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 processs
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 * For the <a href="http://www.alsa-project.org/">ALSA</a> API, 8- and 16- bit
29 * stereo and mono are supported, with any sample rate (within the limits that
30 * 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 * This program deliberately does not use the garbage collector even though it
43 * might be convenient to do so. This is for two reasons. Firstly some sound
44 * APIs use thread threads and we do not want to have to deal with potential
45 * interactions between threading and garbage collection. Secondly this
46 * process needs to be able to respond quickly and this is not compatible with
47 * the collector hanging the program even relatively briefly.
63 #include <sys/select.h>
68 #include <sys/socket.h>
73 #include "configuration.h"
85 #include <alsa/asoundlib.h>
88 #ifdef WORDS_BIGENDIAN
89 # define MACHINE_AO_FMT AO_FMT_BIG
91 # define MACHINE_AO_FMT AO_FMT_LITTLE
94 /** @brief How many seconds of input to buffer
96 * While any given connection has this much audio buffered, no more reads will
97 * be issued for that connection. The decoder will have to wait.
99 #define BUFFER_SECONDS 5
101 #define FRAMES 4096 /* Frame batch size */
103 /** @brief Bytes to send per network packet
105 * Don't make this too big or arithmetic will start to overflow.
107 #define NETWORK_BYTES (1024+sizeof(struct rtp_header))
109 /** @brief Maximum RTP playahead (ms) */
110 #define RTP_AHEAD_MS 1000
112 /** @brief Maximum number of FDs to poll for */
115 /** @brief Track structure
117 * Known tracks are kept in a linked list. Usually there will be at most two
118 * of these but rearranging the queue can cause there to be more.
120 static struct track
{
121 struct track
*next
; /* next track */
122 int fd
; /* input FD */
123 char id
[24]; /* ID */
124 size_t start
, used
; /* start + bytes used */
125 int eof
; /* input is at EOF */
126 int got_format
; /* got format yet? */
127 ao_sample_format format
; /* sample format */
128 unsigned long long played
; /* number of frames played */
129 char *buffer
; /* sample buffer */
130 size_t size
; /* sample buffer size */
131 int slot
; /* poll array slot */
132 } *tracks
, *playing
; /* all tracks + playing track */
134 static time_t last_report
; /* when we last reported */
135 static int paused
; /* pause status */
136 static ao_sample_format pcm_format
; /* current format if aodev != 0 */
137 static size_t bpf
; /* bytes per frame */
138 static struct pollfd fds
[NFDS
]; /* if we need more than that */
139 static int fdno
; /* fd number */
140 static size_t bufsize
; /* buffer size */
142 static snd_pcm_t
*pcm
; /* current pcm handle */
143 static snd_pcm_uframes_t last_pcm_bufsize
; /* last seen buffer size */
145 static int ready
; /* ready to send audio */
146 static int forceplay
; /* frames to force play */
147 static int cmdfd
= -1; /* child process input */
148 static int bfd
= -1; /* broadcast FD */
150 /** @brief RTP timestamp
152 * This counts the number of samples played (NB not the number of frames
155 * The timestamp in the packet header is only 32 bits wide. With 44100Hz
156 * stereo, that only gives about half a day before wrapping, which is not
157 * particularly convenient for certain debugging purposes. Therefore the
158 * timestamp is maintained as a 64-bit integer, giving around six million years
159 * before wrapping, and truncated to 32 bits when transmitting.
161 static uint64_t rtp_time
;
163 /** @brief RTP base timestamp
165 * This is the real time correspoding to an @ref rtp_time of 0. It is used
166 * to recalculate the timestamp after idle periods.
168 static struct timeval rtp_time_0
;
170 static uint16_t rtp_seq
; /* frame sequence number */
171 static uint32_t rtp_id
; /* RTP SSRC */
172 static int idled
; /* set when idled */
173 static int audio_errors
; /* audio error counter */
175 /** @brief Structure of a backend */
176 struct speaker_backend
{
177 /** @brief Which backend this is
179 * @c -1 terminates the list.
183 /** @brief Initialization
185 * Called once at startup.
189 /** @brief Activation
190 * @return 0 on success, non-0 on error
192 * Called to activate the output device.
