xref: /freebsd/sys/kern/uipc_usrreq.c (revision f0a75d274af375d15b97b830966b99a02b7db911)
1 /*-
2  * Copyright (c) 1982, 1986, 1989, 1991, 1993
3  *	The Regents of the University of California.
4  * Copyright (c) 2004-2007 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  *	SEQPACKET, RDM
54  *	rethink name space problems
55  *	need a proper out-of-band
56  *	lock pushdown
57  */
58 
59 #include <sys/cdefs.h>
60 __FBSDID("$FreeBSD$");
61 
62 #include "opt_mac.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/jail.h>
72 #include <sys/kernel.h>
73 #include <sys/lock.h>
74 #include <sys/mbuf.h>
75 #include <sys/mount.h>
76 #include <sys/mutex.h>
77 #include <sys/namei.h>
78 #include <sys/proc.h>
79 #include <sys/protosw.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 <security/mac/mac_framework.h>
95 
96 #include <vm/uma.h>
97 
98 static uma_zone_t	unp_zone;
99 static unp_gen_t	unp_gencnt;
100 static u_int		unp_count;	/* Count of local sockets. */
101 static ino_t		unp_ino;	/* Prototype for fake inode numbers. */
102 static int		unp_rights;	/* File descriptors in flight. */
103 static struct unp_head	unp_shead;	/* List of local stream sockets. */
104 static struct unp_head	unp_dhead;	/* List of local datagram sockets. */
105 
106 static const struct sockaddr	sun_noname = { sizeof(sun_noname), AF_LOCAL };
107 
108 /*
109  * Garbage collection of cyclic file descriptor/socket references occurs
110  * asynchronously in a taskqueue context in order to avoid recursion and
111  * reentrance in the UNIX domain socket, file descriptor, and socket layer
112  * code.  See unp_gc() for a full description.
113  */
114 static struct task	unp_gc_task;
115 
116 /*
117  * Both send and receive buffers are allocated PIPSIZ bytes of buffering for
118  * stream sockets, although the total for sender and receiver is actually
119  * only PIPSIZ.
120  *
121  * Datagram sockets really use the sendspace as the maximum datagram size,
122  * and don't really want to reserve the sendspace.  Their recvspace should be
123  * large enough for at least one max-size datagram plus address.
124  */
125 #ifndef PIPSIZ
126 #define	PIPSIZ	8192
127 #endif
128 static u_long	unpst_sendspace = PIPSIZ;
129 static u_long	unpst_recvspace = PIPSIZ;
130 static u_long	unpdg_sendspace = 2*1024;	/* really max datagram size */
131 static u_long	unpdg_recvspace = 4*1024;
132 
133 SYSCTL_NODE(_net, PF_LOCAL, local, CTLFLAG_RW, 0, "Local domain");
134 SYSCTL_NODE(_net_local, SOCK_STREAM, stream, CTLFLAG_RW, 0, "SOCK_STREAM");
135 SYSCTL_NODE(_net_local, SOCK_DGRAM, dgram, CTLFLAG_RW, 0, "SOCK_DGRAM");
136 
137 SYSCTL_ULONG(_net_local_stream, OID_AUTO, sendspace, CTLFLAG_RW,
138 	   &unpst_sendspace, 0, "");
139 SYSCTL_ULONG(_net_local_stream, OID_AUTO, recvspace, CTLFLAG_RW,
140 	   &unpst_recvspace, 0, "");
141 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, maxdgram, CTLFLAG_RW,
142 	   &unpdg_sendspace, 0, "");
143 SYSCTL_ULONG(_net_local_dgram, OID_AUTO, recvspace, CTLFLAG_RW,
144 	   &unpdg_recvspace, 0, "");
145 SYSCTL_INT(_net_local, OID_AUTO, inflight, CTLFLAG_RD, &unp_rights, 0, "");
146 
147 /*-
148  * Locking and synchronization:
149  *
150  * The global UNIX domain socket rwlock (unp_global_rwlock) protects all
151  * global variables, including the linked lists tracking the set of allocated
152  * UNIX domain sockets.  The global rwlock also serves to prevent deadlock
153  * when more than one PCB lock is acquired at a time (i.e., during
154  * connect()).  Finally, the global rwlock protects uncounted references from
155  * vnodes to sockets bound to those vnodes: to safely dereference the
156  * v_socket pointer, the global rwlock must be held while a full reference is
157  * acquired.
158  *
159  * UNIX domain sockets each have an unpcb hung off of their so_pcb pointer,
160  * allocated in pru_attach() and freed in pru_detach().  The validity of that
161  * pointer is an invariant, so no lock is required to dereference the so_pcb
162  * pointer if a valid socket reference is held by the caller.  In practice,
163  * this is always true during operations performed on a socket.  Each unpcb
164  * has a back-pointer to its socket, unp_socket, which will be stable under
165  * the same circumstances.
166  *
167  * This pointer may only be safely dereferenced as long as a valid reference
168  * to the unpcb is held.  Typically, this reference will be from the socket,
169  * or from another unpcb when the referring unpcb's lock is held (in order
170  * that the reference not be invalidated during use).  For example, to follow
171  * unp->unp_conn->unp_socket, you need unlock the lock on unp, not unp_conn,
172  * as unp_socket remains valid as long as the reference to unp_conn is valid.
173  *
174  * Fields of unpcbss are locked using a per-unpcb lock, unp_mtx.  Individual
175  * atomic reads without the lock may be performed "lockless", but more
176  * complex reads and read-modify-writes require the mutex to be held.  No
177  * lock order is defined between unpcb locks -- multiple unpcb locks may be
178  * acquired at the same time only when holding the global UNIX domain socket
179  * rwlock exclusively, which prevents deadlocks.
180  *
181  * Blocking with UNIX domain sockets is a tricky issue: unlike most network
182  * protocols, bind() is a non-atomic operation, and connect() requires
183  * potential sleeping in the protocol, due to potentially waiting on local or
184  * distributed file systems.  We try to separate "lookup" operations, which
185  * may sleep, and the IPC operations themselves, which typically can occur
186  * with relative atomicity as locks can be held over the entire operation.
187  *
188  * Another tricky issue is simultaneous multi-threaded or multi-process
189  * access to a single UNIX domain socket.  These are handled by the flags
190  * UNP_CONNECTING and UNP_BINDING, which prevent concurrent connecting or
191  * binding, both of which involve dropping UNIX domain socket locks in order
192  * to perform namei() and other file system operations.
193  */
194 static struct rwlock	unp_global_rwlock;
195 
196 #define	UNP_GLOBAL_LOCK_INIT()		rw_init(&unp_global_rwlock,	\
197 					    "unp_global_rwlock")
198 
199 #define	UNP_GLOBAL_LOCK_ASSERT()	rw_assert(&unp_global_rwlock,	\
200 					    RA_LOCKED)
201 #define	UNP_GLOBAL_UNLOCK_ASSERT()	rw_assert(&unp_global_rwlock,	\
202 					    RA_UNLOCKED)
203 
204 #define	UNP_GLOBAL_WLOCK()		rw_wlock(&unp_global_rwlock)
205 #define	UNP_GLOBAL_WUNLOCK()		rw_wunlock(&unp_global_rwlock)
206 #define	UNP_GLOBAL_WLOCK_ASSERT()	rw_assert(&unp_global_rwlock,	\
207 					    RA_WLOCKED)
208 #define	UNP_GLOBAL_WOWNED()		rw_wowned(&unp_global_rwlock)
209 
210 #define	UNP_GLOBAL_RLOCK()		rw_rlock(&unp_global_rwlock)
211 #define	UNP_GLOBAL_RUNLOCK()		rw_runlock(&unp_global_rwlock)
212 #define	UNP_GLOBAL_RLOCK_ASSERT()	rw_assert(&unp_global_rwlock,	\
213 					    RA_RLOCKED)
214 
215 #define UNP_PCB_LOCK_INIT(unp)		mtx_init(&(unp)->unp_mtx,	\
216 					    "unp_mtx", "unp_mtx",	\
217 					    MTX_DUPOK|MTX_DEF|MTX_RECURSE)
218 #define	UNP_PCB_LOCK_DESTROY(unp)	mtx_destroy(&(unp)->unp_mtx)
219 #define	UNP_PCB_LOCK(unp)		mtx_lock(&(unp)->unp_mtx)
220 #define	UNP_PCB_UNLOCK(unp)		mtx_unlock(&(unp)->unp_mtx)
221 #define	UNP_PCB_LOCK_ASSERT(unp)	mtx_assert(&(unp)->unp_mtx, MA_OWNED)
222 
223 static int	unp_connect(struct socket *, struct sockaddr *,
224 		    struct thread *);
225 static int	unp_connect2(struct socket *so, struct socket *so2, int);
226 static void	unp_disconnect(struct unpcb *unp, struct unpcb *unp2);
227 static void	unp_shutdown(struct unpcb *);
228 static void	unp_drop(struct unpcb *, int);
229 static void	unp_gc(__unused void *, int);
230 static void	unp_scan(struct mbuf *, void (*)(struct file *));
231 static void	unp_mark(struct file *);
232 static void	unp_discard(struct file *);
233 static void	unp_freerights(struct file **, int);
234 static int	unp_internalize(struct mbuf **, struct thread *);
235 static struct mbuf	*unp_addsockcred(struct thread *, struct mbuf *);
236 
237 /*
238  * Definitions of protocols supported in the LOCAL domain.
239  */
240 static struct domain localdomain;
241 static struct protosw localsw[] = {
242 {
243 	.pr_type =		SOCK_STREAM,
244 	.pr_domain =		&localdomain,
245 	.pr_flags =		PR_CONNREQUIRED|PR_WANTRCVD|PR_RIGHTS,
246 	.pr_ctloutput =		&uipc_ctloutput,
247 	.pr_usrreqs =		&uipc_usrreqs
248 },
249 {
250 	.pr_type =		SOCK_DGRAM,
251 	.pr_domain =		&localdomain,
252 	.pr_flags =		PR_ATOMIC|PR_ADDR|PR_RIGHTS,
253 	.pr_usrreqs =		&uipc_usrreqs
254 },
255 };
256 
257 static struct domain localdomain = {
258 	.dom_family =		AF_LOCAL,
259 	.dom_name =		"local",
260 	.dom_init =		unp_init,
261 	.dom_externalize =	unp_externalize,
262 	.dom_dispose =		unp_dispose,
263 	.dom_protosw =		localsw,
264 	.dom_protoswNPROTOSW =	&localsw[sizeof(localsw)/sizeof(localsw[0])]
265 };
266 DOMAIN_SET(local);
267 
268 static void
269 uipc_abort(struct socket *so)
270 {
271 	struct unpcb *unp, *unp2;
272 
273 	unp = sotounpcb(so);
274 	KASSERT(unp != NULL, ("uipc_abort: unp == NULL"));
275 
276 	UNP_GLOBAL_WLOCK();
277 	UNP_PCB_LOCK(unp);
278 	unp2 = unp->unp_conn;
279 	if (unp2 != NULL) {
280 		UNP_PCB_LOCK(unp2);
281 		unp_drop(unp2, ECONNABORTED);
282 		UNP_PCB_UNLOCK(unp2);
283 	}
284 	UNP_PCB_UNLOCK(unp);
285 	UNP_GLOBAL_WUNLOCK();
286 }
287 
288 static int
289 uipc_accept(struct socket *so, struct sockaddr **nam)
290 {
291 	struct unpcb *unp, *unp2;
292 	const struct sockaddr *sa;
293 
294 	/*
295 	 * Pass back name of connected socket, if it was bound and we are
296 	 * still connected (our peer may have closed already!).
