Clean-up: consumerd: reduce duplication of stream output close code
[lttng-tools.git] / src / common / consumer / consumer-stream.cpp
1 /*
2 * Copyright (C) 2011 EfficiOS Inc.
3 * Copyright (C) 2011 Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 * Copyright (C) 2013 David Goulet <dgoulet@efficios.com>
5 *
6 * SPDX-License-Identifier: GPL-2.0-only
7 *
8 */
9
10 #define _LGPL_SOURCE
11 #include <inttypes.h>
12 #include <sys/mman.h>
13 #include <unistd.h>
14
15 #include <common/common.hpp>
16 #include <common/consumer/consumer-timer.hpp>
17 #include <common/consumer/consumer-timer.hpp>
18 #include <common/consumer/consumer.hpp>
19 #include <common/consumer/consumer.hpp>
20 #include <common/consumer/metadata-bucket.hpp>
21 #include <common/consumer/metadata-bucket.hpp>
22 #include <common/index/index.hpp>
23 #include <common/kernel-consumer/kernel-consumer.hpp>
24 #include <common/kernel-ctl/kernel-ctl.hpp>
25 #include <common/macros.hpp>
26 #include <common/relayd/relayd.hpp>
27 #include <common/ust-consumer/ust-consumer.hpp>
28 #include <common/utils.hpp>
29
30 #include "consumer-stream.hpp"
31
32 /*
33 * RCU call to free stream. MUST only be used with call_rcu().
34 */
35 static void free_stream_rcu(struct rcu_head *head)
36 {
37 struct lttng_ht_node_u64 *node =
38 lttng::utils::container_of(head, &lttng_ht_node_u64::head);
39 struct lttng_consumer_stream *stream =
40 lttng::utils::container_of(node, &lttng_consumer_stream::node);
41
42 pthread_mutex_destroy(&stream->lock);
43 free(stream);
44 }
45
46 static void consumer_stream_data_lock_all(struct lttng_consumer_stream *stream)
47 {
48 pthread_mutex_lock(&stream->chan->lock);
49 pthread_mutex_lock(&stream->lock);
50 }
51
52 static void consumer_stream_data_unlock_all(struct lttng_consumer_stream *stream)
53 {
54 pthread_mutex_unlock(&stream->lock);
55 pthread_mutex_unlock(&stream->chan->lock);
56 }
57
58 static void consumer_stream_data_assert_locked_all(struct lttng_consumer_stream *stream)
59 {
60 ASSERT_LOCKED(stream->lock);
61 ASSERT_LOCKED(stream->chan->lock);
62 }
63
64 static void consumer_stream_metadata_lock_all(struct lttng_consumer_stream *stream)
65 {
66 consumer_stream_data_lock_all(stream);
67 pthread_mutex_lock(&stream->metadata_rdv_lock);
68 }
69
70 static void consumer_stream_metadata_unlock_all(struct lttng_consumer_stream *stream)
71 {
72 pthread_mutex_unlock(&stream->metadata_rdv_lock);
73 consumer_stream_data_unlock_all(stream);
74 }
75
76 static void consumer_stream_metadata_assert_locked_all(struct lttng_consumer_stream *stream)
77 {
78 ASSERT_LOCKED(stream->metadata_rdv_lock);
79 consumer_stream_data_assert_locked_all(stream);
80 }
81
82 /* Only used for data streams. */
83 static int consumer_stream_update_stats(struct lttng_consumer_stream *stream,
84 const struct stream_subbuffer *subbuf)
85 {
86 int ret = 0;
87 uint64_t sequence_number;
88 const uint64_t discarded_events = subbuf->info.data.events_discarded;
89
90 if (!subbuf->info.data.sequence_number.is_set) {
91 /* Command not supported by the tracer. */
92 sequence_number = -1ULL;
93 stream->sequence_number_unavailable = true;
94 } else {
95 sequence_number = subbuf->info.data.sequence_number.value;
96 }
97
98 /*
99 * Start the sequence when we extract the first packet in case we don't
100 * start at 0 (for example if a consumer is not connected to the
101 * session immediately after the beginning).
102 */
103 if (stream->last_sequence_number == -1ULL) {
104 stream->last_sequence_number = sequence_number;
105 } else if (sequence_number > stream->last_sequence_number) {
106 stream->chan->lost_packets += sequence_number -
107 stream->last_sequence_number - 1;
108 } else {
109 /* seq <= last_sequence_number */
110 ERR("Sequence number inconsistent : prev = %" PRIu64
111 ", current = %" PRIu64,
112 stream->last_sequence_number, sequence_number);
113 ret = -1;
114 goto end;
115 }
116 stream->last_sequence_number = sequence_number;
117
118 if (discarded_events < stream->last_discarded_events) {
119 /*
120 * Overflow has occurred. We assume only one wrap-around
121 * has occurred.
122 */
123 stream->chan->discarded_events +=
124 (1ULL << (CAA_BITS_PER_LONG - 1)) -
125 stream->last_discarded_events +
126 discarded_events;
127 } else {
128 stream->chan->discarded_events += discarded_events -
129 stream->last_discarded_events;
130 }
131 stream->last_discarded_events = discarded_events;
132 ret = 0;
133
134 end:
135 return ret;
136 }
137
138 static
139 void ctf_packet_index_populate(struct ctf_packet_index *index,
140 off_t offset, const struct stream_subbuffer *subbuffer)
141 {
142 *index = (typeof(*index)){
143 .offset = htobe64(offset),
144 .packet_size = htobe64(subbuffer->info.data.packet_size),
145 .content_size = htobe64(subbuffer->info.data.content_size),
146 .timestamp_begin = htobe64(
147 subbuffer->info.data.timestamp_begin),
148 .timestamp_end = htobe64(
149 subbuffer->info.data.timestamp_end),
150 .events_discarded = htobe64(
151 subbuffer->info.data.events_discarded),
152 .stream_id = htobe64(subbuffer->info.data.stream_id),
153 .stream_instance_id = htobe64(
154 subbuffer->info.data.stream_instance_id.is_set ?
