Fix: put subbuffer back in kernel snapshot error path
[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 #include <common/consumer-stream.h>
41
42 #include "kernel-consumer.h"
43
44 extern struct lttng_consumer_global_data consumer_data;
45 extern int consumer_poll_timeout;
46 extern volatile int consumer_quit;
47
48 /*
49 * Take a snapshot for a specific fd
50 *
51 * Returns 0 on success, < 0 on error
52 */
53 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream *stream)
54 {
55 int ret = 0;
56 int infd = stream->wait_fd;
57
58 ret = kernctl_snapshot(infd);
59 if (ret != 0) {
60 errno = -ret;
61 perror("Getting sub-buffer snapshot.");
62 }
63
64 return ret;
65 }
66
67 /*
68 * Get the produced position
69 *
70 * Returns 0 on success, < 0 on error
71 */
72 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream *stream,
73 unsigned long *pos)
74 {
75 int ret;
76 int infd = stream->wait_fd;
77
78 ret = kernctl_snapshot_get_produced(infd, pos);
79 if (ret != 0) {
80 errno = -ret;
81 perror("kernctl_snapshot_get_produced");
82 }
83
84 return ret;
85 }
86
87 /*
88 * Get the consumerd position
89 *
90 * Returns 0 on success, < 0 on error
91 */
92 int lttng_kconsumer_get_consumed_snapshot(struct lttng_consumer_stream *stream,
93 unsigned long *pos)
94 {
95 int ret;
96 int infd = stream->wait_fd;
97
98 ret = kernctl_snapshot_get_consumed(infd, pos);
99 if (ret != 0) {
100 errno = -ret;
101 perror("kernctl_snapshot_get_consumed");
102 }
103
104 return ret;
105 }
106
107 /*
108 * Find a relayd and send the stream
109 *
110 * Returns 0 on success, < 0 on error
111 */
112 static
113 int send_relayd_stream(struct lttng_consumer_stream *stream, char *path)
114 {
115 struct consumer_relayd_sock_pair *relayd;
116 int ret = 0;
117 char *stream_path;
118
119 if (path != NULL) {
120 stream_path = path;
121 } else {
122 stream_path = stream->chan->pathname;
123 }
124 /* The stream is not metadata. Get relayd reference if exists. */
125 rcu_read_lock();
126 relayd = consumer_find_relayd(stream->net_seq_idx);
127 if (relayd != NULL) {
128 /* Add stream on the relayd */
129 pthread_mutex_lock(&relayd->ctrl_sock_mutex);
130 ret = relayd_add_stream(&relayd->control_sock,
131 stream->name, stream_path,
132 &stream->relayd_stream_id,
133 stream->chan->tracefile_size,
134 stream->chan->tracefile_count);
135 pthread_mutex_unlock(&relayd->ctrl_sock_mutex);
136 if (ret < 0) {
137 goto end;
138 }
139 uatomic_inc(&relayd->refcount);
140 } else if (stream->net_seq_idx != (uint64_t) -1ULL) {
141 ERR("Network sequence index %" PRIu64 " unknown. Not adding stream.",
142 stream->net_seq_idx);
143 ret = -1;
144 goto end;
145 }
146
147 end:
148 rcu_read_unlock();
149 return ret;
150 }
151
152 /*
153 * Find a relayd and close the stream
154 */
155 static
156 void close_relayd_stream(struct lttng_consumer_stream *stream)
157 {
158 struct consumer_relayd_sock_pair *relayd;
159
160 /* The stream is not metadata. Get relayd reference if exists. */
161 rcu_read_lock();
162 relayd = consumer_find_relayd(stream->net_seq_idx);
163 if (relayd != NULL) {
164 consumer_stream_relayd_close(stream, relayd);
165 }
166 rcu_read_unlock();
167 }
168
169 /*
170 * Take a snapshot of all the stream of a channel
171 *
172 * Returns 0 on success, < 0 on error
173 */
174 int lttng_kconsumer_snapshot_channel(uint64_t key, char *path,
175 uint64_t relayd_id, struct lttng_consumer_local_data *ctx)
176 {
177 int ret;
178 unsigned long consumed_pos, produced_pos;
179 struct lttng_consumer_channel *channel;
180 struct lttng_consumer_stream *stream;
181
182 DBG("Kernel consumer snapshot channel %lu", key);
183
184 rcu_read_lock();
185
186 channel = consumer_find_channel(key);
187 if (!channel) {
188 ERR("No channel found for key %lu", key);
189 ret = -1;
190 goto end;
191 }
192
193 /* Splice is not supported yet for channel snapshot. */
194 if (channel->output != CONSUMER_CHANNEL_MMAP) {
195 ERR("Unsupported output %d", channel->output);
196 ret = -1;
197 goto end;
198 }
199
200 cds_list_for_each_entry(stream, &channel->stream_no_monitor_list.head,
201 no_monitor_node) {
202 /*
203 * Lock stream because we are about to change its state.
