2 * Copyright (c) 1982, 1986, 1989, 1991, 1993
3 * The Regents of the University of California.
4 * Copyright 2004-2006 Robert N. M. Watson
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 * 4. Neither the name of the University nor the names of its contributors
16 * may be used to endorse or promote products derived from this software
17 * without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * From: @(#)uipc_usrreq.c 8.3 (Berkeley) 1/4/94
35 * UNIX Domain (Local) Sockets
37 * This is an implementation of UNIX (local) domain sockets. Each socket has
38 * an associated struct unpcb (UNIX protocol control block). Stream sockets
39 * may be connected to 0 or 1 other socket. Datagram sockets may be
40 * connected to 0, 1, or many other sockets. Sockets may be created and
41 * connected in pairs (socketpair(2)), or bound/connected to using the file
42 * system name space. For most purposes, only the receive socket buffer is
43 * used, as sending on one socket delivers directly to the receive socket
44 * buffer of a second socket. The implementation is substantially
45 * complicated by the fact that "ancillary data", such as file descriptors or
46 * credentials, may be passed across UNIX domain sockets. The potential for
47 * passing UNIX domain sockets over other UNIX domain sockets requires the
48 * implementation of a simple garbage collector to find and tear down cycles
49 * of disconnected sockets.
52 #include <sys/cdefs.h>
53 __FBSDID("$FreeBSD$");
57 #include <sys/param.h>
58 #include <sys/domain.h>
59 #include <sys/fcntl.h>
60 #include <sys/malloc.h> /* XXX must be before <sys/file.h> */
61 #include <sys/eventhandler.h>
63 #include <sys/filedesc.h>
65 #include <sys/kernel.h>
69 #include <sys/mount.h>
70 #include <sys/mutex.h>
71 #include <sys/namei.h>
73 #include <sys/protosw.h>
74 #include <sys/resourcevar.h>
75 #include <sys/socket.h>
76 #include <sys/socketvar.h>
77 #include <sys/signalvar.h>
80 #include <sys/sysctl.h>
81 #include <sys/systm.h>
82 #include <sys/taskqueue.h>
84 #include <sys/unpcb.h>
85 #include <sys/vnode.h>
89 static uma_zone_t unp_zone;
90 static unp_gen_t unp_gencnt;
91 static u_int unp_count;
93 static struct unp_head unp_shead, unp_dhead;
96 * Unix communications domain.
100 * rethink name space problems
101 * need a proper out-of-band
104 static const struct sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
105 static ino_t unp_ino; /* prototype for fake inode numbers */
106 struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
109 * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
110 * stream sockets, although the total for sender and receiver is actually
113 * Datagram sockets really use the sendspace as the maximum datagram size,
114 * and don't really want to reserve the sendspace. Their recvspace should be
115 * large enough for at least one max-size datagram plus address.
120 static u_long unpst_sendspace = PIPSIZ;
121 static u_long unpst_recvspace = PIPSIZ;
122 static u_long unpdg_sendspace = 2*1024; /* really max datagram size */
123 static u_long unpdg_recvspace = 4*1024;
125 static int unp_rights; /* file descriptors in flight */
127 SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
128 SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, "SOCK_STREAM");
129 SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
131 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
132 &unpst_sendspace, 0, "");
133 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
134 &unpst_recvspace, 0, "");
135 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
136 &unpdg_sendspace, 0, "");
137 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
138 &unpdg_recvspace, 0, "");
139 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
142 * Currently, UNIX domain sockets are protected by a single subsystem lock,
143 * which covers global data structures and variables, the contents of each
144 * per-socket unpcb structure, and the so_pcb field in sockets attached to
145 * the UNIX domain. This provides for a moderate degree of paralellism, as
146 * receive operations on UNIX domain sockets do not need to acquire the
147 * subsystem lock. Finer grained locking to permit send() without acquiring
148 * a global lock would be a logical next step.
150 * The UNIX domain socket lock preceds all socket layer locks, including the
151 * socket lock and socket buffer lock, permitting UNIX domain socket code to
152 * call into socket support routines without releasing its locks.
154 * Some caution is required in areas where the UNIX domain socket code enters
155 * VFS in order to create or find rendezvous points. This results in
156 * dropping of the UNIX domain socket subsystem lock, acquisition of the
157 * Giant lock, and potential sleeping. This increases the chances of races,
158 * and exposes weaknesses in the socket->protocol API by offering poor
161 static struct mtx unp_mtx;
162 #define UNP_LOCK_INIT() \
163 mtx_init(&unp_mtx, "unp", NULL, MTX_DEF)
164 #define UNP_LOCK() mtx_lock(&unp_mtx)
165 #define UNP_UNLOCK() mtx_unlock(&unp_mtx)
166 #define UNP_LOCK_ASSERT() mtx_assert(&unp_mtx, MA_OWNED)
167 #define UNP_UNLOCK_ASSERT() mtx_assert(&unp_mtx, MA_NOTOWNED)
170 * Garbage collection of cyclic file descriptor/socket references occurs
171 * asynchronously in a taskqueue context in order to avoid recursion and
172 * reentrance in the UNIX domain socket, file descriptor, and socket layer
173 * code. See unp_gc() for a full description.
