f23fc9c5711f18101e1b7b3fb6cd7245ea378cee
[lttng-tools.git] / src / common / kernel-consumer / kernel-consumer.c
1 /*
2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
4 *
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.
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 *
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.
17 */
18
19 #define _GNU_SOURCE
20 #include <assert.h>
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>
28 #include <inttypes.h>
29 #include <unistd.h>
30 #include <sys/stat.h>
31
32 #include <common/common.h>
33 #include <common/kernel-ctl/kernel-ctl.h>
34 #include <common/sessiond-comm/sessiond-comm.h>
35 #include <common/sessiond-comm/relayd.h>
36 #include <common/compat/fcntl.h>
37 #include <common/pipe.h>
38 #include <common/relayd/relayd.h>
39 #include <common/utils.h>
40
41 #include "kernel-consumer.h"
42
43 extern struct lttng_consumer_global_data consumer_data;
44 extern int consumer_poll_timeout;
45 extern volatile int consumer_quit;
46
47 /*
48 * Take a snapshot for a specific fd
49 *
50 * Returns 0 on success, < 0 on error
51 */
52 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
53 {
54 int ret = 0;
55 int infd = stream->wait_fd;
56
57 ret = kernctl_snapshot(infd);
58 if (ret != 0) {
59 errno = -ret;
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 */
71 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
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) {
79 errno = -ret;
80 perror("kernctl_snapshot_get_produced");
81 }
82
83 return ret;
84 }
85
86 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
87 int sock, struct pollfd *consumer_sockpoll)
88 {
89 ssize_t ret;
90 enum lttng_error_code ret_code = LTTNG_OK;
91 struct lttcomm_consumer_msg msg;
92
93 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
94 if (ret != sizeof(msg)) {
95 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
96 return ret;
97 }
98 if (msg.cmd_type == LTTNG_CONSUMER_STOP) {
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);
107 return -ENOENT;
108 }
109
110 /* relayd needs RCU read-side protection */
111 rcu_read_lock();
112
113 switch (msg.cmd_type) {
114 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
115 {
116 /* Session daemon status message are handled in the following call. */
117 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
118 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
119 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id);
120 goto end_nosignal;
121 }
122 case LTTNG_CONSUMER_ADD_CHANNEL:
123 {
124 struct lttng_consumer_channel *new_channel;
125
126 /* First send a status message before receiving the fds. */
127 ret = consumer_send_status_msg(sock, ret_code);
128 if (ret < 0) {
129 /* Somehow, the session daemon is not responding anymore. */
130 goto end_nosignal;
131 }
132
133 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
134 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
135 msg.u.channel.session_id, msg.u.channel.pathname,
136 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
137 msg.u.channel.relayd_id, msg.u.channel.output,
138 msg.u.channel.tracefile_size,
139 msg.u.channel.tracefile_count);
140 if (new_channel == NULL) {
141 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
142 goto end_nosignal;
143 }
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
157 if (ctx->on_recv_channel != NULL) {
158 ret = ctx->on_recv_channel(new_channel);
159 if (ret == 0) {
160 consumer_add_channel(new_channel, ctx);
161 } else if (ret < 0) {
162 goto end_nosignal;
163 }
164 } else {
165 consumer_add_channel(new_channel, ctx);
166 }
167 goto end_nosignal;
168 }
169 case LTTNG_CONSUMER_ADD_STREAM:
170 {
171 int fd, stream_pipe;
172 struct consumer_relayd_sock_pair *relayd = NULL;
173 struct lttng_consumer_stream *new_stream;
174 struct lttng_consumer_channel *channel;
175 int alloc_ret = 0;
176
177 /*
178 * Get stream's channel reference. Needed when adding the stream to the
179 * global hash table.
180 */
181 channel = consumer_find_channel(msg.u.stream.channel_key);
182 if (!channel) {
183 /*
184 * We could not find the channel. Can happen if cpu hotplug
185 * happens while tearing down.
