socket.c

/*
 * Copyright (C) 1998-2000  Internet Software Consortium.
 *
 * Permission to use, copy, modify, and distribute this software for any
 * purpose with or without fee is hereby granted, provided that the above
 * copyright notice and this permission notice appear in all copies.
 *
 * THE SOFTWARE IS PROVIDED "AS IS" AND INTERNET SOFTWARE CONSORTIUM
 * DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE INCLUDING ALL
 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL
 * INTERNET SOFTWARE CONSORTIUM BE LIABLE FOR ANY SPECIAL, DIRECT,
 * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING
 * FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT,
 * NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION
 * WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
 */

/* $Id: socket.c,v 1.172 2000/11/22 23:48:14 gson Exp $ */

#include <config.h>

#include <sys/param.h>
#include <sys/types.h>
#include <sys/socket.h>
#include <sys/time.h>
#include <sys/uio.h>

#include <errno.h>
#include <stddef.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <fcntl.h>

#include <isc/buffer.h>
#include <isc/bufferlist.h>
#include <isc/condition.h>
#include <isc/list.h>
#include <isc/log.h>
#include <isc/mem.h>
#include <isc/mutex.h>
#include <isc/net.h>
#include <isc/platform.h>
#include <isc/print.h>
#include <isc/region.h>
#include <isc/socket.h>
#include <isc/task.h>
#include <isc/thread.h>
#include <isc/util.h>

#ifndef ISC_PLATFORM_USETHREADS
#include "socket_p.h"
#endif /* ISC_PLATFORM_USETHREADS */

/*
 * Some systems define the socket length argument as an int, some as size_t,
 * some as socklen_t.  This is here so it can be easily changed if needed.
 */
#ifndef ISC_SOCKADDR_LEN_T
#define ISC_SOCKADDR_LEN_T unsigned int
#endif

/*
 * Define what the possible "soft" errors can be.  These are non-fatal returns
 * of various network related functions, like recv() and so on.
 *
 * For some reason, BSDI (and perhaps others) will sometimes return <0
 * from recv() but will have errno==0.  This is broken, but we have to
 * work around it here.
 */
#define SOFT_ERROR(e)	((e) == EAGAIN || \
			 (e) == EWOULDBLOCK || \
			 (e) == EINTR || \
			 (e) == 0)

#define DLVL(x) ISC_LOGCATEGORY_GENERAL, ISC_LOGMODULE_SOCKET, ISC_LOG_DEBUG(x)

/*
 * DLVL(90)  --  Function entry/exit and other tracing.
 * DLVL(70)  --  Socket "correctness" -- including returning of events, etc.
 * DLVL(60)  --  Socket data send/receive
 * DLVL(50)  --  Event tracing, including receiving/sending completion events.
 * DLVL(20)  --  Socket creation/destruction.
 */
#define TRACE_LEVEL		90
#define CORRECTNESS_LEVEL	70
#define IOEVENT_LEVEL		60
#define EVENT_LEVEL		50
#define CREATION_LEVEL		20

#define TRACE		DLVL(TRACE_LEVEL)
#define CORRECTNESS	DLVL(CORRECTNESS_LEVEL)
#define IOEVENT		DLVL(IOEVENT_LEVEL)
#define EVENT		DLVL(EVENT_LEVEL)
#define CREATION	DLVL(CREATION_LEVEL)

typedef isc_event_t intev_t;

#define SOCKET_MAGIC		0x494f696fU	/* IOio */
#define VALID_SOCKET(t)		((t) != NULL && (t)->magic == SOCKET_MAGIC)

/*
 * IPv6 control information.  If the socket is an IPv6 socket we want
 * to collect the destination address and interface so the client can
 * set them on outgoing packets.
 */
#ifdef ISC_PLATFORM_HAVEIPV6
#ifndef USE_CMSG
#define USE_CMSG	1
#endif
#endif

/*
 * NetBSD and FreeBSD can timestamp packets.  XXXMLG Should we have
 * a setsockopt() like interface to request timestamps, and if the OS
 * doesn't do it for us, call gettimeofday() on every UDP receive?
 */
#ifdef SO_TIMESTAMP
#ifndef USE_CMSG
#define USE_CMSG	1
#endif
#endif

/*
 * Check to see if we have even basic support for cracking messages from
 * the control data returned from/sent via recvmsg()/sendmsg().
 */
#if defined(USE_CMSG) && (!defined(CMSG_LEN) || !defined(CMSG_SPACE))
#undef USE_CMSG
#endif

struct isc_socket {
	/* Not locked. */
	unsigned int		magic;
	isc_socketmgr_t	       *manager;
	isc_mutex_t		lock;
	isc_sockettype_t	type;

	/* Locked by socket lock. */
	ISC_LINK(isc_socket_t)	link;
	unsigned int		references;
	int			fd;
	int			pf;

