xref: /freebsd/sys/kern/uipc_usrreq.c (revision ff0ba87247820afbdfdc1b307c803f7923d0e4d3)
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  * 4. 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  * Three types of locks exit in the local domain socket implementation: a
193  * global list mutex, a global linkage rwlock, and per-unpcb mutexes.  Of the
194  * global locks, the list lock protects the socket count, global generation
195  * number, and stream/datagram global lists.  The linkage lock protects the
196  * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be
197  * held exclusively over the acquisition of multiple unpcb locks to prevent
198  * deadlock.
199  *
200  * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
201  * allocated in pru_attach() and freed in pru_detach().  The validity of that
202  * pointer is an invariant, so no lock is required to dereference the so_pcb
203  * pointer if a valid socket reference is held by the caller.  In practice,
204  * this is always true during operations performed on a socket.  Each unpcb
205  * has a back-pointer to its socket, unp_socket, which will be stable under
206  * the same circumstances.
207  *
208  * This pointer may only be safely dereferenced as long as a valid reference
209  * to the unpcb is held.  Typically, this reference will be from the socket,
210  * or from another unpcb when the referring unpcb's lock is held (in order
211  * that the reference not be invalidated during use).  For example, to follow
212  * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
213  * as unp_socket remains valid as long as the reference to unp_conn is valid.
214  *
215  * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx.  Individual
216  * atomic reads without the lock may be performed "lockless", but more
217  * complex reads and read-modify-writes require the mutex to be held.  No
218  * lock order is defined between unpcb locks -- multiple unpcb locks may be
219  * acquired at the same time only when holding the linkage rwlock
220  * exclusively, which prevents deadlocks.
221  *
222  * Blocking with UNIX domain sockets is a tricky issue: unlike most network
223  * protocols, bind() is a non-atomic operation, and connect() requires
224  * potential sleeping in the protocol, due to potentially waiting on local or
225  * distributed file systems.  We try to separate "lookup" operations, which
226  * may sleep, and the IPC operations themselves, which typically can occur
227  * with relative atomicity as locks can be held over the entire operation.
228  *
229  * Another tricky issue is simultaneous multi-threaded or multi-process
230  * access to a single UNIX domain socket.  These are handled by the flags
231  * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
232  * binding, both of which involve dropping UNIX domain socket locks in order
233  * to perform namei() and other file system operations.
234  */
235 static struct rwlock	unp_link_rwlock;
236 static struct mtx	unp_list_lock;
237 static struct mtx	unp_defers_lock;
238 
239 #define	UNP_LINK_LOCK_INIT()		rw_init(&unp_link_rwlock,	\
240 					    "unp_link_rwlock")
241 
242 #define	UNP_LINK_LOCK_ASSERT()	rw_assert(&unp_link_rwlock,	\
243 					    RA_LOCKED)
244 #define	UNP_LINK_UNLOCK_ASSERT()	rw_assert(&unp_link_rwlock,	\
245 					    RA_UNLOCKED)
246 
247 #define	UNP_LINK_RLOCK()		rw_rlock(&unp_link_rwlock)
248 #define	UNP_LINK_RUNLOCK()		rw_runlock(&unp_link_rwlock)
249 #define	UNP_LINK_WLOCK()		rw_wlock(&unp_link_rwlock)
250 #define	UNP_LINK_WUNLOCK()		rw_wunlock(&unp_link_rwlock)
251 #define	UNP_LINK_WLOCK_ASSERT()		rw_assert(&unp_link_rwlock,	\
252 					    RA_WLOCKED)
253 
254 #define	UNP_LIST_LOCK_INIT()		mtx_init(&unp_list_lock,	\
255 					    "unp_list_lock", NULL, MTX_DEF)
256 #define	UNP_LIST_LOCK()			mtx_lock(&unp_list_lock)
257 #define	UNP_LIST_UNLOCK()		mtx_unlock(&unp_list_lock)
258 
259 #define	UNP_DEFERRED_LOCK_INIT()	mtx_init(&unp_defers_lock, \
260 					    "unp_defer", NULL, MTX_DEF)
261 #define	UNP_DEFERRED_LOCK()		mtx_lock(&unp_defers_lock)
262 #define	UNP_DEFERRED_UNLOCK()		mtx_unlock(&unp_defers_lock)
263 
264 #define UNP_PCB_LOCK_INIT(unp)		mtx_init(&(unp)->unp_mtx,	\
265 					    "unp_mtx", "unp_mtx",	\
266 					    MTX_DUPOK|MTX_DEF|MTX_RECURSE)
267 #define	UNP_PCB_LOCK_DESTROY(unp)	mtx_destroy(&(unp)->unp_mtx)
268 #define	UNP_PCB_LOCK(unp)		mtx_lock(&(unp)->unp_mtx)
269 #define	UNP_PCB_UNLOCK(unp)		mtx_unlock(&(unp)->unp_mtx)
270 #define	UNP_PCB_LOCK_ASSERT(unp)	mtx_assert(&(unp)->unp_mtx, MA_OWNED)
271 
272 static int	uipc_connect2(struct socket *, struct socket *);
273 static int	uipc_ctloutput(struct socket *, struct sockopt *);
274 static int	unp_connect(struct socket *, struct sockaddr *,
275 		    struct thread *);
276 static int	unp_connectat(int, struct socket *, struct sockaddr *,
277 		    struct thread *);
278 static int	unp_connect2(struct socket *so, struct socket *so2, int);
279 static void	unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
280 static void	unp_dispose(struct mbuf *);
281 static void	unp_shutdown(struct unpcb *);
282 static void	unp_drop(struct unpcb *, int);
283 static void	unp_gc(__unused void *, int);
284 static void	unp_scan(struct mbuf *, void (*)(struct filedescent **, int));
285 static void	unp_discard(struct file *);
286 static void	unp_freerights(struct filedescent **, int);
287 static void	unp_init(void);
288 static int	unp_internalize(struct mbuf **, struct thread *);
289 static void	unp_internalize_fp(struct file *);
290 static int	unp_externalize(struct mbuf *, struct mbuf **, int);
291 static int	unp_externalize_fp(struct file *);
292 static struct mbuf	*unp_addsockcred(struct thread *, struct mbuf *);
293 static void	unp_process_defers(void * __unused, int);
294 
295 /*
296  * Definitions of protocols supported in the LOCAL domain.
297  */
298 static struct domain localdomain;
299 static struct pr_usrreqs uipc_usrreqs_dgram, uipc_usrreqs_stream;
300 static struct pr_usrreqs uipc_usrreqs_seqpacket;
301 static struct protosw localsw[] = {
302 {
303 	.pr_type =		SOCK_STREAM,
304 	.pr_domain =		&localdomain,
305 	.pr_flags =		PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
306 	.pr_ctloutput =		&uipc_ctloutput,
307 	.pr_usrreqs =		&uipc_usrreqs_stream
308 },
309 {
310 	.pr_type =		SOCK_DGRAM,
311 	.pr_domain =		&localdomain,
312 	.pr_flags =		PR_ATOMIC|PR_ADDR|PR_RIGHTS,
313 	.pr_ctloutput =		&uipc_ctloutput,
314 	.pr_usrreqs =		&uipc_usrreqs_dgram
315 },
316 {
317 	.pr_type =		SOCK_SEQPACKET,
318 	.pr_domain =		&localdomain,
319 
320 	/*
321 	 * XXXRW: For now, PR_ADDR because soreceive will bump into them
322 	 * due to our use of sbappendaddr.  A new sbappend variants is needed
323 	 * that supports both atomic record writes and control data.
324 	 */
325 	.pr_flags =		PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
326 				    PR_RIGHTS,
327 	.pr_ctloutput =		&uipc_ctloutput,
328 	.pr_usrreqs =		&uipc_usrreqs_seqpacket,
329 },
330 };
331 
332 static struct domain localdomain = {
333 	.dom_family =		AF_LOCAL,
334 	.dom_name =		"local",
335 	.dom_init =		unp_init,
336 	.dom_externalize =	unp_externalize,
337 	.dom_dispose =		unp_dispose,
338 	.dom_protosw =		localsw,
339 	.dom_protoswNPROTOSW =	&localsw[sizeof(localsw)/sizeof(localsw[0])]
340 };
341 DOMAIN_SET(local);
342 
343 static void
344 uipc_abort(struct socket *so)
345 {
346 	struct unpcb *unp, *unp2;
347 
348 	unp = sotounpcb(so);
349 	KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
350 
351 	UNP_LINK_WLOCK();
352 	UNP_PCB_LOCK(unp);
353 	unp2 = unp->unp_conn;
354 	if (unp2 != NULL) {
355 		UNP_PCB_LOCK(unp2);
356 		unp_drop(unp2, ECONNABORTED);
357 		UNP_PCB_UNLOCK(unp2);
358 	}
359 	UNP_PCB_UNLOCK(unp);
360 	UNP_LINK_WUNLOCK();
361 }
362 
363 static int
364 uipc_accept(struct socket *so, struct sockaddr **nam)
365 {
366 	struct unpcb *unp, *unp2;
367 	const struct sockaddr *sa;
368 
369 	/*
370 	 * Pass back name of connected socket, if it was bound and we are
371 	 * still connected (our peer may have closed already!).
