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, write_bytes, written_frames;
566 ssize_t written_bytes;
567 struct rtp_header header;
570 /* Make sure the output device is activated */
575 forceplay = 0; /* Must have called abandon() */
578 D(("play: play %zu/%zu%s %dHz %db %dc", frames, playing->used / bpf,
579 playing->eof ? " EOF" : "",
580 playing->format.rate,
581 playing->format.bits,
582 playing->format.channels));
583 /* If we haven't got enough bytes yet wait until we have. Exception: when
585 if(playing->used < frames * bpf && !playing->eof) {
589 /* We have got enough data so don't force play again */
591 /* Figure out how many frames there are available to write */
592 if(playing->start + playing->used > playing->size)
593 /* The ring buffer is currently wrapped, only play up to the wrap point */
594 avail_bytes = playing->size - playing->start;
596 /* The ring buffer is not wrapped, can play the lot */
597 avail_bytes = playing->used;
598 avail_frames = avail_bytes / bpf;
599 /* Only play up to the requested amount */
600 if(avail_frames > frames)
601 avail_frames = frames;
605 switch(config->speaker_backend) {
608 written_frames = backend->play(avail_frames);
612 case BACKEND_COMMAND:
613 if(avail_bytes > frames * bpf)
614 avail_bytes = frames * bpf;
615 written_bytes = write(cmdfd, playing->buffer + playing->start,
617 D(("actually play %zu bytes, wrote %d",
618 avail_bytes, (int)written_bytes));
619 if(written_bytes < 0) {
622 error(0, "hmm, command died; trying another");
629 written_frames = written_bytes / bpf; /* good enough */
631 case BACKEND_NETWORK:
632 /* We transmit using RTP (RFC3550) and attempt to conform to the internet
633 * AVT profile (RFC3551). */
636 /* There may have been a gap. Fix up the RTP time accordingly. */
639 uint64_t target_rtp_time;
641 /* Find the current time */
642 xgettimeofday(&now, 0);
643 /* Find the number of microseconds elapsed since rtp_time=0 */
644 delta = tvsub_us(now, rtp_time_0);
645 assert(delta <= UINT64_MAX / 88200);
646 target_rtp_time = (delta * playing->format.rate
647 * playing->format.channels) / 1000000;
648 /* Overflows at ~6 years uptime with 44100Hz stereo */
650 /* rtp_time is the number of samples we've played. NB that we play
651 * RTP_AHEAD_MS ahead of ourselves, so it may legitimately be ahead of
652 * the value we deduce from time comparison.
654 * Suppose we have 1s track started at t=0, and another track begins to
655 * play at t=2s. Suppose RTP_AHEAD_MS=1000 and 44100Hz stereo. In that
656 * case we'll send 1s of audio as fast as we can, giving rtp_time=88200.
657 * rtp_time stops at this point.
659 * At t=2s we'll have calculated target_rtp_time=176400. In this case we
660 * set rtp_time=176400 and the player can correctly conclude that it
661 * should leave 1s between the tracks.
663 * Suppose instead that the second track arrives at t=0.5s, and that
664 * we've managed to transmit the whole of the first track already. We'll
665 * have target_rtp_time=44100.
667 * The desired behaviour is to play the second track back to back with
668 * first. In this case therefore we do not modify rtp_time.
670 * Is it ever right to reduce rtp_time? No; for that would imply
671 * transmitting packets with overlapping timestamp ranges, which does not
674 if(target_rtp_time > rtp_time) {
675 /* More time has elapsed than we've transmitted samples. That implies
676 * we've been 'sending' silence. */
677 info("advancing rtp_time by %"PRIu64" samples",
678 target_rtp_time - rtp_time);
679 rtp_time = target_rtp_time;
680 } else if(target_rtp_time < rtp_time) {
681 const int64_t samples_ahead = ((uint64_t)RTP_AHEAD_MS
682 * config->sample_format.rate
683 * config->sample_format.channels
686 if(target_rtp_time + samples_ahead < rtp_time) {
687 info("reversing rtp_time by %"PRIu64" samples",
688 rtp_time - target_rtp_time);
692 header.vpxcc = 2 << 6; /* V=2, P=0, X=0, CC=0 */
693 header.seq = htons(rtp_seq++);
694 header.timestamp = htonl((uint32_t)rtp_time);
695 header.ssrc = rtp_id;
696 header.mpt = (idled ? 0x80 : 0x00) | 10;
697 /* 10 = L16 = 16-bit x 2 x 44100KHz. We ought to deduce this value from
698 * the sample rate (in a library somewhere so that configuration.c can rule
699 * out invalid rates).
