Cleanup: keep the number of pipes used by poll in a variable
[lttng-tools.git] / src / common / kernel-consumer / kernel-consumer.c
CommitLineData
3bd1e081
MD
1/*
2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
b3530820 4 * Copyright (C) 2017 - Jérémie Galarneau <jeremie.galarneau@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
MD
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
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
6c1c0768 20#define _LGPL_SOURCE
3bd1e081 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>
77c7c900 29#include <inttypes.h>
3bd1e081 30#include <unistd.h>
dbb5dfe6 31#include <sys/stat.h>
3bd1e081 32
51a9e1c7 33#include <bin/lttng-consumerd/health-consumerd.h>
990570ed 34#include <common/common.h>
10a8a223 35#include <common/kernel-ctl/kernel-ctl.h>
10a8a223 36#include <common/sessiond-comm/sessiond-comm.h>
00e2e675 37#include <common/sessiond-comm/relayd.h>
dbb5dfe6 38#include <common/compat/fcntl.h>
f263b7fd 39#include <common/compat/endian.h>
acdb9057 40#include <common/pipe.h>
00e2e675 41#include <common/relayd/relayd.h>
fe4477ee 42#include <common/utils.h>
c8fea79c 43#include <common/consumer/consumer-stream.h>
309167d2 44#include <common/index/index.h>
c8fea79c 45#include <common/consumer/consumer-timer.h>
0857097f 46
10a8a223 47#include "kernel-consumer.h"
3bd1e081
MD
48
49extern struct lttng_consumer_global_data consumer_data;
50extern int consumer_poll_timeout;
3bd1e081 51
3bd1e081
MD
52/*
53 * Take a snapshot for a specific fd
54 *
55 * Returns 0 on success, < 0 on error
56 */
ffe60014 57int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
3bd1e081
MD
58{
59 int ret = 0;
60 int infd = stream->wait_fd;
61
62 ret = kernctl_snapshot(infd);
d2d2f190
JD
63 /*
64 * -EAGAIN is not an error, it just means that there is no data to
65 * be read.
66 */
67 if (ret != 0 && ret != -EAGAIN) {
5a510c9f 68 PERROR("Getting sub-buffer snapshot.");
3bd1e081
MD
69 }
70
71 return ret;
72}
73
e9404c27
JG
74/*
75 * Sample consumed and produced positions for a specific fd.
76 *
77 * Returns 0 on success, < 0 on error.
78 */
79int lttng_kconsumer_sample_snapshot_positions(
80 struct lttng_consumer_stream *stream)
81{
82 assert(stream);
83
84 return kernctl_snapshot_sample_positions(stream->wait_fd);
85}
86
3bd1e081
MD
87/*
88 * Get the produced position
89 *
90 * Returns 0 on success, < 0 on error
91 */
ffe60014 92int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
3bd1e081
MD
93 unsigned long *pos)
94{
95 int ret;
96 int infd = stream->wait_fd;
97
98 ret = kernctl_snapshot_get_produced(infd, pos);
99 if (ret != 0) {
5a510c9f 100 PERROR("kernctl_snapshot_get_produced");
3bd1e081
MD
101 }
102
103 return ret;
104}
105
07b86b52
JD
106/*
107 * Get the consumerd position
108 *
109 * Returns 0 on success, < 0 on error
110 */
111int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
112 unsigned long *pos)
113{
114 int ret;
115 int infd = stream->wait_fd;
116
117 ret = kernctl_snapshot_get_consumed(infd, pos);
118 if (ret != 0) {
5a510c9f 119 PERROR("kernctl_snapshot_get_consumed");
07b86b52
JD
120 }
121
122 return ret;
123}
124
07b86b52
JD
125/*
126 * Take a snapshot of all the stream of a channel
127 *
128 * Returns 0 on success, < 0 on error
129 */
130int lttng_kconsumer_snapshot_channel(uint64_t key, char *path,
d07ceecd 131 uint64_t relayd_id, uint64_t nb_packets_per_stream,
5c786ded 132 struct lttng_consumer_local_data *ctx)
07b86b52
JD
133{
134 int ret;
07b86b52
JD
135 struct lttng_consumer_channel *channel;
136 struct lttng_consumer_stream *stream;
137
6a00837f 138 DBG("Kernel consumer snapshot channel %" PRIu64, key);
07b86b52
JD
139
140 rcu_read_lock();
141
142 channel = consumer_find_channel(key);
143 if (!channel) {
6a00837f 144 ERR("No channel found for key %" PRIu64, key);
07b86b52
JD
145 ret = -1;
146 goto end;
147 }
148
149 /* Splice is not supported yet for channel snapshot. */
150 if (channel->output != CONSUMER_CHANNEL_MMAP) {
151 ERR("Unsupported output %d", channel->output);
152 ret = -1;
153 goto end;
154 }
155
10a50311 156 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
923333cd 157 unsigned long consumed_pos, produced_pos;
9ce5646a
MD
158
159 health_code_update();
160
07b86b52
JD
161 /*
162 * Lock stream because we are about to change its state.
163 */
164 pthread_mutex_lock(&stream->lock);
165
29decac3
DG
166 /*
167 * Assign the received relayd ID so we can use it for streaming. The streams
168 * are not visible to anyone so this is OK to change it.
169 */
07b86b52
JD
170 stream->net_seq_idx = relayd_id;
171 channel->relayd_id = relayd_id;
172 if (relayd_id != (uint64_t) -1ULL) {
10a50311 173 ret = consumer_send_relayd_stream(stream, path);
07b86b52
JD
174 if (ret < 0) {
175 ERR("sending stream to relayd");
176 goto end_unlock;
177 }
07b86b52
JD
178 } else {
179 ret = utils_create_stream_file(path, stream->name,
10a50311
JD
180 stream->chan->tracefile_size,
181 stream->tracefile_count_current,
309167d2 182 stream->uid, stream->gid, NULL);
07b86b52
JD
183 if (ret < 0) {
184 ERR("utils_create_stream_file");
185 goto end_unlock;
186 }
187
188 stream->out_fd = ret;
189 stream->tracefile_size_current = 0;
190
81ea21bf
MD
191 DBG("Kernel consumer snapshot stream %s/%s (%" PRIu64 ")",
192 path, stream->name, stream->key);
07b86b52
JD
193 }
194
f22dd891 195 ret = kernctl_buffer_flush_empty(stream->wait_fd);
07b86b52 196 if (ret < 0) {
f22dd891
MD
197 /*
198 * Doing a buffer flush which does not take into
199 * account empty packets. This is not perfect
200 * for stream intersection, but required as a
201 * fall-back when "flush_empty" is not
202 * implemented by lttng-modules.
