xref: /netbsd-src/sys/kern/uipc_usrreq.c (revision 62bee434bcfed81d36d05693bf87fe350107864a)
1 /*	$NetBSD: uipc_usrreq.c,v 1.207 2024/12/07 02:31:14 riastradh Exp $	*/
2 
3 /*-
4  * Copyright (c) 1998, 2000, 2004, 2008, 2009, 2020 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
9  * NASA Ames Research Center, and by Andrew Doran.
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  * 1. Redistributions of source code must retain the above copyright
15  *    notice, this list of conditions and the following disclaimer.
16  * 2. Redistributions in binary form must reproduce the above copyright
17  *    notice, this list of conditions and the following disclaimer in the
18  *    documentation and/or other materials provided with the distribution.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
21  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
22  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
23  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
24  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
25  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
26  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
27  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
28  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
29  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
30  * POSSIBILITY OF SUCH DAMAGE.
31  */
32 
33 /*
34  * Copyright (c) 1982, 1986, 1989, 1991, 1993
35  *	The Regents of the University of California.  All rights reserved.
36  *
37  * Redistribution and use in source and binary forms, with or without
38  * modification, are permitted provided that the following conditions
39  * are met:
40  * 1. Redistributions of source code must retain the above copyright
41  *    notice, this list of conditions and the following disclaimer.
42  * 2. Redistributions in binary form must reproduce the above copyright
43  *    notice, this list of conditions and the following disclaimer in the
44  *    documentation and/or other materials provided with the distribution.
45  * 3. Neither the name of the University nor the names of its contributors
46  *    may be used to endorse or promote products derived from this software
47  *    without specific prior written permission.
48  *
49  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
50  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
51  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
52  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
53  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
54  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
55  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
56  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
57  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
58  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
59  * SUCH DAMAGE.
60  *
61  *	@(#)uipc_usrreq.c	8.9 (Berkeley) 5/14/95
62  */
63 
64 /*
65  * Copyright (c) 1997 Christopher G. Demetriou.  All rights reserved.
66  *
67  * Redistribution and use in source and binary forms, with or without
68  * modification, are permitted provided that the following conditions
69  * are met:
70  * 1. Redistributions of source code must retain the above copyright
71  *    notice, this list of conditions and the following disclaimer.
72  * 2. Redistributions in binary form must reproduce the above copyright
73  *    notice, this list of conditions and the following disclaimer in the
74  *    documentation and/or other materials provided with the distribution.
75  * 3. All advertising materials mentioning features or use of this software
76  *    must display the following acknowledgement:
77  *	This product includes software developed by the University of
78  *	California, Berkeley and its contributors.
79  * 4. Neither the name of the University nor the names of its contributors
80  *    may be used to endorse or promote products derived from this software
81  *    without specific prior written permission.
82  *
83  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
84  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
85  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
86  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
87  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
88  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
89  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
90  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
91  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
92  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
93  * SUCH DAMAGE.
94  *
95  *	@(#)uipc_usrreq.c	8.9 (Berkeley) 5/14/95
96  */
97 
98 #include <sys/cdefs.h>
99 __KERNEL_RCSID(0, "$NetBSD: uipc_usrreq.c,v 1.207 2024/12/07 02:31:14 riastradh Exp $");
100 
101 #ifdef _KERNEL_OPT
102 #include "opt_compat_netbsd.h"
103 #endif
104 
105 #include <sys/param.h>
106 #include <sys/types.h>
107 
108 #include <sys/atomic.h>
109 #include <sys/compat_stub.h>
110 #include <sys/domain.h>
111 #include <sys/file.h>
112 #include <sys/filedesc.h>
113 #include <sys/kauth.h>
114 #include <sys/kernel.h>
115 #include <sys/kmem.h>
116 #include <sys/kthread.h>
117 #include <sys/mbuf.h>
118 #include <sys/namei.h>
119 #include <sys/proc.h>
120 #include <sys/protosw.h>
121 #include <sys/sdt.h>
122 #include <sys/socket.h>
123 #include <sys/socketvar.h>
124 #include <sys/stat.h>
125 #include <sys/systm.h>
126 #include <sys/uidinfo.h>
127 #include <sys/un.h>
128 #include <sys/unpcb.h>
129 #include <sys/vnode.h>
130 
131 #include <compat/net/route_70.h>
132 #include <compat/sys/socket.h>
133 
134 /*
135  * Unix communications domain.
136  *
137  * TODO:
138  *	RDM
139  *	rethink name space problems
140  *	need a proper out-of-band
141  *
142  * Notes on locking:
143  *
144  * The generic rules noted in uipc_socket2.c apply.  In addition:
145  *
146  * o We have a global lock, uipc_lock.
147  *
148  * o All datagram sockets are locked by uipc_lock.
149  *
150  * o For stream socketpairs, the two endpoints are created sharing the same
151  *   independent lock.  Sockets presented to PRU_CONNECT2 must already have
152  *   matching locks.
153  *
154  * o Stream sockets created via socket() start life with their own
155  *   independent lock.
156  *
157  * o Stream connections to a named endpoint are slightly more complicated.
158  *   Sockets that have called listen() have their lock pointer mutated to
159  *   the global uipc_lock.  When establishing a connection, the connecting
160  *   socket also has its lock mutated to uipc_lock, which matches the head
161  *   (listening socket).  We create a new socket for accept() to return, and
162  *   that also shares the head's lock.  Until the connection is completely
163  *   done on both ends, all three sockets are locked by uipc_lock.  Once the
164  *   connection is complete, the association with the head's lock is broken.
165  *   The connecting socket and the socket returned from accept() have their
166  *   lock pointers mutated away from uipc_lock, and back to the connecting
167  *   socket's original, independent lock.  The head continues to be locked
168  *   by uipc_lock.
169  *
170  * o If uipc_lock is determined to be a significant source of contention,
171  *   it could easily be hashed out.  It is difficult to simply make it an
172  *   independent lock because of visibility / garbage collection issues:
173  *   if a socket has been associated with a lock at any point, that lock
174  *   must remain valid until the socket is no longer visible in the system.
175  *   The lock must not be freed or otherwise destroyed until any sockets
176  *   that had referenced it have also been destroyed.
177  */
178 const struct sockaddr_un sun_noname = {
179 	.sun_len = offsetof(struct sockaddr_un, sun_path),
180 	.sun_family = AF_LOCAL,
181 };
182 ino_t	unp_ino;			/* prototype for fake inode numbers */
183 
184 static struct mbuf * unp_addsockcred(struct lwp *, struct mbuf *);
185 static void   unp_discard_later(file_t *);
186 static void   unp_discard_now(file_t *);
187 static void   unp_disconnect1(struct unpcb *);
188 static bool   unp_drop(struct unpcb *, int);
189 static int    unp_internalize(struct mbuf **);
190 static void   unp_mark(file_t *);
191 static void   unp_scan(struct mbuf *, void (*)(file_t *), int);
192 static void   unp_shutdown1(struct unpcb *);
193 static void   unp_thread(void *);
194 static void   unp_thread_kick(void);
195 
196 static kmutex_t *uipc_lock;
197 
198 static kcondvar_t unp_thread_cv;
199 static lwp_t *unp_thread_lwp;
200 static SLIST_HEAD(,file) unp_thread_discard;
201 static int unp_defer;
202 static struct sysctllog *usrreq_sysctllog;
203 static void unp_sysctl_create(void);
204 
205 /* Compat interface */
206 
207 struct mbuf * stub_compat_70_unp_addsockcred(lwp_t *, struct mbuf *);
208 
209 struct mbuf * stub_compat_70_unp_addsockcred(struct lwp *lwp,
210     struct mbuf *control)
211 {
212 
213 /* just copy our initial argument */
214 	return control;
215 }
216 
217 bool compat70_ocreds_valid = false;
218 
219 /*
220  * Initialize Unix protocols.
221  */
222 void
223 uipc_init(void)
224 {
225 	int error;
226 
227 	unp_sysctl_create();
228 
229 	uipc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
230 	cv_init(&unp_thread_cv, "unpgc");
231 
232 	error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, unp_thread,
233 	    NULL, &unp_thread_lwp, "unpgc");
234 	if (error != 0)
235 		panic("uipc_init %d", error);
236 }
237 
238 static void
239 unp_connid(struct lwp *l, struct unpcb *unp, int flags)
240 {
241 	unp->unp_connid.unp_pid = l->l_proc->p_pid;
242 	unp->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred);
243 	unp->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred);
244 	unp->unp_flags |= flags;
245 }
246 
247 /*
248  * A connection succeeded: disassociate both endpoints from the head's
249  * lock, and make them share their own lock.  There is a race here: for
250  * a very brief time one endpoint will be locked by a different lock
251  * than the other end.  However, since the current thread holds the old
252  * lock (the listening socket's lock, the head) access can still only be
253  * made to one side of the connection.
254  */
255 static void
256 unp_setpeerlocks(struct socket *so, struct socket *so2)
257 {
258 	struct unpcb *unp;
259 	kmutex_t *lock;
260 
261 	KASSERT(solocked2(so, so2));
262 
263 	/*
264 	 * Bail out if either end of the socket is not yet fully
265 	 * connected or accepted.  We only break the lock association
266 	 * with the head when the pair of sockets stand completely
267 	 * on their own.
268 	 */
269 	KASSERT(so->so_head == NULL);
270 	if (so2->so_head != NULL)
271 		return;
272 
273 	/*
274 	 * Drop references to old lock.  A third reference (from the
275 	 * queue head) must be held as we still hold its lock.  Bonus:
276 	 * we don't need to worry about garbage collecting the lock.