702 if(avail_bytes > NETWORK_BYTES - sizeof header) {
703 avail_bytes = NETWORK_BYTES - sizeof header;
704 /* Always send a whole number of frames */
705 avail_bytes -= avail_bytes % bpf;
707 /* "The RTP clock rate used for generating the RTP timestamp is independent
708 * of the number of channels and the encoding; it equals the number of
709 * sampling periods per second. For N-channel encodings, each sampling
710 * period (say, 1/8000 of a second) generates N samples. (This terminology
711 * is standard, but somewhat confusing, as the total number of samples
712 * generated per second is then the sampling rate times the channel
715 write_bytes = avail_bytes;
717 vec[0].iov_base = (void *)&header;
718 vec[0].iov_len = sizeof header;
719 vec[1].iov_base = playing->buffer + playing->start;
720 vec[1].iov_len = avail_bytes;
722 written_bytes = writev(bfd,
725 } while(written_bytes < 0 && errno == EINTR);
726 if(written_bytes < 0) {
727 error(errno, "error transmitting audio data");
729 if(audio_errors == 10)
730 fatal(0, "too many audio errors");
735 written_bytes = avail_bytes;
736 written_frames = written_bytes / bpf;
737 /* Advance RTP's notion of the time */
738 rtp_time += written_frames * playing->format.channels;
743 written_bytes = written_frames * bpf;
744 /* written_bytes and written_frames had better both be set and correct by
746 playing->start += written_bytes;
747 playing->used -= written_bytes;
748 playing->played += written_frames;
749 /* If the pointer is at the end of the buffer (or the buffer is completely
750 * empty) wrap it back to the start. */
751 if(!playing->used || playing->start == playing->size)
753 frames -= written_frames;
756 /* Notify the server what we're up to. */
757 static void report(void) {
758 struct speaker_message sm;
760 if(playing && playing->buffer != (void *)&playing->format) {
761 memset(&sm, 0, sizeof sm);
762 sm.type = paused ? SM_PAUSED : SM_PLAYING;
763 strcpy(sm.id, playing->id);
764 sm.data = playing->played / playing->format.rate;
765 speaker_send(1, &sm, 0);
770 static void reap(int __attribute__((unused)) sig) {
775 cmdpid = waitpid(-1, &st, WNOHANG);
777 signal(SIGCHLD, reap);
780 static int addfd(int fd, int events) {
783 fds[fdno].events = events;
790 /** @brief ALSA backend initialization */
791 static void alsa_init(void) {
792 info("selected ALSA backend");
795 /** @brief ALSA backend activation */
796 static int alsa_activate(void) {
797 /* If we need to change format then close the current device. */
798 if(pcm && !formats_equal(&playing->format, &pcm_format))
801 snd_pcm_hw_params_t *hwparams;
802 snd_pcm_sw_params_t *swparams;
803 snd_pcm_uframes_t pcm_bufsize;
805 int sample_format = 0;
809 if((err = snd_pcm_open(&pcm,
811 SND_PCM_STREAM_PLAYBACK,
812 SND_PCM_NONBLOCK))) {
813 error(0, "error from snd_pcm_open: %d", err);
816 snd_pcm_hw_params_alloca(&hwparams);
817 D(("set up hw params"));
818 if((err = snd_pcm_hw_params_any(pcm, hwparams)) < 0)
819 fatal(0, "error from snd_pcm_hw_params_any: %d", err);
820 if((err = snd_pcm_hw_params_set_access(pcm, hwparams,
821 SND_PCM_ACCESS_RW_INTERLEAVED)) < 0)
822 fatal(0, "error from snd_pcm_hw_params_set_access: %d", err);
823 switch(playing->format.bits) {
825 sample_format = SND_PCM_FORMAT_S8;
828 switch(playing->format.byte_format) {
829 case AO_FMT_NATIVE: sample_format = SND_PCM_FORMAT_S16; break;
830 case AO_FMT_LITTLE: sample_format = SND_PCM_FORMAT_S16_LE; break;
831 case AO_FMT_BIG: sample_format = SND_PCM_FORMAT_S16_BE; break;
832 error(0, "unrecognized byte format %d", playing->format.byte_format);