175 static struct task unp_gc_task;
177 static int unp_connect(struct socket *,struct sockaddr *, struct thread *);
178 static int unp_connect2(struct socket *so, struct socket *so2, int);
179 static void unp_disconnect(struct unpcb *);
180 static void unp_shutdown(struct unpcb *);
181 static void unp_drop(struct unpcb *, int);
182 static void unp_gc(__unused void *, int);
183 static void unp_scan(struct mbuf *, void (*)(struct file *));
184 static void unp_mark(struct file *);
185 static void unp_discard(struct file *);
186 static void unp_freerights(struct file **, int);
187 static int unp_internalize(struct mbuf **, struct thread *);
188 static int unp_listen(struct socket *, struct unpcb *, int,
192 * Definitions of protocols supported in the LOCAL domain.
194 static struct domain localdomain;
195 static struct protosw localsw[] = {
197 .pr_type = SOCK_STREAM,
198 .pr_domain = &localdomain,
199 .pr_flags = PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
200 .pr_ctloutput = &uipc_ctloutput,
201 .pr_usrreqs = &uipc_usrreqs
204 .pr_type = SOCK_DGRAM,
205 .pr_domain = &localdomain,
206 .pr_flags = PR_ATOMIC|PR_ADDR|PR_RIGHTS,
207 .pr_usrreqs = &uipc_usrreqs
211 static struct domain localdomain = {
212 .dom_family = AF_LOCAL,
214 .dom_init = unp_init,
215 .dom_externalize = unp_externalize,
216 .dom_dispose = unp_dispose,
217 .dom_protosw = localsw,
218 .dom_protoswNPROTOSW = &localsw[sizeof(localsw)/sizeof(localsw[0])]
223 uipc_abort(struct socket *so)
228 KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
230 unp_drop(unp, ECONNABORTED);
235 uipc_accept(struct socket *so, struct sockaddr **nam)
238 const struct sockaddr *sa;
241 * Pass back name of connected socket, if it was bound and we are
242 * still connected (our peer may have closed already!).
245 KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
246 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
248 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL)
249 sa = (struct sockaddr *) unp->unp_conn->unp_addr;
252 bcopy(sa, *nam, sa->sa_len);
258 uipc_attach(struct socket *so, int proto, struct thread *td)
263 KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
264 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
265 switch (so->so_type) {
267 error = soreserve(so, unpst_sendspace, unpst_recvspace);
271 error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
280 unp = uma_zalloc(unp_zone, M_WAITOK | M_ZERO);
283 LIST_INIT(&unp->unp_refs);
284 unp->unp_socket = so;
288 unp->unp_gencnt = ++unp_gencnt;
290 LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead : &unp_shead,
298 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
300 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
310 KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
312 namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
317 * We don't allow simultaneous bind() calls on a single UNIX domain
318 * socket, so flag in-progress operations, and return an error if an
319 * operation is already in progress.
321 * Historically, we have not allowed a socket to be rebound, so this
322 * also returns an error. Not allowing re-binding certainly
323 * simplifies the implementation and avoids a great many possible
327 if (unp->unp_vnode != NULL) {
331 if (unp->unp_flags & UNP_BINDING) {
335 unp->unp_flags |= UNP_BINDING;
338 buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
339 strlcpy(buf, soun->sun_path, namelen + 1);
343 mtx_assert(&Giant, MA_OWNED);
344 NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE,
346 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
351 if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
352 NDFREE(&nd, NDF_ONLY_PNBUF);
362 error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
368 vattr.va_type = VSOCK;
369 vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
371 error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
375 VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
376 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
378 NDFREE(&nd, NDF_ONLY_PNBUF);
381 vn_finished_write(mp);
385 ASSERT_VOP_LOCKED(vp, "uipc_bind");
386 soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
388 vp->v_socket = unp->unp_socket;
390 unp->unp_addr = soun;
391 unp->unp_flags &= ~UNP_BINDING;
393 VOP_UNLOCK(vp, 0, td);
394 vn_finished_write(mp);
400 unp->unp_flags &= ~UNP_BINDING;
408 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
412 KASSERT(td == curthread, ("uipc_connect: td != curthread"));
414 error = unp_connect(so, nam, td);
420 * XXXRW: Should also unbind?
