Add channel wakeup fd to monitor close
[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>
3bd1e081 31
990570ed 32#include <common/common.h>
fb3a43a9
DG
33#include <common/utils.h>
34#include <common/compat/poll.h>
10a8a223 35#include <common/kernel-ctl/kernel-ctl.h>
00e2e675 36#include <common/sessiond-comm/relayd.h>
10a8a223
DG
37#include <common/sessiond-comm/sessiond-comm.h>
38#include <common/kernel-consumer/kernel-consumer.h>
00e2e675 39#include <common/relayd/relayd.h>
10a8a223
DG
40#include <common/ust-consumer/ust-consumer.h>
41
42#include "consumer.h"
3bd1e081
MD
43
44struct lttng_consumer_global_data consumer_data = {
3bd1e081
MD
45 .stream_count = 0,
46 .need_update = 1,
47 .type = LTTNG_CONSUMER_UNKNOWN,
48};
49
3bd1e081
MD
50/*
51 * Flag to inform the polling thread to quit when all fd hung up. Updated by
52 * the consumer_thread_receive_fds when it notices that all fds has hung up.
53 * Also updated by the signal handler (consumer_should_exit()). Read by the
54 * polling threads.
55 */
a98dae5f 56volatile int consumer_quit;
3bd1e081 57
43c34bc3 58/*
43c34bc3
DG
59 * Global hash table containing respectively metadata and data streams. The
60 * stream element in this ht should only be updated by the metadata poll thread
61 * for the metadata and the data poll thread for the data.
62 */
40dc48e0
DG
63static struct lttng_ht *metadata_ht;
64static struct lttng_ht *data_ht;
43c34bc3 65
8994307f
DG
66/*
67 * Notify a thread pipe to poll back again. This usually means that some global
68 * state has changed so we just send back the thread in a poll wait call.
69 */
70static void notify_thread_pipe(int wpipe)
71{
72 int ret;
73
74 do {
75 struct lttng_consumer_stream *null_stream = NULL;
76
77 ret = write(wpipe, &null_stream, sizeof(null_stream));
78 } while (ret < 0 && errno == EINTR);
79}
80
3bd1e081
MD
81/*
82 * Find a stream. The consumer_data.lock must be locked during this
83 * call.
84 */
d88aee68 85static struct lttng_consumer_stream *find_stream(uint64_t key,
8389e4f8 86 struct lttng_ht *ht)
3bd1e081 87{
e4421fec 88 struct lttng_ht_iter iter;
d88aee68 89 struct lttng_ht_node_u64 *node;
e4421fec 90 struct lttng_consumer_stream *stream = NULL;
3bd1e081 91
8389e4f8
DG
92 assert(ht);
93
d88aee68
DG
94 /* -1ULL keys are lookup failures */
95 if (key == (uint64_t) -1ULL) {
7ad0a0cb 96 return NULL;
7a57cf92 97 }
e4421fec 98
6065ceec
DG
99 rcu_read_lock();
100
d88aee68
DG
101 lttng_ht_lookup(ht, &key, &iter);
102 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
103 if (node != NULL) {
104 stream = caa_container_of(node, struct lttng_consumer_stream, node);
3bd1e081 105 }
e4421fec 106
6065ceec
DG
107 rcu_read_unlock();
108
e4421fec 109 return stream;
3bd1e081
MD
110}
111
ffe60014 112static void steal_stream_key(int key, struct lttng_ht *ht)
7ad0a0cb
MD
113{
114 struct lttng_consumer_stream *stream;
115
04253271 116 rcu_read_lock();
ffe60014 117 stream = find_stream(key, ht);
04253271 118 if (stream) {
d88aee68 119 stream->key = -1ULL;
04253271
MD
120 /*
121 * We don't want the lookup to match, but we still need
122 * to iterate on this stream when iterating over the hash table. Just
123 * change the node key.
124 */
d88aee68 125 stream->node.key = -1ULL;
04253271
MD
126 }
127 rcu_read_unlock();
7ad0a0cb
MD
128}
129
d56db448
DG
130/*
131 * Return a channel object for the given key.
132 *
133 * RCU read side lock MUST be acquired before calling this function and
134 * protects the channel ptr.
135 */
d88aee68 136struct lttng_consumer_channel *consumer_find_channel(uint64_t key)
3bd1e081 137{
e4421fec 138 struct lttng_ht_iter iter;
d88aee68 139 struct lttng_ht_node_u64 *node;
e4421fec 140 struct lttng_consumer_channel *channel = NULL;
3bd1e081 141
d88aee68
DG
142 /* -1ULL keys are lookup failures */
143 if (key == (uint64_t) -1ULL) {
7ad0a0cb 144 return NULL;
7a57cf92 145 }
e4421fec 146
d88aee68
DG
147 lttng_ht_lookup(consumer_data.channel_ht, &key, &iter);
148 node = lttng_ht_iter_get_node_u64(&iter);
e4421fec
DG
149 if (node != NULL) {
150 channel = caa_container_of(node, struct lttng_consumer_channel, node);
3bd1e081 151 }
e4421fec
DG
152
153 return channel;
3bd1e081
MD
154}
155
ffe60014 156static void free_stream_rcu(struct rcu_head *head)
7ad0a0cb 157{
d88aee68
DG
158 struct lttng_ht_node_u64 *node =
159 caa_container_of(head, struct lttng_ht_node_u64, head);
ffe60014
DG
160 struct lttng_consumer_stream *stream =
161 caa_container_of(node, struct lttng_consumer_stream, node);
7ad0a0cb 162
ffe60014 163 free(stream);
7ad0a0cb
MD
164}
165
ffe60014 166static void free_channel_rcu(struct rcu_head *head)
702b1ea4 167{
d88aee68
DG
168 struct lttng_ht_node_u64 *node =
169 caa_container_of(head, struct lttng_ht_node_u64, head);
ffe60014
DG
170 struct lttng_consumer_channel *channel =
171 caa_container_of(node, struct lttng_consumer_channel, node);
702b1ea4 172
ffe60014 173 free(channel);
702b1ea4
MD
174}
175
00e2e675
DG
176/*
177 * RCU protected relayd socket pair free.
178 */
ffe60014 179static void free_relayd_rcu(struct rcu_head *head)
00e2e675 180{
d88aee68
DG
181 struct lttng_ht_node_u64 *node =
182 caa_container_of(head, struct lttng_ht_node_u64, head);
00e2e675
DG
183 struct consumer_relayd_sock_pair *relayd =
184 caa_container_of(node, struct consumer_relayd_sock_pair, node);
185
8994307f
DG
186 /*
187 * Close all sockets. This is done in the call RCU since we don't want the
188 * socket fds to be reassigned thus potentially creating bad state of the
189 * relayd object.
190 *
191 * We do not have to lock the control socket mutex here since at this stage
192 * there is no one referencing to this relayd object.
193 */
194 (void) relayd_close(&relayd->control_sock);
195 (void) relayd_close(&relayd->data_sock);
196
00e2e675
DG
197 free(relayd);
198}
199
200/*
201 * Destroy and free relayd socket pair object.
202 *
203 * This function MUST be called with the consumer_data lock acquired.
204 */
d09e1200 205static void destroy_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
206{
207 int ret;
208 struct lttng_ht_iter iter;
209
173af62f
DG
210 if (relayd == NULL) {
211 return;
212 }
213
00e2e675
DG
214 DBG("Consumer destroy and close relayd socket pair");
215
216 iter.iter.node = &relayd->node.node;
217 ret = lttng_ht_del(consumer_data.relayd_ht, &iter);
173af62f 218 if (ret != 0) {
8994307f 219 /* We assume the relayd is being or is destroyed */
173af62f
DG
220 return;
221 }
00e2e675 222
00e2e675 223 /* RCU free() call */
ffe60014
DG
224 call_rcu(&relayd->node.head, free_relayd_rcu);
225}
226
227/*
228 * Remove a channel from the global list protected by a mutex. This function is
229 * also responsible for freeing its data structures.
230 */
231void consumer_del_channel(struct lttng_consumer_channel *channel)
232{
233 int ret;
234 struct lttng_ht_iter iter;
235
d88aee68 236 DBG("Consumer delete channel key %" PRIu64, channel->key);
ffe60014
DG
237
238 pthread_mutex_lock(&consumer_data.lock);
239
240 switch (consumer_data.type) {
241 case LTTNG_CONSUMER_KERNEL:
242 break;
243 case LTTNG_CONSUMER32_UST:
244 case LTTNG_CONSUMER64_UST:
245 lttng_ustconsumer_del_channel(channel);
246 break;
247 default:
248 ERR("Unknown consumer_data type");
249 assert(0);
250 goto end;
251 }
252
253 rcu_read_lock();
254 iter.iter.node = &channel->node.node;
255 ret = lttng_ht_del(consumer_data.channel_ht, &iter);
256 assert(!ret);
257 rcu_read_unlock();
258
259 call_rcu(&channel->node.head, free_channel_rcu);
260end:
261 pthread_mutex_unlock(&consumer_data.lock);
00e2e675
DG
262}
263
228b5bf7
DG
264/*
265 * Iterate over the relayd hash table and destroy each element. Finally,
266 * destroy the whole hash table.
267 */
268static void cleanup_relayd_ht(void)
269{
270 struct lttng_ht_iter iter;
271 struct consumer_relayd_sock_pair *relayd;
272
273 rcu_read_lock();
274
275 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
276 node.node) {
277 destroy_relayd(relayd);
278 }
279
280 lttng_ht_destroy(consumer_data.relayd_ht);
281
282 rcu_read_unlock();
283}
284
8994307f
DG
285/*
286 * Update the end point status of all streams having the given network sequence
287 * index (relayd index).
288 *
289 * It's atomically set without having the stream mutex locked which is fine
290 * because we handle the write/read race with a pipe wakeup for each thread.
291 */
292static void update_endpoint_status_by_netidx(int net_seq_idx,
293 enum consumer_endpoint_status status)
294{
295 struct lttng_ht_iter iter;
296 struct lttng_consumer_stream *stream;
297
298 DBG("Consumer set delete flag on stream by idx %d", net_seq_idx);
299
300 rcu_read_lock();
301
302 /* Let's begin with metadata */
303 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
304 if (stream->net_seq_idx == net_seq_idx) {
305 uatomic_set(&stream->endpoint_status, status);
306 DBG("Delete flag set to metadata stream %d", stream->wait_fd);
307 }
308 }
309
310 /* Follow up by the data streams */
311 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
312 if (stream->net_seq_idx == net_seq_idx) {
313 uatomic_set(&stream->endpoint_status, status);
314 DBG("Delete flag set to data stream %d", stream->wait_fd);
315 }
316 }
317 rcu_read_unlock();
318}
319
320/*
321 * Cleanup a relayd object by flagging every associated streams for deletion,
322 * destroying the object meaning removing it from the relayd hash table,
323 * closing the sockets and freeing the memory in a RCU call.
324 *
325 * If a local data context is available, notify the threads that the streams'
326 * state have changed.
327 */
328static void cleanup_relayd(struct consumer_relayd_sock_pair *relayd,
329 struct lttng_consumer_local_data *ctx)
330{
331 int netidx;
332
333 assert(relayd);
334
9617607b
DG
335 DBG("Cleaning up relayd sockets");
336
8994307f
DG
337 /* Save the net sequence index before destroying the object */
338 netidx = relayd->net_seq_idx;
339
340 /*
341 * Delete the relayd from the relayd hash table, close the sockets and free
342 * the object in a RCU call.
343 */
344 destroy_relayd(relayd);
345
346 /* Set inactive endpoint to all streams */
347 update_endpoint_status_by_netidx(netidx, CONSUMER_ENDPOINT_INACTIVE);
348
349 /*
350 * With a local data context, notify the threads that the streams' state
351 * have changed. The write() action on the pipe acts as an "implicit"
352 * memory barrier ordering the updates of the end point status from the
353 * read of this status which happens AFTER receiving this notify.
354 */
355 if (ctx) {
356 notify_thread_pipe(ctx->consumer_data_pipe[1]);
357 notify_thread_pipe(ctx->consumer_metadata_pipe[1]);
358 }
359}
360
a6ba4fe1
DG
361/*
362 * Flag a relayd socket pair for destruction. Destroy it if the refcount
363 * reaches zero.
364 *
365 * RCU read side lock MUST be aquired before calling this function.
366 */
367void consumer_flag_relayd_for_destroy(struct consumer_relayd_sock_pair *relayd)
368{
369 assert(relayd);
370
371 /* Set destroy flag for this object */
372 uatomic_set(&relayd->destroy_flag, 1);
373
374 /* Destroy the relayd if refcount is 0 */
375 if (uatomic_read(&relayd->refcount) == 0) {
d09e1200 376 destroy_relayd(relayd);
a6ba4fe1
DG
377 }
378}
379
3bd1e081
MD
380/*
381 * Remove a stream from the global list protected by a mutex. This
382 * function is also responsible for freeing its data structures.
383 */
e316aad5
DG
384void consumer_del_stream(struct lttng_consumer_stream *stream,
385 struct lttng_ht *ht)
3bd1e081
MD
386{
387 int ret;
e4421fec 388 struct lttng_ht_iter iter;
3bd1e081 389 struct lttng_consumer_channel *free_chan = NULL;
00e2e675
DG
390 struct consumer_relayd_sock_pair *relayd;
391
392 assert(stream);
3bd1e081 393
8994307f
DG
394 DBG("Consumer del stream %d", stream->wait_fd);
395
e316aad5
DG
396 if (ht == NULL) {
397 /* Means the stream was allocated but not successfully added */
ffe60014 398 goto free_stream_rcu;
e316aad5
DG
399 }
400
3bd1e081 401 pthread_mutex_lock(&consumer_data.lock);
74251bb8 402 pthread_mutex_lock(&stream->lock);
3bd1e081
MD
403
404 switch (consumer_data.type) {
405 case LTTNG_CONSUMER_KERNEL:
406 if (stream->mmap_base != NULL) {
407 ret = munmap(stream->mmap_base, stream->mmap_len);
408 if (ret != 0) {
7a57cf92 409 PERROR("munmap");
3bd1e081
MD
410 }
411 }
412 break;
7753dea8
MD
413 case LTTNG_CONSUMER32_UST:
414 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
415 lttng_ustconsumer_del_stream(stream);
416 break;
417 default:
418 ERR("Unknown consumer_data type");
419 assert(0);
420 goto end;
421 }
422
6065ceec 423 rcu_read_lock();
04253271 424 iter.iter.node = &stream->node.node;
e316aad5 425 ret = lttng_ht_del(ht, &iter);
04253271 426 assert(!ret);
ca22feea
DG
427
428 /* Remove node session id from the consumer_data stream ht */
429 iter.iter.node = &stream->node_session_id.node;
430 ret = lttng_ht_del(consumer_data.stream_list_ht, &iter);
431 assert(!ret);
6065ceec
DG
432 rcu_read_unlock();
433
50f8ae69 434 assert(consumer_data.stream_count > 0);
3bd1e081 435 consumer_data.stream_count--;
50f8ae69 436
3bd1e081 437 if (stream->out_fd >= 0) {
4c462e79
MD
438 ret = close(stream->out_fd);
439 if (ret) {
440 PERROR("close");
441 }
3bd1e081 442 }
00e2e675
DG
443
444 /* Check and cleanup relayd */
b0b335c8 445 rcu_read_lock();
00e2e675
DG
446 relayd = consumer_find_relayd(stream->net_seq_idx);
447 if (relayd != NULL) {
b0b335c8
MD
448 uatomic_dec(&relayd->refcount);
449 assert(uatomic_read(&relayd->refcount) >= 0);
173af62f 450
3f8e211f
DG
451 /* Closing streams requires to lock the control socket. */
452 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
173af62f
DG
453 ret = relayd_send_close_stream(&relayd->control_sock,
454 stream->relayd_stream_id,
455 stream->next_net_seq_num - 1);
3f8e211f 456 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
173af62f 457 if (ret < 0) {
a4b92340
DG
458 DBG("Unable to close stream on the relayd. Continuing");
459 /*
460 * Continue here. There is nothing we can do for the relayd.
