X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~mdw/git/disorder/blobdiff_plain/ae5b28b95df354760fc05130db1d816691cc58ad..3c68b773ae196c2691ed97d987be6221b1cf79b7:/server/speaker.c diff --git a/server/speaker.c b/server/speaker.c index 3aef0b0..f97bb69 100644 --- a/server/speaker.c +++ b/server/speaker.c @@ -133,16 +133,25 @@ static struct track { static time_t last_report; /* when we last reported */ static int paused; /* pause status */ -static ao_sample_format pcm_format; /* current format if aodev != 0 */ static size_t bpf; /* bytes per frame */ static struct pollfd fds[NFDS]; /* if we need more than that */ static int fdno; /* fd number */ static size_t bufsize; /* buffer size */ #if API_ALSA -static snd_pcm_t *pcm; /* current pcm handle */ +/** @brief The current PCM handle */ +static snd_pcm_t *pcm; static snd_pcm_uframes_t last_pcm_bufsize; /* last seen buffer size */ +static ao_sample_format pcm_format; /* current format if aodev != 0 */ #endif -static int ready; /* ready to send audio */ + +/** @brief Ready to send audio + * + * This is set when the destination is ready to receive audio. Generally + * this implies that the sound device is open. In the ALSA backend it + * does @b not necessarily imply that is has the right sample format. + */ +static int ready; + static int forceplay; /* frames to force play */ static int cmdfd = -1; /* child process input */ static int bfd = -1; /* broadcast FD */ @@ -172,6 +181,62 @@ static uint32_t rtp_id; /* RTP SSRC */ static int idled; /* set when idled */ static int audio_errors; /* audio error counter */ +/** @brief Structure of a backend */ +struct speaker_backend { + /** @brief Which backend this is + * + * @c -1 terminates the list. + */ + int backend; + + /** @brief Flags + * + * Possible values + * - @ref FIXED_FORMAT + */ + unsigned flags; +/** @brief Lock to configured sample format */ +#define FIXED_FORMAT 0x0001 + + /** @brief Initialization + * + * Called once at startup. This is responsible for one-time setup + * operations, for instance opening a network socket to transmit to. + * + * When writing to a native sound API this might @b not imply opening the + * native sound device - that might be done by @c activate below. + */ + void (*init)(void); + + /** @brief Activation + * @return 0 on success, non-0 on error + * + * Called to activate the output device. + * + * After this function succeeds, @ref ready should be non-0. As well as + * opening the audio device, this function is responsible for reconfiguring + * if it necessary to cope with different samples formats (for backends that + * don't demand a single fixed sample format for the lifetime of the server). + */ + int (*activate)(void); + + /** @brief Play sound + * @param frames Number of frames to play + * @return Number of frames actually played + */ + size_t (*play)(size_t frames); + + /** @brief Deactivation + * + * Called to deactivate the sound device. This is the inverse of + * @c activate above. + */ + void (*deactivate)(void); +}; + +/** @brief Selected backend */ +static const struct speaker_backend *backend; + static const struct option options[] = { { "help", no_argument, 0, 'h' }, { "version", no_argument, 0, 'V' }, @@ -309,16 +374,8 @@ static void soxargs(const char ***pp, char **qq, ao_sample_format *ao) { * to a sox invocation, which performs the required translation. */ static void enable_translation(struct track *t) { - switch(config->speaker_backend) { - case BACKEND_COMMAND: - case BACKEND_NETWORK: - /* These backends need a specific sample format */ - break; - case