postmaster.c

关系型数据库 Postgresql 6.5.2

		fprintf(stderr, "%s: cannot create UNIX stream port\n",
				progname);
		exit(1);
	}
#endif
	/* set up shared memory and semaphores */
	EnableMemoryContext(TRUE);
	reset_shared(PostPortName);

	/*
	 * Initialize the list of active backends.	This list is only used for
	 * garbage collecting the backend processes.
	 */
	BackendList = DLNewList();
	PortList = DLNewList();

	if (silentflag)
		pmdaemonize();

	/*
	 * Set up signal handlers for the postmaster process.
	 */

	pqsignal(SIGHUP, pmdie);	/* send SIGHUP, don't die */
	pqsignal(SIGINT, pmdie);	/* die */
	pqsignal(SIGQUIT, pmdie);	/* send SIGTERM and die */
	pqsignal(SIGTERM, pmdie);	/* send SIGTERM,SIGKILL and die */
	pqsignal(SIGPIPE, SIG_IGN); /* ignored */
	pqsignal(SIGUSR1, pmdie);	/* send SIGUSR1 and die */
	pqsignal(SIGUSR2, pmdie);	/* send SIGUSR2, don't die */
	pqsignal(SIGCHLD, reaper);	/* handle child termination */
	pqsignal(SIGTTIN, SIG_IGN); /* ignored */
	pqsignal(SIGTTOU, SIG_IGN); /* ignored */
	pqsignal(SIGWINCH, dumpstatus);		/* dump port status */

	status = ServerLoop();

	ExitPostmaster(status != STATUS_OK);
	return 0;					/* not reached */
}

static void
pmdaemonize(void)
{
	int			i;

	if (fork())
		_exit(0);
/* GH: If there's no setsid(), we hopefully don't need silent mode.
 * Until there's a better solution.
 */
#ifdef HAVE_SETSID
	if (setsid() < 0)
	{
		fprintf(stderr, "%s: ", progname);
		perror("cannot disassociate from controlling TTY");
		exit(1);
	}
#endif
#ifndef __CYGWIN32__
	i = open(NULL_DEV, O_RDWR);
#else
	i = open(NULL_DEV, O_RDWR | O_BINARY);
#endif
	dup2(i, 0);
	dup2(i, 1);
	dup2(i, 2);
	close(i);
}

static void
usage(const char *progname)
{
	fprintf(stderr, "usage: %s [options]\n", progname);
#ifdef USE_ASSERT_CHECKING
	fprintf(stderr, "\t-A [1|0]\tenable/disable runtime assert checking\n");
#endif
	fprintf(stderr, "\t-B nbufs\tset number of shared buffers\n");
	fprintf(stderr, "\t-D datadir\tset data directory\n");
	fprintf(stderr, "\t-S \t\tsilent mode (disassociate from tty)\n");
	fprintf(stderr, "\t-a system\tuse this authentication system\n");
	fprintf(stderr, "\t-b backend\tuse a specific backend server executable\n");
	fprintf(stderr, "\t-d [1|2|3]\tset debugging level\n");
	fprintf(stderr, "\t-i \t\tlisten on TCP/IP sockets as well as Unix domain socket\n");
	fprintf(stderr, "\t-N nprocs\tset max number of backends (1..%d, default %d)\n",
			MAXBACKENDS, DEF_MAXBACKENDS);
	fprintf(stderr, "\t-n \t\tdon't reinitialize shared memory after abnormal exit\n");
	fprintf(stderr, "\t-o option\tpass 'option' to each backend servers\n");
	fprintf(stderr, "\t-p port\tspecify port for postmaster to listen on\n");
	fprintf(stderr, "\t-s \t\tsend SIGSTOP to all backend servers if one dies\n");
	exit(1);
}

static int
ServerLoop(void)
{
	fd_set		readmask,
				writemask;
	int			nSockets;
	Dlelem	   *curr;
	struct timeval now,
				later;
	struct timezone tz;

	gettimeofday(&now, &tz);

	nSockets = initMasks(&readmask, &writemask);

#ifdef HAVE_SIGPROCMASK
	sigprocmask(0, NULL, &oldsigmask);
	sigemptyset(&newsigmask);
	sigaddset(&newsigmask, SIGCHLD);
#endif

	for (;;)
	{
		Port	   *port;
		fd_set		rmask,
					wmask;

