xref: /freebsd/sys/kern/uipc_usrreq.c (revision ca9ac06c99bfd0150b85d4d83c396ce6237c0e05)
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 
85 /*
86  * Currently, UNIX domain sockets are protected by a single subsystem lock,
87  * which covers global data structures and variables, the contents of each
88  * per-socket unpcb structure, and the so_pcb field in sockets attached to
89  * the UNIX domain.  This provides for a moderate degree of paralellism, as
90  * receive operations on UNIX domain sockets do not need to acquire the
91  * subsystem lock.  Finer grained locking to permit send() without acquiring
92  * a global lock would be a logical next step.
93  *
94  * The UNIX domain socket lock preceds all socket layer locks, including the
95  * socket lock and socket buffer lock, permitting UNIX domain socket code to
96  * call into socket support routines without releasing its locks.
97  *
98  * Some caution is required in areas where the UNIX domain socket code enters
99  * VFS in order to create or find rendezvous points.  This results in
100  * dropping of the UNIX domain socket subsystem lock, acquisition of the
101  * Giant lock, and potential sleeping.  This increases the chances of races,
102  * and exposes weaknesses in the socket->protocol API by offering poor
103  * failure modes.
104  */
105 static struct mtx unp_mtx;
106 #define	UNP_LOCK_INIT() \
107 	mtx_init(&unp_mtx, "unp", NULL, MTX_DEF)
108 #define	UNP_LOCK()		mtx_lock(&unp_mtx)
109 #define	UNP_UNLOCK()		mtx_unlock(&unp_mtx)
110 #define	UNP_LOCK_ASSERT()	mtx_assert(&unp_mtx, MA_OWNED)
111 #define	UNP_UNLOCK_ASSERT()	mtx_assert(&unp_mtx, MA_NOTOWNED)
112 
113 static int     unp_attach(struct socket *);
114 static void    unp_detach(struct unpcb *);
115 static int     unp_bind(struct unpcb *,struct sockaddr *, struct thread *);
116 static int     unp_connect(struct socket *,struct sockaddr *, struct thread *);
117 static int     unp_connect2(struct socket *so, struct socket *so2);
118 static void    unp_disconnect(struct unpcb *);
119 static void    unp_shutdown(struct unpcb *);
120 static void    unp_drop(struct unpcb *, int);
121 static void    unp_gc(void);
122 static void    unp_scan(struct mbuf *, void (*)(struct file *));
123 static void    unp_mark(struct file *);
124 static void    unp_discard(struct file *);
125 static void    unp_freerights(struct file **, int);
126 static int     unp_internalize(struct mbuf **, struct thread *);
127 static int     unp_listen(struct socket *, struct unpcb *, struct thread *);
128 
129 static int
130 uipc_abort(struct socket *so)
131 {
132 	struct unpcb *unp;
133 
134 	UNP_LOCK();
135 	unp = sotounpcb(so);
136 	if (unp == NULL) {
137 		UNP_UNLOCK();
138 		return (EINVAL);
139 	}
140 	unp_drop(unp, ECONNABORTED);
141 	unp_detach(unp);
142 	UNP_UNLOCK_ASSERT();
143 	ACCEPT_LOCK();
144 	SOCK_LOCK(so);
145 	sotryfree(so);
146 	return (0);
147 }
148 
149 static int
150 uipc_accept(struct socket *so, struct sockaddr **nam)
151 {
152 	struct unpcb *unp;
153 	const struct sockaddr *sa;
154 
155 	/*
156 	 * Pass back name of connected socket,
157 	 * if it was bound and we are still connected
158 	 * (our peer may have closed already!).
159 	 */
160 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
161 	UNP_LOCK();
162 	unp = sotounpcb(so);
163 	if (unp == NULL) {
164 		UNP_UNLOCK();
165 		free(*nam, M_SONAME);
166 		*nam = NULL;
167 		return (EINVAL);
168 	}
169 	if (unp->unp_conn != NULL && unp->unp_conn->unp_addr != NULL)
170 		sa = (struct sockaddr *) unp->unp_conn->unp_addr;
171 	else
172 		sa = &sun_noname;
173 	bcopy(sa, *nam, sa->sa_len);
174 	UNP_UNLOCK();
175 	return (0);
176 }
177 
178 static int
179 uipc_attach(struct socket *so, int proto, struct thread *td)
180 {
181 	struct unpcb *unp = sotounpcb(so);
182 
183 	if (unp != NULL)
184 		return (EISCONN);
185 	return (unp_attach(so));
186 }
187 
188 static int
189 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
190 {
191 	struct unpcb *unp;
192 	int error;
193 
194 	UNP_LOCK();
195 	unp = sotounpcb(so);
196 	if (unp == NULL) {
197 		UNP_UNLOCK();
198 		return (EINVAL);
199 	}
200 	error = unp_bind(unp, nam, td);
201 	UNP_UNLOCK();
202 	return (error);
203 }
204 
205 static int
206 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
207 {
208 	struct unpcb *unp;
209 	int error;
210 
211 	KASSERT(td == curthread, ("uipc_connect: td != curthread"));
212 
213 	UNP_LOCK();
214 	unp = sotounpcb(so);
215 	if (unp == NULL) {
216 		UNP_UNLOCK();
217 		return (EINVAL);
218 	}
219 	error = unp_connect(so, nam, td);
220 	UNP_UNLOCK();
221 	return (error);
222 }
223 
224 int
225 uipc_connect2(struct socket *so1, struct socket *so2)
226 {
227 	struct unpcb *unp;
228 	int error;
229 
230 	UNP_LOCK();
231 	unp = sotounpcb(so1);
232 	if (unp == NULL) {
233 		UNP_UNLOCK();
234 		return (EINVAL);
235 	}
236 	error = unp_connect2(so1, so2);
237 	UNP_UNLOCK();
238 	return (error);
239 }
240 
241 /* control is EOPNOTSUPP */
242 
243 static int
244 uipc_detach(struct socket *so)
245 {
246 	struct unpcb *unp;
247 
248 	UNP_LOCK();
249 	unp = sotounpcb(so);
250 	if (unp == NULL) {
251 		UNP_UNLOCK();
252 		return (EINVAL);
253 	}
254 	unp_detach(unp);
255 	UNP_UNLOCK_ASSERT();
256 	return (0);
257 }
258 
259 static int
260 uipc_disconnect(struct socket *so)
261 {
262 	struct unpcb *unp;
263 
264 	UNP_LOCK();
265 	unp = sotounpcb(so);
266 	if (unp == NULL) {
267 		UNP_UNLOCK();
268 		return (EINVAL);
269 	}
270 	unp_disconnect(unp);
271 	UNP_UNLOCK();
272 	return (0);
273 }
274 
275 static int
276 uipc_listen(struct socket *so, struct thread *td)
277 {
278 	struct unpcb *unp;
279 	int error;
280 
281 	UNP_LOCK();
282 	unp = sotounpcb(so);
283 	if (unp == NULL || unp->unp_vnode == NULL) {
284 		UNP_UNLOCK();
285 		return (EINVAL);
286 	}
287 	error = unp_listen(so, unp, td);
288 	UNP_UNLOCK();
289 	return (error);
290 }
291 
292 static int
293 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
294 {
295 	struct unpcb *unp;
296 	const struct sockaddr *sa;
297 
298 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
299 	UNP_LOCK();
300 	unp = sotounpcb(so);
301 	if (unp == NULL) {
302 		UNP_UNLOCK();
303 		free(*nam, M_SONAME);
304 		*nam = NULL;
305 		return (EINVAL);
306 	}
307 	if (unp->unp_conn != NULL && unp->unp_conn->unp_addr!= NULL)
308 		sa = (struct sockaddr *) unp->unp_conn->unp_addr;
309 	else {
310 		/*
311 		 * XXX: It seems that this test always fails even when
312 		 * connection is established.  So, this else clause is
313 		 * added as workaround to return PF_LOCAL sockaddr.
