xref: /freebsd/sys/kern/uipc_usrreq.c (revision ceaec73d406831b1251babb61675df0a1aa54a31)
1 /*-
2  * Copyright 2004-2005 Robert N. M. Watson
3  * Copyright (c) 1982, 1986, 1989, 1991, 1993
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  * 1. Redistributions of source code must retain the above copyright
10  *    notice, this list of conditions and the following disclaimer.
11  * 2. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  * 4. Neither the name of the University nor the names of its contributors
15  *    may be used to endorse or promote products derived from this software
16  *    without specific prior written permission.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
19  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
22  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
28  * SUCH DAMAGE.
29  *
30  *	From: @(#)uipc_usrreq.c	8.3 (Berkeley) 1/4/94
31  */
32 
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
35 
36 #include "opt_mac.h"
37 
38 #include <sys/param.h>
39 #include <sys/domain.h>
40 #include <sys/fcntl.h>
41 #include <sys/malloc.h>		/* XXX must be before <sys/file.h> */
42 #include <sys/file.h>
43 #include <sys/filedesc.h>
44 #include <sys/jail.h>
45 #include <sys/kernel.h>
46 #include <sys/lock.h>
47 #include <sys/mac.h>
48 #include <sys/mbuf.h>
49 #include <sys/mutex.h>
50 #include <sys/namei.h>
51 #include <sys/proc.h>
52 #include <sys/protosw.h>
53 #include <sys/resourcevar.h>
54 #include <sys/socket.h>
55 #include <sys/socketvar.h>
56 #include <sys/signalvar.h>
57 #include <sys/stat.h>
58 #include <sys/sx.h>
59 #include <sys/sysctl.h>
60 #include <sys/systm.h>
61 #include <sys/un.h>
62 #include <sys/unpcb.h>
63 #include <sys/vnode.h>
64 
65 #include <vm/uma.h>
66 
67 static uma_zone_t unp_zone;
68 static	unp_gen_t unp_gencnt;
69 static	u_int unp_count;
70 
71 static	struct unp_head unp_shead, unp_dhead;
72 
73 /*
74  * Unix communications domain.
75  *
76  * TODO:
77  *	SEQPACKET, RDM
78  *	rethink name space problems
79  *	need a proper out-of-band
80  *	lock pushdown
81  */
82 static const struct	sockaddr sun_noname = { sizeof(sun_noname), AF_LOCAL };
83 static ino_t	unp_ino;		/* prototype for fake inode numbers */
84 struct mbuf *unp_addsockcred(struct thread *, struct mbuf *);
85 
86 /*
87  * Currently, UNIX domain sockets are protected by a single subsystem lock,
88  * which covers global data structures and variables, the contents of each
89  * per-socket unpcb structure, and the so_pcb field in sockets attached to
90  * the UNIX domain.  This provides for a moderate degree of paralellism, as
91  * receive operations on UNIX domain sockets do not need to acquire the
92  * subsystem lock.  Finer grained locking to permit send() without acquiring
93  * a global lock would be a logical next step.
94  *
95  * The UNIX domain socket lock preceds all socket layer locks, including the
96  * socket lock and socket buffer lock, permitting UNIX domain socket code to
97  * call into socket support routines without releasing its locks.
98  *
99  * Some caution is required in areas where the UNIX domain socket code enters
100  * VFS in order to create or find rendezvous points.  This results in
101  * dropping of the UNIX domain socket subsystem lock, acquisition of the
102  * Giant lock, and potential sleeping.  This increases the chances of races,
103  * and exposes weaknesses in the socket->protocol API by offering poor
104  * failure modes.
105  */
106 static struct mtx unp_mtx;
107 #define	UNP_LOCK_INIT() \
108 	mtx_init(&unp_mtx, "unp", NULL, MTX_DEF)
109 #define	UNP_LOCK()		mtx_lock(&unp_mtx)
110 #define	UNP_UNLOCK()		mtx_unlock(&unp_mtx)
111 #define	UNP_LOCK_ASSERT()	mtx_assert(&unp_mtx, MA_OWNED)
112 #define	UNP_UNLOCK_ASSERT()	mtx_assert(&unp_mtx, MA_NOTOWNED)
113 
114 static int     unp_attach(struct socket *);
115 static void    unp_detach(struct unpcb *);
116 static int     unp_bind(struct unpcb *,struct sockaddr *, struct thread *);
117 static int     unp_connect(struct socket *,struct sockaddr *, struct thread *);
118 static int     unp_connect2(struct socket *so, struct socket *so2, int);
119 static void    unp_disconnect(struct unpcb *);
120 static void    unp_shutdown(struct unpcb *);
121 static void    unp_drop(struct unpcb *, int);
122 static void    unp_gc(void);
123 static void    unp_scan(struct mbuf *, void (*)(struct file *));
124 static void    unp_mark(struct file *);
125 static void    unp_discard(struct file *);
126 static void    unp_freerights(struct file **, int);
127 static int     unp_internalize(struct mbuf **, struct thread *);
128 static int     unp_listen(struct socket *, struct unpcb *, struct thread *);
129 
130 static int
131 uipc_abort(struct socket *so)
132 {
133 	struct unpcb *unp;
134 
135 	UNP_LOCK();
136 	unp = sotounpcb(so);
137 	if (unp == NULL) {
138 		UNP_UNLOCK();
139 		return (EINVAL);
140 	}
141 	unp_drop(unp, ECONNABORTED);
142 	unp_detach(unp);
143 	UNP_UNLOCK_ASSERT();
144 	ACCEPT_LOCK();
145 	SOCK_LOCK(so);
146 	sotryfree(so);
147 	return (0);
148 }
149 
150 static int
151 uipc_accept(struct socket *so, struct sockaddr **nam)
152 {
153 	struct unpcb *unp;
154 	const struct sockaddr *sa;
155 
156 	/*
157 	 * Pass back name of connected socket,
158 	 * if it was bound and we are still connected
159 	 * (our peer may have closed already!).
