xref: /freebsd/sys/kern/uipc_usrreq.c (revision 1c05a6ea6b849ff95e539c31adea887c644a6a01)
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 	/*
1060 	 * In case of PRUS_NOTREADY, uipc_ready() is responsible
1061 	 * for freeing memory.
1062 	 */
1063 	if (m != NULL && (flags & PRUS_NOTREADY) == 0)
1064 		m_freem(m);
1065 	return (error);
1066 }
1067 
1068 static int
1069 uipc_ready(struct socket *so, struct mbuf *m, int count)
1070 {
1071 	struct unpcb *unp, *unp2;
1072 	struct socket *so2;
1073 	int error;
1074 
1075 	unp = sotounpcb(so);
1076 
1077 	UNP_LINK_RLOCK();
1078 	if ((unp2 = unp->unp_conn) == NULL) {
1079 		UNP_LINK_RUNLOCK();
1080 		for (int i = 0; i < count; i++)
1081 			m = m_free(m);
1082 		return (ECONNRESET);
1083 	}
1084 	UNP_PCB_LOCK(unp2);
1085 	so2 = unp2->unp_socket;
1086 
1087 	SOCKBUF_LOCK(&so2->so_rcv);
1088 	if ((error = sbready(&so2->so_rcv, m, count)) == 0)
1089 		sorwakeup_locked(so2);
1090 	else
1091 		SOCKBUF_UNLOCK(&so2->so_rcv);
1092 
1093 	UNP_PCB_UNLOCK(unp2);
1094 	UNP_LINK_RUNLOCK();
1095 
1096 	return (error);
1097 }
1098 
1099 static int
1100 uipc_sense(struct socket *so, struct stat *sb)
1101 {
1102 	struct unpcb *unp;
1103 
1104 	unp = sotounpcb(so);
1105 	KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1106 
1107 	sb->st_blksize = so->so_snd.sb_hiwat;
1108 	UNP_PCB_LOCK(unp);
1109 	sb->st_dev = NODEV;
1110 	if (unp->unp_ino == 0)
1111 		unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1112 	sb->st_ino = unp->unp_ino;
1113 	UNP_PCB_UNLOCK(unp);
1114 	return (0);
1115 }
1116 
1117 static int
1118 uipc_shutdown(struct socket *so)
1119 {
1120 	struct unpcb *unp;
1121 
1122 	unp = sotounpcb(so);
1123 	KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1124 
1125 	UNP_LINK_WLOCK();
1126 	UNP_PCB_LOCK(unp);
1127 	socantsendmore(so);
1128 	unp_shutdown(unp);
1129 	UNP_PCB_UNLOCK(unp);
1130 	UNP_LINK_WUNLOCK();
1131 	return (0);
1132 }
1133 
1134 static int
1135 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1136 {
1137 	struct unpcb *unp;
1138 	const struct sockaddr *sa;
1139 
1140 	unp = sotounpcb(so);
1141 	KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1142 
1143 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1144 	UNP_PCB_LOCK(unp);
1145 	if (unp->unp_addr != NULL)
1146 		sa = (struct sockaddr *) unp->unp_addr;
1147 	else
1148 		sa = &sun_noname;
1149 	bcopy(sa, *nam, sa->sa_len);
1150 	UNP_PCB_UNLOCK(unp);
1151 	return (0);
1152 }
1153 
1154 static struct pr_usrreqs uipc_usrreqs_dgram = {
1155 	.pru_abort = 		uipc_abort,
1156 	.pru_accept =		uipc_accept,
1157 	.pru_attach =		uipc_attach,
1158 	.pru_bind =		uipc_bind,
1159 	.pru_bindat =		uipc_bindat,
1160 	.pru_connect =		uipc_connect,
1161 	.pru_connectat =	uipc_connectat,
1162 	.pru_connect2 =		uipc_connect2,
1163 	.pru_detach =		uipc_detach,
1164 	.pru_disconnect =	uipc_disconnect,
1165 	.pru_listen =		uipc_listen,
1166 	.pru_peeraddr =		uipc_peeraddr,
1167 	.pru_rcvd =		uipc_rcvd,
1168 	.pru_send =		uipc_send,
1169 	.pru_sense =		uipc_sense,
1170 	.pru_shutdown =		uipc_shutdown,
1171 	.pru_sockaddr =		uipc_sockaddr,
1172 	.pru_soreceive =	soreceive_dgram,
1173 	.pru_close =		uipc_close,
1174 };
1175 
1176 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1177 	.pru_abort =		uipc_abort,
1178 	.pru_accept =		uipc_accept,
1179 	.pru_attach =		uipc_attach,
1180 	.pru_bind =		uipc_bind,
1181 	.pru_bindat =		uipc_bindat,
1182 	.pru_connect =		uipc_connect,
1183 	.pru_connectat =	uipc_connectat,
1184 	.pru_connect2 =		uipc_connect2,
1185 	.pru_detach =		uipc_detach,
1186 	.pru_disconnect =	uipc_disconnect,
1187 	.pru_listen =		uipc_listen,
1188 	.pru_peeraddr =		uipc_peeraddr,
1189 	.pru_rcvd =		uipc_rcvd,
1190 	.pru_send =		uipc_send,
1191 	.pru_sense =		uipc_sense,
1192 	.pru_shutdown =		uipc_shutdown,
1193 	.pru_sockaddr =		uipc_sockaddr,
1194 	.pru_soreceive =	soreceive_generic,	/* XXX: or...? */
1195 	.pru_close =		uipc_close,
1196 };
1197 
1198 static struct pr_usrreqs uipc_usrreqs_stream = {
1199 	.pru_abort = 		uipc_abort,
1200 	.pru_accept =		uipc_accept,
1201 	.pru_attach =		uipc_attach,
1202 	.pru_bind =		uipc_bind,
1203 	.pru_bindat =		uipc_bindat,
1204 	.pru_connect =		uipc_connect,
1205 	.pru_connectat =	uipc_connectat,
1206 	.pru_connect2 =		uipc_connect2,
1207 	.pru_detach =		uipc_detach,
1208 	.pru_disconnect =	uipc_disconnect,
1209 	.pru_listen =		uipc_listen,
1210 	.pru_peeraddr =		uipc_peeraddr,
1211 	.pru_rcvd =		uipc_rcvd,
1212 	.pru_send =		uipc_send,
1213 	.pru_ready =		uipc_ready,
1214 	.pru_sense =		uipc_sense,
1215 	.pru_shutdown =		uipc_shutdown,
1216 	.pru_sockaddr =		uipc_sockaddr,
1217 	.pru_soreceive =	soreceive_generic,
1218 	.pru_close =		uipc_close,
1219 };
1220 
1221 static int
1222 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1223 {
1224 	struct unpcb *unp;
1225 	struct xucred xu;
1226 	int error, optval;
1227 
1228 	if (sopt->sopt_level != 0)
1229 		return (EINVAL);
1230 
1231 	unp = sotounpcb(so);
1232 	KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1233 	error = 0;
1234 	switch (sopt->sopt_dir) {
1235 	case SOPT_GET:
1236 		switch (sopt->sopt_name) {
1237 		case LOCAL_PEERCRED:
1238 			UNP_PCB_LOCK(unp);
1239 			if (unp->unp_flags & UNP_HAVEPC)
1240 				xu = unp->unp_peercred;
1241 			else {
