Fix: increment channel refcount on add_stream
[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>
4 *
d14d33bf
AM
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License, version 2 only,
7 * as published by the Free Software Foundation.
3bd1e081
MD
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
d14d33bf
AM
14 * You should have received a copy of the GNU General Public License along
15 * with this program; if not, write to the Free Software Foundation, Inc.,
16 * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
3bd1e081
MD
17 */
18
19#define _GNU_SOURCE
20#include <assert.h>
3bd1e081
MD
21#include <poll.h>
22#include <pthread.h>
23#include <stdlib.h>
24#include <string.h>
25#include <sys/mman.h>
26#include <sys/socket.h>
27#include <sys/types.h>
77c7c900 28#include <inttypes.h>
3bd1e081 29#include <unistd.h>
dbb5dfe6 30#include <sys/stat.h>
3bd1e081 31
990570ed 32#include <common/common.h>
10a8a223 33#include <common/kernel-ctl/kernel-ctl.h>
10a8a223 34#include <common/sessiond-comm/sessiond-comm.h>
00e2e675 35#include <common/sessiond-comm/relayd.h>
dbb5dfe6 36#include <common/compat/fcntl.h>
acdb9057 37#include <common/pipe.h>
00e2e675 38#include <common/relayd/relayd.h>
fe4477ee 39#include <common/utils.h>
0857097f 40
10a8a223 41#include "kernel-consumer.h"
3bd1e081
MD
42
43extern struct lttng_consumer_global_data consumer_data;
44extern int consumer_poll_timeout;
45extern volatile int consumer_quit;
46
3bd1e081
MD
47/*
48 * Take a snapshot for a specific fd
49 *
50 * Returns 0 on success, < 0 on error
51 */
ffe60014 52int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
3bd1e081
MD
53{
54 int ret = 0;
55 int infd = stream->wait_fd;
56
57 ret = kernctl_snapshot(infd);
58 if (ret != 0) {
87dc6a9c 59 errno = -ret;
3bd1e081
MD
60 perror("Getting sub-buffer snapshot.");
61 }
62
63 return ret;
64}
65
66/*
67 * Get the produced position
68 *
69 * Returns 0 on success, < 0 on error
70 */
ffe60014 71int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
3bd1e081
MD
72 unsigned long *pos)
73{
74 int ret;
75 int infd = stream->wait_fd;
76
77 ret = kernctl_snapshot_get_produced(infd, pos);
78 if (ret != 0) {
87dc6a9c 79 errno = -ret;
3bd1e081
MD
80 perror("kernctl_snapshot_get_produced");
81 }
82
83 return ret;
84}
85
86int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
87 int sock, struct pollfd *consumer_sockpoll)
88{
89 ssize_t ret;
f50f23d9 90 enum lttng_error_code ret_code = LTTNG_OK;
3bd1e081
MD
91 struct lttcomm_consumer_msg msg;
92
93 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
94 if (ret != sizeof(msg)) {
f73fabfd 95 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
3bd1e081
MD
96 return ret;
97 }
98 if (msg.cmd_type == LTTNG_CONSUMER_STOP) {
f50f23d9
DG
99 /*
100 * Notify the session daemon that the command is completed.
101 *
102 * On transport layer error, the function call will print an error
103 * message so handling the returned code is a bit useless since we
104 * return an error code anyway.