160 	 */
161 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
162 	UNP_LOCK();
163 	unp = sotounpcb(so);
164 	if (unp == NULL) {
165 		UNP_UNLOCK();
166 		free(*nam, M_SONAME);
167 		*nam = NULL;
168 		return (EINVAL);
169 	}
170 	if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL)
171 		sa = (struct sockaddr *) unp->unp_conn->unp_addr;
172 	else
173 		sa = &sun_noname;
174 	bcopy(sa, *nam, sa->sa_len);
175 	UNP_UNLOCK();
176 	return (0);
177 }
178 
179 static int
180 uipc_attach(struct socket *so, int proto, struct thread *td)
181 {
182 	struct unpcb *unp = sotounpcb(so);
183 
184 	if (unp != NULL)
185 		return (EISCONN);
186 	return (unp_attach(so));
187 }
188 
189 static int
190 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
191 {
192 	struct unpcb *unp;
193 	int error;
194 
195 	UNP_LOCK();
196 	unp = sotounpcb(so);
197 	if (unp == NULL) {
198 		UNP_UNLOCK();
199 		return (EINVAL);
200 	}
201 	error = unp_bind(unp, nam, td);
202 	UNP_UNLOCK();
203 	return (error);
204 }
205 
206 static int
207 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
208 {
209 	struct unpcb *unp;
210 	int error;
211 
212 	KASSERT(td == curthread, ("uipc_connect: td != curthread"));
213 
214 	UNP_LOCK();
215 	unp = sotounpcb(so);
216 	if (unp == NULL) {
217 		UNP_UNLOCK();
218 		return (EINVAL);
219 	}
220 	error = unp_connect(so, nam, td);
221 	UNP_UNLOCK();
222 	return (error);
223 }
224 
225 int
226 uipc_connect2(struct socket *so1, struct socket *so2)
227 {
228 	struct unpcb *unp;
229 	int error;
230 
231 	UNP_LOCK();
232 	unp = sotounpcb(so1);
233 	if (unp == NULL) {
234 		UNP_UNLOCK();
235 		return (EINVAL);
236 	}
237 	error = unp_connect2(so1, so2, PRU_CONNECT2);
238 	UNP_UNLOCK();
239 	return (error);
240 }
241 
242 /* control is EOPNOTSUPP */
243 
244 static int
245 uipc_detach(struct socket *so)
246 {
247 	struct unpcb *unp;
248 
249 	UNP_LOCK();
250 	unp = sotounpcb(so);
251 	if (unp == NULL) {
252 		UNP_UNLOCK();
253 		return (EINVAL);
254 	}
255 	unp_detach(unp);
256 	UNP_UNLOCK_ASSERT();
257 	return (0);
258 }
259 
260 static int
261 uipc_disconnect(struct socket *so)
262 {
263 	struct unpcb *unp;
264 
265 	UNP_LOCK();
266 	unp = sotounpcb(so);
267 	if (unp == NULL) {
268 		UNP_UNLOCK();
269 		return (EINVAL);
270 	}
271 	unp_disconnect(unp);
272 	UNP_UNLOCK();
273 	return (0);
274 }
275 
276 static int
277 uipc_listen(struct socket *so, struct thread *td)
278 {
279 	struct unpcb *unp;
280 	int error;
281 
282 	UNP_LOCK();
283 	unp = sotounpcb(so);
284 	if (unp == NULL || unp->unp_vnode == NULL) {
285 		UNP_UNLOCK();
286 		return (EINVAL);
287 	}
288 	error = unp_listen(so, unp, td);
289 	UNP_UNLOCK();
290 	return (error);
291 }
292 
293 static int
294 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
295 {
296 	struct unpcb *unp;
297 	const struct sockaddr *sa;
298 
299 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
300 	UNP_LOCK();
301 	unp = sotounpcb(so);
302 	if (unp == NULL) {
303 		UNP_UNLOCK();
304 		free(*nam, M_SONAME);
305 		*nam = NULL;
306 		return (EINVAL);
307 	}
308 	if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL)
309 		sa = (struct sockaddr *) unp->unp_conn->unp_addr;
310 	else {
311 		/*
312 		 * XXX: It seems that this test always fails even when
313 		 * connection is established.  So, this else clause is
314 		 * added as workaround to return PF_LOCAL sockaddr.
315 		 */
316 		sa = &sun_noname;
317 	}
318 	bcopy(sa, *nam, sa->sa_len);
319 	UNP_UNLOCK();
320 	return (0);
321 }
322 
323 static int
324 uipc_rcvd(struct socket *so, int flags)
325 {
326 	struct unpcb *unp;
327 	struct socket *so2;
328 	u_long newhiwat;
329 
330 	UNP_LOCK();
331 	unp = sotounpcb(so);
332 	if (unp == NULL) {
333 		UNP_UNLOCK();
334 		return (EINVAL);
335 	}
336 	switch (so->so_type) {
337 	case SOCK_DGRAM:
338 		panic("uipc_rcvd DGRAM?");
339 		/*NOTREACHED*/
340 
341 	case SOCK_STREAM:
342 		if (unp->unp_conn == NULL)
343 			break;
344 		so2 = unp->unp_conn->unp_socket;
345 		SOCKBUF_LOCK(&so2->so_snd);
346 		SOCKBUF_LOCK(&so->so_rcv);
347 		/*
348 		 * Adjust backpressure on sender
349 		 * and wakeup any waiting to write.
350 		 */
351 		so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt;
352 		unp->unp_mbcnt = so->so_rcv.sb_mbcnt;
353 		newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc -
354 		    so->so_rcv.sb_cc;
355 		(void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
356 		    newhiwat, RLIM_INFINITY);
357 		unp->unp_cc = so->so_rcv.sb_cc;
358 		SOCKBUF_UNLOCK(&so->so_rcv);
359 		sowwakeup_locked(so2);
360 		break;
361 
362 	default:
363 		panic("uipc_rcvd unknown socktype");
364 	}
365 	UNP_UNLOCK();
366 	return (0);
367 }
368 
369 /* pru_rcvoob is EOPNOTSUPP */
370 
371 static int
372 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
373     struct mbuf *control, struct thread *td)
374 {
375 	int error = 0;
376 	struct unpcb *unp;
377 	struct socket *so2;
378 	u_long newhiwat;
379 
380 	unp = sotounpcb(so);
381 	if (unp == NULL) {
382 		error = EINVAL;
383 		goto release;
384 	}
385 	if (flags & PRUS_OOB) {
386 		error = EOPNOTSUPP;
387 		goto release;
388 	}
389 
390 	if (control != NULL && (error = unp_internalize(&control, td)))
391 		goto release;
392 
393 	UNP_LOCK();
394 	unp = sotounpcb(so);
395 	if (unp == NULL) {
396 		UNP_UNLOCK();
397 		error = EINVAL;
398 		goto dispose_release;
399 	}
400 
401 	switch (so->so_type) {
402 	case SOCK_DGRAM:
403 	{
404 		const struct sockaddr *from;
405 
406 		if (nam != NULL) {
407 			if (unp->unp_conn != NULL) {
408 				error = EISCONN;
409 				break;
410 			}
411 			error = unp_connect(so, nam, td);
412 			if (error)
413 				break;
414 		} else {
415 			if (unp->unp_conn == NULL) {
416 				error = ENOTCONN;
417 				break;
418 			}
419 		}
420 		so2 = unp->unp_conn->unp_socket;
421 		if (unp->unp_addr != NULL)
422 			from = (struct sockaddr *)unp->unp_addr;
423 		else
424 			from = &sun_noname;
425 		if (unp->unp_conn->unp_flags & UNP_WANTCRED)
426 			control = unp_addsockcred(td, control);
427 		SOCKBUF_LOCK(&so2->so_rcv);
428 		if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
429 			sorwakeup_locked(so2);
430 			m = NULL;
431 			control = NULL;
432 		} else {
433 			SOCKBUF_UNLOCK(&so2->so_rcv);
434 			error = ENOBUFS;
435 		}
436 		if (nam != NULL)
437 			unp_disconnect(unp);
438 		break;
439 	}
440 
441 	case SOCK_STREAM:
442 		/* Connect if not connected yet. */
443 		/*
444 		 * Note: A better implementation would complain
445 		 * if not equal to the peer's address.
446 		 */
447 		if ((so->so_state & SS_ISCONNECTED) == 0) {
448 			if (nam != NULL) {
449 				error = unp_connect(so, nam, td);
450 				if (error)
451 					break;	/* XXX */
452 			} else {
453 				error = ENOTCONN;
454 				break;
455 			}
456 		}
457 
458 		SOCKBUF_LOCK(&so->so_snd);
459 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
460 			SOCKBUF_UNLOCK(&so->so_snd);
461 			error = EPIPE;
462 			break;
463 		}
464 		if (unp->unp_conn == NULL)
465 			panic("uipc_send connected but no connection?");
466 		so2 = unp->unp_conn->unp_socket;
467 		SOCKBUF_LOCK(&so2->so_rcv);
468 		if (unp->unp_conn->unp_flags & UNP_WANTCRED) {
469 			/*
470 			 * Credentials are passed only once on
471 			 * SOCK_STREAM.
472 			 */
473 			unp->unp_conn->unp_flags &= ~UNP_WANTCRED;
474 			control = unp_addsockcred(td, control);
475 		}
476 		/*
477 		 * Send to paired receive port, and then reduce
478 		 * send buffer hiwater marks to maintain backpressure.
479 		 * Wake up readers.
