xref: /freebsd/sys/kern/uipc_usrreq.c (revision 884a2a699669ec61e2366e3e358342dbc94be24a)
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  *	distinguish datagram size limits from flow control limits in SEQPACKET
55  *	rethink name space problems
56  *	need a proper out-of-band
57  */
58 
59 #include <sys/cdefs.h>
60 __FBSDID("$FreeBSD$");
61 
62 #include "opt_ddb.h"
63 
64 #include <sys/param.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 /*
105  * Locking key:
106  * (l)	Locked using list lock
107  * (g)	Locked using linkage lock
108  */
109 
110 static uma_zone_t	unp_zone;
111 static unp_gen_t	unp_gencnt;	/* (l) */
112 static u_int		unp_count;	/* (l) Count of local sockets. */
113 static ino_t		unp_ino;	/* Prototype for fake inode numbers. */
114 static int		unp_rights;	/* (g) File descriptors in flight. */
115 static struct unp_head	unp_shead;	/* (l) List of stream sockets. */
116 static struct unp_head	unp_dhead;	/* (l) List of datagram sockets. */
117 static struct unp_head	unp_sphead;	/* (l) List of seqpacket sockets. */
118 
119 struct unp_defer {
120 	SLIST_ENTRY(unp_defer) ud_link;
121 	struct file *ud_fp;
122 };
123 static SLIST_HEAD(, unp_defer) unp_defers;
124 static int unp_defers_count;
125 
126 static const struct sockaddr	sun_noname = { sizeof(sun_noname), AF_LOCAL };
127 
128 /*
129  * Garbage collection of cyclic file descriptor/socket references occurs
130  * asynchronously in a taskqueue context in order to avoid recursion and
131  * reentrance in the UNIX domain socket, file descriptor, and socket layer
132  * code.  See unp_gc() for a full description.
133  */
134 static struct task	unp_gc_task;
135 
136 /*
137  * The close of unix domain sockets attached as SCM_RIGHTS is
138  * postponed to the taskqueue, to avoid arbitrary recursion depth.
139  * The attached sockets might have another sockets attached.
140  */
141 static struct task	unp_defer_task;
142 
143 /*
144  * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
145  * stream sockets, although the total for sender and receiver is actually
146  * only PIPSIZ.
147  *
148  * Datagram sockets really use the sendspace as the maximum datagram size,
149  * and don't really want to reserve the sendspace.  Their recvspace should be
150  * large enough for at least one max-size datagram plus address.
151  */
152 #ifndef PIPSIZ
153 #define	PIPSIZ	8192
154 #endif
155 static u_long	unpst_sendspace = PIPSIZ;
156 static u_long	unpst_recvspace = PIPSIZ;
157 static u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
158 static u_long	unpdg_recvspace = 4*1024;
159 static u_long	unpsp_sendspace = PIPSIZ;	/* really max datagram size */
160 static u_long	unpsp_recvspace = PIPSIZ;
161 
162 SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
163 SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, "SOCK_STREAM");
164 SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
165 SYSCTL_NODE(_net_local, SOCK_SEQPACKET, seqpacket, CTLFLAG_RW, 0,
166     "SOCK_SEQPACKET");
167 
168 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
169 	   &unpst_sendspace, 0, "Default stream send space.");
170 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
171 	   &unpst_recvspace, 0, "Default stream receive space.");
172 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
173 	   &unpdg_sendspace, 0, "Default datagram send space.");
174 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
175 	   &unpdg_recvspace, 0, "Default datagram receive space.");
176 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, maxseqpacket, CTLFLAG_RW,
177 	   &unpsp_sendspace, 0, "Default seqpacket send space.");
178 SYSCTL_ULONG(_net_local_seqpacket, OID_AUTO, recvspace, CTLFLAG_RW,
179 	   &unpsp_recvspace, 0, "Default seqpacket receive space.");
180 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0,
181     "File descriptors in flight.");
182 SYSCTL_INT(_net_local, OID_AUTO, deferred, CTLFLAG_RD,
183     &unp_defers_count, 0,
184     "File descriptors deferred to taskqueue for close.");
185 
186 /*
187  * Locking and synchronization:
188  *
189  * Three types of locks exit in the local domain socket implementation: a
190  * global list mutex, a global linkage rwlock, and per-unpcb mutexes.  Of the
191  * global locks, the list lock protects the socket count, global generation
192  * number, and stream/datagram global lists.  The linkage lock protects the
193  * interconnection of unpcbs, the v_socket and unp_vnode pointers, and can be
194  * held exclusively over the acquisition of multiple unpcb locks to prevent
195  * deadlock.
196  *
197  * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
198  * allocated in pru_attach() and freed in pru_detach().  The validity of that
199  * pointer is an invariant, so no lock is required to dereference the so_pcb
200  * pointer if a valid socket reference is held by the caller.  In practice,
201  * this is always true during operations performed on a socket.  Each unpcb
202  * has a back-pointer to its socket, unp_socket, which will be stable under
203  * the same circumstances.
204  *
205  * This pointer may only be safely dereferenced as long as a valid reference
206  * to the unpcb is held.  Typically, this reference will be from the socket,
207  * or from another unpcb when the referring unpcb's lock is held (in order
208  * that the reference not be invalidated during use).  For example, to follow
209  * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
210  * as unp_socket remains valid as long as the reference to unp_conn is valid.
211  *
212  * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx.  Individual
213  * atomic reads without the lock may be performed "lockless", but more
214  * complex reads and read-modify-writes require the mutex to be held.  No
215  * lock order is defined between unpcb locks -- multiple unpcb locks may be
216  * acquired at the same time only when holding the linkage rwlock
217  * exclusively, which prevents deadlocks.
218  *
219  * Blocking with UNIX domain sockets is a tricky issue: unlike most network
220  * protocols, bind() is a non-atomic operation, and connect() requires
221  * potential sleeping in the protocol, due to potentially waiting on local or
222  * distributed file systems.  We try to separate "lookup" operations, which
223  * may sleep, and the IPC operations themselves, which typically can occur
224  * with relative atomicity as locks can be held over the entire operation.
225  *
226  * Another tricky issue is simultaneous multi-threaded or multi-process
227  * access to a single UNIX domain socket.  These are handled by the flags
228  * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
229  * binding, both of which involve dropping UNIX domain socket locks in order
230  * to perform namei() and other file system operations.
231  */
232 static struct rwlock	unp_link_rwlock;
233 static struct mtx	unp_list_lock;
234 static struct mtx	unp_defers_lock;
235 
236 #define	UNP_LINK_LOCK_INIT()		rw_init(&unp_link_rwlock,	\
237 					    "unp_link_rwlock")
238 
239 #define	UNP_LINK_LOCK_ASSERT()	rw_assert(&unp_link_rwlock,	\
240 					    RA_LOCKED)
241 #define	UNP_LINK_UNLOCK_ASSERT()	rw_assert(&unp_link_rwlock,	\
242 					    RA_UNLOCKED)
243 
244 #define	UNP_LINK_RLOCK()		rw_rlock(&unp_link_rwlock)
245 #define	UNP_LINK_RUNLOCK()		rw_runlock(&unp_link_rwlock)
246 #define	UNP_LINK_WLOCK()		rw_wlock(&unp_link_rwlock)
247 #define	UNP_LINK_WUNLOCK()		rw_wunlock(&unp_link_rwlock)
248 #define	UNP_LINK_WLOCK_ASSERT()		rw_assert(&unp_link_rwlock,	\
249 					    RA_WLOCKED)
250 
251 #define	UNP_LIST_LOCK_INIT()		mtx_init(&unp_list_lock,	\
252 					    "unp_list_lock", NULL, MTX_DEF)
253 #define	UNP_LIST_LOCK()			mtx_lock(&unp_list_lock)
254 #define	UNP_LIST_UNLOCK()		mtx_unlock(&unp_list_lock)
255 
256 #define	UNP_DEFERRED_LOCK_INIT()	mtx_init(&unp_defers_lock, \
257 					    "unp_defer", NULL, MTX_DEF)
258 #define	UNP_DEFERRED_LOCK()		mtx_lock(&unp_defers_lock)
259 #define	UNP_DEFERRED_UNLOCK()		mtx_unlock(&unp_defers_lock)
260 
261 #define UNP_PCB_LOCK_INIT(unp)		mtx_init(&(unp)->unp_mtx,	\
262 					    "unp_mtx", "unp_mtx",	\
263 					    MTX_DUPOK|MTX_DEF|MTX_RECURSE)
264 #define	UNP_PCB_LOCK_DESTROY(unp)	mtx_destroy(&(unp)->unp_mtx)
265 #define	UNP_PCB_LOCK(unp)		mtx_lock(&(unp)->unp_mtx)
266 #define	UNP_PCB_UNLOCK(unp)		mtx_unlock(&(unp)->unp_mtx)
267 #define	UNP_PCB_LOCK_ASSERT(unp)	mtx_assert(&(unp)->unp_mtx, MA_OWNED)
268 
269 static int	uipc_connect2(struct socket *, struct socket *);
270 static int	uipc_ctloutput(struct socket *, struct sockopt *);
271 static int	unp_connect(struct socket *, struct sockaddr *,
272 		    struct thread *);
273 static int	unp_connect2(struct socket *so, struct socket *so2, int);
274 static void	unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
275 static void	unp_dispose(struct mbuf *);
276 static void	unp_shutdown(struct unpcb *);
277 static void	unp_drop(struct unpcb *, int);
278 static void	unp_gc(__unused void *, int);
279 static void	unp_scan(struct mbuf *, void (*)(struct file *));
280 static void	unp_discard(struct file *);
281 static void	unp_freerights(struct file **, 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 **);
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_usrreqs =		&uipc_usrreqs_dgram
309 },
310 {
311 	.pr_type =		SOCK_SEQPACKET,
312 	.pr_domain =		&localdomain,
313 
314 	/*
315 	 * XXXRW: For now, PR_ADDR because soreceive will bump into them
316 	 * due to our use of sbappendaddr.  A new sbappend variants is needed
317 	 * that supports both atomic record writes and control data.
