xref: /illumos-gate/usr/src/uts/common/fs/sockfs/socksubr.c (revision ddb365bfc9e868ad24ccdcb0dc91af18b10df082)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 1995, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright 2016 Nexenta Systems, Inc.  All rights reserved.
25  * Copyright 2015, Joyent, Inc. All rights reserved.
26  * Copyright 2020 OmniOS Community Edition (OmniOSce) Association.
27  * Copyright 2022 Garrett D'Amore
28  */
29 
30 #include <sys/types.h>
31 #include <sys/t_lock.h>
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/buf.h>
35 #include <sys/conf.h>
36 #include <sys/cred.h>
37 #include <sys/kmem.h>
38 #include <sys/sysmacros.h>
39 #include <sys/vfs.h>
40 #include <sys/vfs_opreg.h>
41 #include <sys/vnode.h>
42 #include <sys/debug.h>
43 #include <sys/errno.h>
44 #include <sys/time.h>
45 #include <sys/file.h>
46 #include <sys/open.h>
47 #include <sys/user.h>
48 #include <sys/termios.h>
49 #include <sys/stream.h>
50 #include <sys/strsubr.h>
51 #include <sys/strsun.h>
52 #include <sys/esunddi.h>
53 #include <sys/flock.h>
54 #include <sys/modctl.h>
55 #include <sys/cmn_err.h>
56 #include <sys/mkdev.h>
57 #include <sys/pathname.h>
58 #include <sys/ddi.h>
59 #include <sys/stat.h>
60 #include <sys/fs/snode.h>
61 #include <sys/fs/dv_node.h>
62 #include <sys/zone.h>
63 
64 #include <sys/socket.h>
65 #include <sys/socketvar.h>
66 #include <netinet/in.h>
67 #include <sys/un.h>
68 #include <sys/ucred.h>
69 
70 #include <sys/tiuser.h>
71 #define	_SUN_TPI_VERSION	2
72 #include <sys/tihdr.h>
73 
74 #include <c2/audit.h>
75 
76 #include <fs/sockfs/sockcommon.h>
77 #include <fs/sockfs/sockfilter_impl.h>
78 #include <fs/sockfs/socktpi.h>
79 #include <fs/sockfs/socktpi_impl.h>
80 #include <fs/sockfs/sodirect.h>
81 
82 /*
83  * Macros that operate on struct cmsghdr.
84  * The CMSG_VALID macro does not assume that the last option buffer is padded.
85  */
86 #define	CMSG_CONTENT(cmsg)	(&((cmsg)[1]))
87 #define	CMSG_CONTENTLEN(cmsg)	((cmsg)->cmsg_len - sizeof (struct cmsghdr))
88 #define	CMSG_VALID(cmsg, start, end)					\
89 	(ISALIGNED_cmsghdr(cmsg) &&					\
90 	((uintptr_t)(cmsg) >= (uintptr_t)(start)) &&			\
91 	((uintptr_t)(cmsg) < (uintptr_t)(end)) &&			\
92 	((ssize_t)(cmsg)->cmsg_len >= sizeof (struct cmsghdr)) &&	\
93 	((uintptr_t)(cmsg) + (cmsg)->cmsg_len <= (uintptr_t)(end)))
94 #define	SO_LOCK_WAKEUP_TIME	3000	/* Wakeup time in milliseconds */
95 
96 dev_t sockdev;	/* For fsid in getattr */
97 
98 struct socklist socklist;
99 
100 struct kmem_cache *socket_cache;
101 
102 /*
103  * sockconf_lock protects the socket configuration (socket types and
104  * socket filters) which is changed via the sockconfig system call.
105  */
106 krwlock_t sockconf_lock;
107 
108 static int sockfs_update(kstat_t *, int);
109 static int sockfs_snapshot(kstat_t *, void *, int);
110 extern smod_info_t *sotpi_smod_create(void);
111 
112 extern void sendfile_init();
113 
114 extern int modrootloaded;
115 
116 /*
117  * Translate from a device pathname (e.g. "/dev/tcp") to a vnode.
118  * Returns with the vnode held.
