consumerd: refactor: split read_subbuf into sub-operations
[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>
e5148e25 46#include <common/optional.h>
6f1177cf
JG
47#include <common/buffer-view.h>
48#include <common/consumer/consumer.h>
49#include <stdint.h>
0857097f 50
10a8a223 51#include "kernel-consumer.h"
3bd1e081
MD
52
53extern struct lttng_consumer_global_data consumer_data;
54extern int consumer_poll_timeout;
3bd1e081 55
3bd1e081
MD
56/*
57 * Take a snapshot for a specific fd
58 *
59 * Returns 0 on success, < 0 on error
60 */
ffe60014 61int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
3bd1e081
MD
62{
63 int ret = 0;
64 int infd = stream->wait_fd;
65
66 ret = kernctl_snapshot(infd);
d2d2f190
JD
67 /*
68 * -EAGAIN is not an error, it just means that there is no data to
69 * be read.
70 */
71 if (ret != 0 && ret != -EAGAIN) {
5a510c9f 72 PERROR("Getting sub-buffer snapshot.");
3bd1e081
MD
73 }
74
75 return ret;
76}
77
e9404c27
JG
78/*
79 * Sample consumed and produced positions for a specific fd.
80 *
81 * Returns 0 on success, < 0 on error.
82 */
83int lttng_kconsumer_sample_snapshot_positions(
84 struct lttng_consumer_stream *stream)
85{
86 assert(stream);
87
88 return kernctl_snapshot_sample_positions(stream->wait_fd);
89}
90
3bd1e081
MD
91/*
92 * Get the produced position
93 *
94 * Returns 0 on success, < 0 on error
95 */
ffe60014 96int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
3bd1e081
MD
97 unsigned long *pos)
98{
99 int ret;
100 int infd = stream->wait_fd;
101
102 ret = kernctl_snapshot_get_produced(infd, pos);
103 if (ret != 0) {
5a510c9f 104 PERROR("kernctl_snapshot_get_produced");
3bd1e081
MD
105 }
106
107 return ret;
108}
109
07b86b52
JD
110/*
111 * Get the consumerd position
112 *
113 * Returns 0 on success, < 0 on error
114 */
115int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
116 unsigned long *pos)
117{
118 int ret;
119 int infd = stream->wait_fd;
120
121 ret = kernctl_snapshot_get_consumed(infd, pos);
122 if (ret != 0) {
5a510c9f 123 PERROR("kernctl_snapshot_get_consumed");
07b86b52
JD
124 }
125
126 return ret;
127}
128
7775df52
JG
129static
130int get_current_subbuf_addr(struct lttng_consumer_stream *stream,
131 const char **addr)
132{
133 int ret;
134 unsigned long mmap_offset;
135 const char *mmap_base = stream->mmap_base;
136
137 ret = kernctl_get_mmap_read_offset(stream->wait_fd, &mmap_offset);
138 if (ret < 0) {
139 PERROR("Failed to get mmap read offset");
140 goto error;
141 }
142
143 *addr = mmap_base + mmap_offset;
144error:
145 return ret;
146}
147
07b86b52
JD
148/*
149 * Take a snapshot of all the stream of a channel
b0226bd4 150 * RCU read-side lock must be held across this function to ensure existence of
e5148e25 151 * channel. The channel lock must be held by the caller.
07b86b52
JD
152 *
153 * Returns 0 on success, < 0 on error
154 */
f0f048c9
JG
155static int lttng_kconsumer_snapshot_channel(
156 struct lttng_consumer_channel *channel,
157 uint64_t key, char *path, uint64_t relayd_id,
158 uint64_t nb_packets_per_stream,
5c786ded 159 struct lttng_consumer_local_data *ctx)
07b86b52
JD
160{
161 int ret;
07b86b52
JD
162 struct lttng_consumer_stream *stream;
163
6a00837f 164 DBG("Kernel consumer snapshot channel %" PRIu64, key);
07b86b52
JD
165
166 rcu_read_lock();
167
07b86b52
JD
168 /* Splice is not supported yet for channel snapshot. */
169 if (channel->output != CONSUMER_CHANNEL_MMAP) {
7318a78f
JG
170 ERR("Unsupported output type for channel \"%s\": mmap output is required to record a snapshot",
171 channel->name);
07b86b52
JD
172 ret = -1;
173 goto end;
174 }
175
10a50311 176 cds_list_for_each_entry(stream, &channel->streams.head, send_node) {
923333cd 177 unsigned long consumed_pos, produced_pos;
9ce5646a
MD
178
179 health_code_update();
180
07b86b52
JD
181 /*
182 * Lock stream because we are about to change its state.
183 */
184 pthread_mutex_lock(&stream->lock);
185
e5148e25
JG
186 assert(channel->trace_chunk);
187 if (!lttng_trace_chunk_get(channel->trace_chunk)) {
188 /*
189 * Can't happen barring an internal error as the channel
190 * holds a reference to the trace chunk.
191 */
192 ERR("Failed to acquire reference to channel's trace chunk");
193 ret = -1;
194 goto end_unlock;
195 }
196 assert(!stream->trace_chunk);
197 stream->trace_chunk = channel->trace_chunk;
198
29decac3
DG
199 /*
200 * Assign the received relayd ID so we can use it for streaming. The streams
201 * are not visible to anyone so this is OK to change it.
202 */
07b86b52
JD
203 stream->net_seq_idx = relayd_id;
204 channel->relayd_id = relayd_id;
205 if (relayd_id != (uint64_t) -1ULL) {
10a50311 206 ret = consumer_send_relayd_stream(stream, path);
07b86b52
JD
207 if (ret < 0) {
208 ERR("sending stream to relayd");
209 goto end_unlock;
210 }
07b86b52 211 } else {
e5148e25
JG
212 ret = consumer_stream_create_output_files(stream,
213 false);
07b86b52 214 if (ret < 0) {
07b86b52
JD
215 goto end_unlock;
216 }
e5148e25
JG
217 DBG("Kernel consumer snapshot stream (%" PRIu64 ")",
218 stream->key);
07b86b52
JD
219 }
220
f22dd891 221 ret = kernctl_buffer_flush_empty(stream->wait_fd);
07b86b52 222 if (ret < 0) {
f22dd891
MD
223 /*
224 * Doing a buffer flush which does not take into
225 * account empty packets. This is not perfect
226 * for stream intersection, but required as a
227 * fall-back when "flush_empty" is not
228 * implemented by lttng-modules.
229 */
230 ret = kernctl_buffer_flush(stream->wait_fd);
231 if (ret < 0) {
232 ERR("Failed to flush kernel stream");
233 goto end_unlock;
234 }
07b86b52
JD
235 goto end_unlock;
236 }
237
238 ret = lttng_kconsumer_take_snapshot(stream);
239 if (ret < 0) {
240 ERR("Taking kernel snapshot");
241 goto end_unlock;
242 }
243
244 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
245 if (ret < 0) {
246 ERR("Produced kernel snapshot position");
247 goto end_unlock;
248 }
249
250 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
251 if (ret < 0) {
252 ERR("Consumerd kernel snapshot position");
253 goto end_unlock;
254 }
255
d07ceecd
MD
256 consumed_pos = consumer_get_consume_start_pos(consumed_pos,
257 produced_pos, nb_packets_per_stream,
258 stream->max_sb_size);
5c786ded 259
0fdaf1ed 260 while ((long) (consumed_pos - produced_pos) < 0) {
07b86b52
JD
261 ssize_t read_len;
262 unsigned long len, padded_len;
7775df52 263 const char *subbuf_addr;
b770aa7f 264 struct lttng_buffer_view subbuf_view;
07b86b52 265
9ce5646a 266 health_code_update();
07b86b52
JD
267 DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
268
269 ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
270 if (ret < 0) {
32af2c95 271 if (ret != -EAGAIN) {
07b86b52
JD
272 PERROR("kernctl_get_subbuf snapshot");
273 goto end_unlock;
274 }
275 DBG("Kernel consumer get subbuf failed. Skipping it.");
276 consumed_pos += stream->max_sb_size;
ddc93ee4 277 stream->chan->lost_packets++;
07b86b52
JD
278 continue;
279 }
280
281 ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
282 if (ret < 0) {
283 ERR("Snapshot kernctl_get_subbuf_size");
29decac3 284 goto error_put_subbuf;
07b86b52
JD
285 }
286
287 ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
288 if (ret < 0) {
289 ERR("Snapshot kernctl_get_padded_subbuf_size");
29decac3 290 goto error_put_subbuf;
07b86b52
JD
291 }
292
7775df52
JG
293 ret = get_current_subbuf_addr(stream, &subbuf_addr);
294 if (ret) {
295 goto error_put_subbuf;
296 }
297
b770aa7f
JG
298 subbuf_view = lttng_buffer_view_init(
299 subbuf_addr, 0, padded_len);
7775df52 300 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx,
b770aa7f 301 stream, &subbuf_view,
6f1177cf 302 padded_len - len);
07b86b52 303 /*
29decac3
DG
304 * We write the padded len in local tracefiles but the data len
305 * when using a relay. Display the error but continue processing
306 * to try to release the subbuffer.
