xref: /freebsd/sys/kern/uipc_usrreq.c (revision ce834215a70ff69e7e222827437116eee2f9ac6f)
1 /*
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 3. All advertising materials mentioning features or use of this software
14  *    must display the following acknowledgement:
15  *	This product includes software developed by the University of
16  *	California, Berkeley and its contributors.
17  * 4. Neither the name of the University nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  *
33  *	From: @(#)uipc_usrreq.c	8.3 (Berkeley) 1/4/94
34  *	$Id: uipc_usrreq.c,v 1.22 1997/03/23 03:36:33 bde Exp $
35  */
36 
37 #include <sys/param.h>
38 #include <sys/queue.h>
39 #include <sys/systm.h>
40 #include <sys/kernel.h>
41 #include <sys/domain.h>
42 #include <sys/fcntl.h>
43 #include <sys/file.h>
44 #include <sys/filedesc.h>
45 #include <sys/mbuf.h>
46 #include <sys/namei.h>
47 #include <sys/proc.h>
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/stat.h>
52 #include <sys/sysctl.h>
53 #include <sys/un.h>
54 #include <sys/unpcb.h>
55 #include <sys/vnode.h>
56 
57 /*
58  * Unix communications domain.
59  *
60  * TODO:
61  *	SEQPACKET, RDM
62  *	rethink name space problems
63  *	need a proper out-of-band
64  */
65 static struct	sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
66 static ino_t	unp_ino;		/* prototype for fake inode numbers */
67 
68 static int     unp_attach __P((struct socket *));
69 static void    unp_detach __P((struct unpcb *));
70 static int     unp_bind __P((struct unpcb *,struct mbuf *, struct proc *));
71 static int     unp_connect __P((struct socket *,struct mbuf *, struct proc *));
72 static void    unp_disconnect __P((struct unpcb *));
73 static void    unp_shutdown __P((struct unpcb *));
74 static void    unp_drop __P((struct unpcb *, int));
75 static void    unp_gc __P((void));
76 static void    unp_scan __P((struct mbuf *, void (*)(struct file *)));
77 static void    unp_mark __P((struct file *));
78 static void    unp_discard __P((struct file *));
79 static int     unp_internalize __P((struct mbuf *, struct proc *));
80 
81 static int
82 uipc_abort(struct socket *so)
83 {
84 	struct unpcb *unp = sotounpcb(so);
85 
86 	if (unp == 0)
87 		return EINVAL;
88 	unp_drop(unp, ECONNABORTED);
89 	return 0;
90 }
91 
92 static int
93 uipc_accept(struct socket *so, struct mbuf *nam)
94 {
95 	struct unpcb *unp = sotounpcb(so);
96 
97 	if (unp == 0)
98 		return EINVAL;
99 
100 	/*
101 	 * Pass back name of connected socket,
102 	 * if it was bound and we are still connected
103 	 * (our peer may have closed already!).
104 	 */
105 	if (unp->unp_conn && unp->unp_conn->unp_addr) {
106 		nam->m_len = unp->unp_conn->unp_addr->m_len;
107 		bcopy(mtod(unp->unp_conn->unp_addr, caddr_t),
108 		      mtod(nam, caddr_t), (unsigned)nam->m_len);
109 	} else {
110 		nam->m_len = sizeof(sun_noname);
111 		*(mtod(nam, struct sockaddr *)) = sun_noname;
112 	}
113 	return 0;
114 }
115 
116 static int
117 uipc_attach(struct socket *so, int proto, struct proc *p)
118 {
119 	struct unpcb *unp = sotounpcb(so);
120 
121 	if (unp != 0)
122 		return EISCONN;
123 	return unp_attach(so);
124 }
125 
126 static int
127 uipc_bind(struct socket *so, struct mbuf *nam, struct proc *p)
128 {
129 	struct unpcb *unp = sotounpcb(so);
130 
131 	if (unp == 0)
132 		return EINVAL;
133 
134 	return unp_bind(unp, nam, p);
135 }
136 
137 static int
138 uipc_connect(struct socket *so, struct mbuf *nam, struct proc *p)
139 {
140 	struct unpcb *unp = sotounpcb(so);
141 
142 	if (unp == 0)
143 		return EINVAL;
144 	return unp_connect(so, nam, curproc);
145 }
146 
147 static int