BACKEND_ALSA: - /* ALSA can cope */ - return; - } - if(!formats_equal(&t->format, &config->sample_format)) { + if((backend->flags & FIXED_FORMAT) + && !formats_equal(&t->format, &config->sample_format)) { char argbuf[1024], *q = argbuf; const char *av[18], **pp = av; int soxpipe[2]; @@ -353,7 +410,6 @@ static void enable_translation(struct track *t) { close(soxpipe[1]); t->fd = soxpipe[0]; t->format = config->sample_format; - ready = 0; } } @@ -414,21 +470,8 @@ static int fill(struct track *t) { /** @brief Close the sound device */ static void idle(void) { D(("idle")); -#if API_ALSA - if(config->speaker_backend == BACKEND_ALSA && pcm) { - int err; - - if((err = snd_pcm_nonblock(pcm, 0)) < 0) - fatal(0, "error calling snd_pcm_nonblock: %d", err); - D(("draining pcm")); - snd_pcm_drain(pcm); - D(("closing pcm")); - snd_pcm_close(pcm); - pcm = 0; - forceplay = 0; - D(("released audio device")); - } -#endif + if(backend->deactivate) + backend->deactivate(); idled = 1; ready = 0; } @@ -487,123 +530,7 @@ static int activate(void) { D((" - not got format for %s", playing->id)); return -1; } - switch(config->speaker_backend) { - case BACKEND_COMMAND: - case BACKEND_NETWORK: - if(!ready) { - pcm_format = config->sample_format; - bufsize = 3 * FRAMES; - bpf = bytes_per_frame(&config->sample_format); - D(("acquired audio device")); - ready = 1; - } - return 0; - case BACKEND_ALSA: -#if API_ALSA - /* If we need to change format then close the current device. */ - if(pcm && !formats_equal(&playing->format, &pcm_format)) - idle(); - if(!pcm) { - snd_pcm_hw_params_t *hwparams; - snd_pcm_sw_params_t *swparams; - snd_pcm_uframes_t pcm_bufsize; - int err; - int sample_format = 0; - unsigned rate; - - D(("snd_pcm_open")); - if((err = snd_pcm_open(&pcm, - config->device, - SND_PCM_STREAM_PLAYBACK, - SND_PCM_NONBLOCK))) { - error(0, "error from snd_pcm_open: %d", err); - goto error; - } - snd_pcm_hw_params_alloca(&hwparams); - D(("set up hw params")); - if((err = snd_pcm_hw_params_any(pcm, hwparams)) < 0) - fatal(0, "error from snd_pcm_hw_params_any: %d", err); - if((err = snd_pcm_hw_params_set_access(pcm, hwparams, - SND_PCM_ACCESS_RW_INTERLEAVED)) < 0) - fatal(0, "error from snd_pcm_hw_params_set_access: %d", err); - switch(playing->format.bits) { - case 8: - sample_format = SND_PCM_FORMAT_S8; - break; - case 16: - switch(playing->format.byte_format) { - case AO_FMT_NATIVE: sample_format = SND_PCM_FORMAT_S16; break; - case AO_FMT_LITTLE: sample_format = SND_PCM_FORMAT_S16_LE; break; - case AO_FMT_BIG: sample_format = SND_PCM_FORMAT_S16_BE; break; - error(0, "unrecognized byte format %d", playing->format.byte_format); - goto fatal; - } - break; - default: - error(0, "unsupported sample size %d", playing->format.bits); - goto fatal; - } - if((err = snd_pcm_hw_params_set_format(pcm, hwparams, - sample_format)) < 0) { - error(0, "error from snd_pcm_hw_params_set_format (%d): %d", - sample_format, err); - goto fatal; - } - rate = playing->format.rate; - if((err = snd_pcm_hw_params_set_rate_near(pcm, hwparams, &rate, 0)) < 0) { - error(0, "error from snd_pcm_hw_params_set_rate (%d): %d", - playing->format.rate, err); - goto fatal; - } - if(rate != (unsigned)playing->format.rate) - info("want rate %d, got %u", playing->format.rate, rate); - if((err = snd_pcm_hw_params_set_channels(pcm, hwparams, - playing->format.channels)) < 0) { - error(0, "error from snd_pcm_hw_params_set_channels (%d): %d", - playing->format.channels, err); - goto fatal; - } - bufsize = 3 * FRAMES; - pcm_bufsize = bufsize; - if((err = snd_pcm_hw_params_set_buffer_size_near(pcm, hwparams, - &pcm_bufsize)) < 