/ext/socket/socket.c

Large files files are truncated, but you can click here to view the full file

/************************************************

  socket.c -

  $Author: akr $
  created at: Thu Mar 31 12:21:29 JST 1994

  Copyright (C) 1993-2007 Yukihiro Matsumoto

************************************************/

#include "macruby_internal.h"
#include "ruby/io.h"
#include "ruby/signal.h"
#include "ruby/util.h"
#include <stdio.h>
#include <sys/types.h>
#include <sys/stat.h>

#ifdef HAVE_UNISTD_H
#include <unistd.h>
#endif

#ifdef HAVE_SYS_UIO_H
#include <sys/uio.h>
#endif

#ifdef HAVE_XTI_H
#include <xti.h>
#endif

#ifndef _WIN32
#if defined(__BEOS__)
# include <net/socket.h>
#else
# include <sys/socket.h>
#endif
#include <netinet/in.h>
#ifdef HAVE_NETINET_IN_SYSTM_H
# include <netinet/in_systm.h>
#endif
#ifdef HAVE_NETINET_TCP_H
# include <netinet/tcp.h>
#endif
#ifdef HAVE_NETINET_UDP_H
# include <netinet/udp.h>
#endif
#ifdef HAVE_ARPA_INET_H
# include <arpa/inet.h>
#endif
#include <netdb.h>
#endif
#include <errno.h>
#ifdef HAVE_SYS_UN_H
#include <sys/un.h>
#endif

#if defined(HAVE_FCNTL)
#ifdef HAVE_SYS_SELECT_H
#include <sys/select.h>
#endif
#ifdef HAVE_SYS_TYPES_H
#include <sys/types.h>
#endif
#ifdef HAVE_SYS_TIME_H
#include <sys/time.h>
#endif
#ifdef HAVE_FCNTL_H
#include <fcntl.h>
#endif
#endif

#ifdef HAVE_IFADDRS_H
#include <ifaddrs.h>
#endif
#ifdef HAVE_SYS_IOCTL_H
#include <sys/ioctl.h>
#endif
#ifdef HAVE_SYS_SOCKIO_H
#include <sys/sockio.h>
#endif
#ifdef HAVE_NET_IF_H
#include <net/if.h>
#endif

#ifdef HAVE_SYS_PARAM_H
#include <sys/param.h>
#endif
#ifdef HAVE_SYS_UCRED_H
#include <sys/ucred.h>
#endif
#ifdef HAVE_UCRED_H
#include <ucred.h>
#endif

#ifndef EWOULDBLOCK
#define EWOULDBLOCK EAGAIN
#endif
#ifndef HAVE_GETADDRINFO
# include "addrinfo.h"
#endif
#include "sockport.h"


static int do_not_reverse_lookup = 1;
#define FMODE_NOREVLOOKUP 0x100

VALUE rb_cBasicSocket;
VALUE rb_cIPSocket;
VALUE rb_cTCPSocket;
VALUE rb_cTCPServer;
VALUE rb_cUDPSocket;
#ifdef AF_UNIX
VALUE rb_cUNIXSocket;
VALUE rb_cUNIXServer;
#endif
VALUE rb_cSocket;
static VALUE rb_cAddrinfo;

static VALUE rb_eSocket;

#define INET_CLIENT 0
#define INET_SERVER 1
#define INET_SOCKS  2

#ifndef NI_MAXHOST
# define NI_MAXHOST 1025
#endif
#ifndef NI_MAXSERV
# define NI_MAXSERV 32
#endif

#ifdef AF_INET6
# define IS_IP_FAMILY(af) ((af) == AF_INET || (af) == AF_INET6)
#else
# define IS_IP_FAMILY(af) ((af) == AF_INET)
#endif

#ifndef HAVE_SOCKADDR_STORAGE
/*
 * RFC 2553: protocol-independent placeholder for socket addresses
 */
#define _SS_MAXSIZE	128
#define _SS_ALIGNSIZE	(sizeof(double))
#define _SS_PAD1SIZE	(_SS_ALIGNSIZE - sizeof(unsigned char) * 2)
#define _SS_PAD2SIZE	(_SS_MAXSIZE - sizeof(unsigned char) * 2 - \
				_SS_PAD1SIZE - _SS_ALIGNSIZE)

struct sockaddr_storage {
#ifdef HAVE_SA_LEN
	unsigned char ss_len;		/* address length */
	unsigned char ss_family;	/* address family */
#else
	unsigned short ss_family;
#endif
	char	__ss_pad1[_SS_PAD1SIZE];
	double	__ss_align;	/* force desired structure storage alignment */
	char	__ss_pad2[_SS_PAD2SIZE];
};
#endif

static void sock_define_const(char *name, int value);
static void sock_define_uconst(const char *name, unsigned int value);
#include "constants.h"

static int str_is_number(const char *);

/* fix [ruby-core:29427] */
static int
ruby_getaddrinfo__darwin(const char *nodename, const char *servname,
			 struct addrinfo *hints, struct addrinfo **res)
{
    const char *tmp_servname;
    struct addrinfo tmp_hints;
    tmp_servname = servname;
    MEMCPY(&tmp_hints, hints, struct addrinfo, 1);
    if (nodename && servname) {
	if (str_is_number(tmp_servname) && atoi(servname) == 0) {
	    tmp_servname = NULL;
#ifdef AI_NUMERICSERV
	    if (tmp_hints.ai_flags) tmp_hints.ai_flags &= ~AI_NUMERICSERV;
#endif
	}
    }
    int error = getaddrinfo(nodename, tmp_servname, &tmp_hints, res);
    return error;
}
#undef getaddrinfo
#define getaddrinfo(node,serv,hints,res) ruby_getaddrinfo__darwin((node),(serv),(hints),(res))

#if defined(INET6) && (defined(LOOKUP_ORDER_HACK_INET) || defined(LOOKUP_ORDER_HACK_INET6))
#define LOOKUP_ORDERS		3
static int lookup_order_table[LOOKUP_ORDERS] = {
#if defined(LOOKUP_ORDER_HACK_INET)
    PF_INET, PF_INET6, PF_UNSPEC,
#elif defined(LOOKUP_ORDER_HACK_INET6)
    PF_INET6, PF_INET, PF_UNSPEC,
#else
    /* should not happen */
#endif
};

static int
ruby_getaddrinfo(char *nodename, char *servname,
		 struct addrinfo *hints, struct addrinfo **res)
{
    struct addrinfo tmp_hints;
    int i, af, error;

    if (hints->ai_family != PF_UNSPEC) {
	return getaddrinfo(nodename, servname, hints, res);
    }

    for (i = 0; i < LOOKUP_ORDERS; i++) {
	af = lookup_order_table[i];
	MEMCPY(&tmp_hints, hints, struct addrinfo, 1);
	tmp_hints.ai_family = af;
	error = getaddrinfo(nodename, servname, &tmp_hints, res);
	if (error) {
	    if (tmp_hints.ai_family == PF_UNSPEC) {
		break;
	    }
	}
	else {
	    break;
	}
    }

    return error;
}
#define getaddrinfo(node,serv,hints,res) ruby_getaddrinfo((node),(serv),(hints),(res))
#endif

#if defined(_AIX)
static int
ruby_getaddrinfo__aix(char *nodename, char *servname,
		      struct addrinfo *hints, struct addrinfo **res)
{
    int error = getaddrinfo(nodename, servname, hints, res);
    struct addrinfo *r;
    if (error)
	return error;
    for (r = *res; r != NULL; r = r->ai_next) {
	if (r->ai_addr->sa_family == 0)
	    r->ai_addr->sa_family = r->ai_family;
	if (r->ai_addr->sa_len == 0)
	    r->ai_addr->sa_len = r->ai_addrlen;
    }
    return 0;
}
#undef getaddrinfo
#define getaddrinfo(node,serv,hints,res) ruby_getaddrinfo__aix((node),(serv),(hints),(res))
static int
ruby_getnameinfo__aix(const struct sockaddr *sa, size_t salen,
		      char *host, size_t hostlen,
		      char *serv, size_t servlen, int flags)
{
    struct sockaddr_in6 *sa6;
    u_int32_t *a6;

    if (sa->sa_family == AF_INET6) {
	sa6 = (struct sockaddr_in6 *)sa;
	a6 = sa6->sin6_addr.u6_addr.u6_addr32;

	if (a6[0] == 0 && a6[1] == 0 && a6[2] == 0 && a6[3] == 0) {
	    strncpy(host, "::", hostlen);
	    snprintf(serv, servlen, "%d", sa6->sin6_port);
	    return 0;
	}
    }
    return getnameinfo(sa, salen, host, hostlen, serv, servlen, flags);
}
#undef getnameinfo
#define getnameinfo(sa, salen, host, hostlen, serv, servlen, flags) \
            ruby_getnameinfo__aix((sa), (salen), (host), (hostlen), (serv), (servlen), (flags))