372 	 */
373 	unp = sotounpcb(so);
374 	KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
375 
376 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
377 	UNP_LINK_RLOCK();
378 	unp2 = unp->unp_conn;
379 	if (unp2 != NULL && unp2->unp_addr != NULL) {
380 		UNP_PCB_LOCK(unp2);
381 		sa = (struct sockaddr *) unp2->unp_addr;
382 		bcopy(sa, *nam, sa->sa_len);
383 		UNP_PCB_UNLOCK(unp2);
384 	} else {
385 		sa = &sun_noname;
386 		bcopy(sa, *nam, sa->sa_len);
387 	}
388 	UNP_LINK_RUNLOCK();
389 	return (0);
390 }
391 
392 static int
393 uipc_attach(struct socket *so, int proto, struct thread *td)
394 {
395 	u_long sendspace, recvspace;
396 	struct unpcb *unp;
397 	int error;
398 
399 	KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
400 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
401 		switch (so->so_type) {
402 		case SOCK_STREAM:
403 			sendspace = unpst_sendspace;
404 			recvspace = unpst_recvspace;
405 			break;
406 
407 		case SOCK_DGRAM:
408 			sendspace = unpdg_sendspace;
409 			recvspace = unpdg_recvspace;
410 			break;
411 
412 		case SOCK_SEQPACKET:
413 			sendspace = unpsp_sendspace;
414 			recvspace = unpsp_recvspace;
415 			break;
416 
417 		default:
418 			panic("uipc_attach");
419 		}
420 		error = soreserve(so, sendspace, recvspace);
421 		if (error)
422 			return (error);
423 	}
424 	unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
425 	if (unp == NULL)
426 		return (ENOBUFS);
427 	LIST_INIT(&unp->unp_refs);
428 	UNP_PCB_LOCK_INIT(unp);
429 	unp->unp_socket = so;
430 	so->so_pcb = unp;
431 	unp->unp_refcount = 1;
432 
433 	UNP_LIST_LOCK();
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 	UNP_LIST_UNLOCK();
453 
454 	return (0);
455 }
456 
457 static int
458 uipc_bindat(int fd, struct socket *so, struct sockaddr *nam, struct thread *td)
459 {
460 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
461 	struct vattr vattr;
462 	int error, namelen;
463 	struct nameidata nd;
464 	struct unpcb *unp;
465 	struct vnode *vp;
466 	struct mount *mp;
467 	cap_rights_t rights;
468 	char *buf;
469 
470 	if (nam->sa_family != AF_UNIX)
471 		return (EAFNOSUPPORT);
472 
473 	unp = sotounpcb(so);
474 	KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
475 
476 	if (soun->sun_len > sizeof(struct sockaddr_un))
477 		return (EINVAL);
478 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
479 	if (namelen <= 0)
480 		return (EINVAL);
481 
482 	/*
483 	 * We don't allow simultaneous bind() calls on a single UNIX domain
484 	 * socket, so flag in-progress operations, and return an error if an
485 	 * operation is already in progress.
486 	 *
487 	 * Historically, we have not allowed a socket to be rebound, so this
488 	 * also returns an error.  Not allowing re-binding simplifies the
489 	 * implementation and avoids a great many possible failure modes.
490 	 */
491 	UNP_PCB_LOCK(unp);
492 	if (unp->unp_vnode != NULL) {
493 		UNP_PCB_UNLOCK(unp);
494 		return (EINVAL);
495 	}
496 	if (unp->unp_flags & UNP_BINDING) {
497 		UNP_PCB_UNLOCK(unp);
498 		return (EALREADY);
499 	}
500 	unp->unp_flags |= UNP_BINDING;
501 	UNP_PCB_UNLOCK(unp);
502 
503 	buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
504 	bcopy(soun->sun_path, buf, namelen);
505 	buf[namelen] = 0;
506 
507 restart:
508 	NDINIT_ATRIGHTS(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME,
509 	    UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_BINDAT), td);
510 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
511 	error = namei(&nd);
512 	if (error)
513 		goto error;
514 	vp = nd.ni_vp;
515 	if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
516 		NDFREE(&nd, NDF_ONLY_PNBUF);
517 		if (nd.ni_dvp == vp)
518 			vrele(nd.ni_dvp);
519 		else
520 			vput(nd.ni_dvp);
521 		if (vp != NULL) {
522 			vrele(vp);
523 			error = EADDRINUSE;
524 			goto error;
525 		}
526 		error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
527 		if (error)
528 			goto error;
529 		goto restart;
530 	}
531 	VATTR_NULL(&vattr);
532 	vattr.va_type = VSOCK;
533 	vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
534 #ifdef MAC
535 	error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
536 	    &vattr);
537 #endif
538 	if (error == 0)
539 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
540 	NDFREE(&nd, NDF_ONLY_PNBUF);
541 	vput(nd.ni_dvp);
542 	if (error) {
543 		vn_finished_write(mp);
544 		goto error;
545 	}
546 	vp = nd.ni_vp;
547 	ASSERT_VOP_ELOCKED(vp, "uipc_bind");
548 	soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
549 
550 	UNP_LINK_WLOCK();
551 	UNP_PCB_LOCK(unp);
552 	VOP_UNP_BIND(vp, unp->unp_socket);
553 	unp->unp_vnode = vp;
554 	unp->unp_addr = soun;
555 	unp->unp_flags &= ~UNP_BINDING;
556 	UNP_PCB_UNLOCK(unp);
557 	UNP_LINK_WUNLOCK();
558 	VOP_UNLOCK(vp, 0);
559 	vn_finished_write(mp);
560 	free(buf, M_TEMP);
561 	return (0);
562 
563 error:
564 	UNP_PCB_LOCK(unp);
565 	unp->unp_flags &= ~UNP_BINDING;
566 	UNP_PCB_UNLOCK(unp);
567 	free(buf, M_TEMP);
568 	return (error);
569 }
570 
571 static int
572 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
573 {
574 
575 	return (uipc_bindat(AT_FDCWD, so, nam, td));
576 }
577 
578 static int
579 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
580 {
581 	int error;
582 
583 	KASSERT(td == curthread, ("uipc_connect: td != curthread"));
584 	UNP_LINK_WLOCK();
585 	error = unp_connect(so, nam, td);
586 	UNP_LINK_WUNLOCK();
587 	return (error);
588 }
589 
590 static int
591 uipc_connectat(int fd, struct socket *so, struct sockaddr *nam,
592     struct thread *td)
593 {
594 	int error;
595 
596 	KASSERT(td == curthread, ("uipc_connectat: td != curthread"));
597 	UNP_LINK_WLOCK();
598 	error = unp_connectat(fd, so, nam, td);
599 	UNP_LINK_WUNLOCK();
600 	return (error);
601 }
602 
603 static void
604 uipc_close(struct socket *so)
605 {
606 	struct unpcb *unp, *unp2;
607 
608 	unp = sotounpcb(so);
609 	KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
610 
611 	UNP_LINK_WLOCK();
612 	UNP_PCB_LOCK(unp);
613 	unp2 = unp->unp_conn;
614 	if (unp2 != NULL) {
615 		UNP_PCB_LOCK(unp2);
616 		unp_disconnect(unp, unp2);
617 		UNP_PCB_UNLOCK(unp2);
618 	}
619 	UNP_PCB_UNLOCK(unp);
620 	UNP_LINK_WUNLOCK();
621 }
622 
623 static int
624 uipc_connect2(struct socket *so1, struct socket *so2)
625 {
626 	struct unpcb *unp, *unp2;
627 	int error;
628 
629 	UNP_LINK_WLOCK();
630 	unp = so1->so_pcb;
631 	KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
632 	UNP_PCB_LOCK(unp);
633 	unp2 = so2->so_pcb;
634 	KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
635 	UNP_PCB_LOCK(unp2);
636 	error = unp_connect2(so1, so2, PRU_CONNECT2);
637 	UNP_PCB_UNLOCK(unp2);
638 	UNP_PCB_UNLOCK(unp);
639 	UNP_LINK_WUNLOCK();
640 	return (error);
641 }
642 
643 static void
644 uipc_detach(struct socket *so)
645 {
646 	struct unpcb *unp, *unp2;
647 	struct sockaddr_un *saved_unp_addr;
648 	struct vnode *vp;
649 	int freeunp, local_unp_rights;
650 
651 	unp = sotounpcb(so);
652 	KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
653 
654 	UNP_LINK_WLOCK();
655 	UNP_LIST_LOCK();
656 	UNP_PCB_LOCK(unp);
657 	LIST_REMOVE(unp, unp_link);
658 	unp->unp_gencnt = ++unp_gencnt;
659 	--unp_count;
660 	UNP_LIST_UNLOCK();
661 
662 	/*
663 	 * XXXRW: Should assert vp->v_socket == so.
664 	 */
665 	if ((vp = unp->unp_vnode) != NULL) {
666 		VOP_UNP_DETACH(vp);
667 		unp->unp_vnode = NULL;
668 	}
669 	unp2 = unp->unp_conn;
670 	if (unp2 != NULL) {
671 		UNP_PCB_LOCK(unp2);
672 		unp_disconnect(unp, unp2);
673 		UNP_PCB_UNLOCK(unp2);
674 	}
675 
676 	/*
677 	 * We hold the linkage lock exclusively, so it's OK to acquire
678 	 * multiple pcb locks at a time.
679 	 */
680 	while (!LIST_EMPTY(&unp->unp_refs)) {
681 		struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
682 
683 		UNP_PCB_LOCK(ref);
684 		unp_drop(ref, ECONNRESET);
685 		UNP_PCB_UNLOCK(ref);
686 	}
687 	local_unp_rights = unp_rights;
688 	UNP_LINK_WUNLOCK();
689 	unp->unp_socket->so_pcb = NULL;
690 	saved_unp_addr = unp->unp_addr;
691 	unp->unp_addr = NULL;
692 	unp->unp_refcount--;
693 	freeunp = (unp->unp_refcount == 0);
694 	if (saved_unp_addr != NULL)
695 		free(saved_unp_addr, M_SONAME);
696 	if (freeunp) {
697 		UNP_PCB_LOCK_DESTROY(unp);
698 		uma_zfree(unp_zone, unp);
699 	} else
700 		UNP_PCB_UNLOCK(unp);
701 	if (vp)
702 		vrele(vp);
703 	if (local_unp_rights)
704 		taskqueue_enqueue_timeout(taskqueue_thread, &unp_gc_task, -1);
705 }
706 
707 static int
708 uipc_disconnect(struct socket *so)
709 {
710 	struct unpcb *unp, *unp2;
711 
712 	unp = sotounpcb(so);
713 	KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
714 
715 	UNP_LINK_WLOCK();
716 	UNP_PCB_LOCK(unp);
717 	unp2 = unp->unp_conn;
718 	if (unp2 != NULL) {
719 		UNP_PCB_LOCK(unp2);
720 		unp_disconnect(unp, unp2);
721 		UNP_PCB_UNLOCK(unp2);
722 	}
723 	UNP_PCB_UNLOCK(unp);
724 	UNP_LINK_WUNLOCK();
725 	return (0);
726 }
727 
728 static int
729 uipc_listen(struct socket *so, int backlog, struct thread *td)
730 {
731 	struct unpcb *unp;
732 	int error;
733 
734 	unp = sotounpcb(so);
735 	KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
736 
737 	UNP_PCB_LOCK(unp);
738 	if (unp->unp_vnode == NULL) {
739 		UNP_PCB_UNLOCK(unp);
740 		return (EINVAL);
741 	}
742 
743 	SOCK_LOCK(so);
744 	error = solisten_proto_check(so);
745 	if (error == 0) {
746 		cru2x(td->td_ucred, &unp->unp_peercred);
747 		unp->unp_flags |= UNP_HAVEPCCACHED;
748 		solisten_proto(so, backlog);
749 	}
750 	SOCK_UNLOCK(so);
751 	UNP_PCB_UNLOCK(unp);
752 	return (error);
753 }
754 
755 static int
756 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
757 {
758 	struct unpcb *unp, *unp2;
759 	const struct sockaddr *sa;
760 
761 	unp = sotounpcb(so);
762 	KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
763 
764 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
765 	UNP_LINK_RLOCK();
766 	/*
767 	 * XXX: It seems that this test always fails even when connection is
768 	 * established.  So, this else clause is added as workaround to
769 	 * return PF_LOCAL sockaddr.
