X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~mdw/git/disorder/blobdiff_plain/6ba5f1eaee022b36d76f245c7cbcf78b72e9cf35..0cf685f6a9ce58259b542dc4102261205a53a6f4:/server/speaker.c diff --git a/server/speaker.c b/server/speaker.c index 994c4b4..a8524b4 100644 --- a/server/speaker.c +++ b/server/speaker.c @@ -1,43 +1,57 @@ /* * This file is part of DisOrder - * Copyright (C) 2005, 2006, 2007 Richard Kettlewell + * Copyright (C) 2005-2010 Richard Kettlewell + * Portions (C) 2007 Mark Wooding * - * This program is free software; you can redistribute it and/or modify + * This program is free software: you can redistribute it and/or modify * it under the terms of the GNU General Public License as published by - * the Free Software Foundation; either version 2 of the License, or + * the Free Software Foundation, either version 3 of the License, or * (at your option) any later version. - * - * This program is distributed in the hope that it will be useful, but - * WITHOUT ANY WARRANTY; without even the implied warranty of - * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU - * General Public License for more details. - * + * + * This program is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the + * GNU General Public License for more details. + * * You should have received a copy of the GNU General Public License - * along with this program; if not, write to the Free Software - * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 - * USA + * along with this program. If not, see . */ /** @file server/speaker.c - * @brief Speaker processs + * @brief Speaker process * * This program is responsible for transmitting a single coherent audio stream * to its destination (over the network, to some sound API, to some - * subprocess). It receives connections from decoders via file descriptor - * passing from the main server and plays them in the right order. + * subprocess). It receives connections from decoders (or rather from the + * process that is about to become disorder-normalize) and plays them in the + * right order. + * + * @b Model. mainloop() implements a select loop awaiting commands from the + * main server, new connections to the speaker socket, and audio data on those + * connections. Each connection starts with a queue ID (with a 32-bit + * native-endian length word), allowing it to be referred to in commands from + * the server. + * + * Data read on connections is buffered, up to a limit (currently 1Mbyte per + * track). No attempt is made here to limit the number of tracks, it is + * assumed that the main server won't start outrageously many decoders. + * + * Audio is supplied from this buffer to the uaudio play callback. Playback is + * enabled when a track is to be played and disabled when the its last bytes + * have been returned by the callback; pause and resume is implemented the + * obvious way. If the callback finds itself required to play when there is no + * playing track it returns dead air. * - * @b Encodings. For the ALSA API, - * 8- and 16- bit stereo and mono are supported, with any sample rate (within - * the limits that ALSA can deal with.) + * To implement gapless playback, the server is notified that a track has + * finished slightly early. @ref SM_PLAY is therefore allowed to arrive while + * the previous track is still playing provided an early @ref SM_FINISHED has + * been sent for it. * - * When communicating with a subprocess, sox is invoked to convert the inbound - * data to a single consistent format. The same applies for network (RTP) - * play, though in that case currently only 44.1KHz 16-bit stereo is supported. + * @b Encodings. The encodings supported depend entirely on the uaudio backend + * chosen. See @ref uaudio.h, etc. * - * The inbound data starts with a structure defining the data format. Note - * that this is NOT portable between different platforms or even necessarily - * between versions; the speaker is assumed to be built from the same source - * and run on the same host as the main server. + * Inbound data is expected to match @c config->sample_format. In normal use + * this is arranged by the @c disorder-normalize program (see @ref + * server/normalize.c). * * @b Garbage @b Collection. This program deliberately does not use the * garbage collector even though it might be convenient to do so. This is for @@ -54,202 +68,160 @@ * 2-byte samples. */ -#include -#include "types.h" +#include "common.h" #include -#include -#include #include #include #include #include -#include -#include -#include #include #include #include #include #include -#include -#include +#include +#include +#include +#include #include -#include #include "configuration.h" #include "syscalls.h" #include "log.h" #include "defs.h" #include "mem.h" -#include "speaker.h" +#include "speaker-protocol.h" #include "user.h" -#include "addr.h" -#include "timeval.h" -#include "rtp.h" +#include "printf.h" +#include "version.h" +#include "uaudio.h" -#if API_ALSA -#include -#endif - -#ifdef WORDS_BIGENDIAN -# define MACHINE_AO_FMT AO_FMT_BIG -#else -# define MACHINE_AO_FMT AO_FMT_LITTLE -#endif - -/** @brief How many seconds of input to buffer - * - * While any given connection has this much audio buffered, no more reads will - * be issued for that connection. The decoder will have to wait. - */ -#define BUFFER_SECONDS 5 - -#define FRAMES 4096 /* Frame batch size */ +/** @brief Maximum number of FDs to poll for */ +#define NFDS 1024 -/** @brief Bytes to send per network packet +/** @brief Number of bytes before end of track to send SM_FINISHED * - * Don't make this too big or arithmetic will start to overflow. + * Generally set to 1 second. */ -#define NETWORK_BYTES (1024+sizeof(struct