204 */
205 pthread_mutex_lock(&stream->lock);
206
207 /*
208 * Assign the received relayd ID so we can use it for streaming. The streams
209 * are not visible to anyone so this is OK to change it.
210 */
211 stream->net_seq_idx = relayd_id;
212 channel->relayd_id = relayd_id;
213 if (relayd_id != (uint64_t) -1ULL) {
214 ret = send_relayd_stream(stream, path);
215 if (ret < 0) {
216 ERR("sending stream to relayd");
217 goto end_unlock;
218 }
219 DBG("Stream %s sent to the relayd", stream->name);
220 } else {
221 ret = utils_create_stream_file(path, stream->name,
222 stream->chan->tracefile_size, stream->tracefile_count_current,
223 stream->uid, stream->gid);
224 if (ret < 0) {
225 ERR("utils_create_stream_file");
226 goto end_unlock;
227 }
228
229 stream->out_fd = ret;
230 stream->tracefile_size_current = 0;
231
232 DBG("Kernel consumer snapshot stream %s/%s (%lu)", path,
233 stream->name, stream->key);
234 }
235
236 ret = kernctl_buffer_flush(stream->wait_fd);
237 if (ret < 0) {
238 ERR("Failed to flush kernel metadata stream");
239 goto end_unlock;
240 }
241
242 ret = lttng_kconsumer_take_snapshot(stream);
243 if (ret < 0) {
244 ERR("Taking kernel snapshot");
245 goto end_unlock;
246 }
247
248 ret = lttng_kconsumer_get_produced_snapshot(stream, &produced_pos);
249 if (ret < 0) {
250 ERR("Produced kernel snapshot position");
251 goto end_unlock;
252 }
253
254 ret = lttng_kconsumer_get_consumed_snapshot(stream, &consumed_pos);
255 if (ret < 0) {
256 ERR("Consumerd kernel snapshot position");
257 goto end_unlock;
258 }
259
260 if (stream->max_sb_size == 0) {
261 ret = kernctl_get_max_subbuf_size(stream->wait_fd,
262 &stream->max_sb_size);
263 if (ret < 0) {
264 ERR("Getting kernel max_sb_size");
265 goto end_unlock;
266 }
267 }
268
269 while (consumed_pos < produced_pos) {
270 ssize_t read_len;
271 unsigned long len, padded_len;
272
273 DBG("Kernel consumer taking snapshot at pos %lu", consumed_pos);
274
275 ret = kernctl_get_subbuf(stream->wait_fd, &consumed_pos);
276 if (ret < 0) {
277 if (errno != EAGAIN) {
278 PERROR("kernctl_get_subbuf snapshot");
279 goto end_unlock;
280 }
281 DBG("Kernel consumer get subbuf failed. Skipping it.");
282 consumed_pos += stream->max_sb_size;
283 continue;
284 }
285
286 ret = kernctl_get_subbuf_size(stream->wait_fd, &len);
287 if (ret < 0) {
288 ERR("Snapshot kernctl_get_subbuf_size");
289 goto error_put_subbuf;
290 }
291
292 ret = kernctl_get_padded_subbuf_size(stream->wait_fd, &padded_len);
293 if (ret < 0) {
294 ERR("Snapshot kernctl_get_padded_subbuf_size");
295 goto error_put_subbuf;
296 }
297
298 read_len = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, len,
299 padded_len - len);
300 /*
301 * We write the padded len in local tracefiles but the data len
302 * when using a relay. Display the error but continue processing
303 * to try to release the subbuffer.
