Fix: ust-consumer: segfault on snapshot after regenerate metadata
[lttng-tools.git] / src / common / consumer / consumer-stream.c
1 /*
2 * Copyright (C) 2011 Julien Desfossez <julien.desfossez@polymtl.ca>
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 <assert.h>
12 #include <inttypes.h>
13 #include <sys/mman.h>
14 #include <unistd.h>
15
16 #include <common/common.h>
17 #include <common/consumer/consumer-timer.h>
18 #include <common/consumer/consumer-timer.h>
19 #include <common/consumer/consumer.h>
20 #include <common/consumer/consumer.h>
21 #include <common/consumer/metadata-bucket.h>
22 #include <common/consumer/metadata-bucket.h>
23 #include <common/index/index.h>
24 #include <common/kernel-consumer/kernel-consumer.h>
25 #include <common/kernel-ctl/kernel-ctl.h>
26 #include <common/macros.h>
27 #include <common/relayd/relayd.h>
28 #include <common/ust-consumer/ust-consumer.h>
29 #include <common/utils.h>
30
31 #include "consumer-stream.h"
32
33 /*
34 * RCU call to free stream. MUST only be used with call_rcu().
35 */
36 static void free_stream_rcu(struct rcu_head *head)
37 {
38 struct lttng_ht_node_u64 *node =
39 caa_container_of(head, struct lttng_ht_node_u64, head);
40 struct lttng_consumer_stream *stream =
41 caa_container_of(node, struct lttng_consumer_stream, node);
42
43 pthread_mutex_destroy(&stream->lock);
44 free(stream);
45 }
46
47 static void consumer_stream_data_lock_all(struct lttng_consumer_stream *stream)
48 {
49 pthread_mutex_lock(&stream->chan->lock);
50 pthread_mutex_lock(&stream->lock);
51 }
52
53 static void consumer_stream_data_unlock_all(struct lttng_consumer_stream *stream)
54 {
55 pthread_mutex_unlock(&stream->lock);
56 pthread_mutex_unlock(&stream->chan->lock);
57 }
58
59 static void consumer_stream_data_assert_locked_all(struct lttng_consumer_stream *stream)
60 {
61 ASSERT_LOCKED(stream->lock);
62 ASSERT_LOCKED(stream->chan->lock);
63 }
64
65 static void consumer_stream_metadata_lock_all(struct lttng_consumer_stream *stream)
66 {
67 consumer_stream_data_lock_all(stream);
68 pthread_mutex_lock(&stream->metadata_rdv_lock);
69 }
70
71 static void consumer_stream_metadata_unlock_all(struct lttng_consumer_stream *stream)
72 {
73 pthread_mutex_unlock(&stream->metadata_rdv_lock);
74 consumer_stream_data_unlock_all(stream);
75 }
76
77 static void consumer_stream_metadata_assert_locked_all(struct lttng_consumer_stream *stream)
78 {
79 ASSERT_LOCKED(stream->metadata_rdv_lock);
80 consumer_stream_data_assert_locked_all(stream);
81 }
82
83 /* Only used for data streams. */
84 static int consumer_stream_update_stats(struct lttng_consumer_stream *stream,
85 const struct stream_subbuffer *subbuf)
86 {
87 int ret = 0;
88 uint64_t sequence_number;
89 const uint64_t discarded_events = subbuf->info.data.events_discarded;
90
91 if (!subbuf->info.data.sequence_number.is_set) {
92 /* Command not supported by the tracer. */
93 sequence_number = -1ULL;
94 stream->sequence_number_unavailable = true;
95 } else {
96 sequence_number = subbuf->info.data.sequence_number.value;
97 }
98
99 /*
100 * Start the sequence when we extract the first packet in case we don't
101 * start at 0 (for example if a consumer is not connected to the
102 * session immediately after the beginning).
103 */
104 if (stream->last_sequence_number == -1ULL) {
105 stream->last_sequence_number = sequence_number;
106 } else if (sequence_number > stream->last_sequence_number) {
107 stream->chan->lost_packets += sequence_number -
108 stream->last_sequence_number - 1;
109 } else {
110 /* seq <= last_sequence_number */
111 ERR("Sequence number inconsistent : prev = %" PRIu64
112 ", current = %" PRIu64,
113 stream->last_sequence_number, sequence_number);
114 ret = -1;
115 goto end;
116 }
117 stream->last_sequence_number = sequence_number;
118
119 if (discarded_events < stream->last_discarded_events) {
120 /*
121 * Overflow has occurred. We assume only one wrap-around
122 * has occurred.
123 */
124 stream->chan->discarded_events +=
125 (1ULL << (CAA_BITS_PER_LONG - 1)) -
126 stream->last_discarded_events +
127 discarded_events;
128 } else {
129 stream->chan->discarded_events += discarded_events -
130 stream->last_discarded_events;
131 }
132 stream->last_discarded_events = discarded_events;
133 ret = 0;
134
135 end:
136 return ret;
137 }
138
139 static
140 void ctf_packet_index_populate(struct ctf_packet_index *index,
141 off_t offset, const struct stream_subbuffer *subbuffer)
142 {
143 *index = (typeof(*index)){
144 .offset = htobe64(offset),
145 .packet_size = htobe64(subbuffer->info.data.packet_size),
146 .content_size = htobe64(subbuffer->info.data.content_size),
147 .timestamp_begin = htobe64(
148 subbuffer->info.data.timestamp_begin),
149 .timestamp_end = htobe64(
150 subbuffer->info.data.timestamp_end),
151 .events_discarded = htobe64(
152 subbuffer->info.data.events_discarded),
153 .stream_id = htobe64(subbuffer->info.data.stream_id),
154 .stream_instance_id = htobe64(
155 subbuffer->info.data.stream_instance_id.is_set ?
156 subbuffer->info.data.stream_instance_id.value : -1ULL),
157 .packet_seq_num = htobe64(
158 subbuffer->info.data.sequence_number.is_set ?