461 * Chances are that the relayd has closed the socket so we just
462 * continue cleaning up.
463 */
173af62f
DG
464 }
465
466 /* Both conditions are met, we destroy the relayd. */
467 if (uatomic_read(&relayd->refcount) == 0 &&
468 uatomic_read(&relayd->destroy_flag)) {
d09e1200 469 destroy_relayd(relayd);
00e2e675 470 }
00e2e675 471 }
b0b335c8 472 rcu_read_unlock();
00e2e675 473
c30aaa51
MD
474 uatomic_dec(&stream->chan->refcount);
475 if (!uatomic_read(&stream->chan->refcount)
ffe60014 476 && !uatomic_read(&stream->chan->nb_init_stream_left)) {
3bd1e081 477 free_chan = stream->chan;
00e2e675
DG
478 }
479
3bd1e081
MD
480end:
481 consumer_data.need_update = 1;
8994307f 482 pthread_mutex_unlock(&stream->lock);
74251bb8 483 pthread_mutex_unlock(&consumer_data.lock);
3bd1e081 484
c30aaa51 485 if (free_chan) {
3bd1e081 486 consumer_del_channel(free_chan);
c30aaa51 487 }
e316aad5 488
ffe60014
DG
489free_stream_rcu:
490 call_rcu(&stream->node.head, free_stream_rcu);
3bd1e081
MD
491}
492
d88aee68
DG
493struct lttng_consumer_stream *consumer_allocate_stream(uint64_t channel_key,
494 uint64_t stream_key,
3bd1e081 495 enum lttng_consumer_stream_state state,
ffe60014 496 const char *channel_name,
6df2e2c9 497 uid_t uid,
00e2e675 498 gid_t gid,
ffe60014 499 int relayd_id,
53632229 500 uint64_t session_id,
ffe60014
DG
501 int cpu,
502 int *alloc_ret,
503 enum consumer_channel_type type)
3bd1e081 504{
ffe60014 505 int ret;
3bd1e081 506 struct lttng_consumer_stream *stream;
3bd1e081 507
effcf122 508 stream = zmalloc(sizeof(*stream));
3bd1e081 509 if (stream == NULL) {
7a57cf92 510 PERROR("malloc struct lttng_consumer_stream");
ffe60014 511 ret = -ENOMEM;
7a57cf92 512 goto end;
3bd1e081 513 }
7a57cf92 514
d56db448
DG
515 rcu_read_lock();
516
3bd1e081 517 stream->key = stream_key;
3bd1e081
MD
518 stream->out_fd = -1;
519 stream->out_fd_offset = 0;
520 stream->state = state;
6df2e2c9
MD
521 stream->uid = uid;
522 stream->gid = gid;
ffe60014 523 stream->net_seq_idx = relayd_id;
53632229 524 stream->session_id = session_id;
53632229 525 pthread_mutex_init(&stream->lock, NULL);
58b1f425 526
ffe60014
DG
527 /* If channel is the metadata, flag this stream as metadata. */
528 if (type == CONSUMER_CHANNEL_TYPE_METADATA) {
529 stream->metadata_flag = 1;
530 /* Metadata is flat out. */
531 strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name));
58b1f425 532 } else {
ffe60014
DG
533 /* Format stream name to <channel_name>_<cpu_number> */
534 ret = snprintf(stream->name, sizeof(stream->name), "%s_%d",
535 channel_name, cpu);
536 if (ret < 0) {
537 PERROR("snprintf stream name");
538 goto error;
539 }
58b1f425 540 }
c30aaa51 541
ffe60014 542 /* Key is always the wait_fd for streams. */
d88aee68 543 lttng_ht_node_init_u64(&stream->node, stream->key);
ffe60014 544
53632229 545 /* Init session id node with the stream session id */
d88aee68 546 lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id);
53632229 547
d88aee68 548 DBG3("Allocated stream %s (key %" PRIu64 ", relayd_id %" PRIu64 ", session_id %" PRIu64,
ffe60014 549 stream->name, stream->key, stream->net_seq_idx, stream->session_id);
d56db448
DG
550
551 rcu_read_unlock();
3bd1e081 552 return stream;
c80048c6
MD
553
554error:
d56db448 555 rcu_read_unlock();
c80048c6 556 free(stream);
7a57cf92 557end:
ffe60014
DG
558 if (alloc_ret) {
559 *alloc_ret = ret;
560 }
c80048c6 561 return NULL;
3bd1e081
MD
562}
563
564/*
565 * Add a stream to the global list protected by a mutex.
566 */
ffe60014 567static int add_stream(struct lttng_consumer_stream *stream,
43c34bc3 568 struct lttng_ht *ht)
3bd1e081
MD
569{
570 int ret = 0;
00e2e675 571 struct consumer_relayd_sock_pair *relayd;
3bd1e081 572
e316aad5 573 assert(stream);
43c34bc3 574 assert(ht);
c77fc10a 575
d88aee68 576 DBG3("Adding consumer stream %" PRIu64, stream->key);
e316aad5
DG
577
578 pthread_mutex_lock(&consumer_data.lock);
2e818a6a 579 pthread_mutex_lock(&stream->lock);
b0b335c8 580 rcu_read_lock();
e316aad5 581
43c34bc3 582 /* Steal stream identifier to avoid having streams with the same key */
ffe60014 583 steal_stream_key(stream->key, ht);
43c34bc3 584
d88aee68 585 lttng_ht_add_unique_u64(ht, &stream->node);
00e2e675 586
ca22feea
DG
587 /*
588 * Add stream to the stream_list_ht of the consumer data. No need to steal
589 * the key since the HT does not use it and we allow to add redundant keys
590 * into this table.
591 */
d88aee68 592 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 593
00e2e675
DG
594 /* Check and cleanup relayd */
595 relayd = consumer_find_relayd(stream->net_seq_idx);
596 if (relayd != NULL) {
b0b335c8 597 uatomic_inc(&relayd->refcount);
00e2e675
DG
598 }
599
e316aad5
DG
600 /* Update channel refcount once added without error(s). */
601 uatomic_inc(&stream->chan->refcount);
602
603 /*
ffe60014
DG
604 * When nb_init_stream_left reaches 0, we don't need to trigger any action
605 * in terms of destroying the associated channel, because the action that
e316aad5
DG
606 * causes the count to become 0 also causes a stream to be added. The
607 * channel deletion will thus be triggered by the following removal of this
608 * stream.
609 */
ffe60014
DG
610 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
611 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
612 }
613
614 /* Update consumer data once the node is inserted. */
3bd1e081
MD
615 consumer_data.stream_count++;
616 consumer_data.need_update = 1;
617
e316aad5 618 rcu_read_unlock();
2e818a6a 619 pthread_mutex_unlock(&stream->lock);
3bd1e081 620 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 621
3bd1e081
MD
622 return ret;
623}
624
00e2e675 625/*
3f8e211f
DG
626 * Add relayd socket to global consumer data hashtable. RCU read side lock MUST
627 * be acquired before calling this.
00e2e675 628 */
d09e1200 629static int add_relayd(struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
630{
631 int ret = 0;
d88aee68 632 struct lttng_ht_node_u64 *node;
00e2e675
DG
633 struct lttng_ht_iter iter;
634
ffe60014 635 assert(relayd);
00e2e675 636
00e2e675 637 lttng_ht_lookup(consumer_data.relayd_ht,
d88aee68
DG
638 &relayd->net_seq_idx, &iter);
639 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675 640 if (node != NULL) {
00e2e675
DG
641 goto end;
642 }
d88aee68 643 lttng_ht_add_unique_u64(consumer_data.relayd_ht, &relayd->node);
00e2e675 644
00e2e675
DG
645end:
646 return ret;
647}
648
649/*
650 * Allocate and return a consumer relayd socket.
651 */
652struct consumer_relayd_sock_pair *consumer_allocate_relayd_sock_pair(
653 int net_seq_idx)
654{
655 struct consumer_relayd_sock_pair *obj = NULL;
656
657 /* Negative net sequence index is a failure */
658 if (net_seq_idx < 0) {
659 goto error;
660 }
661
662 obj = zmalloc(sizeof(struct consumer_relayd_sock_pair));
663 if (obj == NULL) {
664 PERROR("zmalloc relayd sock");
665 goto error;
666 }
667
668 obj->net_seq_idx = net_seq_idx;
669 obj->refcount = 0;
173af62f 670 obj->destroy_flag = 0;
d88aee68 671 lttng_ht_node_init_u64(&obj->node, obj->net_seq_idx);
00e2e675
DG
672 pthread_mutex_init(&obj->ctrl_sock_mutex, NULL);
673
674error:
675 return obj;
676}
677
678/*
679 * Find a relayd socket pair in the global consumer data.
680 *
681 * Return the object if found else NULL.
b0b335c8
MD
682 * RCU read-side lock must be held across this call and while using the
683 * returned object.
00e2e675 684 */
d88aee68 685struct consumer_relayd_sock_pair *consumer_find_relayd(uint64_t key)
00e2e675
DG
686{
687 struct lttng_ht_iter iter;
d88aee68 688 struct lttng_ht_node_u64 *node;
00e2e675
DG
689 struct consumer_relayd_sock_pair *relayd = NULL;
690
691 /* Negative keys are lookup failures */
d88aee68 692 if (key == (uint64_t) -1ULL) {
00e2e675
DG
693 goto error;
694 }
695
d88aee68 696 lttng_ht_lookup(consumer_data.relayd_ht, &key,
00e2e675 697 &iter);
d88aee68 698 node = lttng_ht_iter_get_node_u64(&iter);
00e2e675
DG
699 if (node != NULL) {
700 relayd = caa_container_of(node, struct consumer_relayd_sock_pair, node);
701 }
702
00e2e675
DG
703error:
704 return relayd;
705}
706
707/*
708 * Handle stream for relayd transmission if the stream applies for network
709 * streaming where the net sequence index is set.
710 *
711 * Return destination file descriptor or negative value on error.
712 */
6197aea7 713static int write_relayd_stream_header(struct lttng_consumer_stream *stream,
1d4dfdef
DG
714 size_t data_size, unsigned long padding,
715 struct consumer_relayd_sock_pair *relayd)
00e2e675
DG
716{
717 int outfd = -1, ret;
00e2e675
DG
718 struct lttcomm_relayd_data_hdr data_hdr;
719
720 /* Safety net */
721 assert(stream);
6197aea7 722 assert(relayd);
00e2e675
DG
723
724 /* Reset data header */
725 memset(&data_hdr, 0, sizeof(data_hdr));
726
00e2e675
DG
727 if (stream->metadata_flag) {
728 /* Caller MUST acquire the relayd control socket lock */
729 ret = relayd_send_metadata(&relayd->control_sock, data_size);
730 if (ret < 0) {
731 goto error;
732 }
733
734 /* Metadata are always sent on the control socket. */
735 outfd = relayd->control_sock.fd;
736 } else {
737 /* Set header with stream information */
738 data_hdr.stream_id = htobe64(stream->relayd_stream_id);
739 data_hdr.data_size = htobe32(data_size);
1d4dfdef 740 data_hdr.padding_size = htobe32(padding);
39df6d9f
DG
741 /*
742 * Note that net_seq_num below is assigned with the *current* value of
743 * next_net_seq_num and only after that the next_net_seq_num will be
744 * increment. This is why when issuing a command on the relayd using
745 * this next value, 1 should always be substracted in order to compare
746 * the last seen sequence number on the relayd side to the last sent.
747 */
3604f373 748 data_hdr.net_seq_num = htobe64(stream->next_net_seq_num);
00e2e675
DG
749 /* Other fields are zeroed previously */
750
751 ret = relayd_send_data_hdr(&relayd->data_sock, &data_hdr,
752 sizeof(data_hdr));
753 if (ret < 0) {
754 goto error;
755 }
756
3604f373
DG
757 ++stream->next_net_seq_num;
758
00e2e675
DG
759 /* Set to go on data socket */
760 outfd = relayd->data_sock.fd;
761 }
762
763error:
764 return outfd;
765}
766
3bd1e081 767/*
ffe60014
DG
768 * Allocate and return a new lttng_consumer_channel object using the given key
769 * to initialize the hash table node.
770 *
771 * On error, return NULL.
3bd1e081 772 */
ffe60014
DG
773struct lttng_consumer_channel *consumer_allocate_channel(unsigned long key,
774 uint64_t session_id,
775 const char *pathname,
776 const char *name,
777 uid_t uid,
778 gid_t gid,
779 int relayd_id,
780 enum lttng_event_output output)
3bd1e081
MD
781{
782 struct lttng_consumer_channel *channel;
3bd1e081 783
276b26d1 784 channel = zmalloc(sizeof(*channel));
3bd1e081 785 if (channel == NULL) {
7a57cf92 786 PERROR("malloc struct lttng_consumer_channel");
3bd1e081
MD
787 goto end;
788 }
ffe60014
DG
789
790 channel->key = key;
3bd1e081 791 channel->refcount = 0;
ffe60014
DG
792 channel->session_id = session_id;
793 channel->uid = uid;
794 channel->gid = gid;
795 channel->relayd_id = relayd_id;
796 channel->output = output;
797
798 strncpy(channel->pathname, pathname, sizeof(channel->pathname));
799 channel->pathname[sizeof(channel->pathname) - 1] = '\0';
800
801 strncpy(channel->name, name, sizeof(channel->name));
802 channel->name[sizeof(channel->name) - 1] = '\0';
803
d88aee68 804 lttng_ht_node_init_u64(&channel->node, channel->key);
ffe60014
DG
805 CDS_INIT_LIST_HEAD(&channel->streams.head);
806
d88aee68 807 DBG("Allocated channel (key %" PRIu64 ")", channel->key)
3bd1e081 808
3bd1e081
MD
809end:
810 return channel;
811}
812
813/*
814 * Add a channel to the global list protected by a mutex.