rtp_header)) - -/** @brief Maximum RTP playahead (ms) */ -#define RTP_AHEAD_MS 1000 - -/** @brief Maximum number of FDs to poll for */ -#define NFDS 256 +static size_t early_finish; /** @brief Track structure * * Known tracks are kept in a linked list. Usually there will be at most two * of these but rearranging the queue can cause there to be more. */ -static struct track { - struct track *next; /* next track */ - int fd; /* input FD */ - char id[24]; /* ID */ - size_t start, used; /* start + bytes used */ - int eof; /* input is at EOF */ - int got_format; /* got format yet? */ - ao_sample_format format; /* sample format */ - unsigned long long played; /* number of frames played */ - char *buffer; /* sample buffer */ - size_t size; /* sample buffer size */ - int slot; /* poll array slot */ -} *tracks, *playing; /* all tracks + playing track */ +struct track { + /** @brief Next track */ + struct track *next; -static time_t last_report; /* when we last reported */ -static int paused; /* pause status */ -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 -/** @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 + /** @brief Input file descriptor */ + int fd; -/** @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; + /** @brief Track ID */ + char id[24]; -static int forceplay; /* frames to force play */ -static int cmdfd = -1; /* child process input */ -static int bfd = -1; /* broadcast FD */ + /** @brief Start position of data in buffer */ + size_t start; -/** @brief RTP timestamp - * - * This counts the number of samples played (NB not the number of frames - * played). - * - * The timestamp in the packet header is only 32 bits wide. With 44100Hz - * stereo, that only gives about half a day before wrapping, which is not - * particularly convenient for certain debugging purposes. Therefore the - * timestamp is maintained as a 64-bit integer, giving around six million years - * before wrapping, and truncated to 32 bits when transmitting. - */ -static uint64_t rtp_time; + /** @brief Number of bytes of data in buffer */ + size_t used; -/** @brief RTP base timestamp - * - * This is the real time correspoding to an @ref rtp_time of 0. It is used - * to recalculate the timestamp after idle periods. - */ -static struct timeval rtp_time_0; + /** @brief Set @c fd is at EOF */ + int eof; -static uint16_t rtp_seq; /* frame sequence number */ -static uint32_t rtp_id; /* RTP SSRC */ -static int idled; /* set when idled */ -static int audio_errors; /* audio error counter */ + /** @brief Total number of samples played */ + unsigned long long played; -/** @brief Structure of a backend */ -struct speaker_backend { - /** @brief Which backend this is + /** @brief Slot in @ref fds */ + int slot; + + /** @brief Set when playable * - * @c -1 terminates the list. + * A track becomes playable whenever it fills its buffer or reaches EOF; it + * stops being playable when it entirely empties its buffer. Tracks start + * out life not playable. */ - int backend; + int playable; - /** @brief Flags + /** @brief Set when finished * - * Possible values - * - @ref FIXED_FORMAT + * This is set when we've notified the server that the track is finished. + * Once this has happened (typically very late in the track's lifetime) the + * track cannot be paused or cancelled. */ - unsigned flags; -/** @brief Lock to configured sample format */ -#define FIXED_FORMAT 0x0001 + int finished; - /** @brief Initialization - * - * Called once at startup. This is responsible for one-time setup - * operations, for instance opening a network socket to transmit to. + /** @brief Input buffer * - * 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. + * 1Mbyte is enough for nearly 6s of 44100Hz 16-bit stereo */ - void (*init)(void); + char buffer[1048576]; +}; - /** @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 Lock protecting data structures + * + * This lock protects values shared between the main thread and the callback. + * + * It is held 'all' the time by the main thread, the exceptions being when + * called activate/deactivate callbacks and when calling (potentially) slow + * system calls (in particular poll(), where in fact the main thread will spend + * most of its time blocked). + * + * The callback holds it when it's running. + */ +static pthread_mutex_t lock = PTHREAD_MUTEX_INITIALIZER; - /** @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 Linked list of all prepared tracks + * + * This includes @ref playing and @ref pending_playing. + */ +static struct track *tracks; - /** @brief Called before poll() - * - * Called before the call to poll(). Should call addfd() to update the FD - * array and stash the slot number somewhere safe. - */ - void (*beforepoll)(void); -}; +/** @brief Playing track, or NULL + * + * This means the track the speaker process intends to play. It does not + * reflect any other state (e.g. activation of uaudio backend). + * + * This track remains on @ref track. + */ +static struct track *playing; + +/** @brief Pending playing track, or NULL + * + * This means the track the server wants the speaker to play. + * + * This track remains on @p track. + */ +static struct track *pending_playing; + +/** @brief Array of file descriptors for poll() */ +static struct pollfd fds[NFDS]; + +/** @brief Next free slot in @ref fds + * + * This is used when filling in the @ref fds array each iteration through the + * event loop. + */ +static int fdno; + +/** @brief Listen socket */ +static int listenfd; + +/** @brief Timestamp of last potential report to server */ +static time_t last_report; + +/** @brief Set when paused */ +static int paused; + +/** @brief Set when back end activated */ +static int activated; + +/** @brief Signal pipe back into the poll() loop */ +static int sigpipe[2]; /** @brief Selected backend */ -static const struct speaker_backend *backend; +static const struct uaudio *backend; static const struct option options[] = { { "help", no_argument, 0, 'h' }, @@ -257,6 +229,8 @@ static const struct option options[] = { { "config", required_argument, 0, 'c' }, { "debug", no_argument, 0, 'd' }, { "no-debug", no_argument, 0, 'D' }, + { "syslog", no_argument, 0, 's' }, + { "no-syslog", no_argument, 0, 'S' }, { 0, 0, 0, 0 } }; @@ -269,6 +243,7 @@ static void help(void) { " --version, -V Display version number\n" " --config PATH, -c PATH Set configuration file\n" " --debug, -d Turn on debugging\n" + " --[no-]syslog Force logging\n" "\n" "Speaker process for DisOrder. Not intended to be run\n" "directly.