480 		 */
481 		if (control != NULL) {
482 			if (sbappendcontrol_locked(&so2->so_rcv, m, control))
483 				control = NULL;
484 		} else {
485 			sbappend_locked(&so2->so_rcv, m);
486 		}
487 		so->so_snd.sb_mbmax -=
488 			so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt;
489 		unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt;
490 		newhiwat = so->so_snd.sb_hiwat -
491 		    (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc);
492 		(void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
493 		    newhiwat, RLIM_INFINITY);
494 		SOCKBUF_UNLOCK(&so->so_snd);
495 		unp->unp_conn->unp_cc = so2->so_rcv.sb_cc;
496 		sorwakeup_locked(so2);
497 		m = NULL;
498 		break;
499 
500 	default:
501 		panic("uipc_send unknown socktype");
502 	}
503 
504 	/*
505 	 * SEND_EOF is equivalent to a SEND followed by
506 	 * a SHUTDOWN.
507 	 */
508 	if (flags & PRUS_EOF) {
509 		socantsendmore(so);
510 		unp_shutdown(unp);
511 	}
512 	UNP_UNLOCK();
513 
514 dispose_release:
515 	if (control != NULL && error != 0)
516 		unp_dispose(control);
517 
518 release:
519 	if (control != NULL)
520 		m_freem(control);
521 	if (m != NULL)
522 		m_freem(m);
523 	return (error);
524 }
525 
526 static int
527 uipc_sense(struct socket *so, struct stat *sb)
528 {
529 	struct unpcb *unp;
530 	struct socket *so2;
531 
532 	UNP_LOCK();
533 	unp = sotounpcb(so);
534 	if (unp == NULL) {
535 		UNP_UNLOCK();
536 		return (EINVAL);
537 	}
538 	sb->st_blksize = so->so_snd.sb_hiwat;
539 	if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) {
540 		so2 = unp->unp_conn->unp_socket;
541 		sb->st_blksize += so2->so_rcv.sb_cc;
542 	}
543 	sb->st_dev = NODEV;
544 	if (unp->unp_ino == 0)
545 		unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
546 	sb->st_ino = unp->unp_ino;
547 	UNP_UNLOCK();
548 	return (0);
549 }
550 
551 static int
552 uipc_shutdown(struct socket *so)
553 {
554 	struct unpcb *unp;
555 
556 	UNP_LOCK();
557 	unp = sotounpcb(so);
558 	if (unp == NULL) {
559 		UNP_UNLOCK();
560 		return (EINVAL);
561 	}
562 	socantsendmore(so);
563 	unp_shutdown(unp);
564 	UNP_UNLOCK();
565 	return (0);
566 }
567 
568 static int
569 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
570 {
571 	struct unpcb *unp;
572 	const struct sockaddr *sa;
573 
574 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
575 	UNP_LOCK();
576 	unp = sotounpcb(so);
577 	if (unp == NULL) {
578 		UNP_UNLOCK();
579 		free(*nam, M_SONAME);
580 		*nam = NULL;
581 		return (EINVAL);
582 	}
583 	if (unp->unp_addr != NULL)
584 		sa = (struct sockaddr *) unp->unp_addr;
585 	else
586 		sa = &sun_noname;
587 	bcopy(sa, *nam, sa->sa_len);
588 	UNP_UNLOCK();
589 	return (0);
590 }
591 
592 struct pr_usrreqs uipc_usrreqs = {
593 	.pru_abort = 		uipc_abort,
594 	.pru_accept =		uipc_accept,
595 	.pru_attach =		uipc_attach,
596 	.pru_bind =		uipc_bind,
597 	.pru_connect =		uipc_connect,
598 	.pru_connect2 =		uipc_connect2,
599 	.pru_detach =		uipc_detach,
600 	.pru_disconnect =	uipc_disconnect,
601 	.pru_listen =		uipc_listen,
602 	.pru_peeraddr =		uipc_peeraddr,
603 	.pru_rcvd =		uipc_rcvd,
604 	.pru_send =		uipc_send,
605 	.pru_sense =		uipc_sense,
606 	.pru_shutdown =		uipc_shutdown,
607 	.pru_sockaddr =		uipc_sockaddr,
608 	.pru_sosend =		sosend,
609 	.pru_soreceive =	soreceive,
610 	.pru_sopoll =		sopoll,
611 };
612 
613 int
614 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
615 {
616 	struct unpcb *unp;
617 	struct xucred xu;
618 	int error, optval;
619 
620 	UNP_LOCK();
621 	unp = sotounpcb(so);
622 	if (unp == NULL) {
623 		UNP_UNLOCK();
624 		return (EINVAL);
625 	}
626 	error = 0;
627 
628 	switch (sopt->sopt_dir) {
629 	case SOPT_GET:
630 		switch (sopt->sopt_name) {
631 		case LOCAL_PEERCRED:
632 			if (unp->unp_flags & UNP_HAVEPC)
633 				xu = unp->unp_peercred;
634 			else {
635 				if (so->so_type == SOCK_STREAM)
636 					error = ENOTCONN;
637 				else
638 					error = EINVAL;
639 			}
640 			if (error == 0)
641 				error = sooptcopyout(sopt, &xu, sizeof(xu));
642 			break;
643 		case LOCAL_CREDS:
644 			optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
645 			error = sooptcopyout(sopt, &optval, sizeof(optval));
646 			break;
647 		case LOCAL_CONNWAIT:
648 			optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
649 			error = sooptcopyout(sopt, &optval, sizeof(optval));
650 			break;
651 		default:
652 			error = EOPNOTSUPP;
653 			break;
654 		}
655 		break;
656 	case SOPT_SET:
657 		switch (sopt->sopt_name) {
658 		case LOCAL_CREDS:
659 		case LOCAL_CONNWAIT:
660 			error = sooptcopyin(sopt, &optval, sizeof(optval),
661 					    sizeof(optval));
662 			if (error)
663 				break;
664 
665 #define	OPTSET(bit) \
666 	if (optval) \
667 		unp->unp_flags |= bit; \
668 	else \
669 		unp->unp_flags &= ~bit;
670 
671 			switch (sopt->sopt_name) {
672 			case LOCAL_CREDS:
673 				OPTSET(UNP_WANTCRED);
674 				break;
675 			case LOCAL_CONNWAIT:
676 				OPTSET(UNP_CONNWAIT);
677 				break;
678 			default:
679 				break;
680 			}
681 			break;
682 #undef	OPTSET
683 		default:
684 			error = ENOPROTOOPT;
685 			break;
686 		}
687 	default:
688 		error = EOPNOTSUPP;
689 		break;
690 	}
691 	UNP_UNLOCK();
692 	return (error);
693 }
694 
695 /*
696  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
697  * for stream sockets, although the total for sender and receiver is
698  * actually only PIPSIZ.
699  * Datagram sockets really use the sendspace as the maximum datagram size,
700  * and don't really want to reserve the sendspace.  Their recvspace should
701  * be large enough for at least one max-size datagram plus address.