#ifndef CMSG_SPACE
# define CMSG_SPACE(len) (_CMSG_ALIGN(sizeof(struct cmsghdr)) + _CMSG_ALIGN(len))
#endif
#ifndef CMSG_LEN
# define CMSG_LEN(len) (_CMSG_ALIGN(sizeof(struct cmsghdr)) + (len))
#endif
#endif

#ifdef __BEOS__
#undef close
#define close closesocket
#endif

#define MakeOpenFile(obj, fp) \
    do { \
	fp = ALLOC(rb_io_t); \
	GC_WB(&RFILE(obj)->fptr, fp); \
	fp->fd = fp->read_fd = fp->write_fd = -1; \
	fp->pid = -1; \
    } \
    while (0)

static int
constant_arg(VALUE arg, int (*str_to_int)(char*, int, int*), const char *errmsg)
{
    VALUE tmp;
    char *ptr;
    int ret;

    if (SYMBOL_P(arg)) {
        arg = rb_sym_to_s(arg);
        goto str;
    }
    else if (!NIL_P(tmp = rb_check_string_type(arg))) {
	arg = tmp;
      str:
	rb_check_safe_obj(arg);
        ptr = RSTRING_PTR(arg);
        if (str_to_int(ptr, RSTRING_LEN(arg), &ret) == -1)
	    rb_raise(rb_eSocket, "%s: %s", errmsg, ptr);
    }
    else {
	ret = NUM2INT(arg);
    }
    return ret;
}

static int
family_arg(VALUE domain)
{
    /* convert AF_INET, etc. */
    return constant_arg(domain, family_to_int, "unknown socket domain");
}

static int
socktype_arg(VALUE type)
{
    /* convert SOCK_STREAM, etc. */
    return constant_arg(type, socktype_to_int, "unknown socket type");
}

static int
level_arg(VALUE level)
{
    /* convert SOL_SOCKET, IPPROTO_TCP, etc. */
    return constant_arg(level, level_to_int, "unknown protocol level");
}

static int
optname_arg(int level, VALUE optname)
{
    switch (level) {
      case SOL_SOCKET:
        return constant_arg(optname, so_optname_to_int, "unknown socket level option name");
      case IPPROTO_IP:
        return constant_arg(optname, ip_optname_to_int, "unknown IP level option name");
#ifdef IPPROTO_IPV6
      case IPPROTO_IPV6:
        return constant_arg(optname, ipv6_optname_to_int, "unknown IPv6 level option name");
#endif
      case IPPROTO_TCP:
        return constant_arg(optname, tcp_optname_to_int, "unknown TCP level option name");
      case IPPROTO_UDP:
        return constant_arg(optname, udp_optname_to_int, "unknown UDP level option name");
      default:
        return NUM2INT(optname);
    }
}

static int
shutdown_how_arg(VALUE how)
{
    /* convert SHUT_RD, SHUT_WR, SHUT_RDWR. */
    return constant_arg(how, shutdown_how_to_int, "unknown shutdown argument");
}

static VALUE
init_sock(VALUE sock, int fd)
{
    rb_io_t *fp;

#ifdef S_ISSOCK
    struct stat sbuf;
    if (fstat(fd, &sbuf) < 0) {
        rb_sys_fail(0);
    }
    if (!S_ISSOCK(sbuf.st_mode)) {
        rb_raise(rb_eArgError, "not a socket file descriptor");
    }
#endif

    MakeOpenFile(sock, fp);
    fp->fd = fp->read_fd = fp->write_fd = fd;
    fp->mode = FMODE_READWRITE|FMODE_DUPLEX;
    rb_io_ascii8bit_binmode(sock);
    if (do_not_reverse_lookup) {
	fp->mode |= FMODE_NOREVLOOKUP;
    }
    fp->mode |= FMODE_SYNC;

    return sock;
}

/*
 * call-seq:
 *   BasicSocket.for_fd(fd) => basicsocket
 *
 * Returns a socket object which contains the file descriptor, _fd_.
 *
 *   # If invoked by inetd, STDIN/STDOUT/STDERR is a socket.
 *   STDIN_SOCK = Socket.for_fd(STDIN.fileno)
 *   p STDIN_SOCK.remote_address
 *
 */
static VALUE
bsock_s_for_fd(VALUE klass, SEL sel, VALUE fd)
{
    rb_io_t *fptr;
    VALUE sock = init_sock(rb_obj_alloc(klass), NUM2INT(fd));

    GetOpenFile(sock, fptr);

    return sock;
}

/*
 * call-seq:
 *   basicsocket.shutdown([how]) => 0
 *
 * Calls shutdown(2) system call.
 *
 * s.shutdown(Socket::SHUT_RD) disallows further read.
 *
 * s.shutdown(Socket::SHUT_WR) disallows further write.
 *
 * s.shutdown(Socket::SHUT_RDWR) disallows further read and write.
 *
 * _how_ can be symbol or string:
 * - :RD, :SHUT_RD, "RD" and "SHUT_RD" are accepted as Socket::SHUT_RD.
 * - :WR, :SHUT_WR, "WR" and "SHUT_WR" are accepted as Socket::SHUT_WR.
 * - :RDWR, :SHUT_RDWR, "RDWR" and "SHUT_RDWR" are accepted as Socket::SHUT_RDWR.
 *
 *   UNIXSocket.pair {|s1, s2|
 *     s1.puts "ping"
 *     s1.shutdown(:WR)
 *     p s2.read          #=> "ping\n"
 *     s2.puts "pong"
 *     s2.close
 *     p s1.read          #=> "pong\n"
 *   }
 *
 */
static VALUE
bsock_shutdown(VALUE sock, SEL sel, int argc, VALUE *argv)
{
    VALUE howto;
    int how;
    rb_io_t *fptr;

    if (rb_safe_level() >= 4 && !OBJ_TAINTED(sock)) {
	rb_raise(rb_eSecurityError, "Insecure: can't shutdown socket");
    }
    rb_scan_args(argc, argv, "01", &howto);
    if (howto == Qnil)
	how = SHUT_RDWR;
    else {
	how = shutdown_how_arg(howto);
	if (how != SHUT_WR && how != SHUT_RD && how != SHUT_RDWR) {
	    rb_raise(rb_eArgError, "`how' should be either :SHUT_RD, :SHUT_WR, :SHUT_RDWR");
	}
    }
    GetOpenFile(sock, fptr);
    if (shutdown(fptr->fd, how) == -1)
	rb_sys_fail(0);

    return INT2FIX(0);
}

/*
 * call-seq:
 *   basicsocket.close_read => nil
 *
 * Disallows further read using shutdown system call.
 *
 *   s1, s2 = UNIXSocket.pair
 *   s1.close_read
 *   s2.puts #=> Broken pipe (Errno::EPIPE)
 */
static VALUE
bsock_close_read(VALUE sock, SEL sel)
{
    rb_io_t *fptr;

    if (rb_safe_level() >= 4 && !OBJ_TAINTED(sock)) {
	rb_raise(rb_eSecurityError, "Insecure: can't close socket");
    }
    GetOpenFile(sock, fptr);
    shutdown(fptr->fd, 0);
    if (!(fptr->mode & FMODE_WRITABLE)) {
	return rb_io_close(sock);
    }
    fptr->mode &= ~FMODE_READABLE;
    fptr->read_fd = -1;