159 subbuffer->info.data.sequence_number.value : -1ULL),
160 };
161 }
162
163 static ssize_t consumer_stream_consume_mmap(
164 struct lttng_consumer_local_data *ctx,
165 struct lttng_consumer_stream *stream,
166 const struct stream_subbuffer *subbuffer)
167 {
168 const unsigned long padding_size =
169 subbuffer->info.data.padded_subbuf_size -
170 subbuffer->info.data.subbuf_size;
171 const ssize_t written_bytes = lttng_consumer_on_read_subbuffer_mmap(
172 stream, &subbuffer->buffer.buffer, padding_size);
173
174 if (stream->net_seq_idx == -1ULL) {
175 /*
176 * When writing on disk, check that only the subbuffer (no
177 * padding) was written to disk.
178 */
179 if (written_bytes != subbuffer->info.data.padded_subbuf_size) {
180 DBG("Failed to write the entire padded subbuffer on disk (written_bytes: %zd, padded subbuffer size %lu)",
181 written_bytes,
182 subbuffer->info.data.padded_subbuf_size);
183 }
184 } else {
185 /*
186 * When streaming over the network, check that the entire
187 * subbuffer including padding was successfully written.
188 */
189 if (written_bytes != subbuffer->info.data.subbuf_size) {
190 DBG("Failed to write only the subbuffer over the network (written_bytes: %zd, subbuffer size %lu)",
191 written_bytes,
192 subbuffer->info.data.subbuf_size);
193 }
194 }
195
196 /*
197 * If `lttng_consumer_on_read_subbuffer_mmap()` returned an error, pass
198 * it along to the caller, else return zero.
199 */
200 if (written_bytes < 0) {
201 ERR("Error reading mmap subbuffer: %zd", written_bytes);
202 }
203
204 return written_bytes;
205 }
206
207 static ssize_t consumer_stream_consume_splice(
208 struct lttng_consumer_local_data *ctx,
209 struct lttng_consumer_stream *stream,
210 const struct stream_subbuffer *subbuffer)
211 {
212 const ssize_t written_bytes = lttng_consumer_on_read_subbuffer_splice(
213 ctx, stream, subbuffer->info.data.padded_subbuf_size, 0);
214
215 if (written_bytes != subbuffer->info.data.padded_subbuf_size) {
216 DBG("Failed to write the entire padded subbuffer (written_bytes: %zd, padded subbuffer size %lu)",
217 written_bytes,
218 subbuffer->info.data.padded_subbuf_size);
219 }
220
221 /*
222 * If `lttng_consumer_on_read_subbuffer_splice()` returned an error,
223 * pass it along to the caller, else return zero.
224 */
225 if (written_bytes < 0) {
226 ERR("Error reading splice subbuffer: %zd", written_bytes);
227 }
228
229 return written_bytes;
230 }
231
232 static int consumer_stream_send_index(
233 struct lttng_consumer_stream *stream,
234 const struct stream_subbuffer *subbuffer,
235 struct lttng_consumer_local_data *ctx)
236 {
237 off_t packet_offset = 0;
238 struct ctf_packet_index index = {};
239
240 /*
241 * This is called after consuming the sub-buffer; substract the
242 * effect this sub-buffer from the offset.
243 */
244 if (stream->net_seq_idx == (uint64_t) -1ULL) {
245 packet_offset = stream->out_fd_offset -
246 subbuffer->info.data.padded_subbuf_size;
247 }
248
249 ctf_packet_index_populate(&index, packet_offset, subbuffer);
250 return consumer_stream_write_index(stream, &index);
251 }
252
253 /*
254 * Actually do the metadata sync using the given metadata stream.
255 *
256 * Return 0 on success else a negative value. ENODATA can be returned also
257 * indicating that there is no metadata available for that stream.
258 */
259 static int do_sync_metadata(struct lttng_consumer_stream *metadata,
260 struct lttng_consumer_local_data *ctx)
261 {
262 int ret;
263 enum sync_metadata_status status;
264
265 assert(metadata);
266 assert(metadata->metadata_flag);
267 assert(ctx);
268
269 /*
270 * In UST, since we have to write the metadata from the cache packet
271 * by packet, we might need to start this procedure multiple times
272 * until all the metadata from the cache has been extracted.
273 */
274 do {
275 /*
276 * Steps :
277 * - Lock the metadata stream
278 * - Check if metadata stream node was deleted before locking.
279 * - if yes, release and return success
280 * - Check if new metadata is ready (flush + snapshot pos)
281 * - If nothing : release and return.
282 * - Lock the metadata_rdv_lock
283 * - Unlock the metadata stream
284 * - cond_wait on metadata_rdv to wait the wakeup from the
285 * metadata thread
286 * - Unlock the metadata_rdv_lock
287 */
288 pthread_mutex_lock(&metadata->lock);
289
290 /*
291 * There is a possibility that we were able to acquire a reference on the
292 * stream from the RCU hash table but between then and now, the node might
293 * have been deleted just before the lock is acquired. Thus, after locking,
294 * we make sure the metadata node has not been deleted which means that the
295 * buffers are closed.
296 *
297 * In that case, there is no need to sync the metadata hence returning a
298 * success return code.
299 */
300 ret = cds_lfht_is_node_deleted(&metadata->node.node);
301 if (ret) {
302 ret = 0;
303 goto end_unlock_mutex;
304 }
305
306 switch (ctx->type) {
307 case LTTNG_CONSUMER_KERNEL:
308 /*
309 * Empty the metadata cache and flush the current stream.
310 */
311 status = lttng_kconsumer_sync_metadata(metadata);
312 break;
313 case LTTNG_CONSUMER32_UST:
314 case LTTNG_CONSUMER64_UST:
315 /*
316 * Ask the sessiond if we have new metadata waiting and update the
317 * consumer metadata cache.