815 */
816int consumer_add_channel(struct lttng_consumer_channel *channel)
817{
ffe60014 818 int ret = 0;
d88aee68 819 struct lttng_ht_node_u64 *node;
c77fc10a
DG
820 struct lttng_ht_iter iter;
821
3bd1e081 822 pthread_mutex_lock(&consumer_data.lock);
6065ceec 823 rcu_read_lock();
c77fc10a
DG
824
825 lttng_ht_lookup(consumer_data.channel_ht,
d88aee68
DG
826 &channel->key, &iter);
827 node = lttng_ht_iter_get_node_u64(&iter);
c77fc10a
DG
828 if (node != NULL) {
829 /* Channel already exist. Ignore the insertion */
d88aee68
DG
830 ERR("Consumer add channel key %" PRIu64 " already exists!",
831 channel->key);
ffe60014 832 ret = -1;
c77fc10a
DG
833 goto end;
834 }
835
d88aee68 836 lttng_ht_add_unique_u64(consumer_data.channel_ht, &channel->node);
c77fc10a
DG
837
838end:
6065ceec 839 rcu_read_unlock();
3bd1e081 840 pthread_mutex_unlock(&consumer_data.lock);
702b1ea4 841
ffe60014 842 return ret;
3bd1e081
MD
843}
844
845/*
846 * Allocate the pollfd structure and the local view of the out fds to avoid
847 * doing a lookup in the linked list and concurrency issues when writing is
848 * needed. Called with consumer_data.lock held.
849 *
850 * Returns the number of fds in the structures.
851 */
ffe60014
DG
852static int update_poll_array(struct lttng_consumer_local_data *ctx,
853 struct pollfd **pollfd, struct lttng_consumer_stream **local_stream,
854 struct lttng_ht *ht)
3bd1e081 855{
3bd1e081 856 int i = 0;
e4421fec
DG
857 struct lttng_ht_iter iter;
858 struct lttng_consumer_stream *stream;
3bd1e081 859
ffe60014
DG
860 assert(ctx);
861 assert(ht);
862 assert(pollfd);
863 assert(local_stream);
864
3bd1e081 865 DBG("Updating poll fd array");
481d6c57 866 rcu_read_lock();
43c34bc3 867 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
8994307f
DG
868 /*
869 * Only active streams with an active end point can be added to the
870 * poll set and local stream storage of the thread.
871 *
872 * There is a potential race here for endpoint_status to be updated
873 * just after the check. However, this is OK since the stream(s) will
874 * be deleted once the thread is notified that the end point state has
875 * changed where this function will be called back again.
876 */
877 if (stream->state != LTTNG_CONSUMER_ACTIVE_STREAM ||
79d4ffb7 878 stream->endpoint_status == CONSUMER_ENDPOINT_INACTIVE) {
3bd1e081
MD
879 continue;
880 }
e4421fec
DG
881 DBG("Active FD %d", stream->wait_fd);
882 (*pollfd)[i].fd = stream->wait_fd;
3bd1e081 883 (*pollfd)[i].events = POLLIN | POLLPRI;
e4421fec 884 local_stream[i] = stream;
3bd1e081
MD
885 i++;
886 }
481d6c57 887 rcu_read_unlock();
3bd1e081
MD
888
889 /*
50f8ae69 890 * Insert the consumer_data_pipe at the end of the array and don't
3bd1e081
MD
891 * increment i so nb_fd is the number of real FD.
892 */
50f8ae69 893 (*pollfd)[i].fd = ctx->consumer_data_pipe[0];
509bb1cf 894 (*pollfd)[i].events = POLLIN | POLLPRI;
3bd1e081
MD
895 return i;
896}
897
898/*
899 * Poll on the should_quit pipe and the command socket return -1 on error and
900 * should exit, 0 if data is available on the command socket
901 */
902int lttng_consumer_poll_socket(struct pollfd *consumer_sockpoll)
903{
904 int num_rdy;
905
88f2b785 906restart:
3bd1e081
MD
907 num_rdy = poll(consumer_sockpoll, 2, -1);
908 if (num_rdy == -1) {
88f2b785
MD
909 /*
910 * Restart interrupted system call.
911 */
912 if (errno == EINTR) {
913 goto restart;
914 }
7a57cf92 915 PERROR("Poll error");
3bd1e081
MD
916 goto exit;
917 }
509bb1cf 918 if (consumer_sockpoll[0].revents & (POLLIN | POLLPRI)) {
3bd1e081
MD
919 DBG("consumer_should_quit wake up");
920 goto exit;
921 }
922 return 0;
923
924exit:
925 return -1;
926}
927
928/*
929 * Set the error socket.
930 */
ffe60014
DG
931void lttng_consumer_set_error_sock(struct lttng_consumer_local_data *ctx,
932 int sock)
3bd1e081
MD
933{
934 ctx->consumer_error_socket = sock;
935}
936
937/*
938 * Set the command socket path.
939 */
3bd1e081
MD
940void lttng_consumer_set_command_sock_path(
941 struct lttng_consumer_local_data *ctx, char *sock)
942{
943 ctx->consumer_command_sock_path = sock;
944}
945
946/*
947 * Send return code to the session daemon.
948 * If the socket is not defined, we return 0, it is not a fatal error
949 */
ffe60014 950int lttng_consumer_send_error(struct lttng_consumer_local_data *ctx, int cmd)
3bd1e081
MD
951{
952 if (ctx->consumer_error_socket > 0) {
953 return lttcomm_send_unix_sock(ctx->consumer_error_socket, &cmd,
954 sizeof(enum lttcomm_sessiond_command));
955 }
956
957 return 0;
958}
959
960/*
228b5bf7
DG
961 * Close all the tracefiles and stream fds and MUST be called when all
962 * instances are destroyed i.e. when all threads were joined and are ended.
3bd1e081
MD
963 */
964void lttng_consumer_cleanup(void)
965{
e4421fec 966 struct lttng_ht_iter iter;
ffe60014 967 struct lttng_consumer_channel *channel;
6065ceec
DG
968
969 rcu_read_lock();
3bd1e081 970
ffe60014
DG
971 cds_lfht_for_each_entry(consumer_data.channel_ht->ht, &iter.iter, channel,
972 node.node) {
702b1ea4 973 consumer_del_channel(channel);
3bd1e081 974 }
6065ceec
DG
975
976 rcu_read_unlock();
d6ce1df2 977
d6ce1df2 978 lttng_ht_destroy(consumer_data.channel_ht);
228b5bf7
DG
979
980 cleanup_relayd_ht();
981
982 /*
983 * This HT contains streams that are freed by either the metadata thread or
984 * the data thread so we do *nothing* on the hash table and simply destroy
985 * it.
986 */
987 lttng_ht_destroy(consumer_data.stream_list_ht);
3bd1e081
MD
988}
989
990/*
991 * Called from signal handler.
992 */
993void lttng_consumer_should_exit(struct lttng_consumer_local_data *ctx)
994{
995 int ret;
996 consumer_quit = 1;
6f94560a
MD
997 do {
998 ret = write(ctx->consumer_should_quit[1], "4", 1);
999 } while (ret < 0 && errno == EINTR);
4cec016f 1000 if (ret < 0 || ret != 1) {
7a57cf92 1001 PERROR("write consumer quit");
3bd1e081 1002 }
ab1027f4
DG
1003
1004 DBG("Consumer flag that it should quit");
3bd1e081
MD
1005}
1006
00e2e675
DG
1007void lttng_consumer_sync_trace_file(struct lttng_consumer_stream *stream,
1008 off_t orig_offset)
3bd1e081
MD
1009{
1010 int outfd = stream->out_fd;
1011
1012 /*
1013 * This does a blocking write-and-wait on any page that belongs to the
1014 * subbuffer prior to the one we just wrote.
1015 * Don't care about error values, as these are just hints and ways to
1016 * limit the amount of page cache used.
1017 */
ffe60014 1018 if (orig_offset < stream->max_sb_size) {
3bd1e081
MD
1019 return;
1020 }
ffe60014
DG
1021 lttng_sync_file_range(outfd, orig_offset - stream->max_sb_size,
1022 stream->max_sb_size,
3bd1e081
MD
1023 SYNC_FILE_RANGE_WAIT_BEFORE
1024 | SYNC_FILE_RANGE_WRITE
1025 | SYNC_FILE_RANGE_WAIT_AFTER);
1026 /*
1027 * Give hints to the kernel about how we access the file:
1028 * POSIX_FADV_DONTNEED : we won't re-access data in a near future after
1029 * we write it.
1030 *
1031 * We need to call fadvise again after the file grows because the
1032 * kernel does not seem to apply fadvise to non-existing parts of the
1033 * file.
1034 *
1035 * Call fadvise _after_ having waited for the page writeback to
1036 * complete because the dirty page writeback semantic is not well
1037 * defined. So it can be expected to lead to lower throughput in
1038 * streaming.
1039 */
ffe60014
DG
1040 posix_fadvise(outfd, orig_offset - stream->max_sb_size,
1041 stream->max_sb_size, POSIX_FADV_DONTNEED);
3bd1e081
MD
1042}
1043
1044/*
1045 * Initialise the necessary environnement :
1046 * - create a new context
1047 * - create the poll_pipe
1048 * - create the should_quit pipe (for signal handler)
1049 * - create the thread pipe (for splice)
1050 *
1051 * Takes a function pointer as argument, this function is called when data is
1052 * available on a buffer. This function is responsible to do the
1053 * kernctl_get_next_subbuf, read the data with mmap or splice depending on the
1054 * buffer configuration and then kernctl_put_next_subbuf at the end.
1055 *
1056 * Returns a pointer to the new context or NULL on error.
1057 */
1058struct lttng_consumer_local_data *lttng_consumer_create(
1059 enum lttng_consumer_type type,
4078b776 1060 ssize_t (*buffer_ready)(struct lttng_consumer_stream *stream,
d41f73b7 1061 struct lttng_consumer_local_data *ctx),
3bd1e081
MD
1062 int (*recv_channel)(struct lttng_consumer_channel *channel),
1063 int (*recv_stream)(struct lttng_consumer_stream *stream),
1064 int (*update_stream)(int stream_key, uint32_t state))
1065{
1066 int ret, i;
1067 struct lttng_consumer_local_data *ctx;
1068
1069 assert(consumer_data.type == LTTNG_CONSUMER_UNKNOWN ||
1070 consumer_data.type == type);
1071 consumer_data.type = type;
1072
effcf122 1073 ctx = zmalloc(sizeof(struct lttng_consumer_local_data));
3bd1e081 1074 if (ctx == NULL) {
7a57cf92 1075 PERROR("allocating context");
3bd1e081
MD
1076 goto error;
1077 }
1078
1079 ctx->consumer_error_socket = -1;
1080 /* assign the callbacks */
1081 ctx->on_buffer_ready = buffer_ready;
1082 ctx->on_recv_channel = recv_channel;
1083 ctx->on_recv_stream = recv_stream;
1084 ctx->on_update_stream = update_stream;
1085
50f8ae69 1086 ret = pipe(ctx->consumer_data_pipe);
3bd1e081 1087 if (ret < 0) {
7a57cf92 1088 PERROR("Error creating poll pipe");
3bd1e081
MD
1089 goto error_poll_pipe;
1090 }
1091
04fdd819 1092 /* set read end of the pipe to non-blocking */
50f8ae69 1093 ret = fcntl(ctx->consumer_data_pipe[0], F_SETFL, O_NONBLOCK);
04fdd819 1094 if (ret < 0) {
7a57cf92 1095 PERROR("fcntl O_NONBLOCK");
04fdd819
MD
1096 goto error_poll_fcntl;
1097 }
1098
1099 /* set write end of the pipe to non-blocking */
50f8ae69 1100 ret = fcntl(ctx->consumer_data_pipe[1], F_SETFL, O_NONBLOCK);
04fdd819 1101 if (ret < 0) {
7a57cf92 1102 PERROR("fcntl O_NONBLOCK");
04fdd819
MD
1103 goto error_poll_fcntl;
1104 }
1105
3bd1e081
MD
1106 ret = pipe(ctx->consumer_should_quit);
1107 if (ret < 0) {
7a57cf92 1108 PERROR("Error creating recv pipe");
3bd1e081
MD
1109 goto error_quit_pipe;
1110 }
1111
1112 ret = pipe(ctx->consumer_thread_pipe);
1113 if (ret < 0) {
7a57cf92 1114 PERROR("Error creating thread pipe");
3bd1e081
MD
1115 goto error_thread_pipe;
1116 }
1117
fb3a43a9
DG
1118 ret = utils_create_pipe(ctx->consumer_metadata_pipe);
1119 if (ret < 0) {
1120 goto error_metadata_pipe;
1121 }
3bd1e081 1122
fb3a43a9
DG
1123 ret = utils_create_pipe(ctx->consumer_splice_metadata_pipe);
1124 if (ret < 0) {
1125 goto error_splice_pipe;
1126 }
1127
1128 return ctx;
3bd1e081 1129
fb3a43a9
DG
1130error_splice_pipe:
1131 utils_close_pipe(ctx->consumer_metadata_pipe);
1132error_metadata_pipe:
1133 utils_close_pipe(ctx->consumer_thread_pipe);
3bd1e081
MD
1134error_thread_pipe:
1135 for (i = 0; i < 2; i++) {
1136 int err;
1137
1138 err = close(ctx->consumer_should_quit[i]);
4c462e79
MD
1139 if (err) {
1140 PERROR("close");
1141 }
3bd1e081 1142 }
04fdd819 1143error_poll_fcntl:
3bd1e081
MD
1144error_quit_pipe:
1145 for (i = 0; i < 2; i++) {
1146 int err;
1147
50f8ae69 1148 err = close(ctx->consumer_data_pipe[i]);
4c462e79
MD
1149 if (err) {
1150 PERROR("close");
1151 }
3bd1e081
MD
1152 }
1153error_poll_pipe:
1154 free(ctx);
1155error:
1156 return NULL;
1157}
1158
1159/*
1160 * Close all fds associated with the instance and free the context.
1161 */
1162void lttng_consumer_destroy(struct lttng_consumer_local_data *ctx)
1163{
4c462e79
MD
1164 int ret;
1165
ab1027f4
DG
1166 DBG("Consumer destroying it. Closing everything.");
1167
4c462e79
MD
1168 ret = close(ctx->consumer_error_socket);
1169 if (ret) {
1170 PERROR("close");
1171 }
1172 ret = close(ctx->consumer_thread_pipe[0]);
1173 if (ret) {
1174 PERROR("close");
1175 }
1176 ret = close(ctx->consumer_thread_pipe[1]);
1177 if (ret) {
1178 PERROR("close");
1179 }
50f8ae69 1180 ret = close(ctx->consumer_data_pipe[0]);
4c462e79
MD
1181 if (ret) {
1182 PERROR("close");
1183 }
50f8ae69 1184 ret = close(ctx->consumer_data_pipe[1]);
4c462e79
MD
1185 if (ret) {
1186 PERROR("close");
1187 }
1188 ret = close(ctx->consumer_should_quit[0]);
1189 if (ret) {
1190 PERROR("close");
1191 }
1192 ret = close(ctx->consumer_should_quit[1]);
1193 if (ret) {
1194 PERROR("close");
1195 }
fb3a43a9
DG
1196 utils_close_pipe(ctx->consumer_splice_metadata_pipe);
1197
3bd1e081
MD
1198 unlink(ctx->consumer_command_sock_path);
1199 free(ctx);
1200}
1201
6197aea7
DG
1202/*
1203 * Write the metadata stream id on the specified file descriptor.
