2 * This file is part of DisOrder
3 * Copyright (C) 2005, 2006, 2007 Richard Kettlewell
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307
20 /** @file server/speaker.c
21 * @brief Speaker processs
23 * This program is responsible for transmitting a single coherent audio stream
24 * to its destination (over the network, to some sound API, to some
25 * subprocess). It receives connections from decoders via file descriptor
26 * passing from the main server and plays them in the right order.
28 * For the <a href="http://www.alsa-project.org/">ALSA</a> API, 8- and 16- bit
29 * stereo and mono are supported, with any sample rate (within the limits that
30 * ALSA can deal with.)
32 * When communicating with a subprocess, <a
33 * href="http://sox.sourceforge.net/">sox</a> is invoked to convert the inbound
34 * data to a single consistent format. The same applies for network (RTP)
35 * play, though in that case currently only 44.1KHz 16-bit stereo is supported.
37 * The inbound data starts with a structure defining the data format. Note
38 * that this is NOT portable between different platforms or even necessarily
39 * between versions; the speaker is assumed to be built from the same source
40 * and run on the same host as the main server.
42 * This program deliberately does not use the garbage collector even though it
43 * might be convenient to do so. This is for two reasons. Firstly some sound
44 * APIs use thread threads and we do not want to have to deal with potential
45 * interactions between threading and garbage collection. Secondly this
46 * process needs to be able to respond quickly and this is not compatible with
47 * the collector hanging the program even relatively briefly.
63 #include <sys/select.h>
68 #include <sys/socket.h>
73 #include "configuration.h"
85 #include <alsa/asoundlib.h>
88 #ifdef WORDS_BIGENDIAN
89 # define MACHINE_AO_FMT AO_FMT_BIG
91 # define MACHINE_AO_FMT AO_FMT_LITTLE
94 /** @brief How many seconds of input to buffer
96 * While any given connection has this much audio buffered, no more reads will
97 * be issued for that connection. The decoder will have to wait.
99 #define BUFFER_SECONDS 5
101 #define FRAMES 4096 /* Frame batch size */
103 /** @brief Bytes to send per network packet
105 * Don't make this too big or arithmetic will start to overflow.
107 #define NETWORK_BYTES (1024+sizeof(struct rtp_header))
109 /** @brief Maximum RTP playahead (ms) */
110 #define RTP_AHEAD_MS 1000
112 /** @brief Maximum number of FDs to poll for */
115 /** @brief Track structure
117 * Known tracks are kept in a linked list. Usually there will be at most two
118 * of these but rearranging the queue can cause there to be more.
120 static struct track {
121 struct track *next; /* next track */
122 int fd; /* input FD */
123 char id[24]; /* ID */
124 size_t start, used; /* start + bytes used */
125 int eof; /* input is at EOF */
126 int got_format; /* got format yet? */
127 ao_sample_format format; /* sample format */
128 unsigned long long played; /* number of frames played */
129 char *buffer; /* sample buffer */
130 size_t size; /* sample buffer size */
131 int slot; /* poll array slot */
132 } *tracks, *playing; /* all tracks + playing track */
134 static time_t last_report; /* when we last reported */
135 static int paused; /* pause status */
136 static size_t bpf; /* bytes per frame */
137 static struct pollfd fds[NFDS]; /* if we need more than that */
138 static int fdno; /* fd number */
139 static size_t bufsize; /* buffer size */
141 /** @brief The current PCM handle */
142 static snd_pcm_t *pcm;
143 static snd_pcm_uframes_t last_pcm_bufsize; /* last seen buffer size */
144 static ao_sample_format pcm_format; /* current format if aodev != 0 */
147 /** @brief Ready to send audio
149 * This is set when the destination is ready to receive audio. Generally
150 * this implies that the sound device is open. In the ALSA backend it
151 * does @b not necessarily imply that is has the right sample format.
155 static int forceplay; /* frames to force play */
156 static int cmdfd = -1; /* child process input */
157 static int bfd = -1; /* broadcast FD */
159 /** @brief RTP timestamp
161 * This counts the number of samples played (NB not the number of frames
164 * The timestamp in the packet header is only 32 bits wide. With 44100Hz
165 * stereo, that only gives about half a day before wrapping, which is not
166 * particularly convenient for certain debugging purposes. Therefore the
167 * timestamp is maintained as a 64-bit integer, giving around six million years
168 * before wrapping, and truncated to 32 bits when transmitting.
170 static uint64_t rtp_time;
172 /** @brief RTP base timestamp
174 * This is the real time correspoding to an @ref rtp_time of 0. It is used
175 * to recalculate the timestamp after idle periods.
