BSD 4_4_Lite2 release
[unix-history] / usr / src / sys / kern / uipc_usrreq.c
index c917863..c6bcbfd 100644 (file)
@@ -1,33 +1,87 @@
-/*     uipc_usrreq.c   1.7     83/01/13        */
-
-#include "../h/param.h"
-#include "../h/dir.h"
-#include "../h/user.h"
-#include "../h/mbuf.h"
-#include "../h/protosw.h"
-#include "../h/socket.h"
-#include "../h/socketvar.h"
-#include "../h/unpcb.h"
-#include "../h/un.h"
-#include "../h/inode.h"
-#include "../h/nami.h"
+/*
+ * Copyright (c) 1982, 1986, 1989, 1991, 1993
+ *     The Regents of the University of California.  All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions
+ * are met:
+ * 1. Redistributions of source code must retain the above copyright
+ *    notice, this list of conditions and the following disclaimer.
+ * 2. Redistributions in binary form must reproduce the above copyright
+ *    notice, this list of conditions and the following disclaimer in the
+ *    documentation and/or other materials provided with the distribution.
+ * 3. All advertising materials mentioning features or use of this software
+ *    must display the following acknowledgement:
+ *     This product includes software developed by the University of
+ *     California, Berkeley and its contributors.
+ * 4. Neither the name of the University nor the names of its contributors
+ *    may be used to endorse or promote products derived from this software
+ *    without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
+ * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
+ * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
+ * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
+ * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
+ * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
+ * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
+ * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
+ * SUCH DAMAGE.
+ *
+ *     @(#)uipc_usrreq.c       8.9 (Berkeley) 5/14/95
+ */
+
+#include <sys/param.h>
+#include <sys/systm.h>
+#include <sys/proc.h>
+#include <sys/filedesc.h>
+#include <sys/domain.h>
+#include <sys/protosw.h>
+#include <sys/socket.h>
+#include <sys/socketvar.h>
+#include <sys/unpcb.h>
+#include <sys/un.h>
+#include <sys/namei.h>
+#include <sys/vnode.h>
+#include <sys/file.h>
+#include <sys/stat.h>
+#include <sys/mbuf.h>
 
 /*
  * Unix communications domain.
 
 /*
  * Unix communications domain.
+ *
+ * TODO:
+ *     SEQPACKET, RDM
+ *     rethink name space problems
+ *     need a proper out-of-band
  */
  */
+struct sockaddr sun_noname = { sizeof(sun_noname), AF_UNIX };
+ino_t  unp_ino;                        /* prototype for fake inode numbers */
 
 /*ARGSUSED*/
 
 /*ARGSUSED*/
-uipc_usrreq(so, req, m, nam)
+int
+uipc_usrreq(so, req, m, nam, control)
        struct socket *so;
        int req;
        struct socket *so;
        int req;
-       struct mbuf *m, *nam;
+       struct mbuf *m, *nam, *control;
 {
        struct unpcb *unp = sotounpcb(so);
        register struct socket *so2;
 {
        struct unpcb *unp = sotounpcb(so);
        register struct socket *so2;
-       int error = 0;
+       register int error = 0;
+       struct proc *p = curproc;       /* XXX */
 
 
-       if (unp == 0 && req != PRU_ATTACH)
-               return (EINVAL);                        /* XXX */
+       if (req == PRU_CONTROL)
+               return (EOPNOTSUPP);
+       if (req != PRU_SEND && control && control->m_len) {
+               error = EOPNOTSUPP;
+               goto release;
+       }
+       if (unp == 0 && req != PRU_ATTACH) {
+               error = EINVAL;
+               goto release;
+       }
        switch (req) {
 
        case PRU_ATTACH:
        switch (req) {
 
        case PRU_ATTACH:
@@ -43,16 +97,20 @@ uipc_usrreq(so, req, m, nam)
                break;
 
        case PRU_BIND:
                break;
 
        case PRU_BIND:
-               error = unp_bind(unp, nam);
+               error = unp_bind(unp, nam, p);
                break;
 
        case PRU_LISTEN:
                break;
 
        case PRU_LISTEN:
-               if (unp->unp_inode == 0)
+               if (unp->unp_vnode == 0)
                        error = EINVAL;
                break;
 
        case PRU_CONNECT:
                        error = EINVAL;
                break;
 