318 */
319 status = lttng_ustconsumer_sync_metadata(ctx, metadata);
320 break;
321 default:
322 abort();
323 }
324
325 switch (status) {
326 case SYNC_METADATA_STATUS_NEW_DATA:
327 break;
328 case SYNC_METADATA_STATUS_NO_DATA:
329 ret = 0;
330 goto end_unlock_mutex;
331 case SYNC_METADATA_STATUS_ERROR:
332 ret = -1;
333 goto end_unlock_mutex;
334 default:
335 abort();
336 }
337
338 /*
339 * At this point, new metadata have been flushed, so we wait on the
340 * rendez-vous point for the metadata thread to wake us up when it
341 * finishes consuming the metadata and continue execution.
342 */
343
344 pthread_mutex_lock(&metadata->metadata_rdv_lock);
345
346 /*
347 * Release metadata stream lock so the metadata thread can process it.
348 */
349 pthread_mutex_unlock(&metadata->lock);
350
351 /*
352 * Wait on the rendez-vous point. Once woken up, it means the metadata was
353 * consumed and thus synchronization is achieved.
354 */
355 pthread_cond_wait(&metadata->metadata_rdv, &metadata->metadata_rdv_lock);
356 pthread_mutex_unlock(&metadata->metadata_rdv_lock);
357 } while (status == SYNC_METADATA_STATUS_NEW_DATA);
358
359 /* Success */
360 return 0;
361
362 end_unlock_mutex:
363 pthread_mutex_unlock(&metadata->lock);
364 return ret;
365 }
366
367 /*
368 * Synchronize the metadata using a given session ID. A successful acquisition
369 * of a metadata stream will trigger a request to the session daemon and a
370 * snapshot so the metadata thread can consume it.
371 *
372 * This function call is a rendez-vous point between the metadata thread and
373 * the data thread.
374 *
375 * Return 0 on success or else a negative value.
376 */
377 int consumer_stream_sync_metadata(struct lttng_consumer_local_data *ctx,
378 uint64_t session_id)
379 {
380 int ret;
381 struct lttng_consumer_stream *stream = NULL;
382 struct lttng_ht_iter iter;
383 struct lttng_ht *ht;
384
385 assert(ctx);
386
387 /* Ease our life a bit. */
388 ht = the_consumer_data.stream_list_ht;
389
390 rcu_read_lock();
391
392 /* Search the metadata associated with the session id of the given stream. */
393
394 cds_lfht_for_each_entry_duplicate(ht->ht,
395 ht->hash_fct(&session_id, lttng_ht_seed), ht->match_fct,
396 &session_id, &iter.iter, stream, node_session_id.node) {
397 if (!stream->metadata_flag) {
398 continue;
399 }
400
401 ret = do_sync_metadata(stream, ctx);
402 if (ret < 0) {
403 goto end;
404 }
405 }
406
407 /*
408 * Force return code to 0 (success) since ret might be ENODATA for instance
409 * which is not an error but rather that we should come back.
410 */
411 ret = 0;
412
413 end:
414 rcu_read_unlock();
415 return ret;
416 }
417
418 static int consumer_stream_sync_metadata_index(
419 struct lttng_consumer_stream *stream,
420 const struct stream_subbuffer *subbuffer,
421 struct lttng_consumer_local_data *ctx)
422 {
423 int ret;
424
425 /* Block until all the metadata is sent. */
426 pthread_mutex_lock(&stream->metadata_timer_lock);
427 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 if (stream->missed_metadata_flush) {
436 stream->missed_metadata_flush = false;
437 pthread_mutex_unlock(&stream->metadata_timer_lock);
438 (void) stream->read_subbuffer_ops.send_live_beacon(stream);
439 } else {
440 pthread_mutex_unlock(&stream->metadata_timer_lock);
441 }
442 if (ret < 0) {
443 goto end;
444 }
445
446 ret = consumer_stream_send_index(stream, subbuffer, ctx);
447 end:
448 return ret;
449 }
450
451 /*
452 * Check if the local version of the metadata stream matches with the version
453 * of the metadata stream in the kernel. If it was updated, set the reset flag
454 * on the stream.
455 */
456 static
457 int metadata_stream_check_version(struct lttng_consumer_stream *stream,
458 const struct stream_subbuffer *subbuffer)
459 {
460 if (stream->metadata_version == subbuffer->info.metadata.version) {
461 goto end;
462 }
463
464 DBG("New metadata version detected");
465 consumer_stream_metadata_set_version(stream,
466 subbuffer->info.metadata.version);
467
468 if (stream->read_subbuffer_ops.reset_metadata) {
469 stream->read_subbuffer_ops.reset_metadata(stream);
470 }
471
472 end:
473 return 0;
474 }
475
476 static
477 bool stream_is_rotating_to_null_chunk(
478 const struct lttng_consumer_stream *stream)
479 {
480 bool rotating_to_null_chunk = false;
481
482 if (stream->rotate_position == -1ULL) {
483 /* No rotation ongoing. */
484 goto end;
485 }
486
487 if (stream->trace_chunk == stream->chan->trace_chunk ||
488 !stream->chan->trace_chunk) {
489 rotating_to_null_chunk = true;
490 }
491 end:
492 return rotating_to_null_chunk;
493 }
494
495 enum consumer_stream_open_packet_status consumer_stream_open_packet(
496 struct lttng_consumer_stream *stream)
497 {
498 int ret;
499 enum consumer_stream_open_packet_status status;
500 unsigned long produced_pos_before, produced_pos_after;
501
502 ret = lttng_consumer_sample_snapshot_positions(stream);
503 if (ret < 0) {
504 ERR("Failed to snapshot positions before post-rotation empty packet flush: stream id = %" PRIu64
505 ", channel name = %s, session id = %" PRIu64,
506 stream->key, stream->chan->name,
507 stream->chan->session_id);
508 status = CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR;
509 goto end;
510 }
511
512 ret = lttng_consumer_get_produced_snapshot(
513 stream, &produced_pos_before);
514 if (ret < 0) {
515 ERR("Failed to read produced position before post-rotation empty packet flush: stream id = %" PRIu64
516 ", channel name = %s, session id = %" PRIu64,
517 stream->key, stream->chan->name,
518 stream->chan->session_id);
519 status = CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR;
520 goto end;
521 }
522
523 ret = consumer_stream_flush_buffer(stream, 0);
524 if (ret) {
525 ERR("Failed to flush an empty packet at rotation point: stream id = %" PRIu64
526 ", channel name = %s, session id = %" PRIu64,
527 stream->key, stream->chan->name,
528 stream->chan->session_id);
529 status = CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR;
530 goto end;
531 }
532
533 ret = lttng_consumer_sample_snapshot_positions(stream);
534 if (ret < 0) {
535 ERR("Failed to snapshot positions after post-rotation empty packet flush: stream id = %" PRIu64
536 ", channel name = %s, session id = %" PRIu64,
537 stream->key, stream->chan->name,
538 stream->chan->session_id);
539 status = CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR;
540 goto end;
541 }
542
543 ret = lttng_consumer_get_produced_snapshot(stream, &produced_pos_after);
544 if (ret < 0) {
545 ERR("Failed to read produced position after post-rotation empty packet flush: stream id = %" PRIu64
546 ", channel name = %s, session id = %" PRIu64,
547 stream->key, stream->chan->name,
548 stream->chan->session_id);
549 status = CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR;
550 goto end;
551 }
552
553 /*
554 * Determine if the flush had an effect by comparing the produced
555 * positons before and after the flush.
