2 * This file is part of DisOrder
3 * Copyright (C) 2005, 2006, 2007 Richard Kettlewell
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
20 /** @file server/speaker.c
21 * @brief Speaker process
23 * This program is responsible for transmitting a single coherent audio stream
24 * to its destination (over the network, to some sound API, to some
25 * subprocess). It receives connections from decoders via file descriptor
26 * passing from the main server and plays them in the right order.
28 * @b Encodings. For the <a href="http://www.alsa-project.org/">ALSA</a> API,
29 * 8- and 16- bit stereo and mono are supported, with any sample rate (within
30 * the limits that ALSA can deal with.)
32 * When communicating with a subprocess, <a
33 * href="http://sox.sourceforge.net/">sox</a> is invoked to convert the inbound
34 * data to a single consistent format. The same applies for network (RTP)
35 * play, though in that case currently only 44.1KHz 16-bit stereo is supported.
37 * The inbound data starts with a structure defining the data format. Note
38 * that this is NOT portable between different platforms or even necessarily
39 * between versions; the speaker is assumed to be built from the same source
40 * and run on the same host as the main server.
42 * @b Garbage @b Collection. This program deliberately does not use the
43 * garbage collector even though it might be convenient to do so. This is for
44 * two reasons. Firstly some sound APIs use thread threads and we do not want
45 * to have to deal with potential interactions between threading and garbage
46 * collection. Secondly this process needs to be able to respond quickly and
47 * this is not compatible with the collector hanging the program even
50 * @b Units. This program thinks at various times in three different units.
51 * Bytes are obvious. A sample is a single sample on a single channel. A
52 * frame is several samples on different channels at the same point in time.
53 * So (for instance) a 16-bit stereo frame is 4 bytes and consists of a pair of
70 #include <sys/select.h>
75 #include <sys/socket.h>
80 #include "configuration.h"
85 #include "speaker-protocol.h"
93 #include <alsa/asoundlib.h>
96 /** @brief Linked list of all prepared tracks */
99 /** @brief Playing track, or NULL */
100 struct track
*playing
;
102 static time_t last_report
; /* when we last reported */
103 static int paused
; /* pause status */
104 static size_t bpf
; /* bytes per frame */
105 static struct pollfd fds
[NFDS
]; /* if we need more than that */
106 static int fdno
; /* fd number */
107 static size_t bufsize
; /* buffer size */
109 /** @brief The current PCM handle */
110 static snd_pcm_t
*pcm
;
111 static snd_pcm_uframes_t last_pcm_bufsize
; /* last seen buffer size */
112 static ao_sample_format pcm_format
; /* current format if aodev != 0 */
115 /** @brief Ready to send audio
117 * This is set when the destination is ready to receive audio. Generally
118 * this implies that the sound device is open. In the ALSA backend it
119 * does @b not necessarily imply that is has the right sample format.
123 /** @brief Frames to force-play
125 * If this is nonzero, and playing is enabled, then the main loop will attempt
126 * to play this many frames without checking whether the output device is
129 static int forceplay
;
131 /** @brief Pipe to subprocess
133 * This is the file descriptor to write to for @ref BACKEND_COMMAND.
135 static int cmdfd
= -1;
137 /** @brief Network socket
139 * This is the file descriptor to write to for @ref BACKEND_NETWORK.
143 /** @brief RTP timestamp
145 * This counts the number of samples played (NB not the number of frames
148 * The timestamp in the packet header is only 32 bits wide. With 44100Hz
149 * stereo, that only gives about half a day before wrapping, which is not
150 * particularly convenient for certain debugging purposes. Therefore the
151 * timestamp is maintained as a 64-bit integer, giving around six million years
152 * before wrapping, and truncated to 32 bits when transmitting.
154 static uint64_t rtp_time
;
156 /** @brief RTP base timestamp
158 * This is the real time correspoding to an @ref rtp_time of 0. It is used
159 * to recalculate the timestamp after idle periods.
161 static struct timeval rtp_time_0
;
163 /** @brief RTP packet sequence number */
164 static uint16_t rtp_seq
;
166 /** @brief RTP SSRC */
167 static uint32_t rtp_id
;
169 /** @brief Set when idled
171 * This is set when the sound device is deliberately closed by idle().
172 * @ref ready is set to 0 at the same time.