314 		 */
315 		sa = &sun_noname;
316 	}
317 	bcopy(sa, *nam, sa->sa_len);
318 	UNP_UNLOCK();
319 	return (0);
320 }
321 
322 static int
323 uipc_rcvd(struct socket *so, int flags)
324 {
325 	struct unpcb *unp;
326 	struct socket *so2;
327 	u_long newhiwat;
328 
329 	UNP_LOCK();
330 	unp = sotounpcb(so);
331 	if (unp == NULL) {
332 		UNP_UNLOCK();
333 		return (EINVAL);
334 	}
335 	switch (so->so_type) {
336 	case SOCK_DGRAM:
337 		panic("uipc_rcvd DGRAM?");
338 		/*NOTREACHED*/
339 
340 	case SOCK_STREAM:
341 		if (unp->unp_conn == NULL)
342 			break;
343 		so2 = unp->unp_conn->unp_socket;
344 		SOCKBUF_LOCK(&so2->so_snd);
345 		SOCKBUF_LOCK(&so->so_rcv);
346 		/*
347 		 * Adjust backpressure on sender
348 		 * and wakeup any waiting to write.
349 		 */
350 		so2->so_snd.sb_mbmax += unp->unp_mbcnt - so->so_rcv.sb_mbcnt;
351 		unp->unp_mbcnt = so->so_rcv.sb_mbcnt;
352 		newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc -
353 		    so->so_rcv.sb_cc;
354 		(void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
355 		    newhiwat, RLIM_INFINITY);
356 		unp->unp_cc = so->so_rcv.sb_cc;
357 		SOCKBUF_UNLOCK(&so->so_rcv);
358 		sowwakeup_locked(so2);
359 		break;
360 
361 	default:
362 		panic("uipc_rcvd unknown socktype");
363 	}
364 	UNP_UNLOCK();
365 	return (0);
366 }
367 
368 /* pru_rcvoob is EOPNOTSUPP */
369 
370 static int
371 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
372     struct mbuf *control, struct thread *td)
373 {
374 	int error = 0;
375 	struct unpcb *unp;
376 	struct socket *so2;
377 	u_long newhiwat;
378 
379 	unp = sotounpcb(so);
380 	if (unp == NULL) {
381 		error = EINVAL;
382 		goto release;
383 	}
384 	if (flags & PRUS_OOB) {
385 		error = EOPNOTSUPP;
386 		goto release;
387 	}
388 
389 	if (control != NULL && (error = unp_internalize(&control, td)))
390 		goto release;
391 
392 	UNP_LOCK();
393 	unp = sotounpcb(so);
394 	if (unp == NULL) {
395 		UNP_UNLOCK();
396 		error = EINVAL;
397 		goto dispose_release;
398 	}
399 
400 	switch (so->so_type) {
401 	case SOCK_DGRAM:
402 	{
403 		const struct sockaddr *from;
404 
405 		if (nam != NULL) {
406 			if (unp->unp_conn != NULL) {
407 				error = EISCONN;
408 				break;
409 			}
410 			error = unp_connect(so, nam, td);
411 			if (error)
412 				break;
413 		} else {
414 			if (unp->unp_conn == NULL) {
415 				error = ENOTCONN;
416 				break;
417 			}
418 		}
419 		so2 = unp->unp_conn->unp_socket;
420 		if (unp->unp_addr != NULL)
421 			from = (struct sockaddr *)unp->unp_addr;
422 		else
423 			from = &sun_noname;
424 		SOCKBUF_LOCK(&so2->so_rcv);
425 		if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
426 			sorwakeup_locked(so2);
427 			m = NULL;
428 			control = NULL;
429 		} else {
430 			SOCKBUF_UNLOCK(&so2->so_rcv);
431 			error = ENOBUFS;
432 		}
433 		if (nam != NULL)
434 			unp_disconnect(unp);
435 		break;
436 	}
437 
438 	case SOCK_STREAM:
439 		/* Connect if not connected yet. */
440 		/*
441 		 * Note: A better implementation would complain
442 		 * if not equal to the peer's address.
443 		 */
444 		if ((so->so_state & SS_ISCONNECTED) == 0) {
445 			if (nam != NULL) {
446 				error = unp_connect(so, nam, td);
447 				if (error)
448 					break;	/* XXX */
449 			} else {
450 				error = ENOTCONN;
451 				break;
452 			}
453 		}
454 
455 		SOCKBUF_LOCK(&so->so_snd);
456 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
457 			SOCKBUF_UNLOCK(&so->so_snd);
458 			error = EPIPE;
459 			break;
460 		}
461 		if (unp->unp_conn == NULL)
462 			panic("uipc_send connected but no connection?");
463 		so2 = unp->unp_conn->unp_socket;
464 		SOCKBUF_LOCK(&so2->so_rcv);
465 		/*
466 		 * Send to paired receive port, and then reduce
467 		 * send buffer hiwater marks to maintain backpressure.
468 		 * Wake up readers.
469 		 */
470 		if (control != NULL) {
471 			if (sbappendcontrol_locked(&so2->so_rcv, m, control))
472 				control = NULL;
473 		} else {
474 			sbappend_locked(&so2->so_rcv, m);
475 		}
476 		so->so_snd.sb_mbmax -=
477 			so2->so_rcv.sb_mbcnt - unp->unp_conn->unp_mbcnt;
478 		unp->unp_conn->unp_mbcnt = so2->so_rcv.sb_mbcnt;
479 		newhiwat = so->so_snd.sb_hiwat -
480 		    (so2->so_rcv.sb_cc - unp->unp_conn->unp_cc);
481 		(void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
482 		    newhiwat, RLIM_INFINITY);
483 		SOCKBUF_UNLOCK(&so->so_snd);
484 		unp->unp_conn->unp_cc = so2->so_rcv.sb_cc;
485 		sorwakeup_locked(so2);
486 		m = NULL;
487 		break;
488 
489 	default:
490 		panic("uipc_send unknown socktype");
491 	}
492 
493 	/*
494 	 * SEND_EOF is equivalent to a SEND followed by
495 	 * a SHUTDOWN.