277 	 */
278 	lock = so->so_lock;
279 	KASSERT(lock == uipc_lock);
280 	mutex_obj_free(lock);
281 	mutex_obj_free(lock);
282 
283 	/*
284 	 * Grab stream lock from the initiator and share between the two
285 	 * endpoints.  Issue memory barrier to ensure all modifications
286 	 * become globally visible before the lock change.  so2 is
287 	 * assumed not to have a stream lock, because it was created
288 	 * purely for the server side to accept this connection and
289 	 * started out life using the domain-wide lock.
290 	 */
291 	unp = sotounpcb(so);
292 	KASSERT(unp->unp_streamlock != NULL);
293 	KASSERT(sotounpcb(so2)->unp_streamlock == NULL);
294 	lock = unp->unp_streamlock;
295 	unp->unp_streamlock = NULL;
296 	mutex_obj_hold(lock);
297 	/*
298 	 * Ensure lock is initialized before publishing it with
299 	 * solockreset.  Pairs with atomic_load_consume in solock and
300 	 * various loops to reacquire lock after wakeup.
301 	 */
302 	membar_release();
303 	/*
304 	 * possible race if lock is not held - see comment in
305 	 * uipc_usrreq(PRU_ACCEPT).
306 	 */
307 	KASSERT(mutex_owned(lock));
308 	solockreset(so, lock);
309 	solockreset(so2, lock);
310 }
311 
312 /*
313  * Reset a socket's lock back to the domain-wide lock.
314  */
315 static void
316 unp_resetlock(struct socket *so)
317 {
318 	kmutex_t *olock, *nlock;
319 	struct unpcb *unp;
320 
321 	KASSERT(solocked(so));
322 
323 	olock = so->so_lock;
324 	nlock = uipc_lock;
325 	if (olock == nlock)
326 		return;
327 	unp = sotounpcb(so);
328 	KASSERT(unp->unp_streamlock == NULL);
329 	unp->unp_streamlock = olock;
330 	mutex_obj_hold(nlock);
331 	mutex_enter(nlock);
332 	solockreset(so, nlock);
333 	mutex_exit(olock);
334 }
335 
336 static void
337 unp_free(struct unpcb *unp)
338 {
339 	if (unp->unp_addr)
340 		free(unp->unp_addr, M_SONAME);
341 	if (unp->unp_streamlock != NULL)
342 		mutex_obj_free(unp->unp_streamlock);
343 	kmem_free(unp, sizeof(*unp));
344 }
345 
346 static int
347 unp_output(struct mbuf *m, struct mbuf *control, struct unpcb *unp)
348 {
349 	struct socket *so2;
350 	const struct sockaddr_un *sun;
351 
352 	/* XXX: server side closed the socket */
353 	if (unp->unp_conn == NULL)
354 		return SET_ERROR(ECONNREFUSED);
355 	so2 = unp->unp_conn->unp_socket;
356 
357 	KASSERT(solocked(so2));
358 
359 	if (unp->unp_addr)
360 		sun = unp->unp_addr;
361 	else
362 		sun = &sun_noname;
363 	if (unp->unp_conn->unp_flags & UNP_WANTCRED)
364 		control = unp_addsockcred(curlwp, control);
365 	if (unp->unp_conn->unp_flags & UNP_OWANTCRED)
366 		MODULE_HOOK_CALL(uipc_unp_70_hook, (curlwp, control),
367 		    stub_compat_70_unp_addsockcred(curlwp, control), control);
368 	if (sbappendaddr(&so2->so_rcv, (const struct sockaddr *)sun, m,
369 	    control) == 0) {
370 		unp_dispose(control);
371 		m_freem(control);
372 		m_freem(m);
373 		/* Don't call soroverflow because we're returning this
374 		 * error directly to the sender. */
375 		so2->so_rcv.sb_overflowed++;
376 		return SET_ERROR(ENOBUFS);
377 	} else {
378 		sorwakeup(so2);
379 		return 0;
380 	}
381 }
382 
383 static void
384 unp_setaddr(struct socket *so, struct sockaddr *nam, bool peeraddr)
385 {
386 	const struct sockaddr_un *sun = NULL;
387 	struct unpcb *unp;
388 
389 	KASSERT(solocked(so));
390 	unp = sotounpcb(so);
391 
392 	if (peeraddr) {
393 		if (unp->unp_conn && unp->unp_conn->unp_addr)
394 			sun = unp->unp_conn->unp_addr;
395 	} else {
396 		if (unp->unp_addr)
397 			sun = unp->unp_addr;
398 	}
399 	if (sun == NULL)
400 		sun = &sun_noname;
401 
402 	memcpy(nam, sun, sun->sun_len);
403 }
404 
405 static int
406 unp_rcvd(struct socket *so, int flags, struct lwp *l)
407 {
408 	struct unpcb *unp = sotounpcb(so);
409 	struct socket *so2;
410 	u_int newhiwat;
411 
412 	KASSERT(solocked(so));
413 	KASSERT(unp != NULL);
414 
415 	switch (so->so_type) {
416 
417 	case SOCK_DGRAM:
418 		panic("uipc 1");
419 		/*NOTREACHED*/
420 
421 	case SOCK_SEQPACKET: /* FALLTHROUGH */
422 	case SOCK_STREAM:
423 #define	rcv (&so->so_rcv)
424 #define snd (&so2->so_snd)
425 		if (unp->unp_conn == 0)
426 			break;
427 		so2 = unp->unp_conn->unp_socket;
428 		KASSERT(solocked2(so, so2));
429 		/*
430 		 * Adjust backpressure on sender
431 		 * and wakeup any waiting to write.
432 		 */
433 		snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt;
434 		unp->unp_mbcnt = rcv->sb_mbcnt;
435 		newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc;
436 		(void)chgsbsize(so2->so_uidinfo,
437 		    &snd->sb_hiwat, newhiwat, RLIM_INFINITY);
438 		unp->unp_cc = rcv->sb_cc;
439 		sowwakeup(so2);
440 #undef snd
441 #undef rcv
442 		break;
443 
444 	default:
445 		panic("uipc 2");
446 	}
447 
448 	return 0;
449 }
450 
451 static int
452 unp_recvoob(struct socket *so, struct mbuf *m, int flags)
453 {
454 	KASSERT(solocked(so));
455 
456 	return SET_ERROR(EOPNOTSUPP);
457 }
458 
459 static int
460 unp_send(struct socket *so, struct mbuf *m, struct sockaddr *nam,
461     struct mbuf *control, struct lwp *l)
462 {
463 	struct unpcb *unp = sotounpcb(so);
464 	int error = 0;
465 	u_int newhiwat;
466 	struct socket *so2;
467 
468 	KASSERT(solocked(so));
469 	KASSERT(unp != NULL);
470 	KASSERT(m != NULL);
471 
472 	/*
473 	 * Note: unp_internalize() rejects any control message
474 	 * other than SCM_RIGHTS, and only allows one.  This
475 	 * has the side-effect of preventing a caller from
476 	 * forging SCM_CREDS.
477 	 */
478 	if (control) {
479 		sounlock(so);
480 		error = unp_internalize(&control);
481 		solock(so);
482 		if (error != 0) {
483 			m_freem(control);
484 			m_freem(m);
485 			return error;
486 		}
487 	}
488 
489 	switch (so->so_type) {
490 
491 	case SOCK_DGRAM: {
492 		KASSERT(so->so_lock == uipc_lock);
493 		if (nam) {
494 			if ((so->so_state & SS_ISCONNECTED) != 0)
495 				error = SET_ERROR(EISCONN);
496 			else {
497 				/*
498 				 * Note: once connected, the
499 				 * socket's lock must not be
500 				 * dropped until we have sent
501 				 * the message and disconnected.
502 				 * This is necessary to prevent
503 				 * intervening control ops, like
504 				 * another connection.
505 				 */
506 				error = unp_connect(so, nam, l);
507 			}
508 		} else {
509 			if ((so->so_state & SS_ISCONNECTED) == 0)
510 				error = SET_ERROR(ENOTCONN);
511 		}
512 		if (error) {
513 			unp_dispose(control);
514 			m_freem(control);
515 			m_freem(m);
516 			return error;
517 		}
518 		error = unp_output(m, control, unp);
519 		if (nam)
520 			unp_disconnect1(unp);
521 		break;
522 	}
523 
524 	case SOCK_SEQPACKET: /* FALLTHROUGH */
525 	case SOCK_STREAM:
526 #define	rcv (&so2->so_rcv)
527 #define	snd (&so->so_snd)
528 		if (unp->unp_conn == NULL) {
529 			error = SET_ERROR(ENOTCONN);
530 			break;
531 		}
532 		so2 = unp->unp_conn->unp_socket;
533 		KASSERT(solocked2(so, so2));
534 		if (unp->unp_conn->unp_flags & UNP_WANTCRED) {
535 			/*
536 			 * Credentials are passed only once on
537 			 * SOCK_STREAM and SOCK_SEQPACKET.
538 			 */
539 			unp->unp_conn->unp_flags &= ~UNP_WANTCRED;
540 			control = unp_addsockcred(l, control);
541 		}
542 		if (unp->unp_conn->unp_flags & UNP_OWANTCRED) {
543 			/*
544 			 * Credentials are passed only once on
545 			 * SOCK_STREAM and SOCK_SEQPACKET.
