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 /** @brief The current PCM handle */
143 static snd_pcm_t
*pcm
;
144 static snd_pcm_uframes_t last_pcm_bufsize
; /* last seen buffer size */
147 /** @brief Ready to send audio
149 * This is set when the destination is ready to receive audio. Generally
150 * this implies that the sound device is open. In the ALSA backend it
151 * does @b not necessarily imply that is has the right sample format.
155 static int forceplay
; /* frames to force play */
156 static int cmdfd
= -1; /* child process input */
157 static int bfd
= -1; /* broadcast FD */
159 /** @brief RTP timestamp
161 * This counts the number of samples played (NB not the number of frames
164 * The timestamp in the packet header is only 32 bits wide. With 44100Hz
165 * stereo, that only gives about half a day before wrapping, which is not
166 * particularly convenient for certain debugging purposes. Therefore the
167 * timestamp is maintained as a 64-bit integer, giving around six million years
168 * before wrapping, and truncated to 32 bits when transmitting.
170 static uint64_t rtp_time
;
172 /** @brief RTP base timestamp
174 * This is the real time correspoding to an @ref rtp_time of 0. It is used
175 * to recalculate the timestamp after idle periods.
177 static struct timeval rtp_time_0
;
179 static uint16_t rtp_seq
; /* frame sequence number */
180 static uint32_t rtp_id
; /* RTP SSRC */
181 static int idled
; /* set when idled */
182 static int audio_errors
; /* audio error counter */
184 /** @brief Structure of a backend */
185 struct speaker_backend
{
186 /** @brief Which backend this is
188 * @c -1 terminates the list.
192 /** @brief Initialization
194 * Called once at startup. This is responsible for one-time setup
195 * operations, for instance opening a network socket to transmit to.
197 * When writing to a native sound API this might @b not imply opening the
198 * native sound device - that might be done by @c activate below.
202 /** @brief Activation
203 * @return 0 on success, non-0 on error
205 * Called to activate the output device.
207 * After this function succeeds, @ref ready should be non-0. As well as
208 * opening the audio device, this function is responsible for reconfiguring
209 * if it necessary to cope with different samples formats (for backends that
210 * don't demand a single fixed sample format for the lifetime of the server).
212 int (*activate
)(void);
215 /** @brief Selected backend */
216 static const struct speaker_backend
*backend
;
218 static const struct option options
[] = {
219 { "help", no_argument
, 0, 'h' },
220 { "version", no_argument
, 0, 'V' },
221 { "config", required_argument
, 0, 'c' },
222 { "debug", no_argument
, 0, 'd' },
223 { "no-debug", no_argument
, 0, 'D' },
227 /* Display usage message and terminate. */
228 static void help(void) {
230 " disorder-speaker [OPTIONS]\n"
232 " --help, -h Display usage message\n"
233 " --version, -V Display version number\n"
234 " --config PATH, -c PATH Set configuration file\n"
235 " --debug, -d Turn on debugging\n"
237 "Speaker process for DisOrder. Not intended to be run\n"
243 /* Display version number and terminate. */
244 static void version(void) {
245 xprintf("disorder-speaker version %s\n", disorder_version_string
);
250 /** @brief Return the number of bytes per frame in @p format */
251 static size_t bytes_per_frame(const ao_sample_format
*format
) {
252 return format
->channels
* format
->bits
/ 8;
255 /** @brief Find track @p id, maybe creating it if not found */
256 static struct track