07b86b52
JD
307 */
308 if (relayd_id != (uint64_t) -1ULL) {
309 if (read_len != len) {
310 ERR("Error sending to the relay (ret: %zd != len: %lu)",
311 read_len, len);
312 }
313 } else {
314 if (read_len != padded_len) {
315 ERR("Error writing to tracefile (ret: %zd != len: %lu)",
316 read_len, padded_len);
317 }
318 }
319
320 ret = kernctl_put_subbuf(stream->wait_fd);
321 if (ret < 0) {
322 ERR("Snapshot kernctl_put_subbuf");
323 goto end_unlock;
324 }
325 consumed_pos += stream->max_sb_size;
326 }
327
328 if (relayd_id == (uint64_t) -1ULL) {
fdf9986c
MD
329 if (stream->out_fd >= 0) {
330 ret = close(stream->out_fd);
331 if (ret < 0) {
332 PERROR("Kernel consumer snapshot close out_fd");
333 goto end_unlock;
334 }
335 stream->out_fd = -1;
07b86b52 336 }
07b86b52
JD
337 } else {
338 close_relayd_stream(stream);
339 stream->net_seq_idx = (uint64_t) -1ULL;
340 }
e5148e25
JG
341 lttng_trace_chunk_put(stream->trace_chunk);
342 stream->trace_chunk = NULL;
07b86b52
JD
343 pthread_mutex_unlock(&stream->lock);
344 }
345
346 /* All good! */
347 ret = 0;
348 goto end;
349
29decac3
DG
350error_put_subbuf:
351 ret = kernctl_put_subbuf(stream->wait_fd);
352 if (ret < 0) {
353 ERR("Snapshot kernctl_put_subbuf error path");
354 }
07b86b52
JD
355end_unlock:
356 pthread_mutex_unlock(&stream->lock);
357end:
358 rcu_read_unlock();
359 return ret;
360}
361
362/*
363 * Read the whole metadata available for a snapshot.
b0226bd4 364 * RCU read-side lock must be held across this function to ensure existence of
e5148e25 365 * metadata_channel. The channel lock must be held by the caller.
07b86b52
JD
366 *
367 * Returns 0 on success, < 0 on error
368 */
e5148e25
JG
369static int lttng_kconsumer_snapshot_metadata(
370 struct lttng_consumer_channel *metadata_channel,
b0226bd4
MD
371 uint64_t key, char *path, uint64_t relayd_id,
372 struct lttng_consumer_local_data *ctx)
07b86b52 373{
d771f832
DG
374 int ret, use_relayd = 0;
375 ssize_t ret_read;
07b86b52 376 struct lttng_consumer_stream *metadata_stream;
d771f832
DG
377
378 assert(ctx);
07b86b52
JD
379
380 DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
381 key, path);
382
383 rcu_read_lock();
384
07b86b52
JD
385 metadata_stream = metadata_channel->metadata_stream;
386 assert(metadata_stream);
e5148e25 387
c55fe3e3 388 pthread_mutex_lock(&metadata_stream->lock);
e5148e25
JG
389 assert(metadata_channel->trace_chunk);
390 assert(metadata_stream->trace_chunk);
07b86b52 391
d771f832 392 /* Flag once that we have a valid relayd for the stream. */
e2039c7a 393 if (relayd_id != (uint64_t) -1ULL) {
d771f832
DG
394 use_relayd = 1;
395 }
396
397 if (use_relayd) {
10a50311 398 ret = consumer_send_relayd_stream(metadata_stream, path);
e2039c7a 399 if (ret < 0) {
c55fe3e3 400 goto error_snapshot;
e2039c7a 401 }
e2039c7a 402 } else {
e5148e25
JG
403 ret = consumer_stream_create_output_files(metadata_stream,
404 false);
e2039c7a 405 if (ret < 0) {
c55fe3e3 406 goto error_snapshot;
e2039c7a 407 }
07b86b52 408 }
07b86b52 409
d771f832 410 do {
9ce5646a
MD
411 health_code_update();
412
6f1177cf 413 ret_read = lttng_consumer_read_subbuffer(metadata_stream, ctx, true);
d771f832 414 if (ret_read < 0) {
56591bac 415 if (ret_read != -EAGAIN) {
6a00837f 416 ERR("Kernel snapshot reading metadata subbuffer (ret: %zd)",
d771f832 417 ret_read);
c55fe3e3
JG
418 ret = ret_read;
419 goto error_snapshot;
07b86b52 420 }
d771f832 421 /* ret_read is negative at this point so we will exit the loop. */
07b86b52
JD
422 continue;
423 }
d771f832 424 } while (ret_read >= 0);
07b86b52 425
d771f832
DG
426 if (use_relayd) {
427 close_relayd_stream(metadata_stream);
428 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
429 } else {
fdf9986c
MD
430 if (metadata_stream->out_fd >= 0) {
431 ret = close(metadata_stream->out_fd);
432 if (ret < 0) {
433 PERROR("Kernel consumer snapshot metadata close out_fd");
434 /*
435 * Don't go on error here since the snapshot was successful at this
436 * point but somehow the close failed.
437 */
438 }
439 metadata_stream->out_fd = -1;
e5148e25
JG
440 lttng_trace_chunk_put(metadata_stream->trace_chunk);
441 metadata_stream->trace_chunk = NULL;
e2039c7a 442 }
e2039c7a
JD
443 }
444
07b86b52 445 ret = 0;
c55fe3e3
JG
446error_snapshot:
447 pthread_mutex_unlock(&metadata_stream->lock);
cf53a8a6
JD
448 cds_list_del(&metadata_stream->send_node);
449 consumer_stream_destroy(metadata_stream, NULL);
450 metadata_channel->metadata_stream = NULL;
07b86b52
JD
451 rcu_read_unlock();
452 return ret;
453}
454
1803a064
MD
455/*
456 * Receive command from session daemon and process it.
457 *
458 * Return 1 on success else a negative value or 0.