1204 */
1205static int write_relayd_metadata_id(int fd,
1206 struct lttng_consumer_stream *stream,
ffe60014 1207 struct consumer_relayd_sock_pair *relayd, unsigned long padding)
6197aea7
DG
1208{
1209 int ret;
1d4dfdef 1210 struct lttcomm_relayd_metadata_payload hdr;
6197aea7 1211
1d4dfdef
DG
1212 hdr.stream_id = htobe64(stream->relayd_stream_id);
1213 hdr.padding_size = htobe32(padding);
6197aea7 1214 do {
1d4dfdef 1215 ret = write(fd, (void *) &hdr, sizeof(hdr));
6197aea7 1216 } while (ret < 0 && errno == EINTR);
4cec016f 1217 if (ret < 0 || ret != sizeof(hdr)) {
d7b75ec8
DG
1218 /*
1219 * This error means that the fd's end is closed so ignore the perror
1220 * not to clubber the error output since this can happen in a normal
1221 * code path.
1222 */
1223 if (errno != EPIPE) {
1224 PERROR("write metadata stream id");
1225 }
1226 DBG3("Consumer failed to write relayd metadata id (errno: %d)", errno);
534d2592
DG
1227 /*
1228 * Set ret to a negative value because if ret != sizeof(hdr), we don't
1229 * handle writting the missing part so report that as an error and
1230 * don't lie to the caller.
1231 */
1232 ret = -1;
6197aea7
DG
1233 goto end;
1234 }
1d4dfdef
DG
1235 DBG("Metadata stream id %" PRIu64 " with padding %lu written before data",
1236 stream->relayd_stream_id, padding);
6197aea7
DG
1237
1238end:
1239 return ret;
1240}
1241
3bd1e081 1242/*
09e26845
DG
1243 * Mmap the ring buffer, read it and write the data to the tracefile. This is a
1244 * core function for writing trace buffers to either the local filesystem or
1245 * the network.
1246 *
79d4ffb7
DG
1247 * It must be called with the stream lock held.
1248 *
09e26845 1249 * Careful review MUST be put if any changes occur!
3bd1e081
MD
1250 *
1251 * Returns the number of bytes written
1252 */
4078b776 1253ssize_t lttng_consumer_on_read_subbuffer_mmap(
3bd1e081 1254 struct lttng_consumer_local_data *ctx,
1d4dfdef
DG
1255 struct lttng_consumer_stream *stream, unsigned long len,
1256 unsigned long padding)
3bd1e081 1257{
f02e1e8a 1258 unsigned long mmap_offset;
ffe60014 1259 void *mmap_base;
f02e1e8a
DG
1260 ssize_t ret = 0, written = 0;
1261 off_t orig_offset = stream->out_fd_offset;
1262 /* Default is on the disk */
1263 int outfd = stream->out_fd;
f02e1e8a 1264 struct consumer_relayd_sock_pair *relayd = NULL;
8994307f 1265 unsigned int relayd_hang_up = 0;
f02e1e8a
DG
1266
1267 /* RCU lock for the relayd pointer */
1268 rcu_read_lock();
1269
1270 /* Flag that the current stream if set for network streaming. */
1271 if (stream->net_seq_idx != -1) {
1272 relayd = consumer_find_relayd(stream->net_seq_idx);
1273 if (relayd == NULL) {
1274 goto end;
1275 }
1276 }
1277
1278 /* get the offset inside the fd to mmap */
3bd1e081
MD
1279 switch (consumer_data.type) {
1280 case LTTNG_CONSUMER_KERNEL:
ffe60014 1281 mmap_base = stream->mmap_base;
f02e1e8a
DG
1282 ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset);
1283 break;
7753dea8
MD
1284 case LTTNG_CONSUMER32_UST:
1285 case LTTNG_CONSUMER64_UST:
ffe60014
DG
1286 mmap_base = lttng_ustctl_get_mmap_base(stream);
1287 if (!mmap_base) {
1288 ERR("read mmap get mmap base for stream %s", stream->name);
1289 written = -1;
1290 goto end;
1291 }
1292 ret = lttng_ustctl_get_mmap_read_offset(stream, &mmap_offset);
f02e1e8a 1293 break;
3bd1e081
MD
1294 default:
1295 ERR("Unknown consumer_data type");
1296 assert(0);
1297 }
f02e1e8a
DG
1298 if (ret != 0) {
1299 errno = -ret;
1300 PERROR("tracer ctl get_mmap_read_offset");
1301 written = ret;
1302 goto end;
1303 }
b9182dd9 1304
f02e1e8a
DG
1305 /* Handle stream on the relayd if the output is on the network */
1306 if (relayd) {
1307 unsigned long netlen = len;
1308
1309 /*
1310 * Lock the control socket for the complete duration of the function
1311 * since from this point on we will use the socket.
1312 */
1313 if (stream->metadata_flag) {
1314 /* Metadata requires the control socket. */
1315 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1d4dfdef 1316 netlen += sizeof(struct lttcomm_relayd_metadata_payload);
f02e1e8a
DG
1317 }
1318
1d4dfdef 1319 ret = write_relayd_stream_header(stream, netlen, padding, relayd);
f02e1e8a
DG
1320 if (ret >= 0) {
1321 /* Use the returned socket. */
1322 outfd = ret;
1323
1324 /* Write metadata stream id before payload */
1325 if (stream->metadata_flag) {
1d4dfdef 1326 ret = write_relayd_metadata_id(outfd, stream, relayd, padding);
f02e1e8a 1327 if (ret < 0) {
f02e1e8a 1328 written = ret;
8994307f
DG
1329 /* Socket operation failed. We consider the relayd dead */
1330 if (ret == -EPIPE || ret == -EINVAL) {
1331 relayd_hang_up = 1;
1332 goto write_error;
1333 }
f02e1e8a
DG
1334 goto end;
1335 }
f02e1e8a 1336 }
8994307f
DG
1337 } else {
1338 /* Socket operation failed. We consider the relayd dead */
1339 if (ret == -EPIPE || ret == -EINVAL) {
1340 relayd_hang_up = 1;
1341 goto write_error;
1342 }
1343 /* Else, use the default set before which is the filesystem. */
f02e1e8a 1344 }
1d4dfdef
DG
1345 } else {
1346 /* No streaming, we have to set the len with the full padding */
1347 len += padding;
f02e1e8a
DG
1348 }
1349
1350 while (len > 0) {
1351 do {
ffe60014 1352 ret = write(outfd, mmap_base + mmap_offset, len);
f02e1e8a 1353 } while (ret < 0 && errno == EINTR);
1d4dfdef 1354 DBG("Consumer mmap write() ret %zd (len %lu)", ret, len);
f02e1e8a 1355 if (ret < 0) {
c5c45efa
DG
1356 /*
1357 * This is possible if the fd is closed on the other side (outfd)
1358 * or any write problem. It can be verbose a bit for a normal
1359 * execution if for instance the relayd is stopped abruptly. This
1360 * can happen so set this to a DBG statement.
1361 */
1362 DBG("Error in file write mmap");
f02e1e8a
DG
1363 if (written == 0) {
1364 written = ret;
1365 }
8994307f
DG
1366 /* Socket operation failed. We consider the relayd dead */
1367 if (errno == EPIPE || errno == EINVAL) {
1368 relayd_hang_up = 1;
1369 goto write_error;
1370 }
f02e1e8a
DG
1371 goto end;
1372 } else if (ret > len) {
77c7c900 1373 PERROR("Error in file write (ret %zd > len %lu)", ret, len);
f02e1e8a
DG
1374 written += ret;
1375 goto end;
1376 } else {
1377 len -= ret;
1378 mmap_offset += ret;
1379 }
f02e1e8a
DG
1380
1381 /* This call is useless on a socket so better save a syscall. */
1382 if (!relayd) {
1383 /* This won't block, but will start writeout asynchronously */
1384 lttng_sync_file_range(outfd, stream->out_fd_offset, ret,
1385 SYNC_FILE_RANGE_WRITE);
1386 stream->out_fd_offset += ret;
1387 }
1388 written += ret;
1389 }
1390 lttng_consumer_sync_trace_file(stream, orig_offset);
1391
8994307f
DG
1392write_error:
1393 /*
1394 * This is a special case that the relayd has closed its socket. Let's
1395 * cleanup the relayd object and all associated streams.
1396 */
1397 if (relayd && relayd_hang_up) {
1398 cleanup_relayd(relayd, ctx);
1399 }
1400
f02e1e8a
DG
1401end:
1402 /* Unlock only if ctrl socket used */
1403 if (relayd && stream->metadata_flag) {
1404 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1405 }
1406
1407 rcu_read_unlock();
1408 return written;
3bd1e081
MD
1409}
1410
1411/*
1412 * Splice the data from the ring buffer to the tracefile.
1413 *
79d4ffb7
DG
1414 * It must be called with the stream lock held.
1415 *
3bd1e081
MD
1416 * Returns the number of bytes spliced.
1417 */
4078b776 1418ssize_t lttng_consumer_on_read_subbuffer_splice(
3bd1e081 1419 struct lttng_consumer_local_data *ctx,
1d4dfdef
DG
1420 struct lttng_consumer_stream *stream, unsigned long len,
1421 unsigned long padding)
3bd1e081 1422{
f02e1e8a
DG
1423 ssize_t ret = 0, written = 0, ret_splice = 0;
1424 loff_t offset = 0;
1425 off_t orig_offset = stream->out_fd_offset;
1426 int fd = stream->wait_fd;
1427 /* Default is on the disk */
1428 int outfd = stream->out_fd;
f02e1e8a 1429 struct consumer_relayd_sock_pair *relayd = NULL;
fb3a43a9 1430 int *splice_pipe;
8994307f 1431 unsigned int relayd_hang_up = 0;
f02e1e8a 1432
3bd1e081
MD
1433 switch (consumer_data.type) {
1434 case LTTNG_CONSUMER_KERNEL:
f02e1e8a 1435 break;
7753dea8
MD
1436 case LTTNG_CONSUMER32_UST:
1437 case LTTNG_CONSUMER64_UST:
f02e1e8a 1438 /* Not supported for user space tracing */
3bd1e081
MD
1439 return -ENOSYS;
1440 default:
1441 ERR("Unknown consumer_data type");
1442 assert(0);
3bd1e081
MD
1443 }
1444
f02e1e8a
DG
1445 /* RCU lock for the relayd pointer */
1446 rcu_read_lock();
1447
1448 /* Flag that the current stream if set for network streaming. */
1449 if (stream->net_seq_idx != -1) {
1450 relayd = consumer_find_relayd(stream->net_seq_idx);
1451 if (relayd == NULL) {
1452 goto end;
1453 }
1454 }
1455
fb3a43a9
DG
1456 /*
1457 * Choose right pipe for splice. Metadata and trace data are handled by
1458 * different threads hence the use of two pipes in order not to race or
1459 * corrupt the written data.
1460 */
1461 if (stream->metadata_flag) {
1462 splice_pipe = ctx->consumer_splice_metadata_pipe;
1463 } else {
1464 splice_pipe = ctx->consumer_thread_pipe;
1465 }
1466
f02e1e8a 1467 /* Write metadata stream id before payload */
1d4dfdef
DG
1468 if (relayd) {
1469 int total_len = len;
f02e1e8a 1470
1d4dfdef
DG
1471 if (stream->metadata_flag) {
1472 /*
1473 * Lock the control socket for the complete duration of the function
1474 * since from this point on we will use the socket.
1475 */
1476 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1477
1478 ret = write_relayd_metadata_id(splice_pipe[1], stream, relayd,
1479 padding);
1480 if (ret < 0) {
1481 written = ret;
8994307f
DG
1482 /* Socket operation failed. We consider the relayd dead */
1483 if (ret == -EBADF) {
1484 WARN("Remote relayd disconnected. Stopping");
1485 relayd_hang_up = 1;
1486 goto write_error;
1487 }
1d4dfdef
DG
1488 goto end;
1489 }
1490
1491 total_len += sizeof(struct lttcomm_relayd_metadata_payload);
1492 }
1493
1494 ret = write_relayd_stream_header(stream, total_len, padding, relayd);
1495 if (ret >= 0) {
1496 /* Use the returned socket. */
1497 outfd = ret;
1498 } else {
8994307f
DG
1499 /* Socket operation failed. We consider the relayd dead */
1500 if (ret == -EBADF) {
1501 WARN("Remote relayd disconnected. Stopping");
1502 relayd_hang_up = 1;
1503 goto write_error;
1504 }
f02e1e8a
DG
1505 goto end;
1506 }
1d4dfdef
DG
1507 } else {
1508 /* No streaming, we have to set the len with the full padding */
1509 len += padding;
f02e1e8a
DG
1510 }
1511
1512 while (len > 0) {
1d4dfdef
DG
1513 DBG("splice chan to pipe offset %lu of len %lu (fd : %d, pipe: %d)",
1514 (unsigned long)offset, len, fd, splice_pipe[1]);
fb3a43a9 1515 ret_splice = splice(fd, &offset, splice_pipe[1], NULL, len,
f02e1e8a
DG
1516 SPLICE_F_MOVE | SPLICE_F_MORE);
1517 DBG("splice chan to pipe, ret %zd", ret_splice);
1518 if (ret_splice < 0) {
1519 PERROR("Error in relay splice");
1520 if (written == 0) {
1521 written = ret_splice;
1522 }
1523 ret = errno;
1524 goto splice_error;
1525 }
1526
1527 /* Handle stream on the relayd if the output is on the network */
1528 if (relayd) {
1529 if (stream->metadata_flag) {
1d4dfdef
DG
1530 size_t metadata_payload_size =
1531 sizeof(struct lttcomm_relayd_metadata_payload);
1532
f02e1e8a 1533 /* Update counter to fit the spliced data */
1d4dfdef
DG
1534 ret_splice += metadata_payload_size;
1535 len += metadata_payload_size;
f02e1e8a
DG
1536 /*
1537 * We do this so the return value can match the len passed as
1538 * argument to this function.