203 */
204 ret = kernctl_buffer_flush(stream->wait_fd);
205 if (ret < 0) {
206 ERR("Failed to flush kernel stream");
207 goto end_unlock;
208 }
07b86b52
JD
209 goto end_unlock;
210 }
211
212 ret = lttng_kconsumer_take_snapshot(stream);
213 if (ret < 0) {
214 ERR("Taking kernel snapshot");
215 goto end_unlock;
216 }
217
218 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
219 if (ret < 0) {
220 ERR("Produced kernel snapshot position");
221 goto end_unlock;
222 }
223
224 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
225 if (ret < 0) {
226 ERR("Consumerd kernel snapshot position");
227 goto end_unlock;
228 }
229
230 if (stream->max_sb_size == 0) {
231 ret = kernctl_get_max_subbuf_size(stream->wait_fd,
232 &stream->max_sb_size);
233 if (ret < 0) {
234 ERR("Getting kernel max_sb_size");
235 goto end_unlock;
236 }
237 }
238
d07ceecd
MD
239 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
240 produced_pos, nb_packets_per_stream,
241 stream->max_sb_size);
5c786ded 242
07b86b52
JD
243 while (consumed_pos < produced_pos) {
244 ssize_t read_len;
245 unsigned long len, padded_len;
246
9ce5646a
MD
247 health_code_update();
248
07b86b52
JD
249 DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
250
251 ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
252 if (ret < 0) {
32af2c95 253 if (ret != -EAGAIN) {
07b86b52
JD
254 PERROR("kernctl_get_subbuf snapshot");
255 goto end_unlock;
256 }
257 DBG("Kernel consumer get subbuf failed. Skipping it.");
258 consumed_pos += stream->max_sb_size;
ddc93ee4 259 stream->chan->lost_packets++;
07b86b52
JD
260 continue;
261 }
262
263 ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
264 if (ret < 0) {
265 ERR("Snapshot kernctl_get_subbuf_size");
29decac3 266 goto error_put_subbuf;
07b86b52
JD
267 }
268
269 ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
270 if (ret < 0) {
271 ERR("Snapshot kernctl_get_padded_subbuf_size");
29decac3 272 goto error_put_subbuf;
07b86b52
JD
273 }
274
275 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
309167d2 276 padded_len - len, NULL);
07b86b52 277 /*
29decac3
DG
278 * We write the padded len in local tracefiles but the data len
279 * when using a relay. Display the error but continue processing
280 * to try to release the subbuffer.
07b86b52
JD
281 */
282 if (relayd_id != (uint64_t) -1ULL) {
283 if (read_len != len) {
284 ERR("Error sending to the relay (ret: %zd != len: %lu)",
285 read_len, len);
286 }
287 } else {
288 if (read_len != padded_len) {
289 ERR("Error writing to tracefile (ret: %zd != len: %lu)",
290 read_len, padded_len);
291 }
292 }
293
294 ret = kernctl_put_subbuf(stream->wait_fd);
295 if (ret < 0) {
296 ERR("Snapshot kernctl_put_subbuf");
297 goto end_unlock;
298 }
299 consumed_pos += stream->max_sb_size;
300 }
301
302 if (relayd_id == (uint64_t) -1ULL) {
fdf9986c
MD
303 if (stream->out_fd >= 0) {
304 ret = close(stream->out_fd);
305 if (ret < 0) {
306 PERROR("Kernel consumer snapshot close out_fd");
307 goto end_unlock;
308 }
309 stream->out_fd = -1;
07b86b52 310 }
07b86b52
JD
311 } else {
312 close_relayd_stream(stream);
313 stream->net_seq_idx = (uint64_t) -1ULL;
314 }
315 pthread_mutex_unlock(&stream->lock);
316 }
317
318 /* All good! */
319 ret = 0;
320 goto end;
321
29decac3
DG
322error_put_subbuf:
323 ret = kernctl_put_subbuf(stream->wait_fd);
324 if (ret < 0) {
325 ERR("Snapshot kernctl_put_subbuf error path");
326 }
07b86b52
JD
327end_unlock:
328 pthread_mutex_unlock(&stream->lock);
329end:
330 rcu_read_unlock();
331 return ret;
332}
333
334/*
335 * Read the whole metadata available for a snapshot.
336 *
337 * Returns 0 on success, < 0 on error
338 */
339int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
e2039c7a 340 uint64_t relayd_id, struct lttng_consumer_local_data *ctx)
07b86b52 341{
d771f832
DG
342 int ret, use_relayd = 0;
343 ssize_t ret_read;
07b86b52
JD
344 struct lttng_consumer_channel *metadata_channel;
345 struct lttng_consumer_stream *metadata_stream;
d771f832
DG
346
347 assert(ctx);
07b86b52
JD
348
349 DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
350 key, path);
351
352 rcu_read_lock();
353
354 metadata_channel = consumer_find_channel(key);
355 if (!metadata_channel) {
d771f832 356 ERR("Kernel snapshot metadata not found for key %" PRIu64, key);
07b86b52 357 ret = -1;
d771f832 358 goto error;
07b86b52
JD
359 }
360
361 metadata_stream = metadata_channel->metadata_stream;
362 assert(metadata_stream);
363
d771f832 364 /* Flag once that we have a valid relayd for the stream. */
e2039c7a 365 if (relayd_id != (uint64_t) -1ULL) {
d771f832
DG
366 use_relayd = 1;
367 }
368
369 if (use_relayd) {
10a50311 370 ret = consumer_send_relayd_stream(metadata_stream, path);
e2039c7a 371 if (ret < 0) {
d771f832 372 goto error;
e2039c7a 373 }
e2039c7a
JD
374 } else {
375 ret = utils_create_stream_file(path, metadata_stream->name,
376 metadata_stream->chan->tracefile_size,
377 metadata_stream->tracefile_count_current,
309167d2 378 metadata_stream->uid, metadata_stream->gid, NULL);
e2039c7a 379 if (ret < 0) {
d771f832 380 goto error;
e2039c7a
JD
381 }
382 metadata_stream->out_fd = ret;
07b86b52 383 }
07b86b52 384
d771f832 385 do {
9ce5646a
MD
386 health_code_update();
387
d771f832
DG
388 ret_read = lttng_kconsumer_read_subbuffer(metadata_stream, ctx);
389 if (ret_read < 0) {
56591bac 390 if (ret_read != -EAGAIN) {
6a00837f 391 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
d771f832
DG
392 ret_read);
393 goto error;
07b86b52 394 }
d771f832 395 /* ret_read is negative at this point so we will exit the loop. */
07b86b52
JD
396 continue;
397 }
d771f832 398 } while (ret_read >= 0);
07b86b52 399
d771f832
DG
400 if (use_relayd) {
401 close_relayd_stream(metadata_stream);
402 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
403 } else {
fdf9986c
MD
404 if (metadata_stream->out_fd >= 0) {
405 ret = close(metadata_stream->out_fd);
406 if (ret < 0) {
407 PERROR("Kernel consumer snapshot metadata close out_fd");
408 /*
409 * Don't go on error here since the snapshot was successful at this
410 * point but somehow the close failed.
411 */
412 }
413 metadata_stream->out_fd = -1;
e2039c7a 414 }
e2039c7a
JD
415 }
416
07b86b52 417 ret = 0;
d771f832 418
cf53a8a6
JD
419 cds_list_del(&metadata_stream->send_node);
420 consumer_stream_destroy(metadata_stream, NULL);
421 metadata_channel->metadata_stream = NULL;
d771f832 422error:
07b86b52
JD
423 rcu_read_unlock();
424 return ret;
425}
426
1803a064
MD
427/*
428 * Receive command from session daemon and process it.
429 *
430 * Return 1 on success else a negative value or 0.