194 int (*activate
)(void);
197 /** @brief Selected backend */
198 static const struct speaker_backend
*backend
;
200 static const struct option options
[] = {
201 { "help", no_argument
, 0, 'h' },
202 { "version", no_argument
, 0, 'V' },
203 { "config", required_argument
, 0, 'c' },
204 { "debug", no_argument
, 0, 'd' },
205 { "no-debug", no_argument
, 0, 'D' },
209 /* Display usage message and terminate. */
210 static void help(void) {
212 " disorder-speaker [OPTIONS]\n"
214 " --help, -h Display usage message\n"
215 " --version, -V Display version number\n"
216 " --config PATH, -c PATH Set configuration file\n"
217 " --debug, -d Turn on debugging\n"
219 "Speaker process for DisOrder. Not intended to be run\n"
225 /* Display version number and terminate. */
226 static void version(void) {
227 xprintf("disorder-speaker version %s\n", disorder_version_string
);
232 /** @brief Return the number of bytes per frame in @p format */
233 static size_t bytes_per_frame(const ao_sample_format
*format
) {
234 return format
->channels
* format
->bits
/ 8;
237 /** @brief Find track @p id, maybe creating it if not found */
238 static struct track
*findtrack(const char *id
, int create
) {
241 D(("findtrack %s %d", id
, create
));
242 for(t
= tracks
; t
&& strcmp(id
, t
->id
); t
= t
->next
)
245 t
= xmalloc(sizeof *t
);
250 /* The initial input buffer will be the sample format. */
251 t
->buffer
= (void *)&t
->format
;
252 t
->size
= sizeof t
->format
;
257 /** @brief Remove track @p id (but do not destroy it) */
258 static struct track
*removetrack(const char *id
) {
259 struct track
*t
, **tt
;
261 D(("removetrack %s", id
));
262 for(tt
= &tracks
; (t
= *tt
) && strcmp(id
, t
->id
); tt
= &t
->next
)
269 /** @brief Destroy a track */
270 static void destroy(struct track
*t
) {
271 D(("destroy %s", t
->id
));
272 if(t
->fd
!= -1) xclose(t
->fd
);
273 if(t
->buffer
!= (void *)&t
->format
) free(t
->buffer
);
277 /** @brief Notice a new connection */
278 static void acquire(struct track
*t
, int fd
) {
279 D(("acquire %s %d", t
->id
, fd
));
286 /** @brief Return true if A and B denote identical libao formats, else false */
287 static int formats_equal(const ao_sample_format
*a
,
288 const ao_sample_format
*b
) {
289 return (a
->bits
== b
->bits
290 && a
->rate
== b
->rate
291 && a
->channels
== b
->channels
292 && a
->byte_format
== b
->byte_format
);
295 /** @brief Compute arguments to sox */
296 static void soxargs(const char ***pp
, char **qq
, ao_sample_format
*ao
) {
301 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-r%d", ao
->rate
); *qq
+= n
+ 1;
302 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-c%d", ao
->channels
); *qq
+= n
+ 1;
303 /* sox 12.17.9 insists on -b etc; CVS sox insists on -<n> etc; both are
305 switch(config
->sox_generation
) {
308 && ao
->byte_format
!= AO_FMT_NATIVE
309 && ao
->byte_format
!= MACHINE_AO_FMT
) {
313 case 8: *(*pp
)++ = "-b"; break;
314 case 16: *(*pp
)++ = "-w"; break;
315 case 32: *(*pp
)++ = "-l"; break;
316 case 64: *(*pp
)++ = "-d"; break;
317 default: fatal(0, "cannot handle sample size %d", (int)ao
->bits
);
321 switch(ao
->byte_format
) {
322 case AO_FMT_NATIVE
: break;
323 case AO_FMT_BIG
: *(*pp
)++ = "-B"; break;
324 case AO_FMT_LITTLE
: *(*pp
)++ = "-L"; break;
326 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-%d", ao
->bits
/8); *qq
+= n
+ 1;
331 /** @brief Enable format translation
333 * If necessary, replaces a tracks inbound file descriptor with one connected
334 * to a sox invocation, which performs the required translation.
336 static void enable_translation(struct track
*t
) {
337 switch(config
->speaker_backend
) {
338 case BACKEND_COMMAND
:
339 case BACKEND_NETWORK
:
340 /* These backends need a specific sample format */
346 if(!formats_equal(&t
->format
, &config
->sample_format
)) {
347 char argbuf
[1024], *q
= argbuf
;
348 const char *av
[18], **pp
= av
;
353 soxargs(&pp
, &q
, &t
->format
);
355 soxargs(&pp
, &q
, &config
->sample_format
);
359 for(pp
= av
; *pp
; pp
++)
360 D(("sox arg[%d] = %s", pp
- av
, *pp
));
366 signal(SIGPIPE
, SIG_DFL
);
368 xdup2(soxpipe
[1], 1);
369 fcntl(0, F_SETFL
, fcntl(0, F_GETFL
) & ~O_NONBLOCK
);
373 execvp("sox", (char **)av
);
376 D(("forking sox for format conversion (kid = %d)", soxkid
));
380 t
->format
= config
->sample_format
;
384 /** @brief Read data into a sample buffer
385 * @param t Pointer to track
386 * @return 0 on success, -1 on EOF
388 * This is effectively the read callback on @c t->fd.
