fix
[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);
6e0be6cc
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;
6af2ba3b 661 consumer_stream_copy_ro_channel_values(new_stream, channel);
07b86b52
JD
662 switch (channel->output) {
663 case CONSUMER_CHANNEL_SPLICE:
664 new_stream->output = LTTNG_EVENT_SPLICE;
a2361a61
JD
665 ret = utils_create_pipe(new_stream->splice_pipe);
666 if (ret < 0) {
667 goto end_nosignal;
668 }
07b86b52
JD
669 break;
670 case CONSUMER_CHANNEL_MMAP:
671 new_stream->output = LTTNG_EVENT_MMAP;
672 break;
673 default:
674 ERR("Stream output unknown %d", channel->output);
675 goto end_nosignal;
676 }
00e2e675 677
a0c83db9
DG
678 /*
679 * We've just assigned the channel to the stream so increment the
07b86b52
JD
680 * refcount right now. We don't need to increment the refcount for
681 * streams in no monitor because we handle manually the cleanup of
682 * those. It is very important to make sure there is NO prior
683 * consumer_del_stream() calls or else the refcount will be unbalanced.
a0c83db9 684 */
07b86b52
JD
685 if (channel->monitor) {
686 uatomic_inc(&new_stream->chan->refcount);
687 }
9d9353f9 688
fb3a43a9
DG
689 /*
690 * The buffer flush is done on the session daemon side for the kernel
691 * so no need for the stream "hangup_flush_done" variable to be
692 * tracked. This is important for a kernel stream since we don't rely
693 * on the flush state of the stream to read data. It's not the case for
694 * user space tracing.
695 */
696 new_stream->hangup_flush_done = 0;
697
9ce5646a
MD
698 health_code_update();
699
633d0084
DG
700 if (ctx->on_recv_stream) {
701 ret = ctx->on_recv_stream(new_stream);
702 if (ret < 0) {
d771f832 703 consumer_stream_free(new_stream);
633d0084 704 goto end_nosignal;
fb3a43a9 705 }
633d0084 706 }
fb3a43a9 707
9ce5646a
MD
708 health_code_update();
709
07b86b52
JD
710 if (new_stream->metadata_flag) {
711 channel->metadata_stream = new_stream;
712 }
713
2bba9e53
DG
714 /* Do not monitor this stream. */
715 if (!channel->monitor) {
5eecee74 716 DBG("Kernel consumer add stream %s in no monitor mode with "
6dc3064a 717 "relayd id %" PRIu64, new_stream->name,
5eecee74 718 new_stream->net_seq_idx);
10a50311 719 cds_list_add(&new_stream->send_node, &channel->streams.head);
6dc3064a
DG
720 break;
721 }
722
e1b71bdc
DG
723 /* Send stream to relayd if the stream has an ID. */
724 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
194ee077
DG
725 ret = consumer_send_relayd_stream(new_stream,
726 new_stream->chan->pathname);
e1b71bdc
DG
727 if (ret < 0) {
728 consumer_stream_free(new_stream);
729 goto end_nosignal;
730 }
001b7e62
MD
731
732 /*
733 * If adding an extra stream to an already
734 * existing channel (e.g. cpu hotplug), we need
735 * to send the "streams_sent" command to relayd.
736 */
737 if (channel->streams_sent_to_relayd) {
738 ret = consumer_send_relayd_streams_sent(
739 new_stream->net_seq_idx);
740 if (ret < 0) {
741 goto end_nosignal;
742 }
743 }
e2039c7a
JD
744 }
745
50f8ae69 746 /* Get the right pipe where the stream will be sent. */
633d0084 747 if (new_stream->metadata_flag) {
5ab66908
MD
748 ret = consumer_add_metadata_stream(new_stream);
749 if (ret) {
750 ERR("Consumer add metadata stream %" PRIu64 " failed. Continuing",
751 new_stream->key);
752 consumer_stream_free(new_stream);
753 goto end_nosignal;
754 }
dae10966 755 stream_pipe = ctx->consumer_metadata_pipe;
3bd1e081 756 } else {
5ab66908
MD
757 ret = consumer_add_data_stream(new_stream);
758 if (ret) {
759 ERR("Consumer add stream %" PRIu64 " failed. Continuing",
760 new_stream->key);
761 consumer_stream_free(new_stream);
762 goto end_nosignal;
763 }
dae10966 764 stream_pipe = ctx->consumer_data_pipe;
50f8ae69
DG
765 }
766
5ab66908
MD
767 /* Vitible to other threads */
768 new_stream->globally_visible = 1;
769
9ce5646a
MD
770 health_code_update();
771
dae10966 772 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
50f8ae69 773 if (ret < 0) {
dae10966 774 ERR("Consumer write %s stream to pipe %d",
50f8ae69 775 new_stream->metadata_flag ? "metadata" : "data",
dae10966 776 lttng_pipe_get_writefd(stream_pipe));
5ab66908
MD
777 if (new_stream->metadata_flag) {
778 consumer_del_stream_for_metadata(new_stream);
779 } else {
780 consumer_del_stream_for_data(new_stream);
781 }
50f8ae69 782 goto end_nosignal;
3bd1e081 783 }
00e2e675 784
50f8ae69 785 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
ffe60014 786 new_stream->name, fd, new_stream->relayd_stream_id);
3bd1e081
MD
787 break;
788 }
a4baae1b
JD
789 case LTTNG_CONSUMER_STREAMS_SENT:
790 {
791 struct lttng_consumer_channel *channel;
792
793 /*
794 * Get stream's channel reference. Needed when adding the stream to the
795 * global hash table.
