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 * @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"
92 #include <alsa/asoundlib.h>
95 #ifdef WORDS_BIGENDIAN
96 # define MACHINE_AO_FMT AO_FMT_BIG
98 # define MACHINE_AO_FMT AO_FMT_LITTLE
101 /** @brief How many seconds of input to buffer
103 * While any given connection has this much audio buffered, no more reads will
104 * be issued for that connection. The decoder will have to wait.
106 #define BUFFER_SECONDS 5
108 #define FRAMES 4096 /* Frame batch size */
110 /** @brief Bytes to send per network packet
112 * Don't make this too big or arithmetic will start to overflow.
114 #define NETWORK_BYTES (1024+sizeof(struct rtp_header))
116 /** @brief Maximum RTP playahead (ms) */
117 #define RTP_AHEAD_MS 1000
119 /** @brief Maximum number of FDs to poll for */
122 /** @brief Track structure
124 * Known tracks are kept in a linked list. Usually there will be at most two
125 * of these but rearranging the queue can cause there to be more.
127 static struct track
{
128 struct track
*next
; /* next track */
129 int fd
; /* input FD */
130 char id
[24]; /* ID */
131 size_t start
, used
; /* start + bytes used */
132 int eof
; /* input is at EOF */
133 int got_format
; /* got format yet? */
134 ao_sample_format format
; /* sample format */
135 unsigned long long played
; /* number of frames played */
136 char *buffer
; /* sample buffer */
137 size_t size
; /* sample buffer size */
138 int slot
; /* poll array slot */
139 } *tracks
, *playing
; /* all tracks + playing track */
141 static time_t last_report
; /* when we last reported */
142 static int paused
; /* pause status */
143 static size_t bpf
; /* bytes per frame */
144 static struct pollfd fds
[NFDS
]; /* if we need more than that */
145 static int fdno
; /* fd number */
146 static size_t bufsize
; /* buffer size */
148 /** @brief The current PCM handle */
149 static snd_pcm_t
*pcm
;
150 static snd_pcm_uframes_t last_pcm_bufsize
; /* last seen buffer size */
151 static ao_sample_format pcm_format
; /* current format if aodev != 0 */
154 /** @brief Ready to send audio
156 * This is set when the destination is ready to receive audio. Generally
157 * this implies that the sound device is open. In the ALSA backend it
158 * does @b not necessarily imply that is has the right sample format.
162 static int forceplay
; /* frames to force play */
163 static int cmdfd
= -1; /* child process input */
164 static int bfd
= -1; /* broadcast FD */
166 /** @brief RTP timestamp
168 * This counts the number of samples played (NB not the number of frames
171 * The timestamp in the packet header is only 32 bits wide. With 44100Hz
172 * stereo, that only gives about half a day before wrapping, which is not
173 * particularly convenient for certain debugging purposes. Therefore the
174 * timestamp is maintained as a 64-bit integer, giving around six million years
175 * before wrapping, and truncated to 32 bits when transmitting.
177 static uint64_t rtp_time
;
179 /** @brief RTP base timestamp
181 * This is the real time correspoding to an @ref rtp_time of 0. It is used
182 * to recalculate the timestamp after idle periods.
184 static struct timeval rtp_time_0
;
186 static uint16_t rtp_seq
; /* frame sequence number */
187 static uint32_t rtp_id
; /* RTP SSRC */
188 static int idled
; /* set when idled */
189 static int audio_errors
; /* audio error counter */
191 /** @brief Structure of a backend */
192 struct speaker_backend
{
193 /** @brief Which backend this is
195 * @c -1 terminates the list.
202 * - @ref FIXED_FORMAT
205 /** @brief Lock to configured sample format */
206 #define FIXED_FORMAT 0x0001
208 /** @brief Initialization
210 * Called once at startup. This is responsible for one-time setup
211 * operations, for instance opening a network socket to transmit to.
213 * When writing to a native sound API this might @b not imply opening the
214 * native sound device - that might be done by @c activate below.
218 /** @brief Activation
219 * @return 0 on success, non-0 on error
221 * Called to activate the output device.
223 * After this function succeeds, @ref ready should be non-0. As well as
224 * opening the audio device, this function is responsible for reconfiguring
225 * if it necessary to cope with different samples formats (for backends that
226 * don't demand a single fixed sample format for the lifetime of the server).
