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);
250 /** @brief Called after poll()
251 * @return 0 if we could play, non-0 if not
253 * Called after the call to poll(). Should arrange to play some audio if the
254 * output device is ready.
256 * The return value should be 0 if the device was ready to play, or nonzero
259 int (*afterpoll
)(void);
262 /** @brief Selected backend */
263 static const struct speaker_backend
*backend
;
265 static const struct option options
[] = {
266 { "help", no_argument
, 0, 'h' },
267 { "version", no_argument
, 0, 'V' },
268 { "config", required_argument
, 0, 'c' },
269 { "debug", no_argument
, 0, 'd' },
270 { "no-debug", no_argument
, 0, 'D' },
274 /* Display usage message and terminate. */
275 static void help(void) {
277 " disorder-speaker [OPTIONS]\n"
279 " --help, -h Display usage message\n"
280 " --version, -V Display version number\n"
281 " --config PATH, -c PATH Set configuration file\n"
282 " --debug, -d Turn on debugging\n"
284 "Speaker process for DisOrder. Not intended to be run\n"
290 /* Display version number and terminate. */
291 static void version(void) {
292 xprintf("disorder-speaker version %s\n", disorder_version_string
);
297 /** @brief Return the number of bytes per frame in @p format */
298 static size_t bytes_per_frame(const ao_sample_format
*format
) {
299 return format
->channels
* format
->bits
/ 8;
302 /** @brief Find track @p id, maybe creating it if not found */
303 static struct track
*findtrack(const char *id
, int create
) {
306 D(("findtrack %s %d", id
, create
));
307 for(t
= tracks
; t
&& strcmp(id
, t
->id
); t
= t
->next
)
310 t
= xmalloc(sizeof *t
);
315 /* The initial input buffer will be the sample format. */
316 t
->buffer
= (void *)&t
->format
;
317 t
->size
= sizeof t
->format
;
322 /** @brief Remove track @p id (but do not destroy it) */
323 static struct track
*removetrack(const char *id
) {
324 struct track
*t
, **tt
;
326 D(("removetrack %s", id
));
327 for(tt
= &tracks
; (t
= *tt
) && strcmp(id
, t
->id
); tt
= &t
->next
)
334 /** @brief Destroy a track */
335 static void destroy(struct track
*t
) {
336 D(("destroy %s", t
->id
));
337 if(t
->fd
!= -1) xclose(t
->fd
);
338 if(t
->buffer
!= (void *)&t
->format
) free(t
->buffer
);
342 /** @brief Notice a new connection */
343 static void acquire(struct track
*t
, int fd
) {
344 D(("acquire %s %d", t
->id
, fd
));
351 /** @brief Return true if A and B denote identical libao formats, else false */
352 static int formats_equal(const ao_sample_format
*a
,
353 const ao_sample_format
*b
) {
354 return (a
->bits
== b
->bits
355 && a
->rate
== b
->rate
356 && a
->channels
== b
->channels
357 && a
->byte_format
== b
->byte_format
);
360 /** @brief Compute arguments to sox */
361 static void soxargs(const char ***pp
, char **qq
, ao_sample_format
*ao
) {
366 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-r%d", ao
->rate
); *qq
+= n
+ 1;
367 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-c%d", ao
->channels
); *qq
+= n
+ 1;
368 /* sox 12.17.9 insists on -b etc; CVS sox insists on -<n> etc; both are
370 switch(config
->sox_generation
) {
373 && ao
->byte_format
!= AO_FMT_NATIVE
374 && ao
->byte_format
!= MACHINE_AO_FMT
) {
378 case 8: *(*pp
)++ = "-b"; break;
379 case 16: *(*pp
)++ = "-w"; break;
380 case 32: *(*pp
)++ = "-l"; break;
381 case 64: *(*pp
)++ = "-d"; break;
382 default: fatal(0, "cannot handle sample size %d", (int)ao
->bits
);
386 switch(ao
->byte_format
) {
387 case AO_FMT_NATIVE
: break;
388 case AO_FMT_BIG
: *(*pp
)++ = "-B"; break;
389 case AO_FMT_LITTLE
: *(*pp
)++ = "-L"; break;
391 *(*pp
)++ = *qq
; n
= sprintf(*qq
, "-%d", ao
->bits
/8); *qq
+= n
+ 1;
396 /** @brief Enable format translation
398 * If necessary, replaces a tracks inbound file descriptor with one connected
399 * to a sox invocation, which performs the required translation.