        case PRU_CONNECT:
-               error = unp_connect(so, nam);
+               error = unp_connect(so, nam, p);
+               break;
+
+       case PRU_CONNECT2:
+               error = unp_connect2(so, (struct socket *)nam);
                break;
 
        case PRU_DISCONNECT:
                break;
 
        case PRU_DISCONNECT:
@@ -60,14 +118,24 @@ uipc_usrreq(so, req, m, nam)
                break;
 
        case PRU_ACCEPT:
                break;
 
        case PRU_ACCEPT:
-               nam->m_len = unp->unp_remaddr->m_len;
-               bcopy(mtod(unp->unp_remaddr, caddr_t),
-                   mtod(nam, caddr_t), (unsigned)nam->m_len);
+               /*
+                * Pass back name of connected socket,
+                * if it was bound and we are still connected
+                * (our peer may have closed already!).
+                */
+               if (unp->unp_conn && unp->unp_conn->unp_addr) {
+                       nam->m_len = unp->unp_conn->unp_addr->m_len;
+                       bcopy(mtod(unp->unp_conn->unp_addr, caddr_t),
+                           mtod(nam, caddr_t), (unsigned)nam->m_len);
+               } else {
+                       nam->m_len = sizeof(sun_noname);
+                       *(mtod(nam, struct sockaddr *)) = sun_noname;
+               }
                break;
 
        case PRU_SHUTDOWN:
                socantsendmore(so);
                break;
 
        case PRU_SHUTDOWN:
                socantsendmore(so);
-               unp_usrclosed(unp);
+               unp_shutdown(unp);
                break;
 
        case PRU_RCVD:
                break;
 
        case PRU_RCVD:
@@ -84,14 +152,14 @@ uipc_usrreq(so, req, m, nam)
                                break;
                        so2 = unp->unp_conn->unp_socket;
                        /*
                                break;
                        so2 = unp->unp_conn->unp_socket;
                        /*
-                        * Transfer resources back to send port
+                        * Adjust backpressure on sender
                         * and wakeup any waiting to write.
                         */
                         * and wakeup any waiting to write.
                         */
-                       snd->sb_mbmax += rcv->sb_mbmax - rcv->sb_mbcnt;
-                       rcv->sb_mbmax = rcv->sb_mbcnt;
-                       snd->sb_hiwat += rcv->sb_hiwat - rcv->sb_cc;
-                       rcv->sb_hiwat = rcv->sb_cc;
-                       sbwakeup(snd);
+                       snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt;
+                       unp->unp_mbcnt = rcv->sb_mbcnt;
+                       snd->sb_hiwat += unp->unp_cc - rcv->sb_cc;
+                       unp->unp_cc = rcv->sb_cc;
+                       sowwakeup(so2);
 #undef snd
 #undef rcv
                        break;
 #undef snd
 #undef rcv
                        break;
@@ -102,15 +170,19 @@ uipc_usrreq(so, req, m, nam)
                break;
 
        case PRU_SEND:
                break;
 
        case PRU_SEND:
+               if (control && (error = unp_internalize(control, p)))
+                       break;
                switch (so->so_type) {
 
                switch (so->so_type) {
 
-               case SOCK_DGRAM:
+               case SOCK_DGRAM: {
+                       struct sockaddr *from;
+
                        if (nam) {
                                if (unp->unp_conn) {
                                        error = EISCONN;
                                        break;
                                }
                        if (nam) {
                                if (unp->unp_conn) {
                                        error = EISCONN;
                                        break;
                                }
-                               error = unp_connect(so, nam);
+                               error = unp_connect(so, nam, p);
                                if (error)
                                        break;
                        } else {
                                if (error)
                                        break;
                        } else {
@@ -120,32 +192,48 @@ uipc_usrreq(so, req, m, nam)
                                }
                        }
                        so2 = unp->unp_conn->unp_socket;
                                }
                        }
                        so2 = unp->unp_conn->unp_socket;
-                       /* BEGIN XXX */
-                       if (sbspace(&so2->so_rcv) > 0)
-                               (void) sbappendaddr(&so2->so_rcv,
-                                       mtod(nam, struct sockaddr *), m);
-                       /* END XXX */
+                       if (unp->unp_addr)
+                               from = mtod(unp->unp_addr, struct sockaddr *);
+                       else
+                               from = &sun_noname;
+                       if (sbappendaddr(&so2->so_rcv, from, m, control)) {
+                               sorwakeup(so2);
+                               m = 0;
+                               control = 0;
+                       } else
+                               error = ENOBUFS;
                        if (nam)
                                unp_disconnect(unp);
                        break;
                        if (nam)
                                unp_disconnect(unp);
                        break;
+               }
 