546 			 */
547 			unp->unp_conn->unp_flags &= ~UNP_OWANTCRED;
548 			MODULE_HOOK_CALL(uipc_unp_70_hook, (curlwp, control),
549 			    stub_compat_70_unp_addsockcred(curlwp, control),
550 			    control);
551 		}
552 		/*
553 		 * Send to paired receive port, and then reduce
554 		 * send buffer hiwater marks to maintain backpressure.
555 		 * Wake up readers.
556 		 */
557 		if (control) {
558 			if (sbappendcontrol(rcv, m, control) != 0)
559 				control = NULL;
560 		} else {
561 			switch(so->so_type) {
562 			case SOCK_SEQPACKET:
563 				sbappendrecord(rcv, m);
564 				break;
565 			case SOCK_STREAM:
566 				sbappend(rcv, m);
567 				break;
568 			default:
569 				panic("uipc_usrreq");
570 				break;
571 			}
572 		}
573 		snd->sb_mbmax -=
574 		    rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt;
575 		unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt;
576 		newhiwat = snd->sb_hiwat -
577 		    (rcv->sb_cc - unp->unp_conn->unp_cc);
578 		(void)chgsbsize(so->so_uidinfo,
579 		    &snd->sb_hiwat, newhiwat, RLIM_INFINITY);
580 		unp->unp_conn->unp_cc = rcv->sb_cc;
581 		sorwakeup(so2);
582 #undef snd
583 #undef rcv
584 		if (control != NULL) {
585 			unp_dispose(control);
586 			m_freem(control);
587 		}
588 		break;
589 
590 	default:
591 		panic("uipc 4");
592 	}
593 
594 	return error;
595 }
596 
597 static int
598 unp_sendoob(struct socket *so, struct mbuf *m, struct mbuf * control)
599 {
600 	KASSERT(solocked(so));
601 
602 	m_freem(m);
603 	m_freem(control);
604 
605 	return SET_ERROR(EOPNOTSUPP);
606 }
607 
608 /*
609  * Unix domain socket option processing.
610  */
611 int
612 uipc_ctloutput(int op, struct socket *so, struct sockopt *sopt)
613 {
614 	struct unpcb *unp = sotounpcb(so);
615 	int optval = 0, error = 0;
616 
617 	KASSERT(solocked(so));
618 
619 	if (sopt->sopt_level != SOL_LOCAL) {
620 		error = SET_ERROR(ENOPROTOOPT);
621 	} else switch (op) {
622 
623 	case PRCO_SETOPT:
624 		switch (sopt->sopt_name) {
625 		case LOCAL_OCREDS:
626 			if (!compat70_ocreds_valid)  {
627 				error = SET_ERROR(ENOPROTOOPT);
628 				break;
629 			}
630 			/* FALLTHROUGH */
631 		case LOCAL_CREDS:
632 		case LOCAL_CONNWAIT:
633 			error = sockopt_getint(sopt, &optval);
634 			if (error)
635 				break;
636 			switch (sopt->sopt_name) {
637 #define	OPTSET(bit) \
638 	if (optval) \
639 		unp->unp_flags |= (bit); \
640 	else \
641 		unp->unp_flags &= ~(bit);
642 
643 			case LOCAL_CREDS:
644 				OPTSET(UNP_WANTCRED);
645 				break;
646 			case LOCAL_CONNWAIT:
647 				OPTSET(UNP_CONNWAIT);
648 				break;
649 			case LOCAL_OCREDS:
650 				OPTSET(UNP_OWANTCRED);
651 				break;
652 			}
653 			break;
654 #undef OPTSET
655 
656 		default:
657 			error = SET_ERROR(ENOPROTOOPT);
658 			break;
659 		}
660 		break;
661 
662 	case PRCO_GETOPT:
663 		sounlock(so);
664 		switch (sopt->sopt_name) {
665 		case LOCAL_PEEREID:
666 			if (unp->unp_flags & UNP_EIDSVALID) {
667 				error = sockopt_set(sopt, &unp->unp_connid,
668 				    sizeof(unp->unp_connid));
669 			} else {
670 				error = SET_ERROR(EINVAL);
671 			}
672 			break;
673 		case LOCAL_CREDS:
674 #define	OPTBIT(bit)	(unp->unp_flags & (bit) ? 1 : 0)
675 
676 			optval = OPTBIT(UNP_WANTCRED);
677 			error = sockopt_setint(sopt, optval);
678 			break;
679 		case LOCAL_OCREDS:
680 			if (compat70_ocreds_valid) {
681 				optval = OPTBIT(UNP_OWANTCRED);
682 				error = sockopt_setint(sopt, optval);
683 				break;
684 			}
685 #undef OPTBIT
686 			/* FALLTHROUGH */
687 		default:
688 			error = SET_ERROR(ENOPROTOOPT);
689 			break;
690 		}
691 		solock(so);
692 		break;
693 	}
694 	return (error);
695 }
696 
697 /*
698  * Both send and receive buffers are allocated PIPSIZ bytes of buffering
699  * for stream sockets, although the total for sender and receiver is
700  * actually only PIPSIZ.
701  * Datagram sockets really use the sendspace as the maximum datagram size,
702  * and don't really want to reserve the sendspace.  Their recvspace should
703  * be large enough for at least one max-size datagram plus address.
704  */
705 #ifndef PIPSIZ
706 #define	PIPSIZ	8192
707 #endif
708 u_long	unpst_sendspace = PIPSIZ;
709 u_long	unpst_recvspace = PIPSIZ;
710 u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
711 u_long	unpdg_recvspace = 16*1024;
712 
713 u_int	unp_rights;			/* files in flight */
714 u_int	unp_rights_ratio = 2;		/* limit, fraction of maxfiles */
715 
716 static int
717 unp_attach(struct socket *so, int proto)
718 {
719 	struct unpcb *unp = sotounpcb(so);
720 	u_long sndspc, rcvspc;
721 	int error;
722 
723 	KASSERT(unp == NULL);
724 
725 	switch (so->so_type) {
726 	case SOCK_SEQPACKET:
727 		/* FALLTHROUGH */
728 	case SOCK_STREAM:
729 		if (so->so_lock == NULL) {
730 			so->so_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
731 			solock(so);
732 		}
733 		sndspc = unpst_sendspace;
734 		rcvspc = unpst_recvspace;
735 		break;
736 
737 	case SOCK_DGRAM:
738 		if (so->so_lock == NULL) {
739 			mutex_obj_hold(uipc_lock);
740 			so->so_lock = uipc_lock;
741 			solock(so);
742 		}
743 		sndspc = unpdg_sendspace;
744 		rcvspc = unpdg_recvspace;
745 		break;
746 
747 	default:
748 		panic("unp_attach");
749 	}
750 
751 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
752 		error = soreserve(so, sndspc, rcvspc);
753 		if (error) {
754 			return error;
755 		}
756 	}
757 
758 	unp = kmem_zalloc(sizeof(*unp), KM_SLEEP);
759 	nanotime(&unp->unp_ctime);
760 	unp->unp_socket = so;
761 	so->so_pcb = unp;
762 
763 	KASSERT(solocked(so));
764 	return 0;
765 }
766 
767 static void
768 unp_detach(struct socket *so)
769 {
770 	struct unpcb *unp;
771 	vnode_t *vp;
772 
773 	unp = sotounpcb(so);
774 	KASSERT(unp != NULL);
775 	KASSERT(solocked(so));
776  retry:
777 	if ((vp = unp->unp_vnode) != NULL) {
778 		sounlock(so);
779 		/* Acquire v_interlock to protect against unp_connect(). */
780 		/* XXXAD racy */
781 		mutex_enter(vp->v_interlock);
782 		vp->v_socket = NULL;
783 		mutex_exit(vp->v_interlock);
784 		vrele(vp);
785 		solock(so);
786 		unp->unp_vnode = NULL;
787 	}
788 	if (unp->unp_conn)
789 		unp_disconnect1(unp);
790 	while (unp->unp_refs) {
791 		KASSERT(solocked2(so, unp->unp_refs->unp_socket));
792 		if (unp_drop(unp->unp_refs, SET_ERROR(ECONNRESET))) {
793 			solock(so);
794 			goto retry;
795 		}
796 	}
797 	soisdisconnected(so);
798 	so->so_pcb = NULL;
799 	if (unp_rights) {
800 		/*
801 		 * Normally the receive buffer is flushed later, in sofree,
802 		 * but if our receive buffer holds references to files that
803 		 * are now garbage, we will enqueue those file references to
804 		 * the garbage collector and kick it into action.
805 		 */
806 		sorflush(so);
807 		unp_free(unp);
808 		unp_thread_kick();
809 	} else
810 		unp_free(unp);
811 }
812 
813 static int
814 unp_accept(struct socket *so, struct sockaddr *nam)
815 {
816 	struct unpcb *unp = sotounpcb(so);
817 	struct socket *so2;
818 
819 	KASSERT(solocked(so));
820 	KASSERT(nam != NULL);
821 
822 	/* XXX code review required to determine if unp can ever be NULL */
823 	if (unp == NULL)
824 		return SET_ERROR(EINVAL);
825 
826 	KASSERT(so->so_lock == uipc_lock);
827 	/*
828 	 * Mark the initiating STREAM socket as connected *ONLY*
829 	 * after it's been accepted.  This prevents a client from
830 	 * overrunning a server and receiving ECONNREFUSED.
831 	 */
832 	if (unp->unp_conn == NULL) {
833 		/*
834 		 * This will use the empty socket and will not
835 		 * allocate.