431 */
3bd1e081
MD
432int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
433 int sock, struct pollfd *consumer_sockpoll)
434{
435 ssize_t ret;
0c759fc9 436 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
3bd1e081
MD
437 struct lttcomm_consumer_msg msg;
438
9ce5646a
MD
439 health_code_update();
440
3bd1e081
MD
441 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
442 if (ret != sizeof(msg)) {
1803a064 443 if (ret > 0) {
c6857fcf 444 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1803a064
MD
445 ret = -1;
446 }
3bd1e081
MD
447 return ret;
448 }
9ce5646a
MD
449
450 health_code_update();
451
84382d49
MD
452 /* Deprecated command */
453 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
3bd1e081 454
9ce5646a
MD
455 health_code_update();
456
b0b335c8
MD
457 /* relayd needs RCU read-side protection */
458 rcu_read_lock();
459
3bd1e081 460 switch (msg.cmd_type) {
00e2e675
DG
461 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
462 {
f50f23d9 463 /* Session daemon status message are handled in the following call. */
2527bf85 464 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
7735ef9e 465 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
d3e2ba59 466 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
2527bf85 467 msg.u.relayd_sock.relayd_session_id);
00e2e675
DG
468 goto end_nosignal;
469 }
3bd1e081
MD
470 case LTTNG_CONSUMER_ADD_CHANNEL:
471 {
472 struct lttng_consumer_channel *new_channel;
e43c41c5 473 int ret_recv;
3bd1e081 474
9ce5646a
MD
475 health_code_update();
476
f50f23d9
DG
477 /* First send a status message before receiving the fds. */
478 ret = consumer_send_status_msg(sock, ret_code);
479 if (ret < 0) {
480 /* Somehow, the session daemon is not responding anymore. */
1803a064 481 goto error_fatal;
f50f23d9 482 }
9ce5646a
MD
483
484 health_code_update();
485
d88aee68 486 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
3bd1e081 487 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
ffe60014
DG
488 msg.u.channel.session_id, msg.u.channel.pathname,
489 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
1624d5b7
JD
490 msg.u.channel.relayd_id, msg.u.channel.output,
491 msg.u.channel.tracefile_size,
1950109e 492 msg.u.channel.tracefile_count, 0,
ecc48a90 493 msg.u.channel.monitor,
d7ba1388 494 msg.u.channel.live_timer_interval,
3d071855 495 NULL, NULL);
3bd1e081 496 if (new_channel == NULL) {
f73fabfd 497 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
3bd1e081
MD
498 goto end_nosignal;
499 }
ffe60014 500 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
95a1109b
JD
501 switch (msg.u.channel.output) {
502 case LTTNG_EVENT_SPLICE:
503 new_channel->output = CONSUMER_CHANNEL_SPLICE;
504 break;
505 case LTTNG_EVENT_MMAP:
506 new_channel->output = CONSUMER_CHANNEL_MMAP;
507 break;
508 default:
509 ERR("Channel output unknown %d", msg.u.channel.output);
510 goto end_nosignal;
511 }
ffe60014
DG
512
513 /* Translate and save channel type. */
514 switch (msg.u.channel.type) {
515 case CONSUMER_CHANNEL_TYPE_DATA:
516 case CONSUMER_CHANNEL_TYPE_METADATA:
517 new_channel->type = msg.u.channel.type;
518 break;
519 default:
520 assert(0);
521 goto end_nosignal;
522 };
523
9ce5646a
MD
524 health_code_update();
525
3bd1e081 526 if (ctx->on_recv_channel != NULL) {
e43c41c5
JD
527 ret_recv = ctx->on_recv_channel(new_channel);
528 if (ret_recv == 0) {
529 ret = consumer_add_channel(new_channel, ctx);
530 } else if (ret_recv < 0) {
3bd1e081
MD
531 goto end_nosignal;
532 }
533 } else {
e43c41c5 534 ret = consumer_add_channel(new_channel, ctx);
3bd1e081 535 }
e9404c27
JG
536 if (msg.u.channel.type == CONSUMER_CHANNEL_TYPE_DATA && !ret) {
537 int monitor_start_ret;
538
539 DBG("Consumer starting monitor timer");
94d49140
JD
540 consumer_timer_live_start(new_channel,
541 msg.u.channel.live_timer_interval);
e9404c27
JG
542 monitor_start_ret = consumer_timer_monitor_start(
543 new_channel,
544 msg.u.channel.monitor_timer_interval);
545 if (monitor_start_ret < 0) {
546 ERR("Starting channel monitoring timer failed");
547 goto end_nosignal;
548 }
549
94d49140 550 }
e43c41c5 551
9ce5646a
MD
552 health_code_update();
553
e43c41c5 554 /* If we received an error in add_channel, we need to report it. */
821fffb2 555 if (ret < 0) {
1803a064
MD
556 ret = consumer_send_status_msg(sock, ret);
557 if (ret < 0) {
558 goto error_fatal;
559 }
e43c41c5
JD
560 goto end_nosignal;
561 }
562
3bd1e081
MD
563 goto end_nosignal;
564 }
565 case LTTNG_CONSUMER_ADD_STREAM:
566 {
dae10966
DG
567 int fd;
568 struct lttng_pipe *stream_pipe;
00e2e675 569 struct lttng_consumer_stream *new_stream;
ffe60014 570 struct lttng_consumer_channel *channel;
c80048c6 571 int alloc_ret = 0;
3bd1e081 572
ffe60014
DG
573 /*
574 * Get stream's channel reference. Needed when adding the stream to the
575 * global hash table.
576 */
577 channel = consumer_find_channel(msg.u.stream.channel_key);
578 if (!channel) {
579 /*
580 * We could not find the channel. Can happen if cpu hotplug
581 * happens while tearing down.
582 */
d88aee68 583 ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
e462382a 584 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
ffe60014
DG
585 }
586
9ce5646a
MD
587 health_code_update();
588
f50f23d9
DG
589 /* First send a status message before receiving the fds. */
590 ret = consumer_send_status_msg(sock, ret_code);
1803a064 591 if (ret < 0) {
d771f832 592 /* Somehow, the session daemon is not responding anymore. */
1803a064
MD
593 goto error_fatal;
594 }
9ce5646a
MD
595
596 health_code_update();
597
0c759fc9 598 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
d771f832 599 /* Channel was not found. */
f50f23d9
DG
600 goto end_nosignal;
601 }
602
d771f832 603 /* Blocking call */
9ce5646a
MD
604 health_poll_entry();
605 ret = lttng_consumer_poll_socket(consumer_sockpoll);
606 health_poll_exit();
84382d49
MD
607 if (ret) {
608 goto error_fatal;
3bd1e081 609 }
00e2e675 610
9ce5646a
MD
611 health_code_update();
612
00e2e675 613 /* Get stream file descriptor from socket */
f2fc6720
MD
614 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
615 if (ret != sizeof(fd)) {
f73fabfd 616 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
3f8e211f 617 rcu_read_unlock();
3bd1e081
MD
618 return ret;
619 }
3bd1e081 620
9ce5646a
MD
621 health_code_update();
622
f50f23d9
DG
623 /*
624 * Send status code to session daemon only if the recv works. If the
625 * above recv() failed, the session daemon is notified through the
626 * error socket and the teardown is eventually done.
627 */
628 ret = consumer_send_status_msg(sock, ret_code);
629 if (ret < 0) {
630 /* Somehow, the session daemon is not responding anymore. */
631 goto end_nosignal;
632 }
633
9ce5646a
MD
634 health_code_update();
635
ffe60014
DG
636 new_stream = consumer_allocate_stream(channel->key,
637 fd,
638 LTTNG_CONSUMER_ACTIVE_STREAM,
639 channel->name,
640 channel->uid,
641 channel->gid,
642 channel->relayd_id,
643 channel->session_id,
644 msg.u.stream.cpu,
645 &alloc_ret,
4891ece8
DG
646 channel->type,
647 channel->monitor);
3bd1e081 648 if (new_stream == NULL) {
c80048c6
MD
649 switch (alloc_ret) {
650 case -ENOMEM:
651 case -EINVAL:
652 default:
653 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
654 break;
c80048c6 655 }
3f8e211f 656 goto end_nosignal;
3bd1e081 657 }
d771f832 658
ffe60014
DG
659 new_stream->chan = channel;
660 new_stream->wait_fd = fd;
07b86b52
JD
661 switch (channel->output) {
662 case CONSUMER_CHANNEL_SPLICE:
663 new_stream->output = LTTNG_EVENT_SPLICE;
a2361a61
JD
664 ret = utils_create_pipe(new_stream->splice_pipe);
665 if (ret < 0) {
666 goto end_nosignal;
667 }
07b86b52
JD
668 break;
669 case CONSUMER_CHANNEL_MMAP:
670 new_stream->output = LTTNG_EVENT_MMAP;
671 break;
672 default:
673 ERR("Stream output unknown %d", channel->output);
674 goto end_nosignal;
675 }
00e2e675 676
a0c83db9
DG
677 /*
678 * We've just assigned the channel to the stream so increment the
07b86b52
JD
679 * refcount right now. We don't need to increment the refcount for
680 * streams in no monitor because we handle manually the cleanup of
681 * those. It is very important to make sure there is NO prior
682 * consumer_del_stream() calls or else the refcount will be unbalanced.