796 */
797 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
798 if (!channel) {
799 /*
800 * We could not find the channel. Can happen if cpu hotplug
801 * happens while tearing down.
802 */
803 ERR("Unable to find channel key %" PRIu64,
804 msg.u.sent_streams.channel_key);
e462382a 805 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
a4baae1b
JD
806 }
807
808 health_code_update();
809
810 /*
811 * Send status code to session daemon.
812 */
813 ret = consumer_send_status_msg(sock, ret_code);
f261ad0a 814 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
a4baae1b
JD
815 /* Somehow, the session daemon is not responding anymore. */
816 goto end_nosignal;
817 }
818
819 health_code_update();
820
821 /*
822 * We should not send this message if we don't monitor the
823 * streams in this channel.
824 */
825 if (!channel->monitor) {
826 break;
827 }
828
829 health_code_update();
830 /* Send stream to relayd if the stream has an ID. */
831 if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
832 ret = consumer_send_relayd_streams_sent(
833 msg.u.sent_streams.net_seq_idx);
834 if (ret < 0) {
835 goto end_nosignal;
836 }
001b7e62 837 channel->streams_sent_to_relayd = true;
a4baae1b
JD
838 }
839 break;
840 }
3bd1e081
MD
841 case LTTNG_CONSUMER_UPDATE_STREAM:
842 {
3f8e211f
DG
843 rcu_read_unlock();
844 return -ENOSYS;
845 }
846 case LTTNG_CONSUMER_DESTROY_RELAYD:
847 {
a6ba4fe1 848 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
3f8e211f
DG
849 struct consumer_relayd_sock_pair *relayd;
850
a6ba4fe1 851 DBG("Kernel consumer destroying relayd %" PRIu64, index);
3f8e211f
DG
852
853 /* Get relayd reference if exists. */
a6ba4fe1 854 relayd = consumer_find_relayd(index);
3f8e211f 855 if (relayd == NULL) {
3448e266 856 DBG("Unable to find relayd %" PRIu64, index);
e462382a 857 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
3bd1e081 858 }
3f8e211f 859
a6ba4fe1
DG
860 /*
861 * Each relayd socket pair has a refcount of stream attached to it
862 * which tells if the relayd is still active or not depending on the
863 * refcount value.
864 *
865 * This will set the destroy flag of the relayd object and destroy it
866 * if the refcount reaches zero when called.
867 *
868 * The destroy can happen either here or when a stream fd hangs up.
869 */
f50f23d9
DG
870 if (relayd) {
871 consumer_flag_relayd_for_destroy(relayd);
872 }
873
9ce5646a
MD
874 health_code_update();
875
f50f23d9
DG
876 ret = consumer_send_status_msg(sock, ret_code);
877 if (ret < 0) {
878 /* Somehow, the session daemon is not responding anymore. */
1803a064 879 goto error_fatal;
f50f23d9 880 }
3f8e211f 881
3f8e211f 882 goto end_nosignal;
3bd1e081 883 }
6d805429 884 case LTTNG_CONSUMER_DATA_PENDING:
53632229 885 {
c8f59ee5 886 int32_t ret;
6d805429 887 uint64_t id = msg.u.data_pending.session_id;
c8f59ee5 888
6d805429 889 DBG("Kernel consumer data pending command for id %" PRIu64, id);
c8f59ee5 890
6d805429 891 ret = consumer_data_pending(id);
c8f59ee5 892
9ce5646a
MD
893 health_code_update();
894
c8f59ee5
DG
895 /* Send back returned value to session daemon */
896 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
897 if (ret < 0) {
6d805429 898 PERROR("send data pending ret code");
1803a064 899 goto error_fatal;
c8f59ee5 900 }
f50f23d9
DG
901
902 /*
903 * No need to send back a status message since the data pending
904 * returned value is the response.