496 	 */
497 	if (flags & PRUS_EOF) {
498 		socantsendmore(so);
499 		unp_shutdown(unp);
500 	}
501 	UNP_UNLOCK();
502 
503 dispose_release:
504 	if (control != NULL && error != 0)
505 		unp_dispose(control);
506 
507 release:
508 	if (control != NULL)
509 		m_freem(control);
510 	if (m != NULL)
511 		m_freem(m);
512 	return (error);
513 }
514 
515 static int
516 uipc_sense(struct socket *so, struct stat *sb)
517 {
518 	struct unpcb *unp;
519 	struct socket *so2;
520 
521 	UNP_LOCK();
522 	unp = sotounpcb(so);
523 	if (unp == NULL) {
524 		UNP_UNLOCK();
525 		return (EINVAL);
526 	}
527 	sb->st_blksize = so->so_snd.sb_hiwat;
528 	if (so->so_type == SOCK_STREAM && unp->unp_conn != NULL) {
529 		so2 = unp->unp_conn->unp_socket;
530 		sb->st_blksize += so2->so_rcv.sb_cc;
531 	}
532 	sb->st_dev = NODEV;
533 	if (unp->unp_ino == 0)
534 		unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
535 	sb->st_ino = unp->unp_ino;
536 	UNP_UNLOCK();
537 	return (0);
538 }
539 
540 static int
541 uipc_shutdown(struct socket *so)
542 {
543 	struct unpcb *unp;
544 
545 	UNP_LOCK();
546 	unp = sotounpcb(so);
547 	if (unp == NULL) {
548 		UNP_UNLOCK();
549 		return (EINVAL);
550 	}
551 	socantsendmore(so);
552 	unp_shutdown(unp);
553 	UNP_UNLOCK();
554 	return (0);
555 }
556 
557 static int
558 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
559 {
560 	struct unpcb *unp;
561 	const struct sockaddr *sa;
562 
563 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
564 	UNP_LOCK();
565 	unp = sotounpcb(so);
566 	if (unp == NULL) {
567 		UNP_UNLOCK();
568 		free(*nam, M_SONAME);
569 		*nam = NULL;
570 		return (EINVAL);
571 	}
572 	if (unp->unp_addr != NULL)
573 		sa = (struct sockaddr *) unp->unp_addr;
574 	else
575 		sa = &sun_noname;
576 	bcopy(sa, *nam, sa->sa_len);
577 	UNP_UNLOCK();
578 	return (0);
579 }
580 
581 struct pr_usrreqs uipc_usrreqs = {
582 	.pru_abort = 		uipc_abort,
583 	.pru_accept =		uipc_accept,
584 	.pru_attach =		uipc_attach,
585 	.pru_bind =		uipc_bind,
586 	.pru_connect =		uipc_connect,
587 	.pru_connect2 =		uipc_connect2,
588 	.pru_detach =		uipc_detach,
589 	.pru_disconnect =	uipc_disconnect,
590 	.pru_listen =		uipc_listen,
591 	.pru_peeraddr =		uipc_peeraddr,
592 	.pru_rcvd =		uipc_rcvd,
593 	.pru_send =		uipc_send,
594 	.pru_sense =		uipc_sense,
595 	.pru_shutdown =		uipc_shutdown,
596 	.pru_sockaddr =		uipc_sockaddr,
597 	.pru_sosend =		sosend,
598 	.pru_soreceive =	soreceive,
599 	.pru_sopoll =		sopoll,
600 };
601 
602 int
603 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
604 {
605 	struct unpcb *unp;
606 	struct xucred xu;
607 	int error;
608 
609 	switch (sopt->sopt_dir) {
610 	case SOPT_GET:
611 		switch (sopt->sopt_name) {
612 		case LOCAL_PEERCRED:
613 			error = 0;
614 			UNP_LOCK();
615 			unp = sotounpcb(so);
616 			if (unp == NULL) {
617 				UNP_UNLOCK();
618 				error = EINVAL;
619 				break;
620 			}
621 			if (unp->unp_flags & UNP_HAVEPC)
622 				xu = unp->unp_peercred;
623 			else {
624 				if (so->so_type == SOCK_STREAM)
625 					error = ENOTCONN;
626 				else
627 					error = EINVAL;
628 			}
629 			UNP_UNLOCK();
630 			if (error == 0)
631 				error = sooptcopyout(sopt, &xu, sizeof(xu));
632 			break;
633 		default:
634 			error = EOPNOTSUPP;
635 			break;
636 		}
637 		break;
638 	case SOPT_SET:
639 	default:
640 		error = EOPNOTSUPP;
641 		break;
642 	}
643 	return (error);
644 }
645 
646 /*
647  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
648  * for stream sockets, although the total for sender and receiver is
649  * actually only PIPSIZ.
650  * Datagram sockets really use the sendspace as the maximum datagram size,
651  * and don't really want to reserve the sendspace.  Their recvspace should
652  * be large enough for at least one max-size datagram plus address.
653  */
654 #ifndef PIPSIZ
655 #define	PIPSIZ	8192
656 #endif
657 static u_long	unpst_sendspace = PIPSIZ;
658 static u_long	unpst_recvspace = PIPSIZ;
659 static u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
660 static u_long	unpdg_recvspace = 4*1024;
661 
662 static int	unp_rights;			/* file descriptors in flight */
663 
664 SYSCTL_DECL(_net_local_stream);
665 SYSCTL_INT(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
666 	   &unpst_sendspace, 0, "");
667 SYSCTL_INT(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
668 	   &unpst_recvspace, 0, "");
669 SYSCTL_DECL(_net_local_dgram);
670 SYSCTL_INT(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
671 	   &unpdg_sendspace, 0, "");
672 SYSCTL_INT(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
673 	   &unpdg_recvspace, 0, "");
674 SYSCTL_DECL(_net_local);
675 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
676 
677 static int
678 unp_attach(struct socket *so)
679 {
680 	struct unpcb *unp;
681 	int error;
682 
683 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
684 		switch (so->so_type) {
685 
686 		case SOCK_STREAM:
687 			error = soreserve(so, unpst_sendspace, unpst_recvspace);
688 			break;
689 
690 		case SOCK_DGRAM:
691 			error = soreserve(so, unpdg_sendspace, unpdg_recvspace);
692 			break;
693 
694 		default:
695 			panic("unp_attach");
696 		}
697 		if (error)
698 			return (error);
699 	}
700 	unp = uma_zalloc(unp_zone, M_WAITOK | M_ZERO);
701 	if (unp == NULL)
702 		return (ENOBUFS);
703 	LIST_INIT(&unp->unp_refs);
704 	unp->unp_socket = so;
705 	so->so_pcb = unp;
706 
707 	UNP_LOCK();
708 	unp->unp_gencnt = ++unp_gencnt;
709 	unp_count++;
710 	LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead
711 			 : &unp_shead, unp, unp_link);
712 	UNP_UNLOCK();
713 
714 	return (0);
715 }
716 
717 static void
718 unp_detach(struct unpcb *unp)
719 {
720 	struct vnode *vp;
721 
722 	UNP_LOCK_ASSERT();
723 
724 	LIST_REMOVE(unp, unp_link);
725 	unp->unp_gencnt = ++unp_gencnt;
726 	--unp_count;
727 	if ((vp = unp->unp_vnode) != NULL) {
728 		/*
729 		 * XXXRW: should v_socket be frobbed only while holding
730 		 * Giant?
731 		 */
732 		unp->unp_vnode->v_socket = NULL;
733 		unp->unp_vnode = NULL;
734 	}
735 	if (unp->unp_conn != NULL)
736 		unp_disconnect(unp);
737 	while (!LIST_EMPTY(&unp->unp_refs)) {
738 		struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
739 		unp_drop(ref, ECONNRESET);
740 	}
741 	soisdisconnected(unp->unp_socket);
742 	unp->unp_socket->so_pcb = NULL;
743 	if (unp_rights) {
744 		/*
745 		 * Normally the receive buffer is flushed later,
746 		 * in sofree, but if our receive buffer holds references
747 		 * to descriptors that are now garbage, we will dispose
748 		 * of those descriptor references after the garbage collector
749 		 * gets them (resulting in a "panic: closef: count < 0").
