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);
243 /** @brief Called before poll()
245 * Called before the call to poll(). Should call addfd() to update the FD
246 * array and stash the slot number somewhere safe.
248 void (*beforepoll
)(void);
251 /** @brief Selected backend */
252 static const struct speaker_backend
*backend
;
254 static const struct option options
[] = {
255 { "help", no_argument
, 0, 'h' },
256 { "version", no_argument
, 0, 'V' },
257 { "config", required_argument
, 0, 'c' },
258 { "debug", no_argument
, 0, 'd' },
259 { "no-debug", no_argument
, 0, 'D' },
263 /* Display usage message and terminate. */
264 static void help(void) {
266 " disorder-speaker [OPTIONS]\n"
268 " --help, -h Display usage message\n"
269 " --version, -V Display version number\n"
270 " --config PATH, -c PATH Set configuration file\n"
271 " --debug, -d Turn on debugging\n"
273 "Speaker process for DisOrder. Not intended to be run\n"
279 /* Display version number and terminate. */
280 static void version(void) {
281 xprintf("disorder-speaker version %s\n", disorder_version_string
);
286 /** @brief Return the number of bytes per frame in @p format */
287 static size_t bytes_per_frame(const ao_sample_format
*format
) {
288 return format
->channels
* format
->bits
/ 8;
291 /** @brief Find track @p id, maybe creating it if not found */
292 static struct track
*findtrack(const char *id
, int create
) {
295 D(("findtrack %s %d", id
, create
));
296 for(t
= tracks
; t
&& strcmp(id
, t
->id
); t
= t
->next
)
299 t
= xmalloc(sizeof *t
);
304 /* The initial input buffer will be the sample format. */
305 t
->buffer
= (void *)&t
->format
;
306 t
->size
= sizeof t
->format
;
311 /** @brief Remove track @p id (but do not destroy it) */
312 static struct track
*removetrack(const char *id
) {
313 struct track
*t
, **tt
;
315 D(("removetrack %s", id
));
316 for(tt
= &tracks
; (t
= *tt
) && strcmp(id
, t
->id
); tt
= &t
->next
)
323 /** @brief Destroy a track */
324 static void destroy(struct track
*t
) {
325 D(("destroy %s", t
->id
));
326 if(t
->fd
!= -1) xclose(t
->fd
);
327 if(t
->buffer
!= (void *)&t
->format
) free(t
->buffer
);
331 /** @brief Notice a new connection */
332 static void acquire(struct track
*t
, int fd
) {
333 D(("acquire %s %d", t
->id
, fd
));
340 /** @brief Return true if A and B denote identical libao formats, else false */
341 static int formats_equal(const ao_sample_format
*a
,
342 const ao_sample_format
*b
) {
343 return (a
->bits
== b
->bits
344 && a
->rate
== b
->rate
345 && a
->channels
== b
->channels
346 && a
->byte_format
== b
->byte_format
);
349 /** @brief Compute arguments to sox */
350 static void soxargs(const char ***pp
, char **qq
, ao_sample_format
*ao
) {
355 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-r%d", ao
->rate
); *qq
+= n
+ 1;
356 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-c%d", ao
->channels
); *qq
+= n
+ 1;
357 /* sox 12.17.9 insists on -b etc; CVS sox insists on -<n> etc; both are
359 switch(config
->sox_generation
) {
362 && ao
->byte_format
!= AO_FMT_NATIVE
363 && ao
->byte_format
!= MACHINE_AO_FMT
) {
367 case 8: *(*pp
)++ = "-b"; break;
368 case 16: *(*pp
)++ = "-w"; break;
369 case 32: *(*pp
)++ = "-l"; break;
370 case 64: *(*pp
)++ = "-d"; break;
371 default: fatal(0, "cannot handle sample size %d", (int)ao
->bits
);
375 switch(ao
->byte_format
) {
376 case AO_FMT_NATIVE
: break;
377 case AO_FMT_BIG
: *(*pp
)++ = "-B"; break;
378 case AO_FMT_LITTLE
: *(*pp
)++ = "-L"; break;
380 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-%d", ao
->bits
/8); *qq
+= n
+ 1;
385 /** @brief Enable format translation
387 * If necessary, replaces a tracks inbound file descriptor with one connected
388 * to a sox invocation, which performs the required translation.