155 subbuffer->info.data.stream_instance_id.value : -1ULL),
156 .packet_seq_num = htobe64(
157 subbuffer->info.data.sequence_number.is_set ?
158 subbuffer->info.data.sequence_number.value : -1ULL),
159 };
160 }
161
162 static ssize_t consumer_stream_consume_mmap(
163 struct lttng_consumer_local_data *ctx __attribute__((unused)),
164 struct lttng_consumer_stream *stream,
165 const struct stream_subbuffer *subbuffer)
166 {
167 const unsigned long padding_size =
168 subbuffer->info.data.padded_subbuf_size -
169 subbuffer->info.data.subbuf_size;
170 const ssize_t written_bytes = lttng_consumer_on_read_subbuffer_mmap(
171 stream, &subbuffer->buffer.buffer, padding_size);
172
173 if (stream->net_seq_idx == -1ULL) {
174 /*
175 * When writing on disk, check that only the subbuffer (no
176 * padding) was written to disk.
177 */
178 if (written_bytes != subbuffer->info.data.padded_subbuf_size) {
179 DBG("Failed to write the entire padded subbuffer on disk (written_bytes: %zd, padded subbuffer size %lu)",
180 written_bytes,
181 subbuffer->info.data.padded_subbuf_size);
182 }
183 } else {
184 /*
185 * When streaming over the network, check that the entire
186 * subbuffer including padding was successfully written.
187 */
188 if (written_bytes != subbuffer->info.data.subbuf_size) {
189 DBG("Failed to write only the subbuffer over the network (written_bytes: %zd, subbuffer size %lu)",
190 written_bytes,
191 subbuffer->info.data.subbuf_size);
192 }
193 }
194
195 /*
196 * If `lttng_consumer_on_read_subbuffer_mmap()` returned an error, pass
197 * it along to the caller, else return zero.
198 */
199 if (written_bytes < 0) {
200 ERR("Error reading mmap subbuffer: %zd", written_bytes);
201 }
202
203 return written_bytes;
204 }
205
206 static ssize_t consumer_stream_consume_splice(
207 struct lttng_consumer_local_data *ctx,
208 struct lttng_consumer_stream *stream,
209 const struct stream_subbuffer *subbuffer)
210 {
211 const ssize_t written_bytes = lttng_consumer_on_read_subbuffer_splice(
212 ctx, stream, subbuffer->info.data.padded_subbuf_size, 0);
213
214 if (written_bytes != subbuffer->info.data.padded_subbuf_size) {
215 DBG("Failed to write the entire padded subbuffer (written_bytes: %zd, padded subbuffer size %lu)",
216 written_bytes,
217 subbuffer->info.data.padded_subbuf_size);
218 }
219
220 /*
221 * If `lttng_consumer_on_read_subbuffer_splice()` returned an error,
222 * pass it along to the caller, else return zero.
223 */
224 if (written_bytes < 0) {
225 ERR("Error reading splice subbuffer: %zd", written_bytes);
226 }
227
228 return written_bytes;
229 }
230
231 static int consumer_stream_send_index(
232 struct lttng_consumer_stream *stream,
233 const struct stream_subbuffer *subbuffer,
234 struct lttng_consumer_local_data *ctx __attribute__((unused)))
235 {
236 off_t packet_offset = 0;
237 struct ctf_packet_index index = {};
238
239 /*
240 * This is called after consuming the sub-buffer; substract the
241 * effect this sub-buffer from the offset.
242 */
243 if (stream->net_seq_idx == (uint64_t) -1ULL) {
244 packet_offset = stream->out_fd_offset -
245 subbuffer->info.data.padded_subbuf_size;
246 }
247
248 ctf_packet_index_populate(&index, packet_offset, subbuffer);
249 return consumer_stream_write_index(stream, &index);
250 }
251
252 /*
253 * Actually do the metadata sync using the given metadata stream.
254 *
255 * Return 0 on success else a negative value. ENODATA can be returned also
256 * indicating that there is no metadata available for that stream.
257 */
258 static int do_sync_metadata(struct lttng_consumer_stream *metadata,
259 struct lttng_consumer_local_data *ctx)
260 {
261 int ret;
262 enum sync_metadata_status status;
263
264 LTTNG_ASSERT(metadata);
265 LTTNG_ASSERT(metadata->metadata_flag);
266 LTTNG_ASSERT(ctx);
267
268 /*
269 * In UST, since we have to write the metadata from the cache packet
270 * by packet, we might need to start this procedure multiple times
271 * until all the metadata from the cache has been extracted.
272 */
273 do {
274 /*
275 * Steps :
276 * - Lock the metadata stream
277 * - Check if metadata stream node was deleted before locking.
278 * - if yes, release and return success
279 * - Check if new metadata is ready (flush + snapshot pos)
280 * - If nothing : release and return.
281 * - Lock the metadata_rdv_lock
282 * - Unlock the metadata stream
283 * - cond_wait on metadata_rdv to wait the wakeup from the
284 * metadata thread
285 * - Unlock the metadata_rdv_lock
286 */
287 pthread_mutex_lock(&metadata->lock);
288
289 /*
290 * There is a possibility that we were able to acquire a reference on the
291 * stream from the RCU hash table but between then and now, the node might
292 * have been deleted just before the lock is acquired. Thus, after locking,
293 * we make sure the metadata node has not been deleted which means that the
294 * buffers are closed.
295 *
296 * In that case, there is no need to sync the metadata hence returning a
297 * success return code.
298 */
299 ret = cds_lfht_is_node_deleted(&metadata->node.node);
300 if (ret) {
301 ret = 0;
302 goto end_unlock_mutex;
303 }
304
305 switch (ctx->type) {
306 case LTTNG_CONSUMER_KERNEL:
307 /*
308 * Empty the metadata cache and flush the current stream.
309 */
310 status = lttng_kconsumer_sync_metadata(metadata);
311 break;
312 case LTTNG_CONSUMER32_UST:
313 case LTTNG_CONSUMER64_UST:
314 /*
315 * Ask the sessiond if we have new metadata waiting and update the
316 * consumer metadata cache.