    return Qnil;
}

/*
 * call-seq:
 *   basicsocket.close_write => nil
 *
 * Disallows further write using shutdown system call.
 *
 *   UNIXSocket.pair {|s1, s2|
 *     s1.print "ping"
 *     s1.close_write
 *     p s2.read        #=> "ping"
 *     s2.print "pong"
 *     s2.close
 *     p s1.read        #=> "pong"
 *   }
 */
static VALUE
bsock_close_write(VALUE sock, SEL sel)
{
    rb_io_t *fptr;

    if (rb_safe_level() >= 4 && !OBJ_TAINTED(sock)) {
	rb_raise(rb_eSecurityError, "Insecure: can't close socket");
    }
    GetOpenFile(sock, fptr);
    if (!(fptr->mode & FMODE_READABLE)) {
	return rb_io_close(sock);
    }
    shutdown(fptr->fd, 1);
    fptr->mode &= ~FMODE_WRITABLE;
    fptr->write_fd = -1;

    return Qnil;
}

/*
 * Document-method: setsockopt
 * call-seq:
 *   setsockopt(level, optname, optval)
 *   setsockopt(socketoption)
 *
 * Sets a socket option. These are protocol and system specific, see your
 * local system documentation for details.
 *
 * === Parameters
 * * +level+ is an integer, usually one of the SOL_ constants such as
 *   Socket::SOL_SOCKET, or a protocol level.
 *   A string or symbol of the name, possibly without prefix, is also
 *   accepted.
 * * +optname+ is an integer, usually one of the SO_ constants, such
 *   as Socket::SO_REUSEADDR.
 *   A string or symbol of the name, possibly without prefix, is also
 *   accepted.
 * * +optval+ is the value of the option, it is passed to the underlying
 *   setsockopt() as a pointer to a certain number of bytes. How this is
 *   done depends on the type:
 *   - Fixnum: value is assigned to an int, and a pointer to the int is
 *     passed, with length of sizeof(int).
 *   - true or false: 1 or 0 (respectively) is assigned to an int, and the
 *     int is passed as for a Fixnum. Note that +false+ must be passed,
 *     not +nil+.
 *   - String: the string's data and length is passed to the socket.
 * * +socketoption+ is an instance of Socket::Option
 *
 * === Examples
 *
 * Some socket options are integers with boolean values, in this case
 * #setsockopt could be called like this:
 *   sock.setsockopt(:SOCKET, :REUSEADDR, true)
 *   sock.setsockopt(Socket::SOL_SOCKET,Socket::SO_REUSEADDR, true)
 *   sock.setsockopt(Socket::Option.bool(:INET, :SOCKET, :REUSEADDR, true))
 *
 * Some socket options are integers with numeric values, in this case
 * #setsockopt could be called like this:
 *   sock.setsockopt(:IP, :TTL, 255)
 *   sock.setsockopt(Socket::IPPROTO_IP, Socket::IP_TTL, 255)
 *   sock.setsockopt(Socket::Option.int(:INET, :IP, :TTL, 255))
 *
 * Option values may be structs. Passing them can be complex as it involves
 * examining your system headers to determine the correct definition. An
 * example is an +ip_mreq+, which may be defined in your system headers as:
 *   struct ip_mreq {
 *     struct  in_addr imr_multiaddr;
 *     struct  in_addr imr_interface;
 *   };
 *
 * In this case #setsockopt could be called like this:
 *   optval = IPAddr.new("224.0.0.251").hton +
 *            IPAddr.new(Socket::INADDR_ANY, Socket::AF_INET).hton
 *   sock.setsockopt(Socket::IPPROTO_IP, Socket::IP_ADD_MEMBERSHIP, optval)
 *
*/
static VALUE
bsock_setsockopt(VALUE sock, SEL sel, VALUE lev, VALUE optname, VALUE val)
{
    int level, option;
    rb_io_t *fptr;
    int i;
    const char *v;
    int vlen;

    rb_secure(2);
    level = level_arg(lev);
    option = optname_arg(level, optname);

    switch (TYPE(val)) {
      case T_FIXNUM:
	i = FIX2INT(val);
	goto numval;
      case T_FALSE:
	i = 0;
	goto numval;
      case T_TRUE:
	i = 1;
      numval:
	v = (char*)&i; vlen = (int)sizeof(i);
	break;
      default:
	StringValue(val);
	v = RSTRING_PTR(val);
	vlen = RSTRING_LENINT(val);
	break;
    }

#define rb_sys_fail_path(path) rb_sys_fail(path == 0 ? NULL : RSTRING_PTR(path));

    GetOpenFile(sock, fptr);
    if (setsockopt(fptr->fd, level, option, v, vlen) < 0)
	rb_sys_fail_path(fptr->path);

    return INT2FIX(0);
}

/*
 * Document-method: getsockopt
 * call-seq:
 *   getsockopt(level, optname) => socketoption
 *
 * Gets a socket option. These are protocol and system specific, see your
 * local system documentation for details. The option is returned as
 * a Socket::Option object.
 *
 * === Parameters
 * * +level+ is an integer, usually one of the SOL_ constants such as
 *   Socket::SOL_SOCKET, or a protocol level.
 *   A string or symbol of the name, possibly without prefix, is also
 *   accepted.
 * * +optname+ is an integer, usually one of the SO_ constants, such
 *   as Socket::SO_REUSEADDR.
 *   A string or symbol of the name, possibly without prefix, is also
 *   accepted.
 *
 * === Examples
 *
 * Some socket options are integers with boolean values, in this case
 * #getsockopt could be called like this:
 *
 *   reuseaddr = sock.getsockopt(:SOCKET, :REUSEADDR).bool
 *
 *   optval = sock.getsockopt(Socket::SOL_SOCKET,Socket::SO_REUSEADDR)
 *   optval = optval.unpack "i"
 *   reuseaddr = optval[0] == 0 ? false : true
 *
 * Some socket options are integers with numeric values, in this case
 * #getsockopt could be called like this:
 *
 *   ipttl = sock.getsockopt(:IP, :TTL).int
 *
 *   optval = sock.getsockopt(Socket::IPPROTO_IP, Socket::IP_TTL)
 *   ipttl = optval.unpack("i")[0]
 *
 * Option values may be structs. Decoding them can be complex as it involves
 * examining your system headers to determine the correct definition. An
 * example is a +struct linger+, which may be defined in your system headers
 * as:
 *   struct linger {
 *     int l_onoff;
 *     int l_linger;
 *   };
 *
 * In this case #getsockopt could be called like this:
 *
 *   # Socket::Option knows linger structure.
 *   onoff, linger = sock.getsockopt(:SOCKET, :LINGER).linger
 *
 *   optval =  sock.getsockopt(Socket::SOL_SOCKET, Socket::SO_LINGER)
 *   onoff, linger = optval.unpack "ii"
 *   onoff = onoff == 0 ? false : true
*/
static VALUE
bsock_getsockopt(VALUE sock, SEL sel, VALUE lev, VALUE optname)
{
#if !defined(__BEOS__)
    int level, option;
    socklen_t len;
    char *buf;
    rb_io_t *fptr;

    level = level_arg(lev);
    option = optname_arg(level, optname);
    len = 256;
    buf = ALLOCA_N(char,len);

    GetOpenFile(sock, fptr);
    if (getsockopt(fptr->fd, level, option, buf, &len) < 0)
	rb_sys_fail_path(fptr->path);

    return rb_str_new(buf, len);
#else
    rb_notimplement();
#endif
}

/*
 * call-seq:
 *   basicsocket.getsockname => sockaddr
 *
 * Returns the local address of the socket as a sockaddr string.
 *
 *   TCPServer.open("127.0.0.1", 15120) {|serv|
 *     p serv.getsockname #=> "\x02\x00;\x10\x7F\x00\x00\x01\x00\x00\x00\x00\x00\x00\x00\x00"
 *   }
 *
 * If Addrinfo object is preferred over the binary string,
 * use BasicSocket#local_address.
 */
static VALUE
bsock_getsockname(VALUE sock, SEL sel)
{
    struct sockaddr_storage buf;
    socklen_t len = (socklen_t)sizeof buf;
    rb_io_t *fptr;

    GetOpenFile(sock, fptr);
    if (getsockname(fptr->fd, (struct sockaddr*)&buf, &len) < 0)
	rb_sys_fail("getsockname(2)");
    return rb_str_new((char*)&buf, len);
}