304 */
305 if (relayd_id != (uint64_t) -1ULL) {
306 if (read_len != len) {
307 ERR("Error sending to the relay (ret: %zd != len: %lu)",
308 read_len, len);
309 }
310 } else {
311 if (read_len != padded_len) {
312 ERR("Error writing to tracefile (ret: %zd != len: %lu)",
313 read_len, padded_len);
314 }
315 }
316
317 ret = kernctl_put_subbuf(stream->wait_fd);
318 if (ret < 0) {
319 ERR("Snapshot kernctl_put_subbuf");
320 goto end_unlock;
321 }
322 consumed_pos += stream->max_sb_size;
323 }
324
325 if (relayd_id == (uint64_t) -1ULL) {
326 ret = close(stream->out_fd);
327 if (ret < 0) {
328 PERROR("Kernel consumer snapshot close out_fd");
329 goto end_unlock;
330 }
331 stream->out_fd = -1;
332 } else {
333 close_relayd_stream(stream);
334 stream->net_seq_idx = (uint64_t) -1ULL;
335 }
336 pthread_mutex_unlock(&stream->lock);
337 }
338
339 /* All good! */
340 ret = 0;
341 goto end;
342
343 error_put_subbuf:
344 ret = kernctl_put_subbuf(stream->wait_fd);
345 if (ret < 0) {
346 ERR("Snapshot kernctl_put_subbuf error path");
347 }
348 end_unlock:
349 pthread_mutex_unlock(&stream->lock);
350 end:
351 rcu_read_unlock();
352 return ret;
353 }
354
355 /*
356 * Read the whole metadata available for a snapshot.
357 *
358 * Returns 0 on success, < 0 on error
359 */
360 int lttng_kconsumer_snapshot_metadata(uint64_t key, char *path,
361 uint64_t relayd_id, struct lttng_consumer_local_data *ctx)
362 {
363 struct lttng_consumer_channel *metadata_channel;
364 struct lttng_consumer_stream *metadata_stream;
365 int ret;
366
367 DBG("Kernel consumer snapshot metadata with key %" PRIu64 " at path %s",
368 key, path);
369
370 rcu_read_lock();
371
372 metadata_channel = consumer_find_channel(key);
373 if (!metadata_channel) {
374 ERR("Snapshot kernel metadata channel not found for key %lu", key);
375 ret = -1;
376 goto end;
377 }
378
379 metadata_stream = metadata_channel->metadata_stream;
380 assert(metadata_stream);
381
382 if (relayd_id != (uint64_t) -1ULL) {
383 ret = send_relayd_stream(metadata_stream, path);
384 if (ret < 0) {
385 ERR("sending stream to relayd");
386 }
387 DBG("Stream %s sent to the relayd", metadata_stream->name);
388 } else {
389 ret = utils_create_stream_file(path, metadata_stream->name,
390 metadata_stream->chan->tracefile_size,
391 metadata_stream->tracefile_count_current,
392 metadata_stream->uid, metadata_stream->gid);
393 if (ret < 0) {
394 goto end;
395 }
396 metadata_stream->out_fd = ret;
397 }
398
399 ret = 0;
400 while (ret >= 0) {
401 ret = lttng_kconsumer_read_subbuffer(metadata_stream, ctx);
402 if (ret < 0) {
403 if (ret != -EPERM) {
404 ERR("Kernel snapshot reading subbuffer");
405 goto end;
406 }
407 /* "ret" is negative at this point so we will exit the loop. */
408 continue;
409 }
410 }
411
412 if (relayd_id == (uint64_t) -1ULL) {
413 ret = close(metadata_stream->out_fd);
414 if (ret < 0) {
415 PERROR("Kernel consumer snapshot close out_fd");
416 goto end;
417 }
418 metadata_stream->out_fd = -1;
419 } else {
420 close_relayd_stream(metadata_stream);
421 metadata_stream->net_seq_idx = (uint64_t) -1ULL;
422 }
423
424 ret = 0;
425 end:
426 rcu_read_unlock();
427 return ret;
428 }
429
430 /*
431 * Receive command from session daemon and process it.
