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");
87 * Receive command from session daemon and process it.
89 * Return 1 on success else a negative value or 0.
91 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
92 int sock
, struct pollfd
*consumer_sockpoll
)
95 enum lttng_error_code ret_code
= LTTNG_OK
;
96 struct lttcomm_consumer_msg msg
;
98 ret
= lttcomm_recv_unix_sock(sock
, &msg
, sizeof(msg
));
99 if (ret
!= sizeof(msg
)) {
100 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
106 if (msg
.cmd_type
== LTTNG_CONSUMER_STOP
) {
108 * Notify the session daemon that the command is completed.
110 * On transport layer error, the function call will print an error
111 * message so handling the returned code is a bit useless since we
112 * return an error code anyway.
114 (void) consumer_send_status_msg(sock
, ret_code
);
118 /* relayd needs RCU read-side protection */
121 switch (msg
.cmd_type
) {
122 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET
:
124 /* Session daemon status message are handled in the following call. */
125 ret
= consumer_add_relayd_socket(msg
.u
.relayd_sock
.net_index
,
126 msg
.u
.relayd_sock
.type
, ctx
, sock
, consumer_sockpoll
,
127 &msg
.u
.relayd_sock
.sock
, msg
.u
.relayd_sock
.session_id
);
130 case LTTNG_CONSUMER_ADD_CHANNEL
:
132 struct lttng_consumer_channel
*new_channel
;
135 /* First send a status message before receiving the fds. */
136 ret
= consumer_send_status_msg(sock
, ret_code
);
138 /* Somehow, the session daemon is not responding anymore. */
141 DBG("consumer_add_channel %" PRIu64
, msg
.u
.channel
.channel_key
);
142 new_channel
= consumer_allocate_channel(msg
.u
.channel
.channel_key
,
143 msg
.u
.channel
.session_id
, msg
.u
.channel
.pathname
,
144 msg
.u
.channel
.name
, msg
.u
.channel
.uid
, msg
.u
.channel
.gid
,
145 msg
.u
.channel
.relayd_id
, msg
.u
.channel
.output
,
146 msg
.u
.channel
.tracefile_size
,
147 msg
.u
.channel
.tracefile_count
);
148 if (new_channel
== NULL
) {
149 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
152 new_channel
->nb_init_stream_left
= msg
.u
.channel
.nb_init_streams
;
154 /* Translate and save channel type. */
155 switch (msg
.u
.channel
.type
) {
156 case CONSUMER_CHANNEL_TYPE_DATA
:
157 case CONSUMER_CHANNEL_TYPE_METADATA
:
158 new_channel
->type
= msg
.u
.channel
.type
;
165 if (ctx
->on_recv_channel
!= NULL
) {
166 ret_recv
= ctx
->on_recv_channel(new_channel
);
168 ret
= consumer_add_channel(new_channel
, ctx
);
169 } else if (ret_recv
< 0) {
173 ret
= consumer_add_channel(new_channel
, ctx
);
176 /* If we received an error in add_channel, we need to report it. */
178 ret
= consumer_send_status_msg(sock
, ret
);
187 case LTTNG_CONSUMER_ADD_STREAM
:
190 struct lttng_pipe
*stream_pipe
;
191 struct consumer_relayd_sock_pair
*relayd
= NULL
;
192 struct lttng_consumer_stream
*new_stream
;
193 struct lttng_consumer_channel
*channel
;
197 * Get stream's channel reference. Needed when adding the stream to the
200 channel
= consumer_find_channel(msg
.u
.stream
.channel_key
);
203 * We could not find the channel. Can happen if cpu hotplug
204 * happens while tearing down.
206 ERR("Unable to find channel key %" PRIu64
, msg
.u
.stream
.channel_key
);
207 ret_code
= LTTNG_ERR_KERN_CHAN_NOT_FOUND
;
210 /* First send a status message before receiving the fds. */
211 ret
= consumer_send_status_msg(sock
, ret_code
);
214 * Somehow, the session daemon is not responding
219 if (ret_code
!= LTTNG_OK
) {
221 * Channel was not found.
227 if (lttng_consumer_poll_socket(consumer_sockpoll
) < 0) {
232 /* Get stream file descriptor from socket */
233 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
234 if (ret
!= sizeof(fd
)) {
235 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_FD
);
241 * Send status code to session daemon only if the recv works. If the
242 * above recv() failed, the session daemon is notified through the
243 * error socket and the teardown is eventually done.