556 */
557 status = produced_pos_before != produced_pos_after ?
558 CONSUMER_STREAM_OPEN_PACKET_STATUS_OPENED :
559 CONSUMER_STREAM_OPEN_PACKET_STATUS_NO_SPACE;
560 if (status == CONSUMER_STREAM_OPEN_PACKET_STATUS_OPENED) {
561 stream->opened_packet_in_current_trace_chunk = true;
562 }
563
564 end:
565 return status;
566 }
567
568 /*
569 * An attempt to open a new packet is performed after a rotation completes to
570 * get a begin timestamp as close as possible to the rotation point.
571 *
572 * However, that initial attempt at opening a packet can fail due to a full
573 * ring-buffer. In that case, a second attempt is performed after consuming
574 * a packet since that will have freed enough space in the ring-buffer.
575 */
576 static
577 int post_consume_open_new_packet(struct lttng_consumer_stream *stream,
578 const struct stream_subbuffer *subbuffer,
579 struct lttng_consumer_local_data *ctx)
580 {
581 int ret = 0;
582
583 if (!stream->opened_packet_in_current_trace_chunk &&
584 stream->trace_chunk &&
585 !stream_is_rotating_to_null_chunk(stream)) {
586 const enum consumer_stream_open_packet_status status =
587 consumer_stream_open_packet(stream);
588
589 switch (status) {
590 case CONSUMER_STREAM_OPEN_PACKET_STATUS_OPENED:
591 DBG("Opened a packet after consuming a packet rotation: stream id = %" PRIu64
592 ", channel name = %s, session id = %" PRIu64,
593 stream->key, stream->chan->name,
594 stream->chan->session_id);
595 stream->opened_packet_in_current_trace_chunk = true;
596 break;
597 case CONSUMER_STREAM_OPEN_PACKET_STATUS_NO_SPACE:
598 /*
599 * Can't open a packet as there is no space left.
600 * This means that new events were produced, resulting
601 * in a packet being opened, which is what we want
602 * anyhow.
603 */
604 DBG("No space left to open a packet after consuming a packet: stream id = %" PRIu64
605 ", channel name = %s, session id = %" PRIu64,
606 stream->key, stream->chan->name,
607 stream->chan->session_id);
608 stream->opened_packet_in_current_trace_chunk = true;
609 break;
610 case CONSUMER_STREAM_OPEN_PACKET_STATUS_ERROR:
611 /* Logged by callee. */
612 ret = -1;
613 goto end;
614 default:
615 abort();
616 }
617
618 stream->opened_packet_in_current_trace_chunk = true;
619 }
620
621 end:
622 return ret;
623 }
624
625 struct lttng_consumer_stream *consumer_stream_create(
626 struct lttng_consumer_channel *channel,
627 uint64_t channel_key,
628 uint64_t stream_key,
629 const char *channel_name,
630 uint64_t relayd_id,
631 uint64_t session_id,
632 struct lttng_trace_chunk *trace_chunk,
633 int cpu,
634 int *alloc_ret,
635 enum consumer_channel_type type,
636 unsigned int monitor)
637 {
638 int ret;
639 struct lttng_consumer_stream *stream;
640
641 stream = zmalloc(sizeof(*stream));
642 if (stream == NULL) {
643 PERROR("malloc struct lttng_consumer_stream");
644 ret = -ENOMEM;
645 goto end;
646 }
647
648 rcu_read_lock();
649
650 if (trace_chunk && !lttng_trace_chunk_get(trace_chunk)) {
651 ERR("Failed to acquire trace chunk reference during the creation of a stream");
652 ret = -1;
653 goto error;
654 }
655
656 stream->chan = channel;
657 stream->key = stream_key;
658 stream->trace_chunk = trace_chunk;
659 stream->out_fd = -1;
660 stream->out_fd_offset = 0;
661 stream->output_written = 0;
662 stream->net_seq_idx = relayd_id;
663 stream->session_id = session_id;
664 stream->monitor = monitor;
665 stream->endpoint_status = CONSUMER_ENDPOINT_ACTIVE;
666 stream->index_file = NULL;
667 stream->last_sequence_number = -1ULL;
668 stream->rotate_position = -1ULL;
669 /* Buffer is created with an open packet. */
670 stream->opened_packet_in_current_trace_chunk = true;
671 pthread_mutex_init(&stream->lock, NULL);
672 pthread_mutex_init(&stream->metadata_timer_lock, NULL);
673
674 /* If channel is the metadata, flag this stream as metadata. */
675 if (type == CONSUMER_CHANNEL_TYPE_METADATA) {
676 stream->metadata_flag = 1;
677 /* Metadata is flat out. */
678 strncpy(stream->name, DEFAULT_METADATA_NAME, sizeof(stream->name));
679 /* Live rendez-vous point. */
680 pthread_cond_init(&stream->metadata_rdv, NULL);
681 pthread_mutex_init(&stream->metadata_rdv_lock, NULL);
682 } else {
683 /* Format stream name to <channel_name>_<cpu_number> */
684 ret = snprintf(stream->name, sizeof(stream->name), "%s_%d",
685 channel_name, cpu);
686 if (ret < 0) {
687 PERROR("snprintf stream name");
688 goto error;
689 }
690 }
691
