xref: /freebsd/sys/kern/uipc_usrreq.c (revision c99b67a7947ea215f9c1d44ec022680e98920cd1)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.
4  * Copyright (c) 2004-2009 Robert N. M. Watson
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  * 3. Neither the name of the University nor the names of its contributors
16  *    may be used to endorse or promote products derived from this software
17  *    without specific prior written permission.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  *	From: @(#)uipc_usrreq.c	8.3 (Berkeley) 1/4/94
32  */
33 
34 /*
35  * UNIX Domain (Local) Sockets
36  *
37  * This is an implementation of UNIX (local) domain sockets.  Each socket has
38  * an associated struct unpcb (UNIX protocol control block).  Stream sockets
39  * may be connected to 0 or 1 other socket.  Datagram sockets may be
40  * connected to 0, 1, or many other sockets.  Sockets may be created and
41  * connected in pairs (socketpair(2)), or bound/connected to using the file
42  * system name space.  For most purposes, only the receive socket buffer is
43  * used, as sending on one socket delivers directly to the receive socket
44  * buffer of a second socket.
45  *
46  * The implementation is substantially complicated by the fact that
47  * "ancillary data", such as file descriptors or credentials, may be passed
48  * across UNIX domain sockets.  The potential for passing UNIX domain sockets
49  * over other UNIX domain sockets requires the implementation of a simple
50  * garbage collector to find and tear down cycles of disconnected sockets.
51  *
52  * TODO:
53  *	RDM
54  *	rethink name space problems
55  *	need a proper out-of-band
56  */
57 
58 #include <sys/cdefs.h>
59 __FBSDID("$FreeBSD$");
60 
61 #include "opt_ddb.h"
62 
63 #include <sys/param.h>
64 #include <sys/capsicum.h>
65 #include <sys/domain.h>
66 #include <sys/fcntl.h>
67 #include <sys/malloc.h>		/* XXX must be before <sys/file.h> */
68 #include <sys/eventhandler.h>
69 #include <sys/file.h>
70 #include <sys/filedesc.h>
71 #include <sys/kernel.h>
72 #include <sys/lock.h>
73 #include <sys/mbuf.h>
74 #include <sys/mount.h>
75 #include <sys/mutex.h>
76 #include <sys/namei.h>
77 #include <sys/proc.h>
78 #include <sys/protosw.h>
79 #include <sys/queue.h>
80 #include <sys/resourcevar.h>
81 #include <sys/rwlock.h>
82 #include <sys/socket.h>
83 #include <sys/socketvar.h>
84 #include <sys/signalvar.h>
85 #include <sys/stat.h>
86 #include <sys/sx.h>
87 #include <sys/sysctl.h>
88 #include <sys/systm.h>
89 #include <sys/taskqueue.h>
90 #include <sys/un.h>
91 #include <sys/unpcb.h>
92 #include <sys/vnode.h>
93 
94 #include <net/vnet.h>
95 
96 #ifdef DDB
97 #include <ddb/ddb.h>
98 #endif
99 
100 #include <security/mac/mac_framework.h>
101 
102 #include <vm/uma.h>
103 
104 MALLOC_DECLARE(M_FILECAPS);
105 
106 /*
107  * Locking key:
108  * (l)	Locked using list lock
109  * (g)	Locked using linkage lock
110  */
111 
112 static uma_zone_t	unp_zone;
113 static unp_gen_t	unp_gencnt;	/* (l) */
114 static u_int		unp_count;	/* (l) Count of local sockets. */
115 static ino_t		unp_ino;	/* Prototype for fake inode numbers. */
116 static int		unp_rights;	/* (g) File descriptors in flight. */
117 static struct unp_head	unp_shead;	/* (l) List of stream sockets. */
118 static struct unp_head	unp_dhead;	/* (l) List of datagram sockets. */
119 static struct unp_head	unp_sphead;	/* (l) List of seqpacket sockets. */
120 
121 struct unp_defer {
122 	SLIST_ENTRY(unp_defer) ud_link;
123 	struct file *ud_fp;
124 };
125 static SLIST_HEAD(, unp_defer) unp_defers;
126 static int unp_defers_count;
127 
128 static const struct sockaddr	sun_noname = { sizeof(sun_noname), AF_LOCAL };
129 
130 /*
131  * Garbage collection of cyclic file descriptor/socket references occurs
132  * asynchronously in a taskqueue context in order to avoid recursion and
133  * reentrance in the UNIX domain socket, file descriptor, and socket layer
134  * code.  See unp_gc() for a full description.
135  */
136 static struct timeout_task unp_gc_task;
137 
138 /*
139  * The close of unix domain sockets attached as SCM_RIGHTS is
140  * postponed to the taskqueue, to avoid arbitrary recursion depth.
141  * The attached sockets might have another sockets attached.
142  */
143 static struct task	unp_defer_task;
144 
145 /*
146  * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
147  * stream sockets, although the total for sender and receiver is actually
148  * only PIPSIZ.
149  *
150  * Datagram sockets really use the sendspace as the maximum datagram size,
151  * and don't really want to reserve the sendspace.  Their recvspace should be
152  * large enough for at least one max-size datagram plus address.
153  */
154 #ifndef PIPSIZ
155 #define	PIPSIZ	8192
156 #endif
157 static u_long	unpst_sendspace = PIPSIZ;
158 static u_long	unpst_recvspace = PIPSIZ;
159 static u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
160 static u_long	unpdg_recvspace = 4*1024;
161 static u_long	unpsp_sendspace = PIPSIZ;	/* really max datagram size */
162 static u_long	unpsp_recvspace = PIPSIZ;
163 
164 static SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
165 static SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0,
166     "SOCK_STREAM");
167 static SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
168 static SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
169     "SOCK_SEQPACKET");
170 
171 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
172 	   &unpst_sendspace, 0, "Default stream send space.");
173 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
174 	   &unpst_recvspace, 0, "Default stream receive space.");
175 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
176 	   &unpdg_sendspace, 0, "Default datagram send space.");
177 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
178 	   &unpdg_recvspace, 0, "Default datagram receive space.");
179 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
180 	   &unpsp_sendspace, 0, "Default seqpacket send space.");
181 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
182 	   &unpsp_recvspace, 0, "Default seqpacket receive space.");
183 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
184     "File descriptors in flight.");
185 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
186     &unp_defers_count, 0,
187     "File descriptors deferred to taskqueue for close.");
188 
189 /*
190  * Locking and synchronization:
191  *
192  * Two types of locks exist in the local domain socket implementation: a
193  * a global linkage rwlock and per-unpcb mutexes.  The linkage lock protects
194  * the socket count, global generation number, stream/datagram global lists and
195  * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be
196  * held exclusively over the acquisition of multiple unpcb locks to prevent
197  * deadlock.
198  *
199  * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
200  * allocated in pru_attach() and freed in pru_detach().  The validity of that
201  * pointer is an invariant, so no lock is required to dereference the so_pcb
202  * pointer if a valid socket reference is held by the caller.  In practice,
203  * this is always true during operations performed on a socket.  Each unpcb
204  * has a back-pointer to its socket, unp_socket, which will be stable under
205  * the same circumstances.
206  *
207  * This pointer may only be safely dereferenced as long as a valid reference
208  * to the unpcb is held.  Typically, this reference will be from the socket,
209  * or from another unpcb when the referring unpcb's lock is held (in order
210  * that the reference not be invalidated during use).  For example, to follow
211  * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
212  * as unp_socket remains valid as long as the reference to unp_conn is valid.
213  *
214  * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx.  Individual
215  * atomic reads without the lock may be performed "lockless", but more
216  * complex reads and read-modify-writes require the mutex to be held.  No
217  * lock order is defined between unpcb locks -- multiple unpcb locks may be
218  * acquired at the same time only when holding the linkage rwlock
219  * exclusively, which prevents deadlocks.
220  *
221  * Blocking with UNIX domain sockets is a tricky issue: unlike most network
222  * protocols, bind() is a non-atomic operation, and connect() requires
223  * potential sleeping in the protocol, due to potentially waiting on local or
224  * distributed file systems.  We try to separate "lookup" operations, which
225  * may sleep, and the IPC operations themselves, which typically can occur
226  * with relative atomicity as locks can be held over the entire operation.
227  *
228  * Another tricky issue is simultaneous multi-threaded or multi-process
229  * access to a single UNIX domain socket.  These are handled by the flags
230  * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
231  * binding, both of which involve dropping UNIX domain socket locks in order
232  * to perform namei() and other file system operations.
233  */
234 static struct rwlock	unp_link_rwlock;
235 static struct mtx	unp_defers_lock;
236 
237 #define	UNP_LINK_LOCK_INIT()		rw_init(&unp_link_rwlock,	\
238 					    "unp_link_rwlock")
239 
240 #define	UNP_LINK_LOCK_ASSERT()	rw_assert(&unp_link_rwlock,	\
241 					    RA_LOCKED)
242 #define	UNP_LINK_UNLOCK_ASSERT()	rw_assert(&unp_link_rwlock,	\
243 					    RA_UNLOCKED)
244 
245 #define	UNP_LINK_RLOCK()		rw_rlock(&unp_link_rwlock)
246 #define	UNP_LINK_RUNLOCK()		rw_runlock(&unp_link_rwlock)
247 #define	UNP_LINK_WLOCK()		rw_wlock(&unp_link_rwlock)
248 #define	UNP_LINK_WUNLOCK()		rw_wunlock(&unp_link_rwlock)
249 #define	UNP_LINK_WLOCK_ASSERT()		rw_assert(&unp_link_rwlock,	\
250 					    RA_WLOCKED)
251 #define	UNP_LINK_WOWNED()		rw_wowned(&unp_link_rwlock)
252 
253 #define	UNP_DEFERRED_LOCK_INIT()	mtx_init(&unp_defers_lock, \
254 					    "unp_defer", NULL, MTX_DEF)
255 #define	UNP_DEFERRED_LOCK()		mtx_lock(&unp_defers_lock)
256 #define	UNP_DEFERRED_UNLOCK()		mtx_unlock(&unp_defers_lock)
257 
258 #define UNP_PCB_LOCK_INIT(unp)		mtx_init(&(unp)->unp_mtx,	\
259 					    "unp_mtx", "unp_mtx",	\
260 					    MTX_DUPOK|MTX_DEF|MTX_RECURSE)
261 #define	UNP_PCB_LOCK_DESTROY(unp)	mtx_destroy(&(unp)->unp_mtx)
262 #define	UNP_PCB_LOCK(unp)		mtx_lock(&(unp)->unp_mtx)
263 #define	UNP_PCB_UNLOCK(unp)		mtx_unlock(&(unp)->unp_mtx)
264 #define	UNP_PCB_LOCK_ASSERT(unp)	mtx_assert(&(unp)->unp_mtx, MA_OWNED)
265 
266 static int	uipc_connect2(struct socket *, struct socket *);
267 static int	uipc_ctloutput(struct socket *, struct sockopt *);
268 static int	unp_connect(struct socket *, struct sockaddr *,
269 		    struct thread *);
270 static int	unp_connectat(int, struct socket *, struct sockaddr *,
271 		    struct thread *);
272 static int	unp_connect2(struct socket *so, struct socket *so2, int);
273 static void	unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
274 static void	unp_dispose(struct socket *so);
275 static void	unp_dispose_mbuf(struct mbuf *);
276 static void	unp_shutdown(struct unpcb *);
277 static void	unp_drop(struct unpcb *);
278 static void	unp_gc(__unused void *, int);
279 static void	unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
280 static void	unp_discard(struct file *);
281 static void	unp_freerights(struct filedescent **, int);
282 static void	unp_init(void);
283 static int	unp_internalize(struct mbuf **, struct thread *);
284 static void	unp_internalize_fp(struct file *);
285 static int	unp_externalize(struct mbuf *, struct mbuf **, int);
286 static int	unp_externalize_fp(struct file *);
287 static struct mbuf	*unp_addsockcred(struct thread *, struct mbuf *);
288 static void	unp_process_defers(void * __unused, int);
289 
290 /*
291  * Definitions of protocols supported in the LOCAL domain.