                case SOCK_STREAM:
 #define        rcv (&so2->so_rcv)
 #define        snd (&so->so_snd)
 
                case SOCK_STREAM:
 #define        rcv (&so2->so_rcv)
 #define        snd (&so->so_snd)
+                       if (so->so_state & SS_CANTSENDMORE) {
+                               error = EPIPE;
+                               break;
+                       }
                        if (unp->unp_conn == 0)
                                panic("uipc 3");
                        so2 = unp->unp_conn->unp_socket;
                        /*
                        if (unp->unp_conn == 0)
                                panic("uipc 3");
                        so2 = unp->unp_conn->unp_socket;
                        /*
-                        * Send to paired receive port, and then
-                        * give it enough resources to hold what it already has.
+                        * Send to paired receive port, and then reduce
+                        * send buffer hiwater marks to maintain backpressure.
                         * Wake up readers.
                         */
                         * Wake up readers.
                         */
-                       sbappend(rcv, m);
-                       snd->sb_mbmax -= rcv->sb_mbcnt - rcv->sb_mbmax;
-                       rcv->sb_mbmax = rcv->sb_mbcnt;
-                       snd->sb_hiwat -= rcv->sb_cc - rcv->sb_hiwat;
-                       rcv->sb_hiwat = rcv->sb_cc;
-                       sbwakeup(rcv);
+                       if (control) {
+                               if (sbappendcontrol(rcv, m, control))
+                                       control = 0;
+                       } else
+                               sbappend(rcv, m);
+                       snd->sb_mbmax -=
+                           rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt;
+                       unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt;
+                       snd->sb_hiwat -= rcv->sb_cc - unp->unp_conn->unp_cc;
+                       unp->unp_conn->unp_cc = rcv->sb_cc;
+                       sorwakeup(so2);
+                       m = 0;
 #undef snd
 #undef rcv
                        break;
 #undef snd
 #undef rcv
                        break;
@@ -159,23 +247,41 @@ uipc_usrreq(so, req, m, nam)
                unp_drop(unp, ECONNABORTED);
                break;
 
                unp_drop(unp, ECONNABORTED);
                break;
 
-/* SOME AS YET UNIMPLEMENTED HOOKS */
-       case PRU_CONTROL:
-               error = EOPNOTSUPP;
-               break;
-
        case PRU_SENSE:
        case PRU_SENSE:
-               error = EOPNOTSUPP;
-               break;
-/* END UNIMPLEMENTED HOOKS */
+               ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat;
+               if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) {
+                       so2 = unp->unp_conn->unp_socket;
+                       ((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc;
+               }
+               ((struct stat *) m)->st_dev = NODEV;
+               if (unp->unp_ino == 0)
+                       unp->unp_ino = unp_ino++;
+               ((struct stat *) m)->st_ino = unp->unp_ino;
+               return (0);
 
        case PRU_RCVOOB:
 
        case PRU_RCVOOB:
-               break;
+               return (EOPNOTSUPP);
 
        case PRU_SENDOOB:
 
        case PRU_SENDOOB:
+               error = EOPNOTSUPP;
                break;
 
        case PRU_SOCKADDR:
                break;
 
        case PRU_SOCKADDR:
+               if (unp->unp_addr) {
+                       nam->m_len = unp->unp_addr->m_len;
+                       bcopy(mtod(unp->unp_addr, caddr_t),
+                           mtod(nam, caddr_t), (unsigned)nam->m_len);
+               } else
+                       nam->m_len = 0;
+               break;
+
+       case PRU_PEERADDR:
+               if (unp->unp_conn && unp->unp_conn->unp_addr) {
+                       nam->m_len = unp->unp_conn->unp_addr->m_len;
+                       bcopy(mtod(unp->unp_conn->unp_addr, caddr_t),
+                           mtod(nam, caddr_t), (unsigned)nam->m_len);
+               } else
+                       nam->m_len = 0;
                break;
 
        case PRU_SLOWTIMO:
                break;
 
        case PRU_SLOWTIMO:
@@ -184,12 +290,31 @@ uipc_usrreq(so, req, m, nam)
        default:
                panic("piusrreq");
        }
        default:
                panic("piusrreq");
        }
-       return (0);
+release:
+       if (control)
+               m_freem(control);
+       if (m)
+               m_freem(m);
+       return (error);
 }
 