148 uipc_connect2(struct socket *so1, struct socket *so2)
149 {
150 	struct unpcb *unp = sotounpcb(so1);
151 
152 	if (unp == 0)
153 		return EINVAL;
154 
155 	return unp_connect2(so1, so2);
156 }
157 
158 /* control is EOPNOTSUPP */
159 
160 static int
161 uipc_detach(struct socket *so)
162 {
163 	struct unpcb *unp = sotounpcb(so);
164 
165 	if (unp == 0)
166 		return EINVAL;
167 
168 	unp_detach(unp);
169 	return 0;
170 }
171 
172 static int
173 uipc_disconnect(struct socket *so)
174 {
175 	struct unpcb *unp = sotounpcb(so);
176 
177 	if (unp == 0)
178 		return EINVAL;
179 	unp_disconnect(unp);
180 	return 0;
181 }
182 
183 static int
184 uipc_listen(struct socket *so, struct proc *p)
185 {
186 	struct unpcb *unp = sotounpcb(so);
187 
188 	if (unp == 0 || unp->unp_vnode == 0)
189 		return EINVAL;
190 	return 0;
191 }
192 
193 static int
194 uipc_peeraddr(struct socket *so, struct mbuf *nam)
195 {
196 	struct unpcb *unp = sotounpcb(so);
197 
198 	if (unp == 0)
199 		return EINVAL;
200 	if (unp->unp_conn && unp->unp_conn->unp_addr) {
201 		nam->m_len = unp->unp_conn->unp_addr->m_len;
202 		bcopy(mtod(unp->unp_conn->unp_addr, caddr_t),
203 		      mtod(nam, caddr_t), (unsigned)nam->m_len);
204 	} else
205 		nam->m_len = 0;
206 	return 0;
207 }
208 
209 static int
210 uipc_rcvd(struct socket *so, int flags)
211 {
212 	struct unpcb *unp = sotounpcb(so);
213 	struct socket *so2;
214 
215 	if (unp == 0)
216 		return EINVAL;
217 	switch (so->so_type) {
218 	case SOCK_DGRAM:
219 		panic("uipc_rcvd DGRAM?");
220 		/*NOTREACHED*/
221 
222 	case SOCK_STREAM:
223 #define	rcv (&so->so_rcv)
224 #define snd (&so2->so_snd)
225 		if (unp->unp_conn == 0)
226 			break;
227 		so2 = unp->unp_conn->unp_socket;
228 		/*
229 		 * Adjust backpressure on sender
230 		 * and wakeup any waiting to write.
231 		 */
232 		snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt;
233 		unp->unp_mbcnt = rcv->sb_mbcnt;
234 		snd->sb_hiwat += unp->unp_cc - rcv->sb_cc;
235 		unp->unp_cc = rcv->sb_cc;
236 		sowwakeup(so2);
237 #undef snd
238 #undef rcv
239 		break;
240 
241 	default:
242 		panic("uipc_rcvd unknown socktype");
243 	}
244 	return 0;
245 }
246 
247 /* pru_rcvoob is EOPNOTSUPP */
248 
249 static int
250 uipc_send(struct socket *so, int flags, struct mbuf *m, struct mbuf *nam,
251 	  struct mbuf *control, struct proc *p)
252 {
253 	int error = 0;
254 	struct unpcb *unp = sotounpcb(so);
255 	struct socket *so2;
256 
257 	if (unp == 0) {
258 		error = EINVAL;
259 		goto release;
260 	}
261 	if (flags & PRUS_OOB) {
262 		error = EOPNOTSUPP;
263 		goto release;
264 	}
265 
266 	if (control && (error = unp_internalize(control, p)))
267 		goto release;
268 
269 	switch (so->so_type) {
270 	case SOCK_DGRAM:
271 	{
272 		struct sockaddr *from;
273 
274 		if (nam) {
275 			if (unp->unp_conn) {
276 				error = EISCONN;
277 				break;
278 			}
279 			error = unp_connect(so, nam, p);
280 			if (error)
281 				break;
282 		} else {
283 			if (unp->unp_conn == 0) {
284 				error = ENOTCONN;
285 				break;
286 			}
287 		}
288 		so2 = unp->unp_conn->unp_socket;
289 		if (unp->unp_addr)
290 			from = mtod(unp->unp_addr, struct sockaddr *);
291 		else
292 			from = &sun_noname;
293 		if (sbappendaddr(&so2->so_rcv, from, m, control)) {
294 			sorwakeup(so2);
295 			m = 0;
296 			control = 0;
297 		} else
298 			error = ENOBUFS;
299 		if (nam)
300 			unp_disconnect(unp);
301 		break;
302 	}
303 
304 	case SOCK_STREAM:
305 #define	rcv (&so2->so_rcv)
306 #define	snd (&so->so_snd)
307 		/* Connect if not connected yet. */
308 		/*
309 		 * Note: A better implementation would complain
310 		 * if not equal to the peer's address.