177 static struct timeval rtp_time_0;
179 static uint16_t rtp_seq; /* frame sequence number */
180 static uint32_t rtp_id; /* RTP SSRC */
181 static int idled; /* set when idled */
182 static int audio_errors; /* audio error counter */
184 /** @brief Structure of a backend */
185 struct speaker_backend {
186 /** @brief Which backend this is
188 * @c -1 terminates the list.
195 * - @ref FIXED_FORMAT
198 /** @brief Lock to configured sample format */
199 #define FIXED_FORMAT 0x0001
201 /** @brief Initialization
203 * Called once at startup. This is responsible for one-time setup
204 * operations, for instance opening a network socket to transmit to.
206 * When writing to a native sound API this might @b not imply opening the
207 * native sound device - that might be done by @c activate below.
211 /** @brief Activation
212 * @return 0 on success, non-0 on error
214 * Called to activate the output device.
216 * After this function succeeds, @ref ready should be non-0. As well as
217 * opening the audio device, this function is responsible for reconfiguring
218 * if it necessary to cope with different samples formats (for backends that
219 * don't demand a single fixed sample format for the lifetime of the server).
221 int (*activate)(void);
223 /** @brief Play sound
224 * @param frames Number of frames to play
225 * @return Number of frames actually played
227 size_t (*play)(size_t frames);
229 /** @brief Deactivation
231 * Called to deactivate the sound device. This is the inverse of
234 void (*deactivate)(void);
237 /** @brief Selected backend */
238 static const struct speaker_backend *backend;
240 static const struct option options[] = {
241 { "help", no_argument, 0, 'h' },
242 { "version", no_argument, 0, 'V' },
243 { "config", required_argument, 0, 'c' },
244 { "debug", no_argument, 0, 'd' },
245 { "no-debug", no_argument, 0, 'D' },
249 /* Display usage message and terminate. */
250 static void help(void) {
252 " disorder-speaker [OPTIONS]\n"
254 " --help, -h Display usage message\n"
255 " --version, -V Display version number\n"
256 " --config PATH, -c PATH Set configuration file\n"
257 " --debug, -d Turn on debugging\n"
259 "Speaker process for DisOrder. Not intended to be run\n"
265 /* Display version number and terminate. */
266 static void version(void) {
267 xprintf("disorder-speaker version %s\n", disorder_version_string);
272 /** @brief Return the number of bytes per frame in @p format */
273 static size_t bytes_per_frame(const ao_sample_format *format) {
274 return format->channels * format->bits / 8;
277 /** @brief Find track @p id, maybe creating it if not found */
278 static struct track *findtrack(const char *id, int create) {
281 D(("findtrack %s %d", id, create));
282 for(t = tracks; t && strcmp(id, t->id); t = t->next)
285 t = xmalloc(sizeof *t);
290 /* The initial input buffer will be the sample format. */
291 t->buffer = (void *)&t->format;
292 t->size = sizeof t->format;
297 /** @brief Remove track @p id (but do not destroy it) */
298 static struct track *removetrack(const char *id) {
299 struct track *t, **tt;
301 D(("removetrack %s", id));
302 for(tt = &tracks; (t = *tt) && strcmp(id, t->id); tt = &t->next)
309 /** @brief Destroy a track */
310 static void destroy(struct track *t) {
311 D(("destroy %s", t->id));
312 if(t->fd != -1) xclose(t->fd);
313 if(t->buffer != (void *)&t->format) free(t->buffer);
317 /** @brief Notice a new connection */
318 static void acquire(struct track *t, int fd) {
319 D(("acquire %s %d", t->id, fd));
326 /** @brief Return true if A and B denote identical libao formats, else false */
327 static int formats_equal(const ao_sample_format *a,
328 const ao_sample_format *b) {
329 return (a->bits == b->bits
330 && a->rate == b->rate
331 && a->channels == b->channels
332 && a->byte_format == b->byte_format);
335 /** @brief Compute arguments to sox */
336 static void soxargs(const char ***pp, char **qq, ao_sample_format *ao) {
341 *(*pp)++ = *qq; n = sprintf(*qq, "-r%d", ao->rate); *qq += n + 1;
342 *(*pp)++ = *qq; n = sprintf(*qq, "-c%d", ao->channels); *qq += n + 1;
343 /* sox 12.17.9 insists on -b etc; CVS sox insists on -<n> etc; both are
345 switch(config->sox_generation) {
348 && ao->byte_format != AO_FMT_NATIVE
349 && ao->byte_format != MACHINE_AO_FMT) {
353 case 8: *(*pp)++ = "-b"; break;
354 case 16: *(*pp)++ = "-w"; break;
355 case 32: *(*pp)++ = "-l"; break;
356 case 64: *(*pp)++ = "-d"; break;
357 default: fatal(0, "cannot handle sample size %d", (int)ao->bits);
361 switch(ao->byte_format) {
362 case AO_FMT_NATIVE: break;
363 case AO_FMT_BIG: *(*pp)++ = "-B"; break;
364 case AO_FMT_LITTLE: *(*pp)++ = "-L"; break;
366 *(*pp)++ = *qq; n = sprintf(*qq, "-%d", ao->bits/8); *qq += n + 1;
371 /** @brief Enable format translation
373 * If necessary, replaces a tracks inbound file descriptor with one connected
374 * to a sox invocation, which performs the required translation.