	ISC_LIST(isc_socketevent_t)		send_list;
	ISC_LIST(isc_socketevent_t)		recv_list;
	ISC_LIST(isc_socket_newconnev_t)	accept_list;
	isc_socket_connev_t		       *connect_ev;

	/*
	 * Internal events.  Posted when a descriptor is readable or
	 * writable.  These are statically allocated and never freed.
	 * They will be set to non-purgable before use.
	 */
	intev_t			readable_ev;
	intev_t			writable_ev;

	isc_sockaddr_t		address;  /* remote address */

	unsigned int		pending_recv : 1,
				pending_send : 1,
				pending_accept : 1,
				listener : 1, /* listener socket */
				connected : 1,
				connecting : 1, /* connect pending */
				bound : 1; /* bound to local addr */

#ifdef ISC_NET_RECVOVERFLOW
	unsigned char		overflow; /* used for MSG_TRUNC fake */
#endif
#ifdef USE_CMSG
	unsigned char	       *cmsg;
	unsigned int		cmsglen;
#endif
};

#define SOCKET_MANAGER_MAGIC	0x494f6d67U	/* IOmg */
#define VALID_MANAGER(m)	((m) != NULL && \
				 (m)->magic == SOCKET_MANAGER_MAGIC)
struct isc_socketmgr {
	/* Not locked. */
	unsigned int		magic;
	isc_mem_t	       *mctx;
	isc_mutex_t		lock;
	/* Locked by manager lock. */
	ISC_LIST(isc_socket_t)	socklist;
	fd_set			read_fds;
	fd_set			write_fds;
	isc_socket_t	       *fds[FD_SETSIZE];
	int			fdstate[FD_SETSIZE];
	int			maxfd;
#ifdef ISC_PLATFORM_USETHREADS
	isc_thread_t		watcher;
	isc_condition_t		shutdown_ok;
	int			pipe_fds[2];
#else /* ISC_PLATFORM_USETHREADS */
	unsigned int		refs;
#endif /* ISC_PLATFORM_USETHREADS */
};

#ifndef ISC_PLATFORM_USETHREADS
static isc_socketmgr_t *socketmgr = NULL;
#endif /* ISC_PLATFORM_USETHREADS */

#define CLOSED		0	/* this one must be zero */
#define MANAGED		1
#define CLOSE_PENDING	2

/*
 * send() and recv() iovec counts
 */
#define MAXSCATTERGATHER_SEND	(ISC_SOCKET_MAXSCATTERGATHER)
#ifdef ISC_NET_RECVOVERFLOW
# define MAXSCATTERGATHER_RECV	(ISC_SOCKET_MAXSCATTERGATHER + 1)
#else
# define MAXSCATTERGATHER_RECV	(ISC_SOCKET_MAXSCATTERGATHER)
#endif

static void send_recvdone_event(isc_socket_t *, isc_socketevent_t **,
				isc_result_t);
static void send_senddone_event(isc_socket_t *, isc_socketevent_t **,
				isc_result_t);
static void free_socket(isc_socket_t **);
static isc_result_t allocate_socket(isc_socketmgr_t *, isc_sockettype_t,
				    isc_socket_t **);
static void destroy(isc_socket_t **);
static void internal_accept(isc_task_t *, isc_event_t *);
static void internal_connect(isc_task_t *, isc_event_t *);
static void internal_recv(isc_task_t *, isc_event_t *);
static void internal_send(isc_task_t *, isc_event_t *);
static void process_cmsg(isc_socket_t *, struct msghdr *, isc_socketevent_t *);
static void build_msghdr_send(isc_socket_t *, isc_socketevent_t *,
			      struct msghdr *, struct iovec *, size_t *);
static void build_msghdr_recv(isc_socket_t *, isc_socketevent_t *,
			      struct msghdr *, struct iovec *, size_t *);

#define SELECT_POKE_SHUTDOWN		(-1)
#define SELECT_POKE_NOTHING		(-2)

#define SOCK_DEAD(s)			((s)->references == 0)

static void
manager_log(isc_socketmgr_t *sockmgr,
	    isc_logcategory_t *category, isc_logmodule_t *module, int level,
	    const char *fmt, ...)
{
	char msgbuf[2048];
	va_list ap;

	if (! isc_log_wouldlog(isc_lctx, level))
		return;

	va_start(ap, fmt);
	vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
	va_end(ap);

	isc_log_write(isc_lctx, category, module, level,
		      "sockmgr %p: %s", sockmgr, msgbuf);
}

static void
socket_log(isc_socket_t *sock, isc_sockaddr_t *address,
	   isc_logcategory_t *category, isc_logmodule_t *module, int level,
	   const char *fmt, ...)
{
	char msgbuf[2048];
	char peerbuf[256];
	va_list ap;

	if (! isc_log_wouldlog(isc_lctx, level))
		return;

	va_start(ap, fmt);
	vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
	va_end(ap);

	if (address == NULL) {
		isc_log_write(isc_lctx, category, module, level,
			      "socket %p: %s", sock, msgbuf);
	} else {
		isc_sockaddr_format(address, peerbuf, sizeof peerbuf);
		isc_log_write(isc_lctx, category, module, level,
			      "socket %p %s: %s", sock, peerbuf, msgbuf);
	}
}

static void
wakeup_socket(isc_socketmgr_t *manager, int fd) {
	isc_event_t *ev2;
	isc_socketevent_t *rev;
	isc_socket_t *sock;