836 		 */
837 		unp_setaddr(so, nam, true);
838 		return 0;
839 	}
840 	so2 = unp->unp_conn->unp_socket;
841 	if (so2->so_state & SS_ISCONNECTING) {
842 		KASSERT(solocked2(so, so->so_head));
843 		KASSERT(solocked2(so2, so->so_head));
844 		soisconnected(so2);
845 	}
846 	/*
847 	 * If the connection is fully established, break the
848 	 * association with uipc_lock and give the connected
849 	 * pair a separate lock to share.
850 	 * There is a race here: sotounpcb(so2)->unp_streamlock
851 	 * is not locked, so when changing so2->so_lock
852 	 * another thread can grab it while so->so_lock is still
853 	 * pointing to the (locked) uipc_lock.
854 	 * this should be harmless, except that this makes
855 	 * solocked2() and solocked() unreliable.
856 	 * Another problem is that unp_setaddr() expects the
857 	 * the socket locked. Grabbing sotounpcb(so2)->unp_streamlock
858 	 * fixes both issues.
859 	 */
860 	mutex_enter(sotounpcb(so2)->unp_streamlock);
861 	unp_setpeerlocks(so2, so);
862 	/*
863 	 * Only now return peer's address, as we may need to
864 	 * block in order to allocate memory.
865 	 *
866 	 * XXX Minor race: connection can be broken while
867 	 * lock is dropped in unp_setaddr().  We will return
868 	 * error == 0 and sun_noname as the peer address.
869 	 */
870 	unp_setaddr(so, nam, true);
871 	/* so_lock now points to unp_streamlock */
872 	mutex_exit(so2->so_lock);
873 	return 0;
874 }
875 
876 static int
877 unp_ioctl(struct socket *so, u_long cmd, void *nam, struct ifnet *ifp)
878 {
879 	return SET_ERROR(EOPNOTSUPP);
880 }
881 
882 static int
883 unp_stat(struct socket *so, struct stat *ub)
884 {
885 	struct unpcb *unp;
886 	struct socket *so2;
887 
888 	KASSERT(solocked(so));
889 
890 	unp = sotounpcb(so);
891 	if (unp == NULL)
892 		return SET_ERROR(EINVAL);
893 
894 	ub->st_blksize = so->so_snd.sb_hiwat;
895 	switch (so->so_type) {
896 	case SOCK_SEQPACKET: /* FALLTHROUGH */
897 	case SOCK_STREAM:
898 		if (unp->unp_conn == 0)
899 			break;
900 
901 		so2 = unp->unp_conn->unp_socket;
902 		KASSERT(solocked2(so, so2));
903 		ub->st_blksize += so2->so_rcv.sb_cc;
904 		break;
905 	default:
906 		break;
907 	}
908 	ub->st_dev = NODEV;
909 	if (unp->unp_ino == 0)
910 		unp->unp_ino = unp_ino++;
911 	ub->st_atimespec = ub->st_mtimespec = ub->st_ctimespec = unp->unp_ctime;
912 	ub->st_ino = unp->unp_ino;
913 	ub->st_uid = so->so_uidinfo->ui_uid;
914 	ub->st_gid = so->so_egid;
915 	return (0);
916 }
917 
918 static int
919 unp_peeraddr(struct socket *so, struct sockaddr *nam)
920 {
921 	KASSERT(solocked(so));
922 	KASSERT(sotounpcb(so) != NULL);
923 	KASSERT(nam != NULL);
924 
925 	unp_setaddr(so, nam, true);
926 	return 0;
927 }
928 
929 static int
930 unp_sockaddr(struct socket *so, struct sockaddr *nam)
931 {
932 	KASSERT(solocked(so));
933 	KASSERT(sotounpcb(so) != NULL);
934 	KASSERT(nam != NULL);
935 
936 	unp_setaddr(so, nam, false);
937 	return 0;
938 }
939 
940 /*
941  * we only need to perform this allocation until syscalls other than
942  * bind are adjusted to use sockaddr_big.
943  */
944 static struct sockaddr_un *
945 makeun_sb(struct sockaddr *nam, size_t *addrlen)
946 {
947 	struct sockaddr_un *sun;
948 
949 	*addrlen = nam->sa_len + 1;
950 	sun = malloc(*addrlen, M_SONAME, M_WAITOK);
951 	memcpy(sun, nam, nam->sa_len);
952 	*(((char *)sun) + nam->sa_len) = '\0';
953 	return sun;
954 }
955 
956 static int
957 unp_bind(struct socket *so, struct sockaddr *nam, struct lwp *l)
958 {
959 	struct sockaddr_un *sun;
960 	struct unpcb *unp;
961 	vnode_t *vp;
962 	struct vattr vattr;
963 	size_t addrlen;
964 	int error;
965 	struct pathbuf *pb;
966 	struct nameidata nd;
967 	proc_t *p;
968 
969 	unp = sotounpcb(so);
970 
971 	KASSERT(solocked(so));
972 	KASSERT(unp != NULL);
973 	KASSERT(nam != NULL);
974 
975 	if (unp->unp_vnode != NULL)
976 		return SET_ERROR(EINVAL);
977 	if ((unp->unp_flags & UNP_BUSY) != 0) {
978 		/*
979 		 * EALREADY may not be strictly accurate, but since this
980 		 * is a major application error it's hardly a big deal.
981 		 */
982 		return SET_ERROR(EALREADY);
983 	}
984 	unp->unp_flags |= UNP_BUSY;
985 	sounlock(so);
986 
987 	p = l->l_proc;
988 	sun = makeun_sb(nam, &addrlen);
989 
990 	pb = pathbuf_create(sun->sun_path);
991 	if (pb == NULL) {
992 		error = SET_ERROR(ENOMEM);
993 		goto bad;
994 	}
995 	NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT | TRYEMULROOT, pb);
996 
997 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
998 	if ((error = namei(&nd)) != 0) {
999 		pathbuf_destroy(pb);
1000 		goto bad;
1001 	}
1002 	vp = nd.ni_vp;
1003 	if (vp != NULL) {
1004 		VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd);
1005 		if (nd.ni_dvp == vp)
1006 			vrele(nd.ni_dvp);
1007 		else
1008 			vput(nd.ni_dvp);
1009 		vrele(vp);
1010 		pathbuf_destroy(pb);
1011 		error = SET_ERROR(EADDRINUSE);
1012 		goto bad;
1013 	}
1014 	vattr_null(&vattr);
1015 	vattr.va_type = VSOCK;
1016 	vattr.va_mode = ACCESSPERMS & ~(p->p_cwdi->cwdi_cmask);
1017 	error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
1018 	if (error) {
1019 		vput(nd.ni_dvp);
1020 		pathbuf_destroy(pb);
1021 		goto bad;
1022 	}
1023 	vp = nd.ni_vp;
1024 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1025 	solock(so);
1026 	vp->v_socket = unp->unp_socket;
1027 	unp->unp_vnode = vp;
1028 	unp->unp_addrlen = addrlen;
1029 	unp->unp_addr = sun;
1030 	VOP_UNLOCK(vp);
1031 	vput(nd.ni_dvp);
1032 	unp->unp_flags &= ~UNP_BUSY;
1033 	pathbuf_destroy(pb);
1034 	return (0);
1035 
1036  bad:
1037 	free(sun, M_SONAME);
1038 	solock(so);
1039 	unp->unp_flags &= ~UNP_BUSY;
1040 	return (error);
1041 }
1042 
1043 static int
1044 unp_listen(struct socket *so, struct lwp *l)
1045 {
1046 	struct unpcb *unp = sotounpcb(so);
1047 
1048 	KASSERT(solocked(so));
1049 	KASSERT(unp != NULL);
1050 
1051 	/*
1052 	 * If the socket can accept a connection, it must be
1053 	 * locked by uipc_lock.
1054 	 */
1055 	unp_resetlock(so);
1056 	if (unp->unp_vnode == NULL)
1057 		return SET_ERROR(EINVAL);
1058 
1059 	unp_connid(l, unp, UNP_EIDSBIND);
1060 	return 0;
1061 }
1062 
1063 static int
1064 unp_disconnect(struct socket *so)
1065 {
1066 	KASSERT(solocked(so));
1067 	KASSERT(sotounpcb(so) != NULL);
1068 
1069 	unp_disconnect1(sotounpcb(so));
1070 	return 0;
1071 }
1072 
1073 static int
1074 unp_shutdown(struct socket *so)
1075 {
1076 	KASSERT(solocked(so));
1077 	KASSERT(sotounpcb(so) != NULL);
1078 
1079 	socantsendmore(so);
1080 	unp_shutdown1(sotounpcb(so));
1081 	return 0;
1082 }
1083 
1084 static int
1085 unp_abort(struct socket *so)
1086 {
1087 	KASSERT(solocked(so));
1088 	KASSERT(sotounpcb(so) != NULL);
1089 
1090 	(void)unp_drop(sotounpcb(so), SET_ERROR(ECONNABORTED));
1091 	KASSERT(so->so_head == NULL);
1092 	KASSERT(so->so_pcb != NULL);
1093 	unp_detach(so);
1094 	return 0;
1095 }
1096 
1097 static int
1098 unp_connect1(struct socket *so, struct socket *so2, struct lwp *l)
1099 {
1100 	struct unpcb *unp = sotounpcb(so);
1101 	struct unpcb *unp2;
1102 
1103 	if (so2->so_type != so->so_type)
1104 		return SET_ERROR(EPROTOTYPE);
1105 
1106 	/*
1107 	 * All three sockets involved must be locked by same lock:
1108 	 *
1109 	 * local endpoint (so)
1110 	 * remote endpoint (so2)
1111 	 * queue head (so2->so_head, only if PR_CONNREQUIRED)
1112 	 */
1113 	KASSERT(solocked2(so, so2));
1114 	KASSERT(so->so_head == NULL);
1115 	if (so2->so_head != NULL) {
1116 		KASSERT(so2->so_lock == uipc_lock);
1117 		KASSERT(solocked2(so2, so2->so_head));
1118 	}
1119 
1120 	unp2 = sotounpcb(so2);
1121 	unp->unp_conn = unp2;
1122 
1123 	switch (so->so_type) {
1124 
1125 	case SOCK_DGRAM:
1126 		unp->unp_nextref = unp2->unp_refs;
1127 		unp2->unp_refs = unp;
1128 		soisconnected(so);
1129 		break;
1130 
1131 	case SOCK_SEQPACKET: /* FALLTHROUGH */
1132 	case SOCK_STREAM:
1133 
1134 		/*
1135 		 * SOCK_SEQPACKET and SOCK_STREAM cases are handled by callers
1136 		 * which are unp_connect() or unp_connect2().