432 *
433 * Return 1 on success else a negative value or 0.
434 */
435 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data *ctx,
436 int sock, struct pollfd *consumer_sockpoll)
437 {
438 ssize_t ret;
439 enum lttng_error_code ret_code = LTTNG_OK;
440 struct lttcomm_consumer_msg msg;
441
442 ret = lttcomm_recv_unix_sock(sock, &msg, sizeof(msg));
443 if (ret != sizeof(msg)) {
444 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_CMD);
445 if (ret > 0) {
446 ret = -1;
447 }
448 return ret;
449 }
450 if (msg.cmd_type == LTTNG_CONSUMER_STOP) {
451 /*
452 * Notify the session daemon that the command is completed.
453 *
454 * On transport layer error, the function call will print an error
455 * message so handling the returned code is a bit useless since we
456 * return an error code anyway.
457 */
458 (void) consumer_send_status_msg(sock, ret_code);
459 return -ENOENT;
460 }
461
462 /* relayd needs RCU read-side protection */
463 rcu_read_lock();
464
465 switch (msg.cmd_type) {
466 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET:
467 {
468 /* Session daemon status message are handled in the following call. */
469 ret = consumer_add_relayd_socket(msg.u.relayd_sock.net_index,
470 msg.u.relayd_sock.type, ctx, sock, consumer_sockpoll,
471 &msg.u.relayd_sock.sock, msg.u.relayd_sock.session_id);
472 goto end_nosignal;
473 }
474 case LTTNG_CONSUMER_ADD_CHANNEL:
475 {
476 struct lttng_consumer_channel *new_channel;
477 int ret_recv;
478
479 /* First send a status message before receiving the fds. */
480 ret = consumer_send_status_msg(sock, ret_code);
481 if (ret < 0) {
482 /* Somehow, the session daemon is not responding anymore. */
483 goto error_fatal;
484 }
485 DBG("consumer_add_channel %" PRIu64, msg.u.channel.channel_key);
486 new_channel = consumer_allocate_channel(msg.u.channel.channel_key,
487 msg.u.channel.session_id, msg.u.channel.pathname,
488 msg.u.channel.name, msg.u.channel.uid, msg.u.channel.gid,
489 msg.u.channel.relayd_id, msg.u.channel.output,
490 msg.u.channel.tracefile_size,
491 msg.u.channel.tracefile_count, 0,
492 msg.u.channel.monitor);
493 if (new_channel == NULL) {
494 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
495 goto end_nosignal;
496 }
497 new_channel->nb_init_stream_left = msg.u.channel.nb_init_streams;
498 switch (msg.u.channel.output) {
499 case LTTNG_EVENT_SPLICE:
500 new_channel->output = CONSUMER_CHANNEL_SPLICE;
501 break;
502 case LTTNG_EVENT_MMAP:
503 new_channel->output = CONSUMER_CHANNEL_MMAP;
504 break;
505 default:
506 ERR("Channel output unknown %d", msg.u.channel.output);
507 goto end_nosignal;
508 }
509
510 /* Translate and save channel type. */
511 switch (msg.u.channel.type) {
512 case CONSUMER_CHANNEL_TYPE_DATA:
513 case CONSUMER_CHANNEL_TYPE_METADATA:
514 new_channel->type = msg.u.channel.type;
515 break;
516 default:
517 assert(0);
518 goto end_nosignal;
519 };
520
521 if (ctx->on_recv_channel != NULL) {
522 ret_recv = ctx->on_recv_channel(new_channel);
523 if (ret_recv == 0) {
524 ret = consumer_add_channel(new_channel, ctx);
525 } else if (ret_recv < 0) {
526 goto end_nosignal;
527 }
528 } else {
529 ret = consumer_add_channel(new_channel, ctx);
530 }
531
532 /* If we received an error in add_channel, we need to report it. */
533 if (ret < 0) {
534 ret = consumer_send_status_msg(sock, ret);
535 if (ret < 0) {
536 goto error_fatal;
537 }
538 goto end_nosignal;
539 }
540
541 goto end_nosignal;
542 }
543 case LTTNG_CONSUMER_ADD_STREAM:
544 {
545 int fd;
546 struct lttng_pipe *stream_pipe;
547 struct lttng_consumer_stream *new_stream;
548 struct lttng_consumer_channel *channel;
549 int alloc_ret = 0;
550
551 /*
552 * Get stream's channel reference. Needed when adding the stream to the
553 * global hash table.
