[disorder] / clients / playrtp.c

/*
 * This file is part of DisOrder.
 * Copyright (C) 2007-2009 Richard Kettlewell
 *
 * 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 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.
 * 
 * You should have received a copy of the GNU General Public License
 * along with this program.  If not, see <http://www.gnu.org/licenses/>.
 */
/** @file clients/playrtp.c
 * @brief RTP player
 *
 * This player supports Linux (<a href="http://www.alsa-project.org/">ALSA</a>)
 * and Apple Mac (<a
 * href="http://developer.apple.com/audio/coreaudio.html">Core Audio</a>)
 * systems.  There is no support for Microsoft Windows yet, and that will in
 * fact probably an entirely separate program.
 *
 * The program runs (at least) three threads:
 *
 * listen_thread() is responsible for reading RTP packets off the wire and
 * adding them to the linked list @ref received_packets, assuming they are
 * basically sound.
 *
 * queue_thread() takes packets off this linked list and adds them to @ref
 * packets (an operation which might be much slower due to contention for @ref
 * lock).
 *
 * control_thread() accepts commands from Disobedience (or anything else).
 *
 * The main thread activates and deactivates audio playing via the @ref
 * lib/uaudio.h API (which probably implies at least one further thread).
 *
 * Sometimes it happens that there is no audio available to play.  This may
 * because the server went away, or a packet was dropped, or the server
 * deliberately did not send any sound because it encountered a silence.
 *
 * Assumptions:
 * - it is safe to read uint32_t values without a lock protecting them
 */

#include "common.h"

#include <getopt.h>
#include <sys/socket.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <netdb.h>
#include <pthread.h>
#include <locale.h>
#include <sys/uio.h>
#include <errno.h>
#include <netinet/in.h>
#include <sys/time.h>
#include <sys/un.h>
#include <unistd.h>
#include <sys/mman.h>
#include <fcntl.h>
#include <math.h>

#include "log.h"
#include "mem.h"
#include "configuration.h"
#include "addr.h"
#include "syscalls.h"
#include "rtp.h"
#include "defs.h"
#include "vector.h"
#include "heap.h"
#include "timeval.h"
#include "client.h"
#include "playrtp.h"
#include "inputline.h"
#include "version.h"
#include "uaudio.h"

/** @brief Obsolete synonym */
#ifndef IPV6_JOIN_GROUP
# define IPV6_JOIN_GROUP IPV6_ADD_MEMBERSHIP
#endif

/** @brief RTP socket */
static int rtpfd;

/** @brief Log output */
static FILE *logfp;

/** @brief Output device */

/** @brief Buffer low watermark in samples */
unsigned minbuffer = 4 * (2 * 44100) / 10;  /* 0.4 seconds */

/** @brief Maximum buffer size in samples
 *
 * We'll stop reading from the network if we have this many samples.
 */
static unsigned maxbuffer;

/** @brief Received packets
 * Protected by @ref receive_lock
 *
 * Received packets are added to this list, and queue_thread() picks them off
 * it and adds them to @ref packets.  Whenever a packet is added to it, @ref
 * receive_cond is signalled.
 */
struct packet *received_packets;

/** @brief Tail of @ref received_packets
 * Protected by @ref receive_lock
 */
struct packet **received_tail = &received_packets;

/** @brief Lock protecting @ref received_packets 
 *
 * Only listen_thread() and queue_thread() ever hold this lock.  It is vital
 * that queue_thread() not hold it any longer than it strictly has to. */
pthread_mutex_t receive_lock = PTHREAD_MUTEX_INITIALIZER;

/** @brief Condition variable signalled when @ref received_packets is updated
 *
 * Used by listen_thread() to notify queue_thread() that it has added another
 * packet to @ref received_packets. */
pthread_cond_t receive_cond = PTHREAD_COND_INITIALIZER;

/** @brief Length of @ref received_packets */
uint32_t nreceived;

/** @brief Binary heap of received packets */
struct pheap packets;

/** @brief Total number of samples available
 *
 * We make this volatile because we inspect it without a protecting lock,
 * so the usual pthread_* guarantees aren't available.
 */
volatile uint32_t nsamples;