#ifdef HAVE_SIGPROCMASK
		sigprocmask(SIG_SETMASK, &oldsigmask, 0);
#else
		sigsetmask(orgsigmask);
#endif

		memmove((char *) &rmask, (char *) &readmask, sizeof(fd_set));
		memmove((char *) &wmask, (char *) &writemask, sizeof(fd_set));
		if (select(nSockets, &rmask, &wmask, (fd_set *) NULL,
				   (struct timeval *) NULL) < 0)
		{
			if (errno == EINTR)
				continue;
			fprintf(stderr, "%s: ServerLoop: select failed: %s\n",
					progname, strerror(errno));
			return STATUS_ERROR;
		}

		/*
		 * Select a random seed at the time of first receiving a request.
		 */
		while (random_seed == 0)
		{
			gettimeofday(&later, &tz);

			/*
			 * We are not sure how much precision is in tv_usec, so we
			 * swap the nibbles of 'later' and XOR them with 'now'. On the
			 * off chance that the result is 0, we loop until it isn't.
			 */
			random_seed = now.tv_usec ^
				((later.tv_usec << 16) |
				 ((later.tv_usec >> 16) & 0xffff));
		}

		/*
		 * [TRH] To avoid race conditions, block SIGCHLD signals while we
		 * are handling the request. (both reaper() and ConnCreate()
		 * manipulate the BackEnd list, and reaper() calls free() which is
		 * usually non-reentrant.)
		 */
#ifdef HAVE_SIGPROCMASK
		sigprocmask(SIG_BLOCK, &newsigmask, &oldsigmask);
#else
		sigblock(sigmask(SIGCHLD));		/* XXX[TRH] portability */
#endif

		/* new connection pending on our well-known port's socket */

#ifndef __CYGWIN32__
		if (ServerSock_UNIX != INVALID_SOCK &&
			FD_ISSET(ServerSock_UNIX, &rmask) &&
			(port = ConnCreate(ServerSock_UNIX)) != NULL)
			PacketReceiveSetup(&port->pktInfo,
							   readStartupPacket,
							   (void *) port);
#endif

		if (ServerSock_INET != INVALID_SOCK &&
			FD_ISSET(ServerSock_INET, &rmask) &&
			(port = ConnCreate(ServerSock_INET)) != NULL)
			PacketReceiveSetup(&port->pktInfo,
							   readStartupPacket,
							   (void *) port);

		/* Build up new masks for select(). */

		nSockets = initMasks(&readmask, &writemask);

		curr = DLGetHead(PortList);

		while (curr)
		{
			Port	   *port = (Port *) DLE_VAL(curr);
			int			status = STATUS_OK;
			Dlelem	   *next;

			if (FD_ISSET(port->sock, &rmask))
			{
				if (DebugLvl > 1)
					fprintf(stderr, "%s: ServerLoop:\t\thandling reading %d\n",
							progname, port->sock);

				if (PacketReceiveFragment(&port->pktInfo, port->sock) != STATUS_OK)
					status = STATUS_ERROR;
			}

			if (FD_ISSET(port->sock, &wmask))
			{
				if (DebugLvl > 1)
					fprintf(stderr, "%s: ServerLoop:\t\thandling writing %d\n",
							progname, port->sock);

				if (PacketSendFragment(&port->pktInfo, port->sock) != STATUS_OK)
					status = STATUS_ERROR;
			}

			/* Get this before the connection might be closed. */

			next = DLGetSucc(curr);

			/*
			 * If there is no error and no outstanding data transfer going
			 * on, then the authentication handshake must be complete to
			 * the postmaster's satisfaction.  So, start the backend.
			 */

			if (status == STATUS_OK && port->pktInfo.state == Idle)
			{
				/* Can't start backend if max backend count is exceeded. */
				if (CountChildren() >= MaxBackends)
					PacketSendError(&port->pktInfo,
									"Sorry, too many clients already");
				else
				{

					/*
					 * If the backend start fails then keep the connection
					 * open to report it.  Otherwise, pretend there is an
					 * error to close the connection which will now be
					 * managed by the backend.
					 */
					if (BackendStartup(port) != STATUS_OK)
						PacketSendError(&port->pktInfo,
										"Backend startup failed");
					else
						status = STATUS_ERROR;
				}
			}