554 */
555 channel = consumer_find_channel(msg.u.stream.channel_key);
556 if (!channel) {
557 /*
558 * We could not find the channel. Can happen if cpu hotplug
559 * happens while tearing down.
560 */
561 ERR("Unable to find channel key %" PRIu64, msg.u.stream.channel_key);
562 ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
563 }
564
565 /* First send a status message before receiving the fds. */
566 ret = consumer_send_status_msg(sock, ret_code);
567 if (ret < 0) {
568 /*
569 * Somehow, the session daemon is not responding
570 * anymore.
571 */
572 goto error_fatal;
573 }
574 if (ret_code != LTTNG_OK) {
575 /*
576 * Channel was not found.
577 */
578 goto end_nosignal;
579 }
580
581 /* block */
582 if (lttng_consumer_poll_socket(consumer_sockpoll) < 0) {
583 rcu_read_unlock();
584 return -EINTR;
585 }
586
587 /* Get stream file descriptor from socket */
588 ret = lttcomm_recv_fds_unix_sock(sock, &fd, 1);
589 if (ret != sizeof(fd)) {
590 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_ERROR_RECV_FD);
591 rcu_read_unlock();
592 return ret;
593 }
594
595 /*
596 * Send status code to session daemon only if the recv works. If the
597 * above recv() failed, the session daemon is notified through the
598 * error socket and the teardown is eventually done.
599 */
600 ret = consumer_send_status_msg(sock, ret_code);
601 if (ret < 0) {
602 /* Somehow, the session daemon is not responding anymore. */
603 goto end_nosignal;
604 }
605
606 new_stream = consumer_allocate_stream(channel->key,
607 fd,
608 LTTNG_CONSUMER_ACTIVE_STREAM,
609 channel->name,
610 channel->uid,
611 channel->gid,
612 channel->relayd_id,
613 channel->session_id,
614 msg.u.stream.cpu,
615 &alloc_ret,
616 channel->type);
617 if (new_stream == NULL) {
618 switch (alloc_ret) {
619 case -ENOMEM:
620 case -EINVAL:
621 default:
622 lttng_consumer_send_error(ctx, LTTCOMM_CONSUMERD_OUTFD_ERROR);
623 break;
624 }
625 goto end_nosignal;
626 }
627 new_stream->chan = channel;
628 new_stream->wait_fd = fd;
629 switch (channel->output) {
630 case CONSUMER_CHANNEL_SPLICE:
631 new_stream->output = LTTNG_EVENT_SPLICE;
632 break;
633 case CONSUMER_CHANNEL_MMAP:
634 new_stream->output = LTTNG_EVENT_MMAP;
635 break;
636 default:
637 ERR("Stream output unknown %d", channel->output);
638 goto end_nosignal;
639 }
640
641 /*
642 * We've just assigned the channel to the stream so increment the
643 * refcount right now. We don't need to increment the refcount for
644 * streams in no monitor because we handle manually the cleanup of
645 * those. It is very important to make sure there is NO prior
646 * consumer_del_stream() calls or else the refcount will be unbalanced.
647 */
648 if (channel->monitor) {
649 uatomic_inc(&new_stream->chan->refcount);
650 }
651
652 /*
653 * The buffer flush is done on the session daemon side for the kernel
654 * so no need for the stream "hangup_flush_done" variable to be
655 * tracked. This is important for a kernel stream since we don't rely
656 * on the flush state of the stream to read data. It's not the case for
657 * user space tracing.