0) - fatal(0, "error from snd_pcm_hw_params_set_buffer_size (%d): %d", - 3 * FRAMES, err); - if(pcm_bufsize != 3 * FRAMES && pcm_bufsize != last_pcm_bufsize) - info("asked for PCM buffer of %d frames, got %d", - 3 * FRAMES, (int)pcm_bufsize); - last_pcm_bufsize = pcm_bufsize; - if((err = snd_pcm_hw_params(pcm, hwparams)) < 0) - fatal(0, "error calling snd_pcm_hw_params: %d", err); - D(("set up sw params")); - snd_pcm_sw_params_alloca(&swparams); - if((err = snd_pcm_sw_params_current(pcm, swparams)) < 0) - fatal(0, "error calling snd_pcm_sw_params_current: %d", err); - if((err = snd_pcm_sw_params_set_avail_min(pcm, swparams, FRAMES)) < 0) - fatal(0, "error calling snd_pcm_sw_params_set_avail_min %d: %d", - FRAMES, err); - if((err = snd_pcm_sw_params(pcm, swparams)) < 0) - fatal(0, "error calling snd_pcm_sw_params: %d", err); - pcm_format = playing->format; - bpf = bytes_per_frame(&pcm_format); - D(("acquired audio device")); - log_params(hwparams, swparams); - ready = 1; - } - return 0; - fatal: - abandon(); - error: - /* We assume the error is temporary and that we'll retry in a bit. */ - if(pcm) { - snd_pcm_close(pcm); - pcm = 0; - } - return -1; -#endif - default: - assert(!"reached"); - } + return backend->activate(); } /* Check to see whether the current track has finished playing */ @@ -635,11 +562,10 @@ static void fork_cmd(void) { } static void play(size_t frames) { - size_t avail_bytes, write_bytes, written_frames; + size_t avail_frames, avail_bytes, written_frames; ssize_t written_bytes; - struct rtp_header header; - struct iovec vec[2]; + /* Make sure the output device is activated */ if(activate()) { if(playing) forceplay = frames; @@ -662,142 +588,20 @@ static void play(size_t frames) { forceplay = 0; /* Figure out how many frames there are available to write */ if(playing->start + playing->used > playing->size) + /* The ring buffer is currently wrapped, only play up to the wrap point */ avail_bytes = playing->size - playing->start; else + /* The ring buffer is not wrapped, can play the lot */ avail_bytes = playing->used; - - switch(config->speaker_backend) { -#if API_ALSA - case BACKEND_ALSA: { - snd_pcm_sframes_t pcm_written_frames; - size_t avail_frames; - int err; - - avail_frames = avail_bytes / bpf; - if(avail_frames > frames) - avail_frames = frames; - if(!avail_frames) - return; - pcm_written_frames = snd_pcm_writei(pcm, - playing->buffer + playing->start, - avail_frames); - D(("actually play %zu frames, wrote %d", - avail_frames, (int)pcm_written_frames)); - if(pcm_written_frames < 0) { - switch(pcm_written_frames) { - case -EPIPE: /* underrun */ - error(0, "snd_pcm_writei reports underrun"); - if((err = snd_pcm_prepare(pcm)) < 0) - fatal(0, "error calling snd_pcm_prepare: %d", err); - return; - case -EAGAIN: - return; - default: - fatal(0, "error calling snd_pcm_writei: %d", - (int)pcm_written_frames); - } - } - written_frames = pcm_written_frames; - written_bytes = written_frames * bpf; - break; - } -#endif - case BACKEND_COMMAND: - if(avail_bytes > frames * bpf) - avail_bytes = frames * bpf; - written_bytes = write(cmdfd, playing->buffer + playing->start, - avail_bytes); - D(("actually play %zu bytes, wrote %d", - avail_bytes, (int)written_bytes)); - if(written_bytes < 0) { - switch(errno) { - case EPIPE: - error(0, "hmm, command died; trying another"); - fork_cmd(); - return; - case EAGAIN: - return; - } - } - written_frames = written_bytes / bpf; /* good enough */ - break; - case BACKEND_NETWORK: - /* We transmit using RTP (RFC3550) and attempt to