750 		 */
751 		sorflush(unp->unp_socket);
752 		unp_gc();	/* Will unlock UNP. */
753 	} else
754 		UNP_UNLOCK();
755 	UNP_UNLOCK_ASSERT();
756 	if (unp->unp_addr != NULL)
757 		FREE(unp->unp_addr, M_SONAME);
758 	uma_zfree(unp_zone, unp);
759 	if (vp) {
760 		mtx_lock(&Giant);
761 		vrele(vp);
762 		mtx_unlock(&Giant);
763 	}
764 }
765 
766 static int
767 unp_bind(struct unpcb *unp, struct sockaddr *nam, struct thread *td)
768 {
769 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
770 	struct vnode *vp;
771 	struct mount *mp;
772 	struct vattr vattr;
773 	int error, namelen;
774 	struct nameidata nd;
775 	char *buf;
776 
777 	UNP_LOCK_ASSERT();
778 
779 	/*
780 	 * XXXRW: This test-and-set of unp_vnode is non-atomic; the
781 	 * unlocked read here is fine, but the value of unp_vnode needs
782 	 * to be tested again after we do all the lookups to see if the
783 	 * pcb is still unbound?
784 	 */
785 	if (unp->unp_vnode != NULL)
786 		return (EINVAL);
787 
788 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
789 	if (namelen <= 0)
790 		return (EINVAL);
791 
792 	UNP_UNLOCK();
793 
794 	buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
795 	strlcpy(buf, soun->sun_path, namelen + 1);
796 
797 	mtx_lock(&Giant);
798 restart:
799 	mtx_assert(&Giant, MA_OWNED);
800 	NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE,
801 	    buf, td);
802 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
803 	error = namei(&nd);
804 	if (error)
805 		goto done;
806 	vp = nd.ni_vp;
807 	if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
808 		NDFREE(&nd, NDF_ONLY_PNBUF);
809 		if (nd.ni_dvp == vp)
810 			vrele(nd.ni_dvp);
811 		else
812 			vput(nd.ni_dvp);
813 		if (vp != NULL) {
814 			vrele(vp);
815 			error = EADDRINUSE;
816 			goto done;
817 		}
818 		error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
819 		if (error)
820 			goto done;
821 		goto restart;
822 	}
823 	VATTR_NULL(&vattr);
824 	vattr.va_type = VSOCK;
825 	vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
826 #ifdef MAC
827 	error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
828 	    &vattr);
829 #endif
830 	if (error == 0) {
831 		VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
832 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
833 	}
834 	NDFREE(&nd, NDF_ONLY_PNBUF);
835 	vput(nd.ni_dvp);
836 	if (error) {
837 		vn_finished_write(mp);
838 		goto done;
839 	}
840 	vp = nd.ni_vp;
841 	ASSERT_VOP_LOCKED(vp, "unp_bind");
842 	soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
843 	UNP_LOCK();
844 	vp->v_socket = unp->unp_socket;
845 	unp->unp_vnode = vp;
846 	unp->unp_addr = soun;
847 	UNP_UNLOCK();
848 	VOP_UNLOCK(vp, 0, td);
849 	vn_finished_write(mp);
850 done:
851 	mtx_unlock(&Giant);
852 	free(buf, M_TEMP);
853 	UNP_LOCK();
854 	return (error);
855 }
856 
857 static int
858 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
859 {
860 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
861 	struct vnode *vp;
862 	struct socket *so2, *so3;
863 	struct unpcb *unp, *unp2, *unp3;
864 	int error, len;
865 	struct nameidata nd;
866 	char buf[SOCK_MAXADDRLEN];
867 	struct sockaddr *sa;
868 
869 	UNP_LOCK_ASSERT();
870 	unp = sotounpcb(so);
871 
872 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
873 	if (len <= 0)
874 		return (EINVAL);
875 	strlcpy(buf, soun->sun_path, len + 1);
876 	UNP_UNLOCK();
877 	sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
878 	mtx_lock(&Giant);
879 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td);
880 	error = namei(&nd);
881 	if (error)
882 		vp = NULL;
883 	else
884 		vp = nd.ni_vp;
885 	ASSERT_VOP_LOCKED(vp, "unp_connect");
886 	NDFREE(&nd, NDF_ONLY_PNBUF);
887 	if (error)
888 		goto bad;
889 
890 	if (vp->v_type != VSOCK) {
891 		error = ENOTSOCK;
892 		goto bad;
893 	}
894 	error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
895 	if (error)
896 		goto bad;
897 	mtx_unlock(&Giant);
898 	UNP_LOCK();
899 	unp = sotounpcb(so);
900 	if (unp == NULL) {
901 		error = EINVAL;
902 		goto bad2;
903 	}
904 	so2 = vp->v_socket;
905 	if (so2 == NULL) {
906 		error = ECONNREFUSED;
907 		goto bad2;
908 	}
909 	if (so->so_type != so2->so_type) {
910 		error = EPROTOTYPE;
911 		goto bad2;
912 	}
913 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
914 		if (so2->so_options & SO_ACCEPTCONN) {
915 			/*
916 			 * NB: drop locks here so unp_attach is entered
917 			 *     w/o locks; this avoids a recursive lock
918 			 *     of the head and holding sleep locks across
919 			 *     a (potentially) blocking malloc.
920 			 */
921 			UNP_UNLOCK();
922 			so3 = sonewconn(so2, 0);
923 			UNP_LOCK();
924 		} else
925 			so3 = NULL;
926 		if (so3 == NULL) {
927 			error = ECONNREFUSED;
928 			goto bad2;
929 		}
930 		unp = sotounpcb(so);
931 		unp2 = sotounpcb(so2);
932 		unp3 = sotounpcb(so3);
933 		if (unp2->unp_addr != NULL) {
934 			bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
935 			unp3->unp_addr = (struct sockaddr_un *) sa;
936 			sa = NULL;
937 		}
938 		/*
939 		 * unp_peercred management:
940 		 *
941 		 * The connecter's (client's) credentials are copied
942 		 * from its process structure at the time of connect()
943 		 * (which is now).
944 		 */
945 		cru2x(td->td_ucred, &unp3->unp_peercred);
946 		unp3->unp_flags |= UNP_HAVEPC;
947 		/*
948 		 * The receiver's (server's) credentials are copied
949 		 * from the unp_peercred member of socket on which the
950 		 * former called listen(); unp_listen() cached that
951 		 * process's credentials at that time so we can use
952 		 * them now.