228 int (*activate
)(void);
230 /** @brief Play sound
231 * @param frames Number of frames to play
232 * @return Number of frames actually played
234 size_t (*play
)(size_t frames
);
236 /** @brief Deactivation
238 * Called to deactivate the sound device. This is the inverse of
241 void (*deactivate
)(void);
244 /** @brief Selected backend */
245 static const struct speaker_backend
*backend
;
247 static const struct option options
[] = {
248 { "help", no_argument
, 0, 'h' },
249 { "version", no_argument
, 0, 'V' },
250 { "config", required_argument
, 0, 'c' },
251 { "debug", no_argument
, 0, 'd' },
252 { "no-debug", no_argument
, 0, 'D' },
256 /* Display usage message and terminate. */
257 static void help(void) {
259 " disorder-speaker [OPTIONS]\n"
261 " --help, -h Display usage message\n"
262 " --version, -V Display version number\n"
263 " --config PATH, -c PATH Set configuration file\n"
264 " --debug, -d Turn on debugging\n"
266 "Speaker process for DisOrder. Not intended to be run\n"
272 /* Display version number and terminate. */
273 static void version(void) {
274 xprintf("disorder-speaker version %s\n", disorder_version_string
);
279 /** @brief Return the number of bytes per frame in @p format */
280 static size_t bytes_per_frame(const ao_sample_format
*format
) {
281 return format
->channels
* format
->bits
/ 8;
284 /** @brief Find track @p id, maybe creating it if not found */
285 static struct track
*findtrack(const char *id
, int create
) {
288 D(("findtrack %s %d", id
, create
));
289 for(t
= tracks
; t
&& strcmp(id
, t
->id
); t
= t
->next
)
292 t
= xmalloc(sizeof *t
);
297 /* The initial input buffer will be the sample format. */
298 t
->buffer
= (void *)&t
->format
;
299 t
->size
= sizeof t
->format
;
304 /** @brief Remove track @p id (but do not destroy it) */
305 static struct track
*removetrack(const char *id
) {
306 struct track
*t
, **tt
;
308 D(("removetrack %s", id
));
309 for(tt
= &tracks
; (t
= *tt
) && strcmp(id
, t
->id
); tt
= &t
->next
)
316 /** @brief Destroy a track */
317 static void destroy(struct track
*t
) {
318 D(("destroy %s", t
->id
));
319 if(t
->fd
!= -1) xclose(t
->fd
);
320 if(t
->buffer
!= (void *)&t
->format
) free(t
->buffer
);
324 /** @brief Notice a new connection */
325 static void acquire(struct track
*t
, int fd
) {
326 D(("acquire %s %d", t
->id
, fd
));
333 /** @brief Return true if A and B denote identical libao formats, else false */
334 static int formats_equal(const ao_sample_format
*a
,
335 const ao_sample_format
*b
) {
336 return (a
->bits
== b
->bits
337 && a
->rate
== b
->rate
338 && a
->channels
== b
->channels
339 && a
->byte_format
== b
->byte_format
);
342 /** @brief Compute arguments to sox */
343 static void soxargs(const char ***pp
, char **qq
, ao_sample_format
*ao
) {
348 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-r%d", ao
->rate
); *qq
+= n
+ 1;
349 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-c%d", ao
->channels
); *qq
+= n
+ 1;
350 /* sox 12.17.9 insists on -b etc; CVS sox insists on -<n> etc; both are
352 switch(config
->sox_generation
) {
355 && ao
->byte_format
!= AO_FMT_NATIVE
356 && ao
->byte_format
!= MACHINE_AO_FMT
) {
360 case 8: *(*pp
)++ = "-b"; break;
361 case 16: *(*pp
)++ = "-w"; break;
362 case 32: *(*pp
)++ = "-l"; break;
363 case 64: *(*pp
)++ = "-d"; break;
364 default: fatal(0, "cannot handle sample size %d", (int)ao
->bits
);
368 switch(ao
->byte_format
) {
369 case AO_FMT_NATIVE
: break;
370 case AO_FMT_BIG
: *(*pp
)++ = "-B"; break;
371 case AO_FMT_LITTLE
: *(*pp
)++ = "-L"; break;
373 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-%d", ao
->bits
/8); *qq
+= n
+ 1;
378 /** @brief Enable format translation
380 * If necessary, replaces a tracks inbound file descriptor with one connected
381 * to a sox invocation, which performs the required translation.
383 static void enable_translation(struct track
*t
) {
384 if((backend
->flags
& FIXED_FORMAT
)
385 && !formats_equal(&t
->format
, &config
->sample_format
)) {
386 char argbuf
[1024], *q
= argbuf
;
387 const char *av
[18], **pp
= av
;
392 soxargs(&pp
, &q
, &t
->format
);
394 soxargs(&pp
, &q
, &config
->sample_format
);
398 for(pp
= av
; *pp
; pp
++)
399 D(("sox arg[%d] = %s", pp
- av
, *pp
));
405 signal(SIGPIPE
, SIG_DFL
);
407 xdup2(soxpipe
[1], 1);
408 fcntl(0, F_SETFL
, fcntl(0, F_GETFL
) & ~O_NONBLOCK
);
412 execvp("sox", (char **)av
);
415 D(("forking sox for format conversion (kid = %d)", soxkid
));
419 t
->format
= config
->sample_format
;
423 /** @brief Read data into a sample buffer
424 * @param t Pointer to track
425 * @return 0 on success, -1 on EOF
427 * This is effectively the read callback on @c t->fd.