692 switch (channel->output) {
693 case CONSUMER_CHANNEL_SPLICE:
694 stream->output = LTTNG_EVENT_SPLICE;
695 ret = utils_create_pipe(stream->splice_pipe);
696 if (ret < 0) {
697 goto error;
698 }
699 break;
700 case CONSUMER_CHANNEL_MMAP:
701 stream->output = LTTNG_EVENT_MMAP;
702 break;
703 default:
704 abort();
705 }
706
707 /* Key is always the wait_fd for streams. */
708 lttng_ht_node_init_u64(&stream->node, stream->key);
709
710 /* Init node per channel id key */
711 lttng_ht_node_init_u64(&stream->node_channel_id, channel_key);
712
713 /* Init session id node with the stream session id */
714 lttng_ht_node_init_u64(&stream->node_session_id, stream->session_id);
715
716 DBG3("Allocated stream %s (key %" PRIu64 ", chan_key %" PRIu64
717 " relayd_id %" PRIu64 ", session_id %" PRIu64,
718 stream->name, stream->key, channel_key,
719 stream->net_seq_idx, stream->session_id);
720
721 rcu_read_unlock();
722
723 lttng_dynamic_array_init(&stream->read_subbuffer_ops.post_consume_cbs,
724 sizeof(post_consume_cb), NULL);
725
726 if (type == CONSUMER_CHANNEL_TYPE_METADATA) {
727 stream->read_subbuffer_ops.lock =
728 consumer_stream_metadata_lock_all;
729 stream->read_subbuffer_ops.unlock =
730 consumer_stream_metadata_unlock_all;
731 stream->read_subbuffer_ops.assert_locked =
732 consumer_stream_metadata_assert_locked_all;
733 stream->read_subbuffer_ops.pre_consume_subbuffer =
734 metadata_stream_check_version;
735 } else {
736 const post_consume_cb post_consume_index_op = channel->is_live ?
737 consumer_stream_sync_metadata_index :
738 consumer_stream_send_index;
739
740 ret = lttng_dynamic_array_add_element(
741 &stream->read_subbuffer_ops.post_consume_cbs,
742 &post_consume_index_op);
743 if (ret) {
744 PERROR("Failed to add `send index` callback to stream's post consumption callbacks");
745 goto error;
746 }
747
748 ret = lttng_dynamic_array_add_element(
749 &stream->read_subbuffer_ops.post_consume_cbs,
750 &(post_consume_cb) { post_consume_open_new_packet });
751 if (ret) {
752 PERROR("Failed to add `open new packet` callback to stream's post consumption callbacks");
753 goto error;
754 }
755
756 stream->read_subbuffer_ops.lock = consumer_stream_data_lock_all;
757 stream->read_subbuffer_ops.unlock =
758 consumer_stream_data_unlock_all;
759 stream->read_subbuffer_ops.assert_locked =
760 consumer_stream_data_assert_locked_all;
761 stream->read_subbuffer_ops.pre_consume_subbuffer =
762 consumer_stream_update_stats;
763 }
764
765 if (channel->output == CONSUMER_CHANNEL_MMAP) {
766 stream->read_subbuffer_ops.consume_subbuffer =
767 consumer_stream_consume_mmap;
768 } else {
769 stream->read_subbuffer_ops.consume_subbuffer =
770 consumer_stream_consume_splice;
771 }
772
773 return stream;
774
775 error:
776 rcu_read_unlock();
777 lttng_trace_chunk_put(stream->trace_chunk);
778 lttng_dynamic_array_reset(&stream->read_subbuffer_ops.post_consume_cbs);
779 free(stream);
780 end:
781 if (alloc_ret) {
782 *alloc_ret = ret;
783 }
784 return NULL;
785 }
786
787 /*
788 * Close stream on the relayd side. This call can destroy a relayd if the
789 * conditions are met.
790 *
791 * A RCU read side lock MUST be acquired if the relayd object was looked up in
792 * a hash table before calling this.
793 */
794 void consumer_stream_relayd_close(struct lttng_consumer_stream *stream,
795 struct consumer_relayd_sock_pair *relayd)
796 {
797 int ret;
798
799 assert(stream);
800 assert(relayd);
801
802 if (stream->sent_to_relayd) {
803 uatomic_dec(&relayd->refcount);
804 assert(uatomic_read(&relayd->refcount) >= 0);
805 }
806
807 /* Closing streams requires to lock the control socket. */
808 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
809 ret = relayd_send_close_stream(&relayd->control_sock,
810 stream->relayd_stream_id,
811 stream->next_net_seq_num - 1);
812 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
813 if (ret < 0) {
814 ERR("Relayd send close stream failed. Cleaning up relayd %" PRIu64 ".", relayd->net_seq_idx);
815 lttng_consumer_cleanup_relayd(relayd);
816 }
817
818 /* Both conditions are met, we destroy the relayd. */
819 if (uatomic_read(&relayd->refcount) == 0 &&
820 uatomic_read(&relayd->destroy_flag)) {
821 consumer_destroy_relayd(relayd);
822 }
823 stream->net_seq_idx = (uint64_t) -1ULL;
824 stream->sent_to_relayd = 0;
825 }
826
827 /*
828 * Close stream's file descriptors and, if needed, close stream also on the
829 * relayd side.