	/*
	 * This is a wakeup on a socket.  Look at the event queue for both
	 * read and write, and decide if we need to watch on it now or not.
	 */
	INSIST(fd < FD_SETSIZE);

	if (manager->fdstate[fd] == CLOSE_PENDING) {
		manager->fdstate[fd] = CLOSED;
		FD_CLR(fd, &manager->read_fds);
		FD_CLR(fd, &manager->write_fds);
		close(fd);
		return;
	}
	if (manager->fdstate[fd] != MANAGED)
		return;

	sock = manager->fds[fd];

	/*
	 * If there are no events, or there is an event but we
	 * have already queued up the internal event on a task's
	 * queue, clear the bit.  Otherwise, set it.
	 */
	rev = ISC_LIST_HEAD(sock->recv_list);
	ev2 = (isc_event_t *) ISC_LIST_HEAD(sock->accept_list);
	if ((rev == NULL && ev2 == NULL)
	    || sock->pending_recv || sock->pending_accept)
		FD_CLR(sock->fd, &manager->read_fds);
	else
		FD_SET(sock->fd, &manager->read_fds);

	rev = ISC_LIST_HEAD(sock->send_list);
	if ((rev == NULL || sock->pending_send) && !sock->connecting)
		FD_CLR(sock->fd, &manager->write_fds);
	else
		FD_SET(sock->fd, &manager->write_fds);
}

#ifdef ISC_PLATFORM_USETHREADS
/*
 * Poke the select loop when there is something for us to do.
 * We assume that if a write completes here, it will be inserted into the
 * queue fully.  That is, we will not get partial writes.
 */
static void
select_poke(isc_socketmgr_t *mgr, int msg) {
	int cc;

	do {
		cc = write(mgr->pipe_fds[1], &msg, sizeof(int));
	} while (cc < 0 && SOFT_ERROR(errno));
			        
	if (cc < 0)
		FATAL_ERROR(__FILE__, __LINE__,
			    "write() failed during watcher poke: %s",
			    strerror(errno));

	INSIST(cc == sizeof(int));
}

/*
 * Read a message on the internal fd.
 */
static int
select_readmsg(isc_socketmgr_t *mgr) {
	int msg;
	int cc;

	cc = read(mgr->pipe_fds[0], &msg, sizeof(int));
	if (cc < 0) {
		if (SOFT_ERROR(errno))
			return (SELECT_POKE_NOTHING);

		FATAL_ERROR(__FILE__, __LINE__,
			    "read() failed during watcher poke: %s",
			    strerror(errno));
		
		return (SELECT_POKE_NOTHING);
	}

	return (msg);
}
#else /* ISC_PLATFORM_USETHREADS */
/*
 * Update the state of the socketmgr when something changes.
 */
static void
select_poke(isc_socketmgr_t *manager, int msg) {
	if (msg == SELECT_POKE_SHUTDOWN)
		return;
	else if (msg >= 0)
		wakeup_socket(manager, msg);
	return;
}
#endif /* ISC_PLATFORM_USETHREADS */

/*
 * Make a fd non-blocking.
 */
static isc_result_t
make_nonblock(int fd) {
	int ret;
	int flags;

	flags = fcntl(fd, F_GETFL, 0);
	flags |= O_NONBLOCK;
	ret = fcntl(fd, F_SETFL, flags);

	if (ret == -1) {
		UNEXPECTED_ERROR(__FILE__, __LINE__,
				 "fcntl(%d, F_SETFL, %d): %s",
				 fd, flags, strerror(errno));

		return (ISC_R_UNEXPECTED);
	}

	return (ISC_R_SUCCESS);
}

/*
 * Process control messages received on a socket.
 */
static void
process_cmsg(isc_socket_t *sock, struct msghdr *msg, isc_socketevent_t *dev) {
#ifdef USE_CMSG
	struct cmsghdr *cmsgp;
#ifdef ISC_PLATFORM_HAVEIPV6
	struct in6_pktinfo *pktinfop;
#endif
#ifdef SO_TIMESTAMP
	struct timeval *timevalp;
#endif
#endif