a0c83db9 683 */
07b86b52
JD
684 if (channel->monitor) {
685 uatomic_inc(&new_stream->chan->refcount);
686 }
9d9353f9 687
fb3a43a9
DG
688 /*
689 * The buffer flush is done on the session daemon side for the kernel
690 * so no need for the stream "hangup_flush_done" variable to be
691 * tracked. This is important for a kernel stream since we don't rely
692 * on the flush state of the stream to read data. It's not the case for
693 * user space tracing.
694 */
695 new_stream->hangup_flush_done = 0;
696
9ce5646a
MD
697 health_code_update();
698
633d0084
DG
699 if (ctx->on_recv_stream) {
700 ret = ctx->on_recv_stream(new_stream);
701 if (ret < 0) {
d771f832 702 consumer_stream_free(new_stream);
633d0084 703 goto end_nosignal;
fb3a43a9 704 }
633d0084 705 }
fb3a43a9 706
9ce5646a
MD
707 health_code_update();
708
07b86b52
JD
709 if (new_stream->metadata_flag) {
710 channel->metadata_stream = new_stream;
711 }
712
2bba9e53
DG
713 /* Do not monitor this stream. */
714 if (!channel->monitor) {
5eecee74 715 DBG("Kernel consumer add stream %s in no monitor mode with "
6dc3064a 716 "relayd id %" PRIu64, new_stream->name,
5eecee74 717 new_stream->net_seq_idx);
10a50311 718 cds_list_add(&new_stream->send_node, &channel->streams.head);
6dc3064a
DG
719 break;
720 }
721
e1b71bdc
DG
722 /* Send stream to relayd if the stream has an ID. */
723 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
194ee077
DG
724 ret = consumer_send_relayd_stream(new_stream,
725 new_stream->chan->pathname);
e1b71bdc
DG
726 if (ret < 0) {
727 consumer_stream_free(new_stream);
728 goto end_nosignal;
729 }
001b7e62
MD
730
731 /*
732 * If adding an extra stream to an already
733 * existing channel (e.g. cpu hotplug), we need
734 * to send the "streams_sent" command to relayd.
735 */
736 if (channel->streams_sent_to_relayd) {
737 ret = consumer_send_relayd_streams_sent(
738 new_stream->net_seq_idx);
739 if (ret < 0) {
740 goto end_nosignal;
741 }
742 }
e2039c7a
JD
743 }
744
50f8ae69 745 /* Get the right pipe where the stream will be sent. */
633d0084 746 if (new_stream->metadata_flag) {
66d583dc 747 consumer_add_metadata_stream(new_stream);
dae10966 748 stream_pipe = ctx->consumer_metadata_pipe;
3bd1e081 749 } else {
66d583dc 750 consumer_add_data_stream(new_stream);
dae10966 751 stream_pipe = ctx->consumer_data_pipe;
50f8ae69
DG
752 }
753
66d583dc 754 /* Visible to other threads */
5ab66908
MD
755 new_stream->globally_visible = 1;
756
9ce5646a
MD
757 health_code_update();
758
dae10966 759 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
50f8ae69 760 if (ret < 0) {
dae10966 761 ERR("Consumer write %s stream to pipe %d",
50f8ae69 762 new_stream->metadata_flag ? "metadata" : "data",
dae10966 763 lttng_pipe_get_writefd(stream_pipe));
5ab66908
MD
764 if (new_stream->metadata_flag) {
765 consumer_del_stream_for_metadata(new_stream);
766 } else {
767 consumer_del_stream_for_data(new_stream);
768 }
50f8ae69 769 goto end_nosignal;
3bd1e081 770 }
00e2e675 771
50f8ae69 772 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
ffe60014 773 new_stream->name, fd, new_stream->relayd_stream_id);
3bd1e081
MD
774 break;
775 }
a4baae1b
JD
776 case LTTNG_CONSUMER_STREAMS_SENT:
777 {
778 struct lttng_consumer_channel *channel;
779
780 /*
781 * Get stream's channel reference. Needed when adding the stream to the
782 * global hash table.
783 */
784 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
785 if (!channel) {
786 /*
787 * We could not find the channel. Can happen if cpu hotplug
788 * happens while tearing down.
789 */
790 ERR("Unable to find channel key %" PRIu64,
791 msg.u.sent_streams.channel_key);
e462382a 792 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
a4baae1b
JD
793 }
794
795 health_code_update();
796
797 /*
798 * Send status code to session daemon.
799 */
800 ret = consumer_send_status_msg(sock, ret_code);
f261ad0a 801 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
a4baae1b
JD
802 /* Somehow, the session daemon is not responding anymore. */
803 goto end_nosignal;
804 }
805
806 health_code_update();
807
808 /*
809 * We should not send this message if we don't monitor the
810 * streams in this channel.
811 */
812 if (!channel->monitor) {
813 break;
814 }
815
816 health_code_update();
817 /* Send stream to relayd if the stream has an ID. */
818 if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
819 ret = consumer_send_relayd_streams_sent(
820 msg.u.sent_streams.net_seq_idx);
821 if (ret < 0) {
822 goto end_nosignal;
823 }
001b7e62 824 channel->streams_sent_to_relayd = true;
a4baae1b
JD
825 }
826 break;
827 }
3bd1e081
MD
828 case LTTNG_CONSUMER_UPDATE_STREAM:
829 {
3f8e211f
DG
830 rcu_read_unlock();
831 return -ENOSYS;
832 }
833 case LTTNG_CONSUMER_DESTROY_RELAYD:
834 {
a6ba4fe1 835 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
3f8e211f
DG
836 struct consumer_relayd_sock_pair *relayd;
837
a6ba4fe1 838 DBG("Kernel consumer destroying relayd %" PRIu64, index);
3f8e211f
DG
839
840 /* Get relayd reference if exists. */
a6ba4fe1 841 relayd = consumer_find_relayd(index);
3f8e211f 842 if (relayd == NULL) {
3448e266 843 DBG("Unable to find relayd %" PRIu64, index);
e462382a 844 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
3bd1e081 845 }
3f8e211f 846
a6ba4fe1
DG
847 /*
848 * Each relayd socket pair has a refcount of stream attached to it
849 * which tells if the relayd is still active or not depending on the
850 * refcount value.
851 *
852 * This will set the destroy flag of the relayd object and destroy it
853 * if the refcount reaches zero when called.
854 *
855 * The destroy can happen either here or when a stream fd hangs up.