105 */
106 (void) consumer_send_status_msg(sock, ret_code);
3bd1e081
MD
107 return -ENOENT;
108 }
109
b0b335c8
MD
110 /* relayd needs RCU read-side protection */
111 rcu_read_lock();
112
3bd1e081 113 switch (msg.cmd_type) {
00e2e675
DG
114 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
115 {
f50f23d9 116 /* Session daemon status message are handled in the following call. */
7735ef9e
DG
117 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
118 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
46e6455f 119 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id);
00e2e675
DG
120 goto end_nosignal;
121 }
3bd1e081
MD
122 case LTTNG_CONSUMER_ADD_CHANNEL:
123 {
124 struct lttng_consumer_channel *new_channel;
e43c41c5 125 int ret_recv;
3bd1e081 126
f50f23d9
DG
127 /* First send a status message before receiving the fds. */
128 ret = consumer_send_status_msg(sock, ret_code);
129 if (ret < 0) {
130 /* Somehow, the session daemon is not responding anymore. */
131 goto end_nosignal;
132 }
d88aee68 133 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
3bd1e081 134 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
ffe60014
DG
135 msg.u.channel.session_id, msg.u.channel.pathname,
136 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
1624d5b7
JD
137 msg.u.channel.relayd_id, msg.u.channel.output,
138 msg.u.channel.tracefile_size,
139 msg.u.channel.tracefile_count);
3bd1e081 140 if (new_channel == NULL) {
f73fabfd 141 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
3bd1e081
MD
142 goto end_nosignal;
143 }
ffe60014
DG
144 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
145
146 /* Translate and save channel type. */
147 switch (msg.u.channel.type) {
148 case CONSUMER_CHANNEL_TYPE_DATA:
149 case CONSUMER_CHANNEL_TYPE_METADATA:
150 new_channel->type = msg.u.channel.type;
151 break;
152 default:
153 assert(0);
154 goto end_nosignal;
155 };
156
3bd1e081 157 if (ctx->on_recv_channel != NULL) {
e43c41c5
JD
158 ret_recv = ctx->on_recv_channel(new_channel);
159 if (ret_recv == 0) {
160 ret = consumer_add_channel(new_channel, ctx);
161 } else if (ret_recv < 0) {
3bd1e081
MD
162 goto end_nosignal;
163 }
164 } else {
e43c41c5 165 ret = consumer_add_channel(new_channel, ctx);
3bd1e081 166 }
e43c41c5
JD
167
168 /* If we received an error in add_channel, we need to report it. */
169 if (ret != 0) {
170 consumer_send_status_msg(sock, ret);
171 goto end_nosignal;
172 }
173
3bd1e081
MD
174 goto end_nosignal;
175 }
176 case LTTNG_CONSUMER_ADD_STREAM:
177 {
dae10966
DG
178 int fd;
179 struct lttng_pipe *stream_pipe;
00e2e675
DG
180 struct consumer_relayd_sock_pair *relayd = NULL;
181 struct lttng_consumer_stream *new_stream;
ffe60014 182 struct lttng_consumer_channel *channel;
c80048c6 183 int alloc_ret = 0;
3bd1e081 184
ffe60014
DG
185 /*
186 * Get stream's channel reference. Needed when adding the stream to the
187 * global hash table.
188 */
189 channel = consumer_find_channel(msg.u.stream.channel_key);
190 if (!channel) {
191 /*
192 * We could not find the channel. Can happen if cpu hotplug
193 * happens while tearing down.
194 */
d88aee68 195 ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
ffe60014
DG
196 ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
197 }
198
f50f23d9
DG
199 /* First send a status message before receiving the fds. */
200 ret = consumer_send_status_msg(sock, ret_code);
ffe60014
DG
201 if (ret < 0 || ret_code != LTTNG_OK) {
202 /*
203 * Somehow, the session daemon is not responding anymore or the
204 * channel was not found.
205 */
f50f23d9
DG
206 goto end_nosignal;
207 }
208
3bd1e081
MD
209 /* block */
210 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
3f8e211f 211 rcu_read_unlock();
3bd1e081
MD
212 return -EINTR;
213 }
00e2e675
DG
214
215 /* Get stream file descriptor from socket */
f2fc6720
MD
216 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
217 if (ret != sizeof(fd)) {
f73fabfd 218 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
3f8e211f 219 rcu_read_unlock();
3bd1e081
MD
220 return ret;
221 }
3bd1e081 222
f50f23d9
DG
223 /*
224 * Send status code to session daemon only if the recv works. If the
225 * above recv() failed, the session daemon is notified through the
226 * error socket and the teardown is eventually done.