297 	 */
298 	unp = sotounpcb(so);
299 	KASSERT(unp != NULL, ("uipc_accept: unp == NULL"));
300 
301 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
302 	UNP_GLOBAL_RLOCK();
303 	unp2 = unp->unp_conn;
304 	if (unp2 != NULL && unp2->unp_addr != NULL) {
305 		UNP_PCB_LOCK(unp2);
306 		sa = (struct sockaddr *) unp2->unp_addr;
307 		bcopy(sa, *nam, sa->sa_len);
308 		UNP_PCB_UNLOCK(unp2);
309 	} else {
310 		sa = &sun_noname;
311 		bcopy(sa, *nam, sa->sa_len);
312 	}
313 	UNP_GLOBAL_RUNLOCK();
314 	return (0);
315 }
316 
317 static int
318 uipc_attach(struct socket *so, int proto, struct thread *td)
319 {
320 	u_long sendspace, recvspace;
321 	struct unpcb *unp;
322 	int error, locked;
323 
324 	KASSERT(so->so_pcb == NULL, ("uipc_attach: so_pcb != NULL"));
325 	if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) {
326 		switch (so->so_type) {
327 		case SOCK_STREAM:
328 			sendspace = unpst_sendspace;
329 			recvspace = unpst_recvspace;
330 			break;
331 
332 		case SOCK_DGRAM:
333 			sendspace = unpdg_sendspace;
334 			recvspace = unpdg_recvspace;
335 			break;
336 
337 		default:
338 			panic("uipc_attach");
339 		}
340 		error = soreserve(so, sendspace, recvspace);
341 		if (error)
342 			return (error);
343 	}
344 	unp = uma_zalloc(unp_zone, M_NOWAIT | M_ZERO);
345 	if (unp == NULL)
346 		return (ENOBUFS);
347 	LIST_INIT(&unp->unp_refs);
348 	UNP_PCB_LOCK_INIT(unp);
349 	unp->unp_socket = so;
350 	so->so_pcb = unp;
351 	unp->unp_refcount = 1;
352 	locked = 0;
353 
354 	/*
355 	 * uipc_attach() may be called indirectly from within the UNIX domain
356 	 * socket code via sonewconn() in unp_connect().  Since rwlocks can
357 	 * not be recursed, we do the closest thing.
358 	 */
359 	if (!UNP_GLOBAL_WOWNED()) {
360 		UNP_GLOBAL_WLOCK();
361 		locked = 1;
362 	}
363 	unp->unp_gencnt = ++unp_gencnt;
364 	unp_count++;
365 	LIST_INSERT_HEAD(so->so_type == SOCK_DGRAM ? &unp_dhead : &unp_shead,
366 	    unp, unp_link);
367 	if (locked)
368 		UNP_GLOBAL_WUNLOCK();
369 
370 	return (0);
371 }
372 
373 static int
374 uipc_bind(struct socket *so, struct sockaddr *nam, struct thread *td)
375 {
376 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
377 	struct vattr vattr;
378 	int error, namelen;
379 	struct nameidata nd;
380 	struct unpcb *unp;
381 	struct vnode *vp;
382 	struct mount *mp;
383 	char *buf;
384 
385 	unp = sotounpcb(so);
386 	KASSERT(unp != NULL, ("uipc_bind: unp == NULL"));
387 
388 	namelen = soun->sun_len - offsetof(struct sockaddr_un, sun_path);
389 	if (namelen <= 0)
390 		return (EINVAL);
391 
392 	/*
393 	 * We don't allow simultaneous bind() calls on a single UNIX domain
394 	 * socket, so flag in-progress operations, and return an error if an
395 	 * operation is already in progress.
396 	 *
397 	 * Historically, we have not allowed a socket to be rebound, so this
398 	 * also returns an error.  Not allowing re-binding certainly
399 	 * simplifies the implementation and avoids a great many possible
400 	 * failure modes.
401 	 */
402 	UNP_PCB_LOCK(unp);
403 	if (unp->unp_vnode != NULL) {
404 		UNP_PCB_UNLOCK(unp);
405 		return (EINVAL);
406 	}
407 	if (unp->unp_flags & UNP_BINDING) {
408 		UNP_PCB_UNLOCK(unp);
409 		return (EALREADY);
410 	}
411 	unp->unp_flags |= UNP_BINDING;
412 	UNP_PCB_UNLOCK(unp);
413 
414 	buf = malloc(namelen + 1, M_TEMP, M_WAITOK);
415 	strlcpy(buf, soun->sun_path, namelen + 1);
416 
417 	mtx_lock(&Giant);
418 restart:
419 	mtx_assert(&Giant, MA_OWNED);
420 	NDINIT(&nd, CREATE, NOFOLLOW | LOCKPARENT | SAVENAME, UIO_SYSSPACE,
421 	    buf, td);
422 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */
423 	error = namei(&nd);
424 	if (error)
425 		goto error;
426 	vp = nd.ni_vp;
427 	if (vp != NULL || vn_start_write(nd.ni_dvp, &mp, V_NOWAIT) != 0) {
428 		NDFREE(&nd, NDF_ONLY_PNBUF);
429 		if (nd.ni_dvp == vp)
430 			vrele(nd.ni_dvp);
431 		else
432 			vput(nd.ni_dvp);
433 		if (vp != NULL) {
434 			vrele(vp);
435 			error = EADDRINUSE;
436 			goto error;
437 		}
438 		error = vn_start_write(NULL, &mp, V_XSLEEP | PCATCH);
439 		if (error)
440 			goto error;
441 		goto restart;
442 	}
443 	VATTR_NULL(&vattr);
444 	vattr.va_type = VSOCK;
445 	vattr.va_mode = (ACCESSPERMS & ~td->td_proc->p_fd->fd_cmask);
446 #ifdef MAC
447 	error = mac_check_vnode_create(td->td_ucred, nd.ni_dvp, &nd.ni_cnd,
448 	    &vattr);
449 #endif
450 	if (error == 0) {
451 		VOP_LEASE(nd.ni_dvp, td, td->td_ucred, LEASE_WRITE);
452 		error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr);
453 	}
454 	NDFREE(&nd, NDF_ONLY_PNBUF);
455 	vput(nd.ni_dvp);
456 	if (error) {
457 		vn_finished_write(mp);
458 		goto error;
459 	}
460 	vp = nd.ni_vp;
461 	ASSERT_VOP_LOCKED(vp, "uipc_bind");
462 	soun = (struct sockaddr_un *)sodupsockaddr(nam, M_WAITOK);
463 
464 	UNP_GLOBAL_WLOCK();
465 	UNP_PCB_LOCK(unp);
466 	vp->v_socket = unp->unp_socket;
467 	unp->unp_vnode = vp;
468 	unp->unp_addr = soun;
469 	unp->unp_flags &= ~UNP_BINDING;
470 	UNP_PCB_UNLOCK(unp);
471 	UNP_GLOBAL_WUNLOCK();
472 	VOP_UNLOCK(vp, 0, td);
473 	vn_finished_write(mp);
474 	mtx_unlock(&Giant);
475 	free(buf, M_TEMP);
476 	return (0);
477 
478 error:
479 	UNP_PCB_LOCK(unp);
480 	unp->unp_flags &= ~UNP_BINDING;
481 	UNP_PCB_UNLOCK(unp);
482 	mtx_unlock(&Giant);
483 	free(buf, M_TEMP);
484 	return (error);
485 }
486 
487 static int
488 uipc_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
489 {
490 	int error;
491 
492 	KASSERT(td == curthread, ("uipc_connect: td != curthread"));
493 	UNP_GLOBAL_WLOCK();
494 	error = unp_connect(so, nam, td);
495 	UNP_GLOBAL_WUNLOCK();
496 	return (error);
497 }
498 
499 static void
500 uipc_close(struct socket *so)
501 {
502 	struct unpcb *unp, *unp2;
503 
504 	unp = sotounpcb(so);
505 	KASSERT(unp != NULL, ("uipc_close: unp == NULL"));
506 
507 	UNP_GLOBAL_WLOCK();
508 	UNP_PCB_LOCK(unp);
509 	unp2 = unp->unp_conn;
510 	if (unp2 != NULL) {
511 		UNP_PCB_LOCK(unp2);
512 		unp_disconnect(unp, unp2);
513 		UNP_PCB_UNLOCK(unp2);
514 	}
515 	UNP_PCB_UNLOCK(unp);
516 	UNP_GLOBAL_WUNLOCK();
517 }
518 
519 int
520 uipc_connect2(struct socket *so1, struct socket *so2)
521 {
522 	struct unpcb *unp, *unp2;
523 	int error;
524 
525 	UNP_GLOBAL_WLOCK();
526 	unp = so1->so_pcb;
527 	KASSERT(unp != NULL, ("uipc_connect2: unp == NULL"));
528 	UNP_PCB_LOCK(unp);
529 	unp2 = so2->so_pcb;
530 	KASSERT(unp2 != NULL, ("uipc_connect2: unp2 == NULL"));
531 	UNP_PCB_LOCK(unp2);
532 	error = unp_connect2(so1, so2, PRU_CONNECT2);
533 	UNP_PCB_UNLOCK(unp2);
534 	UNP_PCB_UNLOCK(unp);
535 	UNP_GLOBAL_WUNLOCK();
536 	return (error);
537 }
538 
539 /* control is EOPNOTSUPP */
540 
541 static void
542 uipc_detach(struct socket *so)
543 {
544 	struct unpcb *unp, *unp2;
545 	struct sockaddr_un *saved_unp_addr;
546 	struct vnode *vp;
547 	int freeunp, local_unp_rights;
548 
549 	unp = sotounpcb(so);
550 	KASSERT(unp != NULL, ("uipc_detach: unp == NULL"));
551 
552 	UNP_GLOBAL_WLOCK();
553 	UNP_PCB_LOCK(unp);
554 
555 	LIST_REMOVE(unp, unp_link);
556 	unp->unp_gencnt = ++unp_gencnt;
557 	--unp_count;
558 
559 	/*
560 	 * XXXRW: Should assert vp->v_socket == so.