856 */
f50f23d9
DG
857 if (relayd) {
858 consumer_flag_relayd_for_destroy(relayd);
859 }
860
9ce5646a
MD
861 health_code_update();
862
f50f23d9
DG
863 ret = consumer_send_status_msg(sock, ret_code);
864 if (ret < 0) {
865 /* Somehow, the session daemon is not responding anymore. */
1803a064 866 goto error_fatal;
f50f23d9 867 }
3f8e211f 868
3f8e211f 869 goto end_nosignal;
3bd1e081 870 }
6d805429 871 case LTTNG_CONSUMER_DATA_PENDING:
53632229 872 {
c8f59ee5 873 int32_t ret;
6d805429 874 uint64_t id = msg.u.data_pending.session_id;
c8f59ee5 875
6d805429 876 DBG("Kernel consumer data pending command for id %" PRIu64, id);
c8f59ee5 877
6d805429 878 ret = consumer_data_pending(id);
c8f59ee5 879
9ce5646a
MD
880 health_code_update();
881
c8f59ee5
DG
882 /* Send back returned value to session daemon */
883 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
884 if (ret < 0) {
6d805429 885 PERROR("send data pending ret code");
1803a064 886 goto error_fatal;
c8f59ee5 887 }
f50f23d9
DG
888
889 /*
890 * No need to send back a status message since the data pending
891 * returned value is the response.
892 */
c8f59ee5 893 break;
53632229 894 }
6dc3064a
DG
895 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
896 {
07b86b52
JD
897 if (msg.u.snapshot_channel.metadata == 1) {
898 ret = lttng_kconsumer_snapshot_metadata(msg.u.snapshot_channel.key,
e2039c7a
JD
899 msg.u.snapshot_channel.pathname,
900 msg.u.snapshot_channel.relayd_id, ctx);
07b86b52
JD
901 if (ret < 0) {
902 ERR("Snapshot metadata failed");
e462382a 903 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
07b86b52
JD
904 }
905 } else {
906 ret = lttng_kconsumer_snapshot_channel(msg.u.snapshot_channel.key,
907 msg.u.snapshot_channel.pathname,
5c786ded 908 msg.u.snapshot_channel.relayd_id,
d07ceecd 909 msg.u.snapshot_channel.nb_packets_per_stream,
5c786ded 910 ctx);
07b86b52
JD
911 if (ret < 0) {
912 ERR("Snapshot channel failed");
e462382a 913 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
07b86b52
JD
914 }
915 }
916
9ce5646a
MD
917 health_code_update();
918
6dc3064a
DG
919 ret = consumer_send_status_msg(sock, ret_code);
920 if (ret < 0) {
921 /* Somehow, the session daemon is not responding anymore. */
922 goto end_nosignal;
923 }
924 break;
925 }
07b86b52
JD
926 case LTTNG_CONSUMER_DESTROY_CHANNEL:
927 {
928 uint64_t key = msg.u.destroy_channel.key;
929 struct lttng_consumer_channel *channel;
930
931 channel = consumer_find_channel(key);
932 if (!channel) {
933 ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
e462382a 934 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
07b86b52
JD
935 }
936
9ce5646a
MD
937 health_code_update();
938
07b86b52
JD
939 ret = consumer_send_status_msg(sock, ret_code);
940 if (ret < 0) {
941 /* Somehow, the session daemon is not responding anymore. */
942 goto end_nosignal;
943 }
944
9ce5646a
MD
945 health_code_update();
946
15dc512a
DG
947 /* Stop right now if no channel was found. */
948 if (!channel) {
949 goto end_nosignal;
950 }
951
07b86b52
JD
952 /*
953 * This command should ONLY be issued for channel with streams set in
954 * no monitor mode.
955 */
956 assert(!channel->monitor);
957
958 /*
959 * The refcount should ALWAYS be 0 in the case of a channel in no
960 * monitor mode.
961 */
962 assert(!uatomic_sub_return(&channel->refcount, 1));
963
964 consumer_del_channel(channel);
965
966 goto end_nosignal;
967 }
fb83fe64
JD
968 case LTTNG_CONSUMER_DISCARDED_EVENTS:
969 {
66ab32be
JD
970 ssize_t ret;
971 uint64_t count;
fb83fe64
JD
972 struct lttng_consumer_channel *channel;
973 uint64_t id = msg.u.discarded_events.session_id;
974 uint64_t key = msg.u.discarded_events.channel_key;
975
e5742757
MD
976 DBG("Kernel consumer discarded events command for session id %"
977 PRIu64 ", channel key %" PRIu64, id, key);
978
fb83fe64
JD
979 channel = consumer_find_channel(key);
980 if (!channel) {
981 ERR("Kernel consumer discarded events channel %"
982 PRIu64 " not found", key);
66ab32be 983 count = 0;
e5742757 984 } else {
66ab32be 985 count = channel->discarded_events;
fb83fe64
JD
986 }
987
fb83fe64
JD
988 health_code_update();
989
990 /* Send back returned value to session daemon */
66ab32be 991 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
fb83fe64
JD
992 if (ret < 0) {
993 PERROR("send discarded events");
994 goto error_fatal;
995 }
996
997 break;
998 }
999 case LTTNG_CONSUMER_LOST_PACKETS:
1000 {
66ab32be
JD
1001 ssize_t ret;
1002 uint64_t count;
fb83fe64
JD
1003 struct lttng_consumer_channel *channel;
1004 uint64_t id = msg.u.lost_packets.session_id;
1005 uint64_t key = msg.u.lost_packets.channel_key;
1006
e5742757
MD
1007 DBG("Kernel consumer lost packets command for session id %"
1008 PRIu64 ", channel key %" PRIu64, id, key);
1009
fb83fe64
JD
1010 channel = consumer_find_channel(key);
1011 if (!channel) {
1012 ERR("Kernel consumer lost packets channel %"
1013 PRIu64 " not found", key);
66ab32be 1014 count = 0;
e5742757 1015 } else {
66ab32be 1016 count = channel->lost_packets;
fb83fe64
JD
1017 }
1018
fb83fe64
JD
1019 health_code_update();
1020
1021 /* Send back returned value to session daemon */
66ab32be 1022 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
fb83fe64
JD
1023 if (ret < 0) {
1024 PERROR("send lost packets");
1025 goto error_fatal;
1026 }
1027
1028 break;
1029 }
b3530820
JG
1030 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1031 {
1032 int channel_monitor_pipe;
1033
1034 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1035 /* Successfully received the command's type. */
1036 ret = consumer_send_status_msg(sock, ret_code);
1037 if (ret < 0) {
1038 goto error_fatal;
1039 }
1040
1041 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
1042 1);
1043 if (ret != sizeof(channel_monitor_pipe)) {
1044 ERR("Failed to receive channel monitor pipe");
1045 goto error_fatal;
1046 }
1047
1048 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1049 ret = consumer_timer_thread_set_channel_monitor_pipe(
1050 channel_monitor_pipe);
1051 if (!ret) {
1052 int flags;
1053
1054 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1055 /* Set the pipe as non-blocking. */
1056 ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
1057 if (ret == -1) {
1058 PERROR("fcntl get flags of the channel monitoring pipe");
1059 goto error_fatal;
1060 }
1061 flags = ret;
1062
1063 ret = fcntl(channel_monitor_pipe, F_SETFL,
1064 flags | O_NONBLOCK);
1065 if (ret == -1) {
1066 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1067 goto error_fatal;
1068 }
1069 DBG("Channel monitor pipe set as non-blocking");
1070 } else {
1071 ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1072 }
1073 ret = consumer_send_status_msg(sock, ret_code);
1074 if (ret < 0) {
1075 goto error_fatal;
1076 }
1077 break;
1078 }
00fb02ac
JD
1079 case LTTNG_CONSUMER_ROTATE_RENAME:
1080 {
1081 DBG("Consumer rename session %" PRIu64 " after rotation, old path = \"%s\", new path = \"%s\"",
1082 msg.u.rotate_rename.session_id,
1083 msg.u.rotate_rename.old_path,
1084 msg.u.rotate_rename.new_path);
1085 ret = lttng_consumer_rotate_rename(msg.u.rotate_rename.old_path,
1086 msg.u.rotate_rename.new_path,
1087 msg.u.rotate_rename.uid,
1088 msg.u.rotate_rename.gid,
1089 msg.u.rotate_rename.relayd_id);
1090 if (ret < 0) {
1091 ERR("Rotate rename failed");
1092 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1093 }
1094
1095 health_code_update();
1096
1097 ret = consumer_send_status_msg(sock, ret_code);
1098 if (ret < 0) {
1099 /* Somehow, the session daemon is not responding anymore. */
1100 goto end_nosignal;
1101 }
1102 break;
1103 }
a1ae2ea5
JD
1104 case LTTNG_CONSUMER_MKDIR:
1105 {
1106 DBG("Consumer mkdir %s in session %" PRIu64,
1107 msg.u.mkdir.path,
1108 msg.u.mkdir.session_id);
1109 ret = lttng_consumer_mkdir(msg.u.mkdir.path,
1110 msg.u.mkdir.uid,
1111 msg.u.mkdir.gid,
1112 msg.u.mkdir.relayd_id);
1113 if (ret < 0) {
1114 ERR("consumer mkdir failed");
1115 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1116 }
1117
1118 health_code_update();
1119
1120 ret = consumer_send_status_msg(sock, ret_code);
1121 if (ret < 0) {
1122 /* Somehow, the session daemon is not responding anymore. */
1123 goto end_nosignal;
1124 }
1125 break;
1126 }
3bd1e081 1127 default:
3f8e211f 1128 goto end_nosignal;
3bd1e081 1129 }
3f8e211f 1130
3bd1e081 1131end_nosignal:
b0b335c8 1132 rcu_read_unlock();
4cbc1a04
DG
1133
1134 /*
1135 * Return 1 to indicate success since the 0 value can be a socket
1136 * shutdown during the recv() or send() call.