423 uipc_close(struct socket *so)
428 KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
435 uipc_connect2(struct socket *so1, struct socket *so2)
440 unp = sotounpcb(so1);
441 KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
443 error = unp_connect2(so1, so2, PRU_CONNECT2);
448 /* control is EOPNOTSUPP */
451 uipc_detach(struct socket *so)
453 int local_unp_rights;
458 KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
460 LIST_REMOVE(unp, unp_link);
461 unp->unp_gencnt = ++unp_gencnt;
463 if ((vp = unp->unp_vnode) != NULL) {
465 * XXXRW: should v_socket be frobbed only while holding
468 unp->unp_vnode->v_socket = NULL;
469 unp->unp_vnode = NULL;
471 if (unp->unp_conn != NULL)
473 while (!LIST_EMPTY(&unp->unp_refs)) {
474 struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
475 unp_drop(ref, ECONNRESET);
477 unp->unp_socket->so_pcb = NULL;
478 local_unp_rights = unp_rights;
480 if (unp->unp_addr != NULL)
481 FREE(unp->unp_addr, M_SONAME);
482 uma_zfree(unp_zone, unp);
486 vfslocked = VFS_LOCK_GIANT(vp->v_mount);
488 VFS_UNLOCK_GIANT(vfslocked);
490 if (local_unp_rights)
491 taskqueue_enqueue(taskqueue_thread, &unp_gc_task);
495 uipc_disconnect(struct socket *so)
500 KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
508 uipc_listen(struct socket *so, int backlog, struct thread *td)
514 KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
516 if (unp->unp_vnode == NULL) {
520 error = unp_listen(so, unp, backlog, td);
526 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
529 const struct sockaddr *sa;
532 KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
533 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
535 if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL)
536 sa = (struct sockaddr *) unp->unp_conn->unp_addr;
539 * XXX: It seems that this test always fails even when
540 * connection is established. So, this else clause is
541 * added as workaround to return PF_LOCAL sockaddr.
545 bcopy(sa, *nam, sa->sa_len);
551 uipc_rcvd(struct socket *so, int flags)
559 KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
560 switch (so->so_type) {
562 panic("uipc_rcvd DGRAM?");
567 * Adjust backpressure on sender and wakeup any waiting to
570 SOCKBUF_LOCK(&so->so_rcv);
571 mbcnt = so->so_rcv.sb_mbcnt;
572 sbcc = so->so_rcv.sb_cc;
573 SOCKBUF_UNLOCK(&so->so_rcv);
575 if (unp->unp_conn == NULL) {
579 so2 = unp->unp_conn->unp_socket;
580 SOCKBUF_LOCK(&so2->so_snd);
581 so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt;
582 newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc;
583 (void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
584 newhiwat, RLIM_INFINITY);
585 sowwakeup_locked(so2);
586 unp->unp_mbcnt = mbcnt;
592 panic("uipc_rcvd unknown socktype");
597 /* pru_rcvoob is EOPNOTSUPP */
600 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
601 struct mbuf *control, struct thread *td)
603 struct unpcb *unp, *unp2;
610 KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
611 if (flags & PRUS_OOB) {
616 if (control != NULL && (error = unp_internalize(&control, td)))
620 switch (so->so_type) {
623 const struct sockaddr *from;
626 if (unp->unp_conn != NULL) {
630 error = unp_connect(so, nam, td);
635 * Because connect() and send() are non-atomic in a sendto()
636 * with a target address, it's possible that the socket will
637 * have disconnected before the send() can run. In that case
638 * return the slightly counter-intuitive but otherwise
639 * correct error that the socket is not connected.
641 unp2 = unp->unp_conn;
646 so2 = unp2->unp_socket;
647 if (unp->unp_addr != NULL)
648 from = (struct sockaddr *)unp->unp_addr;
651 if (unp2->unp_flags & UNP_WANTCRED)
652 control = unp_addsockcred(td, control);
653 SOCKBUF_LOCK(&so2->so_rcv);
654 if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
655 sorwakeup_locked(so2);
659 SOCKBUF_UNLOCK(&so2->so_rcv);
669 * Connect if not connected yet.
671 * Note: A better implementation would complain if not equal
672 * to the peer's address.
674 if ((so->so_state & SS_ISCONNECTED) == 0) {
676 error = unp_connect(so, nam, td);
686 if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
691 * Because connect() and send() are non-atomic in a sendto()
692 * with a target address, it's possible that the socket will
693 * have disconnected before the send() can run. In that case
694 * return the slightly counter-intuitive but otherwise
695 * correct error that the socket is not connected.
697 unp2 = unp->unp_conn;
702 so2 = unp2->unp_socket;
703 SOCKBUF_LOCK(&so2->so_rcv);
704 if (unp2->unp_flags & UNP_WANTCRED) {
706 * Credentials are passed only once on
709 unp2->unp_flags &= ~UNP_WANTCRED;
710 control = unp_addsockcred(td, control);
713 * Send to paired receive port, and then reduce send buffer
714 * hiwater marks to maintain backpressure. Wake up readers.