459 */
3bd1e081
MD
460int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
461 int sock, struct pollfd *consumer_sockpoll)
462{
463 ssize_t ret;
0c759fc9 464 enum lttcomm_return_code ret_code = LTTCOMM_CONSUMERD_SUCCESS;
3bd1e081
MD
465 struct lttcomm_consumer_msg msg;
466
9ce5646a
MD
467 health_code_update();
468
3bd1e081
MD
469 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
470 if (ret != sizeof(msg)) {
1803a064 471 if (ret > 0) {
c6857fcf 472 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
1803a064
MD
473 ret = -1;
474 }
3bd1e081
MD
475 return ret;
476 }
9ce5646a
MD
477
478 health_code_update();
479
84382d49
MD
480 /* Deprecated command */
481 assert(msg.cmd_type != LTTNG_CONSUMER_STOP);
3bd1e081 482
9ce5646a
MD
483 health_code_update();
484
b0b335c8
MD
485 /* relayd needs RCU read-side protection */
486 rcu_read_lock();
487
3bd1e081 488 switch (msg.cmd_type) {
00e2e675
DG
489 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
490 {
f50f23d9 491 /* Session daemon status message are handled in the following call. */
2527bf85 492 consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
7735ef9e 493 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
d3e2ba59 494 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id,
2527bf85 495 msg.u.relayd_sock.relayd_session_id);
00e2e675
DG
496 goto end_nosignal;
497 }
3bd1e081
MD
498 case LTTNG_CONSUMER_ADD_CHANNEL:
499 {
500 struct lttng_consumer_channel *new_channel;
e43c41c5 501 int ret_recv;
e5148e25 502 const uint64_t chunk_id = msg.u.channel.chunk_id.value;
3bd1e081 503
9ce5646a
MD
504 health_code_update();
505
f50f23d9
DG
506 /* First send a status message before receiving the fds. */
507 ret = consumer_send_status_msg(sock, ret_code);
508 if (ret < 0) {
509 /* Somehow, the session daemon is not responding anymore. */
1803a064 510 goto error_fatal;
f50f23d9 511 }
9ce5646a
MD
512
513 health_code_update();
514
d88aee68 515 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
3bd1e081 516 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
e5148e25
JG
517 msg.u.channel.session_id,
518 msg.u.channel.chunk_id.is_set ?
519 &chunk_id : NULL,
520 msg.u.channel.pathname,
521 msg.u.channel.name,
1624d5b7
JD
522 msg.u.channel.relayd_id, msg.u.channel.output,
523 msg.u.channel.tracefile_size,
1950109e 524 msg.u.channel.tracefile_count, 0,
ecc48a90 525 msg.u.channel.monitor,
d7ba1388 526 msg.u.channel.live_timer_interval,
3ef395a9 527 msg.u.channel.is_live,
3d071855 528 NULL, NULL);
3bd1e081 529 if (new_channel == NULL) {
f73fabfd 530 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
3bd1e081
MD
531 goto end_nosignal;
532 }
ffe60014 533 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
95a1109b
JD
534 switch (msg.u.channel.output) {
535 case LTTNG_EVENT_SPLICE:
536 new_channel->output = CONSUMER_CHANNEL_SPLICE;
537 break;
538 case LTTNG_EVENT_MMAP:
539 new_channel->output = CONSUMER_CHANNEL_MMAP;
540 break;
541 default:
542 ERR("Channel output unknown %d", msg.u.channel.output);
543 goto end_nosignal;
544 }
ffe60014
DG
545
546 /* Translate and save channel type. */
547 switch (msg.u.channel.type) {
548 case CONSUMER_CHANNEL_TYPE_DATA:
549 case CONSUMER_CHANNEL_TYPE_METADATA:
550 new_channel->type = msg.u.channel.type;
551 break;
552 default:
553 assert(0);
554 goto end_nosignal;
555 };
556
9ce5646a
MD
557 health_code_update();
558
3bd1e081 559 if (ctx->on_recv_channel != NULL) {
e43c41c5
JD
560 ret_recv = ctx->on_recv_channel(new_channel);
561 if (ret_recv == 0) {
562 ret = consumer_add_channel(new_channel, ctx);
563 } else if (ret_recv < 0) {
3bd1e081
MD
564 goto end_nosignal;
565 }
566 } else {
e43c41c5 567 ret = consumer_add_channel(new_channel, ctx);
3bd1e081 568 }
e9404c27
JG
569 if (msg.u.channel.type == CONSUMER_CHANNEL_TYPE_DATA && !ret) {
570 int monitor_start_ret;
571
572 DBG("Consumer starting monitor timer");
94d49140
JD
573 consumer_timer_live_start(new_channel,
574 msg.u.channel.live_timer_interval);
e9404c27
JG
575 monitor_start_ret = consumer_timer_monitor_start(
576 new_channel,
577 msg.u.channel.monitor_timer_interval);
578 if (monitor_start_ret < 0) {
579 ERR("Starting channel monitoring timer failed");
580 goto end_nosignal;
581 }
582
94d49140 583 }
e43c41c5 584
9ce5646a
MD
585 health_code_update();
586
e43c41c5 587 /* If we received an error in add_channel, we need to report it. */
821fffb2 588 if (ret < 0) {
1803a064
MD
589 ret = consumer_send_status_msg(sock, ret);
590 if (ret < 0) {
591 goto error_fatal;
592 }
e43c41c5
JD
593 goto end_nosignal;
594 }
595
3bd1e081
MD
596 goto end_nosignal;
597 }
598 case LTTNG_CONSUMER_ADD_STREAM:
599 {
dae10966
DG
600 int fd;
601 struct lttng_pipe *stream_pipe;
00e2e675 602 struct lttng_consumer_stream *new_stream;
ffe60014 603 struct lttng_consumer_channel *channel;
c80048c6 604 int alloc_ret = 0;
3bd1e081 605
ffe60014
DG
606 /*
607 * Get stream's channel reference. Needed when adding the stream to the
608 * global hash table.
609 */
610 channel = consumer_find_channel(msg.u.stream.channel_key);
611 if (!channel) {
612 /*
613 * We could not find the channel. Can happen if cpu hotplug
614 * happens while tearing down.
615 */
d88aee68 616 ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
e462382a 617 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
ffe60014
DG
618 }
619
9ce5646a
MD
620 health_code_update();
621
f50f23d9
DG
622 /* First send a status message before receiving the fds. */
623 ret = consumer_send_status_msg(sock, ret_code);
1803a064 624 if (ret < 0) {
d771f832 625 /* Somehow, the session daemon is not responding anymore. */
e0e7757d 626 goto error_add_stream_fatal;
1803a064 627 }
9ce5646a
MD
628
629 health_code_update();
630
0c759fc9 631 if (ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
d771f832 632 /* Channel was not found. */
e0e7757d 633 goto error_add_stream_nosignal;
f50f23d9
DG
634 }
635
d771f832 636 /* Blocking call */
9ce5646a
MD
637 health_poll_entry();
638 ret = lttng_consumer_poll_socket(consumer_sockpoll);
639 health_poll_exit();
84382d49 640 if (ret) {
e0e7757d 641 goto error_add_stream_fatal;
3bd1e081 642 }
00e2e675 643
9ce5646a
MD
644 health_code_update();
645
00e2e675 646 /* Get stream file descriptor from socket */
f2fc6720
MD
647 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
648 if (ret != sizeof(fd)) {
f73fabfd 649 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
e0e7757d 650 goto end;
3bd1e081 651 }
3bd1e081 652
9ce5646a
MD
653 health_code_update();
654
f50f23d9
DG
655 /*
656 * Send status code to session daemon only if the recv works. If the
657 * above recv() failed, the session daemon is notified through the
658 * error socket and the teardown is eventually done.
659 */
660 ret = consumer_send_status_msg(sock, ret_code);
661 if (ret < 0) {
662 /* Somehow, the session daemon is not responding anymore. */
e0e7757d 663 goto error_add_stream_nosignal;
f50f23d9
DG
664 }
665
9ce5646a
MD
666 health_code_update();
667
e5148e25 668 pthread_mutex_lock(&channel->lock);
6f1177cf 669 new_stream = consumer_stream_create(
c0c85e4f
JG
670 channel,
671 channel->key,
ffe60014 672 fd,
ffe60014 673 channel->name,
ffe60014
DG
674 channel->relayd_id,
675 channel->session_id,
e5148e25 676 channel->trace_chunk,
ffe60014
DG
677 msg.u.stream.cpu,
678 &alloc_ret,
4891ece8 679 channel->type,
e5148e25 680 channel->monitor);
3bd1e081 681 if (new_stream == NULL) {
c80048c6
MD
682 switch (alloc_ret) {
683 case -ENOMEM:
684 case -EINVAL:
685 default:
686 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
687 break;
c80048c6 688 }
e5148e25 689 pthread_mutex_unlock(&channel->lock);
e0e7757d 690 goto error_add_stream_nosignal;
3bd1e081 691 }
d771f832 692
ffe60014 693 new_stream->wait_fd = fd;
2c42aa5c
JG
694 ret = kernctl_get_max_subbuf_size(new_stream->wait_fd,
695 &new_stream->max_sb_size);
696 if (ret < 0) {
697 pthread_mutex_unlock(&channel->lock);
698 ERR("Failed to get kernel maximal subbuffer size");
e0e7757d 699 goto error_add_stream_nosignal;
2c42aa5c
JG
700 }
701
d9a2e16e
JD
702 consumer_stream_update_channel_attributes(new_stream,
703 channel);
00e2e675 704
a0c83db9
DG
705 /*
706 * We've just assigned the channel to the stream so increment the
07b86b52
JD
707 * refcount right now. We don't need to increment the refcount for
708 * streams in no monitor because we handle manually the cleanup of
709 * those. It is very important to make sure there is NO prior
710 * consumer_del_stream() calls or else the refcount will be unbalanced.
a0c83db9 711 */
07b86b52
JD
712 if (channel->monitor) {
713 uatomic_inc(&new_stream->chan->refcount);
714 }
9d9353f9 715
fb3a43a9
DG
716 /*
717 * The buffer flush is done on the session daemon side for the kernel
718 * so no need for the stream "hangup_flush_done" variable to be
719 * tracked. This is important for a kernel stream since we don't rely
720 * on the flush state of the stream to read data. It's not the case for
721 * user space tracing.