318 	 */
319 	.pr_flags =		PR_ADDR|PR_ATOMIC|PR_CONNREQUIRED|PR_WANTRCVD|
320 				    PR_RIGHTS,
321 	.pr_usrreqs =		&uipc_usrreqs_seqpacket,
322 },
323 };
324 
325 static struct domain localdomain = {
326 	.dom_family =		AF_LOCAL,
327 	.dom_name =		"local",
328 	.dom_init =		unp_init,
329 	.dom_externalize =	unp_externalize,
330 	.dom_dispose =		unp_dispose,
331 	.dom_protosw =		localsw,
332 	.dom_protoswNPROTOSW =	&localsw[sizeof(localsw)/sizeof(localsw[0])]
333 };
334 DOMAIN_SET(local);
335 
336 static void
337 uipc_abort(struct socket *so)
338 {
339 	struct unpcb *unp, *unp2;
340 
341 	unp = sotounpcb(so);
342 	KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
343 
344 	UNP_LINK_WLOCK();
345 	UNP_PCB_LOCK(unp);
346 	unp2 = unp->unp_conn;
347 	if (unp2 != NULL) {
348 		UNP_PCB_LOCK(unp2);
349 		unp_drop(unp2, ECONNABORTED);
350 		UNP_PCB_UNLOCK(unp2);
351 	}
352 	UNP_PCB_UNLOCK(unp);
353 	UNP_LINK_WUNLOCK();
354 }
355 
356 static int
357 uipc_accept(struct socket *so, struct sockaddr **nam)
358 {
359 	struct unpcb *unp, *unp2;
360 	const struct sockaddr *sa;
361 
362 	/*
363 	 * Pass back name of connected socket, if it was bound and we are
364 	 * still connected (our peer may have closed already!).
365 	 */
366 	unp = sotounpcb(so);
367 	KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
368 
369 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
370 	UNP_LINK_RLOCK();
371 	unp2 = unp->unp_conn;
372 	if (unp2 != NULL && unp2->unp_addr != NULL) {
373 		UNP_PCB_LOCK(unp2);
374 		sa = (struct sockaddr *) unp2->unp_addr;
375 		bcopy(sa, *nam, sa->sa_len);
376 		UNP_PCB_UNLOCK(unp2);
377 	} else {
378 		sa = &sun_noname;
379 		bcopy(sa, *nam, sa->sa_len);
380 	}
381 	UNP_LINK_RUNLOCK();
382 	return (0);
383 }
384 
385 static int
386 uipc_attach(struct socket *so, int proto, struct thread *td)
387 {
388 	u_long sendspace, recvspace;
389 	struct unpcb *unp;
390 	int error;
391 
392 	KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
393 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
394 		switch (so->so_type) {
395 		case SOCK_STREAM:
396 			sendspace = unpst_sendspace;
397 			recvspace = unpst_recvspace;
398 			break;
399 
400 		case SOCK_DGRAM:
401 			sendspace = unpdg_sendspace;
402 			recvspace = unpdg_recvspace;
403 			break;
404 
405 		case SOCK_SEQPACKET:
406 			sendspace = unpsp_sendspace;
407 			recvspace = unpsp_recvspace;
408 			break;
409 
410 		default:
411 			panic("uipc_attach");
412 		}
413 		error = soreserve(so, sendspace, recvspace);
414 		if (error)
415 			return (error);
416 	}
417 	unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
418 	if (unp == NULL)
419 		return (ENOBUFS);
420 	LIST_INIT(&unp->unp_refs);
421 	UNP_PCB_LOCK_INIT(unp);
422 	unp->unp_socket = so;
423 	so->so_pcb = unp;
424 	unp->unp_refcount = 1;
425 
426 	UNP_LIST_LOCK();
427 	unp->unp_gencnt = ++unp_gencnt;
428 	unp_count++;
429 	switch (so->so_type) {
430 	case SOCK_STREAM:
431 		LIST_INSERT_HEAD(&unp_shead, unp, unp_link);
432 		break;
433 
434 	case SOCK_DGRAM:
435 		LIST_INSERT_HEAD(&unp_dhead, unp, unp_link);
436 		break;
437 
438 	case SOCK_SEQPACKET:
439 		LIST_INSERT_HEAD(&unp_sphead, unp, unp_link);
440 		break;
441 
442 	default:
443 		panic("uipc_attach");
444 	}
445 	UNP_LIST_UNLOCK();
446 
447 	return (0);
448 }
449 
450 static int
451 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
452 {
453 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
454 	struct vattr vattr;
455 	int error, namelen, vfslocked;
456 	struct nameidata nd;
457 	struct unpcb *unp;
458 	struct vnode *vp;
459 	struct mount *mp;
460 	char *buf;
461 
462 	unp = sotounpcb(so);
463 	KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
464 
465 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
466 	if (namelen <= 0)
467 		return (EINVAL);
468 
469 	/*
470 	 * We don't allow simultaneous bind() calls on a single UNIX domain
471 	 * socket, so flag in-progress operations, and return an error if an
472 	 * operation is already in progress.
473 	 *
474 	 * Historically, we have not allowed a socket to be rebound, so this
475 	 * also returns an error.  Not allowing re-binding simplifies the
476 	 * implementation and avoids a great many possible failure modes.
477 	 */
478 	UNP_PCB_LOCK(unp);
479 	if (unp->unp_vnode != NULL) {
480 		UNP_PCB_UNLOCK(unp);
481 		return (EINVAL);
482 	}
483 	if (unp->unp_flags & UNP_BINDING) {
484 		UNP_PCB_UNLOCK(unp);
485 		return (EALREADY);
486 	}
487 	unp->unp_flags |= UNP_BINDING;
488 	UNP_PCB_UNLOCK(unp);
489 
490 	buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
491 	bcopy(soun->sun_path, buf, namelen);
492 	buf[namelen] = 0;
493 
494 restart:
495 	vfslocked = 0;
496 	NDINIT(&nd, CREATE, MPSAFE | NOFOLLOW | LOCKPARENT | SAVENAME,
497 	    UIO_SYSSPACE, buf, td);
498 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
499 	error = namei(&nd);
500 	if (error)
501 		goto error;
502 	vp = nd.ni_vp;
503 	vfslocked = NDHASGIANT(&nd);
504 	if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
505 		NDFREE(&nd, NDF_ONLY_PNBUF);
506 		if (nd.ni_dvp == vp)
507 			vrele(nd.ni_dvp);
508 		else
509 			vput(nd.ni_dvp);
510 		if (vp != NULL) {
511 			vrele(vp);
512 			error = EADDRINUSE;
513 			goto error;
514 		}
515 		error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
516 		if (error)
517 			goto error;
518 		VFS_UNLOCK_GIANT(vfslocked);
519 		goto restart;
520 	}
521 	VATTR_NULL(&vattr);
522 	vattr.va_type = VSOCK;
523 	vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
524 #ifdef MAC
525 	error = mac_vnode_check_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
526 	    &vattr);
527 #endif
528 	if (error == 0)
529 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
530 	NDFREE(&nd, NDF_ONLY_PNBUF);
531 	vput(nd.ni_dvp);
532 	if (error) {
533 		vn_finished_write(mp);
534 		goto error;
535 	}
536 	vp = nd.ni_vp;
537 	ASSERT_VOP_ELOCKED(vp, "uipc_bind");
538 	soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
539 
540 	UNP_LINK_WLOCK();
541 	UNP_PCB_LOCK(unp);
542 	vp->v_socket = unp->unp_socket;
543 	unp->unp_vnode = vp;
544 	unp->unp_addr = soun;
545 	unp->unp_flags &= ~UNP_BINDING;
546 	UNP_PCB_UNLOCK(unp);
547 	UNP_LINK_WUNLOCK();
548 	VOP_UNLOCK(vp, 0);
549 	vn_finished_write(mp);
550 	VFS_UNLOCK_GIANT(vfslocked);
551 	free(buf, M_TEMP);
552 	return (0);
553 
554 error:
555 	VFS_UNLOCK_GIANT(vfslocked);
556 	UNP_PCB_LOCK(unp);
557 	unp->unp_flags &= ~UNP_BINDING;
558 	UNP_PCB_UNLOCK(unp);
559 	free(buf, M_TEMP);
560 	return (error);
561 }
562 
563 static int
564 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
565 {
566 	int error;
567 
568 	KASSERT(td == curthread, ("uipc_connect: td != curthread"));
569 	UNP_LINK_WLOCK();
570 	error = unp_connect(so, nam, td);
571 	UNP_LINK_WUNLOCK();
572 	return (error);
573 }
574 
575 static void
576 uipc_close(struct socket *so)
577 {
578 	struct unpcb *unp, *unp2;
579 
580 	unp = sotounpcb(so);
581 	KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
582 
583 	UNP_LINK_WLOCK();
584 	UNP_PCB_LOCK(unp);
585 	unp2 = unp->unp_conn;
586 	if (unp2 != NULL) {
587 		UNP_PCB_LOCK(unp2);
588 		unp_disconnect(unp, unp2);
589 		UNP_PCB_UNLOCK(unp2);
590 	}
591 	UNP_PCB_UNLOCK(unp);
592 	UNP_LINK_WUNLOCK();
593 }
594 
595 static int
596 uipc_connect2(struct socket *so1, struct socket *so2)
597 {
598 	struct unpcb *unp, *unp2;
599 	int error;
600 
601 	UNP_LINK_WLOCK();
602 	unp = so1->so_pcb;
603 	KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
604 	UNP_PCB_LOCK(unp);
605 	unp2 = so2->so_pcb;
606 	KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
607 	UNP_PCB_LOCK(unp2);
608 	error = unp_connect2(so1, so2, PRU_CONNECT2);
609 	UNP_PCB_UNLOCK(unp2);
610 	UNP_PCB_UNLOCK(unp);
611 	UNP_LINK_WUNLOCK();
612 	return (error);
613 }
614 
615 static void
616 uipc_detach(struct socket *so)
617 {
618 	struct unpcb *unp, *unp2;
619 	struct sockaddr_un *saved_unp_addr;
620 	struct vnode *vp;
621 	int freeunp, local_unp_rights;
622 
623 	unp = sotounpcb(so);
624 	KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
625 
626 	UNP_LINK_WLOCK();
627 	UNP_LIST_LOCK();
628 	UNP_PCB_LOCK(unp);
629 	LIST_REMOVE(unp, unp_link);
630 	unp->unp_gencnt = ++unp_gencnt;
631 	--unp_count;
632 	UNP_LIST_UNLOCK();
633 
634 	/*
635 	 * XXXRW: Should assert vp->v_socket == so.
636 	 */
637 	if ((vp = unp->unp_vnode) != NULL) {
638 		unp->unp_vnode->v_socket = NULL;
639 		unp->unp_vnode = NULL;
640 	}
641 	unp2 = unp->unp_conn;
642 	if (unp2 != NULL) {
643 		UNP_PCB_LOCK(unp2);
644 		unp_disconnect(unp, unp2);
645 		UNP_PCB_UNLOCK(unp2);
646 	}
647 
648 	/*
649 	 * We hold the linkage lock exclusively, so it's OK to acquire
650 	 * multiple pcb locks at a time.