*findtrack(const char *id
, int create
) {
259 D(("findtrack %s %d", id
, create
));
260 for(t
= tracks
; t
&& strcmp(id
, t
->id
); t
= t
->next
)
263 t
= xmalloc(sizeof *t
);
268 /* The initial input buffer will be the sample format. */
269 t
->buffer
= (void *)&t
->format
;
270 t
->size
= sizeof t
->format
;
275 /** @brief Remove track @p id (but do not destroy it) */
276 static struct track
*removetrack(const char *id
) {
277 struct track
*t
, **tt
;
279 D(("removetrack %s", id
));
280 for(tt
= &tracks
; (t
= *tt
) && strcmp(id
, t
->id
); tt
= &t
->next
)
287 /** @brief Destroy a track */
288 static void destroy(struct track
*t
) {
289 D(("destroy %s", t
->id
));
290 if(t
->fd
!= -1) xclose(t
->fd
);
291 if(t
->buffer
!= (void *)&t
->format
) free(t
->buffer
);
295 /** @brief Notice a new connection */
296 static void acquire(struct track
*t
, int fd
) {
297 D(("acquire %s %d", t
->id
, fd
));
304 /** @brief Return true if A and B denote identical libao formats, else false */
305 static int formats_equal(const ao_sample_format
*a
,
306 const ao_sample_format
*b
) {
307 return (a
->bits
== b
->bits
308 && a
->rate
== b
->rate
309 && a
->channels
== b
->channels
310 && a
->byte_format
== b
->byte_format
);
313 /** @brief Compute arguments to sox */
314 static void soxargs(const char ***pp
, char **qq
, ao_sample_format
*ao
) {
319 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-r%d", ao
->rate
); *qq
+= n
+ 1;
320 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-c%d", ao
->channels
); *qq
+= n
+ 1;
321 /* sox 12.17.9 insists on -b etc; CVS sox insists on -<n> etc; both are
323 switch(config
->sox_generation
) {
326 && ao
->byte_format
!= AO_FMT_NATIVE
327 && ao
->byte_format
!= MACHINE_AO_FMT
) {
331 case 8: *(*pp
)++ = "-b"; break;
332 case 16: *(*pp
)++ = "-w"; break;
333 case 32: *(*pp
)++ = "-l"; break;
334 case 64: *(*pp
)++ = "-d"; break;
335 default: fatal(0, "cannot handle sample size %d", (int)ao
->bits
);
339 switch(ao
->byte_format
) {
340 case AO_FMT_NATIVE
: break;
341 case AO_FMT_BIG
: *(*pp
)++ = "-B"; break;
342 case AO_FMT_LITTLE
: *(*pp
)++ = "-L"; break;
344 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-%d", ao
->bits
/8); *qq
+= n
+ 1;
349 /** @brief Enable format translation
351 * If necessary, replaces a tracks inbound file descriptor with one connected
352 * to a sox invocation, which performs the required translation.
354 static void enable_translation(struct track
*t
) {
355 switch(config
->speaker_backend
) {
356 case BACKEND_COMMAND
:
357 case BACKEND_NETWORK
:
358 /* These backends need a specific sample format */
364 if(!formats_equal(&t
->format
, &config
->sample_format
)) {
365 char argbuf
[1024], *q
= argbuf
;
366 const char *av
[18], **pp
= av
;
371 soxargs(&pp
, &q
, &t
->format
);
373 soxargs(&pp
, &q
, &config
->sample_format
);
377 for(pp
= av
; *pp
; pp
++)
378 D(("sox arg[%d] = %s", pp
- av
, *pp
));
384 signal(SIGPIPE
, SIG_DFL
);
386 xdup2(soxpipe
[1], 1);
387 fcntl(0, F_SETFL
, fcntl(0, F_GETFL
) & ~O_NONBLOCK
);
391 execvp("sox", (char **)av
);
394 D(("forking sox for format conversion (kid = %d)", soxkid
));
398 t
->format
= config
->sample_format
;
402 /** @brief Read data into a sample buffer
403 * @param t Pointer to track
404 * @return 0 on success, -1 on EOF
406 * This is effectively the read callback on @c t->fd.
408 static int fill(struct track
*t
) {
412 D(("fill %s: eof=%d used=%zu size=%zu got_format=%d",
413 t
->id
, t
->eof
, t
->used
, t
->size
, t
->got_format
));
414 if(t
->eof
) return -1;
415 if(t
->used
< t
->size
) {