722 */
723 new_stream->hangup_flush_done = 0;
724
9ce5646a
MD
725 health_code_update();
726
e5148e25 727 pthread_mutex_lock(&new_stream->lock);
633d0084
DG
728 if (ctx->on_recv_stream) {
729 ret = ctx->on_recv_stream(new_stream);
730 if (ret < 0) {
e5148e25
JG
731 pthread_mutex_unlock(&new_stream->lock);
732 pthread_mutex_unlock(&channel->lock);
d771f832 733 consumer_stream_free(new_stream);
e0e7757d 734 goto error_add_stream_nosignal;
fb3a43a9 735 }
633d0084 736 }
9ce5646a
MD
737 health_code_update();
738
07b86b52
JD
739 if (new_stream->metadata_flag) {
740 channel->metadata_stream = new_stream;
741 }
742
2bba9e53
DG
743 /* Do not monitor this stream. */
744 if (!channel->monitor) {
5eecee74 745 DBG("Kernel consumer add stream %s in no monitor mode with "
6dc3064a 746 "relayd id %" PRIu64, new_stream->name,
5eecee74 747 new_stream->net_seq_idx);
10a50311 748 cds_list_add(&new_stream->send_node, &channel->streams.head);
e5148e25
JG
749 pthread_mutex_unlock(&new_stream->lock);
750 pthread_mutex_unlock(&channel->lock);
e0e7757d 751 goto end_add_stream;
6dc3064a
DG
752 }
753
e1b71bdc
DG
754 /* Send stream to relayd if the stream has an ID. */
755 if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
194ee077
DG
756 ret = consumer_send_relayd_stream(new_stream,
757 new_stream->chan->pathname);
e1b71bdc 758 if (ret < 0) {
e5148e25
JG
759 pthread_mutex_unlock(&new_stream->lock);
760 pthread_mutex_unlock(&channel->lock);
e1b71bdc 761 consumer_stream_free(new_stream);
e0e7757d 762 goto error_add_stream_nosignal;
e1b71bdc 763 }
001b7e62
MD
764
765 /*
766 * If adding an extra stream to an already
767 * existing channel (e.g. cpu hotplug), we need
768 * to send the "streams_sent" command to relayd.
769 */
770 if (channel->streams_sent_to_relayd) {
771 ret = consumer_send_relayd_streams_sent(
772 new_stream->net_seq_idx);
773 if (ret < 0) {
e5148e25
JG
774 pthread_mutex_unlock(&new_stream->lock);
775 pthread_mutex_unlock(&channel->lock);
e0e7757d 776 goto error_add_stream_nosignal;
001b7e62
MD
777 }
778 }
e2039c7a 779 }
e5148e25
JG
780 pthread_mutex_unlock(&new_stream->lock);
781 pthread_mutex_unlock(&channel->lock);
e2039c7a 782
50f8ae69 783 /* Get the right pipe where the stream will be sent. */
633d0084 784 if (new_stream->metadata_flag) {
66d583dc 785 consumer_add_metadata_stream(new_stream);
dae10966 786 stream_pipe = ctx->consumer_metadata_pipe;
3bd1e081 787 } else {
66d583dc 788 consumer_add_data_stream(new_stream);
dae10966 789 stream_pipe = ctx->consumer_data_pipe;
50f8ae69
DG
790 }
791
66d583dc 792 /* Visible to other threads */
5ab66908
MD
793 new_stream->globally_visible = 1;
794
9ce5646a
MD
795 health_code_update();
796
dae10966 797 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
50f8ae69 798 if (ret < 0) {
dae10966 799 ERR("Consumer write %s stream to pipe %d",
50f8ae69 800 new_stream->metadata_flag ? "metadata" : "data",
dae10966 801 lttng_pipe_get_writefd(stream_pipe));
5ab66908
MD
802 if (new_stream->metadata_flag) {
803 consumer_del_stream_for_metadata(new_stream);
804 } else {
805 consumer_del_stream_for_data(new_stream);
806 }
e0e7757d 807 goto error_add_stream_nosignal;
3bd1e081 808 }
00e2e675 809
02d02e31
JD
810 DBG("Kernel consumer ADD_STREAM %s (fd: %d) %s with relayd id %" PRIu64,
811 new_stream->name, fd, new_stream->chan->pathname, new_stream->relayd_stream_id);
e0e7757d 812end_add_stream:
3bd1e081 813 break;
e0e7757d
JG
814error_add_stream_nosignal:
815 goto end_nosignal;
816error_add_stream_fatal:
817 goto error_fatal;
3bd1e081 818 }
a4baae1b
JD
819 case LTTNG_CONSUMER_STREAMS_SENT:
820 {
821 struct lttng_consumer_channel *channel;
822
823 /*
824 * Get stream's channel reference. Needed when adding the stream to the
825 * global hash table.
826 */
827 channel = consumer_find_channel(msg.u.sent_streams.channel_key);
828 if (!channel) {
829 /*
830 * We could not find the channel. Can happen if cpu hotplug
831 * happens while tearing down.
832 */
833 ERR("Unable to find channel key %" PRIu64,
834 msg.u.sent_streams.channel_key);
e462382a 835 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
a4baae1b
JD
836 }
837
838 health_code_update();
839
840 /*
841 * Send status code to session daemon.
842 */
843 ret = consumer_send_status_msg(sock, ret_code);
f261ad0a 844 if (ret < 0 || ret_code != LTTCOMM_CONSUMERD_SUCCESS) {
a4baae1b 845 /* Somehow, the session daemon is not responding anymore. */
6ec577e9 846 goto error_streams_sent_nosignal;
a4baae1b
JD
847 }
848
849 health_code_update();
850
851 /*
852 * We should not send this message if we don't monitor the
853 * streams in this channel.
854 */
855 if (!channel->monitor) {
6ec577e9 856 goto end_error_streams_sent;
a4baae1b
JD
857 }
858
859 health_code_update();
860 /* Send stream to relayd if the stream has an ID. */
861 if (msg.u.sent_streams.net_seq_idx != (uint64_t) -1ULL) {
862 ret = consumer_send_relayd_streams_sent(
863 msg.u.sent_streams.net_seq_idx);
864 if (ret < 0) {
6ec577e9 865 goto error_streams_sent_nosignal;
a4baae1b 866 }
001b7e62 867 channel->streams_sent_to_relayd = true;
a4baae1b 868 }
6ec577e9 869end_error_streams_sent:
a4baae1b 870 break;
6ec577e9
JG
871error_streams_sent_nosignal:
872 goto end_nosignal;
a4baae1b 873 }
3bd1e081
MD
874 case LTTNG_CONSUMER_UPDATE_STREAM:
875 {
3f8e211f
DG
876 rcu_read_unlock();
877 return -ENOSYS;
878 }
879 case LTTNG_CONSUMER_DESTROY_RELAYD:
880 {
a6ba4fe1 881 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
3f8e211f
DG
882 struct consumer_relayd_sock_pair *relayd;
883
a6ba4fe1 884 DBG("Kernel consumer destroying relayd %" PRIu64, index);
3f8e211f
DG
885
886 /* Get relayd reference if exists. */
a6ba4fe1 887 relayd = consumer_find_relayd(index);
3f8e211f 888 if (relayd == NULL) {
3448e266 889 DBG("Unable to find relayd %" PRIu64, index);
e462382a 890 ret_code = LTTCOMM_CONSUMERD_RELAYD_FAIL;
3bd1e081 891 }
3f8e211f 892
a6ba4fe1
DG
893 /*
894 * Each relayd socket pair has a refcount of stream attached to it
895 * which tells if the relayd is still active or not depending on the
896 * refcount value.