651 	 */
652 	while (!LIST_EMPTY(&unp->unp_refs)) {
653 		struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
654 
655 		UNP_PCB_LOCK(ref);
656 		unp_drop(ref, ECONNRESET);
657 		UNP_PCB_UNLOCK(ref);
658 	}
659 	local_unp_rights = unp_rights;
660 	UNP_LINK_WUNLOCK();
661 	unp->unp_socket->so_pcb = NULL;
662 	saved_unp_addr = unp->unp_addr;
663 	unp->unp_addr = NULL;
664 	unp->unp_refcount--;
665 	freeunp = (unp->unp_refcount == 0);
666 	if (saved_unp_addr != NULL)
667 		free(saved_unp_addr, M_SONAME);
668 	if (freeunp) {
669 		UNP_PCB_LOCK_DESTROY(unp);
670 		uma_zfree(unp_zone, unp);
671 	} else
672 		UNP_PCB_UNLOCK(unp);
673 	if (vp) {
674 		int vfslocked;
675 
676 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
677 		vrele(vp);
678 		VFS_UNLOCK_GIANT(vfslocked);
679 	}
680 	if (local_unp_rights)
681 		taskqueue_enqueue(taskqueue_thread, &unp_gc_task);
682 }
683 
684 static int
685 uipc_disconnect(struct socket *so)
686 {
687 	struct unpcb *unp, *unp2;
688 
689 	unp = sotounpcb(so);
690 	KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
691 
692 	UNP_LINK_WLOCK();
693 	UNP_PCB_LOCK(unp);
694 	unp2 = unp->unp_conn;
695 	if (unp2 != NULL) {
696 		UNP_PCB_LOCK(unp2);
697 		unp_disconnect(unp, unp2);
698 		UNP_PCB_UNLOCK(unp2);
699 	}
700 	UNP_PCB_UNLOCK(unp);
701 	UNP_LINK_WUNLOCK();
702 	return (0);
703 }
704 
705 static int
706 uipc_listen(struct socket *so, int backlog, struct thread *td)
707 {
708 	struct unpcb *unp;
709 	int error;
710 
711 	unp = sotounpcb(so);
712 	KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
713 
714 	UNP_PCB_LOCK(unp);
715 	if (unp->unp_vnode == NULL) {
716 		UNP_PCB_UNLOCK(unp);
717 		return (EINVAL);
718 	}
719 
720 	SOCK_LOCK(so);
721 	error = solisten_proto_check(so);
722 	if (error == 0) {
723 		cru2x(td->td_ucred, &unp->unp_peercred);
724 		unp->unp_flags |= UNP_HAVEPCCACHED;
725 		solisten_proto(so, backlog);
726 	}
727 	SOCK_UNLOCK(so);
728 	UNP_PCB_UNLOCK(unp);
729 	return (error);
730 }
731 
732 static int
733 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
734 {
735 	struct unpcb *unp, *unp2;
736 	const struct sockaddr *sa;
737 
738 	unp = sotounpcb(so);
739 	KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
740 
741 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
742 	UNP_LINK_RLOCK();
743 	/*
744 	 * XXX: It seems that this test always fails even when connection is
745 	 * established.  So, this else clause is added as workaround to
746 	 * return PF_LOCAL sockaddr.
747 	 */
748 	unp2 = unp->unp_conn;
749 	if (unp2 != NULL) {
750 		UNP_PCB_LOCK(unp2);
751 		if (unp2->unp_addr != NULL)
752 			sa = (struct sockaddr *) unp2->unp_addr;
753 		else
754 			sa = &sun_noname;
755 		bcopy(sa, *nam, sa->sa_len);
756 		UNP_PCB_UNLOCK(unp2);
757 	} else {
758 		sa = &sun_noname;
759 		bcopy(sa, *nam, sa->sa_len);
760 	}
761 	UNP_LINK_RUNLOCK();
762 	return (0);
763 }
764 
765 static int
766 uipc_rcvd(struct socket *so, int flags)
767 {
768 	struct unpcb *unp, *unp2;
769 	struct socket *so2;
770 	u_int mbcnt, sbcc;
771 	u_long newhiwat;
772 
773 	unp = sotounpcb(so);
774 	KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
775 
776 	if (so->so_type != SOCK_STREAM && so->so_type != SOCK_SEQPACKET)
777 		panic("uipc_rcvd socktype %d", so->so_type);
778 
779 	/*
780 	 * Adjust backpressure on sender and wakeup any waiting to write.
781 	 *
782 	 * The unp lock is acquired to maintain the validity of the unp_conn
783 	 * pointer; no lock on unp2 is required as unp2->unp_socket will be
784 	 * static as long as we don't permit unp2 to disconnect from unp,
785 	 * which is prevented by the lock on unp.  We cache values from
786 	 * so_rcv to avoid holding the so_rcv lock over the entire
787 	 * transaction on the remote so_snd.
788 	 */
789 	SOCKBUF_LOCK(&so->so_rcv);
790 	mbcnt = so->so_rcv.sb_mbcnt;
791 	sbcc = so->so_rcv.sb_cc;
792 	SOCKBUF_UNLOCK(&so->so_rcv);
793 	UNP_PCB_LOCK(unp);
794 	unp2 = unp->unp_conn;
795 	if (unp2 == NULL) {
796 		UNP_PCB_UNLOCK(unp);
797 		return (0);
798 	}
799 	so2 = unp2->unp_socket;
800 	SOCKBUF_LOCK(&so2->so_snd);
801 	so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt;
802 	newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc;
803 	(void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
804 	    newhiwat, RLIM_INFINITY);
805 	sowwakeup_locked(so2);
806 	unp->unp_mbcnt = mbcnt;
807 	unp->unp_cc = sbcc;
808 	UNP_PCB_UNLOCK(unp);
809 	return (0);
810 }
811 
812 static int
813 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
814     struct mbuf *control, struct thread *td)
815 {
816 	struct unpcb *unp, *unp2;
817 	struct socket *so2;
818 	u_int mbcnt_delta, sbcc;
819 	u_long newhiwat;
820 	int error = 0;
821 
822 	unp = sotounpcb(so);
823 	KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
824 
825 	if (flags & PRUS_OOB) {
826 		error = EOPNOTSUPP;
827 		goto release;
828 	}
829 	if (control != NULL && (error = unp_internalize(&control, td)))
830 		goto release;
831 	if ((nam != NULL) || (flags & PRUS_EOF))
832 		UNP_LINK_WLOCK();
833 	else
834 		UNP_LINK_RLOCK();
835 	switch (so->so_type) {
836 	case SOCK_DGRAM:
837 	{
838 		const struct sockaddr *from;
839 
840 		unp2 = unp->unp_conn;
841 		if (nam != NULL) {
842 			UNP_LINK_WLOCK_ASSERT();
843 			if (unp2 != NULL) {
844 				error = EISCONN;
845 				break;
846 			}
847 			error = unp_connect(so, nam, td);
848 			if (error)
849 				break;
850 			unp2 = unp->unp_conn;
851 		}
852 
853 		/*
854 		 * Because connect() and send() are non-atomic in a sendto()
855 		 * with a target address, it's possible that the socket will
856 		 * have disconnected before the send() can run.  In that case
857 		 * return the slightly counter-intuitive but otherwise
858 		 * correct error that the socket is not connected.
859 		 */
860 		if (unp2 == NULL) {
861 			error = ENOTCONN;
862 			break;
863 		}
864 		/* Lockless read. */
865 		if (unp2->unp_flags & UNP_WANTCRED)
866 			control = unp_addsockcred(td, control);
867 		UNP_PCB_LOCK(unp);
868 		if (unp->unp_addr != NULL)
869 			from = (struct sockaddr *)unp->unp_addr;
870 		else
871 			from = &sun_noname;
872 		so2 = unp2->unp_socket;
873 		SOCKBUF_LOCK(&so2->so_rcv);
874 		if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
875 			sorwakeup_locked(so2);
876 			m = NULL;
877 			control = NULL;
878 		} else {
879 			SOCKBUF_UNLOCK(&so2->so_rcv);
880 			error = ENOBUFS;
881 		}
882 		if (nam != NULL) {
883 			UNP_LINK_WLOCK_ASSERT();
884 			UNP_PCB_LOCK(unp2);
885 			unp_disconnect(unp, unp2);
886 			UNP_PCB_UNLOCK(unp2);
887 		}
888 		UNP_PCB_UNLOCK(unp);
889 		break;
890 	}
891 
892 	case SOCK_SEQPACKET:
893 	case SOCK_STREAM:
894 		if ((so->so_state & SS_ISCONNECTED) == 0) {
895 			if (nam != NULL) {
896 				UNP_LINK_WLOCK_ASSERT();
897 				error = unp_connect(so, nam, td);
898 				if (error)
899 					break;	/* XXX */
900 			} else {
901 				error = ENOTCONN;
902 				break;
903 			}
904 		}
905 
906 		/* Lockless read. */
907 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
908 			error = EPIPE;
909 			break;
910 		}
911 
912 		/*
913 		 * Because connect() and send() are non-atomic in a sendto()
914 		 * with a target address, it's possible that the socket will
915 		 * have disconnected before the send() can run.  In that case
916 		 * return the slightly counter-intuitive but otherwise
917 		 * correct error that the socket is not connected.
918 		 *
919 		 * Locking here must be done carefully: the linkage lock
920 		 * prevents interconnections between unpcbs from changing, so
921 		 * we can traverse from unp to unp2 without acquiring unp's
922 		 * lock.  Socket buffer locks follow unpcb locks, so we can
923 		 * acquire both remote and lock socket buffer locks.
924 		 */
925 		unp2 = unp->unp_conn;
926 		if (unp2 == NULL) {
927 			error = ENOTCONN;
928 			break;
929 		}
930 		so2 = unp2->unp_socket;
931 		UNP_PCB_LOCK(unp2);
932 		SOCKBUF_LOCK(&so2->so_rcv);
933 		if (unp2->unp_flags & UNP_WANTCRED) {
934 			/*
935 			 * Credentials are passed only once on SOCK_STREAM.
936 			 */
937 			unp2->unp_flags &= ~UNP_WANTCRED;
938 			control = unp_addsockcred(td, control);
939 		}
940 		/*
941 		 * Send to paired receive port, and then reduce send buffer
942 		 * hiwater marks to maintain backpressure.  Wake up readers.
