daemon.c

/* $OpenLDAP$ */
/* This work is part of OpenLDAP Software <http://www.openldap.org/>.
 *
 * Copyright 1998-2020 The OpenLDAP Foundation.
 * Portions Copyright 2007 by Howard Chu, Symas Corporation.
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted only as authorized by the OpenLDAP
 * Public License.
 *
 * A copy of this license is available in the file LICENSE in the
 * top-level directory of the distribution or, alternatively, at
 * <http://www.OpenLDAP.org/license.html>.
 */
/* Portions Copyright (c) 1995 Regents of the University of Michigan.
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms are permitted
 * provided that this notice is preserved and that due credit is given
 * to the University of Michigan at Ann Arbor. The name of the University
 * may not be used to endorse or promote products derived from this
 * software without specific prior written permission. This software
 * is provided ``as is'' without express or implied warranty.
 */

#include "portable.h"

#include <stdio.h>

#include <ac/ctype.h>
#include <ac/errno.h>
#include <ac/socket.h>
#include <ac/string.h>
#include <ac/time.h>
#include <ac/unistd.h>

#include "slap.h"
#include "ldap_pvt_thread.h"
#include "lutil.h"

#include "ldap_rq.h"

#if defined(HAVE_SYS_EPOLL_H) && defined(HAVE_EPOLL)
# include <sys/epoll.h>
#elif defined(SLAP_X_DEVPOLL) && defined(HAVE_SYS_DEVPOLL_H) && defined(HAVE_DEVPOLL)
# include <sys/types.h>
# include <sys/stat.h>
# include <fcntl.h>
# include <sys/devpoll.h>
#endif /* ! epoll && ! /dev/poll */

#ifdef HAVE_TCPD
int allow_severity = LOG_INFO;
int deny_severity = LOG_NOTICE;
#endif /* TCP Wrappers */

#ifdef LDAP_PF_LOCAL
# include <sys/stat.h>
/* this should go in <ldap.h> as soon as it is accepted */
# define LDAPI_MOD_URLEXT		"x-mod"
#endif /* LDAP_PF_LOCAL */

#ifdef LDAP_PF_INET6
int slap_inet4or6 = AF_UNSPEC;
#else /* ! INETv6 */
int slap_inet4or6 = AF_INET;
#endif /* ! INETv6 */

/* globals */
time_t starttime;
ber_socket_t dtblsize;
slap_ssf_t local_ssf = LDAP_PVT_SASL_LOCAL_SSF;
struct runqueue_s slapd_rq;

#ifndef SLAPD_MAX_DAEMON_THREADS
#define SLAPD_MAX_DAEMON_THREADS	16
#endif
int slapd_daemon_threads = 1;
int slapd_daemon_mask;

#ifdef LDAP_TCP_BUFFER
int slapd_tcp_rmem;
int slapd_tcp_wmem;
#endif /* LDAP_TCP_BUFFER */

Listener **slap_listeners = NULL;
static volatile sig_atomic_t listening = 1; /* 0 when slap_listeners closed */
static ldap_pvt_thread_t *listener_tid;

#ifndef SLAPD_LISTEN_BACKLOG
#define SLAPD_LISTEN_BACKLOG 1024
#endif /* ! SLAPD_LISTEN_BACKLOG */

#define	DAEMON_ID(fd)	(fd & slapd_daemon_mask)

static ber_socket_t wake_sds[SLAPD_MAX_DAEMON_THREADS][2];
static int emfile;

static time_t chk_writetime;

static volatile int waking;
#ifdef NO_THREADS
#define WAKE_LISTENER(l,w)	do { \
	if ((w) && ++waking < 5) { \
		tcp_write( SLAP_FD2SOCK(wake_sds[l][1]), "0", 1 ); \
	} \
} while (0)
#else /* ! NO_THREADS */
#define WAKE_LISTENER(l,w)	do { \
	if (w) { \
		tcp_write( SLAP_FD2SOCK(wake_sds[l][1]), "0", 1 ); \
	} \
} while (0)
#endif /* ! NO_THREADS */

volatile sig_atomic_t slapd_shutdown = 0;
volatile sig_atomic_t slapd_gentle_shutdown = 0;
volatile sig_atomic_t slapd_abrupt_shutdown = 0;

#ifdef HAVE_WINSOCK
ldap_pvt_thread_mutex_t slapd_ws_mutex;
SOCKET *slapd_ws_sockets;
#define	SD_READ 1
#define	SD_WRITE	2
#define	SD_ACTIVE	4
#define	SD_LISTENER	8
#endif

#ifdef HAVE_TCPD
static ldap_pvt_thread_mutex_t	sd_tcpd_mutex;
#endif /* TCP Wrappers */

typedef struct slap_daemon_st {
	ldap_pvt_thread_mutex_t	sd_mutex;

	ber_socket_t		sd_nactives;
	int			sd_nwriters;
	int			sd_nfds;

#if defined(HAVE_EPOLL)
	struct epoll_event	*sd_epolls;
	int			*sd_index;
	int			sd_epfd;
#elif defined(SLAP_X_DEVPOLL) && defined(HAVE_DEVPOLL)
	/* eXperimental */
	struct pollfd		*sd_pollfd;
	int			*sd_index;
	Listener		**sd_l;
	int			sd_dpfd;
#else /* ! epoll && ! /dev/poll */
#ifdef HAVE_WINSOCK
	char	*sd_flags;
	char	*sd_rflags;
#else /* ! HAVE_WINSOCK */
	fd_set			sd_actives;
	fd_set			sd_readers;
	fd_set			sd_writers;
#endif /* ! HAVE_WINSOCK */
#endif /* ! epoll && ! /dev/poll */
} slap_daemon_st;

static slap_daemon_st slap_daemon[SLAPD_MAX_DAEMON_THREADS];