376 static void enable_translation(struct track *t) {
377 if((backend->flags & FIXED_FORMAT)
378 && !formats_equal(&t->format, &config->sample_format)) {
379 char argbuf[1024], *q = argbuf;
380 const char *av[18], **pp = av;
385 soxargs(&pp, &q, &t->format);
387 soxargs(&pp, &q, &config->sample_format);
391 for(pp = av; *pp; pp++)
392 D(("sox arg[%d] = %s", pp - av, *pp));
398 signal(SIGPIPE, SIG_DFL);
400 xdup2(soxpipe[1], 1);
401 fcntl(0, F_SETFL, fcntl(0, F_GETFL) & ~O_NONBLOCK);
405 execvp("sox", (char **)av);
408 D(("forking sox for format conversion (kid = %d)", soxkid));
412 t->format = config->sample_format;
416 /** @brief Read data into a sample buffer
417 * @param t Pointer to track
418 * @return 0 on success, -1 on EOF
420 * This is effectively the read callback on @c t->fd.
422 static int fill(struct track *t) {
426 D(("fill %s: eof=%d used=%zu size=%zu got_format=%d",
427 t->id, t->eof, t->used, t->size, t->got_format));
428 if(t->eof) return -1;
429 if(t->used < t->size) {
430 /* there is room left in the buffer */
431 where = (t->start + t->used) % t->size;
433 /* We are reading audio data, get as much as we can */
434 if(where >= t->start) left = t->size - where;
435 else left = t->start - where;
437 /* We are still waiting for the format, only get that */
438 left = sizeof (ao_sample_format) - t->used;
440 n = read(t->fd, t->buffer + where, left);
441 } while(n < 0 && errno == EINTR);
443 if(errno != EAGAIN) fatal(errno, "error reading sample stream");
447 D(("fill %s: eof detected", t->id));
452 if(!t->got_format && t->used >= sizeof (ao_sample_format)) {
453 assert(t->used == sizeof (ao_sample_format));
454 /* Check that our assumptions are met. */
455 if(t->format.bits & 7)
456 fatal(0, "bits per sample not a multiple of 8");
457 /* If the input format is unsuitable, arrange to translate it */
458 enable_translation(t);
459 /* Make a new buffer for audio data. */
460 t->size = bytes_per_frame(&t->format) * t->format.rate * BUFFER_SECONDS;
461 t->buffer = xmalloc(t->size);
464 D(("got format for %s", t->id));
470 /** @brief Close the sound device */
471 static void idle(void) {
473 if(backend->deactivate)
474 backend->deactivate();
479 /** @brief Abandon the current track */
480 static void abandon(void) {
481 struct speaker_message sm;
484 memset(&sm, 0, sizeof sm);
485 sm.type = SM_FINISHED;
486 strcpy(sm.id, playing->id);
487 speaker_send(1, &sm, 0);
488 removetrack(playing->id);
495 /** @brief Log ALSA parameters */
496 static void log_params(snd_pcm_hw_params_t *hwparams,
497 snd_pcm_sw_params_t *swparams) {
501 return; /* too verbose */
506 snd_pcm_sw_params_get_silence_size(swparams, &f);
507 info("sw silence_size=%lu", (unsigned long)f);
508 snd_pcm_sw_params_get_silence_threshold(swparams, &f);
509 info("sw silence_threshold=%lu", (unsigned long)f);
510 snd_pcm_sw_params_get_sleep_min(swparams, &u);
511 info("sw sleep_min=%lu", (unsigned long)u);
512 snd_pcm_sw_params_get_start_threshold(swparams, &f);
513 info("sw start_threshold=%lu", (unsigned long)f);
514 snd_pcm_sw_params_get_stop_threshold(swparams, &f);
515 info("sw stop_threshold=%lu", (unsigned long)f);
516 snd_pcm_sw_params_get_xfer_align(swparams, &f);
517 info("sw xfer_align=%lu", (unsigned long)f);
522 /** @brief Enable sound output
524 * Makes sure the sound device is open and has the right sample format. Return
525 * 0 on success and -1 on error.
