Beispiel #1
0
static void
faulted (void)
{
  struct
    {
      mach_msg_header_t head;
      char buf[64];
    } request;
  mig_reply_header_t reply;
  extern int _hurdsig_fault_exc_server (mach_msg_header_t *,
					mach_msg_header_t *);

 /* Wait for the exception_raise message forwarded by the proc server.  */

 if (__mach_msg (&request.head, MACH_RCV_MSG, 0,
		  sizeof request, forward_sigexc,
		  MACH_MSG_TIMEOUT_NONE, MACH_PORT_NULL)
      != MACH_MSG_SUCCESS)
    __libc_fatal ("msg receive failed on signal thread exc\n");

  /* Run the exc demuxer which should call the server function above.
     That function returns 0 if the exception was expected.  */
  _hurdsig_fault_exc_server (&request.head, &reply.Head);
  if (reply.Head.msgh_remote_port != MACH_PORT_NULL)
    __mach_msg (&reply.Head, MACH_SEND_MSG, reply.Head.msgh_size,
		0, MACH_PORT_NULL, MACH_MSG_TIMEOUT_NONE, MACH_PORT_NULL);
  if (reply.RetCode == MIG_BAD_ID)
    __mach_msg_destroy (&request.head);

  if (reply.RetCode)
    __libc_fatal ("BUG: unexpected fault in signal thread\n");

  _hurdsig_fault_preemptor.signals = 0;
  longjmp (_hurdsig_fault_env, 1);
}
Beispiel #2
0
mach_msg_return_t
__mach_msg_receive (mach_msg_header_t *msg)
{
  return __mach_msg (msg, MACH_RCV_MSG,
		     0, msg->msgh_size, msg->msgh_local_port,
		     MACH_MSG_TIMEOUT_NONE, MACH_PORT_NULL);
}
Beispiel #3
0
mach_msg_return_t
__mach_msg_send	(mach_msg_header_t *msg)
{
  return __mach_msg (msg, MACH_SEND_MSG,
		     msg->msgh_size, 0, MACH_PORT_NULL,
		     MACH_MSG_TIMEOUT_NONE, MACH_PORT_NULL);
}
Beispiel #4
0
static void
timer_thread (void)
{
  while (1)
    {
      error_t err;
      /* The only message we ever expect to receive is the reply from the
         signal thread to a sig_post call we did.  We never examine the
	 contents.  */
      struct
	{
	  mach_msg_header_t header;
	  error_t return_code;
	} msg;

      /* Wait for a message on a port that noone sends to.  The purpose is
	 the receive timeout.  Notice interrupts so that if we are
	 thread_abort'd, we will loop around and fetch new values from
	 _hurd_itimerval.  */
      err = __mach_msg (&msg.header,
			MACH_RCV_MSG|MACH_RCV_TIMEOUT|MACH_RCV_INTERRUPT,
			0, 0, _hurd_itimer_port,
			_hurd_itimerval.it_value.tv_sec * 1000 +
			_hurd_itimerval.it_value.tv_usec / 1000,
			MACH_PORT_NULL);
      switch (err)
	{
	case MACH_RCV_TIMED_OUT:
	  /* We got the expected timeout.  Send a message to the signal
	     thread to tell it to post a SIGALRM signal.  We use
	     _hurd_itimer_port as the reply port just so we will block until
	     the signal thread has frobnicated things to reload the itimer or
	     has terminated this thread.  */
	  __msg_sig_post_request (_hurd_msgport,
				  _hurd_itimer_port,
				  MACH_MSG_TYPE_MAKE_SEND_ONCE,
				  SIGALRM, 0, __mach_task_self ());
	  break;

	case MACH_RCV_INTERRUPTED:
	  /* We were thread_abort'd.  This is to tell us that
	     _hurd_itimerval has changed and we need to reexamine it
	     and start waiting with the new timeout value.  */
	  break;

	case MACH_MSG_SUCCESS:
	  /* We got the reply message from the sig_post_request above.
	     Ignore it and reexamine the timer value.  */
	  __mach_msg_destroy (&msg.header); /* Just in case.  */
	  break;