/*
 * NOTE: naming convention for macros:
 *
 * - SLAP_SOCK_* and SLAP_EVENT_* for public interface that deals
 *   with file descriptors and events respectively
 *
 * - SLAP_<type>_* for private interface; type by now is one of
 *   EPOLL, DEVPOLL, SELECT
 *
 * private interface should not be used in the code.
 */
#if defined(HAVE_EPOLL)
/***************************************
 * Use epoll infrastructure - epoll(4) *
 ***************************************/
# define SLAP_EVENT_FNAME		"epoll"
# define SLAP_EVENTS_ARE_INDEXED	0
# define SLAP_EPOLL_SOCK_IX(t,s)		(slap_daemon[t].sd_index[(s)])
# define SLAP_EPOLL_SOCK_EP(t,s)		(slap_daemon[t].sd_epolls[SLAP_EPOLL_SOCK_IX(t,s)])
# define SLAP_EPOLL_SOCK_EV(t,s)		(SLAP_EPOLL_SOCK_EP(t,s).events)
# define SLAP_SOCK_IS_ACTIVE(t,s)		(SLAP_EPOLL_SOCK_IX(t,s) != -1)
# define SLAP_SOCK_NOT_ACTIVE(t,s)	(SLAP_EPOLL_SOCK_IX(t,s) == -1)
# define SLAP_EPOLL_SOCK_IS_SET(t,s, mode)	(SLAP_EPOLL_SOCK_EV(t,s) & (mode))

# define SLAP_SOCK_IS_READ(t,s)		SLAP_EPOLL_SOCK_IS_SET(t,(s), EPOLLIN)
# define SLAP_SOCK_IS_WRITE(t,s)		SLAP_EPOLL_SOCK_IS_SET(t,(s), EPOLLOUT)

# define SLAP_EPOLL_SOCK_SET(t,s, mode)	do { \
	if ( (SLAP_EPOLL_SOCK_EV(t,s) & (mode)) != (mode) ) {	\
		SLAP_EPOLL_SOCK_EV(t,s) |= (mode); \
		epoll_ctl( slap_daemon[t].sd_epfd, EPOLL_CTL_MOD, (s), \
			&SLAP_EPOLL_SOCK_EP(t,s) ); \
	} \
} while (0)

# define SLAP_EPOLL_SOCK_CLR(t,s, mode)	do { \
	if ( (SLAP_EPOLL_SOCK_EV(t,s) & (mode)) ) { \
		SLAP_EPOLL_SOCK_EV(t,s) &= ~(mode);	\
		epoll_ctl( slap_daemon[t].sd_epfd, EPOLL_CTL_MOD, s, \
			&SLAP_EPOLL_SOCK_EP(t,s) ); \
	} \
} while (0)

# define SLAP_SOCK_SET_READ(t,s)		SLAP_EPOLL_SOCK_SET(t,s, EPOLLIN)
# define SLAP_SOCK_SET_WRITE(t,s)		SLAP_EPOLL_SOCK_SET(t,s, EPOLLOUT)

# define SLAP_SOCK_CLR_READ(t,s)		SLAP_EPOLL_SOCK_CLR(t,(s), EPOLLIN)
# define SLAP_SOCK_CLR_WRITE(t,s)		SLAP_EPOLL_SOCK_CLR(t,(s), EPOLLOUT)

#  define SLAP_SOCK_SET_SUSPEND(t,s) \
	( slap_daemon[t].sd_suspend[SLAP_EPOLL_SOCK_IX(t,s)] = 1 )
#  define SLAP_SOCK_CLR_SUSPEND(t,s) \
	( slap_daemon[t].sd_suspend[SLAP_EPOLL_SOCK_IX(t,s)] = 0 )
#  define SLAP_SOCK_IS_SUSPEND(t,s) \
	( slap_daemon[t].sd_suspend[SLAP_EPOLL_SOCK_IX(t,s)] == 1 )

# define SLAP_EPOLL_EVENT_CLR(i, mode)	(revents[(i)].events &= ~(mode))

# define SLAP_EVENT_MAX(t)			slap_daemon[t].sd_nfds

/* If a Listener address is provided, store that as the epoll data.
 * Otherwise, store the address of this socket's slot in the
 * index array. If we can't do this add, the system is out of
 * resources and we need to shutdown.
 */
# define SLAP_SOCK_ADD(t, s, l)		do { \
	int rc; \
	SLAP_EPOLL_SOCK_IX(t,(s)) = slap_daemon[t].sd_nfds; \
	SLAP_EPOLL_SOCK_EP(t,(s)).data.ptr = (l) ? (l) : (void *)(&SLAP_EPOLL_SOCK_IX(t,s)); \
	SLAP_EPOLL_SOCK_EV(t,(s)) = EPOLLIN; \
	rc = epoll_ctl(slap_daemon[t].sd_epfd, EPOLL_CTL_ADD, \
		(s), &SLAP_EPOLL_SOCK_EP(t,(s))); \
	if ( rc == 0 ) { \
		slap_daemon[t].sd_nfds++; \
	} else { \
		Debug( LDAP_DEBUG_ANY, \
			"daemon: epoll_ctl(ADD,fd=%d) failed, errno=%d, shutting down\n", \
			s, errno, 0 ); \
		slapd_shutdown = 2; \
	} \
} while (0)

# define SLAP_EPOLL_EV_LISTENER(t,ptr) \
	(((int *)(ptr) >= slap_daemon[t].sd_index && \
	(int *)(ptr) <= &slap_daemon[t].sd_index[dtblsize]) ? 0 : 1 )

# define SLAP_EPOLL_EV_PTRFD(t,ptr)		(SLAP_EPOLL_EV_LISTENER(t,ptr) ? \
	((Listener *)ptr)->sl_sd : \
	(ber_socket_t) ((int *)(ptr) - slap_daemon[t].sd_index))

# define SLAP_SOCK_DEL(t,s)		do { \
	int fd, rc, index = SLAP_EPOLL_SOCK_IX(t,(s)); \
	if ( index < 0 ) break; \
	rc = epoll_ctl(slap_daemon[t].sd_epfd, EPOLL_CTL_DEL, \
		(s), &SLAP_EPOLL_SOCK_EP(t,(s))); \
	slap_daemon[t].sd_epolls[index] = \
		slap_daemon[t].sd_epolls[slap_daemon[t].sd_nfds-1]; \
	fd = SLAP_EPOLL_EV_PTRFD(t,slap_daemon[t].sd_epolls[index].data.ptr); \
	slap_daemon[t].sd_index[fd] = index; \
	slap_daemon[t].sd_index[(s)] = -1; \
	slap_daemon[t].sd_nfds--; \
} while (0)