527 static int activate(void) {
528 /* If we don't know the format yet we cannot start. */
529 if(!playing->got_format) {
530 D((" - not got format for %s", playing->id));
533 return backend->activate();
536 /* Check to see whether the current track has finished playing */
537 static void maybe_finished(void) {
540 && (!playing->got_format
541 || playing->used < bytes_per_frame(&playing->format)))
545 static void fork_cmd(void) {
548 if(cmdfd != -1) close(cmdfd);
552 signal(SIGPIPE, SIG_DFL);
556 execl("/bin/sh", "sh", "-c", config->speaker_command, (char *)0);
557 fatal(errno, "error execing /bin/sh");
561 D(("forked cmd %d, fd = %d", cmdpid, cmdfd));
564 static void play(size_t frames) {
565 size_t avail_frames, avail_bytes, written_frames;
566 ssize_t written_bytes;
568 /* Make sure the output device is activated */
573 forceplay = 0; /* Must have called abandon() */
576 D(("play: play %zu/%zu%s %dHz %db %dc", frames, playing->used / bpf,
577 playing->eof ? " EOF" : "",
578 playing->format.rate,
579 playing->format.bits,
580 playing->format.channels));
581 /* If we haven't got enough bytes yet wait until we have. Exception: when
583 if(playing->used < frames * bpf && !playing->eof) {
587 /* We have got enough data so don't force play again */
589 /* Figure out how many frames there are available to write */
590 if(playing->start + playing->used > playing->size)
591 /* The ring buffer is currently wrapped, only play up to the wrap point */
592 avail_bytes = playing->size - playing->start;
594 /* The ring buffer is not wrapped, can play the lot */
595 avail_bytes = playing->used;
596 avail_frames = avail_bytes / bpf;
597 /* Only play up to the requested amount */
598 if(avail_frames > frames)
599 avail_frames = frames;
603 written_frames = backend->play(avail_frames);
604 written_bytes = written_frames * bpf;
605 /* written_bytes and written_frames had better both be set and correct by
607 playing->start += written_bytes;
608 playing->used -= written_bytes;
609 playing->played += written_frames;
610 /* If the pointer is at the end of the buffer (or the buffer is completely
611 * empty) wrap it back to the start. */
612 if(!playing->used || playing->start == playing->size)
614 frames -= written_frames;
617 /* Notify the server what we're up to. */
618 static void report(void) {
619 struct speaker_message sm;
621 if(playing && playing->buffer != (void *)&playing->format) {
622 memset(&sm, 0, sizeof sm);
623 sm.type = paused ? SM_PAUSED : SM_PLAYING;
624 strcpy(sm.id, playing->id);
625 sm.data = playing->played / playing->format.rate;
626 speaker_send(1, &sm, 0);
631 static void reap(int __attribute__((unused)) sig) {
636 cmdpid = waitpid(-1, &st, WNOHANG);
638 signal(SIGCHLD, reap);
641 static int addfd(int fd, int events) {
644 fds[fdno].events = events;
651 /** @brief ALSA backend initialization */
652 static void alsa_init(void) {
653 info("selected ALSA backend");
656 /** @brief ALSA backend activation */
657 static int alsa_activate(void) {
658 /* If we need to change format then close the current device. */
659 if(pcm && !formats_equal(&playing->format, &pcm_format))
662 snd_pcm_hw_params_t *hwparams;
663 snd_pcm_sw_params_t *swparams;
664 snd_pcm_uframes_t pcm_bufsize;
666 int sample_format = 0;
670 if((err = snd_pcm_open(&pcm,
672 SND_PCM_STREAM_PLAYBACK,
673 SND_PCM_NONBLOCK))) {
674 error(0, "error from snd_pcm_open: %d", err);
677 snd_pcm_hw_params_alloca(&hwparams);
678 D(("set up hw params"));
679 if((err = snd_pcm_hw_params_any(pcm, hwparams)) < 0)
680 fatal(0, "error from snd_pcm_hw_params_any: %d", err);
681 if((err = snd_pcm_hw_params_set_access(pcm, hwparams,
682 SND_PCM_ACCESS_RW_INTERLEAVED)) < 0)
683 fatal(0, "error from snd_pcm_hw_params_set_access: %d", err);
684 switch(playing->format.bits) {
686 sample_format = SND_PCM_FORMAT_S8;
689 switch(playing->format.byte_format) {
690 case AO_FMT_NATIVE: sample_format = SND_PCM_FORMAT_S16; break;
691 case AO_FMT_LITTLE: sample_format = SND_PCM_FORMAT_S16_LE; break;
692 case AO_FMT_BIG: sample_format = SND_PCM_FORMAT_S16_BE; break;