227 */
228 ret = consumer_send_status_msg(sock, ret_code);
229 if (ret < 0) {
230 /* Somehow, the session daemon is not responding anymore. */
231 goto end_nosignal;
232 }
233
ffe60014
DG
234 new_stream = consumer_allocate_stream(channel->key,
235 fd,
236 LTTNG_CONSUMER_ACTIVE_STREAM,
237 channel->name,
238 channel->uid,
239 channel->gid,
240 channel->relayd_id,
241 channel->session_id,
242 msg.u.stream.cpu,
243 &alloc_ret,
244 channel->type);
3bd1e081 245 if (new_stream == NULL) {
c80048c6
MD
246 switch (alloc_ret) {
247 case -ENOMEM:
248 case -EINVAL:
249 default:
250 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
251 break;
c80048c6 252 }
3f8e211f 253 goto end_nosignal;
3bd1e081 254 }
ffe60014
DG
255 new_stream->chan = channel;
256 new_stream->wait_fd = fd;
00e2e675 257
9d9353f9
JD
258 /* Metadata chan refcount is increment in add_metadata_stream */
259 if (new_stream->chan->type != CONSUMER_CHANNEL_TYPE_METADATA) {
260 /* Update channel refcount */
261 uatomic_inc(&new_stream->chan->refcount);
262 }
263
fb3a43a9
DG
264 /*
265 * The buffer flush is done on the session daemon side for the kernel
266 * so no need for the stream "hangup_flush_done" variable to be
267 * tracked. This is important for a kernel stream since we don't rely
268 * on the flush state of the stream to read data. It's not the case for
269 * user space tracing.
270 */
271 new_stream->hangup_flush_done = 0;
272
00e2e675 273 /* The stream is not metadata. Get relayd reference if exists. */
ffe60014 274 relayd = consumer_find_relayd(new_stream->net_seq_idx);
00e2e675
DG
275 if (relayd != NULL) {
276 /* Add stream on the relayd */
277 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
278 ret = relayd_add_stream(&relayd->control_sock,
ffe60014 279 new_stream->name, new_stream->chan->pathname,
0f907de1
JD
280 &new_stream->relayd_stream_id,
281 new_stream->chan->tracefile_size,
282 new_stream->chan->tracefile_count);
00e2e675
DG
283 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
284 if (ret < 0) {
e316aad5 285 consumer_del_stream(new_stream, NULL);
3f8e211f 286 goto end_nosignal;
00e2e675 287 }
d88aee68
DG
288 } else if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
289 ERR("Network sequence index %" PRIu64 " unknown. Not adding stream.",
ffe60014 290 new_stream->net_seq_idx);
e316aad5 291 consumer_del_stream(new_stream, NULL);
3f8e211f 292 goto end_nosignal;
00e2e675
DG
293 }
294
633d0084
DG
295 if (ctx->on_recv_stream) {
296 ret = ctx->on_recv_stream(new_stream);
297 if (ret < 0) {
e316aad5 298 consumer_del_stream(new_stream, NULL);
633d0084 299 goto end_nosignal;
fb3a43a9 300 }
633d0084 301 }
fb3a43a9 302
50f8ae69 303 /* Get the right pipe where the stream will be sent. */
633d0084 304 if (new_stream->metadata_flag) {
dae10966 305 stream_pipe = ctx->consumer_metadata_pipe;
3bd1e081 306 } else {
dae10966 307 stream_pipe = ctx->consumer_data_pipe;
50f8ae69
DG
308 }
309
dae10966 310 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
50f8ae69 311 if (ret < 0) {
dae10966 312 ERR("Consumer write %s stream to pipe %d",
50f8ae69 313 new_stream->metadata_flag ? "metadata" : "data",
dae10966 314 lttng_pipe_get_writefd(stream_pipe));
50f8ae69
DG
315 consumer_del_stream(new_stream, NULL);
316 goto end_nosignal;
3bd1e081 317 }
00e2e675 318
50f8ae69 319 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
ffe60014 320 new_stream->name, fd, new_stream->relayd_stream_id);
3bd1e081
MD
321 break;
322 }
323 case LTTNG_CONSUMER_UPDATE_STREAM:
324 {
3f8e211f
DG
325 rcu_read_unlock();
326 return -ENOSYS;
327 }
328 case LTTNG_CONSUMER_DESTROY_RELAYD:
329 {
a6ba4fe1 330 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
3f8e211f
DG
331 struct consumer_relayd_sock_pair *relayd;
332
a6ba4fe1 333 DBG("Kernel consumer destroying relayd %" PRIu64, index);
3f8e211f
DG
334
335 /* Get relayd reference if exists. */
a6ba4fe1 336 relayd = consumer_find_relayd(index);
3f8e211f 337 if (relayd == NULL) {
3448e266 338 DBG("Unable to find relayd %" PRIu64, index);
f50f23d9 339 ret_code = LTTNG_ERR_NO_CONSUMER;
3bd1e081 340 }
3f8e211f 341
a6ba4fe1
DG
342 /*
343 * Each relayd socket pair has a refcount of stream attached to it
344 * which tells if the relayd is still active or not depending on the
345 * refcount value.