# define SLAP_EVENT_CLR_READ(i)		SLAP_EPOLL_EVENT_CLR((i), EPOLLIN)
# define SLAP_EVENT_CLR_WRITE(i)	SLAP_EPOLL_EVENT_CLR((i), EPOLLOUT)

# define SLAP_EPOLL_EVENT_CHK(i, mode)	(revents[(i)].events & mode)

# define SLAP_EVENT_IS_READ(i)		SLAP_EPOLL_EVENT_CHK((i), EPOLLIN)
# define SLAP_EVENT_IS_WRITE(i)		SLAP_EPOLL_EVENT_CHK((i), EPOLLOUT)
# define SLAP_EVENT_IS_LISTENER(t,i)	SLAP_EPOLL_EV_LISTENER(t,revents[(i)].data.ptr)
# define SLAP_EVENT_LISTENER(t,i)		((Listener *)(revents[(i)].data.ptr))

# define SLAP_EVENT_FD(t,i)		SLAP_EPOLL_EV_PTRFD(t,revents[(i)].data.ptr)

# define SLAP_SOCK_INIT(t)		do { \
	int j; \
	slap_daemon[t].sd_epolls = ch_calloc(1, \
		( sizeof(struct epoll_event) * 2 \
			+ sizeof(int) ) * dtblsize * 2); \
	slap_daemon[t].sd_index = (int *)&slap_daemon[t].sd_epolls[ 2 * dtblsize ]; \
	slap_daemon[t].sd_epfd = epoll_create( dtblsize / slapd_daemon_threads ); \
	for ( j = 0; j < dtblsize; j++ ) slap_daemon[t].sd_index[j] = -1; \
} while (0)

# define SLAP_SOCK_DESTROY(t)		do { \
	if ( slap_daemon[t].sd_epolls != NULL ) { \
		ch_free( slap_daemon[t].sd_epolls ); \
		slap_daemon[t].sd_epolls = NULL; \
		slap_daemon[t].sd_index = NULL; \
		close( slap_daemon[t].sd_epfd ); \
	} \
} while ( 0 )

# define SLAP_EVENT_DECL		struct epoll_event *revents

# define SLAP_EVENT_INIT(t)		do { \
	revents = slap_daemon[t].sd_epolls + dtblsize; \
} while (0)

# define SLAP_EVENT_WAIT(t, tvp, nsp)	do { \
	*(nsp) = epoll_wait( slap_daemon[t].sd_epfd, revents, \
		dtblsize, (tvp) ? ((tvp)->tv_sec * 1000 + (tvp)->tv_usec / 1000) : -1 ); \
} while (0)

#elif defined(SLAP_X_DEVPOLL) && defined(HAVE_DEVPOLL)

/*************************************************************
 * Use Solaris' (>= 2.7) /dev/poll infrastructure - poll(7d) *
 *************************************************************/
# define SLAP_EVENT_FNAME		"/dev/poll"
# define SLAP_EVENTS_ARE_INDEXED	0
/*
 * - sd_index	is used much like with epoll()
 * - sd_l	is maintained as an array containing the address
 *		of the listener; the index is the fd itself
 * - sd_pollfd	is used to keep track of what data has been
 *		registered in /dev/poll
 */
# define SLAP_DEVPOLL_SOCK_IX(t,s)	(slap_daemon[t].sd_index[(s)])
# define SLAP_DEVPOLL_SOCK_LX(t,s)	(slap_daemon[t].sd_l[(s)])
# define SLAP_DEVPOLL_SOCK_EP(t,s)	(slap_daemon[t].sd_pollfd[SLAP_DEVPOLL_SOCK_IX(t,(s))])
# define SLAP_DEVPOLL_SOCK_FD(t,s)	(SLAP_DEVPOLL_SOCK_EP(t,(s)).fd)
# define SLAP_DEVPOLL_SOCK_EV(t,s)	(SLAP_DEVPOLL_SOCK_EP(t,(s)).events)
# define SLAP_SOCK_IS_ACTIVE(t,s)		(SLAP_DEVPOLL_SOCK_IX(t,(s)) != -1)
# define SLAP_SOCK_NOT_ACTIVE(t,s)	(SLAP_DEVPOLL_SOCK_IX(t,(s)) == -1)
# define SLAP_SOCK_IS_SET(t,s, mode)	(SLAP_DEVPOLL_SOCK_EV(t,(s)) & (mode))

# define SLAP_SOCK_IS_READ(t,s)		SLAP_SOCK_IS_SET(t,(s), POLLIN)
# define SLAP_SOCK_IS_WRITE(t,s)		SLAP_SOCK_IS_SET(t,(s), POLLOUT)

/* as far as I understand, any time we need to communicate with the kernel
 * about the number and/or properties of a file descriptor we need it to
 * wait for, we have to rewrite the whole set */
# define SLAP_DEVPOLL_WRITE_POLLFD(t,s, pfd, n, what, shdn)	do { \
	int rc; \
	size_t size = (n) * sizeof( struct pollfd ); \
	/* FIXME: use pwrite? */ \
	rc = write( slap_daemon[t].sd_dpfd, (pfd), size ); \
	if ( rc != size ) { \
		Debug( LDAP_DEBUG_ANY, "daemon: " SLAP_EVENT_FNAME ": " \
			"%s fd=%d failed errno=%d\n", \
			(what), (s), errno ); \
		if ( (shdn) ) { \
			slapd_shutdown = 2; \
		} \
	} \
} while (0)

# define SLAP_DEVPOLL_SOCK_SET(t,s, mode) 	do { \
	Debug( LDAP_DEBUG_CONNS, "SLAP_SOCK_SET_%s(%d)=%d\n", \
		(mode) == POLLIN ? "READ" : "WRITE", (s), \
		( (SLAP_DEVPOLL_SOCK_EV(t,(s)) & (mode)) != (mode) ) ); \
	if ( (SLAP_DEVPOLL_SOCK_EV(t,(s)) & (mode)) != (mode) ) { \
		struct pollfd pfd; \
		SLAP_DEVPOLL_SOCK_EV(t,(s)) |= (mode); \
		pfd.fd = SLAP_DEVPOLL_SOCK_FD(t,(s)); \
		pfd.events = /* (mode) */ SLAP_DEVPOLL_SOCK_EV(t,(s)); \
		SLAP_DEVPOLL_WRITE_POLLFD(t,(s), &pfd, 1, "SET", 0); \
	} \
} while (0)