/** @brief Timestamp of next packet to play.
 *
 * This is set to the timestamp of the last packet, plus the number of
 * samples it contained.  Only valid if @ref active is nonzero.
 */
uint32_t next_timestamp;

/** @brief True if actively playing
 *
 * This is true when playing and false when just buffering. */
int active;

/** @brief Lock protecting @ref packets */
pthread_mutex_t lock = PTHREAD_MUTEX_INITIALIZER;

/** @brief Condition variable signalled whenever @ref packets is changed */
pthread_cond_t cond = PTHREAD_COND_INITIALIZER;

/** @brief Backend to play with */
static const struct uaudio *backend;

HEAP_DEFINE(pheap, struct packet *, lt_packet);

/** @brief Control socket or NULL */
const char *control_socket;

/** @brief Buffer for debugging dump
 *
 * The debug dump is enabled by the @c --dump option.  It records the last 20s
 * of audio to the specified file (which will be about 3.5Mbytes).  The file is
 * written as as ring buffer, so the start point will progress through it.
 *
 * Use clients/dump2wav to convert this to a WAV file, which can then be loaded
 * into (e.g.) Audacity for further inspection.
 *
 * All three backends (ALSA, OSS, Core Audio) now support this option.
 *
 * The idea is to allow the user a few seconds to react to an audible artefact.
 */
int16_t *dump_buffer;

/** @brief Current index within debugging dump */
size_t dump_index;

/** @brief Size of debugging dump in samples */
size_t dump_size = 44100/*Hz*/ * 2/*channels*/ * 20/*seconds*/;

static const struct option options[] = {
  { "help", no_argument, 0, 'h' },
  { "version", no_argument, 0, 'V' },
  { "debug", no_argument, 0, 'd' },
  { "device", required_argument, 0, 'D' },
  { "min", required_argument, 0, 'm' },
  { "max", required_argument, 0, 'x' },
  { "rcvbuf", required_argument, 0, 'R' },
#if HAVE_SYS_SOUNDCARD_H || EMPEG_HOST
  { "oss", no_argument, 0, 'o' },
#endif
#if HAVE_ALSA_ASOUNDLIB_H
  { "alsa", no_argument, 0, 'a' },
#endif
#if HAVE_COREAUDIO_AUDIOHARDWARE_H
  { "core-audio", no_argument, 0, 'c' },
#endif
  { "dump", required_argument, 0, 'r' },
  { "command", required_argument, 0, 'e' },
  { "pause-mode", required_argument, 0, 'P' },
  { "socket", required_argument, 0, 's' },
  { "config", required_argument, 0, 'C' },
  { "monitor", no_argument, 0, 'M' },
  { 0, 0, 0, 0 }
};

/** @brief Control thread
 *
 * This thread is responsible for accepting control commands from Disobedience
 * (or other controllers) over an AF_UNIX stream socket with a path specified
 * by the @c --socket option.  The protocol uses simple string commands and
 * replies:
 *
 * - @c stop will shut the player down
 * - @c query will send back the reply @c running
 * - anything else is ignored
 *
 * Commands and response strings terminated by shutting down the connection or
 * by a newline.  No attempt is made to multiplex multiple clients so it is
 * important that the command be sent as soon as the connection is made - it is
 * assumed that both parties to the protocol are entirely cooperating with one
 * another.
 */
static void *control_thread(void attribute((unused)) *arg) {
  struct sockaddr_un sa;
  int sfd, cfd;
  char *line;
  socklen_t salen;
  FILE *fp;