897 *
898 * This will set the destroy flag of the relayd object and destroy it
899 * if the refcount reaches zero when called.
900 *
901 * The destroy can happen either here or when a stream fd hangs up.
902 */
f50f23d9
DG
903 if (relayd) {
904 consumer_flag_relayd_for_destroy(relayd);
905 }
906
9ce5646a
MD
907 health_code_update();
908
f50f23d9
DG
909 ret = consumer_send_status_msg(sock, ret_code);
910 if (ret < 0) {
911 /* Somehow, the session daemon is not responding anymore. */
1803a064 912 goto error_fatal;
f50f23d9 913 }
3f8e211f 914
3f8e211f 915 goto end_nosignal;
3bd1e081 916 }
6d805429 917 case LTTNG_CONSUMER_DATA_PENDING:
53632229 918 {
c8f59ee5 919 int32_t ret;
6d805429 920 uint64_t id = msg.u.data_pending.session_id;
c8f59ee5 921
6d805429 922 DBG("Kernel consumer data pending command for id %" PRIu64, id);
c8f59ee5 923
6d805429 924 ret = consumer_data_pending(id);
c8f59ee5 925
9ce5646a
MD
926 health_code_update();
927
c8f59ee5
DG
928 /* Send back returned value to session daemon */
929 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
930 if (ret < 0) {
6d805429 931 PERROR("send data pending ret code");
1803a064 932 goto error_fatal;
c8f59ee5 933 }
f50f23d9
DG
934
935 /*
936 * No need to send back a status message since the data pending
937 * returned value is the response.
938 */
c8f59ee5 939 break;
53632229 940 }
6dc3064a
DG
941 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
942 {
b0226bd4
MD
943 struct lttng_consumer_channel *channel;
944 uint64_t key = msg.u.snapshot_channel.key;
945
946 channel = consumer_find_channel(key);
947 if (!channel) {
948 ERR("Channel %" PRIu64 " not found", key);
949 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
07b86b52 950 } else {
e5148e25 951 pthread_mutex_lock(&channel->lock);
b0226bd4
MD
952 if (msg.u.snapshot_channel.metadata == 1) {
953 ret = lttng_kconsumer_snapshot_metadata(channel, key,
954 msg.u.snapshot_channel.pathname,
955 msg.u.snapshot_channel.relayd_id, ctx);
956 if (ret < 0) {
957 ERR("Snapshot metadata failed");
958 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
959 }
960 } else {
961 ret = lttng_kconsumer_snapshot_channel(channel, key,
962 msg.u.snapshot_channel.pathname,
963 msg.u.snapshot_channel.relayd_id,
964 msg.u.snapshot_channel.nb_packets_per_stream,
965 ctx);
966 if (ret < 0) {
967 ERR("Snapshot channel failed");
968 ret_code = LTTCOMM_CONSUMERD_SNAPSHOT_FAILED;
969 }
07b86b52 970 }
e5148e25 971 pthread_mutex_unlock(&channel->lock);
07b86b52 972 }
9ce5646a
MD
973 health_code_update();
974
6dc3064a
DG
975 ret = consumer_send_status_msg(sock, ret_code);
976 if (ret < 0) {
977 /* Somehow, the session daemon is not responding anymore. */
978 goto end_nosignal;
979 }
980 break;
981 }
07b86b52
JD
982 case LTTNG_CONSUMER_DESTROY_CHANNEL:
983 {
984 uint64_t key = msg.u.destroy_channel.key;
985 struct lttng_consumer_channel *channel;
986
987 channel = consumer_find_channel(key);
988 if (!channel) {
989 ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
e462382a 990 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
07b86b52
JD
991 }
992
9ce5646a
MD
993 health_code_update();
994
07b86b52
JD
995 ret = consumer_send_status_msg(sock, ret_code);
996 if (ret < 0) {
997 /* Somehow, the session daemon is not responding anymore. */
09d3dbc0 998 goto end_destroy_channel;
07b86b52
JD
999 }
1000
9ce5646a
MD
1001 health_code_update();
1002
15dc512a
DG
1003 /* Stop right now if no channel was found. */
1004 if (!channel) {
09d3dbc0 1005 goto end_destroy_channel;
15dc512a
DG
1006 }
1007
07b86b52
JD
1008 /*
1009 * This command should ONLY be issued for channel with streams set in
1010 * no monitor mode.
1011 */
1012 assert(!channel->monitor);
1013
1014 /*
1015 * The refcount should ALWAYS be 0 in the case of a channel in no
1016 * monitor mode.
1017 */
1018 assert(!uatomic_sub_return(&channel->refcount, 1));
1019
1020 consumer_del_channel(channel);
09d3dbc0 1021end_destroy_channel:
07b86b52
JD
1022 goto end_nosignal;
1023 }
fb83fe64
JD
1024 case LTTNG_CONSUMER_DISCARDED_EVENTS:
1025 {
66ab32be
JD
1026 ssize_t ret;
1027 uint64_t count;
fb83fe64
JD
1028 struct lttng_consumer_channel *channel;
1029 uint64_t id = msg.u.discarded_events.session_id;
1030 uint64_t key = msg.u.discarded_events.channel_key;
1031
e5742757
MD
1032 DBG("Kernel consumer discarded events command for session id %"
1033 PRIu64 ", channel key %" PRIu64, id, key);
1034
fb83fe64
JD
1035 channel = consumer_find_channel(key);
1036 if (!channel) {
1037 ERR("Kernel consumer discarded events channel %"
1038 PRIu64 " not found", key);
66ab32be 1039 count = 0;
e5742757 1040 } else {
66ab32be 1041 count = channel->discarded_events;
fb83fe64
JD
1042 }
1043
fb83fe64
JD
1044 health_code_update();
1045
1046 /* Send back returned value to session daemon */
66ab32be 1047 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
fb83fe64
JD
1048 if (ret < 0) {
1049 PERROR("send discarded events");
1050 goto error_fatal;
1051 }
1052
1053 break;
1054 }
1055 case LTTNG_CONSUMER_LOST_PACKETS:
1056 {
66ab32be
JD
1057 ssize_t ret;
1058 uint64_t count;
fb83fe64
JD
1059 struct lttng_consumer_channel *channel;
1060 uint64_t id = msg.u.lost_packets.session_id;
1061 uint64_t key = msg.u.lost_packets.channel_key;
1062
e5742757
MD
1063 DBG("Kernel consumer lost packets command for session id %"
1064 PRIu64 ", channel key %" PRIu64, id, key);
1065
fb83fe64
JD
1066 channel = consumer_find_channel(key);
1067 if (!channel) {
1068 ERR("Kernel consumer lost packets channel %"
1069 PRIu64 " not found", key);
66ab32be 1070 count = 0;
e5742757 1071 } else {
66ab32be 1072 count = channel->lost_packets;
fb83fe64
JD
1073 }
1074
fb83fe64
JD
1075 health_code_update();
1076
1077 /* Send back returned value to session daemon */
66ab32be 1078 ret = lttcomm_send_unix_sock(sock, &count, sizeof(count));
fb83fe64
JD
1079 if (ret < 0) {
1080 PERROR("send lost packets");
1081 goto error_fatal;
1082 }
1083
1084 break;
1085 }
b3530820
JG
1086 case LTTNG_CONSUMER_SET_CHANNEL_MONITOR_PIPE:
1087 {
1088 int channel_monitor_pipe;
1089
1090 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1091 /* Successfully received the command's type. */