# define SLAP_DEVPOLL_SOCK_CLR(t,s, mode)		do { \
	Debug( LDAP_DEBUG_CONNS, "SLAP_SOCK_CLR_%s(%d)=%d\n", \
		(mode) == POLLIN ? "READ" : "WRITE", (s), \
		( (SLAP_DEVPOLL_SOCK_EV(t,(s)) & (mode)) == (mode) ) ); \
	if ((SLAP_DEVPOLL_SOCK_EV(t,(s)) & (mode)) == (mode) ) { \
		struct pollfd pfd[2]; \
		SLAP_DEVPOLL_SOCK_EV(t,(s)) &= ~(mode); \
		pfd[0].fd = SLAP_DEVPOLL_SOCK_FD(t,(s)); \
		pfd[0].events = POLLREMOVE; \
		pfd[1] = SLAP_DEVPOLL_SOCK_EP(t,(s)); \
		SLAP_DEVPOLL_WRITE_POLLFD(t,(s), &pfd[0], 2, "CLR", 0); \
	} \
} while (0)

# define SLAP_SOCK_SET_READ(t,s)		SLAP_DEVPOLL_SOCK_SET(t,s, POLLIN)
# define SLAP_SOCK_SET_WRITE(t,s)		SLAP_DEVPOLL_SOCK_SET(t,s, POLLOUT)

# define SLAP_SOCK_CLR_READ(t,s)		SLAP_DEVPOLL_SOCK_CLR(t,(s), POLLIN)
# define SLAP_SOCK_CLR_WRITE(t,s)		SLAP_DEVPOLL_SOCK_CLR(t,(s), POLLOUT)

#  define SLAP_SOCK_SET_SUSPEND(t,s) \
	( slap_daemon[t].sd_suspend[SLAP_DEVPOLL_SOCK_IX(t,(s))] = 1 )
#  define SLAP_SOCK_CLR_SUSPEND(t,s) \
	( slap_daemon[t].sd_suspend[SLAP_DEVPOLL_SOCK_IX(t,(s))] = 0 )
#  define SLAP_SOCK_IS_SUSPEND(t,s) \
	( slap_daemon[t].sd_suspend[SLAP_DEVPOLL_SOCK_IX(t,(s))] == 1 )

# define SLAP_DEVPOLL_EVENT_CLR(i, mode)	(revents[(i)].events &= ~(mode))

# define SLAP_EVENT_MAX(t)			slap_daemon[t].sd_nfds

/* If a Listener address is provided, store that in the sd_l array.
 * If we can't do this add, the system is out of resources and we 
 * need to shutdown.
 */
# define SLAP_SOCK_ADD(t, s, l)		do { \
	Debug( LDAP_DEBUG_CONNS, "SLAP_SOCK_ADD(%d, %p)\n", (s), (l), 0 ); \
	SLAP_DEVPOLL_SOCK_IX(t,(s)) = slap_daemon[t].sd_nfds; \
	SLAP_DEVPOLL_SOCK_LX(t,(s)) = (l); \
	SLAP_DEVPOLL_SOCK_FD(t,(s)) = (s); \
	SLAP_DEVPOLL_SOCK_EV(t,(s)) = POLLIN; \
	SLAP_DEVPOLL_WRITE_POLLFD(t,(s), &SLAP_DEVPOLL_SOCK_EP((s)), 1, "ADD", 1); \
	slap_daemon[t].sd_nfds++; \
} while (0)

# define SLAP_DEVPOLL_EV_LISTENER(ptr)	((ptr) != NULL)

# define SLAP_SOCK_DEL(t,s)		do { \
	int fd, index = SLAP_DEVPOLL_SOCK_IX(t,(s)); \
	Debug( LDAP_DEBUG_CONNS, "SLAP_SOCK_DEL(%d)\n", (s), 0, 0 ); \
	if ( index < 0 ) break; \
	if ( index < slap_daemon[t].sd_nfds - 1 ) { \
		struct pollfd pfd = slap_daemon[t].sd_pollfd[index]; \
		fd = slap_daemon[t].sd_pollfd[slap_daemon[t].sd_nfds - 1].fd; \
		slap_daemon[t].sd_pollfd[index] = slap_daemon[t].sd_pollfd[slap_daemon[t].sd_nfds - 1]; \
		slap_daemon[t].sd_pollfd[slap_daemon[t].sd_nfds - 1] = pfd; \
		slap_daemon[t].sd_index[fd] = index; \
	} \
	slap_daemon[t].sd_index[(s)] = -1; \
	slap_daemon[t].sd_pollfd[slap_daemon[t].sd_nfds - 1].events = POLLREMOVE; \
	SLAP_DEVPOLL_WRITE_POLLFD(t,(s), &slap_daemon[t].sd_pollfd[slap_daemon[t].sd_nfds - 1], 1, "DEL", 0); \
	slap_daemon[t].sd_pollfd[slap_daemon[t].sd_nfds - 1].events = 0; \
	slap_daemon[t].sd_nfds--; \
} while (0)

# define SLAP_EVENT_CLR_READ(i)		SLAP_DEVPOLL_EVENT_CLR((i), POLLIN)
# define SLAP_EVENT_CLR_WRITE(i)	SLAP_DEVPOLL_EVENT_CLR((i), POLLOUT)

# define SLAP_DEVPOLL_EVENT_CHK(i, mode)	(revents[(i)].events & (mode))

# define SLAP_EVENT_FD(t,i)		(revents[(i)].fd)

# define SLAP_EVENT_IS_READ(i)		SLAP_DEVPOLL_EVENT_CHK((i), POLLIN)
# define SLAP_EVENT_IS_WRITE(i)		SLAP_DEVPOLL_EVENT_CHK((i), POLLOUT)
# define SLAP_EVENT_IS_LISTENER(t,i)	SLAP_DEVPOLL_EV_LISTENER(SLAP_DEVPOLL_SOCK_LX(SLAP_EVENT_FD(t,(i))))
# define SLAP_EVENT_LISTENER(t,i)		SLAP_DEVPOLL_SOCK_LX(SLAP_EVENT_FD(t,(i)))