401 static void enable_translation(struct track
*t
) {
402 if((backend
->flags
& FIXED_FORMAT
)
403 && !formats_equal(&t
->format
, &config
->sample_format
)) {
404 char argbuf
[1024], *q
= argbuf
;
405 const char *av
[18], **pp
= av
;
410 soxargs(&pp
, &q
, &t
->format
);
412 soxargs(&pp
, &q
, &config
->sample_format
);
416 for(pp
= av
; *pp
; pp
++)
417 D(("sox arg[%d] = %s", pp
- av
, *pp
));
423 signal(SIGPIPE
, SIG_DFL
);
425 xdup2(soxpipe
[1], 1);
426 fcntl(0, F_SETFL
, fcntl(0, F_GETFL
) & ~O_NONBLOCK
);
430 execvp("sox", (char **)av
);
433 D(("forking sox for format conversion (kid = %d)", soxkid
));
437 t
->format
= config
->sample_format
;
441 /** @brief Read data into a sample buffer
442 * @param t Pointer to track
443 * @return 0 on success, -1 on EOF
445 * This is effectively the read callback on @c t->fd.
447 static int fill(struct track
*t
) {
451 D(("fill %s: eof=%d used=%zu size=%zu got_format=%d",
452 t
->id
, t
->eof
, t
->used
, t
->size
, t
->got_format
));
453 if(t
->eof
) return -1;
454 if(t
->used
< t
->size
) {
455 /* there is room left in the buffer */
456 where
= (t
->start
+ t
->used
) % t
->size
;
458 /* We are reading audio data, get as much as we can */
459 if(where
>= t
->start
) left
= t
->size
- where
;
460 else left
= t
->start
- where
;
462 /* We are still waiting for the format, only get that */
463 left
= sizeof (ao_sample_format
) - t
->used
;
465 n
= read(t
->fd
, t
->buffer
+ where
, left
);
466 } while(n
< 0 && errno
== EINTR
);
468 if(errno
!= EAGAIN
) fatal(errno
, "error reading sample stream");
472 D(("fill %s: eof detected", t
->id
));
477 if(!t
->got_format
&& t
->used
>= sizeof (ao_sample_format
)) {
478 assert(t
->used
== sizeof (ao_sample_format
));
479 /* Check that our assumptions are met. */
480 if(t
->format
.bits
& 7)
481 fatal(0, "bits per sample not a multiple of 8");
482 /* If the input format is unsuitable, arrange to translate it */
483 enable_translation(t
);
484 /* Make a new buffer for audio data. */
485 t
->size
= bytes_per_frame(&t
->format
) * t
->format
.rate
* BUFFER_SECONDS
;
486 t
->buffer
= xmalloc(t
->size
);
489 D(("got format for %s", t
->id
));
495 /** @brief Close the sound device */
496 static void idle(void) {
498 if(backend
->deactivate
)
499 backend
->deactivate();
504 /** @brief Abandon the current track */
505 static void abandon(void) {
506 struct speaker_message sm
;
509 memset(&sm
, 0, sizeof sm
);
510 sm
.type
= SM_FINISHED
;
511 strcpy(sm
.id
, playing
->id
);
512 speaker_send(1, &sm
, 0);
513 removetrack(playing
->id
);
520 /** @brief Log ALSA parameters */
521 static void log_params(snd_pcm_hw_params_t
*hwparams
,
522 snd_pcm_sw_params_t
*swparams
) {
526 return; /* too verbose */
531 snd_pcm_sw_params_get_silence_size(swparams
, &f
);
532 info("sw silence_size=%lu", (unsigned long)f
);
533 snd_pcm_sw_params_get_silence_threshold(swparams
, &f
);
534 info("sw silence_threshold=%lu", (unsigned long)f
);
535 snd_pcm_sw_params_get_sleep_min(swparams
, &u
);
536 info("sw sleep_min=%lu", (unsigned long)u
);
537 snd_pcm_sw_params_get_start_threshold(swparams
, &f
);
538 info("sw start_threshold=%lu", (unsigned long)f
);
539 snd_pcm_sw_params_get_stop_threshold(swparams
, &f
);
540 info("sw stop_threshold=%lu", (unsigned long)f
);
541 snd_pcm_sw_params_get_xfer_align(swparams
, &f
);
542 info("sw xfer_align=%lu", (unsigned long)f
);
547 /** @brief Enable sound output
549 * Makes sure the sound device is open and has the right sample format. Return
550 * 0 on success and -1 on error.