174 static int idled
; /* set when idled */
176 /** @brief Error counter */
177 static int audio_errors
;
179 /** @brief Selected backend */
180 static const struct speaker_backend
*backend
;
182 static const struct option options
[] = {
183 { "help", no_argument
, 0, 'h' },
184 { "version", no_argument
, 0, 'V' },
185 { "config", required_argument
, 0, 'c' },
186 { "debug", no_argument
, 0, 'd' },
187 { "no-debug", no_argument
, 0, 'D' },
191 /* Display usage message and terminate. */
192 static void help(void) {
194 " disorder-speaker [OPTIONS]\n"
196 " --help, -h Display usage message\n"
197 " --version, -V Display version number\n"
198 " --config PATH, -c PATH Set configuration file\n"
199 " --debug, -d Turn on debugging\n"
201 "Speaker process for DisOrder. Not intended to be run\n"
207 /* Display version number and terminate. */
208 static void version(void) {
209 xprintf("disorder-speaker version %s\n", disorder_version_string
);
214 /** @brief Return the number of bytes per frame in @p format */
215 static size_t bytes_per_frame(const ao_sample_format
*format
) {
216 return format
->channels
* format
->bits
/ 8;
219 /** @brief Find track @p id, maybe creating it if not found */
220 static struct track
*findtrack(const char *id
, int create
) {
223 D(("findtrack %s %d", id
, create
));
224 for(t
= tracks
; t
&& strcmp(id
, t
->id
); t
= t
->next
)
227 t
= xmalloc(sizeof *t
);
232 /* The initial input buffer will be the sample format. */
233 t
->buffer
= (void *)&t
->format
;
234 t
->size
= sizeof t
->format
;
239 /** @brief Remove track @p id (but do not destroy it) */
240 static struct track
*removetrack(const char *id
) {
241 struct track
*t
, **tt
;
243 D(("removetrack %s", id
));
244 for(tt
= &tracks
; (t
= *tt
) && strcmp(id
, t
->id
); tt
= &t
->next
)
251 /** @brief Destroy a track */
252 static void destroy(struct track
*t
) {
253 D(("destroy %s", t
->id
));
254 if(t
->fd
!= -1) xclose(t
->fd
);
255 if(t
->buffer
!= (void *)&t
->format
) free(t
->buffer
);
259 /** @brief Notice a new connection */
260 static void acquire(struct track
*t
, int fd
) {
261 D(("acquire %s %d", t
->id
, fd
));
268 /** @brief Return true if A and B denote identical libao formats, else false */
269 static int formats_equal(const ao_sample_format
*a
,
270 const ao_sample_format
*b
) {
271 return (a
->bits
== b
->bits
272 && a
->rate
== b
->rate
273 && a
->channels
== b
->channels
274 && a
->byte_format
== b
->byte_format
);
277 /** @brief Compute arguments to sox */
278 static void soxargs(const char ***pp
, char **qq
, ao_sample_format
*ao
) {
283 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-r%d", ao
->rate
); *qq
+= n
+ 1;
284 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-c%d", ao
->channels
); *qq
+= n
+ 1;
285 /* sox 12.17.9 insists on -b etc; CVS sox insists on -<n> etc; both are
287 switch(config
->sox_generation
) {
290 && ao
->byte_format
!= AO_FMT_NATIVE
291 && ao
->byte_format
!= MACHINE_AO_FMT
) {
295 case 8: *(*pp
)++ = "-b"; break;
296 case 16: *(*pp
)++ = "-w"; break;
297 case 32: *(*pp
)++ = "-l"; break;
298 case 64: *(*pp
)++ = "-d"; break;
299 default: fatal(0, "cannot handle sample size %d", (int)ao
->bits
);
303 switch(ao
->byte_format
) {
304 case AO_FMT_NATIVE
: break;
305 case AO_FMT_BIG
: *(*pp
)++ = "-B"; break;
306 case AO_FMT_LITTLE
: *(*pp
)++ = "-L"; break;
308 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-%d", ao
->bits
/8); *qq
+= n
+ 1;
313 /** @brief Enable format translation
315 * If necessary, replaces a tracks inbound file descriptor with one connected
316 * to a sox invocation, which performs the required translation.
318 static void enable_translation(struct track
*t
) {
319 if((backend
->flags
& FIXED_FORMAT
)
320 && !formats_equal(&t
->format
, &config
->sample_format
)) {
321 char argbuf
[1024], *q
= argbuf
;
322 const char *av
[18], **pp
= av
;
327 soxargs(&pp
, &q
, &t
->format
);
329 soxargs(&pp
, &q
, &config
->sample_format
);
333 for(pp
= av
; *pp
; pp
++)
334 D(("sox arg[%d] = %s", pp
- av
, *pp
));
340 signal(SIGPIPE
, SIG_DFL
);
342 xdup2(soxpipe
[1], 1);
343 fcntl(0, F_SETFL
, fcntl(0, F_GETFL
) & ~O_NONBLOCK
);
347 execvp("sox", (char **)av
);
350 D(("forking sox for format conversion (kid = %d)", soxkid
));
354 t
->format
= config
->sample_format
;
358 /** @brief Read data into a sample buffer
359 * @param t Pointer to track
360 * @return 0 on success, -1 on EOF
362 * This is effectively the read callback on @c t->fd. It is called from the
363 * main loop whenever the track's file descriptor is readable, assuming the
364 * buffer has not reached the maximum allowed occupancy.