390 static int fill(struct track
*t
) {
394 D(("fill %s: eof=%d used=%zu size=%zu got_format=%d",
395 t
->id
, t
->eof
, t
->used
, t
->size
, t
->got_format
));
396 if(t
->eof
) return -1;
397 if(t
->used
< t
->size
) {
398 /* there is room left in the buffer */
399 where
= (t
->start
+ t
->used
) % t
->size
;
401 /* We are reading audio data, get as much as we can */
402 if(where
>= t
->start
) left
= t
->size
- where
;
403 else left
= t
->start
- where
;
405 /* We are still waiting for the format, only get that */
406 left
= sizeof (ao_sample_format
) - t
->used
;
408 n
= read(t
->fd
, t
->buffer
+ where
, left
);
409 } while(n
< 0 && errno
== EINTR
);
411 if(errno
!= EAGAIN
) fatal(errno
, "error reading sample stream");
415 D(("fill %s: eof detected", t
->id
));
420 if(!t
->got_format
&& t
->used
>= sizeof (ao_sample_format
)) {
421 assert(t
->used
== sizeof (ao_sample_format
));
422 /* Check that our assumptions are met. */
423 if(t
->format
.bits
& 7)
424 fatal(0, "bits per sample not a multiple of 8");
425 /* If the input format is unsuitable, arrange to translate it */
426 enable_translation(t
);
427 /* Make a new buffer for audio data. */
428 t
->size
= bytes_per_frame(&t
->format
) * t
->format
.rate
* BUFFER_SECONDS
;
429 t
->buffer
= xmalloc(t
->size
);
432 D(("got format for %s", t
->id
));
438 /** @brief Close the sound device */
439 static void idle(void) {
442 if(config
->speaker_backend
== BACKEND_ALSA
&& pcm
) {
445 if((err
= snd_pcm_nonblock(pcm
, 0)) < 0)
446 fatal(0, "error calling snd_pcm_nonblock: %d", err
);
453 D(("released audio device"));
460 /** @brief Abandon the current track */
461 static void abandon(void) {
462 struct speaker_message sm
;
465 memset(&sm
, 0, sizeof sm
);
466 sm
.type
= SM_FINISHED
;
467 strcpy(sm
.id
, playing
->id
);
468 speaker_send(1, &sm
, 0);
469 removetrack(playing
->id
);
476 /** @brief Log ALSA parameters */
477 static void log_params(snd_pcm_hw_params_t
*hwparams
,
478 snd_pcm_sw_params_t
*swparams
) {
482 return; /* too verbose */
487 snd_pcm_sw_params_get_silence_size(swparams
, &f
);
488 info("sw silence_size=%lu", (unsigned long)f
);
489 snd_pcm_sw_params_get_silence_threshold(swparams
, &f
);
490 info("sw silence_threshold=%lu", (unsigned long)f
);
491 snd_pcm_sw_params_get_sleep_min(swparams
, &u
);
492 info("sw sleep_min=%lu", (unsigned long)u
);
493 snd_pcm_sw_params_get_start_threshold(swparams
, &f
);
494 info("sw start_threshold=%lu", (unsigned long)f
);
495 snd_pcm_sw_params_get_stop_threshold(swparams
, &f
);
496 info("sw stop_threshold=%lu", (unsigned long)f
);
497 snd_pcm_sw_params_get_xfer_align(swparams
, &f
);
498 info("sw xfer_align=%lu", (unsigned long)f
);
503 /** @brief Enable sound output
505 * Makes sure the sound device is open and has the right sample format. Return
506 * 0 on success and -1 on error.
508 static int activate(void) {
509 /* If we don't know the format yet we cannot start. */
510 if(!playing
->got_format
) {
511 D((" - not got format for %s", playing
->id
));
514 return backend
->activate();
517 /* Check to see whether the current track has finished playing */
518 static void maybe_finished(void) {
521 && (!playing
->got_format
522 || playing
->used
< bytes_per_frame(&playing
->format
)))
526 static void fork_cmd(void) {
529 if(cmdfd
!= -1) close(cmdfd
);
533 signal(SIGPIPE
, SIG_DFL
);
537 execl("/bin/sh", "sh", "-c", config
->speaker_command
, (char *)0);
538 fatal(errno
, "error execing /bin/sh");
542 D(("forked cmd %d, fd = %d", cmdpid
, cmdfd
));
545 static void play(size_t frames
) {
546 size_t avail_bytes
, write_bytes
, written_frames
;
547 ssize_t written_bytes
;
548 struct rtp_header header
;
555 forceplay
= 0; /* Must have called abandon() */
558 D(("play: play %zu/%zu%s %dHz %db %dc", frames
, playing
->used
/ bpf
,
559 playing
->eof ?