770 	 */
771 	unp2 = unp->unp_conn;
772 	if (unp2 != NULL) {
773 		UNP_PCB_LOCK(unp2);
774 		if (unp2->unp_addr != NULL)
775 			sa = (struct sockaddr *) unp2->unp_addr;
776 		else
777 			sa = &sun_noname;
778 		bcopy(sa, *nam, sa->sa_len);
779 		UNP_PCB_UNLOCK(unp2);
780 	} else {
781 		sa = &sun_noname;
782 		bcopy(sa, *nam, sa->sa_len);
783 	}
784 	UNP_LINK_RUNLOCK();
785 	return (0);
786 }
787 
788 static int
789 uipc_rcvd(struct socket *so, int flags)
790 {
791 	struct unpcb *unp, *unp2;
792 	struct socket *so2;
793 	u_int mbcnt, sbcc;
794 
795 	unp = sotounpcb(so);
796 	KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
797 	KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET,
798 	    ("%s: socktype %d", __func__, so->so_type));
799 
800 	/*
801 	 * Adjust backpressure on sender and wakeup any waiting to write.
802 	 *
803 	 * The unp lock is acquired to maintain the validity of the unp_conn
804 	 * pointer; no lock on unp2 is required as unp2->unp_socket will be
805 	 * static as long as we don't permit unp2 to disconnect from unp,
806 	 * which is prevented by the lock on unp.  We cache values from
807 	 * so_rcv to avoid holding the so_rcv lock over the entire
808 	 * transaction on the remote so_snd.
809 	 */
810 	SOCKBUF_LOCK(&so->so_rcv);
811 	mbcnt = so->so_rcv.sb_mbcnt;
812 	sbcc = sbavail(&so->so_rcv);
813 	SOCKBUF_UNLOCK(&so->so_rcv);
814 	/*
815 	 * There is a benign race condition at this point.  If we're planning to
816 	 * clear SB_STOP, but uipc_send is called on the connected socket at
817 	 * this instant, it might add data to the sockbuf and set SB_STOP.  Then
818 	 * we would erroneously clear SB_STOP below, even though the sockbuf is
819 	 * full.  The race is benign because the only ill effect is to allow the
820 	 * sockbuf to exceed its size limit, and the size limits are not
821 	 * strictly guaranteed anyway.
822 	 */
823 	UNP_PCB_LOCK(unp);
824 	unp2 = unp->unp_conn;
825 	if (unp2 == NULL) {
826 		UNP_PCB_UNLOCK(unp);
827 		return (0);
828 	}
829 	so2 = unp2->unp_socket;
830 	SOCKBUF_LOCK(&so2->so_snd);
831 	if (sbcc < so2->so_snd.sb_hiwat && mbcnt < so2->so_snd.sb_mbmax)
832 		so2->so_snd.sb_flags &= ~SB_STOP;
833 	sowwakeup_locked(so2);
834 	UNP_PCB_UNLOCK(unp);
835 	return (0);
836 }
837 
838 static int
839 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
840     struct mbuf *control, struct thread *td)
841 {
842 	struct unpcb *unp, *unp2;
843 	struct socket *so2;
844 	u_int mbcnt, sbcc;
845 	int error = 0;
846 
847 	unp = sotounpcb(so);
848 	KASSERT(unp != NULL, ("%s: unp == NULL", __func__));
849 	KASSERT(so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM ||
850 	    so->so_type == SOCK_SEQPACKET,
851 	    ("%s: socktype %d", __func__, so->so_type));
852 
853 	if (flags & PRUS_OOB) {
854 		error = EOPNOTSUPP;
855 		goto release;
856 	}
857 	if (control != NULL && (error = unp_internalize(&control, td)))
858 		goto release;
859 	if ((nam != NULL) || (flags & PRUS_EOF))
860 		UNP_LINK_WLOCK();
861 	else
862 		UNP_LINK_RLOCK();
863 	switch (so->so_type) {
864 	case SOCK_DGRAM:
865 	{
866 		const struct sockaddr *from;
867 
868 		unp2 = unp->unp_conn;
869 		if (nam != NULL) {
870 			UNP_LINK_WLOCK_ASSERT();
871 			if (unp2 != NULL) {
872 				error = EISCONN;
873 				break;
874 			}
875 			error = unp_connect(so, nam, td);
876 			if (error)
877 				break;
878 			unp2 = unp->unp_conn;
879 		}
880 
881 		/*
882 		 * Because connect() and send() are non-atomic in a sendto()
883 		 * with a target address, it's possible that the socket will
884 		 * have disconnected before the send() can run.  In that case
885 		 * return the slightly counter-intuitive but otherwise
886 		 * correct error that the socket is not connected.
887 		 */
888 		if (unp2 == NULL) {
889 			error = ENOTCONN;
890 			break;
891 		}
892 		/* Lockless read. */
893 		if (unp2->unp_flags & UNP_WANTCRED)
894 			control = unp_addsockcred(td, control);
895 		UNP_PCB_LOCK(unp);
896 		if (unp->unp_addr != NULL)
897 			from = (struct sockaddr *)unp->unp_addr;
898 		else
899 			from = &sun_noname;
900 		so2 = unp2->unp_socket;
901 		SOCKBUF_LOCK(&so2->so_rcv);
902 		if (sbappendaddr_locked(&so2->so_rcv, from, m,
903 		    control)) {
904 			sorwakeup_locked(so2);
905 			m = NULL;
906 			control = NULL;
907 		} else {
908 			SOCKBUF_UNLOCK(&so2->so_rcv);
909 			error = ENOBUFS;
910 		}
911 		if (nam != NULL) {
912 			UNP_LINK_WLOCK_ASSERT();
913 			UNP_PCB_LOCK(unp2);
914 			unp_disconnect(unp, unp2);
915 			UNP_PCB_UNLOCK(unp2);
916 		}
917 		UNP_PCB_UNLOCK(unp);
918 		break;
919 	}
920 
921 	case SOCK_SEQPACKET:
922 	case SOCK_STREAM:
923 		if ((so->so_state & SS_ISCONNECTED) == 0) {
924 			if (nam != NULL) {
925 				UNP_LINK_WLOCK_ASSERT();
926 				error = unp_connect(so, nam, td);
927 				if (error)
928 					break;	/* XXX */
929 			} else {
930 				error = ENOTCONN;
931 				break;
932 			}
933 		}
934 
935 		/* Lockless read. */
936 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
937 			error = EPIPE;
938 			break;
939 		}
940 
941 		/*
942 		 * Because connect() and send() are non-atomic in a sendto()
943 		 * with a target address, it's possible that the socket will
944 		 * have disconnected before the send() can run.  In that case
945 		 * return the slightly counter-intuitive but otherwise
946 		 * correct error that the socket is not connected.
947 		 *
948 		 * Locking here must be done carefully: the linkage lock
949 		 * prevents interconnections between unpcbs from changing, so
950 		 * we can traverse from unp to unp2 without acquiring unp's
951 		 * lock.  Socket buffer locks follow unpcb locks, so we can
952 		 * acquire both remote and lock socket buffer locks.
953 		 */
954 		unp2 = unp->unp_conn;
955 		if (unp2 == NULL) {
956 			error = ENOTCONN;
957 			break;
958 		}
959 		so2 = unp2->unp_socket;
960 		UNP_PCB_LOCK(unp2);
961 		SOCKBUF_LOCK(&so2->so_rcv);
962 		if (unp2->unp_flags & UNP_WANTCRED) {
963 			/*
964 			 * Credentials are passed only once on SOCK_STREAM
965 			 * and SOCK_SEQPACKET.
966 			 */
967 			unp2->unp_flags &= ~UNP_WANTCRED;
968 			control = unp_addsockcred(td, control);
969 		}
970 		/*
971 		 * Send to paired receive port, and then reduce send buffer
972 		 * hiwater marks to maintain backpressure.  Wake up readers.
973 		 */
974 		switch (so->so_type) {
975 		case SOCK_STREAM:
976 			if (control != NULL) {
977 				if (sbappendcontrol_locked(&so2->so_rcv, m,
978 				    control))
979 					control = NULL;
980 			} else
981 				sbappend_locked(&so2->so_rcv, m);
982 			break;
983 
984 		case SOCK_SEQPACKET: {
985 			const struct sockaddr *from;
986 
987 			from = &sun_noname;
988 			/*
989 			 * Don't check for space available in so2->so_rcv.
990 			 * Unix domain sockets only check for space in the
991 			 * sending sockbuf, and that check is performed one
992 			 * level up the stack.
