xref: /freebsd/sys/kern/uipc_usrreq.c (revision 6e8394b8baa7d5d9153ab90de6824bcd19b3b4e1)
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 1.44 1999/05/10 18:09:39 truckman Exp $
35  */
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.h>
40 #include <sys/domain.h>
41 #include <sys/fcntl.h>
42 #include <sys/malloc.h>		/* XXX must be before <sys/file.h> */
43 #include <sys/file.h>
44 #include <sys/filedesc.h>
45 #include <sys/lock.h>
46 #include <sys/mbuf.h>
47 #include <sys/namei.h>
48 #include <sys/proc.h>
49 #include <sys/protosw.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/stat.h>
53 #include <sys/sysctl.h>
54 #include <sys/un.h>
55 #include <sys/unpcb.h>
56 #include <sys/vnode.h>
57 
58 #include <vm/vm_zone.h>
59 
60 static	struct vm_zone *unp_zone;
61 static	unp_gen_t unp_gencnt;
62 static	u_int unp_count;
63 
64 static	struct unp_head unp_shead, unp_dhead;
65 
66 /*
67  * Unix communications domain.
68  *
69  * TODO:
70  *	SEQPACKET, RDM
71  *	rethink name space problems
72  *	need a proper out-of-band
73  *	lock pushdown
74  */
75 static struct	sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
76 static ino_t	unp_ino;		/* prototype for fake inode numbers */
77 
78 static int     unp_attach __P((struct socket *));
79 static void    unp_detach __P((struct unpcb *));
80 static int     unp_bind __P((struct unpcb *,struct sockaddr *, struct proc *));
81 static int     unp_connect __P((struct socket *,struct sockaddr *,
82 				struct proc *));
83 static void    unp_disconnect __P((struct unpcb *));
84 static void    unp_shutdown __P((struct unpcb *));
85 static void    unp_drop __P((struct unpcb *, int));
86 static void    unp_gc __P((void));
87 static void    unp_scan __P((struct mbuf *, void (*)(struct file *)));
88 static void    unp_mark __P((struct file *));
89 static void    unp_discard __P((struct file *));
90 static int     unp_internalize __P((struct mbuf *, struct proc *));
91 
92 static int
93 uipc_abort(struct socket *so)
94 {
95 	struct unpcb *unp = sotounpcb(so);
96 
97 	if (unp == 0)
98 		return EINVAL;
99 	unp_drop(unp, ECONNABORTED);
100 	return 0;
101 }
102 
103 static int
104 uipc_accept(struct socket *so, struct sockaddr **nam)
105 {
106 	struct unpcb *unp = sotounpcb(so);
107 
108 	if (unp == 0)
109 		return EINVAL;
110 
111 	/*
112 	 * Pass back name of connected socket,
113 	 * if it was bound and we are still connected
114 	 * (our peer may have closed already!).
115 	 */
116 	if (unp->unp_conn && unp->unp_conn->unp_addr) {
117 		*nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr,
118 				    1);
119 	} else {
120 		*nam = dup_sockaddr((struct sockaddr *)&sun_noname, 1);
121 	}
122 	return 0;
123 }
124 
125 static int
126 uipc_attach(struct socket *so, int proto, struct proc *p)
127 {
128 	struct unpcb *unp = sotounpcb(so);
129 
130 	if (unp != 0)
131 		return EISCONN;
132 	return unp_attach(so);
133 }
134 
135 static int
136 uipc_bind(struct socket *so, struct sockaddr *nam, struct proc *p)
137 {
138 	struct unpcb *unp = sotounpcb(so);
139 
140 	if (unp == 0)
141 		return EINVAL;
142 
143 	return unp_bind(unp, nam, p);
144 }
145 
146 static int
147 uipc_connect(struct socket *so, struct sockaddr *nam, struct proc *p)
148 {
149 	struct unpcb *unp = sotounpcb(so);
150 
151 	if (unp == 0)
152 		return EINVAL;
153 	return unp_connect(so, nam, curproc);
154 }
155 
156 static int
157 uipc_connect2(struct socket *so1, struct socket *so2)
158 {
159 	struct unpcb *unp = sotounpcb(so1);
160 
161 	if (unp == 0)
162 		return EINVAL;
163 
164 	return unp_connect2(so1, so2);
165 }
166 
167 /* control is EOPNOTSUPP */
168 
169 static int
170 uipc_detach(struct socket *so)
171 {
172 	struct unpcb *unp = sotounpcb(so);
173 
174 	if (unp == 0)
175 		return EINVAL;
176 
177 	unp_detach(unp);
178 	return 0;
179 }
180 
181 static int
182 uipc_disconnect(struct socket *so)
183 {
184 	struct unpcb *unp = sotounpcb(so);
185 
186 	if (unp == 0)
187 		return EINVAL;
188 	unp_disconnect(unp);
189 	return 0;
190 }
191 
192 static int
193 uipc_listen(struct socket *so, struct proc *p)
194 {
195 	struct unpcb *unp = sotounpcb(so);
196 
197 	if (unp == 0 || unp->unp_vnode == 0)
198 		return EINVAL;
199 	return 0;
200 }
201 
202 static int
203 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
204 {
205 	struct unpcb *unp = sotounpcb(so);
206 
207 	if (unp == 0)
208 		return EINVAL;
209 	if (unp->unp_conn && unp->unp_conn->unp_addr)
210 		*nam = dup_sockaddr((struct sockaddr *)unp->unp_conn->unp_addr,
211 				    1);
212 	return 0;
213 }
214 
215 static int
216 uipc_rcvd(struct socket *so, int flags)
217 {
218 	struct unpcb *unp = sotounpcb(so);
219 	struct socket *so2;
220 
221 	if (unp == 0)
222 		return EINVAL;
223 	switch (so->so_type) {
224 	case SOCK_DGRAM:
225 		panic("uipc_rcvd DGRAM?");
226 		/*NOTREACHED*/
227 
228 	case SOCK_STREAM:
229 #define	rcv (&so->so_rcv)
230 #define snd (&so2->so_snd)
231 		if (unp->unp_conn == 0)
232 			break;
233 		so2 = unp->unp_conn->unp_socket;
234 		/*
235 		 * Adjust backpressure on sender
236 		 * and wakeup any waiting to write.