390 static void enable_translation(struct track
*t
) {
391 if((backend
->flags
& FIXED_FORMAT
)
392 && !formats_equal(&t
->format
, &config
->sample_format
)) {
393 char argbuf
[1024], *q
= argbuf
;
394 const char *av
[18], **pp
= av
;
399 soxargs(&pp
, &q
, &t
->format
);
401 soxargs(&pp
, &q
, &config
->sample_format
);
405 for(pp
= av
; *pp
; pp
++)
406 D(("sox arg[%d] = %s", pp
- av
, *pp
));
412 signal(SIGPIPE
, SIG_DFL
);
414 xdup2(soxpipe
[1], 1);
415 fcntl(0, F_SETFL
, fcntl(0, F_GETFL
) & ~O_NONBLOCK
);
419 execvp("sox", (char **)av
);
422 D(("forking sox for format conversion (kid = %d)", soxkid
));
426 t
->format
= config
->sample_format
;
430 /** @brief Read data into a sample buffer
431 * @param t Pointer to track
432 * @return 0 on success, -1 on EOF
434 * This is effectively the read callback on @c t->fd.
436 static int fill(struct track
*t
) {
440 D(("fill %s: eof=%d used=%zu size=%zu got_format=%d",
441 t
->id
, t
->eof
, t
->used
, t
->size
, t
->got_format
));
442 if(t
->eof
) return -1;
443 if(t
->used
< t
->size
) {
444 /* there is room left in the buffer */
445 where
= (t
->start
+ t
->used
) % t
->size
;
447 /* We are reading audio data, get as much as we can */
448 if(where
>= t
->start
) left
= t
->size
- where
;
449 else left
= t
->start
- where
;
451 /* We are still waiting for the format, only get that */
452 left
= sizeof (ao_sample_format
) - t
->used
;
454 n
= read(t
->fd
, t
->buffer
+ where
, left
);
455 } while(n
< 0 && errno
== EINTR
);
457 if(errno
!= EAGAIN
) fatal(errno
, "error reading sample stream");
461 D(("fill %s: eof detected", t
->id
));
466 if(!t
->got_format
&& t
->used
>= sizeof (ao_sample_format
)) {
467 assert(t
->used
== sizeof (ao_sample_format
));
468 /* Check that our assumptions are met. */
469 if(t
->format
.bits
& 7)
470 fatal(0, "bits per sample not a multiple of 8");
471 /* If the input format is unsuitable, arrange to translate it */
472 enable_translation(t
);
473 /* Make a new buffer for audio data. */
474 t
->size
= bytes_per_frame(&t
->format
) * t
->format
.rate
* BUFFER_SECONDS
;
475 t
->buffer
= xmalloc(t
->size
);
478 D(("got format for %s", t
->id
));
484 /** @brief Close the sound device */
485 static void idle(void) {
487 if(backend
->deactivate
)
488 backend
->deactivate();
493 /** @brief Abandon the current track */
494 static void abandon(void) {
495 struct speaker_message sm
;
498 memset(&sm
, 0, sizeof sm
);
499 sm
.type
= SM_FINISHED
;
500 strcpy(sm
.id
, playing
->id
);
501 speaker_send(1, &sm
, 0);
502 removetrack(playing
->id
);
509 /** @brief Log ALSA parameters */
510 static void log_params(snd_pcm_hw_params_t
*hwparams
,
511 snd_pcm_sw_params_t
*swparams
) {
515 return; /* too verbose */
520 snd_pcm_sw_params_get_silence_size(swparams
, &f
);
521 info("sw silence_size=%lu", (unsigned long)f
);
522 snd_pcm_sw_params_get_silence_threshold(swparams
, &f
);
523 info("sw silence_threshold=%lu", (unsigned long)f
);
524 snd_pcm_sw_params_get_sleep_min(swparams
, &u
);
525 info("sw sleep_min=%lu", (unsigned long)u
);
526 snd_pcm_sw_params_get_start_threshold(swparams
, &f
);
527 info("sw start_threshold=%lu", (unsigned long)f
);
528 snd_pcm_sw_params_get_stop_threshold(swparams
, &f
);
529 info("sw stop_threshold=%lu", (unsigned long)f
);
530 snd_pcm_sw_params_get_xfer_align(swparams
, &f
);
531 info("sw xfer_align=%lu", (unsigned long)f
);
536 /** @brief Enable sound output
538 * Makes sure the sound device is open and has the right sample format. Return
539 * 0 on success and -1 on error.