	default:
	  /* Unexpected lossage.  Oh well, keep trying.  */
	  break;
	}
    }
}
Beispiel #5
0
/* Block THREAD.  */
void
__pthread_block (struct __pthread *thread)
{
  mach_msg_header_t msg;
  error_t err;

  err = __mach_msg (&msg, MACH_RCV_MSG, 0, sizeof msg,
		    thread->wakeupmsg.msgh_remote_port,
		    MACH_MSG_TIMEOUT_NONE, MACH_PORT_NULL);
  assert_perror (err);
}
Beispiel #6
0
/* Make the process sleep for SECONDS seconds, or until a signal arrives
   and is not ignored.  The function returns the number of seconds less
   than SECONDS which it actually slept (zero if it slept the full time).
   There is no return value to indicate error, but if `sleep' returns
   SECONDS, it probably didn't work.  */
unsigned int
__sleep (unsigned int seconds)
{
  time_t before, after;
  mach_port_t recv;

  recv = __mach_reply_port ();

  before = time ((time_t *) NULL);
  (void) __mach_msg (NULL, MACH_RCV_MSG|MACH_RCV_TIMEOUT|MACH_RCV_INTERRUPT,
		     0, 0, recv, seconds * 1000, MACH_PORT_NULL);
  after = time ((time_t *) NULL);
  __mach_port_destroy (__mach_task_self (), recv);

  return seconds - (after - before);
}
Beispiel #7
0
/* Sleep USECONDS microseconds, or until a previously set timer goes off.  */
int
usleep (useconds_t useconds)
{
  mach_port_t recv;
  struct timeval before, after;

  recv = __mach_reply_port ();

  if (__gettimeofday (&before, NULL) < 0)
    return -1;
  (void) __mach_msg (NULL, MACH_RCV_MSG|MACH_RCV_TIMEOUT|MACH_RCV_INTERRUPT,
		     0, 0, recv, (useconds + 999) / 1000, MACH_PORT_NULL);
  __mach_port_destroy (mach_task_self (), recv);
  if (__gettimeofday (&after, NULL) < 0)
    return -1;

  return 0;
}
Beispiel #8
0
int
__libc_nanosleep (const struct timespec *requested_time,
	     struct timespec *remaining)
{
  mach_port_t recv;
  struct timeval before, after;

  if (requested_time->tv_sec < 0
      || requested_time->tv_nsec < 0
      || requested_time->tv_nsec >= 1000000000)
    {
      errno = EINVAL;
      return -1;
    }

  const mach_msg_timeout_t ms
    = requested_time->tv_sec * 1000
    + (requested_time->tv_nsec + 999999) / 1000000;

  recv = __mach_reply_port ();

  if (remaining && __gettimeofday (&before, NULL) < 0)
    return -1;
  error_t err = __mach_msg (NULL, MACH_RCV_MSG|MACH_RCV_TIMEOUT|MACH_RCV_INTERRUPT,
			    0, 0, recv, ms, MACH_PORT_NULL);
  __mach_port_destroy (mach_task_self (), recv);
  if (err == EMACH_RCV_INTERRUPTED)
    {
      if (remaining && __gettimeofday (&after, NULL) >= 0)
	{
	  struct timeval req_time, elapsed, rem;
	  TIMESPEC_TO_TIMEVAL (&req_time, requested_time);
	  timersub (&after, &before, &elapsed);
	  timersub (&req_time, &elapsed, &rem);
	  TIMEVAL_TO_TIMESPEC (&rem, remaining);
	}

      errno = EINTR;
      return -1;
    }

  return 0;
}
Beispiel #9
0
/* Wake up any sigsuspend call that is blocking SS->thread.  SS must be
   locked.  */
static void
wake_sigsuspend (struct hurd_sigstate *ss)
{
  error_t err;
  mach_msg_header_t msg;

  if (ss->suspended == MACH_PORT_NULL)
    return;