292  */
293 static struct domain localdomain;
294 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
295 static struct pr_usrreqs uipc_usrreqs_seqpacket;
296 static struct protosw localsw[] = {
297 {
298 	.pr_type =		SOCK_STREAM,
299 	.pr_domain =		&localdomain,
300 	.pr_flags =		PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
301 	.pr_ctloutput =		&uipc_ctloutput,
302 	.pr_usrreqs =		&uipc_usrreqs_stream
303 },
304 {
305 	.pr_type =		SOCK_DGRAM,
306 	.pr_domain =		&localdomain,
307 	.pr_flags =		PR_ATOMIC|PR_ADDR|PR_RIGHTS,
308 	.pr_ctloutput =		&uipc_ctloutput,
309 	.pr_usrreqs =		&uipc_usrreqs_dgram
310 },
311 {
312 	.pr_type =		SOCK_SEQPACKET,
313 	.pr_domain =		&localdomain,
314 
315 	/*
316 	 * XXXRW: For now, PR_ADDR because soreceive will bump into them
317 	 * due to our use of sbappendaddr.  A new sbappend variants is needed
318 	 * that supports both atomic record writes and control data.
319 	 */
320 	.pr_flags =		PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
321 				    PR_RIGHTS,
322 	.pr_ctloutput =		&uipc_ctloutput,
323 	.pr_usrreqs =		&uipc_usrreqs_seqpacket,
324 },
325 };
326 
327 static struct domain localdomain = {
328 	.dom_family =		AF_LOCAL,
329 	.dom_name =		"local",
330 	.dom_init =		unp_init,
331 	.dom_externalize =	unp_externalize,
332 	.dom_dispose =		unp_dispose,
333 	.dom_protosw =		localsw,
334 	.dom_protoswNPROTOSW =	&localsw[nitems(localsw)]
335 };
336 DOMAIN_SET(local);
337 
338 static void
339 uipc_abort(struct socket *so)
340 {
341 	struct unpcb *unp, *unp2;
342 
343 	unp = sotounpcb(so);
344 	KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
345 
346 	UNP_LINK_WLOCK();
347 	UNP_PCB_LOCK(unp);
348 	unp2 = unp->unp_conn;
349 	if (unp2 != NULL) {
350 		UNP_PCB_LOCK(unp2);
351 		unp_drop(unp2);
352 		UNP_PCB_UNLOCK(unp2);
353 	}
354 	UNP_PCB_UNLOCK(unp);
355 	UNP_LINK_WUNLOCK();
356 }
357 
358 static int
359 uipc_accept(struct socket *so, struct sockaddr **nam)
360 {
361 	struct unpcb *unp, *unp2;
362 	const struct sockaddr *sa;
363 
364 	/*
365 	 * Pass back name of connected socket, if it was bound and we are
366 	 * still connected (our peer may have closed already!).
367 	 */
368 	unp = sotounpcb(so);
369 	KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
370 
371 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
372 	UNP_LINK_RLOCK();
373 	unp2 = unp->unp_conn;
374 	if (unp2 != NULL && unp2->unp_addr != NULL) {
375 		UNP_PCB_LOCK(unp2);
376 		sa = (struct sockaddr *) unp2->unp_addr;
377 		bcopy(sa, *nam, sa->sa_len);
378 		UNP_PCB_UNLOCK(unp2);
379 	} else {
380 		sa = &sun_noname;
381 		bcopy(sa, *nam, sa->sa_len);
382 	}
383 	UNP_LINK_RUNLOCK();
384 	return (0);
385 }
386 
387 static int
388 uipc_attach(struct socket *so, int proto, struct thread *td)
389 {
390 	u_long sendspace, recvspace;
391 	struct unpcb *unp;
392 	int error;
393 	bool locked;
394 
395 	KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
396 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
397 		switch (so->so_type) {
398 		case SOCK_STREAM:
399 			sendspace = unpst_sendspace;
400 			recvspace = unpst_recvspace;
401 			break;
402 
403 		case SOCK_DGRAM:
404 			sendspace = unpdg_sendspace;
405 			recvspace = unpdg_recvspace;
406 			break;
407 
408 		case SOCK_SEQPACKET:
409 			sendspace = unpsp_sendspace;
410 			recvspace = unpsp_recvspace;
411 			break;
412 
413 		default:
414 			panic("uipc_attach");
415 		}
416 		error = soreserve(so, sendspace, recvspace);
417 		if (error)
418 			return (error);
419 	}
420 	unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
421 	if (unp == NULL)
422 		return (ENOBUFS);
423 	LIST_INIT(&unp->unp_refs);
424 	UNP_PCB_LOCK_INIT(unp);
425 	unp->unp_socket = so;
426 	so->so_pcb = unp;
427 	unp->unp_refcount = 1;
428 	if (so->so_listen != NULL)
429 		unp->unp_flags |= UNP_NASCENT;
430 
431 	if ((locked = UNP_LINK_WOWNED()) == false)
432 		UNP_LINK_WLOCK();
433 
434 	unp->unp_gencnt = ++unp_gencnt;
435 	unp_count++;
436 	switch (so->so_type) {
437 	case SOCK_STREAM:
438 		LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
439 		break;
440 
441 	case SOCK_DGRAM:
442 		LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
443 		break;
444 
445 	case SOCK_SEQPACKET:
446 		LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
447 		break;
448 
449 	default:
450 		panic("uipc_attach");
451 	}
452 
453 	if (locked == false)
454 		UNP_LINK_WUNLOCK();
455 
456 	return (0);
457 }
458 
459 static int
460 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
461 {
462 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
463 	struct vattr vattr;
464 	int error, namelen;
465 	struct nameidata nd;
466 	struct unpcb *unp;
467 	struct vnode *vp;
468 	struct mount *mp;
469 	cap_rights_t rights;
470 	char *buf;
471 
472 	if (nam->sa_family != AF_UNIX)
473 		return (EAFNOSUPPORT);
474 
475 	unp = sotounpcb(so);
476 	KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
477 
478 	if (soun->sun_len > sizeof(struct sockaddr_un))
479 		return (EINVAL);
480 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
481 	if (namelen <= 0)
482 		return (EINVAL);
483 
484 	/*
485 	 * We don't allow simultaneous bind() calls on a single UNIX domain
486 	 * socket, so flag in-progress operations, and return an error if an
487 	 * operation is already in progress.
488 	 *
489 	 * Historically, we have not allowed a socket to be rebound, so this
490 	 * also returns an error.  Not allowing re-binding simplifies the
491 	 * implementation and avoids a great many possible failure modes.
492 	 */
493 	UNP_PCB_LOCK(unp);
494 	if (unp->unp_vnode != NULL) {
495 		UNP_PCB_UNLOCK(unp);
496 		return (EINVAL);
497 	}
498 	if (unp->unp_flags & UNP_BINDING) {
499 		UNP_PCB_UNLOCK(unp);
500 		return (EALREADY);
501 	}
502 	unp->unp_flags |= UNP_BINDING;
503 	UNP_PCB_UNLOCK(unp);
504 
505 	buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
506 	bcopy(soun->sun_path, buf, namelen);
507 	buf[namelen] = 0;
508 
509 restart:
510 	NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME | NOCACHE,
511 	    UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td);
512 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
513 	error = namei(&nd);
514 	if (error)
515 		goto error;
516 	vp = nd.ni_vp;
517 	if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
518 		NDFREE(&nd, NDF_ONLY_PNBUF);
519 		if (nd.ni_dvp == vp)
520 			vrele(nd.ni_dvp);
521 		else
522 			vput(nd.ni_dvp);
523 		if (vp != NULL) {
524 			vrele(vp);
525 			error = EADDRINUSE;
526 			goto error;
527 		}
528 		error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
529 		if (error)
530 			goto error;
531 		goto restart;
532 	}
533 	VATTR_NULL(&vattr);
534 	vattr.va_type = VSOCK;
535 	vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
536 #ifdef MAC
537 	error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
538 	    &vattr);
539 #endif
540 	if (error == 0)
541 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
542 	NDFREE(&nd, NDF_ONLY_PNBUF);
543 	vput(nd.ni_dvp);
544 	if (error) {
545 		vn_finished_write(mp);
546 		goto error;
547 	}
548 	vp = nd.ni_vp;
549 	ASSERT_VOP_ELOCKED(vp, "uipc_bind");
550 	soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
551 
552 	UNP_LINK_WLOCK();
553 	UNP_PCB_LOCK(unp);
554 	VOP_UNP_BIND(vp, unp);
555 	unp->unp_vnode = vp;
556 	unp->unp_addr = soun;
557 	unp->unp_flags &= ~UNP_BINDING;
558 	UNP_PCB_UNLOCK(unp);
559 	UNP_LINK_WUNLOCK();
560 	VOP_UNLOCK(vp, 0);
561 	vn_finished_write(mp);
562 	free(buf, M_TEMP);
563 	return (0);
564 
565 error:
566 	UNP_PCB_LOCK(unp);
567 	unp->unp_flags &= ~UNP_BINDING;
568 	UNP_PCB_UNLOCK(unp);
569 	free(buf, M_TEMP);
570 	return (error);
571 }
572 
573 static int
574 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
575 {
576 
577 	return (uipc_bindat(AT_FDCWD, so, nam, td));
578 }
579 
580 static int
581 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
582 {
583 	int error;
584 
585 	KASSERT(td == curthread, ("uipc_connect: td != curthread"));
586 	UNP_LINK_WLOCK();
587 	error = unp_connect(so, nam, td);
588 	UNP_LINK_WUNLOCK();
589 	return (error);
590 }
591 
592 static int
593 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
594     struct thread *td)
595 {
596 	int error;
597 
598 	KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
599 	UNP_LINK_WLOCK();
600 	error = unp_connectat(fd, so, nam, td);
601 	UNP_LINK_WUNLOCK();
602 	return (error);
603 }
604 
605 static void
606 uipc_close(struct socket *so)
607 {
608 	struct unpcb *unp, *unp2;
609 	struct vnode *vp = NULL;
610 
611 	unp = sotounpcb(so);
612 	KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
613 
614 	UNP_LINK_WLOCK();
615 	UNP_PCB_LOCK(unp);
616 	unp2 = unp->unp_conn;
617 	if (unp2 != NULL) {
618 		UNP_PCB_LOCK(unp2);
619 		unp_disconnect(unp, unp2);
620 		UNP_PCB_UNLOCK(unp2);
621 	}
622 	if (SOLISTENING(so) && ((vp = unp->unp_vnode) != NULL)) {
623 		VOP_UNP_DETACH(vp);
624 		unp->unp_vnode = NULL;
625 	}
626 	UNP_PCB_UNLOCK(unp);
627 	UNP_LINK_WUNLOCK();
628 	if (vp)
629 		vrele(vp);
630 }
631 
632 static int
633 uipc_connect2(struct socket *so1, struct socket *so2)
634 {
635 	struct unpcb *unp, *unp2;
636 	int error;
637 
638 	UNP_LINK_WLOCK();
639 	unp = so1->so_pcb;
640 	KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
641 	UNP_PCB_LOCK(unp);
642 	unp2 = so2->so_pcb;
643 	KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
644 	UNP_PCB_LOCK(unp2);
645 	error = unp_connect2(so1, so2, PRU_CONNECT2);
646 	UNP_PCB_UNLOCK(unp2);
647 	UNP_PCB_UNLOCK(unp);
648 	UNP_LINK_WUNLOCK();
649 	return (error);
650 }
651 
652 static void
653 uipc_detach(struct socket *so)
654 {
655 	struct unpcb *unp, *unp2;
656 	struct sockaddr_un *saved_unp_addr;
657 	struct vnode *vp;
658 	int freeunp, local_unp_rights;
659 
660 	unp = sotounpcb(so);
661 	KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
662 
663 	vp = NULL;
664 	local_unp_rights = 0;
665 
666 	UNP_LINK_WLOCK();
667 	LIST_REMOVE(unp, unp_link);
668 	unp->unp_gencnt = ++unp_gencnt;
669 	--unp_count;
670 	UNP_PCB_LOCK(unp);
671 	if ((unp->unp_flags & UNP_NASCENT) != 0)
672 		goto teardown;
673 
674 	if ((vp = unp->unp_vnode) != NULL) {
675 		VOP_UNP_DETACH(vp);
676 		unp->unp_vnode = NULL;
677 	}
678 	unp2 = unp->unp_conn;
679 	if (unp2 != NULL) {
680 		UNP_PCB_LOCK(unp2);
681 		unp_disconnect(unp, unp2);
682 		UNP_PCB_UNLOCK(unp2);
683 	}
684 
685 	/*
686 	 * We hold the linkage lock exclusively, so it's OK to acquire
687 	 * multiple pcb locks at a time.