245 ret
= consumer_send_status_msg(sock
, ret_code
);
247 /* Somehow, the session daemon is not responding anymore. */
251 new_stream
= consumer_allocate_stream(channel
->key
,
253 LTTNG_CONSUMER_ACTIVE_STREAM
,
262 if (new_stream
== NULL
) {
267 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
272 new_stream
->chan
= channel
;
273 new_stream
->wait_fd
= fd
;
276 * We've just assigned the channel to the stream so increment the
277 * refcount right now.
279 uatomic_inc(&new_stream
->chan
->refcount
);
282 * The buffer flush is done on the session daemon side for the kernel
283 * so no need for the stream "hangup_flush_done" variable to be
284 * tracked. This is important for a kernel stream since we don't rely
285 * on the flush state of the stream to read data. It's not the case for
286 * user space tracing.
288 new_stream
->hangup_flush_done
= 0;
290 /* The stream is not metadata. Get relayd reference if exists. */
291 relayd
= consumer_find_relayd(new_stream
->net_seq_idx
);
292 if (relayd
!= NULL
) {
293 /* Add stream on the relayd */
294 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
295 ret
= relayd_add_stream(&relayd
->control_sock
,
296 new_stream
->name
, new_stream
->chan
->pathname
,
297 &new_stream
->relayd_stream_id
,
298 new_stream
->chan
->tracefile_size
,
299 new_stream
->chan
->tracefile_count
);
300 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
302 consumer_del_stream(new_stream
, NULL
);
305 } else if (new_stream
->net_seq_idx
!= (uint64_t) -1ULL) {
306 ERR("Network sequence index %" PRIu64
" unknown. Not adding stream.",
307 new_stream
->net_seq_idx
);
308 consumer_del_stream(new_stream
, NULL
);
312 if (ctx
->on_recv_stream
) {
313 ret
= ctx
->on_recv_stream(new_stream
);
315 consumer_del_stream(new_stream
, NULL
);
320 /* Get the right pipe where the stream will be sent. */
321 if (new_stream
->metadata_flag
) {
322 stream_pipe
= ctx
->consumer_metadata_pipe
;
324 stream_pipe
= ctx
->consumer_data_pipe
;
327 ret
= lttng_pipe_write(stream_pipe
, &new_stream
, sizeof(new_stream
));
329 ERR("Consumer write %s stream to pipe %d",
330 new_stream
->metadata_flag
? "metadata" : "data",
331 lttng_pipe_get_writefd(stream_pipe
));
332 consumer_del_stream(new_stream
, NULL
);
336 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64
,
337 new_stream
->name
, fd
, new_stream
->relayd_stream_id
);
340 case LTTNG_CONSUMER_UPDATE_STREAM
:
345 case LTTNG_CONSUMER_DESTROY_RELAYD
:
347 uint64_t index
= msg
.u
.destroy_relayd
.net_seq_idx
;
348 struct consumer_relayd_sock_pair
*relayd
;
350 DBG("Kernel consumer destroying relayd %" PRIu64
, index
);
352 /* Get relayd reference if exists. */
353 relayd
= consumer_find_relayd(index
);
354 if (relayd
== NULL
) {
355 DBG("Unable to find relayd %" PRIu64
, index
);
356 ret_code
= LTTNG_ERR_NO_CONSUMER
;
360 * Each relayd socket pair has a refcount of stream attached to it
361 * which tells if the relayd is still active or not depending on the
364 * This will set the destroy flag of the relayd object and destroy it
365 * if the refcount reaches zero when called.
367 * The destroy can happen either here or when a stream fd hangs up.
370 consumer_flag_relayd_for_destroy(relayd
);
373 ret
= consumer_send_status_msg(sock
, ret_code
);
375 /* Somehow, the session daemon is not responding anymore. */
381 case LTTNG_CONSUMER_DATA_PENDING
:
384 uint64_t id
= msg
.u
.data_pending
.session_id
;
386 DBG("Kernel consumer data pending command for id %" PRIu64
, id
);
388 ret
= consumer_data_pending(id
);
390 /* Send back returned value to session daemon */
391 ret
= lttcomm_send_unix_sock(sock
, &ret
, sizeof(ret
));
393 PERROR("send data pending ret code");
398 * No need to send back a status message since the data pending
399 * returned value is the response.
411 * Return 1 to indicate success since the 0 value can be a socket
412 * shutdown during the recv() or send() call.
418 /* This will issue a consumer stop. */
423 * Consume data on a file descriptor and write it on a trace file.