993 			 */
994 			if (sbappendaddr_nospacecheck_locked(&so2->so_rcv,
995 				from, m, control))
996 				control = NULL;
997 			break;
998 			}
999 		}
1000 
1001 		mbcnt = so2->so_rcv.sb_mbcnt;
1002 		sbcc = sbavail(&so2->so_rcv);
1003 		if (sbcc)
1004 			sorwakeup_locked(so2);
1005 		else
1006 			SOCKBUF_UNLOCK(&so2->so_rcv);
1007 
1008 		/*
1009 		 * The PCB lock on unp2 protects the SB_STOP flag.  Without it,
1010 		 * it would be possible for uipc_rcvd to be called at this
1011 		 * point, drain the receiving sockbuf, clear SB_STOP, and then
1012 		 * we would set SB_STOP below.  That could lead to an empty
1013 		 * sockbuf having SB_STOP set
1014 		 */
1015 		SOCKBUF_LOCK(&so->so_snd);
1016 		if (sbcc >= so->so_snd.sb_hiwat || mbcnt >= so->so_snd.sb_mbmax)
1017 			so->so_snd.sb_flags |= SB_STOP;
1018 		SOCKBUF_UNLOCK(&so->so_snd);
1019 		UNP_PCB_UNLOCK(unp2);
1020 		m = NULL;
1021 		break;
1022 	}
1023 
1024 	/*
1025 	 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
1026 	 */
1027 	if (flags & PRUS_EOF) {
1028 		UNP_PCB_LOCK(unp);
1029 		socantsendmore(so);
1030 		unp_shutdown(unp);
1031 		UNP_PCB_UNLOCK(unp);
1032 	}
1033 
1034 	if ((nam != NULL) || (flags & PRUS_EOF))
1035 		UNP_LINK_WUNLOCK();
1036 	else
1037 		UNP_LINK_RUNLOCK();
1038 
1039 	if (control != NULL && error != 0)
1040 		unp_dispose(control);
1041 
1042 release:
1043 	if (control != NULL)
1044 		m_freem(control);
1045 	if (m != NULL)
1046 		m_freem(m);
1047 	return (error);
1048 }
1049 
1050 static int
1051 uipc_sense(struct socket *so, struct stat *sb)
1052 {
1053 	struct unpcb *unp;
1054 
1055 	unp = sotounpcb(so);
1056 	KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1057 
1058 	sb->st_blksize = so->so_snd.sb_hiwat;
1059 	UNP_PCB_LOCK(unp);
1060 	sb->st_dev = NODEV;
1061 	if (unp->unp_ino == 0)
1062 		unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1063 	sb->st_ino = unp->unp_ino;
1064 	UNP_PCB_UNLOCK(unp);
1065 	return (0);
1066 }
1067 
1068 static int
1069 uipc_shutdown(struct socket *so)
1070 {
1071 	struct unpcb *unp;
1072 
1073 	unp = sotounpcb(so);
1074 	KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1075 
1076 	UNP_LINK_WLOCK();
1077 	UNP_PCB_LOCK(unp);
1078 	socantsendmore(so);
1079 	unp_shutdown(unp);
1080 	UNP_PCB_UNLOCK(unp);
1081 	UNP_LINK_WUNLOCK();
1082 	return (0);
1083 }
1084 
1085 static int
1086 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1087 {
1088 	struct unpcb *unp;
1089 	const struct sockaddr *sa;
1090 
1091 	unp = sotounpcb(so);
1092 	KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1093 
1094 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1095 	UNP_PCB_LOCK(unp);
1096 	if (unp->unp_addr != NULL)
1097 		sa = (struct sockaddr *) unp->unp_addr;
1098 	else
1099 		sa = &sun_noname;
1100 	bcopy(sa, *nam, sa->sa_len);
1101 	UNP_PCB_UNLOCK(unp);
1102 	return (0);
1103 }
1104 
1105 static struct pr_usrreqs uipc_usrreqs_dgram = {
1106 	.pru_abort = 		uipc_abort,
1107 	.pru_accept =		uipc_accept,
1108 	.pru_attach =		uipc_attach,
1109 	.pru_bind =		uipc_bind,
1110 	.pru_bindat =		uipc_bindat,
1111 	.pru_connect =		uipc_connect,
1112 	.pru_connectat =	uipc_connectat,
1113 	.pru_connect2 =		uipc_connect2,
1114 	.pru_detach =		uipc_detach,
1115 	.pru_disconnect =	uipc_disconnect,
1116 	.pru_listen =		uipc_listen,
1117 	.pru_peeraddr =		uipc_peeraddr,
1118 	.pru_rcvd =		uipc_rcvd,
1119 	.pru_send =		uipc_send,
1120 	.pru_sense =		uipc_sense,
1121 	.pru_shutdown =		uipc_shutdown,
1122 	.pru_sockaddr =		uipc_sockaddr,
1123 	.pru_soreceive =	soreceive_dgram,
1124 	.pru_close =		uipc_close,
1125 };
1126 
1127 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1128 	.pru_abort =		uipc_abort,
1129 	.pru_accept =		uipc_accept,
1130 	.pru_attach =		uipc_attach,
1131 	.pru_bind =		uipc_bind,
1132 	.pru_bindat =		uipc_bindat,
1133 	.pru_connect =		uipc_connect,
1134 	.pru_connectat =	uipc_connectat,
1135 	.pru_connect2 =		uipc_connect2,
1136 	.pru_detach =		uipc_detach,
1137 	.pru_disconnect =	uipc_disconnect,
1138 	.pru_listen =		uipc_listen,
1139 	.pru_peeraddr =		uipc_peeraddr,
1140 	.pru_rcvd =		uipc_rcvd,
1141 	.pru_send =		uipc_send,
1142 	.pru_sense =		uipc_sense,
1143 	.pru_shutdown =		uipc_shutdown,
1144 	.pru_sockaddr =		uipc_sockaddr,
1145 	.pru_soreceive =	soreceive_generic,	/* XXX: or...? */
1146 	.pru_close =		uipc_close,
1147 };
1148 
1149 static struct pr_usrreqs uipc_usrreqs_stream = {
1150 	.pru_abort = 		uipc_abort,
1151 	.pru_accept =		uipc_accept,
1152 	.pru_attach =		uipc_attach,
1153 	.pru_bind =		uipc_bind,
1154 	.pru_bindat =		uipc_bindat,
1155 	.pru_connect =		uipc_connect,
1156 	.pru_connectat =	uipc_connectat,
1157 	.pru_connect2 =		uipc_connect2,
1158 	.pru_detach =		uipc_detach,
1159 	.pru_disconnect =	uipc_disconnect,
1160 	.pru_listen =		uipc_listen,
1161 	.pru_peeraddr =		uipc_peeraddr,
1162 	.pru_rcvd =		uipc_rcvd,
1163 	.pru_send =		uipc_send,
1164 	.pru_sense =		uipc_sense,
1165 	.pru_shutdown =		uipc_shutdown,
1166 	.pru_sockaddr =		uipc_sockaddr,
1167 	.pru_soreceive =	soreceive_generic,
1168 	.pru_close =		uipc_close,
1169 };
1170 
1171 static int
1172 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1173 {
1174 	struct unpcb *unp;
1175 	struct xucred xu;
1176 	int error, optval;
1177 
1178 	if (sopt->sopt_level != 0)
1179 		return (EINVAL);
1180 
1181 	unp = sotounpcb(so);
1182 	KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1183 	error = 0;
1184 	switch (sopt->sopt_dir) {
1185 	case SOPT_GET:
1186 		switch (sopt->sopt_name) {
1187 		case LOCAL_PEERCRED:
1188 			UNP_PCB_LOCK(unp);
1189 			if (unp->unp_flags & UNP_HAVEPC)
1190 				xu = unp->unp_peercred;
1191 			else {
1192 				if (so->so_type == SOCK_STREAM)
1193 					error = ENOTCONN;
1194 				else
1195 					error = EINVAL;
1196 			}
1197 			UNP_PCB_UNLOCK(unp);
1198 			if (error == 0)
1199 				error = sooptcopyout(sopt, &xu, sizeof(xu));
1200 			break;
1201 
1202 		case LOCAL_CREDS:
1203 			/* Unlocked read. */
1204 			optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1205 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1206 			break;
1207 
1208 		case LOCAL_CONNWAIT:
1209 			/* Unlocked read. */
1210 			optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1211 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1212 			break;
1213 
1214 		default:
1215 			error = EOPNOTSUPP;
1216 			break;
1217 		}
1218 		break;
1219 
1220 	case SOPT_SET:
1221 		switch (sopt->sopt_name) {
1222 		case LOCAL_CREDS:
1223 		case LOCAL_CONNWAIT:
1224 			error = sooptcopyin(sopt, &optval, sizeof(optval),
1225 					    sizeof(optval));
1226 			if (error)
1227 				break;
1228 
1229 #define	OPTSET(bit) do {						\
1230 	UNP_PCB_LOCK(unp);						\
1231 	if (optval)							\
1232 		unp->unp_flags |= bit;					\
1233 	else								\
1234 		unp->unp_flags &= ~bit;					\
1235 	UNP_PCB_UNLOCK(unp);						\
1236 } while (0)
1237 
1238 			switch (sopt->sopt_name) {
1239 			case LOCAL_CREDS:
1240 				OPTSET(UNP_WANTCRED);
1241 				break;
1242 
1243 			case LOCAL_CONNWAIT:
1244 				OPTSET(UNP_CONNWAIT);
1245 				break;
1246 
1247 			default:
1248 				break;
1249 			}
1250 			break;
1251 #undef	OPTSET
1252 		default:
1253 			error = ENOPROTOOPT;
1254 			break;
1255 		}
1256 		break;
1257 
1258 	default:
1259 		error = EOPNOTSUPP;
1260 		break;
1261 	}
1262 	return (error);
1263 }
1264 
1265 static int
1266 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1267 {
1268 
1269 	return (unp_connectat(AT_FDCWD, so, nam, td));
1270 }
1271 
1272 static int
1273 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1274     struct thread *td)
1275 {
1276 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1277 	struct vnode *vp;
1278 	struct socket *so2, *so3;
1279 	struct unpcb *unp, *unp2, *unp3;
1280 	struct nameidata nd;
1281 	char buf[SOCK_MAXADDRLEN];
1282 	struct sockaddr *sa;
1283 	cap_rights_t rights;
1284 	int error, len;
1285 
1286 	if (nam->sa_family != AF_UNIX)
1287 		return (EAFNOSUPPORT);
1288 
1289 	UNP_LINK_WLOCK_ASSERT();
1290 
1291 	unp = sotounpcb(so);
1292 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1293 
1294 	if (nam->sa_len > sizeof(struct sockaddr_un))
1295 		return (EINVAL);
1296 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1297 	if (len <= 0)
1298 		return (EINVAL);
1299 	bcopy(soun->sun_path, buf, len);
1300 	buf[len] = 0;
1301 
1302 	UNP_PCB_LOCK(unp);
1303 	if (unp->unp_flags & UNP_CONNECTING) {
1304 		UNP_PCB_UNLOCK(unp);
1305 		return (EALREADY);
1306 	}
1307 	UNP_LINK_WUNLOCK();
1308 	unp->unp_flags |= UNP_CONNECTING;
1309 	UNP_PCB_UNLOCK(unp);
1310 
1311 	sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1312 	NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1313 	    UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1314 	error = namei(&nd);
1315 	if (error)
1316 		vp = NULL;
1317 	else
1318 		vp = nd.ni_vp;
1319 	ASSERT_VOP_LOCKED(vp, "unp_connect");
1320 	NDFREE(&nd, NDF_ONLY_PNBUF);
1321 	if (error)
1322 		goto bad;
1323 
1324 	if (vp->v_type != VSOCK) {
1325 		error = ENOTSOCK;
1326 		goto bad;
1327 	}
1328 #ifdef MAC
1329 	error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1330 	if (error)
1331 		goto bad;
1332 #endif
1333 	error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1334 	if (error)
1335 		goto bad;
1336 
1337 	unp = sotounpcb(so);
1338 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1339 
1340 	/*
1341 	 * Lock linkage lock for two reasons: make sure v_socket is stable,
1342 	 * and to protect simultaneous locking of multiple pcbs.