\n"); @@ -276,19 +251,11 @@ static void help(void) { exit(0); } -/* Display version number and terminate. */ -static void version(void) { - xprintf("disorder-speaker version %s\n", disorder_version_string); - xfclose(stdout); - exit(0); -} - -/** @brief Return the number of bytes per frame in @p format */ -static size_t bytes_per_frame(const ao_sample_format *format) { - return format->channels * format->bits / 8; -} - -/** @brief Find track @p id, maybe creating it if not found */ +/** @brief Find track @p id, maybe creating it if not found + * @param id Track ID to find + * @param create If nonzero, create track structure of @p id if not found + * @return Pointer to track structure or NULL + */ static struct track *findtrack(const char *id, int create) { struct track *t; @@ -301,14 +268,14 @@ static struct track *findtrack(const char *id, int create) { strcpy(t->id, id); t->fd = -1; tracks = t; - /* The initial input buffer will be the sample format. */ - t->buffer = (void *)&t->format; - t->size = sizeof t->format; } return t; } -/** @brief Remove track @p id (but do not destroy it) */ +/** @brief Remove track @p id (but do not destroy it) + * @param id Track ID to remove + * @return Track structure or NULL if not found + */ static struct track *removetrack(const char *id) { struct track *t, **tt; @@ -320,338 +287,117 @@ static struct track *removetrack(const char *id) { return t; } -/** @brief Destroy a track */ +/** @brief Destroy a track + * @param t Track structure + */ static void destroy(struct track *t) { D(("destroy %s", t->id)); - if(t->fd != -1) xclose(t->fd); - if(t->buffer != (void *)&t->format) free(t->buffer); - free(t); -} - -/** @brief Notice a new connection */ -static void acquire(struct track *t, int fd) { - D(("acquire %s %d", t->id, fd)); if(t->fd != -1) xclose(t->fd); - t->fd = fd; - nonblock(fd); -} - -/** @brief Return true if A and B denote identical libao formats, else false */ -static int formats_equal(const ao_sample_format *a, - const ao_sample_format *b) { - return (a->bits == b->bits - && a->rate == b->rate - && a->channels == b->channels - && a->byte_format == b->byte_format); -} - -/** @brief Compute arguments to sox */ -static void soxargs(const char ***pp, char **qq, ao_sample_format *ao) { - int n; - - *(*pp)++ = "-t.raw"; - *(*pp)++ = "-s"; - *(*pp)++ = *qq; n = sprintf(*qq, "-r%d", ao->rate); *qq += n + 1; - *(*pp)++ = *qq; n = sprintf(*qq, "-c%d", ao->channels); *qq += n + 1; - /* sox 12.17.9 insists on -b etc; CVS sox insists on - etc; both are - * deployed! */ - switch(config->sox_generation) { - case 0: - if(ao->bits != 8 - && ao->byte_format != AO_FMT_NATIVE - && ao->byte_format != MACHINE_AO_FMT) { - *(*pp)++ = "-x"; - } - switch(ao->bits) { - case 8: *(*pp)++ = "-b"; break; - case 16: *(*pp)++ = "-w"; break; - case 32: *(*pp)++ = "-l"; break; - case 64: *(*pp)++ = "-d"; break; - default: fatal(0, "cannot handle sample size %d", (int)ao->bits); - } - break; - case 1: - switch(ao->byte_format) { - case AO_FMT_NATIVE: break; - case AO_FMT_BIG: *(*pp)++ = "-B"; break; - case AO_FMT_LITTLE: *(*pp)++ = "-L"; break; - } - *(*pp)++ = *qq; n = sprintf(*qq, "-%d", ao->bits/8); *qq += n + 1; - break; - } -} - -/** @brief Enable format translation - * - * If necessary, replaces a tracks inbound file descriptor with one connected - * to a sox invocation, which performs the required translation. - */ -static void enable_translation(struct track *t) { - 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]; - pid_t soxkid; - - *pp++ = "sox"; - soxargs(&pp, &q, &t->format); - *pp++ = "-"; - soxargs(&pp, &q, &config->sample_format); - *pp++ = "-"; - *pp++ = 0; - if(debugging) { - for(pp = av; *pp; pp++) - D(("sox arg[%d] = %s", pp - av, *pp)); - D(("end args")); - } - xpipe(soxpipe); - soxkid = xfork(); - if(soxkid == 0) { - signal(SIGPIPE, SIG_DFL); - xdup2(t->fd, 0); - xdup2(soxpipe[1], 1); - fcntl(0, F_SETFL, fcntl(0, F_GETFL) & ~O_NONBLOCK); - close(soxpipe[0]); - close(soxpipe[1]); - close(t->fd); - execvp("sox", (char **)av); - _exit(1); - } - D(("forking sox for format conversion (kid = %d)", soxkid)); - close(t->fd); - close(soxpipe[1]); - t->fd = soxpipe[0]; - t->format = config->sample_format; - } + free(t); } /** @brief Read data into a sample buffer * @param t Pointer to track * @return 0 on success, -1 on EOF * - * This is effectively the read callback on @c t->fd. + * This is effectively the read callback on @c t->fd. It is called from the + * main loop whenever the track's file descriptor is readable, assuming the + * buffer has not reached the maximum allowed occupancy. + * + * Errors count as EOF. */ -static int fill(struct track *t) { +static int speaker_fill(struct track *t) { size_t where, left; - int n; + int n, rc; - D(("fill %s: eof=%d used=%zu size=%zu got_format=%d", - t->id, t->eof, t->used, t->size, t->got_format)); - if(t->eof) return -1; - if(t->used < t->size) { + D(("fill %s: eof=%d used=%zu", + t->id, t->eof, t->used)); + if(t->eof) + return -1; + if(t->used < sizeof t->buffer) { /* there is room left in the buffer */ - where = (t->start + t->used) % t->size; - if(t->got_format) { - /* We are reading audio data, get as much as we can */ - if(where >= t->start) left = t->size - where; - else left = t->start - where; - } else - /* We are still waiting for the format, only get that */ - left = sizeof (ao_sample_format) - t->used; + where = (t->start + t->used) % sizeof t->buffer; + /* Get as much data as we can */ + if(where >= t->start) + left = (sizeof t->buffer) - where; + else + left = t->start - where; + pthread_mutex_unlock(&lock); do { n = read(t->fd, t->buffer + where, left); } while(n < 0 && errno == EINTR); - if(n < 0) { - if(errno != EAGAIN) fatal(errno, "error reading sample stream"); - return 0; - } - if(n == 0) { - D(("fill %s: eof detected", t->id)); + pthread_mutex_lock(&lock); + if(n < 0 && errno == EAGAIN) { + /* EAGAIN means more later */ + rc = 0; + } else if(n <= 0) { + /* n=0 means EOF. n<0 means some error occurred. We log the error but + * otherwise treat it as identical to EOF. */ + if(n < 0) + disorder_error(errno, "error reading sample stream for %s", t->id); + else + D(("fill %s: eof detected", t->id)); t->eof = 1; - return -1; - } - t->used += n; - if(!t->got_format && t->used >= sizeof (ao_sample_format)) { - assert(t->used == sizeof (ao_sample_format)); - /* Check that our assumptions are met. */ - if(t->format.bits & 7) - fatal(0, "bits per sample not a multiple of 8"); - /* If the input format is unsuitable, arrange to translate it */ - enable_translation(t); - /* Make a new buffer for audio data. */ - t->size = bytes_per_frame(&t->format) * t->format.rate * BUFFER_SECONDS; - t->buffer = xmalloc(t->size); - t->used = 0; - t->got_format = 1; - D(("got format for %s", t->id)); + /* A track always becomes playable at EOF; we're not going to see any + * more data. */ + t->playable = 1; + rc = -1; + } else { + t->used += n; + /* A track becomes playable when it (first) fills its buffer. For + * 44.1KHz 16-bit stereo this is ~6s of audio data. The latency will + * depend how long that takes to decode (hopefuly not very!) */ + if(t->used == sizeof t->buffer) + t->playable = 1; + rc = 0; } - } - return 0; -} - -/** @brief Close the sound device */ -static void idle(void) { - D(("idle")); - if(backend->deactivate) - backend->deactivate(); - idled = 1; - ready = 0; -} - -/** @brief Abandon the current track */ -static void abandon(void) { - struct speaker_message sm; - - D(("abandon")); - memset(&sm, 0, sizeof sm); - sm.type = SM_FINISHED; - strcpy(sm.id, playing->id); - speaker_send(1, &sm, 0); - removetrack(playing->id); - destroy(playing); - playing = 0; - forceplay = 0; -} - -#if API_ALSA -/** @brief Log ALSA parameters */ -static void log_params(snd_pcm_hw_params_t *hwparams, - snd_pcm_sw_params_t *swparams) { - snd_pcm_uframes_t f; - unsigned u; - - return; /* too verbose */ - if(hwparams) { - /* TODO */ - } - if(swparams) { - snd_pcm_sw_params_get_silence_size(swparams, &f); - info("sw silence_size=%lu", (unsigned long)f); - snd_pcm_sw_params_get_silence_threshold(swparams, &f); - info("sw silence_threshold=%lu", (unsigned long)f); - snd_pcm_sw_params_get_sleep_min(swparams, &u); - info("sw sleep_min=%lu", (unsigned long)u); - snd_pcm_sw_params_get_start_threshold(swparams, &f); - info("sw start_threshold=%lu", (unsigned long)f); - snd_pcm_sw_params_get_stop_threshold(swparams, &f); - info("sw stop_threshold=%lu", (unsigned long)f); - snd_pcm_sw_params_get_xfer_align(swparams, &f); - info("sw xfer_align=%lu", (unsigned long)f); - } + } else + rc = 0; + return rc; } -#endif -/** @brief Enable sound output +/** @brief Return nonzero if we want to play some audio * - * Makes sure the sound device is open and has the right sample format. Return - * 0 on success and -1 on error. + * We want to play audio if there is a current track; and it is not paused; and + * it is playable according to the rules for @ref track::playable. + * + * We don't allow tracks to be paused if we've already told the server we've + * finished them; that would cause such tracks to survive much longer than the + * few samples they're supposed to, with report() remaining silent for the + * duration. The effect is that if you hit pause towards the end of a track, + * what should happen is that it finished but the next one is paused right at + * its start. */ -static int activate(void) { - /* If we don't know the format yet we cannot start. */ - if(!playing->got_format) { - D((" - not got format for %s", playing->id)); - return -1; - } - return backend->activate(); -} - -/* Check to see whether the current track has finished playing */ -static void maybe_finished(void) { - if(playing - && playing->eof - && (!playing->got_format - || playing->used < bytes_per_frame(&playing->format))) - abandon(); -} - -static void fork_cmd(void) { - pid_t cmdpid; - int pfd[2]; - if(cmdfd != -1) close(cmdfd); - xpipe(pfd); - cmdpid = xfork(); - if(!cmdpid) { - signal(SIGPIPE, SIG_DFL); - xdup2(pfd[0], 0); - close(pfd[0]); - close(pfd[1]); - execl("/bin/sh", "sh", "-c", config->speaker_command, (char *)0); - fatal(errno, "error execing /bin/sh"); - } - close(pfd[0]); - cmdfd = pfd[1]; - D(("forked cmd %d, fd = %d", cmdpid, cmdfd)); +static int playable(void) { + return playing + && (!paused || playing->finished) + && playing->playable; } -static void play(size_t frames) { - size_t avail_frames, avail_bytes, written_frames; - ssize_t written_bytes; - - /* Make sure the output device is activated */ - if(activate()) { - if(playing) - forceplay = frames; - else - forceplay = 0; /* Must have called abandon() */ - return; - } - D(("play: play %zu/%zu%s %dHz %db %dc", frames, playing->used / bpf, - playing->eof ? " EOF" : "", - playing->format.rate, - playing->format.bits, - playing->format.channels)); - /* If we haven't got enough bytes yet wait until we have. Exception: when - * we are at eof. */ - if(playing->used < frames * bpf && !playing->eof) { - forceplay = frames; - return; - } - /* We have got enough data so don't force play again */ - 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; - 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; - playing->used -= written_bytes; - playing->played += written_frames; - /* If the pointer is at the end of the buffer (or the buffer is completely - * empty) wrap it back to the start. */ - if(!playing->used || playing->start == playing->size) - playing->start = 0; - frames -= written_frames; -} - -/* Notify the server what we're up to. */ +/** @brief Notify the server what we're up to */ static void report(void) { struct speaker_message sm; - if(playing && playing->buffer != (void *)&playing->format) { + if(playing) { + /* Had better not send a report for a track that the server thinks has + * finished, that would be confusing. */ + if(playing->finished) + return; memset(&sm, 0, sizeof sm); sm.type = paused ? SM_PAUSED : SM_PLAYING; strcpy(sm.id, playing->id); - sm.data = playing->played / playing->format.rate; - speaker_send(1, &sm, 0); + sm.data = playing->played / (uaudio_rate * uaudio_channels); + speaker_send(1, &sm); + xtime(&last_report); } - time(&last_report); -} - -static void reap(int __attribute__((unused)) sig) { - pid_t cmdpid; - int st; - - do - cmdpid = waitpid(-1, &st, WNOHANG); - while(cmdpid > 0); - signal(SIGCHLD, reap); } +/** @brief Add a file descriptor to the set to poll() for + * @param fd File descriptor + * @param events Events to wait for e.g. @c POLLIN + * @return Slot number + */ static int addfd(int fd, int events) { if(fdno < NFDS) { fds[fdno].fd = fd; @@ -661,679 +407,419 @@ static int addfd(int fd, int events) { return -1; } -#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; +/** @brief Callback to return some sampled data + * @param buffer Where to put sample data + * @param max_samples How many samples to return + * @param userdata User data + * @return Number of samples written + * + * See uaudio_callback(). + */ +static size_t speaker_callback(void *buffer, + size_t max_samples, + void attribute((unused)) *userdata) { + const size_t max_bytes = max_samples * uaudio_sample_size; + size_t provided_samples = 0; + + pthread_mutex_lock(&lock); + /* TODO perhaps we should immediately go silent if we've been asked to pause + * or cancel the playing track (maybe block in the cancel case and see what + * else turns up?) */ + if(playing) { + if(playing->used > 0) { + size_t bytes; + /* Compute size of largest contiguous chunk. We get called as often as + * necessary so there's no need for cleverness here. */ + if(playing->start + playing->used > sizeof playing->buffer) + bytes = sizeof playing->buffer - playing->start; + else + bytes = playing->used; + /* Limit to what we were asked for */ + if(bytes > max_bytes) + bytes = max_bytes; + /* Provide it */ + memcpy(buffer, playing->buffer + playing->start, bytes); + playing->start += bytes; + playing->used -= bytes; + /* Wrap around to start of buffer */ + if(playing->start == sizeof playing->buffer) + playing->start = 0; + /* See if we've reached the end of the track; if so make sure the event + * loop wakes up. */ + if(playing->used == 0 && playing->eof) { + int ignored = write(sigpipe[1], "", 1); + (void) ignored; } - 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; + provided_samples = bytes / uaudio_sample_size; + playing->played += provided_samples; } - 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; -} - -static int alsa_slots, alsa_nslots = -1; - -/** @brief Fill in poll fd array for ALSA */ -static void alsa_beforepoll(void) { - /* We send sample data to ALSA as fast as it can accept it, relying on - * the fact that it has a relatively small buffer to minimize pause - * latency. */ - int retry = 3, err; - - alsa_slots = fdno; - do { - retry = 0; - alsa_nslots = snd_pcm_poll_descriptors(pcm, &fds[fdno], NFDS - fdno); - if((alsa_nslots <= 0 - || !(fds[alsa_slots].events & POLLOUT)) - && snd_pcm_state(pcm) == SND_PCM_STATE_XRUN) { - error(0, "underrun detected after call to snd_pcm_poll_descriptors()"); - if((err = snd_pcm_prepare(pcm))) - fatal(0, "error calling snd_pcm_prepare: %d", err); - } else - break; - } while(retry-- > 0); - if(alsa_nslots >= 0) - fdno += alsa_nslots; -} - -/** @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; -} - -static int cmdfd_slot; - -/** @brief Update poll array for writing to subprocess */ -static void command_beforepoll(void) { - /* We send sample data to the subprocess as fast as it can accept it. - * This isn't ideal as pause latency can be very high as a result. */ - if(cmdfd >= 0) - cmdfd_slot = addfd(cmdfd, POLLOUT); -} - -/** @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, - PF_INET, - SOCK_DGRAM, - IPPROTO_UDP, - 0, - 0, - 0, - 0 - }; - static const struct addrinfo prefbind = { - AI_PASSIVE, - PF_INET, - SOCK_DGRAM, - IPPROTO_UDP, - 0, - 0, - 0, - 0 - }; - static const int one = 1; - 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); + /* If we couldn't provide anything at all, play dead air */ + /* TODO maybe it would be better to block, in some cases? */ + if(!provided_samples) { + memset(buffer, 0, max_bytes); + provided_samples = max_samples; + if(playing) + disorder_info("%zu samples