1137 		 */
1138 
1139 		break;
1140 
1141 	default:
1142 		panic("unp_connect1");
1143 	}
1144 
1145 	return 0;
1146 }
1147 
1148 int
1149 unp_connect(struct socket *so, struct sockaddr *nam, struct lwp *l)
1150 {
1151 	struct sockaddr_un *sun;
1152 	vnode_t *vp;
1153 	struct socket *so2, *so3;
1154 	struct unpcb *unp, *unp2, *unp3;
1155 	size_t addrlen;
1156 	int error;
1157 	struct pathbuf *pb;
1158 	struct nameidata nd;
1159 
1160 	unp = sotounpcb(so);
1161 	if ((unp->unp_flags & UNP_BUSY) != 0) {
1162 		/*
1163 		 * EALREADY may not be strictly accurate, but since this
1164 		 * is a major application error it's hardly a big deal.
1165 		 */
1166 		return SET_ERROR(EALREADY);
1167 	}
1168 	unp->unp_flags |= UNP_BUSY;
1169 	sounlock(so);
1170 
1171 	sun = makeun_sb(nam, &addrlen);
1172 	pb = pathbuf_create(sun->sun_path);
1173 	if (pb == NULL) {
1174 		error = SET_ERROR(ENOMEM);
1175 		goto bad2;
1176 	}
1177 
1178 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb);
1179 
1180 	if ((error = namei(&nd)) != 0) {
1181 		pathbuf_destroy(pb);
1182 		goto bad2;
1183 	}
1184 	vp = nd.ni_vp;
1185 	pathbuf_destroy(pb);
1186 	if (vp->v_type != VSOCK) {
1187 		error = SET_ERROR(ENOTSOCK);
1188 		goto bad;
1189 	}
1190 	if ((error = VOP_ACCESS(vp, VWRITE, l->l_cred)) != 0)
1191 		goto bad;
1192 	/* Acquire v_interlock to protect against unp_detach(). */
1193 	mutex_enter(vp->v_interlock);
1194 	so2 = vp->v_socket;
1195 	if (so2 == NULL) {
1196 		mutex_exit(vp->v_interlock);
1197 		error = SET_ERROR(ECONNREFUSED);
1198 		goto bad;
1199 	}
1200 	if (so->so_type != so2->so_type) {
1201 		mutex_exit(vp->v_interlock);
1202 		error = SET_ERROR(EPROTOTYPE);
1203 		goto bad;
1204 	}
1205 	solock(so);
1206 	unp_resetlock(so);
1207 	mutex_exit(vp->v_interlock);
1208 	if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) {
1209 		/*
1210 		 * This may seem somewhat fragile but is OK: if we can
1211 		 * see SO_ACCEPTCONN set on the endpoint, then it must
1212 		 * be locked by the domain-wide uipc_lock.
1213 		 */
1214 		KASSERT((so2->so_options & SO_ACCEPTCONN) == 0 ||
1215 		    so2->so_lock == uipc_lock);
1216 		if ((so2->so_options & SO_ACCEPTCONN) == 0 ||
1217 		    (so3 = sonewconn(so2, false)) == NULL) {
1218 			error = SET_ERROR(ECONNREFUSED);
1219 			sounlock(so);
1220 			goto bad;
1221 		}
1222 		unp2 = sotounpcb(so2);
1223 		unp3 = sotounpcb(so3);
1224 		if (unp2->unp_addr) {
1225 			unp3->unp_addr = malloc(unp2->unp_addrlen,
1226 			    M_SONAME, M_WAITOK);
1227 			memcpy(unp3->unp_addr, unp2->unp_addr,
1228 			    unp2->unp_addrlen);
1229 			unp3->unp_addrlen = unp2->unp_addrlen;
1230 		}
1231 		unp3->unp_flags = unp2->unp_flags;
1232 		so2 = so3;
1233 		/*
1234 		 * The connector's (client's) credentials are copied from its
1235 		 * process structure at the time of connect() (which is now).
1236 		 */
1237 		unp_connid(l, unp3, UNP_EIDSVALID);
1238 		 /*
1239 		  * The receiver's (server's) credentials are copied from the
1240 		  * unp_peercred member of socket on which the former called
1241 		  * listen(); unp_listen() cached that process's credentials
1242 		  * at that time so we can use them now.
1243 		  */
1244 		if (unp2->unp_flags & UNP_EIDSBIND) {
1245 			memcpy(&unp->unp_connid, &unp2->unp_connid,
1246 			    sizeof(unp->unp_connid));
1247 			unp->unp_flags |= UNP_EIDSVALID;
1248 		}
1249 	}
1250 	error = unp_connect1(so, so2, l);
1251 	if (error) {
1252 		sounlock(so);
1253 		goto bad;
1254 	}
1255 	unp2 = sotounpcb(so2);
1256 	switch (so->so_type) {
1257 
1258 	/*
1259 	 * SOCK_DGRAM and default cases are handled in prior call to
1260 	 * unp_connect1(), do not add a default case without fixing
1261 	 * unp_connect1().
1262 	 */
1263 
1264 	case SOCK_SEQPACKET: /* FALLTHROUGH */
1265 	case SOCK_STREAM:
1266 		unp2->unp_conn = unp;
1267 		if ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)
1268 			soisconnecting(so);
1269 		else
1270 			soisconnected(so);
1271 		soisconnected(so2);
1272 		/*
1273 		 * If the connection is fully established, break the
1274 		 * association with uipc_lock and give the connected
1275 		 * pair a separate lock to share.
1276 		 */
1277 		KASSERT(so2->so_head != NULL);
1278 		unp_setpeerlocks(so, so2);
1279 		break;
1280 
1281 	}
1282 	sounlock(so);
1283  bad:
1284 	vput(vp);
1285  bad2:
1286 	free(sun, M_SONAME);
1287 	solock(so);
1288 	unp->unp_flags &= ~UNP_BUSY;
1289 	return (error);
1290 }
1291 
1292 int
1293 unp_connect2(struct socket *so, struct socket *so2)
1294 {
1295 	struct unpcb *unp = sotounpcb(so);
1296 	struct unpcb *unp2;
1297 	int error = 0;
1298 
1299 	KASSERT(solocked2(so, so2));
1300 
1301 	error = unp_connect1(so, so2, curlwp);
1302 	if (error)
1303 		return error;
1304 
1305 	unp2 = sotounpcb(so2);
1306 	switch (so->so_type) {
1307 
1308 	/*
1309 	 * SOCK_DGRAM and default cases are handled in prior call to
1310 	 * unp_connect1(), do not add a default case without fixing
1311 	 * unp_connect1().