693 error(0, "unrecognized byte format %d", playing->format.byte_format);
698 error(0, "unsupported sample size %d", playing->format.bits);
701 if((err = snd_pcm_hw_params_set_format(pcm, hwparams,
702 sample_format)) < 0) {
703 error(0, "error from snd_pcm_hw_params_set_format (%d): %d",
707 rate = playing->format.rate;
708 if((err = snd_pcm_hw_params_set_rate_near(pcm, hwparams, &rate, 0)) < 0) {
709 error(0, "error from snd_pcm_hw_params_set_rate (%d): %d",
710 playing->format.rate, err);
713 if(rate != (unsigned)playing->format.rate)
714 info("want rate %d, got %u", playing->format.rate, rate);
715 if((err = snd_pcm_hw_params_set_channels(pcm, hwparams,
716 playing->format.channels)) < 0) {
717 error(0, "error from snd_pcm_hw_params_set_channels (%d): %d",
718 playing->format.channels, err);
721 bufsize = 3 * FRAMES;
722 pcm_bufsize = bufsize;
723 if((err = snd_pcm_hw_params_set_buffer_size_near(pcm, hwparams,
725 fatal(0, "error from snd_pcm_hw_params_set_buffer_size (%d): %d",
727 if(pcm_bufsize != 3 * FRAMES && pcm_bufsize != last_pcm_bufsize)
728 info("asked for PCM buffer of %d frames, got %d",
729 3 * FRAMES, (int)pcm_bufsize);
730 last_pcm_bufsize = pcm_bufsize;
731 if((err = snd_pcm_hw_params(pcm, hwparams)) < 0)
732 fatal(0, "error calling snd_pcm_hw_params: %d", err);
733 D(("set up sw params"));
734 snd_pcm_sw_params_alloca(&swparams);
735 if((err = snd_pcm_sw_params_current(pcm, swparams)) < 0)
736 fatal(0, "error calling snd_pcm_sw_params_current: %d", err);
737 if((err = snd_pcm_sw_params_set_avail_min(pcm, swparams, FRAMES)) < 0)
738 fatal(0, "error calling snd_pcm_sw_params_set_avail_min %d: %d",
740 if((err = snd_pcm_sw_params(pcm, swparams)) < 0)
741 fatal(0, "error calling snd_pcm_sw_params: %d", err);
742 pcm_format = playing->format;
743 bpf = bytes_per_frame(&pcm_format);
744 D(("acquired audio device"));
745 log_params(hwparams, swparams);
752 /* We assume the error is temporary and that we'll retry in a bit. */
760 /** @brief Play via ALSA */
761 static size_t alsa_play(size_t frames) {
762 snd_pcm_sframes_t pcm_written_frames;
765 pcm_written_frames = snd_pcm_writei(pcm,
766 playing->buffer + playing->start,
768 D(("actually play %zu frames, wrote %d",
769 frames, (int)pcm_written_frames));
770 if(pcm_written_frames < 0) {
771 switch(pcm_written_frames) {
772 case -EPIPE: /* underrun */
773 error(0, "snd_pcm_writei reports underrun");
774 if((err = snd_pcm_prepare(pcm)) < 0)
775 fatal(0, "error calling snd_pcm_prepare: %d", err);
780 fatal(0, "error calling snd_pcm_writei: %d",
781 (int)pcm_written_frames);
784 return pcm_written_frames;
787 /** @brief ALSA deactivation */
788 static void alsa_deactivate(void) {
792 if((err = snd_pcm_nonblock(pcm, 0)) < 0)
793 fatal(0, "error calling snd_pcm_nonblock: %d", err);
800 D(("released audio device"));
805 /** @brief Command backend initialization */
806 static void command_init(void) {
807 info("selected command backend");
811 /** @brief Play to a subprocess */
812 static size_t command_play(size_t frames) {
813 size_t bytes = frames * bpf;
816 written_bytes = write(cmdfd, playing->buffer + playing->start, bytes);
817 D(("actually play %zu bytes, wrote %d",
818 bytes, written_bytes));
819 if(written_bytes < 0) {
822 error(0, "hmm, command died; trying another");
828 fatal(errno, "error writing to subprocess");
831 return written_bytes / bpf;
834 /** @brief Command/network backend activation */
835 static int generic_activate(void) {
837 bufsize = 3 * FRAMES;
838 bpf = bytes_per_frame(&config->sample_format);
839 D(("acquired audio device"));
845 /** @brief Network backend initialization */
846 static void network_init(void) {
847 struct addrinfo *res, *sres;
848 static const struct addrinfo pref = {
858 static const struct addrinfo prefbind = {
868 static const int one = 1;
869 int sndbuf, target_sndbuf = 131072;
871 char *sockname, *ssockname;
873 res = get_address(&config->broadcast, &pref, &sockname);
875 if(config->broadcast_from.n) {
876 sres = get_address(&config->broadcast_from, &prefbind, &ssockname);
880 if((bfd = socket(res->ai_family,