716 if (control != NULL) {
717 if (sbappendcontrol_locked(&so2->so_rcv, m, control))
720 sbappend_locked(&so2->so_rcv, m);
722 mbcnt = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt;
723 unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt;
724 sbcc = so2->so_rcv.sb_cc;
725 sorwakeup_locked(so2);
727 SOCKBUF_LOCK(&so->so_snd);
728 newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc);
729 (void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
730 newhiwat, RLIM_INFINITY);
731 so->so_snd.sb_mbmax -= mbcnt;
732 SOCKBUF_UNLOCK(&so->so_snd);
739 panic("uipc_send unknown socktype");
743 * SEND_EOF is equivalent to a SEND followed by
746 if (flags & PRUS_EOF) {
752 if (control != NULL && error != 0)
753 unp_dispose(control);
764 uipc_sense(struct socket *so, struct stat *sb)
770 KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
772 sb->st_blksize = so->so_snd.sb_hiwat;
773 if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) {
774 so2 = unp->unp_conn->unp_socket;
775 sb->st_blksize += so2->so_rcv.sb_cc;
778 if (unp->unp_ino == 0)
779 unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
780 sb->st_ino = unp->unp_ino;
786 uipc_shutdown(struct socket *so)
791 KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
800 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
803 const struct sockaddr *sa;
806 KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
807 *nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
809 if (unp->unp_addr != NULL)
810 sa = (struct sockaddr *) unp->unp_addr;
813 bcopy(sa, *nam, sa->sa_len);
818 struct pr_usrreqs uipc_usrreqs = {
819 .pru_abort = uipc_abort,
820 .pru_accept = uipc_accept,
821 .pru_attach = uipc_attach,
822 .pru_bind = uipc_bind,
823 .pru_connect = uipc_connect,
824 .pru_connect2 = uipc_connect2,
825 .pru_detach = uipc_detach,
826 .pru_disconnect = uipc_disconnect,
827 .pru_listen = uipc_listen,
828 .pru_peeraddr = uipc_peeraddr,
829 .pru_rcvd = uipc_rcvd,
830 .pru_send = uipc_send,
831 .pru_sense = uipc_sense,
832 .pru_shutdown = uipc_shutdown,
833 .pru_sockaddr = uipc_sockaddr,
834 .pru_close = uipc_close,
838 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
844 if (sopt->sopt_level != 0)
848 KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
851 switch (sopt->sopt_dir) {
853 switch (sopt->sopt_name) {
855 if (unp->unp_flags & UNP_HAVEPC)
856 xu = unp->unp_peercred;
858 if (so->so_type == SOCK_STREAM)
864 error = sooptcopyout(sopt, &xu, sizeof(xu));
867 optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
868 error = sooptcopyout(sopt, &optval, sizeof(optval));
871 optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
872 error = sooptcopyout(sopt, &optval, sizeof(optval));
880 switch (sopt->sopt_name) {
883 error = sooptcopyin(sopt, &optval, sizeof(optval),
888 #define OPTSET(bit) \
890 unp->unp_flags |= bit; \
892 unp->unp_flags &= ~bit;
894 switch (sopt->sopt_name) {
896 OPTSET(UNP_WANTCRED);
899 OPTSET(UNP_CONNWAIT);
920 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
922 struct sockaddr_un *soun = (struct sockaddr_un *)nam;
924 struct socket *so2, *so3;
925 struct unpcb *unp, *unp2, *unp3;
928 char buf[SOCK_MAXADDRLEN];
934 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
935 len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
938 strlcpy(buf, soun->sun_path, len + 1);
939 if (unp->unp_flags & UNP_CONNECTING) {
944 sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
946 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td);
952 ASSERT_VOP_LOCKED(vp, "unp_connect");
953 NDFREE(&nd, NDF_ONLY_PNBUF);
957 if (vp->v_type != VSOCK) {
961 error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
967 KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
970 error = ECONNREFUSED;
973 if (so->so_type != so2->so_type) {
977 if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
978 if (so2->so_options & SO_ACCEPTCONN) {
980 * NB: drop locks here so unp_attach is entered w/o
981 * locks; this avoids a recursive lock of the head
982 * and holding sleep locks across a (potentially)
986 so3 = sonewconn(so2, 0);
991 error = ECONNREFUSED;
995 unp2 = sotounpcb(so2);
996 unp3 = sotounpcb(so3);
997 if (unp2->unp_addr != NULL) {
998 bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
999 unp3->unp_addr = (struct sockaddr_un *) sa;
1003 * unp_peercred management:
1005 * The connecter's (client's) credentials are copied from its
1006 * process structure at the time of connect() (which is now).
1008 cru2x(td->td_ucred, &unp3->unp_peercred);
1009 unp3->unp_flags |= UNP_HAVEPC;
1011 * The receiver's (server's) credentials are copied from the
1012 * unp_peercred member of socket on which the former called
1013 * listen(); unp_listen() cached that process's credentials
1014 * at that time so we can use them now.