237 		 */
238 		snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt;
239 		unp->unp_mbcnt = rcv->sb_mbcnt;
240 		snd->sb_hiwat += unp->unp_cc - rcv->sb_cc;
241 		unp->unp_cc = rcv->sb_cc;
242 		sowwakeup(so2);
243 #undef snd
244 #undef rcv
245 		break;
246 
247 	default:
248 		panic("uipc_rcvd unknown socktype");
249 	}
250 	return 0;
251 }
252 
253 /* pru_rcvoob is EOPNOTSUPP */
254 
255 static int
256 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
257 	  struct mbuf *control, struct proc *p)
258 {
259 	int error = 0;
260 	struct unpcb *unp = sotounpcb(so);
261 	struct socket *so2;
262 
263 	if (unp == 0) {
264 		error = EINVAL;
265 		goto release;
266 	}
267 	if (flags & PRUS_OOB) {
268 		error = EOPNOTSUPP;
269 		goto release;
270 	}
271 
272 	if (control && (error = unp_internalize(control, p)))
273 		goto release;
274 
275 	switch (so->so_type) {
276 	case SOCK_DGRAM:
277 	{
278 		struct sockaddr *from;
279 
280 		if (nam) {
281 			if (unp->unp_conn) {
282 				error = EISCONN;
283 				break;
284 			}
285 			error = unp_connect(so, nam, p);
286 			if (error)
287 				break;
288 		} else {
289 			if (unp->unp_conn == 0) {
290 				error = ENOTCONN;
291 				break;
292 			}
293 		}
294 		so2 = unp->unp_conn->unp_socket;
295 		if (unp->unp_addr)
296 			from = (struct sockaddr *)unp->unp_addr;
297 		else
298 			from = &sun_noname;
299 		if (sbappendaddr(&so2->so_rcv, from, m, control)) {
300 			sorwakeup(so2);
301 			m = 0;
302 			control = 0;
303 		} else
304 			error = ENOBUFS;
305 		if (nam)
306 			unp_disconnect(unp);
307 		break;
308 	}
309 
310 	case SOCK_STREAM:
311 #define	rcv (&so2->so_rcv)
312 #define	snd (&so->so_snd)
313 		/* Connect if not connected yet. */
314 		/*
315 		 * Note: A better implementation would complain
316 		 * if not equal to the peer's address.
317 		 */
318 		if ((so->so_state & SS_ISCONNECTED) == 0) {
319 			if (nam) {
320 				error = unp_connect(so, nam, p);
321 				if (error)
322 					break;	/* XXX */
323 			} else {
324 				error = ENOTCONN;
325 				break;
326 			}
327 		}
328 
329 		if (so->so_state & SS_CANTSENDMORE) {
330 			error = EPIPE;
331 			break;
332 		}
333 		if (unp->unp_conn == 0)
334 			panic("uipc_send connected but no connection?");
335 		so2 = unp->unp_conn->unp_socket;
336 		/*
337 		 * Send to paired receive port, and then reduce
338 		 * send buffer hiwater marks to maintain backpressure.
339 		 * Wake up readers.
340 		 */
341 		if (control) {
342 			if (sbappendcontrol(rcv, m, control))
343 				control = 0;
344 		} else
345 			sbappend(rcv, m);
346 		snd->sb_mbmax -=
347 			rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt;
348 		unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt;
349 		snd->sb_hiwat -= rcv->sb_cc - unp->unp_conn->unp_cc;
350 		unp->unp_conn->unp_cc = rcv->sb_cc;
351 		sorwakeup(so2);
352 		m = 0;
353 #undef snd
354 #undef rcv
355 		break;
356 
357 	default:
358 		panic("uipc_send unknown socktype");
359 	}
360 
361 	/*
362 	 * SEND_EOF is equivalent to a SEND followed by
363 	 * a SHUTDOWN.