	/*
	 * sock is used only when ISC_NET_BSD44MSGHDR and USE_CMSG are defined.
	 * msg and dev are used only when ISC_NET_BSD44MSGHDR is defined.
	 * They are all here, outside of the CPP tests, because it is
	 * more consistent with the usual ISC coding style.
	 */
	UNUSED(sock);
	UNUSED(msg);
	UNUSED(dev);

#ifndef ISC_NET_BSD44MSGHDR
	return;

#else  /* defined ISC_NET_BSD44MSGHDR */

#ifdef MSG_TRUNC
	if ((msg->msg_flags & MSG_TRUNC) == MSG_TRUNC)
		dev->attributes |= ISC_SOCKEVENTATTR_TRUNC;
#endif

#ifdef MSG_CTRUNC
	if ((msg->msg_flags & MSG_CTRUNC) == MSG_CTRUNC)
		dev->attributes |= ISC_SOCKEVENTATTR_CTRUNC;
#endif

	/*
	 * Check for multicast.
	 */
	if (isc_sockaddr_ismulticast(&dev->address))
		dev->attributes |= ISC_SOCKEVENTATTR_MULTICAST;

#ifndef USE_CMSG
	return;
#else
	if (msg->msg_controllen == 0 || msg->msg_control == NULL)
		return;

#ifdef SO_TIMESTAMP
	timevalp = NULL;
#endif
#ifdef ISC_PLATFORM_HAVEIPV6
	pktinfop = NULL;
#endif

	cmsgp = CMSG_FIRSTHDR(msg);
	while (cmsgp != NULL) {
		socket_log(sock, NULL, TRACE, "processing cmsg %p", cmsgp);

#ifdef ISC_PLATFORM_HAVEIPV6
		if (cmsgp->cmsg_level == IPPROTO_IPV6
		    && cmsgp->cmsg_type == IPV6_PKTINFO) {

			pktinfop = (struct in6_pktinfo *)CMSG_DATA(cmsgp);
			memcpy(&dev->pktinfo, pktinfop,
			       sizeof(struct in6_pktinfo));
			dev->attributes |= ISC_SOCKEVENTATTR_PKTINFO;
			socket_log(sock, NULL, TRACE,
				   "interface received on ifindex %u",
				   dev->pktinfo.ipi6_ifindex);
			goto next;
		}
#endif

#ifdef SO_TIMESTAMP
		if (cmsgp->cmsg_level == SOL_SOCKET
		    && cmsgp->cmsg_type == SCM_TIMESTAMP) {
			timevalp = (struct timeval *)CMSG_DATA(cmsgp);
			dev->timestamp.seconds = timevalp->tv_sec;
			dev->timestamp.nanoseconds = timevalp->tv_usec * 1000;
			dev->attributes |= ISC_SOCKEVENTATTR_TIMESTAMP;
			goto next;
		}
#endif

	next:
		cmsgp = CMSG_NXTHDR(msg, cmsgp);
	}
#endif /* USE_CMSG */

#endif /* ISC_NET_BSD44MSGHDR */

}

/*
 * Construct an iov array and attach it to the msghdr passed in.  Return
 * 0 on success, non-zero on failure.  This is the SEND constructor, which
 * will used the used region of the buffer (if using a buffer list) or
 * will use the internal region (if a single buffer I/O is requested).
 *
 * Nothing can be NULL, and the done event must list at least one buffer
 * on the buffer linked list for this function to be meaningful.
 *
 * If write_countp != NULL, *write_countp will hold the number of bytes
 * this transaction can send.
 */
static void
build_msghdr_send(isc_socket_t *sock, isc_socketevent_t *dev,
		  struct msghdr *msg, struct iovec *iov, size_t *write_countp)
{
	unsigned int iovcount;
	isc_buffer_t *buffer;
	isc_region_t used;
	size_t write_count;
	size_t skip_count;

	memset(msg, 0, sizeof (*msg));

	if (sock->type == isc_sockettype_udp) {
		msg->msg_name = (void *)&dev->address.type.sa;
		msg->msg_namelen = dev->address.length;
	} else {
		msg->msg_name = NULL;
		msg->msg_namelen = 0;
	}

	buffer = ISC_LIST_HEAD(dev->bufferlist);
	write_count = 0;
	iovcount = 0;