119  */
120 int
121 sogetvp(char *devpath, vnode_t **vpp, int uioflag)
122 {
123 	struct snode *csp;
124 	vnode_t *vp, *dvp;
125 	major_t maj;
126 	int error;
127 
128 	ASSERT(uioflag == UIO_SYSSPACE || uioflag == UIO_USERSPACE);
129 
130 	/*
131 	 * Lookup the underlying filesystem vnode.
132 	 */
133 	error = lookupname(devpath, uioflag, FOLLOW, NULLVPP, &vp);
134 	if (error)
135 		return (error);
136 
137 	/* Check that it is the correct vnode */
138 	if (vp->v_type != VCHR) {
139 		VN_RELE(vp);
140 		return (ENOTSOCK);
141 	}
142 
143 	/*
144 	 * If devpath went through devfs, the device should already
145 	 * be configured. If devpath is a mknod file, however, we
146 	 * need to make sure the device is properly configured.
147 	 * To do this, we do something similar to spec_open()
148 	 * except that we resolve to the minor/leaf level since
149 	 * we need to return a vnode.
150 	 */
151 	csp = VTOS(VTOS(vp)->s_commonvp);
152 	if (!(csp->s_flag & SDIPSET)) {
153 		char *pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
154 		error = ddi_dev_pathname(vp->v_rdev, S_IFCHR, pathname);
155 		if (error == 0)
156 			error = devfs_lookupname(pathname, NULLVPP, &dvp);
157 		VN_RELE(vp);
158 		kmem_free(pathname, MAXPATHLEN);
159 		if (error != 0)
160 			return (ENXIO);
161 		vp = dvp;	/* use the devfs vp */
162 	}
163 
164 	/* device is configured at this point */
165 	maj = getmajor(vp->v_rdev);
166 	if (!STREAMSTAB(maj)) {
167 		VN_RELE(vp);
168 		return (ENOSTR);
169 	}
170 
171 	*vpp = vp;
172 	return (0);
173 }
174 
175 /*
176  * Update the accessed, updated, or changed times in an sonode
177  * with the current time.
178  *
179  * Note that both SunOS 4.X and 4.4BSD sockets do not present reasonable
180  * attributes in a fstat call. (They return the current time and 0 for
181  * all timestamps, respectively.) We maintain the current timestamps
182  * here primarily so that should sockmod be popped the resulting
183  * file descriptor will behave like a stream w.r.t. the timestamps.
184  */
185 void
186 so_update_attrs(struct sonode *so, int flag)
187 {
188 	time_t now = gethrestime_sec();
189 
190 	if (SOCK_IS_NONSTR(so))
191 		return;
192 
193 	mutex_enter(&so->so_lock);
194 	so->so_flag |= flag;
195 	if (flag & SOACC)
196 		SOTOTPI(so)->sti_atime = now;
197 	if (flag & SOMOD)
198 		SOTOTPI(so)->sti_mtime = now;
199 	mutex_exit(&so->so_lock);
200 }
201 
202 extern so_create_func_t sock_comm_create_function;
203 extern so_destroy_func_t sock_comm_destroy_function;
204 /*
205  * Init function called when sockfs is loaded.
206  */
207 int
208 sockinit(int fstype, char *name)
209 {
210 	static const fs_operation_def_t sock_vfsops_template[] = {
211 		NULL, NULL
212 	};
213 	int error;
214 	major_t dev;
215 	char *err_str;
216 
217 	error = vfs_setfsops(fstype, sock_vfsops_template, NULL);
218 	if (error != 0) {
219 		zcmn_err(GLOBAL_ZONEID, CE_WARN,
220 		    "sockinit: bad vfs ops template");
221 		return (error);
222 	}
223 
224 	error = vn_make_ops(name, socket_vnodeops_template,
225 	    &socket_vnodeops);
226 	if (error != 0) {
227 		err_str = "sockinit: bad socket vnode ops template";
228 		/* vn_make_ops() does not reset socktpi_vnodeops on failure. */
229 		socket_vnodeops = NULL;
230 		goto failure;
231 	}
232 
233 	socket_cache = kmem_cache_create("socket_cache",
234 	    sizeof (struct sonode), 0, sonode_constructor,
235 	    sonode_destructor, NULL, NULL, NULL, 0);
236 
237 	rw_init(&sockconf_lock, NULL, RW_DEFAULT, NULL);
238 
239 	error = socktpi_init();
240 	if (error != 0) {
241 		err_str = NULL;
242 		goto failure;
243 	}
244 
245 	error = sod_init();
246 	if (error != 0) {
247 		err_str = NULL;
248 		goto failure;
249 	}
250 
251 	/*
252 	 * Set up the default create and destroy functions
253 	 */
254 	sock_comm_create_function = socket_sonode_create;
255 	sock_comm_destroy_function = socket_sonode_destroy;
256 
257 	/*
258 	 * Build initial list mapping socket parameters to vnode.