364 	 */
365 	if (flags & PRUS_EOF) {
366 		socantsendmore(so);
367 		unp_shutdown(unp);
368 	}
369 
370 	if (control && error != 0)
371 		unp_dispose(control);
372 
373 release:
374 	if (control)
375 		m_freem(control);
376 	if (m)
377 		m_freem(m);
378 	return error;
379 }
380 
381 static int
382 uipc_sense(struct socket *so, struct stat *sb)
383 {
384 	struct unpcb *unp = sotounpcb(so);
385 	struct socket *so2;
386 
387 	if (unp == 0)
388 		return EINVAL;
389 	sb->st_blksize = so->so_snd.sb_hiwat;
390 	if (so->so_type == SOCK_STREAM && unp->unp_conn != 0) {
391 		so2 = unp->unp_conn->unp_socket;
392 		sb->st_blksize += so2->so_rcv.sb_cc;
393 	}
394 	sb->st_dev = NOUDEV;
395 	if (unp->unp_ino == 0)
396 		unp->unp_ino = unp_ino++;
397 	sb->st_ino = unp->unp_ino;
398 	return (0);
399 }
400 
401 static int
402 uipc_shutdown(struct socket *so)
403 {
404 	struct unpcb *unp = sotounpcb(so);
405 
406 	if (unp == 0)
407 		return EINVAL;
408 	socantsendmore(so);
409 	unp_shutdown(unp);
410 	return 0;
411 }
412 
413 static int
414 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
415 {
416 	struct unpcb *unp = sotounpcb(so);
417 
418 	if (unp == 0)
419 		return EINVAL;
420 	if (unp->unp_addr)
421 		*nam = dup_sockaddr((struct sockaddr *)unp->unp_addr, 1);
422 	return 0;
423 }
424 
425 struct pr_usrreqs uipc_usrreqs = {
426 	uipc_abort, uipc_accept, uipc_attach, uipc_bind, uipc_connect,
427 	uipc_connect2, pru_control_notsupp, uipc_detach, uipc_disconnect,
428 	uipc_listen, uipc_peeraddr, uipc_rcvd, pru_rcvoob_notsupp,
429 	uipc_send, uipc_sense, uipc_shutdown, uipc_sockaddr,
430 	sosend, soreceive, sopoll
431 };
432 
433 /*
434  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
435  * for stream sockets, although the total for sender and receiver is
436  * actually only PIPSIZ.
437  * Datagram sockets really use the sendspace as the maximum datagram size,
438  * and don't really want to reserve the sendspace.  Their recvspace should
439  * be large enough for at least one max-size datagram plus address.
440  */
441 #ifndef PIPSIZ
442 #define	PIPSIZ	8192
443 #endif
444 static u_long	unpst_sendspace = PIPSIZ;
445 static u_long	unpst_recvspace = PIPSIZ;
446 static u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
447 static u_long	unpdg_recvspace = 4*1024;
448 
449 static int	unp_rights;			/* file descriptors in flight */
450 
451 SYSCTL_DECL(_net_local_stream);
452 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
453 	   &unpst_sendspace, 0, "");
454 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
455 	   &unpst_recvspace, 0, "");
456 SYSCTL_DECL(_net_local_dgram);
457 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
458 	   &unpdg_sendspace, 0, "");
459 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
460 	   &unpdg_recvspace, 0, "");
461 SYSCTL_DECL(_net_local);
462 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
463 
464 static int
465 unp_attach(so)
466 	struct socket *so;
467 {
468 	register struct unpcb *unp;
469 	int error;
470 
471 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
472 		switch (so->so_type) {
473 
474 		case SOCK_STREAM:
475 			error = soreserve(so, unpst_sendspace, unpst_recvspace);
476 			break;
477 
478 		case SOCK_DGRAM:
479 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
480 			break;
481 
482 		default:
483 			panic("unp_attach");
484 		}
485 		if (error)
486 			return (error);
487 	}
488 	unp = zalloc(unp_zone);
489 	if (unp == NULL)
490 		return (ENOBUFS);
491 	bzero(unp, sizeof *unp);
492 	unp->unp_gencnt = ++unp_gencnt;
493 	unp_count++;
494 	LIST_INIT(&unp->unp_refs);
495 	unp->unp_socket = so;
496 	unp->unp_rvnode = curproc->p_fd->fd_rdir;
497 	LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
498 			 : &unp_shead, unp, unp_link);
499 	so->so_pcb = (caddr_t)unp;
500 	return (0);
501 }
502 
503 static void
504 unp_detach(unp)
505 	register struct unpcb *unp;
506 {
507 	LIST_REMOVE(unp, unp_link);
508 	unp->unp_gencnt = ++unp_gencnt;
509 	--unp_count;
510 	if (unp->unp_vnode) {
511 		unp->unp_vnode->v_socket = 0;
512 		vrele(unp->unp_vnode);
513 		unp->unp_vnode = 0;
514 	}
515 	if (unp->unp_conn)
516 		unp_disconnect(unp);
517 	while (unp->unp_refs.lh_first)
518 		unp_drop(unp->unp_refs.lh_first, ECONNRESET);
519 	soisdisconnected(unp->unp_socket);
520 	unp->unp_socket->so_pcb = 0;
521 	if (unp_rights) {
522 		/*
523 		 * Normally the receive buffer is flushed later,
524 		 * in sofree, but if our receive buffer holds references
525 		 * to descriptors that are now garbage, we will dispose
526 		 * of those descriptor references after the garbage collector
527 		 * gets them (resulting in a "panic: closef: count < 0").