silence, playing->used=%zu", + provided_samples, playing->used); 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; + disorder_info("%zu samples silence, playing=NULL", provided_samples); } + pthread_mutex_unlock(&lock); + return provided_samples; } -/** @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; -} - -static int bfd_slot; - -/** @brief Set up poll array for network play */ -static void network_beforepoll(void) { - struct timeval now; - uint64_t target_us; - uint64_t target_rtp_time; - const int64_t samples_ahead = ((uint64_t)RTP_AHEAD_MS - * config->sample_format.rate - * config->sample_format.channels - / 1000); - - /* If we're starting then initialize the base time */ - if(!rtp_time) - xgettimeofday(&rtp_time_0, 0); - /* We send audio data whenever we get RTP_AHEAD seconds or more - * behind */ - xgettimeofday(&now, 0); - target_us = tvsub_us(now, rtp_time_0); - assert(target_us <= UINT64_MAX / 88200); - target_rtp_time = (target_us * config->sample_format.rate - * config->sample_format.channels) - / 1000000; - if((int64_t)(rtp_time - target_rtp_time) < samples_ahead) - bfd_slot = addfd(bfd, POLLOUT); -} - -/** @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, - alsa_beforepoll - }, -#endif - { - BACKEND_COMMAND, - FIXED_FORMAT, - command_init, - generic_activate, - command_play, - 0, /* deactivate */ - command_beforepoll - }, - { - BACKEND_NETWORK, - FIXED_FORMAT, - network_init, - generic_activate, - network_play, - 0, /* deactivate */ - network_beforepoll - }, - { -1, 0, 0, 0, 0, 0, 0 } -}; - -int main(int argc, char **argv) { - int n, fd, stdin_slot, poke, timeout; +/** @brief Main event loop */ +static void mainloop(void) { struct track *t; struct speaker_message sm; -#if API_ALSA - int err; -#endif + int n, fd, stdin_slot, timeout, listen_slot, sigpipe_slot; - set_progname(argv); - if(!setlocale(LC_CTYPE, "")) fatal(errno, "error calling setlocale"); - while((n = getopt_long(argc, argv, "hVc:dD", options, 0)) >= 0) { - switch(n) { - case 'h': help(); - case 'V': version(); - case 'c': configfile = optarg; break; - case 'd': debugging = 1; break; - case 'D': debugging = 0; break; - default: fatal(0, "invalid option"); - } - } - if(getenv("DISORDER_DEBUG_SPEAKER")) debugging = 1; - /* If stderr is a TTY then log there, otherwise to syslog. */ - if(!isatty(2)) { - openlog(progname, LOG_PID, LOG_DAEMON); - log_default = &log_syslog; - } - if(config_read()) fatal(0, "cannot read configuration"); - /* ignore SIGPIPE */ - signal(SIGPIPE, SIG_IGN); - /* reap kids */ - signal(SIGCHLD, reap); - /* set nice value */ - xnice(config->nice_speaker); - /* change user */ - become_mortal(); - /* make sure we're not root, whatever the config says */ - if(getuid() == 0 || geteuid() == 0) fatal(0, "do not run as root"); - /* 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(); + pthread_mutex_lock(&lock); + /* Keep going while our parent process is alive */ while(getppid() != 1) { + int force_report = 0; + fdno = 0; + /* By default we will wait up to half a second before thinking about + * current state. */ + timeout = 500; /* Always ready for commands from the main server. */ stdin_slot = addfd(0, POLLIN); + /* Also always ready for inbound connections */ + listen_slot = addfd(listenfd, POLLIN); /* Try to read sample data for the currently playing track if there is * buffer space. */ - if(playing && !playing->eof && playing->used < playing->size) { + if(playing + && playing->fd >= 0 + && !playing->eof + && playing->used < (sizeof playing->buffer)) playing->slot = addfd(playing->fd, POLLIN); - } else if(playing) + else if(playing) playing->slot = -1; - /* If forceplay is set then wait until it succeeds before waiting on the - * sound device. */ - alsa_slots = -1; - cmdfd_slot = -1; - bfd_slot = -1; - /* By default we will wait up to a second before thinking about current - * state. */ - timeout = 1000; - /* We'll break the poll as soon as the underlying sound device is ready for - * more data */ - if(ready && !forceplay) - backend->beforepoll(); + /* Allow the poll() to be interrupted at the end of a track */ + sigpipe_slot = addfd(sigpipe[0], POLLIN); /* If any other tracks don't have a full buffer, try to read sample data - * from them. */ + * from them. We do this last of all, so that if we run out of slots, + * nothing important can't be monitored. */ for(t = tracks; t; t = t->next) if(t != playing) { - if(!t->eof && t->used < t->size) { + if(t->fd >= 0 + && !t->eof + && t->used < sizeof t->buffer) { t->slot = addfd(t->fd, POLLIN | POLLHUP); } else t->slot = -1; } /* Wait for something interesting to happen */ + pthread_mutex_unlock(&lock); n = poll(fds, fdno, timeout); + pthread_mutex_lock(&lock); if(n < 0) { if(errno == EINTR) continue; - fatal(errno, "error calling poll"); + disorder_fatal(errno, "error calling poll"); } - /* Play some sound before doing anything else */ - poke = 0; - switch(config->speaker_backend) { -#if API_ALSA - case BACKEND_ALSA: - if(alsa_slots != -1) { - unsigned short alsa_revents; - - if((err = snd_pcm_poll_descriptors_revents(pcm, - &fds[alsa_slots], - alsa_nslots, - &alsa_revents)) < 0) - fatal(0, "error calling snd_pcm_poll_descriptors_revents: %d", err); - if(alsa_revents & (POLLOUT | POLLERR)) - play(3 * FRAMES); - } else - poke = 1; - break; -#endif - case BACKEND_COMMAND: - if(cmdfd_slot != -1) { - if(fds[cmdfd_slot].revents & (POLLOUT | POLLERR)) - play(3 * FRAMES); - } else - poke = 1; - break; - case BACKEND_NETWORK: - if(bfd_slot != -1) { - if(fds[bfd_slot].revents & (POLLOUT | POLLERR)) - play(3 * FRAMES); + /* Perhaps a connection has arrived */ + if(fds[listen_slot].revents & POLLIN) { + struct sockaddr_un addr; + socklen_t addrlen = sizeof addr; + uint32_t l; + char id[24]; + + if((fd = accept(listenfd, (struct sockaddr *)&addr, &addrlen)) >= 0) { + /* We do blocking reads for the header. In theory this means that the + * connecting process could wedge the speaker indefinitely. In + * practice that would mean that the main server was broken anyway. + * Still, this is ugly, and a rewrite would be nice. */ + blocking(fd); + if(read(fd, &l, sizeof l) < 4) { + disorder_error(errno, "reading length from inbound connection"); + xclose(fd); + } else if(l >= sizeof id) { + disorder_error(0, "id length too long"); + xclose(fd); + } else if(read(fd, id, l) < (ssize_t)l) { + disorder_error(errno, "reading id from inbound connection"); + xclose(fd); + } else { + id[l] = 0; + D(("id %s fd %d", id, fd)); + t = findtrack(id, 1/*create*/); + if (write(fd, "", 1) < 0) /* write an ack */ + disorder_error(errno, "writing ack to inbound connection for %s", + id); + if(t->fd != -1) { + disorder_error(0, "%s: already got a connection", id); + xclose(fd); + } else { + nonblock(fd); + t->fd = fd; /* yay */ + } + /* Notify the server that the connection arrived */ + sm.type = SM_ARRIVED; + strcpy(sm.id, id); + speaker_send(1, &sm); + } } else - poke = 1; - break; - } - if(poke) { - /* Some attempt to play must have failed */ - if(playing && !paused) - play(forceplay); - else - forceplay = 0; /* just in case */ + disorder_error(errno, "accept"); } /* Perhaps we have a command to process */ if(fds[stdin_slot].revents & POLLIN) { - n = speaker_recv(0, &sm, &fd); + /* There might (in theory) be several commands queued up, but in general + * this won't be the case, so we don't bother looping around to pick them + * all up. */ + n = speaker_recv(0, &sm); if(n > 0) + /* As a rule we don't send success replies to most commands - we just + * force the regular status update to be sent immediately rather than + * on schedule. */ switch(sm.type) { - case SM_PREPARE: - D(("SM_PREPARE %s %d", sm.id, fd)); - if(fd == -1) fatal(0, "got SM_PREPARE but no file descriptor"); - t = findtrack(sm.id, 1); - acquire(t, fd); - break; case SM_PLAY: - D(("SM_PLAY %s %d", sm.id, fd)); - if(playing) fatal(0, "got SM_PLAY but already playing something"); + /* SM_PLAY is only allowed if the server reasonably believes that + * nothing is playing */ + if(playing) { + /* If finished isn't set then the server can't believe that this + * track has finished */ + if(!playing->finished) + disorder_fatal(0, "got SM_PLAY but already playing something"); + /* If pending_playing is set then the server must believe that that + * is playing */ + if(pending_playing) + disorder_fatal(0, "got SM_PLAY but have a pending playing track"); + } t = findtrack(sm.id, 1); - if(fd != -1) acquire(t, fd); - playing = t; - play(bufsize); - report(); + D(("SM_PLAY %s fd %d", t->id, t->fd)); + if(t->fd == -1) + disorder_error(0, + "cannot play track because no connection arrived"); + /* TODO as things stand we often report this error message but then + * appear to proceed successfully. Understanding why requires a look + * at play.c: we call prepare() which makes the connection in a child + * process, and then sends the SM_PLAY in the parent process. The + * latter may well be faster. As it happens this is harmless; we'll + * just sit around sending silence until the decoder connects and + * starts sending some sample data. But is is annoying and ought to + * be fixed. */ + pending_playing = t; + /* If nothing is currently playing then we'll switch to the pending + * track below so there's no point distinguishing the situations + * here. */ break; case SM_PAUSE: D(("SM_PAUSE")); paused = 1; - report(); + force_report = 1; break; case SM_RESUME: D(("SM_RESUME")); - if(paused) { - paused = 0; - if(playing) - play(bufsize); - } - report(); + paused = 0; + force_report = 1; break; case SM_CANCEL: - D(("SM_CANCEL %s", sm.id)); + D(("SM_CANCEL %s", sm.id)); t = removetrack(sm.id); if(t) { - if(t == playing) { + if(t == playing || t == pending_playing) { + /* Scratching the track that the server believes is playing, + * which might either be the actual playing track or a pending + * playing track */ sm.type = SM_FINISHED; - strcpy(sm.id, playing->id); - speaker_send(1, &sm, 0); - playing = 0; + if(t == playing) + playing = 0; + else + pending_playing = 0; + } else { + /* Could be scratching the playing track before it's quite got + * going, or could be just removing a track from the queue. We + * log more because there's been a bug here recently than because + * it's particularly interesting; the log message will be removed + * if no further problems show up. */ + disorder_info("SM_CANCEL for nonplaying track %s", sm.id); + sm.type = SM_STILLBORN; } + strcpy(sm.id, t->id); destroy(t); - } else - error(0, "SM_CANCEL for unknown track %s", sm.id); - report(); + } else { + /* Probably scratching the playing track well before it's got + * going, but could indicate a bug, so we log this as an error. */ + sm.type = SM_UNKNOWN; + disorder_error(0, "SM_CANCEL for unknown track %s", sm.id); + } + speaker_send(1, &sm); + force_report = 1; break; case SM_RELOAD: D(("SM_RELOAD")); - if(config_read()) error(0, "cannot read configuration"); - info("reloaded configuration"); + if(config_read(1, NULL)) + disorder_error(0, "cannot read configuration"); + disorder_info("reloaded configuration"); break; default: - error(0, "unknown message type %d", sm.type); + disorder_error(0, "unknown message type %d", sm.type); } } /* Read in any buffered data */ for(t = tracks; t; t = t->next) - if(t->slot != -1 && (fds[t->slot].revents & (POLLIN | POLLHUP))) - fill(t); - /* We might be able to play now */ - if(ready && forceplay && playing && !paused) - play(forceplay); - /* Maybe we finished playing a track somewhere in the above */ - maybe_finished(); - /* If we don't need the sound device for now then close it for the benefit - * of anyone else who wants it. */ - if((!playing || paused) && ready) - idle(); - /* If we've not reported out state for a second do so now. */ - if(time(0) > last_report) + if(t->fd != -1 + && t->slot != -1 + && (fds[t->slot].revents & (POLLIN | POLLHUP))) + speaker_fill(t); + /* Drain the signal pipe. We don't care about its contents, merely that it + * interrupted poll(). */ + if(fds[sigpipe_slot].revents & POLLIN) { + char buffer[64]; + int ignored; (void)ignored; + + ignored = read(sigpipe[0], buffer, sizeof buffer); + } + /* Send SM_FINISHED when we're near the end of the track. + * + * This is how we implement gapless play; we hope that the SM_PLAY from the + * server arrives before the remaining bytes of the track play out. + */ + if(playing + && playing->eof + && !playing->finished + && playing->used <= early_finish) { + memset(&sm, 0, sizeof sm); + sm.type = SM_FINISHED; + strcpy(sm.id, playing->id); + speaker_send(1, &sm); + playing->finished = 1; + } + /* When the track is actually finished, deconfigure it */ + if(playing && playing->eof && !playing->used) { + if(!playing->finished) { + /* should never happen but we'd like to know if it does */ + disorder_fatal(0, "track finish state inconsistent"); + } + removetrack(playing->id); + destroy(playing); + playing = 0; + } + /* Act on the pending SM_PLAY */ + if(!playing && pending_playing) { + playing = pending_playing; + pending_playing = 0; + force_report = 1; + } + /* Impose any state change required by the above */ + if(playable()) { + if(!activated) { + activated = 1; + pthread_mutex_unlock(&lock); + backend->activate(); + pthread_mutex_lock(&lock); + } + } else { + if(activated) { + activated = 0; + pthread_mutex_unlock(&lock); + backend->deactivate(); + pthread_mutex_lock(&lock); + } + } + /* If we've not reported our state for a second do so now. */ + if(force_report || xtime(0) > last_report) report(); } - info("stopped (parent terminated)"); +} + +int main(int argc, char **argv) { + int n, logsyslog = !isatty(2); + struct sockaddr_un addr; + static const int one = 1; + struct speaker_message sm; + const char *d; + char *dir; + struct rlimit rl[1]; + + set_progname(argv); + if(!setlocale(LC_CTYPE, "")) disorder_fatal(errno, "error calling setlocale"); + while((n = getopt_long(argc, argv, "hVc:dDSs", options, 0)) >= 0) { + switch(n) { + case 'h': help(); + case 'V': version("disorder-speaker"); + case 'c': configfile = optarg; break; + case 'd': debugging = 1; break; + case 'D': debugging = 0; break; + case 'S': logsyslog = 0; break; + case 's': logsyslog = 1; break; + default: disorder_fatal(0, "invalid option"); + } + } + if((d = getenv("DISORDER_DEBUG_SPEAKER"))) debugging = atoi(d); + if(logsyslog) { + openlog(progname, LOG_PID, LOG_DAEMON); + log_default = &log_syslog; + } + config_uaudio_apis = uaudio_apis; + if(config_read(1, NULL)) disorder_fatal(0, "cannot read configuration"); + /* ignore SIGPIPE */ + signal(SIGPIPE, SIG_IGN); + /* set nice value */ + xnice(config->nice_speaker); + /* change user */ + become_mortal(); + /* make sure we're not root, whatever the config says */ + if(getuid() == 0 || geteuid() == 0) + disorder_fatal(0, "do not run as root"); + /* Make sure we can't have more than NFDS files open (it would bust our + * poll() array) */ + if(getrlimit(RLIMIT_NOFILE, rl) < 0) + disorder_fatal(errno, "getrlimit RLIMIT_NOFILE"); + if(rl->rlim_cur > NFDS) { + rl->rlim_cur = NFDS; + if(setrlimit(RLIMIT_NOFILE, rl) < 0) + disorder_fatal(errno, "setrlimit to reduce RLIMIT_NOFILE to %lu", + (unsigned long)rl->rlim_cur); + disorder_info("set RLIM_NOFILE to %lu", (unsigned long)rl->rlim_cur); + } else + disorder_info("RLIM_NOFILE is %lu", (unsigned long)rl->rlim_cur); + /* gcrypt initialization */ + if(!gcry_check_version(NULL)) + disorder_fatal(0, "gcry_check_version failed"); + gcry_control(GCRYCTL_INIT_SECMEM, 0); + gcry_control (GCRYCTL_INITIALIZATION_FINISHED, 0); + /* create a pipe between the backend callback and the poll() loop */ + xpipe(sigpipe); + nonblock(sigpipe[0]); + /* set up audio backend */ + uaudio_set_format(config->sample_format.rate, + config->sample_format.channels, + config->sample_format.bits, + config->sample_format.bits != 8); + early_finish = uaudio_sample_size * uaudio_channels * uaudio_rate; + /* TODO other parameters! */ + backend = uaudio_find(config->api); + /* backend-specific initialization */ + if(backend->configure) + backend->configure(); + backend->start(speaker_callback, NULL); + /* create the socket directory */ + byte_xasprintf(&dir, "%s/speaker", config->home); + unlink(dir); /* might be a leftover socket */ + if(mkdir(dir, 0700) < 0 && errno != EEXIST) + disorder_fatal(errno, "error creating %s", dir); + /* set up the listen socket */ + listenfd = xsocket(PF_UNIX, SOCK_STREAM, 0); + memset(&addr, 0, sizeof addr); + addr.sun_family = AF_UNIX; + snprintf(addr.sun_path, sizeof addr.sun_path, "%s/speaker/socket", + config->home); + if(unlink(addr.sun_path) < 0 && errno != ENOENT) + disorder_error(errno, "removing %s", addr.sun_path); + xsetsockopt(listenfd, SOL_SOCKET, SO_REUSEADDR, &one, sizeof one); + if(bind(listenfd, (const struct sockaddr *)&addr, sizeof addr) < 0) + disorder_fatal(errno, "error binding socket to %s", addr.sun_path); + xlisten(listenfd, 128); + nonblock(listenfd); + disorder_info("listening on %s", addr.sun_path); + memset(&sm, 0, sizeof sm); + sm.type = SM_READY; + speaker_send(1, &sm); + mainloop(); + disorder_info("stopped (parent terminated)"); exit(0); }