 }
 
-int    unp_sendspace = 1024*2;
-int    unp_recvspace = 1024*2;
+/*
+ * Both send and receive buffers are allocated PIPSIZ bytes of buffering
+ * for stream sockets, although the total for sender and receiver is
+ * actually only PIPSIZ.
+ * Datagram sockets really use the sendspace as the maximum datagram size,
+ * and don't really want to reserve the sendspace.  Their recvspace should
+ * be large enough for at least one max-size datagram plus address.
+ */
+#define        PIPSIZ  4096
+u_long unpst_sendspace = PIPSIZ;
+u_long unpst_recvspace = PIPSIZ;
+u_long unpdg_sendspace = 2*1024;       /* really max datagram size */
+u_long unpdg_recvspace = 4*1024;
 
 
+int    unp_rights;                     /* file descriptors in flight */
+
+int
 unp_attach(so)
        struct socket *so;
 {
 unp_attach(so)
        struct socket *so;
 {
@@ -197,9 +322,23 @@ unp_attach(so)
        register struct unpcb *unp;
        int error;
        
        register struct unpcb *unp;
        int error;
        
-       error = soreserve(so, unp_sendspace, unp_recvspace);
-       if (error)
-               return (error);
+       if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
+               switch (so->so_type) {
+
+               case SOCK_STREAM:
+                       error = soreserve(so, unpst_sendspace, unpst_recvspace);
+                       break;
+
+               case SOCK_DGRAM:
+                       error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
+                       break;
+
+               default:
+                       panic("unp_attach");
+               }
+               if (error)
+                       return (error);
+       }
        m = m_getclr(M_DONTWAIT, MT_PCB);
        if (m == NULL)
                return (ENOBUFS);
        m = m_getclr(M_DONTWAIT, MT_PCB);
        if (m == NULL)
                return (ENOBUFS);
@@ -209,13 +348,15 @@ unp_attach(so)
        return (0);
 }
 
        return (0);
 }
 
+void
 unp_detach(unp)
        register struct unpcb *unp;
 {
        
 unp_detach(unp)
        register struct unpcb *unp;
 {
        
-       if (unp->unp_inode) {
-               irele(unp->unp_inode);
-               unp->unp_inode = 0;
+       if (unp->unp_vnode) {
+               unp->unp_vnode->v_socket = 0;
+               vrele(unp->unp_vnode);
+               unp->unp_vnode = 0;
        }
        if (unp->unp_conn)
                unp_disconnect(unp);
        }
        if (unp->unp_conn)
                unp_disconnect(unp);
@@ -223,103 +364,158 @@ unp_detach(unp)
                unp_drop(unp->unp_refs, ECONNRESET);
        soisdisconnected(unp->unp_socket);
        unp->unp_socket->so_pcb = 0;
                unp_drop(unp->unp_refs, ECONNRESET);
        soisdisconnected(unp->unp_socket);
        unp->unp_socket->so_pcb = 0;
-       m_freem(unp->unp_remaddr);
+       m_freem(unp->unp_addr);
        (void) m_free(dtom(unp));
        (void) m_free(dtom(unp));
+       if (unp_rights) {
+               /*
+                * Normally the receive buffer is flushed later,
+                * in sofree, but if our receive buffer holds references
+                * to descriptors that are now garbage, we will dispose
+                * of those descriptor references after the garbage collector
+                * gets them (resulting in a "panic: closef: count < 0").
+                */
+               sorflush(unp->unp_socket);
+               unp_gc();
+       }
 }
 