528 		 */
529 		sorflush(unp->unp_socket);
530 		unp_gc();
531 	}
532 	if (unp->unp_addr)
533 		FREE(unp->unp_addr, M_SONAME);
534 	zfree(unp_zone, unp);
535 }
536 
537 static int
538 unp_bind(unp, nam, p)
539 	struct unpcb *unp;
540 	struct sockaddr *nam;
541 	struct proc *p;
542 {
543 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
544 	register struct vnode *vp;
545 	struct vattr vattr;
546 	int error, namelen;
547 	struct nameidata nd;
548 	char buf[SOCK_MAXADDRLEN];
549 
550 	if (unp->unp_vnode != NULL)
551 		return (EINVAL);
552 #define offsetof(s, e) ((char *)&((s *)0)->e - (char *)((s *)0))
553 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
554 	if (namelen <= 0)
555 		return EINVAL;
556 	strncpy(buf, soun->sun_path, namelen);
557 	buf[namelen] = 0;	/* null-terminate the string */
558 	NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT, UIO_SYSSPACE,
559 	    buf, p);
560 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
561 	error = namei(&nd);
562 	if (error)
563 		return (error);
564 	vp = nd.ni_vp;
565 	if (vp != NULL) {
566 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
567 		if (nd.ni_dvp == vp)
568 			vrele(nd.ni_dvp);
569 		else
570 			vput(nd.ni_dvp);
571 		vrele(vp);
572 		return (EADDRINUSE);
573 	}
574 	VATTR_NULL(&vattr);
575 	vattr.va_type = VSOCK;
576 	vattr.va_mode = (ACCESSPERMS & ~p->p_fd->fd_cmask);
577 	VOP_LEASE(nd.ni_dvp, p, p->p_ucred, LEASE_WRITE);
578 	error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
579 	vput(nd.ni_dvp);
580 	if (error)
581 		return (error);
582 	vp = nd.ni_vp;
583 	vp->v_socket = unp->unp_socket;
584 	unp->unp_vnode = vp;
585 	unp->unp_addr = (struct sockaddr_un *)dup_sockaddr(nam, 1);
586 	VOP_UNLOCK(vp, 0, p);
587 	return (0);
588 }
589 
590 static int
591 unp_connect(so, nam, p)
592 	struct socket *so;
593 	struct sockaddr *nam;
594 	struct proc *p;
595 {
596 	register struct sockaddr_un *soun = (struct sockaddr_un *)nam;
597 	register struct vnode *vp;
598 	register struct socket *so2, *so3;
599 	struct unpcb *unp2, *unp3;
600 	int error, len;
601 	struct nameidata nd;
602 	char buf[SOCK_MAXADDRLEN];
603 
604 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
605 	if (len <= 0)
606 		return EINVAL;
607 	strncpy(buf, soun->sun_path, len);
608 	buf[len] = 0;
609 
610 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, p);
611 	error = namei(&nd);
612 	if (error)
613 		return (error);
614 	vp = nd.ni_vp;
615 	if (vp->v_type != VSOCK) {
616 		error = ENOTSOCK;
617 		goto bad;
618 	}
619 	error = VOP_ACCESS(vp, VWRITE, p->p_ucred, p);
620 	if (error)
621 		goto bad;
622 	so2 = vp->v_socket;
623 	if (so2 == 0) {
624 		error = ECONNREFUSED;
625 		goto bad;
626 	}
627 	if (so->so_type != so2->so_type) {
628 		error = EPROTOTYPE;
629 		goto bad;
630 	}
631 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
632 		if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
633 		    (so3 = sonewconn(so2, 0)) == 0) {
634 			error = ECONNREFUSED;
635 			goto bad;
636 		}
637 		unp2 = sotounpcb(so2);
638 		unp3 = sotounpcb(so3);
639 		if (unp2->unp_addr)
640 			unp3->unp_addr = (struct sockaddr_un *)
641 				dup_sockaddr((struct sockaddr *)
642 					     unp2->unp_addr, 1);
643 		so2 = so3;
644 	}
645 	error = unp_connect2(so, so2);
646 bad:
647 	vput(vp);
648 	return (error);
649 }
650 
651 int
652 unp_connect2(so, so2)
653 	register struct socket *so;
654 	register struct socket *so2;
655 {
656 	register struct unpcb *unp = sotounpcb(so);