  assert(control_socket);
  unlink(control_socket);
  memset(&sa, 0, sizeof sa);
  sa.sun_family = AF_UNIX;
  strcpy(sa.sun_path, control_socket);
  sfd = xsocket(PF_UNIX, SOCK_STREAM, 0);
  if(bind(sfd, (const struct sockaddr *)&sa, sizeof sa) < 0)
    disorder_fatal(errno, "error binding to %s", control_socket);
  if(listen(sfd, 128) < 0)
    disorder_fatal(errno, "error calling listen on %s", control_socket);
  disorder_info("listening on %s", control_socket);
  for(;;) {
    salen = sizeof sa;
    cfd = accept(sfd, (struct sockaddr *)&sa, &salen);
    if(cfd < 0) {
      switch(errno) {
      case EINTR:
      case EAGAIN:
        break;
      default:
        disorder_fatal(errno, "error calling accept on %s", control_socket);
      }
    }
    if(!(fp = fdopen(cfd, "r+"))) {
      disorder_error(errno, "error calling fdopen for %s connection", control_socket);
      close(cfd);
      continue;
    }
    if(!inputline(control_socket, fp, &line, '\n')) {
      if(!strcmp(line, "stop")) {
        disorder_info("stopped via %s", control_socket);
        exit(0);                          /* terminate immediately */
      }
      if(!strcmp(line, "query"))
        fprintf(fp, "running");
      xfree(line);
    }
    if(fclose(fp) < 0)
      disorder_error(errno, "error closing %s connection", control_socket);
  }
}

/** @brief Drop the first packet
 *
 * Assumes that @ref lock is held. 
 */
static void drop_first_packet(void) {
  if(pheap_count(&packets)) {
    struct packet *const p = pheap_remove(&packets);
    nsamples -= p->nsamples;
    playrtp_free_packet(p);
    pthread_cond_broadcast(&cond);
  }
}

/** @brief Background thread adding packets to heap
 *
 * This just transfers packets from @ref received_packets to @ref packets.  It
 * is important that it holds @ref receive_lock for as little time as possible,
 * in order to minimize the interval between calls to read() in
 * listen_thread().
 */
static void *queue_thread(void attribute((unused)) *arg) {
  struct packet *p;

  for(;;) {
    /* Get the next packet */
    pthread_mutex_lock(&receive_lock);
    while(!received_packets) {
      pthread_cond_wait(&receive_cond, &receive_lock);
    }
    p = received_packets;
    received_packets = p->next;
    if(!received_packets)
      received_tail = &received_packets;
    --nreceived;
    pthread_mutex_unlock(&receive_lock);
    /* Add it to the heap */
    pthread_mutex_lock(&lock);
    pheap_insert(&packets, p);
    nsamples += p->nsamples;
    pthread_cond_broadcast(&cond);
    pthread_mutex_unlock(&lock);
  }
#if HAVE_STUPID_GCC44
  return NULL;
#endif
}

/** @brief Background thread collecting samples
 *
 * This function collects samples, perhaps converts them to the target format,
 * and adds them to the packet list.
 *
 * It is crucial that the gap between successive calls to read() is as small as
 * possible: otherwise packets will be dropped.
 *
 * We use a binary heap to ensure that the unavoidable effort is at worst
 * logarithmic in the total number of packets - in fact if packets are mostly
 * received in order then we will largely do constant work per packet since the
 * newest packet will always be last.
 *
 * Of more concern is that we must acquire the lock on the heap to add a packet
 * to it.  If this proves a problem in practice then the answer would be
 * (probably doubly) linked list with new packets added the end and a second
 * thread which reads packets off the list and adds them to the heap.
 *
 * We keep memory allocation (mostly) very fast by keeping pre-allocated
 * packets around; see @ref playrtp_new_packet().
 */
static void *listen_thread(void attribute((unused)) *arg) {
  struct packet *p = 0;
  int n;
  struct rtp_header header;
  uint16_t seq;
  uint32_t timestamp;
  struct iovec iov[2];