688 	 */
689 	while (!LIST_EMPTY(&unp->unp_refs)) {
690 		struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
691 
692 		UNP_PCB_LOCK(ref);
693 		unp_drop(ref);
694 		UNP_PCB_UNLOCK(ref);
695 	}
696 	local_unp_rights = unp_rights;
697 teardown:
698 	UNP_LINK_WUNLOCK();
699 	unp->unp_socket->so_pcb = NULL;
700 	saved_unp_addr = unp->unp_addr;
701 	unp->unp_addr = NULL;
702 	unp->unp_refcount--;
703 	freeunp = (unp->unp_refcount == 0);
704 	if (saved_unp_addr != NULL)
705 		free(saved_unp_addr, M_SONAME);
706 	if (freeunp) {
707 		UNP_PCB_LOCK_DESTROY(unp);
708 		uma_zfree(unp_zone, unp);
709 	} else
710 		UNP_PCB_UNLOCK(unp);
711 	if (vp)
712 		vrele(vp);
713 	if (local_unp_rights)
714 		taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
715 }
716 
717 static int
718 uipc_disconnect(struct socket *so)
719 {
720 	struct unpcb *unp, *unp2;
721 
722 	unp = sotounpcb(so);
723 	KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
724 
725 	UNP_LINK_WLOCK();
726 	UNP_PCB_LOCK(unp);
727 	unp2 = unp->unp_conn;
728 	if (unp2 != NULL) {
729 		UNP_PCB_LOCK(unp2);
730 		unp_disconnect(unp, unp2);
731 		UNP_PCB_UNLOCK(unp2);
732 	}
733 	UNP_PCB_UNLOCK(unp);
734 	UNP_LINK_WUNLOCK();
735 	return (0);
736 }
737 
738 static int
739 uipc_listen(struct socket *so, int backlog, struct thread *td)
740 {
741 	struct unpcb *unp;
742 	int error;
743 
744 	if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
745 		return (EOPNOTSUPP);
746 
747 	unp = sotounpcb(so);
748 	KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
749 
750 	UNP_PCB_LOCK(unp);
751 	if (unp->unp_vnode == NULL) {
752 		/* Already connected or not bound to an address. */
753 		error = unp->unp_conn != NULL ? EINVAL : EDESTADDRREQ;
754 		UNP_PCB_UNLOCK(unp);
755 		return (error);
756 	}
757 
758 	SOCK_LOCK(so);
759 	error = solisten_proto_check(so);
760 	if (error == 0) {
761 		cru2x(td->td_ucred, &unp->unp_peercred);
762 		solisten_proto(so, backlog);
763 	}
764 	SOCK_UNLOCK(so);
765 	UNP_PCB_UNLOCK(unp);
766 	return (error);
767 }
768 
769 static int
770 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
771 {
772 	struct unpcb *unp, *unp2;
773 	const struct sockaddr *sa;
774 
775 	unp = sotounpcb(so);
776 	KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
777 
778 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
779 	UNP_LINK_RLOCK();
780 	/*
781 	 * XXX: It seems that this test always fails even when connection is
782 	 * established.  So, this else clause is added as workaround to
783 	 * return PF_LOCAL sockaddr.
784 	 */
785 	unp2 = unp->unp_conn;
786 	if (unp2 != NULL) {
787 		UNP_PCB_LOCK(unp2);
788 		if (unp2->unp_addr != NULL)
789 			sa = (struct sockaddr *) unp2->unp_addr;
790 		else
791 			sa = &sun_noname;
792 		bcopy(sa, *nam, sa->sa_len);
793 		UNP_PCB_UNLOCK(unp2);
794 	} else {
795 		sa = &sun_noname;
796 		bcopy(sa, *nam, sa->sa_len);
797 	}
798 	UNP_LINK_RUNLOCK();
799 	return (0);
800 }
801 
802 static int
803 uipc_rcvd(struct socket *so, int flags)
804 {
805 	struct unpcb *unp, *unp2;
806 	struct socket *so2;
807 	u_int mbcnt, sbcc;
808 
809 	unp = sotounpcb(so);
810 	KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
811 	KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
812 	    ("%s: socktype %d", __func__, so->so_type));
813 
814 	/*
815 	 * Adjust backpressure on sender and wakeup any waiting to write.
816 	 *
817 	 * The unp lock is acquired to maintain the validity of the unp_conn
818 	 * pointer; no lock on unp2 is required as unp2->unp_socket will be
819 	 * static as long as we don't permit unp2 to disconnect from unp,
820 	 * which is prevented by the lock on unp.  We cache values from
821 	 * so_rcv to avoid holding the so_rcv lock over the entire
822 	 * transaction on the remote so_snd.
823 	 */
824 	SOCKBUF_LOCK(&so->so_rcv);
825 	mbcnt = so->so_rcv.sb_mbcnt;
826 	sbcc = sbavail(&so->so_rcv);
827 	SOCKBUF_UNLOCK(&so->so_rcv);
828 	/*
829 	 * There is a benign race condition at this point.  If we're planning to
830 	 * clear SB_STOP, but uipc_send is called on the connected socket at
831 	 * this instant, it might add data to the sockbuf and set SB_STOP.  Then
832 	 * we would erroneously clear SB_STOP below, even though the sockbuf is
833 	 * full.  The race is benign because the only ill effect is to allow the
834 	 * sockbuf to exceed its size limit, and the size limits are not
835 	 * strictly guaranteed anyway.
836 	 */
837 	UNP_PCB_LOCK(unp);
838 	unp2 = unp->unp_conn;
839 	if (unp2 == NULL) {
840 		UNP_PCB_UNLOCK(unp);
841 		return (0);
842 	}
843 	so2 = unp2->unp_socket;
844 	SOCKBUF_LOCK(&so2->so_snd);
845 	if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
846 		so2->so_snd.sb_flags &= ~SB_STOP;
847 	sowwakeup_locked(so2);
848 	UNP_PCB_UNLOCK(unp);
849 	return (0);
850 }
851 
852 static int
853 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
854     struct mbuf *control, struct thread *td)
855 {
856 	struct unpcb *unp, *unp2;
857 	struct socket *so2;
858 	u_int mbcnt, sbcc;
859 	int error = 0;
860 
861 	unp = sotounpcb(so);
862 	KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
863 	KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
864 	    so->so_type == SOCK_SEQPACKET,
865 	    ("%s: socktype %d", __func__, so->so_type));
866 
867 	if (flags & PRUS_OOB) {
868 		error = EOPNOTSUPP;
869 		goto release;
870 	}
871 	if (control != NULL && (error = unp_internalize(&control, td)))
872 		goto release;
873 	if ((nam != NULL) || (flags & PRUS_EOF))
874 		UNP_LINK_WLOCK();
875 	else
876 		UNP_LINK_RLOCK();
877 	switch (so->so_type) {
878 	case SOCK_DGRAM:
879 	{
880 		const struct sockaddr *from;
881 
882 		unp2 = unp->unp_conn;
883 		if (nam != NULL) {
884 			UNP_LINK_WLOCK_ASSERT();
885 			if (unp2 != NULL) {
886 				error = EISCONN;
887 				break;
888 			}
889 			error = unp_connect(so, nam, td);
890 			if (error)
891 				break;
892 			unp2 = unp->unp_conn;
893 		}
894 
895 		/*
896 		 * Because connect() and send() are non-atomic in a sendto()
897 		 * with a target address, it's possible that the socket will
898 		 * have disconnected before the send() can run.  In that case
899 		 * return the slightly counter-intuitive but otherwise
900 		 * correct error that the socket is not connected.
901 		 */
902 		if (unp2 == NULL) {
903 			error = ENOTCONN;
904 			break;
905 		}
906 		/* Lockless read. */
907 		if (unp2->unp_flags & UNP_WANTCRED)
908 			control = unp_addsockcred(td, control);
909 		UNP_PCB_LOCK(unp);
910 		if (unp->unp_addr != NULL)
911 			from = (struct sockaddr *)unp->unp_addr;
912 		else
913 			from = &sun_noname;
914 		so2 = unp2->unp_socket;
915 		SOCKBUF_LOCK(&so2->so_rcv);
916 		if (sbappendaddr_locked(&so2->so_rcv, from, m,
917 		    control)) {
918 			sorwakeup_locked(so2);
919 			m = NULL;
920 			control = NULL;
921 		} else {
922 			SOCKBUF_UNLOCK(&so2->so_rcv);
923 			error = ENOBUFS;
924 		}
925 		if (nam != NULL) {
926 			UNP_LINK_WLOCK_ASSERT();
927 			UNP_PCB_LOCK(unp2);
928 			unp_disconnect(unp, unp2);
929 			UNP_PCB_UNLOCK(unp2);
930 		}
931 		UNP_PCB_UNLOCK(unp);
932 		break;
933 	}
934 
935 	case SOCK_SEQPACKET:
936 	case SOCK_STREAM:
937 		if ((so->so_state & SS_ISCONNECTED) == 0) {
938 			if (nam != NULL) {
939 				UNP_LINK_WLOCK_ASSERT();
940 				error = unp_connect(so, nam, td);
941 				if (error)
942 					break;	/* XXX */
943 			} else {
944 				error = ENOTCONN;
945 				break;
946 			}
947 		}
948 
949 		/* Lockless read. */
950 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
951 			error = EPIPE;
952 			break;
953 		}
954 
955 		/*
956 		 * Because connect() and send() are non-atomic in a sendto()
957 		 * with a target address, it's possible that the socket will
958 		 * have disconnected before the send() can run.  In that case
959 		 * return the slightly counter-intuitive but otherwise
960 		 * correct error that the socket is not connected.