1539 */
1d4dfdef 1540 written -= metadata_payload_size;
f02e1e8a
DG
1541 }
1542 }
1543
1544 /* Splice data out */
fb3a43a9 1545 ret_splice = splice(splice_pipe[0], NULL, outfd, NULL,
f02e1e8a 1546 ret_splice, SPLICE_F_MOVE | SPLICE_F_MORE);
1d4dfdef 1547 DBG("Consumer splice pipe to file, ret %zd", ret_splice);
f02e1e8a
DG
1548 if (ret_splice < 0) {
1549 PERROR("Error in file splice");
1550 if (written == 0) {
1551 written = ret_splice;
1552 }
8994307f 1553 /* Socket operation failed. We consider the relayd dead */
00c8752b 1554 if (errno == EBADF || errno == EPIPE) {
8994307f
DG
1555 WARN("Remote relayd disconnected. Stopping");
1556 relayd_hang_up = 1;
1557 goto write_error;
1558 }
f02e1e8a
DG
1559 ret = errno;
1560 goto splice_error;
1561 } else if (ret_splice > len) {
1562 errno = EINVAL;
1563 PERROR("Wrote more data than requested %zd (len: %lu)",
1564 ret_splice, len);
1565 written += ret_splice;
1566 ret = errno;
1567 goto splice_error;
1568 }
1569 len -= ret_splice;
1570
1571 /* This call is useless on a socket so better save a syscall. */
1572 if (!relayd) {
1573 /* This won't block, but will start writeout asynchronously */
1574 lttng_sync_file_range(outfd, stream->out_fd_offset, ret_splice,
1575 SYNC_FILE_RANGE_WRITE);
1576 stream->out_fd_offset += ret_splice;
1577 }
1578 written += ret_splice;
1579 }
1580 lttng_consumer_sync_trace_file(stream, orig_offset);
1581
1582 ret = ret_splice;
1583
1584 goto end;
1585
8994307f
DG
1586write_error:
1587 /*
1588 * This is a special case that the relayd has closed its socket. Let's
1589 * cleanup the relayd object and all associated streams.
1590 */
1591 if (relayd && relayd_hang_up) {
1592 cleanup_relayd(relayd, ctx);
1593 /* Skip splice error so the consumer does not fail */
1594 goto end;
1595 }
1596
f02e1e8a
DG
1597splice_error:
1598 /* send the appropriate error description to sessiond */
1599 switch (ret) {
f02e1e8a 1600 case EINVAL:
f73fabfd 1601 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_EINVAL);
f02e1e8a
DG
1602 break;
1603 case ENOMEM:
f73fabfd 1604 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ENOMEM);
f02e1e8a
DG
1605 break;
1606 case ESPIPE:
f73fabfd 1607 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_SPLICE_ESPIPE);
f02e1e8a
DG
1608 break;
1609 }
1610
1611end:
1612 if (relayd && stream->metadata_flag) {
1613 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1614 }
1615
1616 rcu_read_unlock();
1617 return written;
3bd1e081
MD
1618}
1619
1620/*
1621 * Take a snapshot for a specific fd
1622 *
1623 * Returns 0 on success, < 0 on error
1624 */
ffe60014 1625int lttng_consumer_take_snapshot(struct lttng_consumer_stream *stream)
3bd1e081
MD
1626{
1627 switch (consumer_data.type) {
1628 case LTTNG_CONSUMER_KERNEL:
ffe60014 1629 return lttng_kconsumer_take_snapshot(stream);
7753dea8
MD
1630 case LTTNG_CONSUMER32_UST:
1631 case LTTNG_CONSUMER64_UST:
ffe60014 1632 return lttng_ustconsumer_take_snapshot(stream);
3bd1e081
MD
1633 default:
1634 ERR("Unknown consumer_data type");
1635 assert(0);
1636 return -ENOSYS;
1637 }
3bd1e081
MD
1638}
1639
1640/*
1641 * Get the produced position
1642 *
1643 * Returns 0 on success, < 0 on error
1644 */
ffe60014 1645int lttng_consumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
3bd1e081
MD
1646 unsigned long *pos)
1647{
1648 switch (consumer_data.type) {
1649 case LTTNG_CONSUMER_KERNEL:
ffe60014 1650 return lttng_kconsumer_get_produced_snapshot(stream, pos);
7753dea8
MD
1651 case LTTNG_CONSUMER32_UST:
1652 case LTTNG_CONSUMER64_UST:
ffe60014 1653 return lttng_ustconsumer_get_produced_snapshot(stream, pos);
3bd1e081
MD
1654 default:
1655 ERR("Unknown consumer_data type");
1656 assert(0);
1657 return -ENOSYS;
1658 }
1659}
1660
1661int lttng_consumer_recv_cmd(struct lttng_consumer_local_data *ctx,
1662 int sock, struct pollfd *consumer_sockpoll)
1663{
1664 switch (consumer_data.type) {
1665 case LTTNG_CONSUMER_KERNEL:
1666 return lttng_kconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
7753dea8
MD
1667 case LTTNG_CONSUMER32_UST:
1668 case LTTNG_CONSUMER64_UST:
3bd1e081
MD
1669 return lttng_ustconsumer_recv_cmd(ctx, sock, consumer_sockpoll);
1670 default:
1671 ERR("Unknown consumer_data type");
1672 assert(0);
1673 return -ENOSYS;
1674 }
1675}
1676
43c34bc3
DG
1677/*
1678 * Iterate over all streams of the hashtable and free them properly.
1679 *
1680 * WARNING: *MUST* be used with data stream only.
1681 */
1682static void destroy_data_stream_ht(struct lttng_ht *ht)
1683{
43c34bc3
DG
1684 struct lttng_ht_iter iter;
1685 struct lttng_consumer_stream *stream;
1686
1687 if (ht == NULL) {
1688 return;
1689 }
1690
1691 rcu_read_lock();
1692 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
5c540210
DG
1693 /*
1694 * Ignore return value since we are currently cleaning up so any error
1695 * can't be handled.
1696 */
1697 (void) consumer_del_stream(stream, ht);
43c34bc3
DG
1698 }
1699 rcu_read_unlock();
1700
1701 lttng_ht_destroy(ht);
1702}
1703
fb3a43a9 1704/*
f724d81e 1705 * Iterate over all streams of the hashtable and free them properly.
e316aad5
DG
1706 *
1707 * XXX: Should not be only for metadata stream or else use an other name.
fb3a43a9
DG
1708 */
1709static void destroy_stream_ht(struct lttng_ht *ht)
1710{
fb3a43a9
DG
1711 struct lttng_ht_iter iter;
1712 struct lttng_consumer_stream *stream;
1713
1714 if (ht == NULL) {
1715 return;
1716 }
1717
d09e1200 1718 rcu_read_lock();
58b1f425 1719 cds_lfht_for_each_entry(ht->ht, &iter.iter, stream, node.node) {
5c540210
DG
1720 /*
1721 * Ignore return value since we are currently cleaning up so any error
1722 * can't be handled.
1723 */
1724 (void) consumer_del_metadata_stream(stream, ht);
fb3a43a9 1725 }
d09e1200 1726 rcu_read_unlock();
fb3a43a9
DG
1727
1728 lttng_ht_destroy(ht);
1729}
1730
d88aee68
DG
1731void lttng_consumer_close_metadata(void)
1732{
1733 switch (consumer_data.type) {
1734 case LTTNG_CONSUMER_KERNEL:
1735 /*
1736 * The Kernel consumer has a different metadata scheme so we don't
1737 * close anything because the stream will be closed by the session
1738 * daemon.
1739 */
1740 break;
1741 case LTTNG_CONSUMER32_UST:
1742 case LTTNG_CONSUMER64_UST:
1743 /*
1744 * Close all metadata streams. The metadata hash table is passed and
1745 * this call iterates over it by closing all wakeup fd. This is safe
1746 * because at this point we are sure that the metadata producer is
1747 * either dead or blocked.
1748 */
1749 lttng_ustconsumer_close_metadata(metadata_ht);
1750 break;
1751 default:
1752 ERR("Unknown consumer_data type");
1753 assert(0);
1754 }
1755}
1756
fb3a43a9
DG
1757/*
1758 * Clean up a metadata stream and free its memory.
1759 */
e316aad5
DG
1760void consumer_del_metadata_stream(struct lttng_consumer_stream *stream,
1761 struct lttng_ht *ht)
fb3a43a9
DG
1762{
1763 int ret;
e316aad5
DG
1764 struct lttng_ht_iter iter;
1765 struct lttng_consumer_channel *free_chan = NULL;
fb3a43a9
DG
1766 struct consumer_relayd_sock_pair *relayd;
1767
1768 assert(stream);
1769 /*
1770 * This call should NEVER receive regular stream. It must always be
1771 * metadata stream and this is crucial for data structure synchronization.
1772 */
1773 assert(stream->metadata_flag);
1774
e316aad5
DG
1775 DBG3("Consumer delete metadata stream %d", stream->wait_fd);
1776
1777 if (ht == NULL) {
1778 /* Means the stream was allocated but not successfully added */
ffe60014 1779 goto free_stream_rcu;
e316aad5
DG
1780 }
1781
74251bb8 1782 pthread_mutex_lock(&consumer_data.lock);
8994307f
DG
1783 pthread_mutex_lock(&stream->lock);
1784
fb3a43a9
DG
1785 switch (consumer_data.type) {
1786 case LTTNG_CONSUMER_KERNEL:
1787 if (stream->mmap_base != NULL) {
1788 ret = munmap(stream->mmap_base, stream->mmap_len);
1789 if (ret != 0) {
1790 PERROR("munmap metadata stream");
1791 }
1792 }
1793 break;
1794 case LTTNG_CONSUMER32_UST:
1795 case LTTNG_CONSUMER64_UST:
1796 lttng_ustconsumer_del_stream(stream);
1797 break;
1798 default:
1799 ERR("Unknown consumer_data type");
1800 assert(0);
e316aad5 1801 goto end;
fb3a43a9 1802 }
fb3a43a9 1803
c869f647 1804 rcu_read_lock();
58b1f425 1805 iter.iter.node = &stream->node.node;
c869f647
DG
1806 ret = lttng_ht_del(ht, &iter);
1807 assert(!ret);
ca22feea
DG
1808
1809 /* Remove node session id from the consumer_data stream ht */
1810 iter.iter.node = &stream->node_session_id.node;
1811 ret = lttng_ht_del(consumer_data.stream_list_ht, &iter);
1812 assert(!ret);
c869f647
DG
1813 rcu_read_unlock();
1814
fb3a43a9
DG
1815 if (stream->out_fd >= 0) {
1816 ret = close(stream->out_fd);
1817 if (ret) {
1818 PERROR("close");
1819 }
1820 }
1821
fb3a43a9
DG
1822 /* Check and cleanup relayd */
1823 rcu_read_lock();
1824 relayd = consumer_find_relayd(stream->net_seq_idx);
1825 if (relayd != NULL) {
1826 uatomic_dec(&relayd->refcount);
1827 assert(uatomic_read(&relayd->refcount) >= 0);
1828
1829 /* Closing streams requires to lock the control socket. */
1830 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1831 ret = relayd_send_close_stream(&relayd->control_sock,
1832 stream->relayd_stream_id, stream->next_net_seq_num - 1);
1833 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1834 if (ret < 0) {
1835 DBG("Unable to close stream on the relayd. Continuing");
1836 /*
1837 * Continue here. There is nothing we can do for the relayd.
1838 * Chances are that the relayd has closed the socket so we just
1839 * continue cleaning up.
1840 */
1841 }
1842
1843 /* Both conditions are met, we destroy the relayd. */
1844 if (uatomic_read(&relayd->refcount) == 0 &&
1845 uatomic_read(&relayd->destroy_flag)) {
d09e1200 1846 destroy_relayd(relayd);
fb3a43a9
DG
1847 }
1848 }
1849 rcu_read_unlock();
1850
1851 /* Atomically decrement channel refcount since other threads can use it. */
1852 uatomic_dec(&stream->chan->refcount);
c30aaa51 1853 if (!uatomic_read(&stream->chan->refcount)
ffe60014 1854 && !uatomic_read(&stream->chan->nb_init_stream_left)) {
c30aaa51 1855 /* Go for channel deletion! */
e316aad5 1856 free_chan = stream->chan;
fb3a43a9
DG
1857 }
1858
e316aad5 1859end:
8994307f 1860 pthread_mutex_unlock(&stream->lock);
74251bb8 1861 pthread_mutex_unlock(&consumer_data.lock);
e316aad5
DG
1862
1863 if (free_chan) {
1864 consumer_del_channel(free_chan);
1865 }
1866
ffe60014
DG
1867free_stream_rcu:
1868 call_rcu(&stream->node.head, free_stream_rcu);
fb3a43a9
DG
1869}
1870
1871/*
1872 * Action done with the metadata stream when adding it to the consumer internal
1873 * data structures to handle it.
1874 */
ffe60014 1875static int add_metadata_stream(struct lttng_consumer_stream *stream,
e316aad5 1876 struct lttng_ht *ht)
fb3a43a9 1877{
e316aad5 1878 int ret = 0;
fb3a43a9 1879 struct consumer_relayd_sock_pair *relayd;
76082088 1880 struct lttng_ht_iter iter;
d88aee68 1881 struct lttng_ht_node_u64 *node;
fb3a43a9 1882
e316aad5
DG
1883 assert(stream);
1884 assert(ht);
1885
d88aee68 1886 DBG3("Adding metadata stream %" PRIu64 " to hash table", stream->key);
e316aad5
DG
1887
1888 pthread_mutex_lock(&consumer_data.lock);
2e818a6a 1889 pthread_mutex_lock(&stream->lock);
e316aad5 1890
e316aad5
DG
1891 /*
1892 * From here, refcounts are updated so be _careful_ when returning an error
1893 * after this point.
1894 */
1895
fb3a43a9 1896 rcu_read_lock();
76082088
DG
1897
1898 /*
1899 * Lookup the stream just to make sure it does not exist in our internal
1900 * state. This should NEVER happen.
1901 */
d88aee68
DG
1902 lttng_ht_lookup(ht, &stream->key, &iter);
1903 node = lttng_ht_iter_get_node_u64(&iter);
76082088
DG
1904 assert(!node);
1905
e316aad5 1906 /* Find relayd and, if one is found, increment refcount. */
fb3a43a9
DG
1907 relayd = consumer_find_relayd(stream->net_seq_idx);
1908 if (relayd != NULL) {
1909 uatomic_inc(&relayd->refcount);
1910 }
e316aad5
DG
1911
1912 /* Update channel refcount once added without error(s). */
1913 uatomic_inc(&stream->chan->refcount);
1914
1915 /*
ffe60014
DG
1916 * When nb_init_stream_left reaches 0, we don't need to trigger any action
1917 * in terms of destroying the associated channel, because the action that
e316aad5
DG
1918 * causes the count to become 0 also causes a stream to be added. The
1919 * channel deletion will thus be triggered by the following removal of this
1920 * stream.
1921 */
ffe60014
DG
1922 if (uatomic_read(&stream->chan->nb_init_stream_left) > 0) {
1923 uatomic_dec(&stream->chan->nb_init_stream_left);
e316aad5
DG
1924 }
1925
d88aee68 1926 lttng_ht_add_unique_u64(ht, &stream->node);
ca22feea
DG
1927
1928 /*
1929 * Add stream to the stream_list_ht of the consumer data. No need to steal
1930 * the key since the HT does not use it and we allow to add redundant keys
1931 * into this table.