658 */
659 new_stream->hangup_flush_done = 0;
660
661 if (ctx->on_recv_stream) {
662 ret = ctx->on_recv_stream(new_stream);
663 if (ret < 0) {
664 consumer_del_stream(new_stream, NULL);
665 goto end_nosignal;
666 }
667 }
668
669 if (new_stream->metadata_flag) {
670 channel->metadata_stream = new_stream;
671 }
672
673 /* Do not monitor this stream. */
674 if (!channel->monitor) {
675 DBG("Kernel consumer add stream %s in no monitor mode with "
676 "relayd id %" PRIu64, new_stream->name,
677 new_stream->net_seq_idx);
678 cds_list_add(&new_stream->no_monitor_node,
679 &channel->stream_no_monitor_list.head);
680 break;
681 }
682
683 ret = send_relayd_stream(new_stream, NULL);
684 if (ret < 0) {
685 consumer_del_stream(new_stream, NULL);
686 goto end_nosignal;
687 }
688
689 /* Get the right pipe where the stream will be sent. */
690 if (new_stream->metadata_flag) {
691 stream_pipe = ctx->consumer_metadata_pipe;
692 } else {
693 stream_pipe = ctx->consumer_data_pipe;
694 }
695
696 ret = lttng_pipe_write(stream_pipe, &new_stream, sizeof(new_stream));
697 if (ret < 0) {
698 ERR("Consumer write %s stream to pipe %d",
699 new_stream->metadata_flag ? "metadata" : "data",
700 lttng_pipe_get_writefd(stream_pipe));
701 consumer_del_stream(new_stream, NULL);
702 goto end_nosignal;
703 }
704
705 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64,
706 new_stream->name, fd, new_stream->relayd_stream_id);
707 break;
708 }
709 case LTTNG_CONSUMER_UPDATE_STREAM:
710 {
711 rcu_read_unlock();
712 return -ENOSYS;
713 }
714 case LTTNG_CONSUMER_DESTROY_RELAYD:
715 {
716 uint64_t index = msg.u.destroy_relayd.net_seq_idx;
717 struct consumer_relayd_sock_pair *relayd;
718
719 DBG("Kernel consumer destroying relayd %" PRIu64, index);
720
721 /* Get relayd reference if exists. */
722 relayd = consumer_find_relayd(index);
723 if (relayd == NULL) {
724 DBG("Unable to find relayd %" PRIu64, index);
725 ret_code = LTTNG_ERR_NO_CONSUMER;
726 }
727
728 /*
729 * Each relayd socket pair has a refcount of stream attached to it
730 * which tells if the relayd is still active or not depending on the
731 * refcount value.
732 *
733 * This will set the destroy flag of the relayd object and destroy it
734 * if the refcount reaches zero when called.
735 *
736 * The destroy can happen either here or when a stream fd hangs up.
737 */
738 if (relayd) {
739 consumer_flag_relayd_for_destroy(relayd);
740 }
741
742 ret = consumer_send_status_msg(sock, ret_code);
743 if (ret < 0) {
744 /* Somehow, the session daemon is not responding anymore. */
745 goto error_fatal;
746 }
747
748 goto end_nosignal;
749 }
750 case LTTNG_CONSUMER_DATA_PENDING:
751 {
752 int32_t ret;
753 uint64_t id = msg.u.data_pending.session_id;
754
755 DBG("Kernel consumer data pending command for id %" PRIu64, id);
756
757 ret = consumer_data_pending(id);
758
759 /* Send back returned value to session daemon */
760 ret = lttcomm_send_unix_sock(sock, &ret, sizeof(ret));
761 if (ret < 0) {
762 PERROR("send data pending ret code");
763 goto error_fatal;
764 }
765
766 /*
767 * No need to send back a status message since the data pending
768 * returned value is the response.