311 		 */
312 		if ((so->so_state & SS_ISCONNECTED) == 0) {
313 			if (nam) {
314 				error = unp_connect(so, nam, p);
315 				if (error)
316 					break;	/* XXX */
317 			} else {
318 				error = ENOTCONN;
319 				break;
320 			}
321 		}
322 
323 		if (so->so_state & SS_CANTSENDMORE) {
324 			error = EPIPE;
325 			break;
326 		}
327 		if (unp->unp_conn == 0)
328 			panic("uipc_send connected but no connection?");
329 		so2 = unp->unp_conn->unp_socket;
330 		/*
331 		 * Send to paired receive port, and then reduce
332 		 * send buffer hiwater marks to maintain backpressure.
333 		 * Wake up readers.
334 		 */
335 		if (control) {
336 			if (sbappendcontrol(rcv, m, control))
337 				control = 0;
338 		} else
339 			sbappend(rcv, m);
340 		snd->sb_mbmax -=
341 			rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt;
342 		unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt;
343 		snd->sb_hiwat -= rcv->sb_cc - unp->unp_conn->unp_cc;
344 		unp->unp_conn->unp_cc = rcv->sb_cc;
345 		sorwakeup(so2);
346 		m = 0;
347 #undef snd
348 #undef rcv
349 		break;
350 
351 	default:
352 		panic("uipc_send unknown socktype");
353 	}
354 
355 	/*
356 	 * SEND_EOF is equivalent to a SEND followed by
357 	 * a SHUTDOWN.
358 	 */
359 	if (flags & PRUS_EOF) {
360 		socantsendmore(so);
361 		unp_shutdown(unp);
362 	}
363 
364 release:
365 	if (control)
366 		m_freem(control);
367 	if (m)
368 		m_freem(m);
369 	return error;
370 }
371 
372 static int
373 uipc_sense(struct socket *so, struct stat *sb)
374 {
375 	struct unpcb *unp = sotounpcb(so);
376 	struct socket *so2;
377 
378 	if (unp == 0)
379 		return EINVAL;
380 	sb->st_blksize = so->so_snd.sb_hiwat;
381 	if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) {
382 		so2 = unp->unp_conn->unp_socket;
383 		sb->st_blksize += so2->so_rcv.sb_cc;
384 	}
385 	sb->st_dev = NODEV;
386 	if (unp->unp_ino == 0)
387 		unp->unp_ino = unp_ino++;
388 	sb->st_ino = unp->unp_ino;
389 	return (0);
390 }
391 
392 static int
393 uipc_shutdown(struct socket *so)
394 {
395 	struct unpcb *unp = sotounpcb(so);
396 
397 	if (unp == 0)
398 		return EINVAL;
399 	socantsendmore(so);
400 	unp_shutdown(unp);
401 	return 0;
402 }
403 
404 static int
405 uipc_sockaddr(struct socket *so, struct mbuf *nam)
406 {
407 	struct unpcb *unp = sotounpcb(so);
408 
409 	if (unp == 0)
410 		return EINVAL;
411 	if (unp->unp_addr) {
412 		nam->m_len = unp->unp_addr->m_len;
413 		bcopy(mtod(unp->unp_addr, caddr_t),
414 		      mtod(nam, caddr_t), (unsigned)nam->m_len);
415 	} else
416 		nam->m_len = 0;
417 	return 0;
418 }
419 
420 struct pr_usrreqs uipc_usrreqs = {
421 	uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect,
422 	uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect,
423 	uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp,
424 	uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr,
425 	sosend, soreceive, soselect
426 };
427 
428 /*
429  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
430  * for stream sockets, although the total for sender and receiver is
431  * actually only PIPSIZ.