1343 	 */
1344 	UNP_LINK_WLOCK();
1345 	VOP_UNP_CONNECT(vp, &so2);
1346 	if (so2 == NULL) {
1347 		error = ECONNREFUSED;
1348 		goto bad2;
1349 	}
1350 	if (so->so_type != so2->so_type) {
1351 		error = EPROTOTYPE;
1352 		goto bad2;
1353 	}
1354 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1355 		if (so2->so_options & SO_ACCEPTCONN) {
1356 			CURVNET_SET(so2->so_vnet);
1357 			so3 = sonewconn(so2, 0);
1358 			CURVNET_RESTORE();
1359 		} else
1360 			so3 = NULL;
1361 		if (so3 == NULL) {
1362 			error = ECONNREFUSED;
1363 			goto bad2;
1364 		}
1365 		unp = sotounpcb(so);
1366 		unp2 = sotounpcb(so2);
1367 		unp3 = sotounpcb(so3);
1368 		UNP_PCB_LOCK(unp);
1369 		UNP_PCB_LOCK(unp2);
1370 		UNP_PCB_LOCK(unp3);
1371 		if (unp2->unp_addr != NULL) {
1372 			bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1373 			unp3->unp_addr = (struct sockaddr_un *) sa;
1374 			sa = NULL;
1375 		}
1376 
1377 		/*
1378 		 * The connector's (client's) credentials are copied from its
1379 		 * process structure at the time of connect() (which is now).
1380 		 */
1381 		cru2x(td->td_ucred, &unp3->unp_peercred);
1382 		unp3->unp_flags |= UNP_HAVEPC;
1383 
1384 		/*
1385 		 * The receiver's (server's) credentials are copied from the
1386 		 * unp_peercred member of socket on which the former called
1387 		 * listen(); uipc_listen() cached that process's credentials
1388 		 * at that time so we can use them now.
1389 		 */
1390 		KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1391 		    ("unp_connect: listener without cached peercred"));
1392 		memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1393 		    sizeof(unp->unp_peercred));
1394 		unp->unp_flags |= UNP_HAVEPC;
1395 		if (unp2->unp_flags & UNP_WANTCRED)
1396 			unp3->unp_flags |= UNP_WANTCRED;
1397 		UNP_PCB_UNLOCK(unp3);
1398 		UNP_PCB_UNLOCK(unp2);
1399 		UNP_PCB_UNLOCK(unp);
1400 #ifdef MAC
1401 		mac_socketpeer_set_from_socket(so, so3);
1402 		mac_socketpeer_set_from_socket(so3, so);
1403 #endif
1404 
1405 		so2 = so3;
1406 	}
1407 	unp = sotounpcb(so);
1408 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1409 	unp2 = sotounpcb(so2);
1410 	KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1411 	UNP_PCB_LOCK(unp);
1412 	UNP_PCB_LOCK(unp2);
1413 	error = unp_connect2(so, so2, PRU_CONNECT);
1414 	UNP_PCB_UNLOCK(unp2);
1415 	UNP_PCB_UNLOCK(unp);
1416 bad2:
1417 	UNP_LINK_WUNLOCK();
1418 bad:
1419 	if (vp != NULL)
1420 		vput(vp);
1421 	free(sa, M_SONAME);
1422 	UNP_LINK_WLOCK();
1423 	UNP_PCB_LOCK(unp);
1424 	unp->unp_flags &= ~UNP_CONNECTING;
1425 	UNP_PCB_UNLOCK(unp);
1426 	return (error);
1427 }
1428 
1429 static int
1430 unp_connect2(struct socket *so, struct socket *so2, int req)
1431 {
1432 	struct unpcb *unp;
1433 	struct unpcb *unp2;
1434 
1435 	unp = sotounpcb(so);
1436 	KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1437 	unp2 = sotounpcb(so2);
1438 	KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1439 
1440 	UNP_LINK_WLOCK_ASSERT();
1441 	UNP_PCB_LOCK_ASSERT(unp);
1442 	UNP_PCB_LOCK_ASSERT(unp2);
1443 
1444 	if (so2->so_type != so->so_type)
1445 		return (EPROTOTYPE);
1446 	unp->unp_conn = unp2;
1447 
1448 	switch (so->so_type) {
1449 	case SOCK_DGRAM:
1450 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1451 		soisconnected(so);
1452 		break;
1453 
1454 	case SOCK_STREAM:
1455 	case SOCK_SEQPACKET:
1456 		unp2->unp_conn = unp;
1457 		if (req == PRU_CONNECT &&
1458 		    ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1459 			soisconnecting(so);
1460 		else
1461 			soisconnected(so);
1462 		soisconnected(so2);
1463 		break;
1464 
1465 	default:
1466 		panic("unp_connect2");
1467 	}
1468 	return (0);
1469 }
1470 
1471 static void
1472 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1473 {
1474 	struct socket *so;
1475 
1476 	KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1477 
1478 	UNP_LINK_WLOCK_ASSERT();
1479 	UNP_PCB_LOCK_ASSERT(unp);
1480 	UNP_PCB_LOCK_ASSERT(unp2);
1481 
1482 	unp->unp_conn = NULL;
1483 	switch (unp->unp_socket->so_type) {
1484 	case SOCK_DGRAM:
1485 		LIST_REMOVE(unp, unp_reflink);
1486 		so = unp->unp_socket;
1487 		SOCK_LOCK(so);
1488 		so->so_state &= ~SS_ISCONNECTED;
1489 		SOCK_UNLOCK(so);
1490 		break;
1491 
1492 	case SOCK_STREAM:
1493 	case SOCK_SEQPACKET:
1494 		soisdisconnected(unp->unp_socket);
1495 		unp2->unp_conn = NULL;
1496 		soisdisconnected(unp2->unp_socket);
1497 		break;
1498 	}
1499 }
1500 
1501 /*
1502  * unp_pcblist() walks the global list of struct unpcb's to generate a
1503  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1504  * sequentially, validating the generation number on each to see if it has
1505  * been detached.  All of this is necessary because copyout() may sleep on
1506  * disk I/O.
1507  */
1508 static int
1509 unp_pcblist(SYSCTL_HANDLER_ARGS)
1510 {
1511 	int error, i, n;
1512 	int freeunp;
1513 	struct unpcb *unp, **unp_list;
1514 	unp_gen_t gencnt;
1515 	struct xunpgen *xug;
1516 	struct unp_head *head;
1517 	struct xunpcb *xu;
1518 
1519 	switch ((intptr_t)arg1) {
1520 	case SOCK_STREAM:
1521 		head = &unp_shead;
1522 		break;
1523 
1524 	case SOCK_DGRAM:
1525 		head = &unp_dhead;
1526 		break;
1527 
1528 	case SOCK_SEQPACKET:
1529 		head = &unp_sphead;
1530 		break;
1531 
1532 	default:
1533 		panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1534 	}
1535 
1536 	/*
1537 	 * The process of preparing the PCB list is too time-consuming and
1538 	 * resource-intensive to repeat twice on every request.
1539 	 */
1540 	if (req->oldptr == NULL) {
1541 		n = unp_count;
1542 		req->oldidx = 2 * (sizeof *xug)
1543 			+ (n + n/8) * sizeof(struct xunpcb);
1544 		return (0);
1545 	}
1546 
1547 	if (req->newptr != NULL)
1548 		return (EPERM);
1549 
1550 	/*
1551 	 * OK, now we're committed to doing something.
1552 	 */
1553 	xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1554 	UNP_LIST_LOCK();
1555 	gencnt = unp_gencnt;
1556 	n = unp_count;
1557 	UNP_LIST_UNLOCK();
1558 
1559 	xug->xug_len = sizeof *xug;
1560 	xug->xug_count = n;
1561 	xug->xug_gen = gencnt;
1562 	xug->xug_sogen = so_gencnt;
1563 	error = SYSCTL_OUT(req, xug, sizeof *xug);
1564 	if (error) {
1565 		free(xug, M_TEMP);
1566 		return (error);
1567 	}
1568 
1569 	unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1570 
1571 	UNP_LIST_LOCK();
1572 	for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1573 	     unp = LIST_NEXT(unp, unp_link)) {
1574 		UNP_PCB_LOCK(unp);
1575 		if (unp->unp_gencnt <= gencnt) {
1576 			if (cr_cansee(req->td->td_ucred,
1577 			    unp->unp_socket->so_cred)) {
1578 				UNP_PCB_UNLOCK(unp);
1579 				continue;
1580 			}
1581 			unp_list[i++] = unp;
1582 			unp->unp_refcount++;
1583 		}
1584 		UNP_PCB_UNLOCK(unp);
1585 	}
1586 	UNP_LIST_UNLOCK();
1587 	n = i;			/* In case we lost some during malloc. */
1588 
1589 	error = 0;
1590 	xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1591 	for (i = 0; i < n; i++) {
1592 		unp = unp_list[i];
1593 		UNP_PCB_LOCK(unp);
1594 		unp->unp_refcount--;
1595 	        if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1596 			xu->xu_len = sizeof *xu;
1597 			xu->xu_unpp = unp;
1598 			/*
1599 			 * XXX - need more locking here to protect against
1600 			 * connect/disconnect races for SMP.