561 	 */
562 	if ((vp = unp->unp_vnode) != NULL) {
563 		unp->unp_vnode->v_socket = NULL;
564 		unp->unp_vnode = NULL;
565 	}
566 	unp2 = unp->unp_conn;
567 	if (unp2 != NULL) {
568 		UNP_PCB_LOCK(unp2);
569 		unp_disconnect(unp, unp2);
570 		UNP_PCB_UNLOCK(unp2);
571 	}
572 
573 	/*
574 	 * We hold the global lock, so it's OK to acquire multiple pcb locks
575 	 * at a time.
576 	 */
577 	while (!LIST_EMPTY(&unp->unp_refs)) {
578 		struct unpcb *ref = LIST_FIRST(&unp->unp_refs);
579 
580 		UNP_PCB_LOCK(ref);
581 		unp_drop(ref, ECONNRESET);
582 		UNP_PCB_UNLOCK(ref);
583 	}
584 	UNP_GLOBAL_WUNLOCK();
585 	unp->unp_socket->so_pcb = NULL;
586 	local_unp_rights = unp_rights;
587 	saved_unp_addr = unp->unp_addr;
588 	unp->unp_addr = NULL;
589 	unp->unp_refcount--;
590 	freeunp = (unp->unp_refcount == 0);
591 	if (saved_unp_addr != NULL)
592 		FREE(saved_unp_addr, M_SONAME);
593 	if (freeunp) {
594 		UNP_PCB_LOCK_DESTROY(unp);
595 		uma_zfree(unp_zone, unp);
596 	} else
597 		UNP_PCB_UNLOCK(unp);
598 	if (vp) {
599 		int vfslocked;
600 
601 		vfslocked = VFS_LOCK_GIANT(vp->v_mount);
602 		vrele(vp);
603 		VFS_UNLOCK_GIANT(vfslocked);
604 	}
605 	if (local_unp_rights)
606 		taskqueue_enqueue(taskqueue_thread, &unp_gc_task);
607 }
608 
609 static int
610 uipc_disconnect(struct socket *so)
611 {
612 	struct unpcb *unp, *unp2;
613 
614 	unp = sotounpcb(so);
615 	KASSERT(unp != NULL, ("uipc_disconnect: unp == NULL"));
616 
617 	UNP_GLOBAL_WLOCK();
618 	UNP_PCB_LOCK(unp);
619 	unp2 = unp->unp_conn;
620 	if (unp2 != NULL) {
621 		UNP_PCB_LOCK(unp2);
622 		unp_disconnect(unp, unp2);
623 		UNP_PCB_UNLOCK(unp2);
624 	}
625 	UNP_PCB_UNLOCK(unp);
626 	UNP_GLOBAL_WUNLOCK();
627 	return (0);
628 }
629 
630 static int
631 uipc_listen(struct socket *so, int backlog, struct thread *td)
632 {
633 	struct unpcb *unp;
634 	int error;
635 
636 	unp = sotounpcb(so);
637 	KASSERT(unp != NULL, ("uipc_listen: unp == NULL"));
638 
639 	UNP_PCB_LOCK(unp);
640 	if (unp->unp_vnode == NULL) {
641 		UNP_PCB_UNLOCK(unp);
642 		return (EINVAL);
643 	}
644 
645 	SOCK_LOCK(so);
646 	error = solisten_proto_check(so);
647 	if (error == 0) {
648 		cru2x(td->td_ucred, &unp->unp_peercred);
649 		unp->unp_flags |= UNP_HAVEPCCACHED;
650 		solisten_proto(so, backlog);
651 	}
652 	SOCK_UNLOCK(so);
653 	UNP_PCB_UNLOCK(unp);
654 	return (error);
655 }
656 
657 static int
658 uipc_peeraddr(struct socket *so, struct sockaddr **nam)
659 {
660 	struct unpcb *unp, *unp2;
661 	const struct sockaddr *sa;
662 
663 	unp = sotounpcb(so);
664 	KASSERT(unp != NULL, ("uipc_peeraddr: unp == NULL"));
665 
666 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
667 	UNP_PCB_LOCK(unp);
668 	/*
669 	 * XXX: It seems that this test always fails even when connection is
670 	 * established.  So, this else clause is added as workaround to
671 	 * return PF_LOCAL sockaddr.
672 	 */
673 	unp2 = unp->unp_conn;
674 	if (unp2 != NULL) {
675 		UNP_PCB_LOCK(unp2);
676 		if (unp2->unp_addr != NULL)
677 			sa = (struct sockaddr *) unp->unp_conn->unp_addr;
678 		else
679 			sa = &sun_noname;
680 		bcopy(sa, *nam, sa->sa_len);
681 		UNP_PCB_UNLOCK(unp2);
682 	} else {
683 		sa = &sun_noname;
684 		bcopy(sa, *nam, sa->sa_len);
685 	}
686 	UNP_PCB_UNLOCK(unp);
687 	return (0);
688 }
689 
690 static int
691 uipc_rcvd(struct socket *so, int flags)
692 {
693 	struct unpcb *unp, *unp2;
694 	struct socket *so2;
695 	u_int mbcnt, sbcc;
696 	u_long newhiwat;
697 
698 	unp = sotounpcb(so);
699 	KASSERT(unp != NULL, ("uipc_rcvd: unp == NULL"));
700 
701 	if (so->so_type == SOCK_DGRAM)
702 		panic("uipc_rcvd DGRAM?");
703 
704 	if (so->so_type != SOCK_STREAM)
705 		panic("uipc_rcvd unknown socktype");
706 
707 	/*
708 	 * Adjust backpressure on sender and wakeup any waiting to write.
709 	 *
710 	 * The consistency requirements here are a bit complex: we must
711 	 * acquire the lock for our own unpcb in order to prevent it from
712 	 * disconnecting while in use, changing the unp_conn peer.  We do not
713 	 * need unp2's lock, since the unp2->unp_socket pointer will remain
714 	 * static as long as the unp2 pcb is valid, which it will be until we
715 	 * release unp's lock to allow a disconnect.  We do need socket
716 	 * mutexes for both socket endpoints since we manipulate fields in
717 	 * both; we hold both locks at once since we access both
718 	 * simultaneously.
719 	 */
720 	SOCKBUF_LOCK(&so->so_rcv);
721 	mbcnt = so->so_rcv.sb_mbcnt;
722 	sbcc = so->so_rcv.sb_cc;
723 	SOCKBUF_UNLOCK(&so->so_rcv);
724 	UNP_PCB_LOCK(unp);
725 	unp2 = unp->unp_conn;
726 	if (unp2 == NULL) {
727 		UNP_PCB_UNLOCK(unp);
728 		return (0);
729 	}
730 	so2 = unp2->unp_socket;
731 	SOCKBUF_LOCK(&so2->so_snd);
732 	so2->so_snd.sb_mbmax += unp->unp_mbcnt - mbcnt;
733 	newhiwat = so2->so_snd.sb_hiwat + unp->unp_cc - sbcc;
734 	(void)chgsbsize(so2->so_cred->cr_uidinfo, &so2->so_snd.sb_hiwat,
735 	    newhiwat, RLIM_INFINITY);
736 	sowwakeup_locked(so2);
737 	unp->unp_mbcnt = mbcnt;
738 	unp->unp_cc = sbcc;
739 	UNP_PCB_UNLOCK(unp);
740 	return (0);
741 }
742 
743 /* pru_rcvoob is EOPNOTSUPP */
744 
745 static int
746 uipc_send(struct socket *so, int flags, struct mbuf *m, struct sockaddr *nam,
747     struct mbuf *control, struct thread *td)
748 {
749 	struct unpcb *unp, *unp2;
750 	struct socket *so2;
751 	u_int mbcnt, sbcc;
752 	u_long newhiwat;
753 	int error = 0;
754 
755 	unp = sotounpcb(so);
756 	KASSERT(unp != NULL, ("uipc_send: unp == NULL"));
757 
758 	if (flags & PRUS_OOB) {
759 		error = EOPNOTSUPP;
760 		goto release;
761 	}
762 
763 	if (control != NULL && (error = unp_internalize(&control, td)))
764 		goto release;
765 
766 	if ((nam != NULL) || (flags & PRUS_EOF))
767 		UNP_GLOBAL_WLOCK();
768 	else
769 		UNP_GLOBAL_RLOCK();
770 
771 	switch (so->so_type) {
772 	case SOCK_DGRAM:
773 	{
774 		const struct sockaddr *from;
775 
776 		unp2 = unp->unp_conn;
777 		if (nam != NULL) {
778 			UNP_GLOBAL_WLOCK_ASSERT();
779 			if (unp2 != NULL) {
780 				error = EISCONN;
781 				break;
782 			}
783 			error = unp_connect(so, nam, td);
784 			if (error)
785 				break;
786 			unp2 = unp->unp_conn;
787 		} else {
788 			if (unp2 == NULL) {
789 				error = ENOTCONN;
790 				break;
791 			}
792 		}
793 		/*
794 		 * Because connect() and send() are non-atomic in a sendto()
795 		 * with a target address, it's possible that the socket will
796 		 * have disconnected before the send() can run.  In that case
797 		 * return the slightly counter-intuitive but otherwise
798 		 * correct error that the socket is not connected.
799 		 */
800 		if (unp2 == NULL) {
801 			error = ENOTCONN;
802 			break;
803 		}
804 		/* Lockless read. */
805 		if (unp2->unp_flags & UNP_WANTCRED)
806 			control = unp_addsockcred(td, control);
807 		UNP_PCB_LOCK(unp);
808 		if (unp->unp_addr != NULL)
809 			from = (struct sockaddr *)unp->unp_addr;
810 		else
811 			from = &sun_noname;
812 		so2 = unp2->unp_socket;
813 		SOCKBUF_LOCK(&so2->so_rcv);
814 		if (sbappendaddr_locked(&so2->so_rcv, from, m, control)) {
815 			sorwakeup_locked(so2);
816 			m = NULL;
817 			control = NULL;
818 		} else {
819 			SOCKBUF_UNLOCK(&so2->so_rcv);
820 			error = ENOBUFS;
821 		}
822 		if (nam != NULL) {
823 			UNP_GLOBAL_WLOCK_ASSERT();
824 			UNP_PCB_LOCK(unp2);
825 			unp_disconnect(unp, unp2);
826 			UNP_PCB_UNLOCK(unp2);
827 		}
828 		UNP_PCB_UNLOCK(unp);
829 		break;
830 	}
831 
832 	case SOCK_STREAM:
833 		/*
834 		 * Connect if not connected yet.