943 		 */
944 		switch (so->so_type) {
945 		case SOCK_STREAM:
946 			if (control != NULL) {
947 				if (sbappendcontrol_locked(&so2->so_rcv, m,
948 				    control))
949 					control = NULL;
950 			} else
951 				sbappend_locked(&so2->so_rcv, m);
952 			break;
953 
954 		case SOCK_SEQPACKET: {
955 			const struct sockaddr *from;
956 
957 			from = &sun_noname;
958 			if (sbappendaddr_locked(&so2->so_rcv, from, m,
959 			    control))
960 				control = NULL;
961 			break;
962 			}
963 		}
964 
965 		/*
966 		 * XXXRW: While fine for SOCK_STREAM, this conflates maximum
967 		 * datagram size and back-pressure for SOCK_SEQPACKET, which
968 		 * can lead to undesired return of EMSGSIZE on send instead
969 		 * of more desirable blocking.
970 		 */
971 		mbcnt_delta = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt;
972 		unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt;
973 		sbcc = so2->so_rcv.sb_cc;
974 		sorwakeup_locked(so2);
975 
976 		SOCKBUF_LOCK(&so->so_snd);
977 		newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc);
978 		(void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
979 		    newhiwat, RLIM_INFINITY);
980 		so->so_snd.sb_mbmax -= mbcnt_delta;
981 		SOCKBUF_UNLOCK(&so->so_snd);
982 		unp2->unp_cc = sbcc;
983 		UNP_PCB_UNLOCK(unp2);
984 		m = NULL;
985 		break;
986 
987 	default:
988 		panic("uipc_send unknown socktype");
989 	}
990 
991 	/*
992 	 * PRUS_EOF is equivalent to pru_send followed by pru_shutdown.
993 	 */
994 	if (flags & PRUS_EOF) {
995 		UNP_PCB_LOCK(unp);
996 		socantsendmore(so);
997 		unp_shutdown(unp);
998 		UNP_PCB_UNLOCK(unp);
999 	}
1000 
1001 	if ((nam != NULL) || (flags & PRUS_EOF))
1002 		UNP_LINK_WUNLOCK();
1003 	else
1004 		UNP_LINK_RUNLOCK();
1005 
1006 	if (control != NULL && error != 0)
1007 		unp_dispose(control);
1008 
1009 release:
1010 	if (control != NULL)
1011 		m_freem(control);
1012 	if (m != NULL)
1013 		m_freem(m);
1014 	return (error);
1015 }
1016 
1017 static int
1018 uipc_sense(struct socket *so, struct stat *sb)
1019 {
1020 	struct unpcb *unp, *unp2;
1021 	struct socket *so2;
1022 
1023 	unp = sotounpcb(so);
1024 	KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
1025 
1026 	sb->st_blksize = so->so_snd.sb_hiwat;
1027 	UNP_LINK_RLOCK();
1028 	UNP_PCB_LOCK(unp);
1029 	unp2 = unp->unp_conn;
1030 	if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) &&
1031 	    unp2 != NULL) {
1032 		so2 = unp2->unp_socket;
1033 		sb->st_blksize += so2->so_rcv.sb_cc;
1034 	}
1035 	sb->st_dev = NODEV;
1036 	if (unp->unp_ino == 0)
1037 		unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
1038 	sb->st_ino = unp->unp_ino;
1039 	UNP_PCB_UNLOCK(unp);
1040 	UNP_LINK_RUNLOCK();
1041 	return (0);
1042 }
1043 
1044 static int
1045 uipc_shutdown(struct socket *so)
1046 {
1047 	struct unpcb *unp;
1048 
1049 	unp = sotounpcb(so);
1050 	KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
1051 
1052 	UNP_LINK_WLOCK();
1053 	UNP_PCB_LOCK(unp);
1054 	socantsendmore(so);
1055 	unp_shutdown(unp);
1056 	UNP_PCB_UNLOCK(unp);
1057 	UNP_LINK_WUNLOCK();
1058 	return (0);
1059 }
1060 
1061 static int
1062 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
1063 {
1064 	struct unpcb *unp;
1065 	const struct sockaddr *sa;
1066 
1067 	unp = sotounpcb(so);
1068 	KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
1069 
1070 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1071 	UNP_PCB_LOCK(unp);
1072 	if (unp->unp_addr != NULL)
1073 		sa = (struct sockaddr *) unp->unp_addr;
1074 	else
1075 		sa = &sun_noname;
1076 	bcopy(sa, *nam, sa->sa_len);
1077 	UNP_PCB_UNLOCK(unp);
1078 	return (0);
1079 }
1080 
1081 static struct pr_usrreqs uipc_usrreqs_dgram = {
1082 	.pru_abort = 		uipc_abort,
1083 	.pru_accept =		uipc_accept,
1084 	.pru_attach =		uipc_attach,
1085 	.pru_bind =		uipc_bind,
1086 	.pru_connect =		uipc_connect,
1087 	.pru_connect2 =		uipc_connect2,
1088 	.pru_detach =		uipc_detach,
1089 	.pru_disconnect =	uipc_disconnect,
1090 	.pru_listen =		uipc_listen,
1091 	.pru_peeraddr =		uipc_peeraddr,
1092 	.pru_rcvd =		uipc_rcvd,
1093 	.pru_send =		uipc_send,
1094 	.pru_sense =		uipc_sense,
1095 	.pru_shutdown =		uipc_shutdown,
1096 	.pru_sockaddr =		uipc_sockaddr,
1097 	.pru_soreceive =	soreceive_dgram,
1098 	.pru_close =		uipc_close,
1099 };
1100 
1101 static struct pr_usrreqs uipc_usrreqs_seqpacket = {
1102 	.pru_abort =		uipc_abort,
1103 	.pru_accept =		uipc_accept,
1104 	.pru_attach =		uipc_attach,
1105 	.pru_bind =		uipc_bind,
1106 	.pru_connect =		uipc_connect,
1107 	.pru_connect2 =		uipc_connect2,
1108 	.pru_detach =		uipc_detach,
1109 	.pru_disconnect =	uipc_disconnect,
1110 	.pru_listen =		uipc_listen,
1111 	.pru_peeraddr =		uipc_peeraddr,
1112 	.pru_rcvd =		uipc_rcvd,
1113 	.pru_send =		uipc_send,
1114 	.pru_sense =		uipc_sense,
1115 	.pru_shutdown =		uipc_shutdown,
1116 	.pru_sockaddr =		uipc_sockaddr,
1117 	.pru_soreceive =	soreceive_generic,	/* XXX: or...? */
1118 	.pru_close =		uipc_close,
1119 };
1120 
1121 static struct pr_usrreqs uipc_usrreqs_stream = {
1122 	.pru_abort = 		uipc_abort,
1123 	.pru_accept =		uipc_accept,
1124 	.pru_attach =		uipc_attach,
1125 	.pru_bind =		uipc_bind,
1126 	.pru_connect =		uipc_connect,
1127 	.pru_connect2 =		uipc_connect2,
1128 	.pru_detach =		uipc_detach,
1129 	.pru_disconnect =	uipc_disconnect,
1130 	.pru_listen =		uipc_listen,
1131 	.pru_peeraddr =		uipc_peeraddr,
1132 	.pru_rcvd =		uipc_rcvd,
1133 	.pru_send =		uipc_send,
1134 	.pru_sense =		uipc_sense,
1135 	.pru_shutdown =		uipc_shutdown,
1136 	.pru_sockaddr =		uipc_sockaddr,
1137 	.pru_soreceive =	soreceive_generic,
1138 	.pru_close =		uipc_close,
1139 };
1140 
1141 static int
1142 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1143 {
1144 	struct unpcb *unp;
1145 	struct xucred xu;
1146 	int error, optval;
1147 
1148 	if (sopt->sopt_level != 0)
1149 		return (EINVAL);
1150 
1151 	unp = sotounpcb(so);
1152 	KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1153 	error = 0;
1154 	switch (sopt->sopt_dir) {
1155 	case SOPT_GET:
1156 		switch (sopt->sopt_name) {
1157 		case LOCAL_PEERCRED:
1158 			UNP_PCB_LOCK(unp);
1159 			if (unp->unp_flags & UNP_HAVEPC)
1160 				xu = unp->unp_peercred;
1161 			else {
1162 				if (so->so_type == SOCK_STREAM)
1163 					error = ENOTCONN;
1164 				else
1165 					error = EINVAL;
1166 			}
1167 			UNP_PCB_UNLOCK(unp);
1168 			if (error == 0)
1169 				error = sooptcopyout(sopt, &xu, sizeof(xu));
1170 			break;
1171 
1172 		case LOCAL_CREDS:
1173 			/* Unlocked read. */
1174 			optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1175 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1176 			break;
1177 
1178 		case LOCAL_CONNWAIT:
1179 			/* Unlocked read. */
1180 			optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1181 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1182 			break;
1183 
1184 		default:
1185 			error = EOPNOTSUPP;
1186 			break;
1187 		}
1188 		break;
1189 
1190 	case SOPT_SET:
1191 		switch (sopt->sopt_name) {
1192 		case LOCAL_CREDS:
1193 		case LOCAL_CONNWAIT:
1194 			error = sooptcopyin(sopt, &optval, sizeof(optval),
1195 					    sizeof(optval));
1196 			if (error)
1197 				break;
1198 
1199 #define	OPTSET(bit) do {						\
1200 	UNP_PCB_LOCK(unp);						\
1201 	if (optval)							\
1202 		unp->unp_flags |= bit;					\
1203 	else								\
1204 		unp->unp_flags &= ~bit;					\
1205 	UNP_PCB_UNLOCK(unp);						\
1206 } while (0)
1207 
1208 			switch (sopt->sopt_name) {
1209 			case LOCAL_CREDS:
1210 				OPTSET(UNP_WANTCRED);
1211 				break;
1212 
1213 			case LOCAL_CONNWAIT:
1214 				OPTSET(UNP_CONNWAIT);
1215 				break;
1216 
1217 			default:
1218 				break;
1219 			}
1220 			break;
1221 #undef	OPTSET
1222 		default:
1223 			error = ENOPROTOOPT;
1224 			break;
1225 		}
1226 		break;
1227 
1228 	default:
1229 		error = EOPNOTSUPP;
1230 		break;
1231 	}
1232 	return (error);
1233 }
1234 
1235 static int
1236 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1237 {
1238 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1239 	struct vnode *vp;
1240 	struct socket *so2, *so3;
1241 	struct unpcb *unp, *unp2, *unp3;
1242 	int error, len, vfslocked;
1243 	struct nameidata nd;
1244 	char buf[SOCK_MAXADDRLEN];
1245 	struct sockaddr *sa;
1246 
1247 	UNP_LINK_WLOCK_ASSERT();
1248 
1249 	unp = sotounpcb(so);
1250 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1251 
1252 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1253 	if (len <= 0)
1254 		return (EINVAL);
1255 	bcopy(soun->sun_path, buf, len);
1256 	buf[len] = 0;
1257 
1258 	UNP_PCB_LOCK(unp);
1259 	if (unp->unp_flags & UNP_CONNECTING) {
1260 		UNP_PCB_UNLOCK(unp);
1261 		return (EALREADY);
1262 	}
1263 	UNP_LINK_WUNLOCK();
1264 	unp->unp_flags |= UNP_CONNECTING;
1265 	UNP_PCB_UNLOCK(unp);
1266 
1267 	sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1268 	NDINIT(&nd, LOOKUP, MPSAFE | FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf,
1269 	    td);
1270 	error = namei(&nd);
1271 	if (error)
1272 		vp = NULL;
1273 	else
1274 		vp = nd.ni_vp;
1275 	ASSERT_VOP_LOCKED(vp, "unp_connect");
1276 	vfslocked = NDHASGIANT(&nd);
1277 	NDFREE(&nd, NDF_ONLY_PNBUF);
1278 	if (error)
1279 		goto bad;
1280 
1281 	if (vp->v_type != VSOCK) {
1282 		error = ENOTSOCK;
1283 		goto bad;
1284 	}
1285 #ifdef MAC
1286 	error = mac_vnode_check_open(td->td_ucred, vp, VWRITE | VREAD);
1287 	if (error)
1288 		goto bad;
1289 #endif
1290 	error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1291 	if (error)
1292 		goto bad;
1293 	VFS_UNLOCK_GIANT(vfslocked);
1294 
1295 	unp = sotounpcb(so);
1296 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1297 
1298 	/*
1299 	 * Lock linkage lock for two reasons: make sure v_socket is stable,
1300 	 * and to protect simultaneous locking of multiple pcbs.