366 static int fill(struct track
*t
) {
370 D(("fill %s: eof=%d used=%zu size=%zu got_format=%d",
371 t
->id
, t
->eof
, t
->used
, t
->size
, t
->got_format
));
372 if(t
->eof
) return -1;
373 if(t
->used
< t
->size
) {
374 /* there is room left in the buffer */
375 where
= (t
->start
+ t
->used
) % t
->size
;
377 /* We are reading audio data, get as much as we can */
378 if(where
>= t
->start
) left
= t
->size
- where
;
379 else left
= t
->start
- where
;
381 /* We are still waiting for the format, only get that */
382 left
= sizeof (ao_sample_format
) - t
->used
;
384 n
= read(t
->fd
, t
->buffer
+ where
, left
);
385 } while(n
< 0 && errno
== EINTR
);
387 if(errno
!= EAGAIN
) fatal(errno
, "error reading sample stream");
391 D(("fill %s: eof detected", t
->id
));
396 if(!t
->got_format
&& t
->used
>= sizeof (ao_sample_format
)) {
397 assert(t
->used
== sizeof (ao_sample_format
));
398 /* Check that our assumptions are met. */
399 if(t
->format
.bits
& 7)
400 fatal(0, "bits per sample not a multiple of 8");
401 /* If the input format is unsuitable, arrange to translate it */
402 enable_translation(t
);
403 /* Make a new buffer for audio data. */
404 t
->size
= bytes_per_frame(&t
->format
) * t
->format
.rate
* BUFFER_SECONDS
;
405 t
->buffer
= xmalloc(t
->size
);
408 D(("got format for %s", t
->id
));
414 /** @brief Close the sound device
416 * This is called to deactivate the output device when pausing, and also by the
417 * ALSA backend when changing encoding (in which case the sound device will be
418 * immediately reactivated).
420 static void idle(void) {
422 if(backend
->deactivate
)
423 backend
->deactivate();
428 /** @brief Abandon the current track */
429 static void abandon(void) {
430 struct speaker_message sm
;
433 memset(&sm
, 0, sizeof sm
);
434 sm
.type
= SM_FINISHED
;
435 strcpy(sm
.id
, playing
->id
);
436 speaker_send(1, &sm
, 0);
437 removetrack(playing
->id
);
444 /** @brief Log ALSA parameters */
445 static void log_params(snd_pcm_hw_params_t
*hwparams
,
446 snd_pcm_sw_params_t
*swparams
) {
450 return; /* too verbose */
455 snd_pcm_sw_params_get_silence_size(swparams
, &f
);
456 info("sw silence_size=%lu", (unsigned long)f
);
457 snd_pcm_sw_params_get_silence_threshold(swparams
, &f
);
458 info("sw silence_threshold=%lu", (unsigned long)f
);
459 snd_pcm_sw_params_get_sleep_min(swparams
, &u
);
460 info("sw sleep_min=%lu", (unsigned long)u
);
461 snd_pcm_sw_params_get_start_threshold(swparams
, &f
);
462 info("sw start_threshold=%lu", (unsigned long)f
);
463 snd_pcm_sw_params_get_stop_threshold(swparams
, &f
);
464 info("sw stop_threshold=%lu", (unsigned long)f
);
465 snd_pcm_sw_params_get_xfer_align(swparams
, &f
);
466 info("sw xfer_align=%lu", (unsigned long)f
);
471 /** @brief Enable sound output
473 * Makes sure the sound device is open and has the right sample format. Return
474 * 0 on success and -1 on error.
476 static int activate(void) {
477 /* If we don't know the format yet we cannot start. */
478 if(!playing
->got_format
) {
479 D((" - not got format for %s", playing
->id
));
482 return backend
->activate();
485 /** @brief Check whether the current track has finished
487 * The current track is determined to have finished either if the input stream
488 * eded before the format could be determined (i.e. it is malformed) or the
489 * input is at end of file and there is less than a frame left unplayed. (So
490 * it copes with decoders that crash mid-frame.)
492 static void maybe_finished(void) {
495 && (!playing
->got_format
496 || playing
->used
< bytes_per_frame(&playing
->format
)))
500 /** @brief Start the subprocess for @ref BACKEND_COMMAND */
501 static void fork_cmd(void) {
504 if(cmdfd
!= -1) close(cmdfd
);
508 signal(SIGPIPE
, SIG_DFL
);
512 execl("/bin/sh", "sh", "-c", config
->speaker_command
, (char *)0);
513 fatal(errno
, "error execing /bin/sh");
517 D(("forked cmd %d, fd = %d", cmdpid
, cmdfd
));
520 /** @brief Play up to @p frames frames of audio */
521 static void play(size_t frames
) {
522 size_t avail_frames
, avail_bytes
, written_frames
;
523 ssize_t written_bytes
;
525 /* Make sure the output device is activated */
530 forceplay
= 0; /* Must have called abandon() */
533 D(("play: play %zu/%zu%s %dHz %db %dc", frames
, playing
->used
/ bpf
,
534 playing
->eof ?