835 		 *
836 		 * Note: A better implementation would complain if not equal
837 		 * to the peer's address.
838 		 */
839 		if ((so->so_state & SS_ISCONNECTED) == 0) {
840 			if (nam != NULL) {
841 				UNP_GLOBAL_WLOCK_ASSERT();
842 				error = unp_connect(so, nam, td);
843 				if (error)
844 					break;	/* XXX */
845 			} else {
846 				error = ENOTCONN;
847 				break;
848 			}
849 		}
850 
851 		/* Lockless read. */
852 		if (so->so_snd.sb_state & SBS_CANTSENDMORE) {
853 			error = EPIPE;
854 			break;
855 		}
856 		/*
857 		 * Because connect() and send() are non-atomic in a sendto()
858 		 * with a target address, it's possible that the socket will
859 		 * have disconnected before the send() can run.  In that case
860 		 * return the slightly counter-intuitive but otherwise
861 		 * correct error that the socket is not connected.
862 		 *
863 		 * Lock order here has to be handled carefully: we hold the
864 		 * global lock, so acquiring two unpcb locks is OK.  We must
865 		 * acquire both before acquiring any socket mutexes.  We must
866 		 * also acquire the local socket send mutex before the remote
867 		 * socket receive mutex.  The only tricky thing is making
868 		 * sure to acquire the unp2 lock before the local socket send
869 		 * lock, or we will experience deadlocks.
870 		 */
871 		unp2 = unp->unp_conn;
872 		if (unp2 == NULL) {
873 			error = ENOTCONN;
874 			break;
875 		}
876 		so2 = unp2->unp_socket;
877 		UNP_PCB_LOCK(unp2);
878 		SOCKBUF_LOCK(&so2->so_rcv);
879 		if (unp2->unp_flags & UNP_WANTCRED) {
880 			/*
881 			 * Credentials are passed only once on SOCK_STREAM.
882 			 */
883 			unp2->unp_flags &= ~UNP_WANTCRED;
884 			control = unp_addsockcred(td, control);
885 		}
886 		/*
887 		 * Send to paired receive port, and then reduce send buffer
888 		 * hiwater marks to maintain backpressure.  Wake up readers.
889 		 */
890 		if (control != NULL) {
891 			if (sbappendcontrol_locked(&so2->so_rcv, m, control))
892 				control = NULL;
893 		} else
894 			sbappend_locked(&so2->so_rcv, m);
895 		mbcnt = so2->so_rcv.sb_mbcnt - unp2->unp_mbcnt;
896 		unp2->unp_mbcnt = so2->so_rcv.sb_mbcnt;
897 		sbcc = so2->so_rcv.sb_cc;
898 		sorwakeup_locked(so2);
899 
900 		SOCKBUF_LOCK(&so->so_snd);
901 		newhiwat = so->so_snd.sb_hiwat - (sbcc - unp2->unp_cc);
902 		(void)chgsbsize(so->so_cred->cr_uidinfo, &so->so_snd.sb_hiwat,
903 		    newhiwat, RLIM_INFINITY);
904 		so->so_snd.sb_mbmax -= mbcnt;
905 		SOCKBUF_UNLOCK(&so->so_snd);
906 		unp2->unp_cc = sbcc;
907 		UNP_PCB_UNLOCK(unp2);
908 		m = NULL;
909 		break;
910 
911 	default:
912 		panic("uipc_send unknown socktype");
913 	}
914 
915 	/*
916 	 * SEND_EOF is equivalent to a SEND followed by a SHUTDOWN.
917 	 */
918 	if (flags & PRUS_EOF) {
919 		UNP_PCB_LOCK(unp);
920 		socantsendmore(so);
921 		unp_shutdown(unp);
922 		UNP_PCB_UNLOCK(unp);
923 	}
924 
925 	if ((nam != NULL) || (flags & PRUS_EOF))
926 		UNP_GLOBAL_WUNLOCK();
927 	else
928 		UNP_GLOBAL_RUNLOCK();
929 
930 	if (control != NULL && error != 0)
931 		unp_dispose(control);
932 
933 release:
934 	if (control != NULL)
935 		m_freem(control);
936 	if (m != NULL)
937 		m_freem(m);
938 	return (error);
939 }
940 
941 static int
942 uipc_sense(struct socket *so, struct stat *sb)
943 {
944 	struct unpcb *unp, *unp2;
945 	struct socket *so2;
946 
947 	unp = sotounpcb(so);
948 	KASSERT(unp != NULL, ("uipc_sense: unp == NULL"));
949 
950 	sb->st_blksize = so->so_snd.sb_hiwat;
951 	UNP_GLOBAL_RLOCK();
952 	UNP_PCB_LOCK(unp);
953 	unp2 = unp->unp_conn;
954 	if (so->so_type == SOCK_STREAM && unp2 != NULL) {
955 		so2 = unp2->unp_socket;
956 		sb->st_blksize += so2->so_rcv.sb_cc;
957 	}
958 	sb->st_dev = NODEV;
959 	if (unp->unp_ino == 0)
960 		unp->unp_ino = (++unp_ino == 0) ? ++unp_ino : unp_ino;
961 	sb->st_ino = unp->unp_ino;
962 	UNP_PCB_UNLOCK(unp);
963 	UNP_GLOBAL_RUNLOCK();
964 	return (0);
965 }
966 
967 static int
968 uipc_shutdown(struct socket *so)
969 {
970 	struct unpcb *unp;
971 
972 	unp = sotounpcb(so);
973 	KASSERT(unp != NULL, ("uipc_shutdown: unp == NULL"));
974 
975 	UNP_GLOBAL_WLOCK();
976 	UNP_PCB_LOCK(unp);
977 	socantsendmore(so);
978 	unp_shutdown(unp);
979 	UNP_PCB_UNLOCK(unp);
980 	UNP_GLOBAL_WUNLOCK();
981 	return (0);
982 }
983 
984 static int
985 uipc_sockaddr(struct socket *so, struct sockaddr **nam)
986 {
987 	struct unpcb *unp;
988 	const struct sockaddr *sa;
989 
990 	unp = sotounpcb(so);
991 	KASSERT(unp != NULL, ("uipc_sockaddr: unp == NULL"));
992 
993 	*nam = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
994 	UNP_PCB_LOCK(unp);
995 	if (unp->unp_addr != NULL)
996 		sa = (struct sockaddr *) unp->unp_addr;
997 	else
998 		sa = &sun_noname;
999 	bcopy(sa, *nam, sa->sa_len);
1000 	UNP_PCB_UNLOCK(unp);
1001 	return (0);
1002 }
1003 
1004 struct pr_usrreqs uipc_usrreqs = {
1005 	.pru_abort = 		uipc_abort,
1006 	.pru_accept =		uipc_accept,
1007 	.pru_attach =		uipc_attach,
1008 	.pru_bind =		uipc_bind,
1009 	.pru_connect =		uipc_connect,
1010 	.pru_connect2 =		uipc_connect2,
1011 	.pru_detach =		uipc_detach,
1012 	.pru_disconnect =	uipc_disconnect,
1013 	.pru_listen =		uipc_listen,
1014 	.pru_peeraddr =		uipc_peeraddr,
1015 	.pru_rcvd =		uipc_rcvd,
1016 	.pru_send =		uipc_send,
1017 	.pru_sense =		uipc_sense,
1018 	.pru_shutdown =		uipc_shutdown,
1019 	.pru_sockaddr =		uipc_sockaddr,
1020 	.pru_close =		uipc_close,
1021 };
1022 
1023 int
1024 uipc_ctloutput(struct socket *so, struct sockopt *sopt)
1025 {
1026 	struct unpcb *unp;
1027 	struct xucred xu;
1028 	int error, optval;
1029 
1030 	if (sopt->sopt_level != 0)
1031 		return (EINVAL);
1032 
1033 	unp = sotounpcb(so);
1034 	KASSERT(unp != NULL, ("uipc_ctloutput: unp == NULL"));
1035 	error = 0;
1036 	switch (sopt->sopt_dir) {
1037 	case SOPT_GET:
1038 		switch (sopt->sopt_name) {
1039 		case LOCAL_PEERCRED:
1040 			UNP_PCB_LOCK(unp);
1041 			if (unp->unp_flags & UNP_HAVEPC)
1042 				xu = unp->unp_peercred;
1043 			else {
1044 				if (so->so_type == SOCK_STREAM)
1045 					error = ENOTCONN;
1046 				else
1047 					error = EINVAL;
1048 			}
1049 			UNP_PCB_UNLOCK(unp);
1050 			if (error == 0)
1051 				error = sooptcopyout(sopt, &xu, sizeof(xu));
1052 			break;
1053 
1054 		case LOCAL_CREDS:
1055 			/* Unocked read. */
1056 			optval = unp->unp_flags & UNP_WANTCRED ? 1 : 0;
1057 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1058 			break;
1059 
1060 		case LOCAL_CONNWAIT:
1061 			/* Unocked read. */
1062 			optval = unp->unp_flags & UNP_CONNWAIT ? 1 : 0;
1063 			error = sooptcopyout(sopt, &optval, sizeof(optval));
1064 			break;
1065 
1066 		default:
1067 			error = EOPNOTSUPP;
1068 			break;
1069 		}
1070 		break;
1071 
1072 	case SOPT_SET:
1073 		switch (sopt->sopt_name) {
1074 		case LOCAL_CREDS:
1075 		case LOCAL_CONNWAIT:
1076 			error = sooptcopyin(sopt, &optval, sizeof(optval),
1077 					    sizeof(optval));
1078 			if (error)
1079 				break;
1080 
1081 #define	OPTSET(bit) do {						\
1082 	UNP_PCB_LOCK(unp);						\
1083 	if (optval)							\
1084 		unp->unp_flags |= bit;					\
1085 	else								\
1086 		unp->unp_flags &= ~bit;					\
1087 	UNP_PCB_UNLOCK(unp);						\
1088 } while (0)
1089 
1090 			switch (sopt->sopt_name) {
1091 			case LOCAL_CREDS:
1092 				OPTSET(UNP_WANTCRED);
1093 				break;
1094 
1095 			case LOCAL_CONNWAIT:
1096 				OPTSET(UNP_CONNWAIT);
1097 				break;
1098 
1099 			default:
1100 				break;
1101 			}
1102 			break;
1103 #undef	OPTSET
1104 		default:
1105 			error = ENOPROTOOPT;
1106 			break;
1107 		}
1108 		break;
1109 
1110 	default:
1111 		error = EOPNOTSUPP;
1112 		break;
1113 	}
1114 	return (error);
1115 }
1116 
1117 static int
1118 unp_connect(struct socket *so, struct sockaddr *nam, struct thread *td)
1119 {
1120 	struct sockaddr_un *soun = (struct sockaddr_un *)nam;
1121 	struct vnode *vp;
1122 	struct socket *so2, *so3;
1123 	struct unpcb *unp, *unp2, *unp3;
1124 	int error, len;
1125 	struct nameidata nd;
1126 	char buf[SOCK_MAXADDRLEN];
1127 	struct sockaddr *sa;
1128 
1129 	UNP_GLOBAL_WLOCK_ASSERT();
1130 	UNP_GLOBAL_WUNLOCK();
1131 
1132 	unp = sotounpcb(so);
1133 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1134 
1135 	len = nam->sa_len - offsetof(struct sockaddr_un, sun_path);
1136 	if (len <= 0)
1137 		return (EINVAL);
1138 	strlcpy(buf, soun->sun_path, len + 1);
1139 
1140 	UNP_PCB_LOCK(unp);
1141 	if (unp->unp_flags & UNP_CONNECTING) {
1142 		UNP_PCB_UNLOCK(unp);
1143 		return (EALREADY);
1144 	}
1145 	unp->unp_flags |= UNP_CONNECTING;
1146 	UNP_PCB_UNLOCK(unp);
1147 
1148 	sa = malloc(sizeof(struct sockaddr_un), M_SONAME, M_WAITOK);
1149 	mtx_lock(&Giant);
1150 	NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, buf, td);
1151 	error = namei(&nd);
1152 	if (error)
1153 		vp = NULL;
1154 	else
1155 		vp = nd.ni_vp;
1156 	ASSERT_VOP_LOCKED(vp, "unp_connect");
1157 	NDFREE(&nd, NDF_ONLY_PNBUF);
1158 	if (error)
1159 		goto bad;
1160 
1161 	if (vp->v_type != VSOCK) {
1162 		error = ENOTSOCK;
1163 		goto bad;
1164 	}
1165 #ifdef MAC
1166 	error = mac_check_vnode_open(td->td_ucred, vp, VWRITE | VREAD);
1167 	if (error)
1168 		goto bad;
1169 #endif
1170 	error = VOP_ACCESS(vp, VWRITE, td->td_ucred, td);
1171 	if (error)
1172 		goto bad;
1173 	mtx_unlock(&Giant);
1174 
1175 	unp = sotounpcb(so);
1176 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1177 
1178 	/*
1179 	 * Lock global lock for two reasons: make sure v_socket is stable,
1180 	 * and to protect simultaneous locking of multiple pcbs.