# define SLAP_SOCK_INIT(t)		do { \
	slap_daemon[t].sd_pollfd = ch_calloc( 1, \
		( sizeof(struct pollfd) * 2 \
			+ sizeof( int ) \
			+ sizeof( Listener * ) ) * dtblsize ); \
	slap_daemon[t].sd_index = (int *)&slap_daemon[t].sd_pollfd[ 2 * dtblsize ]; \
	slap_daemon[t].sd_l = (Listener **)&slap_daemon[t].sd_index[ dtblsize ]; \
	slap_daemon[t].sd_dpfd = open( SLAP_EVENT_FNAME, O_RDWR ); \
	if ( slap_daemon[t].sd_dpfd == -1 ) { \
		Debug( LDAP_DEBUG_ANY, "daemon: " SLAP_EVENT_FNAME ": " \
			"open(\"" SLAP_EVENT_FNAME "\") failed errno=%d\n", \
			errno, 0, 0 ); \
		SLAP_SOCK_DESTROY; \
		return -1; \
	} \
	for ( i = 0; i < dtblsize; i++ ) { \
		slap_daemon[t].sd_pollfd[i].fd = -1; \
		slap_daemon[t].sd_index[i] = -1; \
	} \
} while (0)

# define SLAP_SOCK_DESTROY(t)		do { \
	if ( slap_daemon[t].sd_pollfd != NULL ) { \
		ch_free( slap_daemon[t].sd_pollfd ); \
		slap_daemon[t].sd_pollfd = NULL; \
		slap_daemon[t].sd_index = NULL; \
		slap_daemon[t].sd_l = NULL; \
		close( slap_daemon[t].sd_dpfd ); \
	} \
} while ( 0 )

# define SLAP_EVENT_DECL		struct pollfd *revents

# define SLAP_EVENT_INIT(t)		do { \
	revents = &slap_daemon[t].sd_pollfd[ dtblsize ]; \
} while (0)

# define SLAP_EVENT_WAIT(t, tvp, nsp)	do { \
	struct dvpoll		sd_dvpoll; \
	sd_dvpoll.dp_timeout = (tvp) ? ((tvp)->tv_sec * 1000 + (tvp)->tv_usec / 1000) : -1; \
	sd_dvpoll.dp_nfds = dtblsize; \
	sd_dvpoll.dp_fds = revents; \
	*(nsp) = ioctl( slap_daemon[t].sd_dpfd, DP_POLL, &sd_dvpoll ); \
} while (0)

#else /* ! epoll && ! /dev/poll */
# ifdef HAVE_WINSOCK
# define SLAP_EVENT_FNAME		"WSselect"
/* Winsock provides a "select" function but its fd_sets are
 * actually arrays of sockets. Since these sockets are handles
 * and not a contiguous range of small integers, we manage our
 * own "fd" table of socket handles and use their indices as
 * descriptors.
 *
 * All of our listener/connection structures use fds; the actual
 * I/O functions use sockets. The SLAP_FD2SOCK macro in proto-slap.h
 * handles the mapping.
 *
 * Despite the mapping overhead, this is about 45% more efficient
 * than just using Winsock's select and FD_ISSET directly.
 *
 * Unfortunately Winsock's select implementation doesn't scale well
 * as the number of connections increases. This probably needs to be
 * rewritten to use the Winsock overlapped/asynchronous I/O functions.
 */
# define SLAP_EVENTS_ARE_INDEXED	1
# define SLAP_EVENT_DECL		fd_set readfds, writefds; char *rflags
# define SLAP_EVENT_INIT(t)	do { \
	int i; \
	FD_ZERO( &readfds ); \
	FD_ZERO( &writefds ); \
	rflags = slap_daemon[t].sd_rflags; \
	memset( rflags, 0, slap_daemon[t].sd_nfds ); \
	for ( i=0; i<slap_daemon[t].sd_nfds; i++ ) { \
		if ( slap_daemon[t].sd_flags[i] & SD_READ ) \
			FD_SET( slapd_ws_sockets[i], &readfds );\
		if ( slap_daemon[t].sd_flags[i] & SD_WRITE ) \
			FD_SET( slapd_ws_sockets[i], &writefds ); \
	} } while ( 0 )

# define SLAP_EVENT_MAX(t)		slap_daemon[t].sd_nfds

# define SLAP_EVENT_WAIT(t, tvp, nsp)	do { \
	int i; \
	*(nsp) = select( SLAP_EVENT_MAX(t), &readfds, \
		nwriters > 0 ? &writefds : NULL, NULL, (tvp) ); \
	for ( i=0; i<readfds.fd_count; i++) { \
		int fd = slapd_sock2fd(readfds.fd_array[i]); \
		if ( fd >= 0 ) { \
			slap_daemon[t].sd_rflags[fd] = SD_READ; \
			if ( fd >= *(nsp)) *(nsp) = fd+1; \
		} \
	} \
	for ( i=0; i<writefds.fd_count; i++) { \
		int fd = slapd_sock2fd(writefds.fd_array[i]); \
		if ( fd >= 0 ) { \
			slap_daemon[t].sd_rflags[fd] = SD_WRITE; \
			if ( fd >= *(nsp)) *(nsp) = fd+1; \
		} \
	} \
} while (0)