" EOF" : "",
535 playing
->format
.rate
,
536 playing
->format
.bits
,
537 playing
->format
.channels
));
538 /* If we haven't got enough bytes yet wait until we have. Exception: when
540 if(playing
->used
< frames
* bpf
&& !playing
->eof
) {
544 /* We have got enough data so don't force play again */
546 /* Figure out how many frames there are available to write */
547 if(playing
->start
+ playing
->used
> playing
->size
)
548 /* The ring buffer is currently wrapped, only play up to the wrap point */
549 avail_bytes
= playing
->size
- playing
->start
;
551 /* The ring buffer is not wrapped, can play the lot */
552 avail_bytes
= playing
->used
;
553 avail_frames
= avail_bytes
/ bpf
;
554 /* Only play up to the requested amount */
555 if(avail_frames
> frames
)
556 avail_frames
= frames
;
560 written_frames
= backend
->play(avail_frames
);
561 written_bytes
= written_frames
* bpf
;
562 /* written_bytes and written_frames had better both be set and correct by
564 playing
->start
+= written_bytes
;
565 playing
->used
-= written_bytes
;
566 playing
->played
+= written_frames
;
567 /* If the pointer is at the end of the buffer (or the buffer is completely
568 * empty) wrap it back to the start. */
569 if(!playing
->used
|| playing
->start
== playing
->size
)
571 frames
-= written_frames
;
574 /* Notify the server what we're up to. */
575 static void report(void) {
576 struct speaker_message sm
;
578 if(playing
&& playing
->buffer
!= (void *)&playing
->format
) {
579 memset(&sm
, 0, sizeof sm
);
580 sm
.type
= paused ? SM_PAUSED
: SM_PLAYING
;
581 strcpy(sm
.id
, playing
->id
);
582 sm
.data
= playing
->played
/ playing
->format
.rate
;
583 speaker_send(1, &sm
, 0);
588 static void reap(int __attribute__((unused
)) sig
) {
593 cmdpid
= waitpid(-1, &st
, WNOHANG
);
595 signal(SIGCHLD
, reap
);
598 static int addfd(int fd
, int events
) {
601 fds
[fdno
].events
= events
;
608 /** @brief ALSA backend initialization */
609 static void alsa_init(void) {
610 info("selected ALSA backend");
613 /** @brief ALSA backend activation */
614 static int alsa_activate(void) {
615 /* If we need to change format then close the current device. */
616 if(pcm
&& !formats_equal(&playing
->format
, &pcm_format
))
619 snd_pcm_hw_params_t
*hwparams
;
620 snd_pcm_sw_params_t
*swparams
;
621 snd_pcm_uframes_t pcm_bufsize
;
623 int sample_format
= 0;
627 if((err
= snd_pcm_open(&pcm
,
629 SND_PCM_STREAM_PLAYBACK
,
630 SND_PCM_NONBLOCK
))) {
631 error(0, "error from snd_pcm_open: %d", err
);
634 snd_pcm_hw_params_alloca(&hwparams
);
635 D(("set up hw params"));
636 if((err
= snd_pcm_hw_params_any(pcm
, hwparams
)) < 0)
637 fatal(0, "error from snd_pcm_hw_params_any: %d", err
);
638 if((err
= snd_pcm_hw_params_set_access(pcm
, hwparams
,
639 SND_PCM_ACCESS_RW_INTERLEAVED
)) < 0)
640 fatal(0, "error from snd_pcm_hw_params_set_access: %d", err
);
641 switch(playing
->format
.bits
) {
643 sample_format
= SND_PCM_FORMAT_S8
;
646 switch(playing
->format
.byte_format
) {
647 case AO_FMT_NATIVE
: sample_format
= SND_PCM_FORMAT_S16
; break;
648 case AO_FMT_LITTLE
: sample_format
= SND_PCM_FORMAT_S16_LE
; break;
649 case AO_FMT_BIG
: sample_format
= SND_PCM_FORMAT_S16_BE
; break;
650 error(0, "unrecognized byte format %d", playing
->format
.byte_format
);
655 error(0, "unsupported sample size %d", playing
->format
.bits
);
658 if((err
= snd_pcm_hw_params_set_format(pcm
, hwparams
,
659 sample_format
)) < 0) {
660 error(0, "error from snd_pcm_hw_params_set_format (%d): %d",
664 rate
= playing
->format
.rate
;
665 if((err
= snd_pcm_hw_params_set_rate_near(pcm
, hwparams
, &rate
, 0)) < 0) {
666 error(0, "error from snd_pcm_hw_params_set_rate (%d): %d",
667 playing
->format
.rate
, err
);
670 if(rate
!= (unsigned)playing
->format
.rate
)
671 info("want rate %d, got %u", playing
->format
.rate
, rate
);
672 if((err
= snd_pcm_hw_params_set_channels(pcm
, hwparams
,
673 playing
->format
.channels