1137 */
9ce5646a 1138 health_code_update();
4cbc1a04 1139 return 1;
1803a064
MD
1140
1141error_fatal:
1142 rcu_read_unlock();
1143 /* This will issue a consumer stop. */
1144 return -1;
3bd1e081 1145}
d41f73b7 1146
309167d2
JD
1147/*
1148 * Populate index values of a kernel stream. Values are set in big endian order.
1149 *
1150 * Return 0 on success or else a negative value.
1151 */
50adc264 1152static int get_index_values(struct ctf_packet_index *index, int infd)
309167d2
JD
1153{
1154 int ret;
1155
1156 ret = kernctl_get_timestamp_begin(infd, &index->timestamp_begin);
1157 if (ret < 0) {
1158 PERROR("kernctl_get_timestamp_begin");
1159 goto error;
1160 }
1161 index->timestamp_begin = htobe64(index->timestamp_begin);
1162
1163 ret = kernctl_get_timestamp_end(infd, &index->timestamp_end);
1164 if (ret < 0) {
1165 PERROR("kernctl_get_timestamp_end");
1166 goto error;
1167 }
1168 index->timestamp_end = htobe64(index->timestamp_end);
1169
1170 ret = kernctl_get_events_discarded(infd, &index->events_discarded);
1171 if (ret < 0) {
1172 PERROR("kernctl_get_events_discarded");
1173 goto error;
1174 }
1175 index->events_discarded = htobe64(index->events_discarded);
1176
1177 ret = kernctl_get_content_size(infd, &index->content_size);
1178 if (ret < 0) {
1179 PERROR("kernctl_get_content_size");
1180 goto error;
1181 }
1182 index->content_size = htobe64(index->content_size);
1183
1184 ret = kernctl_get_packet_size(infd, &index->packet_size);
1185 if (ret < 0) {
1186 PERROR("kernctl_get_packet_size");
1187 goto error;
1188 }
1189 index->packet_size = htobe64(index->packet_size);
1190
1191 ret = kernctl_get_stream_id(infd, &index->stream_id);
1192 if (ret < 0) {
1193 PERROR("kernctl_get_stream_id");
1194 goto error;
1195 }
1196 index->stream_id = htobe64(index->stream_id);
1197
234cd636
JD
1198 ret = kernctl_get_instance_id(infd, &index->stream_instance_id);
1199 if (ret < 0) {
f0b03c22
MD
1200 if (ret == -ENOTTY) {
1201 /* Command not implemented by lttng-modules. */
1202 index->stream_instance_id = -1ULL;
f0b03c22
MD
1203 } else {
1204 PERROR("kernctl_get_instance_id");
1205 goto error;
1206 }
234cd636
JD
1207 }
1208 index->stream_instance_id = htobe64(index->stream_instance_id);
1209
1210 ret = kernctl_get_sequence_number(infd, &index->packet_seq_num);
1211 if (ret < 0) {
f0b03c22
MD
1212 if (ret == -ENOTTY) {
1213 /* Command not implemented by lttng-modules. */
1214 index->packet_seq_num = -1ULL;
1215 ret = 0;
1216 } else {
1217 PERROR("kernctl_get_sequence_number");
1218 goto error;
1219 }
234cd636
JD
1220 }
1221 index->packet_seq_num = htobe64(index->packet_seq_num);
1222
309167d2
JD
1223error:
1224 return ret;
1225}
94d49140
JD
1226/*
1227 * Sync metadata meaning request them to the session daemon and snapshot to the
1228 * metadata thread can consumer them.
1229 *
1230 * Metadata stream lock MUST be acquired.
1231 *
1232 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1233 * is empty or a negative value on error.
1234 */
1235int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata)
1236{
1237 int ret;
1238
1239 assert(metadata);
1240
1241 ret = kernctl_buffer_flush(metadata->wait_fd);
1242 if (ret < 0) {
1243 ERR("Failed to flush kernel stream");
1244 goto end;
1245 }
1246
1247 ret = kernctl_snapshot(metadata->wait_fd);
1248 if (ret < 0) {
32af2c95 1249 if (ret != -EAGAIN) {
94d49140
JD
1250 ERR("Sync metadata, taking kernel snapshot failed.");
1251 goto end;
1252 }
1253 DBG("Sync metadata, no new kernel metadata");
1254 /* No new metadata, exit. */
1255 ret = ENODATA;
1256 goto end;
1257 }
1258
1259end:
1260 return ret;
1261}
309167d2 1262
fb83fe64
JD
1263static
1264int update_stream_stats(struct lttng_consumer_stream *stream)
1265{
1266 int ret;
1267 uint64_t seq, discarded;
1268
1269 ret = kernctl_get_sequence_number(stream->wait_fd, &seq);
1270 if (ret < 0) {
f0b03c22
MD
1271 if (ret == -ENOTTY) {
1272 /* Command not implemented by lttng-modules. */
1273 seq = -1ULL;
f0b03c22
MD
1274 } else {
1275 PERROR("kernctl_get_sequence_number");
1276 goto end;
1277 }
fb83fe64
JD
1278 }
1279
1280 /*
1281 * Start the sequence when we extract the first packet in case we don't
1282 * start at 0 (for example if a consumer is not connected to the
1283 * session immediately after the beginning).