432  * Datagram sockets really use the sendspace as the maximum datagram size,
433  * and don't really want to reserve the sendspace.  Their recvspace should
434  * be large enough for at least one max-size datagram plus address.
435  */
436 #ifndef PIPSIZ
437 #define	PIPSIZ	8192
438 #endif
439 static u_long	unpst_sendspace = PIPSIZ;
440 static u_long	unpst_recvspace = PIPSIZ;
441 static u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
442 static u_long	unpdg_recvspace = 4*1024;
443 
444 static int	unp_rights;			/* file descriptors in flight */
445 
446 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
447 	   &unpst_sendspace, 0, "");
448 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
449 	   &unpst_recvspace, 0, "");
450 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
451 	   &unpdg_sendspace, 0, "");
452 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
453 	   &unpdg_recvspace, 0, "");
454 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
455 
456 static int
457 unp_attach(so)
458 	struct socket *so;
459 {
460 	register struct mbuf *m;
461 	register struct unpcb *unp;
462 	int error;
463 
464 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
465 		switch (so->so_type) {
466 
467 		case SOCK_STREAM:
468 			error = soreserve(so, unpst_sendspace, unpst_recvspace);
469 			break;
470 
471 		case SOCK_DGRAM:
472 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
473 			break;
474 
475 		default:
476 			panic("unp_attach");
477 		}
478 		if (error)
479 			return (error);
480 	}
481 	m = m_getclr(M_DONTWAIT, MT_PCB);
482 	if (m == NULL)
483 		return (ENOBUFS);
484 	unp = mtod(m, struct unpcb *);
485 	so->so_pcb = (caddr_t)unp;
486 	unp->unp_socket = so;
487 	return (0);
488 }
489 
490 static void
491 unp_detach(unp)
492 	register struct unpcb *unp;
493 {
494 
495 	if (unp->unp_vnode) {
496 		unp->unp_vnode->v_socket = 0;
497 		vrele(unp->unp_vnode);
498 		unp->unp_vnode = 0;
499 	}
500 	if (unp->unp_conn)
501 		unp_disconnect(unp);
502 	while (unp->unp_refs)
503 		unp_drop(unp->unp_refs, ECONNRESET);
504 	soisdisconnected(unp->unp_socket);
505 	unp->unp_socket->so_pcb = 0;
506 	if (unp_rights) {
507 		/*
508 		 * Normally the receive buffer is flushed later,
509 		 * in sofree, but if our receive buffer holds references
510 		 * to descriptors that are now garbage, we will dispose
511 		 * of those descriptor references after the garbage collector
512 		 * gets them (resulting in a "panic: closef: count < 0").
513 		 */
514 		sorflush(unp->unp_socket);
515 		unp_gc();
516 	}
517 	m_freem(unp->unp_addr);
518 	(void) m_free(dtom(unp));
519 }
520 
521 static int
522 unp_bind(unp, nam, p)
523 	struct unpcb *unp;
524 	struct mbuf *nam;
525 	struct proc *p;
526 {
527 	struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *);
528 	register struct vnode *vp;
529 	struct vattr vattr;
530 	int error;
531 	struct nameidata nd;
532 
533 	NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT, UIO_SYSSPACE,
534 	    soun->sun_path, p);
535 	if (unp->unp_vnode != NULL)
536 		return (EINVAL);
537 	if (nam->m_len == MLEN) {
538 		if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0)
539 			return (EINVAL);
540 	} else
541 		*(mtod(nam, caddr_t) + nam->m_len) = 0;
542 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
543 	error = namei(&nd);
544 	if (error)
545 		return (error);
546 	vp = nd.ni_vp;
547 	if (vp != NULL) {
548 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
549 		if (nd.ni_dvp == vp)
550 			vrele(nd.ni_dvp);
551 		else
552 			vput(nd.ni_dvp);
553 		vrele(vp);
554 		return (EADDRINUSE);