953 		 */
954 		KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
955 		    ("unp_connect: listener without cached peercred"));
956 		memcpy(&unp->unp_peercred, &unp2->unp_peercred,
957 		    sizeof(unp->unp_peercred));
958 		unp->unp_flags |= UNP_HAVEPC;
959 #ifdef MAC
960 		SOCK_LOCK(so);
961 		mac_set_socket_peer_from_socket(so, so3);
962 		mac_set_socket_peer_from_socket(so3, so);
963 		SOCK_UNLOCK(so);
964 #endif
965 
966 		so2 = so3;
967 	}
968 	error = unp_connect2(so, so2);
969 bad2:
970 	UNP_UNLOCK();
971 	mtx_lock(&Giant);
972 bad:
973 	mtx_assert(&Giant, MA_OWNED);
974 	if (vp != NULL)
975 		vput(vp);
976 	mtx_unlock(&Giant);
977 	free(sa, M_SONAME);
978 	UNP_LOCK();
979 	return (error);
980 }
981 
982 static int
983 unp_connect2(struct socket *so, struct socket *so2)
984 {
985 	struct unpcb *unp = sotounpcb(so);
986 	struct unpcb *unp2;
987 
988 	UNP_LOCK_ASSERT();
989 
990 	if (so2->so_type != so->so_type)
991 		return (EPROTOTYPE);
992 	unp2 = sotounpcb(so2);
993 	unp->unp_conn = unp2;
994 	switch (so->so_type) {
995 
996 	case SOCK_DGRAM:
997 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
998 		soisconnected(so);
999 		break;
1000 
1001 	case SOCK_STREAM:
1002 		unp2->unp_conn = unp;
1003 		soisconnected(so);
1004 		soisconnected(so2);
1005 		break;
1006 
1007 	default:
1008 		panic("unp_connect2");
1009 	}
1010 	return (0);
1011 }
1012 
1013 static void
1014 unp_disconnect(struct unpcb *unp)
1015 {
1016 	struct unpcb *unp2 = unp->unp_conn;
1017 	struct socket *so;
1018 
1019 	UNP_LOCK_ASSERT();
1020 
1021 	if (unp2 == NULL)
1022 		return;
1023 	unp->unp_conn = NULL;
1024 	switch (unp->unp_socket->so_type) {
1025 
1026 	case SOCK_DGRAM:
1027 		LIST_REMOVE(unp, unp_reflink);
1028 		so = unp->unp_socket;
1029 		SOCK_LOCK(so);
1030 		so->so_state &= ~SS_ISCONNECTED;
1031 		SOCK_UNLOCK(so);
1032 		break;
1033 
1034 	case SOCK_STREAM:
1035 		soisdisconnected(unp->unp_socket);
1036 		unp2->unp_conn = NULL;
1037 		soisdisconnected(unp2->unp_socket);
1038 		break;
1039 	}
1040 }
1041 
1042 #ifdef notdef
1043 void
1044 unp_abort(struct unpcb *unp)
1045 {
1046 
1047 	unp_detach(unp);
1048 	UNP_UNLOCK_ASSERT();
1049 }
1050 #endif
1051 
1052 /*
1053  * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed
1054  * by the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers
1055  * are safe to reference.  It first scans the list of struct unpcb's to
1056  * generate a pointer list, then it rescans its list one entry at a time to
1057  * externalize and copyout.  It checks the generation number to see if a
1058  * struct unpcb has been reused, and will skip it if so.
1059  */
1060 static int
1061 unp_pcblist(SYSCTL_HANDLER_ARGS)
1062 {
1063 	int error, i, n;
1064 	struct unpcb *unp, **unp_list;
1065 	unp_gen_t gencnt;
1066 	struct xunpgen *xug;
1067 	struct unp_head *head;
1068 	struct xunpcb *xu;
1069 
1070 	head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1071 
1072 	/*
1073 	 * The process of preparing the PCB list is too time-consuming and
1074 	 * resource-intensive to repeat twice on every request.
1075 	 */
1076 	if (req->oldptr == NULL) {
1077 		n = unp_count;
1078 		req->oldidx = 2 * (sizeof *xug)
1079 			+ (n + n/8) * sizeof(struct xunpcb);
1080 		return (0);
1081 	}
1082 
1083 	if (req->newptr != NULL)
1084 		return (EPERM);
1085 
1086 	/*
1087 	 * OK, now we're committed to doing something.
1088 	 */
1089 	xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1090 	UNP_LOCK();
1091 	gencnt = unp_gencnt;
1092 	n = unp_count;
1093 	UNP_UNLOCK();
1094 
1095 	xug->xug_len = sizeof *xug;
1096 	xug->xug_count = n;
1097 	xug->xug_gen = gencnt;
1098 	xug->xug_sogen = so_gencnt;
1099 	error = SYSCTL_OUT(req, xug, sizeof *xug);
1100 	if (error) {
1101 		free(xug, M_TEMP);
1102 		return (error);
1103 	}
1104 
1105 	unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1106 
1107 	UNP_LOCK();
1108 	for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1109 	     unp = LIST_NEXT(unp, unp_link)) {
1110 		if (unp->unp_gencnt <= gencnt) {
1111 			if (cr_cansee(req->td->td_ucred,
1112 			    unp->unp_socket->so_cred))
1113 				continue;
1114 			unp_list[i++] = unp;
1115 		}
1116 	}
1117 	UNP_UNLOCK();
1118 	n = i;			/* in case we lost some during malloc */
1119 
1120 	error = 0;
1121 	xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK);
1122 	for (i = 0; i < n; i++) {
1123 		unp = unp_list[i];
1124 		if (unp->unp_gencnt <= gencnt) {
1125 			xu->xu_len = sizeof *xu;
1126 			xu->xu_unpp = unp;
1127 			/*
1128 			 * XXX - need more locking here to protect against
1129 			 * connect/disconnect races for SMP.
1130 			 */
1131 			if (unp->unp_addr != NULL)
1132 				bcopy(unp->unp_addr, &xu->xu_addr,
1133 				      unp->unp_addr->sun_len);
1134 			if (unp->unp_conn != NULL &&
1135 			    unp->unp_conn->unp_addr != NULL)
1136 				bcopy(unp->unp_conn->unp_addr,
1137 				      &xu->xu_caddr,
1138 				      unp->unp_conn->unp_addr->sun_len);
1139 			bcopy(unp, &xu->xu_unp, sizeof *unp);
1140 			sotoxsocket(unp->unp_socket, &xu->xu_socket);
1141 			error = SYSCTL_OUT(req, xu, sizeof *xu);
1142 		}
1143 	}
1144 	free(xu, M_TEMP);
1145 	if (!error) {
1146 		/*
1147 		 * Give the user an updated idea of our state.
1148 		 * If the generation differs from what we told
1149 		 * her before, she knows that something happened
1150 		 * while we were processing this request, and it
1151 		 * might be necessary to retry.