830 *
831 * The consumer data lock MUST be acquired.
832 * The stream lock MUST be acquired.
833 */
834 void consumer_stream_close(struct lttng_consumer_stream *stream)
835 {
836 int ret;
837 struct consumer_relayd_sock_pair *relayd;
838
839 assert(stream);
840
841 switch (the_consumer_data.type) {
842 case LTTNG_CONSUMER_KERNEL:
843 if (stream->mmap_base != NULL) {
844 ret = munmap(stream->mmap_base, stream->mmap_len);
845 if (ret != 0) {
846 PERROR("munmap");
847 }
848 }
849
850 if (stream->wait_fd >= 0) {
851 ret = close(stream->wait_fd);
852 if (ret) {
853 PERROR("close");
854 }
855 stream->wait_fd = -1;
856 }
857 if (stream->chan->output == CONSUMER_CHANNEL_SPLICE) {
858 utils_close_pipe(stream->splice_pipe);
859 }
860 break;
861 case LTTNG_CONSUMER32_UST:
862 case LTTNG_CONSUMER64_UST:
863 {
864 /*
865 * Special case for the metadata since the wait fd is an internal pipe
866 * polled in the metadata thread.
867 */
868 if (stream->metadata_flag && stream->chan->monitor) {
869 int rpipe = stream->ust_metadata_poll_pipe[0];
870
871 /*
872 * This will stop the channel timer if one and close the write side
873 * of the metadata poll pipe.
874 */
875 lttng_ustconsumer_close_metadata(stream->chan);
876 if (rpipe >= 0) {
877 ret = close(rpipe);
878 if (ret < 0) {
879 PERROR("closing metadata pipe read side");
880 }
881 stream->ust_metadata_poll_pipe[0] = -1;
882 }
883 }
884 break;
885 }
886 default:
887 ERR("Unknown consumer_data type");
888 assert(0);
889 }
890
891 /* Close output fd. Could be a socket or local file at this point. */
892 if (stream->out_fd >= 0) {
893 ret = close(stream->out_fd);
894 if (ret) {
895 PERROR("close");
896 }
897 stream->out_fd = -1;
898 }
899
900 if (stream->index_file) {
901 lttng_index_file_put(stream->index_file);
902 stream->index_file = NULL;
903 }
904
905 lttng_trace_chunk_put(stream->trace_chunk);
906 stream->trace_chunk = NULL;
907
908 /* Check and cleanup relayd if needed. */
909 rcu_read_lock();
910 relayd = consumer_find_relayd(stream->net_seq_idx);
911 if (relayd != NULL) {
912 consumer_stream_relayd_close(stream, relayd);
913 }
914 rcu_read_unlock();
915 }
916
917 /*
918 * Delete the stream from all possible hash tables.
919 *
920 * The consumer data lock MUST be acquired.
921 * The stream lock MUST be acquired.
922 */
923 void consumer_stream_delete(struct lttng_consumer_stream *stream,
924 struct lttng_ht *ht)
925 {
926 int ret;
927 struct lttng_ht_iter iter;
928
929 assert(stream);
930 /* Should NEVER be called not in monitor mode. */
931 assert(stream->chan->monitor);
932
933 rcu_read_lock();
934
935 if (ht) {
936 iter.iter.node = &stream->node.node;
937 ret = lttng_ht_del(ht, &iter);
938 assert(!ret);
939 }
940
941 /* Delete from stream per channel ID hash table. */
942 iter.iter.node = &stream->node_channel_id.node;
943 /*
944 * The returned value is of no importance. Even if the node is NOT in the
945 * hash table, we continue since we may have been called by a code path
946 * that did not add the stream to a (all) hash table. Same goes for the
947 * next call ht del call.
948 */
949 (void) lttng_ht_del(the_consumer_data.stream_per_chan_id_ht, &iter);
950
951 /* Delete from the global stream list. */
952 iter.iter.node = &stream->node_session_id.node;
953 /* See the previous ht del on why we ignore the returned value. */
954 (void) lttng_ht_del(the_consumer_data.stream_list_ht, &iter);
955
956 rcu_read_unlock();
957
958 if (!stream->metadata_flag) {
959 /* Decrement the stream count of the global consumer data. */
960 assert(the_consumer_data.stream_count > 0);
961 the_consumer_data.stream_count--;
962 }
963 }
964
965 /*
966 * Free the given stream within a RCU call.
967 */
968 void consumer_stream_free(struct lttng_consumer_stream *stream)
969 {
970 assert(stream);
971
972 metadata_bucket_destroy(stream->metadata_bucket);
973 call_rcu(&stream->node.head, free_stream_rcu);
974 }
975
976 /*
977 * Destroy the stream's buffers of the tracer.
978 */
979 void consumer_stream_destroy_buffers(struct lttng_consumer_stream *stream)
980 {
981 assert(stream);
982
983 switch (the_consumer_data.type) {
984 case LTTNG_CONSUMER_KERNEL:
985 break;
986 case LTTNG_CONSUMER32_UST:
987 case LTTNG_CONSUMER64_UST:
988 lttng_ustconsumer_del_stream(stream);
989 break;
990 default:
991 ERR("Unknown consumer_data type");
992 assert(0);
993 }
994 }
995
996 /*
997 * Destroy and close a already created stream.
998 */
999 static void destroy_close_stream(struct lttng_consumer_stream *stream)
1000 {
1001 assert(stream);
1002
1003 DBG("Consumer stream destroy monitored key: %" PRIu64, stream->key);
1004
1005 /* Destroy tracer buffers of the stream. */
1006 consumer_stream_destroy_buffers(stream);
1007 /* Close down everything including the relayd if one. */
1008 consumer_stream_close(stream);
1009 }
1010
1011 /*
1012 * Decrement the stream's channel refcount and if down to 0, return the channel
1013 * pointer so it can be destroyed by the caller or NULL if not.