1242 				if (so->so_type == SOCK_STREAM)
1243 					error = ENOTCONN;
1244 				else
1245 					error = EINVAL;
1246 			}
1247 			UNP_PCB_UNLOCK(unp);
1248 			if (error == 0)
1249 				error = sooptcopyout(sopt, &xu, sizeof(xu));
1250 			break;
1251 
1252 		case LOCAL_CREDS:
1253 			/* Unlocked read. */
1254 			optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1255 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1256 			break;
1257 
1258 		case LOCAL_CONNWAIT:
1259 			/* Unlocked read. */
1260 			optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1261 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1262 			break;
1263 
1264 		default:
1265 			error = EOPNOTSUPP;
1266 			break;
1267 		}
1268 		break;
1269 
1270 	case SOPT_SET:
1271 		switch (sopt->sopt_name) {
1272 		case LOCAL_CREDS:
1273 		case LOCAL_CONNWAIT:
1274 			error = sooptcopyin(sopt, &optval, sizeof(optval),
1275 					    sizeof(optval));
1276 			if (error)
1277 				break;
1278 
1279 #define	OPTSET(bit) do {						\
1280 	UNP_PCB_LOCK(unp);						\
1281 	if (optval)							\
1282 		unp->unp_flags |= bit;					\
1283 	else								\
1284 		unp->unp_flags &= ~bit;					\
1285 	UNP_PCB_UNLOCK(unp);						\
1286 } while (0)
1287 
1288 			switch (sopt->sopt_name) {
1289 			case LOCAL_CREDS:
1290 				OPTSET(UNP_WANTCRED);
1291 				break;
1292 
1293 			case LOCAL_CONNWAIT:
1294 				OPTSET(UNP_CONNWAIT);
1295 				break;
1296 
1297 			default:
1298 				break;
1299 			}
1300 			break;
1301 #undef	OPTSET
1302 		default:
1303 			error = ENOPROTOOPT;
1304 			break;
1305 		}
1306 		break;
1307 
1308 	default:
1309 		error = EOPNOTSUPP;
1310 		break;
1311 	}
1312 	return (error);
1313 }
1314 
1315 static int
1316 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1317 {
1318 
1319 	return (unp_connectat(AT_FDCWD, so, nam, td));
1320 }
1321 
1322 static int
1323 unp_connectat(int fd, struct socket *so, struct sockaddr *nam,
1324     struct thread *td)
1325 {
1326 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1327 	struct vnode *vp;
1328 	struct socket *so2;
1329 	struct unpcb *unp, *unp2, *unp3;
1330 	struct nameidata nd;
1331 	char buf[SOCK_MAXADDRLEN];
1332 	struct sockaddr *sa;
1333 	cap_rights_t rights;
1334 	int error, len;
1335 
1336 	if (nam->sa_family != AF_UNIX)
1337 		return (EAFNOSUPPORT);
1338 
1339 	UNP_LINK_WLOCK_ASSERT();
1340 
1341 	unp = sotounpcb(so);
1342 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1343 
1344 	if (nam->sa_len > sizeof(struct sockaddr_un))
1345 		return (EINVAL);
1346 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1347 	if (len <= 0)
1348 		return (EINVAL);
1349 	bcopy(soun->sun_path, buf, len);
1350 	buf[len] = 0;
1351 
1352 	UNP_PCB_LOCK(unp);
1353 	if (unp->unp_flags & UNP_CONNECTING) {
1354 		UNP_PCB_UNLOCK(unp);
1355 		return (EALREADY);
1356 	}
1357 	UNP_LINK_WUNLOCK();
1358 	unp->unp_flags |= UNP_CONNECTING;
1359 	UNP_PCB_UNLOCK(unp);
1360 
1361 	sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1362 	NDINIT_ATRIGHTS(&nd, LOOKUP, FOLLOW | LOCKSHARED | LOCKLEAF,
1363 	    UIO_SYSSPACE, buf, fd, cap_rights_init(&rights, CAP_CONNECTAT), td);
1364 	error = namei(&nd);
1365 	if (error)
1366 		vp = NULL;
1367 	else
1368 		vp = nd.ni_vp;
1369 	ASSERT_VOP_LOCKED(vp, "unp_connect");
1370 	NDFREE(&nd, NDF_ONLY_PNBUF);
1371 	if (error)
1372 		goto bad;
1373 
1374 	if (vp->v_type != VSOCK) {
1375 		error = ENOTSOCK;
1376 		goto bad;
1377 	}
1378 #ifdef MAC
1379 	error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1380 	if (error)
1381 		goto bad;
1382 #endif
1383 	error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1384 	if (error)
1385 		goto bad;
1386 
1387 	unp = sotounpcb(so);
1388 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1389 
1390 	/*
1391 	 * Lock linkage lock for two reasons: make sure v_socket is stable,
1392 	 * and to protect simultaneous locking of multiple pcbs.
1393 	 */
1394 	UNP_LINK_WLOCK();
1395 	VOP_UNP_CONNECT(vp, &unp2);
1396 	if (unp2 == NULL) {
1397 		error = ECONNREFUSED;
1398 		goto bad2;
1399 	}
1400 	so2 = unp2->unp_socket;
1401 	if (so->so_type != so2->so_type) {
1402 		error = EPROTOTYPE;
1403 		goto bad2;
1404 	}
1405 	UNP_PCB_LOCK(unp);
1406 	UNP_PCB_LOCK(unp2);
1407 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1408 		if (so2->so_options & SO_ACCEPTCONN) {
1409 			CURVNET_SET(so2->so_vnet);
1410 			so2 = sonewconn(so2, 0);
1411 			CURVNET_RESTORE();
1412 		} else
1413 			so2 = NULL;
1414 		if (so2 == NULL) {
1415 			error = ECONNREFUSED;
1416 			goto bad3;
1417 		}
1418 		unp3 = sotounpcb(so2);
1419 		UNP_PCB_LOCK(unp3);
1420 		if (unp2->unp_addr != NULL) {
1421 			bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1422 			unp3->unp_addr = (struct sockaddr_un *) sa;
1423 			sa = NULL;
1424 		}
1425 
1426 		/*
1427 		 * The connector's (client's) credentials are copied from its
1428 		 * process structure at the time of connect() (which is now).
1429 		 */
1430 		cru2x(td->td_ucred, &unp3->unp_peercred);
1431 		unp3->unp_flags |= UNP_HAVEPC;
1432 
1433 		/*
1434 		 * The receiver's (server's) credentials are copied from the
1435 		 * unp_peercred member of socket on which the former called
1436 		 * listen(); uipc_listen() cached that process's credentials
1437 		 * at that time so we can use them now.