702  */
703 #ifndef PIPSIZ
704 #define	PIPSIZ	8192
705 #endif
706 static u_long	unpst_sendspace = PIPSIZ;
707 static u_long	unpst_recvspace = PIPSIZ;
708 static u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
709 static u_long	unpdg_recvspace = 4*1024;
710 
711 static int	unp_rights;			/* file descriptors in flight */
712 
713 SYSCTL_DECL(_net_local_stream);
714 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
715 	   &unpst_sendspace, 0, "");
716 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
717 	   &unpst_recvspace, 0, "");
718 SYSCTL_DECL(_net_local_dgram);
719 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
720 	   &unpdg_sendspace, 0, "");
721 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
722 	   &unpdg_recvspace, 0, "");
723 SYSCTL_DECL(_net_local);
724 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
725 
726 static int
727 unp_attach(struct socket *so)
728 {
729 	struct unpcb *unp;
730 	int error;
731 
732 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
733 		switch (so->so_type) {
734 
735 		case SOCK_STREAM:
736 			error = soreserve(so, unpst_sendspace, unpst_recvspace);
737 			break;
738 
739 		case SOCK_DGRAM:
740 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
741 			break;
742 
743 		default:
744 			panic("unp_attach");
745 		}
746 		if (error)
747 			return (error);
748 	}
749 	unp = uma_zalloc(unp_zone, M_WAITOK | M_ZERO);
750 	if (unp == NULL)
751 		return (ENOBUFS);
752 	LIST_INIT(&unp->unp_refs);
753 	unp->unp_socket = so;
754 	so->so_pcb = unp;
755 
756 	UNP_LOCK();
757 	unp->unp_gencnt = ++unp_gencnt;
758 	unp_count++;
759 	LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
760 			 : &unp_shead, unp, unp_link);
761 	UNP_UNLOCK();
762 
763 	return (0);
764 }
765 
766 static void
767 unp_detach(struct unpcb *unp)
768 {
769 	struct vnode *vp;
770 
771 	UNP_LOCK_ASSERT();
772 
773 	LIST_REMOVE(unp, unp_link);
774 	unp->unp_gencnt = ++unp_gencnt;
775 	--unp_count;
776 	if ((vp = unp->unp_vnode) != NULL) {
777 		/*
778 		 * XXXRW: should v_socket be frobbed only while holding
779 		 * Giant?
780 		 */
781 		unp->unp_vnode->v_socket = NULL;
782 		unp->unp_vnode = NULL;
783 	}
784 	if (unp->unp_conn != NULL)
785 		unp_disconnect(unp);
786 	while (!LIST_EMPTY(&unp->unp_refs)) {
787 		struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
788 		unp_drop(ref, ECONNRESET);
789 	}
790 	soisdisconnected(unp->unp_socket);
791 	unp->unp_socket->so_pcb = NULL;
792 	if (unp_rights) {
793 		/*
794 		 * Normally the receive buffer is flushed later,
795 		 * in sofree, but if our receive buffer holds references
796 		 * to descriptors that are now garbage, we will dispose
797 		 * of those descriptor references after the garbage collector
798 		 * gets them (resulting in a "panic: closef: count < 0").
799 		 */
800 		sorflush(unp->unp_socket);
801 		unp_gc();	/* Will unlock UNP. */
802 	} else
803 		UNP_UNLOCK();
804 	UNP_UNLOCK_ASSERT();
805 	if (unp->unp_addr != NULL)
806 		FREE(unp->unp_addr, M_SONAME);
807 	uma_zfree(unp_zone, unp);
808 	if (vp) {
809 		mtx_lock(&Giant);
810 		vrele(vp);
811 		mtx_unlock(&Giant);
812 	}
813 }
814 
815 static int
816 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td)
817 {
818 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
819 	struct vnode *vp;
820 	struct mount *mp;
821 	struct vattr vattr;
822 	int error, namelen;
823 	struct nameidata nd;
824 	char *buf;
825 
826 	UNP_LOCK_ASSERT();
827 
828 	/*
829 	 * XXXRW: This test-and-set of unp_vnode is non-atomic; the
830 	 * unlocked read here is fine, but the value of unp_vnode needs
831 	 * to be tested again after we do all the lookups to see if the
832 	 * pcb is still unbound?
833 	 */
834 	if (unp->unp_vnode != NULL)
835 		return (EINVAL);
836 
837 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
838 	if (namelen <= 0)
839 		return (EINVAL);
840 
841 	UNP_UNLOCK();
842 
843 	buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
844 	strlcpy(buf, soun->sun_path, namelen + 1);
845 
846 	mtx_lock(&Giant);
847 restart:
848 	mtx_assert(&Giant, MA_OWNED);
849 	NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE,
850 	    buf, td);
851 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
852 	error = namei(&nd);
853 	if (error)
854 		goto done;
855 	vp = nd.ni_vp;
856 	if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
857 		NDFREE(&nd, NDF_ONLY_PNBUF);
858 		if (nd.ni_dvp == vp)
859 			vrele(nd.ni_dvp);
860 		else
861 			vput(nd.ni_dvp);
862 		if (vp != NULL) {
863 			vrele(vp);
864 			error = EADDRINUSE;
865 			goto done;
866 		}
867 		error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
868 		if (error)
869 			goto done;
870 		goto restart;
871 	}
872 	VATTR_NULL(&vattr);
873 	vattr.va_type = VSOCK;
874 	vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
875 #ifdef MAC
876 	error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
877 	    &vattr);
878 #endif
879 	if (error == 0) {
880 		VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
881 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
882 	}
883 	NDFREE(&nd, NDF_ONLY_PNBUF);
884 	vput(nd.ni_dvp);
885 	if (error) {
886 		vn_finished_write(mp);
887 		goto done;
888 	}
889 	vp = nd.ni_vp;
890 	ASSERT_VOP_LOCKED(vp, "unp_bind");
891 	soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
892 	UNP_LOCK();
893 	vp->v_socket = unp->unp_socket;
894 	unp->unp_vnode = vp;
895 	unp->unp_addr = soun;
896 	UNP_UNLOCK();
897 	VOP_UNLOCK(vp, 0, td);
898 	vn_finished_write(mp);
899 done:
900 	mtx_unlock(&Giant);
901 	free(buf, M_TEMP);
902 	UNP_LOCK();
903 	return (error);
904 }
905 
906 static int
907 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
908 {
909 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
910 	struct vnode *vp;
911 	struct socket *so2, *so3;
912 	struct unpcb *unp, *unp2, *unp3;
913 	int error, len;
914 	struct nameidata nd;
915 	char buf[SOCK_MAXADDRLEN];
916 	struct sockaddr *sa;
917 
918 	UNP_LOCK_ASSERT();
919 	unp = sotounpcb(so);
920 
921 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
922 	if (len <= 0)
923 		return (EINVAL);
924 	strlcpy(buf, soun->sun_path, len + 1);
925 	UNP_UNLOCK();
926 	sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
927 	mtx_lock(&Giant);
928 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td);
929 	error = namei(&nd);
930 	if (error)
931 		vp = NULL;
932 	else
933 		vp = nd.ni_vp;
934 	ASSERT_VOP_LOCKED(vp, "unp_connect");
935 	NDFREE(&nd, NDF_ONLY_PNBUF);
936 	if (error)
937 		goto bad;
938 
939 	if (vp->v_type != VSOCK) {
940 		error = ENOTSOCK;
941 		goto bad;
942 	}
943 	error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
944 	if (error)
945 		goto bad;
946 	mtx_unlock(&Giant);
947 	UNP_LOCK();
948 	unp = sotounpcb(so);
949 	if (unp == NULL) {
950 		error = EINVAL;
951 		goto bad2;
952 	}
953 	so2 = vp->v_socket;
954 	if (so2 == NULL) {
955 		error = ECONNREFUSED;
956 		goto bad2;
957 	}
958 	if (so->so_type != so2->so_type) {
959 		error = EPROTOTYPE;
960 		goto bad2;
961 	}
962 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
963 		if (so2->so_options & SO_ACCEPTCONN) {
964 			/*
965 			 * NB: drop locks here so unp_attach is entered
966 			 *     w/o locks; this avoids a recursive lock
967 			 *     of the head and holding sleep locks across
968 			 *     a (potentially) blocking malloc.
969 			 */
970 			UNP_UNLOCK();
971 			so3 = sonewconn(so2, 0);
972 			UNP_LOCK();
973 		} else
974 			so3 = NULL;
975 		if (so3 == NULL) {
976 			error = ECONNREFUSED;
977 			goto bad2;
978 		}
979 		unp = sotounpcb(so);
980 		unp2 = sotounpcb(so2);
981 		unp3 = sotounpcb(so3);
982 		if (unp2->unp_addr != NULL) {
983 			bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
984 			unp3->unp_addr = (struct sockaddr_un *) sa;
985 			sa = NULL;
986 		}
987 		/*
988 		 * unp_peercred management:
989 		 *
990 		 * The connecter's (client's) credentials are copied
991 		 * from its process structure at the time of connect()
992 		 * (which is now).