 }
 
-unp_bind(unp, nam)
+int
+unp_bind(unp, nam, p)
        struct unpcb *unp;
        struct mbuf *nam;
        struct unpcb *unp;
        struct mbuf *nam;
+       struct proc *p;
 {
        struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *);
 {
        struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *);
-       register struct inode *ip;
-       extern schar();
+       register struct vnode *vp;
+       struct vattr vattr;
        int error;
        int error;
-
-       u.u_dirp = soun->sun_path;
-       soun->sun_path[sizeof(soun->sun_path)-1] = 0;
-       ip = namei(schar, CREATE, 1);
-       if (ip) {
-               iput(ip);
+       struct nameidata nd;
+
+       NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT, UIO_SYSSPACE,
+           soun->sun_path, p);
+       if (unp->unp_vnode != NULL)
+               return (EINVAL);
+       if (nam->m_len == MLEN) {
+               if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0)
+                       return (EINVAL);
+       } else
+               *(mtod(nam, caddr_t) + nam->m_len) = 0;
+/* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
+       if (error = namei(&nd))
+               return (error);
+       vp = nd.ni_vp;
+       if (vp != NULL) {
+               VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
+               if (nd.ni_dvp == vp)
+                       vrele(nd.ni_dvp);
+               else
+                       vput(nd.ni_dvp);
+               vrele(vp);
                return (EADDRINUSE);
        }
                return (EADDRINUSE);
        }
-       ip = maknode(IFSOCK | 0777);
-       if (ip == NULL) {
-               error = u.u_error;              /* XXX */
-               u.u_error = 0;                  /* XXX */
+       VATTR_NULL(&vattr);
+       vattr.va_type = VSOCK;
+       vattr.va_mode = ACCESSPERMS;
+       VOP_LEASE(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE);
+       if (error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr))
                return (error);
                return (error);
-       }
-       ip->i_socket = unp->unp_socket;
-       unp->unp_inode = ip;
-       iunlock(ip);                    /* but keep reference */
+       vp = nd.ni_vp;
+       vp->v_socket = unp->unp_socket;
+       unp->unp_vnode = vp;
+       unp->unp_addr = m_copy(nam, 0, (int)M_COPYALL);
+       VOP_UNLOCK(vp, 0, p);
        return (0);
 }
 
        return (0);
 }
 
-unp_connect(so, nam)
+int
+unp_connect(so, nam, p)
        struct socket *so;
        struct mbuf *nam;
        struct socket *so;
        struct mbuf *nam;
+       struct proc *p;
 {
        register struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *);
 {
        register struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *);
-       struct unpcb *unp = sotounpcb(so);
-       register struct inode *ip;
+       register struct vnode *vp;
+       register struct socket *so2, *so3;
+       struct unpcb *unp2, *unp3;
        int error;
        int error;
-       struct socket *so2;
-       struct unpcb *unp2;
-
-       u.u_dirp = soun->sun_path;
-       soun->sun_path[sizeof(soun->sun_path)-1] = 0;
-       ip = namei(schar, LOOKUP, 1);
-       if (ip == 0) {
-               error = u.u_error;
-               u.u_error = 0;
-               return (error);         /* XXX */
-       }
-       if ((ip->i_mode&IFMT) != IFSOCK) {
+       struct nameidata nd;
+
+       NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, soun->sun_path, p);
+       if (nam->m_data + nam->m_len == &nam->m_dat[MLEN]) {    /* XXX */
+               if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0)
+                       return (EMSGSIZE);
+       } else
+               *(mtod(nam, caddr_t) + nam->m_len) = 0;
+       if (error = namei(&nd))
+               return (error);
+       vp = nd.ni_vp;
+       if (vp->v_type != VSOCK) {
                error = ENOTSOCK;
                goto bad;
        }
                error = ENOTSOCK;
                goto bad;
        }
-       so2 = ip->i_socket;
+       if (error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p))
+               goto bad;
+       so2 = vp->v_socket;
        if (so2 == 0) {
                error = ECONNREFUSED;
                goto bad;
        }
        if (so2 == 0) {
                error = ECONNREFUSED;
                goto bad;
        }
-       if (so2->so_type != so->so_type) {
+       if (so->so_type != so2->so_type) {
                error = EPROTOTYPE;
                goto bad;
        }
                error = EPROTOTYPE;
                goto bad;
        }
+       if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
+               if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
+                   (so3 = sonewconn(so2, 0)) == 0) {
+                       error = ECONNREFUSED;
+                       goto bad;
+               }
+               unp2 = sotounpcb(so2);
+               unp3 = sotounpcb(so3);
+               if (unp2->unp_addr)
+                       unp3->unp_addr =
+                                 m_copy(unp2->unp_addr, 0, (int)M_COPYALL);
+               so2 = so3;
+       }
+       error = unp_connect2(so, so2);
+bad:
+       vput(vp);
+       return (error);
+}
+
+int
+unp_connect2(so, so2)
+       register struct socket *so;
+       register struct socket *so2;
+{
+       register struct unpcb *unp = sotounpcb(so);
+       register struct unpcb *unp2;
+
+       if (so2->so_type != so->so_type)
+               return (EPROTOTYPE);
+       unp2 = sotounpcb(so2);
+       unp->unp_conn = unp2;
        switch (so->so_type) {
 