186 */
187 ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
188 ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
189 }
190
191 /* First send a status message before receiving the fds. */
192 ret = consumer_send_status_msg(sock, ret_code);
193 if (ret < 0 || ret_code != LTTNG_OK) {
194 /*
195 * Somehow, the session daemon is not responding anymore or the
196 * channel was not found.
197 */
198 goto end_nosignal;
199 }
200
201 /* block */
202 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
203 rcu_read_unlock();
204 return -EINTR;
205 }
206
207 /* Get stream file descriptor from socket */
208 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
209 if (ret != sizeof(fd)) {
210 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
211 rcu_read_unlock();
212 return ret;
213 }
214
215 /*
216 * Send status code to session daemon only if the recv works. If the
217 * above recv() failed, the session daemon is notified through the
218 * error socket and the teardown is eventually done.
219 */
220 ret = consumer_send_status_msg(sock, ret_code);
221 if (ret < 0) {
222 /* Somehow, the session daemon is not responding anymore. */
223 goto end_nosignal;
224 }
225
226 new_stream = consumer_allocate_stream(channel->key,
227 fd,
228 LTTNG_CONSUMER_ACTIVE_STREAM,
229 channel->name,
230 channel->uid,
231 channel->gid,
232 channel->relayd_id,
233 channel->session_id,
234 msg.u.stream.cpu,
235 &alloc_ret,
236 channel->type);
237 if (new_stream == NULL) {
238 switch (alloc_ret) {
239 case -ENOMEM:
240 case -EINVAL:
241 default:
242 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
243 break;
244 }
245 goto end_nosignal;
246 }
247 new_stream->chan = channel;
248 new_stream->wait_fd = fd;
249
250 /*
251 * The buffer flush is done on the session daemon side for the kernel
252 * so no need for the stream "hangup_flush_done" variable to be
253 * tracked. This is important for a kernel stream since we don't rely
254 * on the flush state of the stream to read data. It's not the case for
255 * user space tracing.
256 */
257 new_stream->hangup_flush_done = 0;
258
259 /* The stream is not metadata. Get relayd reference if exists. */
260 relayd = consumer_find_relayd(new_stream->net_seq_idx);
261 if (relayd != NULL) {
262 /* Add stream on the relayd */
263 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
264 ret = relayd_add_stream(&relayd->control_sock,
265 new_stream->name, new_stream->chan->pathname,
266 &new_stream->relayd_stream_id,
267 new_stream->chan->tracefile_size,
268 new_stream->chan->tracefile_count);
269 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
270 if (ret < 0) {
271 consumer_del_stream(new_stream, NULL);
272 goto end_nosignal;
273 }
274 } else if (new_stream->net_seq_idx != (uint64_t) -1ULL) {
275 ERR("Network sequence index %" PRIu64 " unknown. Not adding stream.",
276 new_stream->net_seq_idx);
277 consumer_del_stream(new_stream, NULL);
278 goto end_nosignal;
279 }
280
281 if (ctx->on_recv_stream) {
282 ret = ctx->on_recv_stream(new_stream);
283 if (ret < 0) {
284 consumer_del_stream(new_stream, NULL);
285 goto end_nosignal;
286 }
287 }
288
289 /* Get the right pipe where the stream will be sent. */
290 if (new_stream->metadata_flag) {
291 stream_pipe = lttng_pipe_get_writefd(ctx->consumer_metadata_pipe);
292 } else {
293 stream_pipe = lttng_pipe_get_writefd(ctx->consumer_data_pipe);
294 }
295
296 do {
297 ret = write(stream_pipe, &new_stream, sizeof(new_stream));
298 } while (ret < 0 && errno == EINTR);
299 if (ret < 0) {
300 PERROR("Consumer write %s stream to pipe %d",
301 new_stream->metadata_flag ? "metadata" : "data",
302 stream_pipe);
303 consumer_del_stream(new_stream, NULL);
304 goto end_nosignal;
305 }
306
307 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
308 new_stream->name, fd, new_stream->relayd_stream_id);
309 break;
310 }
311 case LTTNG_CONSUMER_UPDATE_STREAM:
312 {
313 rcu_read_unlock();
314 return -ENOSYS;
315 }
316 case LTTNG_CONSUMER_DESTROY_RELAYD:
317 {
318 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
319 struct consumer_relayd_sock_pair *relayd;
320
321 DBG("Kernel consumer destroying relayd %" PRIu64, index);
322
323 /* Get relayd reference if exists. */
324 relayd = consumer_find_relayd(index);
325 if (relayd == NULL) {
326 DBG("Unable to find relayd %" PRIu64, index);
327 ret_code = LTTNG_ERR_NO_CONSUMER;
328 }
329
330 /*
331 * Each relayd socket pair has a refcount of stream attached to it
332 * which tells if the relayd is still active or not depending on the
333 * refcount value.