541 static int activate(void) {
542 /* If we don't know the format yet we cannot start. */
543 if(!playing
->got_format
) {
544 D((" - not got format for %s", playing
->id
));
547 return backend
->activate();
550 /* Check to see whether the current track has finished playing */
551 static void maybe_finished(void) {
554 && (!playing
->got_format
555 || playing
->used
< bytes_per_frame(&playing
->format
)))
559 static void fork_cmd(void) {
562 if(cmdfd
!= -1) close(cmdfd
);
566 signal(SIGPIPE
, SIG_DFL
);
570 execl("/bin/sh", "sh", "-c", config
->speaker_command
, (char *)0);
571 fatal(errno
, "error execing /bin/sh");
575 D(("forked cmd %d, fd = %d", cmdpid
, cmdfd
));
578 static void play(size_t frames
) {
579 size_t avail_frames
, avail_bytes
, written_frames
;
580 ssize_t written_bytes
;
582 /* Make sure the output device is activated */
587 forceplay
= 0; /* Must have called abandon() */
590 D(("play: play %zu/%zu%s %dHz %db %dc", frames
, playing
->used
/ bpf
,
591 playing
->eof ?
" EOF" : "",
592 playing
->format
.rate
,
593 playing
->format
.bits
,
594 playing
->format
.channels
));
595 /* If we haven't got enough bytes yet wait until we have. Exception: when
597 if(playing
->used
< frames
* bpf
&& !playing
->eof
) {
601 /* We have got enough data so don't force play again */
603 /* Figure out how many frames there are available to write */
604 if(playing
->start
+ playing
->used
> playing
->size
)
605 /* The ring buffer is currently wrapped, only play up to the wrap point */
606 avail_bytes
= playing
->size
- playing
->start
;
608 /* The ring buffer is not wrapped, can play the lot */
609 avail_bytes
= playing
->used
;
610 avail_frames
= avail_bytes
/ bpf
;
611 /* Only play up to the requested amount */
612 if(avail_frames
> frames
)
613 avail_frames
= frames
;
617 written_frames
= backend
->play(avail_frames
);
618 written_bytes
= written_frames
* bpf
;
619 /* written_bytes and written_frames had better both be set and correct by
621 playing
->start
+= written_bytes
;
622 playing
->used
-= written_bytes
;
623 playing
->played
+= written_frames
;
624 /* If the pointer is at the end of the buffer (or the buffer is completely
625 * empty) wrap it back to the start. */
626 if(!playing
->used
|| playing
->start
== playing
->size
)
628 frames
-= written_frames
;
631 /* Notify the server what we're up to. */
632 static void report(void) {
633 struct speaker_message sm
;
635 if(playing
&& playing
->buffer
!= (void *)&playing
->format
) {
636 memset(&sm
, 0, sizeof sm
);
637 sm
.type
= paused ? SM_PAUSED
: SM_PLAYING
;
638 strcpy(sm
.id
, playing
->id
);
639 sm
.data
= playing
->played
/ playing
->format
.rate
;
640 speaker_send(1, &sm
, 0);
645 static void reap(int __attribute__((unused
)) sig
) {
650 cmdpid
= waitpid(-1, &st
, WNOHANG
);
652 signal(SIGCHLD
, reap
);
655 static int addfd(int fd
, int events
) {
658 fds
[fdno
].events
= events
;
665 /** @brief ALSA backend initialization */
666 static void alsa_init(void) {
667 info("selected ALSA backend");
670 /** @brief ALSA backend activation */
671 static int alsa_activate(void) {
672 /* If we need to change format then close the current device. */
673 if(pcm
&& !formats_equal(&playing
->format
, &pcm_format
))
676 snd_pcm_hw_params_t
*hwparams
;
677 snd_pcm_sw_params_t
*swparams
;
678 snd_pcm_uframes_t pcm_bufsize
;
680 int sample_format
= 0;
684 if((err
= snd_pcm_open(&pcm