905 */
c8f59ee5 906 break;
53632229 907 }
6dc3064a
DG
908 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
909 {
07b86b52
JD
910 if (msg.u.snapshot_channel.metadata == 1) {
911 ret = lttng_kconsumer_snapshot_metadata(msg.u.snapshot_channel.key,
e2039c7a
JD
912 msg.u.snapshot_channel.pathname,
913 msg.u.snapshot_channel.relayd_id, ctx);
07b86b52
JD
914 if (ret < 0) {
915 ERR("Snapshot metadata failed");
e462382a 916 ret_code = LTTCOMM_CONSUMERD_ERROR_METADATA;
07b86b52
JD
917 }
918 } else {
919 ret = lttng_kconsumer_snapshot_channel(msg.u.snapshot_channel.key,
920 msg.u.snapshot_channel.pathname,
5c786ded 921 msg.u.snapshot_channel.relayd_id,
d07ceecd 922 msg.u.snapshot_channel.nb_packets_per_stream,
5c786ded 923 ctx);
07b86b52
JD
924 if (ret < 0) {
925 ERR("Snapshot channel failed");
e462382a 926 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
07b86b52
JD
927 }
928 }
929
9ce5646a
MD
930 health_code_update();
931
6dc3064a
DG
932 ret = consumer_send_status_msg(sock, ret_code);
933 if (ret < 0) {
934 /* Somehow, the session daemon is not responding anymore. */
935 goto end_nosignal;
936 }
937 break;
938 }
07b86b52
JD
939 case LTTNG_CONSUMER_DESTROY_CHANNEL:
940 {
941 uint64_t key = msg.u.destroy_channel.key;
942 struct lttng_consumer_channel *channel;
943
944 channel = consumer_find_channel(key);
945 if (!channel) {
946 ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
e462382a 947 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
07b86b52
JD
948 }
949
9ce5646a
MD
950 health_code_update();
951
07b86b52
JD
952 ret = consumer_send_status_msg(sock, ret_code);
953 if (ret < 0) {
954 /* Somehow, the session daemon is not responding anymore. */
955 goto end_nosignal;
956 }
957
9ce5646a
MD
958 health_code_update();
959
15dc512a
DG
960 /* Stop right now if no channel was found. */
961 if (!channel) {
962 goto end_nosignal;
963 }
964
07b86b52
JD
965 /*
966 * This command should ONLY be issued for channel with streams set in
967 * no monitor mode.
968 */
969 assert(!channel->monitor);
970
971 /*
972 * The refcount should ALWAYS be 0 in the case of a channel in no
973 * monitor mode.
974 */
975 assert(!uatomic_sub_return(&channel->refcount, 1));
976
977 consumer_del_channel(channel);
978
979 goto end_nosignal;
980 }
fb83fe64
JD
981 case LTTNG_CONSUMER_DISCARDED_EVENTS:
982 {
52042398
JD
983 ssize_t ret;
984 uint64_t count;
fb83fe64
JD
985 struct lttng_consumer_channel *channel;
986 uint64_t id = msg.u.discarded_events.session_id;
987 uint64_t key = msg.u.discarded_events.channel_key;
988
e5742757
MD
989 DBG("Kernel consumer discarded events command for session id %"
990 PRIu64 ", channel key %" PRIu64, id, key);
991
fb83fe64
JD
992 channel = consumer_find_channel(key);
993 if (!channel) {
994 ERR("Kernel consumer discarded events channel %"
995 PRIu64 " not found", key);
52042398 996 count = 0;
e5742757 997 } else {
52042398 998 count = channel->discarded_events;
fb83fe64
JD
999 }
1000
fb83fe64
JD
1001 health_code_update();
1002
1003 /* Send back returned value to session daemon */
52042398 1004 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
fb83fe64
JD
1005 if (ret < 0) {
1006 PERROR("send discarded events");
1007 goto error_fatal;
1008 }
1009
1010 break;
1011 }
1012 case LTTNG_CONSUMER_LOST_PACKETS:
1013 {
52042398
JD
1014 ssize_t ret;
1015 uint64_t count;
fb83fe64
JD
1016 struct lttng_consumer_channel *channel;
1017 uint64_t id = msg.u.lost_packets.session_id;
1018 uint64_t key = msg.u.lost_packets.channel_key;
1019
e5742757
MD
1020 DBG("Kernel consumer lost packets command for session id %"
1021 PRIu64 ", channel key %" PRIu64, id, key);
1022
fb83fe64
JD
1023 channel = consumer_find_channel(key);
1024 if (!channel) {
1025 ERR("Kernel consumer lost packets channel %"
1026 PRIu64 " not found", key);
52042398 1027 count = 0;
e5742757 1028 } else {
52042398 1029 count = channel->lost_packets;
fb83fe64
JD
1030 }
1031
fb83fe64
JD
1032 health_code_update();
1033
1034 /* Send back returned value to session daemon */
52042398 1035 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
fb83fe64
JD
1036 if (ret < 0) {
1037 PERROR("send lost packets");
1038 goto error_fatal;
1039 }
1040
1041 break;
1042 }
b3530820
JG
1043 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1044 {
1045 int channel_monitor_pipe;
1046
1047 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1048 /* Successfully received the command's type. */