1601 			 */
1602 			if (unp->unp_addr != NULL)
1603 				bcopy(unp->unp_addr, &xu->xu_addr,
1604 				      unp->unp_addr->sun_len);
1605 			if (unp->unp_conn != NULL &&
1606 			    unp->unp_conn->unp_addr != NULL)
1607 				bcopy(unp->unp_conn->unp_addr,
1608 				      &xu->xu_caddr,
1609 				      unp->unp_conn->unp_addr->sun_len);
1610 			bcopy(unp, &xu->xu_unp, sizeof *unp);
1611 			sotoxsocket(unp->unp_socket, &xu->xu_socket);
1612 			UNP_PCB_UNLOCK(unp);
1613 			error = SYSCTL_OUT(req, xu, sizeof *xu);
1614 		} else {
1615 			freeunp = (unp->unp_refcount == 0);
1616 			UNP_PCB_UNLOCK(unp);
1617 			if (freeunp) {
1618 				UNP_PCB_LOCK_DESTROY(unp);
1619 				uma_zfree(unp_zone, unp);
1620 			}
1621 		}
1622 	}
1623 	free(xu, M_TEMP);
1624 	if (!error) {
1625 		/*
1626 		 * Give the user an updated idea of our state.  If the
1627 		 * generation differs from what we told her before, she knows
1628 		 * that something happened while we were processing this
1629 		 * request, and it might be necessary to retry.
1630 		 */
1631 		xug->xug_gen = unp_gencnt;
1632 		xug->xug_sogen = so_gencnt;
1633 		xug->xug_count = unp_count;
1634 		error = SYSCTL_OUT(req, xug, sizeof *xug);
1635 	}
1636 	free(unp_list, M_TEMP);
1637 	free(xug, M_TEMP);
1638 	return (error);
1639 }
1640 
1641 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1642     (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1643     "List of active local datagram sockets");
1644 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1645     (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1646     "List of active local stream sockets");
1647 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1648     CTLTYPE_OPAQUE | CTLFLAG_RD,
1649     (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1650     "List of active local seqpacket sockets");
1651 
1652 static void
1653 unp_shutdown(struct unpcb *unp)
1654 {
1655 	struct unpcb *unp2;
1656 	struct socket *so;
1657 
1658 	UNP_LINK_WLOCK_ASSERT();
1659 	UNP_PCB_LOCK_ASSERT(unp);
1660 
1661 	unp2 = unp->unp_conn;
1662 	if ((unp->unp_socket->so_type == SOCK_STREAM ||
1663 	    (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1664 		so = unp2->unp_socket;
1665 		if (so != NULL)
1666 			socantrcvmore(so);
1667 	}
1668 }
1669 
1670 static void
1671 unp_drop(struct unpcb *unp, int errno)
1672 {
1673 	struct socket *so = unp->unp_socket;
1674 	struct unpcb *unp2;
1675 
1676 	UNP_LINK_WLOCK_ASSERT();
1677 	UNP_PCB_LOCK_ASSERT(unp);
1678 
1679 	so->so_error = errno;
1680 	unp2 = unp->unp_conn;
1681 	if (unp2 == NULL)
1682 		return;
1683 	UNP_PCB_LOCK(unp2);
1684 	unp_disconnect(unp, unp2);
1685 	UNP_PCB_UNLOCK(unp2);
1686 }
1687 
1688 static void
1689 unp_freerights(struct filedescent **fdep, int fdcount)
1690 {
1691 	struct file *fp;
1692 	int i;
1693 
1694 	KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1695 
1696 	for (i = 0; i < fdcount; i++) {
1697 		fp = fdep[i]->fde_file;
1698 		filecaps_free(&fdep[i]->fde_caps);
1699 		unp_discard(fp);
1700 	}
1701 	free(fdep[0], M_FILECAPS);
1702 }
1703 
1704 static int
1705 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1706 {
1707 	struct thread *td = curthread;		/* XXX */
1708 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1709 	int i;
1710 	int *fdp;
1711 	struct filedesc *fdesc = td->td_proc->p_fd;
1712 	struct filedescent *fde, **fdep;
1713 	void *data;
1714 	socklen_t clen = control->m_len, datalen;
1715 	int error, newfds;
1716 	u_int newlen;
1717 
1718 	UNP_LINK_UNLOCK_ASSERT();
1719 
1720 	error = 0;
1721 	if (controlp != NULL) /* controlp == NULL => free control messages */
1722 		*controlp = NULL;
1723 	while (cm != NULL) {
1724 		if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1725 			error = EINVAL;
1726 			break;
1727 		}
1728 		data = CMSG_DATA(cm);
1729 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1730 		if (cm->cmsg_level == SOL_SOCKET
1731 		    && cm->cmsg_type == SCM_RIGHTS) {
1732 			newfds = datalen / sizeof(*fdep);
1733 			if (newfds == 0)
1734 				goto next;
1735 			fdep = data;
1736 
1737 			/* If we're not outputting the descriptors free them. */
1738 			if (error || controlp == NULL) {
1739 				unp_freerights(fdep, newfds);
1740 				goto next;
1741 			}
1742 			FILEDESC_XLOCK(fdesc);
1743 
1744 			/*
1745 			 * Now change each pointer to an fd in the global
1746 			 * table to an integer that is the index to the local
1747 			 * fd table entry that we set up to point to the
1748 			 * global one we are transferring.
1749 			 */
1750 			newlen = newfds * sizeof(int);
1751 			*controlp = sbcreatecontrol(NULL, newlen,
1752 			    SCM_RIGHTS, SOL_SOCKET);
1753 			if (*controlp == NULL) {
1754 				FILEDESC_XUNLOCK(fdesc);
1755 				error = E2BIG;
1756 				unp_freerights(fdep, newfds);
1757 				goto next;
1758 			}
1759 
1760 			fdp = (int *)
1761 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1762 			if (fdallocn(td, 0, fdp, newfds) != 0) {
1763 				FILEDESC_XUNLOCK(td->td_proc->p_fd);
1764 				error = EMSGSIZE;
1765 				unp_freerights(fdep, newfds);
1766 				m_freem(*controlp);
1767 				*controlp = NULL;
1768 				goto next;
1769 			}
1770 			for (i = 0; i < newfds; i++, fdp++) {
1771 				fde = &fdesc->fd_ofiles[*fdp];
1772 				fde->fde_file = fdep[i]->fde_file;
1773 				filecaps_move(&fdep[i]->fde_caps,
1774 				    &fde->fde_caps);
1775 				if ((flags & MSG_CMSG_CLOEXEC) != 0)
1776 					fde->fde_flags |= UF_EXCLOSE;
1777 				unp_externalize_fp(fde->fde_file);
1778 			}
1779 			FILEDESC_XUNLOCK(fdesc);
1780 			free(fdep[0], M_FILECAPS);
1781 		} else {
1782 			/* We can just copy anything else across. */
1783 			if (error || controlp == NULL)
1784 				goto next;
1785 			*controlp = sbcreatecontrol(NULL, datalen,
1786 			    cm->cmsg_type, cm->cmsg_level);
1787 			if (*controlp == NULL) {
1788 				error = ENOBUFS;
1789 				goto next;
1790 			}
1791 			bcopy(data,
1792 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1793 			    datalen);
1794 		}
1795 		controlp = &(*controlp)->m_next;
1796 
1797 next:
1798 		if (CMSG_SPACE(datalen) < clen) {
1799 			clen -= CMSG_SPACE(datalen);
1800 			cm = (struct cmsghdr *)
1801 			    ((caddr_t)cm + CMSG_SPACE(datalen));
1802 		} else {
1803 			clen = 0;
1804 			cm = NULL;
1805 		}
1806 	}
1807 
1808 	m_freem(control);
1809 	return (error);
1810 }
1811 
1812 static void
1813 unp_zone_change(void *tag)
1814 {
1815 
1816 	uma_zone_set_max(unp_zone, maxsockets);
1817 }
1818 
1819 static void
1820 unp_init(void)
1821 {
1822 
1823 #ifdef VIMAGE
1824 	if (!IS_DEFAULT_VNET(curvnet))
1825 		return;
1826 #endif
1827 	unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1828 	    NULL, NULL, UMA_ALIGN_PTR, 0);
1829 	if (unp_zone == NULL)
1830 		panic("unp_init");
1831 	uma_zone_set_max(unp_zone, maxsockets);
1832 	uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1833 	EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1834 	    NULL, EVENTHANDLER_PRI_ANY);
1835 	LIST_INIT(&unp_dhead);
1836 	LIST_INIT(&unp_shead);
1837 	LIST_INIT(&unp_sphead);
1838 	SLIST_INIT(&unp_defers);
1839 	TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1840 	TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1841 	UNP_LINK_LOCK_INIT();
1842 	UNP_LIST_LOCK_INIT();
1843 	UNP_DEFERRED_LOCK_INIT();
1844 }
1845 
1846 static int
1847 unp_internalize(struct mbuf **controlp, struct thread *td)
1848 {
1849 	struct mbuf *control = *controlp;
1850 	struct proc *p = td->td_proc;
1851 	struct filedesc *fdesc = p->p_fd;
1852 	struct bintime *bt;
1853 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1854 	struct cmsgcred *cmcred;
1855 	struct filedescent *fde, **fdep, *fdev;
1856 	struct file *fp;
1857 	struct timeval *tv;
1858 	int i, *fdp;
1859 	void *data;
1860 	socklen_t clen = control->m_len, datalen;
1861 	int error, oldfds;
1862 	u_int newlen;
1863 
1864 	UNP_LINK_UNLOCK_ASSERT();
1865 
1866 	error = 0;
1867 	*controlp = NULL;
1868 	while (cm != NULL) {
1869 		if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1870 		    || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
1871 			error = EINVAL;
1872 			goto out;
1873 		}
1874 		data = CMSG_DATA(cm);
1875 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1876 
1877 		switch (cm->cmsg_type) {
1878 		/*
1879 		 * Fill in credential information.
