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/relayd/relayd.h>
39 #include "kernel-consumer.h"
41 extern struct lttng_consumer_global_data consumer_data
;
42 extern int consumer_poll_timeout
;
43 extern volatile int consumer_quit
;
46 * Take a snapshot for a specific fd
48 * Returns 0 on success, < 0 on error
50 int lttng_kconsumer_take_snapshot(struct lttng_consumer_stream
*stream
)
53 int infd
= stream
->wait_fd
;
55 ret
= kernctl_snapshot(infd
);
58 perror("Getting sub-buffer snapshot.");
65 * Get the produced position
67 * Returns 0 on success, < 0 on error
69 int lttng_kconsumer_get_produced_snapshot(struct lttng_consumer_stream
*stream
,
73 int infd
= stream
->wait_fd
;
75 ret
= kernctl_snapshot_get_produced(infd
, pos
);
78 perror("kernctl_snapshot_get_produced");
84 int lttng_kconsumer_recv_cmd(struct lttng_consumer_local_data
*ctx
,
85 int sock
, struct pollfd
*consumer_sockpoll
)
88 enum lttng_error_code ret_code
= LTTNG_OK
;
89 struct lttcomm_consumer_msg msg
;
91 ret
= lttcomm_recv_unix_sock(sock
, &msg
, sizeof(msg
));
92 if (ret
!= sizeof(msg
)) {
93 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_CMD
);
96 if (msg
.cmd_type
== LTTNG_CONSUMER_STOP
) {
98 * Notify the session daemon that the command is completed.
100 * On transport layer error, the function call will print an error
101 * message so handling the returned code is a bit useless since we
102 * return an error code anyway.
104 (void) consumer_send_status_msg(sock
, ret_code
);
108 /* relayd needs RCU read-side protection */
111 switch (msg
.cmd_type
) {
112 case LTTNG_CONSUMER_ADD_RELAYD_SOCKET
:
114 /* Session daemon status message are handled in the following call. */
115 ret
= consumer_add_relayd_socket(msg
.u
.relayd_sock
.net_index
,
116 msg
.u
.relayd_sock
.type
, ctx
, sock
, consumer_sockpoll
,
117 &msg
.u
.relayd_sock
.sock
, msg
.u
.relayd_sock
.session_id
);
120 case LTTNG_CONSUMER_ADD_CHANNEL
:
122 struct lttng_consumer_channel
*new_channel
;
124 /* First send a status message before receiving the fds. */
125 ret
= consumer_send_status_msg(sock
, ret_code
);
127 /* Somehow, the session daemon is not responding anymore. */
131 DBG("consumer_add_channel %" PRIu64
, msg
.u
.channel
.channel_key
);
132 new_channel
= consumer_allocate_channel(msg
.u
.channel
.channel_key
,
133 msg
.u
.channel
.session_id
, msg
.u
.channel
.pathname
,
134 msg
.u
.channel
.name
, msg
.u
.channel
.uid
, msg
.u
.channel
.gid
,
135 msg
.u
.channel
.relayd_id
, msg
.u
.channel
.output
,
136 msg
.u
.channel
.tracefile_size
,
137 msg
.u
.channel
.tracefile_count
);
138 if (new_channel
== NULL
) {
139 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
142 new_channel
->nb_init_stream_left
= msg
.u
.channel
.nb_init_streams
;
144 /* Translate and save channel type. */
145 switch (msg
.u
.channel
.type
) {
146 case CONSUMER_CHANNEL_TYPE_DATA
:
147 case CONSUMER_CHANNEL_TYPE_METADATA
:
148 new_channel
->type
= msg
.u
.channel
.type
;
155 if (ctx
->on_recv_channel
!= NULL
) {
156 ret
= ctx
->on_recv_channel(new_channel
);
158 consumer_add_channel(new_channel
, ctx
);
159 } else if (ret
< 0) {
163 consumer_add_channel(new_channel
, ctx
);
167 case LTTNG_CONSUMER_ADD_STREAM
:
170 struct consumer_relayd_sock_pair
*relayd
= NULL
;
171 struct lttng_consumer_stream
*new_stream
;
172 struct lttng_consumer_channel
*channel
;
176 * Get stream's channel reference. Needed when adding the stream to the
179 channel
= consumer_find_channel(msg
.u
.stream
.channel_key
);
182 * We could not find the channel. Can happen if cpu hotplug
183 * happens while tearing down.
185 ERR("Unable to find channel key %" PRIu64
, msg
.u
.stream
.channel_key
);
186 ret_code
= LTTNG_ERR_KERN_CHAN_NOT_FOUND
;
189 /* First send a status message before receiving the fds. */
190 ret
= consumer_send_status_msg(sock
, ret_code
);
191 if (ret
< 0 || ret_code
!= LTTNG_OK
) {
193 * Somehow, the session daemon is not responding anymore or the
194 * channel was not found.