	/*
	 * Single buffer I/O?  Skip what we've done so far in this region.
	 */
	if (buffer == NULL) {
		write_count = dev->region.length - dev->n;
		iov[0].iov_base = (void *)(dev->region.base + dev->n);
		iov[0].iov_len = write_count;
		iovcount = 1;

		goto config;
	}

	/*
	 * Multibuffer I/O.
	 * Skip the data in the buffer list that we have already written.
	 */
	skip_count = dev->n;
	while (buffer != NULL) {
		REQUIRE(ISC_BUFFER_VALID(buffer));
		if (skip_count < isc_buffer_usedlength(buffer))
			break;
		skip_count -= isc_buffer_usedlength(buffer);
		buffer = ISC_LIST_NEXT(buffer, link);
	}

	while (buffer != NULL) {
		INSIST(iovcount < MAXSCATTERGATHER_SEND);

		isc_buffer_usedregion(buffer, &used);

		if (used.length > 0) {
			iov[iovcount].iov_base = (void *)(used.base
							  + skip_count);
			iov[iovcount].iov_len = used.length - skip_count;
			write_count += (used.length - skip_count);
			skip_count = 0;
			iovcount++;
		}
		buffer = ISC_LIST_NEXT(buffer, link);
	}

	INSIST(skip_count == 0);

 config:
	msg->msg_iov = iov;
	msg->msg_iovlen = iovcount;

#ifdef ISC_NET_BSD44MSGHDR
	msg->msg_control = NULL;
	msg->msg_controllen = 0;
	msg->msg_flags = 0;
#if defined(USE_CMSG)
	if ((sock->type == isc_sockettype_udp)
	    && ((dev->attributes & ISC_SOCKEVENTATTR_PKTINFO) != 0)) {
		struct cmsghdr *cmsgp;
		struct in6_pktinfo *pktinfop;

		socket_log(sock, NULL, TRACE,
			   "sendto pktinfo data, ifindex %u",
			   dev->pktinfo.ipi6_ifindex);

		msg->msg_controllen = CMSG_SPACE(sizeof(struct in6_pktinfo));
		msg->msg_control = (void *)sock->cmsg;

		cmsgp = (struct cmsghdr *)sock->cmsg;
		cmsgp->cmsg_level = IPPROTO_IPV6;
		cmsgp->cmsg_type = IPV6_PKTINFO;
		cmsgp->cmsg_len = CMSG_LEN(sizeof(struct in6_pktinfo));
		pktinfop = (struct in6_pktinfo *)CMSG_DATA(cmsgp);
		memcpy(pktinfop, &dev->pktinfo, sizeof(struct in6_pktinfo));
	}
#endif /* USE_CMSG */
#else /* ISC_NET_BSD44MSGHDR */
	msg->msg_accrights = NULL;
	msg->msg_accrightslen = 0;
#endif /* ISC_NET_BSD44MSGHDR */

	if (write_countp != NULL)
		*write_countp = write_count;
}

/*
 * Construct an iov array and attach it to the msghdr passed in.  Return
 * 0 on success, non-zero on failure.  This is the RECV constructor, which
 * will use the avialable region of the buffer (if using a buffer list) or
 * will use the internal region (if a single buffer I/O is requested).
 *
 * Nothing can be NULL, and the done event must list at least one buffer
 * on the buffer linked list for this function to be meaningful.
 *
 * If read_countp != NULL, *read_countp will hold the number of bytes
 * this transaction can receive.
 */
static void
build_msghdr_recv(isc_socket_t *sock, isc_socketevent_t *dev,
		  struct msghdr *msg, struct iovec *iov, size_t *read_countp)
{
	unsigned int iovcount;
	isc_buffer_t *buffer;
	isc_region_t available;
	size_t read_count;

	memset(msg, 0, sizeof (struct msghdr));

	if (sock->type == isc_sockettype_udp) {
		memset(&dev->address, 0, sizeof(dev->address));
		msg->msg_name = (void *)&dev->address.type.sa;
		msg->msg_namelen = sizeof(dev->address.type);
#ifdef ISC_NET_RECVOVERFLOW
		/* If needed, steal one iovec for overflow detection. */
		maxiov--;
#endif
	} else { /* TCP */
		msg->msg_name = NULL;
		msg->msg_namelen = 0;
		dev->address = sock->address;
	}

	buffer = ISC_LIST_HEAD(dev->bufferlist);
	read_count = 0;

	/*
	 * Single buffer I/O?  Skip what we've done so far in this region.
	 */
	if (buffer == NULL) {
		read_count = dev->region.length - dev->n;
		iov[0].iov_base = (void *)(dev->region.base + dev->n);
		iov[0].iov_len = read_count;
		iovcount = 1;

		goto config;
	}

	/*
	 * Multibuffer I/O.
	 * Skip empty buffers.
	 */
	while (buffer != NULL) {
		REQUIRE(ISC_BUFFER_VALID(buffer));
		if (isc_buffer_availablelength(buffer) != 0)
			break;
		buffer = ISC_LIST_NEXT(buffer, link);
	}

	iovcount = 0;
	while (buffer != NULL) {
		INSIST(iovcount < MAXSCATTERGATHER_RECV);

		isc_buffer_availableregion(buffer, &available);

		if (available.length > 0) {
			iov[iovcount].iov_base = (void *)(available.base);
			iov[iovcount].iov_len = available.length;
			read_count += available.length;
			iovcount++;
		}
		buffer = ISC_LIST_NEXT(buffer, link);
	}

 config:

	/*
	 * If needed, set up to receive that one extra byte.  Note that
	 * we know there is at least one iov left, since we stole it
	 * at the top of this function.
	 */
#ifdef ISC_NET_RECVOVERFLOW
	if (sock->type == isc_sockettype_udp) {
		iov[iovcount].iov_base = (void *)(&sock->overflow);
		iov[iovcount].iov_len = 1;
		iovcount++;
	}
#endif

	msg->msg_iov = iov;
	msg->msg_iovlen = iovcount;

#ifdef ISC_NET_BSD44MSGHDR
	msg->msg_control = NULL;
	msg->msg_controllen = 0;
	msg->msg_flags = 0;
#if defined(USE_CMSG)
	if (sock->type == isc_sockettype_udp) {
		msg->msg_control = (void *)sock->cmsg;
		msg->msg_controllen = sock->cmsglen;
	}
#endif /* USE_CMSG */
#else /* ISC_NET_BSD44MSGHDR */
	msg->msg_accrights = NULL;
	msg->msg_accrightslen = 0;
#endif /* ISC_NET_BSD44MSGHDR */

	if (read_countp != NULL)
		*read_countp = read_count;
}

static void
set_dev_address(isc_sockaddr_t *address, isc_socket_t *sock,
		isc_socketevent_t *dev)
{
	if (sock->type == isc_sockettype_udp) {
		if (address != NULL)
			dev->address = *address;
		else
			dev->address = sock->address;
	} else if (sock->type == isc_sockettype_tcp) {
		INSIST(address == NULL);
		dev->address = sock->address;
	}
}

static isc_socketevent_t *
allocate_socketevent(isc_socket_t *sock, isc_eventtype_t eventtype,
		     isc_taskaction_t action, const void *arg)
{
	isc_socketevent_t *ev;

	ev = (isc_socketevent_t *)isc_event_allocate(sock->manager->mctx,
						     sock, eventtype,
						     action, arg,
						     sizeof (*ev));

	if (ev == NULL)
		return (NULL);

	ev->result = ISC_R_UNEXPECTED;
	ISC_LINK_INIT(ev, ev_link);
	ISC_LIST_INIT(ev->bufferlist);
	ev->region.base = NULL;
	ev->n = 0;
	ev->offset = 0;
	ev->attributes = 0;

	return (ev);
}

#if defined(ISC_SOCKET_DEBUG)
static void
dump_msg(struct msghdr *msg) {
	unsigned int i;

	printf("MSGHDR %p\n", msg);
	printf("\tname %p, namelen %d\n", msg->msg_name, msg->msg_namelen);
	printf("\tiov %p, iovlen %d\n", msg->msg_iov, msg->msg_iovlen);
	for (i = 0 ; i < (unsigned int)msg->msg_iovlen ; i++)
		printf("\t\t%d\tbase %p, len %d\n", i,
		       msg->msg_iov[i].iov_base,
		       msg->msg_iov[i].iov_len);
#ifdef ISC_NET_BSD44MSGHDR
	printf("\tcontrol %p, controllen %d\n", msg->msg_control,
	       msg->msg_controllen);
#endif
}
#endif

#define DOIO_SUCCESS		0	/* i/o ok, event sent */
#define DOIO_SOFT		1	/* i/o ok, soft error, no event sent */
#define DOIO_HARD		2	/* i/o error, event sent */
#define DOIO_EOF		3	/* EOF, no event sent */

static int
doio_recv(isc_socket_t *sock, isc_socketevent_t *dev) {
	int cc;
	struct iovec iov[MAXSCATTERGATHER_RECV];
	size_t read_count;
	size_t actual_count;
	struct msghdr msghdr;
	isc_buffer_t *buffer;

	build_msghdr_recv(sock, dev, &msghdr, iov, &read_count);

#if defined(ISC_SOCKET_DEBUG)
	dump_msg(&msghdr);
#endif

	cc = recvmsg(sock->fd, &msghdr, 0);

	if (cc < 0) {
		if (SOFT_ERROR(errno))
			return (DOIO_SOFT);

		if (isc_log_wouldlog(isc_lctx, IOEVENT_LEVEL))
			socket_log(sock, NULL, IOEVENT,
				   "doio_recv: recvmsg(%d) %d bytes, err %d/%s",
				   sock->fd, cc, errno, strerror(errno));

#define SOFT_OR_HARD(_system, _isc) \
	if (errno == _system) { \
		if (sock->connected) { \
			send_recvdone_event(sock, &dev, _isc); \
			return (DOIO_HARD); \
		} \
		return (DOIO_SOFT); \
	}
#define ALWAYS_HARD(_system, _isc) \
	if (errno == _system) { \
		send_recvdone_event(sock, &dev, _isc); \
		return (DOIO_HARD); \
	}

