X-Git-Url: http://www.chiark.greenend.org.uk/ucgi/~mdw/git/disorder/blobdiff_plain/22b9fa74de8e80471a5033ea067d3b360930b91d..e99d42b153659e7ac644bb93700acb81514998e5:/server/speaker.c

diff --git a/server/speaker.c b/server/speaker.c
index 1dc90e4..7f4578d 100644
--- a/server/speaker.c
+++ b/server/speaker.c
@@ -22,22 +22,17 @@
  *
  * 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 Encodings.  For the <a href="http://www.alsa-project.org/">ALSA</a> API,
  * 8- and 16- bit stereo and mono are supported, with any sample rate (within
  * the limits that ALSA can deal with.)
  *
- * When communicating with a subprocess, <a
- * href="http://sox.sourceforge.net/">sox</a> 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.
- *
- * 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
@@ -72,6 +67,7 @@
 #include <time.h>
 #include <fcntl.h>
 #include <poll.h>
+#include <sys/un.h>
 
 #include "configuration.h"
 #include "syscalls.h"
@@ -89,7 +85,7 @@ struct track *tracks;
 struct track *playing;
 
 /** @brief Number of bytes pre frame */
-size_t device_bpf;
+size_t bpf;
 
 /** @brief Array of file descriptors for poll() */
 struct pollfd fds[NFDS];
@@ -97,20 +93,15 @@ struct pollfd fds[NFDS];
 /** @brief Next free slot in @ref fds */
 int fdno;
 
+/** @brief Listen socket */
+static int listenfd;
+
 static time_t last_report;              /* when we last reported */
 static int paused;                      /* pause status */
 
 /** @brief The current device state */
 enum device_states device_state;
 
-/** @brief The current device sample format
- *
- * Only meaningful if @ref device_state = @ref device_open or perhaps @ref
- * device_error.  For @ref FIXED_FORMAT backends, this should always match @c
- * config->sample_format.
- */
-ao_sample_format device_format;
-
 /** @brief Set when idled
  *
  * This is set when the sound device is deliberately closed by idle().
@@ -153,7 +144,7 @@ static void version(void) {
 }
 
 /** @brief Return the number of bytes per frame in @p format */
-static size_t bytes_per_frame(const ao_sample_format *format) {
+static size_t bytes_per_frame(const struct stream_header *format) {
   return format->channels * format->bits / 8;
 }
 
@@ -170,9 +161,6 @@ 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;
 }
@@ -193,109 +181,9 @@ static struct track *removetrack(const char *id) {
 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 */
-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 -<n> 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;
-  }
-}
-
 /** @brief Read data into a sample buffer
  * @param t Pointer to track
  * @return 0 on success, -1 on EOF
@@ -308,19 +196,15 @@ static int fill(struct track *t) {
   size_t where, left;
   int n;
 
-  D(("fill %s: eof=%d used=%zu size=%zu  got_format=%d",
-     t->id, t->eof, t->used, t->size, t->got_format));
+  D(("fill %s: eof=%d used=%zu",
+     t->id, t->eof, t->used));
   if(t->eof) return -1;
-  if(t->used < t->size) {
+  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;
     do {
       n = read(t->fd, t->buffer + where, left);
     } while(n < 0 && errno == EINTR);
@@ -334,20 +218,6 @@ static int fill(struct track *t) {
       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));
-    }
   }
   return 0;
 }
@@ -375,7 +245,7 @@ void abandon(void) {
   memset(&sm, 0, sizeof sm);
   sm.type = SM_FINISHED;
   strcpy(sm.id, playing->id);
-  speaker_send(1, &sm, 0);
+  speaker_send(1, &sm);
   removetrack(playing->id);
   destroy(playing);
   playing = 0;
@@ -387,22 +257,10 @@ void abandon(void) {
  * 0 on success and -1 on error.
  */
 static void 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;
-  }
-  if(backend->flags & FIXED_FORMAT)
-    device_format = config->sample_format;
-  if(backend->activate) {
+  if(backend->activate)
     backend->activate();
-  } else {
-    assert(backend->flags & FIXED_FORMAT);
-    /* ...otherwise device_format not set */
+  else
     device_state = device_open;
-  }
-  if(device_state == device_open)
-    device_bpf = bytes_per_frame(&device_format);
 }
 