657 	register struct unpcb *unp2;
658 
659 	if (so2->so_type != so->so_type)
660 		return (EPROTOTYPE);
661 	unp2 = sotounpcb(so2);
662 	unp->unp_conn = unp2;
663 	switch (so->so_type) {
664 
665 	case SOCK_DGRAM:
666 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
667 		soisconnected(so);
668 		break;
669 
670 	case SOCK_STREAM:
671 		unp2->unp_conn = unp;
672 		soisconnected(so);
673 		soisconnected(so2);
674 		break;
675 
676 	default:
677 		panic("unp_connect2");
678 	}
679 	return (0);
680 }
681 
682 static void
683 unp_disconnect(unp)
684 	struct unpcb *unp;
685 {
686 	register struct unpcb *unp2 = unp->unp_conn;
687 
688 	if (unp2 == 0)
689 		return;
690 	unp->unp_conn = 0;
691 	switch (unp->unp_socket->so_type) {
692 
693 	case SOCK_DGRAM:
694 		LIST_REMOVE(unp, unp_reflink);
695 		unp->unp_socket->so_state &= ~SS_ISCONNECTED;
696 		break;
697 
698 	case SOCK_STREAM:
699 		soisdisconnected(unp->unp_socket);
700 		unp2->unp_conn = 0;
701 		soisdisconnected(unp2->unp_socket);
702 		break;
703 	}
704 }
705 
706 #ifdef notdef
707 void
708 unp_abort(unp)
709 	struct unpcb *unp;
710 {
711 
712 	unp_detach(unp);
713 }
714 #endif
715 
716 static int
717 prison_unpcb(struct proc *p, struct unpcb *unp)
718 {
719 	if (!p->p_prison)
720 		return (0);
721 	if (p->p_fd->fd_rdir == unp->unp_rvnode)
722 		return (0);
723 	return (1);
724 }
725 
726 static int
727 unp_pcblist SYSCTL_HANDLER_ARGS
728 {
729 	int error, i, n;
730 	struct unpcb *unp, **unp_list;
731 	unp_gen_t gencnt;
732 	struct xunpgen xug;
733 	struct unp_head *head;
734 
735 	head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
736 
737 	/*
738 	 * The process of preparing the PCB list is too time-consuming and
739 	 * resource-intensive to repeat twice on every request.
740 	 */
741 	if (req->oldptr == 0) {
742 		n = unp_count;
743 		req->oldidx = 2 * (sizeof xug)
744 			+ (n + n/8) * sizeof(struct xunpcb);
745 		return 0;
746 	}
747 
748 	if (req->newptr != 0)
749 		return EPERM;
750 
751 	/*
752 	 * OK, now we're committed to doing something.
753 	 */
754 	gencnt = unp_gencnt;
755 	n = unp_count;
756 
757 	xug.xug_len = sizeof xug;
758 	xug.xug_count = n;
759 	xug.xug_gen = gencnt;
760 	xug.xug_sogen = so_gencnt;
761 	error = SYSCTL_OUT(req, &xug, sizeof xug);
762 	if (error)
763 		return error;
764 
765 	unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
766 	if (unp_list == 0)
767 		return ENOMEM;
768 
769 	for (unp = head->lh_first, i = 0; unp && i < n;
770 	     unp = unp->unp_link.le_next) {
771 		if (unp->unp_gencnt <= gencnt && !prison_unpcb(req->p, unp))
772 			unp_list[i++] = unp;
773 	}
774 	n = i;			/* in case we lost some during malloc */
775 
776 	error = 0;
777 	for (i = 0; i < n; i++) {
778 		unp = unp_list[i];
779 		if (unp->unp_gencnt <= gencnt) {
780 			struct xunpcb xu;
781 			xu.xu_len = sizeof xu;
782 			xu.xu_unpp = unp;
783 			/*
784 			 * XXX - need more locking here to protect against
785 			 * connect/disconnect races for SMP.
786 			 */
787 			if (unp->unp_addr)
788 				bcopy(unp->unp_addr, &xu.xu_addr,
789 				      unp->unp_addr->sun_len);
790 			if (unp->unp_conn && unp->unp_conn->unp_addr)
791 				bcopy(unp->unp_conn->unp_addr,
792 				      &xu.xu_caddr,
793 				      unp->unp_conn->unp_addr->sun_len);
794 			bcopy(unp, &xu.xu_unp, sizeof *unp);
795 			sotoxsocket(unp->unp_socket, &xu.xu_socket);
796 			error = SYSCTL_OUT(req, &xu, sizeof xu);
797 		}
798 	}
799 	if (!error) {
800 		/*
801 		 * Give the user an updated idea of our state.