			/* Close the connection if required. */

			if (status != STATUS_OK)
			{
				StreamClose(port->sock);
				DLRemove(curr);
				free(port);
				DLFreeElem(curr);
			}
			else
			{
				/* Set the masks for this connection. */

				if (nSockets <= port->sock)
					nSockets = port->sock + 1;

				if (port->pktInfo.state == WritingPacket)
					FD_SET(port->sock, &writemask);
				else
					FD_SET(port->sock, &readmask);
			}

			curr = next;
		}
	}
}


/*
 * Initialise the read and write masks for select() for the well-known ports
 * we are listening on.  Return the number of sockets to listen on.
 */

static int
initMasks(fd_set *rmask, fd_set *wmask)
{
	int			nsocks = -1;

	FD_ZERO(rmask);
	FD_ZERO(wmask);

#ifndef __CYGWIN32__
	if (ServerSock_UNIX != INVALID_SOCK)
	{
		FD_SET(ServerSock_UNIX, rmask);

		if (ServerSock_UNIX > nsocks)
			nsocks = ServerSock_UNIX;
	}
#endif

	if (ServerSock_INET != INVALID_SOCK)
	{
		FD_SET(ServerSock_INET, rmask);

		if (ServerSock_INET > nsocks)
			nsocks = ServerSock_INET;
	}

	return nsocks + 1;
}


/*
 * Called when the startup packet has been read.
 */

static int
readStartupPacket(void *arg, PacketLen len, void *pkt)
{
	Port	   *port;
	StartupPacket *si;

	port = (Port *) arg;
	si = (StartupPacket *) pkt;

	/*
	 * The first field is either a protocol version number or a special
	 * request code.
	 */

	port->proto = ntohl(si->protoVersion);

	if (port->proto == CANCEL_REQUEST_CODE)
		return processCancelRequest(port, len, pkt);

	/* Could add additional special packet types here */

	/* Check we can handle the protocol the frontend is using. */

	if (PG_PROTOCOL_MAJOR(port->proto) < PG_PROTOCOL_MAJOR(PG_PROTOCOL_EARLIEST) ||
		PG_PROTOCOL_MAJOR(port->proto) > PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) ||
		(PG_PROTOCOL_MAJOR(port->proto) == PG_PROTOCOL_MAJOR(PG_PROTOCOL_LATEST) &&
		 PG_PROTOCOL_MINOR(port->proto) > PG_PROTOCOL_MINOR(PG_PROTOCOL_LATEST)))
	{
		PacketSendError(&port->pktInfo, "Unsupported frontend protocol.");
		return STATUS_OK;		/* don't close the connection yet */
	}

	/*
	 * Get the parameters from the startup packet as C strings.  The
	 * packet destination was cleared first so a short packet has zeros
	 * silently added and a long packet is silently truncated.
	 */

	StrNCpy(port->database, si->database, sizeof(port->database) - 1);
	StrNCpy(port->user, si->user, sizeof(port->user) - 1);
	StrNCpy(port->options, si->options, sizeof(port->options) - 1);
	StrNCpy(port->tty, si->tty, sizeof(port->tty) - 1);

	/* The database defaults to the user name. */

	if (port->database[0] == '\0')
		StrNCpy(port->database, si->user, sizeof(port->database) - 1);

	/* Check a user name was given. */

	if (port->user[0] == '\0')
	{
		PacketSendError(&port->pktInfo,
					"No Postgres username specified in startup packet.");
		return STATUS_OK;		/* don't close the connection yet */
	}

	/* Start the authentication itself. */

	be_recvauth(port);

	return STATUS_OK;			/* don't close the connection yet */
}


/*
 * The client has sent a cancel request packet, not a normal
 * start-a-new-backend packet.	Perform the necessary processing.
 * Note that in any case, we return STATUS_ERROR to close the
 * connection immediately.	Nothing is sent back to the client.
 */

static int
processCancelRequest(Port *port, PacketLen len, void *pkt)
{
	CancelRequestPacket *canc = (CancelRequestPacket *) pkt;
	int			backendPID;
	long		cancelAuthCode;
	Dlelem	   *curr;
	Backend    *bp;

	backendPID = (int) ntohl(canc->backendPID);
	cancelAuthCode = (long) ntohl(canc->cancelAuthCode);