1312 	 */
1313 
1314 	case SOCK_SEQPACKET: /* FALLTHROUGH */
1315 	case SOCK_STREAM:
1316 		unp2->unp_conn = unp;
1317 		soisconnected(so);
1318 		soisconnected(so2);
1319 		break;
1320 
1321 	}
1322 	return error;
1323 }
1324 
1325 static void
1326 unp_disconnect1(struct unpcb *unp)
1327 {
1328 	struct unpcb *unp2 = unp->unp_conn;
1329 	struct socket *so;
1330 
1331 	if (unp2 == 0)
1332 		return;
1333 	unp->unp_conn = 0;
1334 	so = unp->unp_socket;
1335 	switch (so->so_type) {
1336 	case SOCK_DGRAM:
1337 		if (unp2->unp_refs == unp)
1338 			unp2->unp_refs = unp->unp_nextref;
1339 		else {
1340 			unp2 = unp2->unp_refs;
1341 			for (;;) {
1342 				KASSERT(solocked2(so, unp2->unp_socket));
1343 				if (unp2 == 0)
1344 					panic("unp_disconnect1");
1345 				if (unp2->unp_nextref == unp)
1346 					break;
1347 				unp2 = unp2->unp_nextref;
1348 			}
1349 			unp2->unp_nextref = unp->unp_nextref;
1350 		}
1351 		unp->unp_nextref = 0;
1352 		so->so_state &= ~SS_ISCONNECTED;
1353 		break;
1354 
1355 	case SOCK_SEQPACKET: /* FALLTHROUGH */
1356 	case SOCK_STREAM:
1357 		KASSERT(solocked2(so, unp2->unp_socket));
1358 		soisdisconnected(so);
1359 		unp2->unp_conn = 0;
1360 		soisdisconnected(unp2->unp_socket);
1361 		break;
1362 	}
1363 }
1364 
1365 static void
1366 unp_shutdown1(struct unpcb *unp)
1367 {
1368 	struct socket *so;
1369 
1370 	switch(unp->unp_socket->so_type) {
1371 	case SOCK_SEQPACKET: /* FALLTHROUGH */
1372 	case SOCK_STREAM:
1373 		if (unp->unp_conn && (so = unp->unp_conn->unp_socket))
1374 			socantrcvmore(so);
1375 		break;
1376 	default:
1377 		break;
1378 	}
1379 }
1380 
1381 static bool
1382 unp_drop(struct unpcb *unp, int errno)
1383 {
1384 	struct socket *so = unp->unp_socket;
1385 
1386 	KASSERT(solocked(so));
1387 
1388 	so->so_error = errno;
1389 	unp_disconnect1(unp);
1390 	if (so->so_head) {
1391 		so->so_pcb = NULL;
1392 		/* sofree() drops the socket lock */
1393 		sofree(so);
1394 		unp_free(unp);
1395 		return true;
1396 	}
1397 	return false;
1398 }
1399 
1400 #ifdef notdef
1401 unp_drain(void)
1402 {
1403 
1404 }
1405 #endif
1406 
1407 int
1408 unp_externalize(struct mbuf *rights, struct lwp *l, int flags)
1409 {
1410 	struct cmsghdr * const cm = mtod(rights, struct cmsghdr *);
1411 	struct proc * const p = l->l_proc;
1412 	file_t **rp;
1413 	int error = 0;
1414 
1415 	const size_t nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) /
1416 	    sizeof(file_t *);
1417 	if (nfds == 0)
1418 		goto noop;
1419 
1420 	int * const fdp = kmem_alloc(nfds * sizeof(int), KM_SLEEP);
1421 	rw_enter(&p->p_cwdi->cwdi_lock, RW_READER);
1422 
1423 	/* Make sure the recipient should be able to see the files.. */
1424 	rp = (file_t **)CMSG_DATA(cm);
1425 	for (size_t i = 0; i < nfds; i++) {
1426 		file_t * const fp = *rp++;
1427 		if (fp == NULL) {
1428 			error = SET_ERROR(EINVAL);
1429 			goto out;
1430 		}
1431 		/*
1432 		 * If we are in a chroot'ed directory, and
1433 		 * someone wants to pass us a directory, make
1434 		 * sure it's inside the subtree we're allowed
1435 		 * to access.
1436 		 */
1437 		if (p->p_cwdi->cwdi_rdir != NULL && fp->f_type == DTYPE_VNODE) {
1438 			vnode_t *vp = fp->f_vnode;
1439 			if ((vp->v_type == VDIR) &&
1440 			    !vn_isunder(vp, p->p_cwdi->cwdi_rdir, l)) {
1441 				error = SET_ERROR(EPERM);
1442 				goto out;
1443 			}
1444 		}
1445 	}
1446 
1447  restart:
1448 	/*
1449 	 * First loop -- allocate file descriptor table slots for the
1450 	 * new files.
1451 	 */
1452 	for (size_t i = 0; i < nfds; i++) {
1453 		if ((error = fd_alloc(p, 0, &fdp[i])) != 0) {
1454 			/*
1455 			 * Back out what we've done so far.
1456 			 */
1457 			while (i-- > 0) {
1458 				fd_abort(p, NULL, fdp[i]);
1459 			}
1460 			if (error == ENOSPC) {
1461 				fd_tryexpand(p);
1462 				error = 0;
1463 				goto restart;
1464 			}
1465 			/*
1466 			 * This is the error that has historically
1467 			 * been returned, and some callers may
1468 			 * expect it.
1469 			 */
1470 			error = SET_ERROR(EMSGSIZE);
1471 			goto out;
1472 		}
1473 	}
1474 
1475 	/*
1476 	 * Now that adding them has succeeded, update all of the
1477 	 * file passing state and affix the descriptors.
1478 	 */
1479 	rp = (file_t **)CMSG_DATA(cm);
1480 	int *ofdp = (int *)CMSG_DATA(cm);
1481 	for (size_t i = 0; i < nfds; i++) {
1482 		file_t * const fp = *rp++;
1483 		const int fd = fdp[i];
1484 		atomic_dec_uint(&unp_rights);
1485 		fd_set_exclose(l, fd, (flags & O_CLOEXEC) != 0);
1486 		fd_affix(p, fp, fd);
1487 		/*
1488 		 * Done with this file pointer, replace it with a fd;
1489 		 */
1490 		*ofdp++ = fd;
1491 		mutex_enter(&fp->f_lock);
1492 		fp->f_msgcount--;
1493 		mutex_exit(&fp->f_lock);
1494 		/*
1495 		 * Note that fd_affix() adds a reference to the file.
1496 		 * The file may already have been closed by another
1497 		 * LWP in the process, so we must drop the reference
1498 		 * added by unp_internalize() with closef().
1499 		 */
1500 		closef(fp);
1501 	}
1502 
1503 	/*
1504 	 * Adjust length, in case of transition from large file_t
1505 	 * pointers to ints.
1506 	 */
1507 	if (sizeof(file_t *) != sizeof(int)) {
1508 		cm->cmsg_len = CMSG_LEN(nfds * sizeof(int));
1509 		rights->m_len = CMSG_SPACE(nfds * sizeof(int));
1510 	}
1511  out:
1512 	if (__predict_false(error != 0)) {
1513 		file_t **const fpp = (file_t **)CMSG_DATA(cm);
1514 		for (size_t i = 0; i < nfds; i++)
1515 			unp_discard_now(fpp[i]);
1516 		/*
1517 		 * Truncate the array so that nobody will try to interpret
1518 		 * what is now garbage in it.
1519 		 */
1520 		cm->cmsg_len = CMSG_LEN(0);
1521 		rights->m_len = CMSG_SPACE(0);
1522 	}
1523 	rw_exit(&p->p_cwdi->cwdi_lock);
1524 	kmem_free(fdp, nfds * sizeof(int));
1525 
1526  noop:
1527 	/*
1528 	 * Don't disclose kernel memory in the alignment space.
1529 	 */
1530 	KASSERT(cm->cmsg_len <= rights->m_len);
1531 	memset(&mtod(rights, char *)[cm->cmsg_len], 0, rights->m_len -
1532 	    cm->cmsg_len);
1533 	return error;
1534 }
1535 
1536 static int
1537 unp_internalize(struct mbuf **controlp)
1538 {
1539 	filedesc_t *fdescp = curlwp->l_fd;
1540 	fdtab_t *dt;
1541 	struct mbuf *control = *controlp;
1542 	struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *);
1543 	file_t **rp, **files;
1544 	file_t *fp;
1545 	int i, fd, *fdp;
1546 	int nfds, error;
1547 	u_int maxmsg;
1548 
1549 	error = 0;
1550 	newcm = NULL;
1551 
1552 	/* Sanity check the control message header. */
1553 	if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET ||
1554 	    cm->cmsg_len > control->m_len ||
1555 	    cm->cmsg_len < CMSG_ALIGN(sizeof(*cm)))
1556 		return SET_ERROR(EINVAL);
1557 
1558 	/*
1559 	 * Verify that the file descriptors are valid, and acquire
1560 	 * a reference to each.
1561 	 */
1562 	nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int);
1563 	fdp = (int *)CMSG_DATA(cm);
1564 	maxmsg = maxfiles / unp_rights_ratio;
1565 	for (i = 0; i < nfds; i++) {
1566 		fd = *fdp++;
1567 		if (atomic_inc_uint_nv(&unp_rights) > maxmsg) {
1568 			atomic_dec_uint(&unp_rights);
1569 			nfds = i;
1570 			error = SET_ERROR(EAGAIN);
1571 			goto out;
1572 		}
1573 		if ((fp = fd_getfile(fd)) == NULL
1574 		    || fp->f_type == DTYPE_KQUEUE) {
1575 			if (fp)
1576 				fd_putfile(fd);
1577 			atomic_dec_uint(&unp_rights);
1578 			nfds = i;
1579 			error = SET_ERROR(EBADF);
1580 			goto out;
1581 		}
1582 	}
1583 
1584 	/* Allocate new space and copy header into it. */
1585 	newcm = malloc(CMSG_SPACE(nfds * sizeof(file_t *)), M_MBUF, M_WAITOK);
1586 	if (newcm == NULL) {
1587 		error = SET_ERROR(E2BIG);
1588 		goto out;
1589 	}
1590 	memcpy(newcm, cm, sizeof(struct cmsghdr));
1591 	memset(newcm + 1, 0, CMSG_LEN(0) - sizeof(struct cmsghdr));
1592 	files = (file_t **)CMSG_DATA(newcm);
1593 
1594 	/*
1595 	 * Transform the file descriptors into file_t pointers, in
1596 	 * reverse order so that if pointers are bigger than ints, the
1597 	 * int won't get until we're done.  No need to lock, as we have
1598 	 * already validated the descriptors with fd_getfile().