        case SOCK_DGRAM:
        switch (so->so_type) {
 
        case SOCK_DGRAM:
-               unp->unp_conn = sotounpcb(so2);
-               unp2 = sotounpcb(so2);
                unp->unp_nextref = unp2->unp_refs;
                unp2->unp_refs = unp;
                unp->unp_nextref = unp2->unp_refs;
                unp2->unp_refs = unp;
+               soisconnected(so);
                break;
 
        case SOCK_STREAM:
                break;
 
        case SOCK_STREAM:
-               if ((so2->so_options&SO_ACCEPTCONN) == 0 ||
-                   (so2 = sonewconn(so2)) == 0) {
-                       error = ECONNREFUSED;
-                       goto bad;
-               }
-               unp2 = sotounpcb(so2);
-               unp->unp_conn = unp2;
                unp2->unp_conn = unp;
                unp2->unp_conn = unp;
-               unp2->unp_remaddr = m_copy(nam, 0, (int)M_COPYALL);
+               soisconnected(so);
+               soisconnected(so2);
                break;
 
        default:
                break;
 
        default:
-               panic("uipc connip");
+               panic("unp_connect2");
        }
        }
-       soisconnected(so2);
-       soisconnected(so);
-       iput(ip);
        return (0);
        return (0);
-bad:
-       iput(ip);
-       return (error);
 }
 
 }
 
+void
 unp_disconnect(unp)
        struct unpcb *unp;
 {
 unp_disconnect(unp)
        struct unpcb *unp;
 {
@@ -328,7 +524,6 @@ unp_disconnect(unp)
        if (unp2 == 0)
                return;
        unp->unp_conn = 0;
        if (unp2 == 0)
                return;
        unp->unp_conn = 0;
-       soisdisconnected(unp->unp_socket);
        switch (unp->unp_socket->so_type) {
 
        case SOCK_DGRAM:
        switch (unp->unp_socket->so_type) {
 
        case SOCK_DGRAM:
@@ -346,39 +541,299 @@ unp_disconnect(unp)
                        unp2->unp_nextref = unp->unp_nextref;
                }
                unp->unp_nextref = 0;
                        unp2->unp_nextref = unp->unp_nextref;
                }
                unp->unp_nextref = 0;
+               unp->unp_socket->so_state &= ~SS_ISCONNECTED;
                break;
 
        case SOCK_STREAM:
                break;
 
        case SOCK_STREAM:
+               soisdisconnected(unp->unp_socket);
                unp2->unp_conn = 0;
                soisdisconnected(unp2->unp_socket);
                break;
        }
 }
 
                unp2->unp_conn = 0;
                soisdisconnected(unp2->unp_socket);
                break;
        }
 }
 
+#ifdef notdef
+void
 unp_abort(unp)
        struct unpcb *unp;
 {
 
        unp_detach(unp);
 }
 unp_abort(unp)
        struct unpcb *unp;
 {
 
        unp_detach(unp);
 }
+#endif
 
 
-/*ARGSUSED*/
-unp_usrclosed(unp)
+void
+unp_shutdown(unp)
        struct unpcb *unp;
 {
        struct unpcb *unp;
 {
+       struct socket *so;
 
 
+       if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
+           (so = unp->unp_conn->unp_socket))
+               socantrcvmore(so);
 }
 
 }
 
+void
 unp_drop(unp, errno)
        struct unpcb *unp;
        int errno;
 {
 unp_drop(unp, errno)
        struct unpcb *unp;
        int errno;
 {
+       struct socket *so = unp->unp_socket;
 