  for(;;) {
    if(!p)
      p = playrtp_new_packet();
    iov[0].iov_base = &header;
    iov[0].iov_len = sizeof header;
    iov[1].iov_base = p->samples_raw;
    iov[1].iov_len = sizeof p->samples_raw / sizeof *p->samples_raw;
    n = readv(rtpfd, iov, 2);
    if(n < 0) {
      switch(errno) {
      case EINTR:
        continue;
      default:
        disorder_fatal(errno, "error reading from socket");
      }
    }
    /* Ignore too-short packets */
    if((size_t)n <= sizeof (struct rtp_header)) {
      disorder_info("ignored a short packet");
      continue;
    }
    timestamp = htonl(header.timestamp);
    seq = htons(header.seq);
    /* Ignore packets in the past */
    if(active && lt(timestamp, next_timestamp)) {
      disorder_info("dropping old packet, timestamp=%"PRIx32" < %"PRIx32,
           timestamp, next_timestamp);
      continue;
    }
    /* Ignore packets with the extension bit set. */
    if(header.vpxcc & 0x10)
      continue;
    p->next = 0;
    p->flags = 0;
    p->timestamp = timestamp;
    /* Convert to target format */
    if(header.mpt & 0x80)
      p->flags |= IDLE;
    switch(header.mpt & 0x7F) {
    case 10:                            /* L16 */
      p->nsamples = (n - sizeof header) / sizeof(uint16_t);
      break;
      /* TODO support other RFC3551 media types (when the speaker does) */
    default:
      disorder_fatal(0, "unsupported RTP payload type %d", header.mpt & 0x7F);
    }
    /* See if packet is silent */
    const uint16_t *s = p->samples_raw;
    n = p->nsamples;
    for(; n > 0; --n)
      if(*s++)
        break;
    if(!n)
      p->flags |= SILENT;
    if(logfp)
      fprintf(logfp, "sequence %u timestamp %"PRIx32" length %"PRIx32" end %"PRIx32"\n",
              seq, timestamp, p->nsamples, timestamp + p->nsamples);
    /* Stop reading if we've reached the maximum.
     *
     * This is rather unsatisfactory: it means that if packets get heavily
     * out of order then we guarantee dropouts.  But for now... */
    if(nsamples >= maxbuffer) {
      pthread_mutex_lock(&lock);
      while(nsamples >= maxbuffer) {
        pthread_cond_wait(&cond, &lock);
      }
      pthread_mutex_unlock(&lock);
    }
    /* Add the packet to the receive queue */
    pthread_mutex_lock(&receive_lock);
    *received_tail = p;
    received_tail = &p->next;
    ++nreceived;
    pthread_cond_signal(&receive_cond);
    pthread_mutex_unlock(&receive_lock);
    /* We'll need a new packet */
    p = 0;
  }
}

/** @brief Wait until the buffer is adequately full
 *
 * Must be called with @ref lock held.
 */
void playrtp_fill_buffer(void) {
  /* Discard current buffer contents */
  while(nsamples) {
    //fprintf(stderr, "%8u/%u (%u) DROPPING\n", nsamples, maxbuffer, minbuffer);
    drop_first_packet();
  }
  disorder_info("Buffering...");
  /* Wait until there's at least minbuffer samples available */
  while(nsamples < minbuffer) {
    //fprintf(stderr, "%8u/%u (%u) FILLING\n", nsamples, maxbuffer, minbuffer);
    pthread_cond_wait(&cond, &lock);
  }
  /* Start from whatever is earliest */
  next_timestamp = pheap_first(&packets)->timestamp;
  active = 1;
}

/** @brief Find next packet
 * @return Packet to play or NULL if none found
 *
 * The return packet is merely guaranteed not to be in the past: it might be
 * the first packet in the future rather than one that is actually suitable to
 * play.
 *
 * Must be called with @ref lock held.
 */
struct packet *playrtp_next_packet(void) {
  while(pheap_count(&packets)) {
    struct packet *const p = pheap_first(&packets);
    if(le(p->timestamp + p->nsamples, next_timestamp)) {
      /* This packet is in the past.  Drop it and try another one. */
      drop_first_packet();
    } else
      /* This packet is NOT in the past.  (It might be in the future
       * however.) */
      return p;
  }
  return 0;
}

/* display usage message and terminate */
static void help(void) {
  xprintf("Usage:\n"
	  "  disorder-playrtp [OPTIONS] [[ADDRESS] PORT]\n"
	  "Options:\n"
          "  --device, -D DEVICE     Output device\n"
          "  --min, -m FRAMES        Buffer low water mark\n"
          "  --max, -x FRAMES        Buffer maximum size\n"
          "  --rcvbuf, -R BYTES      Socket receive buffer size\n"
          "  --config, -C PATH       Set configuration file\n"
#if HAVE_ALSA_ASOUNDLIB_H
          "  --alsa, -a              Use ALSA to play audio\n"
#endif
#if HAVE_SYS_SOUNDCARD_H || EMPEG_HOST
          "  --oss, -o               Use OSS to play audio\n"
#endif
#if HAVE_COREAUDIO_AUDIOHARDWARE_H
          "  --core-audio, -c        Use Core Audio to play audio\n"
#endif
          "  --command, -e COMMAND   Pipe audio to command.\n"
          "  --pause-mode, -P silence  For -e: pauses send silence (default)\n"
          "  --pause-mode, -P suspend  For -e: pauses suspend writes\n"
	  "  --help, -h              Display usage message\n"
	  "  --version, -V           Display version number\n"
          );
  xfclose(stdout);
  exit(0);
}