961 		 *
962 		 * Locking here must be done carefully: the linkage lock
963 		 * prevents interconnections between unpcbs from changing, so
964 		 * we can traverse from unp to unp2 without acquiring unp's
965 		 * lock.  Socket buffer locks follow unpcb locks, so we can
966 		 * acquire both remote and lock socket buffer locks.
967 		 */
968 		unp2 = unp->unp_conn;
969 		if (unp2 == NULL) {
970 			error = ENOTCONN;
971 			break;
972 		}
973 		so2 = unp2->unp_socket;
974 		UNP_PCB_LOCK(unp2);
975 		SOCKBUF_LOCK(&so2->so_rcv);
976 		if (unp2->unp_flags & UNP_WANTCRED) {
977 			/*
978 			 * Credentials are passed only once on SOCK_STREAM
979 			 * and SOCK_SEQPACKET.
980 			 */
981 			unp2->unp_flags &= ~UNP_WANTCRED;
982 			control = unp_addsockcred(td, control);
983 		}
984 		/*
985 		 * Send to paired receive port, and then reduce send buffer
986 		 * hiwater marks to maintain backpressure.  Wake up readers.
987 		 */
988 		switch (so->so_type) {
989 		case SOCK_STREAM:
990 			if (control != NULL) {
991 				if (sbappendcontrol_locked(&so2->so_rcv, m,
992 				    control))
993 					control = NULL;
994 			} else
995 				sbappend_locked(&so2->so_rcv, m, flags);
996 			break;
997 
998 		case SOCK_SEQPACKET: {
999 			const struct sockaddr *from;
1000 
1001 			from = &sun_noname;
1002 			/*
1003 			 * Don't check for space available in so2->so_rcv.
1004 			 * Unix domain sockets only check for space in the
1005 			 * sending sockbuf, and that check is performed one
1006 			 * level up the stack.
1007 			 */
1008 			if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
1009 				from, m, control))
1010 				control = NULL;
1011 			break;
1012 			}
1013 		}
1014 
1015 		mbcnt = so2->so_rcv.sb_mbcnt;
1016 		sbcc = sbavail(&so2->so_rcv);
1017 		if (sbcc)
1018 			sorwakeup_locked(so2);
1019 		else
1020 			SOCKBUF_UNLOCK(&so2->so_rcv);
1021 
1022 		/*
1023 		 * The PCB lock on unp2 protects the SB_STOP flag.  Without it,
1024 		 * it would be possible for uipc_rcvd to be called at this
1025 		 * point, drain the receiving sockbuf, clear SB_STOP, and then
1026 		 * we would set SB_STOP below.  That could lead to an empty
1027 		 * sockbuf having SB_STOP set
1028 		 */
1029 		SOCKBUF_LOCK(&so->so_snd);
1030 		if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1031 			so->so_snd.sb_flags |= SB_STOP;
1032 		SOCKBUF_UNLOCK(&so->so_snd);
1033 		UNP_PCB_UNLOCK(unp2);
1034 		m = NULL;
1035 		break;
1036 	}
1037 
1038 	/*
1039 	 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1040 	 */
1041 	if (flags & PRUS_EOF) {
1042 		UNP_PCB_LOCK(unp);
1043 		socantsendmore(so);
1044 		unp_shutdown(unp);
1045 		UNP_PCB_UNLOCK(unp);
1046 	}
1047 
1048 	if ((nam != NULL) || (flags & PRUS_EOF))
1049 		UNP_LINK_WUNLOCK();
1050 	else
1051 		UNP_LINK_RUNLOCK();
1052 
1053 	if (control != NULL && error != 0)
1054 		unp_dispose_mbuf(control);
1055 
1056 release:
1057 	if (control != NULL)
1058 		m_freem(control);
1059 	if (m != NULL)
1060 		m_freem(m);
1061 	return (error);
1062 }
1063 
1064 static int
1065 uipc_ready(struct socket *so, struct mbuf *m, int count)
1066 {
1067 	struct unpcb *unp, *unp2;
1068 	struct socket *so2;
1069 	int error;
1070 
1071 	unp = sotounpcb(so);
1072 
1073 	UNP_LINK_RLOCK();
1074 	unp2 = unp->unp_conn;
1075 	UNP_PCB_LOCK(unp2);
1076 	so2 = unp2->unp_socket;
1077 
1078 	SOCKBUF_LOCK(&so2->so_rcv);
1079 	if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1080 		sorwakeup_locked(so2);
1081 	else
1082 		SOCKBUF_UNLOCK(&so2->so_rcv);
1083 
1084 	UNP_PCB_UNLOCK(unp2);
1085 	UNP_LINK_RUNLOCK();
1086 
1087 	return (error);
1088 }
1089 
1090 static int
1091 uipc_sense(struct socket *so, struct stat *sb)
1092 {
1093 	struct unpcb *unp;
1094 
1095 	unp = sotounpcb(so);
1096 	KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1097 
1098 	sb->st_blksize = so->so_snd.sb_hiwat;
1099 	UNP_PCB_LOCK(unp);
1100 	sb->st_dev = NODEV;
1101 	if (unp->unp_ino == 0)
1102 		unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1103 	sb->st_ino = unp->unp_ino;
1104 	UNP_PCB_UNLOCK(unp);
1105 	return (0);
1106 }
1107 
1108 static int
1109 uipc_shutdown(struct socket *so)
1110 {
1111 	struct unpcb *unp;
1112 
1113 	unp = sotounpcb(so);
1114 	KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1115 
1116 	UNP_LINK_WLOCK();
1117 	UNP_PCB_LOCK(unp);
1118 	socantsendmore(so);
1119 	unp_shutdown(unp);
1120 	UNP_PCB_UNLOCK(unp);
1121 	UNP_LINK_WUNLOCK();
1122 	return (0);
1123 }
1124 
1125 static int
1126 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1127 {
1128 	struct unpcb *unp;
1129 	const struct sockaddr *sa;
1130 
1131 	unp = sotounpcb(so);
1132 	KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1133 
1134 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1135 	UNP_PCB_LOCK(unp);
1136 	if (unp->unp_addr != NULL)
1137 		sa = (struct sockaddr *) unp->unp_addr;
1138 	else
1139 		sa = &sun_noname;
1140 	bcopy(sa, *nam, sa->sa_len);
1141 	UNP_PCB_UNLOCK(unp);
1142 	return (0);
1143 }
1144 
1145 static struct pr_usrreqs uipc_usrreqs_dgram = {
1146 	.pru_abort = 		uipc_abort,
1147 	.pru_accept =		uipc_accept,
1148 	.pru_attach =		uipc_attach,
1149 	.pru_bind =		uipc_bind,
1150 	.pru_bindat =		uipc_bindat,
1151 	.pru_connect =		uipc_connect,
1152 	.pru_connectat =	uipc_connectat,
1153 	.pru_connect2 =		uipc_connect2,
1154 	.pru_detach =		uipc_detach,
1155 	.pru_disconnect =	uipc_disconnect,
1156 	.pru_listen =		uipc_listen,
1157 	.pru_peeraddr =		uipc_peeraddr,
1158 	.pru_rcvd =		uipc_rcvd,
1159 	.pru_send =		uipc_send,
1160 	.pru_sense =		uipc_sense,
1161 	.pru_shutdown =		uipc_shutdown,
1162 	.pru_sockaddr =		uipc_sockaddr,
1163 	.pru_soreceive =	soreceive_dgram,
1164 	.pru_close =		uipc_close,
1165 };
1166 
1167 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1168 	.pru_abort =		uipc_abort,
1169 	.pru_accept =		uipc_accept,
1170 	.pru_attach =		uipc_attach,
1171 	.pru_bind =		uipc_bind,
1172 	.pru_bindat =		uipc_bindat,
1173 	.pru_connect =		uipc_connect,
1174 	.pru_connectat =	uipc_connectat,
1175 	.pru_connect2 =		uipc_connect2,
1176 	.pru_detach =		uipc_detach,
1177 	.pru_disconnect =	uipc_disconnect,
1178 	.pru_listen =		uipc_listen,
1179 	.pru_peeraddr =		uipc_peeraddr,
1180 	.pru_rcvd =		uipc_rcvd,
1181 	.pru_send =		uipc_send,
1182 	.pru_sense =		uipc_sense,
1183 	.pru_shutdown =		uipc_shutdown,
1184 	.pru_sockaddr =		uipc_sockaddr,
1185 	.pru_soreceive =	soreceive_generic,	/* XXX: or...? */
1186 	.pru_close =		uipc_close,
1187 };
1188 
1189 static struct pr_usrreqs uipc_usrreqs_stream = {
1190 	.pru_abort = 		uipc_abort,
1191 	.pru_accept =		uipc_accept,
1192 	.pru_attach =		uipc_attach,
1193 	.pru_bind =		uipc_bind,
1194 	.pru_bindat =		uipc_bindat,
1195 	.pru_connect =		uipc_connect,
1196 	.pru_connectat =	uipc_connectat,
1197 	.pru_connect2 =		uipc_connect2,
1198 	.pru_detach =		uipc_detach,
1199 	.pru_disconnect =	uipc_disconnect,
1200 	.pru_listen =		uipc_listen,
1201 	.pru_peeraddr =		uipc_peeraddr,
1202 	.pru_rcvd =		uipc_rcvd,
1203 	.pru_send =		uipc_send,
1204 	.pru_ready =		uipc_ready,
1205 	.pru_sense =		uipc_sense,
1206 	.pru_shutdown =		uipc_shutdown,
1207 	.pru_sockaddr =		uipc_sockaddr,
1208 	.pru_soreceive =	soreceive_generic,
1209 	.pru_close =		uipc_close,
1210 };
1211 
1212 static int
1213 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1214 {
1215 	struct unpcb *unp;
1216 	struct xucred xu;
1217 	int error, optval;
1218 
1219 	if (sopt->sopt_level != 0)
1220 		return (EINVAL);
1221 
1222 	unp = sotounpcb(so);
1223 	KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1224 	error = 0;
1225 	switch (sopt->sopt_dir) {
1226 	case SOPT_GET:
1227 		switch (sopt->sopt_name) {
1228 		case LOCAL_PEERCRED:
1229 			UNP_PCB_LOCK(unp);
1230 			if (unp->unp_flags & UNP_HAVEPC)
1231 				xu = unp->unp_peercred;
1232 			else {
1233 				if (so->so_type == SOCK_STREAM)
1234 					error = ENOTCONN;
1235 				else
1236 					error = EINVAL;
1237 			}
1238 			UNP_PCB_UNLOCK(unp);
1239 			if (error == 0)
1240 				error = sooptcopyout(sopt, &xu, sizeof(xu));
1241 			break;
1242 
1243 		case LOCAL_CREDS:
1244 			/* Unlocked read. */
1245 			optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1246 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1247 			break;
1248 
1249 		case LOCAL_CONNWAIT:
1250 			/* Unlocked read. */
1251 			optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1252 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1253 			break;
1254 
1255 		default:
1256 			error = EOPNOTSUPP;
1257 			break;
1258 		}
1259 		break;
1260 
1261 	case SOPT_SET:
1262 		switch (sopt->sopt_name) {
1263 		case LOCAL_CREDS:
1264 		case LOCAL_CONNWAIT:
1265 			error = sooptcopyin(sopt, &optval, sizeof(optval),
1266 					    sizeof(optval));
1267 			if (error)
1268 				break;
1269 
1270 #define	OPTSET(bit) do {						\
1271 	UNP_PCB_LOCK(unp);						\
1272 	if (optval)							\
1273 		unp->unp_flags |= bit;					\
1274 	else								\
1275 		unp->unp_flags &= ~bit;					\
1276 	UNP_PCB_UNLOCK(unp);						\
1277 } while (0)
1278 
1279 			switch (sopt->sopt_name) {
1280 			case LOCAL_CREDS:
1281 				OPTSET(UNP_WANTCRED);
1282 				break;
1283 
1284 			case LOCAL_CONNWAIT:
1285 				OPTSET(UNP_CONNWAIT);
1286 				break;
1287 
1288 			default:
1289 				break;
1290 			}
1291 			break;
1292 #undef	OPTSET
1293 		default:
1294 			error = ENOPROTOOPT;
1295 			break;
1296 		}
1297 		break;
1298 
1299 	default:
1300 		error = EOPNOTSUPP;
1301 		break;
1302 	}
1303 	return (error);
1304 }
1305 
1306 static int
1307 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1308 {
1309 
1310 	return (unp_connectat(AT_FDCWD, so, nam, td));
1311 }
1312 
1313 static int
1314 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1315     struct thread *td)
1316 {
1317 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1318 	struct vnode *vp;
1319 	struct socket *so2;
1320 	struct unpcb *unp, *unp2, *unp3;
1321 	struct nameidata nd;
1322 	char buf[SOCK_MAXADDRLEN];
1323 	struct sockaddr *sa;
1324 	cap_rights_t rights;
1325 	int error, len;
1326 
1327 	if (nam->sa_family != AF_UNIX)
1328 		return (EAFNOSUPPORT);
1329 
1330 	UNP_LINK_WLOCK_ASSERT();
1331 
1332 	unp = sotounpcb(so);
1333 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1334 
1335 	if (nam->sa_len > sizeof(struct sockaddr_un))
1336 		return (EINVAL);
1337 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1338 	if (len <= 0)
1339 		return (EINVAL);
1340 	bcopy(soun->sun_path, buf, len);
1341 	buf[len] = 0;
1342 
1343 	UNP_PCB_LOCK(unp);
1344 	if (unp->unp_flags & UNP_CONNECTING) {
1345 		UNP_PCB_UNLOCK(unp);
1346 		return (EALREADY);
1347 	}
1348 	UNP_LINK_WUNLOCK();
1349 	unp->unp_flags |= UNP_CONNECTING;
1350 	UNP_PCB_UNLOCK(unp);
1351 
1352 	sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1353 	NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1354 	    UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1355 	error = namei(&nd);
1356 	if (error)
1357 		vp = NULL;
1358 	else
1359 		vp = nd.ni_vp;
1360 	ASSERT_VOP_LOCKED(vp, "unp_connect");
1361 	NDFREE(&nd, NDF_ONLY_PNBUF);
1362 	if (error)
1363 		goto bad;
1364 
1365 	if (vp->v_type != VSOCK) {
1366 		error = ENOTSOCK;
1367 		goto bad;
1368 	}
1369 #ifdef MAC
1370 	error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1371 	if (error)
1372 		goto bad;
1373 #endif
1374 	error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1375 	if (error)
1376 		goto bad;
1377 
1378 	unp = sotounpcb(so);
1379 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1380 
1381 	/*
1382 	 * Lock linkage lock for two reasons: make sure v_socket is stable,
1383 	 * and to protect simultaneous locking of multiple pcbs.
1384 	 */
1385 	UNP_LINK_WLOCK();
1386 	VOP_UNP_CONNECT(vp, &unp2);
1387 	if (unp2 == NULL) {
1388 		error = ECONNREFUSED;
1389 		goto bad2;
1390 	}
1391 	so2 = unp2->unp_socket;
1392 	if (so->so_type != so2->so_type) {
1393 		error = EPROTOTYPE;
1394 		goto bad2;
1395 	}
1396 	UNP_PCB_LOCK(unp);
1397 	UNP_PCB_LOCK(unp2);
1398 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1399 		if (so2->so_options & SO_ACCEPTCONN) {
1400 			CURVNET_SET(so2->so_vnet);
1401 			so2 = sonewconn(so2, 0);
1402 			CURVNET_RESTORE();
1403 		} else
1404 			so2 = NULL;
1405 		if (so2 == NULL) {
1406 			error = ECONNREFUSED;
1407 			goto bad3;
1408 		}
1409 		unp3 = sotounpcb(so2);
1410 		UNP_PCB_LOCK(unp3);
1411 		if (unp2->unp_addr != NULL) {
1412 			bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1413 			unp3->unp_addr = (struct sockaddr_un *) sa;
1414 			sa = NULL;
1415 		}
1416 
1417 		/*
1418 		 * The connector's (client's) credentials are copied from its
1419 		 * process structure at the time of connect() (which is now).
1420 		 */
1421 		cru2x(td->td_ucred, &unp3->unp_peercred);
1422 		unp3->unp_flags |= UNP_HAVEPC;
1423 
1424 		/*
1425 		 * The receiver's (server's) credentials are copied from the
1426 		 * unp_peercred member of socket on which the former called
1427 		 * listen(); uipc_listen() cached that process's credentials
1428 		 * at that time so we can use them now.
1429 		 */
1430 		memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1431 		    sizeof(unp->unp_peercred));
1432 		unp->unp_flags |= UNP_HAVEPC;
1433 		if (unp2->unp_flags & UNP_WANTCRED)
1434 			unp3->unp_flags |= UNP_WANTCRED;
1435 		UNP_PCB_UNLOCK(unp2);
1436 		unp2 = unp3;
1437 #ifdef MAC
1438 		mac_socketpeer_set_from_socket(so, so2);
1439 		mac_socketpeer_set_from_socket(so2, so);
1440 #endif
1441 	}
1442 
1443 	KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 &&
1444 	    sotounpcb(so2) == unp2,
1445 	    ("%s: unp2 %p so2 %p", __func__, unp2, so2));
1446 	error = unp_connect2(so, so2, PRU_CONNECT);
1447 bad3:
1448 	UNP_PCB_UNLOCK(unp2);
1449 	UNP_PCB_UNLOCK(unp);
1450 bad2:
1451 	UNP_LINK_WUNLOCK();
1452 bad:
1453 	if (vp != NULL)
1454 		vput(vp);
1455 	free(sa, M_SONAME);
1456 	UNP_LINK_WLOCK();
1457 	UNP_PCB_LOCK(unp);
1458 	unp->unp_flags &= ~UNP_CONNECTING;
1459 	UNP_PCB_UNLOCK(unp);
1460 	return (error);
1461 }
1462 
1463 static int
1464 unp_connect2(struct socket *so, struct socket *so2, int req)
1465 {
1466 	struct unpcb *unp;
1467 	struct unpcb *unp2;
1468 
1469 	unp = sotounpcb(so);
1470 	KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1471 	unp2 = sotounpcb(so2);
1472 	KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1473 
1474 	UNP_LINK_WLOCK_ASSERT();
1475 	UNP_PCB_LOCK_ASSERT(unp);
1476 	UNP_PCB_LOCK_ASSERT(unp2);
1477 
1478 	if (so2->so_type != so->so_type)
1479 		return (EPROTOTYPE);
1480 	unp2->unp_flags &= ~UNP_NASCENT;
1481 	unp->unp_conn = unp2;
1482 
1483 	switch (so->so_type) {
1484 	case SOCK_DGRAM:
1485 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1486 		soisconnected(so);
1487 		break;
1488 
1489 	case SOCK_STREAM:
1490 	case SOCK_SEQPACKET:
1491 		unp2->unp_conn = unp;
1492 		if (req == PRU_CONNECT &&
1493 		    ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1494 			soisconnecting(so);
1495 		else
1496 			soisconnected(so);
1497 		soisconnected(so2);
1498 		break;
1499 
1500 	default:
1501 		panic("unp_connect2");
1502 	}
1503 	return (0);
1504 }
1505 
1506 static void
1507 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1508 {
1509 	struct socket *so;
1510 
1511 	KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1512 
1513 	UNP_LINK_WLOCK_ASSERT();
1514 	UNP_PCB_LOCK_ASSERT(unp);
1515 	UNP_PCB_LOCK_ASSERT(unp2);
1516 
1517 	unp->unp_conn = NULL;
1518 	switch (unp->unp_socket->so_type) {
1519 	case SOCK_DGRAM:
1520 		LIST_REMOVE(unp, unp_reflink);
1521 		so = unp->unp_socket;
1522 		SOCK_LOCK(so);
1523 		so->so_state &= ~SS_ISCONNECTED;
1524 		SOCK_UNLOCK(so);
1525 		break;
1526 
1527 	case SOCK_STREAM:
1528 	case SOCK_SEQPACKET:
1529 		soisdisconnected(unp->unp_socket);
1530 		unp2->unp_conn = NULL;
1531 		soisdisconnected(unp2->unp_socket);
1532 		break;
1533 	}
1534 }
1535 
1536 /*
1537  * unp_pcblist() walks the global list of struct unpcb's to generate a
1538  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1539  * sequentially, validating the generation number on each to see if it has
1540  * been detached.  All of this is necessary because copyout() may sleep on
1541  * disk I/O.
1542  */
1543 static int
1544 unp_pcblist(SYSCTL_HANDLER_ARGS)
1545 {
1546 	int error, i, n;
1547 	int freeunp;
1548 	struct unpcb *unp, **unp_list;
1549 	unp_gen_t gencnt;
1550 	struct xunpgen *xug;
1551 	struct unp_head *head;
1552 	struct xunpcb *xu;
1553 
1554 	switch ((intptr_t)arg1) {
1555 	case SOCK_STREAM:
1556 		head = &unp_shead;
1557 		break;
1558 
1559 	case SOCK_DGRAM:
1560 		head = &unp_dhead;
1561 		break;
1562 
1563 	case SOCK_SEQPACKET:
1564 		head = &unp_sphead;
1565 		break;
1566 
1567 	default:
1568 		panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1569 	}
1570 
1571 	/*
1572 	 * The process of preparing the PCB list is too time-consuming and
1573 	 * resource-intensive to repeat twice on every request.
1574 	 */
1575 	if (req->oldptr == NULL) {
1576 		n = unp_count;
1577 		req->oldidx = 2 * (sizeof *xug)
1578 			+ (n + n/8) * sizeof(struct xunpcb);
1579 		return (0);
1580 	}
1581 
1582 	if (req->newptr != NULL)
1583 		return (EPERM);
1584 
1585 	/*
1586 	 * OK, now we're committed to doing something.