)) < 0) {
674 error(0, "error from snd_pcm_hw_params_set_channels (%d): %d",
675 playing
->format
.channels
, err
);
678 bufsize
= 3 * FRAMES
;
679 pcm_bufsize
= bufsize
;
680 if((err
= snd_pcm_hw_params_set_buffer_size_near(pcm
, hwparams
,
682 fatal(0, "error from snd_pcm_hw_params_set_buffer_size (%d): %d",
684 if(pcm_bufsize
!= 3 * FRAMES
&& pcm_bufsize
!= last_pcm_bufsize
)
685 info("asked for PCM buffer of %d frames, got %d",
686 3 * FRAMES
, (int)pcm_bufsize
);
687 last_pcm_bufsize
= pcm_bufsize
;
688 if((err
= snd_pcm_hw_params(pcm
, hwparams
)) < 0)
689 fatal(0, "error calling snd_pcm_hw_params: %d", err
);
690 D(("set up sw params"));
691 snd_pcm_sw_params_alloca(&swparams
);
692 if((err
= snd_pcm_sw_params_current(pcm
, swparams
)) < 0)
693 fatal(0, "error calling snd_pcm_sw_params_current: %d", err
);
694 if((err
= snd_pcm_sw_params_set_avail_min(pcm
, swparams
, FRAMES
)) < 0)
695 fatal(0, "error calling snd_pcm_sw_params_set_avail_min %d: %d",
697 if((err
= snd_pcm_sw_params(pcm
, swparams
)) < 0)
698 fatal(0, "error calling snd_pcm_sw_params: %d", err
);
699 pcm_format
= playing
->format
;
700 bpf
= bytes_per_frame(&pcm_format
);
701 D(("acquired audio device"));
702 log_params(hwparams
, swparams
);
709 /* We assume the error is temporary and that we'll retry in a bit. */
717 /** @brief Play via ALSA */
718 static size_t alsa_play(size_t frames
) {
719 snd_pcm_sframes_t pcm_written_frames
;
722 pcm_written_frames
= snd_pcm_writei(pcm
,
723 playing
->buffer
+ playing
->start
,
725 D(("actually play %zu frames, wrote %d",
726 frames
, (int)pcm_written_frames
));
727 if(pcm_written_frames
< 0) {
728 switch(pcm_written_frames
) {
729 case -EPIPE
: /* underrun */
730 error(0, "snd_pcm_writei reports underrun");
731 if((err
= snd_pcm_prepare(pcm
)) < 0)
732 fatal(0, "error calling snd_pcm_prepare: %d", err
);
737 fatal(0, "error calling snd_pcm_writei: %d",
738 (int)pcm_written_frames
);
741 return pcm_written_frames
;
744 static int alsa_slots
, alsa_nslots
= -1;
746 /** @brief Fill in poll fd array for ALSA */
747 static void alsa_beforepoll(void) {
748 /* We send sample data to ALSA as fast as it can accept it, relying on
749 * the fact that it has a relatively small buffer to minimize pause
756 alsa_nslots
= snd_pcm_poll_descriptors(pcm
, &fds
[fdno
], NFDS
- fdno
);
758 || !(fds
[alsa_slots
].events
& POLLOUT
))
759 && snd_pcm_state(pcm
) == SND_PCM_STATE_XRUN
) {
760 error(0, "underrun detected after call to snd_pcm_poll_descriptors()");
761 if((err
= snd_pcm_prepare(pcm
)))
762 fatal(0, "error calling snd_pcm_prepare: %d", err
);
765 } while(retry
-- > 0);
770 /** @brief Process poll() results for ALSA */
771 static int alsa_afterpoll(void) {
774 if(alsa_slots
!= -1) {
775 unsigned short alsa_revents
;
777 if((err
= snd_pcm_poll_descriptors_revents(pcm
,
781 fatal(0, "error calling snd_pcm_poll_descriptors_revents: %d", err
);
782 if(alsa_revents
& (POLLOUT
| POLLERR
))
789 /** @brief ALSA deactivation */
790 static void alsa_deactivate(void) {
794 if((err
= snd_pcm_nonblock(pcm
, 0)) < 0)
795 fatal(0, "error calling snd_pcm_nonblock: %d", err
);
802 D(("released audio device"));
807 /** @brief Command backend initialization */
808 static void command_init(void) {
809 info("selected command backend");
813 /** @brief Play to a subprocess */
814 static size_t command_play(size_t frames
) {
815 size_t bytes
= frames
* bpf
;
818 written_bytes
= write(cmdfd
, playing
->buffer
+ playing
->start
, bytes
);
819 D(("actually play %zu bytes, wrote %d",
820 bytes
, written_bytes
));
821 if(written_bytes
< 0) {
824 error(0, "hmm, command died; trying another");
830 fatal(errno
, "error writing to subprocess");
833 return written_bytes
/ bpf
;
836 static int cmdfd_slot
;
838 /** @brief Update poll array for writing to subprocess */
839 static void command_beforepoll(void) {
840 /* We send sample data to the subprocess as fast as it can accept it.
