER32_UST:
3022 case LTTNG_CONSUMER64_UST:
74251bb8
DG
3023 ret = lttng_ustconsumer_read_subbuffer(stream, ctx);
3024 break;
d41f73b7
MD
3025 default:
3026 ERR("Unknown consumer_data type");
3027 assert(0);
74251bb8
DG
3028 ret = -ENOSYS;
3029 break;
d41f73b7 3030 }
74251bb8
DG
3031
3032 pthread_mutex_unlock(&stream->lock);
3033 return ret;
d41f73b7
MD
3034}
3035
3036int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream)
3037{
3038 switch (consumer_data.type) {
3039 case LTTNG_CONSUMER_KERNEL:
3040 return lttng_kconsumer_on_recv_stream(stream);
7753dea8
MD
3041 case LTTNG_CONSUMER32_UST:
3042 case LTTNG_CONSUMER64_UST:
d41f73b7
MD
3043 return lttng_ustconsumer_on_recv_stream(stream);
3044 default:
3045 ERR("Unknown consumer_data type");
3046 assert(0);
3047 return -ENOSYS;
3048 }
3049}
e4421fec
DG
3050
3051/*
3052 * Allocate and set consumer data hash tables.
3053 */
3054void lttng_consumer_init(void)
3055{
d88aee68
DG
3056 consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3057 consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
3058 consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
d8ef542d 3059 consumer_data.stream_per_chan_id_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
e4421fec 3060}
7735ef9e
DG
3061
3062/*
3063 * Process the ADD_RELAYD command receive by a consumer.
3064 *
3065 * This will create a relayd socket pair and add it to the relayd hash table.
3066 * The caller MUST acquire a RCU read side lock before calling it.
3067 */
da009f2c 3068int consumer_add_relayd_socket(uint64_t net_seq_idx, int sock_type,
7735ef9e 3069 struct lttng_consumer_local_data *ctx, int sock,
6151a90f 3070 struct pollfd *consumer_sockpoll,
30319bcb 3071 struct lttcomm_relayd_sock *relayd_sock, uint64_t sessiond_id)
7735ef9e 3072{
cd2b09ed 3073 int fd = -1, ret = -1, relayd_created = 0;
f50f23d9 3074 enum lttng_error_code ret_code = LTTNG_OK;
d4298c99 3075 struct consumer_relayd_sock_pair *relayd = NULL;
7735ef9e 3076
6151a90f
JD
3077 assert(ctx);
3078 assert(relayd_sock);
3079
da009f2c 3080 DBG("Consumer adding relayd socket (idx: %" PRIu64 ")", net_seq_idx);
7735ef9e
DG
3081
3082 /* Get relayd reference if exists. */
3083 relayd = consumer_find_relayd(net_seq_idx);
3084 if (relayd == NULL) {
da009f2c 3085 assert(sock_type == LTTNG_STREAM_CONTROL);
7735ef9e
DG
3086 /* Not found. Allocate one. */
3087 relayd = consumer_allocate_relayd_sock_pair(net_seq_idx);
3088 if (relayd == NULL) {
0d08d75e 3089 ret = -ENOMEM;
618a6a28
MD
3090 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
3091 goto error;
0d08d75e 3092 } else {
30319bcb 3093 relayd->sessiond_session_id = sessiond_id;
0d08d75e 3094 relayd_created = 1;
7735ef9e 3095 }
0d08d75e
DG
3096
3097 /*
3098 * This code path MUST continue to the consumer send status message to
3099 * we can notify the session daemon and continue our work without
3100 * killing everything.
3101 */
da009f2c
MD
3102 } else {
3103 /*
3104 * relayd key should never be found for control socket.