259 	 */
260 	smod_init();
261 	smod_add(sotpi_smod_create());
262 
263 	sockparams_init();
264 
265 	/*
266 	 * If sockets are needed before init runs /sbin/soconfig
267 	 * it is possible to preload the sockparams list here using
268 	 * calls like:
269 	 *	sockconfig(1,2,3, "/dev/tcp", 0);
270 	 */
271 
272 	/*
273 	 * Create a unique dev_t for use in so_fsid.
274 	 */
275 
276 	if ((dev = getudev()) == (major_t)-1)
277 		dev = 0;
278 	sockdev = makedevice(dev, 0);
279 
280 	mutex_init(&socklist.sl_lock, NULL, MUTEX_DEFAULT, NULL);
281 	sendfile_init();
282 
283 	/* Initialize socket filters */
284 	sof_init();
285 
286 	return (0);
287 
288 failure:
289 	(void) vfs_freevfsops_by_type(fstype);
290 	if (socket_vnodeops != NULL)
291 		vn_freevnodeops(socket_vnodeops);
292 	if (err_str != NULL)
293 		zcmn_err(GLOBAL_ZONEID, CE_WARN, err_str);
294 	return (error);
295 }
296 
297 /*
298  * Caller must hold the mutex. Used to set SOLOCKED.
299  */
300 void
301 so_lock_single(struct sonode *so)
302 {
303 	ASSERT(MUTEX_HELD(&so->so_lock));
304 
305 	while (so->so_flag & (SOLOCKED | SOASYNC_UNBIND)) {
306 		cv_wait_stop(&so->so_single_cv, &so->so_lock,
307 		    SO_LOCK_WAKEUP_TIME);
308 	}
309 	so->so_flag |= SOLOCKED;
310 }
311 
312 /*
313  * Caller must hold the mutex and pass in SOLOCKED or SOASYNC_UNBIND.
314  * Used to clear SOLOCKED or SOASYNC_UNBIND.
315  */
316 void
317 so_unlock_single(struct sonode *so, int flag)
318 {
319 	ASSERT(MUTEX_HELD(&so->so_lock));
320 	ASSERT(flag & (SOLOCKED|SOASYNC_UNBIND));
321 	ASSERT((flag & ~(SOLOCKED|SOASYNC_UNBIND)) == 0);
322 	ASSERT(so->so_flag & flag);
323 	/*
324 	 * Process the T_DISCON_IND on sti_discon_ind_mp.
325 	 *
326 	 * Call to so_drain_discon_ind will result in so_lock
327 	 * being dropped and re-acquired later.
328 	 */
329 	if (!SOCK_IS_NONSTR(so)) {
330 		sotpi_info_t *sti = SOTOTPI(so);
331 
332 		if (sti->sti_discon_ind_mp != NULL)
333 			so_drain_discon_ind(so);
334 	}
335 
336 	cv_signal(&so->so_single_cv);
337 	so->so_flag &= ~flag;
338 }
339 
340 /*
341  * Caller must hold the mutex. Used to set SOREADLOCKED.
342  * If the caller wants nonblocking behavior it should set fmode.
343  */
344 int
345 so_lock_read(struct sonode *so, int fmode)
346 {
347 	ASSERT(MUTEX_HELD(&so->so_lock));
348 
349 	while (so->so_flag & SOREADLOCKED) {
350 		if (fmode & (FNDELAY|FNONBLOCK))
351 			return (EWOULDBLOCK);
352 		cv_wait_stop(&so->so_read_cv, &so->so_lock,
353 		    SO_LOCK_WAKEUP_TIME);
354 	}
355 	so->so_flag |= SOREADLOCKED;
356 	return (0);
357 }
358 
359 /*
360  * Like so_lock_read above but allows signals.