346 *
347 * This will set the destroy flag of the relayd object and destroy it
348 * if the refcount reaches zero when called.
349 *
350 * The destroy can happen either here or when a stream fd hangs up.
351 */
f50f23d9
DG
352 if (relayd) {
353 consumer_flag_relayd_for_destroy(relayd);
354 }
355
356 ret = consumer_send_status_msg(sock, ret_code);
357 if (ret < 0) {
358 /* Somehow, the session daemon is not responding anymore. */
359 goto end_nosignal;
360 }
3f8e211f 361
3f8e211f 362 goto end_nosignal;
3bd1e081 363 }
6d805429 364 case LTTNG_CONSUMER_DATA_PENDING:
53632229 365 {
c8f59ee5 366 int32_t ret;
6d805429 367 uint64_t id = msg.u.data_pending.session_id;
c8f59ee5 368
6d805429 369 DBG("Kernel consumer data pending command for id %" PRIu64, id);
c8f59ee5 370
6d805429 371 ret = consumer_data_pending(id);
c8f59ee5
DG
372
373 /* Send back returned value to session daemon */
374 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
375 if (ret < 0) {
6d805429 376 PERROR("send data pending ret code");
c8f59ee5 377 }
f50f23d9
DG
378
379 /*
380 * No need to send back a status message since the data pending
381 * returned value is the response.
382 */
c8f59ee5 383 break;
53632229 384 }
3bd1e081 385 default:
3f8e211f 386 goto end_nosignal;
3bd1e081 387 }
3f8e211f 388
3bd1e081 389end_nosignal:
b0b335c8 390 rcu_read_unlock();
4cbc1a04
DG
391
392 /*
393 * Return 1 to indicate success since the 0 value can be a socket
394 * shutdown during the recv() or send() call.
395 */
396 return 1;
3bd1e081 397}
d41f73b7
MD
398
399/*
400 * Consume data on a file descriptor and write it on a trace file.
401 */
4078b776 402ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
d41f73b7
MD
403 struct lttng_consumer_local_data *ctx)
404{
1d4dfdef 405 unsigned long len, subbuf_size, padding;
d41f73b7 406 int err;
4078b776 407 ssize_t ret = 0;
d41f73b7
MD
408 int infd = stream->wait_fd;
409
410 DBG("In read_subbuffer (infd : %d)", infd);
411 /* Get the next subbuffer */
412 err = kernctl_get_next_subbuf(infd);
413 if (err != 0) {
1d4dfdef 414 ret = err;
d41f73b7
MD
415 /*
416 * This is a debug message even for single-threaded consumer,
417 * because poll() have more relaxed criterions than get subbuf,
418 * so get_subbuf may fail for short race windows where poll()
419 * would issue wakeups.
420 */
421 DBG("Reserving sub buffer failed (everything is normal, "
422 "it is due to concurrency)");
423 goto end;
424 }
425
1d4dfdef
DG
426 /* Get the full subbuffer size including padding */
427 err = kernctl_get_padded_subbuf_size(infd, &len);
428 if (err != 0) {
429 errno = -err;
430 perror("Getting sub-buffer len failed.");
431 ret = err;
432 goto end;
433 }
434
ffe60014 435 switch (stream->chan->output) {
1d4dfdef 436 case LTTNG_EVENT_SPLICE:
d41f73b7 437
1d4dfdef
DG
438 /*
439 * XXX: The lttng-modules splice "actor" does not handle copying
440 * partial pages hence only using the subbuffer size without the
441 * padding makes the splice fail.
442 */
443 subbuf_size = len;
444 padding = 0;
445
446 /* splice the subbuffer to the tracefile */
91dfef6e
DG
447 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
448 padding);
449 /*
450 * XXX: Splice does not support network streaming so the return value
451 * is simply checked against subbuf_size and not like the mmap() op.