416 /* there is room left in the buffer */
417 where
= (t
->start
+ t
->used
) % t
->size
;
419 /* We are reading audio data, get as much as we can */
420 if(where
>= t
->start
) left
= t
->size
- where
;
421 else left
= t
->start
- where
;
423 /* We are still waiting for the format, only get that */
424 left
= sizeof (ao_sample_format
) - t
->used
;
426 n
= read(t
->fd
, t
->buffer
+ where
, left
);
427 } while(n
< 0 && errno
== EINTR
);
429 if(errno
!= EAGAIN
) fatal(errno
, "error reading sample stream");
433 D(("fill %s: eof detected", t
->id
));
438 if(!t
->got_format
&& t
->used
>= sizeof (ao_sample_format
)) {
439 assert(t
->used
== sizeof (ao_sample_format
));
440 /* Check that our assumptions are met. */
441 if(t
->format
.bits
& 7)
442 fatal(0, "bits per sample not a multiple of 8");
443 /* If the input format is unsuitable, arrange to translate it */
444 enable_translation(t
);
445 /* Make a new buffer for audio data. */
446 t
->size
= bytes_per_frame(&t
->format
) * t
->format
.rate
* BUFFER_SECONDS
;
447 t
->buffer
= xmalloc(t
->size
);
450 D(("got format for %s", t
->id
));
456 /** @brief Close the sound device */
457 static void idle(void) {
460 if(config
->speaker_backend
== BACKEND_ALSA
&& pcm
) {
463 if((err
= snd_pcm_nonblock(pcm
, 0)) < 0)
464 fatal(0, "error calling snd_pcm_nonblock: %d", err
);
471 D(("released audio device"));
478 /** @brief Abandon the current track */
479 static void abandon(void) {
480 struct speaker_message sm
;
483 memset(&sm
, 0, sizeof sm
);
484 sm
.type
= SM_FINISHED
;
485 strcpy(sm
.id
, playing
->id
);
486 speaker_send(1, &sm
, 0);
487 removetrack(playing
->id
);
494 /** @brief Log ALSA parameters */
495 static void log_params(snd_pcm_hw_params_t
*hwparams
,
496 snd_pcm_sw_params_t
*swparams
) {
500 return; /* too verbose */
505 snd_pcm_sw_params_get_silence_size(swparams
, &f
);
506 info("sw silence_size=%lu", (unsigned long)f
);
507 snd_pcm_sw_params_get_silence_threshold(swparams
, &f
);
508 info("sw silence_threshold=%lu", (unsigned long)f
);
509 snd_pcm_sw_params_get_sleep_min(swparams
, &u
);
510 info("sw sleep_min=%lu", (unsigned long)u
);
511 snd_pcm_sw_params_get_start_threshold(swparams
, &f
);
512 info("sw start_threshold=%lu", (unsigned long)f
);
513 snd_pcm_sw_params_get_stop_threshold(swparams
, &f
);
514 info("sw stop_threshold=%lu", (unsigned long)f
);
515 snd_pcm_sw_params_get_xfer_align(swparams
, &f
);
516 info("sw xfer_align=%lu", (unsigned long)f
);
521 /** @brief Enable sound output
523 * Makes sure the sound device is open and has the right sample format. Return
524 * 0 on success and -1 on error.
526 static int activate(void) {
527 /* If we don't know the format yet we cannot start. */
528 if(!playing
->got_format
) {
529 D((" - not got format for %s", playing
->id
));
532 return backend
->activate();
535 /* Check to see whether the current track has finished playing */
536 static void maybe_finished(void) {
539 && (!playing
->got_format
540 || playing
->used
< bytes_per_frame(&playing
->format
)))
544 static void fork_cmd(void) {
547 if(cmdfd
!= -1) close(cmdfd
);
551 signal(SIGPIPE
, SIG_DFL
);
555 execl("/bin/sh", "sh", "-c", config
->speaker_command
, (char *)0);
556 fatal(errno
, "error execing /bin/sh");
560 D(("forked cmd %d, fd = %d", cmdpid
, cmdfd
));
563 static void play(size_t frames
) {
564 size_t avail_bytes
, write_bytes
, written_frames
;
565 ssize_t written_bytes
;
566 struct rtp_header header
;
573 forceplay
= 0; /* Must have called abandon() */
576 D(("play: play %zu/%zu%s %dHz %db %dc", frames
, playing
->used
/ bpf
,
577 playing
->eof ?