conform to the internet - * AVT profile (RFC3551). */ - - if(idled) { - /* There's been a gap. Fix up the RTP time accordingly. */ - struct timeval now; - uint64_t delta; - uint64_t target_rtp_time; - - /* Find the current time */ - xgettimeofday(&now, 0); - /* Find the number of microseconds elapsed since rtp_time=0 */ - delta = tvsub_us(now, rtp_time_0); - assert(delta <= UINT64_MAX / 88200); - target_rtp_time = (delta * playing->format.rate - * playing->format.channels) / 1000000; - /* Overflows at ~6 years uptime with 44100Hz stereo */ - if(target_rtp_time > rtp_time) - info("advancing rtp_time by %"PRIu64" samples", - target_rtp_time - rtp_time); - else if(target_rtp_time < rtp_time) - info("reversing rtp_time by %"PRIu64" samples", - rtp_time - target_rtp_time); - rtp_time = target_rtp_time; - } - header.vpxcc = 2 << 6; /* V=2, P=0, X=0, CC=0 */ - header.seq = htons(rtp_seq++); - header.timestamp = htonl((uint32_t)rtp_time); - header.ssrc = rtp_id; - header.mpt = (idled ? 0x80 : 0x00) | 10; - /* 10 = L16 = 16-bit x 2 x 44100KHz. We ought to deduce this value from - * the sample rate (in a library somewhere so that configuration.c can rule - * out invalid rates). - */ - idled = 0; - if(avail_bytes > NETWORK_BYTES - sizeof header) { - avail_bytes = NETWORK_BYTES - sizeof header; - /* Always send a whole number of frames */ - avail_bytes -= avail_bytes % bpf; - } - /* "The RTP clock rate used for generating the RTP timestamp is independent - * of the number of channels and the encoding; it equals the number of - * sampling periods per second. For N-channel encodings, each sampling - * period (say, 1/8000 of a second) generates N samples. (This terminology - * is standard, but somewhat confusing, as the total number of samples - * generated per second is then the sampling rate times the channel - * count.)" - */ - write_bytes = avail_bytes; - if(write_bytes) { - vec[0].iov_base = (void *)&header; - vec[0].iov_len = sizeof header; - vec[1].iov_base = playing->buffer + playing->start; - vec[1].iov_len = avail_bytes; - do { - written_bytes = writev(bfd, - vec, - 2); - } while(written_bytes < 0 && errno == EINTR); - if(written_bytes < 0) { - error(errno, "error transmitting audio data"); - ++audio_errors; - if(audio_errors == 10) - fatal(0, "too many audio errors"); - return; - } - } else - audio_errors /= 2; - written_bytes = avail_bytes; - written_frames = written_bytes / bpf; - /* Advance RTP's notion of the time */ - rtp_time += written_frames * playing->format.channels; - break; - default: - assert(!"reached"); - } + avail_frames = avail_bytes / bpf; + /* Only play up to the requested amount */ + if(avail_frames > frames) + avail_frames = frames; + if(!avail_frames) + return; + /* Play it, Sam */ + written_frames = backend->play(avail_frames); + written_bytes = written_frames * bpf; /* written_bytes and written_frames had better both be set and correct by * this point */ playing->start += written_bytes; @@ -843,10 +647,203 @@ static int addfd(int fd, int events) { return -1; } -int main(int argc, char **argv) { - int n, fd, stdin_slot, alsa_slots, cmdfd_slot, bfd_slot, poke, timeout; - struct track *t; - struct speaker_message sm; +#if API_ALSA +/** @brief ALSA backend initialization */ +static void alsa_init(void) { + info("selected ALSA backend"); +} + +/** @brief ALSA backend activation */ +static int alsa_activate(void) { + /* If we need to change format then close the current device. */ + if(pcm && !formats_equal(&playing->format, &pcm_format)) + idle(); + if(!pcm) { + snd_pcm_hw_params_t *hwparams; + snd_pcm_sw_params_t *swparams; + snd_pcm_uframes_t pcm_bufsize; + int err; + int sample_format = 0; + unsigned rate; + + D(("snd_pcm_open")); + if((err = snd_pcm_open(&pcm, + config->device, + SND_PCM_STREAM_PLAYBACK, + SND_PCM_NONBLOCK))) { + error(0, "error from snd_pcm_open: %d", err); + goto error; + } + snd_pcm_hw_params_alloca(&hwparams); + D(("set up hw params")); + if((err = snd_pcm_hw_params_any(pcm, hwparams)) < 0) + fatal(0, "error from snd_pcm_hw_params_any: %d", err); + if((err = snd_pcm_hw_params_set_access(pcm, hwparams, + SND_PCM_ACCESS_RW_INTERLEAVED)) < 0) + fatal(0, "error from snd_pcm_hw_params_set_access: %d", err); + switch(playing->format.bits) { + case 8: + sample_format = SND_PCM_FORMAT_S8; + break; + case 16: + switch(playing->format.byte_format) { + case AO_FMT_NATIVE: sample_format = SND_PCM_FORMAT_S16; break; + case AO_FMT_LITTLE: sample_format = SND_PCM_FORMAT_S16_LE; break; + case AO_FMT_BIG: sample_format = SND_PCM_FORMAT_S16_BE; break; + error(0, "unrecognized byte format %d", playing->format.byte_format); + goto fatal; + } + break; + default: + error(0, "unsupported sample size %d", playing->format.bits); + goto fatal; + } + if((err = snd_pcm_hw_params_set_format(pcm, hwparams, + sample_format)) < 0) { + error(0, "error from snd_pcm_hw_params_set_format (%d): %d", + sample_format, err); + goto fatal; + } + rate = playing->format.rate; + if((err = snd_pcm_hw_params_set_rate_near(pcm, hwparams, &rate, 0)) < 0) { + error(0, "error from snd_pcm_hw_params_set_rate (%d): %d", + playing->format.rate, err); + goto fatal; + } + if(rate != (unsigned)playing->format.rate) + info("want rate %d, got %u", playing->format.rate, rate); + if((err = snd_pcm_hw_params_set_channels(pcm, hwparams, + playing->format.channels)) < 0) { + error(0, "error from snd_pcm_hw_params_set_channels (%d): %d", + playing->format.channels, err); + goto fatal; + } + bufsize = 3 * FRAMES; + pcm_bufsize = bufsize; + if((err = snd_pcm_hw_params_set_buffer_size_near(pcm, hwparams, + &pcm_bufsize)) < 0) + fatal(0, "error from snd_pcm_hw_params_set_buffer_size (%d): %d", + 3 * FRAMES, err); + if(pcm_bufsize != 3 * FRAMES && pcm_bufsize != last_pcm_bufsize) + info("asked for PCM buffer of %d frames, got %d", + 3 * FRAMES, (int)pcm_bufsize); + last_pcm_bufsize = pcm_bufsize; + if((err = snd_pcm_hw_params(pcm, hwparams)) < 0) + fatal(0, "error calling snd_pcm_hw_params: %d", err); + D(("set up sw params")); + snd_pcm_sw_params_alloca(&swparams); + if((err = snd_pcm_sw_params_current(pcm, swparams)) < 0) + fatal(0, "error calling snd_pcm_sw_params_current: %d", err); + if((err = snd_pcm_sw_params_set_avail_min(pcm, swparams, FRAMES)) < 0) + fatal(0, "error calling snd_pcm_sw_params_set_avail_min %d: %d", + FRAMES, err); + if((err = snd_pcm_sw_params(pcm, swparams)) < 0) + fatal(0, "error calling snd_pcm_sw_params: %d", err); + pcm_format = playing->format; + bpf = bytes_per_frame(&pcm_format); + D(("acquired audio device")); + log_params(hwparams, swparams); + ready = 1; + } + return 0; +fatal: + abandon(); +error: + /* We assume the error is temporary and that we'll retry in a bit. */ + if(pcm) { + snd_pcm_close(pcm); + pcm = 0; + } + return -1; +} + +/** @brief Play via ALSA */ +static size_t alsa_play(size_t frames) { + snd_pcm_sframes_t pcm_written_frames; + int err; + + pcm_written_frames = snd_pcm_writei(pcm, + playing->buffer + playing->start, + frames); + D(("actually play %zu frames, wrote %d", + frames, (int)pcm_written_frames)); + if(pcm_written_frames < 0) { + switch(pcm_written_frames) { + case -EPIPE: /* underrun */ + error(0, "snd_pcm_writei reports underrun"); + if((err = snd_pcm_prepare(pcm)) < 0) + fatal(0, "error calling snd_pcm_prepare: %d", err); + return 0; + case -EAGAIN: + return 0; + default: + fatal(0, "error calling snd_pcm_writei: %d", + (int)pcm_written_frames); + } + } else + return pcm_written_frames; +} + +/** @brief ALSA deactivation */ +static void alsa_deactivate(void) { + if(pcm) { + int err; + + if((err = snd_pcm_nonblock(pcm, 0)) < 0) + fatal(0, "error calling snd_pcm_nonblock: %d", err); + D(("draining pcm")); + snd_pcm_drain(pcm); + D(("closing pcm")); + snd_pcm_close(pcm); + pcm = 0; + forceplay = 0; + D(("released audio device")); + } +} +#endif + +/** @brief Command backend initialization */ +static void command_init(void) { + info("selected command backend"); + fork_cmd(); +} + +/** @brief Play to a subprocess */ +static size_t command_play(size_t frames) { + size_t bytes = frames * bpf; + int written_bytes; + + written_bytes = write(cmdfd, playing->buffer + playing->start, bytes); + D(("actually play %zu bytes, wrote %d", + bytes, written_bytes)); + if(written_bytes < 0) { + switch(errno) { + case EPIPE: + error(0, "hmm, command died; trying another"); + fork_cmd(); + return 0; + case EAGAIN: + return 0; + default: + fatal(errno, "error writing to subprocess"); + } + } else + return written_bytes / bpf; +} + +/** @brief Command/network backend activation */ +static int generic_activate(void) { + if(!ready) { + bufsize = 3 * FRAMES; + bpf = bytes_per_frame(&config->sample_format); + D(("acquired audio device")); + ready = 1; + } + return 0; +} + +/** @brief Network backend initialization */ +static void network_init(void) { struct addrinfo *res, *sres; static const struct addrinfo pref = { 0, @@ -872,6 +869,195 @@ int main(int argc, char **argv) { int sndbuf, target_sndbuf = 131072; socklen_t len; char *sockname, *ssockname; + + res = get_address(&config->broadcast, &pref, &sockname); + if(!res) exit(-1); + if(config->broadcast_from.n) { + sres = get_address(&config->broadcast_from, &prefbind, &ssockname); + if(!sres) exit(-1); + } else + sres = 0; + if((bfd = socket(res->ai_family, + res->ai_socktype, + res->ai_protocol)) < 0) + fatal(errno, "error creating broadcast socket"); + if(setsockopt(bfd, SOL_SOCKET, SO_BROADCAST, &one, sizeof one) < 0) + fatal(errno, "error setting SO_BROADCAST on broadcast socket"); + len = sizeof sndbuf; + if(getsockopt(bfd, SOL_SOCKET, SO_SNDBUF, + &sndbuf, &len) < 0) + fatal(errno, "error getting SO_SNDBUF"); + if(target_sndbuf > sndbuf) { + if(setsockopt(bfd, SOL_SOCKET, SO_SNDBUF, + &target_sndbuf, sizeof target_sndbuf) < 0) + error(errno, "error setting SO_SNDBUF to %d", target_sndbuf); + else + info("changed socket send buffer size from %d to %d", + sndbuf, target_sndbuf); + } else + info("default socket send buffer is %d", + sndbuf); + /* We might well want to set additional broadcast- or multicast-related + * options here */ + if(sres && bind(bfd, sres->ai_addr, sres->ai_addrlen) < 0) + fatal(errno, "error binding broadcast socket to %s", ssockname); + if(connect(bfd, res->ai_addr, res->ai_addrlen) < 0) + fatal(errno, "error connecting broadcast socket to %s", sockname); + /* Select an SSRC */ + gcry_randomize(&rtp_id, sizeof rtp_id, GCRY_STRONG_RANDOM); + info("selected network backend, sending to %s", sockname); + if(config->sample_format.byte_format != AO_FMT_BIG) { + info("forcing big-endian sample format"); + config->sample_format.byte_format = AO_FMT_BIG; + } +} + +/** @brief Play over the network */ +static size_t network_play(size_t frames) { + struct