/*
 * call-seq:
 *   basicsocket.getpeername => sockaddr
 *
 * Returns the remote address of the socket as a sockaddr string.
 *
 *   TCPServer.open("127.0.0.1", 1440) {|serv|
 *     c = TCPSocket.new("127.0.0.1", 1440)
 *     s = serv.accept
 *     p s.getpeername #=> "\x02\x00\x82u\x7F\x00\x00\x01\x00\x00\x00\x00\x00\x00\x00\x00"
 *   }
 *
 * If Addrinfo object is preferred over the binary string,
 * use BasicSocket#remote_address.
 *
 */
static VALUE
bsock_getpeername(VALUE sock, SEL sel)
{
    struct sockaddr_storage buf;
    socklen_t len = (socklen_t)sizeof buf;
    rb_io_t *fptr;

    GetOpenFile(sock, fptr);
    if (getpeername(fptr->fd, (struct sockaddr*)&buf, &len) < 0)
	rb_sys_fail("getpeername(2)");
    return rb_str_new((char*)&buf, len);
}

#if defined(HAVE_GETPEEREID) || defined(SO_PEERCRED) || defined(HAVE_GETPEERUCRED)
/*
 * call-seq:
 *   basicsocket.getpeereid => [euid, egid]
 *
 * Returns the user and group on the peer of the UNIX socket.
 * The result is a two element array which contains the effective uid and the effective gid.
 *
 *   Socket.unix_server_loop("/tmp/sock") {|s|
 *     begin
 *       euid, egid = s.getpeereid
 *
 *       # Check the connected client is myself or not.
 *       next if euid != Process.uid
 *
 *       # do something about my resource.
 *
 *     ensure
 *       s.close
 *     end
 *   }
 *
 */
static VALUE
bsock_getpeereid(VALUE self, SEL sel)
{
#if defined(HAVE_GETPEEREID)
    rb_io_t *fptr;
    uid_t euid;
    gid_t egid;
    GetOpenFile(self, fptr);
    if (getpeereid(fptr->fd, &euid, &egid) == -1)
	rb_sys_fail("getpeereid");
    return rb_assoc_new(UIDT2NUM(euid), GIDT2NUM(egid));
#elif defined(SO_PEERCRED) /* GNU/Linux */
    rb_io_t *fptr;
    struct ucred cred;
    socklen_t len = sizeof(cred);
    GetOpenFile(self, fptr);
    if (getsockopt(fptr->fd, SOL_SOCKET, SO_PEERCRED, &cred, &len) == -1)
	rb_sys_fail("getsockopt(SO_PEERCRED)");
    return rb_assoc_new(UIDT2NUM(cred.uid), GIDT2NUM(cred.gid));
#elif defined(HAVE_GETPEERUCRED) /* Solaris */
    rb_io_t *fptr;
    ucred_t *uc = NULL;
    VALUE ret;
    GetOpenFile(self, fptr);
    if (getpeerucred(fptr->fd, &uc) == -1)
	rb_sys_fail("getpeerucred");
    ret = rb_assoc_new(UIDT2NUM(ucred_geteuid(uc)), GIDT2NUM(ucred_getegid(uc)));
    ucred_free(uc);
    return ret;
#endif
}
#else
#define bsock_getpeereid rb_f_notimplement
#endif

static VALUE addrinfo_new(struct sockaddr *addr, socklen_t len, int family, int socktype, int protocol, VALUE canonname, VALUE inspectname);
static VALUE fd_socket_addrinfo(int fd, struct sockaddr *addr, socklen_t len);
static VALUE io_socket_addrinfo(VALUE io, struct sockaddr *addr, socklen_t len);

/*
 * call-seq:
 *   bsock.local_address => addrinfo
 *
 * Returns an Addrinfo object for local address obtained by getsockname.
 *
 * Note that addrinfo.protocol is filled by 0.
 *
 *   TCPSocket.open("www.ruby-lang.org", 80) {|s|
 *     p s.local_address #=> #<Addrinfo: 192.168.0.129:36873 TCP>
 *   }
 *
 *   TCPServer.open("127.0.0.1", 1512) {|serv|
 *     p serv.local_address #=> #<Addrinfo: 127.0.0.1:1512 TCP>
 *   }
 *
 */
static VALUE
bsock_local_address(VALUE sock, SEL sel)
{
    struct sockaddr_storage buf;
    socklen_t len = (socklen_t)sizeof buf;
    rb_io_t *fptr;

    GetOpenFile(sock, fptr);
    if (getsockname(fptr->fd, (struct sockaddr*)&buf, &len) < 0)
	rb_sys_fail("getsockname(2)");
    return fd_socket_addrinfo(fptr->fd, (struct sockaddr *)&buf, len);
}

/*
 * call-seq:
 *   bsock.remote_address => addrinfo
 *
 * Returns an Addrinfo object for remote address obtained by getpeername.
 *
 * Note that addrinfo.protocol is filled by 0.
 *
 *   TCPSocket.open("www.ruby-lang.org", 80) {|s|
 *     p s.remote_address #=> #<Addrinfo: 221.186.184.68:80 TCP>
 *   }
 *
 *   TCPServer.open("127.0.0.1", 1728) {|serv|
 *     c = TCPSocket.new("127.0.0.1", 1728)
 *     s = serv.accept
 *     p s.remote_address #=> #<Addrinfo: 127.0.0.1:36504 TCP>
 *   }
 *
 */
static VALUE
bsock_remote_address(VALUE sock, SEL sel)
{
    struct sockaddr_storage buf;
    socklen_t len = (socklen_t)sizeof buf;
    rb_io_t *fptr;

    GetOpenFile(sock, fptr);
    if (getpeername(fptr->fd, (struct sockaddr*)&buf, &len) < 0)
	rb_sys_fail("getpeername(2)");
    return fd_socket_addrinfo(fptr->fd, (struct sockaddr *)&buf, len);
}

#define SockAddrStringValue(v) sockaddr_string_value(&(v))
#define SockAddrStringValuePtr(v) sockaddr_string_value_ptr(&(v))
static VALUE sockaddr_string_value(volatile VALUE *);
static char *sockaddr_string_value_ptr(volatile VALUE *);

/*
 * call-seq:
 *   basicsocket.send(mesg, flags [, dest_sockaddr]) => numbytes_sent
 *
 * send _mesg_ via _basicsocket_.
 *
 * _mesg_ should be a string.
 *
 * _flags_ should be a bitwise OR of Socket::MSG_* constants.
 *
 * _dest_sockaddr_ should be a packed sockaddr string or an addrinfo.
 *
 *   TCPSocket.open("localhost", 80) {|s|
 *     s.send "GET / HTTP/1.0\r\n\r\n", 0
 *     p s.read
 *   }
 */
static VALUE
bsock_send(VALUE sock, SEL sel, int argc, VALUE *argv)
{
    VALUE mesg, to;
    VALUE flags;
    rb_io_t *fptr;
    int fd, n;

    rb_secure(4);
    rb_scan_args(argc, argv, "21", &mesg, &flags, &to);

    StringValue(mesg);
    if (!NIL_P(to)) SockAddrStringValue(to);
    GetOpenFile(sock, fptr);
    fd = fptr->fd;
  retry:
    rb_thread_fd_writable(fd);
    if (!NIL_P(to)) {
	n = sendto(fd, RSTRING_PTR(mesg), RSTRING_LEN(mesg), NUM2INT(flags),
		   (struct sockaddr*)RSTRING_PTR(to), RSTRING_LEN(to));
    }
    else {
	n = send(fd, RSTRING_PTR(mesg), RSTRING_LEN(mesg), NUM2INT(flags));
    }
    if (n < 0) {
	if (rb_io_wait_writable(fd)) {
	    goto retry;
	}
	rb_sys_fail("send(2)");
    }
    return INT2FIX(n);
}

/*
 * call-seq:
 *   basicsocket.do_not_reverse_lookup => true or false
 *
 * Gets the do_not_reverse_lookup flag of _basicsocket_.
 *
 *   TCPSocket.open("www.ruby-lang.org", 80) {|sock|
 *     p sock.do_not_reverse_lookup      #=> false
 *     p sock.peeraddr                   #=> ["AF_INET", 80, "carbon.ruby-lang.org", "221.186.184.68"]
 *     sock.do_not_reverse_lookup = true
 *     p sock.peeraddr                   #=> ["AF_INET", 80, "221.186.184.68", "221.186.184.68"]
 *   }
 */
static VALUE
bsock_do_not_reverse_lookup(VALUE sock, SEL sel)
{
    rb_io_t *fptr;