1181 	 */
1182 	UNP_GLOBAL_WLOCK();
1183 	so2 = vp->v_socket;
1184 	if (so2 == NULL) {
1185 		error = ECONNREFUSED;
1186 		goto bad2;
1187 	}
1188 	if (so->so_type != so2->so_type) {
1189 		error = EPROTOTYPE;
1190 		goto bad2;
1191 	}
1192 	if (so->so_proto->pr_flags & PR_CONNREQUIRED) {
1193 		if (so2->so_options & SO_ACCEPTCONN) {
1194 			/*
1195 			 * We can't drop the global lock here or 'so2' may
1196 			 * become invalid, meaning that we will later recurse
1197 			 * back into the UNIX domain socket code while
1198 			 * holding the global lock.
1199 			 */
1200 			so3 = sonewconn(so2, 0);
1201 		} else
1202 			so3 = NULL;
1203 		if (so3 == NULL) {
1204 			error = ECONNREFUSED;
1205 			goto bad2;
1206 		}
1207 		unp = sotounpcb(so);
1208 		unp2 = sotounpcb(so2);
1209 		unp3 = sotounpcb(so3);
1210 		UNP_PCB_LOCK(unp);
1211 		UNP_PCB_LOCK(unp2);
1212 		UNP_PCB_LOCK(unp3);
1213 		if (unp2->unp_addr != NULL) {
1214 			bcopy(unp2->unp_addr, sa, unp2->unp_addr->sun_len);
1215 			unp3->unp_addr = (struct sockaddr_un *) sa;
1216 			sa = NULL;
1217 		}
1218 		/*
1219 		 * unp_peercred management:
1220 		 *
1221 		 * The connecter's (client's) credentials are copied from its
1222 		 * process structure at the time of connect() (which is now).
1223 		 */
1224 		cru2x(td->td_ucred, &unp3->unp_peercred);
1225 		unp3->unp_flags |= UNP_HAVEPC;
1226 		/*
1227 		 * The receiver's (server's) credentials are copied from the
1228 		 * unp_peercred member of socket on which the former called
1229 		 * listen(); uipc_listen() cached that process's credentials
1230 		 * at that time so we can use them now.
1231 		 */
1232 		KASSERT(unp2->unp_flags & UNP_HAVEPCCACHED,
1233 		    ("unp_connect: listener without cached peercred"));
1234 		memcpy(&unp->unp_peercred, &unp2->unp_peercred,
1235 		    sizeof(unp->unp_peercred));
1236 		unp->unp_flags |= UNP_HAVEPC;
1237 		if (unp2->unp_flags & UNP_WANTCRED)
1238 			unp3->unp_flags |= UNP_WANTCRED;
1239 		UNP_PCB_UNLOCK(unp3);
1240 		UNP_PCB_UNLOCK(unp2);
1241 		UNP_PCB_UNLOCK(unp);
1242 #ifdef MAC
1243 		SOCK_LOCK(so);
1244 		mac_set_socket_peer_from_socket(so, so3);
1245 		mac_set_socket_peer_from_socket(so3, so);
1246 		SOCK_UNLOCK(so);
1247 #endif
1248 
1249 		so2 = so3;
1250 	}
1251 	unp = sotounpcb(so);
1252 	KASSERT(unp != NULL, ("unp_connect: unp == NULL"));
1253 	unp2 = sotounpcb(so2);
1254 	KASSERT(unp2 != NULL, ("unp_connect: unp2 == NULL"));
1255 	UNP_PCB_LOCK(unp);
1256 	UNP_PCB_LOCK(unp2);
1257 	error = unp_connect2(so, so2, PRU_CONNECT);
1258 	UNP_PCB_UNLOCK(unp2);
1259 	UNP_PCB_UNLOCK(unp);
1260 bad2:
1261 	UNP_GLOBAL_WUNLOCK();
1262 	mtx_lock(&Giant);
1263 bad:
1264 	mtx_assert(&Giant, MA_OWNED);
1265 	if (vp != NULL)
1266 		vput(vp);
1267 	mtx_unlock(&Giant);
1268 	free(sa, M_SONAME);
1269 	UNP_GLOBAL_WLOCK();
1270 	UNP_PCB_LOCK(unp);
1271 	unp->unp_flags &= ~UNP_CONNECTING;
1272 	UNP_PCB_UNLOCK(unp);
1273 	return (error);
1274 }
1275 
1276 static int
1277 unp_connect2(struct socket *so, struct socket *so2, int req)
1278 {
1279 	struct unpcb *unp;
1280 	struct unpcb *unp2;
1281 
1282 	unp = sotounpcb(so);
1283 	KASSERT(unp != NULL, ("unp_connect2: unp == NULL"));
1284 	unp2 = sotounpcb(so2);
1285 	KASSERT(unp2 != NULL, ("unp_connect2: unp2 == NULL"));
1286 
1287 	UNP_GLOBAL_WLOCK_ASSERT();
1288 	UNP_PCB_LOCK_ASSERT(unp);
1289 	UNP_PCB_LOCK_ASSERT(unp2);
1290 
1291 	if (so2->so_type != so->so_type)
1292 		return (EPROTOTYPE);
1293 	unp->unp_conn = unp2;
1294 
1295 	switch (so->so_type) {
1296 	case SOCK_DGRAM:
1297 		LIST_INSERT_HEAD(&unp2->unp_refs, unp, unp_reflink);
1298 		soisconnected(so);
1299 		break;
1300 
1301 	case SOCK_STREAM:
1302 		unp2->unp_conn = unp;
1303 		if (req == PRU_CONNECT &&
1304 		    ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT))
1305 			soisconnecting(so);
1306 		else
1307 			soisconnected(so);
1308 		soisconnected(so2);
1309 		break;
1310 
1311 	default:
1312 		panic("unp_connect2");
1313 	}
1314 	return (0);
1315 }
1316 
1317 static void
1318 unp_disconnect(struct unpcb *unp, struct unpcb *unp2)
1319 {
1320 	struct socket *so;
1321 
1322 	KASSERT(unp2 != NULL, ("unp_disconnect: unp2 == NULL"));
1323 
1324 	UNP_GLOBAL_WLOCK_ASSERT();
1325 	UNP_PCB_LOCK_ASSERT(unp);
1326 	UNP_PCB_LOCK_ASSERT(unp2);
1327 
1328 	unp->unp_conn = NULL;
1329 	switch (unp->unp_socket->so_type) {
1330 	case SOCK_DGRAM:
1331 		LIST_REMOVE(unp, unp_reflink);
1332 		so = unp->unp_socket;
1333 		SOCK_LOCK(so);
1334 		so->so_state &= ~SS_ISCONNECTED;
1335 		SOCK_UNLOCK(so);
1336 		break;
1337 
1338 	case SOCK_STREAM:
1339 		soisdisconnected(unp->unp_socket);
1340 		unp2->unp_conn = NULL;
1341 		soisdisconnected(unp2->unp_socket);
1342 		break;
1343 	}
1344 }
1345 
1346 /*
1347  * unp_pcblist() assumes that UNIX domain socket memory is never reclaimed by
1348  * the zone (UMA_ZONE_NOFREE), and as such potentially stale pointers are
1349  * safe to reference.  It first scans the list of struct unpcb's to generate
1350  * a pointer list, then it rescans its list one entry at a time to
1351  * externalize and copyout.  It checks the generation number to see if a
1352  * struct unpcb has been reused, and will skip it if so.
1353  */
1354 static int
1355 unp_pcblist(SYSCTL_HANDLER_ARGS)
1356 {
1357 	int error, i, n;
1358 	int freeunp;
1359 	struct unpcb *unp, **unp_list;
1360 	unp_gen_t gencnt;
1361 	struct xunpgen *xug;
1362 	struct unp_head *head;
1363 	struct xunpcb *xu;
1364 
1365 	head = ((intptr_t)arg1 == SOCK_DGRAM ? &unp_dhead : &unp_shead);
1366 
1367 	/*
1368 	 * The process of preparing the PCB list is too time-consuming and
1369 	 * resource-intensive to repeat twice on every request.