1438 		 */
1439 		memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1440 		    sizeof(unp->unp_peercred));
1441 		unp->unp_flags |= UNP_HAVEPC;
1442 		if (unp2->unp_flags & UNP_WANTCRED)
1443 			unp3->unp_flags |= UNP_WANTCRED;
1444 		UNP_PCB_UNLOCK(unp2);
1445 		unp2 = unp3;
1446 #ifdef MAC
1447 		mac_socketpeer_set_from_socket(so, so2);
1448 		mac_socketpeer_set_from_socket(so2, so);
1449 #endif
1450 	}
1451 
1452 	KASSERT(unp2 != NULL && so2 != NULL && unp2->unp_socket == so2 &&
1453 	    sotounpcb(so2) == unp2,
1454 	    ("%s: unp2 %p so2 %p", __func__, unp2, so2));
1455 	error = unp_connect2(so, so2, PRU_CONNECT);
1456 bad3:
1457 	UNP_PCB_UNLOCK(unp2);
1458 	UNP_PCB_UNLOCK(unp);
1459 bad2:
1460 	UNP_LINK_WUNLOCK();
1461 bad:
1462 	if (vp != NULL)
1463 		vput(vp);
1464 	free(sa, M_SONAME);
1465 	UNP_LINK_WLOCK();
1466 	UNP_PCB_LOCK(unp);
1467 	unp->unp_flags &= ~UNP_CONNECTING;
1468 	UNP_PCB_UNLOCK(unp);
1469 	return (error);
1470 }
1471 
1472 static int
1473 unp_connect2(struct socket *so, struct socket *so2, int req)
1474 {
1475 	struct unpcb *unp;
1476 	struct unpcb *unp2;
1477 
1478 	unp = sotounpcb(so);
1479 	KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1480 	unp2 = sotounpcb(so2);
1481 	KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1482 
1483 	UNP_LINK_WLOCK_ASSERT();
1484 	UNP_PCB_LOCK_ASSERT(unp);
1485 	UNP_PCB_LOCK_ASSERT(unp2);
1486 
1487 	if (so2->so_type != so->so_type)
1488 		return (EPROTOTYPE);
1489 	unp2->unp_flags &= ~UNP_NASCENT;
1490 	unp->unp_conn = unp2;
1491 
1492 	switch (so->so_type) {
1493 	case SOCK_DGRAM:
1494 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1495 		soisconnected(so);
1496 		break;
1497 
1498 	case SOCK_STREAM:
1499 	case SOCK_SEQPACKET:
1500 		unp2->unp_conn = unp;
1501 		if (req == PRU_CONNECT &&
1502 		    ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1503 			soisconnecting(so);
1504 		else
1505 			soisconnected(so);
1506 		soisconnected(so2);
1507 		break;
1508 
1509 	default:
1510 		panic("unp_connect2");
1511 	}
1512 	return (0);
1513 }
1514 
1515 static void
1516 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1517 {
1518 	struct socket *so;
1519 
1520 	KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1521 
1522 	UNP_LINK_WLOCK_ASSERT();
1523 	UNP_PCB_LOCK_ASSERT(unp);
1524 	UNP_PCB_LOCK_ASSERT(unp2);
1525 
1526 	unp->unp_conn = NULL;
1527 	switch (unp->unp_socket->so_type) {
1528 	case SOCK_DGRAM:
1529 		LIST_REMOVE(unp, unp_reflink);
1530 		so = unp->unp_socket;
1531 		SOCK_LOCK(so);
1532 		so->so_state &= ~SS_ISCONNECTED;
1533 		SOCK_UNLOCK(so);
1534 		break;
1535 
1536 	case SOCK_STREAM:
1537 	case SOCK_SEQPACKET:
1538 		soisdisconnected(unp->unp_socket);
1539 		unp2->unp_conn = NULL;
1540 		soisdisconnected(unp2->unp_socket);
1541 		break;
1542 	}
1543 }
1544 
1545 /*
1546  * unp_pcblist() walks the global list of struct unpcb's to generate a
1547  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1548  * sequentially, validating the generation number on each to see if it has
1549  * been detached.  All of this is necessary because copyout() may sleep on
1550  * disk I/O.
1551  */
1552 static int
1553 unp_pcblist(SYSCTL_HANDLER_ARGS)
1554 {
1555 	int error, i, n;
1556 	int freeunp;
1557 	struct unpcb *unp, **unp_list;
1558 	unp_gen_t gencnt;
1559 	struct xunpgen *xug;
1560 	struct unp_head *head;
1561 	struct xunpcb *xu;
1562 
1563 	switch ((intptr_t)arg1) {
1564 	case SOCK_STREAM:
1565 		head = &unp_shead;
1566 		break;
1567 
1568 	case SOCK_DGRAM:
1569 		head = &unp_dhead;
1570 		break;
1571 
1572 	case SOCK_SEQPACKET:
1573 		head = &unp_sphead;
1574 		break;
1575 
1576 	default:
1577 		panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1578 	}
1579 
1580 	/*
1581 	 * The process of preparing the PCB list is too time-consuming and
1582 	 * resource-intensive to repeat twice on every request.
1583 	 */
1584 	if (req->oldptr == NULL) {
1585 		n = unp_count;
1586 		req->oldidx = 2 * (sizeof *xug)
1587 			+ (n + n/8) * sizeof(struct xunpcb);
1588 		return (0);
1589 	}
1590 
1591 	if (req->newptr != NULL)
1592 		return (EPERM);
1593 
1594 	/*
1595 	 * OK, now we're committed to doing something.
1596 	 */
1597 	xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1598 	UNP_LINK_RLOCK();
1599 	gencnt = unp_gencnt;
1600 	n = unp_count;
1601 	UNP_LINK_RUNLOCK();
1602 
1603 	xug->xug_len = sizeof *xug;
1604 	xug->xug_count = n;
1605 	xug->xug_gen = gencnt;
1606 	xug->xug_sogen = so_gencnt;
1607 	error = SYSCTL_OUT(req, xug, sizeof *xug);
1608 	if (error) {
1609 		free(xug, M_TEMP);
1610 		return (error);
1611 	}
1612 
1613 	unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1614 
1615 	UNP_LINK_RLOCK();
1616 	for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1617 	     unp = LIST_NEXT(unp, unp_link)) {
1618 		UNP_PCB_LOCK(unp);
1619 		if (unp->unp_gencnt <= gencnt) {
1620 			if (cr_cansee(req->td->td_ucred,
1621 			    unp->unp_socket->so_cred)) {
1622 				UNP_PCB_UNLOCK(unp);
1623 				continue;
1624 			}
1625 			unp_list[i++] = unp;
1626 			unp->unp_refcount++;
1627 		}
1628 		UNP_PCB_UNLOCK(unp);
1629 	}
1630 	UNP_LINK_RUNLOCK();
1631 	n = i;			/* In case we lost some during malloc. */
1632 
1633 	error = 0;
1634 	xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1635 	for (i = 0; i < n; i++) {
1636 		unp = unp_list[i];
1637 		UNP_PCB_LOCK(unp);
1638 		unp->unp_refcount--;
1639 	        if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1640 			xu->xu_len = sizeof *xu;
1641 			xu->xu_unpp = unp;
1642 			/*
1643 			 * XXX - need more locking here to protect against
1644 			 * connect/disconnect races for SMP.