3105 */
3106 assert(sock_type != LTTNG_STREAM_CONTROL);
0d08d75e
DG
3107 }
3108
3109 /* First send a status message before receiving the fds. */
618a6a28
MD
3110 ret = consumer_send_status_msg(sock, LTTNG_OK);
3111 if (ret < 0) {
0d08d75e 3112 /* Somehow, the session daemon is not responding anymore. */
618a6a28
MD
3113 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3114 goto error_nosignal;
7735ef9e
DG
3115 }
3116
3117 /* Poll on consumer socket. */
3118 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
0d08d75e 3119 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
7735ef9e 3120 ret = -EINTR;
618a6a28 3121 goto error_nosignal;
7735ef9e
DG
3122 }
3123
3124 /* Get relayd socket from session daemon */
3125 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
3126 if (ret != sizeof(fd)) {
7735ef9e 3127 ret = -1;
4028eeb9 3128 fd = -1; /* Just in case it gets set with an invalid value. */
0d08d75e
DG
3129
3130 /*
3131 * Failing to receive FDs might indicate a major problem such as
3132 * reaching a fd limit during the receive where the kernel returns a
3133 * MSG_CTRUNC and fails to cleanup the fd in the queue. Any case, we
3134 * don't take any chances and stop everything.
3135 *
3136 * XXX: Feature request #558 will fix that and avoid this possible
3137 * issue when reaching the fd limit.
3138 */
3139 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
618a6a28 3140 ret_code = LTTCOMM_CONSUMERD_ERROR_RECV_FD;
f50f23d9
DG
3141 goto error;
3142 }
3143
7735ef9e
DG
3144 /* Copy socket information and received FD */
3145 switch (sock_type) {
3146 case LTTNG_STREAM_CONTROL:
3147 /* Copy received lttcomm socket */
6151a90f
JD
3148 lttcomm_copy_sock(&relayd->control_sock.sock, &relayd_sock->sock);
3149 ret = lttcomm_create_sock(&relayd->control_sock.sock);
4028eeb9 3150 /* Handle create_sock error. */
f66c074c 3151 if (ret < 0) {
618a6a28 3152 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
4028eeb9 3153 goto error;
f66c074c 3154 }
da009f2c
MD
3155 /*
3156 * Close the socket created internally by
3157 * lttcomm_create_sock, so we can replace it by the one
3158 * received from sessiond.
3159 */
3160 if (close(relayd->control_sock.sock.fd)) {
3161 PERROR("close");
3162 }
7735ef9e
DG
3163
3164 /* Assign new file descriptor */
6151a90f 3165 relayd->control_sock.sock.fd = fd;
4b29f1ce 3166 fd = -1; /* For error path */
6151a90f
JD
3167 /* Assign version values. */
3168 relayd->control_sock.major = relayd_sock->major;
3169 relayd->control_sock.minor = relayd_sock->minor;
c5b6f4f0
DG
3170
3171 /*
59e71485
DG
3172 * Create a session on the relayd and store the returned id. Lock the
3173 * control socket mutex if the relayd was NOT created before.
c5b6f4f0 3174 */
59e71485
DG
3175 if (!relayd_created) {
3176 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3177 }
c5b6f4f0 3178 ret = relayd_create_session(&relayd->control_sock,
f7079f67 3179 &relayd->relayd_session_id);
59e71485
DG
3180 if (!relayd_created) {
3181 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
3182 }
c5b6f4f0 3183 if (ret < 0) {
ffe60014
DG
3184 /*
3185 * Close all sockets of a relayd object. It will be freed if it was
3186 * created at the error code path or else it will be garbage
3187 * collect.
3188 */
3189 (void) relayd_close(&relayd->control_sock);
3190 (void) relayd_close(&relayd->data_sock);
618a6a28 3191 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
c5b6f4f0
DG
3192 goto error;
3193 }
3194
7735ef9e
DG
3195 break;
3196 case LTTNG_STREAM_DATA:
3197 /* Copy received lttcomm socket */
6151a90f
JD
3198 lttcomm_copy_sock(&relayd->data_sock.sock, &relayd_sock->sock);
3199 ret = lttcomm_create_sock(&relayd->data_sock.sock);
4028eeb9 3200 /* Handle create_sock error. */
f66c074c 3201 if (ret < 0) {
618a6a28 3202 ret_code = LTTCOMM_CONSUMERD_ENOMEM;
4028eeb9 3203 goto error;
f66c074c 3204 }
da009f2c
MD
3205 /*
3206 * Close the socket created internally by
3207 * lttcomm_create_sock, so we can replace it by the one
3208 * received from sessiond.
