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 process
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>
76 #include "configuration.h"
81 #include "speaker-protocol.h"
85 /** @brief Linked list of all prepared tracks */
88 /** @brief Playing track, or NULL */
89 struct track *playing;
91 /** @brief Number of bytes pre frame */
94 /** @brief Array of file descriptors for poll() */
95 struct pollfd fds[NFDS];
97 /** @brief Next free slot in @ref fds */
100 static time_t last_report; /* when we last reported */
101 static int paused; /* pause status */
103 /** @brief The current device state */
104 enum device_states device_state;
106 /** @brief The current device sample format
108 * Only meaningful if @ref device_state = @ref device_open or perhaps @ref
109 * device_error. For @ref FIXED_FORMAT backends, this should always match @c
110 * config->sample_format.
112 ao_sample_format device_format;
114 /** @brief Set when idled
116 * This is set when the sound device is deliberately closed by idle().
120 /** @brief Selected backend */
121 static const struct speaker_backend *backend;
123 static const struct option options[] = {
124 { "help", no_argument, 0, 'h' },
125 { "version", no_argument, 0, 'V' },
126 { "config", required_argument, 0, 'c' },
127 { "debug", no_argument, 0, 'd' },
128 { "no-debug", no_argument, 0, 'D' },
132 /* Display usage message and terminate. */
133 static void help(void) {
135 " disorder-speaker [OPTIONS]\n"
137 " --help, -h Display usage message\n"
138 " --version, -V Display version number\n"
139 " --config PATH, -c PATH Set configuration file\n"
140 " --debug, -d Turn on debugging\n"
142 "Speaker process for DisOrder. Not intended to be run\n"
148 /* Display version number and terminate. */
149 static void version(void) {
150 xprintf("disorder-speaker version %s\n", disorder_version_string);
155 /** @brief Return the number of bytes per frame in @p format */
156 static size_t bytes_per_frame(const ao_sample_format *format) {
157 return format->channels * format->bits / 8;
160 /** @brief Find track @p id, maybe creating it if not found */
161 static struct track *findtrack(const char *id, int create) {
164 D(("findtrack %s %d", id, create));
165 for(t = tracks; t && strcmp(id, t->id); t = t->next)
168 t = xmalloc(sizeof *t);
173 /* The initial input buffer will be the sample format. */
174 t->buffer = (void *)&t->format;
175 t->size = sizeof t->format;
180 /** @brief Remove track @p id (but do not destroy it) */
181 static struct track *removetrack(const char *id) {
182 struct track *t, **tt;
184 D(("removetrack %s", id));
185 for(tt = &tracks; (t = *tt) && strcmp(id, t->id); tt = &t->next)
192 /** @brief Destroy a track */
193 static void destroy(struct track *t) {
194 D(("destroy %s", t->id));
195 if(t->fd != -1) xclose(t->fd);
196 if(t->buffer != (void *)&t->format) free(t->buffer);
200 /** @brief Notice a new connection */
201 static void acquire(struct track *t, int fd) {
202 D(("acquire %s %d", t->id, fd));
209 /** @brief Return true if A and B denote identical libao formats, else false */
210 int formats_equal(const ao_sample_format *a,
211 const ao_sample_format *b) {
212 return (a->bits == b->bits
213 && a->rate == b->rate
214 && a->channels == b->channels
215 && a->byte_format == b->byte_format);
218 /** @brief Compute arguments to sox */
219 static void soxargs(const char ***pp, char **qq, ao_sample_format *ao) {
224 *(*pp)++ = *qq; n = sprintf(*qq, "-r%d", ao->rate); *qq += n + 1;
225 *(*pp)++ = *qq; n = sprintf(*qq, "-c%d", ao->channels); *qq += n + 1;
226 /* sox 12.17.9 insists on -b etc; CVS sox insists on -<n> etc; both are
228 switch(config->sox_generation) {
231 && ao->byte_format != AO_FMT_NATIVE
232 && ao->byte_format != MACHINE_AO_FMT) {
236 case 8: *(*pp)++ = "-b"; break;
237 case 16: *(*pp)++ = "-w"; break;
238 case 32: *(*pp)++ = "-l"; break;
239 case 64: *(*pp)++ = "-d"; break;
240 default: fatal(0, "cannot handle sample size %d", (int)ao->bits);
244 switch(ao->byte_format) {
245 case AO_FMT_NATIVE: break;
246 case AO_FMT_BIG: *(*pp)++ = "-B"; break;
247 case AO_FMT_LITTLE: *(*pp)++ = "-L"; break;
249 *(*pp)++ = *qq; n = sprintf(*qq, "-%d", ao->bits/8); *qq += n + 1;
254 /** @brief Enable format translation
256 * If necessary, replaces a tracks inbound file descriptor with one connected
257 * to a sox invocation, which performs the required translation.