1599 	 */
1600 	fdp = (int *)CMSG_DATA(cm) + nfds;
1601 	rp = files + nfds;
1602 	for (i = 0; i < nfds; i++) {
1603 		dt = atomic_load_consume(&fdescp->fd_dt);
1604 		fp = atomic_load_consume(&dt->dt_ff[*--fdp]->ff_file);
1605 		KASSERT(fp != NULL);
1606 		mutex_enter(&fp->f_lock);
1607 		*--rp = fp;
1608 		fp->f_count++;
1609 		fp->f_msgcount++;
1610 		mutex_exit(&fp->f_lock);
1611 	}
1612 
1613  out:
1614 	/* Release descriptor references. */
1615 	fdp = (int *)CMSG_DATA(cm);
1616 	for (i = 0; i < nfds; i++) {
1617 		fd_putfile(*fdp++);
1618 		if (error != 0) {
1619 			atomic_dec_uint(&unp_rights);
1620 		}
1621 	}
1622 
1623 	if (error == 0) {
1624 		if (control->m_flags & M_EXT) {
1625 			m_freem(control);
1626 			*controlp = control = m_get(M_WAIT, MT_CONTROL);
1627 		}
1628 		MEXTADD(control, newcm, CMSG_SPACE(nfds * sizeof(file_t *)),
1629 		    M_MBUF, NULL, NULL);
1630 		cm = newcm;
1631 		/*
1632 		 * Adjust message & mbuf to note amount of space
1633 		 * actually used.
1634 		 */
1635 		cm->cmsg_len = CMSG_LEN(nfds * sizeof(file_t *));
1636 		control->m_len = CMSG_SPACE(nfds * sizeof(file_t *));
1637 	}
1638 
1639 	return error;
1640 }
1641 
1642 struct mbuf *
1643 unp_addsockcred(struct lwp *l, struct mbuf *control)
1644 {
1645 	struct sockcred *sc;
1646 	struct mbuf *m;
1647 	void *p;
1648 
1649 	m = sbcreatecontrol1(&p, SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred)),
1650 		SCM_CREDS, SOL_SOCKET, M_WAITOK);
1651 	if (m == NULL)
1652 		return control;
1653 
1654 	sc = p;
1655 	sc->sc_pid = l->l_proc->p_pid;
1656 	sc->sc_uid = kauth_cred_getuid(l->l_cred);
1657 	sc->sc_euid = kauth_cred_geteuid(l->l_cred);
1658 	sc->sc_gid = kauth_cred_getgid(l->l_cred);
1659 	sc->sc_egid = kauth_cred_getegid(l->l_cred);
1660 	sc->sc_ngroups = kauth_cred_ngroups(l->l_cred);
1661 
1662 	for (int i = 0; i < sc->sc_ngroups; i++)
1663 		sc->sc_groups[i] = kauth_cred_group(l->l_cred, i);
1664 
1665 	return m_add(control, m);
1666 }
1667 
1668 /*
1669  * Do a mark-sweep GC of files in the system, to free up any which are
1670  * caught in flight to an about-to-be-closed socket.  Additionally,
1671  * process deferred file closures.
1672  */
1673 static void
1674 unp_gc(file_t *dp)
1675 {
1676 	extern	struct domain unixdomain;
1677 	file_t *fp, *np;
1678 	struct socket *so, *so1;
1679 	u_int i, oflags, rflags;
1680 	bool didwork;
1681 
1682 	KASSERT(curlwp == unp_thread_lwp);
1683 	KASSERT(mutex_owned(&filelist_lock));
1684 
1685 	/*
1686 	 * First, process deferred file closures.
1687 	 */
1688 	while (!SLIST_EMPTY(&unp_thread_discard)) {
1689 		fp = SLIST_FIRST(&unp_thread_discard);
1690 		KASSERT(fp->f_unpcount > 0);
1691 		KASSERT(fp->f_count > 0);
1692 		KASSERT(fp->f_msgcount > 0);
1693 		KASSERT(fp->f_count >= fp->f_unpcount);
1694 		KASSERT(fp->f_count >= fp->f_msgcount);
1695 		KASSERT(fp->f_msgcount >= fp->f_unpcount);
1696 		SLIST_REMOVE_HEAD(&unp_thread_discard, f_unplist);
1697 		i = fp->f_unpcount;
1698 		fp->f_unpcount = 0;
1699 		mutex_exit(&filelist_lock);
1700 		for (; i != 0; i--) {
1701 			unp_discard_now(fp);
1702 		}
1703 		mutex_enter(&filelist_lock);
1704 	}
1705 
1706 	/*
1707 	 * Clear mark bits.  Ensure that we don't consider new files
1708 	 * entering the file table during this loop (they will not have
1709 	 * FSCAN set).
1710 	 */
1711 	unp_defer = 0;
1712 	LIST_FOREACH(fp, &filehead, f_list) {
1713 		for (oflags = fp->f_flag;; oflags = rflags) {
1714 			rflags = atomic_cas_uint(&fp->f_flag, oflags,
1715 			    (oflags | FSCAN) & ~(FMARK|FDEFER));
1716 			if (__predict_true(oflags == rflags)) {
1717 				break;
1718 			}
1719 		}
1720 	}
1721 
1722 	/*
1723 	 * Iterate over the set of sockets, marking ones believed (based on
1724 	 * refcount) to be referenced from a process, and marking for rescan
1725 	 * sockets which are queued on a socket.  Recan continues descending
1726 	 * and searching for sockets referenced by sockets (FDEFER), until
1727 	 * there are no more socket->socket references to be discovered.
1728 	 */
1729 	do {
1730 		didwork = false;
1731 		for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) {
1732 			KASSERT(mutex_owned(&filelist_lock));
1733 			np = LIST_NEXT(fp, f_list);
1734 			mutex_enter(&fp->f_lock);
1735 			if ((fp->f_flag & FDEFER) != 0) {
1736 				atomic_and_uint(&fp->f_flag, ~FDEFER);
1737 				unp_defer--;
1738 				if (fp->f_count == 0) {
1739 					/*
1740 					 * XXX: closef() doesn't pay attention
1741 					 * to FDEFER
1742 					 */
1743 					mutex_exit(&fp->f_lock);
1744 					continue;
1745 				}
1746 			} else {
1747 				if (fp->f_count == 0 ||
1748 				    (fp->f_flag & FMARK) != 0 ||
1749 				    fp->f_count == fp->f_msgcount ||
1750 				    fp->f_unpcount != 0) {
1751 					mutex_exit(&fp->f_lock);
1752 					continue;
1753 				}
1754 			}
1755 			atomic_or_uint(&fp->f_flag, FMARK);
1756 
1757 			if (fp->f_type != DTYPE_SOCKET ||
1758 			    (so = fp->f_socket) == NULL ||
1759 			    so->so_proto->pr_domain != &unixdomain ||
1760 			    (so->so_proto->pr_flags & PR_RIGHTS) == 0) {
1761 				mutex_exit(&fp->f_lock);
1762 				continue;
1763 			}
1764 
1765 			/* Gain file ref, mark our position, and unlock. */
1766 			didwork = true;
1767 			LIST_INSERT_AFTER(fp, dp, f_list);
1768 			fp->f_count++;
1769 			mutex_exit(&fp->f_lock);
1770 			mutex_exit(&filelist_lock);
1771 
1772 			/*
1773 			 * Mark files referenced from sockets queued on the
1774 			 * accept queue as well.
1775 			 */
1776 			solock(so);
1777 			unp_scan(so->so_rcv.sb_mb, unp_mark, 0);
1778 			if ((so->so_options & SO_ACCEPTCONN) != 0) {
1779 				TAILQ_FOREACH(so1, &so->so_q0, so_qe) {
1780 					unp_scan(so1->so_rcv.sb_mb, unp_mark, 0);
1781 				}
1782 				TAILQ_FOREACH(so1, &so->so_q, so_qe) {
1783 					unp_scan(so1->so_rcv.sb_mb, unp_mark, 0);
1784 				}
1785 			}
1786 			sounlock(so);
1787 
1788 			/* Re-lock and restart from where we left off. */
1789 			closef(fp);
1790 			mutex_enter(&filelist_lock);
1791 			np = LIST_NEXT(dp, f_list);
1792 			LIST_REMOVE(dp, f_list);
1793 		}
1794 		/*
1795 		 * Bail early if we did nothing in the loop above.  Could
1796 		 * happen because of concurrent activity causing unp_defer
1797 		 * to get out of sync.
1798 		 */
1799 	} while (unp_defer != 0 && didwork);
1800 
1801 	/*
1802 	 * Sweep pass.
1803 	 *
1804 	 * We grab an extra reference to each of the files that are
1805 	 * not otherwise accessible and then free the rights that are
1806 	 * stored in messages on them.
1807 	 */
1808 	for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) {
1809 		KASSERT(mutex_owned(&filelist_lock));
1810 		np = LIST_NEXT(fp, f_list);
1811 		mutex_enter(&fp->f_lock);
1812 
1813 		/*
1814 		 * Ignore non-sockets.
1815 		 * Ignore dead sockets, or sockets with pending close.
1816 		 * Ignore sockets obviously referenced elsewhere.
1817 		 * Ignore sockets marked as referenced by our scan.
1818 		 * Ignore new sockets that did not exist during the scan.
1819 		 */
1820 		if (fp->f_type != DTYPE_SOCKET ||
1821 		    fp->f_count == 0 || fp->f_unpcount != 0 ||
1822 		    fp->f_count != fp->f_msgcount ||
1823 		    (fp->f_flag & (FMARK | FSCAN)) != FSCAN) {
1824 			mutex_exit(&fp->f_lock);
1825 			continue;
1826 		}
1827 
1828 		/* Gain file ref, mark our position, and unlock. */
1829 		LIST_INSERT_AFTER(fp, dp, f_list);
1830 		fp->f_count++;
1831 		mutex_exit(&fp->f_lock);
1832 		mutex_exit(&filelist_lock);
1833 
1834 		/*
1835 		 * Flush all data from the socket's receive buffer.
