2 * Copyright (C) 2011 - Julien Desfossez <julien.desfossez@polymtl.ca>
3 * Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
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.
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.
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.
26 #include <sys/socket.h>
27 #include <sys/types.h>
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>
41 #include "kernel-consumer.h"
43 extern struct lttng_consumer_global_data consumer_data
;
44 extern int consumer_poll_timeout
;
45 extern volatile int consumer_quit
;
48 * Take a snapshot for a specific fd
50 * Returns 0 on success, < 0 on error
52 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream
*stream
)
55 int infd
= stream
->wait_fd
;
57 ret
= kernctl_snapshot(infd
);
60 perror("Getting sub-buffer snapshot.");
67 * Get the produced position
69 * Returns 0 on success, < 0 on error
71 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream
*stream
,
75 int infd
= stream
->wait_fd
;
77 ret
= kernctl_snapshot_get_produced(infd
, pos
);
80 perror("kernctl_snapshot_get_produced");
86 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
87 int sock
, struct pollfd
*consumer_sockpoll
)
90 enum lttng_error_code ret_code
= LTTNG_OK
;
91 struct lttcomm_consumer_msg msg
;
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
);
98 if (msg
.cmd_type
== LTTNG_CONSUMER_STOP
) {
100 * Notify the session daemon that the command is completed.
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.
106 (void) consumer_send_status_msg(sock
, ret_code
);
110 /* relayd needs RCU read-side protection */
113 switch (msg
.cmd_type
) {
114 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET
:
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
);
122 case LTTNG_CONSUMER_ADD_CHANNEL
:
124 struct lttng_consumer_channel
*new_channel
;
127 /* First send a status message before receiving the fds. */
128 ret
= consumer_send_status_msg(sock
, ret_code
);
130 /* Somehow, the session daemon is not responding anymore. */
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
);
144 new_channel
->nb_init_stream_left
= msg
.u
.channel
.nb_init_streams
;
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
;
157 if (ctx
->on_recv_channel
!= NULL
) {
158 ret_recv
= ctx
->on_recv_channel(new_channel
);
160 ret
= consumer_add_channel(new_channel
, ctx
);
161 } else if (ret_recv
< 0) {
165 ret
= consumer_add_channel(new_channel
, ctx
);
168 /* If we received an error in add_channel, we need to report it. */
170 consumer_send_status_msg(sock
, ret
);
176 case LTTNG_CONSUMER_ADD_STREAM
:
179 struct lttng_pipe
*stream_pipe
;
180 struct consumer_relayd_sock_pair
*relayd
= NULL
;
181 struct lttng_consumer_stream
*new_stream
;
182 struct lttng_consumer_channel
*channel
;
186 * Get stream's channel reference. Needed when adding the stream to the
189 channel
= consumer_find_channel(msg
.u
.stream
.channel_key
);
192 * We could not find the channel. Can happen if cpu hotplug
193 * happens while tearing down.
195 ERR("Unable to find channel key %" PRIu64
, msg
.u
.stream
.channel_key
);
196 ret_code
= LTTNG_ERR_KERN_CHAN_NOT_FOUND
;
199 /* First send a status message before receiving the fds. */
200 ret
= consumer_send_status_msg(sock
, ret_code
);
201 if (ret
< 0 || ret_code
!= LTTNG_OK
) {
203 * Somehow, the session daemon is not responding anymore or the
204 * channel was not found.
210 if (lttng_consumer_poll_socket(consumer_sockpoll
) < 0) {
215 /* Get stream file descriptor from socket */
216 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
217 if (ret
!= sizeof(fd
)) {
218 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_FD
);
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.
228 ret
= consumer_send_status_msg(sock
, ret_code
);
230 /* Somehow, the session daemon is not responding anymore. */
234 new_stream
= consumer_allocate_stream(channel
->key
,
236 LTTNG_CONSUMER_ACTIVE_STREAM
,
245 if (new_stream
== NULL
) {
250 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
255 new_stream
->chan
= channel
;
256 new_stream
->wait_fd
= fd
;
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
);
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.
271 new_stream
->hangup_flush_done
= 0;
273 /* The stream is not metadata. Get relayd reference if exists. */
274 relayd
= consumer_find_relayd(new_stream
->net_seq_idx
);
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
,
279 new_stream
->name
, new_stream
->chan
->pathname
,
280 &new_stream
->relayd_stream_id
,
281 new_stream
->chan
->tracefile_size
,
282 new_stream
->chan
->tracefile_count
);
283 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
285 consumer_del_stream(new_stream
, NULL
);
288 } else if (new_stream
->net_seq_idx
!= (uint64_t) -1ULL) {
289 ERR("Network sequence index %" PRIu64
" unknown. Not adding stream.",
290 new_stream
->net_seq_idx
);
291 consumer_del_stream(new_stream
, NULL
);
295 if (ctx
->on_recv_stream
) {
296 ret
= ctx
->on_recv_stream(new_stream
);
298 consumer_del_stream(new_stream
, NULL
);
303 /* Get the right pipe where the stream will be sent. */
304 if (new_stream
->metadata_flag
) {
305 stream_pipe
= ctx
->consumer_metadata_pipe
;
307 stream_pipe
= ctx
->consumer_data_pipe
;
310 ret
= lttng_pipe_write(stream_pipe
, &new_stream
, sizeof(new_stream
));
312 ERR("Consumer write %s stream to pipe %d",
313 new_stream
->metadata_flag
? "metadata" : "data",
314 lttng_pipe_get_writefd(stream_pipe
));
315 consumer_del_stream(new_stream
, NULL
);
319 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64
,
320 new_stream
->name
, fd
, new_stream
->relayd_stream_id
);
323 case LTTNG_CONSUMER_UPDATE_STREAM
:
328 case LTTNG_CONSUMER_DESTROY_RELAYD
:
330 uint64_t index
= msg
.u
.destroy_relayd
.net_seq_idx
;
331 struct consumer_relayd_sock_pair
*relayd
;
333 DBG("Kernel consumer destroying relayd %" PRIu64
, index
);
335 /* Get relayd reference if exists. */
336 relayd
= consumer_find_relayd(index
);
337 if (relayd
== NULL
) {
338 DBG("Unable to find relayd %" PRIu64
, index
);
339 ret_code
= LTTNG_ERR_NO_CONSUMER
;
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
347 * This will set the destroy flag of the relayd object and destroy it
348 * if the refcount reaches zero when called.