# define SLAP_EVENT_IS_READ(fd)		(rflags[fd] & SD_READ)
# define SLAP_EVENT_IS_WRITE(fd)	(rflags[fd] & SD_WRITE)

# define SLAP_EVENT_CLR_READ(fd) 	rflags[fd] &= ~SD_READ
# define SLAP_EVENT_CLR_WRITE(fd)	rflags[fd] &= ~SD_WRITE

# define SLAP_SOCK_INIT(t)		do { \
	if (!t) { \
	ldap_pvt_thread_mutex_init( &slapd_ws_mutex ); \
	slapd_ws_sockets = ch_malloc( dtblsize * ( sizeof(SOCKET) + 2)); \
	memset( slapd_ws_sockets, -1, dtblsize * sizeof(SOCKET) ); \
	} \
	slap_daemon[t].sd_flags = (char *)(slapd_ws_sockets + dtblsize); \
	slap_daemon[t].sd_rflags = slap_daemon[t].sd_flags + dtblsize; \
	memset( slap_daemon[t].sd_flags, 0, dtblsize ); \
	slapd_ws_sockets[t*2] = wake_sds[t][0]; \
	slapd_ws_sockets[t*2+1] = wake_sds[t][1]; \
	wake_sds[t][0] = t*2; \
	wake_sds[t][1] = t*2+1; \
	slap_daemon[t].sd_nfds = t*2 + 2; \
	} while ( 0 )

# define SLAP_SOCK_DESTROY(t)	do { \
	ch_free( slapd_ws_sockets ); slapd_ws_sockets = NULL; \
	slap_daemon[t].sd_flags = NULL; \
	slap_daemon[t].sd_rflags = NULL; \
	ldap_pvt_thread_mutex_destroy( &slapd_ws_mutex ); \
	} while ( 0 )

# define SLAP_SOCK_IS_ACTIVE(t,fd) ( slap_daemon[t].sd_flags[fd] & SD_ACTIVE )
# define SLAP_SOCK_IS_READ(t,fd) ( slap_daemon[t].sd_flags[fd] & SD_READ )
# define SLAP_SOCK_IS_WRITE(t,fd) ( slap_daemon[t].sd_flags[fd] & SD_WRITE )
# define SLAP_SOCK_NOT_ACTIVE(t,fd)	(!slap_daemon[t].sd_flags[fd])

# define SLAP_SOCK_SET_READ(t,fd)		( slap_daemon[t].sd_flags[fd] |= SD_READ )
# define SLAP_SOCK_SET_WRITE(t,fd)		( slap_daemon[t].sd_flags[fd] |= SD_WRITE )

# define SLAP_SELECT_ADDTEST(t,s)	do { \
	if ((s) >= slap_daemon[t].sd_nfds) slap_daemon[t].sd_nfds = (s)+1; \
} while (0)

# define SLAP_SOCK_CLR_READ(t,fd)		( slap_daemon[t].sd_flags[fd] &= ~SD_READ )
# define SLAP_SOCK_CLR_WRITE(t,fd)		( slap_daemon[t].sd_flags[fd] &= ~SD_WRITE )

# define SLAP_SOCK_ADD(t,s, l)	do { \
	SLAP_SELECT_ADDTEST(t,(s)); \
	slap_daemon[t].sd_flags[s] = SD_ACTIVE|SD_READ; \
} while ( 0 )

# define SLAP_SOCK_DEL(t,s) do { \
	slap_daemon[t].sd_flags[s] = 0; \
	slapd_sockdel( s ); \
} while ( 0 )

# else /* !HAVE_WINSOCK */

/**************************************
 * Use select system call - select(2) *
 **************************************/
# define SLAP_EVENT_FNAME		"select"
/* select */
# define SLAP_EVENTS_ARE_INDEXED	1
# define SLAP_EVENT_DECL		fd_set readfds, writefds

# define SLAP_EVENT_INIT(t)		do { \
	AC_MEMCPY( &readfds, &slap_daemon[t].sd_readers, sizeof(fd_set) );	\
	if ( nwriters )	{ \
		AC_MEMCPY( &writefds, &slap_daemon[t].sd_writers, sizeof(fd_set) ); \
	} else { \
		FD_ZERO( &writefds ); \
	} \
} while (0)

# ifdef FD_SETSIZE
#  define SLAP_SELECT_CHK_SETSIZE	do { \
	if (dtblsize > FD_SETSIZE) dtblsize = FD_SETSIZE; \
} while (0)
# else /* ! FD_SETSIZE */
#  define SLAP_SELECT_CHK_SETSIZE	do { ; } while (0)
# endif /* ! FD_SETSIZE */

# define SLAP_SOCK_INIT(t)			do { \
	SLAP_SELECT_CHK_SETSIZE; \
	FD_ZERO(&slap_daemon[t].sd_actives); \
	FD_ZERO(&slap_daemon[t].sd_readers); \
	FD_ZERO(&slap_daemon[t].sd_writers); \
} while (0)

# define SLAP_SOCK_DESTROY(t)

# define SLAP_SOCK_IS_ACTIVE(t,fd)	FD_ISSET((fd), &slap_daemon[t].sd_actives)
# define SLAP_SOCK_IS_READ(t,fd)		FD_ISSET((fd), &slap_daemon[t].sd_readers)
# define SLAP_SOCK_IS_WRITE(t,fd)		FD_ISSET((fd), &slap_daemon[t].sd_writers)

# define SLAP_SOCK_NOT_ACTIVE(t,fd)	(!SLAP_SOCK_IS_ACTIVE(t,fd) && \
	 !SLAP_SOCK_IS_READ(t,fd) && !SLAP_SOCK_IS_WRITE(t,fd))

# define SLAP_SOCK_SET_READ(t,fd)	FD_SET((fd), &slap_daemon[t].sd_readers)
# define SLAP_SOCK_SET_WRITE(t,fd)	FD_SET((fd), &slap_daemon[t].sd_writers)

# define SLAP_EVENT_MAX(t)		slap_daemon[t].sd_nfds
# define SLAP_SELECT_ADDTEST(t,s)	do { \
	if ((s) >= slap_daemon[t].sd_nfds) slap_daemon[t].sd_nfds = (s)+1; \
} while (0)

# define SLAP_SOCK_CLR_READ(t,fd)		FD_CLR((fd), &slap_daemon[t].sd_readers)
# define SLAP_SOCK_CLR_WRITE(t,fd)	FD_CLR((fd), &slap_daemon[t].sd_writers)

# define SLAP_SOCK_ADD(t,s, l)		do { \
	SLAP_SELECT_ADDTEST(t,(s)); \
	FD_SET((s), &slap_daemon[t].sd_actives); \
	FD_SET((s), &slap_daemon[t].sd_readers); \
} while (0)