 /** @brief Check whether the current track has finished
@@ -415,8 +273,7 @@ static void activate(void) {
 static void maybe_finished(void) {
   if(playing
      && playing->eof
-     && (!playing->got_format
-         || playing->used < bytes_per_frame(&playing->format)))
+     && playing->used < bytes_per_frame(&config->sample_format))
     abandon();
 }
 
@@ -440,26 +297,25 @@ static void play(size_t frames) {
   /* Make sure there's a track to play and it is not pasued */
   if(!playing || paused)
     return;
-  /* Make sure the output device is open and has the right sample format */
-  if(device_state != device_open
-     || !formats_equal(&device_format, &playing->format)) {
+  /* Make sure the output device is open */
+  if(device_state != device_open) {
     activate(); 
     if(device_state != device_open)
       return;
   }
-  D(("play: play %zu/%zu%s %dHz %db %dc",  frames, playing->used / device_bpf,
+  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));
+     config->sample_format.rate,
+     config->sample_format.bits,
+     config->sample_format.channels));
   /* Figure out how many frames there are available to write */
-  if(playing->start + playing->used > playing->size)
+  if(playing->start + playing->used > sizeof playing->buffer)
     /* The ring buffer is currently wrapped, only play up to the wrap point */
-    avail_bytes = playing->size - playing->start;
+    avail_bytes = (sizeof playing->buffer) - playing->start;
   else
     /* The ring buffer is not wrapped, can play the lot */
     avail_bytes = playing->used;
-  avail_frames = avail_bytes / device_bpf;
+  avail_frames = avail_bytes / bpf;
   /* Only play up to the requested amount */
   if(avail_frames > frames)
     avail_frames = frames;
@@ -467,7 +323,7 @@ static void play(size_t frames) {
     return;
   /* Play it, Sam */
   written_frames = backend->play(avail_frames);
-  written_bytes = written_frames * device_bpf;
+  written_bytes = written_frames * bpf;
   /* written_bytes and written_frames had better both be set and correct by
    * this point */
   playing->start += written_bytes;
@@ -475,7 +331,7 @@ static void play(size_t frames) {
   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)
+  if(!playing->used || playing->start == (sizeof playing->buffer))
     playing->start = 0;
   frames -= written_frames;
   return;
@@ -485,12 +341,12 @@ static void play(size_t frames) {
 static void report(void) {
   struct speaker_message sm;
 
-  if(playing && playing->buffer != (void *)&playing->format) {
+  if(playing) {
     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 / config->sample_format.rate;
+    speaker_send(1, &sm);
   }
   time(&last_report);
 }
@@ -521,6 +377,12 @@ static const struct speaker_backend *backends[] = {
 #endif
   &command_backend,
   &network_backend,
+#if HAVE_COREAUDIO_AUDIOHARDWARE_H
+  &coreaudio_backend,
+#endif
+#if HAVE_SYS_SOUNDCARD_H
+  &oss_backend,
+#endif
   0
 };
 
@@ -540,7 +402,7 @@ static int playable(void) {
 static void mainloop(void) {
   struct track *t;
   struct speaker_message sm;
-  int n, fd, stdin_slot, timeout;
+  int n, fd, stdin_slot, timeout, listen_slot;
 