1932 */
d88aee68 1933 lttng_ht_add_u64(consumer_data.stream_list_ht, &stream->node_session_id);
ca22feea 1934
fb3a43a9 1935 rcu_read_unlock();
e316aad5 1936
2e818a6a 1937 pthread_mutex_unlock(&stream->lock);
e316aad5
DG
1938 pthread_mutex_unlock(&consumer_data.lock);
1939 return ret;
fb3a43a9
DG
1940}
1941
8994307f
DG
1942/*
1943 * Delete data stream that are flagged for deletion (endpoint_status).
1944 */
1945static void validate_endpoint_status_data_stream(void)
1946{
1947 struct lttng_ht_iter iter;
1948 struct lttng_consumer_stream *stream;
1949
1950 DBG("Consumer delete flagged data stream");
1951
1952 rcu_read_lock();
1953 cds_lfht_for_each_entry(data_ht->ht, &iter.iter, stream, node.node) {
1954 /* Validate delete flag of the stream */
79d4ffb7 1955 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
1956 continue;
1957 }
1958 /* Delete it right now */
1959 consumer_del_stream(stream, data_ht);
1960 }
1961 rcu_read_unlock();
1962}
1963
1964/*
1965 * Delete metadata stream that are flagged for deletion (endpoint_status).
1966 */
1967static void validate_endpoint_status_metadata_stream(
1968 struct lttng_poll_event *pollset)
1969{
1970 struct lttng_ht_iter iter;
1971 struct lttng_consumer_stream *stream;
1972
1973 DBG("Consumer delete flagged metadata stream");
1974
1975 assert(pollset);
1976
1977 rcu_read_lock();
1978 cds_lfht_for_each_entry(metadata_ht->ht, &iter.iter, stream, node.node) {
1979 /* Validate delete flag of the stream */
79d4ffb7 1980 if (stream->endpoint_status == CONSUMER_ENDPOINT_ACTIVE) {
8994307f
DG
1981 continue;
1982 }
1983 /*
1984 * Remove from pollset so the metadata thread can continue without
1985 * blocking on a deleted stream.
1986 */
1987 lttng_poll_del(pollset, stream->wait_fd);
1988
1989 /* Delete it right now */
1990 consumer_del_metadata_stream(stream, metadata_ht);
1991 }
1992 rcu_read_unlock();
1993}
1994
fb3a43a9
DG
1995/*
1996 * Thread polls on metadata file descriptor and write them on disk or on the
1997 * network.
1998 */
7d980def 1999void *consumer_thread_metadata_poll(void *data)
fb3a43a9
DG
2000{
2001 int ret, i, pollfd;
2002 uint32_t revents, nb_fd;
e316aad5 2003 struct lttng_consumer_stream *stream = NULL;
fb3a43a9 2004 struct lttng_ht_iter iter;
d88aee68 2005 struct lttng_ht_node_u64 *node;
fb3a43a9
DG
2006 struct lttng_poll_event events;
2007 struct lttng_consumer_local_data *ctx = data;
2008 ssize_t len;
2009
2010 rcu_register_thread();
2011
d88aee68 2012 metadata_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
04bb2b64
DG
2013 if (!metadata_ht) {
2014 /* ENOMEM at this point. Better to bail out. */
2015 goto error;
2016 }
2017
fb3a43a9
DG
2018 DBG("Thread metadata poll started");
2019
fb3a43a9
DG
2020 /* Size is set to 1 for the consumer_metadata pipe */
2021 ret = lttng_poll_create(&events, 2, LTTNG_CLOEXEC);
2022 if (ret < 0) {
2023 ERR("Poll set creation failed");
2024 goto end;
2025 }
2026
2027 ret = lttng_poll_add(&events, ctx->consumer_metadata_pipe[0], LPOLLIN);
2028 if (ret < 0) {
2029 goto end;
2030 }
2031
2032 /* Main loop */
2033 DBG("Metadata main loop started");
2034
2035 while (1) {
fb3a43a9 2036 /* Only the metadata pipe is set */
d21b0d71 2037 if (LTTNG_POLL_GETNB(&events) == 0 && consumer_quit == 1) {
fb3a43a9
DG
2038 goto end;
2039 }
2040
2041restart:
d21b0d71 2042 DBG("Metadata poll wait with %d fd(s)", LTTNG_POLL_GETNB(&events));
fb3a43a9
DG
2043 ret = lttng_poll_wait(&events, -1);
2044 DBG("Metadata event catched in thread");
2045 if (ret < 0) {
2046 if (errno == EINTR) {
e316aad5 2047 ERR("Poll EINTR catched");
fb3a43a9
DG
2048 goto restart;
2049 }
2050 goto error;
2051 }
2052
0d9c5d77
DG
2053 nb_fd = ret;
2054
e316aad5 2055 /* From here, the event is a metadata wait fd */
fb3a43a9
DG
2056 for (i = 0; i < nb_fd; i++) {
2057 revents = LTTNG_POLL_GETEV(&events, i);
2058 pollfd = LTTNG_POLL_GETFD(&events, i);
2059
e316aad5
DG
2060 /* Just don't waste time if no returned events for the fd */
2061 if (!revents) {
2062 continue;
2063 }
2064
fb3a43a9 2065 if (pollfd == ctx->consumer_metadata_pipe[0]) {
4adabd61 2066 if (revents & (LPOLLERR | LPOLLHUP )) {
fb3a43a9
DG
2067 DBG("Metadata thread pipe hung up");
2068 /*
2069 * Remove the pipe from the poll set and continue the loop
2070 * since their might be data to consume.
2071 */
2072 lttng_poll_del(&events, ctx->consumer_metadata_pipe[0]);
f66c074c
DG
2073 ret = close(ctx->consumer_metadata_pipe[0]);
2074 if (ret < 0) {
2075 PERROR("close metadata pipe");
2076 }
fb3a43a9
DG
2077 continue;
2078 } else if (revents & LPOLLIN) {
fb3a43a9 2079 do {
633d0084
DG
2080 /* Get the stream pointer received */
2081 ret = read(pollfd, &stream, sizeof(stream));
fb3a43a9 2082 } while (ret < 0 && errno == EINTR);
633d0084
DG
2083 if (ret < 0 ||
2084 ret < sizeof(struct lttng_consumer_stream *)) {
fb3a43a9 2085 PERROR("read metadata stream");
fb3a43a9
DG
2086 /*
2087 * Let's continue here and hope we can still work
2088 * without stopping the consumer. XXX: Should we?
2089 */
2090 continue;
2091 }
2092
8994307f
DG
2093 /* A NULL stream means that the state has changed. */
2094 if (stream == NULL) {
2095 /* Check for deleted streams. */
2096 validate_endpoint_status_metadata_stream(&events);
3714380f 2097 goto restart;
8994307f
DG
2098 }
2099
fb3a43a9
DG
2100 DBG("Adding metadata stream %d to poll set",
2101 stream->wait_fd);
2102
ffe60014 2103 ret = add_metadata_stream(stream, metadata_ht);
e316aad5
DG
2104 if (ret) {
2105 ERR("Unable to add metadata stream");
2106 /* Stream was not setup properly. Continuing. */
2107 consumer_del_metadata_stream(stream, NULL);
2108 continue;
2109 }
fb3a43a9
DG
2110
2111 /* Add metadata stream to the global poll events list */
2112 lttng_poll_add(&events, stream->wait_fd,
2113 LPOLLIN | LPOLLPRI);
fb3a43a9
DG
2114 }
2115
e316aad5 2116 /* Handle other stream */
fb3a43a9
DG
2117 continue;
2118 }
2119
d09e1200 2120 rcu_read_lock();
d88aee68
DG
2121 {
2122 uint64_t tmp_id = (uint64_t) pollfd;
2123
2124 lttng_ht_lookup(metadata_ht, &tmp_id, &iter);
2125 }
2126 node = lttng_ht_iter_get_node_u64(&iter);
e316aad5 2127 assert(node);
fb3a43a9
DG
2128
2129 stream = caa_container_of(node, struct lttng_consumer_stream,
58b1f425 2130 node);
fb3a43a9 2131
e316aad5 2132 /* Check for error event */
4adabd61 2133 if (revents & (LPOLLERR | LPOLLHUP)) {
e316aad5 2134 DBG("Metadata fd %d is hup|err.", pollfd);
fb3a43a9
DG
2135 if (!stream->hangup_flush_done
2136 && (consumer_data.type == LTTNG_CONSUMER32_UST
2137 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2138 DBG("Attempting to flush and consume the UST buffers");
2139 lttng_ustconsumer_on_stream_hangup(stream);
2140
2141 /* We just flushed the stream now read it. */
4bb94b75
DG
2142 do {
2143 len = ctx->on_buffer_ready(stream, ctx);
2144 /*
2145 * We don't check the return value here since if we get
2146 * a negative len, it means an error occured thus we
2147 * simply remove it from the poll set and free the
2148 * stream.
2149 */
2150 } while (len > 0);
fb3a43a9
DG
2151 }
2152
fb3a43a9 2153 lttng_poll_del(&events, stream->wait_fd);
e316aad5
DG
2154 /*
2155 * This call update the channel states, closes file descriptors
2156 * and securely free the stream.
2157 */
2158 consumer_del_metadata_stream(stream, metadata_ht);
2159 } else if (revents & (LPOLLIN | LPOLLPRI)) {
2160 /* Get the data out of the metadata file descriptor */
2161 DBG("Metadata available on fd %d", pollfd);
2162 assert(stream->wait_fd == pollfd);
2163
2164 len = ctx->on_buffer_ready(stream, ctx);
2165 /* It's ok to have an unavailable sub-buffer */
b64403e3 2166 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2167 /* Clean up stream from consumer and free it. */
2168 lttng_poll_del(&events, stream->wait_fd);
2169 consumer_del_metadata_stream(stream, metadata_ht);
e316aad5
DG
2170 } else if (len > 0) {
2171 stream->data_read = 1;
2172 }
fb3a43a9 2173 }
e316aad5
DG
2174
2175 /* Release RCU lock for the stream looked up */
d09e1200 2176 rcu_read_unlock();
fb3a43a9
DG
2177 }
2178 }
2179
2180error:
2181end:
2182 DBG("Metadata poll thread exiting");
2183 lttng_poll_clean(&events);
2184
04bb2b64 2185 destroy_stream_ht(metadata_ht);
fb3a43a9
DG
2186
2187 rcu_unregister_thread();
2188 return NULL;
2189}
2190
3bd1e081 2191/*
e4421fec 2192 * This thread polls the fds in the set to consume the data and write
3bd1e081
MD
2193 * it to tracefile if necessary.
2194 */
7d980def 2195void *consumer_thread_data_poll(void *data)
3bd1e081
MD
2196{
2197 int num_rdy, num_hup, high_prio, ret, i;
2198 struct pollfd *pollfd = NULL;
2199 /* local view of the streams */
c869f647 2200 struct lttng_consumer_stream **local_stream = NULL, *new_stream = NULL;
3bd1e081
MD
2201 /* local view of consumer_data.fds_count */
2202 int nb_fd = 0;
3bd1e081 2203 struct lttng_consumer_local_data *ctx = data;
00e2e675 2204 ssize_t len;
3bd1e081 2205
e7b994a3
DG
2206 rcu_register_thread();
2207
d88aee68 2208 data_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
43c34bc3 2209 if (data_ht == NULL) {
04bb2b64 2210 /* ENOMEM at this point. Better to bail out. */
43c34bc3
DG
2211 goto end;
2212 }
2213
effcf122 2214 local_stream = zmalloc(sizeof(struct lttng_consumer_stream));
3bd1e081
MD
2215
2216 while (1) {
2217 high_prio = 0;
2218 num_hup = 0;
2219
2220 /*
e4421fec 2221 * the fds set has been updated, we need to update our
3bd1e081
MD
2222 * local array as well
2223 */
2224 pthread_mutex_lock(&consumer_data.lock);
2225 if (consumer_data.need_update) {
0e428499
DG
2226 free(pollfd);
2227 pollfd = NULL;
2228
2229 free(local_stream);
2230 local_stream = NULL;
3bd1e081 2231
50f8ae69 2232 /* allocate for all fds + 1 for the consumer_data_pipe */
effcf122 2233 pollfd = zmalloc((consumer_data.stream_count + 1) * sizeof(struct pollfd));
3bd1e081 2234 if (pollfd == NULL) {
7a57cf92 2235 PERROR("pollfd malloc");
3bd1e081
MD
2236 pthread_mutex_unlock(&consumer_data.lock);
2237 goto end;
2238 }
2239
50f8ae69 2240 /* allocate for all fds + 1 for the consumer_data_pipe */
effcf122 2241 local_stream = zmalloc((consumer_data.stream_count + 1) *
3bd1e081
MD
2242 sizeof(struct lttng_consumer_stream));
2243 if (local_stream == NULL) {
7a57cf92 2244 PERROR("local_stream malloc");
3bd1e081
MD
2245 pthread_mutex_unlock(&consumer_data.lock);
2246 goto end;
2247 }
ffe60014 2248 ret = update_poll_array(ctx, &pollfd, local_stream,
43c34bc3 2249 data_ht);
3bd1e081
MD
2250 if (ret < 0) {
2251 ERR("Error in allocating pollfd or local_outfds");
f73fabfd 2252 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2253 pthread_mutex_unlock(&consumer_data.lock);
2254 goto end;
2255 }
2256 nb_fd = ret;
2257 consumer_data.need_update = 0;
2258 }
2259 pthread_mutex_unlock(&consumer_data.lock);
2260
4078b776
MD
2261 /* No FDs and consumer_quit, consumer_cleanup the thread */
2262 if (nb_fd == 0 && consumer_quit == 1) {
2263 goto end;
2264 }
3bd1e081 2265 /* poll on the array of fds */
88f2b785 2266 restart:
3bd1e081 2267 DBG("polling on %d fd", nb_fd + 1);
cb365c03 2268 num_rdy = poll(pollfd, nb_fd + 1, -1);
3bd1e081
MD
2269 DBG("poll num_rdy : %d", num_rdy);
2270 if (num_rdy == -1) {
88f2b785
MD
2271 /*
2272 * Restart interrupted system call.
2273 */
2274 if (errno == EINTR) {
2275 goto restart;
2276 }
7a57cf92 2277 PERROR("Poll error");
f73fabfd 2278 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_POLL_ERROR);
3bd1e081
MD
2279 goto end;
2280 } else if (num_rdy == 0) {
2281 DBG("Polling thread timed out");
2282 goto end;
2283 }
2284
3bd1e081 2285 /*
50f8ae69 2286 * If the consumer_data_pipe triggered poll go directly to the
00e2e675
DG
2287 * beginning of the loop to update the array. We want to prioritize
2288 * array update over low-priority reads.