1016 KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1017 ("unp_connect: listener without cached peercred"));
1018 memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1019 sizeof(unp->unp_peercred));
1020 unp->unp_flags |= UNP_HAVEPC;
1021 if (unp2->unp_flags & UNP_WANTCRED)
1022 unp3->unp_flags |= UNP_WANTCRED;
1025 mac_set_socket_peer_from_socket(so, so3);
1026 mac_set_socket_peer_from_socket(so3, so);
1032 error = unp_connect2(so, so2, PRU_CONNECT);
1037 mtx_assert(&Giant, MA_OWNED);
1043 unp->unp_flags &= ~UNP_CONNECTING;
1048 unp_connect2(struct socket *so, struct socket *so2, int req)
1050 struct unpcb *unp = sotounpcb(so);
1055 if (so2->so_type != so->so_type)
1056 return (EPROTOTYPE);
1057 unp2 = sotounpcb(so2);
1058 KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1059 unp->unp_conn = unp2;
1060 switch (so->so_type) {
1062 LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1067 unp2->unp_conn = unp;
1068 if (req == PRU_CONNECT &&
1069 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1077 panic("unp_connect2");
1083 unp_disconnect(struct unpcb *unp)
1085 struct unpcb *unp2 = unp->unp_conn;
1092 unp->unp_conn = NULL;
1093 switch (unp->unp_socket->so_type) {
1095 LIST_REMOVE(unp, unp_reflink);
1096 so = unp->unp_socket;
1098 so->so_state &= ~SS_ISCONNECTED;
1103 soisdisconnected(unp->unp_socket);
1104 unp2->unp_conn = NULL;
1105 soisdisconnected(unp2->unp_socket);
1111 * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed by
1112 * the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers are
1113 * safe to reference. It first scans the list of struct unpcb's to generate
1114 * a pointer list, then it rescans its list one entry at a time to
1115 * externalize and copyout. It checks the generation number to see if a
1116 * struct unpcb has been reused, and will skip it if so.
1119 unp_pcblist(SYSCTL_HANDLER_ARGS)
1122 struct unpcb *unp, **unp_list;
1124 struct xunpgen *xug;
1125 struct unp_head *head;
1128 head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1131 * The process of preparing the PCB list is too time-consuming and
1132 * resource-intensive to repeat twice on every request.
1134 if (req->oldptr == NULL) {
1136 req->oldidx = 2 * (sizeof *xug)
1137 + (n + n/8) * sizeof(struct xunpcb);
1141 if (req->newptr != NULL)
1145 * OK, now we're committed to doing something.
1147 xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1149 gencnt = unp_gencnt;
1153 xug->xug_len = sizeof *xug;
1155 xug->xug_gen = gencnt;
1156 xug->xug_sogen = so_gencnt;
1157 error = SYSCTL_OUT(req, xug, sizeof *xug);
1163 unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1166 for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1167 unp = LIST_NEXT(unp, unp_link)) {
1168 if (unp->unp_gencnt <= gencnt) {
1169 if (cr_cansee(req->td->td_ucred,
1170 unp->unp_socket->so_cred))
1172 unp_list[i++] = unp;
1176 n = i; /* In case we lost some during malloc. */
1179 xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1180 for (i = 0; i < n; i++) {
1182 if (unp->unp_gencnt <= gencnt) {
1183 xu->xu_len = sizeof *xu;
1186 * XXX - need more locking here to protect against
1187 * connect/disconnect races for SMP.
1189 if (unp->unp_addr != NULL)
1190 bcopy(unp->unp_addr, &xu->xu_addr,
1191 unp->unp_addr->sun_len);
1192 if (unp->unp_conn != NULL &&
1193 unp->unp_conn->unp_addr != NULL)
1194 bcopy(unp->unp_conn->unp_addr,
1196 unp->unp_conn->unp_addr->sun_len);
1197 bcopy(unp, &xu->xu_unp, sizeof *unp);
1198 sotoxsocket(unp->unp_socket, &xu->xu_socket);
1199 error = SYSCTL_OUT(req, xu, sizeof *xu);
1205 * Give the user an updated idea of our state. If the
1206 * generation differs from what we told her before, she knows
1207 * that something happened while we were processing this
1208 * request, and it might be necessary to retry.
1210 xug->xug_gen = unp_gencnt;
1211 xug->xug_sogen = so_gencnt;
1212 xug->xug_count = unp_count;
1213 error = SYSCTL_OUT(req, xug, sizeof *xug);
1215 free(unp_list, M_TEMP);
1220 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1221 (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1222 "List of active local datagram sockets");
1223 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1224 (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1225 "List of active local stream sockets");
1228 unp_shutdown(struct unpcb *unp)
1234 if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
1235 (so = unp->unp_conn->unp_socket))
1240 unp_drop(struct unpcb *unp, int errno)
1242 struct socket *so = unp->unp_socket;
1246 so->so_error = errno;
1247 unp_disconnect(unp);
1251 unp_freerights(struct file **rp, int fdcount)
1256 for (i = 0; i < fdcount; i++) {
1259 * Zero the pointer before calling unp_discard since it may
1260 * end up in unp_gc()..