   while(getppid() != 1) {
     fdno = 0;
@@ -549,9 +411,14 @@ static void mainloop(void) {
     timeout = 1000;
     /* 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)
       playing->slot = -1;
@@ -572,7 +439,9 @@ static void mainloop(void) {
      * 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;
@@ -599,25 +468,55 @@ static void mainloop(void) {
         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) {
+        blocking(fd);
+        if(read(fd, &l, sizeof l) < 4) {
+          error(errno, "reading length from inbound connection");
+          xclose(fd);
+        } else if(l >= sizeof id) {
+          error(0, "id length too long");
+          xclose(fd);
+        } else if(read(fd, id, l) < (ssize_t)l) {
+          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*/);
+          write(fd, "", 1);             /* write an ack */
+          if(t->fd != -1) {
+            error(0, "got a connection for a track that already has one");
+            xclose(fd);
+          } else {
+            nonblock(fd);
+            t->fd = fd;               /* yay */
+          }
+        }
+      } else
+        error(errno, "accept");
+    }
     /* Perhaps we have a command to process */
     if(fds[stdin_slot].revents & POLLIN) {
       /* 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, &fd);
+      n = speaker_recv(0, &sm);
+      /* TODO */
       if(n > 0)
 	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");
 	  t = findtrack(sm.id, 1);
-          if(fd != -1) acquire(t, fd);
+          D(("SM_PLAY %s fd %d", t->id, t->fd));
+          if(t->fd == -1)
+            error(0, "cannot play track because no connection arrived");
           playing = t;
           /* We attempt to play straight away rather than going round the loop.
            * play() is clever enough to perform any activation that is
@@ -647,7 +546,7 @@ static void mainloop(void) {
 	    if(t == playing) {
               sm.type = SM_FINISHED;
               strcpy(sm.id, playing->id);
-              speaker_send(1, &sm, 0);
+              speaker_send(1, &sm);
 	      playing = 0;
             }
 	    destroy(t);
@@ -657,7 +556,7 @@ static void mainloop(void) {
 	  break;
 	case SM_RELOAD:
           D(("SM_RELOAD"));
-	  if(config_read()) error(0, "cannot read configuration");
+	  if(config_read(1)) error(0, "cannot read configuration");
           info("reloaded configuration");
 	  break;
 	default:
@@ -666,7 +565,9 @@ static void mainloop(void) {
     }
     /* Read in any buffered data */
     for(t = tracks; t; t = t->next)
-      if(t->slot != -1 && (fds[t->slot].revents & (POLLIN | POLLHUP)))
+      if(t->fd != -1
+         && t->slot != -1
+         && (fds[t->slot].revents & (POLLIN | POLLHUP)))
          fill(t);
     /* Maybe we finished playing a track somewhere in the above */
     maybe_finished();
@@ -682,6 +583,9 @@ static void mainloop(void) {
 
 int main(int argc, char **argv) {
   int n;
+  struct sockaddr_un addr;
+  static const int one = 1;
+  struct speaker_message sm;
 
   set_progname(argv);
   if(!setlocale(LC_CTYPE, "")) fatal(errno, "error calling setlocale");
@@ -701,7 +605,8 @@ int main(int argc, char **argv) {
     openlog(progname, LOG_PID, LOG_DAEMON);
     log_default = &log_syslog;
   }
-  if(config_read()) fatal(0, "cannot read configuration");
+  if(config_read(1)) fatal(0, "cannot read configuration");
+  bpf = bytes_per_frame(&config->sample_format);
   /* ignore SIGPIPE */
   signal(SIGPIPE, SIG_IGN);
   /* reap kids */
@@ -721,6 +626,23 @@ int main(int argc, char **argv) {
   backend = backends[n];
   /* backend-specific initialization */
   backend->init();
+  /* 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",
+           config->home);
+  if(unlink(addr.sun_path) < 0 && errno != ENOENT)
+    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)
+    fatal(errno, "error binding socket to %s", addr.sun_path);
+  xlisten(listenfd, 128);
+  nonblock(listenfd);
+  info("listening on %s", addr.sun_path);
+  memset(&sm, 0, sizeof sm);
+  sm.type = SM_READY;
+  speaker_send(1, &sm);
   mainloop();
   info("stopped (parent terminated)");
   exit(0);