3bd1e081 2289 */
509bb1cf 2290 if (pollfd[nb_fd].revents & (POLLIN | POLLPRI)) {
ab30f567 2291 ssize_t pipe_readlen;
04fdd819 2292
50f8ae69 2293 DBG("consumer_data_pipe wake up");
04fdd819
MD
2294 /* Consume 1 byte of pipe data */
2295 do {
50f8ae69 2296 pipe_readlen = read(ctx->consumer_data_pipe[0], &new_stream,
c869f647 2297 sizeof(new_stream));
04fdd819 2298 } while (pipe_readlen == -1 && errno == EINTR);
23f5f35d
DG
2299 if (pipe_readlen < 0) {
2300 PERROR("read consumer data pipe");
2301 /* Continue so we can at least handle the current stream(s). */
2302 continue;
2303 }
c869f647
DG
2304
2305 /*
2306 * If the stream is NULL, just ignore it. It's also possible that
2307 * the sessiond poll thread changed the consumer_quit state and is
2308 * waking us up to test it.
2309 */
2310 if (new_stream == NULL) {
8994307f 2311 validate_endpoint_status_data_stream();
c869f647
DG
2312 continue;
2313 }
2314
ffe60014 2315 ret = add_stream(new_stream, data_ht);
c869f647 2316 if (ret) {
d88aee68 2317 ERR("Consumer add stream %" PRIu64 " failed. Continuing",
c869f647
DG
2318 new_stream->key);
2319 /*
2320 * At this point, if the add_stream fails, it is not in the
2321 * hash table thus passing the NULL value here.
2322 */
2323 consumer_del_stream(new_stream, NULL);
2324 }
2325
2326 /* Continue to update the local streams and handle prio ones */
3bd1e081
MD
2327 continue;
2328 }
2329
2330 /* Take care of high priority channels first. */
2331 for (i = 0; i < nb_fd; i++) {
9617607b
DG
2332 if (local_stream[i] == NULL) {
2333 continue;
2334 }
fb3a43a9 2335 if (pollfd[i].revents & POLLPRI) {
d41f73b7
MD
2336 DBG("Urgent read on fd %d", pollfd[i].fd);
2337 high_prio = 1;
4078b776 2338 len = ctx->on_buffer_ready(local_stream[i], ctx);
d41f73b7 2339 /* it's ok to have an unavailable sub-buffer */
b64403e3 2340 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2341 /* Clean the stream and free it. */
2342 consumer_del_stream(local_stream[i], data_ht);
9617607b 2343 local_stream[i] = NULL;
4078b776
MD
2344 } else if (len > 0) {
2345 local_stream[i]->data_read = 1;
d41f73b7 2346 }
3bd1e081
MD
2347 }
2348 }
2349
4078b776
MD
2350 /*
2351 * If we read high prio channel in this loop, try again
2352 * for more high prio data.
2353 */
2354 if (high_prio) {
3bd1e081
MD
2355 continue;
2356 }
2357
2358 /* Take care of low priority channels. */
4078b776 2359 for (i = 0; i < nb_fd; i++) {
9617607b
DG
2360 if (local_stream[i] == NULL) {
2361 continue;
2362 }
4078b776
MD
2363 if ((pollfd[i].revents & POLLIN) ||
2364 local_stream[i]->hangup_flush_done) {
4078b776
MD
2365 DBG("Normal read on fd %d", pollfd[i].fd);
2366 len = ctx->on_buffer_ready(local_stream[i], ctx);
2367 /* it's ok to have an unavailable sub-buffer */
b64403e3 2368 if (len < 0 && len != -EAGAIN && len != -ENODATA) {
ab1027f4
DG
2369 /* Clean the stream and free it. */
2370 consumer_del_stream(local_stream[i], data_ht);
9617607b 2371 local_stream[i] = NULL;
4078b776
MD
2372 } else if (len > 0) {
2373 local_stream[i]->data_read = 1;
2374 }
2375 }
2376 }
2377
2378 /* Handle hangup and errors */
2379 for (i = 0; i < nb_fd; i++) {
9617607b
DG
2380 if (local_stream[i] == NULL) {
2381 continue;
2382 }
4078b776
MD
2383 if (!local_stream[i]->hangup_flush_done
2384 && (pollfd[i].revents & (POLLHUP | POLLERR | POLLNVAL))
2385 && (consumer_data.type == LTTNG_CONSUMER32_UST
2386 || consumer_data.type == LTTNG_CONSUMER64_UST)) {
2387 DBG("fd %d is hup|err|nval. Attempting flush and read.",
9617607b 2388 pollfd[i].fd);
4078b776
MD
2389 lttng_ustconsumer_on_stream_hangup(local_stream[i]);
2390 /* Attempt read again, for the data we just flushed. */
2391 local_stream[i]->data_read = 1;
2392 }
2393 /*
2394 * If the poll flag is HUP/ERR/NVAL and we have
2395 * read no data in this pass, we can remove the
2396 * stream from its hash table.
2397 */
2398 if ((pollfd[i].revents & POLLHUP)) {
2399 DBG("Polling fd %d tells it has hung up.", pollfd[i].fd);
2400 if (!local_stream[i]->data_read) {
43c34bc3 2401 consumer_del_stream(local_stream[i], data_ht);
9617607b 2402 local_stream[i] = NULL;
4078b776
MD
2403 num_hup++;
2404 }
2405 } else if (pollfd[i].revents & POLLERR) {
2406 ERR("Error returned in polling fd %d.", pollfd[i].fd);
2407 if (!local_stream[i]->data_read) {
43c34bc3 2408 consumer_del_stream(local_stream[i], data_ht);
9617607b 2409 local_stream[i] = NULL;
4078b776
MD
2410 num_hup++;
2411 }
2412 } else if (pollfd[i].revents & POLLNVAL) {
2413 ERR("Polling fd %d tells fd is not open.", pollfd[i].fd);
2414 if (!local_stream[i]->data_read) {
43c34bc3 2415 consumer_del_stream(local_stream[i], data_ht);
9617607b 2416 local_stream[i] = NULL;
4078b776 2417 num_hup++;
3bd1e081
MD
2418 }
2419 }
9617607b
DG
2420 if (local_stream[i] != NULL) {
2421 local_stream[i]->data_read = 0;
2422 }
3bd1e081
MD
2423 }
2424 }
2425end:
2426 DBG("polling thread exiting");
0e428499
DG
2427 free(pollfd);
2428 free(local_stream);
fb3a43a9
DG
2429
2430 /*
2431 * Close the write side of the pipe so epoll_wait() in
7d980def
DG
2432 * consumer_thread_metadata_poll can catch it. The thread is monitoring the
2433 * read side of the pipe. If we close them both, epoll_wait strangely does
2434 * not return and could create a endless wait period if the pipe is the
2435 * only tracked fd in the poll set. The thread will take care of closing
2436 * the read side.
fb3a43a9 2437 */
f66c074c
DG
2438 ret = close(ctx->consumer_metadata_pipe[1]);
2439 if (ret < 0) {
2440 PERROR("close data pipe");
2441 }
fb3a43a9 2442
04bb2b64 2443 destroy_data_stream_ht(data_ht);
43c34bc3 2444
e7b994a3 2445 rcu_unregister_thread();
3bd1e081
MD
2446 return NULL;
2447}
2448
2449/*
2450 * This thread listens on the consumerd socket and receives the file
2451 * descriptors from the session daemon.
2452 */
7d980def 2453void *consumer_thread_sessiond_poll(void *data)
3bd1e081 2454{
d96f09c6 2455 int sock = -1, client_socket, ret;
3bd1e081
MD
2456 /*
2457 * structure to poll for incoming data on communication socket avoids
2458 * making blocking sockets.
2459 */
2460 struct pollfd consumer_sockpoll[2];
2461 struct lttng_consumer_local_data *ctx = data;
2462
e7b994a3
DG
2463 rcu_register_thread();
2464
3bd1e081
MD
2465 DBG("Creating command socket %s", ctx->consumer_command_sock_path);
2466 unlink(ctx->consumer_command_sock_path);
2467 client_socket = lttcomm_create_unix_sock(ctx->consumer_command_sock_path);
2468 if (client_socket < 0) {
2469 ERR("Cannot create command socket");
2470 goto end;
2471 }
2472
2473 ret = lttcomm_listen_unix_sock(client_socket);
2474 if (ret < 0) {
2475 goto end;
2476 }
2477
32258573 2478 DBG("Sending ready command to lttng-sessiond");
f73fabfd 2479 ret = lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_COMMAND_SOCK_READY);
3bd1e081
MD
2480 /* return < 0 on error, but == 0 is not fatal */
2481 if (ret < 0) {
32258573 2482 ERR("Error sending ready command to lttng-sessiond");
3bd1e081
MD
2483 goto end;
2484 }
2485
2486 ret = fcntl(client_socket, F_SETFL, O_NONBLOCK);
2487 if (ret < 0) {
7a57cf92 2488 PERROR("fcntl O_NONBLOCK");
3bd1e081
MD
2489 goto end;
2490 }
2491
2492 /* prepare the FDs to poll : to client socket and the should_quit pipe */
2493 consumer_sockpoll[0].fd = ctx->consumer_should_quit[0];
2494 consumer_sockpoll[0].events = POLLIN | POLLPRI;
2495 consumer_sockpoll[1].fd = client_socket;
2496 consumer_sockpoll[1].events = POLLIN | POLLPRI;
2497
2498 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
2499 goto end;
2500 }
2501 DBG("Connection on client_socket");
2502
2503 /* Blocking call, waiting for transmission */
2504 sock = lttcomm_accept_unix_sock(client_socket);
534d2592 2505 if (sock < 0) {
3bd1e081
MD
2506 WARN("On accept");
2507 goto end;
2508 }
2509 ret = fcntl(sock, F_SETFL, O_NONBLOCK);
2510 if (ret < 0) {
7a57cf92 2511 PERROR("fcntl O_NONBLOCK");
3bd1e081
MD
2512 goto end;
2513 }
2514
d96f09c6
DG
2515 /* This socket is not useful anymore. */
2516 ret = close(client_socket);
2517 if (ret < 0) {
2518 PERROR("close client_socket");
2519 }
2520 client_socket = -1;
2521
3bd1e081
MD
2522 /* update the polling structure to poll on the established socket */
2523 consumer_sockpoll[1].fd = sock;
2524 consumer_sockpoll[1].events = POLLIN | POLLPRI;
2525
2526 while (1) {
2527 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
2528 goto end;
2529 }
2530 DBG("Incoming command on sock");
2531 ret = lttng_consumer_recv_cmd(ctx, sock, consumer_sockpoll);
2532 if (ret == -ENOENT) {
2533 DBG("Received STOP command");
2534 goto end;
2535 }
4cbc1a04
DG
2536 if (ret <= 0) {
2537 /*
2538 * This could simply be a session daemon quitting. Don't output
2539 * ERR() here.
2540 */
2541 DBG("Communication interrupted on command socket");
3bd1e081
MD
2542 goto end;
2543 }
2544 if (consumer_quit) {
2545 DBG("consumer_thread_receive_fds received quit from signal");
2546 goto end;
2547 }
ffe60014 2548 DBG("received command on sock");
3bd1e081
MD
2549 }
2550end:
ffe60014 2551 DBG("Consumer thread sessiond poll exiting");
3bd1e081 2552
d88aee68
DG
2553 /*
2554 * Close metadata streams since the producer is the session daemon which
2555 * just died.
2556 *
2557 * NOTE: for now, this only applies to the UST tracer.
2558 */
2559 lttng_consumer_close_metadata();
2560
3bd1e081
MD
2561 /*
2562 * when all fds have hung up, the polling thread
2563 * can exit cleanly
2564 */
2565 consumer_quit = 1;
2566
04fdd819 2567 /*
c869f647 2568 * Notify the data poll thread to poll back again and test the
8994307f 2569 * consumer_quit state that we just set so to quit gracefully.
04fdd819 2570 */
8994307f 2571 notify_thread_pipe(ctx->consumer_data_pipe[1]);
c869f647 2572
d96f09c6
DG
2573 /* Cleaning up possibly open sockets. */
2574 if (sock >= 0) {
2575 ret = close(sock);
2576 if (ret < 0) {
2577 PERROR("close sock sessiond poll");
2578 }
2579 }
2580 if (client_socket >= 0) {
2581 ret = close(sock);
2582 if (ret < 0) {
2583 PERROR("close client_socket sessiond poll");
2584 }
2585 }
2586
e7b994a3 2587 rcu_unregister_thread();
3bd1e081
MD
2588 return NULL;
2589}
d41f73b7 2590
4078b776 2591ssize_t lttng_consumer_read_subbuffer(struct lttng_consumer_stream *stream,
d41f73b7
MD
2592 struct lttng_consumer_local_data *ctx)
2593{
74251bb8
DG
2594 ssize_t ret;
2595
2596 pthread_mutex_lock(&stream->lock);
2597
d41f73b7
MD
2598 switch (consumer_data.type) {
2599 case LTTNG_CONSUMER_KERNEL:
74251bb8
DG
2600 ret = lttng_kconsumer_read_subbuffer(stream, ctx);
2601 break;
7753dea8
MD
2602 case LTTNG_CONSUMER32_UST:
2603 case LTTNG_CONSUMER64_UST:
74251bb8
DG
2604 ret = lttng_ustconsumer_read_subbuffer(stream, ctx);
2605 break;
d41f73b7
MD
2606 default:
2607 ERR("Unknown consumer_data type");
2608 assert(0);
74251bb8
DG
2609 ret = -ENOSYS;
2610 break;
d41f73b7 2611 }
74251bb8
DG
2612
2613 pthread_mutex_unlock(&stream->lock);
2614 return ret;
d41f73b7
MD
2615}
2616
2617int lttng_consumer_on_recv_stream(struct lttng_consumer_stream *stream)
2618{
2619 switch (consumer_data.type) {
2620 case LTTNG_CONSUMER_KERNEL:
2621 return lttng_kconsumer_on_recv_stream(stream);
7753dea8
MD
2622 case LTTNG_CONSUMER32_UST:
2623 case LTTNG_CONSUMER64_UST:
d41f73b7
MD
2624 return lttng_ustconsumer_on_recv_stream(stream);
2625 default:
2626 ERR("Unknown consumer_data type");
2627 assert(0);
2628 return -ENOSYS;
2629 }
2630}
e4421fec
DG
2631
2632/*
2633 * Allocate and set consumer data hash tables.
2634 */
2635void lttng_consumer_init(void)
2636{
d88aee68
DG
2637 consumer_data.channel_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2638 consumer_data.relayd_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
2639 consumer_data.stream_list_ht = lttng_ht_new(0, LTTNG_HT_TYPE_U64);
e4421fec 2640}
7735ef9e
DG
2641
2642/*
2643 * Process the ADD_RELAYD command receive by a consumer.
2644 *
2645 * This will create a relayd socket pair and add it to the relayd hash table.
2646 * The caller MUST acquire a RCU read side lock before calling it.