" EOF" : "",
578 playing
->format
.rate
,
579 playing
->format
.bits
,
580 playing
->format
.channels
));
581 /* If we haven't got enough bytes yet wait until we have. Exception: when
583 if(playing
->used
< frames
* bpf
&& !playing
->eof
) {
587 /* We have got enough data so don't force play again */
589 /* Figure out how many frames there are available to write */
590 if(playing
->start
+ playing
->used
> playing
->size
)
591 avail_bytes
= playing
->size
- playing
->start
;
593 avail_bytes
= playing
->used
;
595 switch(config
->speaker_backend
) {
598 snd_pcm_sframes_t pcm_written_frames
;
602 avail_frames
= avail_bytes
/ bpf
;
603 if(avail_frames
> frames
)
604 avail_frames
= frames
;
607 pcm_written_frames
= snd_pcm_writei(pcm
,
608 playing
->buffer
+ playing
->start
,
610 D(("actually play %zu frames, wrote %d",
611 avail_frames
, (int)pcm_written_frames
));
612 if(pcm_written_frames
< 0) {
613 switch(pcm_written_frames
) {
614 case -EPIPE
: /* underrun */
615 error(0, "snd_pcm_writei reports underrun");
616 if((err
= snd_pcm_prepare(pcm
)) < 0)
617 fatal(0, "error calling snd_pcm_prepare: %d", err
);
622 fatal(0, "error calling snd_pcm_writei: %d",
623 (int)pcm_written_frames
);
626 written_frames
= pcm_written_frames
;
627 written_bytes
= written_frames
* bpf
;
631 case BACKEND_COMMAND
:
632 if(avail_bytes
> frames
* bpf
)
633 avail_bytes
= frames
* bpf
;
634 written_bytes
= write(cmdfd
, playing
->buffer
+ playing
->start
,
636 D(("actually play %zu bytes, wrote %d",
637 avail_bytes
, (int)written_bytes
));
638 if(written_bytes
< 0) {
641 error(0, "hmm, command died; trying another");
648 written_frames
= written_bytes
/ bpf
; /* good enough */
650 case BACKEND_NETWORK
:
651 /* We transmit using RTP (RFC3550) and attempt to conform to the internet
652 * AVT profile (RFC3551). */
655 /* There may have been a gap. Fix up the RTP time accordingly. */
658 uint64_t target_rtp_time
;
660 /* Find the current time */
661 xgettimeofday(&now
, 0);
662 /* Find the number of microseconds elapsed since rtp_time=0 */
663 delta
= tvsub_us(now
, rtp_time_0
);
664 assert(delta
<= UINT64_MAX
/ 88200);
665 target_rtp_time
= (delta
* playing
->format
.rate
666 * playing
->format
.channels
) / 1000000;
667 /* Overflows at ~6 years uptime with 44100Hz stereo */
669 /* rtp_time is the number of samples we've played. NB that we play
670 * RTP_AHEAD_MS ahead of ourselves, so it may legitimately be ahead of
671 * the value we deduce from time comparison.
673 * Suppose we have 1s track started at t=0, and another track begins to
674 * play at t=2s. Suppose RTP_AHEAD_MS=1000 and 44100Hz stereo. In that
675 * case we'll send 1s of audio as fast as we can, giving rtp_time=88200.
676 * rtp_time stops at this point.
678 * At t=2s we'll have calculated target_rtp_time=176400. In this case we
679 * set rtp_time=176400 and the player can correctly conclude that it
680 * should leave 1s between the tracks.
682 * Suppose instead that the second track arrives at t=0.5s, and that
683 * we've managed to transmit the whole of the first track already. We'll
684 * have target_rtp_time=44100.
686 * The desired behaviour is to play the second track back to back with
687 * first. In this case therefore we do not modify rtp_time.
689 * Is it ever right to reduce rtp_time? No; for that would imply
690 * transmitting packets with overlapping timestamp ranges, which does not
693 if(target_rtp_time
> rtp_time
) {
694 /* More time has elapsed than we've transmitted samples. That implies
695 * we've been 'sending' silence. */
696 info("advancing rtp_time by %"PRIu64
" samples",
697 target_rtp_time
- rtp_time
);
698 rtp_time
= target_rtp_time
;
699 } else if(target_rtp_time
< rtp_time
) {
700 const int64_t samples_ahead
= ((uint64_t)RTP_AHEAD_MS
701 * config
->sample_format
.rate
702 * config
->sample_format
.channels
705 if(target_rtp_time
+ samples_ahead
< rtp_time
) {
706 info("reversing rtp_time by %"PRIu64
" samples",
707 rtp_time
- target_rtp_time
);
711 header
.vpxcc
= 2 << 6; /* V=2, P=0, X=0, CC=0 */
712 header
.seq
= htons(rtp_seq
++);
713 header
.timestamp
= htonl((uint32_t)rtp_time
);
714 header
.ssrc
= rtp_id
;
715 header
.mpt
= (idled ?
0x80 : 0x00) | 10;
716 /* 10 = L16 = 16-bit x 2 x 44100KHz. We ought to deduce this value from
717 * the sample rate (in a library somewhere so that configuration.c can rule
718 * out invalid rates).