1092 ret = consumer_send_status_msg(sock, ret_code);
1093 if (ret < 0) {
1094 goto error_fatal;
1095 }
1096
1097 ret = lttcomm_recv_fds_unix_sock(sock, &channel_monitor_pipe,
1098 1);
1099 if (ret != sizeof(channel_monitor_pipe)) {
1100 ERR("Failed to receive channel monitor pipe");
1101 goto error_fatal;
1102 }
1103
1104 DBG("Received channel monitor pipe (%d)", channel_monitor_pipe);
1105 ret = consumer_timer_thread_set_channel_monitor_pipe(
1106 channel_monitor_pipe);
1107 if (!ret) {
1108 int flags;
1109
1110 ret_code = LTTCOMM_CONSUMERD_SUCCESS;
1111 /* Set the pipe as non-blocking. */
1112 ret = fcntl(channel_monitor_pipe, F_GETFL, 0);
1113 if (ret == -1) {
1114 PERROR("fcntl get flags of the channel monitoring pipe");
1115 goto error_fatal;
1116 }
1117 flags = ret;
1118
1119 ret = fcntl(channel_monitor_pipe, F_SETFL,
1120 flags | O_NONBLOCK);
1121 if (ret == -1) {
1122 PERROR("fcntl set O_NONBLOCK flag of the channel monitoring pipe");
1123 goto error_fatal;
1124 }
1125 DBG("Channel monitor pipe set as non-blocking");
1126 } else {
1127 ret_code = LTTCOMM_CONSUMERD_ALREADY_SET;
1128 }
1129 ret = consumer_send_status_msg(sock, ret_code);
1130 if (ret < 0) {
1131 goto error_fatal;
1132 }
1133 break;
1134 }
b99a8d42
JD
1135 case LTTNG_CONSUMER_ROTATE_CHANNEL:
1136 {
e96d66b4
MD
1137 struct lttng_consumer_channel *channel;
1138 uint64_t key = msg.u.rotate_channel.key;
b99a8d42 1139
e96d66b4 1140 DBG("Consumer rotate channel %" PRIu64, key);
b99a8d42 1141
e96d66b4
MD
1142 channel = consumer_find_channel(key);
1143 if (!channel) {
1144 ERR("Channel %" PRIu64 " not found", key);
1145 ret_code = LTTCOMM_CONSUMERD_CHAN_NOT_FOUND;
1146 } else {
1147 /*
1148 * Sample the rotate position of all the streams in this channel.
1149 */
1150 ret = lttng_consumer_rotate_channel(channel, key,
e96d66b4
MD
1151 msg.u.rotate_channel.relayd_id,
1152 msg.u.rotate_channel.metadata,
e96d66b4
MD
1153 ctx);
1154 if (ret < 0) {
1155 ERR("Rotate channel failed");
1156 ret_code = LTTCOMM_CONSUMERD_ROTATION_FAIL;
1157 }
b99a8d42 1158
e96d66b4
MD
1159 health_code_update();
1160 }
b99a8d42
JD
1161 ret = consumer_send_status_msg(sock, ret_code);
1162 if (ret < 0) {
1163 /* Somehow, the session daemon is not responding anymore. */
812e4d26 1164 goto error_rotate_channel;
b99a8d42 1165 }
e96d66b4
MD
1166 if (channel) {
1167 /* Rotate the streams that are ready right now. */
1168 ret = lttng_consumer_rotate_ready_streams(
1169 channel, key, ctx);
1170 if (ret < 0) {
1171 ERR("Rotate ready streams failed");
1172 }
b99a8d42 1173 }
b99a8d42 1174 break;
812e4d26
JG
1175error_rotate_channel:
1176 goto end_nosignal;
b99a8d42 1177 }
e5148e25 1178 case LTTNG_CONSUMER_INIT:
00fb02ac 1179 {
e5148e25
JG
1180 ret_code = lttng_consumer_init_command(ctx,
1181 msg.u.init.sessiond_uuid);
00fb02ac 1182 health_code_update();
00fb02ac
JD
1183 ret = consumer_send_status_msg(sock, ret_code);
1184 if (ret < 0) {
1185 /* Somehow, the session daemon is not responding anymore. */
1186 goto end_nosignal;
1187 }
1188 break;
1189 }
e5148e25 1190 case LTTNG_CONSUMER_CREATE_TRACE_CHUNK:
d88744a4 1191 {
e5148e25 1192 const struct lttng_credentials credentials = {
0ebdafe0
JG
1193 .uid = msg.u.create_trace_chunk.credentials.value.uid,
1194 .gid = msg.u.create_trace_chunk.credentials.value.gid,
e5148e25
JG
1195 };
1196 const bool is_local_trace =
1197 !msg.u.create_trace_chunk.relayd_id.is_set;
1198 const uint64_t relayd_id =
1199 msg.u.create_trace_chunk.relayd_id.value;
1200 const char *chunk_override_name =
1201 *msg.u.create_trace_chunk.override_name ?
1202 msg.u.create_trace_chunk.override_name :
1203 NULL;
1204 LTTNG_OPTIONAL(struct lttng_directory_handle) chunk_directory_handle =
1205 LTTNG_OPTIONAL_INIT;
d88744a4 1206
e5148e25
JG
1207 /*
1208 * The session daemon will only provide a chunk directory file
1209 * descriptor for local traces.
1210 */
1211 if (is_local_trace) {
1212 int chunk_dirfd;
19990ed5 1213
e5148e25
JG
1214 /* Acnowledge the reception of the command. */
1215 ret = consumer_send_status_msg(sock,
1216 LTTCOMM_CONSUMERD_SUCCESS);
1217 if (ret < 0) {
1218 /* Somehow, the session daemon is not responding anymore. */
1219 goto end_nosignal;
1220 }
82528808 1221
e5148e25
JG
1222 ret = lttcomm_recv_fds_unix_sock(sock, &chunk_dirfd, 1);
1223 if (ret != sizeof(chunk_dirfd)) {
1224 ERR("Failed to receive trace chunk directory file descriptor");
1225 goto error_fatal;
1226 }
82528808 1227
e5148e25
JG
1228 DBG("Received trace chunk directory fd (%d)",
1229 chunk_dirfd);
1230 ret = lttng_directory_handle_init_from_dirfd(
1231 &chunk_directory_handle.value,
1232 chunk_dirfd);
1233 if (ret) {
1234 ERR("Failed to initialize chunk directory handle from directory file descriptor");
1235 if (close(chunk_dirfd)) {
1236 PERROR("Failed to close chunk directory file descriptor");
1237 }
1238 goto error_fatal;
1239 }
1240 chunk_directory_handle.is_set = true;
82528808
JG
1241 }
1242
e5148e25
JG
1243 ret_code = lttng_consumer_create_trace_chunk(
1244 !is_local_trace ? &relayd_id : NULL,
1245 msg.u.create_trace_chunk.session_id,
1246 msg.u.create_trace_chunk.chunk_id,
0ebdafe0
JG
1247 (time_t) msg.u.create_trace_chunk
1248 .creation_timestamp,
e5148e25 1249 chunk_override_name,
0ebdafe0
JG
1250 msg.u.create_trace_chunk.credentials.is_set ?
1251 &credentials :
1252 NULL,
e5148e25
JG
1253 chunk_directory_handle.is_set ?
1254 &chunk_directory_handle.value :
1255 NULL);
82528808 1256
e5148e25
JG
1257 if (chunk_directory_handle.is_set) {
1258 lttng_directory_handle_fini(
1259 &chunk_directory_handle.value);
d88744a4 1260 }
e5148e25 1261 goto end_msg_sessiond;
d88744a4 1262 }
e5148e25 1263 case LTTNG_CONSUMER_CLOSE_TRACE_CHUNK:
a1ae2ea5 1264 {
6bbcff33
JG
1265 enum lttng_trace_chunk_command_type close_command =
1266 msg.u.close_trace_chunk.close_command.value;
e5148e25
JG
1267 const uint64_t relayd_id =
1268 msg.u.close_trace_chunk.relayd_id.value;
41b23598
MD
1269 struct lttcomm_consumer_close_trace_chunk_reply reply;
1270 char path[LTTNG_PATH_MAX];
e5148e25
JG
1271
1272 ret_code = lttng_consumer_close_trace_chunk(
1273 msg.u.close_trace_chunk.relayd_id.is_set ?