  /* There is a sigsuspend waiting.  Tell it to wake up.  */
  msg.msgh_bits = MACH_MSGH_BITS (MACH_MSG_TYPE_MAKE_SEND, 0);
  msg.msgh_remote_port = ss->suspended;
  msg.msgh_local_port = MACH_PORT_NULL;
  /* These values do not matter.  */
  msg.msgh_id = 8675309; /* Jenny, Jenny.  */
  ss->suspended = MACH_PORT_NULL;
  err = __mach_msg (&msg, MACH_SEND_MSG, sizeof msg, 0,
      MACH_PORT_NULL, MACH_MSG_TIMEOUT_NONE,
      MACH_PORT_NULL);
  assert_perror (err);
}
Beispiel #10
0
/* This function must be very careful not to depend on Hurd threadvar
   variables.  We arrange that by using special stubs arranged for at the
   end of this file. */
static void
profile_waiter (void)
{
    mach_msg_header_t msg;
    mach_port_t timeout_reply_port;

    profil_reply_port = __mach_reply_port ();
    timeout_reply_port = __mach_reply_port ();

    while (1)
    {
        __spin_lock (&lock);

        fetch_samples ();

        __spin_unlock (&lock);

        __mach_msg (&msg, MACH_RCV_MSG|MACH_RCV_TIMEOUT, 0, sizeof msg,
                    timeout_reply_port, collector_timeout, MACH_PORT_NULL);
    }
}
Beispiel #11
0
/* Block THREAD.  */
error_t
__pthread_timedblock (struct __pthread *thread,
		      const struct timespec *abstime,
		      clockid_t clock_id)
{
  error_t err;
  mach_msg_header_t msg;
  mach_msg_timeout_t timeout;
  struct timeval now;

  /* We have an absolute time and now we have to convert it to a
     relative time.  Arg.  */

  err = gettimeofday(&now, NULL);
  assert (! err);

  if (now.tv_sec > abstime->tv_sec
      || (now.tv_sec == abstime->tv_sec
	  && now.tv_usec > ((abstime->tv_nsec + 999) / 1000)))
    return ETIMEDOUT;

  timeout = (abstime->tv_sec - now.tv_sec) * 1000;

  if (((abstime->tv_nsec + 999) / 1000) >= now.tv_usec)
    timeout += (((abstime->tv_nsec + 999) / 1000) - now.tv_usec + 999) / 1000;
  else
    /* Need to do a carry.  */
    timeout -= (now.tv_usec + 999) / 1000 -
      ((abstime->tv_nsec + 999999) / 1000000);

  err = __mach_msg (&msg, MACH_RCV_MSG | MACH_RCV_TIMEOUT, 0,
		    sizeof msg, thread->wakeupmsg.msgh_remote_port,
		    timeout, MACH_PORT_NULL);
  if (err == EMACH_RCV_TIMED_OUT)
    return ETIMEDOUT;

  assert_perror (err);
  return 0;
}
Beispiel #12
0
/* Check the first NFDS descriptors either in POLLFDS (if nonnnull) or in
   each of READFDS, WRITEFDS, EXCEPTFDS that is nonnull.  If TIMEOUT is not
   NULL, time out after waiting the interval specified therein.  Returns
   the number of ready descriptors, or -1 for errors.  */
int
_hurd_select (int nfds,
	      struct pollfd *pollfds,
	      fd_set *readfds, fd_set *writefds, fd_set *exceptfds,
	      const struct timespec *timeout, const sigset_t *sigmask)
{
  int i;
  mach_port_t portset;
  int got;
  error_t err;
  fd_set rfds, wfds, xfds;
  int firstfd, lastfd;
  mach_msg_timeout_t to = 0;
  struct
    {
      struct hurd_userlink ulink;
      struct hurd_fd *cell;
      mach_port_t io_port;
      int type;
      mach_port_t reply_port;
    } d[nfds];
  sigset_t oset;

  union typeword		/* Use this to avoid unkosher casts.  */
    {
      mach_msg_type_t type;
      uint32_t word;
    };
  assert (sizeof (union typeword) == sizeof (mach_msg_type_t));
  assert (sizeof (uint32_t) == sizeof (mach_msg_type_t));

  if (nfds < 0 || (pollfds == NULL && nfds > FD_SETSIZE))
    {
      errno = EINVAL;
      return -1;
    }

  if (timeout != NULL)
    {
      if (timeout->tv_sec < 0 || timeout->tv_nsec < 0)
	{
	  errno = EINVAL;
	  return -1;
	}