static size_t playrtp_callback(void *buffer,
                               size_t max_samples,
                               void attribute((unused)) *userdata) {
  size_t samples;
  int silent = 0;

  pthread_mutex_lock(&lock);
  /* Get the next packet, junking any that are now in the past */
  const struct packet *p = playrtp_next_packet();
  if(p && contains(p, next_timestamp)) {
    /* This packet is ready to play; the desired next timestamp points
     * somewhere into it. */

    /* Timestamp of end of packet */
    const uint32_t packet_end = p->timestamp + p->nsamples;

    /* Offset of desired next timestamp into current packet */
    const uint32_t offset = next_timestamp - p->timestamp;

    /* Pointer to audio data */
    const uint16_t *ptr = (void *)(p->samples_raw + offset);

    /* Compute number of samples left in packet, limited to output buffer
     * size */
    samples = packet_end - next_timestamp;
    if(samples > max_samples)
      samples = max_samples;

    /* Copy into buffer, converting to native endianness */
    size_t i = samples;
    int16_t *bufptr = buffer;
    while(i > 0) {
      *bufptr++ = (int16_t)ntohs(*ptr++);
      --i;
    }
    silent = !!(p->flags & SILENT);
  } else {
    /* There is no suitable packet.  We introduce 0s up to the next packet, or
     * to fill the buffer if there's no next packet or that's too many.  The
     * comparison with max_samples deals with the otherwise troubling overflow
     * case. */
    samples = p ? p->timestamp - next_timestamp : max_samples;
    if(samples > max_samples)
      samples = max_samples;
    //info("infill by %zu", samples);
    memset(buffer, 0, samples * uaudio_sample_size);
    silent = 1;
  }
  /* Debug dump */
  if(dump_buffer) {
    for(size_t i = 0; i < samples; ++i) {
      dump_buffer[dump_index++] = ((int16_t *)buffer)[i];
      dump_index %= dump_size;
    }
  }
  /* Advance timestamp */
  next_timestamp += samples;
  /* If we're getting behind then try to drop just silent packets
   *
   * In theory this shouldn't be necessary.  The server is supposed to send
   * packets at the right rate and compares the number of samples sent with the
   * time in order to ensure this.
   *
   * However, various things could throw this off:
   *
   * - the server's clock could advance at the wrong rate.  This would cause it
   *   to mis-estimate the right number of samples to have sent and
   *   inappropriately throttle or speed up.
   *
   * - playback could happen at the wrong rate.  If the playback host's sound
   *   card has a slightly incorrect clock then eventually it will get out
   *   of step.
   *
   * So if we play back slightly slower than the server sends for either of
   * these reasons then eventually our buffer, and the socket's buffer, will
   * fill, and the kernel will start dropping packets.  The result is audible
   * and not very nice.
   *
   * Therefore if we're getting behind, we pre-emptively drop silent packets,
   * since a change in the duration of a silence is less noticeable than a
   * dropped packet from the middle of continuous music.
   *
   * (If things go wrong the other way then eventually we run out of packets to
   * play and are forced to play silence.  This doesn't seem to happen in
   * practice but if it does then in the same way we can artificially extend
   * silent packets to compensate.)
   *
   * Dropped packets are always logged; use 'disorder-playrtp --monitor' to
   * track how close to target buffer occupancy we are on a once-a-minute
   * basis.
   */
  if(nsamples > minbuffer && silent) {
    disorder_info("dropping %zu samples (%"PRIu32" > %"PRIu32")",
                  samples, nsamples, minbuffer);
    samples = 0;
  }
  /* Junk obsolete packets */
  playrtp_next_packet();
  pthread_mutex_unlock(&lock);
  return samples;
}