1049 ret = consumer_send_status_msg(sock, ret_code);
1050 if (ret < 0) {
1051 goto error_fatal;
1052 }
1053
1054 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
1055 1);
1056 if (ret != sizeof(channel_monitor_pipe)) {
1057 ERR("Failed to receive channel monitor pipe");
1058 goto error_fatal;
1059 }
1060
1061 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1062 ret = consumer_timer_thread_set_channel_monitor_pipe(
1063 channel_monitor_pipe);
1064 if (!ret) {
1065 int flags;
1066
1067 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1068 /* Set the pipe as non-blocking. */
1069 ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
1070 if (ret == -1) {
1071 PERROR("fcntl get flags of the channel monitoring pipe");
1072 goto error_fatal;
1073 }
1074 flags = ret;
1075
1076 ret = fcntl(channel_monitor_pipe, F_SETFL,
1077 flags | O_NONBLOCK);
1078 if (ret == -1) {
1079 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1080 goto error_fatal;
1081 }
1082 DBG("Channel monitor pipe set as non-blocking");
1083 } else {
1084 ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1085 }
1086 ret = consumer_send_status_msg(sock, ret_code);
1087 if (ret < 0) {
1088 goto error_fatal;
1089 }
1090 break;
1091 }
6e0be6cc
JD
1092 case LTTNG_CONSUMER_SET_CHANNEL_ROTATE_PIPE:
1093 {
1094 int channel_rotate_pipe;
1095 int flags;
1096
1097 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1098 /* Successfully received the command's type. */
1099 ret = consumer_send_status_msg(sock, ret_code);
1100 if (ret < 0) {
1101 goto error_fatal;
1102 }
1103
1104 ret = lttcomm_recv_fds_unix_sock(sock, &channel_rotate_pipe,
1105 1);
1106 if (ret != sizeof(channel_rotate_pipe)) {
1107 ERR("Failed to receive channel rotate pipe");
1108 goto error_fatal;
1109 }
1110
1111 DBG("Received channel rotate pipe (%d)", channel_rotate_pipe);
1112 ctx->channel_rotate_pipe = channel_rotate_pipe;
1113 /* Set the pipe as non-blocking. */
1114 ret = fcntl(channel_rotate_pipe, F_GETFL, 0);
1115 if (ret == -1) {
1116 PERROR("fcntl get flags of the channel rotate pipe");
1117 goto error_fatal;
1118 }
1119 flags = ret;
1120
1121 ret = fcntl(channel_rotate_pipe, F_SETFL,
1122 flags | O_NONBLOCK);
1123 if (ret == -1) {
1124 PERROR("fcntl set O_NONBLOCK flag of the channel rotate pipe");
1125 goto error_fatal;
1126 }
1127 DBG("Channel rotate pipe set as non-blocking");
1128 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1129 ret = consumer_send_status_msg(sock, ret_code);
1130 if (ret < 0) {
1131 goto error_fatal;
1132 }
1133 break;
1134 }
1135 case LTTNG_CONSUMER_ROTATE_CHANNEL:
1136 {
6af2ba3b 1137 ret = lttng_consumer_rotate_channel(msg.u.rotate_channel.key,
6e0be6cc
JD
1138 msg.u.rotate_channel.pathname,
1139 msg.u.rotate_channel.relayd_id,
1140 msg.u.rotate_channel.metadata,
1141 ctx);
1142 if (ret < 0) {
1143 ERR("Rotate channel failed");
1144 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1145 }
1146
1147 health_code_update();
1148
1149 ret = consumer_send_status_msg(sock, ret_code);
1150 if (ret < 0) {
1151 /* Somehow, the session daemon is not responding anymore. */
1152 goto end_nosignal;
1153 }
1154
1155 /*
1156 * Rotate the streams that are ready right now.
1157 * FIXME: this is a second consecutive iteration over the
1158 * streams in a channel, there is probably a better way to
1159 * handle this, but it needs to be after the
1160 * consumer_send_status_msg() call.
1161 */
6af2ba3b 1162 ret = lttng_consumer_rotate_ready_streams(
6e0be6cc
JD
1163 msg.u.rotate_channel.key, ctx);
1164 if (ret < 0) {
1165 ERR("Rotate channel failed");
1166 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1167 }
1168 break;
1169 }
3bd1e081 1170 default:
3f8e211f 1171 goto end_nosignal;
3bd1e081 1172 }
3f8e211f 1173
3bd1e081 1174end_nosignal:
b0b335c8 1175 rcu_read_unlock();
4cbc1a04
DG
1176
1177 /*
1178 * Return 1 to indicate success since the 0 value can be a socket
1179 * shutdown during the recv() or send() call.
1180 */
9ce5646a 1181 health_code_update();
4cbc1a04 1182 return 1;
1803a064
MD
1183
1184error_fatal:
1185 rcu_read_unlock();
1186 /* This will issue a consumer stop. */
1187 return -1;
3bd1e081 1188}
d41f73b7 1189
309167d2
JD
1190/*
1191 * Populate index values of a kernel stream. Values are set in big endian order.
1192 *
1193 * Return 0 on success or else a negative value.