1262 * XXXRW: This is less true than it used to be.
1270 unp_externalize(struct mbuf *control, struct mbuf **controlp)
1272 struct thread *td = curthread; /* XXX */
1273 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1279 socklen_t clen = control->m_len, datalen;
1284 UNP_UNLOCK_ASSERT();
1287 if (controlp != NULL) /* controlp == NULL => free control messages */
1290 while (cm != NULL) {
1291 if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1296 data = CMSG_DATA(cm);
1297 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1299 if (cm->cmsg_level == SOL_SOCKET
1300 && cm->cmsg_type == SCM_RIGHTS) {
1301 newfds = datalen / sizeof(struct file *);
1304 /* If we're not outputting the descriptors free them. */
1305 if (error || controlp == NULL) {
1306 unp_freerights(rp, newfds);
1309 FILEDESC_LOCK(td->td_proc->p_fd);
1310 /* if the new FD's will not fit free them. */
1311 if (!fdavail(td, newfds)) {
1312 FILEDESC_UNLOCK(td->td_proc->p_fd);
1314 unp_freerights(rp, newfds);
1318 * Now change each pointer to an fd in the global
1319 * table to an integer that is the index to the local
1320 * fd table entry that we set up to point to the
1321 * global one we are transferring.
1323 newlen = newfds * sizeof(int);
1324 *controlp = sbcreatecontrol(NULL, newlen,
1325 SCM_RIGHTS, SOL_SOCKET);
1326 if (*controlp == NULL) {
1327 FILEDESC_UNLOCK(td->td_proc->p_fd);
1329 unp_freerights(rp, newfds);
1334 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1335 for (i = 0; i < newfds; i++) {
1336 if (fdalloc(td, 0, &f))
1337 panic("unp_externalize fdalloc failed");
1339 td->td_proc->p_fd->fd_ofiles[f] = fp;
1346 FILEDESC_UNLOCK(td->td_proc->p_fd);
1348 /* We can just copy anything else across. */
1349 if (error || controlp == NULL)
1351 *controlp = sbcreatecontrol(NULL, datalen,
1352 cm->cmsg_type, cm->cmsg_level);
1353 if (*controlp == NULL) {
1358 CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1362 controlp = &(*controlp)->m_next;
1365 if (CMSG_SPACE(datalen) < clen) {
1366 clen -= CMSG_SPACE(datalen);
1367 cm = (struct cmsghdr *)
1368 ((caddr_t)cm + CMSG_SPACE(datalen));
1381 unp_zone_change(void *tag)
1384 uma_zone_set_max(unp_zone, maxsockets);
1391 unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1392 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1393 if (unp_zone == NULL)
1395 uma_zone_set_max(unp_zone, maxsockets);
1396 EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1397 NULL, EVENTHANDLER_PRI_ANY);
1398 LIST_INIT(&unp_dhead);
1399 LIST_INIT(&unp_shead);
1400 TASK_INIT(&unp_gc_task, 0, unp_gc, NULL);
1405 unp_internalize(struct mbuf **controlp, struct thread *td)
1407 struct mbuf *control = *controlp;
1408 struct proc *p = td->td_proc;
1409 struct filedesc *fdescp = p->p_fd;
1410 struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1411 struct cmsgcred *cmcred;
1417 socklen_t clen = control->m_len, datalen;
1421 UNP_UNLOCK_ASSERT();
1426 while (cm != NULL) {
1427 if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1428 || cm->cmsg_len > clen) {
1433 data = CMSG_DATA(cm);
1434 datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1436 switch (cm->cmsg_type) {
1438 * Fill in credential information.
1441 *controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1442 SCM_CREDS, SOL_SOCKET);
1443 if (*controlp == NULL) {
1448 cmcred = (struct cmsgcred *)
1449 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1450 cmcred->cmcred_pid = p->p_pid;
1451 cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1452 cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1453 cmcred->cmcred_euid = td->td_ucred->cr_uid;
1454 cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1456 for (i = 0; i < cmcred->cmcred_ngroups; i++)
1457 cmcred->cmcred_groups[i] =
1458 td->td_ucred->cr_groups[i];
1462 oldfds = datalen / sizeof (int);
1464 * Check that all the FDs passed in refer to legal
1465 * files. If not, reject the entire operation.
1468 FILEDESC_LOCK(fdescp);
1469 for (i = 0; i < oldfds; i++) {
1471 if ((unsigned)fd >= fdescp->fd_nfiles ||
1472 fdescp->fd_ofiles[fd] == NULL) {
1473 FILEDESC_UNLOCK(fdescp);
1477 fp = fdescp->fd_ofiles[fd];
1478 if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1479 FILEDESC_UNLOCK(fdescp);
1486 * Now replace the integer FDs with pointers to the
1487 * associated global file table entry..