837 error(0, "unsupported sample size %d", playing->format.bits);
840 if((err = snd_pcm_hw_params_set_format(pcm, hwparams,
841 sample_format)) < 0) {
842 error(0, "error from snd_pcm_hw_params_set_format (%d): %d",
846 rate = playing->format.rate;
847 if((err = snd_pcm_hw_params_set_rate_near(pcm, hwparams, &rate, 0)) < 0) {
848 error(0, "error from snd_pcm_hw_params_set_rate (%d): %d",
849 playing->format.rate, err);
852 if(rate != (unsigned)playing->format.rate)
853 info("want rate %d, got %u", playing->format.rate, rate);
854 if((err = snd_pcm_hw_params_set_channels(pcm, hwparams,
855 playing->format.channels)) < 0) {
856 error(0, "error from snd_pcm_hw_params_set_channels (%d): %d",
857 playing->format.channels, err);
860 bufsize = 3 * FRAMES;
861 pcm_bufsize = bufsize;
862 if((err = snd_pcm_hw_params_set_buffer_size_near(pcm, hwparams,
864 fatal(0, "error from snd_pcm_hw_params_set_buffer_size (%d): %d",
866 if(pcm_bufsize != 3 * FRAMES && pcm_bufsize != last_pcm_bufsize)
867 info("asked for PCM buffer of %d frames, got %d",
868 3 * FRAMES, (int)pcm_bufsize);
869 last_pcm_bufsize = pcm_bufsize;
870 if((err = snd_pcm_hw_params(pcm, hwparams)) < 0)
871 fatal(0, "error calling snd_pcm_hw_params: %d", err);
872 D(("set up sw params"));
873 snd_pcm_sw_params_alloca(&swparams);
874 if((err = snd_pcm_sw_params_current(pcm, swparams)) < 0)
875 fatal(0, "error calling snd_pcm_sw_params_current: %d", err);
876 if((err = snd_pcm_sw_params_set_avail_min(pcm, swparams, FRAMES)) < 0)
877 fatal(0, "error calling snd_pcm_sw_params_set_avail_min %d: %d",
879 if((err = snd_pcm_sw_params(pcm, swparams)) < 0)
880 fatal(0, "error calling snd_pcm_sw_params: %d", err);
881 pcm_format = playing->format;
882 bpf = bytes_per_frame(&pcm_format);
883 D(("acquired audio device"));
884 log_params(hwparams, swparams);
891 /* We assume the error is temporary and that we'll retry in a bit. */
899 /** @brief Play via ALSA */
900 static size_t alsa_play(size_t frames) {
901 snd_pcm_sframes_t pcm_written_frames;
904 pcm_written_frames = snd_pcm_writei(pcm,
905 playing->buffer + playing->start,
907 D(("actually play %zu frames, wrote %d",
908 frames, (int)pcm_written_frames));
909 if(pcm_written_frames < 0) {
910 switch(pcm_written_frames) {
911 case -EPIPE: /* underrun */
912 error(0, "snd_pcm_writei reports underrun");
913 if((err = snd_pcm_prepare(pcm)) < 0)
914 fatal(0, "error calling snd_pcm_prepare: %d", err);
919 fatal(0, "error calling snd_pcm_writei: %d",
920 (int)pcm_written_frames);
923 return pcm_written_frames;
926 /** @brief ALSA deactivation */
927 static void alsa_deactivate(void) {
931 if((err = snd_pcm_nonblock(pcm, 0)) < 0)
932 fatal(0, "error calling snd_pcm_nonblock: %d", err);
939 D(("released audio device"));
944 /** @brief Command backend initialization */
945 static void command_init(void) {
946 info("selected command backend");
950 /** @brief Play to a subprocess */
951 static size_t command_play(size_t frames) {
955 /** @brief Command/network backend activation */
956 static int generic_activate(void) {
958 bufsize = 3 * FRAMES;
959 bpf = bytes_per_frame(&config->sample_format);
960 D(("acquired audio device"));
966 /** @brief Network backend initialization */
967 static void network_init(void) {
968 struct addrinfo *res, *sres;
969 static const struct addrinfo pref = {
979 static const struct addrinfo prefbind = {
989 static const int one = 1;
990 int sndbuf, target_sndbuf = 131072;
992 char *sockname, *ssockname;
994 res = get_address(&config->broadcast, &pref, &sockname);
996 if(config->broadcast_from.n) {
997 sres = get_address(&config->broadcast_from, &prefbind, &ssockname);