841 * This isn't ideal as pause latency can be very high as a result. */
843 cmdfd_slot
= addfd(cmdfd
, POLLOUT
);
846 /** @brief Process poll() results for subprocess play */
847 static int command_afterpoll(void) {
848 if(cmdfd_slot
!= -1) {
849 if(fds
[cmdfd_slot
].revents
& (POLLOUT
| POLLERR
))
856 /** @brief Command/network backend activation */
857 static int generic_activate(void) {
859 bufsize
= 3 * FRAMES
;
860 bpf
= bytes_per_frame(&config
->sample_format
);
861 D(("acquired audio device"));
867 /** @brief Network backend initialization */
868 static void network_init(void) {
869 struct addrinfo
*res
, *sres
;
870 static const struct addrinfo pref
= {
880 static const struct addrinfo prefbind
= {
890 static const int one
= 1;
891 int sndbuf
, target_sndbuf
= 131072;
893 char *sockname
, *ssockname
;
895 res
= get_address(&config
->broadcast
, &pref
, &sockname
);
897 if(config
->broadcast_from
.n
) {
898 sres
= get_address(&config
->broadcast_from
, &prefbind
, &ssockname
);
902 if((bfd
= socket(res
->ai_family
,
904 res
->ai_protocol
)) < 0)
905 fatal(errno
, "error creating broadcast socket");
906 if(setsockopt(bfd
, SOL_SOCKET
, SO_BROADCAST
, &one
, sizeof one
) < 0)
907 fatal(errno
, "error setting SO_BROADCAST on broadcast socket");
909 if(getsockopt(bfd
, SOL_SOCKET
, SO_SNDBUF
,
911 fatal(errno
, "error getting SO_SNDBUF");
912 if(target_sndbuf
> sndbuf
) {
913 if(setsockopt(bfd
, SOL_SOCKET
, SO_SNDBUF
,
914 &target_sndbuf
, sizeof target_sndbuf
) < 0)
915 error(errno
, "error setting SO_SNDBUF to %d", target_sndbuf
);
917 info("changed socket send buffer size from %d to %d",
918 sndbuf
, target_sndbuf
);
920 info("default socket send buffer is %d",
922 /* We might well want to set additional broadcast- or multicast-related
924 if(sres
&& bind(bfd
, sres
->ai_addr
, sres
->ai_addrlen
) < 0)
925 fatal(errno
, "error binding broadcast socket to %s", ssockname
);
926 if(connect(bfd
, res
->ai_addr
, res
->ai_addrlen
) < 0)
927 fatal(errno
, "error connecting broadcast socket to %s", sockname
);
929 gcry_randomize(&rtp_id
, sizeof rtp_id
, GCRY_STRONG_RANDOM
);
930 info("selected network backend, sending to %s", sockname
);
931 if(config
->sample_format
.byte_format
!= AO_FMT_BIG
) {
932 info("forcing big-endian sample format");
933 config
->sample_format
.byte_format
= AO_FMT_BIG
;
937 /** @brief Play over the network */
938 static size_t network_play(size_t frames
) {
939 struct rtp_header header
;
941 size_t bytes
= frames
* bpf
, written_frames
;
943 /* We transmit using RTP (RFC3550) and attempt to conform to the internet
944 * AVT profile (RFC3551). */
947 /* There may have been a gap. Fix up the RTP time accordingly. */
950 uint64_t target_rtp_time
;
952 /* Find the current time */
953 xgettimeofday(&now
, 0);
954 /* Find the number of microseconds elapsed since rtp_time=0 */
955 delta
= tvsub_us(now
, rtp_time_0
);
956 assert(delta
<= UINT64_MAX
/ 88200);
957 target_rtp_time
= (delta
* playing
->format
.rate
958 * playing
->format
.channels
) / 1000000;
959 /* Overflows at ~6 years uptime with 44100Hz stereo */
961 /* rtp_time is the number of samples we've played. NB that we play
962 * RTP_AHEAD_MS ahead of ourselves, so it may legitimately be ahead of
963 * the value we deduce from time comparison.
965 * Suppose we have 1s track started at t=0, and another track begins to
966 * play at t=2s. Suppose RTP_AHEAD_MS=1000 and 44100Hz stereo. In that
967 * case we'll send 1s of audio as fast as we can, giving rtp_time=88200.
968 * rtp_time stops at this point.
970 * At t=2s we'll have calculated target_rtp_time=176400. In this case we
971 * set rtp_time=176400 and the player can correctly conclude that it
972 * should leave 1s between the tracks.
974 * Suppose instead that the second track arrives at t=0.5s, and that
975 * we've managed to transmit the whole of the first track already. We'll
976 * have target_rtp_time=44100.
978 * The desired behaviour is to play the second track back to back with
979 * first. In this case therefore we do not modify rtp_time.
981 * Is it ever right to reduce rtp_time? No; for that would imply
982 * transmitting packets with overlapping timestamp ranges, which does not
985 if(target_rtp_time
> rtp_time
) {
986 /* More time has elapsed than we've transmitted samples. That implies
987 * we've been 'sending' silence. */
988 info("advancing rtp_time by %"PRIu64
" samples",
989 target_rtp_time
- rtp_time
);
990 rtp_time
= target_rtp_time
;
991 } else if(target_rtp_time
< rtp_time
) {
992 const int64_t samples_ahead
= ((uint64_t)RTP_AHEAD_MS
993 * config
->sample_format
.rate
994 * config
->sample_format
.channels
997 if(target_rtp_time
+ samples_ahead
< rtp_time
) {
998 info("reversing rtp_time by %"PRIu64
" samples",
999 rtp_time
- target_rtp_time
);
1003 header
.vpxcc
= 2 << 6; /* V=2, P=0, X=0, CC=0 */
1004 header
.seq
= htons(rtp_seq
++);
1005 header
.timestamp
= htonl((uint32_t)rtp_time
);
1006 header
.ssrc
= rtp_id
;
1007 header
.mpt
= (idled ?