1152 		 */
1153 		xug->xug_gen = unp_gencnt;
1154 		xug->xug_sogen = so_gencnt;
1155 		xug->xug_count = unp_count;
1156 		error = SYSCTL_OUT(req, xug, sizeof *xug);
1157 	}
1158 	free(unp_list, M_TEMP);
1159 	free(xug, M_TEMP);
1160 	return (error);
1161 }
1162 
1163 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1164 	    (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1165 	    "List of active local datagram sockets");
1166 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1167 	    (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1168 	    "List of active local stream sockets");
1169 
1170 static void
1171 unp_shutdown(struct unpcb *unp)
1172 {
1173 	struct socket *so;
1174 
1175 	UNP_LOCK_ASSERT();
1176 
1177 	if (unp->unp_socket->so_type == SOCK_STREAM && unp->unp_conn &&
1178 	    (so = unp->unp_conn->unp_socket))
1179 		socantrcvmore(so);
1180 }
1181 
1182 static void
1183 unp_drop(struct unpcb *unp, int errno)
1184 {
1185 	struct socket *so = unp->unp_socket;
1186 
1187 	UNP_LOCK_ASSERT();
1188 
1189 	so->so_error = errno;
1190 	unp_disconnect(unp);
1191 }
1192 
1193 #ifdef notdef
1194 void
1195 unp_drain(void)
1196 {
1197 
1198 }
1199 #endif
1200 
1201 static void
1202 unp_freerights(struct file **rp, int fdcount)
1203 {
1204 	int i;
1205 	struct file *fp;
1206 
1207 	for (i = 0; i < fdcount; i++) {
1208 		fp = *rp;
1209 		/*
1210 		 * zero the pointer before calling
1211 		 * unp_discard since it may end up
1212 		 * in unp_gc()..
1213 		 */
1214 		*rp++ = 0;
1215 		unp_discard(fp);
1216 	}
1217 }
1218 
1219 int
1220 unp_externalize(struct mbuf *control, struct mbuf **controlp)
1221 {
1222 	struct thread *td = curthread;		/* XXX */
1223 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1224 	int i;
1225 	int *fdp;
1226 	struct file **rp;
1227 	struct file *fp;
1228 	void *data;
1229 	socklen_t clen = control->m_len, datalen;
1230 	int error, newfds;
1231 	int f;
1232 	u_int newlen;
1233 
1234 	UNP_UNLOCK_ASSERT();
1235 
1236 	error = 0;
1237 	if (controlp != NULL) /* controlp == NULL => free control messages */
1238 		*controlp = NULL;
1239 
1240 	while (cm != NULL) {
1241 		if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1242 			error = EINVAL;
1243 			break;
1244 		}
1245 
1246 		data = CMSG_DATA(cm);
1247 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1248 
1249 		if (cm->cmsg_level == SOL_SOCKET
1250 		    && cm->cmsg_type == SCM_RIGHTS) {
1251 			newfds = datalen / sizeof(struct file *);
1252 			rp = data;
1253 
1254 			/* If we're not outputting the descriptors free them. */
1255 			if (error || controlp == NULL) {
1256 				unp_freerights(rp, newfds);
1257 				goto next;
1258 			}
1259 			FILEDESC_LOCK(td->td_proc->p_fd);
1260 			/* if the new FD's will not fit free them.  */
1261 			if (!fdavail(td, newfds)) {
1262 				FILEDESC_UNLOCK(td->td_proc->p_fd);
1263 				error = EMSGSIZE;
1264 				unp_freerights(rp, newfds);
1265 				goto next;
1266 			}
1267 			/*
1268 			 * now change each pointer to an fd in the global
1269 			 * table to an integer that is the index to the
1270 			 * local fd table entry that we set up to point
1271 			 * to the global one we are transferring.
1272 			 */
1273 			newlen = newfds * sizeof(int);
1274 			*controlp = sbcreatecontrol(NULL, newlen,
1275 			    SCM_RIGHTS, SOL_SOCKET);
1276 			if (*controlp == NULL) {
1277 				FILEDESC_UNLOCK(td->td_proc->p_fd);
1278 				error = E2BIG;
1279 				unp_freerights(rp, newfds);
1280 				goto next;
1281 			}
1282 
1283 			fdp = (int *)
1284 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1285 			for (i = 0; i < newfds; i++) {
1286 				if (fdalloc(td, 0, &f))
1287 					panic("unp_externalize fdalloc failed");
1288 				fp = *rp++;
1289 				td->td_proc->p_fd->fd_ofiles[f] = fp;
1290 				FILE_LOCK(fp);
1291 				fp->f_msgcount--;
1292 				FILE_UNLOCK(fp);
1293 				unp_rights--;
1294 				*fdp++ = f;
1295 			}
1296 			FILEDESC_UNLOCK(td->td_proc->p_fd);
1297 		} else { /* We can just copy anything else across */
1298 			if (error || controlp == NULL)
1299 				goto next;
1300 			*controlp = sbcreatecontrol(NULL, datalen,
1301 			    cm->cmsg_type, cm->cmsg_level);
1302 			if (*controlp == NULL) {
1303 				error = ENOBUFS;
1304 				goto next;
1305 			}
1306 			bcopy(data,
1307 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1308 			    datalen);
1309 		}
1310 
1311 		controlp = &(*controlp)->m_next;
1312 
1313 next:
1314 		if (CMSG_SPACE(datalen) < clen) {
1315 			clen -= CMSG_SPACE(datalen);
1316 			cm = (struct cmsghdr *)
1317 			    ((caddr_t)cm + CMSG_SPACE(datalen));
1318 		} else {
1319 			clen = 0;
1320 			cm = NULL;
1321 		}
1322 	}
1323 
1324 	m_freem(control);
1325 
1326 	return (error);
1327 }
1328 
1329 void
1330 unp_init(void)
1331 {
1332 	unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1333 	    NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
1334 	if (unp_zone == NULL)
1335 		panic("unp_init");
1336 	uma_zone_set_max(unp_zone, nmbclusters);
1337 	LIST_INIT(&unp_dhead);
1338 	LIST_INIT(&unp_shead);
1339 
1340 	UNP_LOCK_INIT();
1341 }
1342 
1343 static int
1344 unp_internalize(struct mbuf **controlp, struct thread *td)
1345 {
1346 	struct mbuf *control = *controlp;
1347 	struct proc *p = td->td_proc;
1348 	struct filedesc *fdescp = p->p_fd;
1349 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1350 	struct cmsgcred *cmcred;
1351 	struct file **rp;
1352 	struct file *fp;
1353 	struct timeval *tv;
1354 	int i, fd, *fdp;
1355 	void *data;
1356 	socklen_t clen = control->m_len, datalen;
1357 	int error, oldfds;
1358 	u_int newlen;
1359 
1360 	UNP_UNLOCK_ASSERT();
1361 
1362 	error = 0;
1363 	*controlp = NULL;
1364 
1365 	while (cm != NULL) {
1366 		if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1367 		    || cm->cmsg_len > clen) {
1368 			error = EINVAL;
1369 			goto out;
1370 		}
1371 
1372 		data = CMSG_DATA(cm);
1373 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1374 
1375 		switch (cm->cmsg_type) {
1376 		/*
1377 		 * Fill in credential information.
1378 		 */
1379 		case SCM_CREDS:
1380 			*controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1381 			    SCM_CREDS, SOL_SOCKET);
1382 			if (*controlp == NULL) {
1383 				error = ENOBUFS;
1384 				goto out;
1385 			}
1386 
1387 			cmcred = (struct cmsgcred *)
1388 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1389 			cmcred->cmcred_pid = p->p_pid;
1390 			cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1391 			cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1392 			cmcred->cmcred_euid = td->td_ucred->cr_uid;
1393 			cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1394 							CMGROUP_MAX);
1395 			for (i = 0; i < cmcred->cmcred_ngroups; i++)
1396 				cmcred->cmcred_groups[i] =
1397 				    td->td_ucred->cr_groups[i];
1398 			break;
1399 
1400 		case SCM_RIGHTS:
1401 			oldfds = datalen / sizeof (int);
1402 			/*
1403 			 * check that all the FDs passed in refer to legal files
1404 			 * If not, reject the entire operation.