882 res->ai_protocol)) < 0)
883 fatal(errno, "error creating broadcast socket");
884 if(setsockopt(bfd, SOL_SOCKET, SO_BROADCAST, &one, sizeof one) < 0)
885 fatal(errno, "error setting SO_BROADCAST on broadcast socket");
887 if(getsockopt(bfd, SOL_SOCKET, SO_SNDBUF,
889 fatal(errno, "error getting SO_SNDBUF");
890 if(target_sndbuf > sndbuf) {
891 if(setsockopt(bfd, SOL_SOCKET, SO_SNDBUF,
892 &target_sndbuf, sizeof target_sndbuf) < 0)
893 error(errno, "error setting SO_SNDBUF to %d", target_sndbuf);
895 info("changed socket send buffer size from %d to %d",
896 sndbuf, target_sndbuf);
898 info("default socket send buffer is %d",
900 /* We might well want to set additional broadcast- or multicast-related
902 if(sres && bind(bfd, sres->ai_addr, sres->ai_addrlen) < 0)
903 fatal(errno, "error binding broadcast socket to %s", ssockname);
904 if(connect(bfd, res->ai_addr, res->ai_addrlen) < 0)
905 fatal(errno, "error connecting broadcast socket to %s", sockname);
907 gcry_randomize(&rtp_id, sizeof rtp_id, GCRY_STRONG_RANDOM);
908 info("selected network backend, sending to %s", sockname);
909 if(config->sample_format.byte_format != AO_FMT_BIG) {
910 info("forcing big-endian sample format");
911 config->sample_format.byte_format = AO_FMT_BIG;
915 /** @brief Play over the network */
916 static size_t network_play(size_t frames) {
917 struct rtp_header header;
919 size_t bytes = frames * bpf, written_frames;
921 /* We transmit using RTP (RFC3550) and attempt to conform to the internet
922 * AVT profile (RFC3551). */
925 /* There may have been a gap. Fix up the RTP time accordingly. */
928 uint64_t target_rtp_time;
930 /* Find the current time */
931 xgettimeofday(&now, 0);
932 /* Find the number of microseconds elapsed since rtp_time=0 */
933 delta = tvsub_us(now, rtp_time_0);
934 assert(delta <= UINT64_MAX / 88200);
935 target_rtp_time = (delta * playing->format.rate
936 * playing->format.channels) / 1000000;
937 /* Overflows at ~6 years uptime with 44100Hz stereo */
939 /* rtp_time is the number of samples we've played. NB that we play
940 * RTP_AHEAD_MS ahead of ourselves, so it may legitimately be ahead of
941 * the value we deduce from time comparison.
943 * Suppose we have 1s track started at t=0, and another track begins to
944 * play at t=2s. Suppose RTP_AHEAD_MS=1000 and 44100Hz stereo. In that
945 * case we'll send 1s of audio as fast as we can, giving rtp_time=88200.
946 * rtp_time stops at this point.
948 * At t=2s we'll have calculated target_rtp_time=176400. In this case we
949 * set rtp_time=176400 and the player can correctly conclude that it
950 * should leave 1s between the tracks.
952 * Suppose instead that the second track arrives at t=0.5s, and that
953 * we've managed to transmit the whole of the first track already. We'll
954 * have target_rtp_time=44100.
956 * The desired behaviour is to play the second track back to back with
957 * first. In this case therefore we do not modify rtp_time.
959 * Is it ever right to reduce rtp_time? No; for that would imply
960 * transmitting packets with overlapping timestamp ranges, which does not
963 if(target_rtp_time > rtp_time) {
964 /* More time has elapsed than we've transmitted samples. That implies
965 * we've been 'sending' silence. */
966 info("advancing rtp_time by %"PRIu64" samples",
967 target_rtp_time - rtp_time);
968 rtp_time = target_rtp_time;
969 } else if(target_rtp_time < rtp_time) {
970 const int64_t samples_ahead = ((uint64_t)RTP_AHEAD_MS
971 * config->sample_format.rate
972 * config->sample_format.channels
975 if(target_rtp_time + samples_ahead < rtp_time) {
976 info("reversing rtp_time by %"PRIu64" samples",
977 rtp_time - target_rtp_time);
981 header.vpxcc = 2 << 6; /* V=2, P=0, X=0, CC=0 */
982 header.seq = htons(rtp_seq++);
983 header.timestamp = htonl((uint32_t)rtp_time);
984 header.ssrc = rtp_id;
985 header.mpt = (idled ? 0x80 : 0x00) | 10;
986 /* 10 = L16 = 16-bit x 2 x 44100KHz. We ought to deduce this value from
987 * the sample rate (in a library somewhere so that configuration.c can rule
988 * out invalid rates).