555 	}
556 	VATTR_NULL(&vattr);
557 	vattr.va_type = VSOCK;
558 	vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask);
559 	VOP_LEASE(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE);
560 	if (error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr))
561 		return (error);
562 	vp = nd.ni_vp;
563 	vp->v_socket = unp->unp_socket;
564 	unp->unp_vnode = vp;
565 	unp->unp_addr = m_copy(nam, 0, (int)M_COPYALL);
566 	VOP_UNLOCK(vp, 0, p);
567 	return (0);
568 }
569 
570 static int
571 unp_connect(so, nam, p)
572 	struct socket *so;
573 	struct mbuf *nam;
574 	struct proc *p;
575 {
576 	register struct sockaddr_un *soun = mtod(nam, struct sockaddr_un *);
577 	register struct vnode *vp;
578 	register struct socket *so2, *so3;
579 	struct unpcb *unp2, *unp3;
580 	int error;
581 	struct nameidata nd;
582 
583 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, soun->sun_path, p);
584 	if (nam->m_data + nam->m_len == &nam->m_dat[MLEN]) {	/* XXX */
585 		if (*(mtod(nam, caddr_t) + nam->m_len - 1) != 0)
586 			return (EMSGSIZE);
587 	} else
588 		*(mtod(nam, caddr_t) + nam->m_len) = 0;
589 	error = namei(&nd);
590 	if (error)
591 		return (error);
592 	vp = nd.ni_vp;
593 	if (vp->v_type != VSOCK) {
594 		error = ENOTSOCK;
595 		goto bad;
596 	}
597 	error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p);
598 	if (error)
599 		goto bad;
600 	so2 = vp->v_socket;
601 	if (so2 == 0) {
602 		error = ECONNREFUSED;
603 		goto bad;
604 	}
605 	if (so->so_type != so2->so_type) {
606 		error = EPROTOTYPE;
607 		goto bad;
608 	}
609 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
610 		if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
611 		    (so3 = sonewconn(so2, 0)) == 0) {
612 			error = ECONNREFUSED;
613 			goto bad;
614 		}
615 		unp2 = sotounpcb(so2);
616 		unp3 = sotounpcb(so3);
617 		if (unp2->unp_addr)
618 			unp3->unp_addr =
619 				  m_copy(unp2->unp_addr, 0, (int)M_COPYALL);
620 		so2 = so3;
621 	}
622 	error = unp_connect2(so, so2);
623 bad:
624 	vput(vp);
625 	return (error);
626 }
627 
628 int
629 unp_connect2(so, so2)
630 	register struct socket *so;
631 	register struct socket *so2;
632 {
633 	register struct unpcb *unp = sotounpcb(so);
634 	register struct unpcb *unp2;
635 
636 	if (so2->so_type != so->so_type)
637 		return (EPROTOTYPE);
638 	unp2 = sotounpcb(so2);
639 	unp->unp_conn = unp2;
640 	switch (so->so_type) {
641 
642 	case SOCK_DGRAM:
643 		unp->unp_nextref = unp2->unp_refs;
644 		unp2->unp_refs = unp;
645 		soisconnected(so);
646 		break;
647 
648 	case SOCK_STREAM:
649 		unp2->unp_conn = unp;
650 		soisconnected(so);
651 		soisconnected(so2);
652 		break;
653 
654 	default:
655 		panic("unp_connect2");
656 	}
657 	return (0);
658 }
659 
660 static void
661 unp_disconnect(unp)
662 	struct unpcb *unp;
663 {
664 	register struct unpcb *unp2 = unp->unp_conn;
665 
666 	if (unp2 == 0)
667 		return;
668 	unp->unp_conn = 0;
669 	switch (unp->unp_socket->so_type) {
670 
671 	case SOCK_DGRAM:
672 		if (unp2->unp_refs == unp)
673 			unp2->unp_refs = unp->unp_nextref;
674 		else {
675 			unp2 = unp2->unp_refs;
676 			for (;;) {
677 				if (unp2 == 0)
678 					panic("unp_disconnect");
679 				if (unp2->unp_nextref == unp)
680 					break;
681 				unp2 = unp2->unp_nextref;
682 			}
683 			unp2->unp_nextref = unp->unp_nextref;
684 		}
685 		unp->unp_nextref = 0;
686 		unp->unp_socket->so_state &= ~SS_ISCONNECTED;
687 		break;
688 
689 	case SOCK_STREAM:
690 		soisdisconnected(unp->unp_socket);
691 		unp2->unp_conn = 0;
692 		soisdisconnected(unp2->unp_socket);
693 		break;
694 	}
695 }