    GetOpenFile(sock, fptr);
    return (fptr->mode & FMODE_NOREVLOOKUP) ? Qtrue : Qfalse;
}

/*
 * call-seq:
 *   basicsocket.do_not_reverse_lookup = bool
 *
 * Sets the do_not_reverse_lookup flag of _basicsocket_.
 *
 *   BasicSocket.do_not_reverse_lookup = false
 *   p TCPSocket.new("127.0.0.1", 80).do_not_reverse_lookup #=> false
 *   BasicSocket.do_not_reverse_lookup = true
 *   p TCPSocket.new("127.0.0.1", 80).do_not_reverse_lookup #=> true
 *
 */
static VALUE
bsock_do_not_reverse_lookup_set(VALUE sock, SEL sel, VALUE state)
{
    rb_io_t *fptr;

    rb_secure(4);
    GetOpenFile(sock, fptr);
    if (RTEST(state)) {
	fptr->mode |= FMODE_NOREVLOOKUP;
    }
    else {
	fptr->mode &= ~FMODE_NOREVLOOKUP;
    }
    return sock;
}

static VALUE ipaddr(struct sockaddr*, int);
#ifdef HAVE_SYS_UN_H
static VALUE unixaddr(struct sockaddr_un*, socklen_t);
#endif

enum sock_recv_type {
    RECV_RECV,			/* BasicSocket#recv(no from) */
    RECV_IP,			/* IPSocket#recvfrom */
    RECV_UNIX,			/* UNIXSocket#recvfrom */
    RECV_SOCKET			/* Socket#recvfrom */
};

static VALUE
s_recvfrom(VALUE sock, int argc, VALUE *argv, enum sock_recv_type from)
{
    rb_io_t *fptr;
    VALUE str;
    struct sockaddr_storage buf;
    socklen_t alen = (socklen_t)sizeof buf;
    VALUE len, flg;
    long buflen;
    long slen;
    int fd, flags;

    rb_scan_args(argc, argv, "11", &len, &flg);

    if (flg == Qnil) flags = 0;
    else             flags = NUM2INT(flg);
    buflen = NUM2INT(len);

    GetOpenFile(sock, fptr);
    if (rb_io_read_pending(fptr)) {
	rb_raise(rb_eIOError, "recv for buffered IO");
    }
    fd = fptr->fd;

    str = rb_bstr_new();
    rb_bstr_resize(str, buflen);

  retry:
    rb_thread_wait_fd(fd);
    rb_io_check_closed(fptr);
    if (rb_bstr_length(str) != buflen) {
	rb_raise(rb_eRuntimeError, "buffer string modified");
    }
    slen = recvfrom(fd, rb_bstr_bytes(str), buflen, flags,
	    (struct sockaddr *)&buf, &alen);

    if (slen < 0) {
	if (rb_io_wait_readable(fd)) {
	    goto retry;
	}
	rb_sys_fail("recvfrom(2)");
    }
    if (slen < rb_bstr_length(str)) {
	rb_bstr_resize(str, slen);
    }
    rb_obj_taint(str);
    switch (from) {
      case RECV_RECV:
	return (VALUE)str;
      case RECV_IP:
#if 0
	if (alen != sizeof(struct sockaddr_in)) {
	    rb_raise(rb_eTypeError, "sockaddr size differs - should not happen");
	}
#endif
	if (alen && alen != sizeof(buf)) /* OSX doesn't return a from result for connection-oriented sockets */
	    return rb_assoc_new(str, ipaddr((struct sockaddr*)&buf, fptr->mode & FMODE_NOREVLOOKUP));
	else
	    return rb_assoc_new(str, Qnil);

#ifdef HAVE_SYS_UN_H
      case RECV_UNIX:
	return rb_assoc_new(str, unixaddr((struct sockaddr_un*)&buf, alen));
#endif
      case RECV_SOCKET:
	return rb_assoc_new(str, io_socket_addrinfo(sock, (struct sockaddr*)&buf, alen));
      default:
	rb_bug("s_recvfrom called with bad value");
    }
}

static VALUE
s_recvfrom_nonblock(VALUE sock, int argc, VALUE *argv, enum sock_recv_type from)
{
    rb_io_t *fptr;
    VALUE str;
    struct sockaddr_storage buf;
    socklen_t alen = (socklen_t)sizeof buf;
    VALUE len, flg;
    long buflen;
    long slen;
    int fd, flags;
    VALUE addr = Qnil;

    rb_scan_args(argc, argv, "11", &len, &flg);

    if (flg == Qnil) flags = 0;
    else             flags = NUM2INT(flg);
    buflen = NUM2INT(len);

#ifdef MSG_DONTWAIT
    /* MSG_DONTWAIT avoids the race condition between fcntl and recvfrom.
       It is not portable, though. */
    flags |= MSG_DONTWAIT;
#endif

    GetOpenFile(sock, fptr);
    if (rb_io_read_pending(fptr)) {
	rb_raise(rb_eIOError, "recvfrom for buffered IO");
    }
    fd = fptr->fd;

    str = rb_bstr_new();
    rb_bstr_resize(str, buflen);

    rb_io_check_closed(fptr);
    rb_io_set_nonblock(fptr);
    slen = recvfrom(fd, rb_bstr_bytes(str), buflen, flags,
	    (struct sockaddr *)&buf, &alen);

    if (slen < 0) {
	switch (errno) {
	  case EAGAIN:
#if defined(EWOULDBLOCK) && EWOULDBLOCK != EAGAIN
	  case EWOULDBLOCK:
#endif
            rb_mod_sys_fail(rb_mWaitReadable, "recvfrom(2) would block");
	}
	rb_sys_fail("recvfrom(2)");
    }
    if (slen < rb_bstr_length(str)) {
	rb_bstr_resize(str, slen);
    }
    rb_obj_taint(str);
    switch (from) {
      case RECV_RECV:
        return str;

      case RECV_IP:
        if (alen && alen != sizeof(buf)) /* connection-oriented socket may not return a from result */
            addr = ipaddr((struct sockaddr*)&buf, fptr->mode & FMODE_NOREVLOOKUP);
        break;

      case RECV_SOCKET:
        addr = io_socket_addrinfo(sock, (struct sockaddr*)&buf, alen);
        break;

      default:
        rb_bug("s_recvfrom_nonblock called with bad value");
    }
    return rb_assoc_new(str, addr);
}

/*
 * call-seq:
 *   basicsocket.recv(maxlen) => mesg
 *   basicsocket.recv(maxlen, flags) => mesg
 *
 * Receives a message.
 *
 * _maxlen_ is the maximum number of bytes to receive.
 *
 * _flags_ should be a bitwise OR of Socket::MSG_* constants.
 *
 *   UNIXSocket.pair {|s1, s2|
 *     s1.puts "Hello World"
 *     p s2.recv(4)                     #=> "Hell"
 *     p s2.recv(4, Socket::MSG_PEEK)   #=> "o Wo"
 *     p s2.recv(4)                     #=> "o Wo"
 *     p s2.recv(10)                    #=> "rld\n"
 *   }
 */
static VALUE
bsock_recv(VALUE sock, SEL sel, int argc, VALUE *argv)
{
    return s_recvfrom(sock, argc, argv, RECV_RECV);
}