721 if(avail_bytes
> NETWORK_BYTES
- sizeof header
) {
722 avail_bytes
= NETWORK_BYTES
- sizeof header
;
723 /* Always send a whole number of frames */
724 avail_bytes
-= avail_bytes
% bpf
;
726 /* "The RTP clock rate used for generating the RTP timestamp is independent
727 * of the number of channels and the encoding; it equals the number of
728 * sampling periods per second. For N-channel encodings, each sampling
729 * period (say, 1/8000 of a second) generates N samples. (This terminology
730 * is standard, but somewhat confusing, as the total number of samples
731 * generated per second is then the sampling rate times the channel
734 write_bytes
= avail_bytes
;
736 vec
[0].iov_base
= (void *)&header
;
737 vec
[0].iov_len
= sizeof header
;
738 vec
[1].iov_base
= playing
->buffer
+ playing
->start
;
739 vec
[1].iov_len
= avail_bytes
;
741 written_bytes
= writev(bfd
,
744 } while(written_bytes
< 0 && errno
== EINTR
);
745 if(written_bytes
< 0) {
746 error(errno
, "error transmitting audio data");
748 if(audio_errors
== 10)
749 fatal(0, "too many audio errors");
754 written_bytes
= avail_bytes
;
755 written_frames
= written_bytes
/ bpf
;
756 /* Advance RTP's notion of the time */
757 rtp_time
+= written_frames
* playing
->format
.channels
;
762 /* written_bytes and written_frames had better both be set and correct by
764 playing
->start
+= written_bytes
;
765 playing
->used
-= written_bytes
;
766 playing
->played
+= written_frames
;
767 /* If the pointer is at the end of the buffer (or the buffer is completely
768 * empty) wrap it back to the start. */
769 if(!playing
->used
|| playing
->start
== playing
->size
)
771 frames
-= written_frames
;
774 /* Notify the server what we're up to. */
775 static void report(void) {
776 struct speaker_message sm
;
778 if(playing
&& playing
->buffer
!= (void *)&playing
->format
) {
779 memset(&sm
, 0, sizeof sm
);
780 sm
.type
= paused ? SM_PAUSED
: SM_PLAYING
;
781 strcpy(sm
.id
, playing
->id
);
782 sm
.data
= playing
->played
/ playing
->format
.rate
;
783 speaker_send(1, &sm
, 0);
788 static void reap(int __attribute__((unused
)) sig
) {
793 cmdpid
= waitpid(-1, &st
, WNOHANG
);
795 signal(SIGCHLD
, reap
);
798 static int addfd(int fd
, int events
) {
801 fds
[fdno
].events
= events
;
808 /** @brief ALSA backend initialization */
809 static void alsa_init(void) {
810 info("selected ALSA backend");
813 /** @brief ALSA backend activation */
814 static int alsa_activate(void) {
815 /* If we need to change format then close the current device. */
816 if(pcm
&& !formats_equal(&playing
->format
, &pcm_format
))
819 snd_pcm_hw_params_t
*hwparams
;
820 snd_pcm_sw_params_t
*swparams
;
821 snd_pcm_uframes_t pcm_bufsize
;
823 int sample_format
= 0;
827 if((err
= snd_pcm_open(&pcm
,
829 SND_PCM_STREAM_PLAYBACK
,
830 SND_PCM_NONBLOCK
))) {
831 error(0, "error from snd_pcm_open: %d", err
);
834 snd_pcm_hw_params_alloca(&hwparams
);
835 D(("set up hw params"));
836 if((err
= snd_pcm_hw_params_any(pcm
, hwparams
)) < 0)
837 fatal(0, "error from snd_pcm_hw_params_any: %d", err
);
838 if((err
= snd_pcm_hw_params_set_access(pcm
, hwparams
,
839 SND_PCM_ACCESS_RW_INTERLEAVED
)) < 0)
840 fatal(0, "error from snd_pcm_hw_params_set_access: %d", err
);
841 switch(playing
->format
.bits
) {
843 sample_format
= SND_PCM_FORMAT_S8
;
846 switch(playing
->format
.byte_format
) {
847 case AO_FMT_NATIVE
: sample_format
= SND_PCM_FORMAT_S16