429 static int fill(struct track
*t
) {
433 D(("fill %s: eof=%d used=%zu size=%zu got_format=%d",
434 t
->id
, t
->eof
, t
->used
, t
->size
, t
->got_format
));
435 if(t
->eof
) return -1;
436 if(t
->used
< t
->size
) {
437 /* there is room left in the buffer */
438 where
= (t
->start
+ t
->used
) % t
->size
;
440 /* We are reading audio data, get as much as we can */
441 if(where
>= t
->start
) left
= t
->size
- where
;
442 else left
= t
->start
- where
;
444 /* We are still waiting for the format, only get that */
445 left
= sizeof (ao_sample_format
) - t
->used
;
447 n
= read(t
->fd
, t
->buffer
+ where
, left
);
448 } while(n
< 0 && errno
== EINTR
);
450 if(errno
!= EAGAIN
) fatal(errno
, "error reading sample stream");
454 D(("fill %s: eof detected", t
->id
));
459 if(!t
->got_format
&& t
->used
>= sizeof (ao_sample_format
)) {
460 assert(t
->used
== sizeof (ao_sample_format
));
461 /* Check that our assumptions are met. */
462 if(t
->format
.bits
& 7)
463 fatal(0, "bits per sample not a multiple of 8");
464 /* If the input format is unsuitable, arrange to translate it */
465 enable_translation(t
);
466 /* Make a new buffer for audio data. */
467 t
->size
= bytes_per_frame(&t
->format
) * t
->format
.rate
* BUFFER_SECONDS
;
468 t
->buffer
= xmalloc(t
->size
);
471 D(("got format for %s", t
->id
));
477 /** @brief Close the sound device */
478 static void idle(void) {
480 if(backend
->deactivate
)
481 backend
->deactivate();
486 /** @brief Abandon the current track */
487 static void abandon(void) {
488 struct speaker_message sm
;
491 memset(&sm
, 0, sizeof sm
);
492 sm
.type
= SM_FINISHED
;
493 strcpy(sm
.id
, playing
->id
);
494 speaker_send(1, &sm
, 0);
495 removetrack(playing
->id
);
502 /** @brief Log ALSA parameters */
503 static void log_params(snd_pcm_hw_params_t
*hwparams
,
504 snd_pcm_sw_params_t
*swparams
) {
508 return; /* too verbose */
513 snd_pcm_sw_params_get_silence_size(swparams
, &f
);
514 info("sw silence_size=%lu", (unsigned long)f
);
515 snd_pcm_sw_params_get_silence_threshold(swparams
, &f
);
516 info("sw silence_threshold=%lu", (unsigned long)f
);
517 snd_pcm_sw_params_get_sleep_min(swparams
, &u
);
518 info("sw sleep_min=%lu", (unsigned long)u
);
519 snd_pcm_sw_params_get_start_threshold(swparams
, &f
);
520 info("sw start_threshold=%lu", (unsigned long)f
);
521 snd_pcm_sw_params_get_stop_threshold(swparams
, &f
);
522 info("sw stop_threshold=%lu", (unsigned long)f
);
523 snd_pcm_sw_params_get_xfer_align(swparams
, &f
);
524 info("sw xfer_align=%lu", (unsigned long)f
);
529 /** @brief Enable sound output
531 * Makes sure the sound device is open and has the right sample format. Return
532 * 0 on success and -1 on error.
534 static int activate(void) {
535 /* If we don't know the format yet we cannot start. */
536 if(!playing
->got_format
) {
537 D((" - not got format for %s", playing
->id
));
540 return backend
->activate();
543 /* Check to see whether the current track has finished playing */
544 static void maybe_finished(void) {
547 && (!playing
->got_format
548 || playing
->used
< bytes_per_frame(&playing
->format
)))
552 static void fork_cmd(void) {
555 if(cmdfd
!= -1) close(cmdfd
);
559 signal(SIGPIPE
, SIG_DFL
);
563 execl("/bin/sh", "sh", "-c", config
->speaker_command
, (char *)0);
564 fatal(errno
, "error execing /bin/sh");
568 D(("forked cmd %d, fd = %d", cmdpid
, cmdfd
));
571 static void play(size_t frames
) {
572 size_t avail_frames
, avail_bytes
, written_frames
;
573 ssize_t written_bytes
;
575 /* Make sure the output device is activated */
580 forceplay
= 0; /* Must have called abandon() */
583 D(("play: play %zu/%zu%s %dHz %db %dc", frames
, playing
->used
/ bpf
,
584 playing
->eof ?