317 */
318 status = lttng_ustconsumer_sync_metadata(ctx, metadata);
319 break;
320 default:
321 abort();
322 }
323
324 switch (status) {
325 case SYNC_METADATA_STATUS_NEW_DATA:
326 break;
327 case SYNC_METADATA_STATUS_NO_DATA:
328 ret = 0;
329 goto end_unlock_mutex;
330 case SYNC_METADATA_STATUS_ERROR:
331 ret = -1;
332 goto end_unlock_mutex;
333 default:
334 abort();
335 }
336
337 /*
338 * At this point, new metadata have been flushed, so we wait on the
339 * rendez-vous point for the metadata thread to wake us up when it
340 * finishes consuming the metadata and continue execution.
341 */
342
343 pthread_mutex_lock(&metadata->metadata_rdv_lock);
344
345 /*
346 * Release metadata stream lock so the metadata thread can process it.
347 */
348 pthread_mutex_unlock(&metadata->lock);
349
350 /*
351 * Wait on the rendez-vous point. Once woken up, it means the metadata was
352 * consumed and thus synchronization is achieved.
353 */
354 pthread_cond_wait(&metadata->metadata_rdv, &metadata->metadata_rdv_lock);
355 pthread_mutex_unlock(&metadata->metadata_rdv_lock);
356 } while (status == SYNC_METADATA_STATUS_NEW_DATA);
357
358 /* Success */
359 return 0;
360
361 end_unlock_mutex:
362 pthread_mutex_unlock(&metadata->lock);
363 return ret;
364 }
365
366 /*
367 * Synchronize the metadata using a given session ID. A successful acquisition
368 * of a metadata stream will trigger a request to the session daemon and a
369 * snapshot so the metadata thread can consume it.
370 *
371 * This function call is a rendez-vous point between the metadata thread and
372 * the data thread.
373 *
374 * Return 0 on success or else a negative value.
375 */
376 int consumer_stream_sync_metadata(struct lttng_consumer_local_data *ctx,
377 uint64_t session_id)
378 {
379 int ret;
380 struct lttng_consumer_stream *stream = NULL;
381 struct lttng_ht_iter iter;
382 struct lttng_ht *ht;
383
384 LTTNG_ASSERT(ctx);
385
386 /* Ease our life a bit. */
387 ht = the_consumer_data.stream_list_ht;
388
389 rcu_read_lock();
390
391 /* Search the metadata associated with the session id of the given stream. */
392
393 cds_lfht_for_each_entry_duplicate(ht->ht,
394 ht->hash_fct(&session_id, lttng_ht_seed), ht->match_fct,
395 &session_id, &iter.iter, stream, node_session_id.node) {
396 if (!stream->metadata_flag) {
397 continue;
398 }
399
400 ret = do_sync_metadata(stream, ctx);
401 if (ret < 0) {
402 goto end;
403 }
404 }
405
406 /*
407 * Force return code to 0 (success) since ret might be ENODATA for instance
408 * which is not an error but rather that we should come back.
409 */
410 ret = 0;
411
412 end:
413 rcu_read_unlock();
414 return ret;
415 }
416
417 static int consumer_stream_sync_metadata_index(
418 struct lttng_consumer_stream *stream,
419 const struct stream_subbuffer *subbuffer,
420 struct lttng_consumer_local_data *ctx)
421 {
422 bool missed_metadata_flush;
423 int ret;
424
425 /* Block until all the metadata is sent. */
426 pthread_mutex_lock(&stream->metadata_timer_lock);
427 LTTNG_ASSERT(!stream->missed_metadata_flush);
428 stream->waiting_on_metadata = true;
429 pthread_mutex_unlock(&stream->metadata_timer_lock);
430
431 ret = consumer_stream_sync_metadata(ctx, stream->session_id);
432
433 pthread_mutex_lock(&stream->metadata_timer_lock);
434 stream->waiting_on_metadata = false;
435 missed_metadata_flush = stream->missed_metadata_flush;
436 if (missed_metadata_flush) {
437 stream->missed_metadata_flush = false;
438 }
439 pthread_mutex_unlock(&stream->metadata_timer_lock);
440 if (ret < 0) {
441 goto end;
442 }
443
444 ret = consumer_stream_send_index(stream, subbuffer, ctx);
445 /*
446 * Send the live inactivity beacon to handle the situation where
447 * the live timer is prevented from sampling this stream
448 * because the stream lock was being held while this stream is
449 * waiting on metadata. This ensures live viewer progress in the
450 * unlikely scenario where a live timer would be prevented from
451 * locking a stream lock repeatedly due to a steady flow of
452 * incoming metadata, for a stream which is mostly inactive.
453 *
454 * It is important to send the inactivity beacon packet to
455 * relayd _after_ sending the index associated with the data
456 * that was just sent, otherwise this can cause live viewers to
457 * observe timestamps going backwards between an inactivity
458 * beacon and a following trace packet.
459 */
460 if (missed_metadata_flush) {
461 (void) stream->read_subbuffer_ops.send_live_beacon(stream);
462 }
463 end:
464 return ret;
465 }
466
467 /*
468 * Check if the local version of the metadata stream matches with the version
469 * of the metadata stream in the kernel. If it was updated, set the reset flag
470 * on the stream.