6bbcff33
JG
1274 &relayd_id :
1275 NULL,
e5148e25
JG
1276 msg.u.close_trace_chunk.session_id,
1277 msg.u.close_trace_chunk.chunk_id,
6bbcff33
JG
1278 (time_t) msg.u.close_trace_chunk.close_timestamp,
1279 msg.u.close_trace_chunk.close_command.is_set ?
1280 &close_command :
41b23598
MD
1281 NULL, path);
1282 reply.ret_code = ret_code;
1283 reply.path_length = strlen(path) + 1;
1284 ret = lttcomm_send_unix_sock(sock, &reply, sizeof(reply));
1285 if (ret != sizeof(reply)) {
1286 goto error_fatal;
1287 }
1288 ret = lttcomm_send_unix_sock(sock, path, reply.path_length);
1289 if (ret != reply.path_length) {
1290 goto error_fatal;
1291 }
1292 goto end_nosignal;
fc181d72 1293 }
e5148e25 1294 case LTTNG_CONSUMER_TRACE_CHUNK_EXISTS:
fc181d72 1295 {
e5148e25
JG
1296 const uint64_t relayd_id =
1297 msg.u.trace_chunk_exists.relayd_id.value;
1298
1299 ret_code = lttng_consumer_trace_chunk_exists(
1300 msg.u.trace_chunk_exists.relayd_id.is_set ?
1301 &relayd_id : NULL,
1302 msg.u.trace_chunk_exists.session_id,
1303 msg.u.trace_chunk_exists.chunk_id);
1304 goto end_msg_sessiond;
a1ae2ea5 1305 }
3bd1e081 1306 default:
3f8e211f 1307 goto end_nosignal;
3bd1e081 1308 }
3f8e211f 1309
3bd1e081 1310end_nosignal:
4cbc1a04
DG
1311 /*
1312 * Return 1 to indicate success since the 0 value can be a socket
1313 * shutdown during the recv() or send() call.
1314 */
e0e7757d
JG
1315 ret = 1;
1316 goto end;
1317error_fatal:
1318 /* This will issue a consumer stop. */
1319 ret = -1;
1320 goto end;
e5148e25
JG
1321end_msg_sessiond:
1322 /*
1323 * The returned value here is not useful since either way we'll return 1 to
1324 * the caller because the session daemon socket management is done
1325 * elsewhere. Returning a negative code or 0 will shutdown the consumer.
1326 */
1327 ret = consumer_send_status_msg(sock, ret_code);
1328 if (ret < 0) {
1329 goto error_fatal;
1330 }
e0e7757d
JG
1331 ret = 1;
1332end:
e5148e25 1333 health_code_update();
1803a064 1334 rcu_read_unlock();
e0e7757d 1335 return ret;
3bd1e081 1336}
d41f73b7 1337
94d49140
JD
1338/*
1339 * Sync metadata meaning request them to the session daemon and snapshot to the
1340 * metadata thread can consumer them.
1341 *
1342 * Metadata stream lock MUST be acquired.
1343 *
1344 * Return 0 if new metadatda is available, EAGAIN if the metadata stream
1345 * is empty or a negative value on error.
1346 */
1347int lttng_kconsumer_sync_metadata(struct lttng_consumer_stream *metadata)
1348{
1349 int ret;
1350
1351 assert(metadata);
1352
1353 ret = kernctl_buffer_flush(metadata->wait_fd);
1354 if (ret < 0) {
1355 ERR("Failed to flush kernel stream");
1356 goto end;
1357 }
1358
1359 ret = kernctl_snapshot(metadata->wait_fd);
1360 if (ret < 0) {
32af2c95 1361 if (ret != -EAGAIN) {
94d49140
JD
1362 ERR("Sync metadata, taking kernel snapshot failed.");
1363 goto end;
1364 }
1365 DBG("Sync metadata, no new kernel metadata");
1366 /* No new metadata, exit. */
1367 ret = ENODATA;
1368 goto end;
1369 }
1370
1371end:
1372 return ret;
1373}
309167d2 1374
fb83fe64 1375static
6f1177cf
JG
1376int extract_common_subbuffer_info(struct lttng_consumer_stream *stream,
1377 struct stream_subbuffer *subbuf)
fb83fe64
JD
1378{
1379 int ret;
fb83fe64 1380
6f1177cf
JG
1381 ret = kernctl_get_subbuf_size(
1382 stream->wait_fd, &subbuf->info.data.subbuf_size);
1383 if (ret) {
fb83fe64
JD
1384 goto end;
1385 }
fb83fe64 1386
6f1177cf
JG
1387 ret = kernctl_get_padded_subbuf_size(
1388 stream->wait_fd, &subbuf->info.data.padded_subbuf_size);
1389 if (ret) {
fb83fe64
JD
1390 goto end;
1391 }
fb83fe64
JD
1392
1393end:
1394 return ret;
1395}
1396
93ec662e 1397static
6f1177cf
JG
1398int extract_metadata_subbuffer_info(struct lttng_consumer_stream *stream,
1399 struct stream_subbuffer *subbuf)
93ec662e
JD
1400{
1401 int ret;
93ec662e 1402
6f1177cf
JG
1403 ret = extract_common_subbuffer_info(stream, subbuf);
1404 if (ret) {
93ec662e
JD
1405 goto end;
1406 }
1407
6f1177cf
JG
1408 ret = kernctl_get_metadata_version(
1409 stream->wait_fd, &subbuf->info.metadata.version);
1410 if (ret) {
93ec662e
JD
1411 goto end;
1412 }
1413
93ec662e
JD
1414end:
1415 return ret;
1416}
1417
6f1177cf
JG
1418static
1419int extract_data_subbuffer_info(struct lttng_consumer_stream *stream,
1420 struct stream_subbuffer *subbuf)
d41f73b7 1421{
6f1177cf 1422 int ret;
d41f73b7 1423
6f1177cf
JG
1424 ret = extract_common_subbuffer_info(stream, subbuf);
1425 if (ret) {
1426 goto end;
1427 }
309167d2 1428
6f1177cf
JG
1429 ret = kernctl_get_packet_size(
1430 stream->wait_fd, &subbuf->info.data.packet_size);
1431 if (ret < 0) {
1432 PERROR("Failed to get sub-buffer packet size");
1433 goto end;
1434 }
02d02e31 1435
6f1177cf
JG
1436 ret = kernctl_get_content_size(
1437 stream->wait_fd, &subbuf->info.data.content_size);
1438 if (ret < 0) {
1439 PERROR("Failed to get sub-buffer content size");
1440 goto end;
d41f73b7
MD
1441 }
1442
6f1177cf
JG
1443 ret = kernctl_get_timestamp_begin(
1444 stream->wait_fd, &subbuf->info.data.timestamp_begin);
1445 if (ret < 0) {
1446 PERROR("Failed to get sub-buffer begin timestamp");
1447 goto end;
1d4dfdef
DG
1448 }
1449
6f1177cf
JG
1450 ret = kernctl_get_timestamp_end(
1451 stream->wait_fd, &subbuf->info.data.timestamp_end);
1452 if (ret < 0) {
1453 PERROR("Failed to get sub-buffer end timestamp");
1454 goto end;
1455 }
1456
1457 ret = kernctl_get_events_discarded(
1458 stream->wait_fd, &subbuf->info.data.events_discarded);
1459 if (ret) {
1460 PERROR("Failed to get sub-buffer events discarded count");
1461 goto end;
1462 }
1463
1464 ret = kernctl_get_sequence_number(stream->wait_fd,
1465 &subbuf->info.data.sequence_number.value);
1466 if (ret) {
1467 /* May not be supported by older LTTng-modules. */
1468 if (ret != -ENOTTY) {
1469 PERROR("Failed to get sub-buffer sequence number");
1470 goto end;
fb83fe64 1471 }
1c20f0e2 1472 } else {
6f1177cf 1473 subbuf->info.data.sequence_number.is_set = true;
309167d2
JD
1474 }
1475
6f1177cf
JG
1476 ret = kernctl_get_stream_id(
1477 stream->wait_fd, &subbuf->info.data.stream_id);
1478 if (ret < 0) {