1194 */
50adc264 1195static int get_index_values(struct ctf_packet_index *index, int infd)
309167d2
JD
1196{
1197 int ret;
1198
1199 ret = kernctl_get_timestamp_begin(infd, &index->timestamp_begin);
1200 if (ret < 0) {
1201 PERROR("kernctl_get_timestamp_begin");
1202 goto error;
1203 }
1204 index->timestamp_begin = htobe64(index->timestamp_begin);
1205
1206 ret = kernctl_get_timestamp_end(infd, &index->timestamp_end);
1207 if (ret < 0) {
1208 PERROR("kernctl_get_timestamp_end");
1209 goto error;
1210 }
1211 index->timestamp_end = htobe64(index->timestamp_end);
1212
1213 ret = kernctl_get_events_discarded(infd, &index->events_discarded);
1214 if (ret < 0) {
1215 PERROR("kernctl_get_events_discarded");
1216 goto error;
1217 }
1218 index->events_discarded = htobe64(index->events_discarded);
1219
1220 ret = kernctl_get_content_size(infd, &index->content_size);
1221 if (ret < 0) {
1222 PERROR("kernctl_get_content_size");
1223 goto error;
1224 }
1225 index->content_size = htobe64(index->content_size);
1226
1227 ret = kernctl_get_packet_size(infd, &index->packet_size);
1228 if (ret < 0) {
1229 PERROR("kernctl_get_packet_size");
1230 goto error;
1231 }
1232 index->packet_size = htobe64(index->packet_size);
1233
1234 ret = kernctl_get_stream_id(infd, &index->stream_id);
1235 if (ret < 0) {
1236 PERROR("kernctl_get_stream_id");
1237 goto error;
1238 }
1239 index->stream_id = htobe64(index->stream_id);
1240
234cd636
JD
1241 ret = kernctl_get_instance_id(infd, &index->stream_instance_id);
1242 if (ret < 0) {
f0b03c22
MD
1243 if (ret == -ENOTTY) {
1244 /* Command not implemented by lttng-modules. */
1245 index->stream_instance_id = -1ULL;
f0b03c22
MD
1246 } else {
1247 PERROR("kernctl_get_instance_id");
1248 goto error;
1249 }
234cd636
JD
1250 }
1251 index->stream_instance_id = htobe64(index->stream_instance_id);
1252
1253 ret = kernctl_get_sequence_number(infd, &index->packet_seq_num);
1254 if (ret < 0) {
f0b03c22
MD
1255 if (ret == -ENOTTY) {
1256 /* Command not implemented by lttng-modules. */
1257 index->packet_seq_num = -1ULL;
1258 ret = 0;
1259 } else {
1260 PERROR("kernctl_get_sequence_number");
1261 goto error;
1262 }
234cd636
JD
1263 }
1264 index->packet_seq_num = htobe64(index->packet_seq_num);
1265
309167d2
JD
1266error:
1267 return ret;
1268}
94d49140
JD
1269/*
1270 * Sync metadata meaning request them to the session daemon and snapshot to the
1271 * metadata thread can consumer them.
1272 *
1273 * Metadata stream lock MUST be acquired.
1274 *
1275 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1276 * is empty or a negative value on error.
1277 */
1278int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata)
1279{
1280 int ret;
1281
1282 assert(metadata);
1283
1284 ret = kernctl_buffer_flush(metadata->wait_fd);
1285 if (ret < 0) {
1286 ERR("Failed to flush kernel stream");
1287 goto end;
1288 }
1289
1290 ret = kernctl_snapshot(metadata->wait_fd);
1291 if (ret < 0) {
32af2c95 1292 if (ret != -EAGAIN) {
94d49140
JD
1293 ERR("Sync metadata, taking kernel snapshot failed.");
1294 goto end;
1295 }
1296 DBG("Sync metadata, no new kernel metadata");
1297 /* No new metadata, exit. */
1298 ret = ENODATA;
1299 goto end;
1300 }
1301
1302end:
1303 return ret;
1304}
309167d2 1305
fb83fe64
JD
1306static
1307int update_stream_stats(struct lttng_consumer_stream *stream)
1308{
1309 int ret;
1310 uint64_t seq, discarded;
1311
1312 ret = kernctl_get_sequence_number(stream->wait_fd, &seq);
1313 if (ret < 0) {
f0b03c22
MD
1314 if (ret == -ENOTTY) {
1315 /* Command not implemented by lttng-modules. */
1316 seq = -1ULL;
f0b03c22
MD
1317 } else {
1318 PERROR("kernctl_get_sequence_number");
1319 goto end;
1320 }
fb83fe64
JD
1321 }
1322
1323 /*
1324 * Start the sequence when we extract the first packet in case we don't
1325 * start at 0 (for example if a consumer is not connected to the
1326 * session immediately after the beginning).
1327 */
1328 if (stream->last_sequence_number == -1ULL) {
1329 stream->last_sequence_number = seq;
1330 } else if (seq > stream->last_sequence_number) {
1331 stream->chan->lost_packets += seq -
1332 stream->last_sequence_number - 1;
1333 } else {
1334 /* seq <= last_sequence_number */
1335 ERR("Sequence number inconsistent : prev = %" PRIu64
1336 ", current = %" PRIu64,
1337 stream->last_sequence_number, seq);
1338 ret = -1;
1339 goto end;
1340 }
1341 stream->last_sequence_number = seq;
1342
1343 ret = kernctl_get_events_discarded(stream->wait_fd, &discarded);
1344 if (ret < 0) {
1345 PERROR("kernctl_get_events_discarded");
1346 goto end;
1347 }
1348 if (discarded < stream->last_discarded_events) {
1349 /*
83f4233d
MJ
1350 * Overflow has occurred. We assume only one wrap-around
1351 * has occurred.
fb83fe64
JD
1352 */
1353 stream->chan->discarded_events += (1ULL << (CAA_BITS_PER_LONG - 1)) -
1354 stream->last_discarded_events + discarded;
1355 } else {
1356 stream->chan->discarded_events += discarded -
1357 stream->last_discarded_events;
1358 }
1359 stream->last_discarded_events = discarded;
1360 ret = 0;
1361
1362end:
1363 return ret;
1364}
1365
93ec662e
JD
1366/*
1367 * Check if the local version of the metadata stream matches with the version
1368 * of the metadata stream in the kernel. If it was updated, set the reset flag
1369 * on the stream.