471 */
472 static
473 int metadata_stream_check_version(struct lttng_consumer_stream *stream,
474 const struct stream_subbuffer *subbuffer)
475 {
476 if (stream->metadata_version == subbuffer->info.metadata.version) {
477 goto end;
478 }
479
480 DBG("New metadata version detected");
481 consumer_stream_metadata_set_version(stream,
482 subbuffer->info.metadata.version);
483
484 if (stream->read_subbuffer_ops.reset_metadata) {
485 stream->read_subbuffer_ops.reset_metadata(stream);
486 }
487
488 end:
489 return 0;
490 }
491
492 static
493 bool stream_is_rotating_to_null_chunk(
494 const struct lttng_consumer_stream *stream)
495 {
496 bool rotating_to_null_chunk = false;
497
498 if (stream->rotate_position == -1ULL) {
499 /* No rotation ongoing. */
500 goto end;
501 }
502
503 if (stream->trace_chunk == stream->chan->trace_chunk ||
504 !stream->chan->trace_chunk) {
505 rotating_to_null_chunk = true;
506 }
507 end:
508 return rotating_to_null_chunk;
509 }
510
511 enum consumer_stream_open_packet_status consumer_stream_open_packet(
512 struct lttng_consumer_stream *stream)
513 {
514 int ret;
515 enum consumer_stream_open_packet_status status;
516 unsigned long produced_pos_before, produced_pos_after;
517
518 ret = lttng_consumer_sample_snapshot_positions(stream);
519 if (ret < 0) {
520 ERR("Failed to snapshot positions before post-rotation empty packet flush: stream id = %" PRIu64
521 ", channel name = %s, session id = %" PRIu64,
522 stream->key, stream->chan->name,
523 stream->chan->session_id);
524 status = CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR;
525 goto end;
526 }
527
528 ret = lttng_consumer_get_produced_snapshot(
529 stream, &produced_pos_before);
530 if (ret < 0) {
531 ERR("Failed to read produced position before post-rotation empty packet flush: stream id = %" PRIu64
532 ", channel name = %s, session id = %" PRIu64,
533 stream->key, stream->chan->name,
534 stream->chan->session_id);
535 status = CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR;
536 goto end;
537 }
538
539 ret = consumer_stream_flush_buffer(stream, 0);
540 if (ret) {
541 ERR("Failed to flush an empty packet at rotation point: stream id = %" PRIu64
542 ", channel name = %s, session id = %" PRIu64,
543 stream->key, stream->chan->name,
544 stream->chan->session_id);
545 status = CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR;
546 goto end;
547 }
548
549 ret = lttng_consumer_sample_snapshot_positions(stream);
550 if (ret < 0) {
551 ERR("Failed to snapshot positions after post-rotation empty packet flush: stream id = %" PRIu64
552 ", channel name = %s, session id = %" PRIu64,
553 stream->key, stream->chan->name,
554 stream->chan->session_id);
555 status = CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR;
556 goto end;
557 }
558
559 ret = lttng_consumer_get_produced_snapshot(stream, &produced_pos_after);
560 if (ret < 0) {
561 ERR("Failed to read produced position after post-rotation empty packet flush: stream id = %" PRIu64
562 ", channel name = %s, session id = %" PRIu64,
563 stream->key, stream->chan->name,
564 stream->chan->session_id);
565 status = CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR;
566 goto end;
567 }
568
569 /*
570 * Determine if the flush had an effect by comparing the produced
571 * positons before and after the flush.
572 */
573 status = produced_pos_before != produced_pos_after ?
574 CONSUMER_STREAM_OPEN_PACKET_STATUS_OPENED :
575 CONSUMER_STREAM_OPEN_PACKET_STATUS_NO_SPACE;
576 if (status == CONSUMER_STREAM_OPEN_PACKET_STATUS_OPENED) {
577 stream->opened_packet_in_current_trace_chunk = true;
578 }
579
580 end:
581 return status;
582 }
583
584 /*
585 * An attempt to open a new packet is performed after a rotation completes to
586 * get a begin timestamp as close as possible to the rotation point.
587 *
588 * However, that initial attempt at opening a packet can fail due to a full
589 * ring-buffer. In that case, a second attempt is performed after consuming
590 * a packet since that will have freed enough space in the ring-buffer.
591 */
592 static
593 int post_consume_open_new_packet(struct lttng_consumer_stream *stream,
594 const struct stream_subbuffer *subbuffer __attribute__((unused)),
595 struct lttng_consumer_local_data *ctx __attribute__((unused)))
596 {
597 int ret = 0;
598
599 if (!stream->opened_packet_in_current_trace_chunk &&
600 stream->trace_chunk &&
601 !stream_is_rotating_to_null_chunk(stream)) {
602 const enum consumer_stream_open_packet_status status =
603 consumer_stream_open_packet(stream);
604
605 switch (status) {
606 case CONSUMER_STREAM_OPEN_PACKET_STATUS_OPENED:
607 DBG("Opened a packet after consuming a packet rotation: stream id = %" PRIu64
608 ", channel name = %s, session id = %" PRIu64,
609 stream->key, stream->chan->name,
610 stream->chan->session_id);
611 stream->opened_packet_in_current_trace_chunk = true;
612 break;
613 case CONSUMER_STREAM_OPEN_PACKET_STATUS_NO_SPACE:
614 /*
615 * Can't open a packet as there is no space left.
616 * This means that new events were produced, resulting
617 * in a packet being opened, which is what we want
618 * anyhow.