1001 if((bfd = socket(res->ai_family,
1003 res->ai_protocol)) < 0)
1004 fatal(errno, "error creating broadcast socket");
1005 if(setsockopt(bfd, SOL_SOCKET, SO_BROADCAST, &one, sizeof one) < 0)
1006 fatal(errno, "error setting SO_BROADCAST on broadcast socket");
1007 len = sizeof sndbuf;
1008 if(getsockopt(bfd, SOL_SOCKET, SO_SNDBUF,
1010 fatal(errno, "error getting SO_SNDBUF");
1011 if(target_sndbuf > sndbuf) {
1012 if(setsockopt(bfd, SOL_SOCKET, SO_SNDBUF,
1013 &target_sndbuf, sizeof target_sndbuf) < 0)
1014 error(errno, "error setting SO_SNDBUF to %d", target_sndbuf);
1016 info("changed socket send buffer size from %d to %d",
1017 sndbuf, target_sndbuf);
1019 info("default socket send buffer is %d",
1021 /* We might well want to set additional broadcast- or multicast-related
1023 if(sres && bind(bfd, sres->ai_addr, sres->ai_addrlen) < 0)
1024 fatal(errno, "error binding broadcast socket to %s", ssockname);
1025 if(connect(bfd, res->ai_addr, res->ai_addrlen) < 0)
1026 fatal(errno, "error connecting broadcast socket to %s", sockname);
1027 /* Select an SSRC */
1028 gcry_randomize(&rtp_id, sizeof rtp_id, GCRY_STRONG_RANDOM);
1029 info("selected network backend, sending to %s", sockname);
1030 if(config->sample_format.byte_format != AO_FMT_BIG) {
1031 info("forcing big-endian sample format");
1032 config->sample_format.byte_format = AO_FMT_BIG;
1036 /** @brief Play over the network */
1037 static size_t network_play(size_t frames) {
1041 /** @brief Table of speaker backends */
1042 static const struct speaker_backend backends[] = {
1069 { -1, 0, 0, 0, 0, 0 }
1072 int main(int argc, char **argv) {
1073 int n, fd, stdin_slot, alsa_slots, cmdfd_slot, bfd_slot, poke, timeout;
1075 struct speaker_message sm;
1077 int alsa_nslots = -1, err;
1081 if(!setlocale(LC_CTYPE, "")) fatal(errno, "error calling setlocale");
1082 while((n = getopt_long(argc, argv, "hVc:dD", options, 0)) >= 0) {
1085 case 'V': version();
1086 case 'c': configfile = optarg; break;
1087 case 'd': debugging = 1; break;
1088 case 'D': debugging = 0; break;
1089 default: fatal(0, "invalid option");
1092 if(getenv("DISORDER_DEBUG_SPEAKER")) debugging = 1;
1093 /* If stderr is a TTY then log there, otherwise to syslog. */
1095 openlog(progname, LOG_PID, LOG_DAEMON);
1096 log_default = &log_syslog;
1098 if(config_read()) fatal(0, "cannot read configuration");
1099 /* ignore SIGPIPE */
1100 signal(SIGPIPE, SIG_IGN);
1102 signal(SIGCHLD, reap);
1103 /* set nice value */
1104 xnice(config->nice_speaker);
1107 /* make sure we're not root, whatever the config says */
1108 if(getuid() == 0 || geteuid() == 0) fatal(0, "do not run as root");
1109 /* identify the backend used to play */
1110 for(n = 0; backends[n].backend != -1; ++n)
1111 if(backends[n].backend == config->speaker_backend)
1113 if(backends[n].backend == -1)
1114 fatal(0, "unsupported backend %d", config->speaker_backend);
1115 backend = &backends[n];
1116 /* backend-specific initialization */
1118 while(getppid() != 1) {
1120 /* Always ready for commands from the main server. */
1121 stdin_slot = addfd(0, POLLIN);
1122 /* Try to read sample data for the currently playing track if there is
1124 if(playing && !playing->eof && playing->used < playing->size) {
1125 playing->slot = addfd(playing->fd, POLLIN);
1128 /* If forceplay is set then wait until it succeeds before waiting on the
1133 /* By default we will wait up to a second before thinking about current
1136 if(ready && !forceplay) {
1137 switch(config->speaker_backend) {
1138 case BACKEND_COMMAND:
1139 /* We send sample data to the subprocess as fast as it can accept it.