991 if(bytes > NETWORK_BYTES - sizeof header) {
992 bytes = NETWORK_BYTES - sizeof header;
993 /* Always send a whole number of frames */
994 bytes -= bytes % bpf;
996 /* "The RTP clock rate used for generating the RTP timestamp is independent
997 * of the number of channels and the encoding; it equals the number of
998 * sampling periods per second. For N-channel encodings, each sampling
999 * period (say, 1/8000 of a second) generates N samples. (This terminology
1000 * is standard, but somewhat confusing, as the total number of samples
1001 * generated per second is then the sampling rate times the channel
1004 vec[0].iov_base = (void *)&header;
1005 vec[0].iov_len = sizeof header;
1006 vec[1].iov_base = playing->buffer + playing->start;
1007 vec[1].iov_len = bytes;
1009 written_bytes = writev(bfd, vec, 2);
1010 } while(written_bytes < 0 && errno == EINTR);
1011 if(written_bytes < 0) {
1012 error(errno, "error transmitting audio data");
1014 if(audio_errors == 10)
1015 fatal(0, "too many audio errors");
1019 written_bytes -= sizeof (struct rtp_header);
1020 written_frames = written_bytes / bpf;
1021 /* Advance RTP's notion of the time */
1022 rtp_time += written_frames * playing->format.channels;
1023 return written_frames;
1026 /** @brief Table of speaker backends */
1027 static const struct speaker_backend backends[] = {
1054 { -1, 0, 0, 0, 0, 0 }
1057 int main(int argc, char **argv) {
1058 int n, fd, stdin_slot, alsa_slots, cmdfd_slot, bfd_slot, poke, timeout;
1060 struct speaker_message sm;
1062 int alsa_nslots = -1, err;
1066 if(!setlocale(LC_CTYPE, "")) fatal(errno, "error calling setlocale");
1067 while((n = getopt_long(argc, argv, "hVc:dD", options, 0)) >= 0) {
1070 case 'V': version();
1071 case 'c': configfile = optarg; break;
1072 case 'd': debugging = 1; break;
1073 case 'D': debugging = 0; break;
1074 default: fatal(0, "invalid option");
1077 if(getenv("DISORDER_DEBUG_SPEAKER")) debugging = 1;
1078 /* If stderr is a TTY then log there, otherwise to syslog. */
1080 openlog(progname, LOG_PID, LOG_DAEMON);
1081 log_default = &log_syslog;
1083 if(config_read()) fatal(0, "cannot read configuration");
1084 /* ignore SIGPIPE */
1085 signal(SIGPIPE, SIG_IGN);
1087 signal(SIGCHLD, reap);
1088 /* set nice value */
1089 xnice(config->nice_speaker);
1092 /* make sure we're not root, whatever the config says */
1093 if(getuid() == 0 || geteuid() == 0) fatal(0, "do not run as root");
1094 /* identify the backend used to play */
1095 for(n = 0; backends[n].backend != -1; ++n)
1096 if(backends[n].backend == config->speaker_backend)
1098 if(backends[n].backend == -1)
1099 fatal(0, "unsupported backend %d", config->speaker_backend);
1100 backend = &backends[n];
1101 /* backend-specific initialization */
1103 while(getppid() != 1) {
1105 /* Always ready for commands from the main server. */
1106 stdin_slot = addfd(0, POLLIN);
1107 /* Try to read sample data for the currently playing track if there is
1109 if(playing && !playing->eof && playing->used < playing->size) {
1110 playing->slot = addfd(playing->fd, POLLIN);
1113 /* If forceplay is set then wait until it succeeds before waiting on the
1118 /* By default we will wait up to a second before thinking about current
1121 if(ready && !forceplay) {
1122 switch(config->speaker_backend) {
1123 case BACKEND_COMMAND:
1124 /* We send sample data to the subprocess as fast as it can accept it.