; break;
848 case AO_FMT_LITTLE
: sample_format
= SND_PCM_FORMAT_S16_LE
; break;
849 case AO_FMT_BIG
: sample_format
= SND_PCM_FORMAT_S16_BE
; break;
850 error(0, "unrecognized byte format %d", playing
->format
.byte_format
);
855 error(0, "unsupported sample size %d", playing
->format
.bits
);
858 if((err
= snd_pcm_hw_params_set_format(pcm
, hwparams
,
859 sample_format
)) < 0) {
860 error(0, "error from snd_pcm_hw_params_set_format (%d): %d",
864 rate
= playing
->format
.rate
;
865 if((err
= snd_pcm_hw_params_set_rate_near(pcm
, hwparams
, &rate
, 0)) < 0) {
866 error(0, "error from snd_pcm_hw_params_set_rate (%d): %d",
867 playing
->format
.rate
, err
);
870 if(rate
!= (unsigned)playing
->format
.rate
)
871 info("want rate %d, got %u", playing
->format
.rate
, rate
);
872 if((err
= snd_pcm_hw_params_set_channels(pcm
, hwparams
,
873 playing
->format
.channels
)) < 0) {
874 error(0, "error from snd_pcm_hw_params_set_channels (%d): %d",
875 playing
->format
.channels
, err
);
878 bufsize
= 3 * FRAMES
;
879 pcm_bufsize
= bufsize
;
880 if((err
= snd_pcm_hw_params_set_buffer_size_near(pcm
, hwparams
,
882 fatal(0, "error from snd_pcm_hw_params_set_buffer_size (%d): %d",
884 if(pcm_bufsize
!= 3 * FRAMES
&& pcm_bufsize
!= last_pcm_bufsize
)
885 info("asked for PCM buffer of %d frames, got %d",
886 3 * FRAMES
, (int)pcm_bufsize
);
887 last_pcm_bufsize
= pcm_bufsize
;
888 if((err
= snd_pcm_hw_params(pcm
, hwparams
)) < 0)
889 fatal(0, "error calling snd_pcm_hw_params: %d", err
);
890 D(("set up sw params"));
891 snd_pcm_sw_params_alloca(&swparams
);
892 if((err
= snd_pcm_sw_params_current(pcm
, swparams
)) < 0)
893 fatal(0, "error calling snd_pcm_sw_params_current: %d", err
);
894 if((err
= snd_pcm_sw_params_set_avail_min(pcm
, swparams
, FRAMES
)) < 0)
895 fatal(0, "error calling snd_pcm_sw_params_set_avail_min %d: %d",
897 if((err
= snd_pcm_sw_params(pcm
, swparams
)) < 0)
898 fatal(0, "error calling snd_pcm_sw_params: %d", err
);
899 pcm_format
= playing
->format
;
900 bpf
= bytes_per_frame(&pcm_format
);
901 D(("acquired audio device"));
902 log_params(hwparams
, swparams
);
909 /* We assume the error is temporary and that we'll retry in a bit. */
918 /** @brief Command backend initialization */
919 static void command_init(void) {
920 info("selected command backend");
924 /** @brief Command backend activation */
925 static int command_activate(void) {
927 pcm_format
= config
->sample_format
;
928 bufsize
= 3 * FRAMES
;
929 bpf
= bytes_per_frame(&config
->sample_format
);
930 D(("acquired audio device"));
936 /** @brief Network backend initialization */
937 static void network_init(void) {
938 struct addrinfo
*res
, *sres
;
939 static const struct addrinfo pref
= {
949 static const struct addrinfo prefbind
= {
959 static const int one
= 1;
960 int sndbuf
, target_sndbuf
= 131072;
962 char *sockname
, *ssockname
;
964 res
= get_address(&config
->broadcast
, &pref
, &sockname
);
966 if(config
->broadcast_from
.n
) {
967 sres
= get_address(&config
->broadcast_from
, &prefbind
, &ssockname
);
971 if((bfd
= socket(res
->ai_family
,
973 res
->ai_protocol
)) < 0)
974 fatal(errno
, "error creating broadcast socket");
975 if(setsockopt(bfd
, SOL_SOCKET
, SO_BROADCAST
, &one
, sizeof one
) < 0)
976 fatal(errno
, "error setting SO_BROADCAST on broadcast socket");
978 if(getsockopt(bfd
, SOL_SOCKET
, SO_SNDBUF
,
980 fatal(errno
, "error getting SO_SNDBUF");