1301 	 */
1302 	UNP_LINK_WLOCK();
1303 	so2 = vp->v_socket;
1304 	if (so2 == NULL) {
1305 		error = ECONNREFUSED;
1306 		goto bad2;
1307 	}
1308 	if (so->so_type != so2->so_type) {
1309 		error = EPROTOTYPE;
1310 		goto bad2;
1311 	}
1312 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1313 		if (so2->so_options & SO_ACCEPTCONN) {
1314 			CURVNET_SET(so2->so_vnet);
1315 			so3 = sonewconn(so2, 0);
1316 			CURVNET_RESTORE();
1317 		} else
1318 			so3 = NULL;
1319 		if (so3 == NULL) {
1320 			error = ECONNREFUSED;
1321 			goto bad2;
1322 		}
1323 		unp = sotounpcb(so);
1324 		unp2 = sotounpcb(so2);
1325 		unp3 = sotounpcb(so3);
1326 		UNP_PCB_LOCK(unp);
1327 		UNP_PCB_LOCK(unp2);
1328 		UNP_PCB_LOCK(unp3);
1329 		if (unp2->unp_addr != NULL) {
1330 			bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1331 			unp3->unp_addr = (struct sockaddr_un *) sa;
1332 			sa = NULL;
1333 		}
1334 
1335 		/*
1336 		 * The connecter's (client's) credentials are copied from its
1337 		 * process structure at the time of connect() (which is now).
1338 		 */
1339 		cru2x(td->td_ucred, &unp3->unp_peercred);
1340 		unp3->unp_flags |= UNP_HAVEPC;
1341 
1342 		/*
1343 		 * The receiver's (server's) credentials are copied from the
1344 		 * unp_peercred member of socket on which the former called
1345 		 * listen(); uipc_listen() cached that process's credentials
1346 		 * at that time so we can use them now.
1347 		 */
1348 		KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1349 		    ("unp_connect: listener without cached peercred"));
1350 		memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1351 		    sizeof(unp->unp_peercred));
1352 		unp->unp_flags |= UNP_HAVEPC;
1353 		if (unp2->unp_flags & UNP_WANTCRED)
1354 			unp3->unp_flags |= UNP_WANTCRED;
1355 		UNP_PCB_UNLOCK(unp3);
1356 		UNP_PCB_UNLOCK(unp2);
1357 		UNP_PCB_UNLOCK(unp);
1358 #ifdef MAC
1359 		mac_socketpeer_set_from_socket(so, so3);
1360 		mac_socketpeer_set_from_socket(so3, so);
1361 #endif
1362 
1363 		so2 = so3;
1364 	}
1365 	unp = sotounpcb(so);
1366 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1367 	unp2 = sotounpcb(so2);
1368 	KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1369 	UNP_PCB_LOCK(unp);
1370 	UNP_PCB_LOCK(unp2);
1371 	error = unp_connect2(so, so2, PRU_CONNECT);
1372 	UNP_PCB_UNLOCK(unp2);
1373 	UNP_PCB_UNLOCK(unp);
1374 bad2:
1375 	UNP_LINK_WUNLOCK();
1376 	if (vfslocked)
1377 		/*
1378 		 * Giant has been previously acquired. This means filesystem
1379 		 * isn't MPSAFE.  Do it once again.
1380 		 */
1381 		mtx_lock(&Giant);
1382 bad:
1383 	if (vp != NULL)
1384 		vput(vp);
1385 	VFS_UNLOCK_GIANT(vfslocked);
1386 	free(sa, M_SONAME);
1387 	UNP_LINK_WLOCK();
1388 	UNP_PCB_LOCK(unp);
1389 	unp->unp_flags &= ~UNP_CONNECTING;
1390 	UNP_PCB_UNLOCK(unp);
1391 	return (error);
1392 }
1393 
1394 static int
1395 unp_connect2(struct socket *so, struct socket *so2, int req)
1396 {
1397 	struct unpcb *unp;
1398 	struct unpcb *unp2;
1399 
1400 	unp = sotounpcb(so);
1401 	KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1402 	unp2 = sotounpcb(so2);
1403 	KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1404 
1405 	UNP_LINK_WLOCK_ASSERT();
1406 	UNP_PCB_LOCK_ASSERT(unp);
1407 	UNP_PCB_LOCK_ASSERT(unp2);
1408 
1409 	if (so2->so_type != so->so_type)
1410 		return (EPROTOTYPE);
1411 	unp->unp_conn = unp2;
1412 
1413 	switch (so->so_type) {
1414 	case SOCK_DGRAM:
1415 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1416 		soisconnected(so);
1417 		break;
1418 
1419 	case SOCK_STREAM:
1420 	case SOCK_SEQPACKET:
1421 		unp2->unp_conn = unp;
1422 		if (req == PRU_CONNECT &&
1423 		    ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1424 			soisconnecting(so);
1425 		else
1426 			soisconnected(so);
1427 		soisconnected(so2);
1428 		break;
1429 
1430 	default:
1431 		panic("unp_connect2");
1432 	}
1433 	return (0);
1434 }
1435 
1436 static void
1437 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1438 {
1439 	struct socket *so;
1440 
1441 	KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1442 
1443 	UNP_LINK_WLOCK_ASSERT();
1444 	UNP_PCB_LOCK_ASSERT(unp);
1445 	UNP_PCB_LOCK_ASSERT(unp2);
1446 
1447 	unp->unp_conn = NULL;
1448 	switch (unp->unp_socket->so_type) {
1449 	case SOCK_DGRAM:
1450 		LIST_REMOVE(unp, unp_reflink);
1451 		so = unp->unp_socket;
1452 		SOCK_LOCK(so);
1453 		so->so_state &= ~SS_ISCONNECTED;
1454 		SOCK_UNLOCK(so);
1455 		break;
1456 
1457 	case SOCK_STREAM:
1458 	case SOCK_SEQPACKET:
1459 		soisdisconnected(unp->unp_socket);
1460 		unp2->unp_conn = NULL;
1461 		soisdisconnected(unp2->unp_socket);
1462 		break;
1463 	}
1464 }
1465 
1466 /*
1467  * unp_pcblist() walks the global list of struct unpcb's to generate a
1468  * pointer list, bumping the refcount on each unpcb.  It then copies them out
1469  * sequentially, validating the generation number on each to see if it has
1470  * been detached.  All of this is necessary because copyout() may sleep on
1471  * disk I/O.
1472  */
1473 static int
1474 unp_pcblist(SYSCTL_HANDLER_ARGS)
1475 {
1476 	int error, i, n;
1477 	int freeunp;
1478 	struct unpcb *unp, **unp_list;
1479 	unp_gen_t gencnt;
1480 	struct xunpgen *xug;
1481 	struct unp_head *head;
1482 	struct xunpcb *xu;
1483 
1484 	switch ((intptr_t)arg1) {
1485 	case SOCK_STREAM:
1486 		head = &unp_shead;
1487 		break;
1488 
1489 	case SOCK_DGRAM:
1490 		head = &unp_dhead;
1491 		break;
1492 
1493 	case SOCK_SEQPACKET:
1494 		head = &unp_sphead;
1495 		break;
1496 
1497 	default:
1498 		panic("unp_pcblist: arg1 %d", (int)(intptr_t)arg1);
1499 	}
1500 
1501 	/*
1502 	 * The process of preparing the PCB list is too time-consuming and
1503 	 * resource-intensive to repeat twice on every request.
1504 	 */
1505 	if (req->oldptr == NULL) {
1506 		n = unp_count;
1507 		req->oldidx = 2 * (sizeof *xug)
1508 			+ (n + n/8) * sizeof(struct xunpcb);
1509 		return (0);
1510 	}
1511 
1512 	if (req->newptr != NULL)
1513 		return (EPERM);
1514 
1515 	/*
1516 	 * OK, now we're committed to doing something.
1517 	 */
1518 	xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1519 	UNP_LIST_LOCK();
1520 	gencnt = unp_gencnt;
1521 	n = unp_count;
1522 	UNP_LIST_UNLOCK();
1523 
1524 	xug->xug_len = sizeof *xug;
1525 	xug->xug_count = n;
1526 	xug->xug_gen = gencnt;
1527 	xug->xug_sogen = so_gencnt;
1528 	error = SYSCTL_OUT(req, xug, sizeof *xug);
1529 	if (error) {
1530 		free(xug, M_TEMP);
1531 		return (error);
1532 	}
1533 
1534 	unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1535 
1536 	UNP_LIST_LOCK();
1537 	for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1538 	     unp = LIST_NEXT(unp, unp_link)) {
1539 		UNP_PCB_LOCK(unp);
1540 		if (unp->unp_gencnt <= gencnt) {
1541 			if (cr_cansee(req->td->td_ucred,
1542 			    unp->unp_socket->so_cred)) {
1543 				UNP_PCB_UNLOCK(unp);
1544 				continue;
1545 			}
1546 			unp_list[i++] = unp;
1547 			unp->unp_refcount++;
1548 		}
1549 		UNP_PCB_UNLOCK(unp);
1550 	}
1551 	UNP_LIST_UNLOCK();
1552 	n = i;			/* In case we lost some during malloc. */
1553 
1554 	error = 0;
1555 	xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1556 	for (i = 0; i < n; i++) {
1557 		unp = unp_list[i];
1558 		UNP_PCB_LOCK(unp);
1559 		unp->unp_refcount--;
1560 	        if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1561 			xu->xu_len = sizeof *xu;
1562 			xu->xu_unpp = unp;
1563 			/*
1564 			 * XXX - need more locking here to protect against
1565 			 * connect/disconnect races for SMP.