      to = (timeout->tv_sec * 1000 +
            (timeout->tv_nsec + 999999) / 1000000);
    }

  if (sigmask && __sigprocmask (SIG_SETMASK, sigmask, &oset))
    return -1;

  if (pollfds)
    {
      /* Collect interesting descriptors from the user's `pollfd' array.
	 We do a first pass that reads the user's array before taking
	 any locks.  The second pass then only touches our own stack,
	 and gets the port references.  */

      for (i = 0; i < nfds; ++i)
	if (pollfds[i].fd >= 0)
	  {
	    int type = 0;
	    if (pollfds[i].events & POLLIN)
	      type |= SELECT_READ;
	    if (pollfds[i].events & POLLOUT)
	      type |= SELECT_WRITE;
	    if (pollfds[i].events & POLLPRI)
	      type |= SELECT_URG;

	    d[i].io_port = pollfds[i].fd;
	    d[i].type = type;
	  }
	else
	  d[i].type = 0;

      HURD_CRITICAL_BEGIN;
      __mutex_lock (&_hurd_dtable_lock);

      for (i = 0; i < nfds; ++i)
	if (d[i].type != 0)
	  {
	    const int fd = (int) d[i].io_port;

	    if (fd < _hurd_dtablesize)
	      {
		d[i].cell = _hurd_dtable[fd];
		d[i].io_port = _hurd_port_get (&d[i].cell->port, &d[i].ulink);
		if (d[i].io_port != MACH_PORT_NULL)
		  continue;
	      }

	    /* If one descriptor is bogus, we fail completely.  */
	    while (i-- > 0)
	      if (d[i].type != 0)
		_hurd_port_free (&d[i].cell->port,
				 &d[i].ulink, d[i].io_port);
	    break;
	  }

      __mutex_unlock (&_hurd_dtable_lock);
      HURD_CRITICAL_END;

      if (i < nfds)
	{
	  if (sigmask)
	    __sigprocmask (SIG_SETMASK, &oset, NULL);
	  errno = EBADF;
	  return -1;
	}

      lastfd = i - 1;
      firstfd = i == 0 ? lastfd : 0;
    }
  else
    {
      /* Collect interested descriptors from the user's fd_set arguments.
	 Use local copies so we can't crash from user bogosity.  */

      if (readfds == NULL)
	FD_ZERO (&rfds);
      else
	rfds = *readfds;
      if (writefds == NULL)
	FD_ZERO (&wfds);
      else
	wfds = *writefds;
      if (exceptfds == NULL)
	FD_ZERO (&xfds);
      else
	xfds = *exceptfds;

      HURD_CRITICAL_BEGIN;
      __mutex_lock (&_hurd_dtable_lock);

      if (nfds > _hurd_dtablesize)
	nfds = _hurd_dtablesize;

      /* Collect the ports for interesting FDs.  */
      firstfd = lastfd = -1;
      for (i = 0; i < nfds; ++i)
	{
	  int type = 0;
	  if (readfds != NULL && FD_ISSET (i, &rfds))
	    type |= SELECT_READ;
	  if (writefds != NULL && FD_ISSET (i, &wfds))
	    type |= SELECT_WRITE;
	  if (exceptfds != NULL && FD_ISSET (i, &xfds))
	    type |= SELECT_URG;
	  d[i].type = type;
	  if (type)
	    {
	      d[i].cell = _hurd_dtable[i];
	      d[i].io_port = _hurd_port_get (&d[i].cell->port, &d[i].ulink);
	      if (d[i].io_port == MACH_PORT_NULL)
		{
		  /* If one descriptor is bogus, we fail completely.  */
		  while (i-- > 0)
		    if (d[i].type != 0)
		      _hurd_port_free (&d[i].cell->port, &d[i].ulink,
				       d[i].io_port);
		  break;
		}
	      lastfd = i;
	      if (firstfd == -1)
		firstfd = i;
	    }
	}

      __mutex_unlock (&_hurd_dtable_lock);
      HURD_CRITICAL_END;

      if (i < nfds)
	{
	  if (sigmask)
	    __sigprocmask (SIG_SETMASK, &oset, NULL);
	  errno = EBADF;
	  return -1;
	}
    }


  err = 0;
  got = 0;