" EOF" : "",
560 playing
->format
.rate
,
561 playing
->format
.bits
,
562 playing
->format
.channels
));
563 /* If we haven't got enough bytes yet wait until we have. Exception: when
565 if(playing
->used
< frames
* bpf
&& !playing
->eof
) {
569 /* We have got enough data so don't force play again */
571 /* Figure out how many frames there are available to write */
572 if(playing
->start
+ playing
->used
> playing
->size
)
573 avail_bytes
= playing
->size
- playing
->start
;
575 avail_bytes
= playing
->used
;
577 switch(config
->speaker_backend
) {
580 snd_pcm_sframes_t pcm_written_frames
;
584 avail_frames
= avail_bytes
/ bpf
;
585 if(avail_frames
> frames
)
586 avail_frames
= frames
;
589 pcm_written_frames
= snd_pcm_writei(pcm
,
590 playing
->buffer
+ playing
->start
,
592 D(("actually play %zu frames, wrote %d",
593 avail_frames
, (int)pcm_written_frames
));
594 if(pcm_written_frames
< 0) {
595 switch(pcm_written_frames
) {
596 case -EPIPE
: /* underrun */
597 error(0, "snd_pcm_writei reports underrun");
598 if((err
= snd_pcm_prepare(pcm
)) < 0)
599 fatal(0, "error calling snd_pcm_prepare: %d", err
);
604 fatal(0, "error calling snd_pcm_writei: %d",
605 (int)pcm_written_frames
);
608 written_frames
= pcm_written_frames
;
609 written_bytes
= written_frames
* bpf
;
613 case BACKEND_COMMAND
:
614 if(avail_bytes
> frames
* bpf
)
615 avail_bytes
= frames
* bpf
;
616 written_bytes
= write(cmdfd
, playing
->buffer
+ playing
->start
,
618 D(("actually play %zu bytes, wrote %d",
619 avail_bytes
, (int)written_bytes
));
620 if(written_bytes
< 0) {
623 error(0, "hmm, command died; trying another");
630 written_frames
= written_bytes
/ bpf
; /* good enough */
632 case BACKEND_NETWORK
:
633 /* We transmit using RTP (RFC3550) and attempt to conform to the internet
634 * AVT profile (RFC3551). */
637 /* There may have been a gap. Fix up the RTP time accordingly. */
640 uint64_t target_rtp_time
;
642 /* Find the current time */
643 xgettimeofday(&now
, 0);
644 /* Find the number of microseconds elapsed since rtp_time=0 */
645 delta
= tvsub_us(now
, rtp_time_0
);
646 assert(delta
<= UINT64_MAX
/ 88200);
647 target_rtp_time
= (delta
* playing
->format
.rate
648 * playing
->format
.channels
) / 1000000;
649 /* Overflows at ~6 years uptime with 44100Hz stereo */
651 /* rtp_time is the number of samples we've played. NB that we play
652 * RTP_AHEAD_MS ahead of ourselves, so it may legitimately be ahead of
653 * the value we deduce from time comparison.
655 * Suppose we have 1s track started at t=0, and another track begins to
656 * play at t=2s. Suppose RTP_AHEAD_MS=1000 and 44100Hz stereo. In that
657 * case we'll send 1s of audio as fast as we can, giving rtp_time=88200.
658 * rtp_time stops at this point.
660 * At t=2s we'll have calculated target_rtp_time=176400. In this case we
661 * set rtp_time=176400 and the player can correctly conclude that it
662 * should leave 1s between the tracks.
664 * Suppose instead that the second track arrives at t=0.5s, and that
665 * we've managed to transmit the whole of the first track already. We'll
666 * have target_rtp_time=44100.
668 * The desired behaviour is to play the second track back to back with
669 * first. In this case therefore we do not modify rtp_time.
671 * Is it ever right to reduce rtp_time? No; for that would imply
672 * transmitting packets with overlapping timestamp ranges, which does not
675 if(target_rtp_time
> rtp_time
) {
676 /* More time has elapsed than we've transmitted samples. That implies
677 * we've been 'sending' silence. */
678 info("advancing rtp_time by %"PRIu64
" samples",
679 target_rtp_time
- rtp_time
);
680 rtp_time
= target_rtp_time
;
681 } else if(target_rtp_time
< rtp_time
) {
682 const int64_t samples_ahead
= ((uint64_t)RTP_AHEAD_MS
683 * config
->sample_format
.rate
684 * config
->sample_format
.channels
687 if(target_rtp_time
+ samples_ahead
< rtp_time
) {
688 info("reversing rtp_time by %"PRIu64
" samples",
689 rtp_time
- target_rtp_time
);
693 header
.vpxcc
= 2 << 6; /* V=2, P=0, X=0, CC=0 */
694 header
.seq
= htons(rtp_seq
++);
695 header
.timestamp
= htonl((uint32_t)rtp_time
);
696 header
.ssrc
= rtp_id
;
697 header
.mpt
= (idled ?