350 * The destroy can happen either here or when a stream fd hangs up.
353 consumer_flag_relayd_for_destroy(relayd
);
356 ret
= consumer_send_status_msg(sock
, ret_code
);
358 /* Somehow, the session daemon is not responding anymore. */
364 case LTTNG_CONSUMER_DATA_PENDING
:
367 uint64_t id
= msg
.u
.data_pending
.session_id
;
369 DBG("Kernel consumer data pending command for id %" PRIu64
, id
);
371 ret
= consumer_data_pending(id
);
373 /* Send back returned value to session daemon */
374 ret
= lttcomm_send_unix_sock(sock
, &ret
, sizeof(ret
));
376 PERROR("send data pending ret code");
380 * No need to send back a status message since the data pending
381 * returned value is the response.
393 * Return 1 to indicate success since the 0 value can be a socket
394 * shutdown during the recv() or send() call.
400 * Consume data on a file descriptor and write it on a trace file.
402 ssize_t
lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
403 struct lttng_consumer_local_data
*ctx
)
405 unsigned long len
, subbuf_size
, padding
;
408 int infd
= stream
->wait_fd
;
410 DBG("In read_subbuffer (infd : %d)", infd
);
411 /* Get the next subbuffer */
412 err
= kernctl_get_next_subbuf(infd
);
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.
421 DBG("Reserving sub buffer failed (everything is normal, "
422 "it is due to concurrency)");
426 /* Get the full subbuffer size including padding */
427 err
= kernctl_get_padded_subbuf_size(infd
, &len
);
430 perror("Getting sub-buffer len failed.");
435 switch (stream
->chan
->output
) {
436 case LTTNG_EVENT_SPLICE
:
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.
446 /* splice the subbuffer to the tracefile */
447 ret
= lttng_consumer_on_read_subbuffer_splice(ctx
, stream
, subbuf_size
,
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.
453 if (ret
!= subbuf_size
) {
455 * display the error but continue processing to try
456 * to release the subbuffer
458 ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
462 case LTTNG_EVENT_MMAP
:
463 /* Get subbuffer size without padding */
464 err
= kernctl_get_subbuf_size(infd
, &subbuf_size
);
467 perror("Getting sub-buffer len failed.");
472 /* Make sure the tracer is not gone mad on us! */
473 assert(len
>= subbuf_size
);
475 padding
= len
- subbuf_size
;
477 /* write the subbuffer to the tracefile */
478 ret
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, subbuf_size
,
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_
485 if ((ret
!= subbuf_size
&& stream
->net_seq_idx
!= (uint64_t) -1ULL) ||
486 (ret
!= len
&& stream
->net_seq_idx
== (uint64_t) -1ULL)) {
488 * Display the error but continue processing to try to release the
491 ERR("Error writing to tracefile "
492 "(ret: %zd != len: %lu != subbuf_size: %lu)",
493 ret
, len
, subbuf_size
);
497 ERR("Unknown output method");
501 err
= kernctl_put_next_subbuf(infd
);
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.");
519 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
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
);
533 stream
->out_fd
= ret
;
534 stream
->tracefile_size_current
= 0;
537 if (stream
->output
== LTTNG_EVENT_MMAP
) {
538 /* get the len of the mmap region */
539 unsigned long mmap_len
;
541 ret
= kernctl_get_mmap_len(stream
->wait_fd
, &mmap_len
);
544 PERROR("kernctl_get_mmap_len");
547 stream
->mmap_len
= (size_t) mmap_len
;
549 stream
->mmap_base
= mmap(NULL
, stream
->mmap_len
, PROT_READ
,
550 MAP_PRIVATE
, stream
->wait_fd
, 0);
551 if (stream
->mmap_base
== MAP_FAILED
) {
552 PERROR("Error mmaping");
558 /* we return 0 to let the library handle the FD internally */
565 err
= close(stream
->out_fd
);
573 * Check if data is still being extracted from the buffers for a specific
574 * stream. Consumer data lock MUST be acquired before calling this function
575 * and the stream lock.
577 * Return 1 if the traced data are still getting read else 0 meaning that the
578 * data is available for trace viewer reading.
580 int lttng_kconsumer_data_pending(struct lttng_consumer_stream
*stream
)
586 ret
= kernctl_get_next_subbuf(stream
->wait_fd
);
588 /* There is still data so let's put back this subbuffer. */
589 ret
= kernctl_put_subbuf(stream
->wait_fd
);
591 ret
= 1; /* Data is pending */
595 /* Data is NOT pending and ready to be read. */