  /* Send them all io_select request messages.  */

  if (firstfd == -1)
    /* But not if there were no ports to deal with at all.
       We are just a pure timeout.  */
    portset = __mach_reply_port ();
  else
    {
      portset = MACH_PORT_NULL;

      for (i = firstfd; i <= lastfd; ++i)
	if (d[i].type)
	  {
	    int type = d[i].type;
	    d[i].reply_port = __mach_reply_port ();
	    err = __io_select (d[i].io_port, d[i].reply_port,
			       /* Poll only if there's a single descriptor.  */
			       (firstfd == lastfd) ? to : 0,
			       &type);
	    switch (err)
	      {
	      case MACH_RCV_TIMED_OUT:
		/* No immediate response.  This is normal.  */
		err = 0;
		if (firstfd == lastfd)
		  /* When there's a single descriptor, we don't need a
		     portset, so just pretend we have one, but really
		     use the single reply port.  */
		  portset = d[i].reply_port;
		else if (got == 0)
		  /* We've got multiple reply ports, so we need a port set to
		     multiplex them.  */
		  {
		    /* We will wait again for a reply later.  */
		    if (portset == MACH_PORT_NULL)
		      /* Create the portset to receive all the replies on.  */
		      err = __mach_port_allocate (__mach_task_self (),
						  MACH_PORT_RIGHT_PORT_SET,
						  &portset);
		    if (! err)
		      /* Put this reply port in the port set.  */
		      __mach_port_move_member (__mach_task_self (),
					       d[i].reply_port, portset);
		  }
		break;

	      default:
		/* No other error should happen.  Callers of select
		   don't expect to see errors, so we simulate
		   readiness of the erring object and the next call
		   hopefully will get the error again.  */
		type = SELECT_ALL;
		/* FALLTHROUGH */

	      case 0:
		/* We got an answer.  */
		if ((type & SELECT_ALL) == 0)
		  /* Bogus answer; treat like an error, as a fake positive.  */
		  type = SELECT_ALL;

		/* This port is already ready already.  */
		d[i].type &= type;
		d[i].type |= SELECT_RETURNED;
		++got;
		break;
	      }
	    _hurd_port_free (&d[i].cell->port, &d[i].ulink, d[i].io_port);
	  }
    }

  /* Now wait for reply messages.  */
  if (!err && got == 0)
    {
      /* Now wait for io_select_reply messages on PORT,
	 timing out as appropriate.  */

      union
	{
	  mach_msg_header_t head;
#ifdef MACH_MSG_TRAILER_MINIMUM_SIZE
	  struct
	    {
	      mach_msg_header_t head;
	      NDR_record_t ndr;
	      error_t err;
	    } error;
	  struct
	    {
	      mach_msg_header_t head;
	      NDR_record_t ndr;
	      error_t err;
	      int result;
	      mach_msg_trailer_t trailer;
	    } success;
#else
	  struct
	    {
	      mach_msg_header_t head;
	      union typeword err_type;
	      error_t err;
	    } error;
	  struct
	    {
	      mach_msg_header_t head;
	      union typeword err_type;
	      error_t err;
	      union typeword result_type;
	      int result;
	    } success;
#endif
	} msg;
      mach_msg_option_t options = (timeout == NULL ? 0 : MACH_RCV_TIMEOUT);
      error_t msgerr;
      while ((msgerr = __mach_msg (&msg.head,
				   MACH_RCV_MSG | MACH_RCV_INTERRUPT | options,
				   0, sizeof msg, portset, to,
				   MACH_PORT_NULL)) == MACH_MSG_SUCCESS)
	{
	  /* We got a message.  Decode it.  */
#define IO_SELECT_REPLY_MSGID (21012 + 100) /* XXX */
#ifdef MACH_MSG_TYPE_BIT
	  const union typeword inttype =
	  { type:
	    { MACH_MSG_TYPE_INTEGER_T, sizeof (integer_t) * 8, 1, 1, 0, 0 }
	  };
#endif
	  if (msg.head.msgh_id == IO_SELECT_REPLY_MSGID &&
	      msg.head.msgh_size >= sizeof msg.error &&
	      !(msg.head.msgh_bits & MACH_MSGH_BITS_COMPLEX) &&
#ifdef MACH_MSG_TYPE_BIT
	      msg.error.err_type.word == inttype.word
#endif
	      )
	    {
	      /* This is a properly formatted message so far.
		 See if it is a success or a failure.  */
	      if (msg.error.err == EINTR &&
		  msg.head.msgh_size == sizeof msg.error)
		{
		  /* EINTR response; poll for further responses
		     and then return quickly.  */
		  err = EINTR;
		  goto poll;
		}
	      if (msg.error.err ||
		  msg.head.msgh_size != sizeof msg.success ||
#ifdef MACH_MSG_TYPE_BIT
		  msg.success.result_type.word != inttype.word ||
#endif
		  (msg.success.result & SELECT_ALL) == 0)
		{
		  /* Error or bogus reply.  Simulate readiness.  */
		  __mach_msg_destroy (&msg.head);
		  msg.success.result = SELECT_ALL;
		}