		SOFT_OR_HARD(ECONNREFUSED, ISC_R_CONNREFUSED);
		SOFT_OR_HARD(ENETUNREACH, ISC_R_NETUNREACH);
		SOFT_OR_HARD(EHOSTUNREACH, ISC_R_HOSTUNREACH);
		ALWAYS_HARD(ENOBUFS, ISC_R_NORESOURCES);

#undef SOFT_OR_HARD
#undef ALWAYS_HARD

		send_recvdone_event(sock, &dev, ISC_R_UNEXPECTED);
		return (DOIO_SUCCESS);
	}

	/*
	 * On TCP, zero length reads indicate EOF, while on
	 * UDP, zero length reads are perfectly valid, although
	 * strange.
	 */
	if ((sock->type == isc_sockettype_tcp) && (cc == 0))
		return (DOIO_EOF);

	if (sock->type == isc_sockettype_udp)
		dev->address.length = msghdr.msg_namelen;

	socket_log(sock, &dev->address, IOEVENT, "packet received correctly");

	/*
	 * Overflow bit detection.  If we received MORE bytes than we should,
	 * this indicates an overflow situation.  Set the flag in the
	 * dev entry and adjust how much we read by one.
	 */
#ifdef ISC_NET_RECVOVERFLOW
	if ((sock->type == isc_sockettype_udp) && ((size_t)cc > read_count)) {
		dev->attributes |= ISC_SOCKEVENTATTR_TRUNC;
		cc--;
	}
#endif

	/*
	 * If there are control messages attached, run through them and pull
	 * out the interesting bits.
	 */
	if (sock->type == isc_sockettype_udp)
		process_cmsg(sock, &msghdr, dev);

	/*
	 * update the buffers (if any) and the i/o count
	 */
	dev->n += cc;
	actual_count = cc;
	buffer = ISC_LIST_HEAD(dev->bufferlist);
	while (buffer != NULL && actual_count > 0) {
		REQUIRE(ISC_BUFFER_VALID(buffer));
		if (isc_buffer_availablelength(buffer) <= actual_count) {
			actual_count -= isc_buffer_availablelength(buffer);
			isc_buffer_add(buffer,
				       isc_buffer_availablelength(buffer));
		} else {
			isc_buffer_add(buffer, actual_count);
			actual_count = 0;
			break;
		}
		buffer = ISC_LIST_NEXT(buffer, link);
		if (buffer == NULL) {
			INSIST(actual_count == 0);
		}
	}

	/*
	 * If we read less than we expected, update counters,
	 * and let the upper layer poke the descriptor.
	 */
	if (((size_t)cc != read_count) && (dev->n < dev->minimum))
		return (DOIO_SOFT);

	/*
	 * Full reads are posted, or partials if partials are ok.
	 */
	send_recvdone_event(sock, &dev, ISC_R_SUCCESS);
	return (DOIO_SUCCESS);
}

/*
 * Returns:
 *	DOIO_SUCCESS	The operation succeeded.  The senddone event
 *			was sent.
 *
 *	DOIO_HARD	A hard or unexpected I/O error was encountered.
 *			The senddone event was sent.
 *
 *	DOIO_SOFT	A soft I/O error was encountered.  No senddone
 *			event was sent.  The operation should be retried.
 *
 *	No other return values are possible.
 */
static int
doio_send(isc_socket_t *sock, isc_socketevent_t *dev) {
	int cc;
	struct iovec iov[MAXSCATTERGATHER_SEND];
	size_t write_count;
	struct msghdr msghdr;
	char addrbuf[ISC_SOCKADDR_FORMATSIZE];

	build_msghdr_send(sock, dev, &msghdr, iov, &write_count);

	cc = sendmsg(sock->fd, &msghdr, 0);

	/*
	 * Check for error or block condition.
	 */
	if (cc < 0) {
		if (SOFT_ERROR(errno))
			return (DOIO_SOFT);

#define SOFT_OR_HARD(_system, _isc) \
	if (errno == _system) { \
		if (sock->connected) { \
			send_senddone_event(sock, &dev, _isc); \
			return (DOIO_HARD); \
		} \
		return (DOIO_SOFT); \
	}
#define ALWAYS_HARD(_system, _isc) \
	if (errno == _system) { \
		send_senddone_event(sock, &dev, _isc); \
		return (DOIO_HARD); \
	}