619 */
620 DBG("No space left to open a packet after consuming a packet: stream id = %" PRIu64
621 ", channel name = %s, session id = %" PRIu64,
622 stream->key, stream->chan->name,
623 stream->chan->session_id);
624 stream->opened_packet_in_current_trace_chunk = true;
625 break;
626 case CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR:
627 /* Logged by callee. */
628 ret = -1;
629 goto end;
630 default:
631 abort();
632 }
633
634 stream->opened_packet_in_current_trace_chunk = true;
635 }
636
637 end:
638 return ret;
639 }
640
641 struct lttng_consumer_stream *consumer_stream_create(
642 struct lttng_consumer_channel *channel,
643 uint64_t channel_key,
644 uint64_t stream_key,
645 const char *channel_name,
646 uint64_t relayd_id,
647 uint64_t session_id,
648 struct lttng_trace_chunk *trace_chunk,
649 int cpu,
650 int *alloc_ret,
651 enum consumer_channel_type type,
652 unsigned int monitor)
653 {
654 int ret;
655 struct lttng_consumer_stream *stream;
656
657 stream = zmalloc<lttng_consumer_stream>();
658 if (stream == NULL) {
659 PERROR("malloc struct lttng_consumer_stream");
660 ret = -ENOMEM;
661 goto end;
662 }
663
664 rcu_read_lock();
665
666 if (trace_chunk && !lttng_trace_chunk_get(trace_chunk)) {
667 ERR("Failed to acquire trace chunk reference during the creation of a stream");
668 ret = -1;
669 goto error;
670 }
671
672 stream->send_node = CDS_LIST_HEAD_INIT(stream->send_node);
673 stream->chan = channel;
674 stream->key = stream_key;
675 stream->trace_chunk = trace_chunk;
676 stream->out_fd = -1;
677 stream->out_fd_offset = 0;
678 stream->output_written = 0;
679 stream->net_seq_idx = relayd_id;
680 stream->session_id = session_id;
681 stream->monitor = monitor;
682 stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE;
683 stream->index_file = NULL;
684 stream->last_sequence_number = -1ULL;
685 stream->rotate_position = -1ULL;
686 /* Buffer is created with an open packet. */
687 stream->opened_packet_in_current_trace_chunk = true;
688 pthread_mutex_init(&stream->lock, NULL);
689 pthread_mutex_init(&stream->metadata_timer_lock, NULL);
690
691 /* If channel is the metadata, flag this stream as metadata. */
692 if (type == CONSUMER_CHANNEL_TYPE_METADATA) {
693 stream->metadata_flag = 1;
694 /* Metadata is flat out. */
695 strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name));
696 /* Live rendez-vous point. */
697 pthread_cond_init(&stream->metadata_rdv, NULL);
698 pthread_mutex_init(&stream->metadata_rdv_lock, NULL);
699 } else {
700 /* Format stream name to <channel_name>_<cpu_number> */
701 ret = snprintf(stream->name, sizeof(stream->name), "%s_%d",
702 channel_name, cpu);
703 if (ret < 0) {
704 PERROR("snprintf stream name");
705 goto error;
706 }
707 }
708
709 switch (channel->output) {
710 case CONSUMER_CHANNEL_SPLICE:
711 stream->output = LTTNG_EVENT_SPLICE;
712 ret = utils_create_pipe(stream->splice_pipe);
713 if (ret < 0) {
714 goto error;
715 }
716 break;
717 case CONSUMER_CHANNEL_MMAP:
718 stream->output = LTTNG_EVENT_MMAP;
719 break;
720 default:
721 abort();
722 }
723
724 /* Key is always the wait_fd for streams. */
725 lttng_ht_node_init_u64(&stream->node, stream->key);
726
727 /* Init node per channel id key */
728 lttng_ht_node_init_u64(&stream->node_channel_id, channel_key);
729
730 /* Init session id node with the stream session id */
731 lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id);
732
733 DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64
734 " relayd_id %" PRIu64 ", session_id %" PRIu64,
735 stream->name, stream->key, channel_key,
736 stream->net_seq_idx, stream->session_id);
737
738 rcu_read_unlock();
739
740 lttng_dynamic_array_init(&stream->read_subbuffer_ops.post_consume_cbs,
741 sizeof(post_consume_cb), NULL);
742
743 if (type == CONSUMER_CHANNEL_TYPE_METADATA) {
744 stream->read_subbuffer_ops.lock =
745 consumer_stream_metadata_lock_all;
746 stream->read_subbuffer_ops.unlock =
747 consumer_stream_metadata_unlock_all;
748 stream->read_subbuffer_ops.assert_locked =
749 consumer_stream_metadata_assert_locked_all;
750 stream->read_subbuffer_ops.pre_consume_subbuffer =
751 metadata_stream_check_version;
752 } else {
753 const post_consume_cb post_consume_index_op = channel->is_live ?
754 consumer_stream_sync_metadata_index :
755 consumer_stream_send_index;
756 const post_consume_cb post_consume_open_new_packet_ =
757 post_consume_open_new_packet;
758
759 ret = lttng_dynamic_array_add_element(
760 &stream->read_subbuffer_ops.post_consume_cbs,
761 &post_consume_index_op);
762 if (ret) {
763 PERROR("Failed to add `send index` callback to stream's post consumption callbacks");
764 goto error;
765 }
766
767 ret = lttng_dynamic_array_add_element(
768 &stream->read_subbuffer_ops.post_consume_cbs,
769 &post_consume_open_new_packet_);
770 if (ret) {
771 PERROR("Failed to add `open new packet` callback to stream's post consumption callbacks");
772 goto error;
773 }
774
775 stream->read_subbuffer_ops.lock = consumer_stream_data_lock_all;
776 stream->read_subbuffer_ops.unlock =
777 consumer_stream_data_unlock_all;
778 stream->read_subbuffer_ops.assert_locked =
779 consumer_stream_data_assert_locked_all;
780 stream->read_subbuffer_ops.pre_consume_subbuffer =
781 consumer_stream_update_stats;
782 }
783
784 if (channel->output == CONSUMER_CHANNEL_MMAP) {
785 stream->read_subbuffer_ops.consume_subbuffer =
786 consumer_stream_consume_mmap;
787 } else {
788 stream->read_subbuffer_ops.consume_subbuffer =
789 consumer_stream_consume_splice;
790 }
791
792 return stream;
793
794 error:
795 rcu_read_unlock();
796 lttng_trace_chunk_put(stream->trace_chunk);
797 lttng_dynamic_array_reset(&stream->read_subbuffer_ops.post_consume_cbs);
798 free(stream);
799 end:
800 if (alloc_ret) {
801 *alloc_ret = ret;
802 }
803 return NULL;
804 }
805
806 /*
807 * Close stream on the relayd side. This call can destroy a relayd if the
808 * conditions are met.
809 *
810 * A RCU read side lock MUST be acquired if the relayd object was looked up in
811 * a hash table before calling this.
812 */
813 void consumer_stream_relayd_close(struct lttng_consumer_stream *stream,
814 struct consumer_relayd_sock_pair *relayd)
815 {
816 int ret;
817
818 LTTNG_ASSERT(stream);
819 LTTNG_ASSERT(relayd);
820
821 if (stream->sent_to_relayd) {
822 uatomic_dec(&relayd->refcount);
823 LTTNG_ASSERT(uatomic_read(&relayd->refcount) >= 0);
824 }
825
826 /* Closing streams requires to lock the control socket. */
827 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
828 ret = relayd_send_close_stream(&relayd->control_sock,
829 stream->relayd_stream_id,
830 stream->next_net_seq_num - 1);
831 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
832 if (ret < 0) {
833 ERR("Relayd send close stream failed. Cleaning up relayd %" PRIu64 ".", relayd->net_seq_idx);
834 lttng_consumer_cleanup_relayd(relayd);
835 }
836
837 /* Both conditions are met, we destroy the relayd. */
838 if (uatomic_read(&relayd->refcount) == 0 &&
839 uatomic_read(&relayd->destroy_flag)) {
840 consumer_destroy_relayd(relayd);
841 }
842 stream->net_seq_idx = (uint64_t) -1ULL;
843 stream->sent_to_relayd = 0;
844 }
845
846 /*
847 * Close stream's file descriptors and, if needed, close stream also on the
848 * relayd side.