3209 */
3210 if (close(relayd->data_sock.sock.fd)) {
3211 PERROR("close");
3212 }
7735ef9e
DG
3213
3214 /* Assign new file descriptor */
6151a90f 3215 relayd->data_sock.sock.fd = fd;
4b29f1ce 3216 fd = -1; /* for eventual error paths */
6151a90f
JD
3217 /* Assign version values. */
3218 relayd->data_sock.major = relayd_sock->major;
3219 relayd->data_sock.minor = relayd_sock->minor;
7735ef9e
DG
3220 break;
3221 default:
3222 ERR("Unknown relayd socket type (%d)", sock_type);
59e71485 3223 ret = -1;
618a6a28 3224 ret_code = LTTCOMM_CONSUMERD_FATAL;
7735ef9e
DG
3225 goto error;
3226 }
3227
d88aee68 3228 DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)",
7735ef9e
DG
3229 sock_type == LTTNG_STREAM_CONTROL ? "control" : "data",
3230 relayd->net_seq_idx, fd);
3231
618a6a28
MD
3232 /* We successfully added the socket. Send status back. */
3233 ret = consumer_send_status_msg(sock, ret_code);
3234 if (ret < 0) {
3235 /* Somehow, the session daemon is not responding anymore. */
3236 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3237 goto error_nosignal;
3238 }
3239
7735ef9e
DG
3240 /*
3241 * Add relayd socket pair to consumer data hashtable. If object already
3242 * exists or on error, the function gracefully returns.
3243 */
d09e1200 3244 add_relayd(relayd);
7735ef9e
DG
3245
3246 /* All good! */
4028eeb9 3247 return 0;
7735ef9e
DG
3248
3249error:
618a6a28
MD
3250 if (consumer_send_status_msg(sock, ret_code) < 0) {
3251 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_FATAL);
3252 }
3253
3254error_nosignal:
4028eeb9
DG
3255 /* Close received socket if valid. */
3256 if (fd >= 0) {
3257 if (close(fd)) {
3258 PERROR("close received socket");
3259 }
3260 }
cd2b09ed
DG
3261
3262 if (relayd_created) {
cd2b09ed
DG
3263 free(relayd);
3264 }
3265
7735ef9e
DG
3266 return ret;
3267}
ca22feea 3268
4e9a4686
DG
3269/*
3270 * Try to lock the stream mutex.
3271 *
3272 * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
3273 */
3274static int stream_try_lock(struct lttng_consumer_stream *stream)
3275{
3276 int ret;
3277
3278 assert(stream);
3279
3280 /*
3281 * Try to lock the stream mutex. On failure, we know that the stream is
3282 * being used else where hence there is data still being extracted.
3283 */
3284 ret = pthread_mutex_trylock(&stream->lock);
3285 if (ret) {
3286 /* For both EBUSY and EINVAL error, the mutex is NOT locked. */
3287 ret = 0;
3288 goto end;
3289 }
3290
3291 ret = 1;
3292
3293end:
3294 return ret;
3295}
3296
f7079f67
DG
3297/*
3298 * Search for a relayd associated to the session id and return the reference.
3299 *
3300 * A rcu read side lock MUST be acquire before calling this function and locked
3301 * until the relayd object is no longer necessary.
3302 */
3303static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id)
3304{
3305 struct lttng_ht_iter iter;
f7079f67 3306 struct consumer_relayd_sock_pair *relayd = NULL;
f7079f67
DG
3307
3308 /* Iterate over all relayd since they are indexed by net_seq_idx. */
3309 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
3310 node.node) {
18261bd1
DG
3311 /*
3312 * Check by sessiond id which is unique here where the relayd session
3313 * id might not be when having multiple relayd.