" EOF" : "",
585 playing
->format
.rate
,
586 playing
->format
.bits
,
587 playing
->format
.channels
));
588 /* If we haven't got enough bytes yet wait until we have. Exception: when
590 if(playing
->used
< frames
* bpf
&& !playing
->eof
) {
594 /* We have got enough data so don't force play again */
596 /* Figure out how many frames there are available to write */
597 if(playing
->start
+ playing
->used
> playing
->size
)
598 /* The ring buffer is currently wrapped, only play up to the wrap point */
599 avail_bytes
= playing
->size
- playing
->start
;
601 /* The ring buffer is not wrapped, can play the lot */
602 avail_bytes
= playing
->used
;
603 avail_frames
= avail_bytes
/ bpf
;
604 /* Only play up to the requested amount */
605 if(avail_frames
> frames
)
606 avail_frames
= frames
;
610 written_frames
= backend
->play(avail_frames
);
611 written_bytes
= written_frames
* bpf
;
612 /* written_bytes and written_frames had better both be set and correct by
614 playing
->start
+= written_bytes
;
615 playing
->used
-= written_bytes
;
616 playing
->played
+= written_frames
;
617 /* If the pointer is at the end of the buffer (or the buffer is completely
618 * empty) wrap it back to the start. */
619 if(!playing
->used
|| playing
->start
== playing
->size
)
621 frames
-= written_frames
;
624 /* Notify the server what we're up to. */
625 static void report(void) {
626 struct speaker_message sm
;
628 if(playing
&& playing
->buffer
!= (void *)&playing
->format
) {
629 memset(&sm
, 0, sizeof sm
);
630 sm
.type
= paused ? SM_PAUSED
: SM_PLAYING
;
631 strcpy(sm
.id
, playing
->id
);
632 sm
.data
= playing
->played
/ playing
->format
.rate
;
633 speaker_send(1, &sm
, 0);
638 static void reap(int __attribute__((unused
)) sig
) {
643 cmdpid
= waitpid(-1, &st
, WNOHANG
);
645 signal(SIGCHLD
, reap
);
648 static int addfd(int fd
, int events
) {
651 fds
[fdno
].events
= events
;
658 /** @brief ALSA backend initialization */
659 static void alsa_init(void) {
660 info("selected ALSA backend");
663 /** @brief ALSA backend activation */
664 static int alsa_activate(void) {
665 /* If we need to change format then close the current device. */
666 if(pcm
&& !formats_equal(&playing
->format
, &pcm_format
))
669 snd_pcm_hw_params_t
*hwparams
;
670 snd_pcm_sw_params_t
*swparams
;
671 snd_pcm_uframes_t pcm_bufsize
;
673 int sample_format
= 0;
677 if((err
= snd_pcm_open(&pcm
,
679 SND_PCM_STREAM_PLAYBACK
,
680 SND_PCM_NONBLOCK
))) {
681 error(0, "error from snd_pcm_open: %d", err
);
684 snd_pcm_hw_params_alloca(&hwparams
);
685 D(("set up hw params"));
686 if((err
= snd_pcm_hw_params_any(pcm
, hwparams
)) < 0)
687 fatal(0, "error from snd_pcm_hw_params_any: %d", err
);
688 if((err
= snd_pcm_hw_params_set_access(pcm
, hwparams
,
689 SND_PCM_ACCESS_RW_INTERLEAVED
)) < 0)
690 fatal(0, "error from snd_pcm_hw_params_set_access: %d", err
);
691 switch(playing
->format
.bits
) {
693 sample_format
= SND_PCM_FORMAT_S8
;
696 switch(playing
->format
.byte_format
) {
697 case AO_FMT_NATIVE
: sample_format
= SND_PCM_FORMAT_S16
; break;
698 case AO_FMT_LITTLE
: sample_format
= SND_PCM_FORMAT_S16_LE
; break;
699 case AO_FMT_BIG
: sample_format
= SND_PCM_FORMAT_S16_BE
; break;
700 error(0, "unrecognized byte format %d", playing
->format
.byte_format
);
705 error(0, "unsupported sample size %d", playing
->format
.bits
);
708 if((err
= snd_pcm_hw_params_set_format(pcm
, hwparams
,
709 sample_format
)) < 0) {
710 error(0, "error from snd_pcm_hw_params_set_format (%d): %d",
714 rate
= playing
->format
.rate
;
715 if((err
= snd_pcm_hw_params_set_rate_near(pcm
, hwparams
, &rate
, 0)) < 0) {
716 error(0, "error from snd_pcm_hw_params_set_rate (%d): %d",
717 playing
->format
.rate
, err
);
720 if(rate
!= (unsigned)playing
->format
.rate
)
721 info("want rate %d, got %u", playing
->format
.rate
, rate
);
722 if((err