1566 			 */
1567 			if (unp->unp_addr != NULL)
1568 				bcopy(unp->unp_addr, &xu->xu_addr,
1569 				      unp->unp_addr->sun_len);
1570 			if (unp->unp_conn != NULL &&
1571 			    unp->unp_conn->unp_addr != NULL)
1572 				bcopy(unp->unp_conn->unp_addr,
1573 				      &xu->xu_caddr,
1574 				      unp->unp_conn->unp_addr->sun_len);
1575 			bcopy(unp, &xu->xu_unp, sizeof *unp);
1576 			sotoxsocket(unp->unp_socket, &xu->xu_socket);
1577 			UNP_PCB_UNLOCK(unp);
1578 			error = SYSCTL_OUT(req, xu, sizeof *xu);
1579 		} else {
1580 			freeunp = (unp->unp_refcount == 0);
1581 			UNP_PCB_UNLOCK(unp);
1582 			if (freeunp) {
1583 				UNP_PCB_LOCK_DESTROY(unp);
1584 				uma_zfree(unp_zone, unp);
1585 			}
1586 		}
1587 	}
1588 	free(xu, M_TEMP);
1589 	if (!error) {
1590 		/*
1591 		 * Give the user an updated idea of our state.  If the
1592 		 * generation differs from what we told her before, she knows
1593 		 * that something happened while we were processing this
1594 		 * request, and it might be necessary to retry.
1595 		 */
1596 		xug->xug_gen = unp_gencnt;
1597 		xug->xug_sogen = so_gencnt;
1598 		xug->xug_count = unp_count;
1599 		error = SYSCTL_OUT(req, xug, sizeof *xug);
1600 	}
1601 	free(unp_list, M_TEMP);
1602 	free(xug, M_TEMP);
1603 	return (error);
1604 }
1605 
1606 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1607     (void *)(intptr_t)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1608     "List of active local datagram sockets");
1609 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLTYPE_OPAQUE | CTLFLAG_RD,
1610     (void *)(intptr_t)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1611     "List of active local stream sockets");
1612 SYSCTL_PROC(_net_local_seqpacket, OID_AUTO, pcblist,
1613     CTLTYPE_OPAQUE | CTLFLAG_RD,
1614     (void *)(intptr_t)SOCK_SEQPACKET, 0, unp_pcblist, "S,xunpcb",
1615     "List of active local seqpacket sockets");
1616 
1617 static void
1618 unp_shutdown(struct unpcb *unp)
1619 {
1620 	struct unpcb *unp2;
1621 	struct socket *so;
1622 
1623 	UNP_LINK_WLOCK_ASSERT();
1624 	UNP_PCB_LOCK_ASSERT(unp);
1625 
1626 	unp2 = unp->unp_conn;
1627 	if ((unp->unp_socket->so_type == SOCK_STREAM ||
1628 	    (unp->unp_socket->so_type == SOCK_SEQPACKET)) && unp2 != NULL) {
1629 		so = unp2->unp_socket;
1630 		if (so != NULL)
1631 			socantrcvmore(so);
1632 	}
1633 }
1634 
1635 static void
1636 unp_drop(struct unpcb *unp, int errno)
1637 {
1638 	struct socket *so = unp->unp_socket;
1639 	struct unpcb *unp2;
1640 
1641 	UNP_LINK_WLOCK_ASSERT();
1642 	UNP_PCB_LOCK_ASSERT(unp);
1643 
1644 	so->so_error = errno;
1645 	unp2 = unp->unp_conn;
1646 	if (unp2 == NULL)
1647 		return;
1648 	UNP_PCB_LOCK(unp2);
1649 	unp_disconnect(unp, unp2);
1650 	UNP_PCB_UNLOCK(unp2);
1651 }
1652 
1653 static void
1654 unp_freerights(struct file **rp, int fdcount)
1655 {
1656 	int i;
1657 	struct file *fp;
1658 
1659 	for (i = 0; i < fdcount; i++) {
1660 		fp = *rp;
1661 		*rp++ = NULL;
1662 		unp_discard(fp);
1663 	}
1664 }
1665 
1666 static int
1667 unp_externalize(struct mbuf *control, struct mbuf **controlp)
1668 {
1669 	struct thread *td = curthread;		/* XXX */
1670 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1671 	int i;
1672 	int *fdp;
1673 	struct file **rp;
1674 	struct file *fp;
1675 	void *data;
1676 	socklen_t clen = control->m_len, datalen;
1677 	int error, newfds;
1678 	int f;
1679 	u_int newlen;
1680 
1681 	UNP_LINK_UNLOCK_ASSERT();
1682 
1683 	error = 0;
1684 	if (controlp != NULL) /* controlp == NULL => free control messages */
1685 		*controlp = NULL;
1686 	while (cm != NULL) {
1687 		if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1688 			error = EINVAL;
1689 			break;
1690 		}
1691 		data = CMSG_DATA(cm);
1692 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1693 		if (cm->cmsg_level == SOL_SOCKET
1694 		    && cm->cmsg_type == SCM_RIGHTS) {
1695 			newfds = datalen / sizeof(struct file *);
1696 			rp = data;
1697 
1698 			/* If we're not outputting the descriptors free them. */
1699 			if (error || controlp == NULL) {
1700 				unp_freerights(rp, newfds);
1701 				goto next;
1702 			}
1703 			FILEDESC_XLOCK(td->td_proc->p_fd);
1704 			/* if the new FD's will not fit free them.  */
1705 			if (!fdavail(td, newfds)) {
1706 				FILEDESC_XUNLOCK(td->td_proc->p_fd);
1707 				error = EMSGSIZE;
1708 				unp_freerights(rp, newfds);
1709 				goto next;
1710 			}
1711 
1712 			/*
1713 			 * Now change each pointer to an fd in the global
1714 			 * table to an integer that is the index to the local
1715 			 * fd table entry that we set up to point to the
1716 			 * global one we are transferring.
1717 			 */
1718 			newlen = newfds * sizeof(int);
1719 			*controlp = sbcreatecontrol(NULL, newlen,
1720 			    SCM_RIGHTS, SOL_SOCKET);
1721 			if (*controlp == NULL) {
1722 				FILEDESC_XUNLOCK(td->td_proc->p_fd);
1723 				error = E2BIG;
1724 				unp_freerights(rp, newfds);
1725 				goto next;
1726 			}
1727 
1728 			fdp = (int *)
1729 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1730 			for (i = 0; i < newfds; i++) {
1731 				if (fdalloc(td, 0, &f))
1732 					panic("unp_externalize fdalloc failed");
1733 				fp = *rp++;
1734 				td->td_proc->p_fd->fd_ofiles[f] = fp;
1735 				unp_externalize_fp(fp);
1736 				*fdp++ = f;
1737 			}
1738 			FILEDESC_XUNLOCK(td->td_proc->p_fd);
1739 		} else {
1740 			/* We can just copy anything else across. */
1741 			if (error || controlp == NULL)
1742 				goto next;
1743 			*controlp = sbcreatecontrol(NULL, datalen,
1744 			    cm->cmsg_type, cm->cmsg_level);
1745 			if (*controlp == NULL) {
1746 				error = ENOBUFS;
1747 				goto next;
1748 			}
1749 			bcopy(data,
1750 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1751 			    datalen);
1752 		}
1753 		controlp = &(*controlp)->m_next;
1754 
1755 next:
1756 		if (CMSG_SPACE(datalen) < clen) {
1757 			clen -= CMSG_SPACE(datalen);
1758 			cm = (struct cmsghdr *)
1759 			    ((caddr_t)cm + CMSG_SPACE(datalen));
1760 		} else {
1761 			clen = 0;
1762 			cm = NULL;
1763 		}
1764 	}
1765 
1766 	m_freem(control);
1767 	return (error);
1768 }
1769 
1770 static void
1771 unp_zone_change(void *tag)
1772 {
1773 
1774 	uma_zone_set_max(unp_zone, maxsockets);
1775 }
1776 
1777 static void
1778 unp_init(void)
1779 {
1780 
1781 #ifdef VIMAGE
1782 	if (!IS_DEFAULT_VNET(curvnet))
1783 		return;
1784 #endif
1785 	unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1786 	    NULL, NULL, UMA_ALIGN_PTR, 0);
1787 	if (unp_zone == NULL)
1788 		panic("unp_init");
1789 	uma_zone_set_max(unp_zone, maxsockets);
1790 	EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1791 	    NULL, EVENTHANDLER_PRI_ANY);
1792 	LIST_INIT(&unp_dhead);
1793 	LIST_INIT(&unp_shead);
1794 	LIST_INIT(&unp_sphead);
1795 	SLIST_INIT(&unp_defers);
1796 	TASK_INIT(&unp_gc_task, 0, unp_gc, NULL);
1797 	TASK_INIT(&unp_defer_task, 0, unp_process_defers, NULL);
1798 	UNP_LINK_LOCK_INIT();
1799 	UNP_LIST_LOCK_INIT();
1800 	UNP_DEFERRED_LOCK_INIT();
1801 }
1802 
1803 static int
1804 unp_internalize(struct mbuf **controlp, struct thread *td)
1805 {
1806 	struct mbuf *control = *controlp;
1807 	struct proc *p = td->td_proc;
1808 	struct filedesc *fdescp = p->p_fd;
1809 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1810 	struct cmsgcred *cmcred;
1811 	struct file **rp;
1812 	struct file *fp;
1813 	struct timeval *tv;
1814 	int i, fd, *fdp;
1815 	void *data;
1816 	socklen_t clen = control->m_len, datalen;
1817 	int error, oldfds;
1818 	u_int newlen;
1819 
1820 	UNP_LINK_UNLOCK_ASSERT();
1821 
1822 	error = 0;
1823 	*controlp = NULL;
1824 	while (cm != NULL) {
1825 		if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1826 		    || cm->cmsg_len > clen) {
1827 			error = EINVAL;
1828 			goto out;
1829 		}
1830 		data = CMSG_DATA(cm);
1831 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1832 
1833 		switch (cm->cmsg_type) {
1834 		/*
1835 		 * Fill in credential information.