1836 		 * This will cause files referenced only by the
1837 		 * socket to be queued for close.
1838 		 */
1839 		so = fp->f_socket;
1840 		solock(so);
1841 		sorflush(so);
1842 		sounlock(so);
1843 
1844 		/* Re-lock and restart from where we left off. */
1845 		closef(fp);
1846 		mutex_enter(&filelist_lock);
1847 		np = LIST_NEXT(dp, f_list);
1848 		LIST_REMOVE(dp, f_list);
1849 	}
1850 }
1851 
1852 /*
1853  * Garbage collector thread.  While SCM_RIGHTS messages are in transit,
1854  * wake once per second to garbage collect.  Run continually while we
1855  * have deferred closes to process.
1856  */
1857 static void
1858 unp_thread(void *cookie)
1859 {
1860 	file_t *dp;
1861 
1862 	/* Allocate a dummy file for our scans. */
1863 	if ((dp = fgetdummy()) == NULL) {
1864 		panic("unp_thread");
1865 	}
1866 
1867 	mutex_enter(&filelist_lock);
1868 	for (;;) {
1869 		KASSERT(mutex_owned(&filelist_lock));
1870 		if (SLIST_EMPTY(&unp_thread_discard)) {
1871 			if (unp_rights != 0) {
1872 				(void)cv_timedwait(&unp_thread_cv,
1873 				    &filelist_lock, hz);
1874 			} else {
1875 				cv_wait(&unp_thread_cv, &filelist_lock);
1876 			}
1877 		}
1878 		unp_gc(dp);
1879 	}
1880 	/* NOTREACHED */
1881 }
1882 
1883 /*
1884  * Kick the garbage collector into action if there is something for
1885  * it to process.
1886  */
1887 static void
1888 unp_thread_kick(void)
1889 {
1890 
1891 	if (!SLIST_EMPTY(&unp_thread_discard) || unp_rights != 0) {
1892 		mutex_enter(&filelist_lock);
1893 		cv_signal(&unp_thread_cv);
1894 		mutex_exit(&filelist_lock);
1895 	}
1896 }
1897 
1898 void
1899 unp_dispose(struct mbuf *m)
1900 {
1901 
1902 	if (m)
1903 		unp_scan(m, unp_discard_later, 1);
1904 }
1905 
1906 void
1907 unp_scan(struct mbuf *m0, void (*op)(file_t *), int discard)
1908 {
1909 	struct mbuf *m;
1910 	file_t **rp, *fp;
1911 	struct cmsghdr *cm;
1912 	int i, qfds;
1913 
1914 	while (m0) {
1915 		for (m = m0; m; m = m->m_next) {
1916 			if (m->m_type != MT_CONTROL ||
1917 			    m->m_len < sizeof(*cm)) {
1918 				continue;
1919 			}
1920 			cm = mtod(m, struct cmsghdr *);
1921 			if (cm->cmsg_level != SOL_SOCKET ||
1922 			    cm->cmsg_type != SCM_RIGHTS)
1923 				continue;
1924 			qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm)))
1925 			    / sizeof(file_t *);
1926 			rp = (file_t **)CMSG_DATA(cm);
1927 			for (i = 0; i < qfds; i++) {
1928 				fp = *rp;
1929 				if (discard) {
1930 					*rp = 0;
1931 				}
1932 				(*op)(fp);
1933 				rp++;
1934 			}
1935 		}
1936 		m0 = m0->m_nextpkt;
1937 	}
1938 }
1939 
1940 void
1941 unp_mark(file_t *fp)
1942 {
1943 
1944 	if (fp == NULL)
1945 		return;
1946 
1947 	/* If we're already deferred, don't screw up the defer count */
1948 	mutex_enter(&fp->f_lock);
1949 	if (fp->f_flag & (FMARK | FDEFER)) {
1950 		mutex_exit(&fp->f_lock);
1951 		return;
1952 	}
1953 
1954 	/*
1955 	 * Minimize the number of deferrals...  Sockets are the only type of
1956 	 * file which can hold references to another file, so just mark
1957 	 * other files, and defer unmarked sockets for the next pass.
1958 	 */
1959 	if (fp->f_type == DTYPE_SOCKET) {
1960 		unp_defer++;
1961 		KASSERT(fp->f_count != 0);
1962 		atomic_or_uint(&fp->f_flag, FDEFER);
1963 	} else {
1964 		atomic_or_uint(&fp->f_flag, FMARK);
1965 	}
1966 	mutex_exit(&fp->f_lock);
1967 }
1968 
1969 static void
1970 unp_discard_now(file_t *fp)
1971 {
1972 
1973 	if (fp == NULL)
1974 		return;
1975 
1976 	KASSERT(fp->f_count > 0);
1977 	KASSERT(fp->f_msgcount > 0);
1978 
1979 	mutex_enter(&fp->f_lock);
1980 	fp->f_msgcount--;
1981 	mutex_exit(&fp->f_lock);
1982 	atomic_dec_uint(&unp_rights);
1983 	(void)closef(fp);
1984 }
1985 
1986 static void
1987 unp_discard_later(file_t *fp)
1988 {
1989 
1990 	if (fp == NULL)
1991 		return;
1992 
1993 	KASSERT(fp->f_count > 0);
1994 	KASSERT(fp->f_msgcount > 0);
1995 
1996 	mutex_enter(&filelist_lock);
1997 	if (fp->f_unpcount++ == 0) {
1998 		SLIST_INSERT_HEAD(&unp_thread_discard, fp, f_unplist);
1999 	}
2000 	mutex_exit(&filelist_lock);
2001 }
2002 
2003 static void
2004 unp_sysctl_create(void)
2005 {
2006 
2007 	KASSERT(usrreq_sysctllog == NULL);
2008 	sysctl_createv(&usrreq_sysctllog, 0, NULL, NULL,
2009 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2010 		       CTLTYPE_LONG, "sendspace",
2011 		       SYSCTL_DESCR("Default stream send space"),
2012 		       NULL, 0, &unpst_sendspace, 0,
2013 		       CTL_NET, PF_LOCAL, SOCK_STREAM, CTL_CREATE, CTL_EOL);
2014 	sysctl_createv(&usrreq_sysctllog, 0, NULL, NULL,
2015 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2016 		       CTLTYPE_LONG, "recvspace",
2017 		       SYSCTL_DESCR("Default stream recv space"),
2018 		       NULL, 0, &unpst_recvspace, 0,
2019 		       CTL_NET, PF_LOCAL, SOCK_STREAM, CTL_CREATE, CTL_EOL);
2020 	sysctl_createv(&usrreq_sysctllog, 0, NULL, NULL,
2021 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2022 		       CTLTYPE_LONG, "sendspace",
2023 		       SYSCTL_DESCR("Default datagram send space"),
2024 		       NULL, 0, &unpdg_sendspace, 0,
2025 		       CTL_NET, PF_LOCAL, SOCK_DGRAM, CTL_CREATE, CTL_EOL);
2026 	sysctl_createv(&usrreq_sysctllog, 0, NULL, NULL,
2027 		       CTLFLAG_PERMANENT|CTLFLAG_READWRITE,
2028 		       CTLTYPE_LONG, "recvspace",
2029 		       SYSCTL_DESCR("Default datagram recv space"),
2030 		       NULL, 0, &unpdg_recvspace, 0,
2031 		       CTL_NET, PF_LOCAL, SOCK_DGRAM, CTL_CREATE, CTL_EOL);
2032 	sysctl_createv(&usrreq_sysctllog, 0, NULL, NULL,
2033 		       CTLFLAG_PERMANENT|CTLFLAG_READONLY,
2034 		       CTLTYPE_INT, "inflight",
2035 		       SYSCTL_DESCR("File descriptors in flight"),
2036 		       NULL, 0, &unp_rights, 0,
2037 		       CTL_NET, PF_LOCAL, CTL_CREATE, CTL_EOL);
2038 	sysctl_createv(&usrreq_sysctllog, 0, NULL, NULL,
2039 		       CTLFLAG_PERMANENT|CTLFLAG_READONLY,
2040 		       CTLTYPE_INT, "deferred",
2041 		       SYSCTL_DESCR("File descriptors deferred for close"),
2042 		       NULL, 0, &unp_defer, 0,
2043 		       CTL_NET, PF_LOCAL, CTL_CREATE, CTL_EOL);
2044 }
2045 
2046 const struct pr_usrreqs unp_usrreqs = {
2047 	.pr_attach	= unp_attach,
2048 	.pr_detach	= unp_detach,
2049 	.pr_accept	= unp_accept,
2050 	.pr_bind	= unp_bind,
2051 	.pr_listen	= unp_listen,
2052 	.pr_connect	= unp_connect,
2053 	.pr_connect2	= unp_connect2,
2054 	.pr_disconnect	= unp_disconnect,
2055 	.pr_shutdown	= unp_shutdown,
2056 	.pr_abort	= unp_abort,
2057 	.pr_ioctl	= unp_ioctl,
2058 	.pr_stat	= unp_stat,
2059 	.pr_peeraddr	= unp_peeraddr,
2060 	.pr_sockaddr	= unp_sockaddr,
2061 	.pr_rcvd	= unp_rcvd,
2062 	.pr_recvoob	= unp_recvoob,
2063 	.pr_send	= unp_send,
2064 	.pr_sendoob	= unp_sendoob,
2065 };
2066