1370 	 */
1371 	if (req->oldptr == NULL) {
1372 		n = unp_count;
1373 		req->oldidx = 2 * (sizeof *xug)
1374 			+ (n + n/8) * sizeof(struct xunpcb);
1375 		return (0);
1376 	}
1377 
1378 	if (req->newptr != NULL)
1379 		return (EPERM);
1380 
1381 	/*
1382 	 * OK, now we're committed to doing something.
1383 	 */
1384 	xug = malloc(sizeof(*xug), M_TEMP, M_WAITOK);
1385 	UNP_GLOBAL_RLOCK();
1386 	gencnt = unp_gencnt;
1387 	n = unp_count;
1388 	UNP_GLOBAL_RUNLOCK();
1389 
1390 	xug->xug_len = sizeof *xug;
1391 	xug->xug_count = n;
1392 	xug->xug_gen = gencnt;
1393 	xug->xug_sogen = so_gencnt;
1394 	error = SYSCTL_OUT(req, xug, sizeof *xug);
1395 	if (error) {
1396 		free(xug, M_TEMP);
1397 		return (error);
1398 	}
1399 
1400 	unp_list = malloc(n * sizeof *unp_list, M_TEMP, M_WAITOK);
1401 
1402 	/*
1403 	 * XXXRW: Note, this code relies very explicitly in pcb's being type
1404 	 * stable.
1405 	 */
1406 	UNP_GLOBAL_RLOCK();
1407 	for (unp = LIST_FIRST(head), i = 0; unp && i < n;
1408 	     unp = LIST_NEXT(unp, unp_link)) {
1409 		UNP_PCB_LOCK(unp);
1410 		if (unp->unp_gencnt <= gencnt) {
1411 			if (cr_cansee(req->td->td_ucred,
1412 			    unp->unp_socket->so_cred)) {
1413 				UNP_PCB_UNLOCK(unp);
1414 				continue;
1415 			}
1416 			unp_list[i++] = unp;
1417 			unp->unp_refcount++;
1418 		}
1419 		UNP_PCB_UNLOCK(unp);
1420 	}
1421 	UNP_GLOBAL_RUNLOCK();
1422 	n = i;			/* In case we lost some during malloc. */
1423 
1424 	/*
1425 	 * XXXRW: The logic below asumes that it is OK to lock a mutex in
1426 	 * an unpcb that may have been freed.
1427 	 */
1428 	error = 0;
1429 	xu = malloc(sizeof(*xu), M_TEMP, M_WAITOK | M_ZERO);
1430 	for (i = 0; i < n; i++) {
1431 		unp = unp_list[i];
1432 		UNP_PCB_LOCK(unp);
1433 		unp->unp_refcount--;
1434 	        if (unp->unp_refcount != 0 && unp->unp_gencnt <= gencnt) {
1435 			xu->xu_len = sizeof *xu;
1436 			xu->xu_unpp = unp;
1437 			/*
1438 			 * XXX - need more locking here to protect against
1439 			 * connect/disconnect races for SMP.
1440 			 */
1441 			if (unp->unp_addr != NULL)
1442 				bcopy(unp->unp_addr, &xu->xu_addr,
1443 				      unp->unp_addr->sun_len);
1444 			if (unp->unp_conn != NULL &&
1445 			    unp->unp_conn->unp_addr != NULL)
1446 				bcopy(unp->unp_conn->unp_addr,
1447 				      &xu->xu_caddr,
1448 				      unp->unp_conn->unp_addr->sun_len);
1449 			bcopy(unp, &xu->xu_unp, sizeof *unp);
1450 			sotoxsocket(unp->unp_socket, &xu->xu_socket);
1451 			UNP_PCB_UNLOCK(unp);
1452 			error = SYSCTL_OUT(req, xu, sizeof *xu);
1453 		} else {
1454 			freeunp = (unp->unp_refcount == 0);
1455 			UNP_PCB_UNLOCK(unp);
1456 			if (freeunp) {
1457 				UNP_PCB_LOCK_DESTROY(unp);
1458 				uma_zfree(unp_zone, unp);
1459 			}
1460 		}
1461 	}
1462 	free(xu, M_TEMP);
1463 	if (!error) {
1464 		/*
1465 		 * Give the user an updated idea of our state.  If the
1466 		 * generation differs from what we told her before, she knows
1467 		 * that something happened while we were processing this
1468 		 * request, and it might be necessary to retry.
1469 		 */
1470 		xug->xug_gen = unp_gencnt;
1471 		xug->xug_sogen = so_gencnt;
1472 		xug->xug_count = unp_count;
1473 		error = SYSCTL_OUT(req, xug, sizeof *xug);
1474 	}
1475 	free(unp_list, M_TEMP);
1476 	free(xug, M_TEMP);
1477 	return (error);
1478 }
1479 
1480 SYSCTL_PROC(_net_local_dgram, OID_AUTO, pcblist, CTLFLAG_RD,
1481 	    (caddr_t)(long)SOCK_DGRAM, 0, unp_pcblist, "S,xunpcb",
1482 	    "List of active local datagram sockets");
1483 SYSCTL_PROC(_net_local_stream, OID_AUTO, pcblist, CTLFLAG_RD,
1484 	    (caddr_t)(long)SOCK_STREAM, 0, unp_pcblist, "S,xunpcb",
1485 	    "List of active local stream sockets");
1486 
1487 static void
1488 unp_shutdown(struct unpcb *unp)
1489 {
1490 	struct unpcb *unp2;
1491 	struct socket *so;
1492 
1493 	UNP_GLOBAL_WLOCK_ASSERT();
1494 	UNP_PCB_LOCK_ASSERT(unp);
1495 
1496 	unp2 = unp->unp_conn;
1497 	if (unp->unp_socket->so_type == SOCK_STREAM && unp2 != NULL) {
1498 		so = unp2->unp_socket;
1499 		if (so != NULL)
1500 			socantrcvmore(so);
1501 	}
1502 }
1503 
1504 static void
1505 unp_drop(struct unpcb *unp, int errno)
1506 {
1507 	struct socket *so = unp->unp_socket;
1508 	struct unpcb *unp2;
1509 
1510 	UNP_GLOBAL_WLOCK_ASSERT();
1511 	UNP_PCB_LOCK_ASSERT(unp);
1512 
1513 	so->so_error = errno;
1514 	unp2 = unp->unp_conn;
1515 	if (unp2 == NULL)
1516 		return;
1517 
1518 	UNP_PCB_LOCK(unp2);
1519 	unp_disconnect(unp, unp2);
1520 	UNP_PCB_UNLOCK(unp2);
1521 }
1522 
1523 static void
1524 unp_freerights(struct file **rp, int fdcount)
1525 {
1526 	int i;
1527 	struct file *fp;
1528 
1529 	for (i = 0; i < fdcount; i++) {
1530 		/*
1531 		 * Zero the pointer before calling unp_discard since it may
1532 		 * end up in unp_gc()..
1533 		 *
1534 		 * XXXRW: This is less true than it used to be.
1535 		 */
1536 		fp = *rp;
1537 		*rp++ = NULL;
1538 		unp_discard(fp);
1539 	}
1540 }
1541 
1542 int
1543 unp_externalize(struct mbuf *control, struct mbuf **controlp)
1544 {
1545 	struct thread *td = curthread;		/* XXX */
1546 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1547 	int i;
1548 	int *fdp;
1549 	struct file **rp;
1550 	struct file *fp;
1551 	void *data;
1552 	socklen_t clen = control->m_len, datalen;
1553 	int error, newfds;
1554 	int f;
1555 	u_int newlen;
1556 
1557 	UNP_GLOBAL_UNLOCK_ASSERT();
1558 
1559 	error = 0;
1560 	if (controlp != NULL) /* controlp == NULL => free control messages */
1561 		*controlp = NULL;
1562 
1563 	while (cm != NULL) {
1564 		if (sizeof(*cm) > clen || cm->cmsg_len > clen) {
1565 			error = EINVAL;
1566 			break;
1567 		}
1568 
1569 		data = CMSG_DATA(cm);
1570 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1571 
1572 		if (cm->cmsg_level == SOL_SOCKET
1573 		    && cm->cmsg_type == SCM_RIGHTS) {
1574 			newfds = datalen / sizeof(struct file *);
1575 			rp = data;
1576 
1577 			/* If we're not outputting the descriptors free them. */
1578 			if (error || controlp == NULL) {
1579 				unp_freerights(rp, newfds);
1580 				goto next;
1581 			}
1582 			FILEDESC_XLOCK(td->td_proc->p_fd);
1583 			/* if the new FD's will not fit free them.  */
1584 			if (!fdavail(td, newfds)) {
1585 				FILEDESC_XUNLOCK(td->td_proc->p_fd);
1586 				error = EMSGSIZE;
1587 				unp_freerights(rp, newfds);
1588 				goto next;
1589 			}
1590 			/*
1591 			 * Now change each pointer to an fd in the global
1592 			 * table to an integer that is the index to the local
1593 			 * fd table entry that we set up to point to the
1594 			 * global one we are transferring.