993 		 */
994 		cru2x(td->td_ucred, &unp3->unp_peercred);
995 		unp3->unp_flags |= UNP_HAVEPC;
996 		/*
997 		 * The receiver's (server's) credentials are copied
998 		 * from the unp_peercred member of socket on which the
999 		 * former called listen(); unp_listen() cached that
1000 		 * process's credentials at that time so we can use
1001 		 * them now.
1002 		 */
1003 		KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1004 		    ("unp_connect: listener without cached peercred"));
1005 		memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1006 		    sizeof(unp->unp_peercred));
1007 		unp->unp_flags |= UNP_HAVEPC;
1008 #ifdef MAC
1009 		SOCK_LOCK(so);
1010 		mac_set_socket_peer_from_socket(so, so3);
1011 		mac_set_socket_peer_from_socket(so3, so);
1012 		SOCK_UNLOCK(so);
1013 #endif
1014 
1015 		so2 = so3;
1016 	}
1017 	error = unp_connect2(so, so2, PRU_CONNECT);
1018 bad2:
1019 	UNP_UNLOCK();
1020 	mtx_lock(&Giant);
1021 bad:
1022 	mtx_assert(&Giant, MA_OWNED);
1023 	if (vp != NULL)
1024 		vput(vp);
1025 	mtx_unlock(&Giant);
1026 	free(sa, M_SONAME);
1027 	UNP_LOCK();
1028 	return (error);
1029 }
1030 
1031 static int
1032 unp_connect2(struct socket *so, struct socket *so2, int req)
1033 {
1034 	struct unpcb *unp = sotounpcb(so);
1035 	struct unpcb *unp2;
1036 
1037 	UNP_LOCK_ASSERT();
1038 
1039 	if (so2->so_type != so->so_type)
1040 		return (EPROTOTYPE);
1041 	unp2 = sotounpcb(so2);
1042 	unp->unp_conn = unp2;
1043 	switch (so->so_type) {
1044 
1045 	case SOCK_DGRAM:
1046 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1047 		soisconnected(so);
1048 		break;
1049 
1050 	case SOCK_STREAM:
1051 		unp2->unp_conn = unp;
1052 		if (req == PRU_CONNECT &&
1053 		    ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1054 			soisconnecting(so);
1055 		else
1056 			soisconnected(so);
1057 		soisconnected(so2);
1058 		break;
1059 
1060 	default:
1061 		panic("unp_connect2");
1062 	}
1063 	return (0);
1064 }
1065 
1066 static void
1067 unp_disconnect(struct unpcb *unp)
1068 {
1069 	struct unpcb *unp2 = unp->unp_conn;
1070 	struct socket *so;
1071 
1072 	UNP_LOCK_ASSERT();
1073 
1074 	if (unp2 == NULL)
1075 		return;
1076 	unp->unp_conn = NULL;
1077 	switch (unp->unp_socket->so_type) {
1078 
1079 	case SOCK_DGRAM:
1080 		LIST_REMOVE(unp, unp_reflink);
1081 		so = unp->unp_socket;
1082 		SOCK_LOCK(so);
1083 		so->so_state &= ~SS_ISCONNECTED;
1084 		SOCK_UNLOCK(so);
1085 		break;
1086 
1087 	case SOCK_STREAM:
1088 		soisdisconnected(unp->unp_socket);
1089 		unp2->unp_conn = NULL;
1090 		soisdisconnected(unp2->unp_socket);
1091 		break;
1092 	}
1093 }
1094 
1095 #ifdef notdef
1096 void
1097 unp_abort(struct unpcb *unp)
1098 {
1099 
1100 	unp_detach(unp);
1101 	UNP_UNLOCK_ASSERT();
1102 }
1103 #endif
1104 
1105 /*
1106  * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed
1107  * by the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers
1108  * are safe to reference.  It first scans the list of struct unpcb's to
1109  * generate a pointer list, then it rescans its list one entry at a time to
1110  * externalize and copyout.  It checks the generation number to see if a
1111  * struct unpcb has been reused, and will skip it if so.
1112  */
1113 static int
1114 unp_pcblist(SYSCTL_HANDLER_ARGS)
1115 {
1116 	int error, i, n;
1117 	struct unpcb *unp, **unp_list;
1118 	unp_gen_t gencnt;
1119 	struct xunpgen *xug;
1120 	struct unp_head *head;
1121 	struct xunpcb *xu;
1122 
1123 	head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1124 
1125 	/*
1126 	 * The process of preparing the PCB list is too time-consuming and
1127 	 * resource-intensive to repeat twice on every request.
1128 	 */
1129 	if (req->oldptr == NULL) {
1130 		n = unp_count;
1131 		req->oldidx = 2 * (sizeof *xug)
1132 			+ (n + n/8) * sizeof(struct xunpcb);
1133 		return (0);
1134 	}
1135 
1136 	if (req->newptr != NULL)
1137 		return (EPERM);
1138 
1139 	/*
1140 	 * OK, now we're committed to doing something.
1141 	 */
1142 	xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1143 	UNP_LOCK();
1144 	gencnt = unp_gencnt;
1145 	n = unp_count;
1146 	UNP_UNLOCK();
1147 
1148 	xug->xug_len = sizeof *xug;
1149 	xug->xug_count = n;
1150 	xug->xug_gen = gencnt;
1151 	xug->xug_sogen = so_gencnt;
1152 	error = SYSCTL_OUT(req, xug, sizeof *xug);
1153 	if (error) {
1154 		free(xug, M_TEMP);
1155 		return (error);
1156 	}
1157 
1158 	unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1159 
1160 	UNP_LOCK();
1161 	for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1162 	     unp = LIST_NEXT(unp, unp_link)) {
1163 		if (unp->unp_gencnt <= gencnt) {
1164 			if (cr_cansee(req->td->td_ucred,
1165 			    unp->unp_socket->so_cred))
1166 				continue;
1167 			unp_list[i++] = unp;
1168 		}
1169 	}
1170 	UNP_UNLOCK();
1171 	n = i;			/* in case we lost some during malloc */
1172 
1173 	error = 0;
1174 	xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK);
1175 	for (i = 0; i < n; i++) {
1176 		unp = unp_list[i];
1177 		if (unp->unp_gencnt <= gencnt) {
1178 			xu->xu_len = sizeof *xu;
1179 			xu->xu_unpp = unp;
1180 			/*
1181 			 * XXX - need more locking here to protect against
1182 			 * connect/disconnect races for SMP.
1183 			 */
1184 			if (unp->unp_addr != NULL)
1185 				bcopy(unp->unp_addr, &xu->xu_addr,
1186 				      unp->unp_addr->sun_len);
1187 			if (unp->unp_conn != NULL &&
1188 			    unp->unp_conn->unp_addr != NULL)
1189 				bcopy(unp->unp_conn->unp_addr,
1190 				      &xu->xu_caddr,
1191 				      unp->unp_conn->unp_addr->sun_len);
1192 			bcopy(unp, &xu->xu_unp, sizeof *unp);
1193 			sotoxsocket(unp->unp_socket, &xu->xu_socket);
1194 			error = SYSCTL_OUT(req, xu, sizeof *xu);
1195 		}
1196 	}
1197 	free(xu, M_TEMP);
1198 	if (!error) {
1199 		/*
1200 		 * Give the user an updated idea of our state.
1201 		 * If the generation differs from what we told
1202 		 * her before, she knows that something happened
1203 		 * while we were processing this request, and it
1204 		 * might be necessary to retry.