981 if(target_sndbuf
> sndbuf
) {
982 if(setsockopt(bfd
, SOL_SOCKET
, SO_SNDBUF
,
983 &target_sndbuf
, sizeof target_sndbuf
) < 0)
984 error(errno
, "error setting SO_SNDBUF to %d", target_sndbuf
);
986 info("changed socket send buffer size from %d to %d",
987 sndbuf
, target_sndbuf
);
989 info("default socket send buffer is %d",
991 /* We might well want to set additional broadcast- or multicast-related
993 if(sres
&& bind(bfd
, sres
->ai_addr
, sres
->ai_addrlen
) < 0)
994 fatal(errno
, "error binding broadcast socket to %s", ssockname
);
995 if(connect(bfd
, res
->ai_addr
, res
->ai_addrlen
) < 0)
996 fatal(errno
, "error connecting broadcast socket to %s", sockname
);
998 gcry_randomize(&rtp_id
, sizeof rtp_id
, GCRY_STRONG_RANDOM
);
999 info("selected network backend, sending to %s", sockname
);
1000 if(config
->sample_format
.byte_format
!= AO_FMT_BIG
) {
1001 info("forcing big-endian sample format");
1002 config
->sample_format
.byte_format
= AO_FMT_BIG
;
1006 /** @brief Network backend activation */
1007 static int network_activate(void) {
1009 pcm_format
= config
->sample_format
;
1010 bufsize
= 3 * FRAMES
;
1011 bpf
= bytes_per_frame(&config
->sample_format
);
1012 D(("acquired audio device"));
1018 /** @brief Table of speaker backends */
1019 static const struct speaker_backend backends
[] = {
1040 int main(int argc
, char **argv
) {
1041 int n
, fd
, stdin_slot
, alsa_slots
, cmdfd_slot
, bfd_slot
, poke
, timeout
;
1043 struct speaker_message sm
;
1045 int alsa_nslots
= -1, err
;
1049 if(!setlocale(LC_CTYPE
, "")) fatal(errno
, "error calling setlocale");
1050 while((n
= getopt_long(argc
, argv
, "hVc:dD", options
, 0)) >= 0) {
1053 case 'V': version();
1054 case 'c': configfile
= optarg
; break;
1055 case 'd': debugging
= 1; break;
1056 case 'D': debugging
= 0; break;
1057 default: fatal(0, "invalid option");
1060 if(getenv("DISORDER_DEBUG_SPEAKER")) debugging
= 1;
1061 /* If stderr is a TTY then log there, otherwise to syslog. */
1063 openlog(progname
, LOG_PID
, LOG_DAEMON
);
1064 log_default
= &log_syslog
;
1066 if(config_read()) fatal(0, "cannot read configuration");
1067 /* ignore SIGPIPE */
1068 signal(SIGPIPE
, SIG_IGN
);
1070 signal(SIGCHLD
, reap
);
1071 /* set nice value */
1072 xnice(config
->nice_speaker
);
1075 /* make sure we're not root, whatever the config says */
1076 if(getuid() == 0 || geteuid() == 0) fatal(0, "do not run as root");
1077 /* identify the backend used to play */
1078 for(n
= 0; backends
[n
].backend
!= -1; ++n
)
1079 if(backends
[n
].backend
== config
->speaker_backend
)
1081 if(backends
[n
].backend
== -1)
1082 fatal(0, "unsupported backend %d", config
->speaker_backend
);
1083 backend
= &backends
[n
];
1084 /* backend-specific initialization */
1086 while(getppid() != 1) {
1088 /* Always ready for commands from the main server. */
1089 stdin_slot
= addfd(0, POLLIN
);
1090 /* Try to read sample data for the currently playing track if there is
1092 if(playing
&& !playing
->eof
&& playing
->used
< playing
->size
) {
1093 playing
->slot
= addfd(playing
->fd
, POLLIN
);
1096 /* If forceplay is set then wait until it succeeds before waiting on the
1101 /* By default we will wait up to a second before thinking about current
1104 if(ready
&& !forceplay
) {
1105 switch(config
->speaker_backend
) {
1106 case BACKEND_COMMAND
:
1107 /* We send sample data to the subprocess as fast as it can accept it.