,
686 SND_PCM_STREAM_PLAYBACK
,
687 SND_PCM_NONBLOCK
))) {
688 error(0, "error from snd_pcm_open: %d", err
);
691 snd_pcm_hw_params_alloca(&hwparams
);
692 D(("set up hw params"));
693 if((err
= snd_pcm_hw_params_any(pcm
, hwparams
)) < 0)
694 fatal(0, "error from snd_pcm_hw_params_any: %d", err
);
695 if((err
= snd_pcm_hw_params_set_access(pcm
, hwparams
,
696 SND_PCM_ACCESS_RW_INTERLEAVED
)) < 0)
697 fatal(0, "error from snd_pcm_hw_params_set_access: %d", err
);
698 switch(playing
->format
.bits
) {
700 sample_format
= SND_PCM_FORMAT_S8
;
703 switch(playing
->format
.byte_format
) {
704 case AO_FMT_NATIVE
: sample_format
= SND_PCM_FORMAT_S16
; break;
705 case AO_FMT_LITTLE
: sample_format
= SND_PCM_FORMAT_S16_LE
; break;
706 case AO_FMT_BIG
: sample_format
= SND_PCM_FORMAT_S16_BE
; break;
707 error(0, "unrecognized byte format %d", playing
->format
.byte_format
);
712 error(0, "unsupported sample size %d", playing
->format
.bits
);
715 if((err
= snd_pcm_hw_params_set_format(pcm
, hwparams
,
716 sample_format
)) < 0) {
717 error(0, "error from snd_pcm_hw_params_set_format (%d): %d",
721 rate
= playing
->format
.rate
;
722 if((err
= snd_pcm_hw_params_set_rate_near(pcm
, hwparams
, &rate
, 0)) < 0) {
723 error(0, "error from snd_pcm_hw_params_set_rate (%d): %d",
724 playing
->format
.rate
, err
);
727 if(rate
!= (unsigned)playing
->format
.rate
)
728 info("want rate %d, got %u", playing
->format
.rate
, rate
);
729 if((err
= snd_pcm_hw_params_set_channels(pcm
, hwparams
,
730 playing
->format
.channels
)) < 0) {
731 error(0, "error from snd_pcm_hw_params_set_channels (%d): %d",
732 playing
->format
.channels
, err
);
735 bufsize
= 3 * FRAMES
;
736 pcm_bufsize
= bufsize
;
737 if((err
= snd_pcm_hw_params_set_buffer_size_near(pcm
, hwparams
,
739 fatal(0, "error from snd_pcm_hw_params_set_buffer_size (%d): %d",
741 if(pcm_bufsize
!= 3 * FRAMES
&& pcm_bufsize
!= last_pcm_bufsize
)
742 info("asked for PCM buffer of %d frames, got %d",
743 3 * FRAMES
, (int)pcm_bufsize
);
744 last_pcm_bufsize
= pcm_bufsize
;
745 if((err
= snd_pcm_hw_params(pcm
, hwparams
)) < 0)
746 fatal(0, "error calling snd_pcm_hw_params: %d", err
);
747 D(("set up sw params"));
748 snd_pcm_sw_params_alloca(&swparams
);
749 if((err
= snd_pcm_sw_params_current(pcm
, swparams
)) < 0)
750 fatal(0, "error calling snd_pcm_sw_params_current: %d", err
);
751 if((err
= snd_pcm_sw_params_set_avail_min(pcm
, swparams
, FRAMES
)) < 0)
752 fatal(0, "error calling snd_pcm_sw_params_set_avail_min %d: %d",
754 if((err
= snd_pcm_sw_params(pcm
, swparams
)) < 0)
755 fatal(0, "error calling snd_pcm_sw_params: %d", err
);
756 pcm_format
= playing
->format
;
757 bpf
= bytes_per_frame(&pcm_format
);
758 D(("acquired audio device"));
759 log_params(hwparams
, swparams
);
766 /* We assume the error is temporary and that we'll retry in a bit. */
774 /** @brief Play via ALSA */
775 static size_t alsa_play(size_t frames
) {
776 snd_pcm_sframes_t pcm_written_frames
;
779 pcm_written_frames
= snd_pcm_writei(pcm
,
780 playing
->buffer
+ playing
->start
,
782 D(("actually play %zu frames, wrote %d",
783 frames
, (int)pcm_written_frames
));
784 if(pcm_written_frames
< 0) {
785 switch(pcm_written_frames
) {
786 case -EPIPE
: /* underrun */
787 error(0, "snd_pcm_writei reports underrun");
788 if((err
= snd_pcm_prepare(pcm
)) < 0)
789 fatal(0, "error calling snd_pcm_prepare: %d", err
);