= snd_pcm_hw_params_set_channels(pcm
, hwparams
,
723 playing
->format
.channels
)) < 0) {
724 error(0, "error from snd_pcm_hw_params_set_channels (%d): %d",
725 playing
->format
.channels
, err
);
728 bufsize
= 3 * FRAMES
;
729 pcm_bufsize
= bufsize
;
730 if((err
= snd_pcm_hw_params_set_buffer_size_near(pcm
, hwparams
,
732 fatal(0, "error from snd_pcm_hw_params_set_buffer_size (%d): %d",
734 if(pcm_bufsize
!= 3 * FRAMES
&& pcm_bufsize
!= last_pcm_bufsize
)
735 info("asked for PCM buffer of %d frames, got %d",
736 3 * FRAMES
, (int)pcm_bufsize
);
737 last_pcm_bufsize
= pcm_bufsize
;
738 if((err
= snd_pcm_hw_params(pcm
, hwparams
)) < 0)
739 fatal(0, "error calling snd_pcm_hw_params: %d", err
);
740 D(("set up sw params"));
741 snd_pcm_sw_params_alloca(&swparams
);
742 if((err
= snd_pcm_sw_params_current(pcm
, swparams
)) < 0)
743 fatal(0, "error calling snd_pcm_sw_params_current: %d", err
);
744 if((err
= snd_pcm_sw_params_set_avail_min(pcm
, swparams
, FRAMES
)) < 0)
745 fatal(0, "error calling snd_pcm_sw_params_set_avail_min %d: %d",
747 if((err
= snd_pcm_sw_params(pcm
, swparams
)) < 0)
748 fatal(0, "error calling snd_pcm_sw_params: %d", err
);
749 pcm_format
= playing
->format
;
750 bpf
= bytes_per_frame(&pcm_format
);
751 D(("acquired audio device"));
752 log_params(hwparams
, swparams
);
759 /* We assume the error is temporary and that we'll retry in a bit. */
767 /** @brief Play via ALSA */
768 static size_t alsa_play(size_t frames
) {
769 snd_pcm_sframes_t pcm_written_frames
;
772 pcm_written_frames
= snd_pcm_writei(pcm
,
773 playing
->buffer
+ playing
->start
,
775 D(("actually play %zu frames, wrote %d",
776 frames
, (int)pcm_written_frames
));
777 if(pcm_written_frames
< 0) {
778 switch(pcm_written_frames
) {
779 case -EPIPE
: /* underrun */
780 error(0, "snd_pcm_writei reports underrun");
781 if((err
= snd_pcm_prepare(pcm
)) < 0)
782 fatal(0, "error calling snd_pcm_prepare: %d", err
);
787 fatal(0, "error calling snd_pcm_writei: %d",
788 (int)pcm_written_frames
);
791 return pcm_written_frames
;
794 /** @brief ALSA deactivation */
795 static void alsa_deactivate(void) {
799 if((err
= snd_pcm_nonblock(pcm
, 0)) < 0)
800 fatal(0, "error calling snd_pcm_nonblock: %d", err
);
807 D(("released audio device"));
812 /** @brief Command backend initialization */
813 static void command_init(void) {
814 info("selected command backend");
818 /** @brief Play to a subprocess */
819 static size_t command_play(size_t frames
) {
820 size_t bytes
= frames
* bpf
;
823 written_bytes
= write(cmdfd
, playing
->buffer
+ playing
->start
, bytes
);
824 D(("actually play %zu bytes, wrote %d",
825 bytes
, written_bytes
));
826 if(written_bytes
< 0) {
829 error(0, "hmm, command died; trying another");
835 fatal(errno
, "error writing to subprocess");
838 return written_bytes
/ bpf
;
841 /** @brief Command/network backend activation */
842 static int generic_activate(void) {
844 bufsize
= 3 * FRAMES
;
845 bpf
= bytes_per_frame(&config
->sample_format
);
846 D(("acquired audio device"));
852 /** @brief Network backend initialization */
853 static void network_init(void) {
854 struct addrinfo
*res
, *sres
;
855 static const struct addrinfo pref
= {
865 static const struct addrinfo prefbind
= {
875 static const int one
= 1;
876 int sndbuf
, target_sndbuf
= 131072;
878 char *sockname
, *ssockname
;
880 res
= get_address(&config
->broadcast
, &pref
, &sockname
);
882 if(config
->broadcast_from
.n
) {
883 sres
= get_address(&config
->broadcast_from
, &prefbind
, &ssockname
);
887 if((bfd
= socket(res
->ai_family
,
889 res
->ai_protocol
)) < 0)
890 fatal(errno
, "error creating broadcast socket");
891 if(setsockopt(bfd
, SOL_SOCKET
, SO_BROADCAST
, &one
, sizeof one
) < 0)
892 fatal(errno
, "error setting SO_BROADCAST on broadcast socket");
894 if(getsockopt(bfd
, SOL_SOCKET
, SO_SNDBUF
,
896 fatal(errno