# define SLAP_SOCK_DEL(t,s)		do { \
	FD_CLR((s), &slap_daemon[t].sd_actives); \
	FD_CLR((s), &slap_daemon[t].sd_readers); \
	FD_CLR((s), &slap_daemon[t].sd_writers); \
} while (0)

# define SLAP_EVENT_IS_READ(fd)		FD_ISSET((fd), &readfds)
# define SLAP_EVENT_IS_WRITE(fd)	FD_ISSET((fd), &writefds)

# define SLAP_EVENT_CLR_READ(fd) 	FD_CLR((fd), &readfds)
# define SLAP_EVENT_CLR_WRITE(fd)	FD_CLR((fd), &writefds)

# define SLAP_EVENT_WAIT(t, tvp, nsp)	do { \
	*(nsp) = select( SLAP_EVENT_MAX(t), &readfds, \
		nwriters > 0 ? &writefds : NULL, NULL, (tvp) ); \
} while (0)
# endif /* !HAVE_WINSOCK */
#endif /* ! epoll && ! /dev/poll */

#ifdef HAVE_SLP
/*
 * SLP related functions
 */
#include <slp.h>

#define LDAP_SRVTYPE_PREFIX "service:ldap://"
#define LDAPS_SRVTYPE_PREFIX "service:ldaps://"
static char** slapd_srvurls = NULL;
static SLPHandle slapd_hslp = 0;
int slapd_register_slp = 0;
const char *slapd_slp_attrs = NULL;

static SLPError slapd_slp_cookie;

static void
slapd_slp_init( const char* urls )
{
	int i;
	SLPError err;

	slapd_srvurls = ldap_str2charray( urls, " " );

	if ( slapd_srvurls == NULL ) return;

	/* find and expand INADDR_ANY URLs */
	for ( i = 0; slapd_srvurls[i] != NULL; i++ ) {
		if ( strcmp( slapd_srvurls[i], "ldap:///" ) == 0 ) {
			slapd_srvurls[i] = (char *) ch_realloc( slapd_srvurls[i],
				global_host_bv.bv_len +
				sizeof( LDAP_SRVTYPE_PREFIX ) );
			strcpy( lutil_strcopy(slapd_srvurls[i],
				LDAP_SRVTYPE_PREFIX ), global_host_bv.bv_val );
		} else if ( strcmp( slapd_srvurls[i], "ldaps:///" ) == 0 ) {
			slapd_srvurls[i] = (char *) ch_realloc( slapd_srvurls[i],
				global_host_bv.bv_len +
				sizeof( LDAPS_SRVTYPE_PREFIX ) );
			strcpy( lutil_strcopy(slapd_srvurls[i],
				LDAPS_SRVTYPE_PREFIX ), global_host_bv.bv_val );
		}
	}

	/* open the SLP handle */
	err = SLPOpen( "en", 0, &slapd_hslp );

	if ( err != SLP_OK ) {
		Debug( LDAP_DEBUG_CONNS, "daemon: SLPOpen() failed with %ld\n",
			(long)err, 0, 0 );
	}
}

static void
slapd_slp_deinit( void )
{
	if ( slapd_srvurls == NULL ) return;

	ldap_charray_free( slapd_srvurls );
	slapd_srvurls = NULL;

	/* close the SLP handle */
	SLPClose( slapd_hslp );
}

static void
slapd_slp_regreport(
	SLPHandle	hslp,
	SLPError	errcode,
	void		*cookie )
{
	/* return the error code in the cookie */
	*(SLPError*)cookie = errcode; 
}

static void
slapd_slp_reg()
{
	int i;
	SLPError err;

	if ( slapd_srvurls == NULL ) return;

	for ( i = 0; slapd_srvurls[i] != NULL; i++ ) {
		if ( strncmp( slapd_srvurls[i], LDAP_SRVTYPE_PREFIX,
				sizeof( LDAP_SRVTYPE_PREFIX ) - 1 ) == 0 ||
			strncmp( slapd_srvurls[i], LDAPS_SRVTYPE_PREFIX,
				sizeof( LDAPS_SRVTYPE_PREFIX ) - 1 ) == 0 )
		{
			err = SLPReg( slapd_hslp,
				slapd_srvurls[i],
				SLP_LIFETIME_MAXIMUM,
				"ldap",
				(slapd_slp_attrs) ? slapd_slp_attrs : "",
				SLP_TRUE,
				slapd_slp_regreport,
				&slapd_slp_cookie );

			if ( err != SLP_OK || slapd_slp_cookie != SLP_OK ) {
				Debug( LDAP_DEBUG_CONNS,
					"daemon: SLPReg(%s) failed with %ld, cookie = %ld\n",
					slapd_srvurls[i], (long)err, (long)slapd_slp_cookie );
			}	
		}
	}
}

static void
slapd_slp_dereg( void )
{
	int i;
	SLPError err;

	if ( slapd_srvurls == NULL ) return;

	for ( i = 0; slapd_srvurls[i] != NULL; i++ ) {
		err = SLPDereg( slapd_hslp,
			slapd_srvurls[i],
			slapd_slp_regreport,
			&slapd_slp_cookie );
		
		if ( err != SLP_OK || slapd_slp_cookie != SLP_OK ) {
			Debug( LDAP_DEBUG_CONNS,
				"daemon: SLPDereg(%s) failed with %ld, cookie = %ld\n",
				slapd_srvurls[i], (long)err, (long)slapd_slp_cookie );
		}
	}
}
#endif /* HAVE_SLP */

#ifdef HAVE_WINSOCK
/* Manage the descriptor to socket table */
ber_socket_t
slapd_socknew( ber_socket_t s )
{
	ber_socket_t i;
	ldap_pvt_thread_mutex_lock( &slapd_ws_mutex );
	for ( i = 0; i < dtblsize && slapd_ws_sockets[i] != INVALID_SOCKET; i++ );
	if ( i == dtblsize ) {
		WSASetLastError( WSAEMFILE );
	} else {
		slapd_ws_sockets[i] = s;
	}
	ldap_pvt_thread_mutex_unlock( &slapd_ws_mutex );
	return i;
}

void
slapd_sockdel( ber_socket_t s )
{
	ldap_pvt_thread_mutex_lock( &slapd_ws_mutex );
	slapd_ws_sockets[s] = INVALID_SOCKET;
	ldap_pvt_thread_mutex_unlock( &slapd_ws_mutex );
}

ber_socket_t
slapd_sock2fd( ber_socket_t s )
{
	ber_socket_t i;
	for ( i=0; i<dtblsize && slapd_ws_sockets[i] != s; i++);
	if ( i == dtblsize )
		i = -1;
	return i;
}
#endif