794 fatal(0, "error calling snd_pcm_writei: %d",
795 (int)pcm_written_frames
);
798 return pcm_written_frames
;
801 static int alsa_slots
, alsa_nslots
= -1;
803 /** @brief Fill in poll fd array for ALSA */
804 static void alsa_beforepoll(void) {
805 /* We send sample data to ALSA as fast as it can accept it, relying on
806 * the fact that it has a relatively small buffer to minimize pause
813 alsa_nslots
= snd_pcm_poll_descriptors(pcm
, &fds
[fdno
], NFDS
- fdno
);
815 || !(fds
[alsa_slots
].events
& POLLOUT
))
816 && snd_pcm_state(pcm
) == SND_PCM_STATE_XRUN
) {
817 error(0, "underrun detected after call to snd_pcm_poll_descriptors()");
818 if((err
= snd_pcm_prepare(pcm
)))
819 fatal(0, "error calling snd_pcm_prepare: %d", err
);
822 } while(retry
-- > 0);
827 /** @brief ALSA deactivation */
828 static void alsa_deactivate(void) {
832 if((err
= snd_pcm_nonblock(pcm
, 0)) < 0)
833 fatal(0, "error calling snd_pcm_nonblock: %d", err
);
840 D(("released audio device"));
845 /** @brief Command backend initialization */
846 static void command_init(void) {
847 info("selected command backend");
851 /** @brief Play to a subprocess */
852 static size_t command_play(size_t frames
) {
853 size_t bytes
= frames
* bpf
;
856 written_bytes
= write(cmdfd
, playing
->buffer
+ playing
->start
, bytes
);
857 D(("actually play %zu bytes, wrote %d",
858 bytes
, written_bytes
));
859 if(written_bytes
< 0) {
862 error(0, "hmm, command died; trying another");
868 fatal(errno
, "error writing to subprocess");
871 return written_bytes
/ bpf
;
874 static int cmdfd_slot
;
876 /** @brief Update poll array for writing to subprocess */
877 static void command_beforepoll(void) {
878 /* We send sample data to the subprocess as fast as it can accept it.
879 * This isn't ideal as pause latency can be very high as a result. */
881 cmdfd_slot
= addfd(cmdfd
, POLLOUT
);
884 /** @brief Command/network backend activation */
885 static int generic_activate(void) {
887 bufsize
= 3 * FRAMES
;
888 bpf
= bytes_per_frame(&config
->sample_format
);
889 D(("acquired audio device"));
895 /** @brief Network backend initialization */
896 static void network_init(void) {
897 struct addrinfo
*res
, *sres
;
898 static const struct addrinfo pref
= {
908 static const struct addrinfo prefbind
= {
918 static const int one
= 1;
919 int sndbuf
, target_sndbuf
= 131072;
921 char *sockname
, *ssockname
;
923 res
= get_address(&config
->broadcast
, &pref
, &sockname
);
925 if(config
->broadcast_from
.n
) {
926 sres
= get_address(&config
->broadcast_from
, &prefbind
, &ssockname
);
930 if((bfd
= socket(res
->ai_family
,
932 res
->ai_protocol
)) < 0)
933 fatal(errno
, "error creating broadcast socket");
934 if(setsockopt(bfd
, SOL_SOCKET
, SO_BROADCAST
, &one
, sizeof one
) < 0)
935 fatal(errno
, "error setting SO_BROADCAST on broadcast socket");
937 if(getsockopt(bfd
, SOL_SOCKET
, SO_SNDBUF
,
939 fatal(errno
, "error getting SO_SNDBUF");
940 if(target_sndbuf
> sndbuf
) {
941 if(setsockopt(bfd
, SOL_SOCKET
, SO_SNDBUF
,
942 &target_sndbuf
, sizeof target_sndbuf
) < 0)
943 error(errno
, "error setting SO_SNDBUF to %d", target_sndbuf
);
945 info("changed socket send buffer size from %d to %d",
946 sndbuf
, target_sndbuf
);
948 info("default socket send buffer is %d",
950 /* We might well want to set additional broadcast- or multicast-related
952 if(sres
&& bind(bfd
, sres
->ai_addr
, sres