1489 newlen = oldfds * sizeof(struct file *);
1490 *controlp = sbcreatecontrol(NULL, newlen,
1491 SCM_RIGHTS, SOL_SOCKET);
1492 if (*controlp == NULL) {
1493 FILEDESC_UNLOCK(fdescp);
1499 rp = (struct file **)
1500 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1501 for (i = 0; i < oldfds; i++) {
1502 fp = fdescp->fd_ofiles[*fdp++];
1510 FILEDESC_UNLOCK(fdescp);
1514 *controlp = sbcreatecontrol(NULL, sizeof(*tv),
1515 SCM_TIMESTAMP, SOL_SOCKET);
1516 if (*controlp == NULL) {
1520 tv = (struct timeval *)
1521 CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1530 controlp = &(*controlp)->m_next;
1532 if (CMSG_SPACE(datalen) < clen) {
1533 clen -= CMSG_SPACE(datalen);
1534 cm = (struct cmsghdr *)
1535 ((caddr_t)cm + CMSG_SPACE(datalen));
1549 unp_addsockcred(struct thread *td, struct mbuf *control)
1551 struct mbuf *m, *n, *n_prev;
1552 struct sockcred *sc;
1553 const struct cmsghdr *cm;
1557 ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1559 m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
1563 sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
1564 sc->sc_uid = td->td_ucred->cr_ruid;
1565 sc->sc_euid = td->td_ucred->cr_uid;
1566 sc->sc_gid = td->td_ucred->cr_rgid;
1567 sc->sc_egid = td->td_ucred->cr_gid;
1568 sc->sc_ngroups = ngroups;
1569 for (i = 0; i < sc->sc_ngroups; i++)
1570 sc->sc_groups[i] = td->td_ucred->cr_groups[i];
1573 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
1574 * created SCM_CREDS control message (struct sockcred) has another
1577 if (control != NULL)
1578 for (n = control, n_prev = NULL; n != NULL;) {
1579 cm = mtod(n, struct cmsghdr *);
1580 if (cm->cmsg_level == SOL_SOCKET &&
1581 cm->cmsg_type == SCM_CREDS) {
1583 control = n->m_next;
1585 n_prev->m_next = n->m_next;
1593 /* Prepend it to the head. */
1594 m->m_next = control;
1600 * unp_defer indicates whether additional work has been defered for a future
1601 * pass through unp_gc(). It is thread local and does not require explicit
1604 static int unp_defer;
1606 static int unp_taskcount;
1607 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, "");
1609 static int unp_recycled;
1610 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, "");
1613 unp_gc(__unused void *arg, int pending)
1615 struct file *fp, *nextfp;
1617 struct file **extra_ref, **fpp;
1620 int nfiles_slack = 20;
1625 * Before going through all this, set all FDs to be NOT defered and
1626 * NOT externally accessible.
1628 sx_slock(&filelist_lock);
1629 LIST_FOREACH(fp, &filehead, f_list)
1630 fp->f_gcflag &= ~(FMARK|FDEFER);
1632 KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer));
1633 LIST_FOREACH(fp, &filehead, f_list) {
1636 * If the file is not open, skip it -- could be a
1637 * file in the process of being opened, or in the
1638 * process of being closed. If the file is
1639 * "closing", it may have been marked for deferred
1640 * consideration. Clear the flag now if so.
1642 if (fp->f_count == 0) {
1643 if (fp->f_gcflag & FDEFER)
1645 fp->f_gcflag &= ~(FMARK|FDEFER);
1650 * If we already marked it as 'defer' in a previous
1651 * pass, then try process it this time and un-mark
1654 if (fp->f_gcflag & FDEFER) {
1655 fp->f_gcflag &= ~FDEFER;
1659 * if it's not defered, then check if it's
1660 * already marked.. if so skip it
1662 if (fp->f_gcflag & FMARK) {
1667 * If all references are from messages in
1668 * transit, then skip it. it's not externally
1671 if (fp->f_count == fp->f_msgcount) {
1676 * If it got this far then it must be
1677 * externally accessible.
1679 fp->f_gcflag |= FMARK;
1682 * Either it was defered, or it is externally
1683 * accessible and not already marked so. Now check
1684 * if it is possibly one of OUR sockets.
1686 if (fp->f_type != DTYPE_SOCKET ||
1687 (so = fp->f_data) == NULL) {
1692 if (so->so_proto->pr_domain != &localdomain ||
1693 (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1696 * So, Ok, it's one of our sockets and it IS
1697 * externally accessible (or was defered). Now we
1698 * look to see if we hold any file descriptors in its
1699 * message buffers. Follow those links and mark them
1700 * as accessible too.
1702 SOCKBUF_LOCK(&so->so_rcv);
1703 unp_scan(so->so_rcv.sb_mb, unp_mark);
1704 SOCKBUF_UNLOCK(&so->so_rcv);
1706 } while (unp_defer);
1707 sx_sunlock(&filelist_lock);
1709 * XXXRW: The following comments need updating for a post-SMPng and
1710 * deferred unp_gc() world, but are still generally accurate.