1587 	 */
1588 	xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1589 	UNP_LINK_RLOCK();
1590 	gencnt = unp_gencnt;
1591 	n = unp_count;
1592 	UNP_LINK_RUNLOCK();
1593 
1594 	xug->xug_len = sizeof *xug;
1595 	xug->xug_count = n;
1596 	xug->xug_gen = gencnt;
1597 	xug->xug_sogen = so_gencnt;
1598 	error = SYSCTL_OUT(req, xug, sizeof *xug);
1599 	if (error) {
1600 		free(xug, M_TEMP);
1601 		return (error);
1602 	}
1603 
1604 	unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1605 
1606 	UNP_LINK_RLOCK();
1607 	for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1608 	     unp = LIST_NEXT(unp, unp_link)) {
1609 		UNP_PCB_LOCK(unp);
1610 		if (unp->unp_gencnt <= gencnt) {
1611 			if (cr_cansee(req->td->td_ucred,
1612 			    unp->unp_socket->so_cred)) {
1613 				UNP_PCB_UNLOCK(unp);
1614 				continue;
1615 			}
1616 			unp_list[i++] = unp;
1617 			unp->unp_refcount++;
1618 		}
1619 		UNP_PCB_UNLOCK(unp);
1620 	}
1621 	UNP_LINK_RUNLOCK();
1622 	n = i;			/* In case we lost some during malloc. */
1623 
1624 	error = 0;
1625 	xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1626 	for (i = 0; i < n; i++) {
1627 		unp = unp_list[i];
1628 		UNP_PCB_LOCK(unp);
1629 		unp->unp_refcount--;
1630 	        if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1631 			xu->xu_len = sizeof *xu;
1632 			xu->xu_unpp = unp;
1633 			/*
1634 			 * XXX - need more locking here to protect against
1635 			 * connect/disconnect races for SMP.
1636 			 */
1637 			if (unp->unp_addr != NULL)
1638 				bcopy(unp->unp_addr, &xu->xu_addr,
1639 				      unp->unp_addr->sun_len);
1640 			if (unp->unp_conn != NULL &&
1641 			    unp->unp_conn->unp_addr != NULL)
1642 				bcopy(unp->unp_conn->unp_addr,
1643 				      &xu->xu_caddr,
1644 				      unp->unp_conn->unp_addr->sun_len);
1645 			bcopy(unp, &xu->xu_unp, sizeof *unp);
1646 			sotoxsocket(unp->unp_socket, &xu->xu_socket);
1647 			UNP_PCB_UNLOCK(unp);
1648 			error = SYSCTL_OUT(req, xu, sizeof *xu);
1649 		} else {
1650 			freeunp = (unp->unp_refcount == 0);
1651 			UNP_PCB_UNLOCK(unp);
1652 			if (freeunp) {
1653 				UNP_PCB_LOCK_DESTROY(unp);
1654 				uma_zfree(unp_zone, unp);
1655 			}
1656 		}
1657 	}
1658 	free(xu, M_TEMP);
1659 	if (!error) {
1660 		/*
1661 		 * Give the user an updated idea of our state.  If the
1662 		 * generation differs from what we told her before, she knows
1663 		 * that something happened while we were processing this
1664 		 * request, and it might be necessary to retry.
1665 		 */
1666 		xug->xug_gen = unp_gencnt;
1667 		xug->xug_sogen = so_gencnt;
1668 		xug->xug_count = unp_count;
1669 		error = SYSCTL_OUT(req, xug, sizeof *xug);
1670 	}
1671 	free(unp_list, M_TEMP);
1672 	free(xug, M_TEMP);
1673 	return (error);
1674 }
1675 
1676 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1677     (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1678     "List of active local datagram sockets");
1679 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1680     (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1681     "List of active local stream sockets");
1682 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1683     CTLTYPE_OPAQUE | CTLFLAG_RD,
1684     (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1685     "List of active local seqpacket sockets");
1686 
1687 static void
1688 unp_shutdown(struct unpcb *unp)
1689 {
1690 	struct unpcb *unp2;
1691 	struct socket *so;
1692 
1693 	UNP_LINK_WLOCK_ASSERT();
1694 	UNP_PCB_LOCK_ASSERT(unp);
1695 
1696 	unp2 = unp->unp_conn;
1697 	if ((unp->unp_socket->so_type == SOCK_STREAM ||
1698 	    (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1699 		so = unp2->unp_socket;
1700 		if (so != NULL)
1701 			socantrcvmore(so);
1702 	}
1703 }
1704 
1705 static void
1706 unp_drop(struct unpcb *unp)
1707 {
1708 	struct socket *so = unp->unp_socket;
1709 	struct unpcb *unp2;
1710 
1711 	UNP_LINK_WLOCK_ASSERT();
1712 	UNP_PCB_LOCK_ASSERT(unp);
1713 
1714 	/*
1715 	 * Regardless of whether the socket's peer dropped the connection
1716 	 * with this socket by aborting or disconnecting, POSIX requires
1717 	 * that ECONNRESET is returned.
1718 	 */
1719 	so->so_error = ECONNRESET;
1720 	unp2 = unp->unp_conn;
1721 	if (unp2 == NULL)
1722 		return;
1723 	UNP_PCB_LOCK(unp2);
1724 	unp_disconnect(unp, unp2);
1725 	UNP_PCB_UNLOCK(unp2);
1726 }
1727 
1728 static void
1729 unp_freerights(struct filedescent **fdep, int fdcount)
1730 {
1731 	struct file *fp;
1732 	int i;
1733 
1734 	KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1735 
1736 	for (i = 0; i < fdcount; i++) {
1737 		fp = fdep[i]->fde_file;
1738 		filecaps_free(&fdep[i]->fde_caps);
1739 		unp_discard(fp);
1740 	}
1741 	free(fdep[0], M_FILECAPS);
1742 }
1743 
1744 static int
1745 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1746 {
1747 	struct thread *td = curthread;		/* XXX */
1748 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1749 	int i;
1750 	int *fdp;
1751 	struct filedesc *fdesc = td->td_proc->p_fd;
1752 	struct filedescent **fdep;
1753 	void *data;
1754 	socklen_t clen = control->m_len, datalen;
1755 	int error, newfds;
1756 	u_int newlen;
1757 
1758 	UNP_LINK_UNLOCK_ASSERT();
1759 
1760 	error = 0;
1761 	if (controlp != NULL) /* controlp == NULL => free control messages */
1762 		*controlp = NULL;
1763 	while (cm != NULL) {
1764 		if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1765 			error = EINVAL;
1766 			break;
1767 		}
1768 		data = CMSG_DATA(cm);
1769 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1770 		if (cm->cmsg_level == SOL_SOCKET
1771 		    && cm->cmsg_type == SCM_RIGHTS) {
1772 			newfds = datalen / sizeof(*fdep);
1773 			if (newfds == 0)
1774 				goto next;
1775 			fdep = data;
1776 
1777 			/* If we're not outputting the descriptors free them. */
1778 			if (error || controlp == NULL) {
1779 				unp_freerights(fdep, newfds);
1780 				goto next;
1781 			}
1782 			FILEDESC_XLOCK(fdesc);
1783 
1784 			/*
1785 			 * Now change each pointer to an fd in the global
1786 			 * table to an integer that is the index to the local
1787 			 * fd table entry that we set up to point to the
1788 			 * global one we are transferring.
1789 			 */
1790 			newlen = newfds * sizeof(int);
1791 			*controlp = sbcreatecontrol(NULL, newlen,
1792 			    SCM_RIGHTS, SOL_SOCKET);
1793 			if (*controlp == NULL) {
1794 				FILEDESC_XUNLOCK(fdesc);
1795 				error = E2BIG;
1796 				unp_freerights(fdep, newfds);
1797 				goto next;
1798 			}
1799 
1800 			fdp = (int *)
1801 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1802 			if (fdallocn(td, 0, fdp, newfds) != 0) {
1803 				FILEDESC_XUNLOCK(fdesc);
1804 				error = EMSGSIZE;
1805 				unp_freerights(fdep, newfds);
1806 				m_freem(*controlp);
1807 				*controlp = NULL;
1808 				goto next;
1809 			}
1810 			for (i = 0; i < newfds; i++, fdp++) {
1811 				_finstall(fdesc, fdep[i]->fde_file, *fdp,
1812 				    (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0,
1813 				    &fdep[i]->fde_caps);
1814 				unp_externalize_fp(fdep[i]->fde_file);
1815 			}
1816 			FILEDESC_XUNLOCK(fdesc);
1817 			free(fdep[0], M_FILECAPS);
1818 		} else {
1819 			/* We can just copy anything else across. */
1820 			if (error || controlp == NULL)
1821 				goto next;
1822 			*controlp = sbcreatecontrol(NULL, datalen,
1823 			    cm->cmsg_type, cm->cmsg_level);
1824 			if (*controlp == NULL) {
1825 				error = ENOBUFS;
1826 				goto next;
1827 			}
1828 			bcopy(data,
1829 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1830 			    datalen);
1831 		}
1832 		controlp = &(*controlp)->m_next;
1833 
1834 next:
1835 		if (CMSG_SPACE(datalen) < clen) {
1836 			clen -= CMSG_SPACE(datalen);
1837 			cm = (struct cmsghdr *)
1838 			    ((caddr_t)cm + CMSG_SPACE(datalen));
1839 		} else {
1840 			clen = 0;
1841 			cm = NULL;
1842 		}
1843 	}
1844 
1845 	m_freem(control);
1846 	return (error);
1847 }
1848 
1849 static void
1850 unp_zone_change(void *tag)
1851 {
1852 
1853 	uma_zone_set_max(unp_zone, maxsockets);
1854 }
1855 
1856 static void
1857 unp_init(void)
1858 {
1859 
1860 #ifdef VIMAGE
1861 	if (!IS_DEFAULT_VNET(curvnet))
1862 		return;
1863 #endif
1864 	unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1865 	    NULL, NULL, UMA_ALIGN_PTR, 0);
1866 	if (unp_zone == NULL)
1867 		panic("unp_init");
1868 	uma_zone_set_max(unp_zone, maxsockets);
1869 	uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1870 	EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1871 	    NULL, EVENTHANDLER_PRI_ANY);
1872 	LIST_INIT(&unp_dhead);
1873 	LIST_INIT(&unp_shead);
1874 	LIST_INIT(&unp_sphead);
1875 	SLIST_INIT(&unp_defers);
1876 	TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1877 	TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1878 	UNP_LINK_LOCK_INIT();
1879 	UNP_DEFERRED_LOCK_INIT();
1880 }
1881 
1882 static int
1883 unp_internalize(struct mbuf **controlp, struct thread *td)
1884 {
1885 	struct mbuf *control = *controlp;
1886 	struct proc *p = td->td_proc;
1887 	struct filedesc *fdesc = p->p_fd;
1888 	struct bintime *bt;
1889 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1890 	struct cmsgcred *cmcred;
1891 	struct filedescent *fde, **fdep, *fdev;
1892 	struct file *fp;
1893 	struct timeval *tv;
1894 	struct timespec *ts;
1895 	int i, *fdp;
1896 	void *data;
1897 	socklen_t clen = control->m_len, datalen;
1898 	int error, oldfds;
1899 	u_int newlen;
1900 
1901 	UNP_LINK_UNLOCK_ASSERT();
1902 
1903 	error = 0;
1904 	*controlp = NULL;
1905 	while (cm != NULL) {
1906 		if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1907 		    || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
1908 			error = EINVAL;
1909 			goto out;
1910 		}
1911 		data = CMSG_DATA(cm);
1912 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1913 
1914 		switch (cm->cmsg_type) {
1915 		/*
1916 		 * Fill in credential information.