->ai_addrlen
) < 0)
953 fatal(errno
, "error binding broadcast socket to %s", ssockname
);
954 if(connect(bfd
, res
->ai_addr
, res
->ai_addrlen
) < 0)
955 fatal(errno
, "error connecting broadcast socket to %s", sockname
);
957 gcry_randomize(&rtp_id
, sizeof rtp_id
, GCRY_STRONG_RANDOM
);
958 info("selected network backend, sending to %s", sockname
);
959 if(config
->sample_format
.byte_format
!= AO_FMT_BIG
) {
960 info("forcing big-endian sample format");
961 config
->sample_format
.byte_format
= AO_FMT_BIG
;
965 /** @brief Play over the network */
966 static size_t network_play(size_t frames
) {
967 struct rtp_header header
;
969 size_t bytes
= frames
* bpf
, written_frames
;
971 /* We transmit using RTP (RFC3550) and attempt to conform to the internet
972 * AVT profile (RFC3551). */
975 /* There may have been a gap. Fix up the RTP time accordingly. */
978 uint64_t target_rtp_time
;
980 /* Find the current time */
981 xgettimeofday(&now
, 0);
982 /* Find the number of microseconds elapsed since rtp_time=0 */
983 delta
= tvsub_us(now
, rtp_time_0
);
984 assert(delta
<= UINT64_MAX
/ 88200);
985 target_rtp_time
= (delta
* playing
->format
.rate
986 * playing
->format
.channels
) / 1000000;
987 /* Overflows at ~6 years uptime with 44100Hz stereo */
989 /* rtp_time is the number of samples we've played. NB that we play
990 * RTP_AHEAD_MS ahead of ourselves, so it may legitimately be ahead of
991 * the value we deduce from time comparison.
993 * Suppose we have 1s track started at t=0, and another track begins to
994 * play at t=2s. Suppose RTP_AHEAD_MS=1000 and 44100Hz stereo. In that
995 * case we'll send 1s of audio as fast as we can, giving rtp_time=88200.
996 * rtp_time stops at this point.
998 * At t=2s we'll have calculated target_rtp_time=176400. In this case we
999 * set rtp_time=176400 and the player can correctly conclude that it
1000 * should leave 1s between the tracks.
1002 * Suppose instead that the second track arrives at t=0.5s, and that
1003 * we've managed to transmit the whole of the first track already. We'll
1004 * have target_rtp_time=44100.
1006 * The desired behaviour is to play the second track back to back with
1007 * first. In this case therefore we do not modify rtp_time.
1009 * Is it ever right to reduce rtp_time? No; for that would imply
1010 * transmitting packets with overlapping timestamp ranges, which does not
1013 if(target_rtp_time
> rtp_time
) {
1014 /* More time has elapsed than we've transmitted samples. That implies
1015 * we've been 'sending' silence. */
1016 info("advancing rtp_time by %"PRIu64
" samples",
1017 target_rtp_time
- rtp_time
);
1018 rtp_time
= target_rtp_time
;
1019 } else if(target_rtp_time
< rtp_time
) {
1020 const int64_t samples_ahead
= ((uint64_t)RTP_AHEAD_MS
1021 * config
->sample_format
.rate
1022 * config
->sample_format
.channels
1025 if(target_rtp_time
+ samples_ahead
< rtp_time
) {
1026 info("reversing rtp_time by %"PRIu64
" samples",
1027 rtp_time
- target_rtp_time
);
1031 header
.vpxcc
= 2 << 6; /* V=2, P=0, X=0, CC=0 */
1032 header
.seq
= htons(rtp_seq
++);
1033 header
.timestamp
= htonl((uint32_t)rtp_time
);
1034 header
.ssrc
= rtp_id
;
1035 header
.mpt
= (idled ?
0x80 : 0x00) | 10;
1036 /* 10 = L16 = 16-bit x 2 x 44100KHz. We ought to deduce this value from
1037 * the sample rate (in a library somewhere so that configuration.c can rule
1038 * out invalid rates).