int main(int argc, char **argv) {
  int n, err;
  struct addrinfo *res;
  struct stringlist sl;
  char *sockname;
  int rcvbuf, target_rcvbuf = 0;
  socklen_t len;
  struct ip_mreq mreq;
  struct ipv6_mreq mreq6;
  disorder_client *c;
  char *address, *port;
  int is_multicast;
  union any_sockaddr {
    struct sockaddr sa;
    struct sockaddr_in in;
    struct sockaddr_in6 in6;
  };
  union any_sockaddr mgroup;
  const char *dumpfile = 0;
  pthread_t ltid;
  int monitor = 0;
  static const int one = 1;

  static const struct addrinfo prefs = {
    .ai_flags = AI_PASSIVE,
    .ai_family = PF_INET,
    .ai_socktype = SOCK_DGRAM,
    .ai_protocol = IPPROTO_UDP
  };

  /* Timing information is often important to debugging playrtp, so we include
   * timestamps in the logs */
  logdate = 1;
  mem_init();
  if(!setlocale(LC_CTYPE, "")) disorder_fatal(errno, "error calling setlocale");
  backend = uaudio_apis[0];
  while((n = getopt_long(argc, argv, "hVdD:m:x:L:R:aocC:re:P:M", options, 0)) >= 0) {
    switch(n) {
    case 'h': help();
    case 'V': version("disorder-playrtp");
    case 'd': debugging = 1; break;
    case 'D': uaudio_set("device", optarg); break;
    case 'm': minbuffer = 2 * atol(optarg); break;
    case 'x': maxbuffer = 2 * atol(optarg); break;
    case 'L': logfp = fopen(optarg, "w"); break;
    case 'R': target_rcvbuf = atoi(optarg); break;
#if HAVE_ALSA_ASOUNDLIB_H
    case 'a': backend = &uaudio_alsa; break;
#endif
#if HAVE_SYS_SOUNDCARD_H || EMPEG_HOST
    case 'o': backend = &uaudio_oss; break;
#endif
#if HAVE_COREAUDIO_AUDIOHARDWARE_H      
    case 'c': backend = &uaudio_coreaudio; break;
#endif
    case 'C': configfile = optarg; break;
    case 's': control_socket = optarg; break;
    case 'r': dumpfile = optarg; break;
    case 'e': backend = &uaudio_command; uaudio_set("command", optarg); break;
    case 'P': uaudio_set("pause-mode", optarg); break;
    case 'M': monitor = 1; break;
    default: disorder_fatal(0, "invalid option");
    }
  }
  if(config_read(0, NULL)) disorder_fatal(0, "cannot read configuration");
  if(backend == &uaudio_rtp) {
    /* This means that you have NO local sound output.  This can happen if you
     * use a non-Apple GCC on a Mac (because it doesn't know how to compile
     * CoreAudio/AudioHardware.h). */
    disorder_fatal(0, "cannot play RTP through RTP");
  }
  if(!maxbuffer)
    maxbuffer = 2 * minbuffer;
  argc -= optind;
  argv += optind;
  switch(argc) {
  case 0:
    /* Get configuration from server */
    if(!(c = disorder_new(1))) exit(EXIT_FAILURE);
    if(disorder_connect(c)) exit(EXIT_FAILURE);
    if(disorder_rtp_address(c, &address, &port)) exit(EXIT_FAILURE);
    sl.n = 2;
    sl.s = xcalloc(2, sizeof *sl.s);
    sl.s[0] = address;
    sl.s[1] = port;
    break;
  case 1:
  case 2:
    /* Use command-line ADDRESS+PORT or just PORT */
    sl.n = argc;
    sl.s = argv;
    break;
  default:
    disorder_fatal(0, "usage: disorder-playrtp [OPTIONS] [[ADDRESS] PORT]");
  }
  /* Look up address and port */
  if(!(res = get_address(&sl, &prefs, &sockname)))
    exit(1);
  /* Create the socket */
  if((rtpfd = socket(res->ai_family,
                     res->ai_socktype,
                     res->ai_protocol)) < 0)
    disorder_fatal(errno, "error creating socket");
  /* Allow multiple listeners */
  xsetsockopt(rtpfd, SOL_SOCKET, SO_REUSEADDR, &one, sizeof one);
  is_multicast = multicast(res->ai_addr);
  /* The multicast and unicast/broadcast cases are different enough that they
   * are totally split.  Trying to find commonality between them causes more
   * trouble that it's worth. */
  if(is_multicast) {
    /* Stash the multicast group address */
    memcpy(&mgroup, res->ai_addr, res->ai_addrlen);
    switch(res->ai_addr->sa_family) {
    case AF_INET:
      mgroup.in.sin_port = 0;
      break;
    case AF_INET6:
      mgroup.in6.sin6_port = 0;
      break;
    default:
      disorder_fatal(0, "unsupported address family %d",
                     (int)res->ai_addr->sa_family);
    }
    /* Bind to to the multicast group address */
    if(bind(rtpfd, res->ai_addr, res->ai_addrlen) < 0)
      disorder_fatal(errno, "error binding socket to %s",
                     format_sockaddr(res->ai_addr));
    /* Add multicast group membership */
    switch(mgroup.sa.sa_family) {
    case PF_INET:
      mreq.imr_multiaddr = mgroup.in.sin_addr;
      mreq.imr_interface.s_addr = 0;      /* use primary interface */
      if(setsockopt(rtpfd, IPPROTO_IP, IP_ADD_MEMBERSHIP,
                    &mreq, sizeof mreq) < 0)
        disorder_fatal(errno, "error calling setsockopt IP_ADD_MEMBERSHIP");
      break;
    case PF_INET6:
      mreq6.ipv6mr_multiaddr = mgroup.in6.sin6_addr;
      memset(&mreq6.ipv6mr_interface, 0, sizeof mreq6.ipv6mr_interface);
      if(setsockopt(rtpfd, IPPROTO_IPV6, IPV6_JOIN_GROUP,
                    &mreq6, sizeof mreq6) < 0)
        disorder_fatal(errno, "error calling setsockopt IPV6_JOIN_GROUP");
      break;
    default:
      disorder_fatal(0, "unsupported address family %d", res->ai_family);
    }
    /* Report what we did */
    disorder_info("listening on %s multicast group %s",
                  format_sockaddr(res->ai_addr), format_sockaddr(&mgroup.sa));
  } else {
    /* Bind to 0/port */
    switch(res->ai_addr->sa_family) {
    case AF_INET: {
      struct sockaddr_in *in = (struct sockaddr_in *)res->ai_addr;
      