1205 		 */
1206 		xug->xug_gen = unp_gencnt;
1207 		xug->xug_sogen = so_gencnt;
1208 		xug->xug_count = unp_count;
1209 		error = SYSCTL_OUT(req, xug, sizeof *xug);
1210 	}
1211 	free(unp_list, M_TEMP);
1212 	free(xug, M_TEMP);
1213 	return (error);
1214 }
1215 
1216 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1217 	    (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1218 	    "List of active local datagram sockets");
1219 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1220 	    (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1221 	    "List of active local stream sockets");
1222 
1223 static void
1224 unp_shutdown(struct unpcb *unp)
1225 {
1226 	struct socket *so;
1227 
1228 	UNP_LOCK_ASSERT();
1229 
1230 	if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
1231 	    (so = unp->unp_conn->unp_socket))
1232 		socantrcvmore(so);
1233 }
1234 
1235 static void
1236 unp_drop(struct unpcb *unp, int errno)
1237 {
1238 	struct socket *so = unp->unp_socket;
1239 
1240 	UNP_LOCK_ASSERT();
1241 
1242 	so->so_error = errno;
1243 	unp_disconnect(unp);
1244 }
1245 
1246 #ifdef notdef
1247 void
1248 unp_drain(void)
1249 {
1250 
1251 }
1252 #endif
1253 
1254 static void
1255 unp_freerights(struct file **rp, int fdcount)
1256 {
1257 	int i;
1258 	struct file *fp;
1259 
1260 	for (i = 0; i < fdcount; i++) {
1261 		fp = *rp;
1262 		/*
1263 		 * zero the pointer before calling
1264 		 * unp_discard since it may end up
1265 		 * in unp_gc()..
1266 		 */
1267 		*rp++ = 0;
1268 		unp_discard(fp);
1269 	}
1270 }
1271 
1272 int
1273 unp_externalize(struct mbuf *control, struct mbuf **controlp)
1274 {
1275 	struct thread *td = curthread;		/* XXX */
1276 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1277 	int i;
1278 	int *fdp;
1279 	struct file **rp;
1280 	struct file *fp;
1281 	void *data;
1282 	socklen_t clen = control->m_len, datalen;
1283 	int error, newfds;
1284 	int f;
1285 	u_int newlen;
1286 
1287 	UNP_UNLOCK_ASSERT();
1288 
1289 	error = 0;
1290 	if (controlp != NULL) /* controlp == NULL => free control messages */
1291 		*controlp = NULL;
1292 
1293 	while (cm != NULL) {
1294 		if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1295 			error = EINVAL;
1296 			break;
1297 		}
1298 
1299 		data = CMSG_DATA(cm);
1300 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1301 
1302 		if (cm->cmsg_level == SOL_SOCKET
1303 		    && cm->cmsg_type == SCM_RIGHTS) {
1304 			newfds = datalen / sizeof(struct file *);
1305 			rp = data;
1306 
1307 			/* If we're not outputting the descriptors free them. */
1308 			if (error || controlp == NULL) {
1309 				unp_freerights(rp, newfds);
1310 				goto next;
1311 			}
1312 			FILEDESC_LOCK(td->td_proc->p_fd);
1313 			/* if the new FD's will not fit free them.  */
1314 			if (!fdavail(td, newfds)) {
1315 				FILEDESC_UNLOCK(td->td_proc->p_fd);
1316 				error = EMSGSIZE;
1317 				unp_freerights(rp, newfds);
1318 				goto next;
1319 			}
1320 			/*
1321 			 * now change each pointer to an fd in the global
1322 			 * table to an integer that is the index to the
1323 			 * local fd table entry that we set up to point
1324 			 * to the global one we are transferring.
1325 			 */
1326 			newlen = newfds * sizeof(int);
1327 			*controlp = sbcreatecontrol(NULL, newlen,
1328 			    SCM_RIGHTS, SOL_SOCKET);
1329 			if (*controlp == NULL) {
1330 				FILEDESC_UNLOCK(td->td_proc->p_fd);
1331 				error = E2BIG;
1332 				unp_freerights(rp, newfds);
1333 				goto next;
1334 			}
1335 
1336 			fdp = (int *)
1337 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1338 			for (i = 0; i < newfds; i++) {
1339 				if (fdalloc(td, 0, &f))
1340 					panic("unp_externalize fdalloc failed");
1341 				fp = *rp++;
1342 				td->td_proc->p_fd->fd_ofiles[f] = fp;
1343 				FILE_LOCK(fp);
1344 				fp->f_msgcount--;
1345 				FILE_UNLOCK(fp);
1346 				unp_rights--;
1347 				*fdp++ = f;
1348 			}
1349 			FILEDESC_UNLOCK(td->td_proc->p_fd);
1350 		} else { /* We can just copy anything else across */
1351 			if (error || controlp == NULL)
1352 				goto next;
1353 			*controlp = sbcreatecontrol(NULL, datalen,
1354 			    cm->cmsg_type, cm->cmsg_level);
1355 			if (*controlp == NULL) {
1356 				error = ENOBUFS;
1357 				goto next;
1358 			}
1359 			bcopy(data,
1360 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1361 			    datalen);
1362 		}
1363 
1364 		controlp = &(*controlp)->m_next;
1365 
1366 next:
1367 		if (CMSG_SPACE(datalen) < clen) {
1368 			clen -= CMSG_SPACE(datalen);
1369 			cm = (struct cmsghdr *)
1370 			    ((caddr_t)cm + CMSG_SPACE(datalen));
1371 		} else {
1372 			clen = 0;
1373 			cm = NULL;
1374 		}
1375 	}
1376 
1377 	m_freem(control);
1378 
1379 	return (error);
1380 }
1381 
1382 void
1383 unp_init(void)
1384 {
1385 	unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1386 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1387 	if (unp_zone == NULL)
1388 		panic("unp_init");
1389 	uma_zone_set_max(unp_zone, nmbclusters);
1390 	LIST_INIT(&unp_dhead);
1391 	LIST_INIT(&unp_shead);
1392 
1393 	UNP_LOCK_INIT();
1394 }
1395 
1396 static int
1397 unp_internalize(struct mbuf **controlp, struct thread *td)
1398 {
1399 	struct mbuf *control = *controlp;
1400 	struct proc *p = td->td_proc;
1401 	struct filedesc *fdescp = p->p_fd;
1402 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1403 	struct cmsgcred *cmcred;
1404 	struct file **rp;
1405 	struct file *fp;
1406 	struct timeval *tv;
1407 	int i, fd, *fdp;
1408 	void *data;
1409 	socklen_t clen = control->m_len, datalen;
1410 	int error, oldfds;
1411 	u_int newlen;
1412 
1413 	UNP_UNLOCK_ASSERT();
1414 
1415 	error = 0;
1416 	*controlp = NULL;
1417 
1418 	while (cm != NULL) {
1419 		if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1420 		    || cm->cmsg_len > clen) {
1421 			error = EINVAL;
1422 			goto out;
1423 		}
1424 
1425 		data = CMSG_DATA(cm);
1426 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1427 
1428 		switch (cm->cmsg_type) {
1429 		/*
1430 		 * Fill in credential information.
1431 		 */
1432 		case SCM_CREDS:
1433 			*controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1434 			    SCM_CREDS, SOL_SOCKET);
1435 			if (*controlp == NULL) {
1436 				error = ENOBUFS;
1437 				goto out;
1438 			}
1439 
1440 			cmcred = (struct cmsgcred *)
1441 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1442 			cmcred->cmcred_pid = p->p_pid;
1443 			cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1444 			cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1445 			cmcred->cmcred_euid = td->td_ucred->cr_uid;
1446 			cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1447 							CMGROUP_MAX);
1448 			for (i = 0; i < cmcred->cmcred_ngroups; i++)
1449 				cmcred->cmcred_groups[i] =
1450 				    td->td_ucred->cr_groups[i];
1451 			break;
1452 
1453 		case SCM_RIGHTS:
1454 			oldfds = datalen / sizeof (int);
1455 			/*
1456 			 * check that all the FDs passed in refer to legal files
1457 			 * If not, reject the entire operation.