849 *
850 * The consumer data lock MUST be acquired.
851 * The stream lock MUST be acquired.
852 */
853 void consumer_stream_close_output(struct lttng_consumer_stream *stream)
854 {
855 struct consumer_relayd_sock_pair *relayd;
856
857 LTTNG_ASSERT(stream);
858
859 /* Close output fd. Could be a socket or local file at this point. */
860 if (stream->out_fd >= 0) {
861 const auto ret = close(stream->out_fd);
862 if (ret) {
863 PERROR("Failed to close stream output file descriptor");
864 }
865
866 stream->out_fd = -1;
867 }
868
869 if (stream->index_file) {
870 lttng_index_file_put(stream->index_file);
871 stream->index_file = NULL;
872 }
873
874 lttng_trace_chunk_put(stream->trace_chunk);
875 stream->trace_chunk = NULL;
876
877 /* Check and cleanup relayd if needed. */
878 rcu_read_lock();
879 relayd = consumer_find_relayd(stream->net_seq_idx);
880 if (relayd != NULL) {
881 consumer_stream_relayd_close(stream, relayd);
882 stream->net_seq_idx = -1ULL;
883 }
884
885 rcu_read_unlock();
886 }
887
888 /*
889 * Delete the stream from all possible hash tables.
890 *
891 * The consumer data lock MUST be acquired.
892 * The stream lock MUST be acquired.
893 */
894 void consumer_stream_delete(struct lttng_consumer_stream *stream,
895 struct lttng_ht *ht)
896 {
897 int ret;
898 struct lttng_ht_iter iter;
899
900 LTTNG_ASSERT(stream);
901 /* Should NEVER be called not in monitor mode. */
902 LTTNG_ASSERT(stream->chan->monitor);
903
904 rcu_read_lock();
905
906 if (ht) {
907 iter.iter.node = &stream->node.node;
908 ret = lttng_ht_del(ht, &iter);
909 LTTNG_ASSERT(!ret);
910 }
911
912 /* Delete from stream per channel ID hash table. */
913 iter.iter.node = &stream->node_channel_id.node;
914 /*
915 * The returned value is of no importance. Even if the node is NOT in the
916 * hash table, we continue since we may have been called by a code path
917 * that did not add the stream to a (all) hash table. Same goes for the
918 * next call ht del call.
919 */
920 (void) lttng_ht_del(the_consumer_data.stream_per_chan_id_ht, &iter);
921
922 /* Delete from the global stream list. */
923 iter.iter.node = &stream->node_session_id.node;
924 /* See the previous ht del on why we ignore the returned value. */
925 (void) lttng_ht_del(the_consumer_data.stream_list_ht, &iter);
926
927 rcu_read_unlock();
928
929 if (!stream->metadata_flag) {
930 /* Decrement the stream count of the global consumer data. */
931 LTTNG_ASSERT(the_consumer_data.stream_count > 0);
932 the_consumer_data.stream_count--;
933 }
934 }
935
936 /*
937 * Free the given stream within a RCU call.
938 */
939 void consumer_stream_free(struct lttng_consumer_stream *stream)
940 {
941 LTTNG_ASSERT(stream);
942
943 metadata_bucket_destroy(stream->metadata_bucket);
944 call_rcu(&stream->node.head, free_stream_rcu);
945 }
946
947 /*
948 * Destroy the stream's buffers of the tracer.
949 */
950 void consumer_stream_destroy_buffers(struct lttng_consumer_stream *stream)
951 {
952 LTTNG_ASSERT(stream);
953
954 switch (the_consumer_data.type) {
955 case LTTNG_CONSUMER_KERNEL:
956 if (stream->mmap_base != NULL) {
957 const auto ret = munmap(stream->mmap_base, stream->mmap_len);
958
959 if (ret != 0) {
960 PERROR("munmap");
961 }
962 }
963
964 if (stream->wait_fd >= 0) {
965 const auto ret = close(stream->wait_fd);
966
967 if (ret) {
968 PERROR("close");
969 }
970
971 stream->wait_fd = -1;
972 }
973
974 if (stream->chan->output == CONSUMER_CHANNEL_SPLICE) {
975 utils_close_pipe(stream->splice_pipe);
976 }
977
978 break;
979 case LTTNG_CONSUMER32_UST:
980 case LTTNG_CONSUMER64_UST:
981 /*
982 * Special case for the metadata since the wait fd is an internal pipe
983 * polled in the metadata thread.
984 */
985 if (stream->metadata_flag && stream->chan->monitor) {
986 const auto rpipe = stream->ust_metadata_poll_pipe[0];
987
988 /*
989 * This will stop the channel timer if one and close the write side
990 * of the metadata poll pipe.
991 */
992 lttng_ustconsumer_close_metadata(stream->chan);
993 if (rpipe >= 0) {
994 const auto ret = close(rpipe);
995
996 if (ret < 0) {
997 PERROR("closing metadata pipe read side");
998 }
999
1000 stream->ust_metadata_poll_pipe[0] = -1;
1001 }
1002 }
1003
1004 lttng_ustconsumer_del_stream(stream);
1005 break;
1006 default:
1007 ERR("Unknown consumer_data type");
1008 abort();
1009 }
1010 }
1011
1012 /*
1013 * Destroy and close a already created stream.