259 static void enable_translation(struct track *t) {
260 if((backend->flags & FIXED_FORMAT)
261 && !formats_equal(&t->format, &config->sample_format)) {
262 char argbuf[1024], *q = argbuf;
263 const char *av[18], **pp = av;
268 soxargs(&pp, &q, &t->format);
270 soxargs(&pp, &q, &config->sample_format);
274 for(pp = av; *pp; pp++)
275 D(("sox arg[%d] = %s", pp - av, *pp));
281 signal(SIGPIPE, SIG_DFL);
283 xdup2(soxpipe[1], 1);
284 fcntl(0, F_SETFL, fcntl(0, F_GETFL) & ~O_NONBLOCK);
288 execvp("sox", (char **)av);
291 D(("forking sox for format conversion (kid = %d)", soxkid));
295 t->format = config->sample_format;
299 /** @brief Read data into a sample buffer
300 * @param t Pointer to track
301 * @return 0 on success, -1 on EOF
303 * This is effectively the read callback on @c t->fd. It is called from the
304 * main loop whenever the track's file descriptor is readable, assuming the
305 * buffer has not reached the maximum allowed occupancy.
307 static int fill(struct track *t) {
311 D(("fill %s: eof=%d used=%zu size=%zu got_format=%d",
312 t->id, t->eof, t->used, t->size, t->got_format));
313 if(t->eof) return -1;
314 if(t->used < t->size) {
315 /* there is room left in the buffer */
316 where = (t->start + t->used) % t->size;
318 /* We are reading audio data, get as much as we can */
319 if(where >= t->start) left = t->size - where;
320 else left = t->start - where;
322 /* We are still waiting for the format, only get that */
323 left = sizeof (ao_sample_format) - t->used;
325 n = read(t->fd, t->buffer + where, left);
326 } while(n < 0 && errno == EINTR);
328 if(errno != EAGAIN) fatal(errno, "error reading sample stream");
332 D(("fill %s: eof detected", t->id));
337 if(!t->got_format && t->used >= sizeof (ao_sample_format)) {
338 assert(t->used == sizeof (ao_sample_format));
339 /* Check that our assumptions are met. */
340 if(t->format.bits & 7)
341 fatal(0, "bits per sample not a multiple of 8");
342 /* If the input format is unsuitable, arrange to translate it */
343 enable_translation(t);
344 /* Make a new buffer for audio data. */
345 t->size = bytes_per_frame(&t->format) * t->format.rate * BUFFER_SECONDS;
346 t->buffer = xmalloc(t->size);
349 D(("got format for %s", t->id));
355 /** @brief Close the sound device
357 * This is called to deactivate the output device when pausing, and also by the
358 * ALSA backend when changing encoding (in which case the sound device will be
359 * immediately reactivated).
361 static void idle(void) {
363 if(backend->deactivate)
364 backend->deactivate();
366 device_state = device_closed;
370 /** @brief Abandon the current track */
372 struct speaker_message sm;
375 memset(&sm, 0, sizeof sm);
376 sm.type = SM_FINISHED;
377 strcpy(sm.id, playing->id);
378 speaker_send(1, &sm, 0);
379 removetrack(playing->id);
384 /** @brief Enable sound output
386 * Makes sure the sound device is open and has the right sample format. Return
387 * 0 on success and -1 on error.