1645 			 */
1646 			if (unp->unp_addr != NULL)
1647 				bcopy(unp->unp_addr, &xu->xu_addr,
1648 				      unp->unp_addr->sun_len);
1649 			else
1650 				bzero(&xu->xu_addr, sizeof(xu->xu_addr));
1651 			if (unp->unp_conn != NULL &&
1652 			    unp->unp_conn->unp_addr != NULL)
1653 				bcopy(unp->unp_conn->unp_addr,
1654 				      &xu->xu_caddr,
1655 				      unp->unp_conn->unp_addr->sun_len);
1656 			else
1657 				bzero(&xu->xu_caddr, sizeof(xu->xu_caddr));
1658 			xu->unp_vnode = unp->unp_vnode;
1659 			xu->unp_conn = unp->unp_conn;
1660 			xu->xu_firstref = LIST_FIRST(&unp->unp_refs);
1661 			xu->xu_nextref = LIST_NEXT(unp, unp_reflink);
1662 			xu->unp_gencnt = unp->unp_gencnt;
1663 			sotoxsocket(unp->unp_socket, &xu->xu_socket);
1664 			UNP_PCB_UNLOCK(unp);
1665 			error = SYSCTL_OUT(req, xu, sizeof *xu);
1666 		} else {
1667 			freeunp = (unp->unp_refcount == 0);
1668 			UNP_PCB_UNLOCK(unp);
1669 			if (freeunp) {
1670 				UNP_PCB_LOCK_DESTROY(unp);
1671 				uma_zfree(unp_zone, unp);
1672 			}
1673 		}
1674 	}
1675 	free(xu, M_TEMP);
1676 	if (!error) {
1677 		/*
1678 		 * Give the user an updated idea of our state.  If the
1679 		 * generation differs from what we told her before, she knows
1680 		 * that something happened while we were processing this
1681 		 * request, and it might be necessary to retry.
1682 		 */
1683 		xug->xug_gen = unp_gencnt;
1684 		xug->xug_sogen = so_gencnt;
1685 		xug->xug_count = unp_count;
1686 		error = SYSCTL_OUT(req, xug, sizeof *xug);
1687 	}
1688 	free(unp_list, M_TEMP);
1689 	free(xug, M_TEMP);
1690 	return (error);
1691 }
1692 
1693 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1694     (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1695     "List of active local datagram sockets");
1696 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1697     (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1698     "List of active local stream sockets");
1699 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1700     CTLTYPE_OPAQUE | CTLFLAG_RD,
1701     (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1702     "List of active local seqpacket sockets");
1703 
1704 static void
1705 unp_shutdown(struct unpcb *unp)
1706 {
1707 	struct unpcb *unp2;
1708 	struct socket *so;
1709 
1710 	UNP_LINK_WLOCK_ASSERT();
1711 	UNP_PCB_LOCK_ASSERT(unp);
1712 
1713 	unp2 = unp->unp_conn;
1714 	if ((unp->unp_socket->so_type == SOCK_STREAM ||
1715 	    (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1716 		so = unp2->unp_socket;
1717 		if (so != NULL)
1718 			socantrcvmore(so);
1719 	}
1720 }
1721 
1722 static void
1723 unp_drop(struct unpcb *unp)
1724 {
1725 	struct socket *so = unp->unp_socket;
1726 	struct unpcb *unp2;
1727 
1728 	UNP_LINK_WLOCK_ASSERT();
1729 	UNP_PCB_LOCK_ASSERT(unp);
1730 
1731 	/*
1732 	 * Regardless of whether the socket's peer dropped the connection
1733 	 * with this socket by aborting or disconnecting, POSIX requires
1734 	 * that ECONNRESET is returned.
1735 	 */
1736 	so->so_error = ECONNRESET;
1737 	unp2 = unp->unp_conn;
1738 	if (unp2 == NULL)
1739 		return;
1740 	UNP_PCB_LOCK(unp2);
1741 	unp_disconnect(unp, unp2);
1742 	UNP_PCB_UNLOCK(unp2);
1743 }
1744 
1745 static void
1746 unp_freerights(struct filedescent **fdep, int fdcount)
1747 {
1748 	struct file *fp;
1749 	int i;
1750 
1751 	KASSERT(fdcount > 0, ("%s: fdcount %d", __func__, fdcount));
1752 
1753 	for (i = 0; i < fdcount; i++) {
1754 		fp = fdep[i]->fde_file;
1755 		filecaps_free(&fdep[i]->fde_caps);
1756 		unp_discard(fp);
1757 	}
1758 	free(fdep[0], M_FILECAPS);
1759 }
1760 
1761 static int
1762 unp_externalize(struct mbuf *control, struct mbuf **controlp, int flags)
1763 {
1764 	struct thread *td = curthread;		/* XXX */
1765 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1766 	int i;
1767 	int *fdp;
1768 	struct filedesc *fdesc = td->td_proc->p_fd;
1769 	struct filedescent **fdep;
1770 	void *data;
1771 	socklen_t clen = control->m_len, datalen;
1772 	int error, newfds;
1773 	u_int newlen;
1774 
1775 	UNP_LINK_UNLOCK_ASSERT();
1776 
1777 	error = 0;
1778 	if (controlp != NULL) /* controlp == NULL => free control messages */
1779 		*controlp = NULL;
1780 	while (cm != NULL) {
1781 		if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1782 			error = EINVAL;
1783 			break;
1784 		}
1785 		data = CMSG_DATA(cm);
1786 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1787 		if (cm->cmsg_level == SOL_SOCKET
1788 		    && cm->cmsg_type == SCM_RIGHTS) {
1789 			newfds = datalen / sizeof(*fdep);
1790 			if (newfds == 0)
1791 				goto next;
1792 			fdep = data;
1793 
1794 			/* If we're not outputting the descriptors free them. */
1795 			if (error || controlp == NULL) {
1796 				unp_freerights(fdep, newfds);
1797 				goto next;
1798 			}
1799 			FILEDESC_XLOCK(fdesc);
1800 
1801 			/*
1802 			 * Now change each pointer to an fd in the global
1803 			 * table to an integer that is the index to the local
1804 			 * fd table entry that we set up to point to the
1805 			 * global one we are transferring.