425 ssize_t
lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
426 struct lttng_consumer_local_data
*ctx
)
428 unsigned long len
, subbuf_size
, padding
;
431 int infd
= stream
->wait_fd
;
433 DBG("In read_subbuffer (infd : %d)", infd
);
434 /* Get the next subbuffer */
435 err
= kernctl_get_next_subbuf(infd
);
439 * This is a debug message even for single-threaded consumer,
440 * because poll() have more relaxed criterions than get subbuf,
441 * so get_subbuf may fail for short race windows where poll()
442 * would issue wakeups.
444 DBG("Reserving sub buffer failed (everything is normal, "
445 "it is due to concurrency)");
449 /* Get the full subbuffer size including padding */
450 err
= kernctl_get_padded_subbuf_size(infd
, &len
);
453 perror("Getting sub-buffer len failed.");
458 switch (stream
->chan
->output
) {
459 case LTTNG_EVENT_SPLICE
:
462 * XXX: The lttng-modules splice "actor" does not handle copying
463 * partial pages hence only using the subbuffer size without the
464 * padding makes the splice fail.
469 /* splice the subbuffer to the tracefile */
470 ret
= lttng_consumer_on_read_subbuffer_splice(ctx
, stream
, subbuf_size
,
473 * XXX: Splice does not support network streaming so the return value
474 * is simply checked against subbuf_size and not like the mmap() op.
476 if (ret
!= subbuf_size
) {
478 * display the error but continue processing to try
479 * to release the subbuffer
481 ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
485 case LTTNG_EVENT_MMAP
:
486 /* Get subbuffer size without padding */
487 err
= kernctl_get_subbuf_size(infd
, &subbuf_size
);
490 perror("Getting sub-buffer len failed.");
495 /* Make sure the tracer is not gone mad on us! */
496 assert(len
>= subbuf_size
);
498 padding
= len
- subbuf_size
;
500 /* write the subbuffer to the tracefile */
501 ret
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, subbuf_size
,
504 * The mmap operation should write subbuf_size amount of data when
505 * network streaming or the full padding (len) size when we are _not_
508 if ((ret
!= subbuf_size
&& stream
->net_seq_idx
!= (uint64_t) -1ULL) ||
509 (ret
!= len
&& stream
->net_seq_idx
== (uint64_t) -1ULL)) {
511 * Display the error but continue processing to try to release the
514 ERR("Error writing to tracefile "
515 "(ret: %zd != len: %lu != subbuf_size: %lu)",
516 ret
, len
, subbuf_size
);
520 ERR("Unknown output method");
524 err
= kernctl_put_next_subbuf(infd
);
527 if (errno
== EFAULT
) {
528 perror("Error in unreserving sub buffer\n");
529 } else if (errno
== EIO
) {
530 /* Should never happen with newer LTTng versions */
531 perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
542 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
548 /* Don't create anything if this is set for streaming. */
549 if (stream
->net_seq_idx
== (uint64_t) -1ULL) {
550 ret
= utils_create_stream_file(stream
->chan
->pathname
, stream
->name
,
551 stream
->chan
->tracefile_size
, stream
->tracefile_count_current
,
552 stream
->uid
, stream
->gid
);
556 stream
->out_fd
= ret
;
557 stream
->tracefile_size_current
= 0;
560 if (stream
->output
== LTTNG_EVENT_MMAP
) {
561 /* get the len of the mmap region */
562 unsigned long mmap_len
;
564 ret
= kernctl_get_mmap_len(stream
->wait_fd
, &mmap_len
);
567 PERROR("kernctl_get_mmap_len");
570 stream
->mmap_len
= (size_t) mmap_len
;
572 stream
->mmap_base
= mmap(NULL
, stream
->mmap_len
, PROT_READ
,
573 MAP_PRIVATE
, stream
->wait_fd
, 0);
574 if (stream
->mmap_base
== MAP_FAILED
) {
575 PERROR("Error mmaping");
581 /* we return 0 to let the library handle the FD internally */
588 err
= close(stream
->out_fd
);
596 * Check if data is still being extracted from the buffers for a specific
597 * stream. Consumer data lock MUST be acquired before calling this function
598 * and the stream lock.
600 * Return 1 if the traced data are still getting read else 0 meaning that the
601 * data is available for trace viewer reading.
603 int lttng_kconsumer_data_pending(struct lttng_consumer_stream
*stream
)
609 ret
= kernctl_get_next_subbuf(stream
->wait_fd
);
611 /* There is still data so let's put back this subbuffer. */
612 ret
= kernctl_put_subbuf(stream
->wait_fd
);
614 ret
= 1; /* Data is pending */
618 /* Data is NOT pending and ready to be read. */