361  */
362 int
363 so_lock_read_intr(struct sonode *so, int fmode)
364 {
365 	ASSERT(MUTEX_HELD(&so->so_lock));
366 
367 	while (so->so_flag & SOREADLOCKED) {
368 		if (fmode & (FNDELAY|FNONBLOCK))
369 			return (EWOULDBLOCK);
370 		if (!cv_wait_sig(&so->so_read_cv, &so->so_lock))
371 			return (EINTR);
372 	}
373 	so->so_flag |= SOREADLOCKED;
374 	return (0);
375 }
376 
377 /*
378  * Caller must hold the mutex. Used to clear SOREADLOCKED,
379  * set in so_lock_read() or so_lock_read_intr().
380  */
381 void
382 so_unlock_read(struct sonode *so)
383 {
384 	ASSERT(MUTEX_HELD(&so->so_lock));
385 	ASSERT(so->so_flag & SOREADLOCKED);
386 
387 	cv_signal(&so->so_read_cv);
388 	so->so_flag &= ~SOREADLOCKED;
389 }
390 
391 /*
392  * Verify that the specified offset falls within the mblk and
393  * that the resulting pointer is aligned.
394  * Returns NULL if not.
395  */
396 void *
397 sogetoff(mblk_t *mp, t_uscalar_t offset,
398     t_uscalar_t length, uint_t align_size)
399 {
400 	uintptr_t ptr1, ptr2;
401 
402 	ASSERT(mp && mp->b_wptr >= mp->b_rptr);
403 	ptr1 = (uintptr_t)mp->b_rptr + offset;
404 	ptr2 = (uintptr_t)ptr1 + length;
405 	if (ptr1 < (uintptr_t)mp->b_rptr || ptr2 > (uintptr_t)mp->b_wptr) {
406 		eprintline(0);
407 		return (NULL);
408 	}
409 	if ((ptr1 & (align_size - 1)) != 0) {
410 		eprintline(0);
411 		return (NULL);
412 	}
413 	return ((void *)ptr1);
414 }
415 
416 /*
417  * Return the AF_UNIX underlying filesystem vnode matching a given name.
418  * Makes sure the sending and the destination sonodes are compatible.
419  * The vnode is returned held.
420  *
421  * The underlying filesystem VSOCK vnode has a v_stream pointer that
422  * references the actual stream head (hence indirectly the actual sonode).
423  */
424 static int
425 so_ux_lookup(struct sonode *so, struct sockaddr_un *soun, int checkaccess,
426     vnode_t **vpp)
427 {
428 	vnode_t		*vp;	/* Underlying filesystem vnode */
429 	vnode_t		*rvp;	/* real vnode */
430 	vnode_t		*svp;	/* sockfs vnode */
431 	struct sonode	*so2;
432 	int		error;
433 
434 	dprintso(so, 1, ("so_ux_lookup(%p) name <%s>\n", (void *)so,
435 	    soun->sun_path));
436 
437 	error = lookupname(soun->sun_path, UIO_SYSSPACE, FOLLOW, NULLVPP, &vp);
438 	if (error) {
439 		eprintsoline(so, error);
440 		return (error);
441 	}
442 
443 	/*
444 	 * Traverse lofs mounts get the real vnode
445 	 */
446 	if (VOP_REALVP(vp, &rvp, NULL) == 0) {
447 		VN_HOLD(rvp);		/* hold the real vnode */
448 		VN_RELE(vp);		/* release hold from lookup */
449 		vp = rvp;
450 	}
451 
452 	if (vp->v_type != VSOCK) {
453 		error = ENOTSOCK;
454 		eprintsoline(so, error);
455 		goto done2;
456 	}
457 
458 	if (checkaccess) {
459 		/*
460 		 * Check that we have permissions to access the destination
461 		 * vnode. This check is not done in BSD but it is required
462 		 * by X/Open.