334 *
335 * This will set the destroy flag of the relayd object and destroy it
336 * if the refcount reaches zero when called.
337 *
338 * The destroy can happen either here or when a stream fd hangs up.
339 */
340 if (relayd) {
341 consumer_flag_relayd_for_destroy(relayd);
342 }
343
344 ret = consumer_send_status_msg(sock, ret_code);
345 if (ret < 0) {
346 /* Somehow, the session daemon is not responding anymore. */
347 goto end_nosignal;
348 }
349
350 goto end_nosignal;
351 }
352 case LTTNG_CONSUMER_DATA_PENDING:
353 {
354 int32_t ret;
355 uint64_t id = msg.u.data_pending.session_id;
356
357 DBG("Kernel consumer data pending command for id %" PRIu64, id);
358
359 ret = consumer_data_pending(id);
360
361 /* Send back returned value to session daemon */
362 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
363 if (ret < 0) {
364 PERROR("send data pending ret code");
365 }
366
367 /*
368 * No need to send back a status message since the data pending
369 * returned value is the response.
370 */
371 break;
372 }
373 default:
374 goto end_nosignal;
375 }
376
377 end_nosignal:
378 rcu_read_unlock();
379
380 /*
381 * Return 1 to indicate success since the 0 value can be a socket
382 * shutdown during the recv() or send() call.
383 */
384 return 1;
385 }
386
387 /*
388 * Consume data on a file descriptor and write it on a trace file.
389 */
390 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
391 struct lttng_consumer_local_data *ctx)
392 {
393 unsigned long len, subbuf_size, padding;
394 int err;
395 ssize_t ret = 0;
396 int infd = stream->wait_fd;
397
398 DBG("In read_subbuffer (infd : %d)", infd);
399 /* Get the next subbuffer */
400 err = kernctl_get_next_subbuf(infd);
401 if (err != 0) {
402 ret = err;
403 /*
404 * This is a debug message even for single-threaded consumer,
405 * because poll() have more relaxed criterions than get subbuf,
406 * so get_subbuf may fail for short race windows where poll()
407 * would issue wakeups.
408 */
409 DBG("Reserving sub buffer failed (everything is normal, "
410 "it is due to concurrency)");
411 goto end;
412 }
413
414 /* Get the full subbuffer size including padding */
415 err = kernctl_get_padded_subbuf_size(infd, &len);
416 if (err != 0) {
417 errno = -err;
418 perror("Getting sub-buffer len failed.");
419 ret = err;
420 goto end;
421 }
422
423 switch (stream->chan->output) {
424 case LTTNG_EVENT_SPLICE:
425
426 /*
427 * XXX: The lttng-modules splice "actor" does not handle copying
428 * partial pages hence only using the subbuffer size without the
429 * padding makes the splice fail.
430 */
431 subbuf_size = len;
432 padding = 0;
433
434 /* splice the subbuffer to the tracefile */
435 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
436 padding);
437 /*
438 * XXX: Splice does not support network streaming so the return value
439 * is simply checked against subbuf_size and not like the mmap() op.