2647 */
2648int consumer_add_relayd_socket(int net_seq_idx, int sock_type,
2649 struct lttng_consumer_local_data *ctx, int sock,
46e6455f
DG
2650 struct pollfd *consumer_sockpoll, struct lttcomm_sock *relayd_sock,
2651 unsigned int sessiond_id)
7735ef9e 2652{
cd2b09ed 2653 int fd = -1, ret = -1, relayd_created = 0;
f50f23d9 2654 enum lttng_error_code ret_code = LTTNG_OK;
7735ef9e
DG
2655 struct consumer_relayd_sock_pair *relayd;
2656
2657 DBG("Consumer adding relayd socket (idx: %d)", net_seq_idx);
2658
f50f23d9
DG
2659 /* First send a status message before receiving the fds. */
2660 ret = consumer_send_status_msg(sock, ret_code);
2661 if (ret < 0) {
2662 /* Somehow, the session daemon is not responding anymore. */
2663 goto error;
2664 }
2665
7735ef9e
DG
2666 /* Get relayd reference if exists. */
2667 relayd = consumer_find_relayd(net_seq_idx);
2668 if (relayd == NULL) {
2669 /* Not found. Allocate one. */
2670 relayd = consumer_allocate_relayd_sock_pair(net_seq_idx);
2671 if (relayd == NULL) {
f73fabfd 2672 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
59e71485 2673 ret = -1;
7735ef9e
DG
2674 goto error;
2675 }
f7079f67 2676 relayd->sessiond_session_id = (uint64_t) sessiond_id;
cd2b09ed 2677 relayd_created = 1;
7735ef9e
DG
2678 }
2679
2680 /* Poll on consumer socket. */
2681 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
2682 ret = -EINTR;
2683 goto error;
2684 }
2685
2686 /* Get relayd socket from session daemon */
2687 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
2688 if (ret != sizeof(fd)) {
f73fabfd 2689 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
7735ef9e 2690 ret = -1;
4028eeb9 2691 fd = -1; /* Just in case it gets set with an invalid value. */
ffe60014 2692 goto error_close;
7735ef9e
DG
2693 }
2694
f50f23d9
DG
2695 /* We have the fds without error. Send status back. */
2696 ret = consumer_send_status_msg(sock, ret_code);
2697 if (ret < 0) {
2698 /* Somehow, the session daemon is not responding anymore. */
2699 goto error;
2700 }
2701
7735ef9e
DG
2702 /* Copy socket information and received FD */
2703 switch (sock_type) {
2704 case LTTNG_STREAM_CONTROL:
2705 /* Copy received lttcomm socket */
2706 lttcomm_copy_sock(&relayd->control_sock, relayd_sock);
2707 ret = lttcomm_create_sock(&relayd->control_sock);
4028eeb9
DG
2708 /* Immediately try to close the created socket if valid. */
2709 if (relayd->control_sock.fd >= 0) {
2710 if (close(relayd->control_sock.fd)) {
2711 PERROR("close relayd control socket");
2712 }
7735ef9e 2713 }
4028eeb9 2714 /* Handle create_sock error. */
f66c074c 2715 if (ret < 0) {
4028eeb9 2716 goto error;
f66c074c 2717 }
7735ef9e
DG
2718
2719 /* Assign new file descriptor */
2720 relayd->control_sock.fd = fd;
c5b6f4f0
DG
2721
2722 /*
59e71485
DG
2723 * Create a session on the relayd and store the returned id. Lock the
2724 * control socket mutex if the relayd was NOT created before.
c5b6f4f0 2725 */
59e71485
DG
2726 if (!relayd_created) {
2727 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
2728 }
c5b6f4f0 2729 ret = relayd_create_session(&relayd->control_sock,
f7079f67 2730 &relayd->relayd_session_id);
59e71485
DG
2731 if (!relayd_created) {
2732 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
2733 }
c5b6f4f0 2734 if (ret < 0) {
ffe60014
DG
2735 /*
2736 * Close all sockets of a relayd object. It will be freed if it was
2737 * created at the error code path or else it will be garbage
2738 * collect.
2739 */
2740 (void) relayd_close(&relayd->control_sock);
2741 (void) relayd_close(&relayd->data_sock);
c5b6f4f0
DG
2742 goto error;
2743 }
2744
7735ef9e
DG
2745 break;
2746 case LTTNG_STREAM_DATA:
2747 /* Copy received lttcomm socket */
2748 lttcomm_copy_sock(&relayd->data_sock, relayd_sock);
2749 ret = lttcomm_create_sock(&relayd->data_sock);
4028eeb9
DG
2750 /* Immediately try to close the created socket if valid. */
2751 if (relayd->data_sock.fd >= 0) {
2752 if (close(relayd->data_sock.fd)) {
2753 PERROR("close relayd data socket");
2754 }
7735ef9e 2755 }
4028eeb9 2756 /* Handle create_sock error. */
f66c074c 2757 if (ret < 0) {
4028eeb9 2758 goto error;
f66c074c 2759 }
7735ef9e
DG
2760
2761 /* Assign new file descriptor */
2762 relayd->data_sock.fd = fd;
2763 break;
2764 default:
2765 ERR("Unknown relayd socket type (%d)", sock_type);
59e71485 2766 ret = -1;
7735ef9e
DG
2767 goto error;
2768 }
2769
d88aee68 2770 DBG("Consumer %s socket created successfully with net idx %" PRIu64 " (fd: %d)",
7735ef9e
DG
2771 sock_type == LTTNG_STREAM_CONTROL ? "control" : "data",
2772 relayd->net_seq_idx, fd);
2773
2774 /*
2775 * Add relayd socket pair to consumer data hashtable. If object already
2776 * exists or on error, the function gracefully returns.
2777 */
d09e1200 2778 add_relayd(relayd);
7735ef9e
DG
2779
2780 /* All good! */
4028eeb9 2781 return 0;
7735ef9e
DG
2782
2783error:
4028eeb9
DG
2784 /* Close received socket if valid. */
2785 if (fd >= 0) {
2786 if (close(fd)) {
2787 PERROR("close received socket");
2788 }
2789 }
cd2b09ed 2790
ffe60014 2791error_close:
cd2b09ed 2792 if (relayd_created) {
cd2b09ed
DG
2793 free(relayd);
2794 }
2795
7735ef9e
DG
2796 return ret;
2797}
ca22feea 2798
4e9a4686
DG
2799/*
2800 * Try to lock the stream mutex.
2801 *
2802 * On success, 1 is returned else 0 indicating that the mutex is NOT lock.
2803 */
2804static int stream_try_lock(struct lttng_consumer_stream *stream)
2805{
2806 int ret;
2807
2808 assert(stream);
2809
2810 /*
2811 * Try to lock the stream mutex. On failure, we know that the stream is
2812 * being used else where hence there is data still being extracted.
2813 */
2814 ret = pthread_mutex_trylock(&stream->lock);
2815 if (ret) {
2816 /* For both EBUSY and EINVAL error, the mutex is NOT locked. */
2817 ret = 0;
2818 goto end;
2819 }
2820
2821 ret = 1;
2822
2823end:
2824 return ret;
2825}
2826
f7079f67
DG
2827/*
2828 * Search for a relayd associated to the session id and return the reference.
2829 *
2830 * A rcu read side lock MUST be acquire before calling this function and locked
2831 * until the relayd object is no longer necessary.
2832 */
2833static struct consumer_relayd_sock_pair *find_relayd_by_session_id(uint64_t id)
2834{
2835 struct lttng_ht_iter iter;
f7079f67 2836 struct consumer_relayd_sock_pair *relayd = NULL;
f7079f67
DG
2837
2838 /* Iterate over all relayd since they are indexed by net_seq_idx. */
2839 cds_lfht_for_each_entry(consumer_data.relayd_ht->ht, &iter.iter, relayd,
2840 node.node) {
18261bd1
DG
2841 /*
2842 * Check by sessiond id which is unique here where the relayd session
2843 * id might not be when having multiple relayd.
2844 */
2845 if (relayd->sessiond_session_id == id) {
f7079f67 2846 /* Found the relayd. There can be only one per id. */
18261bd1 2847 goto found;
f7079f67
DG
2848 }
2849 }
2850
18261bd1
DG
2851 return NULL;
2852
2853found:
f7079f67
DG
2854 return relayd;
2855}
2856
ca22feea
DG
2857/*
2858 * Check if for a given session id there is still data needed to be extract
2859 * from the buffers.
2860 *
6d805429 2861 * Return 1 if data is pending or else 0 meaning ready to be read.
ca22feea 2862 */
6d805429 2863int consumer_data_pending(uint64_t id)
ca22feea
DG
2864{
2865 int ret;
2866 struct lttng_ht_iter iter;
2867 struct lttng_ht *ht;
2868 struct lttng_consumer_stream *stream;
f7079f67 2869 struct consumer_relayd_sock_pair *relayd = NULL;
6d805429 2870 int (*data_pending)(struct lttng_consumer_stream *);
ca22feea 2871
6d805429 2872 DBG("Consumer data pending command on session id %" PRIu64, id);
ca22feea 2873
6f6eda74 2874 rcu_read_lock();
ca22feea
DG
2875 pthread_mutex_lock(&consumer_data.lock);
2876
2877 switch (consumer_data.type) {
2878 case LTTNG_CONSUMER_KERNEL:
6d805429 2879 data_pending = lttng_kconsumer_data_pending;
ca22feea
DG
2880 break;
2881 case LTTNG_CONSUMER32_UST:
2882 case LTTNG_CONSUMER64_UST:
6d805429 2883 data_pending = lttng_ustconsumer_data_pending;
ca22feea
DG
2884 break;
2885 default:
2886 ERR("Unknown consumer data type");
2887 assert(0);
2888 }
2889
2890 /* Ease our life a bit */
2891 ht = consumer_data.stream_list_ht;
2892
f7079f67
DG
2893 relayd = find_relayd_by_session_id(id);
2894 if (relayd) {
2895 /* Send init command for data pending. */
2896 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
2897 ret = relayd_begin_data_pending(&relayd->control_sock,
2898 relayd->relayd_session_id);
2899 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
2900 if (ret < 0) {
2901 /* Communication error thus the relayd so no data pending. */
2902 goto data_not_pending;
2903 }
2904 }
2905
c8f59ee5 2906 cds_lfht_for_each_entry_duplicate(ht->ht,
d88aee68
DG
2907 ht->hash_fct(&id, lttng_ht_seed),
2908 ht->match_fct, &id,
ca22feea 2909 &iter.iter, stream, node_session_id.node) {
4e9a4686
DG
2910 /* If this call fails, the stream is being used hence data pending. */
2911 ret = stream_try_lock(stream);
2912 if (!ret) {
f7079f67 2913 goto data_pending;
ca22feea 2914 }
ca22feea 2915
4e9a4686
DG
2916 /*
2917 * A removed node from the hash table indicates that the stream has
2918 * been deleted thus having a guarantee that the buffers are closed
2919 * on the consumer side. However, data can still be transmitted
2920 * over the network so don't skip the relayd check.
2921 */
2922 ret = cds_lfht_is_node_deleted(&stream->node.node);
2923 if (!ret) {
2924 /* Check the stream if there is data in the buffers. */
6d805429
DG
2925 ret = data_pending(stream);
2926 if (ret == 1) {
4e9a4686 2927 pthread_mutex_unlock(&stream->lock);
f7079f67 2928 goto data_pending;
4e9a4686
DG
2929 }
2930 }
2931
2932 /* Relayd check */
f7079f67 2933 if (relayd) {
c8f59ee5
DG
2934 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
2935 if (stream->metadata_flag) {
ad7051c0
DG
2936 ret = relayd_quiescent_control(&relayd->control_sock,
2937 stream->relayd_stream_id);
c8f59ee5 2938 } else {
6d805429 2939 ret = relayd_data_pending(&relayd->control_sock,
39df6d9f
DG
2940 stream->relayd_stream_id,
2941 stream->next_net_seq_num - 1);
c8f59ee5
DG
2942 }
2943 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
6d805429 2944 if (ret == 1) {
4e9a4686 2945 pthread_mutex_unlock(&stream->lock);
f7079f67 2946 goto data_pending;
c8f59ee5
DG
2947 }
2948 }
4e9a4686 2949 pthread_mutex_unlock(&stream->lock);
c8f59ee5 2950 }
ca22feea 2951
f7079f67
DG
2952 if (relayd) {
2953 unsigned int is_data_inflight = 0;
2954
2955 /* Send init command for data pending. */
2956 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
2957 ret = relayd_end_data_pending(&relayd->control_sock,
2958 relayd->relayd_session_id, &is_data_inflight);
2959 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
bdd88757 2960 if (ret < 0) {
f7079f67
DG
2961 goto data_not_pending;
2962 }
bdd88757
DG
2963 if (is_data_inflight) {
2964 goto data_pending;
2965 }
f7079f67
DG
2966 }
2967
ca22feea 2968 /*
f7079f67
DG
2969 * Finding _no_ node in the hash table and no inflight data means that the
2970 * stream(s) have been removed thus data is guaranteed to be available for
2971 * analysis from the trace files.
ca22feea
DG
2972 */
2973
f7079f67 2974data_not_pending:
ca22feea
DG
2975 /* Data is available to be read by a viewer. */
2976 pthread_mutex_unlock(&consumer_data.lock);
c8f59ee5 2977 rcu_read_unlock();
6d805429 2978 return 0;
ca22feea 2979
f7079f67 2980data_pending:
ca22feea
DG
2981 /* Data is still being extracted from buffers. */
2982 pthread_mutex_unlock(&consumer_data.lock);
c8f59ee5 2983 rcu_read_unlock();
6d805429 2984 return 1;
ca22feea 2985}
f50f23d9
DG
2986
2987/*
2988 * Send a ret code status message to the sessiond daemon.
2989 *
2990 * Return the sendmsg() return value.
2991 */
2992int consumer_send_status_msg(int sock, int ret_code)
2993{
2994 struct lttcomm_consumer_status_msg msg;
2995
2996 msg.ret_code = ret_code;
2997
2998 return lttcomm_send_unix_sock(sock, &msg, sizeof(msg));
2999}
ffe60014
DG
3000
3001/*
3002 * Send a channel status message to the sessiond daemon.
3003 *
3004 * Return the sendmsg() return value.
3005 */
3006int consumer_send_status_channel(int sock,
3007 struct lttng_consumer_channel *channel)
3008{
3009 struct lttcomm_consumer_status_channel msg;
3010
3011 assert(sock >= 0);
3012
3013 if (!channel) {
3014 msg.ret_code = -LTTNG_ERR_UST_CHAN_FAIL;
3015 } else {
3016 msg.ret_code = LTTNG_OK;
3017 msg.key = channel->key;
3018 msg.stream_count = channel->streams.count;
3019 }
3020
3021 return lttcomm_send_unix_sock(sock, &msg, sizeof(msg));
3022}
This page took 0.338354 seconds and 4 git commands to generate.