552 static int activate(void) {
553 /* If we don't know the format yet we cannot start. */
554 if(!playing
->got_format
) {
555 D((" - not got format for %s", playing
->id
));
558 return backend
->activate();
561 /* Check to see whether the current track has finished playing */
562 static void maybe_finished(void) {
565 && (!playing
->got_format
566 || playing
->used
< bytes_per_frame(&playing
->format
)))
570 static void fork_cmd(void) {
573 if(cmdfd
!= -1) close(cmdfd
);
577 signal(SIGPIPE
, SIG_DFL
);
581 execl("/bin/sh", "sh", "-c", config
->speaker_command
, (char *)0);
582 fatal(errno
, "error execing /bin/sh");
586 D(("forked cmd %d, fd = %d", cmdpid
, cmdfd
));
589 static void play(size_t frames
) {
590 size_t avail_frames
, avail_bytes
, written_frames
;
591 ssize_t written_bytes
;
593 /* Make sure the output device is activated */
598 forceplay
= 0; /* Must have called abandon() */
601 D(("play: play %zu/%zu%s %dHz %db %dc", frames
, playing
->used
/ bpf
,
602 playing
->eof ?
" EOF" : "",
603 playing
->format
.rate
,
604 playing
->format
.bits
,
605 playing
->format
.channels
));
606 /* If we haven't got enough bytes yet wait until we have. Exception: when
608 if(playing
->used
< frames
* bpf
&& !playing
->eof
) {
612 /* We have got enough data so don't force play again */
614 /* Figure out how many frames there are available to write */
615 if(playing
->start
+ playing
->used
> playing
->size
)
616 /* The ring buffer is currently wrapped, only play up to the wrap point */
617 avail_bytes
= playing
->size
- playing
->start
;
619 /* The ring buffer is not wrapped, can play the lot */
620 avail_bytes
= playing
->used
;
621 avail_frames
= avail_bytes
/ bpf
;
622 /* Only play up to the requested amount */
623 if(avail_frames
> frames
)
624 avail_frames
= frames
;
628 written_frames
= backend
->play(avail_frames
);
629 written_bytes
= written_frames
* bpf
;
630 /* written_bytes and written_frames had better both be set and correct by
632 playing
->start
+= written_bytes
;
633 playing
->used
-= written_bytes
;
634 playing
->played
+= written_frames
;
635 /* If the pointer is at the end of the buffer (or the buffer is completely
636 * empty) wrap it back to the start. */
637 if(!playing
->used
|| playing
->start
== playing
->size
)
639 frames
-= written_frames
;
642 /* Notify the server what we're up to. */
643 static void report(void) {
644 struct speaker_message sm
;
646 if(playing
&& playing
->buffer
!= (void *)&playing
->format
) {
647 memset(&sm
, 0, sizeof sm
);
648 sm
.type
= paused ? SM_PAUSED
: SM_PLAYING
;
649 strcpy(sm
.id
, playing
->id
);
650 sm
.data
= playing
->played
/ playing
->format
.rate
;
651 speaker_send(1, &sm
, 0);
656 static void reap(int __attribute__((unused
)) sig
) {
661 cmdpid
= waitpid(-1, &st
, WNOHANG
);
663 signal(SIGCHLD
, reap
);
666 static int addfd(int fd
, int events
) {
669 fds
[fdno
].events
= events
;
676 /** @brief ALSA backend initialization */
677 static void alsa_init(void) {
678 info("selected ALSA backend");
681 /** @brief ALSA backend activation */
682 static int alsa_activate(void) {
683 /* If we need to change format then close the current device. */
684 if(pcm
&& !formats_equal(&playing
->format
, &pcm_format
))
687 snd_pcm_hw_params_t
*hwparams
;
688 snd_pcm_sw_params_t
*swparams
;
689 snd_pcm_uframes_t pcm_bufsize
;
691 int sample_format
= 0;
695 if((err
= snd_pcm_open(&pcm
,
697 SND_PCM_STREAM_PLAYBACK
,
698 SND_PCM_NONBLOCK
))) {
699 error(0, "error from snd_pcm_open: %d", err