1917 		 */
1918 		case SCM_CREDS:
1919 			*controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1920 			    SCM_CREDS, SOL_SOCKET);
1921 			if (*controlp == NULL) {
1922 				error = ENOBUFS;
1923 				goto out;
1924 			}
1925 			cmcred = (struct cmsgcred *)
1926 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1927 			cmcred->cmcred_pid = p->p_pid;
1928 			cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1929 			cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1930 			cmcred->cmcred_euid = td->td_ucred->cr_uid;
1931 			cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1932 			    CMGROUP_MAX);
1933 			for (i = 0; i < cmcred->cmcred_ngroups; i++)
1934 				cmcred->cmcred_groups[i] =
1935 				    td->td_ucred->cr_groups[i];
1936 			break;
1937 
1938 		case SCM_RIGHTS:
1939 			oldfds = datalen / sizeof (int);
1940 			if (oldfds == 0)
1941 				break;
1942 			/*
1943 			 * Check that all the FDs passed in refer to legal
1944 			 * files.  If not, reject the entire operation.
1945 			 */
1946 			fdp = data;
1947 			FILEDESC_SLOCK(fdesc);
1948 			for (i = 0; i < oldfds; i++, fdp++) {
1949 				fp = fget_locked(fdesc, *fdp);
1950 				if (fp == NULL) {
1951 					FILEDESC_SUNLOCK(fdesc);
1952 					error = EBADF;
1953 					goto out;
1954 				}
1955 				if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1956 					FILEDESC_SUNLOCK(fdesc);
1957 					error = EOPNOTSUPP;
1958 					goto out;
1959 				}
1960 
1961 			}
1962 
1963 			/*
1964 			 * Now replace the integer FDs with pointers to the
1965 			 * file structure and capability rights.
1966 			 */
1967 			newlen = oldfds * sizeof(fdep[0]);
1968 			*controlp = sbcreatecontrol(NULL, newlen,
1969 			    SCM_RIGHTS, SOL_SOCKET);
1970 			if (*controlp == NULL) {
1971 				FILEDESC_SUNLOCK(fdesc);
1972 				error = E2BIG;
1973 				goto out;
1974 			}
1975 			fdp = data;
1976 			fdep = (struct filedescent **)
1977 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1978 			fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
1979 			    M_WAITOK);
1980 			for (i = 0; i < oldfds; i++, fdev++, fdp++) {
1981 				fde = &fdesc->fd_ofiles[*fdp];
1982 				fdep[i] = fdev;
1983 				fdep[i]->fde_file = fde->fde_file;
1984 				filecaps_copy(&fde->fde_caps,
1985 				    &fdep[i]->fde_caps, true);
1986 				unp_internalize_fp(fdep[i]->fde_file);
1987 			}
1988 			FILEDESC_SUNLOCK(fdesc);
1989 			break;
1990 
1991 		case SCM_TIMESTAMP:
1992 			*controlp = sbcreatecontrol(NULL, sizeof(*tv),
1993 			    SCM_TIMESTAMP, SOL_SOCKET);
1994 			if (*controlp == NULL) {
1995 				error = ENOBUFS;
1996 				goto out;
1997 			}
1998 			tv = (struct timeval *)
1999 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2000 			microtime(tv);
2001 			break;
2002 
2003 		case SCM_BINTIME:
2004 			*controlp = sbcreatecontrol(NULL, sizeof(*bt),
2005 			    SCM_BINTIME, SOL_SOCKET);
2006 			if (*controlp == NULL) {
2007 				error = ENOBUFS;
2008 				goto out;
2009 			}
2010 			bt = (struct bintime *)
2011 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2012 			bintime(bt);
2013 			break;
2014 
2015 		case SCM_REALTIME:
2016 			*controlp = sbcreatecontrol(NULL, sizeof(*ts),
2017 			    SCM_REALTIME, SOL_SOCKET);
2018 			if (*controlp == NULL) {
2019 				error = ENOBUFS;
2020 				goto out;
2021 			}
2022 			ts = (struct timespec *)
2023 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2024 			nanotime(ts);
2025 			break;
2026 
2027 		case SCM_MONOTONIC:
2028 			*controlp = sbcreatecontrol(NULL, sizeof(*ts),
2029 			    SCM_MONOTONIC, SOL_SOCKET);
2030 			if (*controlp == NULL) {
2031 				error = ENOBUFS;
2032 				goto out;
2033 			}
2034 			ts = (struct timespec *)
2035 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2036 			nanouptime(ts);
2037 			break;
2038 
2039 		default:
2040 			error = EINVAL;
2041 			goto out;
2042 		}
2043 
2044 		controlp = &(*controlp)->m_next;
2045 		if (CMSG_SPACE(datalen) < clen) {
2046 			clen -= CMSG_SPACE(datalen);
2047 			cm = (struct cmsghdr *)
2048 			    ((caddr_t)cm + CMSG_SPACE(datalen));
2049 		} else {
2050 			clen = 0;
2051 			cm = NULL;
2052 		}
2053 	}
2054 
2055 out:
2056 	m_freem(control);
2057 	return (error);
2058 }
2059 
2060 static struct mbuf *
2061 unp_addsockcred(struct thread *td, struct mbuf *control)
2062 {
2063 	struct mbuf *m, *n, *n_prev;
2064 	struct sockcred *sc;
2065 	const struct cmsghdr *cm;
2066 	int ngroups;
2067 	int i;
2068 
2069 	ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2070 	m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2071 	if (m == NULL)
2072 		return (control);
2073 
2074 	sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2075 	sc->sc_uid = td->td_ucred->cr_ruid;
2076 	sc->sc_euid = td->td_ucred->cr_uid;
2077 	sc->sc_gid = td->td_ucred->cr_rgid;
2078 	sc->sc_egid = td->td_ucred->cr_gid;
2079 	sc->sc_ngroups = ngroups;
2080 	for (i = 0; i < sc->sc_ngroups; i++)
2081 		sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2082 
2083 	/*
2084 	 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2085 	 * created SCM_CREDS control message (struct sockcred) has another
2086 	 * format.
2087 	 */
2088 	if (control != NULL)
2089 		for (n = control, n_prev = NULL; n != NULL;) {
2090 			cm = mtod(n, struct cmsghdr *);
2091     			if (cm->cmsg_level == SOL_SOCKET &&
2092 			    cm->cmsg_type == SCM_CREDS) {
2093     				if (n_prev == NULL)
2094 					control = n->m_next;
2095 				else
2096 					n_prev->m_next = n->m_next;
2097 				n = m_free(n);
2098 			} else {
2099 				n_prev = n;
2100 				n = n->m_next;
2101 			}
2102 		}
2103 
2104 	/* Prepend it to the head. */
2105 	m->m_next = control;
2106 	return (m);
2107 }
2108 
2109 static struct unpcb *
2110 fptounp(struct file *fp)
2111 {
2112 	struct socket *so;
2113 
2114 	if (fp->f_type != DTYPE_SOCKET)
2115 		return (NULL);
2116 	if ((so = fp->f_data) == NULL)
2117 		return (NULL);
2118 	if (so->so_proto->pr_domain != &localdomain)
2119 		return (NULL);
2120 	return sotounpcb(so);
2121 }
2122 
2123 static void
2124 unp_discard(struct file *fp)
2125 {
2126 	struct unp_defer *dr;
2127 
2128 	if (unp_externalize_fp(fp)) {
2129 		dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2130 		dr->ud_fp = fp;
2131 		UNP_DEFERRED_LOCK();
2132 		SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2133 		UNP_DEFERRED_UNLOCK();
2134 		atomic_add_int(&unp_defers_count, 1);
2135 		taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2136 	} else
2137 		(void) closef(fp, (struct thread *)NULL);
2138 }
2139 
2140 static void
2141 unp_process_defers(void *arg __unused, int pending)
2142 {
2143 	struct unp_defer *dr;
2144 	SLIST_HEAD(, unp_defer) drl;
2145 	int count;
2146 
2147 	SLIST_INIT(&drl);
2148 	for (;;) {
2149 		UNP_DEFERRED_LOCK();
2150 		if (SLIST_FIRST(&unp_defers) == NULL) {
2151 			UNP_DEFERRED_UNLOCK();
2152 			break;
2153 		}
2154 		SLIST_SWAP(&unp_defers, &drl, unp_defer);
2155 		UNP_DEFERRED_UNLOCK();
2156 		count = 0;
2157 		while ((dr = SLIST_FIRST(&drl)) != NULL) {
2158 			SLIST_REMOVE_HEAD(&drl, ud_link);
2159 			closef(dr->ud_fp, NULL);
2160 			free(dr, M_TEMP);
2161 			count++;
2162 		}
2163 		atomic_add_int(&unp_defers_count, -count);
2164 	}
2165 }
2166 
2167 static void
2168 unp_internalize_fp(struct file *fp)
2169 {
2170 	struct unpcb *unp;
2171 
2172 	UNP_LINK_WLOCK();
2173 	if ((unp = fptounp(fp)) != NULL) {
2174 		unp->unp_file = fp;
2175 		unp->unp_msgcount++;
2176 	}
2177 	fhold(fp);
2178 	unp_rights++;
2179 	UNP_LINK_WUNLOCK();
2180 }
2181 
2182 static int
2183 unp_externalize_fp(struct file *fp)
2184 {
2185 	struct unpcb *unp;
2186 	int ret;
2187 
2188 	UNP_LINK_WLOCK();
2189 	if ((unp = fptounp(fp)) != NULL) {
2190 		unp->unp_msgcount--;
2191 		ret = 1;
2192 	} else
2193 		ret = 0;
2194 	unp_rights--;
2195 	UNP_LINK_WUNLOCK();
2196 	return (ret);
2197 }
2198 
2199 /*
2200  * unp_defer indicates whether additional work has been defered for a future
2201  * pass through unp_gc().  It is thread local and does not require explicit
2202  * synchronization.
2203  */
2204 static int	unp_marked;
2205 static int	unp_unreachable;
2206 
2207 static void
2208 unp_accessable(struct filedescent **fdep, int fdcount)
2209 {
2210 	struct unpcb *unp;
2211 	struct file *fp;
2212 	int i;
2213 
2214 	for (i = 0; i < fdcount; i++) {
2215 		fp = fdep[i]->fde_file;
2216 		if ((unp = fptounp(fp)) == NULL)
2217 			continue;
2218 		if (unp->unp_gcflag & UNPGC_REF)
2219 			continue;
2220 		unp->unp_gcflag &= ~UNPGC_DEAD;
2221 		unp->unp_gcflag |= UNPGC_REF;
2222 		unp_marked++;
2223 	}
2224 }
2225 
2226 static void
2227 unp_gc_process(struct unpcb *unp)
2228 {
2229 	struct socket *so, *soa;
2230 	struct file *fp;
2231 
2232 	/* Already processed. */
2233 	if (unp->unp_gcflag & UNPGC_SCANNED)
2234 		return;
2235 	fp = unp->unp_file;
2236 
2237 	/*
2238 	 * Check for a socket potentially in a cycle.  It must be in a
2239 	 * queue as indicated by msgcount, and this must equal the file
2240 	 * reference count.  Note that when msgcount is 0 the file is NULL.