	/* See if we have a matching backend */

	for (curr = DLGetHead(BackendList); curr; curr = DLGetSucc(curr))
	{
		bp = (Backend *) DLE_VAL(curr);
		if (bp->pid == backendPID)
		{
			if (bp->cancel_key == cancelAuthCode)
			{
				/* Found a match; signal that backend to cancel current op */
				if (DebugLvl)
					fprintf(stderr, "%s: processCancelRequest: sending SIGINT to process %d\n",
							progname, bp->pid);
				kill(bp->pid, SIGINT);
			}
			else
			{
				/* Right PID, wrong key: no way, Jose */
				if (DebugLvl)
					fprintf(stderr, "%s: processCancelRequest: bad key in cancel request for process %d\n",
							progname, bp->pid);
			}
			return STATUS_ERROR;
		}
	}

	/* No matching backend */
	if (DebugLvl)
		fprintf(stderr, "%s: processCancelRequest: bad PID in cancel request for process %d\n",
				progname, backendPID);

	return STATUS_ERROR;
}


/*
 * ConnCreate -- create a local connection data structure
 */
static Port *
ConnCreate(int serverFd)
{
	Port	   *port;


	if (!(port = (Port *) calloc(1, sizeof(Port))))
	{
		fprintf(stderr, "%s: ConnCreate: malloc failed\n",
				progname);
		ExitPostmaster(1);
	}

	if (StreamConnection(serverFd, port) != STATUS_OK)
	{
		StreamClose(port->sock);
		free(port);
		port = NULL;
	}
	else
	{
		DLAddHead(PortList, DLNewElem(port));
		RandomSalt(port->salt);
		port->pktInfo.state = Idle;
	}

	return port;
}

/*
 * reset_shared -- reset shared memory and semaphores
 */
static void
reset_shared(unsigned short port)
{
	ipc_key = port * 1000 + shmem_seq * 100;
	CreateSharedMemoryAndSemaphores(ipc_key, MaxBackends);
	ActiveBackends = FALSE;
	shmem_seq += 1;
	if (shmem_seq >= 10)
		shmem_seq -= 10;
}

/*
 * pmdie -- signal handler for cleaning up after a kill signal.
 */
static void
pmdie(SIGNAL_ARGS)
{
	int			i;

	TPRINTF(TRACE_VERBOSE, "pmdie %d", postgres_signal_arg);

	/*
	 * Kill self and/or children processes depending on signal number.
	 */
	switch (postgres_signal_arg)
	{
		case SIGHUP:
			/* Send SIGHUP to all children (update options flags) */
			SignalChildren(SIGHUP);
			/* Don't die */
			return;
		case SIGINT:
			/* Die without killing children */
			break;
		case SIGQUIT:
			/* Shutdown all children with SIGTERM */
			SignalChildren(SIGTERM);
			/* Don't die */
			return;
		case SIGTERM:
			/* Shutdown all children with SIGTERM and SIGKILL, then die */
			SignalChildren(SIGTERM);
			for (i = 0; i < 10; i++)
			{
				if (!DLGetHead(BackendList))
					break;
				sleep(1);
			}
			if (DLGetHead(BackendList))
				SignalChildren(SIGKILL);
			break;
		case SIGUSR1:
			/* Quick die all children with SIGUSR1 and die */
			SignalChildren(SIGUSR1);
			break;
		case SIGUSR2:
			/* Send SIGUSR2 to all children (AsyncNotifyHandler) */
			SignalChildren(SIGUSR2);
			/* Don't die */
			return;
	}

	/* exit postmaster */
	proc_exit(0);
}

/*
 * Reaper -- signal handler to cleanup after a backend (child) dies.
 */
static void
reaper(SIGNAL_ARGS)
{
/* GH: replace waitpid for !HAVE_WAITPID. Does this work ? */
#ifdef HAVE_WAITPID
	int			status;			/* backend exit status */

#else
	union wait	statusp;		/* backend exit status */

#endif
	int			pid;			/* process id of dead backend */

	if (DebugLvl)
		fprintf(stderr, "%s: reaping dead processes...\n",
				progname);
#ifdef HAVE_WAITPID
	while ((pid = waitpid(-1, &status, WNOHANG)) > 0)
	{
		CleanupProc(pid, status);
		pqsignal(SIGCHLD, reaper);
	}
#else
	while ((pid = wait3(&statusp, WNOHANG, NULL)) > 0)
	{
		CleanupProc(pid, statusp.w_status);
		pqsignal(SIGCHLD, reaper);
	}
#endif
}

/*
 * CleanupProc -- cleanup after terminated backend.
 *
 * Remove all local state associated with backend.
 *
 * Dillon's note: should log child's exit status in the system log.
 */
static void
CleanupProc(int pid,