1014 */
1015 static struct lttng_consumer_channel *unref_channel(
1016 struct lttng_consumer_stream *stream)
1017 {
1018 struct lttng_consumer_channel *free_chan = NULL;
1019
1020 assert(stream);
1021 assert(stream->chan);
1022
1023 /* Update refcount of channel and see if we need to destroy it. */
1024 if (!uatomic_sub_return(&stream->chan->refcount, 1)
1025 && !uatomic_read(&stream->chan->nb_init_stream_left)) {
1026 free_chan = stream->chan;
1027 }
1028
1029 return free_chan;
1030 }
1031
1032 /*
1033 * Destroy a stream completely. This will delete, close and free the stream.
1034 * Once return, the stream is NO longer usable. Its channel may get destroyed
1035 * if conditions are met for a monitored stream.
1036 *
1037 * This MUST be called WITHOUT the consumer data and stream lock acquired if
1038 * the stream is in _monitor_ mode else it does not matter.
1039 */
1040 void consumer_stream_destroy(struct lttng_consumer_stream *stream,
1041 struct lttng_ht *ht)
1042 {
1043 assert(stream);
1044
1045 /* Stream is in monitor mode. */
1046 if (stream->monitor) {
1047 struct lttng_consumer_channel *free_chan = NULL;
1048
1049 /*
1050 * This means that the stream was successfully removed from the streams
1051 * list of the channel and sent to the right thread managing this
1052 * stream thus being globally visible.
1053 */
1054 if (stream->globally_visible) {
1055 pthread_mutex_lock(&the_consumer_data.lock);
1056 pthread_mutex_lock(&stream->chan->lock);
1057 pthread_mutex_lock(&stream->lock);
1058 /* Remove every reference of the stream in the consumer. */
1059 consumer_stream_delete(stream, ht);
1060
1061 destroy_close_stream(stream);
1062
1063 /* Update channel's refcount of the stream. */
1064 free_chan = unref_channel(stream);
1065
1066 /* Indicates that the consumer data state MUST be updated after this. */
1067 the_consumer_data.need_update = 1;
1068
1069 pthread_mutex_unlock(&stream->lock);
1070 pthread_mutex_unlock(&stream->chan->lock);
1071 pthread_mutex_unlock(&the_consumer_data.lock);
1072 } else {
1073 /*
1074 * If the stream is not visible globally, this needs to be done
1075 * outside of the consumer data lock section.
1076 */
1077 free_chan = unref_channel(stream);
1078 }
1079
1080 if (free_chan) {
1081 consumer_del_channel(free_chan);
1082 }
1083 } else {
1084 destroy_close_stream(stream);
1085 }
1086
1087 /* Free stream within a RCU call. */
1088 lttng_trace_chunk_put(stream->trace_chunk);
1089 stream->trace_chunk = NULL;
1090 lttng_dynamic_array_reset(&stream->read_subbuffer_ops.post_consume_cbs);
1091 consumer_stream_free(stream);
1092 }
1093
1094 /*
1095 * Write index of a specific stream either on the relayd or local disk.
1096 *
1097 * Return 0 on success or else a negative value.
1098 */
1099 int consumer_stream_write_index(struct lttng_consumer_stream *stream,
1100 struct ctf_packet_index *element)
1101 {
1102 int ret;
1103
1104 assert(stream);
1105 assert(element);
1106
1107 rcu_read_lock();
1108 if (stream->net_seq_idx != (uint64_t) -1ULL) {
1109 struct consumer_relayd_sock_pair *relayd;
1110 relayd = consumer_find_relayd(stream->net_seq_idx);
1111 if (relayd) {
1112 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
1113 ret = relayd_send_index(&relayd->control_sock, element,
1114 stream->relayd_stream_id, stream->next_net_seq_num - 1);
1115 if (ret < 0) {
1116 /*
1117 * Communication error with lttng-relayd,
1118 * perform cleanup now
1119 */
1120 ERR("Relayd send index failed. Cleaning up relayd %" PRIu64 ".", relayd->net_seq_idx);
1121 lttng_consumer_cleanup_relayd(relayd);
1122 ret = -1;
1123 }
1124 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
1125 } else {
1126 ERR("Stream %" PRIu64 " relayd ID %" PRIu64 " unknown. Can't write index.",
1127 stream->key, stream->net_seq_idx);
1128 ret = -1;
1129 }
1130 } else {
1131 if (lttng_index_file_write(stream->index_file, element)) {
1132 ret = -1;
1133 } else {
1134 ret = 0;
1135 }
1136 }
1137 if (ret < 0) {
1138 goto error;
1139 }
1140
1141 error:
1142 rcu_read_unlock();
1143 return ret;
1144 }
1145
1146 int consumer_stream_create_output_files(struct lttng_consumer_stream *stream,
1147 bool create_index)
1148 {
1149 int ret;
1150 enum lttng_trace_chunk_status chunk_status;
1151 const int flags = O_WRONLY | O_CREAT | O_TRUNC;
1152 const mode_t mode = S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP;
1153 char stream_path[LTTNG_PATH_MAX];
1154
1155 ASSERT_LOCKED(stream->lock);
1156 assert(stream->trace_chunk);
1157
1158 ret = utils_stream_file_path(stream->chan->pathname, stream->name,
1159 stream->chan->tracefile_size,
1160 stream->tracefile_count_current, NULL,
1161 stream_path, sizeof(stream_path));
1162 if (ret < 0) {
1163 goto end;
1164 }
1165
1166 if (stream->out_fd >= 0) {
1167 ret = close(stream->out_fd);
1168 if (ret < 0) {
1169 PERROR("Failed to close stream file \"%s\"",
1170 stream->name);
1171 goto end;