389 static void activate(void) {
390 /* If we don't know the format yet we cannot start. */
391 if(!playing->got_format) {
392 D((" - not got format for %s", playing->id));
395 if(backend->flags & FIXED_FORMAT)
396 device_format = config->sample_format;
397 if(backend->activate) {
400 assert(backend->flags & FIXED_FORMAT);
401 /* ...otherwise device_format not set */
402 device_state = device_open;
404 if(device_state == device_open)
405 device_bpf = bytes_per_frame(&device_format);
408 /** @brief Check whether the current track has finished
410 * The current track is determined to have finished either if the input stream
411 * eded before the format could be determined (i.e. it is malformed) or the
412 * input is at end of file and there is less than a frame left unplayed. (So
413 * it copes with decoders that crash mid-frame.)
415 static void maybe_finished(void) {
418 && (!playing->got_format
419 || playing->used < bytes_per_frame(&playing->format)))
423 /** @brief Play up to @p frames frames of audio
425 * It is always safe to call this function.
426 * - If @ref playing is 0 then it will just return
427 * - If @ref paused is non-0 then it will just return
428 * - If @ref device_state != @ref device_open then it will call activate() and
429 * return if it it fails.
430 * - If there is not enough audio to play then it play what is available.
432 * If there are not enough frames to play then whatever is available is played
433 * instead. It is up to mainloop() to ensure that play() is not called when
434 * unreasonably only an small amounts of data is available to play.
436 static void play(size_t frames) {
437 size_t avail_frames, avail_bytes, written_frames;
438 ssize_t written_bytes;
440 /* Make sure there's a track to play and it is not pasued */
441 if(!playing || paused)
443 /* Make sure the output device is open and has the right sample format */
444 if(device_state != device_open
445 || !formats_equal(&device_format, &playing->format)) {
447 if(device_state != device_open)
450 D(("play: play %zu/%zu%s %dHz %db %dc", frames, playing->used / device_bpf,
451 playing->eof ? " EOF" : "",
452 playing->format.rate,
453 playing->format.bits,
454 playing->format.channels));
455 /* Figure out how many frames there are available to write */
456 if(playing->start + playing->used > playing->size)
457 /* The ring buffer is currently wrapped, only play up to the wrap point */
458 avail_bytes = playing->size - playing->start;
460 /* The ring buffer is not wrapped, can play the lot */
461 avail_bytes = playing->used;
462 avail_frames = avail_bytes / device_bpf;
463 /* Only play up to the requested amount */
464 if(avail_frames > frames)
465 avail_frames = frames;
469 written_frames = backend->play(avail_frames);
470 written_bytes = written_frames * device_bpf;
471 /* written_bytes and written_frames had better both be set and correct by
473 playing->start += written_bytes;
474 playing->used -= written_bytes;
475 playing->played += written_frames;
476 /* If the pointer is at the end of the buffer (or the buffer is completely
477 * empty) wrap it back to the start. */
478 if(!playing->used || playing->start == playing->size)
480 frames -= written_frames;
484 /* Notify the server what we're up to. */
485 static void report(void) {
486 struct speaker_message sm;
488 if(playing && playing->buffer != (void *)&playing->format) {
489 memset(&sm, 0, sizeof sm);
490 sm.type = paused ? SM_PAUSED : SM_PLAYING;
491 strcpy(sm.id, playing->id);
492 sm.data = playing->played / playing->format.rate;
493 speaker_send(1, &sm, 0);
498 static void reap(int __attribute__((unused)) sig) {
503 cmdpid = waitpid(-1, &st, WNOHANG);
505 signal(SIGCHLD, reap);
508 int addfd(int fd, int events) {
511 fds[fdno].events = events;
517 /** @brief Table of speaker backends */
518 static const struct speaker_backend *backends[] = {
527 /** @brief Return nonzero if we want to play some audio
529 * We want to play audio if there is a current track; and it is not paused; and
530 * there are at least @ref FRAMES frames of audio to play, or we are in sight
531 * of the end of the current track.