      memset(&in->sin_addr, 0, sizeof (struct in_addr));
      break;
    }
    case AF_INET6: {
      struct sockaddr_in6 *in6 = (struct sockaddr_in6 *)res->ai_addr;
      
      memset(&in6->sin6_addr, 0, sizeof (struct in6_addr));
      break;
    }
    default:
      disorder_fatal(0, "unsupported family %d", (int)res->ai_addr->sa_family);
    }
    if(bind(rtpfd, res->ai_addr, res->ai_addrlen) < 0)
      disorder_fatal(errno, "error binding socket to %s",
                     format_sockaddr(res->ai_addr));
    /* Report what we did */
    disorder_info("listening on %s", format_sockaddr(res->ai_addr));
  }
  len = sizeof rcvbuf;
  if(getsockopt(rtpfd, SOL_SOCKET, SO_RCVBUF, &rcvbuf, &len) < 0)
    disorder_fatal(errno, "error calling getsockopt SO_RCVBUF");
  if(target_rcvbuf > rcvbuf) {
    if(setsockopt(rtpfd, SOL_SOCKET, SO_RCVBUF,
                  &target_rcvbuf, sizeof target_rcvbuf) < 0)
      disorder_error(errno, "error calling setsockopt SO_RCVBUF %d", 
                     target_rcvbuf);
      /* We try to carry on anyway */
    else
      disorder_info("changed socket receive buffer from %d to %d",
                    rcvbuf, target_rcvbuf);
  } else
    disorder_info("default socket receive buffer %d", rcvbuf);
  //info("minbuffer %u maxbuffer %u", minbuffer, maxbuffer);
  if(logfp)
    disorder_info("WARNING: -L option can impact performance");
  if(control_socket) {
    pthread_t tid;