1806 			 */
1807 			newlen = newfds * sizeof(int);
1808 			*controlp = sbcreatecontrol(NULL, newlen,
1809 			    SCM_RIGHTS, SOL_SOCKET);
1810 			if (*controlp == NULL) {
1811 				FILEDESC_XUNLOCK(fdesc);
1812 				error = E2BIG;
1813 				unp_freerights(fdep, newfds);
1814 				goto next;
1815 			}
1816 
1817 			fdp = (int *)
1818 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1819 			if (fdallocn(td, 0, fdp, newfds) != 0) {
1820 				FILEDESC_XUNLOCK(fdesc);
1821 				error = EMSGSIZE;
1822 				unp_freerights(fdep, newfds);
1823 				m_freem(*controlp);
1824 				*controlp = NULL;
1825 				goto next;
1826 			}
1827 			for (i = 0; i < newfds; i++, fdp++) {
1828 				_finstall(fdesc, fdep[i]->fde_file, *fdp,
1829 				    (flags & MSG_CMSG_CLOEXEC) != 0 ? UF_EXCLOSE : 0,
1830 				    &fdep[i]->fde_caps);
1831 				unp_externalize_fp(fdep[i]->fde_file);
1832 			}
1833 			FILEDESC_XUNLOCK(fdesc);
1834 			free(fdep[0], M_FILECAPS);
1835 		} else {
1836 			/* We can just copy anything else across. */
1837 			if (error || controlp == NULL)
1838 				goto next;
1839 			*controlp = sbcreatecontrol(NULL, datalen,
1840 			    cm->cmsg_type, cm->cmsg_level);
1841 			if (*controlp == NULL) {
1842 				error = ENOBUFS;
1843 				goto next;
1844 			}
1845 			bcopy(data,
1846 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1847 			    datalen);
1848 		}
1849 		controlp = &(*controlp)->m_next;
1850 
1851 next:
1852 		if (CMSG_SPACE(datalen) < clen) {
1853 			clen -= CMSG_SPACE(datalen);
1854 			cm = (struct cmsghdr *)
1855 			    ((caddr_t)cm + CMSG_SPACE(datalen));
1856 		} else {
1857 			clen = 0;
1858 			cm = NULL;
1859 		}
1860 	}
1861 
1862 	m_freem(control);
1863 	return (error);
1864 }
1865 
1866 static void
1867 unp_zone_change(void *tag)
1868 {
1869 
1870 	uma_zone_set_max(unp_zone, maxsockets);
1871 }
1872 
1873 static void
1874 unp_init(void)
1875 {
1876 
1877 #ifdef VIMAGE
1878 	if (!IS_DEFAULT_VNET(curvnet))
1879 		return;
1880 #endif
1881 	unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1882 	    NULL, NULL, UMA_ALIGN_PTR, 0);
1883 	if (unp_zone == NULL)
1884 		panic("unp_init");
1885 	uma_zone_set_max(unp_zone, maxsockets);
1886 	uma_zone_set_warning(unp_zone, "kern.ipc.maxsockets limit reached");
1887 	EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1888 	    NULL, EVENTHANDLER_PRI_ANY);
1889 	LIST_INIT(&unp_dhead);
1890 	LIST_INIT(&unp_shead);
1891 	LIST_INIT(&unp_sphead);
1892 	SLIST_INIT(&unp_defers);
1893 	TIMEOUT_TASK_INIT(taskqueue_thread, &unp_gc_task, 0, unp_gc, NULL);
1894 	TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1895 	UNP_LINK_LOCK_INIT();
1896 	UNP_DEFERRED_LOCK_INIT();
1897 }
1898 
1899 static int
1900 unp_internalize(struct mbuf **controlp, struct thread *td)
1901 {
1902 	struct mbuf *control = *controlp;
1903 	struct proc *p = td->td_proc;
1904 	struct filedesc *fdesc = p->p_fd;
1905 	struct bintime *bt;
1906 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1907 	struct cmsgcred *cmcred;
1908 	struct filedescent *fde, **fdep, *fdev;
1909 	struct file *fp;
1910 	struct timeval *tv;
1911 	struct timespec *ts;
1912 	int i, *fdp;
1913 	void *data;
1914 	socklen_t clen = control->m_len, datalen;
1915 	int error, oldfds;
1916 	u_int newlen;
1917 
1918 	UNP_LINK_UNLOCK_ASSERT();
1919 
1920 	error = 0;
1921 	*controlp = NULL;
1922 	while (cm != NULL) {
1923 		if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1924 		    || cm->cmsg_len > clen || cm->cmsg_len < sizeof(*cm)) {
1925 			error = EINVAL;
1926 			goto out;
1927 		}
1928 		data = CMSG_DATA(cm);
1929 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1930 
1931 		switch (cm->cmsg_type) {
1932 		/*
1933 		 * Fill in credential information.
1934 		 */
1935 		case SCM_CREDS:
1936 			*controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1937 			    SCM_CREDS, SOL_SOCKET);
1938 			if (*controlp == NULL) {
1939 				error = ENOBUFS;
1940 				goto out;
1941 			}
1942 			cmcred = (struct cmsgcred *)
1943 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1944 			cmcred->cmcred_pid = p->p_pid;
1945 			cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1946 			cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1947 			cmcred->cmcred_euid = td->td_ucred->cr_uid;
1948 			cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1949 			    CMGROUP_MAX);
1950 			for (i = 0; i < cmcred->cmcred_ngroups; i++)
1951 				cmcred->cmcred_groups[i] =
1952 				    td->td_ucred->cr_groups[i];
1953 			break;
1954 
1955 		case SCM_RIGHTS:
1956 			oldfds = datalen / sizeof (int);
1957 			if (oldfds == 0)
1958 				break;
1959 			/*
1960 			 * Check that all the FDs passed in refer to legal
1961 			 * files.  If not, reject the entire operation.
1962 			 */
1963 			fdp = data;
1964 			FILEDESC_SLOCK(fdesc);
1965 			for (i = 0; i < oldfds; i++, fdp++) {
1966 				fp = fget_locked(fdesc, *fdp);
1967 				if (fp == NULL) {
1968 					FILEDESC_SUNLOCK(fdesc);
1969 					error = EBADF;
1970 					goto out;
1971 				}
1972 				if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1973 					FILEDESC_SUNLOCK(fdesc);
1974 					error = EOPNOTSUPP;
1975 					goto out;
1976 				}
1977 
1978 			}
1979 
1980 			/*
1981 			 * Now replace the integer FDs with pointers to the
1982 			 * file structure and capability rights.