		SOFT_OR_HARD(EACCES, ISC_R_NOPERM);
		SOFT_OR_HARD(EAFNOSUPPORT, ISC_R_ADDRNOTAVAIL);
		SOFT_OR_HARD(ECONNREFUSED, ISC_R_CONNREFUSED);
		ALWAYS_HARD(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
		ALWAYS_HARD(EHOSTUNREACH, ISC_R_HOSTUNREACH);
#ifdef EHOSTDOWN
		ALWAYS_HARD(EHOSTDOWN, ISC_R_HOSTUNREACH);
#endif
		ALWAYS_HARD(ENETUNREACH, ISC_R_NETUNREACH);
		ALWAYS_HARD(ENOBUFS, ISC_R_NORESOURCES);
		ALWAYS_HARD(EPERM, ISC_R_HOSTUNREACH);
		ALWAYS_HARD(EPIPE, ISC_R_NOTCONNECTED);

#undef SOFT_OR_HARD
#undef ALWAYS_HARD

		/*
		 * The other error types depend on whether or not the
		 * socket is UDP or TCP.  If it is UDP, some errors
		 * that we expect to be fatal under TCP are merely
		 * annoying, and are really soft errors.
		 *
		 * However, these soft errors are still returned as
		 * a status.
		 */
		isc_sockaddr_format(&dev->address, addrbuf, sizeof(addrbuf));
		UNEXPECTED_ERROR(__FILE__, __LINE__,
				 "internal_send: %s: %s",
				 addrbuf, strerror(errno));
		send_senddone_event(sock, &dev, ISC_R_UNEXPECTED);
		return (DOIO_HARD);
	}

	if (cc == 0)
		UNEXPECTED_ERROR(__FILE__, __LINE__,
				 "internal_send: send() returned 0");

	/*
	 * If we write less than we expected, update counters, poke.
	 */
	dev->n += cc;
	if ((size_t)cc != write_count)
		return (DOIO_SOFT);

	/*
	 * Exactly what we wanted to write.  We're done with this
	 * entry.  Post its completion event.
	 */
	send_senddone_event(sock, &dev, ISC_R_SUCCESS);
	return (DOIO_SUCCESS);
}

/*
 * Kill.
 *
 * Caller must ensure that the socket is not locked and no external
 * references exist.
 */
static void
destroy(isc_socket_t **sockp) {
	isc_socket_t *sock = *sockp;
	isc_socketmgr_t *manager = sock->manager;

	socket_log(sock, NULL, CREATION, "destroying");

	INSIST(ISC_LIST_EMPTY(sock->accept_list));
	INSIST(ISC_LIST_EMPTY(sock->recv_list));
	INSIST(ISC_LIST_EMPTY(sock->send_list));
	INSIST(sock->connect_ev == NULL);

	LOCK(&manager->lock);

	/*
	 * No one has this socket open, so the watcher doesn't have to be
	 * poked, and the socket doesn't have to be locked.
	 */
	manager->fds[sock->fd] = NULL;
	manager->fdstate[sock->fd] = CLOSE_PENDING;
	select_poke(manager, sock->fd);
	ISC_LIST_UNLINK(manager->socklist, sock, link);

#ifdef ISC_PLATFORM_USETHREADS
	if (ISC_LIST_EMPTY(manager->socklist))
		SIGNAL(&manager->shutdown_ok);
#endif /* ISC_PLATFORM_USETHREADS */

	/*
	 * XXX should reset manager->maxfd here
	 */

	UNLOCK(&manager->lock);

	free_socket(sockp);
}

static isc_result_t
allocate_socket(isc_socketmgr_t *manager, isc_sockettype_t type,
		isc_socket_t **socketp)
{
	isc_socket_t *sock;
	isc_result_t ret;

	sock = isc_mem_get(manager->mctx, sizeof *sock);

	if (sock == NULL)
		return (ISC_R_NOMEMORY);

#if USE_CMSG  /* Let's hope the OSs are sane, and pad correctly XXXMLG */
	sock->cmsglen = 0;
#ifdef ISC_PLATFORM_HAVEIPV6
	sock->cmsglen += CMSG_SPACE(sizeof(struct in6_pktinfo));
#endif
#ifdef SO_TIMESTAMP
	sock->cmsglen += CMSG_SPACE(sizeof(struct timeval));
#endif
	sock->cmsg = isc_mem_get(manager->mctx, sock->cmsglen);
	if (sock->cmsg == NULL) {
		ret = ISC_R_NOMEMORY;
		goto err1;
	}
#endif

	ret = ISC_R_UNEXPECTED;

	sock->magic = 0;
	sock->references = 0;

	sock->manager = manager;
	sock->type = type;
	sock->fd = -1;

	ISC_LINK_INIT(sock, link);

	/*
	 * set up list of readers and writers to be initially empty
	 */
	ISC_LIST_INIT(sock->recv_list);
	ISC_LIST_INIT(sock->send_list);
	ISC_LIST_INIT(sock->accept_list);
	sock->connect_ev = NULL;
	sock->pending_recv = 0;
	sock->pending_send = 0;
	sock->pending_accept = 0;
	sock->listener = 0;
	sock->connected = 0;
	sock->connecting = 0;
	sock->bound = 0;

	/*