	      /* Look up the respondent's reply port and record its
		 readiness.  */
	      {
		int had = got;
		if (firstfd != -1)
		  for (i = firstfd; i <= lastfd; ++i)
		    if (d[i].type
			&& d[i].reply_port == msg.head.msgh_local_port)
		      {
			d[i].type &= msg.success.result;
			d[i].type |= SELECT_RETURNED;
			++got;
		      }
		assert (got > had);
	      }
	    }

	  if (msg.head.msgh_remote_port != MACH_PORT_NULL)
	    __mach_port_deallocate (__mach_task_self (),
				    msg.head.msgh_remote_port);

	  if (got)
	  poll:
	    {
	      /* Poll for another message.  */
	      to = 0;
	      options |= MACH_RCV_TIMEOUT;
	    }
	}

      if (msgerr == MACH_RCV_INTERRUPTED)
	/* Interruption on our side (e.g. signal reception).  */
	err = EINTR;

      if (got)
	/* At least one descriptor is known to be ready now, so we will
	   return success.  */
	err = 0;
    }
Beispiel #13
0
static void
abort_all_rpcs (int signo, struct machine_thread_all_state *state, int live)
{
  /* We can just loop over the sigstates.  Any thread doing something
     interruptible must have one.  We needn't bother locking because all
     other threads are stopped.  */

  struct hurd_sigstate *ss;
  size_t nthreads;
  mach_port_t *reply_ports;

  /* First loop over the sigstates to count them.
     We need to know how big a vector we will need for REPLY_PORTS.  */
  nthreads = 0;
  for (ss = _hurd_sigstates; ss != NULL; ss = ss->next)
    ++nthreads;

  reply_ports = alloca (nthreads * sizeof *reply_ports);

  nthreads = 0;
  for (ss = _hurd_sigstates; ss != NULL; ss = ss->next, ++nthreads)
    if (ss->thread == _hurd_msgport_thread)
      reply_ports[nthreads] = MACH_PORT_NULL;
    else
      {
	int state_changed;
	state->set = 0;		/* Reset scratch area.  */

	/* Abort any operation in progress with interrupt_operation.
	   Record the reply port the thread is waiting on.
	   We will wait for all the replies below.  */
	reply_ports[nthreads] = _hurdsig_abort_rpcs (ss, signo, 1,
						     state, &state_changed,
						     NULL);
	if (live)
	  {
	    if (reply_ports[nthreads] != MACH_PORT_NULL)
	      {
		/* We will wait for the reply to this RPC below, so the
		   thread must issue a new RPC rather than waiting for the
		   reply to the one it sent.  */
		state->basic.SYSRETURN = EINTR;
		state_changed = 1;
	      }
	    if (state_changed)
	      /* Aborting the RPC needed to change this thread's state,
		 and it might ever run again.  So write back its state.  */
	      __thread_set_state (ss->thread, MACHINE_THREAD_STATE_FLAVOR,
				  (natural_t *) &state->basic,
				  MACHINE_THREAD_STATE_COUNT);
	  }
      }