);
702 snd_pcm_hw_params_alloca(&hwparams
);
703 D(("set up hw params"));
704 if((err
= snd_pcm_hw_params_any(pcm
, hwparams
)) < 0)
705 fatal(0, "error from snd_pcm_hw_params_any: %d", err
);
706 if((err
= snd_pcm_hw_params_set_access(pcm
, hwparams
,
707 SND_PCM_ACCESS_RW_INTERLEAVED
)) < 0)
708 fatal(0, "error from snd_pcm_hw_params_set_access: %d", err
);
709 switch(playing
->format
.bits
) {
711 sample_format
= SND_PCM_FORMAT_S8
;
714 switch(playing
->format
.byte_format
) {
715 case AO_FMT_NATIVE
: sample_format
= SND_PCM_FORMAT_S16
; break;
716 case AO_FMT_LITTLE
: sample_format
= SND_PCM_FORMAT_S16_LE
; break;
717 case AO_FMT_BIG
: sample_format
= SND_PCM_FORMAT_S16_BE
; break;
718 error(0, "unrecognized byte format %d", playing
->format
.byte_format
);
723 error(0, "unsupported sample size %d", playing
->format
.bits
);
726 if((err
= snd_pcm_hw_params_set_format(pcm
, hwparams
,
727 sample_format
)) < 0) {
728 error(0, "error from snd_pcm_hw_params_set_format (%d): %d",
732 rate
= playing
->format
.rate
;
733 if((err
= snd_pcm_hw_params_set_rate_near(pcm
, hwparams
, &rate
, 0)) < 0) {
734 error(0, "error from snd_pcm_hw_params_set_rate (%d): %d",
735 playing
->format
.rate
, err
);
738 if(rate
!= (unsigned)playing
->format
.rate
)
739 info("want rate %d, got %u", playing
->format
.rate
, rate
);
740 if((err
= snd_pcm_hw_params_set_channels(pcm
, hwparams
,
741 playing
->format
.channels
)) < 0) {
742 error(0, "error from snd_pcm_hw_params_set_channels (%d): %d",
743 playing
->format
.channels
, err
);
746 bufsize
= 3 * FRAMES
;
747 pcm_bufsize
= bufsize
;
748 if((err
= snd_pcm_hw_params_set_buffer_size_near(pcm
, hwparams
,
750 fatal(0, "error from snd_pcm_hw_params_set_buffer_size (%d): %d",
752 if(pcm_bufsize
!= 3 * FRAMES
&& pcm_bufsize
!= last_pcm_bufsize
)
753 info("asked for PCM buffer of %d frames, got %d",
754 3 * FRAMES
, (int)pcm_bufsize
);
755 last_pcm_bufsize
= pcm_bufsize
;
756 if((err
= snd_pcm_hw_params(pcm
, hwparams
)) < 0)
757 fatal(0, "error calling snd_pcm_hw_params: %d", err
);
758 D(("set up sw params"));
759 snd_pcm_sw_params_alloca(&swparams
);
760 if((err
= snd_pcm_sw_params_current(pcm
, swparams
)) < 0)
761 fatal(0, "error calling snd_pcm_sw_params_current: %d", err
);
762 if((err
= snd_pcm_sw_params_set_avail_min(pcm
, swparams
, FRAMES
)) < 0)
763 fatal(0, "error calling snd_pcm_sw_params_set_avail_min %d: %d",
765 if((err
= snd_pcm_sw_params(pcm
, swparams
)) < 0)
766 fatal(0, "error calling snd_pcm_sw_params: %d", err
);
767 pcm_format
= playing
->format
;
768 bpf
= bytes_per_frame(&pcm_format
);
769 D(("acquired audio device"));
770 log_params(hwparams
, swparams
);
777 /* We assume the error is temporary and that we'll retry in a bit. */
785 /** @brief Play via ALSA */
786 static size_t alsa_play(size_t frames
) {
787 snd_pcm_sframes_t pcm_written_frames
;
790 pcm_written_frames
= snd_pcm_writei(pcm
,
791 playing
->buffer
+ playing
->start
,
793 D(("actually play %zu frames, wrote %d",
794 frames
, (int)pcm_written_frames
));
795 if(pcm_written_frames
< 0) {
796 switch(pcm_written_frames
) {
797 case -EPIPE
: /* underrun */
798 error(0, "snd_pcm_writei reports underrun");
799 if((err
= snd_pcm_prepare(pcm
)) < 0)
800 fatal(0, "error calling snd_pcm_prepare: %d", err
);
805 fatal(0, "error calling snd_pcm_writei: %d",
806 (int)pcm_written_frames
);
809 return pcm_written_frames
;
812 static int alsa_slots
, alsa_nslots
= -1;
814 /** @brief Fill in poll fd array for ALSA */