1880 		 */
1881 		case SCM_CREDS:
1882 			*controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1883 			    SCM_CREDS, SOL_SOCKET);
1884 			if (*controlp == NULL) {
1885 				error = ENOBUFS;
1886 				goto out;
1887 			}
1888 			cmcred = (struct cmsgcred *)
1889 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1890 			cmcred->cmcred_pid = p->p_pid;
1891 			cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1892 			cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1893 			cmcred->cmcred_euid = td->td_ucred->cr_uid;
1894 			cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1895 			    CMGROUP_MAX);
1896 			for (i = 0; i < cmcred->cmcred_ngroups; i++)
1897 				cmcred->cmcred_groups[i] =
1898 				    td->td_ucred->cr_groups[i];
1899 			break;
1900 
1901 		case SCM_RIGHTS:
1902 			oldfds = datalen / sizeof (int);
1903 			if (oldfds == 0)
1904 				break;
1905 			/*
1906 			 * Check that all the FDs passed in refer to legal
1907 			 * files.  If not, reject the entire operation.
1908 			 */
1909 			fdp = data;
1910 			FILEDESC_SLOCK(fdesc);
1911 			for (i = 0; i < oldfds; i++, fdp++) {
1912 				fp = fget_locked(fdesc, *fdp);
1913 				if (fp == NULL) {
1914 					FILEDESC_SUNLOCK(fdesc);
1915 					error = EBADF;
1916 					goto out;
1917 				}
1918 				if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1919 					FILEDESC_SUNLOCK(fdesc);
1920 					error = EOPNOTSUPP;
1921 					goto out;
1922 				}
1923 
1924 			}
1925 
1926 			/*
1927 			 * Now replace the integer FDs with pointers to the
1928 			 * file structure and capability rights.
1929 			 */
1930 			newlen = oldfds * sizeof(fdep[0]);
1931 			*controlp = sbcreatecontrol(NULL, newlen,
1932 			    SCM_RIGHTS, SOL_SOCKET);
1933 			if (*controlp == NULL) {
1934 				FILEDESC_SUNLOCK(fdesc);
1935 				error = E2BIG;
1936 				goto out;
1937 			}
1938 			fdp = data;
1939 			fdep = (struct filedescent **)
1940 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1941 			fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
1942 			    M_WAITOK);
1943 			for (i = 0; i < oldfds; i++, fdev++, fdp++) {
1944 				fde = &fdesc->fd_ofiles[*fdp];
1945 				fdep[i] = fdev;
1946 				fdep[i]->fde_file = fde->fde_file;
1947 				filecaps_copy(&fde->fde_caps,
1948 				    &fdep[i]->fde_caps);
1949 				unp_internalize_fp(fdep[i]->fde_file);
1950 			}
1951 			FILEDESC_SUNLOCK(fdesc);
1952 			break;
1953 
1954 		case SCM_TIMESTAMP:
1955 			*controlp = sbcreatecontrol(NULL, sizeof(*tv),
1956 			    SCM_TIMESTAMP, SOL_SOCKET);
1957 			if (*controlp == NULL) {
1958 				error = ENOBUFS;
1959 				goto out;
1960 			}
1961 			tv = (struct timeval *)
1962 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1963 			microtime(tv);
1964 			break;
1965 
1966 		case SCM_BINTIME:
1967 			*controlp = sbcreatecontrol(NULL, sizeof(*bt),
1968 			    SCM_BINTIME, SOL_SOCKET);
1969 			if (*controlp == NULL) {
1970 				error = ENOBUFS;
1971 				goto out;
1972 			}
1973 			bt = (struct bintime *)
1974 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1975 			bintime(bt);
1976 			break;
1977 
1978 		default:
1979 			error = EINVAL;
1980 			goto out;
1981 		}
1982 
1983 		controlp = &(*controlp)->m_next;
1984 		if (CMSG_SPACE(datalen) < clen) {
1985 			clen -= CMSG_SPACE(datalen);
1986 			cm = (struct cmsghdr *)
1987 			    ((caddr_t)cm + CMSG_SPACE(datalen));
1988 		} else {
1989 			clen = 0;
1990 			cm = NULL;
1991 		}
1992 	}
1993 
1994 out:
1995 	m_freem(control);
1996 	return (error);
1997 }
1998 
1999 static struct mbuf *
2000 unp_addsockcred(struct thread *td, struct mbuf *control)
2001 {
2002 	struct mbuf *m, *n, *n_prev;
2003 	struct sockcred *sc;
2004 	const struct cmsghdr *cm;
2005 	int ngroups;
2006 	int i;
2007 
2008 	ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2009 	m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2010 	if (m == NULL)
2011 		return (control);
2012 
2013 	sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2014 	sc->sc_uid = td->td_ucred->cr_ruid;
2015 	sc->sc_euid = td->td_ucred->cr_uid;
2016 	sc->sc_gid = td->td_ucred->cr_rgid;
2017 	sc->sc_egid = td->td_ucred->cr_gid;
2018 	sc->sc_ngroups = ngroups;
2019 	for (i = 0; i < sc->sc_ngroups; i++)
2020 		sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2021 
2022 	/*
2023 	 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2024 	 * created SCM_CREDS control message (struct sockcred) has another
2025 	 * format.
2026 	 */
2027 	if (control != NULL)
2028 		for (n = control, n_prev = NULL; n != NULL;) {
2029 			cm = mtod(n, struct cmsghdr *);
2030     			if (cm->cmsg_level == SOL_SOCKET &&
2031 			    cm->cmsg_type == SCM_CREDS) {
2032     				if (n_prev == NULL)
2033 					control = n->m_next;
2034 				else
2035 					n_prev->m_next = n->m_next;
2036 				n = m_free(n);
2037 			} else {
2038 				n_prev = n;
2039 				n = n->m_next;
2040 			}
2041 		}
2042 
2043 	/* Prepend it to the head. */
2044 	m->m_next = control;
2045 	return (m);
2046 }
2047 
2048 static struct unpcb *
2049 fptounp(struct file *fp)
2050 {
2051 	struct socket *so;
2052 
2053 	if (fp->f_type != DTYPE_SOCKET)
2054 		return (NULL);
2055 	if ((so = fp->f_data) == NULL)
2056 		return (NULL);
2057 	if (so->so_proto->pr_domain != &localdomain)
2058 		return (NULL);
2059 	return sotounpcb(so);
2060 }
2061 
2062 static void
2063 unp_discard(struct file *fp)
2064 {
2065 	struct unp_defer *dr;
2066 
2067 	if (unp_externalize_fp(fp)) {
2068 		dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2069 		dr->ud_fp = fp;
2070 		UNP_DEFERRED_LOCK();
2071 		SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2072 		UNP_DEFERRED_UNLOCK();
2073 		atomic_add_int(&unp_defers_count, 1);
2074 		taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2075 	} else
2076 		(void) closef(fp, (struct thread *)NULL);
2077 }
2078 
2079 static void
2080 unp_process_defers(void *arg __unused, int pending)
2081 {
2082 	struct unp_defer *dr;
2083 	SLIST_HEAD(, unp_defer) drl;
2084 	int count;
2085 
2086 	SLIST_INIT(&drl);
2087 	for (;;) {
2088 		UNP_DEFERRED_LOCK();
2089 		if (SLIST_FIRST(&unp_defers) == NULL) {
2090 			UNP_DEFERRED_UNLOCK();
2091 			break;
2092 		}
2093 		SLIST_SWAP(&unp_defers, &drl, unp_defer);
2094 		UNP_DEFERRED_UNLOCK();
2095 		count = 0;
2096 		while ((dr = SLIST_FIRST(&drl)) != NULL) {
2097 			SLIST_REMOVE_HEAD(&drl, ud_link);
2098 			closef(dr->ud_fp, NULL);
2099 			free(dr, M_TEMP);
2100 			count++;
2101 		}
2102 		atomic_add_int(&unp_defers_count, -count);
2103 	}
2104 }
2105 
2106 static void
2107 unp_internalize_fp(struct file *fp)
2108 {
2109 	struct unpcb *unp;
2110 
2111 	UNP_LINK_WLOCK();
2112 	if ((unp = fptounp(fp)) != NULL) {
2113 		unp->unp_file = fp;
2114 		unp->unp_msgcount++;
2115 	}
2116 	fhold(fp);
2117 	unp_rights++;
2118 	UNP_LINK_WUNLOCK();
2119 }
2120 
2121 static int
2122 unp_externalize_fp(struct file *fp)
2123 {
2124 	struct unpcb *unp;
2125 	int ret;
2126 
2127 	UNP_LINK_WLOCK();
2128 	if ((unp = fptounp(fp)) != NULL) {
2129 		unp->unp_msgcount--;
2130 		ret = 1;
2131 	} else
2132 		ret = 0;
2133 	unp_rights--;
2134 	UNP_LINK_WUNLOCK();
2135 	return (ret);
2136 }
2137 
2138 /*
2139  * unp_defer indicates whether additional work has been defered for a future
2140  * pass through unp_gc().  It is thread local and does not require explicit
2141  * synchronization.
2142  */
2143 static int	unp_marked;
2144 static int	unp_unreachable;
2145 
2146 static void
2147 unp_accessable(struct filedescent **fdep, int fdcount)
2148 {
2149 	struct unpcb *unp;
2150 	struct file *fp;
2151 	int i;
2152 
2153 	for (i = 0; i < fdcount; i++) {
2154 		fp = fdep[i]->fde_file;
2155 		if ((unp = fptounp(fp)) == NULL)
2156 			continue;
2157 		if (unp->unp_gcflag & UNPGC_REF)
2158 			continue;
2159 		unp->unp_gcflag &= ~UNPGC_DEAD;
2160 		unp->unp_gcflag |= UNPGC_REF;
2161 		unp_marked++;
2162 	}
2163 }
2164 
2165 static void
2166 unp_gc_process(struct unpcb *unp)
2167 {
2168 	struct socket *soa;
2169 	struct socket *so;
2170 	struct file *fp;
2171 
2172 	/* Already processed. */
2173 	if (unp->unp_gcflag & UNPGC_SCANNED)
2174 		return;
2175 	fp = unp->unp_file;
2176 
2177 	/*
2178 	 * Check for a socket potentially in a cycle.  It must be in a
2179 	 * queue as indicated by msgcount, and this must equal the file
2180 	 * reference count.  Note that when msgcount is 0 the file is NULL.
2181 	 */
2182 	if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2183 	    unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2184 		unp->unp_gcflag |= UNPGC_DEAD;
2185 		unp_unreachable++;
2186 		return;
2187 	}
2188 
2189 	/*
2190 	 * Mark all sockets we reference with RIGHTS.