, "error getting SO_SNDBUF");
897 if(target_sndbuf
> sndbuf
) {
898 if(setsockopt(bfd
, SOL_SOCKET
, SO_SNDBUF
,
899 &target_sndbuf
, sizeof target_sndbuf
) < 0)
900 error(errno
, "error setting SO_SNDBUF to %d", target_sndbuf
);
902 info("changed socket send buffer size from %d to %d",
903 sndbuf
, target_sndbuf
);
905 info("default socket send buffer is %d",
907 /* We might well want to set additional broadcast- or multicast-related
909 if(sres
&& bind(bfd
, sres
->ai_addr
, sres
->ai_addrlen
) < 0)
910 fatal(errno
, "error binding broadcast socket to %s", ssockname
);
911 if(connect(bfd
, res
->ai_addr
, res
->ai_addrlen
) < 0)
912 fatal(errno
, "error connecting broadcast socket to %s", sockname
);
914 gcry_randomize(&rtp_id
, sizeof rtp_id
, GCRY_STRONG_RANDOM
);
915 info("selected network backend, sending to %s", sockname
);
916 if(config
->sample_format
.byte_format
!= AO_FMT_BIG
) {
917 info("forcing big-endian sample format");
918 config
->sample_format
.byte_format
= AO_FMT_BIG
;
922 /** @brief Play over the network */
923 static size_t network_play(size_t frames
) {
924 struct rtp_header header
;
926 size_t bytes
= frames
* bpf
, written_frames
;
928 /* We transmit using RTP (RFC3550) and attempt to conform to the internet
929 * AVT profile (RFC3551). */
932 /* There may have been a gap. Fix up the RTP time accordingly. */
935 uint64_t target_rtp_time
;
937 /* Find the current time */
938 xgettimeofday(&now
, 0);
939 /* Find the number of microseconds elapsed since rtp_time=0 */
940 delta
= tvsub_us(now
, rtp_time_0
);
941 assert(delta
<= UINT64_MAX
/ 88200);
942 target_rtp_time
= (delta
* playing
->format
.rate
943 * playing
->format
.channels
) / 1000000;
944 /* Overflows at ~6 years uptime with 44100Hz stereo */
946 /* rtp_time is the number of samples we've played. NB that we play
947 * RTP_AHEAD_MS ahead of ourselves, so it may legitimately be ahead of
948 * the value we deduce from time comparison.
950 * Suppose we have 1s track started at t=0, and another track begins to
951 * play at t=2s. Suppose RTP_AHEAD_MS=1000 and 44100Hz stereo. In that
952 * case we'll send 1s of audio as fast as we can, giving rtp_time=88200.
953 * rtp_time stops at this point.
955 * At t=2s we'll have calculated target_rtp_time=176400. In this case we
956 * set rtp_time=176400 and the player can correctly conclude that it
957 * should leave 1s between the tracks.
959 * Suppose instead that the second track arrives at t=0.5s, and that
960 * we've managed to transmit the whole of the first track already. We'll
961 * have target_rtp_time=44100.
963 * The desired behaviour is to play the second track back to back with
964 * first. In this case therefore we do not modify rtp_time.
966 * Is it ever right to reduce rtp_time? No; for that would imply
967 * transmitting packets with overlapping timestamp ranges, which does not
970 if(target_rtp_time
> rtp_time
) {
971 /* More time has elapsed than we've transmitted samples. That implies
972 * we've been 'sending' silence. */
973 info("advancing rtp_time by %"PRIu64
" samples",
974 target_rtp_time
- rtp_time
);
975 rtp_time
= target_rtp_time
;
976 } else if(target_rtp_time
< rtp_time
) {
977 const int64_t samples_ahead
= ((uint64_t)RTP_AHEAD_MS
978 * config
->sample_format
.rate
979 * config
->sample_format
.channels
982 if(target_rtp_time
+ samples_ahead
< rtp_time
) {
983 info("reversing rtp_time by %"PRIu64
" samples",
984 rtp_time
- target_rtp_time
);
988 header
.vpxcc
= 2 << 6; /* V=2, P=0, X=0, CC=0 */
989 header
.seq
= htons(rtp_seq
++);
990 header
.timestamp
= htonl((uint32_t)rtp_time
);
991 header
.ssrc
= rtp_id
;
992 header
.mpt
= (idled ?
0x80 : 0x00) | 10;
993 /* 10 = L16 = 16-bit x 2 x 44100KHz. We ought to deduce this value from
994 * the sample rate (in a library somewhere so that configuration.c can rule
995 * out invalid rates).