/*
 * call-seq:
 * 	basicsocket.recv_nonblock(maxlen) => mesg
 * 	basicsocket.recv_nonblock(maxlen, flags) => mesg
 *
 * Receives up to _maxlen_ bytes from +socket+ using recvfrom(2) after
 * O_NONBLOCK is set for the underlying file descriptor.
 * _flags_ is zero or more of the +MSG_+ options.
 * The result, _mesg_, is the data received.
 *
 * When recvfrom(2) returns 0, Socket#recv_nonblock returns
 * an empty string as data.
 * The meaning depends on the socket: EOF on TCP, empty packet on UDP, etc.
 *
 * === Parameters
 * * +maxlen+ - the number of bytes to receive from the socket
 * * +flags+ - zero or more of the +MSG_+ options
 *
 * === Example
 * 	serv = TCPServer.new("127.0.0.1", 0)
 * 	af, port, host, addr = serv.addr
 * 	c = TCPSocket.new(addr, port)
 * 	s = serv.accept
 * 	c.send "aaa", 0
 * 	begin # emulate blocking recv.
 * 	  p s.recv_nonblock(10) #=> "aaa"
 * 	rescue IO::WaitReadable
 * 	  IO.select([s])
 * 	  retry
 * 	end
 *
 * Refer to Socket#recvfrom for the exceptions that may be thrown if the call
 * to _recv_nonblock_ fails.
 *
 * BasicSocket#recv_nonblock may raise any error corresponding to recvfrom(2) failure,
 * including Errno::EWOULDBLOCK.
 *
 * If the exception is Errno::EWOULDBLOCK or Errno::AGAIN,
 * it is extended by IO::WaitReadable.
 * So IO::WaitReadable can be used to rescue the exceptions for retrying recv_nonblock.
 *
 * === See
 * * Socket#recvfrom
 */
static VALUE
bsock_recv_nonblock(VALUE sock, SEL sel, int argc, VALUE *argv)
{
    return s_recvfrom_nonblock(sock, argc, argv, RECV_RECV);
}

/*
 * call-seq:
 *   BasicSocket.do_not_reverse_lookup => true or false
 *
 * Gets the global do_not_reverse_lookup flag.
 *
 *   BasicSocket.do_not_reverse_lookup  #=> false
 */
static VALUE
bsock_do_not_rev_lookup(VALUE self, SEL sel)
{
    return do_not_reverse_lookup ? Qtrue : Qfalse;
}

/*
 * call-seq:
 *   BasicSocket.do_not_reverse_lookup = bool
 *
 * Sets the global do_not_reverse_lookup flag.
 *
 * The flag is used for initial value of do_not_reverse_lookup for each socket.
 *
 *   s1 = TCPSocket.new("localhost", 80)
 *   p s1.do_not_reverse_lookup                 #=> true
 *   BasicSocket.do_not_reverse_lookup = false
 *   s2 = TCPSocket.new("localhost", 80)
 *   p s2.do_not_reverse_lookup                 #=> false
 *   p s1.do_not_reverse_lookup                 #=> true
 *
 */
static VALUE
bsock_do_not_rev_lookup_set(VALUE self, SEL sel, VALUE val)
{
    rb_secure(4);
    do_not_reverse_lookup = RTEST(val);
    return val;
}

NORETURN(static void raise_socket_error(char *, int));
static void
raise_socket_error(char *reason, int error)
{
#ifdef EAI_SYSTEM
    if (error == EAI_SYSTEM) rb_sys_fail(reason);
#endif
    rb_raise(rb_eSocket, "%s: %s", reason, gai_strerror(error));
}

static void
make_ipaddr0(struct sockaddr *addr, char *buf, size_t len)
{
    int error;

    error = getnameinfo(addr, SA_LEN(addr), buf, len, NULL, 0, NI_NUMERICHOST);
    if (error) {
        raise_socket_error("getnameinfo", error);
    }
}

static VALUE
make_ipaddr(struct sockaddr *addr)
{
    char hbuf[1024];

    make_ipaddr0(addr, hbuf, sizeof(hbuf));
    return rb_str_new2(hbuf);
}

static void
make_inetaddr(long host, char *buf, size_t len)
{
    struct sockaddr_in sin;

    MEMZERO(&sin, struct sockaddr_in, 1);
    sin.sin_family = AF_INET;
    SET_SIN_LEN(&sin, sizeof(sin));
    sin.sin_addr.s_addr = host;
    make_ipaddr0((struct sockaddr*)&sin, buf, len);
}

static int
str_is_number(const char *p)
{
    char *ep;

    if (!p || *p == '\0')
       return 0;
    ep = NULL;
    (void)STRTOUL(p, &ep, 10);
    if (ep && *ep == '\0')
       return 1;
    else
       return 0;
}

static char*
host_str(VALUE host, char *hbuf, size_t len, int *flags_ptr)
{
    if (NIL_P(host)) {
        return NULL;
    }
    else if (rb_obj_is_kind_of(host, rb_cInteger)) {
        unsigned int i = NUM2UINT(host);

        make_inetaddr(htonl(i), hbuf, len);
        if (flags_ptr) *flags_ptr |= AI_NUMERICHOST;
        return hbuf;
    }
    else {
        const char *name;

        SafeStringValue(host);
        name = RSTRING_PTR(host);
        if (!name || *name == 0 || (name[0] == '<' && strcmp(name, "<any>") == 0)) {
            make_inetaddr(INADDR_ANY, hbuf, len);
            if (flags_ptr) *flags_ptr |= AI_NUMERICHOST;
        }
        else if (name[0] == '<' && strcmp(name, "<broadcast>") == 0) {
            make_inetaddr(INADDR_BROADCAST, hbuf, len);
            if (flags_ptr) *flags_ptr |= AI_NUMERICHOST;
        }
        else if (strlen(name) >= len) {
            rb_raise(rb_eArgError, "hostname too long (%ld)", strlen(name));
        }
        else {
            strcpy(hbuf, name);
        }
        return hbuf;
    }
}

static char*
port_str(VALUE port, char *pbuf, size_t len, int *flags_ptr)
{
    if (NIL_P(port)) {
        return 0;
    }
    else if (FIXNUM_P(port)) {
        snprintf(pbuf, len, "%ld", FIX2LONG(port));
#ifdef AI_NUMERICSERV
        if (flags_ptr) *flags_ptr |= AI_NUMERICSERV;
#endif
        return pbuf;
    }
    else {
        const char *serv;

        SafeStringValue(port);
        serv = RSTRING_PTR(port);
        if (strlen(serv) >= len) {
            rb_raise(rb_eArgError, "service name too long (%ld)", strlen(serv));
        }
        strcpy(pbuf, serv);
        return pbuf;
    }
}

static struct addrinfo*
sock_getaddrinfo(VALUE host, VALUE port, struct addrinfo *hints, int socktype_hack)
{
    struct addrinfo* res = NULL;
    char *hostp, *portp;
    int error;
    char hbuf[NI_MAXHOST], pbuf[NI_MAXSERV];
    int additional_flags = 0;

    hostp = host_str(host, hbuf, sizeof(hbuf), &additional_flags);
    portp = port_str(port, pbuf, sizeof(pbuf), &additional_flags);

    if (socktype_hack && hints->ai_socktype == 0 && str_is_number(portp)) {
        hints->ai_socktype = SOCK_DGRAM;
    }
    hints->ai_flags |= additional_flags;

    error = getaddrinfo(hostp, portp, hints, &res);
    if (error) {
        if (hostp && hostp[strlen(hostp)-1] == '\n') {
            rb_raise(rb_eSocket, "newline at the end of hostname");
        }
        raise_socket_error("getaddrinfo", error);
    }

#if defined(__APPLE__) && defined(__MACH__)
    {
        struct addrinfo *r;
        r = res;
        while (r) {
            if (! r->ai_socktype) r->ai_socktype = hints->ai_socktype;
            if (! r->ai_protocol) {
                if (r->ai_socktype == SOCK_DGRAM) {
                    r->ai_protocol = IPPROTO_UDP;
                }
                else if (r->ai_socktype == SOCK_STREAM) {
                    r->ai_protocol = IPPROTO_TCP;
                }
            }
            r = r->ai_next;
        }
    }
#endif
    return res;
}

static struct addrinfo*
sock_addrinfo(VALUE host, VALUE port, int socktype, int flags)
{
    struct addrinfo hints;