1595 			 */
1596 			newlen = newfds * sizeof(int);
1597 			*controlp = sbcreatecontrol(NULL, newlen,
1598 			    SCM_RIGHTS, SOL_SOCKET);
1599 			if (*controlp == NULL) {
1600 				FILEDESC_XUNLOCK(td->td_proc->p_fd);
1601 				error = E2BIG;
1602 				unp_freerights(rp, newfds);
1603 				goto next;
1604 			}
1605 
1606 			fdp = (int *)
1607 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1608 			for (i = 0; i < newfds; i++) {
1609 				if (fdalloc(td, 0, &f))
1610 					panic("unp_externalize fdalloc failed");
1611 				fp = *rp++;
1612 				td->td_proc->p_fd->fd_ofiles[f] = fp;
1613 				FILE_LOCK(fp);
1614 				fp->f_msgcount--;
1615 				FILE_UNLOCK(fp);
1616 				unp_rights--;
1617 				*fdp++ = f;
1618 			}
1619 			FILEDESC_XUNLOCK(td->td_proc->p_fd);
1620 		} else {
1621 			/* We can just copy anything else across. */
1622 			if (error || controlp == NULL)
1623 				goto next;
1624 			*controlp = sbcreatecontrol(NULL, datalen,
1625 			    cm->cmsg_type, cm->cmsg_level);
1626 			if (*controlp == NULL) {
1627 				error = ENOBUFS;
1628 				goto next;
1629 			}
1630 			bcopy(data,
1631 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *)),
1632 			    datalen);
1633 		}
1634 
1635 		controlp = &(*controlp)->m_next;
1636 
1637 next:
1638 		if (CMSG_SPACE(datalen) < clen) {
1639 			clen -= CMSG_SPACE(datalen);
1640 			cm = (struct cmsghdr *)
1641 			    ((caddr_t)cm + CMSG_SPACE(datalen));
1642 		} else {
1643 			clen = 0;
1644 			cm = NULL;
1645 		}
1646 	}
1647 
1648 	m_freem(control);
1649 
1650 	return (error);
1651 }
1652 
1653 static void
1654 unp_zone_change(void *tag)
1655 {
1656 
1657 	uma_zone_set_max(unp_zone, maxsockets);
1658 }
1659 
1660 void
1661 unp_init(void)
1662 {
1663 
1664 	unp_zone = uma_zcreate("unpcb", sizeof(struct unpcb), NULL, NULL,
1665 	    NULL, NULL, UMA_ALIGN_PTR, 0);
1666 	if (unp_zone == NULL)
1667 		panic("unp_init");
1668 	uma_zone_set_max(unp_zone, maxsockets);
1669 	EVENTHANDLER_REGISTER(maxsockets_change, unp_zone_change,
1670 	    NULL, EVENTHANDLER_PRI_ANY);
1671 	LIST_INIT(&unp_dhead);
1672 	LIST_INIT(&unp_shead);
1673 	TASK_INIT(&unp_gc_task, 0, unp_gc, NULL);
1674 	UNP_GLOBAL_LOCK_INIT();
1675 }
1676 
1677 static int
1678 unp_internalize(struct mbuf **controlp, struct thread *td)
1679 {
1680 	struct mbuf *control = *controlp;
1681 	struct proc *p = td->td_proc;
1682 	struct filedesc *fdescp = p->p_fd;
1683 	struct cmsghdr *cm = mtod(control, struct cmsghdr *);
1684 	struct cmsgcred *cmcred;
1685 	struct file **rp;
1686 	struct file *fp;
1687 	struct timeval *tv;
1688 	int i, fd, *fdp;
1689 	void *data;
1690 	socklen_t clen = control->m_len, datalen;
1691 	int error, oldfds;
1692 	u_int newlen;
1693 
1694 	UNP_GLOBAL_UNLOCK_ASSERT();
1695 
1696 	error = 0;
1697 	*controlp = NULL;
1698 
1699 	while (cm != NULL) {
1700 		if (sizeof(*cm) > clen || cm->cmsg_level != SOL_SOCKET
1701 		    || cm->cmsg_len > clen) {
1702 			error = EINVAL;
1703 			goto out;
1704 		}
1705 
1706 		data = CMSG_DATA(cm);
1707 		datalen = (caddr_t)cm + cm->cmsg_len - (caddr_t)data;
1708 
1709 		switch (cm->cmsg_type) {
1710 		/*
1711 		 * Fill in credential information.
1712 		 */
1713 		case SCM_CREDS:
1714 			*controlp = sbcreatecontrol(NULL, sizeof(*cmcred),
1715 			    SCM_CREDS, SOL_SOCKET);
1716 			if (*controlp == NULL) {
1717 				error = ENOBUFS;
1718 				goto out;
1719 			}
1720 
1721 			cmcred = (struct cmsgcred *)
1722 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1723 			cmcred->cmcred_pid = p->p_pid;
1724 			cmcred->cmcred_uid = td->td_ucred->cr_ruid;
1725 			cmcred->cmcred_gid = td->td_ucred->cr_rgid;
1726 			cmcred->cmcred_euid = td->td_ucred->cr_uid;
1727 			cmcred->cmcred_ngroups = MIN(td->td_ucred->cr_ngroups,
1728 							CMGROUP_MAX);
1729 			for (i = 0; i < cmcred->cmcred_ngroups; i++)
1730 				cmcred->cmcred_groups[i] =
1731 				    td->td_ucred->cr_groups[i];
1732 			break;
1733 
1734 		case SCM_RIGHTS:
1735 			oldfds = datalen / sizeof (int);
1736 			/*
1737 			 * Check that all the FDs passed in refer to legal
1738 			 * files.  If not, reject the entire operation.
1739 			 */
1740 			fdp = data;
1741 			FILEDESC_SLOCK(fdescp);
1742 			for (i = 0; i < oldfds; i++) {
1743 				fd = *fdp++;
1744 				if ((unsigned)fd >= fdescp->fd_nfiles ||
1745 				    fdescp->fd_ofiles[fd] == NULL) {
1746 					FILEDESC_SUNLOCK(fdescp);
1747 					error = EBADF;
1748 					goto out;
1749 				}
1750 				fp = fdescp->fd_ofiles[fd];
1751 				if (!(fp->f_ops->fo_flags & DFLAG_PASSABLE)) {
1752 					FILEDESC_SUNLOCK(fdescp);
1753 					error = EOPNOTSUPP;
1754 					goto out;
1755 				}
1756 
1757 			}
1758 
1759 			/*
1760 			 * Now replace the integer FDs with pointers to
1761 			 * the associated global file table entry..
1762 			 */
1763 			newlen = oldfds * sizeof(struct file *);
1764 			*controlp = sbcreatecontrol(NULL, newlen,
1765 			    SCM_RIGHTS, SOL_SOCKET);
1766 			if (*controlp == NULL) {
1767 				FILEDESC_SUNLOCK(fdescp);
1768 				error = E2BIG;
1769 				goto out;
1770 			}
1771 
1772 			fdp = data;
1773 			rp = (struct file **)
1774 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1775 			for (i = 0; i < oldfds; i++) {
1776 				fp = fdescp->fd_ofiles[*fdp++];
1777 				*rp++ = fp;
1778 				FILE_LOCK(fp);
1779 				fp->f_count++;
1780 				fp->f_msgcount++;
1781 				FILE_UNLOCK(fp);
1782 				unp_rights++;
1783 			}
1784 			FILEDESC_SUNLOCK(fdescp);
1785 			break;
1786 
1787 		case SCM_TIMESTAMP:
1788 			*controlp = sbcreatecontrol(NULL, sizeof(*tv),
1789 			    SCM_TIMESTAMP, SOL_SOCKET);
1790 			if (*controlp == NULL) {
1791 				error = ENOBUFS;
1792 				goto out;
1793 			}
1794 			tv = (struct timeval *)
1795 			    CMSG_DATA(mtod(*controlp, struct cmsghdr *));
1796 			microtime(tv);
1797 			break;
1798 
1799 		default:
1800 			error = EINVAL;
1801 			goto out;
1802 		}
1803 
1804 		controlp = &(*controlp)->m_next;
1805 
1806 		if (CMSG_SPACE(datalen) < clen) {
1807 			clen -= CMSG_SPACE(datalen);
1808 			cm = (struct cmsghdr *)
1809 			    ((caddr_t)cm + CMSG_SPACE(datalen));
1810 		} else {
1811 			clen = 0;
1812 			cm = NULL;
1813 		}
1814 	}
1815 
1816 out:
1817 	m_freem(control);
1818 
1819 	return (error);
1820 }
1821 
1822 static struct mbuf *
1823 unp_addsockcred(struct thread *td, struct mbuf *control)
1824 {
1825 	struct mbuf *m, *n, *n_prev;
1826 	struct sockcred *sc;
1827 	const struct cmsghdr *cm;
1828 	int ngroups;
1829 	int i;
1830 
1831 	ngroups = MIN(td->td_ucred->cr_ngroups, CMGROUP_MAX);
1832 
1833 	m = sbcreatecontrol(NULL, SOCKCREDSIZE(ngroups), SCM_CREDS, SOL_SOCKET);
1834 	if (m == NULL)
1835 		return (control);
1836 
1837 	sc = (struct sockcred *) CMSG_DATA(mtod(m, struct cmsghdr *));
1838 	sc->sc_uid = td->td_ucred->cr_ruid;
1839 	sc->sc_euid = td->td_ucred->cr_uid;
1840 	sc->sc_gid = td->td_ucred->cr_rgid;
1841 	sc->sc_egid = td->td_ucred->cr_gid;
1842 	sc->sc_ngroups = ngroups;
1843 	for (i = 0; i < sc->sc_ngroups; i++)
1844 		sc->sc_groups[i] = td->td_ucred->cr_groups[i];
1845 
1846 	/*
1847 	 * Unlink SCM_CREDS control messages (struct cmsgcred), since just
1848 	 * created SCM_CREDS control message (struct sockcred) has another
1849 	 * format.
1850 	 */
1851 	if (control != NULL)
1852 		for (n = control, n_prev = NULL; n != NULL;) {
1853 			cm = mtod(n, struct cmsghdr *);
1854     			if (cm->cmsg_level == SOL_SOCKET &&
1855 			    cm->cmsg_type == SCM_CREDS) {
1856     				if (n_prev == NULL)
1857 					control = n->m_next;
1858 				else
1859 					n_prev->m_next = n->m_next;
1860 				n = m_free(n);
1861 			} else {
1862 				n_prev = n;
1863 				n = n->m_next;
1864 			}
1865 		}
1866 
1867 	/* Prepend it to the head. */
1868 	m->m_next = control;
1869 
1870 	return (m);
1871 }
1872 
1873 /*
1874  * unp_defer indicates whether additional work has been defered for a future
1875  * pass through unp_gc().  It is thread local and does not require explicit
1876  * synchronization.
1877  */
1878 static int	unp_defer;
1879 
1880 static int unp_taskcount;
1881 SYSCTL_INT(_net_local, OID_AUTO, taskcount, CTLFLAG_RD, &unp_taskcount, 0, "");
1882 
1883 static int unp_recycled;
1884 SYSCTL_INT(_net_local, OID_AUTO, recycled, CTLFLAG_RD, &unp_recycled, 0, "");
1885 
1886 static void
1887 unp_gc(__unused void *arg, int pending)
1888 {
1889 	struct file *fp, *nextfp;
1890 	struct socket *so;
1891 	struct file **extra_ref, **fpp;
1892 	int nunref, i;
1893 	int nfiles_snap;
1894 	int nfiles_slack = 20;
1895 
1896 	unp_taskcount++;
1897 	unp_defer = 0;
1898 	/*
1899 	 * Before going through all this, set all FDs to be NOT deferred and
1900 	 * NOT externally accessible.
1901 	 */
1902 	sx_slock(&filelist_lock);
1903 	LIST_FOREACH(fp, &filehead, f_list)
1904 		fp->f_gcflag &= ~(FMARK|FDEFER);
1905 	do {
1906 		KASSERT(unp_defer >= 0, ("unp_gc: unp_defer %d", unp_defer));
1907 		LIST_FOREACH(fp, &filehead, f_list) {
1908 			FILE_LOCK(fp);
1909 			/*
1910 			 * If the file is not open, skip it -- could be a
1911 			 * file in the process of being opened, or in the
1912 			 * process of being closed.  If the file is
1913 			 * "closing", it may have been marked for deferred
1914 			 * consideration.  Clear the flag now if so.