533 static int playable(void) {
536 && (playing->used >= FRAMES || playing->eof);
539 /** @brief Main event loop */
540 static void mainloop(void) {
542 struct speaker_message sm;
543 int n, fd, stdin_slot, timeout;
545 while(getppid() != 1) {
547 /* By default we will wait up to a second before thinking about current
550 /* Always ready for commands from the main server. */
551 stdin_slot = addfd(0, POLLIN);
552 /* Try to read sample data for the currently playing track if there is
554 if(playing && !playing->eof && playing->used < playing->size)
555 playing->slot = addfd(playing->fd, POLLIN);
559 /* We want to play some audio. If the device is closed then we attempt
561 if(device_state == device_closed)
563 /* If the device is (now) open then we will wait up until it is ready for
564 * more. If something went wrong then we should have device_error
565 * instead, but the post-poll code will cope even if it's
567 if(device_state == device_open)
568 backend->beforepoll();
570 /* If any other tracks don't have a full buffer, try to read sample data
571 * from them. We do this last of all, so that if we run out of slots,
572 * nothing important can't be monitored. */
573 for(t = tracks; t; t = t->next)
575 if(!t->eof && t->used < t->size) {
576 t->slot = addfd(t->fd, POLLIN | POLLHUP);
580 /* Wait for something interesting to happen */
581 n = poll(fds, fdno, timeout);
583 if(errno == EINTR) continue;
584 fatal(errno, "error calling poll");
586 /* Play some sound before doing anything else */
588 /* We want to play some audio */
589 if(device_state == device_open) {
593 /* We must be in _closed or _error, and it should be the latter, but we
596 * We most likely timed out, so now is a good time to retry. play()
597 * knows to re-activate the device if necessary.
602 /* Perhaps we have a command to process */
603 if(fds[stdin_slot].revents & POLLIN) {
604 /* There might (in theory) be several commands queued up, but in general
605 * this won't be the case, so we don't bother looping around to pick them
607 n = speaker_recv(0, &sm, &fd);
611 D(("SM_PREPARE %s %d", sm.id, fd));
612 if(fd == -1) fatal(0, "got SM_PREPARE but no file descriptor");
613 t = findtrack(sm.id, 1);
617 D(("SM_PLAY %s %d", sm.id, fd));
618 if(playing) fatal(0, "got SM_PLAY but already playing something");
619 t = findtrack(sm.id, 1);
620 if(fd != -1) acquire(t, fd);
622 /* We attempt to play straight away rather than going round the loop.
623 * play() is clever enough to perform any activation that is
637 /* As for SM_PLAY we attempt to play straight away. */
644 D(("SM_CANCEL %s", sm.id));
645 t = removetrack(sm.id);
648 sm.type = SM_FINISHED;
649 strcpy(sm.id, playing->id);
650 speaker_send(1, &sm, 0);
655 error(0, "SM_CANCEL for unknown track %s", sm.id);
660 if(config_read()) error(0, "cannot read configuration");
661 info("reloaded configuration");
664 error(0, "unknown message type %d", sm.type);
667 /* Read in any buffered data */
668 for(t = tracks; t; t = t->next)
669 if(t->slot != -1 && (fds[t->slot].revents & (POLLIN | POLLHUP)))
671 /* Maybe we finished playing a track somewhere in the above */
673 /* If we don't need the sound device for now then close it for the benefit
674 * of anyone else who wants it. */
675 if((!playing || paused) && device_state == device_open)
677 /* If we've not reported out state for a second do so now. */
678 if(time(0) > last_report)
683 int main(int argc, char **argv) {
687 if(!setlocale(LC_CTYPE, "")) fatal(errno, "error calling setlocale");
688 while((n = getopt_long(argc, argv, "hVc:dD", options, 0)) >= 0) {
692 case 'c': configfile = optarg; break;
693 case 'd': debugging = 1; break;
694 case 'D': debugging = 0; break;
695 default: fatal(0, "invalid option");
698 if(getenv("DISORDER_DEBUG_SPEAKER")) debugging = 1;
699 /* If stderr is a TTY then log there, otherwise to syslog. */
701 openlog(progname, LOG_PID, LOG_DAEMON);
702 log_default = &log_syslog;
704 if(config_read()) fatal(0, "cannot read configuration");
706 signal(SIGPIPE, SIG_IGN);
708 signal(SIGCHLD, reap);
710 xnice(config->nice_speaker);
713 /* make sure we're not root, whatever the config says */
714 if(getuid() == 0 || geteuid() == 0) fatal(0, "do not run as root");
715 /* identify the backend used to play */
716 for(n = 0; backends[n]; ++n)
717 if(backends[n]->backend == config->speaker_backend)
720 fatal(0, "unsupported backend %d", config->speaker_backend);
721 backend = backends[n];
722 /* backend-specific initialization */
725 info("stopped (parent terminated)");