2191 	 */
2192 	so = unp->unp_socket;
2193 	SOCKBUF_LOCK(&so->so_rcv);
2194 	unp_scan(so->so_rcv.sb_mb, unp_accessable);
2195 	SOCKBUF_UNLOCK(&so->so_rcv);
2196 
2197 	/*
2198 	 * Mark all sockets in our accept queue.
2199 	 */
2200 	ACCEPT_LOCK();
2201 	TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2202 		SOCKBUF_LOCK(&soa->so_rcv);
2203 		unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2204 		SOCKBUF_UNLOCK(&soa->so_rcv);
2205 	}
2206 	ACCEPT_UNLOCK();
2207 	unp->unp_gcflag |= UNPGC_SCANNED;
2208 }
2209 
2210 static int unp_recycled;
2211 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2212     "Number of unreachable sockets claimed by the garbage collector.");
2213 
2214 static int unp_taskcount;
2215 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2216     "Number of times the garbage collector has run.");
2217 
2218 static void
2219 unp_gc(__unused void *arg, int pending)
2220 {
2221 	struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2222 				    NULL };
2223 	struct unp_head **head;
2224 	struct file *f, **unref;
2225 	struct unpcb *unp;
2226 	int i, total;
2227 
2228 	unp_taskcount++;
2229 	UNP_LIST_LOCK();
2230 	/*
2231 	 * First clear all gc flags from previous runs.
2232 	 */
2233 	for (head = heads; *head != NULL; head++)
2234 		LIST_FOREACH(unp, *head, unp_link)
2235 			unp->unp_gcflag = 0;
2236 
2237 	/*
2238 	 * Scan marking all reachable sockets with UNPGC_REF.  Once a socket
2239 	 * is reachable all of the sockets it references are reachable.
2240 	 * Stop the scan once we do a complete loop without discovering
2241 	 * a new reachable socket.
2242 	 */
2243 	do {
2244 		unp_unreachable = 0;
2245 		unp_marked = 0;
2246 		for (head = heads; *head != NULL; head++)
2247 			LIST_FOREACH(unp, *head, unp_link)
2248 				unp_gc_process(unp);
2249 	} while (unp_marked);
2250 	UNP_LIST_UNLOCK();
2251 	if (unp_unreachable == 0)
2252 		return;
2253 
2254 	/*
2255 	 * Allocate space for a local list of dead unpcbs.
2256 	 */
2257 	unref = malloc(unp_unreachable * sizeof(struct file *),
2258 	    M_TEMP, M_WAITOK);
2259 
2260 	/*
2261 	 * Iterate looking for sockets which have been specifically marked
2262 	 * as as unreachable and store them locally.
2263 	 */
2264 	UNP_LINK_RLOCK();
2265 	UNP_LIST_LOCK();
2266 	for (total = 0, head = heads; *head != NULL; head++)
2267 		LIST_FOREACH(unp, *head, unp_link)
2268 			if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2269 				f = unp->unp_file;
2270 				if (unp->unp_msgcount == 0 || f == NULL ||
2271 				    f->f_count != unp->unp_msgcount)
2272 					continue;
2273 				unref[total++] = f;
2274 				fhold(f);
2275 				KASSERT(total <= unp_unreachable,
2276 				    ("unp_gc: incorrect unreachable count."));
2277 			}
2278 	UNP_LIST_UNLOCK();
2279 	UNP_LINK_RUNLOCK();
2280 
2281 	/*
2282 	 * Now flush all sockets, free'ing rights.  This will free the
2283 	 * struct files associated with these sockets but leave each socket
2284 	 * with one remaining ref.
2285 	 */
2286 	for (i = 0; i < total; i++) {
2287 		struct socket *so;
2288 
2289 		so = unref[i]->f_data;
2290 		CURVNET_SET(so->so_vnet);
2291 		sorflush(so);
2292 		CURVNET_RESTORE();
2293 	}
2294 
2295 	/*
2296 	 * And finally release the sockets so they can be reclaimed.
2297 	 */
2298 	for (i = 0; i < total; i++)
2299 		fdrop(unref[i], NULL);
2300 	unp_recycled += total;
2301 	free(unref, M_TEMP);
2302 }
2303 
2304 static void
2305 unp_dispose(struct mbuf *m)
2306 {
2307 
2308 	if (m)
2309 		unp_scan(m, unp_freerights);
2310 }
2311 
2312 static void
2313 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2314 {
2315 	struct mbuf *m;
2316 	struct cmsghdr *cm;
2317 	void *data;
2318 	socklen_t clen, datalen;
2319 
2320 	while (m0 != NULL) {
2321 		for (m = m0; m; m = m->m_next) {
2322 			if (m->m_type != MT_CONTROL)
2323 				continue;
2324 
2325 			cm = mtod(m, struct cmsghdr *);
2326 			clen = m->m_len;
2327 
2328 			while (cm != NULL) {
2329 				if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2330 					break;
2331 
2332 				data = CMSG_DATA(cm);
2333 				datalen = (caddr_t)cm + cm->cmsg_len
2334 				    - (caddr_t)data;
2335 
2336 				if (cm->cmsg_level == SOL_SOCKET &&
2337 				    cm->cmsg_type == SCM_RIGHTS) {
2338 					(*op)(data, datalen /
2339 					    sizeof(struct filedescent *));
2340 				}
2341 
2342 				if (CMSG_SPACE(datalen) < clen) {
2343 					clen -= CMSG_SPACE(datalen);
2344 					cm = (struct cmsghdr *)
2345 					    ((caddr_t)cm + CMSG_SPACE(datalen));
2346 				} else {
2347 					clen = 0;
2348 					cm = NULL;
2349 				}
2350 			}
2351 		}
2352 		m0 = m0->m_nextpkt;
2353 	}
2354 }
2355 
2356 /*
2357  * A helper function called by VFS before socket-type vnode reclamation.
2358  * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2359  * use count.
2360  */
2361 void
2362 vfs_unp_reclaim(struct vnode *vp)
2363 {
2364 	struct socket *so;
2365 	struct unpcb *unp;
2366 	int active;
2367 
2368 	ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2369 	KASSERT(vp->v_type == VSOCK,
2370 	    ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2371 
2372 	active = 0;
2373 	UNP_LINK_WLOCK();
2374 	VOP_UNP_CONNECT(vp, &so);
2375 	if (so == NULL)
2376 		goto done;
2377 	unp = sotounpcb(so);
2378 	if (unp == NULL)
2379 		goto done;
2380 	UNP_PCB_LOCK(unp);
2381 	if (unp->unp_vnode == vp) {
2382 		VOP_UNP_DETACH(vp);
2383 		unp->unp_vnode = NULL;
2384 		active = 1;
2385 	}
2386 	UNP_PCB_UNLOCK(unp);
2387 done:
2388 	UNP_LINK_WUNLOCK();
2389 	if (active)
2390 		vunref(vp);
2391 }
2392 
2393 #ifdef DDB
2394 static void
2395 db_print_indent(int indent)
2396 {
2397 	int i;
2398 
2399 	for (i = 0; i < indent; i++)
2400 		db_printf(" ");
2401 }
2402 
2403 static void
2404 db_print_unpflags(int unp_flags)
2405 {
2406 	int comma;
2407 
2408 	comma = 0;
2409 	if (unp_flags & UNP_HAVEPC) {
2410 		db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2411 		comma = 1;
2412 	}
2413 	if (unp_flags & UNP_HAVEPCCACHED) {
2414 		db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2415 		comma = 1;
2416 	}
2417 	if (unp_flags & UNP_WANTCRED) {
2418 		db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2419 		comma = 1;
2420 	}
2421 	if (unp_flags & UNP_CONNWAIT) {
2422 		db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2423 		comma = 1;
2424 	}
2425 	if (unp_flags & UNP_CONNECTING) {
2426 		db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2427 		comma = 1;
2428 	}
2429 	if (unp_flags & UNP_BINDING) {
2430 		db_printf("%sUNP_BINDING", comma ? ", " : "");
2431 		comma = 1;
2432 	}
2433 }
2434 
2435 static void
2436 db_print_xucred(int indent, struct xucred *xu)
2437 {
2438 	int comma, i;
2439 
2440 	db_print_indent(indent);
2441 	db_printf("cr_version: %u   cr_uid: %u   cr_ngroups: %d\n",
2442 	    xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2443 	db_print_indent(indent);
2444 	db_printf("cr_groups: ");
2445 	comma = 0;
2446 	for (i = 0; i < xu->cr_ngroups; i++) {
2447 		db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2448 		comma = 1;
2449 	}
2450 	db_printf("\n");
2451 }
2452 
2453 static void
2454 db_print_unprefs(int indent, struct unp_head *uh)
2455 {
2456 	struct unpcb *unp;
2457 	int counter;
2458 
2459 	counter = 0;
2460 	LIST_FOREACH(unp, uh, unp_reflink) {
2461 		if (counter % 4 == 0)
2462 			db_print_indent(indent);
2463 		db_printf("%p  ", unp);
2464 		if (counter % 4 == 3)
2465 			db_printf("\n");
2466 		counter++;
2467 	}
2468 	if (counter != 0 && counter % 4 != 0)
2469 		db_printf("\n");
2470 }
2471 
2472 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2473 {
2474 	struct unpcb *unp;
2475 
2476         if (!have_addr) {
2477                 db_printf("usage: show unpcb <addr>\n");
2478                 return;
2479         }
2480         unp = (struct unpcb *)addr;
2481 
2482 	db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2483 	    unp->unp_vnode);
2484 
2485 	db_printf("unp_ino: %ju   unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2486 	    unp->unp_conn);
2487 
2488 	db_printf("unp_refs:\n");
2489 	db_print_unprefs(2, &unp->unp_refs);
2490 
2491 	/* XXXRW: Would be nice to print the full address, if any. */
2492 	db_printf("unp_addr: %p\n", unp->unp_addr);
2493 
2494 	db_printf("unp_gencnt: %llu\n",
2495 	    (unsigned long long)unp->unp_gencnt);
2496 
2497 	db_printf("unp_flags: %x (", unp->unp_flags);
2498 	db_print_unpflags(unp->unp_flags);
2499 	db_printf(")\n");
2500 
2501 	db_printf("unp_peercred:\n");
2502 	db_print_xucred(2, &unp->unp_peercred);
2503 
2504 	db_printf("unp_refcount: %u\n", unp->unp_refcount);
2505 }
2506 #endif
2507