0x80 : 0x00) | 10;
1008 /* 10 = L16 = 16-bit x 2 x 44100KHz. We ought to deduce this value from
1009 * the sample rate (in a library somewhere so that configuration.c can rule
1010 * out invalid rates).
1013 if(bytes
> NETWORK_BYTES
- sizeof header
) {
1014 bytes
= NETWORK_BYTES
- sizeof header
;
1015 /* Always send a whole number of frames */
1016 bytes
-= bytes
% bpf
;
1018 /* "The RTP clock rate used for generating the RTP timestamp is independent
1019 * of the number of channels and the encoding; it equals the number of
1020 * sampling periods per second. For N-channel encodings, each sampling
1021 * period (say, 1/8000 of a second) generates N samples. (This terminology
1022 * is standard, but somewhat confusing, as the total number of samples
1023 * generated per second is then the sampling rate times the channel
1026 vec
[0].iov_base
= (void *)&header
;
1027 vec
[0].iov_len
= sizeof header
;
1028 vec
[1].iov_base
= playing
->buffer
+ playing
->start
;
1029 vec
[1].iov_len
= bytes
;
1031 written_bytes
= writev(bfd
, vec
, 2);
1032 } while(written_bytes
< 0 && errno
== EINTR
);
1033 if(written_bytes
< 0) {
1034 error(errno
, "error transmitting audio data");
1036 if(audio_errors
== 10)
1037 fatal(0, "too many audio errors");
1041 written_bytes
-= sizeof (struct rtp_header
);
1042 written_frames
= written_bytes
/ bpf
;
1043 /* Advance RTP's notion of the time */
1044 rtp_time
+= written_frames
* playing
->format
.channels
;
1045 return written_frames
;
1048 static int bfd_slot
;
1050 /** @brief Set up poll array for network play */
1051 static void network_beforepoll(void) {
1054 uint64_t target_rtp_time
;
1055 const int64_t samples_ahead
= ((uint64_t)RTP_AHEAD_MS
1056 * config
->sample_format
.rate
1057 * config
->sample_format
.channels
1060 /* If we're starting then initialize the base time */
1062 xgettimeofday(&rtp_time_0
, 0);
1063 /* We send audio data whenever we get RTP_AHEAD seconds or more
1065 xgettimeofday(&now
, 0);
1066 target_us
= tvsub_us(now
, rtp_time_0
);
1067 assert(target_us
<= UINT64_MAX
/ 88200);
1068 target_rtp_time
= (target_us
* config
->sample_format
.rate
1069 * config
->sample_format
.channels
)
1071 if((int64_t)(rtp_time
- target_rtp_time
) < samples_ahead
)
1072 bfd_slot
= addfd(bfd
, POLLOUT
);
1075 /** @brief Process poll() results for network play */
1076 static int network_afterpoll(void) {
1077 if(bfd_slot
!= -1) {
1078 if(fds
[bfd_slot
].revents
& (POLLOUT
| POLLERR
))
1085 /** @brief Table of speaker backends */
1086 static const struct speaker_backend backends
[] = {
1119 { -1, 0, 0, 0, 0, 0, 0, 0 } /* end of list */
1122 /** @brief Main event loop
1124 * This has grown in a rather bizarre and ad-hoc way is very sensitive to
1127 * Firstly the loop is terminated when the parent process exits. Therefore the
1128 * speaker process has the same lifetime as the main server. This and the
1129 * reading of data from decoders is comprehensible enough.
1131 * The playing of audio is more complicated however.
1133 * On the first run through when a track is ready to be played, @ref ready and
1134 * @ref forceplay will both be zero. Therefore @c beforepoll is not called.
1136 * @c afterpoll on the other hand @b is called and will return nonzero. The
1137 * result is that we call @c play(0). This will call activate(), setting
1138 * @ref ready nonzero, but otherwise has no immediate effect.
1140 * We then deal with stdin and the decoders.
1142 * We then reach the second place we might play some audio. @ref forceplay is
1143 * 0 so nothing happens here again.
1145 * On the next iteration through however @ref ready is nonzero, and @ref
1146 * forceplay is 0, so we call @c beforepoll. After the @c poll() we call @c
1147 * afterpoll and actually get some audio played.
1149 * This is surely @b far more complicated than it needs to be!
1151 * If at any call to play(), activate() fails, or if there aren't enough bytes
1152 * in the buffer to satisfy the request, then @ref forceplay is set non-0. On
1153 * the next pass through the event loop @c beforepoll is not called. This
1154 * means that (if none of the other FDs trigger) the @c poll() call will block
1155 * for up to a second. @c afterpoll will return nonzero, since @c beforepoll
1156 * wasn't called, and consequently play() is called with @ref forceplay as its
1159 * The effect is to attempt to restart playing audio - including the activate()
1160 * step, which may have failed at the previous attempt - at least once a second
1161 * after an error has disabled it. The delay prevents busy-waiting on whatever
1162 * condition has rendered the audio device uncooperative.