1125 * This isn't ideal as pause latency can be very high as a result. */
1127 cmdfd_slot = addfd(cmdfd, POLLOUT);
1129 case BACKEND_NETWORK: {
1132 uint64_t target_rtp_time;
1133 const int64_t samples_ahead = ((uint64_t)RTP_AHEAD_MS
1134 * config->sample_format.rate
1135 * config->sample_format.channels
1138 static unsigned logit;
1141 /* If we're starting then initialize the base time */
1143 xgettimeofday(&rtp_time_0, 0);
1144 /* We send audio data whenever we get RTP_AHEAD seconds or more
1146 xgettimeofday(&now, 0);
1147 target_us = tvsub_us(now, rtp_time_0);
1148 assert(target_us <= UINT64_MAX / 88200);
1149 target_rtp_time = (target_us * config->sample_format.rate
1150 * config->sample_format.channels)
1154 /* TODO remove logging guff */
1155 if(!(logit++ & 1023))
1156 info("rtp_time %llu target %llu difference %lld [%lld]",
1157 rtp_time, target_rtp_time,
1158 rtp_time - target_rtp_time,
1161 if((int64_t)(rtp_time - target_rtp_time) < samples_ahead)
1162 bfd_slot = addfd(bfd, POLLOUT);
1166 case BACKEND_ALSA: {
1167 /* We send sample data to ALSA as fast as it can accept it, relying on
1168 * the fact that it has a relatively small buffer to minimize pause
1175 alsa_nslots = snd_pcm_poll_descriptors(pcm, &fds[fdno], NFDS - fdno);
1176 if((alsa_nslots <= 0
1177 || !(fds[alsa_slots].events & POLLOUT))
1178 && snd_pcm_state(pcm) == SND_PCM_STATE_XRUN) {
1179 error(0, "underrun detected after call to snd_pcm_poll_descriptors()");
1180 if((err = snd_pcm_prepare(pcm)))
1181 fatal(0, "error calling snd_pcm_prepare: %d", err);
1184 } while(retry-- > 0);
1185 if(alsa_nslots >= 0)
1186 fdno += alsa_nslots;
1191 assert(!"unknown backend");
1194 /* If any other tracks don't have a full buffer, try to read sample data
1196 for(t = tracks; t; t = t->next)
1198 if(!t->eof && t->used < t->size) {
1199 t->slot = addfd(t->fd, POLLIN | POLLHUP);
1203 /* Wait for something interesting to happen */
1204 n = poll(fds, fdno, timeout);
1206 if(errno == EINTR) continue;
1207 fatal(errno, "error calling poll");
1209 /* Play some sound before doing anything else */
1211 switch(config->speaker_backend) {
1214 if(alsa_slots != -1) {
1215 unsigned short alsa_revents;
1217 if((err = snd_pcm_poll_descriptors_revents(pcm,
1220 &alsa_revents)) < 0)
1221 fatal(0, "error calling snd_pcm_poll_descriptors_revents: %d", err);
1222 if(alsa_revents & (POLLOUT | POLLERR))
1228 case BACKEND_COMMAND:
1229 if(cmdfd_slot != -1) {
1230 if(fds[cmdfd_slot].revents & (POLLOUT | POLLERR))
1235 case BACKEND_NETWORK:
1236 if(bfd_slot != -1) {
1237 if(fds[bfd_slot].revents & (POLLOUT | POLLERR))
1244 /* Some attempt to play must have failed */
1245 if(playing && !paused)
1248 forceplay = 0; /* just in case */
1250 /* Perhaps we have a command to process */
1251 if(fds[stdin_slot].revents & POLLIN) {
1252 n = speaker_recv(0, &sm, &fd);
1256 D(("SM_PREPARE %s %d", sm.id, fd));
1257 if(fd == -1) fatal(0, "got SM_PREPARE but no file descriptor");
1258 t = findtrack(sm.id, 1);
1262 D(("SM_PLAY %s %d", sm.id, fd));
1263 if(playing) fatal(0, "got SM_PLAY but already playing something");
1264 t = findtrack(sm.id, 1);
1265 if(fd != -1) acquire(t, fd);
1285 D(("SM_CANCEL %s", sm.id));
1286 t = removetrack(sm.id);
1289 sm.type = SM_FINISHED;
1290 strcpy(sm.id, playing->id);
1291 speaker_send(1, &sm, 0);
1296 error(0, "SM_CANCEL for unknown track %s", sm.id);
1301 if(config_read()) error(0, "cannot read configuration");
1302 info("reloaded configuration");
1305 error(0, "unknown message type %d", sm.type);
1308 /* Read in any buffered data */
1309 for(t = tracks; t; t = t->next)
1310 if(t->slot != -1 && (fds[t->slot].revents & (POLLIN | POLLHUP)))
1312 /* We might be able to play now */
1313 if(ready && forceplay && playing && !paused)
1315 /* Maybe we finished playing a track somewhere in the above */
1317 /* If we don't need the sound device for now then close it for the benefit
1318 * of anyone else who wants it. */
1319 if((!playing || paused) && ready)
1321 /* If we've not reported out state for a second do so now. */
1322 if(time(0) > last_report)
1325 info("stopped (parent terminated)");
1334 indent-tabs-mode:nil