0x80 : 0x00) | 10;
698 /* 10 = L16 = 16-bit x 2 x 44100KHz. We ought to deduce this value from
699 * the sample rate (in a library somewhere so that configuration.c can rule
700 * out invalid rates).
703 if(avail_bytes
> NETWORK_BYTES
- sizeof header
) {
704 avail_bytes
= NETWORK_BYTES
- sizeof header
;
705 /* Always send a whole number of frames */
706 avail_bytes
-= avail_bytes
% bpf
;
708 /* "The RTP clock rate used for generating the RTP timestamp is independent
709 * of the number of channels and the encoding; it equals the number of
710 * sampling periods per second. For N-channel encodings, each sampling
711 * period (say, 1/8000 of a second) generates N samples. (This terminology
712 * is standard, but somewhat confusing, as the total number of samples
713 * generated per second is then the sampling rate times the channel
716 write_bytes
= avail_bytes
;
718 vec
[0].iov_base
= (void *)&header
;
719 vec
[0].iov_len
= sizeof header
;
720 vec
[1].iov_base
= playing
->buffer
+ playing
->start
;
721 vec
[1].iov_len
= avail_bytes
;
723 written_bytes
= writev(bfd
,
726 } while(written_bytes
< 0 && errno
== EINTR
);
727 if(written_bytes
< 0) {
728 error(errno
, "error transmitting audio data");
730 if(audio_errors
== 10)
731 fatal(0, "too many audio errors");
736 written_bytes
= avail_bytes
;
737 written_frames
= written_bytes
/ bpf
;
738 /* Advance RTP's notion of the time */
739 rtp_time
+= written_frames
* playing
->format
.channels
;
744 /* written_bytes and written_frames had better both be set and correct by
746 playing
->start
+= written_bytes
;
747 playing
->used
-= written_bytes
;
748 playing
->played
+= written_frames
;
749 /* If the pointer is at the end of the buffer (or the buffer is completely
750 * empty) wrap it back to the start. */
751 if(!playing
->used
|| playing
->start
== playing
->size
)
753 frames
-= written_frames
;
756 /* Notify the server what we're up to. */
757 static void report(void) {
758 struct speaker_message sm
;
760 if(playing
&& playing
->buffer
!= (void *)&playing
->format
) {
761 memset(&sm
, 0, sizeof sm
);
762 sm
.type
= paused ? SM_PAUSED
: SM_PLAYING
;
763 strcpy(sm
.id
, playing
->id
);
764 sm
.data
= playing
->played
/ playing
->format
.rate
;
765 speaker_send(1, &sm
, 0);
770 static void reap(int __attribute__((unused
)) sig
) {
775 cmdpid
= waitpid(-1, &st
, WNOHANG
);
777 signal(SIGCHLD
, reap
);
780 static int addfd(int fd
, int events
) {
783 fds
[fdno
].events
= events
;
790 /** @brief ALSA backend initialization */
791 static void alsa_init(void) {
792 info("selected ALSA backend");
795 /** @brief ALSA backend activation */
796 static int alsa_activate(void) {
797 /* If we need to change format then close the current device. */
798 if(pcm
&& !formats_equal(&playing
->format
, &pcm_format
))
801 snd_pcm_hw_params_t
*hwparams
;
802 snd_pcm_sw_params_t
*swparams
;
803 snd_pcm_uframes_t pcm_bufsize
;
805 int sample_format
= 0;
809 if((err
= snd_pcm_open(&pcm
,
811 SND_PCM_STREAM_PLAYBACK
,
812 SND_PCM_NONBLOCK
))) {
813 error(0, "error from snd_pcm_open: %d", err
);
816 snd_pcm_hw_params_alloca(&hwparams
);
817 D(("set up hw params"));
818 if((err
= snd_pcm_hw_params_any(pcm
, hwparams
)) < 0)
819 fatal(0, "error from snd_pcm_hw_params_any: %d", err
);
820 if((err
= snd_pcm_hw_params_set_access(pcm
, hwparams
,
821 SND_PCM_ACCESS_RW_INTERLEAVED
)) < 0)
822 fatal(0, "error from snd_pcm_hw_params_set_access: %d", err
);
823 switch(playing
->format
.bits
) {