    if((err = pthread_create(&tid, 0, control_thread, 0)))
      disorder_fatal(err, "pthread_create control_thread");
  }
  if(dumpfile) {
    int fd;
    unsigned char buffer[65536];
    size_t written;

    if((fd = open(dumpfile, O_RDWR|O_TRUNC|O_CREAT, 0666)) < 0)
      disorder_fatal(errno, "opening %s", dumpfile);
    /* Fill with 0s to a suitable size */
    memset(buffer, 0, sizeof buffer);
    for(written = 0; written < dump_size * sizeof(int16_t);
        written += sizeof buffer) {
      if(write(fd, buffer, sizeof buffer) < 0)
        disorder_fatal(errno, "clearing %s", dumpfile);
    }
    /* Map the buffer into memory for convenience */
    dump_buffer = mmap(0, dump_size * sizeof(int16_t), PROT_READ|PROT_WRITE,
                       MAP_SHARED, fd, 0);
    if(dump_buffer == (void *)-1)
      disorder_fatal(errno, "mapping %s", dumpfile);
    disorder_info("dumping to %s", dumpfile);
  }
  /* Set up output.  Currently we only support L16 so there's no harm setting
   * the format before we know what it is! */
  uaudio_set_format(44100/*Hz*/, 2/*channels*/,
                    16/*bits/channel*/, 1/*signed*/);
  backend->start(playrtp_callback, NULL);
  /* We receive and convert audio data in a background thread */
  if((err = pthread_create(&ltid, 0, listen_thread, 0)))
    disorder_fatal(err, "pthread_create listen_thread");
  /* We have a second thread to add received packets to the queue */
  if((err = pthread_create(&ltid, 0, queue_thread, 0)))
    disorder_fatal(err, "pthread_create queue_thread");
  pthread_mutex_lock(&lock);
  time_t lastlog = 0;
  for(;;) {
    /* Wait for the buffer to fill up a bit */
    playrtp_fill_buffer();
    /* Start playing now */
    disorder_info("Playing...");
    next_timestamp = pheap_first(&packets)->timestamp;
    active = 1;
    pthread_mutex_unlock(&lock);
    backend->activate();
    pthread_mutex_lock(&lock);
    /* Wait until the buffer empties out
     *
     * If there's a packet that we can play right now then we definitely
     * continue.
     *
     * Also if there's at least minbuffer samples we carry on regardless and
     * insert silence.  The assumption is there's been a pause but more data
     * is now available.
     */
    while(nsamples >= minbuffer
	  || (nsamples > 0
	      && contains(pheap_first(&packets), next_timestamp))) {
      if(monitor) {
        time_t now = xtime(0);

        if(now >= lastlog + 60) {
          int offset = nsamples - minbuffer;
          double offtime = (double)offset / (uaudio_rate * uaudio_channels);
          disorder_info("%+d samples off (%d.%02ds, %d bytes)",
                        offset,
                        (int)fabs(offtime) * (offtime < 0 ? -1 : 1),
                        (int)(fabs(offtime) * 100) % 100,
                        offset * uaudio_bits / CHAR_BIT);
          lastlog = now;
        }
      }
      //fprintf(stderr, "%8u/%u (%u) PLAYING\n", nsamples, maxbuffer, minbuffer);
      pthread_cond_wait(&cond, &lock);
    }
#if 0
    if(nsamples) {
      struct packet *p = pheap_first(&packets);
      fprintf(stderr, "nsamples=%u (%u) next_timestamp=%"PRIx32", first packet is [%"PRIx32",%"PRIx32")\n",
              nsamples, minbuffer, next_timestamp,p->timestamp,p->timestamp+p->nsamples);
    }
#endif
    /* Stop playing for a bit until the buffer re-fills */
    pthread_mutex_unlock(&lock);
    backend->deactivate();
    pthread_mutex_lock(&lock);
    active = 0;
    /* Go back round */
  }
  return 0;
}

/*
Local Variables:
c-basic-offset:2
comment-column:40
fill-column:79
indent-tabs-mode:nil
End:
*/