769 */
770 break;
771 }
772 case LTTNG_CONSUMER_SNAPSHOT_CHANNEL:
773 {
774 if (msg.u.snapshot_channel.metadata == 1) {
775 ret = lttng_kconsumer_snapshot_metadata(msg.u.snapshot_channel.key,
776 msg.u.snapshot_channel.pathname,
777 msg.u.snapshot_channel.relayd_id, ctx);
778 if (ret < 0) {
779 ERR("Snapshot metadata failed");
780 ret_code = LTTNG_ERR_KERN_META_FAIL;
781 }
782 } else {
783 ret = lttng_kconsumer_snapshot_channel(msg.u.snapshot_channel.key,
784 msg.u.snapshot_channel.pathname,
785 msg.u.snapshot_channel.relayd_id, ctx);
786 if (ret < 0) {
787 ERR("Snapshot channel failed");
788 ret_code = LTTNG_ERR_KERN_CHAN_FAIL;
789 }
790 }
791
792 ret = consumer_send_status_msg(sock, ret_code);
793 if (ret < 0) {
794 /* Somehow, the session daemon is not responding anymore. */
795 goto end_nosignal;
796 }
797 break;
798 }
799 case LTTNG_CONSUMER_DESTROY_CHANNEL:
800 {
801 uint64_t key = msg.u.destroy_channel.key;
802 struct lttng_consumer_channel *channel;
803
804 channel = consumer_find_channel(key);
805 if (!channel) {
806 ERR("Kernel consumer destroy channel %" PRIu64 " not found", key);
807 ret_code = LTTNG_ERR_KERN_CHAN_NOT_FOUND;
808 }
809
810 ret = consumer_send_status_msg(sock, ret_code);
811 if (ret < 0) {
812 /* Somehow, the session daemon is not responding anymore. */
813 goto end_nosignal;
814 }
815
816 /*
817 * This command should ONLY be issued for channel with streams set in
818 * no monitor mode.
819 */
820 assert(!channel->monitor);
821
822 /*
823 * The refcount should ALWAYS be 0 in the case of a channel in no
824 * monitor mode.
825 */
826 assert(!uatomic_sub_return(&channel->refcount, 1));
827
828 consumer_del_channel(channel);
829
830 goto end_nosignal;
831 }
832 default:
833 goto end_nosignal;
834 }
835
836 end_nosignal:
837 rcu_read_unlock();
838
839 /*
840 * Return 1 to indicate success since the 0 value can be a socket
841 * shutdown during the recv() or send() call.
842 */
843 return 1;
844
845 error_fatal:
846 rcu_read_unlock();
847 /* This will issue a consumer stop. */
848 return -1;
849 }
850
851 /*
852 * Consume data on a file descriptor and write it on a trace file.
853 */
854 ssize_t lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream *stream,
855 struct lttng_consumer_local_data *ctx)
856 {
857 unsigned long len, subbuf_size, padding;
858 int err;
859 ssize_t ret = 0;
860 int infd = stream->wait_fd;
861
862 DBG("In read_subbuffer (infd : %d)", infd);
863 /* Get the next subbuffer */
864 err = kernctl_get_next_subbuf(infd);
865 if (err != 0) {
866 ret = err;
867 /*
868 * This is a debug message even for single-threaded consumer,
869 * because poll() have more relaxed criterions than get subbuf,
870 * so get_subbuf may fail for short race windows where poll()
871 * would issue wakeups.
872 */
873 DBG("Reserving sub buffer failed (everything is normal, "
874 "it is due to concurrency)");
875 goto end;
876 }
877
878 /* Get the full subbuffer size including padding */
879 err = kernctl_get_padded_subbuf_size(infd, &len);
880 if (err != 0) {
881 errno = -err;
882 perror("Getting sub-buffer len failed.");
883 ret = err;
884 goto end;
885 }
886
887 switch (stream->chan->output) {
888 case CONSUMER_CHANNEL_SPLICE:
889 /*
890 * XXX: The lttng-modules splice "actor" does not handle copying
891 * partial pages hence only using the subbuffer size without the
892 * padding makes the splice fail.
893 */
894 subbuf_size = len;
895 padding = 0;
896
897 /* splice the subbuffer to the tracefile */
898 ret = lttng_consumer_on_read_subbuffer_splice(ctx, stream, subbuf_size,
899 padding);
900 /*
901 * XXX: Splice does not support network streaming so the return value
902 * is simply checked against subbuf_size and not like the mmap() op.
903 */
904 if (ret != subbuf_size) {
905 /*
906 * display the error but continue processing to try
907 * to release the subbuffer
908 */
909 ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
910 ret, subbuf_size);
911 }
912 break;
913 case CONSUMER_CHANNEL_MMAP:
914 /* Get subbuffer size without padding */
915 err = kernctl_get_subbuf_size(infd, &subbuf_size);
916 if (err != 0) {
917 errno = -err;
918 perror("Getting sub-buffer len failed.");
919 ret = err;
920 goto end;
921 }
922
923 /* Make sure the tracer is not gone mad on us! */
924 assert(len >= subbuf_size);
925
926 padding = len - subbuf_size;
927
928 /* write the subbuffer to the tracefile */
929 ret = lttng_consumer_on_read_subbuffer_mmap(ctx, stream, subbuf_size,
930 padding);
931 /*
932 * The mmap operation should write subbuf_size amount of data when
933 * network streaming or the full padding (len) size when we are _not_
934 * streaming.