1405 			 */
1406 			fdp = data;
1407 			FILEDESC_LOCK(fdescp);
1408 			for (i = 0; i < oldfds; i++) {
1409 				fd = *fdp++;
1410 				if ((unsigned)fd >= fdescp->fd_nfiles ||
1411 				    fdescp->fd_ofiles[fd] == NULL) {
1412 					FILEDESC_UNLOCK(fdescp);
1413 					error = EBADF;
1414 					goto out;
1415 				}
1416 				fp = fdescp->fd_ofiles[fd];
1417 				if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1418 					FILEDESC_UNLOCK(fdescp);
1419 					error = EOPNOTSUPP;
1420 					goto out;
1421 				}
1422 
1423 			}
1424 			/*
1425 			 * Now replace the integer FDs with pointers to
1426 			 * the associated global file table entry..
1427 			 */
1428 			newlen = oldfds * sizeof(struct file *);
1429 			*controlp = sbcreatecontrol(NULL, newlen,
1430 			    SCM_RIGHTS, SOL_SOCKET);
1431 			if (*controlp == NULL) {
1432 				FILEDESC_UNLOCK(fdescp);
1433 				error = E2BIG;
1434 				goto out;
1435 			}
1436 
1437 			fdp = data;
1438 			rp = (struct file **)
1439 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1440 			for (i = 0; i < oldfds; i++) {
1441 				fp = fdescp->fd_ofiles[*fdp++];
1442 				*rp++ = fp;
1443 				FILE_LOCK(fp);
1444 				fp->f_count++;
1445 				fp->f_msgcount++;
1446 				FILE_UNLOCK(fp);
1447 				unp_rights++;
1448 			}
1449 			FILEDESC_UNLOCK(fdescp);
1450 			break;
1451 
1452 		case SCM_TIMESTAMP:
1453 			*controlp = sbcreatecontrol(NULL, sizeof(*tv),
1454 			    SCM_TIMESTAMP, SOL_SOCKET);
1455 			if (*controlp == NULL) {
1456 				error = ENOBUFS;
1457 				goto out;
1458 			}
1459 			tv = (struct timeval *)
1460 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1461 			microtime(tv);
1462 			break;
1463 
1464 		default:
1465 			error = EINVAL;
1466 			goto out;
1467 		}
1468 
1469 		controlp = &(*controlp)->m_next;
1470 
1471 		if (CMSG_SPACE(datalen) < clen) {
1472 			clen -= CMSG_SPACE(datalen);
1473 			cm = (struct cmsghdr *)
1474 			    ((caddr_t)cm + CMSG_SPACE(datalen));
1475 		} else {
1476 			clen = 0;
1477 			cm = NULL;
1478 		}
1479 	}
1480 
1481 out:
1482 	m_freem(control);
1483 
1484 	return (error);
1485 }
1486 
1487 /*
1488  * unp_defer is thread-local during garbage collection, and does not require
1489  * explicit synchronization.  unp_gcing prevents other threads from entering
1490  * garbage collection, and perhaps should be an sx lock instead.
1491  */
1492 static int	unp_defer, unp_gcing;
1493 
1494 static void
1495 unp_gc(void)
1496 {
1497 	struct file *fp, *nextfp;
1498 	struct socket *so;
1499 	struct file **extra_ref, **fpp;
1500 	int nunref, i;
1501 	int nfiles_snap;
1502 	int nfiles_slack = 20;
1503 
1504 	UNP_LOCK_ASSERT();
1505 
1506 	if (unp_gcing) {
1507 		UNP_UNLOCK();
1508 		return;
1509 	}
1510 	unp_gcing = 1;
1511 	unp_defer = 0;
1512 	UNP_UNLOCK();
1513 	/*
1514 	 * before going through all this, set all FDs to
1515 	 * be NOT defered and NOT externally accessible
1516 	 */
1517 	sx_slock(&filelist_lock);
1518 	LIST_FOREACH(fp, &filehead, f_list)
1519 		fp->f_gcflag &= ~(FMARK|FDEFER);
1520 	do {
1521 		LIST_FOREACH(fp, &filehead, f_list) {
1522 			FILE_LOCK(fp);
1523 			/*
1524 			 * If the file is not open, skip it
1525 			 */
1526 			if (fp->f_count == 0) {
1527 				FILE_UNLOCK(fp);
1528 				continue;
1529 			}
1530 			/*
1531 			 * If we already marked it as 'defer'  in a
1532 			 * previous pass, then try process it this time
1533 			 * and un-mark it
1534 			 */
1535 			if (fp->f_gcflag & FDEFER) {
1536 				fp->f_gcflag &= ~FDEFER;
1537 				unp_defer--;
1538 			} else {
1539 				/*
1540 				 * if it's not defered, then check if it's
1541 				 * already marked.. if so skip it
1542 				 */
1543 				if (fp->f_gcflag & FMARK) {
1544 					FILE_UNLOCK(fp);
1545 					continue;
1546 				}
1547 				/*
1548 				 * If all references are from messages
1549 				 * in transit, then skip it. it's not
1550 				 * externally accessible.
1551 				 */
1552 				if (fp->f_count == fp->f_msgcount) {
1553 					FILE_UNLOCK(fp);
1554 					continue;
1555 				}
1556 				/*
1557 				 * If it got this far then it must be
1558 				 * externally accessible.
1559 				 */
1560 				fp->f_gcflag |= FMARK;
1561 			}
1562 			/*
1563 			 * either it was defered, or it is externally
1564 			 * accessible and not already marked so.
1565 			 * Now check if it is possibly one of OUR sockets.
1566 			 */
1567 			if (fp->f_type != DTYPE_SOCKET ||
1568 			    (so = fp->f_data) == NULL) {
1569 				FILE_UNLOCK(fp);
1570 				continue;
1571 			}
1572 			FILE_UNLOCK(fp);
1573 			if (so->so_proto->pr_domain != &localdomain ||
1574 			    (so->so_proto->pr_flags&PR_RIGHTS) == 0)
1575 				continue;
1576 #ifdef notdef
1577 			if (so->so_rcv.sb_flags & SB_LOCK) {
1578 				/*
1579 				 * This is problematical; it's not clear
1580 				 * we need to wait for the sockbuf to be
1581 				 * unlocked (on a uniprocessor, at least),
1582 				 * and it's also not clear what to do
1583 				 * if sbwait returns an error due to receipt
1584 				 * of a signal.  If sbwait does return
1585 				 * an error, we'll go into an infinite
1586 				 * loop.  Delete all of this for now.
1587 				 */
1588 				(void) sbwait(&so->so_rcv);
1589 				goto restart;
1590 			}
1591 #endif
1592 			/*
1593 			 * So, Ok, it's one of our sockets and it IS externally
1594 			 * accessible (or was defered). Now we look
1595 			 * to see if we hold any file descriptors in its
1596 			 * message buffers. Follow those links and mark them
1597 			 * as accessible too.
1598 			 */
1599 			SOCKBUF_LOCK(&so->so_rcv);
1600 			unp_scan(so->so_rcv.sb_mb, unp_mark);
1601 			SOCKBUF_UNLOCK(&so->so_rcv);
1602 		}
1603 	} while (unp_defer);
1604 	sx_sunlock(&filelist_lock);
1605 	/*
1606 	 * We grab an extra reference to each of the file table entries
1607 	 * that are not otherwise accessible and then free the rights
1608 	 * that are stored in messages on them.