1836 		 */
1837 		case SCM_CREDS:
1838 			*controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1839 			    SCM_CREDS, SOL_SOCKET);
1840 			if (*controlp == NULL) {
1841 				error = ENOBUFS;
1842 				goto out;
1843 			}
1844 			cmcred = (struct cmsgcred *)
1845 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1846 			cmcred->cmcred_pid = p->p_pid;
1847 			cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1848 			cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1849 			cmcred->cmcred_euid = td->td_ucred->cr_uid;
1850 			cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1851 			    CMGROUP_MAX);
1852 			for (i = 0; i < cmcred->cmcred_ngroups; i++)
1853 				cmcred->cmcred_groups[i] =
1854 				    td->td_ucred->cr_groups[i];
1855 			break;
1856 
1857 		case SCM_RIGHTS:
1858 			oldfds = datalen / sizeof (int);
1859 			/*
1860 			 * Check that all the FDs passed in refer to legal
1861 			 * files.  If not, reject the entire operation.
1862 			 */
1863 			fdp = data;
1864 			FILEDESC_SLOCK(fdescp);
1865 			for (i = 0; i < oldfds; i++) {
1866 				fd = *fdp++;
1867 				if ((unsigned)fd >= fdescp->fd_nfiles ||
1868 				    fdescp->fd_ofiles[fd] == NULL) {
1869 					FILEDESC_SUNLOCK(fdescp);
1870 					error = EBADF;
1871 					goto out;
1872 				}
1873 				fp = fdescp->fd_ofiles[fd];
1874 				if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1875 					FILEDESC_SUNLOCK(fdescp);
1876 					error = EOPNOTSUPP;
1877 					goto out;
1878 				}
1879 
1880 			}
1881 
1882 			/*
1883 			 * Now replace the integer FDs with pointers to the
1884 			 * associated global file table entry..
1885 			 */
1886 			newlen = oldfds * sizeof(struct file *);
1887 			*controlp = sbcreatecontrol(NULL, newlen,
1888 			    SCM_RIGHTS, SOL_SOCKET);
1889 			if (*controlp == NULL) {
1890 				FILEDESC_SUNLOCK(fdescp);
1891 				error = E2BIG;
1892 				goto out;
1893 			}
1894 			fdp = data;
1895 			rp = (struct file **)
1896 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1897 			for (i = 0; i < oldfds; i++) {
1898 				fp = fdescp->fd_ofiles[*fdp++];
1899 				*rp++ = fp;
1900 				unp_internalize_fp(fp);
1901 			}
1902 			FILEDESC_SUNLOCK(fdescp);
1903 			break;
1904 
1905 		case SCM_TIMESTAMP:
1906 			*controlp = sbcreatecontrol(NULL, sizeof(*tv),
1907 			    SCM_TIMESTAMP, SOL_SOCKET);
1908 			if (*controlp == NULL) {
1909 				error = ENOBUFS;
1910 				goto out;
1911 			}
1912 			tv = (struct timeval *)
1913 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1914 			microtime(tv);
1915 			break;
1916 
1917 		default:
1918 			error = EINVAL;
1919 			goto out;
1920 		}
1921 
1922 		controlp = &(*controlp)->m_next;
1923 		if (CMSG_SPACE(datalen) < clen) {
1924 			clen -= CMSG_SPACE(datalen);
1925 			cm = (struct cmsghdr *)
1926 			    ((caddr_t)cm + CMSG_SPACE(datalen));
1927 		} else {
1928 			clen = 0;
1929 			cm = NULL;
1930 		}
1931 	}
1932 
1933 out:
1934 	m_freem(control);
1935 	return (error);
1936 }
1937 
1938 static struct mbuf *
1939 unp_addsockcred(struct thread *td, struct mbuf *control)
1940 {
1941 	struct mbuf *m, *n, *n_prev;
1942 	struct sockcred *sc;
1943 	const struct cmsghdr *cm;
1944 	int ngroups;
1945 	int i;
1946 
1947 	ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1948 	m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
1949 	if (m == NULL)
1950 		return (control);
1951 
1952 	sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
1953 	sc->sc_uid = td->td_ucred->cr_ruid;
1954 	sc->sc_euid = td->td_ucred->cr_uid;
1955 	sc->sc_gid = td->td_ucred->cr_rgid;
1956 	sc->sc_egid = td->td_ucred->cr_gid;
1957 	sc->sc_ngroups = ngroups;
1958 	for (i = 0; i < sc->sc_ngroups; i++)
1959 		sc->sc_groups[i] = td->td_ucred->cr_groups[i];
1960 
1961 	/*
1962 	 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
1963 	 * created SCM_CREDS control message (struct sockcred) has another
1964 	 * format.
1965 	 */
1966 	if (control != NULL)
1967 		for (n = control, n_prev = NULL; n != NULL;) {
1968 			cm = mtod(n, struct cmsghdr *);
1969     			if (cm->cmsg_level == SOL_SOCKET &&
1970 			    cm->cmsg_type == SCM_CREDS) {
1971     				if (n_prev == NULL)
1972 					control = n->m_next;
1973 				else
1974 					n_prev->m_next = n->m_next;
1975 				n = m_free(n);
1976 			} else {
1977 				n_prev = n;
1978 				n = n->m_next;
1979 			}
1980 		}
1981 
1982 	/* Prepend it to the head. */
1983 	m->m_next = control;
1984 	return (m);
1985 }
1986 
1987 static struct unpcb *
1988 fptounp(struct file *fp)
1989 {
1990 	struct socket *so;
1991 
1992 	if (fp->f_type != DTYPE_SOCKET)
1993 		return (NULL);
1994 	if ((so = fp->f_data) == NULL)
1995 		return (NULL);
1996 	if (so->so_proto->pr_domain != &localdomain)
1997 		return (NULL);
1998 	return sotounpcb(so);
1999 }
2000 
2001 static void
2002 unp_discard(struct file *fp)
2003 {
2004 	struct unp_defer *dr;
2005 
2006 	if (unp_externalize_fp(fp)) {
2007 		dr = malloc(sizeof(*dr), M_TEMP, M_WAITOK);
2008 		dr->ud_fp = fp;
2009 		UNP_DEFERRED_LOCK();
2010 		SLIST_INSERT_HEAD(&unp_defers, dr, ud_link);
2011 		UNP_DEFERRED_UNLOCK();
2012 		atomic_add_int(&unp_defers_count, 1);
2013 		taskqueue_enqueue(taskqueue_thread, &unp_defer_task);
2014 	} else
2015 		(void) closef(fp, (struct thread *)NULL);
2016 }
2017 
2018 static void
2019 unp_process_defers(void *arg __unused, int pending)
2020 {
2021 	struct unp_defer *dr;
2022 	SLIST_HEAD(, unp_defer) drl;
2023 	int count;
2024 
2025 	SLIST_INIT(&drl);
2026 	for (;;) {
2027 		UNP_DEFERRED_LOCK();
2028 		if (SLIST_FIRST(&unp_defers) == NULL) {
2029 			UNP_DEFERRED_UNLOCK();
2030 			break;
2031 		}
2032 		SLIST_SWAP(&unp_defers, &drl, unp_defer);
2033 		UNP_DEFERRED_UNLOCK();
2034 		count = 0;
2035 		while ((dr = SLIST_FIRST(&drl)) != NULL) {
2036 			SLIST_REMOVE_HEAD(&drl, ud_link);
2037 			closef(dr->ud_fp, NULL);
2038 			free(dr, M_TEMP);
2039 			count++;
2040 		}
2041 		atomic_add_int(&unp_defers_count, -count);
2042 	}
2043 }
2044 
2045 static void
2046 unp_internalize_fp(struct file *fp)
2047 {
2048 	struct unpcb *unp;
2049 
2050 	UNP_LINK_WLOCK();
2051 	if ((unp = fptounp(fp)) != NULL) {
2052 		unp->unp_file = fp;
2053 		unp->unp_msgcount++;
2054 	}
2055 	fhold(fp);
2056 	unp_rights++;
2057 	UNP_LINK_WUNLOCK();
2058 }
2059 
2060 static int
2061 unp_externalize_fp(struct file *fp)
2062 {
2063 	struct unpcb *unp;
2064 	int ret;
2065 
2066 	UNP_LINK_WLOCK();
2067 	if ((unp = fptounp(fp)) != NULL) {
2068 		unp->unp_msgcount--;
2069 		ret = 1;
2070 	} else
2071 		ret = 0;
2072 	unp_rights--;
2073 	UNP_LINK_WUNLOCK();
2074 	return (ret);
2075 }
2076 
2077 /*
2078  * unp_defer indicates whether additional work has been defered for a future
2079  * pass through unp_gc().  It is thread local and does not require explicit
2080  * synchronization.
2081  */
2082 static int	unp_marked;
2083 static int	unp_unreachable;
2084 
2085 static void
2086 unp_accessable(struct file *fp)
2087 {
2088 	struct unpcb *unp;
2089 
2090 	if ((unp = fptounp(fp)) == NULL)
2091 		return;
2092 	if (unp->unp_gcflag & UNPGC_REF)
2093 		return;
2094 	unp->unp_gcflag &= ~UNPGC_DEAD;
2095 	unp->unp_gcflag |= UNPGC_REF;
2096 	unp_marked++;
2097 }
2098 
2099 static void
2100 unp_gc_process(struct unpcb *unp)
2101 {
2102 	struct socket *soa;
2103 	struct socket *so;
2104 	struct file *fp;
2105 
2106 	/* Already processed. */
2107 	if (unp->unp_gcflag & UNPGC_SCANNED)
2108 		return;
2109 	fp = unp->unp_file;
2110 
2111 	/*
2112 	 * Check for a socket potentially in a cycle.  It must be in a
2113 	 * queue as indicated by msgcount, and this must equal the file
2114 	 * reference count.  Note that when msgcount is 0 the file is NULL.
2115 	 */
2116 	if ((unp->unp_gcflag & UNPGC_REF) == 0 && fp &&
2117 	    unp->unp_msgcount != 0 && fp->f_count == unp->unp_msgcount) {
2118 		unp->unp_gcflag |= UNPGC_DEAD;
2119 		unp_unreachable++;
2120 		return;
2121 	}
2122 
2123 	/*
2124 	 * Mark all sockets we reference with RIGHTS.