815 static void alsa_beforepoll(void) {
816 /* We send sample data to ALSA as fast as it can accept it, relying on
817 * the fact that it has a relatively small buffer to minimize pause
824 alsa_nslots
= snd_pcm_poll_descriptors(pcm
, &fds
[fdno
], NFDS
- fdno
);
826 || !(fds
[alsa_slots
].events
& POLLOUT
))
827 && snd_pcm_state(pcm
) == SND_PCM_STATE_XRUN
) {
828 error(0, "underrun detected after call to snd_pcm_poll_descriptors()");
829 if((err
= snd_pcm_prepare(pcm
)))
830 fatal(0, "error calling snd_pcm_prepare: %d", err
);
833 } while(retry
-- > 0);
838 /** @brief Process poll() results for ALSA */
839 static int alsa_afterpoll(void) {
842 if(alsa_slots
!= -1) {
843 unsigned short alsa_revents
;
845 if((err
= snd_pcm_poll_descriptors_revents(pcm
,
849 fatal(0, "error calling snd_pcm_poll_descriptors_revents: %d", err
);
850 if(alsa_revents
& (POLLOUT
| POLLERR
))
857 /** @brief ALSA deactivation */
858 static void alsa_deactivate(void) {
862 if((err
= snd_pcm_nonblock(pcm
, 0)) < 0)
863 fatal(0, "error calling snd_pcm_nonblock: %d", err
);
870 D(("released audio device"));
875 /** @brief Command backend initialization */
876 static void command_init(void) {
877 info("selected command backend");
881 /** @brief Play to a subprocess */
882 static size_t command_play(size_t frames
) {
883 size_t bytes
= frames
* bpf
;
886 written_bytes
= write(cmdfd
, playing
->buffer
+ playing
->start
, bytes
);
887 D(("actually play %zu bytes, wrote %d",
888 bytes
, written_bytes
));
889 if(written_bytes
< 0) {
892 error(0, "hmm, command died; trying another");
898 fatal(errno
, "error writing to subprocess");
901 return written_bytes
/ bpf
;
904 static int cmdfd_slot
;
906 /** @brief Update poll array for writing to subprocess */
907 static void command_beforepoll(void) {
908 /* We send sample data to the subprocess as fast as it can accept it.
909 * This isn't ideal as pause latency can be very high as a result. */
911 cmdfd_slot
= addfd(cmdfd
, POLLOUT
);
914 /** @brief Process poll() results for subprocess play */
915 static int command_afterpoll(void) {
916 if(cmdfd_slot
!= -1) {
917 if(fds
[cmdfd_slot
].revents
& (POLLOUT
| POLLERR
))
924 /** @brief Command/network backend activation */
925 static int generic_activate(void) {
927 bufsize
= 3 * FRAMES
;
928 bpf
= bytes_per_frame(&config
->sample_format
);
929 D(("acquired audio device"));
935 /** @brief Network backend initialization */
936 static void network_init(void) {
937 struct addrinfo
*res
, *sres
;
938 static const struct addrinfo pref
= {
948 static const struct addrinfo prefbind
= {
958 static const int one
= 1;
959 int sndbuf
, target_sndbuf
= 131072;
961 char *sockname
, *ssockname
;
963 res
= get_address(&config
->broadcast
, &pref
, &sockname
);
965 if(config
->broadcast_from
.n
) {
966 sres
= get_address(&config
->broadcast_from
, &prefbind
, &ssockname
);
970 if((bfd
= socket(res
->ai_family
,
972 res
->ai_protocol
)) < 0)
973 fatal(errno
, "error creating broadcast socket");
974 if(setsockopt(bfd
, SOL_SOCKET
, SO_BROADCAST
, &one
, sizeof one
) < 0)
975 fatal(errno
, "error setting SO_BROADCAST on broadcast socket");
977 if(getsockopt(bfd
, SOL_SOCKET
, SO_SNDBUF
,
979 fatal(errno
, "error getting SO_SNDBUF");
980 if(target_sndbuf
> sndbuf
) {
981 if(setsockopt(bfd
, SOL_SOCKET
, SO_SNDBUF
,
982 &target_sndbuf
, sizeof target_sndbuf
) < 0)
983 error(errno
, "error setting SO_SNDBUF to %d", target_sndbuf
);