3314 */
3315 if (relayd->sessiond_session_id == id) {
f7079f67 3316 /* Found the relayd. There can be only one per id. */
18261bd1 3317 goto found;
f7079f67
DG
3318 }
3319 }
3320
18261bd1
DG
3321 return NULL;
3322
3323found:
f7079f67
DG
3324 return relayd;
3325}
3326
ca22feea
DG
3327/*
3328 * Check if for a given session id there is still data needed to be extract
3329 * from the buffers.
3330 *
6d805429 3331 * Return 1 if data is pending or else 0 meaning ready to be read.
ca22feea 3332 */
6d805429 3333int consumer_data_pending(uint64_t id)
ca22feea
DG
3334{
3335 int ret;
3336 struct lttng_ht_iter iter;
3337 struct lttng_ht *ht;
3338 struct lttng_consumer_stream *stream;
f7079f67 3339 struct consumer_relayd_sock_pair *relayd = NULL;
6d805429 3340 int (*data_pending)(struct lttng_consumer_stream *);
ca22feea 3341
6d805429 3342 DBG("Consumer data pending command on session id %" PRIu64, id);
ca22feea 3343
6f6eda74 3344 rcu_read_lock();
ca22feea
DG
3345 pthread_mutex_lock(&consumer_data.lock);
3346
3347 switch (consumer_data.type) {
3348 case LTTNG_CONSUMER_KERNEL:
6d805429 3349 data_pending = lttng_kconsumer_data_pending;
ca22feea
DG
3350 break;
3351 case LTTNG_CONSUMER32_UST:
3352 case LTTNG_CONSUMER64_UST:
6d805429 3353 data_pending = lttng_ustconsumer_data_pending;
ca22feea
DG
3354 break;
3355 default:
3356 ERR("Unknown consumer data type");
3357 assert(0);
3358 }
3359
3360 /* Ease our life a bit */
3361 ht = consumer_data.stream_list_ht;
3362
f7079f67
DG
3363 relayd = find_relayd_by_session_id(id);
3364 if (relayd) {
3365 /* Send init command for data pending. */
3366 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3367 ret = relayd_begin_data_pending(&relayd->control_sock,
3368 relayd->relayd_session_id);
3369 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
3370 if (ret < 0) {
3371 /* Communication error thus the relayd so no data pending. */
3372 goto data_not_pending;
3373 }
3374 }
3375
c8f59ee5 3376 cds_lfht_for_each_entry_duplicate(ht->ht,
d88aee68
DG
3377 ht->hash_fct(&id, lttng_ht_seed),
3378 ht->match_fct, &id,
ca22feea 3379 &iter.iter, stream, node_session_id.node) {
4e9a4686
DG
3380 /* If this call fails, the stream is being used hence data pending. */
3381 ret = stream_try_lock(stream);
3382 if (!ret) {
f7079f67 3383 goto data_pending;
ca22feea 3384 }
ca22feea 3385
4e9a4686
DG
3386 /*
3387 * A removed node from the hash table indicates that the stream has
3388 * been deleted thus having a guarantee that the buffers are closed
3389 * on the consumer side. However, data can still be transmitted
3390 * over the network so don't skip the relayd check.