1164 static void mainloop(void) {
1166 struct speaker_message sm
;
1167 int n
, fd
, stdin_slot
, poke
, timeout
;
1169 while(getppid() != 1) {
1171 /* Always ready for commands from the main server. */
1172 stdin_slot
= addfd(0, POLLIN
);
1173 /* Try to read sample data for the currently playing track if there is
1175 if(playing
&& !playing
->eof
&& playing
->used
< playing
->size
) {
1176 playing
->slot
= addfd(playing
->fd
, POLLIN
);
1179 /* If forceplay is set then wait until it succeeds before waiting on the
1186 /* By default we will wait up to a second before thinking about current
1189 /* We'll break the poll as soon as the underlying sound device is ready for
1191 if(ready
&& !forceplay
)
1192 backend
->beforepoll();
1193 /* If any other tracks don't have a full buffer, try to read sample data
1195 for(t
= tracks
; t
; t
= t
->next
)
1197 if(!t
->eof
&& t
->used
< t
->size
) {
1198 t
->slot
= addfd(t
->fd
, POLLIN
| POLLHUP
);
1202 /* Wait for something interesting to happen */
1203 n
= poll(fds
, fdno
, timeout
);
1205 if(errno
== EINTR
) continue;
1206 fatal(errno
, "error calling poll");
1208 /* Play some sound before doing anything else */
1209 poke
= backend
->afterpoll();
1211 /* Some attempt to play must have failed */
1212 if(playing
&& !paused
)
1215 forceplay
= 0; /* just in case */
1217 /* Perhaps we have a command to process */
1218 if(fds
[stdin_slot
].revents
& POLLIN
) {
1219 n
= speaker_recv(0, &sm
, &fd
);
1223 D(("SM_PREPARE %s %d", sm
.id
, fd
));
1224 if(fd
== -1) fatal(0, "got SM_PREPARE but no file descriptor");
1225 t
= findtrack(sm
.id
, 1);
1229 D(("SM_PLAY %s %d", sm
.id
, fd
));
1230 if(playing
) fatal(0, "got SM_PLAY but already playing something");
1231 t
= findtrack(sm
.id
, 1);
1232 if(fd
!= -1) acquire(t
, fd
);
1252 D(("SM_CANCEL %s", sm
.id
));
1253 t
= removetrack(sm
.id
);
1256 sm
.type
= SM_FINISHED
;
1257 strcpy(sm
.id
, playing
->id
);
1258 speaker_send(1, &sm
, 0);
1263 error(0, "SM_CANCEL for unknown track %s", sm
.id
);
1268 if(config_read()) error(0, "cannot read configuration");
1269 info("reloaded configuration");
1272 error(0, "unknown message type %d", sm
.type
);
1275 /* Read in any buffered data */
1276 for(t
= tracks
; t
; t
= t
->next
)
1277 if(t
->slot
!= -1 && (fds
[t
->slot
].revents
& (POLLIN
| POLLHUP
)))
1279 /* We might be able to play now */
1280 if(ready
&& forceplay
&& playing
&& !paused
)
1282 /* Maybe we finished playing a track somewhere in the above */
1284 /* If we don't need the sound device for now then close it for the benefit
1285 * of anyone else who wants it. */
1286 if((!playing
|| paused
) && ready
)
1288 /* If we've not reported out state for a second do so now. */
1289 if(time(0) > last_report
)
1294 int main(int argc
, char **argv
) {
1298 if(!setlocale(LC_CTYPE
, "")) fatal(errno
, "error calling setlocale");
1299 while((n
= getopt_long(argc
, argv
, "hVc:dD", options
, 0)) >= 0) {
1302 case 'V': version();
1303 case 'c': configfile
= optarg
; break;
1304 case 'd': debugging
= 1; break;
1305 case 'D': debugging
= 0; break;
1306 default: fatal(0, "invalid option");
1309 if(getenv("DISORDER_DEBUG_SPEAKER")) debugging
= 1;
1310 /* If stderr is a TTY then log there, otherwise to syslog. */
1312 openlog(progname
, LOG_PID
, LOG_DAEMON
);
1313 log_default
= &log_syslog
;
1315 if(config_read()) fatal(0, "cannot read configuration");
1316 /* ignore SIGPIPE */
1317 signal(SIGPIPE
, SIG_IGN
);
1319 signal(SIGCHLD
, reap
);
1320 /* set nice value */
1321 xnice(config
->nice_speaker
);
1324 /* make sure we're not root, whatever the config says */
1325 if(getuid() == 0 || geteuid() == 0) fatal(0, "do not run as root");
1326 /* identify the backend used to play */
1327 for(n
= 0; backends
[n
].backend
!= -1; ++n
)
1328 if(backends
[n
].backend
== config
->speaker_backend
)
1330 if(backends
[n
].backend
== -1)
1331 fatal(0, "unsupported backend %d", config
->speaker_backend
);
1332 backend
= &backends
[n
];
1333 /* backend-specific initialization */
1336 info("stopped (parent terminated)");
1345 indent-tabs-mode:nil