1915 			 */
1916 			if (fp->f_count == 0) {
1917 				if (fp->f_gcflag & FDEFER)
1918 					unp_defer--;
1919 				fp->f_gcflag &= ~(FMARK|FDEFER);
1920 				FILE_UNLOCK(fp);
1921 				continue;
1922 			}
1923 			/*
1924 			 * If we already marked it as 'defer' in a
1925 			 * previous pass, then try to process it this
1926 			 * time and un-mark it.
1927 			 */
1928 			if (fp->f_gcflag & FDEFER) {
1929 				fp->f_gcflag &= ~FDEFER;
1930 				unp_defer--;
1931 			} else {
1932 				/*
1933 				 * If it's not deferred, then check if it's
1934 				 * already marked.. if so skip it
1935 				 */
1936 				if (fp->f_gcflag & FMARK) {
1937 					FILE_UNLOCK(fp);
1938 					continue;
1939 				}
1940 				/*
1941 				 * If all references are from messages in
1942 				 * transit, then skip it. it's not externally
1943 				 * accessible.
1944 				 */
1945 				if (fp->f_count == fp->f_msgcount) {
1946 					FILE_UNLOCK(fp);
1947 					continue;
1948 				}
1949 				/*
1950 				 * If it got this far then it must be
1951 				 * externally accessible.
1952 				 */
1953 				fp->f_gcflag |= FMARK;
1954 			}
1955 			/*
1956 			 * Either it was deferred, or it is externally
1957 			 * accessible and not already marked so.  Now check
1958 			 * if it is possibly one of OUR sockets.
1959 			 */
1960 			if (fp->f_type != DTYPE_SOCKET ||
1961 			    (so = fp->f_data) == NULL) {
1962 				FILE_UNLOCK(fp);
1963 				continue;
1964 			}
1965 			if (so->so_proto->pr_domain != &localdomain ||
1966 			    (so->so_proto->pr_flags & PR_RIGHTS) == 0) {
1967 				FILE_UNLOCK(fp);
1968 				continue;
1969 			}
1970 
1971 			/*
1972 			 * Tell any other threads that do a subsequent
1973 			 * fdrop() that we are scanning the message
1974 			 * buffers.
1975 			 */
1976 			fp->f_gcflag |= FWAIT;
1977 			FILE_UNLOCK(fp);
1978 
1979 			/*
1980 			 * So, Ok, it's one of our sockets and it IS
1981 			 * externally accessible (or was deferred).  Now we
1982 			 * look to see if we hold any file descriptors in its
1983 			 * message buffers. Follow those links and mark them
1984 			 * as accessible too.
1985 			 */
1986 			SOCKBUF_LOCK(&so->so_rcv);
1987 			unp_scan(so->so_rcv.sb_mb, unp_mark);
1988 			SOCKBUF_UNLOCK(&so->so_rcv);
1989 
1990 			/*
1991 			 * Wake up any threads waiting in fdrop().
1992 			 */
1993 			FILE_LOCK(fp);
1994 			fp->f_gcflag &= ~FWAIT;
1995 			wakeup(&fp->f_gcflag);
1996 			FILE_UNLOCK(fp);
1997 		}
1998 	} while (unp_defer);
1999 	sx_sunlock(&filelist_lock);
2000 	/*
2001 	 * XXXRW: The following comments need updating for a post-SMPng and
2002 	 * deferred unp_gc() world, but are still generally accurate.
2003 	 *
2004 	 * We grab an extra reference to each of the file table entries that
2005 	 * are not otherwise accessible and then free the rights that are
2006 	 * stored in messages on them.
2007 	 *
2008 	 * The bug in the orginal code is a little tricky, so I'll describe
2009 	 * what's wrong with it here.
2010 	 *
2011 	 * It is incorrect to simply unp_discard each entry for f_msgcount
2012 	 * times -- consider the case of sockets A and B that contain
2013 	 * references to each other.  On a last close of some other socket,
2014 	 * we trigger a gc since the number of outstanding rights (unp_rights)
2015 	 * is non-zero.  If during the sweep phase the gc code unp_discards,
2016 	 * we end up doing a (full) closef on the descriptor.  A closef on A
2017 	 * results in the following chain.  Closef calls soo_close, which
2018 	 * calls soclose.   Soclose calls first (through the switch
2019 	 * uipc_usrreq) unp_detach, which re-invokes unp_gc.  Unp_gc simply
2020 	 * returns because the previous instance had set unp_gcing, and we
2021 	 * return all the way back to soclose, which marks the socket with
2022 	 * SS_NOFDREF, and then calls sofree.  Sofree calls sorflush to free
2023 	 * up the rights that are queued in messages on the socket A, i.e.,
2024 	 * the reference on B.  The sorflush calls via the dom_dispose switch
2025 	 * unp_dispose, which unp_scans with unp_discard.  This second
2026 	 * instance of unp_discard just calls closef on B.
2027 	 *
2028 	 * Well, a similar chain occurs on B, resulting in a sorflush on B,
2029 	 * which results in another closef on A.  Unfortunately, A is already
2030 	 * being closed, and the descriptor has already been marked with
2031 	 * SS_NOFDREF, and soclose panics at this point.
2032 	 *
2033 	 * Here, we first take an extra reference to each inaccessible
2034 	 * descriptor.  Then, we call sorflush ourself, since we know it is a
2035 	 * Unix domain socket anyhow.  After we destroy all the rights
2036 	 * carried in messages, we do a last closef to get rid of our extra
2037 	 * reference.  This is the last close, and the unp_detach etc will
2038 	 * shut down the socket.
2039 	 *
2040 	 * 91/09/19, bsy@cs.cmu.edu
2041 	 */
2042 again:
2043 	nfiles_snap = openfiles + nfiles_slack;	/* some slack */
2044 	extra_ref = malloc(nfiles_snap * sizeof(struct file *), M_TEMP,
2045 	    M_WAITOK);
2046 	sx_slock(&filelist_lock);
2047 	if (nfiles_snap < openfiles) {
2048 		sx_sunlock(&filelist_lock);
2049 		free(extra_ref, M_TEMP);
2050 		nfiles_slack += 20;
2051 		goto again;
2052 	}
2053 	for (nunref = 0, fp = LIST_FIRST(&filehead), fpp = extra_ref;
2054 	    fp != NULL; fp = nextfp) {
2055 		nextfp = LIST_NEXT(fp, f_list);
2056 		FILE_LOCK(fp);
2057 		/*
2058 		 * If it's not open, skip it
2059 		 */
2060 		if (fp->f_count == 0) {
2061 			FILE_UNLOCK(fp);
2062 			continue;
2063 		}
2064 		/*
2065 		 * If all refs are from msgs, and it's not marked accessible
2066 		 * then it must be referenced from some unreachable cycle of
2067 		 * (shut-down) FDs, so include it in our list of FDs to
2068 		 * remove.
2069 		 */
2070 		if (fp->f_count == fp->f_msgcount && !(fp->f_gcflag & FMARK)) {
2071 			*fpp++ = fp;
2072 			nunref++;
2073 			fp->f_count++;
2074 		}
2075 		FILE_UNLOCK(fp);
2076 	}
2077 	sx_sunlock(&filelist_lock);
2078 	/*
2079 	 * For each FD on our hit list, do the following two things:
2080 	 */
2081 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
2082 		struct file *tfp = *fpp;
2083 		FILE_LOCK(tfp);
2084 		if (tfp->f_type == DTYPE_SOCKET &&
2085 		    tfp->f_data != NULL) {
2086 			FILE_UNLOCK(tfp);
2087 			sorflush(tfp->f_data);
2088 		} else {
2089 			FILE_UNLOCK(tfp);
2090 		}
2091 	}
2092 	for (i = nunref, fpp = extra_ref; --i >= 0; ++fpp) {
2093 		closef(*fpp, (struct thread *) NULL);
2094 		unp_recycled++;
2095 	}
2096 	free(extra_ref, M_TEMP);
2097 }
2098 
2099 void
2100 unp_dispose(struct mbuf *m)
2101 {
2102 
2103 	if (m)
2104 		unp_scan(m, unp_discard);
2105 }
2106 
2107 static void
2108 unp_scan(struct mbuf *m0, void (*op)(struct file *))
2109 {
2110 	struct mbuf *m;
2111 	struct file **rp;
2112 	struct cmsghdr *cm;
2113 	void *data;
2114 	int i;
2115 	socklen_t clen, datalen;
2116 	int qfds;
2117 
2118 	while (m0 != NULL) {
2119 		for (m = m0; m; m = m->m_next) {
2120 			if (m->m_type != MT_CONTROL)
2121 				continue;
2122 
2123 			cm = mtod(m, struct cmsghdr *);
2124 			clen = m->m_len;
2125 
2126 			while (cm != NULL) {
2127 				if (sizeof(*cm) > clen || cm->cmsg_len > clen)
2128 					break;
2129 
2130 				data = CMSG_DATA(cm);
2131 				datalen = (caddr_t)cm + cm->cmsg_len
2132 				    - (caddr_t)data;
2133 
2134 				if (cm->cmsg_level == SOL_SOCKET &&
2135 				    cm->cmsg_type == SCM_RIGHTS) {
2136 					qfds = datalen / sizeof (struct file *);
2137 					rp = data;
2138 					for (i = 0; i < qfds; i++)
2139 						(*op)(*rp++);
2140 				}
2141 
2142 				if (CMSG_SPACE(datalen) < clen) {
2143 					clen -= CMSG_SPACE(datalen);
2144 					cm = (struct cmsghdr *)
2145 					    ((caddr_t)cm + CMSG_SPACE(datalen));
2146 				} else {
2147 					clen = 0;
2148 					cm = NULL;
2149 				}
2150 			}
2151 		}
2152 		m0 = m0->m_act;
2153 	}
2154 }
2155 
2156 static void
2157 unp_mark(struct file *fp)
2158 {
2159 
2160 	/* XXXRW: Should probably assert file list lock here. */
2161 
2162 	if (fp->f_gcflag & FMARK)
2163 		return;
2164 	unp_defer++;
2165 	fp->f_gcflag |= (FMARK|FDEFER);
2166 }
2167 
2168 static void
2169 unp_discard(struct file *fp)
2170 {
2171 
2172 	UNP_GLOBAL_WLOCK();
2173 	FILE_LOCK(fp);
2174 	fp->f_msgcount--;
2175 	unp_rights--;
2176 	FILE_UNLOCK(fp);
2177 	UNP_GLOBAL_WUNLOCK();
2178 	(void) closef(fp, (struct thread *)NULL);
2179 }
2180