 
-       unp->unp_socket->so_error = errno;
+       so->so_error = errno;
        unp_disconnect(unp);
        unp_disconnect(unp);
+       if (so->so_head) {
+               so->so_pcb = (caddr_t) 0;
+               m_freem(unp->unp_addr);
+               (void) m_free(dtom(unp));
+               sofree(so);
+       }
 }
 
 }
 
+#ifdef notdef
 unp_drain()
 {
 
 }
 unp_drain()
 {
 
 }
+#endif
+
+int
+unp_externalize(rights)
+       struct mbuf *rights;
+{
+       struct proc *p = curproc;               /* XXX */
+       register int i;
+       register struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
+       register struct file **rp = (struct file **)(cm + 1);
+       register struct file *fp;
+       int newfds = (cm->cmsg_len - sizeof(*cm)) / sizeof (int);
+       int f;
+
+       if (!fdavail(p, newfds)) {
+               for (i = 0; i < newfds; i++) {
+                       fp = *rp;
+                       unp_discard(fp);
+                       *rp++ = 0;
+               }
+               return (EMSGSIZE);
+       }
+       for (i = 0; i < newfds; i++) {
+               if (fdalloc(p, 0, &f))
+                       panic("unp_externalize");
+               fp = *rp;
+               p->p_fd->fd_ofiles[f] = fp;
+               fp->f_msgcount--;
+               unp_rights--;
+               *(int *)rp++ = f;
+       }
+       return (0);
+}
+
+int
+unp_internalize(control, p)
+       struct mbuf *control;
+       struct proc *p;
+{
+       struct filedesc *fdp = p->p_fd;
+       register struct cmsghdr *cm = mtod(control, struct cmsghdr *);
+       register struct file **rp;
+       register struct file *fp;
+       register int i, fd;
+       int oldfds;
+
+       if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET ||
+           cm->cmsg_len != control->m_len)
+               return (EINVAL);
+       oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int);
+       rp = (struct file **)(cm + 1);
+       for (i = 0; i < oldfds; i++) {
+               fd = *(int *)rp++;
+               if ((unsigned)fd >= fdp->fd_nfiles ||
+                   fdp->fd_ofiles[fd] == NULL)
+                       return (EBADF);
+       }
+       rp = (struct file **)(cm + 1);
+       for (i = 0; i < oldfds; i++) {
+               fp = fdp->fd_ofiles[*(int *)rp];
+               *rp++ = fp;
+               fp->f_count++;
+               fp->f_msgcount++;
+               unp_rights++;
+       }
+       return (0);
+}
+
+int    unp_defer, unp_gcing;
+extern struct domain unixdomain;
+
+void
+unp_gc()
+{
+       register struct file *fp, *nextfp;
+       register struct socket *so;
+       struct file **extra_ref, **fpp;
+       int nunref, i;
+
+       if (unp_gcing)
+               return;
+       unp_gcing = 1;
+       unp_defer = 0;
+       for (fp = filehead.lh_first; fp != 0; fp = fp->f_list.le_next)
+               fp->f_flag &= ~(FMARK|FDEFER);
+       do {
+               for (fp = filehead.lh_first; fp != 0; fp = fp->f_list.le_next) {
+                       if (fp->f_count == 0)
+                               continue;
+                       if (fp->f_flag & FDEFER) {
+                               fp->f_flag &= ~FDEFER;
+                               unp_defer--;
+                       } else {
+                               if (fp->f_flag & FMARK)
+                                       continue;
+                               if (fp->f_count == fp->f_msgcount)
+                                       continue;
+                               fp->f_flag |= FMARK;
+                       }
+                       if (fp->f_type != DTYPE_SOCKET ||
+                           (so = (struct socket *)fp->f_data) == 0)
+                               continue;
+                       if (so->so_proto->pr_domain != &unixdomain ||
+                           (so->so_proto->pr_flags&PR_RIGHTS) == 0)
+                               continue;
+#ifdef notdef
+                       if (so->so_rcv.sb_flags & SB_LOCK) {
+                               /*
+                                * This is problematical; it's not clear
+                                * we need to wait for the sockbuf to be
+                                * unlocked (on a uniprocessor, at least),
+                                * and it's also not clear what to do
+                                * if sbwait returns an error due to receipt
+                                * of a signal.  If sbwait does return
+                                * an error, we'll go into an infinite
+                                * loop.  Delete all of this for now.
+                                */
+                               (void) sbwait(&so->so_rcv);
+                               goto restart;
+                       }