1370 */
1371static
1372int metadata_stream_check_version(int infd, struct lttng_consumer_stream *stream)
1373{
1374 int ret;
1375 uint64_t cur_version;
1376
1377 ret = kernctl_get_metadata_version(infd, &cur_version);
1378 if (ret < 0) {
f0b03c22
MD
1379 if (ret == -ENOTTY) {
1380 /*
1381 * LTTng-modules does not implement this
1382 * command.
1383 */
1384 ret = 0;
1385 goto end;
1386 }
93ec662e
JD
1387 ERR("Failed to get the metadata version");
1388 goto end;
1389 }
1390
1391 if (stream->metadata_version == cur_version) {
1392 ret = 0;
1393 goto end;
1394 }
1395
1396 DBG("New metadata version detected");
1397 stream->metadata_version = cur_version;
1398 stream->reset_metadata_flag = 1;
1399 ret = 0;
1400
1401end:
1402 return ret;
1403}
1404
d41f73b7
MD
1405/*
1406 * Consume data on a file descriptor and write it on a trace file.
1407 */
4078b776 1408ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
d41f73b7
MD
1409 struct lttng_consumer_local_data *ctx)
1410{
1d4dfdef 1411 unsigned long len, subbuf_size, padding;
6e0be6cc 1412 int err, write_index = 1, rotation_ret;
4078b776 1413 ssize_t ret = 0;
d41f73b7 1414 int infd = stream->wait_fd;
50adc264 1415 struct ctf_packet_index index;
d41f73b7
MD
1416
1417 DBG("In read_subbuffer (infd : %d)", infd);
309167d2 1418
d41f73b7
MD
1419 /* Get the next subbuffer */
1420 err = kernctl_get_next_subbuf(infd);
1421 if (err != 0) {
d41f73b7
MD
1422 /*
1423 * This is a debug message even for single-threaded consumer,
1424 * because poll() have more relaxed criterions than get subbuf,
1425 * so get_subbuf may fail for short race windows where poll()
1426 * would issue wakeups.
1427 */
1428 DBG("Reserving sub buffer failed (everything is normal, "
1429 "it is due to concurrency)");
32af2c95 1430 ret = err;
b9478f8d 1431 goto error;
d41f73b7
MD
1432 }
1433
1d4dfdef
DG
1434 /* Get the full subbuffer size including padding */
1435 err = kernctl_get_padded_subbuf_size(infd, &len);
1436 if (err != 0) {
5a510c9f 1437 PERROR("Getting sub-buffer len failed.");
8265f19e
MD
1438 err = kernctl_put_subbuf(infd);
1439 if (err != 0) {
32af2c95 1440 if (err == -EFAULT) {
5a510c9f 1441 PERROR("Error in unreserving sub buffer\n");
32af2c95 1442 } else if (err == -EIO) {
8265f19e 1443 /* Should never happen with newer LTTng versions */
5a510c9f 1444 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
8265f19e 1445 }
32af2c95 1446 ret = err;
b9478f8d 1447 goto error;
8265f19e 1448 }
32af2c95 1449 ret = err;
b9478f8d 1450 goto error;
1d4dfdef
DG
1451 }
1452
1c20f0e2 1453 if (!stream->metadata_flag) {
309167d2
JD
1454 ret = get_index_values(&index, infd);
1455 if (ret < 0) {
8265f19e
MD
1456 err = kernctl_put_subbuf(infd);
1457 if (err != 0) {
32af2c95 1458 if (err == -EFAULT) {
5a510c9f 1459 PERROR("Error in unreserving sub buffer\n");
32af2c95 1460 } else if (err == -EIO) {
8265f19e 1461 /* Should never happen with newer LTTng versions */
5a510c9f 1462 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
8265f19e 1463 }
32af2c95 1464 ret = err;
b9478f8d 1465 goto error;
8265f19e 1466 }
b9478f8d 1467 goto error;
309167d2 1468 }
fb83fe64
JD
1469 ret = update_stream_stats(stream);
1470 if (ret < 0) {
7b87473d
MD
1471 err = kernctl_put_subbuf(infd);
1472 if (err != 0) {
1473 if (err == -EFAULT) {
1474 PERROR("Error in unreserving sub buffer\n");
1475 } else if (err == -EIO) {
1476 /* Should never happen with newer LTTng versions */
1477 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1478 }
1479 ret = err;
b9478f8d 1480 goto error;
7b87473d 1481 }
b9478f8d 1482 goto error;
fb83fe64 1483 }
1c20f0e2
JD
1484 } else {
1485 write_index = 0;
93ec662e
JD
1486 ret = metadata_stream_check_version(infd, stream);
1487 if (ret < 0) {
7b87473d
MD
1488 err = kernctl_put_subbuf(infd);
1489 if (err != 0) {
1490 if (err == -EFAULT) {
1491 PERROR("Error in unreserving sub buffer\n");
1492 } else if (err == -EIO) {
1493 /* Should never happen with newer LTTng versions */
1494 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
1495 }
1496 ret = err;
b9478f8d 1497 goto error;
7b87473d 1498 }
b9478f8d 1499 goto error;
93ec662e 1500 }
309167d2
JD
1501 }
1502
ffe60014 1503 switch (stream->chan->output) {
07b86b52 1504 case CONSUMER_CHANNEL_SPLICE:
1d4dfdef
DG
1505 /*
1506 * XXX: The lttng-modules splice "actor" does not handle copying
1507 * partial pages hence only using the subbuffer size without the
1508 * padding makes the splice fail.