985 info("changed socket send buffer size from %d to %d",
986 sndbuf
, target_sndbuf
);
988 info("default socket send buffer is %d",
990 /* We might well want to set additional broadcast- or multicast-related
992 if(sres
&& bind(bfd
, sres
->ai_addr
, sres
->ai_addrlen
) < 0)
993 fatal(errno
, "error binding broadcast socket to %s", ssockname
);
994 if(connect(bfd
, res
->ai_addr
, res
->ai_addrlen
) < 0)
995 fatal(errno
, "error connecting broadcast socket to %s", sockname
);
997 gcry_randomize(&rtp_id
, sizeof rtp_id
, GCRY_STRONG_RANDOM
);
998 info("selected network backend, sending to %s", sockname
);
999 if(config
->sample_format
.byte_format
!= AO_FMT_BIG
) {
1000 info("forcing big-endian sample format");
1001 config
->sample_format
.byte_format
= AO_FMT_BIG
;
1005 /** @brief Play over the network */
1006 static size_t network_play(size_t frames
) {
1007 struct rtp_header header
;
1008 struct iovec vec
[2];
1009 size_t bytes
= frames
* bpf
, written_frames
;
1011 /* We transmit using RTP (RFC3550) and attempt to conform to the internet
1012 * AVT profile (RFC3551). */
1015 /* There may have been a gap. Fix up the RTP time accordingly. */
1018 uint64_t target_rtp_time
;
1020 /* Find the current time */
1021 xgettimeofday(&now
, 0);
1022 /* Find the number of microseconds elapsed since rtp_time=0 */
1023 delta
= tvsub_us(now
, rtp_time_0
);
1024 assert(delta
<= UINT64_MAX
/ 88200);
1025 target_rtp_time
= (delta
* playing
->format
.rate
1026 * playing
->format
.channels
) / 1000000;
1027 /* Overflows at ~6 years uptime with 44100Hz stereo */
1029 /* rtp_time is the number of samples we've played. NB that we play
1030 * RTP_AHEAD_MS ahead of ourselves, so it may legitimately be ahead of
1031 * the value we deduce from time comparison.
1033 * Suppose we have 1s track started at t=0, and another track begins to
1034 * play at t=2s. Suppose RTP_AHEAD_MS=1000 and 44100Hz stereo. In that
1035 * case we'll send 1s of audio as fast as we can, giving rtp_time=88200.
1036 * rtp_time stops at this point.
1038 * At t=2s we'll have calculated target_rtp_time=176400. In this case we
1039 * set rtp_time=176400 and the player can correctly conclude that it
1040 * should leave 1s between the tracks.
1042 * Suppose instead that the second track arrives at t=0.5s, and that
1043 * we've managed to transmit the whole of the first track already. We'll
1044 * have target_rtp_time=44100.
1046 * The desired behaviour is to play the second track back to back with
1047 * first. In this case therefore we do not modify rtp_time.
1049 * Is it ever right to reduce rtp_time? No; for that would imply
1050 * transmitting packets with overlapping timestamp ranges, which does not
1053 if(target_rtp_time
> rtp_time
) {
1054 /* More time has elapsed than we've transmitted samples. That implies
1055 * we've been 'sending' silence. */
1056 info("advancing rtp_time by %"PRIu64
" samples",
1057 target_rtp_time
- rtp_time
);
1058 rtp_time
= target_rtp_time
;
1059 } else if(target_rtp_time
< rtp_time
) {
1060 const int64_t samples_ahead
= ((uint64_t)RTP_AHEAD_MS
1061 * config
->sample_format
.rate
1062 * config
->sample_format
.channels
1065 if(target_rtp_time
+ samples_ahead
< rtp_time
) {
1066 info("reversing rtp_time by %"PRIu64
" samples",
1067 rtp_time
- target_rtp_time
);
1071 header
.vpxcc
= 2 << 6; /* V=2, P=0, X=0, CC=0 */
1072 header
.seq
= htons(rtp_seq
++);
1073 header
.timestamp
= htonl((uint32_t)rtp_time
);
1074 header
.ssrc
= rtp_id
;
1075 header
.mpt
= (idled ?