+#endif
+                       unp_scan(so->so_rcv.sb_mb, unp_mark);
+               }
+       } while (unp_defer);
+       /*
+        * We grab an extra reference to each of the file table entries
+        * that are not otherwise accessible and then free the rights
+        * that are stored in messages on them.
+        *
+        * The bug in the orginal code is a little tricky, so I'll describe
+        * what's wrong with it here.
+        *
+        * It is incorrect to simply unp_discard each entry for f_msgcount
+        * times -- consider the case of sockets A and B that contain
+        * references to each other.  On a last close of some other socket,
+        * we trigger a gc since the number of outstanding rights (unp_rights)
+        * is non-zero.  If during the sweep phase the gc code un_discards,
+        * we end up doing a (full) closef on the descriptor.  A closef on A
+        * results in the following chain.  Closef calls soo_close, which
+        * calls soclose.   Soclose calls first (through the switch
+        * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
+        * returns because the previous instance had set unp_gcing, and
+        * we return all the way back to soclose, which marks the socket
+        * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
+        * to free up the rights that are queued in messages on the socket A,
+        * i.e., the reference on B.  The sorflush calls via the dom_dispose
+        * switch unp_dispose, which unp_scans with unp_discard.  This second
+        * instance of unp_discard just calls closef on B.
+        *
+        * Well, a similar chain occurs on B, resulting in a sorflush on B,
+        * which results in another closef on A.  Unfortunately, A is already
+        * being closed, and the descriptor has already been marked with
+        * SS_NOFDREF, and soclose panics at this point.
+        *
+        * Here, we first take an extra reference to each inaccessible
+        * descriptor.  Then, we call sorflush ourself, since we know
+        * it is a Unix domain socket anyhow.  After we destroy all the
+        * rights carried in messages, we do a last closef to get rid
+        * of our extra reference.  This is the last close, and the
+        * unp_detach etc will shut down the socket.
+        *
+        * 91/09/19, bsy@cs.cmu.edu
+        */
+       extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK);
+       for (nunref = 0, fp = filehead.lh_first, fpp = extra_ref; fp != 0;
+           fp = nextfp) {
+               nextfp = fp->f_list.le_next;
+               if (fp->f_count == 0)
+                       continue;
+               if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
+                       *fpp++ = fp;
+                       nunref++;
+                       fp->f_count++;
+               }
+       }
+       for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
+               sorflush((struct socket *)(*fpp)->f_data);
+       for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
+               closef(*fpp, (struct proc *)NULL);
+       free((caddr_t)extra_ref, M_FILE);
+       unp_gcing = 0;
+}
+
+void
+unp_dispose(m)
+       struct mbuf *m;
+{
+
+       if (m)
+               unp_scan(m, unp_discard);
+}
+
+void
+unp_scan(m0, op)
+       register struct mbuf *m0;
+       void (*op) __P((struct file *));
+{
+       register struct mbuf *m;
+       register struct file **rp;
+       register struct cmsghdr *cm;
+       register int i;
+       int qfds;
+
+       while (m0) {
+               for (m = m0; m; m = m->m_next)
+                       if (m->m_type == MT_CONTROL &&
+                           m->m_len >= sizeof(*cm)) {
+                               cm = mtod(m, struct cmsghdr *);
+                               if (cm->cmsg_level != SOL_SOCKET ||
+                                   cm->cmsg_type != SCM_RIGHTS)
+                                       continue;
+                               qfds = (cm->cmsg_len - sizeof *cm)
+                                               / sizeof (struct file *);
+                               rp = (struct file **)(cm + 1);
+                               for (i = 0; i < qfds; i++)
+                                       (*op)(*rp++);
+                               break;          /* XXX, but saves time */
+                       }
+               m0 = m0->m_act;
+       }
+}
+
+void
+unp_mark(fp)
+       struct file *fp;
+{
+
+       if (fp->f_flag & FMARK)
+               return;
+       unp_defer++;
+       fp->f_flag |= (FMARK|FDEFER);
+}
+
+void
+unp_discard(fp)
+       struct file *fp;
+{
+
+       fp->f_msgcount--;
+       unp_rights--;
+       (void) closef(fp, (struct proc *)NULL);
+}