1509 */
1510 subbuf_size = len;
1511 padding = 0;
1512
1513 /* splice the subbuffer to the tracefile */
91dfef6e 1514 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
309167d2 1515 padding, &index);
91dfef6e
DG
1516 /*
1517 * XXX: Splice does not support network streaming so the return value
1518 * is simply checked against subbuf_size and not like the mmap() op.
1519 */
1d4dfdef
DG
1520 if (ret != subbuf_size) {
1521 /*
1522 * display the error but continue processing to try
1523 * to release the subbuffer
1524 */
1525 ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
1526 ret, subbuf_size);
309167d2 1527 write_index = 0;
1d4dfdef
DG
1528 }
1529 break;
07b86b52 1530 case CONSUMER_CHANNEL_MMAP:
1d4dfdef
DG
1531 /* Get subbuffer size without padding */
1532 err = kernctl_get_subbuf_size(infd, &subbuf_size);
1533 if (err != 0) {
5a510c9f 1534 PERROR("Getting sub-buffer len failed.");
8265f19e
MD
1535 err = kernctl_put_subbuf(infd);
1536 if (err != 0) {
32af2c95 1537 if (err == -EFAULT) {
5a510c9f 1538 PERROR("Error in unreserving sub buffer\n");
32af2c95 1539 } else if (err == -EIO) {
8265f19e 1540 /* Should never happen with newer LTTng versions */
5a510c9f 1541 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
8265f19e 1542 }
32af2c95 1543 ret = err;
b9478f8d 1544 goto error;
8265f19e 1545 }
32af2c95 1546 ret = err;
b9478f8d 1547 goto error;
1d4dfdef 1548 }
47e81c02 1549
1d4dfdef
DG
1550 /* Make sure the tracer is not gone mad on us! */
1551 assert(len >= subbuf_size);
1552
1553 padding = len - subbuf_size;
1554
1555 /* write the subbuffer to the tracefile */
91dfef6e 1556 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
309167d2 1557 padding, &index);
91dfef6e
DG
1558 /*
1559 * The mmap operation should write subbuf_size amount of data when
1560 * network streaming or the full padding (len) size when we are _not_
1561 * streaming.
1562 */
d88aee68
DG
1563 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
1564 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1d4dfdef 1565 /*
91dfef6e 1566 * Display the error but continue processing to try to release the
2336629e
DG
1567 * subbuffer. This is a DBG statement since this is possible to
1568 * happen without being a critical error.
1d4dfdef 1569 */
2336629e 1570 DBG("Error writing to tracefile "
91dfef6e
DG
1571 "(ret: %zd != len: %lu != subbuf_size: %lu)",
1572 ret, len, subbuf_size);
309167d2 1573 write_index = 0;
1d4dfdef
DG
1574 }
1575 break;
1576 default:
1577 ERR("Unknown output method");
56591bac 1578 ret = -EPERM;
d41f73b7
MD
1579 }
1580
1581 err = kernctl_put_next_subbuf(infd);
1582 if (err != 0) {
32af2c95 1583 if (err == -EFAULT) {
5a510c9f 1584 PERROR("Error in unreserving sub buffer\n");
32af2c95 1585 } else if (err == -EIO) {
d41f73b7 1586 /* Should never happen with newer LTTng versions */
5a510c9f 1587 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted.");
d41f73b7 1588 }
32af2c95 1589 ret = err;
b9478f8d 1590 goto error;
d41f73b7
MD
1591 }
1592
309167d2 1593 /* Write index if needed. */
1c20f0e2 1594 if (!write_index) {
b9478f8d 1595 goto rotate;
1c20f0e2
JD
1596 }
1597
94d49140
JD
1598 if (stream->chan->live_timer_interval && !stream->metadata_flag) {
1599 /*
1600 * In live, block until all the metadata is sent.