998 if(bytes
> NETWORK_BYTES
- sizeof header
) {
999 bytes
= NETWORK_BYTES
- sizeof header
;
1000 /* Always send a whole number of frames */
1001 bytes
-= bytes
% bpf
;
1003 /* "The RTP clock rate used for generating the RTP timestamp is independent
1004 * of the number of channels and the encoding; it equals the number of
1005 * sampling periods per second. For N-channel encodings, each sampling
1006 * period (say, 1/8000 of a second) generates N samples. (This terminology
1007 * is standard, but somewhat confusing, as the total number of samples
1008 * generated per second is then the sampling rate times the channel
1011 vec
[0].iov_base
= (void *)&header
;
1012 vec
[0].iov_len
= sizeof header
;
1013 vec
[1].iov_base
= playing
->buffer
+ playing
->start
;
1014 vec
[1].iov_len
= bytes
;
1016 written_bytes
= writev(bfd
, vec
, 2);
1017 } while(written_bytes
< 0 && errno
== EINTR
);
1018 if(written_bytes
< 0) {
1019 error(errno
, "error transmitting audio data");
1021 if(audio_errors
== 10)
1022 fatal(0, "too many audio errors");
1026 written_bytes
-= sizeof (struct rtp_header
);
1027 written_frames
= written_bytes
/ bpf
;
1028 /* Advance RTP's notion of the time */
1029 rtp_time
+= written_frames
* playing
->format
.channels
;
1030 return written_frames
;
1033 /** @brief Table of speaker backends */
1034 static const struct speaker_backend backends
[] = {
1061 { -1, 0, 0, 0, 0, 0 }
1064 int main(int argc
, char **argv
) {
1065 int n
, fd
, stdin_slot
, alsa_slots
, cmdfd_slot
, bfd_slot
, poke
, timeout
;
1067 struct speaker_message sm
;
1069 int alsa_nslots
= -1, err
;
1073 if(!setlocale(LC_CTYPE
, "")) fatal(errno
, "error calling setlocale");
1074 while((n
= getopt_long(argc
, argv
, "hVc:dD", options
, 0)) >= 0) {
1077 case 'V': version();
1078 case 'c': configfile
= optarg
; break;
1079 case 'd': debugging
= 1; break;
1080 case 'D': debugging
= 0; break;
1081 default: fatal(0, "invalid option");
1084 if(getenv("DISORDER_DEBUG_SPEAKER")) debugging
= 1;
1085 /* If stderr is a TTY then log there, otherwise to syslog. */
1087 openlog(progname
, LOG_PID
, LOG_DAEMON
);
1088 log_default
= &log_syslog
;
1090 if(config_read()) fatal(0, "cannot read configuration");
1091 /* ignore SIGPIPE */
1092 signal(SIGPIPE
, SIG_IGN
);
1094 signal(SIGCHLD
, reap
);
1095 /* set nice value */
1096 xnice(config
->nice_speaker
);
1099 /* make sure we're not root, whatever the config says */
1100 if(getuid() == 0 || geteuid() == 0) fatal(0, "do not run as root");
1101 /* identify the backend used to play */
1102 for(n
= 0; backends
[n
].backend
!= -1; ++n
)
1103 if(backends
[n
].backend
== config
->speaker_backend
)
1105 if(backends
[n
].backend
== -1)
1106 fatal(0, "unsupported backend %d", config
->speaker_backend
);
1107 backend
= &backends
[n
];
1108 /* backend-specific initialization */
1110 while(getppid() != 1) {
1112 /* Always ready for commands from the main server. */
1113 stdin_slot
= addfd(0, POLLIN
);
1114 /* Try to read sample data for the currently playing track if there is
1116 if(playing
&& !playing
->eof
&& playing
->used
< playing
->size
) {
1117 playing
->slot
= addfd(playing
->fd
, POLLIN
);
1120 /* If forceplay is set then wait until it succeeds before waiting on the
1125 /* By default we will wait up to a second before thinking about current
1128 if(ready
&& !forceplay
) {
1129 switch(config
->speaker_backend
) {
1130 case BACKEND_COMMAND
:
1131 /* We send sample data to the subprocess as fast as it can accept it.