1458 			 */
1459 			fdp = data;
1460 			FILEDESC_LOCK(fdescp);
1461 			for (i = 0; i < oldfds; i++) {
1462 				fd = *fdp++;
1463 				if ((unsigned)fd >= fdescp->fd_nfiles ||
1464 				    fdescp->fd_ofiles[fd] == NULL) {
1465 					FILEDESC_UNLOCK(fdescp);
1466 					error = EBADF;
1467 					goto out;
1468 				}
1469 				fp = fdescp->fd_ofiles[fd];
1470 				if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1471 					FILEDESC_UNLOCK(fdescp);
1472 					error = EOPNOTSUPP;
1473 					goto out;
1474 				}
1475 
1476 			}
1477 			/*
1478 			 * Now replace the integer FDs with pointers to
1479 			 * the associated global file table entry..
1480 			 */
1481 			newlen = oldfds * sizeof(struct file *);
1482 			*controlp = sbcreatecontrol(NULL, newlen,
1483 			    SCM_RIGHTS, SOL_SOCKET);
1484 			if (*controlp == NULL) {
1485 				FILEDESC_UNLOCK(fdescp);
1486 				error = E2BIG;
1487 				goto out;
1488 			}
1489 
1490 			fdp = data;
1491 			rp = (struct file **)
1492 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1493 			for (i = 0; i < oldfds; i++) {
1494 				fp = fdescp->fd_ofiles[*fdp++];
1495 				*rp++ = fp;
1496 				FILE_LOCK(fp);
1497 				fp->f_count++;
1498 				fp->f_msgcount++;
1499 				FILE_UNLOCK(fp);
1500 				unp_rights++;
1501 			}
1502 			FILEDESC_UNLOCK(fdescp);
1503 			break;
1504 
1505 		case SCM_TIMESTAMP:
1506 			*controlp = sbcreatecontrol(NULL, sizeof(*tv),
1507 			    SCM_TIMESTAMP, SOL_SOCKET);
1508 			if (*controlp == NULL) {
1509 				error = ENOBUFS;
1510 				goto out;
1511 			}
1512 			tv = (struct timeval *)
1513 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1514 			microtime(tv);
1515 			break;
1516 
1517 		default:
1518 			error = EINVAL;
1519 			goto out;
1520 		}
1521 
1522 		controlp = &(*controlp)->m_next;
1523 
1524 		if (CMSG_SPACE(datalen) < clen) {
1525 			clen -= CMSG_SPACE(datalen);
1526 			cm = (struct cmsghdr *)
1527 			    ((caddr_t)cm + CMSG_SPACE(datalen));
1528 		} else {
1529 			clen = 0;
1530 			cm = NULL;
1531 		}
1532 	}
1533 
1534 out:
1535 	m_freem(control);
1536 
1537 	return (error);
1538 }
1539 
1540 struct mbuf *
1541 unp_addsockcred(struct thread *td, struct mbuf *control)
1542 {
1543 	struct mbuf *m, *n;
1544 	struct sockcred *sc;
1545 	int ngroups;
1546 	int i;
1547 
1548 	ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1549 
1550 	m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
1551 	if (m == NULL)
1552 		return (control);
1553 	m->m_next = NULL;
1554 
1555 	sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
1556 	sc->sc_uid = td->td_ucred->cr_ruid;
1557 	sc->sc_euid = td->td_ucred->cr_uid;
1558 	sc->sc_gid = td->td_ucred->cr_rgid;
1559 	sc->sc_egid = td->td_ucred->cr_gid;
1560 	sc->sc_ngroups = ngroups;
1561 	for (i = 0; i < sc->sc_ngroups; i++)
1562 		sc->sc_groups[i] = td->td_ucred->cr_groups[i];
1563 
1564 	/*
1565 	 * If a control message already exists, append us to the end.
1566 	 */
1567 	if (control != NULL) {
1568 		for (n = control; n->m_next != NULL; n = n->m_next)
1569 			;
1570 		n->m_next = m;
1571 	} else
1572 		control = m;
1573 
1574 	return (control);
1575 }
1576 
1577 /*
1578  * unp_defer is thread-local during garbage collection, and does not require
1579  * explicit synchronization.  unp_gcing prevents other threads from entering
1580  * garbage collection, and perhaps should be an sx lock instead.
1581  */
1582 static int	unp_defer, unp_gcing;
1583 
1584 static void
1585 unp_gc(void)
1586 {
1587 	struct file *fp, *nextfp;
1588 	struct socket *so;
1589 	struct file **extra_ref, **fpp;
1590 	int nunref, i;
1591 	int nfiles_snap;
1592 	int nfiles_slack = 20;
1593 
1594 	UNP_LOCK_ASSERT();
1595 
1596 	if (unp_gcing) {
1597 		UNP_UNLOCK();
1598 		return;
1599 	}
1600 	unp_gcing = 1;
1601 	unp_defer = 0;
1602 	UNP_UNLOCK();
1603 	/*
1604 	 * before going through all this, set all FDs to
1605 	 * be NOT defered and NOT externally accessible
1606 	 */
1607 	sx_slock(&filelist_lock);
1608 	LIST_FOREACH(fp, &filehead, f_list)
1609 		fp->f_gcflag &= ~(FMARK|FDEFER);
1610 	do {
1611 		LIST_FOREACH(fp, &filehead, f_list) {
1612 			FILE_LOCK(fp);
1613 			/*
1614 			 * If the file is not open, skip it
1615 			 */
1616 			if (fp->f_count == 0) {
1617 				FILE_UNLOCK(fp);
1618 				continue;
1619 			}
1620 			/*
1621 			 * If we already marked it as 'defer'  in a
1622 			 * previous pass, then try process it this time
1623 			 * and un-mark it
1624 			 */
1625 			if (fp->f_gcflag & FDEFER) {
1626 				fp->f_gcflag &= ~FDEFER;
1627 				unp_defer--;
1628 			} else {
1629 				/*
1630 				 * if it's not defered, then check if it's
1631 				 * already marked.. if so skip it
1632 				 */
1633 				if (fp->f_gcflag & FMARK) {
1634 					FILE_UNLOCK(fp);
1635 					continue;
1636 				}
1637 				/*
1638 				 * If all references are from messages
1639 				 * in transit, then skip it. it's not
1640 				 * externally accessible.
1641 				 */
1642 				if (fp->f_count == fp->f_msgcount) {
1643 					FILE_UNLOCK(fp);
1644 					continue;
1645 				}
1646 				/*
1647 				 * If it got this far then it must be
1648 				 * externally accessible.
1649 				 */
1650 				fp->f_gcflag |= FMARK;
1651 			}
1652 			/*
1653 			 * either it was defered, or it is externally
1654 			 * accessible and not already marked so.
1655 			 * Now check if it is possibly one of OUR sockets.
1656 			 */
1657 			if (fp->f_type != DTYPE_SOCKET ||
1658 			    (so = fp->f_data) == NULL) {
1659 				FILE_UNLOCK(fp);
1660 				continue;
1661 			}
1662 			FILE_UNLOCK(fp);
1663 			if (so->so_proto->pr_domain != &localdomain ||
1664 			    (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1665 				continue;
1666 #ifdef notdef
1667 			if (so->so_rcv.sb_flags & SB_LOCK) {
1668 				/*
1669 				 * This is problematical; it's not clear
1670 				 * we need to wait for the sockbuf to be
1671 				 * unlocked (on a uniprocessor, at least),
1672 				 * and it's also not clear what to do
1673 				 * if sbwait returns an error due to receipt
1674 				 * of a signal.  If sbwait does return
1675 				 * an error, we'll go into an infinite
1676 				 * loop.  Delete all of this for now.