825 sample_format
= SND_PCM_FORMAT_S8
;
828 switch(playing
->format
.byte_format
) {
829 case AO_FMT_NATIVE
: sample_format
= SND_PCM_FORMAT_S16
; break;
830 case AO_FMT_LITTLE
: sample_format
= SND_PCM_FORMAT_S16_LE
; break;
831 case AO_FMT_BIG
: sample_format
= SND_PCM_FORMAT_S16_BE
; break;
832 error(0, "unrecognized byte format %d", playing
->format
.byte_format
);
837 error(0, "unsupported sample size %d", playing
->format
.bits
);
840 if((err
= snd_pcm_hw_params_set_format(pcm
, hwparams
,
841 sample_format
)) < 0) {
842 error(0, "error from snd_pcm_hw_params_set_format (%d): %d",
846 rate
= playing
->format
.rate
;
847 if((err
= snd_pcm_hw_params_set_rate_near(pcm
, hwparams
, &rate
, 0)) < 0) {
848 error(0, "error from snd_pcm_hw_params_set_rate (%d): %d",
849 playing
->format
.rate
, err
);
852 if(rate
!= (unsigned)playing
->format
.rate
)
853 info("want rate %d, got %u", playing
->format
.rate
, rate
);
854 if((err
= snd_pcm_hw_params_set_channels(pcm
, hwparams
,
855 playing
->format
.channels
)) < 0) {
856 error(0, "error from snd_pcm_hw_params_set_channels (%d): %d",
857 playing
->format
.channels
, err
);
860 bufsize
= 3 * FRAMES
;
861 pcm_bufsize
= bufsize
;
862 if((err
= snd_pcm_hw_params_set_buffer_size_near(pcm
, hwparams
,
864 fatal(0, "error from snd_pcm_hw_params_set_buffer_size (%d): %d",
866 if(pcm_bufsize
!= 3 * FRAMES
&& pcm_bufsize
!= last_pcm_bufsize
)
867 info("asked for PCM buffer of %d frames, got %d",
868 3 * FRAMES
, (int)pcm_bufsize
);
869 last_pcm_bufsize
= pcm_bufsize
;
870 if((err
= snd_pcm_hw_params(pcm
, hwparams
)) < 0)
871 fatal(0, "error calling snd_pcm_hw_params: %d", err
);
872 D(("set up sw params"));
873 snd_pcm_sw_params_alloca(&swparams
);
874 if((err
= snd_pcm_sw_params_current(pcm
, swparams
)) < 0)
875 fatal(0, "error calling snd_pcm_sw_params_current: %d", err
);
876 if((err
= snd_pcm_sw_params_set_avail_min(pcm
, swparams
, FRAMES
)) < 0)
877 fatal(0, "error calling snd_pcm_sw_params_set_avail_min %d: %d",
879 if((err
= snd_pcm_sw_params(pcm
, swparams
)) < 0)
880 fatal(0, "error calling snd_pcm_sw_params: %d", err
);
881 pcm_format
= playing
->format
;
882 bpf
= bytes_per_frame(&pcm_format
);
883 D(("acquired audio device"));
884 log_params(hwparams
, swparams
);
891 /* We assume the error is temporary and that we'll retry in a bit. */
900 /** @brief Command backend initialization */
901 static void command_init(void) {
902 info("selected command backend");
906 /** @brief Command backend activation */
907 static int command_activate(void) {
909 pcm_format
= config
->sample_format
;
910 bufsize
= 3 * FRAMES
;
911 bpf
= bytes_per_frame(&config
->sample_format
);
912 D(("acquired audio device"));
918 /** @brief Network backend initialization */
919 static void network_init(void) {
920 struct addrinfo
*res
, *sres
;
921 static const struct addrinfo pref
= {
931 static const struct addrinfo prefbind
= {
941 static const int one
= 1;
942 int sndbuf
, target_sndbuf
= 131072;
944 char *sockname
, *ssockname
;
946 res
= get_address(&config
->broadcast
, &pref
, &sockname
);
948 if(config
->broadcast_from
.n
) {
949 sres
= get_address(&config
->broadcast_from
, &prefbind
, &ssockname
);
953 if((bfd
= socket(res
->ai_family
,
955 res
->ai_protocol
)) < 0)
956 fatal(errno
, "error creating broadcast socket");
957 if(setsockopt(bfd
, SOL_SOCKET
, SO_BROADCAST
, &one
, sizeof one
) < 0)
958 fatal(errno
, "error setting SO_BROADCAST on broadcast socket");
960 if(getsockopt(bfd