1983 			 */
1984 			newlen = oldfds * sizeof(fdep[0]);
1985 			*controlp = sbcreatecontrol(NULL, newlen,
1986 			    SCM_RIGHTS, SOL_SOCKET);
1987 			if (*controlp == NULL) {
1988 				FILEDESC_SUNLOCK(fdesc);
1989 				error = E2BIG;
1990 				goto out;
1991 			}
1992 			fdp = data;
1993 			fdep = (struct filedescent **)
1994 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1995 			fdev = malloc(sizeof(*fdev) * oldfds, M_FILECAPS,
1996 			    M_WAITOK);
1997 			for (i = 0; i < oldfds; i++, fdev++, fdp++) {
1998 				fde = &fdesc->fd_ofiles[*fdp];
1999 				fdep[i] = fdev;
2000 				fdep[i]->fde_file = fde->fde_file;
2001 				filecaps_copy(&fde->fde_caps,
2002 				    &fdep[i]->fde_caps, true);
2003 				unp_internalize_fp(fdep[i]->fde_file);
2004 			}
2005 			FILEDESC_SUNLOCK(fdesc);
2006 			break;
2007 
2008 		case SCM_TIMESTAMP:
2009 			*controlp = sbcreatecontrol(NULL, sizeof(*tv),
2010 			    SCM_TIMESTAMP, SOL_SOCKET);
2011 			if (*controlp == NULL) {
2012 				error = ENOBUFS;
2013 				goto out;
2014 			}
2015 			tv = (struct timeval *)
2016 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2017 			microtime(tv);
2018 			break;
2019 
2020 		case SCM_BINTIME:
2021 			*controlp = sbcreatecontrol(NULL, sizeof(*bt),
2022 			    SCM_BINTIME, SOL_SOCKET);
2023 			if (*controlp == NULL) {
2024 				error = ENOBUFS;
2025 				goto out;
2026 			}
2027 			bt = (struct bintime *)
2028 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2029 			bintime(bt);
2030 			break;
2031 
2032 		case SCM_REALTIME:
2033 			*controlp = sbcreatecontrol(NULL, sizeof(*ts),
2034 			    SCM_REALTIME, SOL_SOCKET);
2035 			if (*controlp == NULL) {
2036 				error = ENOBUFS;
2037 				goto out;
2038 			}
2039 			ts = (struct timespec *)
2040 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2041 			nanotime(ts);
2042 			break;
2043 
2044 		case SCM_MONOTONIC:
2045 			*controlp = sbcreatecontrol(NULL, sizeof(*ts),
2046 			    SCM_MONOTONIC, SOL_SOCKET);
2047 			if (*controlp == NULL) {
2048 				error = ENOBUFS;
2049 				goto out;
2050 			}
2051 			ts = (struct timespec *)
2052 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
2053 			nanouptime(ts);
2054 			break;
2055 
2056 		default:
2057 			error = EINVAL;
2058 			goto out;
2059 		}
2060 
2061 		controlp = &(*controlp)->m_next;
2062 		if (CMSG_SPACE(datalen) < clen) {
2063 			clen -= CMSG_SPACE(datalen);
2064 			cm = (struct cmsghdr *)
2065 			    ((caddr_t)cm + CMSG_SPACE(datalen));
2066 		} else {
2067 			clen = 0;
2068 			cm = NULL;
2069 		}
2070 	}
2071 
2072 out:
2073 	m_freem(control);
2074 	return (error);
2075 }
2076 
2077 static struct mbuf *
2078 unp_addsockcred(struct thread *td, struct mbuf *control)
2079 {
2080 	struct mbuf *m, *n, *n_prev;
2081 	struct sockcred *sc;
2082 	const struct cmsghdr *cm;
2083 	int ngroups;
2084 	int i;
2085 
2086 	ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
2087 	m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
2088 	if (m == NULL)
2089 		return (control);
2090 
2091 	sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
2092 	sc->sc_uid = td->td_ucred->cr_ruid;
2093 	sc->sc_euid = td->td_ucred->cr_uid;
2094 	sc->sc_gid = td->td_ucred->cr_rgid;
2095 	sc->sc_egid = td->td_ucred->cr_gid;
2096 	sc->sc_ngroups = ngroups;
2097 	for (i = 0; i < sc->sc_ngroups; i++)
2098 		sc->sc_groups[i] = td->td_ucred->cr_groups[i];
2099 
2100 	/*
2101 	 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
2102 	 * created SCM_CREDS control message (struct sockcred) has another
2103 	 * format.
2104 	 */
2105 	if (control != NULL)
2106 		for (n = control, n_prev = NULL; n != NULL;) {
2107 			cm = mtod(n, struct cmsghdr *);
2108     			if (cm->cmsg_level == SOL_SOCKET &&
2109 			    cm->cmsg_type == SCM_CREDS) {
2110     				if (n_prev == NULL)
2111 					control = n->m_next;
2112 				else
2113 					n_prev->m_next = n->m_next;
2114 				n = m_free(n);
2115 			} else {
2116 				n_prev = n;
2117 				n = n->m_next;
2118 			}
2119 		}
2120 
2121 	/* Prepend it to the head. */
2122 	m->m_next = control;
2123 	return (m);
2124 }
2125 
2126 static struct unpcb *
2127 fptounp(struct file *fp)
2128 {
2129 	struct socket *so;
2130 
2131 	if (fp->f_type != DTYPE_SOCKET)
2132 		return (NULL);
2133 	if ((so = fp->f_data) == NULL)
2134 		return (NULL);
2135 	if (so->so_proto->pr_domain != &localdomain)
2136 		return (NULL);
2137 	return sotounpcb(so);
2138 }
2139 
2140 static void
2141 unp_discard(struct file *fp)
2142 {
2143 	struct unp_defer *dr;
2144 
2145 	if (unp_externalize_fp(fp)) {
2146 		dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2147 		dr->ud_fp = fp;
2148 		UNP_DEFERRED_LOCK();
2149 		SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2150 		UNP_DEFERRED_UNLOCK();
2151 		atomic_add_int(&unp_defers_count, 1);
2152 		taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2153 	} else
2154 		(void) closef(fp, (struct thread *)NULL);
2155 }
2156 
2157 static void
2158 unp_process_defers(void *arg __unused, int pending)
2159 {
2160 	struct unp_defer *dr;
2161 	SLIST_HEAD(, unp_defer) drl;
2162 	int count;
2163 
2164 	SLIST_INIT(&drl);
2165 	for (;;) {
2166 		UNP_DEFERRED_LOCK();
2167 		if (SLIST_FIRST(&unp_defers) == NULL) {
2168 			UNP_DEFERRED_UNLOCK();
2169 			break;
2170 		}
2171 		SLIST_SWAP(&unp_defers, &drl, unp_defer);
2172 		UNP_DEFERRED_UNLOCK();
2173 		count = 0;
2174 		while ((dr = SLIST_FIRST(&drl)) != NULL) {
2175 			SLIST_REMOVE_HEAD(&drl, ud_link);
2176 			closef(dr->ud_fp, NULL);
2177 			free(dr, M_TEMP);
2178 			count++;
2179 		}
2180 		atomic_add_int(&unp_defers_count, -count);
2181 	}
2182 }
2183 
2184 static void
2185 unp_internalize_fp(struct file *fp)
2186 {
2187 	struct unpcb *unp;
2188 
2189 	UNP_LINK_WLOCK();
2190 	if ((unp = fptounp(fp)) != NULL) {
2191 		unp->unp_file = fp;
2192 		unp->unp_msgcount++;
2193 	}
2194 	fhold(fp);
2195 	unp_rights++;
2196 	UNP_LINK_WUNLOCK();
2197 }
2198 
2199 static int
2200 unp_externalize_fp(struct file *fp)
2201 {
2202 	struct unpcb *unp;
2203 	int ret;
2204 
2205 	UNP_LINK_WLOCK();
2206 	if ((unp = fptounp(fp)) != NULL) {
2207 		unp->unp_msgcount--;
2208 		ret = 1;
2209 	} else
2210 		ret = 0;
2211 	unp_rights--;
2212 	UNP_LINK_WUNLOCK();
2213 	return (ret);
2214 }
2215 
2216 /*
2217  * unp_defer indicates whether additional work has been defered for a future
2218  * pass through unp_gc().  It is thread local and does not require explicit
2219  * synchronization.