935 */
936 if ((ret != subbuf_size && stream->net_seq_idx != (uint64_t) -1ULL) ||
937 (ret != len && stream->net_seq_idx == (uint64_t) -1ULL)) {
938 /*
939 * Display the error but continue processing to try to release the
940 * subbuffer
941 */
942 ERR("Error writing to tracefile "
943 "(ret: %zd != len: %lu != subbuf_size: %lu)",
944 ret, len, subbuf_size);
945 }
946 break;
947 default:
948 ERR("Unknown output method");
949 ret = -1;
950 }
951
952 err = kernctl_put_next_subbuf(infd);
953 if (err != 0) {
954 errno = -err;
955 if (errno == EFAULT) {
956 perror("Error in unreserving sub buffer\n");
957 } else if (errno == EIO) {
958 /* Should never happen with newer LTTng versions */
959 perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
960 }
961
962 ret = -err;
963 goto end;
964 }
965
966 end:
967 return ret;
968 }
969
970 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream *stream)
971 {
972 int ret;
973
974 assert(stream);
975
976 /*
977 * Don't create anything if this is set for streaming or should not be
978 * monitored.
979 */
980 if (stream->net_seq_idx == (uint64_t) -1ULL && stream->chan->monitor) {
981 ret = utils_create_stream_file(stream->chan->pathname, stream->name,
982 stream->chan->tracefile_size, stream->tracefile_count_current,
983 stream->uid, stream->gid);
984 if (ret < 0) {
985 goto error;
986 }
987 stream->out_fd = ret;
988 stream->tracefile_size_current = 0;
989 }
990
991 if (stream->output == LTTNG_EVENT_MMAP) {
992 /* get the len of the mmap region */
993 unsigned long mmap_len;
994
995 ret = kernctl_get_mmap_len(stream->wait_fd, &mmap_len);
996 if (ret != 0) {
997 errno = -ret;
998 PERROR("kernctl_get_mmap_len");
999 goto error_close_fd;
1000 }
1001 stream->mmap_len = (size_t) mmap_len;
1002
1003 stream->mmap_base = mmap(NULL, stream->mmap_len, PROT_READ,
1004 MAP_PRIVATE, stream->wait_fd, 0);
1005 if (stream->mmap_base == MAP_FAILED) {
1006 PERROR("Error mmaping");
1007 ret = -1;
1008 goto error_close_fd;
1009 }
1010 }
1011
1012 /* we return 0 to let the library handle the FD internally */
1013 return 0;
1014
1015 error_close_fd:
1016 {
1017 int err;
1018
1019 err = close(stream->out_fd);
1020 assert(!err);
1021 }
1022 error:
1023 return ret;
1024 }
1025
1026 /*
1027 * Check if data is still being extracted from the buffers for a specific
1028 * stream. Consumer data lock MUST be acquired before calling this function
1029 * and the stream lock.
1030 *
1031 * Return 1 if the traced data are still getting read else 0 meaning that the
1032 * data is available for trace viewer reading.
1033 */
1034 int lttng_kconsumer_data_pending(struct lttng_consumer_stream *stream)
1035 {
1036 int ret;
1037
1038 assert(stream);
1039
1040 ret = kernctl_get_next_subbuf(stream->wait_fd);
1041 if (ret == 0) {
1042 /* There is still data so let's put back this subbuffer. */
1043 ret = kernctl_put_subbuf(stream->wait_fd);
1044 assert(ret == 0);
1045 ret = 1; /* Data is pending */
1046 goto end;
1047 }
1048
1049 /* Data is NOT pending and ready to be read. */
1050 ret = 0;
1051
1052 end:
1053 return ret;
1054 }
This page took 0.082067 seconds and 5 git commands to generate.