0x80 : 0x00) | 10;
1076 /* 10 = L16 = 16-bit x 2 x 44100KHz. We ought to deduce this value from
1077 * the sample rate (in a library somewhere so that configuration.c can rule
1078 * out invalid rates).
1081 if(bytes
> NETWORK_BYTES
- sizeof header
) {
1082 bytes
= NETWORK_BYTES
- sizeof header
;
1083 /* Always send a whole number of frames */
1084 bytes
-= bytes
% bpf
;
1086 /* "The RTP clock rate used for generating the RTP timestamp is independent
1087 * of the number of channels and the encoding; it equals the number of
1088 * sampling periods per second. For N-channel encodings, each sampling
1089 * period (say, 1/8000 of a second) generates N samples. (This terminology
1090 * is standard, but somewhat confusing, as the total number of samples
1091 * generated per second is then the sampling rate times the channel
1094 vec
[0].iov_base
= (void *)&header
;
1095 vec
[0].iov_len
= sizeof header
;
1096 vec
[1].iov_base
= playing
->buffer
+ playing
->start
;
1097 vec
[1].iov_len
= bytes
;
1099 written_bytes
= writev(bfd
, vec
, 2);
1100 } while(written_bytes
< 0 && errno
== EINTR
);
1101 if(written_bytes
< 0) {
1102 error(errno
, "error transmitting audio data");
1104 if(audio_errors
== 10)
1105 fatal(0, "too many audio errors");
1109 written_bytes
-= sizeof (struct rtp_header
);
1110 written_frames
= written_bytes
/ bpf
;
1111 /* Advance RTP's notion of the time */
1112 rtp_time
+= written_frames
* playing
->format
.channels
;
1113 return written_frames
;
1116 static int bfd_slot
;
1118 /** @brief Set up poll array for network play */
1119 static void network_beforepoll(void) {
1122 uint64_t target_rtp_time
;
1123 const int64_t samples_ahead
= ((uint64_t)RTP_AHEAD_MS
1124 * config
->sample_format
.rate
1125 * config
->sample_format
.channels
1128 /* If we're starting then initialize the base time */
1130 xgettimeofday(&rtp_time_0
, 0);
1131 /* We send audio data whenever we get RTP_AHEAD seconds or more
1133 xgettimeofday(&now
, 0);
1134 target_us
= tvsub_us(now
, rtp_time_0
);
1135 assert(target_us
<= UINT64_MAX
/ 88200);
1136 target_rtp_time
= (target_us
* config
->sample_format
.rate
1137 * config
->sample_format
.channels
)
1139 if((int64_t)(rtp_time
- target_rtp_time
) < samples_ahead
)
1140 bfd_slot
= addfd(bfd
, POLLOUT
);
1143 /** @brief Process poll() results for network play */
1144 static int network_afterpoll(void) {
1145 if(bfd_slot
!= -1) {
1146 if(fds
[bfd_slot
].revents
& (POLLOUT
| POLLERR
))
1153 /** @brief Table of speaker backends */
1154 static const struct speaker_backend backends
[] = {
1187 { -1, 0, 0, 0, 0, 0, 0, 0 } /* end of list */
1190 int main(int argc
, char **argv
) {
1191 int n
, fd
, stdin_slot
, poke
, timeout
;
1193 struct speaker_message sm
;
1196 if(!setlocale(LC_CTYPE
, "")) fatal(errno
, "error calling setlocale");
1197 while((n
= getopt_long(argc
, argv
, "hVc:dD", options
, 0)) >= 0) {
1200 case 'V': version();
1201 case 'c': configfile
= optarg
; break;
1202 case 'd': debugging
= 1; break;
1203 case 'D': debugging
= 0; break;
1204 default: fatal(0, "invalid option");
1207 if(getenv("DISORDER_DEBUG_SPEAKER")) debugging
= 1;
1208 /* If stderr is a TTY then log there, otherwise to syslog. */
1210 openlog(progname
, LOG_PID
, LOG_DAEMON
);
1211 log_default
= &log_syslog
;
1213 if(config_read()) fatal(0, "cannot read configuration");