696 
697 #ifdef notdef
698 void
699 unp_abort(unp)
700 	struct unpcb *unp;
701 {
702 
703 	unp_detach(unp);
704 }
705 #endif
706 
707 static void
708 unp_shutdown(unp)
709 	struct unpcb *unp;
710 {
711 	struct socket *so;
712 
713 	if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
714 	    (so = unp->unp_conn->unp_socket))
715 		socantrcvmore(so);
716 }
717 
718 static void
719 unp_drop(unp, errno)
720 	struct unpcb *unp;
721 	int errno;
722 {
723 	struct socket *so = unp->unp_socket;
724 
725 	so->so_error = errno;
726 	unp_disconnect(unp);
727 	if (so->so_head) {
728 		so->so_pcb = (caddr_t) 0;
729 		m_freem(unp->unp_addr);
730 		(void) m_free(dtom(unp));
731 		sofree(so);
732 	}
733 }
734 
735 #ifdef notdef
736 void
737 unp_drain()
738 {
739 
740 }
741 #endif
742 
743 int
744 unp_externalize(rights)
745 	struct mbuf *rights;
746 {
747 	struct proc *p = curproc;		/* XXX */
748 	register int i;
749 	register struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
750 	register struct file **rp = (struct file **)(cm + 1);
751 	register struct file *fp;
752 	int newfds = (cm->cmsg_len - sizeof(*cm)) / sizeof (int);
753 	int f;
754 
755 	/*
756 	 * if the new FD's will not fit, then we free them all
757 	 */
758 	if (!fdavail(p, newfds)) {
759 		for (i = 0; i < newfds; i++) {
760 			fp = *rp;
761 			unp_discard(fp);
762 			*rp++ = 0;
763 		}
764 		return (EMSGSIZE);
765 	}
766 	/*
767 	 * now change each pointer to an fd in the global table to
768 	 * an integer that is the index to the local fd table entry
769 	 * that we set up to point to the global one we are transferring.
770 	 * XXX this assumes a pointer and int are the same size...!
771 	 */
772 	for (i = 0; i < newfds; i++) {
773 		if (fdalloc(p, 0, &f))
774 			panic("unp_externalize");
775 		fp = *rp;
776 		p->p_fd->fd_ofiles[f] = fp;
777 		fp->f_msgcount--;
778 		unp_rights--;
779 		*(int *)rp++ = f;
780 	}
781 	return (0);
782 }
783 
784 #ifndef MIN
785 #define	MIN(a,b) (((a)<(b))?(a):(b))
786 #endif
787 
788 static int
789 unp_internalize(control, p)
790 	struct mbuf *control;
791 	struct proc *p;
792 {
793 	struct filedesc *fdp = p->p_fd;
794 	register struct cmsghdr *cm = mtod(control, struct cmsghdr *);
795 	register struct file **rp;
796 	register struct file *fp;
797 	register int i, fd;
798 	register struct cmsgcred *cmcred;
799 	int oldfds;
800 
801 	if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) ||
802 	    cm->cmsg_level != SOL_SOCKET || cm->cmsg_len != control->m_len)
803 		return (EINVAL);
804 
805 	/*
806 	 * Fill in credential information.
807 	 */
808 	if (cm->cmsg_type == SCM_CREDS) {
809 		cmcred = (struct cmsgcred *)(cm + 1);
810 		cmcred->cmcred_pid = p->p_pid;
811 		cmcred->cmcred_uid = p->p_cred->p_ruid;
812 		cmcred->cmcred_gid = p->p_cred->p_rgid;
813 		cmcred->cmcred_euid = p->p_ucred->cr_uid;
814 		cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups,
815 							CMGROUP_MAX);
816 		for (i = 0; i < cmcred->cmcred_ngroups; i++)
817 			cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i];
818 		return(0);
819 	}
820 
821 	oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int);
822 	/*
823 	 * check that all the FDs passed in refer to legal OPEN files
824 	 * If not, reject the entire operation.
825 	 */
826 	rp = (struct file **)(cm + 1);
827 	for (i = 0; i < oldfds; i++) {
828 		fd = *(int *)rp++;
829 		if ((unsigned)fd >= fdp->fd_nfiles ||
830 		    fdp->fd_ofiles[fd] == NULL)
831 			return (EBADF);
832 	}
833 	/*
834 	 * Now replace the integer FDs with pointers to
835 	 * the associated global file table entry..