rtp_header header; + struct iovec vec[2]; + size_t bytes = frames * bpf, written_frames; + int written_bytes; + /* We transmit using RTP (RFC3550) and attempt to conform to the internet + * AVT profile (RFC3551). */ + + if(idled) { + /* There may have been a gap. Fix up the RTP time accordingly. */ + struct timeval now; + uint64_t delta; + uint64_t target_rtp_time; + + /* Find the current time */ + xgettimeofday(&now, 0); + /* Find the number of microseconds elapsed since rtp_time=0 */ + delta = tvsub_us(now, rtp_time_0); + assert(delta <= UINT64_MAX / 88200); + target_rtp_time = (delta * playing->format.rate + * playing->format.channels) / 1000000; + /* Overflows at ~6 years uptime with 44100Hz stereo */ + + /* rtp_time is the number of samples we've played. NB that we play + * RTP_AHEAD_MS ahead of ourselves, so it may legitimately be ahead of + * the value we deduce from time comparison. + * + * Suppose we have 1s track started at t=0, and another track begins to + * play at t=2s. Suppose RTP_AHEAD_MS=1000 and 44100Hz stereo. In that + * case we'll send 1s of audio as fast as we can, giving rtp_time=88200. + * rtp_time stops at this point. + * + * At t=2s we'll have calculated target_rtp_time=176400. In this case we + * set rtp_time=176400 and the player can correctly conclude that it + * should leave 1s between the tracks. + * + * Suppose instead that the second track arrives at t=0.5s, and that + * we've managed to transmit the whole of the first track already. We'll + * have target_rtp_time=44100. + * + * The desired behaviour is to play the second track back to back with + * first. In this case therefore we do not modify rtp_time. + * + * Is it ever right to reduce rtp_time? No; for that would imply + * transmitting packets with overlapping timestamp ranges, which does not + * make sense. + */ + if(target_rtp_time > rtp_time) { + /* More time has elapsed than we've transmitted samples. That implies + * we've been 'sending' silence. */ + info("advancing rtp_time by %"PRIu64" samples", + target_rtp_time - rtp_time); + rtp_time = target_rtp_time; + } else if(target_rtp_time < rtp_time) { + const int64_t samples_ahead = ((uint64_t)RTP_AHEAD_MS + * config->sample_format.rate + * config->sample_format.channels + / 1000); + + if(target_rtp_time + samples_ahead < rtp_time) { + info("reversing rtp_time by %"PRIu64" samples", + rtp_time - target_rtp_time); + } + } + } + header.vpxcc = 2 << 6; /* V=2, P=0, X=0, CC=0 */ + header.seq = htons(rtp_seq++); + header.timestamp = htonl((uint32_t)rtp_time); + header.ssrc = rtp_id; + header.mpt = (idled ? 0x80 : 0x00) | 10; + /* 10 = L16 = 16-bit x 2 x 44100KHz. We ought to deduce this value from + * the sample rate (in a library somewhere so that configuration.c can rule + * out invalid rates). + */ + idled = 0; + if(bytes > NETWORK_BYTES - sizeof header) { + bytes = NETWORK_BYTES - sizeof header; + /* Always send a whole number of frames */ + bytes -= bytes % bpf; + } + /* "The RTP clock rate used for generating the RTP timestamp is independent + * of the number of channels and the encoding; it equals the number of + * sampling periods per second. For N-channel encodings, each sampling + * period (say, 1/8000 of a second) generates N samples. (This terminology + * is standard, but somewhat confusing, as the total number of samples + * generated per second is then the sampling rate times the channel + * count.)" + */ + vec[0].iov_base = (void *)&header; + vec[0].iov_len = sizeof header; + vec[1].iov_base = playing->buffer + playing->start; + vec[1].iov_len = bytes; + do { + written_bytes = writev(bfd, vec, 2); + } while(written_bytes < 0 && errno == EINTR); + if(written_bytes < 0) { + error(errno, "error transmitting audio data"); + ++audio_errors; + if(audio_errors == 10) + fatal(0, "too many audio errors"); + return 0; + } else + audio_errors /= 2; + written_bytes -= sizeof (struct rtp_header); + written_frames = written_bytes / bpf; + /* Advance RTP's notion of the time */ + rtp_time += written_frames * playing->format.channels; + return written_frames; +} + +/** @brief Table of speaker backends */ +static const struct speaker_backend backends[] = { +#if API_ALSA + { + BACKEND_ALSA, + 0, + alsa_init, + alsa_activate, + alsa_play, + alsa_deactivate + }, +#endif + { + BACKEND_COMMAND, + FIXED_FORMAT, + command_init, + generic_activate, + command_play, + 0 /* deactivate */ + }, + { + BACKEND_NETWORK, + FIXED_FORMAT, + network_init, + generic_activate, + network_play, + 0 /* deactivate */ + }, + { -1, 0, 0, 0, 0, 0 } +}; + +int main(int argc, char **argv) { + int n, fd, stdin_slot, alsa_slots, cmdfd_slot, bfd_slot, poke, timeout; + struct track *t; + struct speaker_message sm; #if API_ALSA int alsa_nslots = -1, err; #endif @@ -905,54 +1091,15 @@ int main(int argc, char **argv) { become_mortal(); /* make sure we're not root, whatever the config says */ if(getuid() == 0 || geteuid() == 0) fatal(0, "do not run as root"); - switch(config->speaker_backend) { - case BACKEND_ALSA: - info("selected ALSA backend"); - case BACKEND_COMMAND: - info("selected command backend"); - fork_cmd(); - break; - case BACKEND_NETWORK: - res = get_address(&config->broadcast, &pref, &sockname); - if(!res) return -1; - if(config->broadcast_from.n) { - sres = get_address(&config->broadcast_from, &prefbind, &ssockname); - if(!sres) return -1; - } else - sres = 0; - if((bfd = socket(res->ai_family, - res->ai_socktype, - res->ai_protocol)) < 0) - fatal(errno, "error creating broadcast socket"); - if(setsockopt(bfd, SOL_SOCKET, SO_BROADCAST, &one, sizeof one) < 0) - fatal(errno, "error setting SO_BROADCAST on broadcast socket"); - len = sizeof sndbuf; - if(getsockopt(bfd, SOL_SOCKET, SO_SNDBUF, - &sndbuf, &len) < 0) - fatal(errno, "error getting SO_SNDBUF"); - if(setsockopt(bfd, SOL_SOCKET, SO_SNDBUF, - &target_sndbuf, sizeof target_sndbuf) < 0) - error(errno, "error setting SO_SNDBUF to %d", target_sndbuf); - else - info("changed socket send buffer size from %d to %d", - sndbuf, target_sndbuf); - /* We might well want to set additional broadcast- or multicast-related - * options here */ - if(sres && bind(bfd, sres->ai_addr, sres->ai_addrlen) < 0) - fatal(errno, "error binding broadcast socket to %s", ssockname); - if(connect(bfd, res->ai_addr, res->ai_addrlen) < 0) - fatal(errno, "error connecting broadcast socket to %s", sockname); - /* Select an SSRC */ - gcry_randomize(&rtp_id, sizeof rtp_id, GCRY_STRONG_RANDOM); - info("selected network backend, sending to %s", sockname); - if(config->sample_format.byte_format != AO_FMT_BIG) { - info("forcing big-endian sample format"); - config->sample_format.byte_format = AO_FMT_BIG; - } - break; - default: - fatal(0, "unknown backend %d", config->speaker_backend); - } + /* identify the backend used to play */ + for(n = 0; backends[n].backend != -1; ++n) + if(backends[n].backend == config->speaker_backend) + break; + if(backends[n].backend == -1) + fatal(0, "unsupported backend %d", config->speaker_backend); + backend = &backends[n]; + /* backend-specific initialization */ + backend->init(); while(getppid() != 1) { fdno = 0; /* Always ready for commands from the main server. */