440 */
441 if (ret != subbuf_size) {
442 /*
443 * display the error but continue processing to try
444 * to release the subbuffer
445 */
446 ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
447 ret, subbuf_size);
448 }
449 break;
450 case LTTNG_EVENT_MMAP:
451 /* Get subbuffer size without padding */
452 err = kernctl_get_subbuf_size(infd, &subbuf_size);
453 if (err != 0) {
454 errno = -err;
455 perror("Getting sub-buffer len failed.");
456 ret = err;
457 goto end;
458 }
459
460 /* Make sure the tracer is not gone mad on us! */
461 assert(len >= subbuf_size);
462
463 padding = len - subbuf_size;
464
465 /* write the subbuffer to the tracefile */
466 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
467 padding);
468 /*
469 * The mmap operation should write subbuf_size amount of data when
470 * network streaming or the full padding (len) size when we are _not_
471 * streaming.
472 */
473 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
474 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
475 /*
476 * Display the error but continue processing to try to release the
477 * subbuffer
478 */
479 ERR("Error writing to tracefile "
480 "(ret: %zd != len: %lu != subbuf_size: %lu)",
481 ret, len, subbuf_size);
482 }
483 break;
484 default:
485 ERR("Unknown output method");
486 ret = -1;
487 }
488
489 err = kernctl_put_next_subbuf(infd);
490 if (err != 0) {
491 errno = -err;
492 if (errno == EFAULT) {
493 perror("Error in unreserving sub buffer\n");
494 } else if (errno == EIO) {
495 /* Should never happen with newer LTTng versions */
496 perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
497 }
498
499 ret = -err;
500 goto end;
501 }
502
503 end:
504 return ret;
505 }
506
507 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
508 {
509 int ret;
510
511 assert(stream);
512
513 /* Don't create anything if this is set for streaming. */
514 if (stream->net_seq_idx == (uint64_t) -1ULL) {
515 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
516 stream->chan->tracefile_size, stream->tracefile_count_current,
517 stream->uid, stream->gid);
518 if (ret < 0) {
519 goto error;
520 }
521 stream->out_fd = ret;
522 stream->tracefile_size_current = 0;
523 }
524
525 if (stream->output == LTTNG_EVENT_MMAP) {
526 /* get the len of the mmap region */
527 unsigned long mmap_len;
528
529 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
530 if (ret != 0) {
531 errno = -ret;
532 PERROR("kernctl_get_mmap_len");
533 goto error_close_fd;
534 }
535 stream->mmap_len = (size_t) mmap_len;
536
537 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
538 MAP_PRIVATE, stream->wait_fd, 0);
539 if (stream->mmap_base == MAP_FAILED) {
540 PERROR("Error mmaping");
541 ret = -1;
542 goto error_close_fd;
543 }
544 }
545
546 /* we return 0 to let the library handle the FD internally */
547 return 0;
548
549 error_close_fd:
550 {
551 int err;
552
553 err = close(stream->out_fd);
554 assert(!err);
555 }
556 error:
557 return ret;
558 }
559
560 /*
561 * Check if data is still being extracted from the buffers for a specific
562 * stream. Consumer data lock MUST be acquired before calling this function
563 * and the stream lock.
564 *
565 * Return 1 if the traced data are still getting read else 0 meaning that the
566 * data is available for trace viewer reading.
567 */
568 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
569 {
570 int ret;
571
572 assert(stream);
573
574 ret = kernctl_get_next_subbuf(stream->wait_fd);
575 if (ret == 0) {
576 /* There is still data so let's put back this subbuffer. */
577 ret = kernctl_put_subbuf(stream->wait_fd);
578 assert(ret == 0);
579 ret = 1; /* Data is pending */
580 goto end;
581 }
582
583 /* Data is NOT pending and ready to be read. */
584 ret = 0;
585
586 end:
587 return ret;
588 }
This page took 0.04778 seconds and 4 git commands to generate.