836 	 * XXX this assumes a pointer and an int are the same size!
837 	 */
838 	rp = (struct file **)(cm + 1);
839 	for (i = 0; i < oldfds; i++) {
840 		fp = fdp->fd_ofiles[*(int *)rp];
841 		*rp++ = fp;
842 		fp->f_count++;
843 		fp->f_msgcount++;
844 		unp_rights++;
845 	}
846 	return (0);
847 }
848 
849 static int	unp_defer, unp_gcing;
850 
851 static void
852 unp_gc()
853 {
854 	register struct file *fp, *nextfp;
855 	register struct socket *so;
856 	struct file **extra_ref, **fpp;
857 	int nunref, i;
858 
859 	if (unp_gcing)
860 		return;
861 	unp_gcing = 1;
862 	unp_defer = 0;
863 	/*
864 	 * before going through all this, set all FDs to
865 	 * be NOT defered and NOT externally accessible
866 	 */
867 	for (fp = filehead.lh_first; fp != 0; fp = fp->f_list.le_next)
868 		fp->f_flag &= ~(FMARK|FDEFER);
869 	do {
870 		for (fp = filehead.lh_first; fp != 0; fp = fp->f_list.le_next) {
871 			/*
872 			 * If the file is not open, skip it
873 			 */
874 			if (fp->f_count == 0)
875 				continue;
876 			/*
877 			 * If we already marked it as 'defer'  in a
878 			 * previous pass, then try process it this time
879 			 * and un-mark it
880 			 */
881 			if (fp->f_flag & FDEFER) {
882 				fp->f_flag &= ~FDEFER;
883 				unp_defer--;
884 			} else {
885 				/*
886 				 * if it's not defered, then check if it's
887 				 * already marked.. if so skip it
888 				 */
889 				if (fp->f_flag & FMARK)
890 					continue;
891 				/*
892 				 * If all references are from messages
893 				 * in transit, then skip it. it's not
894 				 * externally accessible.
895 				 */
896 				if (fp->f_count == fp->f_msgcount)
897 					continue;
898 				/*
899 				 * If it got this far then it must be
900 				 * externally accessible.
901 				 */
902 				fp->f_flag |= FMARK;
903 			}
904 			/*
905 			 * either it was defered, or it is externally
906 			 * accessible and not already marked so.
907 			 * Now check if it is possibly one of OUR sockets.
908 			 */
909 			if (fp->f_type != DTYPE_SOCKET ||
910 			    (so = (struct socket *)fp->f_data) == 0)
911 				continue;
912 			if (so->so_proto->pr_domain != &localdomain ||
913 			    (so->so_proto->pr_flags&PR_RIGHTS) == 0)
914 				continue;
915 #ifdef notdef
916 			if (so->so_rcv.sb_flags & SB_LOCK) {
917 				/*
918 				 * This is problematical; it's not clear
919 				 * we need to wait for the sockbuf to be
920 				 * unlocked (on a uniprocessor, at least),
921 				 * and it's also not clear what to do
922 				 * if sbwait returns an error due to receipt
923 				 * of a signal.  If sbwait does return
924 				 * an error, we'll go into an infinite
925 				 * loop.  Delete all of this for now.
926 				 */
927 				(void) sbwait(&so->so_rcv);
928 				goto restart;
929 			}
930 #endif
931 			/*
932 			 * So, Ok, it's one of our sockets and it IS externally
933 			 * accessible (or was defered). Now we look
934 			 * to see if we hold any file descriptors in it's
935 			 * message buffers. Follow those links and mark them
936 			 * as accessible too.
937 			 */
938 			unp_scan(so->so_rcv.sb_mb, unp_mark);
939 		}
940 	} while (unp_defer);
941 	/*
942 	 * We grab an extra reference to each of the file table entries
943 	 * that are not otherwise accessible and then free the rights
944 	 * that are stored in messages on them.
945 	 *
946 	 * The bug in the orginal code is a little tricky, so I'll describe
947 	 * what's wrong with it here.