2241 	 */
2242 	if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2243 	    unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2244 		unp->unp_gcflag |= UNPGC_DEAD;
2245 		unp_unreachable++;
2246 		return;
2247 	}
2248 
2249 	so = unp->unp_socket;
2250 	SOCK_LOCK(so);
2251 	if (SOLISTENING(so)) {
2252 		/*
2253 		 * Mark all sockets in our accept queue.
2254 		 */
2255 		TAILQ_FOREACH(soa, &so->sol_comp, so_list) {
2256 			if (sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS)
2257 				continue;
2258 			SOCKBUF_LOCK(&soa->so_rcv);
2259 			unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2260 			SOCKBUF_UNLOCK(&soa->so_rcv);
2261 		}
2262 	} else {
2263 		/*
2264 		 * Mark all sockets we reference with RIGHTS.
2265 		 */
2266 		if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2267 			SOCKBUF_LOCK(&so->so_rcv);
2268 			unp_scan(so->so_rcv.sb_mb, unp_accessable);
2269 			SOCKBUF_UNLOCK(&so->so_rcv);
2270 		}
2271 	}
2272 	SOCK_UNLOCK(so);
2273 	unp->unp_gcflag |= UNPGC_SCANNED;
2274 }
2275 
2276 static int unp_recycled;
2277 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2278     "Number of unreachable sockets claimed by the garbage collector.");
2279 
2280 static int unp_taskcount;
2281 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2282     "Number of times the garbage collector has run.");
2283 
2284 static void
2285 unp_gc(__unused void *arg, int pending)
2286 {
2287 	struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2288 				    NULL };
2289 	struct unp_head **head;
2290 	struct file *f, **unref;
2291 	struct unpcb *unp;
2292 	int i, total;
2293 
2294 	unp_taskcount++;
2295 	UNP_LINK_RLOCK();
2296 	/*
2297 	 * First clear all gc flags from previous runs, apart from
2298 	 * UNPGC_IGNORE_RIGHTS.
2299 	 */
2300 	for (head = heads; *head != NULL; head++)
2301 		LIST_FOREACH(unp, *head, unp_link)
2302 			unp->unp_gcflag =
2303 			    (unp->unp_gcflag & UNPGC_IGNORE_RIGHTS);
2304 
2305 	/*
2306 	 * Scan marking all reachable sockets with UNPGC_REF.  Once a socket
2307 	 * is reachable all of the sockets it references are reachable.
2308 	 * Stop the scan once we do a complete loop without discovering
2309 	 * a new reachable socket.
2310 	 */
2311 	do {
2312 		unp_unreachable = 0;
2313 		unp_marked = 0;
2314 		for (head = heads; *head != NULL; head++)
2315 			LIST_FOREACH(unp, *head, unp_link)
2316 				unp_gc_process(unp);
2317 	} while (unp_marked);
2318 	UNP_LINK_RUNLOCK();
2319 	if (unp_unreachable == 0)
2320 		return;
2321 
2322 	/*
2323 	 * Allocate space for a local list of dead unpcbs.
2324 	 */
2325 	unref = malloc(unp_unreachable * sizeof(struct file *),
2326 	    M_TEMP, M_WAITOK);
2327 
2328 	/*
2329 	 * Iterate looking for sockets which have been specifically marked
2330 	 * as as unreachable and store them locally.
2331 	 */
2332 	UNP_LINK_RLOCK();
2333 	for (total = 0, head = heads; *head != NULL; head++)
2334 		LIST_FOREACH(unp, *head, unp_link)
2335 			if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2336 				f = unp->unp_file;
2337 				if (unp->unp_msgcount == 0 || f == NULL ||
2338 				    f->f_count != unp->unp_msgcount)
2339 					continue;
2340 				unref[total++] = f;
2341 				fhold(f);
2342 				KASSERT(total <= unp_unreachable,
2343 				    ("unp_gc: incorrect unreachable count."));
2344 			}
2345 	UNP_LINK_RUNLOCK();
2346 
2347 	/*
2348 	 * Now flush all sockets, free'ing rights.  This will free the
2349 	 * struct files associated with these sockets but leave each socket
2350 	 * with one remaining ref.
2351 	 */
2352 	for (i = 0; i < total; i++) {
2353 		struct socket *so;
2354 
2355 		so = unref[i]->f_data;
2356 		CURVNET_SET(so->so_vnet);
2357 		sorflush(so);
2358 		CURVNET_RESTORE();
2359 	}
2360 
2361 	/*
2362 	 * And finally release the sockets so they can be reclaimed.
2363 	 */
2364 	for (i = 0; i < total; i++)
2365 		fdrop(unref[i], NULL);
2366 	unp_recycled += total;
2367 	free(unref, M_TEMP);
2368 }
2369 
2370 static void
2371 unp_dispose_mbuf(struct mbuf *m)
2372 {
2373 
2374 	if (m)
2375 		unp_scan(m, unp_freerights);
2376 }
2377 
2378 /*
2379  * Synchronize against unp_gc, which can trip over data as we are freeing it.
2380  */
2381 static void
2382 unp_dispose(struct socket *so)
2383 {
2384 	struct unpcb *unp;
2385 
2386 	unp = sotounpcb(so);
2387 	UNP_LINK_WLOCK();
2388 	unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2389 	UNP_LINK_WUNLOCK();
2390 	if (!SOLISTENING(so))
2391 		unp_dispose_mbuf(so->so_rcv.sb_mb);
2392 }
2393 
2394 static void
2395 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2396 {
2397 	struct mbuf *m;
2398 	struct cmsghdr *cm;
2399 	void *data;
2400 	socklen_t clen, datalen;
2401 
2402 	while (m0 != NULL) {
2403 		for (m = m0; m; m = m->m_next) {
2404 			if (m->m_type != MT_CONTROL)
2405 				continue;
2406 
2407 			cm = mtod(m, struct cmsghdr *);
2408 			clen = m->m_len;
2409 
2410 			while (cm != NULL) {
2411 				if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2412 					break;
2413 
2414 				data = CMSG_DATA(cm);
2415 				datalen = (caddr_t)cm + cm->cmsg_len
2416 				    - (caddr_t)data;
2417 
2418 				if (cm->cmsg_level == SOL_SOCKET &&
2419 				    cm->cmsg_type == SCM_RIGHTS) {
2420 					(*op)(data, datalen /
2421 					    sizeof(struct filedescent *));
2422 				}
2423 
2424 				if (CMSG_SPACE(datalen) < clen) {
2425 					clen -= CMSG_SPACE(datalen);
2426 					cm = (struct cmsghdr *)
2427 					    ((caddr_t)cm + CMSG_SPACE(datalen));
2428 				} else {
2429 					clen = 0;
2430 					cm = NULL;
2431 				}
2432 			}
2433 		}
2434 		m0 = m0->m_nextpkt;
2435 	}
2436 }
2437 
2438 /*
2439  * A helper function called by VFS before socket-type vnode reclamation.
2440  * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2441  * use count.
2442  */
2443 void
2444 vfs_unp_reclaim(struct vnode *vp)
2445 {
2446 	struct unpcb *unp;
2447 	int active;
2448 
2449 	ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2450 	KASSERT(vp->v_type == VSOCK,
2451 	    ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2452 
2453 	active = 0;
2454 	UNP_LINK_WLOCK();
2455 	VOP_UNP_CONNECT(vp, &unp);
2456 	if (unp == NULL)
2457 		goto done;
2458 	UNP_PCB_LOCK(unp);
2459 	if (unp->unp_vnode == vp) {
2460 		VOP_UNP_DETACH(vp);
2461 		unp->unp_vnode = NULL;
2462 		active = 1;
2463 	}
2464 	UNP_PCB_UNLOCK(unp);
2465 done:
2466 	UNP_LINK_WUNLOCK();
2467 	if (active)
2468 		vunref(vp);
2469 }
2470 
2471 #ifdef DDB
2472 static void
2473 db_print_indent(int indent)
2474 {
2475 	int i;
2476 
2477 	for (i = 0; i < indent; i++)
2478 		db_printf(" ");
2479 }
2480 
2481 static void
2482 db_print_unpflags(int unp_flags)
2483 {
2484 	int comma;
2485 
2486 	comma = 0;
2487 	if (unp_flags & UNP_HAVEPC) {
2488 		db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2489 		comma = 1;
2490 	}
2491 	if (unp_flags & UNP_WANTCRED) {
2492 		db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2493 		comma = 1;
2494 	}
2495 	if (unp_flags & UNP_CONNWAIT) {
2496 		db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2497 		comma = 1;
2498 	}
2499 	if (unp_flags & UNP_CONNECTING) {
2500 		db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2501 		comma = 1;
2502 	}
2503 	if (unp_flags & UNP_BINDING) {
2504 		db_printf("%sUNP_BINDING", comma ? ", " : "");
2505 		comma = 1;
2506 	}
2507 }
2508 
2509 static void
2510 db_print_xucred(int indent, struct xucred *xu)
2511 {
2512 	int comma, i;
2513 
2514 	db_print_indent(indent);
2515 	db_printf("cr_version: %u   cr_uid: %u   cr_ngroups: %d\n",
2516 	    xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2517 	db_print_indent(indent);
2518 	db_printf("cr_groups: ");
2519 	comma = 0;
2520 	for (i = 0; i < xu->cr_ngroups; i++) {
2521 		db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2522 		comma = 1;
2523 	}
2524 	db_printf("\n");
2525 }
2526 
2527 static void
2528 db_print_unprefs(int indent, struct unp_head *uh)
2529 {
2530 	struct unpcb *unp;
2531 	int counter;
2532 
2533 	counter = 0;
2534 	LIST_FOREACH(unp, uh, unp_reflink) {
2535 		if (counter % 4 == 0)
2536 			db_print_indent(indent);
2537 		db_printf("%p  ", unp);
2538 		if (counter % 4 == 3)
2539 			db_printf("\n");
2540 		counter++;
2541 	}
2542 	if (counter != 0 && counter % 4 != 0)
2543 		db_printf("\n");
2544 }
2545 
2546 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2547 {
2548 	struct unpcb *unp;
2549 
2550         if (!have_addr) {
2551                 db_printf("usage: show unpcb <addr>\n");
2552                 return;
2553         }
2554         unp = (struct unpcb *)addr;
2555 
2556 	db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2557 	    unp->unp_vnode);
2558 
2559 	db_printf("unp_ino: %ju   unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2560 	    unp->unp_conn);
2561 
2562 	db_printf("unp_refs:\n");
2563 	db_print_unprefs(2, &unp->unp_refs);
2564 
2565 	/* XXXRW: Would be nice to print the full address, if any. */
2566 	db_printf("unp_addr: %p\n", unp->unp_addr);
2567 
2568 	db_printf("unp_gencnt: %llu\n",
2569 	    (unsigned long long)unp->unp_gencnt);
2570 
2571 	db_printf("unp_flags: %x (", unp->unp_flags);
2572 	db_print_unpflags(unp->unp_flags);
2573 	db_printf(")\n");
2574 
2575 	db_printf("unp_peercred:\n");
2576 	db_print_xucred(2, &unp->unp_peercred);
2577 
2578 	db_printf("unp_refcount: %u\n", unp->unp_refcount);
2579 }
2580 #endif
2581