1014 */
1015 static void destroy_close_stream(struct lttng_consumer_stream *stream)
1016 {
1017 LTTNG_ASSERT(stream);
1018
1019 DBG("Consumer stream destroy monitored key: %" PRIu64, stream->key);
1020
1021 /* Destroy tracer buffers of the stream. */
1022 consumer_stream_destroy_buffers(stream);
1023 /* Close down everything including the relayd if one. */
1024 consumer_stream_close_output(stream);
1025 }
1026
1027 /*
1028 * Decrement the stream's channel refcount and if down to 0, return the channel
1029 * pointer so it can be destroyed by the caller or NULL if not.
1030 */
1031 static struct lttng_consumer_channel *unref_channel(
1032 struct lttng_consumer_stream *stream)
1033 {
1034 struct lttng_consumer_channel *free_chan = NULL;
1035
1036 LTTNG_ASSERT(stream);
1037 LTTNG_ASSERT(stream->chan);
1038
1039 /* Update refcount of channel and see if we need to destroy it. */
1040 if (!uatomic_sub_return(&stream->chan->refcount, 1)
1041 && !uatomic_read(&stream->chan->nb_init_stream_left)) {
1042 free_chan = stream->chan;
1043 }
1044
1045 return free_chan;
1046 }
1047
1048 /*
1049 * Destroy a stream completely. This will delete, close and free the stream.
1050 * Once return, the stream is NO longer usable. Its channel may get destroyed
1051 * if conditions are met for a monitored stream.
1052 *
1053 * This MUST be called WITHOUT the consumer data and stream lock acquired if
1054 * the stream is in _monitor_ mode else it does not matter.
1055 */
1056 void consumer_stream_destroy(struct lttng_consumer_stream *stream,
1057 struct lttng_ht *ht)
1058 {
1059 LTTNG_ASSERT(stream);
1060
1061 cds_list_del_init(&stream->send_node);
1062
1063 /* Stream is in monitor mode. */
1064 if (stream->monitor) {
1065 struct lttng_consumer_channel *free_chan = NULL;
1066
1067 /*
1068 * This means that the stream was successfully removed from the streams
1069 * list of the channel and sent to the right thread managing this
1070 * stream thus being globally visible.
1071 */
1072 if (stream->globally_visible) {
1073 pthread_mutex_lock(&the_consumer_data.lock);
1074 pthread_mutex_lock(&stream->chan->lock);
1075
1076 pthread_mutex_lock(&stream->lock);
1077 /* Remove every reference of the stream in the consumer. */
1078 consumer_stream_delete(stream, ht);
1079
1080
1081 destroy_close_stream(stream);
1082
1083 /* Update channel's refcount of the stream. */
1084 free_chan = unref_channel(stream);
1085
1086 /* Indicates that the consumer data state MUST be updated after this. */
1087 the_consumer_data.need_update = 1;
1088
1089 pthread_mutex_unlock(&stream->lock);
1090 pthread_mutex_unlock(&stream->chan->lock);
1091 pthread_mutex_unlock(&the_consumer_data.lock);
1092 } else {
1093 /*
1094 * If the stream is not visible globally, this needs to be done
1095 * outside of the consumer data lock section.
1096 */
1097 destroy_close_stream(stream);
1098 free_chan = unref_channel(stream);
1099 }
1100
1101 if (free_chan) {
1102 consumer_del_channel(free_chan);
1103 }
1104 } else {
1105 destroy_close_stream(stream);
1106 }
1107
1108 /* Free stream within a RCU call. */
1109 lttng_trace_chunk_put(stream->trace_chunk);
1110 stream->trace_chunk = NULL;
1111 lttng_dynamic_array_reset(&stream->read_subbuffer_ops.post_consume_cbs);
1112 consumer_stream_free(stream);
1113 }
1114
1115 /*
1116 * Write index of a specific stream either on the relayd or local disk.
1117 *
1118 * Return 0 on success or else a negative value.
1119 */
1120 int consumer_stream_write_index(struct lttng_consumer_stream *stream,
1121 struct ctf_packet_index *element)
1122 {
1123 int ret;
1124
1125 LTTNG_ASSERT(stream);
1126 LTTNG_ASSERT(element);
1127
1128 rcu_read_lock();
1129 if (stream->net_seq_idx != (uint64_t) -1ULL) {
1130 struct consumer_relayd_sock_pair *relayd;
1131 relayd = consumer_find_relayd(stream->net_seq_idx);
1132 if (relayd) {
1133 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1134 ret = relayd_send_index(&relayd->control_sock, element,
1135 stream->relayd_stream_id, stream->next_net_seq_num - 1);
1136 if (ret < 0) {
1137 /*
1138 * Communication error with lttng-relayd,
1139 * perform cleanup now
1140 */
1141 ERR("Relayd send index failed. Cleaning up relayd %" PRIu64 ".", relayd->net_seq_idx);
1142 lttng_consumer_cleanup_relayd(relayd);
1143 ret = -1;
1144 }
1145 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1146 } else {
1147 ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't write index.",
1148 stream->key, stream->net_seq_idx);
1149 ret = -1;
1150 }
1151 } else {
1152 if (lttng_index_file_write(stream->index_file, element)) {
1153 ret = -1;
1154 } else {
1155 ret = 0;
1156 }
1157 }
1158 if (ret < 0) {
1159 goto error;
1160 }
1161
1162 error:
1163 rcu_read_unlock();
1164 return ret;
1165 }
1166
1167 int consumer_stream_create_output_files(struct lttng_consumer_stream *stream,
1168 bool create_index)
1169 {
1170 int ret;
1171 enum lttng_trace_chunk_status chunk_status;
1172 const int flags = O_WRONLY | O_CREAT | O_TRUNC;
1173 const mode_t mode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP;
1174 char stream_path[LTTNG_PATH_MAX];
1175
1176 ASSERT_LOCKED(stream->lock);
1177 LTTNG_ASSERT(stream->trace_chunk);
1178
1179 ret = utils_stream_file_path(stream->chan->pathname, stream->name,
1180 stream->chan->tracefile_size,