452 */
1d4dfdef
DG
453 if (ret != subbuf_size) {
454 /*
455 * display the error but continue processing to try
456 * to release the subbuffer
457 */
458 ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
459 ret, subbuf_size);
460 }
461 break;
462 case LTTNG_EVENT_MMAP:
463 /* Get subbuffer size without padding */
464 err = kernctl_get_subbuf_size(infd, &subbuf_size);
465 if (err != 0) {
466 errno = -err;
467 perror("Getting sub-buffer len failed.");
468 ret = err;
469 goto end;
470 }
47e81c02 471
1d4dfdef
DG
472 /* Make sure the tracer is not gone mad on us! */
473 assert(len >= subbuf_size);
474
475 padding = len - subbuf_size;
476
477 /* write the subbuffer to the tracefile */
91dfef6e
DG
478 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
479 padding);
480 /*
481 * The mmap operation should write subbuf_size amount of data when
482 * network streaming or the full padding (len) size when we are _not_
483 * streaming.
484 */
d88aee68
DG
485 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
486 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
1d4dfdef 487 /*
91dfef6e
DG
488 * Display the error but continue processing to try to release the
489 * subbuffer
1d4dfdef 490 */
91dfef6e
DG
491 ERR("Error writing to tracefile "
492 "(ret: %zd != len: %lu != subbuf_size: %lu)",
493 ret, len, subbuf_size);
1d4dfdef
DG
494 }
495 break;
496 default:
497 ERR("Unknown output method");
498 ret = -1;
d41f73b7
MD
499 }
500
501 err = kernctl_put_next_subbuf(infd);
502 if (err != 0) {
21073eaa 503 errno = -err;
d41f73b7
MD
504 if (errno == EFAULT) {
505 perror("Error in unreserving sub buffer\n");
506 } else if (errno == EIO) {
507 /* Should never happen with newer LTTng versions */
508 perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
509 }
21073eaa
DG
510
511 ret = -err;
d41f73b7
MD
512 goto end;
513 }
514
515end:
516 return ret;
517}
518
519int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
520{
521 int ret;
ffe60014
DG
522
523 assert(stream);
524
fe4477ee
JD
525 /* Don't create anything if this is set for streaming. */
526 if (stream->net_seq_idx == (uint64_t) -1ULL) {
527 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
528 stream->chan->tracefile_size, stream->tracefile_count_current,
529 stream->uid, stream->gid);
530 if (ret < 0) {
531 goto error;
532 }
533 stream->out_fd = ret;
534 stream->tracefile_size_current = 0;
ffe60014 535 }
d41f73b7 536
d41f73b7
MD
537 if (stream->output == LTTNG_EVENT_MMAP) {
538 /* get the len of the mmap region */
539 unsigned long mmap_len;
540
541 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
542 if (ret != 0) {
87dc6a9c 543 errno = -ret;
ffe60014 544 PERROR("kernctl_get_mmap_len");
d41f73b7
MD
545 goto error_close_fd;
546 }
547 stream->mmap_len = (size_t) mmap_len;
548
ffe60014
DG
549 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
550 MAP_PRIVATE, stream->wait_fd, 0);
d41f73b7 551 if (stream->mmap_base == MAP_FAILED) {
ffe60014 552 PERROR("Error mmaping");
d41f73b7
MD
553 ret = -1;
554 goto error_close_fd;
555 }
556 }
557
558 /* we return 0 to let the library handle the FD internally */
559 return 0;
560
561error_close_fd:
562 {
563 int err;
564
565 err = close(stream->out_fd);
566 assert(!err);
567 }
568error:
569 return ret;
570}
571
ca22feea
DG
572/*
573 * Check if data is still being extracted from the buffers for a specific
4e9a4686
DG
574 * stream. Consumer data lock MUST be acquired before calling this function
575 * and the stream lock.
ca22feea 576 *
6d805429 577 * Return 1 if the traced data are still getting read else 0 meaning that the
ca22feea
DG
578 * data is available for trace viewer reading.
579 */
6d805429 580int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
ca22feea
DG
581{
582 int ret;
583
584 assert(stream);
585
ca22feea
DG
586 ret = kernctl_get_next_subbuf(stream->wait_fd);
587 if (ret == 0) {
588 /* There is still data so let's put back this subbuffer. */
589 ret = kernctl_put_subbuf(stream->wait_fd);
590 assert(ret == 0);
6d805429 591 ret = 1; /* Data is pending */
4e9a4686 592 goto end;
ca22feea
DG
593 }
594
6d805429
DG
595 /* Data is NOT pending and ready to be read. */
596 ret = 0;
ca22feea 597
6efae65e
DG
598end:
599 return ret;
ca22feea 600}
This page took 0.104742 seconds and 4 git commands to generate.