1041 if(bytes
> NETWORK_BYTES
- sizeof header
) {
1042 bytes
= NETWORK_BYTES
- sizeof header
;
1043 /* Always send a whole number of frames */
1044 bytes
-= bytes
% bpf
;
1046 /* "The RTP clock rate used for generating the RTP timestamp is independent
1047 * of the number of channels and the encoding; it equals the number of
1048 * sampling periods per second. For N-channel encodings, each sampling
1049 * period (say, 1/8000 of a second) generates N samples. (This terminology
1050 * is standard, but somewhat confusing, as the total number of samples
1051 * generated per second is then the sampling rate times the channel
1054 vec
[0].iov_base
= (void *)&header
;
1055 vec
[0].iov_len
= sizeof header
;
1056 vec
[1].iov_base
= playing
->buffer
+ playing
->start
;
1057 vec
[1].iov_len
= bytes
;
1059 written_bytes
= writev(bfd
, vec
, 2);
1060 } while(written_bytes
< 0 && errno
== EINTR
);
1061 if(written_bytes
< 0) {
1062 error(errno
, "error transmitting audio data");
1064 if(audio_errors
== 10)
1065 fatal(0, "too many audio errors");
1069 written_bytes
-= sizeof (struct rtp_header
);
1070 written_frames
= written_bytes
/ bpf
;
1071 /* Advance RTP's notion of the time */
1072 rtp_time
+= written_frames
* playing
->format
.channels
;
1073 return written_frames
;
1076 static int bfd_slot
;
1078 /** @brief Set up poll array for network play */
1079 static void network_beforepoll(void) {
1082 uint64_t target_rtp_time
;
1083 const int64_t samples_ahead
= ((uint64_t)RTP_AHEAD_MS
1084 * config
->sample_format
.rate
1085 * config
->sample_format
.channels
1088 /* If we're starting then initialize the base time */
1090 xgettimeofday(&rtp_time_0
, 0);
1091 /* We send audio data whenever we get RTP_AHEAD seconds or more
1093 xgettimeofday(&now
, 0);
1094 target_us
= tvsub_us(now
, rtp_time_0
);
1095 assert(target_us
<= UINT64_MAX
/ 88200);
1096 target_rtp_time
= (target_us
* config
->sample_format
.rate
1097 * config
->sample_format
.channels
)
1099 if((int64_t)(rtp_time
- target_rtp_time
) < samples_ahead
)
1100 bfd_slot
= addfd(bfd
, POLLOUT
);
1103 /** @brief Table of speaker backends */
1104 static const struct speaker_backend backends
[] = {
1134 { -1, 0, 0, 0, 0, 0, 0 }
1137 int main(int argc
, char **argv
) {
1138 int n
, fd
, stdin_slot
, poke
, timeout
;
1140 struct speaker_message sm
;
1146 if(!setlocale(LC_CTYPE
, "")) fatal(errno
, "error calling setlocale");
1147 while((n
= getopt_long(argc
, argv
, "hVc:dD", options
, 0)) >= 0) {
1150 case 'V': version();
1151 case 'c': configfile
= optarg
; break;
1152 case 'd': debugging
= 1; break;
1153 case 'D': debugging
= 0; break;
1154 default: fatal(0, "invalid option");
1157 if(getenv("DISORDER_DEBUG_SPEAKER")) debugging
= 1;
1158 /* If stderr is a TTY then log there, otherwise to syslog. */
1160 openlog(progname
, LOG_PID
, LOG_DAEMON
);
1161 log_default
= &log_syslog
;
1163 if(config_read()) fatal(0, "cannot read configuration");
1164 /* ignore SIGPIPE */
1165 signal(SIGPIPE
, SIG_IGN
);
1167 signal(SIGCHLD
, reap
);
1168 /* set nice value */
1169 xnice(config
->nice_speaker
);
1172 /* make sure we're not root, whatever the config says */
1173 if(getuid() == 0 || geteuid() == 0) fatal(0, "do not run as root");
1174 /* identify the backend used to play */
1175 for(n
= 0; backends
[n
].backend
!= -1; ++n
)
1176 if(backends
[n
].backend
== config
->speaker_backend