3391 */
3392 ret = cds_lfht_is_node_deleted(&stream->node.node);
3393 if (!ret) {
3394 /* Check the stream if there is data in the buffers. */
6d805429
DG
3395 ret = data_pending(stream);
3396 if (ret == 1) {
4e9a4686 3397 pthread_mutex_unlock(&stream->lock);
f7079f67 3398 goto data_pending;
4e9a4686
DG
3399 }
3400 }
3401
3402 /* Relayd check */
f7079f67 3403 if (relayd) {
c8f59ee5
DG
3404 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3405 if (stream->metadata_flag) {
ad7051c0
DG
3406 ret = relayd_quiescent_control(&relayd->control_sock,
3407 stream->relayd_stream_id);
c8f59ee5 3408 } else {
6d805429 3409 ret = relayd_data_pending(&relayd->control_sock,
39df6d9f
DG
3410 stream->relayd_stream_id,
3411 stream->next_net_seq_num - 1);
c8f59ee5
DG
3412 }
3413 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
6d805429 3414 if (ret == 1) {
4e9a4686 3415 pthread_mutex_unlock(&stream->lock);
f7079f67 3416 goto data_pending;
c8f59ee5
DG
3417 }
3418 }
4e9a4686 3419 pthread_mutex_unlock(&stream->lock);
c8f59ee5 3420 }
ca22feea 3421
f7079f67
DG
3422 if (relayd) {
3423 unsigned int is_data_inflight = 0;
3424
3425 /* Send init command for data pending. */
3426 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
3427 ret = relayd_end_data_pending(&relayd->control_sock,
3428 relayd->relayd_session_id, &is_data_inflight);
3429 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
bdd88757 3430 if (ret < 0) {
f7079f67
DG
3431 goto data_not_pending;
3432 }
bdd88757
DG
3433 if (is_data_inflight) {
3434 goto data_pending;
3435 }
f7079f67
DG
3436 }
3437
ca22feea 3438 /*
f7079f67
DG
3439 * Finding _no_ node in the hash table and no inflight data means that the
3440 * stream(s) have been removed thus data is guaranteed to be available for
3441 * analysis from the trace files.
ca22feea
DG
3442 */
3443
f7079f67 3444data_not_pending:
ca22feea
DG
3445 /* Data is available to be read by a viewer. */
3446 pthread_mutex_unlock(&consumer_data.lock);
c8f59ee5 3447 rcu_read_unlock();
6d805429 3448 return 0;
ca22feea 3449
f7079f67 3450data_pending:
ca22feea
DG
3451 /* Data is still being extracted from buffers. */
3452 pthread_mutex_unlock(&consumer_data.lock);
c8f59ee5 3453 rcu_read_unlock();
6d805429 3454 return 1;
ca22feea 3455}
f50f23d9
DG
3456
3457/*
3458 * Send a ret code status message to the sessiond daemon.
3459 *
3460 * Return the sendmsg() return value.
3461 */
3462int consumer_send_status_msg(int sock, int ret_code)
3463{
3464 struct lttcomm_consumer_status_msg msg;
3465
3466 msg.ret_code = ret_code;
3467
3468 return lttcomm_send_unix_sock(sock, &msg, sizeof(msg));
3469}
ffe60014
DG
3470
3471/*
3472 * Send a channel status message to the sessiond daemon.
3473 *
3474 * Return the sendmsg() return value.
3475 */
3476int consumer_send_status_channel(int sock,
3477 struct lttng_consumer_channel *channel)
3478{
3479 struct lttcomm_consumer_status_channel msg;
3480
3481 assert(sock >= 0);
3482
3483 if (!channel) {
3484 msg.ret_code = -LTTNG_ERR_UST_CHAN_FAIL;
3485 } else {
3486 msg.ret_code = LTTNG_OK;
3487 msg.key = channel->key;
3488 msg.stream_count = channel->streams.count;
3489 }
3490
3491 return lttcomm_send_unix_sock(sock, &msg, sizeof(msg));
3492}
5c786ded
JD
3493
3494/*
3495 * Using a maximum stream size with the produced and consumed position of a
3496 * stream, computes the new consumed position to be as close as possible to the
3497 * maximum possible stream size.
3498 *
3499 * If maximum stream size is lower than the possible buffer size (produced -
3500 * consumed), the consumed_pos given is returned untouched else the new value
3501 * is returned.
3502 */
3503unsigned long consumer_get_consumed_maxsize(unsigned long consumed_pos,
3504 unsigned long produced_pos, uint64_t max_stream_size)
3505{
3506 if (max_stream_size && max_stream_size < (produced_pos - consumed_pos)) {
3507 /* Offset from the produced position to get the latest buffers. */
3508 return produced_pos - max_stream_size;
3509 }
3510
3511 return consumed_pos;
3512}
This page took 0.278523 seconds and 4 git commands to generate.