1132 * This isn't ideal as pause latency can be very high as a result. */
1134 cmdfd_slot
= addfd(cmdfd
, POLLOUT
);
1136 case BACKEND_NETWORK
: {
1139 uint64_t target_rtp_time
;
1140 const int64_t samples_ahead
= ((uint64_t)RTP_AHEAD_MS
1141 * config
->sample_format
.rate
1142 * config
->sample_format
.channels
1145 static unsigned logit
;
1148 /* If we're starting then initialize the base time */
1150 xgettimeofday(&rtp_time_0
, 0);
1151 /* We send audio data whenever we get RTP_AHEAD seconds or more
1153 xgettimeofday(&now
, 0);
1154 target_us
= tvsub_us(now
, rtp_time_0
);
1155 assert(target_us
<= UINT64_MAX
/ 88200);
1156 target_rtp_time
= (target_us
* config
->sample_format
.rate
1157 * config
->sample_format
.channels
)
1161 /* TODO remove logging guff */
1162 if(!(logit
++ & 1023))
1163 info("rtp_time %llu target %llu difference %lld [%lld]",
1164 rtp_time
, target_rtp_time
,
1165 rtp_time
- target_rtp_time
,
1168 if((int64_t)(rtp_time
- target_rtp_time
) < samples_ahead
)
1169 bfd_slot
= addfd(bfd
, POLLOUT
);
1173 case BACKEND_ALSA
: {
1174 /* We send sample data to ALSA as fast as it can accept it, relying on
1175 * the fact that it has a relatively small buffer to minimize pause
1182 alsa_nslots
= snd_pcm_poll_descriptors(pcm
, &fds
[fdno
], NFDS
- fdno
);
1183 if((alsa_nslots
<= 0
1184 || !(fds
[alsa_slots
].events
& POLLOUT
))
1185 && snd_pcm_state(pcm
) == SND_PCM_STATE_XRUN
) {
1186 error(0, "underrun detected after call to snd_pcm_poll_descriptors()");
1187 if((err
= snd_pcm_prepare(pcm
)))
1188 fatal(0, "error calling snd_pcm_prepare: %d", err
);
1191 } while(retry
-- > 0);
1192 if(alsa_nslots
>= 0)
1193 fdno
+= alsa_nslots
;
1198 assert(!"unknown backend");
1201 /* If any other tracks don't have a full buffer, try to read sample data
1203 for(t
= tracks
; t
; t
= t
->next
)
1205 if(!t
->eof
&& t
->used
< t
->size
) {
1206 t
->slot
= addfd(t
->fd
, POLLIN
| POLLHUP
);
1210 /* Wait for something interesting to happen */
1211 n
= poll(fds
, fdno
, timeout
);
1213 if(errno
== EINTR
) continue;
1214 fatal(errno
, "error calling poll");
1216 /* Play some sound before doing anything else */
1218 switch(config
->speaker_backend
) {
1221 if(alsa_slots
!= -1) {
1222 unsigned short alsa_revents
;
1224 if((err
= snd_pcm_poll_descriptors_revents(pcm
,
1227 &alsa_revents
)) < 0)
1228 fatal(0, "error calling snd_pcm_poll_descriptors_revents: %d", err
);
1229 if(alsa_revents
& (POLLOUT
| POLLERR
))
1235 case BACKEND_COMMAND
:
1236 if(cmdfd_slot
!= -1) {
1237 if(fds
[cmdfd_slot
].revents
& (POLLOUT
| POLLERR
))
1242 case BACKEND_NETWORK
:
1243 if(bfd_slot
!= -1) {
1244 if(fds
[bfd_slot
].revents
& (POLLOUT
| POLLERR
))
1251 /* Some attempt to play must have failed */
1252 if(playing
&& !paused
)
1255 forceplay
= 0; /* just in case */
1257 /* Perhaps we have a command to process */
1258 if(fds
[stdin_slot
].revents
& POLLIN
) {
1259 n
= speaker_recv(0, &sm
, &fd
);
1263 D(("SM_PREPARE %s %d", sm
.id
, fd
));
1264 if(fd
== -1) fatal(0, "got SM_PREPARE but no file descriptor");
1265 t
= findtrack(sm
.id
, 1);
1269 D(("SM_PLAY %s %d", sm
.id
, fd
));
1270 if(playing
) fatal(0, "got SM_PLAY but already playing something");
1271 t
= findtrack(sm
.id
, 1);
1272 if(fd
!= -1) acquire(t
, fd
);
1292 D(("SM_CANCEL %s", sm
.id
));
1293 t
= removetrack(sm
.id
);
1296 sm
.type
= SM_FINISHED
;
1297 strcpy(sm
.id
, playing
->id
);
1298 speaker_send(1, &sm
, 0);
1303 error(0, "SM_CANCEL for unknown track %s", sm
.id
);
1308 if(config_read()) error(0, "cannot read configuration");
1309 info("reloaded configuration");
1312 error(0, "unknown message type %d", sm
.type
);
1315 /* Read in any buffered data */
1316 for(t
= tracks
; t
; t
= t
->next
)
1317 if(t
->slot
!= -1 && (fds
[t
->slot
].revents
& (POLLIN
| POLLHUP
)))
1319 /* We might be able to play now */
1320 if(ready
&& forceplay
&& playing
&& !paused
)
1322 /* Maybe we finished playing a track somewhere in the above */
1324 /* If we don't need the sound device for now then close it for the benefit
1325 * of anyone else who wants it. */
1326 if((!playing
|| paused
) && ready
)
1328 /* If we've not reported out state for a second do so now. */
1329 if(time(0) > last_report
)
1332 info("stopped (parent terminated)");
1341 indent-tabs-mode:nil