, SOL_SOCKET
, SO_SNDBUF
,
962 fatal(errno
, "error getting SO_SNDBUF");
963 if(target_sndbuf
> sndbuf
) {
964 if(setsockopt(bfd
, SOL_SOCKET
, SO_SNDBUF
,
965 &target_sndbuf
, sizeof target_sndbuf
) < 0)
966 error(errno
, "error setting SO_SNDBUF to %d", target_sndbuf
);
968 info("changed socket send buffer size from %d to %d",
969 sndbuf
, target_sndbuf
);
971 info("default socket send buffer is %d",
973 /* We might well want to set additional broadcast- or multicast-related
975 if(sres
&& bind(bfd
, sres
->ai_addr
, sres
->ai_addrlen
) < 0)
976 fatal(errno
, "error binding broadcast socket to %s", ssockname
);
977 if(connect(bfd
, res
->ai_addr
, res
->ai_addrlen
) < 0)
978 fatal(errno
, "error connecting broadcast socket to %s", sockname
);
980 gcry_randomize(&rtp_id
, sizeof rtp_id
, GCRY_STRONG_RANDOM
);
981 info("selected network backend, sending to %s", sockname
);
982 if(config
->sample_format
.byte_format
!= AO_FMT_BIG
) {
983 info("forcing big-endian sample format");
984 config
->sample_format
.byte_format
= AO_FMT_BIG
;
988 /** @brief Network backend activation */
989 static int network_activate(void) {
991 pcm_format
= config
->sample_format
;
992 bufsize
= 3 * FRAMES
;
993 bpf
= bytes_per_frame(&config
->sample_format
);
994 D(("acquired audio device"));
1000 /** @brief Table of speaker backends */
1001 static const struct speaker_backend backends
[] = {
1022 int main(int argc
, char **argv
) {
1023 int n
, fd
, stdin_slot
, alsa_slots
, cmdfd_slot
, bfd_slot
, poke
, timeout
;
1025 struct speaker_message sm
;
1027 int alsa_nslots
= -1, err
;
1031 if(!setlocale(LC_CTYPE
, "")) fatal(errno
, "error calling setlocale");
1032 while((n
= getopt_long(argc
, argv
, "hVc:dD", options
, 0)) >= 0) {
1035 case 'V': version();
1036 case 'c': configfile
= optarg
; break;
1037 case 'd': debugging
= 1; break;
1038 case 'D': debugging
= 0; break;
1039 default: fatal(0, "invalid option");
1042 if(getenv("DISORDER_DEBUG_SPEAKER")) debugging
= 1;
1043 /* If stderr is a TTY then log there, otherwise to syslog. */
1045 openlog(progname
, LOG_PID
, LOG_DAEMON
);
1046 log_default
= &log_syslog
;
1048 if(config_read()) fatal(0, "cannot read configuration");
1049 /* ignore SIGPIPE */
1050 signal(SIGPIPE
, SIG_IGN
);
1052 signal(SIGCHLD
, reap
);
1053 /* set nice value */
1054 xnice(config
->nice_speaker
);
1057 /* make sure we're not root, whatever the config says */
1058 if(getuid() == 0 || geteuid() == 0) fatal(0, "do not run as root");
1059 /* identify the backend used to play */
1060 for(n
= 0; backends
[n
].backend
!= -1; ++n
)
1061 if(backends
[n
].backend
== config
->speaker_backend
)
1063 if(backends
[n
].backend
== -1)
1064 fatal(0, "unsupported backend %d", config
->speaker_backend
);
1065 backend
= &backends
[n
];
1066 /* backend-specific initialization */
1068 while(getppid() != 1) {
1070 /* Always ready for commands from the main server. */
1071 stdin_slot
= addfd(0, POLLIN
);
1072 /* Try to read sample data for the currently playing track if there is
1074 if(playing
&& !playing
->eof
&& playing
->used
< playing
->size
) {
1075 playing
->slot
= addfd(playing
->fd
, POLLIN
);
1078 /* If forceplay is set then wait until it succeeds before waiting on the
1083 /* By default we will wait up to a second before thinking about current
1086 if(ready
&& !forceplay
) {
1087 switch(config
->speaker_backend
) {
1088 case BACKEND_COMMAND
:
1089 /* We send sample data to the subprocess as fast as it can accept it.