)
1178 if(backends
[n
].backend
== -1)
1179 fatal(0, "unsupported backend %d", config
->speaker_backend
);
1180 backend
= &backends
[n
];
1181 /* backend-specific initialization */
1183 while(getppid() != 1) {
1185 /* Always ready for commands from the main server. */
1186 stdin_slot
= addfd(0, POLLIN
);
1187 /* Try to read sample data for the currently playing track if there is
1189 if(playing
&& !playing
->eof
&& playing
->used
< playing
->size
) {
1190 playing
->slot
= addfd(playing
->fd
, POLLIN
);
1193 /* If forceplay is set then wait until it succeeds before waiting on the
1198 /* By default we will wait up to a second before thinking about current
1201 /* We'll break the poll as soon as the underlying sound device is ready for
1203 if(ready
&& !forceplay
)
1204 backend
->beforepoll();
1205 /* If any other tracks don't have a full buffer, try to read sample data
1207 for(t
= tracks
; t
; t
= t
->next
)
1209 if(!t
->eof
&& t
->used
< t
->size
) {
1210 t
->slot
= addfd(t
->fd
, POLLIN
| POLLHUP
);
1214 /* Wait for something interesting to happen */
1215 n
= poll(fds
, fdno
, timeout
);
1217 if(errno
== EINTR
) continue;
1218 fatal(errno
, "error calling poll");
1220 /* Play some sound before doing anything else */
1222 switch(config
->speaker_backend
) {
1225 if(alsa_slots
!= -1) {
1226 unsigned short alsa_revents
;
1228 if((err
= snd_pcm_poll_descriptors_revents(pcm
,
1231 &alsa_revents
)) < 0)
1232 fatal(0, "error calling snd_pcm_poll_descriptors_revents: %d", err
);
1233 if(alsa_revents
& (POLLOUT
| POLLERR
))
1239 case BACKEND_COMMAND
:
1240 if(cmdfd_slot
!= -1) {
1241 if(fds
[cmdfd_slot
].revents
& (POLLOUT
| POLLERR
))
1246 case BACKEND_NETWORK
:
1247 if(bfd_slot
!= -1) {
1248 if(fds
[bfd_slot
].revents
& (POLLOUT
| POLLERR
))
1255 /* Some attempt to play must have failed */
1256 if(playing
&& !paused
)
1259 forceplay
= 0; /* just in case */
1261 /* Perhaps we have a command to process */
1262 if(fds
[stdin_slot
].revents
& POLLIN
) {
1263 n
= speaker_recv(0, &sm
, &fd
);
1267 D(("SM_PREPARE %s %d", sm
.id
, fd
));
1268 if(fd
== -1) fatal(0, "got SM_PREPARE but no file descriptor");
1269 t
= findtrack(sm
.id
, 1);
1273 D(("SM_PLAY %s %d", sm
.id
, fd
));
1274 if(playing
) fatal(0, "got SM_PLAY but already playing something");
1275 t
= findtrack(sm
.id
, 1);
1276 if(fd
!= -1) acquire(t
, fd
);
1296 D(("SM_CANCEL %s", sm
.id
));
1297 t
= removetrack(sm
.id
);
1300 sm
.type
= SM_FINISHED
;
1301 strcpy(sm
.id
, playing
->id
);
1302 speaker_send(1, &sm
, 0);
1307 error(0, "SM_CANCEL for unknown track %s", sm
.id
);
1312 if(config_read()) error(0, "cannot read configuration");
1313 info("reloaded configuration");
1316 error(0, "unknown message type %d", sm
.type
);
1319 /* Read in any buffered data */
1320 for(t
= tracks
; t
; t
= t
->next
)
1321 if(t
->slot
!= -1 && (fds
[t
->slot
].revents
& (POLLIN
| POLLHUP
)))
1323 /* We might be able to play now */
1324 if(ready
&& forceplay
&& playing
&& !paused
)
1326 /* Maybe we finished playing a track somewhere in the above */
1328 /* If we don't need the sound device for now then close it for the benefit
1329 * of anyone else who wants it. */
1330 if((!playing
|| paused
) && ready
)
1332 /* If we've not reported out state for a second do so now. */
1333 if(time(0) > last_report
)
1336 info("stopped (parent terminated)");
1345 indent-tabs-mode:nil