1677 				 */
1678 				(void) sbwait(&so->so_rcv);
1679 				goto restart;
1680 			}
1681 #endif
1682 			/*
1683 			 * So, Ok, it's one of our sockets and it IS externally
1684 			 * accessible (or was defered). Now we look
1685 			 * to see if we hold any file descriptors in its
1686 			 * message buffers. Follow those links and mark them
1687 			 * as accessible too.
1688 			 */
1689 			SOCKBUF_LOCK(&so->so_rcv);
1690 			unp_scan(so->so_rcv.sb_mb, unp_mark);
1691 			SOCKBUF_UNLOCK(&so->so_rcv);
1692 		}
1693 	} while (unp_defer);
1694 	sx_sunlock(&filelist_lock);
1695 	/*
1696 	 * We grab an extra reference to each of the file table entries
1697 	 * that are not otherwise accessible and then free the rights
1698 	 * that are stored in messages on them.
1699 	 *
1700 	 * The bug in the orginal code is a little tricky, so I'll describe
1701 	 * what's wrong with it here.
1702 	 *
1703 	 * It is incorrect to simply unp_discard each entry for f_msgcount
1704 	 * times -- consider the case of sockets A and B that contain
1705 	 * references to each other.  On a last close of some other socket,
1706 	 * we trigger a gc since the number of outstanding rights (unp_rights)
1707 	 * is non-zero.  If during the sweep phase the gc code un_discards,
1708 	 * we end up doing a (full) closef on the descriptor.  A closef on A
1709 	 * results in the following chain.  Closef calls soo_close, which
1710 	 * calls soclose.   Soclose calls first (through the switch
1711 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
1712 	 * returns because the previous instance had set unp_gcing, and
1713 	 * we return all the way back to soclose, which marks the socket
1714 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
1715 	 * to free up the rights that are queued in messages on the socket A,
1716 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
1717 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
1718 	 * instance of unp_discard just calls closef on B.
1719 	 *
1720 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1721 	 * which results in another closef on A.  Unfortunately, A is already
1722 	 * being closed, and the descriptor has already been marked with
1723 	 * SS_NOFDREF, and soclose panics at this point.
1724 	 *
1725 	 * Here, we first take an extra reference to each inaccessible
1726 	 * descriptor.  Then, we call sorflush ourself, since we know
1727 	 * it is a Unix domain socket anyhow.  After we destroy all the
1728 	 * rights carried in messages, we do a last closef to get rid
1729 	 * of our extra reference.  This is the last close, and the
1730 	 * unp_detach etc will shut down the socket.
1731 	 *
1732 	 * 91/09/19, bsy@cs.cmu.edu
1733 	 */
1734 again:
1735 	nfiles_snap = openfiles + nfiles_slack;	/* some slack */
1736 	extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP,
1737 	    M_WAITOK);
1738 	sx_slock(&filelist_lock);
1739 	if (nfiles_snap < openfiles) {
1740 		sx_sunlock(&filelist_lock);
1741 		free(extra_ref, M_TEMP);
1742 		nfiles_slack += 20;
1743 		goto again;
1744 	}
1745 	for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref;
1746 	    fp != NULL; fp = nextfp) {
1747 		nextfp = LIST_NEXT(fp, f_list);
1748 		FILE_LOCK(fp);
1749 		/*
1750 		 * If it's not open, skip it
1751 		 */
1752 		if (fp->f_count == 0) {
1753 			FILE_UNLOCK(fp);
1754 			continue;
1755 		}
1756 		/*
1757 		 * If all refs are from msgs, and it's not marked accessible
1758 		 * then it must be referenced from some unreachable cycle
1759 		 * of (shut-down) FDs, so include it in our
1760 		 * list of FDs to remove
1761 		 */
1762 		if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) {
1763 			*fpp++ = fp;
1764 			nunref++;
1765 			fp->f_count++;
1766 		}
1767 		FILE_UNLOCK(fp);
1768 	}
1769 	sx_sunlock(&filelist_lock);
1770 	/*
1771 	 * for each FD on our hit list, do the following two things
1772 	 */
1773 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1774 		struct file *tfp = *fpp;
1775 		FILE_LOCK(tfp);
1776 		if (tfp->f_type == DTYPE_SOCKET &&
1777 		    tfp->f_data != NULL) {
1778 			FILE_UNLOCK(tfp);
1779 			sorflush(tfp->f_data);
1780 		} else {
1781 			FILE_UNLOCK(tfp);
1782 		}
1783 	}
1784 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
1785 		closef(*fpp, (struct thread *) NULL);
1786 	free(extra_ref, M_TEMP);
1787 	unp_gcing = 0;
1788 
1789 	UNP_UNLOCK_ASSERT();
1790 }
1791 
1792 void
1793 unp_dispose(struct mbuf *m)
1794 {
1795 
1796 	if (m)
1797 		unp_scan(m, unp_discard);
1798 }
1799 
1800 static int
1801 unp_listen(struct socket *so, struct unpcb *unp, struct thread *td)
1802 {
1803 	int error;
1804 
1805 	UNP_LOCK_ASSERT();
1806 
1807 	SOCK_LOCK(so);
1808 	error = solisten_proto_check(so);
1809 	if (error == 0) {
1810 		cru2x(td->td_ucred, &unp->unp_peercred);
1811 		unp->unp_flags |= UNP_HAVEPCCACHED;
1812 		solisten_proto(so);
1813 	}
1814 	SOCK_UNLOCK(so);
1815 	return (error);
1816 }
1817 
1818 static void
1819 unp_scan(struct mbuf *m0, void (*op)(struct file *))
1820 {
1821 	struct mbuf *m;
1822 	struct file **rp;
1823 	struct cmsghdr *cm;
1824 	void *data;
1825 	int i;
1826 	socklen_t clen, datalen;
1827 	int qfds;
1828 
1829 	while (m0 != NULL) {
1830 		for (m = m0; m; m = m->m_next) {
1831 			if (m->m_type != MT_CONTROL)
1832 				continue;
1833 
1834 			cm = mtod(m, struct cmsghdr *);
1835 			clen = m->m_len;
1836 
1837 			while (cm != NULL) {
1838 				if (sizeof(*cm) > clen || cm->cmsg_len > clen)
1839 					break;
1840 
1841 				data = CMSG_DATA(cm);
1842 				datalen = (caddr_t)cm + cm->cmsg_len
1843 				    - (caddr_t)data;
1844 
1845 				if (cm->cmsg_level == SOL_SOCKET &&
1846 				    cm->cmsg_type == SCM_RIGHTS) {
1847 					qfds = datalen / sizeof (struct file *);
1848 					rp = data;
1849 					for (i = 0; i < qfds; i++)
1850 						(*op)(*rp++);
1851 				}
1852 
1853 				if (CMSG_SPACE(datalen) < clen) {
1854 					clen -= CMSG_SPACE(datalen);
1855 					cm = (struct cmsghdr *)
1856 					    ((caddr_t)cm + CMSG_SPACE(datalen));
1857 				} else {
1858 					clen = 0;
1859 					cm = NULL;
1860 				}
1861 			}
1862 		}
1863 		m0 = m0->m_act;
1864 	}
1865 }
1866 
1867 static void
1868 unp_mark(struct file *fp)
1869 {
1870 	if (fp->f_gcflag & FMARK)
1871 		return;
1872 	unp_defer++;
1873 	fp->f_gcflag |= (FMARK|FDEFER);
1874 }
1875 
1876 static void
1877 unp_discard(struct file *fp)
1878 {
1879 	FILE_LOCK(fp);
1880 	fp->f_msgcount--;
1881 	unp_rights--;
1882 	FILE_UNLOCK(fp);
1883 	(void) closef(fp, (struct thread *)NULL);
1884 }
1885