200 if (lttng_consumer_poll_socket(consumer_sockpoll
) < 0) {
205 /* Get stream file descriptor from socket */
206 ret
= lttcomm_recv_fds_unix_sock(sock
, &fd
, 1);
207 if (ret
!= sizeof(fd
)) {
208 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_ERROR_RECV_FD
);
214 * Send status code to session daemon only if the recv works. If the
215 * above recv() failed, the session daemon is notified through the
216 * error socket and the teardown is eventually done.
218 ret
= consumer_send_status_msg(sock
, ret_code
);
220 /* Somehow, the session daemon is not responding anymore. */
224 new_stream
= consumer_allocate_stream(channel
->key
,
226 LTTNG_CONSUMER_ACTIVE_STREAM
,
235 if (new_stream
== NULL
) {
240 lttng_consumer_send_error(ctx
, LTTCOMM_CONSUMERD_OUTFD_ERROR
);
245 new_stream
->chan
= channel
;
246 new_stream
->wait_fd
= fd
;
249 * The buffer flush is done on the session daemon side for the kernel
250 * so no need for the stream "hangup_flush_done" variable to be
251 * tracked. This is important for a kernel stream since we don't rely
252 * on the flush state of the stream to read data. It's not the case for
253 * user space tracing.
255 new_stream
->hangup_flush_done
= 0;
257 /* The stream is not metadata. Get relayd reference if exists. */
258 relayd
= consumer_find_relayd(new_stream
->net_seq_idx
);
259 if (relayd
!= NULL
) {
260 /* Add stream on the relayd */
261 pthread_mutex_lock(&relayd
->ctrl_sock_mutex
);
262 ret
= relayd_add_stream(&relayd
->control_sock
,
263 new_stream
->name
, new_stream
->chan
->pathname
,
264 &new_stream
->relayd_stream_id
);
265 pthread_mutex_unlock(&relayd
->ctrl_sock_mutex
);
267 consumer_del_stream(new_stream
, NULL
);
270 } else if (new_stream
->net_seq_idx
!= (uint64_t) -1ULL) {
271 ERR("Network sequence index %" PRIu64
" unknown. Not adding stream.",
272 new_stream
->net_seq_idx
);
273 consumer_del_stream(new_stream
, NULL
);
277 if (ctx
->on_recv_stream
) {
278 ret
= ctx
->on_recv_stream(new_stream
);
280 consumer_del_stream(new_stream
, NULL
);
285 /* Get the right pipe where the stream will be sent. */
286 if (new_stream
->metadata_flag
) {
287 stream_pipe
= ctx
->consumer_metadata_pipe
[1];
289 stream_pipe
= ctx
->consumer_data_pipe
[1];
293 ret
= write(stream_pipe
, &new_stream
, sizeof(new_stream
));
294 } while (ret
< 0 && errno
== EINTR
);
296 PERROR("Consumer write %s stream to pipe %d",
297 new_stream
->metadata_flag
? "metadata" : "data",
299 consumer_del_stream(new_stream
, NULL
);
303 DBG("Kernel consumer ADD_STREAM %s (fd: %d) with relayd id %" PRIu64
,
304 new_stream
->name
, fd
, new_stream
->relayd_stream_id
);
307 case LTTNG_CONSUMER_UPDATE_STREAM
:
312 case LTTNG_CONSUMER_DESTROY_RELAYD
:
314 uint64_t index
= msg
.u
.destroy_relayd
.net_seq_idx
;
315 struct consumer_relayd_sock_pair
*relayd
;
317 DBG("Kernel consumer destroying relayd %" PRIu64
, index
);
319 /* Get relayd reference if exists. */
320 relayd
= consumer_find_relayd(index
);
321 if (relayd
== NULL
) {
322 DBG("Unable to find relayd %" PRIu64
, index
);
323 ret_code
= LTTNG_ERR_NO_CONSUMER
;
327 * Each relayd socket pair has a refcount of stream attached to it
328 * which tells if the relayd is still active or not depending on the
331 * This will set the destroy flag of the relayd object and destroy it
332 * if the refcount reaches zero when called.
334 * The destroy can happen either here or when a stream fd hangs up.
337 consumer_flag_relayd_for_destroy(relayd
);
340 ret
= consumer_send_status_msg(sock
, ret_code
);
342 /* Somehow, the session daemon is not responding anymore. */
348 case LTTNG_CONSUMER_DATA_PENDING
:
351 uint64_t id
= msg
.u
.data_pending
.session_id
;
353 DBG("Kernel consumer data pending command for id %" PRIu64
, id
);
355 ret
= consumer_data_pending(id
);
357 /* Send back returned value to session daemon */
358 ret
= lttcomm_send_unix_sock(sock
, &ret
, sizeof(ret
));
360 PERROR("send data pending ret code");
364 * No need to send back a status message since the data pending
365 * returned value is the response.