  /* Wait for replies from all the successfully interrupted RPCs.  */
  while (nthreads-- > 0)
    if (reply_ports[nthreads] != MACH_PORT_NULL)
      {
	error_t err;
	mach_msg_header_t head;
	err = __mach_msg (&head, MACH_RCV_MSG|MACH_RCV_TIMEOUT, 0, sizeof head,
			  reply_ports[nthreads],
			  _hurd_interrupted_rpc_timeout, MACH_PORT_NULL);
	switch (err)
	  {
	  case MACH_RCV_TIMED_OUT:
	  case MACH_RCV_TOO_LARGE:
	    break;

	  default:
	    assert_perror (err);
	  }
      }
}
Beispiel #14
0
mach_msg_return_t
__mach_msg_server_timeout (boolean_t (*demux) (mach_msg_header_t *request,
					       mach_msg_header_t *reply),
			   mach_msg_size_t max_size,
			   mach_port_t rcv_name,
			   mach_msg_option_t option,
			   mach_msg_timeout_t timeout)
{
  mig_reply_header_t *request, *reply;
  mach_msg_return_t mr;

  if (max_size == 0)
    {
#ifdef MACH_RCV_LARGE
      option |= MACH_RCV_LARGE;
      max_size = 2 * __vm_page_size; /* Generic.  Good? XXX */
#else
      max_size = 4 * __vm_page_size; /* XXX */
#endif
    }

  request = __alloca (max_size);
  reply = __alloca (max_size);

  while (1)
    {
    get_request:
      mr = __mach_msg (&request->Head, MACH_RCV_MSG|option,
		       0, max_size, rcv_name,
		       timeout, MACH_PORT_NULL);
      while (mr == MACH_MSG_SUCCESS)
	{
	  /* We have a request message.
	     Pass it to DEMUX for processing.  */

	  (void) (*demux) (&request->Head, &reply->Head);
	  assert (reply->Head.msgh_size <= max_size);

	  switch (reply->RetCode)
	    {
	    case KERN_SUCCESS:
	      /* Hunky dory.  */
	      break;

	    case MIG_NO_REPLY:
	      /* The server function wanted no reply sent.
		 Loop for another request.  */
	      goto get_request;

	    default:
	      /* Some error; destroy the request message to release any
		 port rights or VM it holds.  Don't destroy the reply port
		 right, so we can send an error message.  */
	      request->Head.msgh_remote_port = MACH_PORT_NULL;
	      __mach_msg_destroy (&request->Head);
	      break;
	    }

	  if (reply->Head.msgh_remote_port == MACH_PORT_NULL)
	    {
	      /* No reply port, so destroy the reply.  */
	      if (reply->Head.msgh_bits & MACH_MSGH_BITS_COMPLEX)
		__mach_msg_destroy (&reply->Head);
	      goto get_request;
	    }

	  /* Send the reply and the get next request.  */

	  {
	    /* Swap the request and reply buffers.  mach_msg will read the
	       reply message from the buffer we pass and write the new
	       request message to the same buffer.  */
	    void *tmp = request;
	    request = reply;
	    reply = tmp;
	  }

	  mr = __mach_msg (&request->Head,
			   MACH_SEND_MSG|MACH_RCV_MSG|option,
			   request->Head.msgh_size, max_size, rcv_name,
			   timeout, MACH_PORT_NULL);
	}

      /* A message error occurred.  */

      switch (mr)
	{
	case MACH_RCV_TOO_LARGE:
#ifdef MACH_RCV_LARGE
	  /* The request message is larger than MAX_SIZE, and has not
	     been dequeued.  The message header has the actual size of
	     the message.  We recurse here in hopes that the compiler
	     will optimize the tail-call and allocate some more stack
	     space instead of way too much.  */
	  return __mach_msg_server_timeout (demux, request->Head.msgh_size,
					    rcv_name, option, timeout);
#else
	  /* XXX the kernel has destroyed the msg */
	  break;
#endif

	case MACH_SEND_INVALID_DEST:
	  /* The reply can't be delivered, so destroy it.  This error
	     indicates only that the requester went away, so we
	     continue and get the next request.  */
	  __mach_msg_destroy (&request->Head);
	  break;

	default:
	  /* Some other form of lossage; return to caller.  */
	  return mr;
	}
    }
}