1108 * This isn't ideal as pause latency can be very high as a result. */
1110 cmdfd_slot
= addfd(cmdfd
, POLLOUT
);
1112 case BACKEND_NETWORK
: {
1115 uint64_t target_rtp_time
;
1116 const int64_t samples_ahead
= ((uint64_t)RTP_AHEAD_MS
1117 * config
->sample_format
.rate
1118 * config
->sample_format
.channels
1121 static unsigned logit
;
1124 /* If we're starting then initialize the base time */
1126 xgettimeofday(&rtp_time_0
, 0);
1127 /* We send audio data whenever we get RTP_AHEAD seconds or more
1129 xgettimeofday(&now
, 0);
1130 target_us
= tvsub_us(now
, rtp_time_0
);
1131 assert(target_us
<= UINT64_MAX
/ 88200);
1132 target_rtp_time
= (target_us
* config
->sample_format
.rate
1133 * config
->sample_format
.channels
)
1137 /* TODO remove logging guff */
1138 if(!(logit
++ & 1023))
1139 info("rtp_time %llu target %llu difference %lld [%lld]",
1140 rtp_time
, target_rtp_time
,
1141 rtp_time
- target_rtp_time
,
1144 if((int64_t)(rtp_time
- target_rtp_time
) < samples_ahead
)
1145 bfd_slot
= addfd(bfd
, POLLOUT
);
1149 case BACKEND_ALSA
: {
1150 /* We send sample data to ALSA as fast as it can accept it, relying on
1151 * the fact that it has a relatively small buffer to minimize pause
1158 alsa_nslots
= snd_pcm_poll_descriptors(pcm
, &fds
[fdno
], NFDS
- fdno
);
1159 if((alsa_nslots
<= 0
1160 || !(fds
[alsa_slots
].events
& POLLOUT
))
1161 && snd_pcm_state(pcm
) == SND_PCM_STATE_XRUN
) {
1162 error(0, "underrun detected after call to snd_pcm_poll_descriptors()");
1163 if((err
= snd_pcm_prepare(pcm
)))
1164 fatal(0, "error calling snd_pcm_prepare: %d", err
);
1167 } while(retry
-- > 0);
1168 if(alsa_nslots
>= 0)
1169 fdno
+= alsa_nslots
;
1174 assert(!"unknown backend");
1177 /* If any other tracks don't have a full buffer, try to read sample data
1179 for(t
= tracks
; t
; t
= t
->next
)
1181 if(!t
->eof
&& t
->used
< t
->size
) {
1182 t
->slot
= addfd(t
->fd
, POLLIN
| POLLHUP
);
1186 /* Wait for something interesting to happen */
1187 n
= poll(fds
, fdno
, timeout
);
1189 if(errno
== EINTR
) continue;
1190 fatal(errno
, "error calling poll");
1192 /* Play some sound before doing anything else */
1194 switch(config
->speaker_backend
) {
1197 if(alsa_slots
!= -1) {
1198 unsigned short alsa_revents
;
1200 if((err
= snd_pcm_poll_descriptors_revents(pcm
,
1203 &alsa_revents
)) < 0)
1204 fatal(0, "error calling snd_pcm_poll_descriptors_revents: %d", err
);
1205 if(alsa_revents
& (POLLOUT
| POLLERR
))
1211 case BACKEND_COMMAND
:
1212 if(cmdfd_slot
!= -1) {
1213 if(fds
[cmdfd_slot
].revents
& (POLLOUT
| POLLERR
))
1218 case BACKEND_NETWORK
:
1219 if(bfd_slot
!= -1) {
1220 if(fds
[bfd_slot
].revents
& (POLLOUT
| POLLERR
))
1227 /* Some attempt to play must have failed */
1228 if(playing
&& !paused
)
1231 forceplay
= 0; /* just in case */
1233 /* Perhaps we have a command to process */
1234 if(fds
[stdin_slot
].revents
& POLLIN
) {
1235 n
= speaker_recv(0, &sm
, &fd
);
1239 D(("SM_PREPARE %s %d", sm
.id
, fd
));
1240 if(fd
== -1) fatal(0, "got SM_PREPARE but no file descriptor");
1241 t
= findtrack(sm
.id
, 1);
1245 D(("SM_PLAY %s %d", sm
.id
, fd
));
1246 if(playing
) fatal(0, "got SM_PLAY but already playing something");
1247 t
= findtrack(sm
.id
, 1);
1248 if(fd
!= -1) acquire(t
, fd
);
1268 D(("SM_CANCEL %s", sm
.id
));
1269 t
= removetrack(sm
.id
);
1272 sm
.type
= SM_FINISHED
;
1273 strcpy(sm
.id
, playing
->id
);
1274 speaker_send(1, &sm
, 0);
1279 error(0, "SM_CANCEL for unknown track %s", sm
.id
);
1284 if(config_read()) error(0, "cannot read configuration");
1285 info("reloaded configuration");
1288 error(0, "unknown message type %d", sm
.type
);
1291 /* Read in any buffered data */
1292 for(t
= tracks
; t
; t
= t
->next
)
1293 if(t
->slot
!= -1 && (fds
[t
->slot
].revents
& (POLLIN
| POLLHUP
)))
1295 /* We might be able to play now */
1296 if(ready
&& forceplay
&& playing
&& !paused
)
1298 /* Maybe we finished playing a track somewhere in the above */
1300 /* If we don't need the sound device for now then close it for the benefit
1301 * of anyone else who wants it. */
1302 if((!playing
|| paused
) && ready
)
1304 /* If we've not reported out state for a second do so now. */
1305 if(time(0) > last_report
)
1308 info("stopped (parent terminated)");
1317 indent-tabs-mode:nil