1090 * This isn't ideal as pause latency can be very high as a result. */
1092 cmdfd_slot
= addfd(cmdfd
, POLLOUT
);
1094 case BACKEND_NETWORK
: {
1097 uint64_t target_rtp_time
;
1098 const int64_t samples_ahead
= ((uint64_t)RTP_AHEAD_MS
1099 * config
->sample_format
.rate
1100 * config
->sample_format
.channels
1103 static unsigned logit
;
1106 /* If we're starting then initialize the base time */
1108 xgettimeofday(&rtp_time_0
, 0);
1109 /* We send audio data whenever we get RTP_AHEAD seconds or more
1111 xgettimeofday(&now
, 0);
1112 target_us
= tvsub_us(now
, rtp_time_0
);
1113 assert(target_us
<= UINT64_MAX
/ 88200);
1114 target_rtp_time
= (target_us
* config
->sample_format
.rate
1115 * config
->sample_format
.channels
)
1119 /* TODO remove logging guff */
1120 if(!(logit
++ & 1023))
1121 info("rtp_time %llu target %llu difference %lld [%lld]",
1122 rtp_time
, target_rtp_time
,
1123 rtp_time
- target_rtp_time
,
1126 if((int64_t)(rtp_time
- target_rtp_time
) < samples_ahead
)
1127 bfd_slot
= addfd(bfd
, POLLOUT
);
1131 case BACKEND_ALSA
: {
1132 /* We send sample data to ALSA as fast as it can accept it, relying on
1133 * the fact that it has a relatively small buffer to minimize pause
1140 alsa_nslots
= snd_pcm_poll_descriptors(pcm
, &fds
[fdno
], NFDS
- fdno
);
1141 if((alsa_nslots
<= 0
1142 || !(fds
[alsa_slots
].events
& POLLOUT
))
1143 && snd_pcm_state(pcm
) == SND_PCM_STATE_XRUN
) {
1144 error(0, "underrun detected after call to snd_pcm_poll_descriptors()");
1145 if((err
= snd_pcm_prepare(pcm
)))
1146 fatal(0, "error calling snd_pcm_prepare: %d", err
);
1149 } while(retry
-- > 0);
1150 if(alsa_nslots
>= 0)
1151 fdno
+= alsa_nslots
;
1156 assert(!"unknown backend");
1159 /* If any other tracks don't have a full buffer, try to read sample data
1161 for(t
= tracks
; t
; t
= t
->next
)
1163 if(!t
->eof
&& t
->used
< t
->size
) {
1164 t
->slot
= addfd(t
->fd
, POLLIN
| POLLHUP
);
1168 /* Wait for something interesting to happen */
1169 n
= poll(fds
, fdno
, timeout
);
1171 if(errno
== EINTR
) continue;
1172 fatal(errno
, "error calling poll");
1174 /* Play some sound before doing anything else */
1176 switch(config
->speaker_backend
) {
1179 if(alsa_slots
!= -1) {
1180 unsigned short alsa_revents
;
1182 if((err
= snd_pcm_poll_descriptors_revents(pcm
,
1185 &alsa_revents
)) < 0)
1186 fatal(0, "error calling snd_pcm_poll_descriptors_revents: %d", err
);
1187 if(alsa_revents
& (POLLOUT
| POLLERR
))
1193 case BACKEND_COMMAND
:
1194 if(cmdfd_slot
!= -1) {
1195 if(fds
[cmdfd_slot
].revents
& (POLLOUT
| POLLERR
))
1200 case BACKEND_NETWORK
:
1201 if(bfd_slot
!= -1) {
1202 if(fds
[bfd_slot
].revents
& (POLLOUT
| POLLERR
))
1209 /* Some attempt to play must have failed */
1210 if(playing
&& !paused
)
1213 forceplay
= 0; /* just in case */
1215 /* Perhaps we have a command to process */
1216 if(fds
[stdin_slot
].revents
& POLLIN
) {
1217 n
= speaker_recv(0, &sm
, &fd
);
1221 D(("SM_PREPARE %s %d", sm
.id
, fd
));
1222 if(fd
== -1) fatal(0, "got SM_PREPARE but no file descriptor");
1223 t
= findtrack(sm
.id
, 1);
1227 D(("SM_PLAY %s %d", sm
.id
, fd
));
1228 if(playing
) fatal(0, "got SM_PLAY but already playing something");
1229 t
= findtrack(sm
.id
, 1);
1230 if(fd
!= -1) acquire(t
, fd
);
1250 D(("SM_CANCEL %s", sm
.id
));
1251 t
= removetrack(sm
.id
);
1254 sm
.type
= SM_FINISHED
;
1255 strcpy(sm
.id
, playing
->id
);
1256 speaker_send(1, &sm
, 0);
1261 error(0, "SM_CANCEL for unknown track %s", sm
.id
);
1266 if(config_read()) error(0, "cannot read configuration");
1267 info("reloaded configuration");
1270 error(0, "unknown message type %d", sm
.type
);
1273 /* Read in any buffered data */
1274 for(t
= tracks
; t
; t
= t
->next
)
1275 if(t
->slot
!= -1 && (fds
[t
->slot
].revents
& (POLLIN
| POLLHUP
)))
1277 /* We might be able to play now */
1278 if(ready
&& forceplay
&& playing
&& !paused
)
1280 /* Maybe we finished playing a track somewhere in the above */
1282 /* If we don't need the sound device for now then close it for the benefit
1283 * of anyone else who wants it. */
1284 if((!playing
|| paused
) && ready
)
1286 /* If we've not reported out state for a second do so now. */
1287 if(time(0) > last_report
)
1290 info("stopped (parent terminated)");
1299 indent-tabs-mode:nil