1609 	 *
1610 	 * The bug in the orginal code is a little tricky, so I'll describe
1611 	 * what's wrong with it here.
1612 	 *
1613 	 * It is incorrect to simply unp_discard each entry for f_msgcount
1614 	 * times -- consider the case of sockets A and B that contain
1615 	 * references to each other.  On a last close of some other socket,
1616 	 * we trigger a gc since the number of outstanding rights (unp_rights)
1617 	 * is non-zero.  If during the sweep phase the gc code un_discards,
1618 	 * we end up doing a (full) closef on the descriptor.  A closef on A
1619 	 * results in the following chain.  Closef calls soo_close, which
1620 	 * calls soclose.   Soclose calls first (through the switch
1621 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
1622 	 * returns because the previous instance had set unp_gcing, and
1623 	 * we return all the way back to soclose, which marks the socket
1624 	 * with SS_NOFDREF, and then calls sofree.  Sofree calls sorflush
1625 	 * to free up the rights that are queued in messages on the socket A,
1626 	 * i.e., the reference on B.  The sorflush calls via the dom_dispose
1627 	 * switch unp_dispose, which unp_scans with unp_discard.  This second
1628 	 * instance of unp_discard just calls closef on B.
1629 	 *
1630 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
1631 	 * which results in another closef on A.  Unfortunately, A is already
1632 	 * being closed, and the descriptor has already been marked with
1633 	 * SS_NOFDREF, and soclose panics at this point.
1634 	 *
1635 	 * Here, we first take an extra reference to each inaccessible
1636 	 * descriptor.  Then, we call sorflush ourself, since we know
1637 	 * it is a Unix domain socket anyhow.  After we destroy all the
1638 	 * rights carried in messages, we do a last closef to get rid
1639 	 * of our extra reference.  This is the last close, and the
1640 	 * unp_detach etc will shut down the socket.
1641 	 *
1642 	 * 91/09/19, bsy@cs.cmu.edu
1643 	 */
1644 again:
1645 	nfiles_snap = openfiles + nfiles_slack;	/* some slack */
1646 	extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP,
1647 	    M_WAITOK);
1648 	sx_slock(&filelist_lock);
1649 	if (nfiles_snap < openfiles) {
1650 		sx_sunlock(&filelist_lock);
1651 		free(extra_ref, M_TEMP);
1652 		nfiles_slack += 20;
1653 		goto again;
1654 	}
1655 	for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref;
1656 	    fp != NULL; fp = nextfp) {
1657 		nextfp = LIST_NEXT(fp, f_list);
1658 		FILE_LOCK(fp);
1659 		/*
1660 		 * If it's not open, skip it
1661 		 */
1662 		if (fp->f_count == 0) {
1663 			FILE_UNLOCK(fp);
1664 			continue;
1665 		}
1666 		/*
1667 		 * If all refs are from msgs, and it's not marked accessible
1668 		 * then it must be referenced from some unreachable cycle
1669 		 * of (shut-down) FDs, so include it in our
1670 		 * list of FDs to remove
1671 		 */
1672 		if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) {
1673 			*fpp++ = fp;
1674 			nunref++;
1675 			fp->f_count++;
1676 		}
1677 		FILE_UNLOCK(fp);
1678 	}
1679 	sx_sunlock(&filelist_lock);
1680 	/*
1681 	 * for each FD on our hit list, do the following two things
1682 	 */
1683 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
1684 		struct file *tfp = *fpp;
1685 		FILE_LOCK(tfp);
1686 		if (tfp->f_type == DTYPE_SOCKET &&
1687 		    tfp->f_data != NULL) {
1688 			FILE_UNLOCK(tfp);
1689 			sorflush(tfp->f_data);
1690 		} else {
1691 			FILE_UNLOCK(tfp);
1692 		}
1693 	}
1694 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp)
1695 		closef(*fpp, (struct thread *) NULL);
1696 	free(extra_ref, M_TEMP);
1697 	unp_gcing = 0;
1698 
1699 	UNP_UNLOCK_ASSERT();
1700 }
1701 
1702 void
1703 unp_dispose(struct mbuf *m)
1704 {
1705 
1706 	if (m)
1707 		unp_scan(m, unp_discard);
1708 }
1709 
1710 static int
1711 unp_listen(struct socket *so, struct unpcb *unp, struct thread *td)
1712 {
1713 	int error;
1714 
1715 	UNP_LOCK_ASSERT();
1716 
1717 	SOCK_LOCK(so);
1718 	error = solisten_proto_check(so);
1719 	if (error == 0) {
1720 		cru2x(td->td_ucred, &unp->unp_peercred);
1721 		unp->unp_flags |= UNP_HAVEPCCACHED;
1722 		solisten_proto(so);
1723 	}
1724 	SOCK_UNLOCK(so);
1725 	return (error);
1726 }
1727 
1728 static void
1729 unp_scan(struct mbuf *m0, void (*op)(struct file *))
1730 {
1731 	struct mbuf *m;
1732 	struct file **rp;
1733 	struct cmsghdr *cm;
1734 	void *data;
1735 	int i;
1736 	socklen_t clen, datalen;
1737 	int qfds;
1738 
1739 	while (m0 != NULL) {
1740 		for (m = m0; m; m = m->m_next) {
1741 			if (m->m_type != MT_CONTROL)
1742 				continue;
1743 
1744 			cm = mtod(m, struct cmsghdr *);
1745 			clen = m->m_len;
1746 
1747 			while (cm != NULL) {
1748 				if (sizeof(*cm) > clen || cm->cmsg_len > clen)
1749 					break;
1750 
1751 				data = CMSG_DATA(cm);
1752 				datalen = (caddr_t)cm + cm->cmsg_len
1753 				    - (caddr_t)data;
1754 
1755 				if (cm->cmsg_level == SOL_SOCKET &&
1756 				    cm->cmsg_type == SCM_RIGHTS) {
1757 					qfds = datalen / sizeof (struct file *);
1758 					rp = data;
1759 					for (i = 0; i < qfds; i++)
1760 						(*op)(*rp++);
1761 				}
1762 
1763 				if (CMSG_SPACE(datalen) < clen) {
1764 					clen -= CMSG_SPACE(datalen);
1765 					cm = (struct cmsghdr *)
1766 					    ((caddr_t)cm + CMSG_SPACE(datalen));
1767 				} else {
1768 					clen = 0;
1769 					cm = NULL;
1770 				}
1771 			}
1772 		}
1773 		m0 = m0->m_act;
1774 	}
1775 }
1776 
1777 static void
1778 unp_mark(struct file *fp)
1779 {
1780 	if (fp->f_gcflag & FMARK)
1781 		return;
1782 	unp_defer++;
1783 	fp->f_gcflag |= (FMARK|FDEFER);
1784 }
1785 
1786 static void
1787 unp_discard(struct file *fp)
1788 {
1789 	FILE_LOCK(fp);
1790 	fp->f_msgcount--;
1791 	unp_rights--;
1792 	FILE_UNLOCK(fp);
1793 	(void) closef(fp, (struct thread *)NULL);
1794 }
1795