802 		 * If the generation differs from what we told
803 		 * her before, she knows that something happened
804 		 * while we were processing this request, and it
805 		 * might be necessary to retry.
806 		 */
807 		xug.xug_gen = unp_gencnt;
808 		xug.xug_sogen = so_gencnt;
809 		xug.xug_count = unp_count;
810 		error = SYSCTL_OUT(req, &xug, sizeof xug);
811 	}
812 	free(unp_list, M_TEMP);
813 	return error;
814 }
815 
816 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
817 	    (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
818 	    "List of active local datagram sockets");
819 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
820 	    (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
821 	    "List of active local stream sockets");
822 
823 static void
824 unp_shutdown(unp)
825 	struct unpcb *unp;
826 {
827 	struct socket *so;
828 
829 	if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
830 	    (so = unp->unp_conn->unp_socket))
831 		socantrcvmore(so);
832 }
833 
834 static void
835 unp_drop(unp, errno)
836 	struct unpcb *unp;
837 	int errno;
838 {
839 	struct socket *so = unp->unp_socket;
840 
841 	so->so_error = errno;
842 	unp_disconnect(unp);
843 	if (so->so_head) {
844 		LIST_REMOVE(unp, unp_link);
845 		unp->unp_gencnt = ++unp_gencnt;
846 		unp_count--;
847 		so->so_pcb = (caddr_t) 0;
848 		if (unp->unp_addr)
849 			FREE(unp->unp_addr, M_SONAME);
850 		zfree(unp_zone, unp);
851 		sofree(so);
852 	}
853 }
854 
855 #ifdef notdef
856 void
857 unp_drain()
858 {
859 
860 }
861 #endif
862 
863 int
864 unp_externalize(rights)
865 	struct mbuf *rights;
866 {
867 	struct proc *p = curproc;		/* XXX */
868 	register int i;
869 	register struct cmsghdr *cm = mtod(rights, struct cmsghdr *);
870 	register struct file **rp = (struct file **)(cm + 1);
871 	register struct file *fp;
872 	int newfds = (cm->cmsg_len - sizeof(*cm)) / sizeof (int);
873 	int f;
874 
875 	/*
876 	 * if the new FD's will not fit, then we free them all
877 	 */
878 	if (!fdavail(p, newfds)) {
879 		for (i = 0; i < newfds; i++) {
880 			fp = *rp;
881 			unp_discard(fp);
882 			*rp++ = 0;
883 		}
884 		return (EMSGSIZE);
885 	}
886 	/*
887 	 * now change each pointer to an fd in the global table to
888 	 * an integer that is the index to the local fd table entry
889 	 * that we set up to point to the global one we are transferring.
890 	 * XXX this assumes a pointer and int are the same size...!
891 	 */
892 	for (i = 0; i < newfds; i++) {
893 		if (fdalloc(p, 0, &f))
894 			panic("unp_externalize");
895 		fp = *rp;
896 		p->p_fd->fd_ofiles[f] = fp;
897 		fp->f_msgcount--;
898 		unp_rights--;
899 		*(int *)rp++ = f;
900 	}
901 	return (0);
902 }
903 
904 void
905 unp_init(void)
906 {
907 	unp_zone = zinit("unpcb", sizeof(struct unpcb), nmbclusters, 0, 0);
908 	if (unp_zone == 0)
909 		panic("unp_init");
910 	LIST_INIT(&unp_dhead);
911 	LIST_INIT(&unp_shead);
912 }
913 
914 #ifndef MIN
915 #define	MIN(a,b) (((a)<(b))?(a):(b))
916 #endif
917 
918 static int
919 unp_internalize(control, p)
920 	struct mbuf *control;
921 	struct proc *p;
922 {
923 	struct filedesc *fdp = p->p_fd;
924 	register struct cmsghdr *cm = mtod(control, struct cmsghdr *);
925 	register struct file **rp;
926 	register struct file *fp;
927 	register int i, fd;
928 	register struct cmsgcred *cmcred;
929 	int oldfds;
930 
931 	if ((cm->cmsg_type != SCM_RIGHTS && cm->cmsg_type != SCM_CREDS) ||
932 	    cm->cmsg_level != SOL_SOCKET || cm->cmsg_len != control->m_len)
933 		return (EINVAL);
934 
935 	/*
936 	 * Fill in credential information.
937 	 */
938 	if (cm->cmsg_type == SCM_CREDS) {
939 		cmcred = (struct cmsgcred *)(cm + 1);
940 		cmcred->cmcred_pid = p->p_pid;
941 		cmcred->cmcred_uid = p->p_cred->p_ruid;
942 		cmcred->cmcred_gid = p->p_cred->p_rgid;
943 		cmcred->cmcred_euid = p->p_ucred->cr_uid;
944 		cmcred->cmcred_ngroups = MIN(p->p_ucred->cr_ngroups,
945 							CMGROUP_MAX);
946 		for (i = 0; i < cmcred->cmcred_ngroups; i++)
947 			cmcred->cmcred_groups[i] = p->p_ucred->cr_groups[i];
948 		return(0);
949 	}
950 
951 	oldfds = (cm->cmsg_len - sizeof (*cm)) / sizeof (int);
952 	/*
953 	 * check that all the FDs passed in refer to legal OPEN files
954 	 * If not, reject the entire operation.