1214 /* ignore SIGPIPE */
1215 signal(SIGPIPE
, SIG_IGN
);
1217 signal(SIGCHLD
, reap
);
1218 /* set nice value */
1219 xnice(config
->nice_speaker
);
1222 /* make sure we're not root, whatever the config says */
1223 if(getuid() == 0 || geteuid() == 0) fatal(0, "do not run as root");
1224 /* identify the backend used to play */
1225 for(n
= 0; backends
[n
].backend
!= -1; ++n
)
1226 if(backends
[n
].backend
== config
->speaker_backend
)
1228 if(backends
[n
].backend
== -1)
1229 fatal(0, "unsupported backend %d", config
->speaker_backend
);
1230 backend
= &backends
[n
];
1231 /* backend-specific initialization */
1233 while(getppid() != 1) {
1235 /* Always ready for commands from the main server. */
1236 stdin_slot
= addfd(0, POLLIN
);
1237 /* Try to read sample data for the currently playing track if there is
1239 if(playing
&& !playing
->eof
&& playing
->used
< playing
->size
) {
1240 playing
->slot
= addfd(playing
->fd
, POLLIN
);
1243 /* If forceplay is set then wait until it succeeds before waiting on the
1248 /* By default we will wait up to a second before thinking about current
1251 /* We'll break the poll as soon as the underlying sound device is ready for
1253 if(ready
&& !forceplay
)
1254 backend
->beforepoll();
1255 /* If any other tracks don't have a full buffer, try to read sample data
1257 for(t
= tracks
; t
; t
= t
->next
)
1259 if(!t
->eof
&& t
->used
< t
->size
) {
1260 t
->slot
= addfd(t
->fd
, POLLIN
| POLLHUP
);
1264 /* Wait for something interesting to happen */
1265 n
= poll(fds
, fdno
, timeout
);
1267 if(errno
== EINTR
) continue;
1268 fatal(errno
, "error calling poll");
1270 /* Play some sound before doing anything else */
1271 poke
= backend
->afterpoll();
1273 /* Some attempt to play must have failed */
1274 if(playing
&& !paused
)
1277 forceplay
= 0; /* just in case */
1279 /* Perhaps we have a command to process */
1280 if(fds
[stdin_slot
].revents
& POLLIN
) {
1281 n
= speaker_recv(0, &sm
, &fd
);
1285 D(("SM_PREPARE %s %d", sm
.id
, fd
));
1286 if(fd
== -1) fatal(0, "got SM_PREPARE but no file descriptor");
1287 t
= findtrack(sm
.id
, 1);
1291 D(("SM_PLAY %s %d", sm
.id
, fd
));
1292 if(playing
) fatal(0, "got SM_PLAY but already playing something");
1293 t
= findtrack(sm
.id
, 1);
1294 if(fd
!= -1) acquire(t
, fd
);
1314 D(("SM_CANCEL %s", sm
.id
));
1315 t
= removetrack(sm
.id
);
1318 sm
.type
= SM_FINISHED
;
1319 strcpy(sm
.id
, playing
->id
);
1320 speaker_send(1, &sm
, 0);
1325 error(0, "SM_CANCEL for unknown track %s", sm
.id
);
1330 if(config_read()) error(0, "cannot read configuration");
1331 info("reloaded configuration");
1334 error(0, "unknown message type %d", sm
.type
);
1337 /* Read in any buffered data */
1338 for(t
= tracks
; t
; t
= t
->next
)
1339 if(t
->slot
!= -1 && (fds
[t
->slot
].revents
& (POLLIN
| POLLHUP
)))
1341 /* We might be able to play now */
1342 if(ready
&& forceplay
&& playing
&& !paused
)
1344 /* Maybe we finished playing a track somewhere in the above */
1346 /* If we don't need the sound device for now then close it for the benefit
1347 * of anyone else who wants it. */
1348 if((!playing
|| paused
) && ready
)
1350 /* If we've not reported out state for a second do so now. */
1351 if(time(0) > last_report
)
1354 info("stopped (parent terminated)");
1363 indent-tabs-mode:nil