2220  */
2221 static int	unp_marked;
2222 static int	unp_unreachable;
2223 
2224 static void
2225 unp_accessable(struct filedescent **fdep, int fdcount)
2226 {
2227 	struct unpcb *unp;
2228 	struct file *fp;
2229 	int i;
2230 
2231 	for (i = 0; i < fdcount; i++) {
2232 		fp = fdep[i]->fde_file;
2233 		if ((unp = fptounp(fp)) == NULL)
2234 			continue;
2235 		if (unp->unp_gcflag & UNPGC_REF)
2236 			continue;
2237 		unp->unp_gcflag &= ~UNPGC_DEAD;
2238 		unp->unp_gcflag |= UNPGC_REF;
2239 		unp_marked++;
2240 	}
2241 }
2242 
2243 static void
2244 unp_gc_process(struct unpcb *unp)
2245 {
2246 	struct socket *so, *soa;
2247 	struct file *fp;
2248 
2249 	/* Already processed. */
2250 	if (unp->unp_gcflag & UNPGC_SCANNED)
2251 		return;
2252 	fp = unp->unp_file;
2253 
2254 	/*
2255 	 * Check for a socket potentially in a cycle.  It must be in a
2256 	 * queue as indicated by msgcount, and this must equal the file
2257 	 * reference count.  Note that when msgcount is 0 the file is NULL.
2258 	 */
2259 	if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2260 	    unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2261 		unp->unp_gcflag |= UNPGC_DEAD;
2262 		unp_unreachable++;
2263 		return;
2264 	}
2265 
2266 	so = unp->unp_socket;
2267 	SOCK_LOCK(so);
2268 	if (SOLISTENING(so)) {
2269 		/*
2270 		 * Mark all sockets in our accept queue.
2271 		 */
2272 		TAILQ_FOREACH(soa, &so->sol_comp, so_list) {
2273 			if (sotounpcb(soa)->unp_gcflag & UNPGC_IGNORE_RIGHTS)
2274 				continue;
2275 			SOCKBUF_LOCK(&soa->so_rcv);
2276 			unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2277 			SOCKBUF_UNLOCK(&soa->so_rcv);
2278 		}
2279 	} else {
2280 		/*
2281 		 * Mark all sockets we reference with RIGHTS.
2282 		 */
2283 		if ((unp->unp_gcflag & UNPGC_IGNORE_RIGHTS) == 0) {
2284 			SOCKBUF_LOCK(&so->so_rcv);
2285 			unp_scan(so->so_rcv.sb_mb, unp_accessable);
2286 			SOCKBUF_UNLOCK(&so->so_rcv);
2287 		}
2288 	}
2289 	SOCK_UNLOCK(so);
2290 	unp->unp_gcflag |= UNPGC_SCANNED;
2291 }
2292 
2293 static int unp_recycled;
2294 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2295     "Number of unreachable sockets claimed by the garbage collector.");
2296 
2297 static int unp_taskcount;
2298 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2299     "Number of times the garbage collector has run.");
2300 
2301 static void
2302 unp_gc(__unused void *arg, int pending)
2303 {
2304 	struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2305 				    NULL };
2306 	struct unp_head **head;
2307 	struct file *f, **unref;
2308 	struct unpcb *unp;
2309 	int i, total;
2310 
2311 	unp_taskcount++;
2312 	UNP_LINK_RLOCK();
2313 	/*
2314 	 * First clear all gc flags from previous runs, apart from
2315 	 * UNPGC_IGNORE_RIGHTS.
2316 	 */
2317 	for (head = heads; *head != NULL; head++)
2318 		LIST_FOREACH(unp, *head, unp_link)
2319 			unp->unp_gcflag =
2320 			    (unp->unp_gcflag & UNPGC_IGNORE_RIGHTS);
2321 
2322 	/*
2323 	 * Scan marking all reachable sockets with UNPGC_REF.  Once a socket
2324 	 * is reachable all of the sockets it references are reachable.
2325 	 * Stop the scan once we do a complete loop without discovering
2326 	 * a new reachable socket.
2327 	 */
2328 	do {
2329 		unp_unreachable = 0;
2330 		unp_marked = 0;
2331 		for (head = heads; *head != NULL; head++)
2332 			LIST_FOREACH(unp, *head, unp_link)
2333 				unp_gc_process(unp);
2334 	} while (unp_marked);
2335 	UNP_LINK_RUNLOCK();
2336 	if (unp_unreachable == 0)
2337 		return;
2338 
2339 	/*
2340 	 * Allocate space for a local list of dead unpcbs.
2341 	 */
2342 	unref = malloc(unp_unreachable * sizeof(struct file *),
2343 	    M_TEMP, M_WAITOK);
2344 
2345 	/*
2346 	 * Iterate looking for sockets which have been specifically marked
2347 	 * as as unreachable and store them locally.
2348 	 */
2349 	UNP_LINK_RLOCK();
2350 	for (total = 0, head = heads; *head != NULL; head++)
2351 		LIST_FOREACH(unp, *head, unp_link)
2352 			if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2353 				f = unp->unp_file;
2354 				if (unp->unp_msgcount == 0 || f == NULL ||
2355 				    f->f_count != unp->unp_msgcount)
2356 					continue;
2357 				unref[total++] = f;
2358 				fhold(f);
2359 				KASSERT(total <= unp_unreachable,
2360 				    ("unp_gc: incorrect unreachable count."));
2361 			}
2362 	UNP_LINK_RUNLOCK();
2363 
2364 	/*
2365 	 * Now flush all sockets, free'ing rights.  This will free the
2366 	 * struct files associated with these sockets but leave each socket
2367 	 * with one remaining ref.
2368 	 */
2369 	for (i = 0; i < total; i++) {
2370 		struct socket *so;
2371 
2372 		so = unref[i]->f_data;
2373 		CURVNET_SET(so->so_vnet);
2374 		sorflush(so);
2375 		CURVNET_RESTORE();
2376 	}
2377 
2378 	/*
2379 	 * And finally release the sockets so they can be reclaimed.
2380 	 */
2381 	for (i = 0; i < total; i++)
2382 		fdrop(unref[i], NULL);
2383 	unp_recycled += total;
2384 	free(unref, M_TEMP);
2385 }
2386 
2387 static void
2388 unp_dispose_mbuf(struct mbuf *m)
2389 {
2390 
2391 	if (m)
2392 		unp_scan(m, unp_freerights);
2393 }
2394 
2395 /*
2396  * Synchronize against unp_gc, which can trip over data as we are freeing it.