1284 */
1285 if (stream->last_sequence_number == -1ULL) {
1286 stream->last_sequence_number = seq;
1287 } else if (seq > stream->last_sequence_number) {
1288 stream->chan->lost_packets += seq -
1289 stream->last_sequence_number - 1;
1290 } else {
1291 /* seq <= last_sequence_number */
1292 ERR("Sequence number inconsistent : prev = %" PRIu64
1293 ", current = %" PRIu64,
1294 stream->last_sequence_number, seq);
1295 ret = -1;
1296 goto end;
1297 }
1298 stream->last_sequence_number = seq;
1299
1300 ret = kernctl_get_events_discarded(stream->wait_fd, &discarded);
1301 if (ret < 0) {
1302 PERROR("kernctl_get_events_discarded");
1303 goto end;
1304 }
1305 if (discarded < stream->last_discarded_events) {
1306 /*
83f4233d
MJ
1307 * Overflow has occurred. We assume only one wrap-around
1308 * has occurred.
fb83fe64
JD
1309 */
1310 stream->chan->discarded_events += (1ULL << (CAA_BITS_PER_LONG - 1)) -
1311 stream->last_discarded_events + discarded;
1312 } else {
1313 stream->chan->discarded_events += discarded -
1314 stream->last_discarded_events;
1315 }
1316 stream->last_discarded_events = discarded;
1317 ret = 0;
1318
1319end:
1320 return ret;
1321}
1322
93ec662e
JD
1323/*
1324 * Check if the local version of the metadata stream matches with the version
1325 * of the metadata stream in the kernel. If it was updated, set the reset flag
1326 * on the stream.
1327 */
1328static
1329int metadata_stream_check_version(int infd, struct lttng_consumer_stream *stream)
1330{
1331 int ret;
1332 uint64_t cur_version;
1333
1334 ret = kernctl_get_metadata_version(infd, &cur_version);
1335 if (ret < 0) {
f0b03c22
MD
1336 if (ret == -ENOTTY) {
1337 /*
1338 * LTTng-modules does not implement this
1339 * command.
1340 */
1341 ret = 0;
1342 goto end;
1343 }
93ec662e
JD
1344 ERR("Failed to get the metadata version");
1345 goto end;
1346 }
1347
1348 if (stream->metadata_version == cur_version) {
1349 ret = 0;
1350 goto end;
1351 }
1352
1353 DBG("New metadata version detected");
1354 stream->metadata_version = cur_version;
1355 stream->reset_metadata_flag = 1;
1356 ret = 0;
1357
1358end:
1359 return ret;
1360}
1361
d41f73b7
MD
1362/*
1363 * Consume data on a file descriptor and write it on a trace file.
1364 */
4078b776 1365ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
d41f73b7
MD
1366 struct lttng_consumer_local_data *ctx)
1367{
1d4dfdef 1368 unsigned long len, subbuf_size, padding;
1c20f0e2 1369 int err, write_index = 1;
4078b776 1370 ssize_t ret = 0;
d41f73b7 1371 int infd = stream->wait_fd;
50adc264 1372 struct ctf_packet_index index;
d41f73b7
MD
1373
1374 DBG("In read_subbuffer (infd : %d)", infd);
309167d2 1375
d41f73b7
MD
1376 /* Get the next subbuffer */
1377 err = kernctl_get_next_subbuf(infd);
1378 if (err != 0) {
d41f73b7
MD
1379 /*
1380 * This is a debug message even for single-threaded consumer,
1381 * because poll() have more relaxed criterions than get subbuf,
1382 * so get_subbuf may fail for short race windows where poll()
1383 * would issue wakeups.
1384 */
1385 DBG("Reserving sub buffer failed (everything is normal, "
1386 "it is due to concurrency)");
32af2c95 1387 ret = err;
d41f73b7
MD
1388 goto end;
1389 }
1390
1d4dfdef
DG
1391 /* Get the full subbuffer size including padding */
1392 err = kernctl_get_padded_subbuf_size(infd, &len);
1393 if (err != 0) {
5a510c9f 1394 PERROR("Getting sub-buffer len failed.");
8265f19e
MD
1395 err = kernctl_put_subbuf(infd);
1396 if (err != 0) {
32af2c95 1397 if (err == -EFAULT) {
5a510c9f 1398 PERROR("Error in unreserving sub buffer\n");
32af2c95 1399 } else if (err == -EIO) {
8265f19e 1400 /* Should never happen with newer LTTng versions */
5a510c9f 1401 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
8265f19e 1402 }
32af2c95 1403 ret = err;
8265f19e
MD
1404 goto end;
1405 }
32af2c95 1406 ret = err;
1d4dfdef
DG
1407 goto end;
1408 }
1409
1c20f0e2 1410 if (!stream->metadata_flag) {
309167d2
JD
1411 ret = get_index_values(&index, infd);
1412 if (ret < 0) {
8265f19e
MD
1413 err = kernctl_put_subbuf(infd);
1414 if (err != 0) {
32af2c95 1415 if (err == -EFAULT) {
5a510c9f 1416 PERROR("Error in unreserving sub buffer\n");
32af2c95 1417 } else if (err == -EIO) {
8265f19e 1418 /* Should never happen with newer LTTng versions */
5a510c9f 1419 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
8265f19e 1420 }
32af2c95 1421 ret = err;
8265f19e
MD
1422 goto end;
1423 }
309167d2
JD
1424 goto end;
1425 }
fb83fe64
JD
1426 ret = update_stream_stats(stream);
1427 if (ret < 0) {
7b87473d
MD
1428 err = kernctl_put_subbuf(infd);
1429 if (err != 0) {
1430 if (err == -EFAULT) {
1431 PERROR("Error in unreserving sub buffer\n");
1432 } else if (err == -EIO) {
1433 /* Should never happen with newer LTTng versions */
1434 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1435 }
1436 ret = err;
1437 goto end;
1438 }
fb83fe64
JD
1439 goto end;
1440 }
1c20f0e2
JD
1441 } else {
1442 write_index = 0;
93ec662e
JD
1443 ret = metadata_stream_check_version(infd, stream);
1444 if (ret < 0) {
7b87473d
MD
1445 err = kernctl_put_subbuf(infd);
1446 if (err != 0) {
1447 if (err == -EFAULT) {
1448 PERROR("Error in unreserving sub buffer\n");
1449 } else if (err == -EIO) {
1450 /* Should never happen with newer LTTng versions */
1451 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1452 }
1453 ret = err;
1454 goto end;
1455 }
93ec662e
JD
1456 goto end;
1457 }
309167d2
JD
1458 }
1459
ffe60014 1460 switch (stream->chan->output) {
07b86b52 1461 case CONSUMER_CHANNEL_SPLICE:
1d4dfdef
DG
1462 /*
1463 * XXX: The lttng-modules splice "actor" does not handle copying
1464 * partial pages hence only using the subbuffer size without the
1465 * padding makes the splice fail.
1466 */
1467 subbuf_size = len;
1468 padding = 0;
1469
1470 /* splice the subbuffer to the tracefile */
91dfef6e 1471 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
309167d2 1472 padding, &index);
91dfef6e
DG
1473 /*
1474 * XXX: Splice does not support network streaming so the return value
1475 * is simply checked against subbuf_size and not like the mmap() op.