    MEMZERO(&hints, struct addrinfo, 1);
    hints.ai_family = AF_UNSPEC;
    hints.ai_socktype = socktype;
    hints.ai_flags = flags;
    return sock_getaddrinfo(host, port, &hints, 1);
}

static VALUE
ipaddr(struct sockaddr *sockaddr, int norevlookup)
{
    VALUE family, port, addr1, addr2;
    VALUE ary;
    int error;
    char hbuf[1024], pbuf[1024];
    ID id;

    id = intern_family(sockaddr->sa_family);
    if (id) {
        family = rb_str_dup(rb_id2str(id));
    }
    else {
        sprintf(pbuf, "unknown:%d", sockaddr->sa_family);
        family = rb_str_new2(pbuf);
    }

    addr1 = Qnil;
    if (!norevlookup) {
        error = getnameinfo(sockaddr, SA_LEN(sockaddr), hbuf, sizeof(hbuf),
                            NULL, 0, 0);
        if (! error) {
            addr1 = rb_str_new2(hbuf);
        }
    }
    error = getnameinfo(sockaddr, SA_LEN(sockaddr), hbuf, sizeof(hbuf),
                        pbuf, sizeof(pbuf), NI_NUMERICHOST | NI_NUMERICSERV);
    if (error) {
        raise_socket_error("getnameinfo", error);
    }
    addr2 = rb_str_new2(hbuf);
    if (addr1 == Qnil) {
	addr1 = addr2;
    }
    port = INT2FIX(atoi(pbuf));
    ary = rb_ary_new3(4, family, port, addr1, addr2);

    return ary;
}

static int
ruby_socket(int domain, int type, int proto)
{
    int fd;

    fd = socket(domain, type, proto);
    if (fd < 0) {
	if (errno == EMFILE || errno == ENFILE) {
	    rb_gc();
	    fd = socket(domain, type, proto);
	}
    }
    return fd;
}

static int
wait_connectable0(int fd, rb_fdset_t *fds_w, rb_fdset_t *fds_e)
{
    int sockerr;
    socklen_t sockerrlen;

    for (;;) {
	rb_fd_zero(fds_w);
	rb_fd_zero(fds_e);

	rb_fd_set(fd, fds_w);
	rb_fd_set(fd, fds_e);

	rb_thread_select(fd+1, 0, rb_fd_ptr(fds_w), rb_fd_ptr(fds_e), 0);

	if (rb_fd_isset(fd, fds_w)) {
	    return 0;
	}
	else if (rb_fd_isset(fd, fds_e)) {
	    sockerrlen = (socklen_t)sizeof(sockerr);
	    if (getsockopt(fd, SOL_SOCKET, SO_ERROR, (void *)&sockerr,
			   &sockerrlen) == 0) {
		if (sockerr == 0)
		    continue;	/* workaround for winsock */
		errno = sockerr;
	    }
	    return -1;
	}
    }
}

struct wait_connectable_arg {
    int fd;
    rb_fdset_t *fds_w;
    rb_fdset_t *fds_e;
};

#ifdef HAVE_RB_FD_INIT
static VALUE
try_wait_connectable(VALUE arg)
{
    struct wait_connectable_arg *p = (struct wait_connectable_arg *)arg;
    return (VALUE)wait_connectable0(p->fd, p->fds_w, p->fds_e);
}

static VALUE
wait_connectable_ensure(VALUE arg)
{
    struct wait_connectable_arg *p = (struct wait_connectable_arg *)arg;
    rb_fd_term(p->fds_w);
    rb_fd_term(p->fds_e);
    return Qnil;
}
#endif

static int
wait_connectable(int fd)
{
    struct wait_connectable_arg *arg;

    arg = (void *)xmalloc(sizeof(struct wait_connectable_arg));
    GC_WB(&arg->fds_w, xmalloc(sizeof(rb_fdset_t)));
    GC_WB(&arg->fds_e, xmalloc(sizeof(rb_fdset_t)));
    rb_fd_init(arg->fds_w);
    rb_fd_init(arg->fds_e);
#ifdef HAVE_RB_FD_INIT
    arg->fd = fd;
    return (int)rb_ensure(try_wait_connectable, (VALUE)arg,
			  wait_connectable_ensure,(VALUE)arg);
#else
    return wait_connectable0(fd, arg->fds_w, arg->fds_e);
#endif
}

#ifdef __CYGWIN__
#define WAIT_IN_PROGRESS 10
#endif
#ifdef __APPLE__
#define WAIT_IN_PROGRESS 10
#endif
#ifdef __linux__
/* returns correct error */
#define WAIT_IN_PROGRESS 0
#endif
#ifndef WAIT_IN_PROGRESS
/* BSD origin code apparently has a problem */
#define WAIT_IN_PROGRESS 1
#endif

static int
ruby_connect(int fd, struct sockaddr *sockaddr, int len, int socks)
{
    int status;
#if WAIT_IN_PROGRESS > 0
    int wait_in_progress = -1;
    int sockerr;
    socklen_t sockerrlen;
#endif

    for (;;) {
	status = connect(fd, sockaddr, len);
	if (status < 0) {
	    switch (errno) {
	      case EINTR:
#if defined(ERESTART)
	      case ERESTART:
#endif
		continue;

	      case EAGAIN:
#ifdef EINPROGRESS
	      case EINPROGRESS:
#endif
#if WAIT_IN_PROGRESS > 0
		sockerrlen = (socklen_t)sizeof(sockerr);
		status = getsockopt(fd, SOL_SOCKET, SO_ERROR, (void *)&sockerr, &sockerrlen);
		if (status) break;
		if (sockerr) {
		    status = -1;
		    errno = sockerr;
		    break;
		}
#endif
#ifdef EALREADY
	      case EALREADY:
#endif
#if WAIT_IN_PROGRESS > 0
		wait_in_progress = WAIT_IN_PROGRESS;
#endif
		status = wait_connectable(fd);
		if (status) {
		    break;
		}
		errno = 0;
		continue;

#if WAIT_IN_PROGRESS > 0
	      case EINVAL:
		if (wait_in_progress-- > 0) {
		    /*
		     * connect() after EINPROGRESS returns EINVAL on
		     * some platforms, need to check true error
		     * status.
		     */
		    sockerrlen = (socklen_t)sizeof(sockerr);
		    status = getsockopt(fd, SOL_SOCKET, SO_ERROR, (void *)&sockerr, &sockerrlen);
		    if (!status && !sockerr) {
			struct timeval tv = {0, 100000};
			rb_thread_wait_for(tv);
			continue;
		    }
		    status = -1;
		    errno = sockerr;
		}
		break;
#endif

#ifdef EISCONN
	      case EISCONN:
		status = 0;
		errno = 0;
		break;
#endif
	      default:
		break;
	    }
	}
	return status;
    }
}

struct inetsock_arg
{
    VALUE sock;
    struct {
	VALUE host, serv;
	struct addrinfo *res;
    } remote, local;
    int type;
    int fd;
};

static VALUE
inetsock_cleanup(struct inetsock_arg *arg)
{
    if (arg->remote.res) {
	freeaddrinfo(arg->remote.res);
	arg->remote.res = 0;
    }
    if (arg->local.res) {
	freeaddrinfo(arg->local.res);
	arg->local.res = 0;
    }
    if (arg->fd >= 0) {
	close(arg->fd);
    }
    return Qnil;
}

static VALUE
init_inetsock_internal(struct inetsock_arg *arg)
{
    int type = arg->type;
    struct addrinfo *res;
    int fd, status = 0;
    char *syscall = NULL;

    arg->remote.res = sock_addrinfo(arg->remote.host, arg->remote.serv, SOCK_STREAM,
				    (type == INET_SERVER) ? AI_PASSIVE : 0);
    /*
     * Maybe also accept a local address
     */

    if (type != INET_SERVER && (!NIL_P(arg->local.host) || !NIL_P(arg->local.serv))) {
	arg->local.res = sock_addrinfo(arg->local.host, arg->local.serv, SOCK_STREAM, 0);
    }

    arg->fd = fd = -1;
    for (res = arg->remote.res; res; res = res->ai_next) {
#if !defined(INET6) && defined(AF_INET6)
	if (res->ai_family == AF_INET6)
	    continue;
#endif
	status = ruby_socket(res->ai_family,res->ai_socktype,res->ai_protocol);
	syscall = "socket(2)";
	fd = status;
	if (fd < 0) {
	    continue;
	}
	arg->fd = fd;
	if (type == INET_SERVER) {
#if !defined(_WIN32) && !defined(__CYGWIN__)
	    status = 1;
	    setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
		       (char*)&status, sizeof(status));
#endif
	    status = bind(fd, res->ai_addr, res->ai_addrlen);
	    syscall = "bind(2)";
	}
	else {
	    if (arg->local.res) {
		status = bind(fd, arg->local.res->ai_addr, arg->local.res->ai_addrlen);
		syscall = "bind(2)";
	    }

	    if (status >= 0) {
		status = ruby_connect(fd, res->ai_addr, res->ai_addrlen,
				      (type == INET_SOCKS));
		syscall = "connect(2)";
	    }
	}

	if (status < 0) {
	    close(fd);
	    arg->fd = fd = -1;
	    continue;
	} else
	    break;
    }
    if (status < 0) {
	rb_sys_fail(syscall);
    }

    arg->fd = -1;

    if (type == INET_SERVER) {
	status = listen(fd, 5);
	if (status < 0) {
	    close(fd);
	    syscall = "listen(2)";
	}
    }