2397  */
2398 static void
2399 unp_dispose(struct socket *so)
2400 {
2401 	struct unpcb *unp;
2402 
2403 	unp = sotounpcb(so);
2404 	UNP_LINK_WLOCK();
2405 	unp->unp_gcflag |= UNPGC_IGNORE_RIGHTS;
2406 	UNP_LINK_WUNLOCK();
2407 	if (!SOLISTENING(so))
2408 		unp_dispose_mbuf(so->so_rcv.sb_mb);
2409 }
2410 
2411 static void
2412 unp_scan(struct mbuf *m0, void (*op)(struct filedescent **, int))
2413 {
2414 	struct mbuf *m;
2415 	struct cmsghdr *cm;
2416 	void *data;
2417 	socklen_t clen, datalen;
2418 
2419 	while (m0 != NULL) {
2420 		for (m = m0; m; m = m->m_next) {
2421 			if (m->m_type != MT_CONTROL)
2422 				continue;
2423 
2424 			cm = mtod(m, struct cmsghdr *);
2425 			clen = m->m_len;
2426 
2427 			while (cm != NULL) {
2428 				if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2429 					break;
2430 
2431 				data = CMSG_DATA(cm);
2432 				datalen = (caddr_t)cm + cm->cmsg_len
2433 				    - (caddr_t)data;
2434 
2435 				if (cm->cmsg_level == SOL_SOCKET &&
2436 				    cm->cmsg_type == SCM_RIGHTS) {
2437 					(*op)(data, datalen /
2438 					    sizeof(struct filedescent *));
2439 				}
2440 
2441 				if (CMSG_SPACE(datalen) < clen) {
2442 					clen -= CMSG_SPACE(datalen);
2443 					cm = (struct cmsghdr *)
2444 					    ((caddr_t)cm + CMSG_SPACE(datalen));
2445 				} else {
2446 					clen = 0;
2447 					cm = NULL;
2448 				}
2449 			}
2450 		}
2451 		m0 = m0->m_nextpkt;
2452 	}
2453 }
2454 
2455 /*
2456  * A helper function called by VFS before socket-type vnode reclamation.
2457  * For an active vnode it clears unp_vnode pointer and decrements unp_vnode
2458  * use count.
2459  */
2460 void
2461 vfs_unp_reclaim(struct vnode *vp)
2462 {
2463 	struct unpcb *unp;
2464 	int active;
2465 
2466 	ASSERT_VOP_ELOCKED(vp, "vfs_unp_reclaim");
2467 	KASSERT(vp->v_type == VSOCK,
2468 	    ("vfs_unp_reclaim: vp->v_type != VSOCK"));
2469 
2470 	active = 0;
2471 	UNP_LINK_WLOCK();
2472 	VOP_UNP_CONNECT(vp, &unp);
2473 	if (unp == NULL)
2474 		goto done;
2475 	UNP_PCB_LOCK(unp);
2476 	if (unp->unp_vnode == vp) {
2477 		VOP_UNP_DETACH(vp);
2478 		unp->unp_vnode = NULL;
2479 		active = 1;
2480 	}
2481 	UNP_PCB_UNLOCK(unp);
2482 done:
2483 	UNP_LINK_WUNLOCK();
2484 	if (active)
2485 		vunref(vp);
2486 }
2487 
2488 #ifdef DDB
2489 static void
2490 db_print_indent(int indent)
2491 {
2492 	int i;
2493 
2494 	for (i = 0; i < indent; i++)
2495 		db_printf(" ");
2496 }
2497 
2498 static void
2499 db_print_unpflags(int unp_flags)
2500 {
2501 	int comma;
2502 
2503 	comma = 0;
2504 	if (unp_flags & UNP_HAVEPC) {
2505 		db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2506 		comma = 1;
2507 	}
2508 	if (unp_flags & UNP_WANTCRED) {
2509 		db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2510 		comma = 1;
2511 	}
2512 	if (unp_flags & UNP_CONNWAIT) {
2513 		db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2514 		comma = 1;
2515 	}
2516 	if (unp_flags & UNP_CONNECTING) {
2517 		db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2518 		comma = 1;
2519 	}
2520 	if (unp_flags & UNP_BINDING) {
2521 		db_printf("%sUNP_BINDING", comma ? ", " : "");
2522 		comma = 1;
2523 	}
2524 }
2525 
2526 static void
2527 db_print_xucred(int indent, struct xucred *xu)
2528 {
2529 	int comma, i;
2530 
2531 	db_print_indent(indent);
2532 	db_printf("cr_version: %u   cr_uid: %u   cr_ngroups: %d\n",
2533 	    xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2534 	db_print_indent(indent);
2535 	db_printf("cr_groups: ");
2536 	comma = 0;
2537 	for (i = 0; i < xu->cr_ngroups; i++) {
2538 		db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2539 		comma = 1;
2540 	}
2541 	db_printf("\n");
2542 }
2543 
2544 static void
2545 db_print_unprefs(int indent, struct unp_head *uh)
2546 {
2547 	struct unpcb *unp;
2548 	int counter;
2549 
2550 	counter = 0;
2551 	LIST_FOREACH(unp, uh, unp_reflink) {
2552 		if (counter % 4 == 0)
2553 			db_print_indent(indent);
2554 		db_printf("%p  ", unp);
2555 		if (counter % 4 == 3)
2556 			db_printf("\n");
2557 		counter++;
2558 	}
2559 	if (counter != 0 && counter % 4 != 0)
2560 		db_printf("\n");
2561 }
2562 
2563 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2564 {
2565 	struct unpcb *unp;
2566 
2567         if (!have_addr) {
2568                 db_printf("usage: show unpcb <addr>\n");
2569                 return;
2570         }
2571         unp = (struct unpcb *)addr;
2572 
2573 	db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2574 	    unp->unp_vnode);
2575 
2576 	db_printf("unp_ino: %ju   unp_conn: %p\n", (uintmax_t)unp->unp_ino,
2577 	    unp->unp_conn);
2578 
2579 	db_printf("unp_refs:\n");
2580 	db_print_unprefs(2, &unp->unp_refs);
2581 
2582 	/* XXXRW: Would be nice to print the full address, if any. */
2583 	db_printf("unp_addr: %p\n", unp->unp_addr);
2584 
2585 	db_printf("unp_gencnt: %llu\n",
2586 	    (unsigned long long)unp->unp_gencnt);
2587 
2588 	db_printf("unp_flags: %x (", unp->unp_flags);
2589 	db_print_unpflags(unp->unp_flags);
2590 	db_printf(")\n");
2591 
2592 	db_printf("unp_peercred:\n");
2593 	db_print_xucred(2, &unp->unp_peercred);
2594 
2595 	db_printf("unp_refcount: %u\n", unp->unp_refcount);
2596 }
2597 #endif
2598