1172 }
1173 stream->out_fd = -1;
1174 }
1175
1176 DBG("Opening stream output file \"%s\"", stream_path);
1177 chunk_status = lttng_trace_chunk_open_file(stream->trace_chunk, stream_path,
1178 flags, mode, &stream->out_fd, false);
1179 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1180 ERR("Failed to open stream file \"%s\"", stream->name);
1181 ret = -1;
1182 goto end;
1183 }
1184
1185 if (!stream->metadata_flag && (create_index || stream->index_file)) {
1186 if (stream->index_file) {
1187 lttng_index_file_put(stream->index_file);
1188 }
1189 chunk_status = lttng_index_file_create_from_trace_chunk(
1190 stream->trace_chunk,
1191 stream->chan->pathname,
1192 stream->name,
1193 stream->chan->tracefile_size,
1194 stream->tracefile_count_current,
1195 CTF_INDEX_MAJOR, CTF_INDEX_MINOR,
1196 false, &stream->index_file);
1197 if (chunk_status != LTTNG_TRACE_CHUNK_STATUS_OK) {
1198 ret = -1;
1199 goto end;
1200 }
1201 }
1202
1203 /* Reset current size because we just perform a rotation. */
1204 stream->tracefile_size_current = 0;
1205 stream->out_fd_offset = 0;
1206 end:
1207 return ret;
1208 }
1209
1210 int consumer_stream_rotate_output_files(struct lttng_consumer_stream *stream)
1211 {
1212 int ret;
1213
1214 stream->tracefile_count_current++;
1215 if (stream->chan->tracefile_count > 0) {
1216 stream->tracefile_count_current %=
1217 stream->chan->tracefile_count;
1218 }
1219
1220 DBG("Rotating output files of stream \"%s\"", stream->name);
1221 ret = consumer_stream_create_output_files(stream, true);
1222 if (ret) {
1223 goto end;
1224 }
1225
1226 end:
1227 return ret;
1228 }
1229
1230 bool consumer_stream_is_deleted(struct lttng_consumer_stream *stream)
1231 {
1232 /*
1233 * This function does not take a const stream since
1234 * cds_lfht_is_node_deleted was not const before liburcu 0.12.
1235 */
1236 assert(stream);
1237 return cds_lfht_is_node_deleted(&stream->node.node);
1238 }
1239
1240 static ssize_t metadata_bucket_flush(
1241 const struct stream_subbuffer *buffer, void *data)
1242 {
1243 ssize_t ret;
1244 struct lttng_consumer_stream *stream = data;
1245
1246 ret = consumer_stream_consume_mmap(NULL, stream, buffer);
1247 if (ret < 0) {
1248 goto end;
1249 }
1250 end:
1251 return ret;
1252 }
1253
1254 static ssize_t metadata_bucket_consume(
1255 struct lttng_consumer_local_data *unused,
1256 struct lttng_consumer_stream *stream,
1257 const struct stream_subbuffer *subbuffer)
1258 {
1259 ssize_t ret;
1260 enum metadata_bucket_status status;
1261
1262 status = metadata_bucket_fill(stream->metadata_bucket, subbuffer);
1263 switch (status) {
1264 case METADATA_BUCKET_STATUS_OK:
1265 /* Return consumed size. */
1266 ret = subbuffer->buffer.buffer.size;
1267 break;
1268 default:
1269 ret = -1;
1270 }
1271
1272 return ret;
1273 }
1274
1275 int consumer_stream_enable_metadata_bucketization(
1276 struct lttng_consumer_stream *stream)
1277 {
1278 int ret = 0;
1279
1280 assert(stream->metadata_flag);
1281 assert(!stream->metadata_bucket);
1282 assert(stream->chan->output == CONSUMER_CHANNEL_MMAP);
1283
1284 stream->metadata_bucket = metadata_bucket_create(
1285 metadata_bucket_flush, stream);
1286 if (!stream->metadata_bucket) {
1287 ret = -1;
1288 goto end;
1289 }
1290
1291 stream->read_subbuffer_ops.consume_subbuffer = metadata_bucket_consume;
1292 end:
1293 return ret;
1294 }
1295
1296 void consumer_stream_metadata_set_version(
1297 struct lttng_consumer_stream *stream, uint64_t new_version)
1298 {
1299 assert(new_version > stream->metadata_version);
1300 stream->metadata_version = new_version;
1301 stream->reset_metadata_flag = 1;
1302
1303 if (stream->metadata_bucket) {
1304 metadata_bucket_reset(stream->metadata_bucket);
1305 }
1306 }
1307
1308 int consumer_stream_flush_buffer(struct lttng_consumer_stream *stream,
1309 bool producer_active)
1310 {
1311 int ret = 0;
1312
1313 switch (the_consumer_data.type) {
1314 case LTTNG_CONSUMER_KERNEL:
1315 if (producer_active) {
1316 ret = kernctl_buffer_flush(stream->wait_fd);
1317 if (ret < 0) {
1318 ERR("Failed to flush kernel stream");
1319 goto end;
1320 }
1321 } else {
1322 ret = kernctl_buffer_flush_empty(stream->wait_fd);
1323 if (ret < 0) {
1324 /*
1325 * Doing a buffer flush which does not take into
1326 * account empty packets. This is not perfect,
1327 * but required as a fall-back when
1328 * "flush_empty" is not implemented by
1329 * lttng-modules.
1330 */
1331 ret = kernctl_buffer_flush(stream->wait_fd);
1332 if (ret < 0) {
1333 ERR("Failed to flush kernel stream");
1334 goto end;
1335 }
1336 }
1337 }
1338 break;
1339 case LTTNG_CONSUMER32_UST:
1340 case LTTNG_CONSUMER64_UST:
1341 ret = lttng_ustconsumer_flush_buffer(stream, (int) producer_active);
1342 break;
1343 default:
1344 ERR("Unknown consumer_data type");
1345 abort();
1346 }
1347
1348 end:
1349 return ret;
1350 }
This page took 0.088112 seconds and 4 git commands to generate.