955 	 */
956 	rp = (struct file **)(cm + 1);
957 	for (i = 0; i < oldfds; i++) {
958 		fd = *(int *)rp++;
959 		if ((unsigned)fd >= fdp->fd_nfiles ||
960 		    fdp->fd_ofiles[fd] == NULL)
961 			return (EBADF);
962 	}
963 	/*
964 	 * Now replace the integer FDs with pointers to
965 	 * the associated global file table entry..
966 	 * XXX this assumes a pointer and an int are the same size!
967 	 */
968 	rp = (struct file **)(cm + 1);
969 	for (i = 0; i < oldfds; i++) {
970 		fp = fdp->fd_ofiles[*(int *)rp];
971 		*rp++ = fp;
972 		fp->f_count++;
973 		fp->f_msgcount++;
974 		unp_rights++;
975 	}
976 	return (0);
977 }
978 
979 static int	unp_defer, unp_gcing;
980 
981 static void
982 unp_gc()
983 {
984 	register struct file *fp, *nextfp;
985 	register struct socket *so;
986 	struct file **extra_ref, **fpp;
987 	int nunref, i;
988 
989 	if (unp_gcing)
990 		return;
991 	unp_gcing = 1;
992 	unp_defer = 0;
993 	/*
994 	 * before going through all this, set all FDs to
995 	 * be NOT defered and NOT externally accessible
996 	 */
997 	for (fp = filehead.lh_first; fp != 0; fp = fp->f_list.le_next)
998 		fp->f_flag &= ~(FMARK|FDEFER);
999 	do {
1000 		for (fp = filehead.lh_first; fp != 0; fp = fp->f_list.le_next) {
1001 			/*
1002 			 * If the file is not open, skip it
1003 			 */
1004 			if (fp->f_count == 0)
1005 				continue;
1006 			/*
1007 			 * If we already marked it as 'defer'  in a
1008 			 * previous pass, then try process it this time
1009 			 * and un-mark it
1010 			 */
1011 			if (fp->f_flag & FDEFER) {
1012 				fp->f_flag &= ~FDEFER;
1013 				unp_defer--;
1014 			} else {
1015 				/*
1016 				 * if it's not defered, then check if it's
1017 				 * already marked.. if so skip it
1018 				 */
1019 				if (fp->f_flag & FMARK)
1020 					continue;
1021 				/*
1022 				 * If all references are from messages
1023 				 * in transit, then skip it. it's not
1024 				 * externally accessible.
1025 				 */
1026 				if (fp->f_count == fp->f_msgcount)
1027 					continue;
1028 				/*
1029 				 * If it got this far then it must be
1030 				 * externally accessible.
1031 				 */
1032 				fp->f_flag |= FMARK;
1033 			}
1034 			/*
1035 			 * either it was defered, or it is externally
1036 			 * accessible and not already marked so.
1037 			 * Now check if it is possibly one of OUR sockets.
1038 			 */
1039 			if (fp->f_type != DTYPE_SOCKET ||
1040 			    (so = (struct socket *)fp->f_data) == 0)
1041 				continue;
1042 			if (so->so_proto->pr_domain != &localdomain ||
1043 			    (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1044 				continue;
1045 #ifdef notdef
1046 			if (so->so_rcv.sb_flags & SB_LOCK) {
1047 				/*
1048 				 * This is problematical; it's not clear
1049 				 * we need to wait for the sockbuf to be
1050 				 * unlocked (on a uniprocessor, at least),
1051 				 * and it's also not clear what to do
1052 				 * if sbwait returns an error due to receipt
1053 				 * of a signal.  If sbwait does return
1054 				 * an error, we'll go into an infinite
1055 				 * loop.  Delete all of this for now.
1056 				 */
1057 				(void) sbwait(&so->so_rcv);
1058 				goto restart;
1059 			}
1060 #endif
1061 			/*
1062 			 * So, Ok, it's one of our sockets and it IS externally
1063 			 * accessible (or was defered). Now we look
1064 			 * to see if we hold any file descriptors in its
1065 			 * message buffers. Follow those links and mark them
1066 			 * as accessible too.
1067 			 */
1068 			unp_scan(so->so_rcv.sb_mb, unp_mark);
1069 		}
1070 	} while (unp_defer);
1071 	/*
1072 	 * We grab an extra reference to each of the file table entries
1073 	 * that are not otherwise accessible and then free the rights
1074 	 * that are stored in messages on them.
1075 	 *
1076 	 * The bug in the orginal code is a little tricky, so I'll describe
1077 	 * what's wrong with it here.