1601 */
c585821b
MD
1602 pthread_mutex_lock(&stream->metadata_timer_lock);
1603 assert(!stream->missed_metadata_flush);
1604 stream->waiting_on_metadata = true;
1605 pthread_mutex_unlock(&stream->metadata_timer_lock);
1606
94d49140 1607 err = consumer_stream_sync_metadata(ctx, stream->session_id);
c585821b
MD
1608
1609 pthread_mutex_lock(&stream->metadata_timer_lock);
1610 stream->waiting_on_metadata = false;
1611 if (stream->missed_metadata_flush) {
1612 stream->missed_metadata_flush = false;
1613 pthread_mutex_unlock(&stream->metadata_timer_lock);
1614 (void) consumer_flush_kernel_index(stream);
1615 } else {
1616 pthread_mutex_unlock(&stream->metadata_timer_lock);
1617 }
94d49140 1618 if (err < 0) {
b9478f8d 1619 goto error;
94d49140
JD
1620 }
1621 }
1622
1c20f0e2
JD
1623 err = consumer_stream_write_index(stream, &index);
1624 if (err < 0) {
b9478f8d 1625 goto error;
309167d2
JD
1626 }
1627
b9478f8d 1628rotate:
6af2ba3b 1629 rotation_ret = lttng_consumer_rotate_stream(ctx, stream);
6e0be6cc 1630 if (rotation_ret < 0) {
6e0be6cc 1631 ERR("Stream rotation error");
b9478f8d 1632 goto error;
6e0be6cc 1633 }
b9478f8d
JD
1634
1635error:
d41f73b7
MD
1636 return ret;
1637}
1638
1639int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1640{
1641 int ret;
ffe60014
DG
1642
1643 assert(stream);
1644
2bba9e53
DG
1645 /*
1646 * Don't create anything if this is set for streaming or should not be
1647 * monitored.
1648 */
1649 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
fe4477ee
JD
1650 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
1651 stream->chan->tracefile_size, stream->tracefile_count_current,
309167d2 1652 stream->uid, stream->gid, NULL);
fe4477ee
JD
1653 if (ret < 0) {
1654 goto error;
1655 }
1656 stream->out_fd = ret;
1657 stream->tracefile_size_current = 0;
309167d2
JD
1658
1659 if (!stream->metadata_flag) {
f8f3885c
MD
1660 struct lttng_index_file *index_file;
1661
1662 index_file = lttng_index_file_create(stream->chan->pathname,
309167d2
JD
1663 stream->name, stream->uid, stream->gid,
1664 stream->chan->tracefile_size,
f8f3885c
MD
1665 stream->tracefile_count_current,
1666 CTF_INDEX_MAJOR, CTF_INDEX_MINOR);
1667 if (!index_file) {
309167d2
JD
1668 goto error;
1669 }
1b47ae58 1670 assert(!stream->index_file);
f8f3885c 1671 stream->index_file = index_file;
309167d2 1672 }
ffe60014 1673 }
d41f73b7 1674
d41f73b7
MD
1675 if (stream->output == LTTNG_EVENT_MMAP) {
1676 /* get the len of the mmap region */
1677 unsigned long mmap_len;
1678
1679 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1680 if (ret != 0) {
ffe60014 1681 PERROR("kernctl_get_mmap_len");
d41f73b7
MD
1682 goto error_close_fd;
1683 }
1684 stream->mmap_len = (size_t) mmap_len;
1685
ffe60014
DG
1686 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1687 MAP_PRIVATE, stream->wait_fd, 0);
d41f73b7 1688 if (stream->mmap_base == MAP_FAILED) {
ffe60014 1689 PERROR("Error mmaping");
d41f73b7
MD
1690 ret = -1;
1691 goto error_close_fd;
1692 }
1693 }
1694
1695 /* we return 0 to let the library handle the FD internally */
1696 return 0;
1697
1698error_close_fd:
2f225ce2 1699 if (stream->out_fd >= 0) {
d41f73b7
MD
1700 int err;
1701
1702 err = close(stream->out_fd);
1703 assert(!err);
2f225ce2 1704 stream->out_fd = -1;
d41f73b7
MD
1705 }
1706error:
1707 return ret;
1708}
1709
ca22feea
DG
1710/*
1711 * Check if data is still being extracted from the buffers for a specific
4e9a4686
DG
1712 * stream. Consumer data lock MUST be acquired before calling this function
1713 * and the stream lock.
ca22feea 1714 *
6d805429 1715 * Return 1 if the traced data are still getting read else 0 meaning that the
ca22feea
DG
1716 * data is available for trace viewer reading.
1717 */
6d805429 1718int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
ca22feea
DG
1719{
1720 int ret;
1721
1722 assert(stream);
1723
873b9e9a
MD
1724 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1725 ret = 0;
1726 goto end;
1727 }
1728
ca22feea
DG
1729 ret = kernctl_get_next_subbuf(stream->wait_fd);
1730 if (ret == 0) {
1731 /* There is still data so let's put back this subbuffer. */
1732 ret = kernctl_put_subbuf(stream->wait_fd);
1733 assert(ret == 0);
6d805429 1734 ret = 1; /* Data is pending */
4e9a4686 1735 goto end;
ca22feea
DG
1736 }
1737
6d805429
DG
1738 /* Data is NOT pending and ready to be read. */
1739 ret = 0;
ca22feea 1740
6efae65e
DG
1741end:
1742 return ret;
ca22feea 1743}
This page took 0.156143 seconds and 5 git commands to generate.