1140 * This isn't ideal as pause latency can be very high as a result. */
1142 cmdfd_slot = addfd(cmdfd, POLLOUT);
1144 case BACKEND_NETWORK: {
1147 uint64_t target_rtp_time;
1148 const int64_t samples_ahead = ((uint64_t)RTP_AHEAD_MS
1149 * config->sample_format.rate
1150 * config->sample_format.channels
1153 static unsigned logit;
1156 /* If we're starting then initialize the base time */
1158 xgettimeofday(&rtp_time_0, 0);
1159 /* We send audio data whenever we get RTP_AHEAD seconds or more
1161 xgettimeofday(&now, 0);
1162 target_us = tvsub_us(now, rtp_time_0);
1163 assert(target_us <= UINT64_MAX / 88200);
1164 target_rtp_time = (target_us * config->sample_format.rate
1165 * config->sample_format.channels)
1169 /* TODO remove logging guff */
1170 if(!(logit++ & 1023))
1171 info("rtp_time %llu target %llu difference %lld [%lld]",
1172 rtp_time, target_rtp_time,
1173 rtp_time - target_rtp_time,
1176 if((int64_t)(rtp_time - target_rtp_time) < samples_ahead)
1177 bfd_slot = addfd(bfd, POLLOUT);
1181 case BACKEND_ALSA: {
1182 /* We send sample data to ALSA as fast as it can accept it, relying on
1183 * the fact that it has a relatively small buffer to minimize pause
1190 alsa_nslots = snd_pcm_poll_descriptors(pcm, &fds[fdno], NFDS - fdno);
1191 if((alsa_nslots <= 0
1192 || !(fds[alsa_slots].events & POLLOUT))
1193 && snd_pcm_state(pcm) == SND_PCM_STATE_XRUN) {
1194 error(0, "underrun detected after call to snd_pcm_poll_descriptors()");
1195 if((err = snd_pcm_prepare(pcm)))
1196 fatal(0, "error calling snd_pcm_prepare: %d", err);
1199 } while(retry-- > 0);
1200 if(alsa_nslots >= 0)
1201 fdno += alsa_nslots;
1206 assert(!"unknown backend");
1209 /* If any other tracks don't have a full buffer, try to read sample data
1211 for(t = tracks; t; t = t->next)
1213 if(!t->eof && t->used < t->size) {
1214 t->slot = addfd(t->fd, POLLIN | POLLHUP);
1218 /* Wait for something interesting to happen */
1219 n = poll(fds, fdno, timeout);
1221 if(errno == EINTR) continue;
1222 fatal(errno, "error calling poll");
1224 /* Play some sound before doing anything else */
1226 switch(config->speaker_backend) {
1229 if(alsa_slots != -1) {
1230 unsigned short alsa_revents;
1232 if((err = snd_pcm_poll_descriptors_revents(pcm,
1235 &alsa_revents)) < 0)
1236 fatal(0, "error calling snd_pcm_poll_descriptors_revents: %d", err);
1237 if(alsa_revents & (POLLOUT | POLLERR))
1243 case BACKEND_COMMAND:
1244 if(cmdfd_slot != -1) {
1245 if(fds[cmdfd_slot].revents & (POLLOUT | POLLERR))
1250 case BACKEND_NETWORK:
1251 if(bfd_slot != -1) {
1252 if(fds[bfd_slot].revents & (POLLOUT | POLLERR))
1259 /* Some attempt to play must have failed */
1260 if(playing && !paused)
1263 forceplay = 0; /* just in case */
1265 /* Perhaps we have a command to process */
1266 if(fds[stdin_slot].revents & POLLIN) {
1267 n = speaker_recv(0, &sm, &fd);
1271 D(("SM_PREPARE %s %d", sm.id, fd));
1272 if(fd == -1) fatal(0, "got SM_PREPARE but no file descriptor");
1273 t = findtrack(sm.id, 1);
1277 D(("SM_PLAY %s %d", sm.id, fd));
1278 if(playing) fatal(0, "got SM_PLAY but already playing something");
1279 t = findtrack(sm.id, 1);
1280 if(fd != -1) acquire(t, fd);
1300 D(("SM_CANCEL %s", sm.id));
1301 t = removetrack(sm.id);
1304 sm.type = SM_FINISHED;
1305 strcpy(sm.id, playing->id);
1306 speaker_send(1, &sm, 0);
1311 error(0, "SM_CANCEL for unknown track %s", sm.id);
1316 if(config_read()) error(0, "cannot read configuration");
1317 info("reloaded configuration");
1320 error(0, "unknown message type %d", sm.type);
1323 /* Read in any buffered data */
1324 for(t = tracks; t; t = t->next)
1325 if(t->slot != -1 && (fds[t->slot].revents & (POLLIN | POLLHUP)))
1327 /* We might be able to play now */
1328 if(ready && forceplay && playing && !paused)
1330 /* Maybe we finished playing a track somewhere in the above */
1332 /* If we don't need the sound device for now then close it for the benefit
1333 * of anyone else who wants it. */
1334 if((!playing || paused) && ready)
1336 /* If we've not reported out state for a second do so now. */
1337 if(time(0) > last_report)
1340 info("stopped (parent terminated)");
1349 indent-tabs-mode:nil