1476 */
1d4dfdef
DG
1477 if (ret != subbuf_size) {
1478 /*
1479 * display the error but continue processing to try
1480 * to release the subbuffer
1481 */
1482 ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
1483 ret, subbuf_size);
309167d2 1484 write_index = 0;
1d4dfdef
DG
1485 }
1486 break;
07b86b52 1487 case CONSUMER_CHANNEL_MMAP:
1d4dfdef
DG
1488 /* Get subbuffer size without padding */
1489 err = kernctl_get_subbuf_size(infd, &subbuf_size);
1490 if (err != 0) {
5a510c9f 1491 PERROR("Getting sub-buffer len failed.");
8265f19e
MD
1492 err = kernctl_put_subbuf(infd);
1493 if (err != 0) {
32af2c95 1494 if (err == -EFAULT) {
5a510c9f 1495 PERROR("Error in unreserving sub buffer\n");
32af2c95 1496 } else if (err == -EIO) {
8265f19e 1497 /* Should never happen with newer LTTng versions */
5a510c9f 1498 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
8265f19e 1499 }
32af2c95 1500 ret = err;
8265f19e
MD
1501 goto end;
1502 }
32af2c95 1503 ret = err;
1d4dfdef
DG
1504 goto end;
1505 }
47e81c02 1506
1d4dfdef
DG
1507 /* Make sure the tracer is not gone mad on us! */
1508 assert(len >= subbuf_size);
1509
1510 padding = len - subbuf_size;
1511
1512 /* write the subbuffer to the tracefile */
91dfef6e 1513 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
309167d2 1514 padding, &index);
91dfef6e
DG
1515 /*
1516 * The mmap operation should write subbuf_size amount of data when
1517 * network streaming or the full padding (len) size when we are _not_
1518 * streaming.
1519 */
d88aee68
DG
1520 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
1521 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1d4dfdef 1522 /*
91dfef6e 1523 * Display the error but continue processing to try to release the
2336629e
DG
1524 * subbuffer. This is a DBG statement since this is possible to
1525 * happen without being a critical error.
1d4dfdef 1526 */
2336629e 1527 DBG("Error writing to tracefile "
91dfef6e
DG
1528 "(ret: %zd != len: %lu != subbuf_size: %lu)",
1529 ret, len, subbuf_size);
309167d2 1530 write_index = 0;
1d4dfdef
DG
1531 }
1532 break;
1533 default:
1534 ERR("Unknown output method");
56591bac 1535 ret = -EPERM;
d41f73b7
MD
1536 }
1537
1538 err = kernctl_put_next_subbuf(infd);
1539 if (err != 0) {
32af2c95 1540 if (err == -EFAULT) {
5a510c9f 1541 PERROR("Error in unreserving sub buffer\n");
32af2c95 1542 } else if (err == -EIO) {
d41f73b7 1543 /* Should never happen with newer LTTng versions */
5a510c9f 1544 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
d41f73b7 1545 }
32af2c95 1546 ret = err;
d41f73b7
MD
1547 goto end;
1548 }
1549
309167d2 1550 /* Write index if needed. */
1c20f0e2
JD
1551 if (!write_index) {
1552 goto end;
1553 }
1554
94d49140
JD
1555 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
1556 /*
1557 * In live, block until all the metadata is sent.
1558 */
c585821b
MD
1559 pthread_mutex_lock(&stream->metadata_timer_lock);
1560 assert(!stream->missed_metadata_flush);
1561 stream->waiting_on_metadata = true;
1562 pthread_mutex_unlock(&stream->metadata_timer_lock);
1563
94d49140 1564 err = consumer_stream_sync_metadata(ctx, stream->session_id);
c585821b
MD
1565
1566 pthread_mutex_lock(&stream->metadata_timer_lock);
1567 stream->waiting_on_metadata = false;
1568 if (stream->missed_metadata_flush) {
1569 stream->missed_metadata_flush = false;
1570 pthread_mutex_unlock(&stream->metadata_timer_lock);
1571 (void) consumer_flush_kernel_index(stream);
1572 } else {
1573 pthread_mutex_unlock(&stream->metadata_timer_lock);
1574 }
94d49140
JD
1575 if (err < 0) {
1576 goto end;
1577 }
1578 }
1579
1c20f0e2
JD
1580 err = consumer_stream_write_index(stream, &index);
1581 if (err < 0) {
1582 goto end;
309167d2
JD
1583 }
1584
d41f73b7
MD
1585end:
1586 return ret;
1587}
1588
1589int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1590{
1591 int ret;
ffe60014
DG
1592
1593 assert(stream);
1594
2bba9e53
DG
1595 /*
1596 * Don't create anything if this is set for streaming or should not be
1597 * monitored.
1598 */
1599 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
fe4477ee
JD
1600 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1601 stream->chan->tracefile_size, stream->tracefile_count_current,
309167d2 1602 stream->uid, stream->gid, NULL);
fe4477ee
JD
1603 if (ret < 0) {
1604 goto error;
1605 }
1606 stream->out_fd = ret;
1607 stream->tracefile_size_current = 0;
309167d2
JD
1608
1609 if (!stream->metadata_flag) {
f8f3885c
MD
1610 struct lttng_index_file *index_file;
1611
1612 index_file = lttng_index_file_create(stream->chan->pathname,
309167d2
JD
1613 stream->name, stream->uid, stream->gid,
1614 stream->chan->tracefile_size,
f8f3885c
MD
1615 stream->tracefile_count_current,
1616 CTF_INDEX_MAJOR, CTF_INDEX_MINOR);
1617 if (!index_file) {
309167d2
JD
1618 goto error;
1619 }
1b47ae58 1620 assert(!stream->index_file);
f8f3885c 1621 stream->index_file = index_file;
309167d2 1622 }
ffe60014 1623 }
d41f73b7 1624
d41f73b7
MD
1625 if (stream->output == LTTNG_EVENT_MMAP) {
1626 /* get the len of the mmap region */
1627 unsigned long mmap_len;
1628
1629 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1630 if (ret != 0) {
ffe60014 1631 PERROR("kernctl_get_mmap_len");
d41f73b7
MD
1632 goto error_close_fd;
1633 }
1634 stream->mmap_len = (size_t) mmap_len;
1635
ffe60014
DG
1636 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1637 MAP_PRIVATE, stream->wait_fd, 0);
d41f73b7 1638 if (stream->mmap_base == MAP_FAILED) {
ffe60014 1639 PERROR("Error mmaping");
d41f73b7
MD
1640 ret = -1;
1641 goto error_close_fd;
1642 }
1643 }
1644
1645 /* we return 0 to let the library handle the FD internally */
1646 return 0;
1647
1648error_close_fd:
2f225ce2 1649 if (stream->out_fd >= 0) {
d41f73b7
MD
1650 int err;
1651
1652 err = close(stream->out_fd);
1653 assert(!err);
2f225ce2 1654 stream->out_fd = -1;
d41f73b7
MD
1655 }
1656error:
1657 return ret;
1658}
1659
ca22feea
DG
1660/*
1661 * Check if data is still being extracted from the buffers for a specific
4e9a4686
DG
1662 * stream. Consumer data lock MUST be acquired before calling this function
1663 * and the stream lock.
ca22feea 1664 *
6d805429 1665 * Return 1 if the traced data are still getting read else 0 meaning that the
ca22feea
DG
1666 * data is available for trace viewer reading.
1667 */
6d805429 1668int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
ca22feea
DG
1669{
1670 int ret;
1671
1672 assert(stream);
1673
873b9e9a
MD
1674 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1675 ret = 0;
1676 goto end;
1677 }
1678
ca22feea
DG
1679 ret = kernctl_get_next_subbuf(stream->wait_fd);
1680 if (ret == 0) {
1681 /* There is still data so let's put back this subbuffer. */
1682 ret = kernctl_put_subbuf(stream->wait_fd);
1683 assert(ret == 0);
6d805429 1684 ret = 1; /* Data is pending */
4e9a4686 1685 goto end;
ca22feea
DG
1686 }
1687
6d805429
DG
1688 /* Data is NOT pending and ready to be read. */
1689 ret = 0;
ca22feea 1690
6efae65e
DG
1691end:
1692 return ret;
ca22feea 1693}
This page took 0.140627 seconds and 4 git commands to generate.