1181 stream->tracefile_count_current, NULL,
1182 stream_path, sizeof(stream_path));
1183 if (ret < 0) {
1184 goto end;
1185 }
1186
1187 if (stream->out_fd >= 0) {
1188 ret = close(stream->out_fd);
1189 if (ret < 0) {
1190 PERROR("Failed to close stream file \"%s\"",
1191 stream->name);
1192 goto end;
1193 }
1194 stream->out_fd = -1;
1195 }
1196
1197 DBG("Opening stream output file \"%s\"", stream_path);
1198 chunk_status = lttng_trace_chunk_open_file(stream->trace_chunk, stream_path,
1199 flags, mode, &stream->out_fd, false);
1200 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1201 ERR("Failed to open stream file \"%s\"", stream->name);
1202 ret = -1;
1203 goto end;
1204 }
1205
1206 if (!stream->metadata_flag && (create_index || stream->index_file)) {
1207 if (stream->index_file) {
1208 lttng_index_file_put(stream->index_file);
1209 }
1210 chunk_status = lttng_index_file_create_from_trace_chunk(
1211 stream->trace_chunk,
1212 stream->chan->pathname,
1213 stream->name,
1214 stream->chan->tracefile_size,
1215 stream->tracefile_count_current,
1216 CTF_INDEX_MAJOR, CTF_INDEX_MINOR,
1217 false, &stream->index_file);
1218 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1219 ret = -1;
1220 goto end;
1221 }
1222 }
1223
1224 /* Reset current size because we just perform a rotation. */
1225 stream->tracefile_size_current = 0;
1226 stream->out_fd_offset = 0;
1227 end:
1228 return ret;
1229 }
1230
1231 int consumer_stream_rotate_output_files(struct lttng_consumer_stream *stream)
1232 {
1233 int ret;
1234
1235 stream->tracefile_count_current++;
1236 if (stream->chan->tracefile_count > 0) {
1237 stream->tracefile_count_current %=
1238 stream->chan->tracefile_count;
1239 }
1240
1241 DBG("Rotating output files of stream \"%s\"", stream->name);
1242 ret = consumer_stream_create_output_files(stream, true);
1243 if (ret) {
1244 goto end;
1245 }
1246
1247 end:
1248 return ret;
1249 }
1250
1251 bool consumer_stream_is_deleted(struct lttng_consumer_stream *stream)
1252 {
1253 /*
1254 * This function does not take a const stream since
1255 * cds_lfht_is_node_deleted was not const before liburcu 0.12.
1256 */
1257 LTTNG_ASSERT(stream);
1258 return cds_lfht_is_node_deleted(&stream->node.node);
1259 }
1260
1261 static ssize_t metadata_bucket_flush(
1262 const struct stream_subbuffer *buffer, void *data)
1263 {
1264 ssize_t ret;
1265 struct lttng_consumer_stream *stream = (lttng_consumer_stream *) data;
1266
1267 ret = consumer_stream_consume_mmap(NULL, stream, buffer);
1268 if (ret < 0) {
1269 goto end;
1270 }
1271 end:
1272 return ret;
1273 }
1274
1275 static ssize_t metadata_bucket_consume(
1276 struct lttng_consumer_local_data *unused __attribute__((unused)),
1277 struct lttng_consumer_stream *stream,
1278 const struct stream_subbuffer *subbuffer)
1279 {
1280 ssize_t ret;
1281 enum metadata_bucket_status status;
1282
1283 status = metadata_bucket_fill(stream->metadata_bucket, subbuffer);
1284 switch (status) {
1285 case METADATA_BUCKET_STATUS_OK:
1286 /* Return consumed size. */
1287 ret = subbuffer->buffer.buffer.size;
1288 break;
1289 default:
1290 ret = -1;
1291 }
1292
1293 return ret;
1294 }
1295
1296 int consumer_stream_enable_metadata_bucketization(
1297 struct lttng_consumer_stream *stream)
1298 {
1299 int ret = 0;
1300
1301 LTTNG_ASSERT(stream->metadata_flag);
1302 LTTNG_ASSERT(!stream->metadata_bucket);
1303 LTTNG_ASSERT(stream->chan->output == CONSUMER_CHANNEL_MMAP);
1304
1305 stream->metadata_bucket = metadata_bucket_create(
1306 metadata_bucket_flush, stream);
1307 if (!stream->metadata_bucket) {
1308 ret = -1;
1309 goto end;
1310 }
1311
1312 stream->read_subbuffer_ops.consume_subbuffer = metadata_bucket_consume;
1313 end:
1314 return ret;
1315 }
1316
1317 void consumer_stream_metadata_set_version(
1318 struct lttng_consumer_stream *stream, uint64_t new_version)
1319 {
1320 LTTNG_ASSERT(new_version > stream->metadata_version);
1321 stream->metadata_version = new_version;
1322 stream->reset_metadata_flag = 1;
1323
1324 if (stream->metadata_bucket) {
1325 metadata_bucket_reset(stream->metadata_bucket);
1326 }
1327 }
1328
1329 int consumer_stream_flush_buffer(struct lttng_consumer_stream *stream,
1330 bool producer_active)
1331 {
1332 int ret = 0;
1333
1334 switch (the_consumer_data.type) {
1335 case LTTNG_CONSUMER_KERNEL:
1336 if (producer_active) {
1337 ret = kernctl_buffer_flush(stream->wait_fd);
1338 if (ret < 0) {
1339 ERR("Failed to flush kernel stream");
1340 goto end;
1341 }
1342 } else {
1343 ret = kernctl_buffer_flush_empty(stream->wait_fd);
1344 if (ret < 0) {
1345 /*
1346 * Doing a buffer flush which does not take into
1347 * account empty packets. This is not perfect,
1348 * but required as a fall-back when
1349 * "flush_empty" is not implemented by
1350 * lttng-modules.
1351 */
1352 ret = kernctl_buffer_flush(stream->wait_fd);
1353 if (ret < 0) {
1354 ERR("Failed to flush kernel stream");
1355 goto end;
1356 }
1357 }
1358 }
1359 break;
1360 case LTTNG_CONSUMER32_UST:
1361 case LTTNG_CONSUMER64_UST:
1362 ret = lttng_ustconsumer_flush_buffer(stream, (int) producer_active);
1363 break;
1364 default:
1365 ERR("Unknown consumer_data type");
1366 abort();
1367 }
1368
1369 end:
1370 return ret;
1371 }
This page took 0.093914 seconds and 5 git commands to generate.