2125 	 */
2126 	so = unp->unp_socket;
2127 	SOCKBUF_LOCK(&so->so_rcv);
2128 	unp_scan(so->so_rcv.sb_mb, unp_accessable);
2129 	SOCKBUF_UNLOCK(&so->so_rcv);
2130 
2131 	/*
2132 	 * Mark all sockets in our accept queue.
2133 	 */
2134 	ACCEPT_LOCK();
2135 	TAILQ_FOREACH(soa, &so->so_comp, so_list) {
2136 		SOCKBUF_LOCK(&soa->so_rcv);
2137 		unp_scan(soa->so_rcv.sb_mb, unp_accessable);
2138 		SOCKBUF_UNLOCK(&soa->so_rcv);
2139 	}
2140 	ACCEPT_UNLOCK();
2141 	unp->unp_gcflag |= UNPGC_SCANNED;
2142 }
2143 
2144 static int unp_recycled;
2145 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0,
2146     "Number of unreachable sockets claimed by the garbage collector.");
2147 
2148 static int unp_taskcount;
2149 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0,
2150     "Number of times the garbage collector has run.");
2151 
2152 static void
2153 unp_gc(__unused void *arg, int pending)
2154 {
2155 	struct unp_head *heads[] = { &unp_dhead, &unp_shead, &unp_sphead,
2156 				    NULL };
2157 	struct unp_head **head;
2158 	struct file *f, **unref;
2159 	struct unpcb *unp;
2160 	int i, total;
2161 
2162 	unp_taskcount++;
2163 	UNP_LIST_LOCK();
2164 	/*
2165 	 * First clear all gc flags from previous runs.
2166 	 */
2167 	for (head = heads; *head != NULL; head++)
2168 		LIST_FOREACH(unp, *head, unp_link)
2169 			unp->unp_gcflag = 0;
2170 
2171 	/*
2172 	 * Scan marking all reachable sockets with UNPGC_REF.  Once a socket
2173 	 * is reachable all of the sockets it references are reachable.
2174 	 * Stop the scan once we do a complete loop without discovering
2175 	 * a new reachable socket.
2176 	 */
2177 	do {
2178 		unp_unreachable = 0;
2179 		unp_marked = 0;
2180 		for (head = heads; *head != NULL; head++)
2181 			LIST_FOREACH(unp, *head, unp_link)
2182 				unp_gc_process(unp);
2183 	} while (unp_marked);
2184 	UNP_LIST_UNLOCK();
2185 	if (unp_unreachable == 0)
2186 		return;
2187 
2188 	/*
2189 	 * Allocate space for a local list of dead unpcbs.
2190 	 */
2191 	unref = malloc(unp_unreachable * sizeof(struct file *),
2192 	    M_TEMP, M_WAITOK);
2193 
2194 	/*
2195 	 * Iterate looking for sockets which have been specifically marked
2196 	 * as as unreachable and store them locally.
2197 	 */
2198 	UNP_LINK_RLOCK();
2199 	UNP_LIST_LOCK();
2200 	for (total = 0, head = heads; *head != NULL; head++)
2201 		LIST_FOREACH(unp, *head, unp_link)
2202 			if ((unp->unp_gcflag & UNPGC_DEAD) != 0) {
2203 				f = unp->unp_file;
2204 				if (unp->unp_msgcount == 0 || f == NULL ||
2205 				    f->f_count != unp->unp_msgcount)
2206 					continue;
2207 				unref[total++] = f;
2208 				fhold(f);
2209 				KASSERT(total <= unp_unreachable,
2210 				    ("unp_gc: incorrect unreachable count."));
2211 			}
2212 	UNP_LIST_UNLOCK();
2213 	UNP_LINK_RUNLOCK();
2214 
2215 	/*
2216 	 * Now flush all sockets, free'ing rights.  This will free the
2217 	 * struct files associated with these sockets but leave each socket
2218 	 * with one remaining ref.
2219 	 */
2220 	for (i = 0; i < total; i++) {
2221 		struct socket *so;
2222 
2223 		so = unref[i]->f_data;
2224 		CURVNET_SET(so->so_vnet);
2225 		sorflush(so);
2226 		CURVNET_RESTORE();
2227 	}
2228 
2229 	/*
2230 	 * And finally release the sockets so they can be reclaimed.
2231 	 */
2232 	for (i = 0; i < total; i++)
2233 		fdrop(unref[i], NULL);
2234 	unp_recycled += total;
2235 	free(unref, M_TEMP);
2236 }
2237 
2238 static void
2239 unp_dispose(struct mbuf *m)
2240 {
2241 
2242 	if (m)
2243 		unp_scan(m, unp_discard);
2244 }
2245 
2246 static void
2247 unp_scan(struct mbuf *m0, void (*op)(struct file *))
2248 {
2249 	struct mbuf *m;
2250 	struct file **rp;
2251 	struct cmsghdr *cm;
2252 	void *data;
2253 	int i;
2254 	socklen_t clen, datalen;
2255 	int qfds;
2256 
2257 	while (m0 != NULL) {
2258 		for (m = m0; m; m = m->m_next) {
2259 			if (m->m_type != MT_CONTROL)
2260 				continue;
2261 
2262 			cm = mtod(m, struct cmsghdr *);
2263 			clen = m->m_len;
2264 
2265 			while (cm != NULL) {
2266 				if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2267 					break;
2268 
2269 				data = CMSG_DATA(cm);
2270 				datalen = (caddr_t)cm + cm->cmsg_len
2271 				    - (caddr_t)data;
2272 
2273 				if (cm->cmsg_level == SOL_SOCKET &&
2274 				    cm->cmsg_type == SCM_RIGHTS) {
2275 					qfds = datalen / sizeof (struct file *);
2276 					rp = data;
2277 					for (i = 0; i < qfds; i++)
2278 						(*op)(*rp++);
2279 				}
2280 
2281 				if (CMSG_SPACE(datalen) < clen) {
2282 					clen -= CMSG_SPACE(datalen);
2283 					cm = (struct cmsghdr *)
2284 					    ((caddr_t)cm + CMSG_SPACE(datalen));
2285 				} else {
2286 					clen = 0;
2287 					cm = NULL;
2288 				}
2289 			}
2290 		}
2291 		m0 = m0->m_act;
2292 	}
2293 }
2294 
2295 #ifdef DDB
2296 static void
2297 db_print_indent(int indent)
2298 {
2299 	int i;
2300 
2301 	for (i = 0; i < indent; i++)
2302 		db_printf(" ");
2303 }
2304 
2305 static void
2306 db_print_unpflags(int unp_flags)
2307 {
2308 	int comma;
2309 
2310 	comma = 0;
2311 	if (unp_flags & UNP_HAVEPC) {
2312 		db_printf("%sUNP_HAVEPC", comma ? ", " : "");
2313 		comma = 1;
2314 	}
2315 	if (unp_flags & UNP_HAVEPCCACHED) {
2316 		db_printf("%sUNP_HAVEPCCACHED", comma ? ", " : "");
2317 		comma = 1;
2318 	}
2319 	if (unp_flags & UNP_WANTCRED) {
2320 		db_printf("%sUNP_WANTCRED", comma ? ", " : "");
2321 		comma = 1;
2322 	}
2323 	if (unp_flags & UNP_CONNWAIT) {
2324 		db_printf("%sUNP_CONNWAIT", comma ? ", " : "");
2325 		comma = 1;
2326 	}
2327 	if (unp_flags & UNP_CONNECTING) {
2328 		db_printf("%sUNP_CONNECTING", comma ? ", " : "");
2329 		comma = 1;
2330 	}
2331 	if (unp_flags & UNP_BINDING) {
2332 		db_printf("%sUNP_BINDING", comma ? ", " : "");
2333 		comma = 1;
2334 	}
2335 }
2336 
2337 static void
2338 db_print_xucred(int indent, struct xucred *xu)
2339 {
2340 	int comma, i;
2341 
2342 	db_print_indent(indent);
2343 	db_printf("cr_version: %u   cr_uid: %u   cr_ngroups: %d\n",
2344 	    xu->cr_version, xu->cr_uid, xu->cr_ngroups);
2345 	db_print_indent(indent);
2346 	db_printf("cr_groups: ");
2347 	comma = 0;
2348 	for (i = 0; i < xu->cr_ngroups; i++) {
2349 		db_printf("%s%u", comma ? ", " : "", xu->cr_groups[i]);
2350 		comma = 1;
2351 	}
2352 	db_printf("\n");
2353 }
2354 
2355 static void
2356 db_print_unprefs(int indent, struct unp_head *uh)
2357 {
2358 	struct unpcb *unp;
2359 	int counter;
2360 
2361 	counter = 0;
2362 	LIST_FOREACH(unp, uh, unp_reflink) {
2363 		if (counter % 4 == 0)
2364 			db_print_indent(indent);
2365 		db_printf("%p  ", unp);
2366 		if (counter % 4 == 3)
2367 			db_printf("\n");
2368 		counter++;
2369 	}
2370 	if (counter != 0 && counter % 4 != 0)
2371 		db_printf("\n");
2372 }
2373 
2374 DB_SHOW_COMMAND(unpcb, db_show_unpcb)
2375 {
2376 	struct unpcb *unp;
2377 
2378         if (!have_addr) {
2379                 db_printf("usage: show unpcb <addr>\n");
2380                 return;
2381         }
2382         unp = (struct unpcb *)addr;
2383 
2384 	db_printf("unp_socket: %p   unp_vnode: %p\n", unp->unp_socket,
2385 	    unp->unp_vnode);
2386 
2387 	db_printf("unp_ino: %d   unp_conn: %p\n", unp->unp_ino,
2388 	    unp->unp_conn);
2389 
2390 	db_printf("unp_refs:\n");
2391 	db_print_unprefs(2, &unp->unp_refs);
2392 
2393 	/* XXXRW: Would be nice to print the full address, if any. */
2394 	db_printf("unp_addr: %p\n", unp->unp_addr);
2395 
2396 	db_printf("unp_cc: %d   unp_mbcnt: %d   unp_gencnt: %llu\n",
2397 	    unp->unp_cc, unp->unp_mbcnt,
2398 	    (unsigned long long)unp->unp_gencnt);
2399 
2400 	db_printf("unp_flags: %x (", unp->unp_flags);
2401 	db_print_unpflags(unp->unp_flags);
2402 	db_printf(")\n");
2403 
2404 	db_printf("unp_peercred:\n");
2405 	db_print_xucred(2, &unp->unp_peercred);
2406 
2407 	db_printf("unp_refcount: %u\n", unp->unp_refcount);
2408 }
2409 #endif
2410