377 * Return 1 to indicate success since the 0 value can be a socket
378 * shutdown during the recv() or send() call.
384 * Consume data on a file descriptor and write it on a trace file.
386 ssize_t
lttng_kconsumer_read_subbuffer(struct lttng_consumer_stream
*stream
,
387 struct lttng_consumer_local_data
*ctx
)
389 unsigned long len
, subbuf_size
, padding
;
392 int infd
= stream
->wait_fd
;
394 DBG("In read_subbuffer (infd : %d)", infd
);
395 /* Get the next subbuffer */
396 err
= kernctl_get_next_subbuf(infd
);
400 * This is a debug message even for single-threaded consumer,
401 * because poll() have more relaxed criterions than get subbuf,
402 * so get_subbuf may fail for short race windows where poll()
403 * would issue wakeups.
405 DBG("Reserving sub buffer failed (everything is normal, "
406 "it is due to concurrency)");
410 /* Get the full subbuffer size including padding */
411 err
= kernctl_get_padded_subbuf_size(infd
, &len
);
414 perror("Getting sub-buffer len failed.");
419 switch (stream
->chan
->output
) {
420 case LTTNG_EVENT_SPLICE
:
423 * XXX: The lttng-modules splice "actor" does not handle copying
424 * partial pages hence only using the subbuffer size without the
425 * padding makes the splice fail.
430 /* splice the subbuffer to the tracefile */
431 ret
= lttng_consumer_on_read_subbuffer_splice(ctx
, stream
, subbuf_size
,
434 * XXX: Splice does not support network streaming so the return value
435 * is simply checked against subbuf_size and not like the mmap() op.
437 if (ret
!= subbuf_size
) {
439 * display the error but continue processing to try
440 * to release the subbuffer
442 ERR("Error splicing to tracefile (ret: %zd != len: %lu)",
446 case LTTNG_EVENT_MMAP
:
447 /* Get subbuffer size without padding */
448 err
= kernctl_get_subbuf_size(infd
, &subbuf_size
);
451 perror("Getting sub-buffer len failed.");
456 /* Make sure the tracer is not gone mad on us! */
457 assert(len
>= subbuf_size
);
459 padding
= len
- subbuf_size
;
461 /* write the subbuffer to the tracefile */
462 ret
= lttng_consumer_on_read_subbuffer_mmap(ctx
, stream
, subbuf_size
,
465 * The mmap operation should write subbuf_size amount of data when
466 * network streaming or the full padding (len) size when we are _not_
469 if ((ret
!= subbuf_size
&& stream
->net_seq_idx
!= (uint64_t) -1ULL) ||
470 (ret
!= len
&& stream
->net_seq_idx
== (uint64_t) -1ULL)) {
472 * Display the error but continue processing to try to release the
475 ERR("Error writing to tracefile "
476 "(ret: %zd != len: %lu != subbuf_size: %lu)",
477 ret
, len
, subbuf_size
);
481 ERR("Unknown output method");
485 err
= kernctl_put_next_subbuf(infd
);
488 if (errno
== EFAULT
) {
489 perror("Error in unreserving sub buffer\n");
490 } else if (errno
== EIO
) {
491 /* Should never happen with newer LTTng versions */
492 perror("Reader has been pushed by the writer, last sub-buffer corrupted.");
503 int lttng_kconsumer_on_recv_stream(struct lttng_consumer_stream
*stream
)
509 ret
= lttng_create_output_file(stream
);
511 ERR("Creating output file");
515 if (stream
->output
== LTTNG_EVENT_MMAP
) {
516 /* get the len of the mmap region */
517 unsigned long mmap_len
;
519 ret
= kernctl_get_mmap_len(stream
->wait_fd
, &mmap_len
);
522 PERROR("kernctl_get_mmap_len");
525 stream
->mmap_len
= (size_t) mmap_len
;
527 stream
->mmap_base
= mmap(NULL
, stream
->mmap_len
, PROT_READ
,
528 MAP_PRIVATE
, stream
->wait_fd
, 0);
529 if (stream
->mmap_base
== MAP_FAILED
) {
530 PERROR("Error mmaping");
536 /* we return 0 to let the library handle the FD internally */
543 err
= close(stream
->out_fd
);
551 * Check if data is still being extracted from the buffers for a specific
552 * stream. Consumer data lock MUST be acquired before calling this function
553 * and the stream lock.
555 * Return 1 if the traced data are still getting read else 0 meaning that the
556 * data is available for trace viewer reading.
558 int lttng_kconsumer_data_pending(struct lttng_consumer_stream
*stream
)
564 ret
= kernctl_get_next_subbuf(stream
->wait_fd
);
566 /* There is still data so let's put back this subbuffer. */
567 ret
= kernctl_put_subbuf(stream
->wait_fd
);
569 ret
= 1; /* Data is pending */
573 /* Data is NOT pending and ready to be read. */