1078 	 *
1079 	 * It is incorrect to simply unp_discard each entry for f_msgcount
1080 	 * times -- consider the case of sockets A and B that contain
1081 	 * references to each other.  On a last close of some other socket,
1082 	 * we trigger a gc since the number of outstanding rights (unp_rights)
1083 	 * is non-zero.  If during the sweep phase the gc code un_discards,
1084 	 * we end up doing a (full) closef on the descriptor.  A closef on A
1085 	 * results in the following chain.  Closef calls soo_close, which
1086 	 * calls soclose.   Soclose calls first (through the switch
1087 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
1088 	 * returns because the previous instance had set unp_gcing, and
1089 	 * we return all the way back to soclose, which marks the socket
1090 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
1091 	 * to free up the rights that are queued in messages on the socket A,
1092 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
1093 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
1094 	 * instance of unp_discard just calls closef on B.
1095 	 *
1096 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1097 	 * which results in another closef on A.  Unfortunately, A is already
1098 	 * being closed, and the descriptor has already been marked with
1099 	 * SS_NOFDREF, and soclose panics at this point.
1100 	 *
1101 	 * Here, we first take an extra reference to each inaccessible
1102 	 * descriptor.  Then, we call sorflush ourself, since we know
1103 	 * it is a Unix domain socket anyhow.  After we destroy all the
1104 	 * rights carried in messages, we do a last closef to get rid
1105 	 * of our extra reference.  This is the last close, and the
1106 	 * unp_detach etc will shut down the socket.
1107 	 *
1108 	 * 91/09/19, bsy@cs.cmu.edu
1109 	 */
1110 	extra_ref = malloc(nfiles * sizeof(struct file *), M_FILE, M_WAITOK);
1111 	for (nunref = 0, fp = filehead.lh_first, fpp = extra_ref; fp != 0;
1112 	    fp = nextfp) {
1113 		nextfp = fp->f_list.le_next;
1114 		/*
1115 		 * If it's not open, skip it
1116 		 */
1117 		if (fp->f_count == 0)
1118 			continue;
1119 		/*
1120 		 * If all refs are from msgs, and it's not marked accessible
1121 		 * then it must be referenced from some unreachable cycle
1122 		 * of (shut-down) FDs, so include it in our
1123 		 * list of FDs to remove
1124 		 */
1125 		if (fp->f_count == fp->f_msgcount && !(fp->f_flag & FMARK)) {
1126 			*fpp++ = fp;
1127 			nunref++;
1128 			fp->f_count++;
1129 		}
1130 	}
1131 	/*
1132 	 * for each FD on our hit list, do the following two things
1133 	 */
1134 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1135 		struct file *tfp = *fpp;
1136 		if (tfp->f_type == DTYPE_SOCKET && tfp->f_data != NULL)
1137 			sorflush((struct socket *)(tfp->f_data));
1138 	}
1139 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
1140 		closef(*fpp, (struct proc *) NULL);
1141 	free((caddr_t)extra_ref, M_FILE);
1142 	unp_gcing = 0;
1143 }
1144 
1145 void
1146 unp_dispose(m)
1147 	struct mbuf *m;
1148 {
1149 
1150 	if (m)
1151 		unp_scan(m, unp_discard);
1152 }
1153 
1154 static void
1155 unp_scan(m0, op)
1156 	register struct mbuf *m0;
1157 	void (*op) __P((struct file *));
1158 {
1159 	register struct mbuf *m;
1160 	register struct file **rp;
1161 	register struct cmsghdr *cm;
1162 	register int i;
1163 	int qfds;
1164 
1165 	while (m0) {
1166 		for (m = m0; m; m = m->m_next)
1167 			if (m->m_type == MT_CONTROL &&
1168 			    m->m_len >= sizeof(*cm)) {
1169 				cm = mtod(m, struct cmsghdr *);
1170 				if (cm->cmsg_level != SOL_SOCKET ||
1171 				    cm->cmsg_type != SCM_RIGHTS)
1172 					continue;
1173 				qfds = (cm->cmsg_len - sizeof *cm)
1174 						/ sizeof (struct file *);
1175 				rp = (struct file **)(cm + 1);
1176 				for (i = 0; i < qfds; i++)
1177 					(*op)(*rp++);
1178 				break;		/* XXX, but saves time */
1179 			}
1180 		m0 = m0->m_act;
1181 	}
1182 }
1183 
1184 static void
1185 unp_mark(fp)
1186 	struct file *fp;
1187 {
1188 
1189 	if (fp->f_flag & FMARK)
1190 		return;
1191 	unp_defer++;
1192 	fp->f_flag |= (FMARK|FDEFER);
1193 }
1194 
1195 static void
1196 unp_discard(fp)
1197 	struct file *fp;
1198 {
1199 
1200 	fp->f_msgcount--;
1201 	unp_rights--;
1202 	(void) closef(fp, (struct proc *)NULL);
1203 }
1204