    /* create new instance */
    return init_sock(arg->sock, fd);
}

static VALUE
init_inetsock(VALUE sock, VALUE remote_host, VALUE remote_serv,
	      VALUE local_host, VALUE local_serv, int type)
{
    struct inetsock_arg arg;
    arg.sock = sock;
    arg.remote.host = remote_host;
    arg.remote.serv = remote_serv;
    arg.remote.res = 0;
    arg.local.host = local_host;
    arg.local.serv = local_serv;
    arg.local.res = 0;
    arg.type = type;
    arg.fd = -1;
    return rb_ensure(init_inetsock_internal, (VALUE)&arg,
		     inetsock_cleanup, (VALUE)&arg);
}

/*
 * call-seq:
 *    TCPSocket.new(remote_host, remote_port, local_host=nil, local_port=nil)
 *
 * Opens a TCP connection to +remote_host+ on +remote_port+.  If +local_host+
 * and +local_port+ are specified, then those parameters are used on the local
 * end to establish the connection.
 */
static VALUE
tcp_init(VALUE sock, SEL sel, int argc, VALUE *argv)
{
    VALUE remote_host, remote_serv;
    VALUE local_host, local_serv;

    rb_scan_args(argc, argv, "22", &remote_host, &remote_serv,
			&local_host, &local_serv);

    return init_inetsock(sock, remote_host, remote_serv,
			local_host, local_serv, INET_CLIENT);
}

struct hostent_arg {
    VALUE host;
    struct addrinfo* addr;
    VALUE (*ipaddr)(struct sockaddr*, size_t);
};

static VALUE
make_hostent_internal(struct hostent_arg *arg)
{
    VALUE host = arg->host;
    struct addrinfo* addr = arg->addr;
    VALUE (*ipaddr)(struct sockaddr*, size_t) = arg->ipaddr;

    struct addrinfo *ai;
    struct hostent *h;
    VALUE ary, names;
    char **pch;
    const char* hostp;
    char hbuf[NI_MAXHOST];

    ary = rb_ary_new();
    if (addr->ai_canonname) {
        hostp = addr->ai_canonname;
    }
    else {
        hostp = host_str(host, hbuf, sizeof(hbuf), NULL);
    }
    rb_ary_push(ary, rb_str_new2(hostp));

    if (addr->ai_canonname && (h = gethostbyname(addr->ai_canonname))) {
        names = rb_ary_new();
        if (h->h_aliases != NULL) {
            for (pch = h->h_aliases; *pch; pch++) {
                rb_ary_push(names, rb_str_new2(*pch));
            }
        }
    }
    else {
        names = rb_ary_new2(0);
    }
    rb_ary_push(ary, names);
    rb_ary_push(ary, INT2NUM(addr->ai_family));
    for (ai = addr; ai; ai = ai->ai_next) {
        rb_ary_push(ary, (*ipaddr)(ai->ai_addr, ai->ai_addrlen));
    }

    return ary;
}

static VALUE
make_hostent(VALUE host, struct addrinfo *addr, VALUE (*ipaddr)(struct sockaddr *, size_t))
{
    struct hostent_arg arg;

    arg.host = host;
    arg.addr = addr;
    arg.ipaddr = ipaddr;
    return rb_ensure(make_hostent_internal, (VALUE)&arg,
		     RUBY_METHOD_FUNC(freeaddrinfo), (VALUE)addr);
}

static VALUE
tcp_sockaddr(struct sockaddr *addr, size_t len)
{
    return make_ipaddr(addr);
}

/*
 * call-seq:
 *   TCPSocket.gethostbyname(hostname) => [official_hostname, alias_hostnames, address_family, *address_list]
 *
 * Lookups host information by _hostname_.
 *
 *   TCPSocket.gethostbyname("localhost")
 *   #=> ["localhost", ["hal"], 2, "127.0.0.1"]
 *
 */
static VALUE
tcp_s_gethostbyname(VALUE obj, SEL sel, VALUE host)
{
    rb_secure(3);
    return make_hostent(host, sock_addrinfo(host, Qnil, SOCK_STREAM,
		AI_CANONNAME), tcp_sockaddr);
}

/*
 * call-seq:
 *   TCPServer.new([hostname,] port)                    => tcpserver
 *
 * Creates a new server socket bound to _port_.
 *
 * If _hostname_ is given, the socket is bound to it.
 *
 *   serv = TCPServer.new("127.0.0.1", 28561)
 *   s = serv.accept
 *   s.puts Time.now
 *   s.close
 */
static VALUE
tcp_svr_init(VALUE sock, SEL sel, int argc, VALUE *argv)
{
    VALUE arg1, arg2;

    if (rb_scan_args(argc, argv, "11", &arg1, &arg2) == 2)
	return init_inetsock(sock, arg1, arg2, Qnil, Qnil, INET_SERVER);
    else
	return init_inetsock(sock, Qnil, arg1, Qnil, Qnil, INET_SERVER);
}

static void
make_fd_nonblock(int fd)
{
    int flags;
#ifdef F_GETFL
    flags = fcntl(fd, F_GETFL);
    if (flags == -1) {
        rb_sys_fail(0);
    }
#else
    flags = 0;
#endif
    flags |= O_NONBLOCK;
    if (fcntl(fd, F_SETFL, flags) == -1) {
        rb_sys_fail(0);
    }
}

static VALUE
s_accept_nonblock(VALUE klass, rb_io_t *fptr, struct sockaddr *sockaddr, socklen_t *len)
{
    int fd2;

    rb_secure(3);
    rb_io_set_nonblock(fptr);
    fd2 = accept(fptr->fd, (struct sockaddr*)sockaddr, len);
    if (fd2 < 0) {
	switch (errno) {
	  case EAGAIN:
#if defined(EWOULDBLOCK) && EWOULDBLOCK != EAGAIN
	  case EWOULDBLOCK:
#endif
	  case ECONNABORTED:
#if defined EPROTO
	  case EPROTO:
#endif
            rb_mod_sys_fail(rb_mWaitReadable, "accept(2) would block");
	}
        rb_sys_fail("accept(2)");
    }
    make_fd_nonblock(fd2);
    return init_sock(rb_obj_alloc(klass), fd2);
}

static VALUE
s_accept(VALUE klass, int fd, struct sockaddr *sockaddr, socklen_t *len)
{
    int fd2;
    int retry = 0;

    rb_secure(3);
  retry:
    rb_thread_wait_fd(fd);
#if defined(_nec_ews)
    fd2 = accept(fd, sockaddr, len);
#else
    fd2 = accept(fd, sockaddr, len);
#endif
    if (fd2 < 0) {
	switch (errno) {
	  case EMFILE:
	  case ENFILE:
	    if (retry) break;
	    rb_gc();
	    retry = 1;
	    goto retry;
	  default:
	    if (!rb_io_wait_readable(fd)) break;
	    retry = 0;
	    goto retry;
	}
	rb_sys_fail(0);
    }
    if (!klass) return INT2NUM(fd2);
    return init_sock(rb_obj_alloc(klass), fd2);
}

/*
 * call-seq:
 *   tcpserver.accept => tcpsocket
 *
 *   TCPServer.open("127.0.0.1", 14641) {|serv|
 *     s = serv.accept
 *     s.puts Time.now
 *     s.close
 *   }
 *
 */
static VALUE
tcp_accept(VALUE soc…