1712 * We grab an extra reference to each of the file table entries that
1713 * are not otherwise accessible and then free the rights that are
1714 * stored in messages on them.
1716 * The bug in the orginal code is a little tricky, so I'll describe
1717 * what's wrong with it here.
1719 * It is incorrect to simply unp_discard each entry for f_msgcount
1720 * times -- consider the case of sockets A and B that contain
1721 * references to each other. On a last close of some other socket,
1722 * we trigger a gc since the number of outstanding rights (unp_rights)
1723 * is non-zero. If during the sweep phase the gc code unp_discards,
1724 * we end up doing a (full) closef on the descriptor. A closef on A
1725 * results in the following chain. Closef calls soo_close, which
1726 * calls soclose. Soclose calls first (through the switch
1727 * uipc_usrreq) unp_detach, which re-invokes unp_gc. Unp_gc simply
1728 * returns because the previous instance had set unp_gcing, and we
1729 * return all the way back to soclose, which marks the socket with
1730 * SS_NOFDREF, and then calls sofree. Sofree calls sorflush to free
1731 * up the rights that are queued in messages on the socket A, i.e.,
1732 * the reference on B. The sorflush calls via the dom_dispose switch
1733 * unp_dispose, which unp_scans with unp_discard. This second
1734 * instance of unp_discard just calls closef on B.
1736 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1737 * which results in another closef on A. Unfortunately, A is already
1738 * being closed, and the descriptor has already been marked with
1739 * SS_NOFDREF, and soclose panics at this point.
1741 * Here, we first take an extra reference to each inaccessible
1742 * descriptor. Then, we call sorflush ourself, since we know it is a
1743 * Unix domain socket anyhow. After we destroy all the rights
1744 * carried in messages, we do a last closef to get rid of our extra
1745 * reference. This is the last close, and the unp_detach etc will
1746 * shut down the socket.
1748 * 91/09/19, bsy@cs.cmu.edu
1751 nfiles_snap = openfiles + nfiles_slack; /* some slack */
1752 extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP,
1754 sx_slock(&filelist_lock);
1755 if (nfiles_snap < openfiles) {
1756 sx_sunlock(&filelist_lock);
1757 free(extra_ref, M_TEMP);
1761 for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref;
1762 fp != NULL; fp = nextfp) {
1763 nextfp = LIST_NEXT(fp, f_list);
1766 * If it's not open, skip it
1768 if (fp->f_count == 0) {
1773 * If all refs are from msgs, and it's not marked accessible
1774 * then it must be referenced from some unreachable cycle of
1775 * (shut-down) FDs, so include it in our list of FDs to
1778 if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) {
1785 sx_sunlock(&filelist_lock);
1787 * For each FD on our hit list, do the following two things:
1789 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1790 struct file *tfp = *fpp;
1792 if (tfp->f_type == DTYPE_SOCKET &&
1793 tfp->f_data != NULL) {
1795 sorflush(tfp->f_data);
1800 for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1801 closef(*fpp, (struct thread *) NULL);
1804 free(extra_ref, M_TEMP);
1808 unp_dispose(struct mbuf *m)
1812 unp_scan(m, unp_discard);
1816 unp_listen(struct socket *so, struct unpcb *unp, int backlog,
1824 error = solisten_proto_check(so);
1826 cru2x(td->td_ucred, &unp->unp_peercred);
1827 unp->unp_flags |= UNP_HAVEPCCACHED;
1828 solisten_proto(so, backlog);
1835 unp_scan(struct mbuf *m0, void (*op)(struct file *))
1842 socklen_t clen, datalen;
1845 while (m0 != NULL) {
1846 for (m = m0; m; m = m->m_next) {
1847 if (m->m_type != MT_CONTROL)
1850 cm = mtod(m, struct cmsghdr *);
1853 while (cm != NULL) {
1854 if (sizeof(*cm) > clen || cm->cmsg_len > clen)
1857 data = CMSG_DATA(cm);
1858 datalen = (caddr_t)cm + cm->cmsg_len
1861 if (cm->cmsg_level == SOL_SOCKET &&
1862 cm->cmsg_type == SCM_RIGHTS) {
1863 qfds = datalen / sizeof (struct file *);
1865 for (i = 0; i < qfds; i++)
1869 if (CMSG_SPACE(datalen) < clen) {
1870 clen -= CMSG_SPACE(datalen);
1871 cm = (struct cmsghdr *)
1872 ((caddr_t)cm + CMSG_SPACE(datalen));
1884 unp_mark(struct file *fp)
1886 if (fp->f_gcflag & FMARK)
1889 fp->f_gcflag |= (FMARK|FDEFER);
1893 unp_discard(struct file *fp)
1901 (void) closef(fp, (struct thread *)NULL);