1479 PERROR("Failed to get stream id");
1480 goto end;
1481 }
1d4dfdef 1482
6f1177cf
JG
1483 ret = kernctl_get_instance_id(stream->wait_fd,
1484 &subbuf->info.data.stream_instance_id.value);
1485 if (ret) {
1486 /* May not be supported by older LTTng-modules. */
1487 if (ret != -ENOTTY) {
1488 PERROR("Failed to get stream instance id");
1489 goto end;
1d4dfdef 1490 }
6f1177cf
JG
1491 } else {
1492 subbuf->info.data.stream_instance_id.is_set = true;
1493 }
1494end:
1495 return ret;
1496}
47e81c02 1497
6f1177cf
JG
1498static
1499int get_subbuffer_common(struct lttng_consumer_stream *stream,
1500 struct stream_subbuffer *subbuffer)
1501{
1502 int ret;
1503
1504 ret = kernctl_get_next_subbuf(stream->wait_fd);
1505 if (ret) {
1506 goto end;
1507 }
1508
1509 ret = stream->read_subbuffer_ops.extract_subbuffer_info(
1510 stream, subbuffer);
1511end:
1512 return ret;
1513}
7775df52 1514
6f1177cf
JG
1515static
1516int get_next_subbuffer_splice(struct lttng_consumer_stream *stream,
1517 struct stream_subbuffer *subbuffer)
1518{
1519 int ret;
1d4dfdef 1520
6f1177cf
JG
1521 ret = get_subbuffer_common(stream, subbuffer);
1522 if (ret) {
1523 goto end;
1524 }
1d4dfdef 1525
6f1177cf
JG
1526 subbuffer->buffer.fd = stream->wait_fd;
1527end:
1528 return ret;
1529}
b770aa7f 1530
6f1177cf
JG
1531static
1532int get_next_subbuffer_mmap(struct lttng_consumer_stream *stream,
1533 struct stream_subbuffer *subbuffer)
1534{
1535 int ret;
1536 const char *addr;
1537
1538 ret = get_subbuffer_common(stream, subbuffer);
1539 if (ret) {
1540 goto end;
7775df52 1541 }
6f1177cf
JG
1542
1543 ret = get_current_subbuf_addr(stream, &addr);
1544 if (ret) {
1545 goto end;
d41f73b7 1546 }
6f1177cf
JG
1547
1548 subbuffer->buffer.buffer = lttng_buffer_view_init(
1549 addr, 0, subbuffer->info.data.padded_subbuf_size);
1550end:
1551 return ret;
1552}
1553
1554static
1555int put_next_subbuffer(struct lttng_consumer_stream *stream,
1556 struct stream_subbuffer *subbuffer)
1557{
1558 const int ret = kernctl_put_next_subbuf(stream->wait_fd);
1559
1560 if (ret) {
1561 if (ret == -EFAULT) {
1562 PERROR("Error in unreserving sub buffer");
1563 } else if (ret == -EIO) {
d41f73b7 1564 /* Should never happen with newer LTTng versions */
6f1177cf 1565 PERROR("Reader has been pushed by the writer, last sub-buffer corrupted");
d41f73b7 1566 }
d41f73b7
MD
1567 }
1568
6f1177cf
JG
1569 return ret;
1570}
1c20f0e2 1571
6f1177cf
JG
1572static void lttng_kconsumer_set_stream_ops(
1573 struct lttng_consumer_stream *stream)
1574{
1575 if (stream->chan->output == CONSUMER_CHANNEL_MMAP) {
1576 stream->read_subbuffer_ops.get_next_subbuffer =
1577 get_next_subbuffer_mmap;
1578 } else {
1579 stream->read_subbuffer_ops.get_next_subbuffer =
1580 get_next_subbuffer_splice;
94d49140
JD
1581 }
1582
6f1177cf
JG
1583 if (stream->metadata_flag) {
1584 stream->read_subbuffer_ops.extract_subbuffer_info =
1585 extract_metadata_subbuffer_info;
1586 } else {
1587 stream->read_subbuffer_ops.extract_subbuffer_info =
1588 extract_data_subbuffer_info;
1589 if (stream->chan->is_live) {
1590 stream->read_subbuffer_ops.send_live_beacon =
1591 consumer_flush_kernel_index;
1592 }
309167d2
JD
1593 }
1594
6f1177cf 1595 stream->read_subbuffer_ops.put_next_subbuffer = put_next_subbuffer;
d41f73b7
MD
1596}
1597
1598int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
1599{
1600 int ret;
ffe60014
DG
1601
1602 assert(stream);
1603
2bba9e53 1604 /*
e5148e25
JG
1605 * Don't create anything if this is set for streaming or if there is
1606 * no current trace chunk on the parent channel.
2bba9e53 1607 */
e5148e25
JG
1608 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor &&
1609 stream->chan->trace_chunk) {
1610 ret = consumer_stream_create_output_files(stream, true);
1611 if (ret) {
fe4477ee
JD
1612 goto error;
1613 }
ffe60014 1614 }
d41f73b7 1615
d41f73b7
MD
1616 if (stream->output == LTTNG_EVENT_MMAP) {
1617 /* get the len of the mmap region */
1618 unsigned long mmap_len;
1619
1620 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
1621 if (ret != 0) {
ffe60014 1622 PERROR("kernctl_get_mmap_len");
d41f73b7
MD
1623 goto error_close_fd;
1624 }
1625 stream->mmap_len = (size_t) mmap_len;
1626
ffe60014
DG
1627 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1628 MAP_PRIVATE, stream->wait_fd, 0);
d41f73b7 1629 if (stream->mmap_base == MAP_FAILED) {
ffe60014 1630 PERROR("Error mmaping");
d41f73b7
MD
1631 ret = -1;
1632 goto error_close_fd;
1633 }
1634 }
1635
6f1177cf
JG
1636 lttng_kconsumer_set_stream_ops(stream);
1637
d41f73b7
MD
1638 /* we return 0 to let the library handle the FD internally */
1639 return 0;
1640
1641error_close_fd:
2f225ce2 1642 if (stream->out_fd >= 0) {
d41f73b7
MD
1643 int err;
1644
1645 err = close(stream->out_fd);
1646 assert(!err);
2f225ce2 1647 stream->out_fd = -1;
d41f73b7
MD
1648 }
1649error:
1650 return ret;
1651}
1652
ca22feea
DG
1653/*
1654 * Check if data is still being extracted from the buffers for a specific
4e9a4686
DG
1655 * stream. Consumer data lock MUST be acquired before calling this function
1656 * and the stream lock.
ca22feea 1657 *
6d805429 1658 * Return 1 if the traced data are still getting read else 0 meaning that the
ca22feea
DG
1659 * data is available for trace viewer reading.
1660 */
6d805429 1661int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
ca22feea
DG
1662{
1663 int ret;
1664
1665 assert(stream);
1666
873b9e9a
MD
1667 if (stream->endpoint_status != CONSUMER_ENDPOINT_ACTIVE) {
1668 ret = 0;
1669 goto end;
1670 }
1671
ca22feea
DG
1672 ret = kernctl_get_next_subbuf(stream->wait_fd);
1673 if (ret == 0) {
1674 /* There is still data so let's put back this subbuffer. */
1675 ret = kernctl_put_subbuf(stream->wait_fd);
1676 assert(ret == 0);
6d805429 1677 ret = 1; /* Data is pending */
4e9a4686 1678 goto end;
ca22feea
DG
1679 }
1680
6d805429
DG
1681 /* Data is NOT pending and ready to be read. */
1682 ret = 0;
ca22feea 1683
6efae65e
DG
1684end:
1685 return ret;
ca22feea 1686}
This page took 0.15715 seconds and 4 git commands to generate.