948 	 *
949 	 * It is incorrect to simply unp_discard each entry for f_msgcount
950 	 * times -- consider the case of sockets A and B that contain
951 	 * references to each other.  On a last close of some other socket,
952 	 * we trigger a gc since the number of outstanding rights (unp_rights)
953 	 * is non-zero.  If during the sweep phase the gc code un_discards,
954 	 * we end up doing a (full) closef on the descriptor.  A closef on A
955 	 * results in the following chain.  Closef calls soo_close, which
956 	 * calls soclose.   Soclose calls first (through the switch
957 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
958 	 * returns because the previous instance had set unp_gcing, and
959 	 * we return all the way back to soclose, which marks the socket
960 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
961 	 * to free up the rights that are queued in messages on the socket A,
962 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
963 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
964 	 * instance of unp_discard just calls closef on B.
965 	 *
966 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
967 	 * which results in another closef on A.  Unfortunately, A is already
968 	 * being closed, and the descriptor has already been marked with
969 	 * SS_NOFDREF, and soclose panics at this point.
970 	 *
971 	 * Here, we first take an extra reference to each inaccessible
972 	 * descriptor.  Then, we call sorflush ourself, since we know
973 	 * it is a Unix domain socket anyhow.  After we destroy all the
974 	 * rights carried in messages, we do a last closef to get rid
975 	 * of our extra reference.  This is the last close, and the
976 	 * unp_detach etc will shut down the socket.
977 	 *
978 	 * 91/09/19, bsy@cs.cmu.edu
979 	 */
980 	extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK);
981 	for (nunref = 0, fp = filehead.lh_first, fpp = extra_ref; fp != 0;
982 	    fp = nextfp) {
983 		nextfp = fp->f_list.le_next;
984 		/*
985 		 * If it's not open, skip it
986 		 */
987 		if (fp->f_count == 0)
988 			continue;
989 		/*
990 		 * If all refs are from msgs, and it's not marked accessible
991 		 * then it must be referenced from some unreachable cycle
992 		 * of (shut-down) FDs, so include it in our
993 		 * list of FDs to remove
994 		 */
995 		if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
996 			*fpp++ = fp;
997 			nunref++;
998 			fp->f_count++;
999 		}
1000 	}
1001 	/*
1002 	 * for each FD on our hit list, do the following two things
1003 	 */
1004 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
1005 		sorflush((struct socket *)(*fpp)->f_data);
1006 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
1007 		closef(*fpp, (struct proc *) NULL);
1008 	free((caddr_t)extra_ref, M_FILE);
1009 	unp_gcing = 0;
1010 }
1011 
1012 void
1013 unp_dispose(m)
1014 	struct mbuf *m;
1015 {
1016 
1017 	if (m)
1018 		unp_scan(m, unp_discard);
1019 }
1020 
1021 static void
1022 unp_scan(m0, op)
1023 	register struct mbuf *m0;
1024 	void (*op) __P((struct file *));
1025 {
1026 	register struct mbuf *m;
1027 	register struct file **rp;
1028 	register struct cmsghdr *cm;
1029 	register int i;
1030 	int qfds;
1031 
1032 	while (m0) {
1033 		for (m = m0; m; m = m->m_next)
1034 			if (m->m_type == MT_CONTROL &&
1035 			    m->m_len >= sizeof(*cm)) {
1036 				cm = mtod(m, struct cmsghdr *);
1037 				if (cm->cmsg_level != SOL_SOCKET ||
1038 				    cm->cmsg_type != SCM_RIGHTS)
1039 					continue;
1040 				qfds = (cm->cmsg_len - sizeof *cm)
1041 						/ sizeof (struct file *);
1042 				rp = (struct file **)(cm + 1);
1043 				for (i = 0; i < qfds; i++)
1044 					(*op)(*rp++);
1045 				break;		/* XXX, but saves time */
1046 			}
1047 		m0 = m0->m_act;
1048 	}
1049 }
1050 
1051 static void
1052 unp_mark(fp)
1053 	struct file *fp;
1054 {
1055 
1056 	if (fp->f_flag & FMARK)
1057 		return;
1058 	unp_defer++;
1059 	fp->f_flag |= (FMARK|FDEFER);
1060 }
1061 
1062 static void
1063 unp_discard(fp)
1064 	struct file *fp;
1065 {
1066 
1067 	fp->f_msgcount--;
1068 	unp_rights--;
1069 	(void) closef(fp, (struct proc *)NULL);
1070 }
1071