#ifdef DEBUG_CLOSE
/* Was used to find a bug causing slapd's descriptors to be closed
 * out from under it. Tracked it down to a long-standing (from 2009)
 * bug in Heimdal https://github.com/heimdal/heimdal/issues/431 .
 * Leaving this here for future use, if necessary.
 */
#include <dlfcn.h>
#ifndef RTLD_NEXT
#define RTLD_NEXT	(void *)-1L
#endif
static char *newconns;
typedef int (closefunc)(int fd);
static closefunc *close_ptr;
int close( int s )
{
	if (newconns) {
		Debug( LDAP_DEBUG_CONNS,
			"daemon: close(%ld)\n", s, 0, 0 );
		if (s >= 0 && s < dtblsize && newconns[s])
			assert(newconns[s] == 2);
	}
	return close_ptr ? close_ptr(s) : -1;
}

void slapd_debug_close()
{
	if (dtblsize)
		newconns = ch_calloc(1, dtblsize);
	close_ptr = dlsym(RTLD_NEXT, "close");
}

void slapd_set_close(int fd)
{
	newconns[fd] = 3;
}
#define SETUP_CLOSE()	slapd_debug_close()
#define SET_CLOSE(fd)	slapd_set_close(fd)
#define CLR_CLOSE(fd)	if (newconns[fd]) newconns[fd]--
#else
#define SETUP_CLOSE(fd)
#define SET_CLOSE(fd)
#define CLR_CLOSE(fd)
#endif

/*
 * Add a descriptor to daemon control
 *
 * If isactive, the descriptor is a live server session and is subject
 * to idletimeout control. Otherwise, the descriptor is a passive
 * listener or an outbound client session, and not subject to
 * idletimeout. The underlying event handler may record the Listener
 * argument to differentiate Listener's from real sessions.
 */
static void
slapd_add( ber_socket_t s, int isactive, Listener *sl, int id )
{
	if (id < 0)
		id = DAEMON_ID(s);
	ldap_pvt_thread_mutex_lock( &slap_daemon[id].sd_mutex );

	assert( SLAP_SOCK_NOT_ACTIVE(id, s) );

	if ( isactive ) slap_daemon[id].sd_nactives++;

	SLAP_SOCK_ADD(id, s, sl);

	Debug( LDAP_DEBUG_CONNS, "daemon: added %ldr%s listener=%p\n",
		(long) s, isactive ? " (active)" : "", (void *)sl );

	ldap_pvt_thread_mutex_unlock( &slap_daemon[id].sd_mutex );

	WAKE_LISTENER(id,1);
}

/*
 * Remove the descriptor from daemon control
 */
void
slapd_remove(
	ber_socket_t s,
	Sockbuf *sb,
	int wasactive,
	int wake,
	int locked )
{
	int waswriter;
	int wasreader;
	int id = DAEMON_ID(s);

	if ( !locked )
		ldap_pvt_thread_mutex_lock( &slap_daemon[id].sd_mutex );

	assert( SLAP_SOCK_IS_ACTIVE( id, s ));

	if ( wasactive ) slap_daemon[id].sd_nactives--;

	waswriter = SLAP_SOCK_IS_WRITE(id, s);
	wasreader = SLAP_SOCK_IS_READ(id, s);

	Debug( LDAP_DEBUG_CONNS, "daemon: removing %ld%s%s\n",
		(long) s,
		wasreader ? "r" : "",
		waswriter ? "w" : "" );

	if ( waswriter ) slap_daemon[id].sd_nwriters--;

	SLAP_SOCK_DEL(id, s);
	CLR_CLOSE(s);

	if ( sb )
		ber_sockbuf_free(sb);

	/* If we ran out of file descriptors, we dropped a listener from
	 * the select() loop. Now that we're removing a session from our
	 * control, we can try to resume a dropped listener to use.
	 */
	if ( emfile && listening ) {
		int i;
		for ( i = 0; slap_listeners[i] != NULL; i++ ) {
			Listener *lr = slap_listeners[i];

			if ( lr->sl_sd == AC_SOCKET_INVALID ) continue;
			if ( lr->sl_sd == s ) continue;
			if ( lr->sl_mute ) {
				lr->sl_mute = 0;
				emfile--;
				if ( DAEMON_ID(lr->sl_sd) != id )
					WAKE_LISTENER(DAEMON_ID(lr->sl_sd), wake);
				break;
			}
		}
		/* Walked the entire list without enabling anything; emfile
		 * counter is stale. Reset it.
		 */
		if ( slap_listeners[i] == NULL ) emfile = 0;
	}
	ldap_pvt_thread_mutex_unlock( &slap_daemon[id].sd_mutex );
	WAKE_LISTENER(id, wake || slapd_gentle_shutdown == 2);
}

void
slapd_clr_write( ber_socket_t s, int wake )
{
	int id = DAEMON_ID(s);
	ldap_pvt_thread_mutex_lock( &slap_daemon[id].sd_mutex );

	if ( SLAP_SOCK_IS_WRITE( id, s )) {
		assert( SLAP_SOCK_IS_ACTIVE( id, s ));

		SLAP_SOCK_CLR_WRITE( id, s );
		slap_daemon[id].sd_nwriters--;
	}

	ldap_pvt_thread_mutex_unlock( &slap_daemon[id].sd_mutex );
	WAKE_LISTENER(id,wake);
}

void
slapd_set_write( ber_socket_t s, int wake )
{
	int id = DAEMON_ID(s);
	ldap_pvt_thread_mutex_lock( &slap_daemon[id].sd_mutex );

	assert( SLAP_SOCK_IS_ACTIVE( id, s ));

	if ( !SLAP_SOCK_IS_WRITE( id, s )) {
		SLAP_SOCK_SET_WRITE( id, s );
		slap_daemon[id].sd_nwriters++;
	}