463 		 */
464 		if (error = VOP_ACCESS(vp, VREAD|VWRITE, 0, CRED(), NULL)) {
465 			eprintsoline(so, error);
466 			goto done2;
467 		}
468 	}
469 
470 	/*
471 	 * Check if the remote socket has been closed.
472 	 *
473 	 * Synchronize with vn_rele_stream by holding v_lock while traversing
474 	 * v_stream->sd_vnode.
475 	 */
476 	mutex_enter(&vp->v_lock);
477 	if (vp->v_stream == NULL) {
478 		mutex_exit(&vp->v_lock);
479 		if (so->so_type == SOCK_DGRAM)
480 			error = EDESTADDRREQ;
481 		else
482 			error = ECONNREFUSED;
483 
484 		eprintsoline(so, error);
485 		goto done2;
486 	}
487 	ASSERT(vp->v_stream->sd_vnode);
488 	svp = vp->v_stream->sd_vnode;
489 	/*
490 	 * holding v_lock on underlying filesystem vnode and acquiring
491 	 * it on sockfs vnode. Assumes that no code ever attempts to
492 	 * acquire these locks in the reverse order.
493 	 */
494 	VN_HOLD(svp);
495 	mutex_exit(&vp->v_lock);
496 
497 	if (svp->v_type != VSOCK) {
498 		error = ENOTSOCK;
499 		eprintsoline(so, error);
500 		goto done;
501 	}
502 
503 	so2 = VTOSO(svp);
504 
505 	if (so->so_type != so2->so_type) {
506 		error = EPROTOTYPE;
507 		eprintsoline(so, error);
508 		goto done;
509 	}
510 
511 	VN_RELE(svp);
512 	*vpp = vp;
513 	return (0);
514 
515 done:
516 	VN_RELE(svp);
517 done2:
518 	VN_RELE(vp);
519 	return (error);
520 }
521 
522 /*
523  * Verify peer address for connect and sendto/sendmsg.
524  * Since sendto/sendmsg would not get synchronous errors from the transport
525  * provider we have to do these ugly checks in the socket layer to
526  * preserve compatibility with SunOS 4.X.
527  */
528 int
529 so_addr_verify(struct sonode *so, const struct sockaddr *name,
530     socklen_t namelen)
531 {
532 	int		family;
533 
534 	dprintso(so, 1, ("so_addr_verify(%p, %p, %d)\n",
535 	    (void *)so, (void *)name, namelen));
536 
537 	ASSERT(name != NULL);
538 
539 	family = so->so_family;
540 	switch (family) {
541 	case AF_INET:
542 		if (name->sa_family != family) {
543 			eprintsoline(so, EAFNOSUPPORT);
544 			return (EAFNOSUPPORT);
545 		}
546 		if (namelen != (socklen_t)sizeof (struct sockaddr_in)) {
547 			eprintsoline(so, EINVAL);
548 			return (EINVAL);
549 		}
550 		break;
551 	case AF_INET6: {
552 #ifdef DEBUG
553 		struct sockaddr_in6 *sin6;
554 #endif /* DEBUG */
555 
556 		if (name->sa_family != family) {
557 			eprintsoline(so, EAFNOSUPPORT);
558 			return (EAFNOSUPPORT);
559 		}
560 		if (namelen != (socklen_t)sizeof (struct sockaddr_in6)) {
561 			eprintsoline(so, EINVAL);
562 			return (EINVAL);
563 		}
564 #ifdef DEBUG
565 		/* Verify that apps don't forget to clear sin6_scope_id etc */
566 		sin6 = (struct sockaddr_in6 *)name;
567 		if (sin6->sin6_scope_id != 0 &&
568 		    !IN6_IS_ADDR_LINKSCOPE(&sin6->sin6_addr)) {
569 			zcmn_err(getzoneid(), CE_WARN,
570 			    "connect/send* with uninitialized sin6_scope_id "
571 			    "(%d) on socket. Pid = %d\n",
572 			    (int)sin6->sin6_scope_id, (int)curproc->p_pid);
573 		}
574 #endif /* DEBUG */
575 		break;
576 	}
577 	case AF_UNIX:
578 		if (SOTOTPI(so)->sti_faddr_noxlate) {
579 			return (0);
580 		}
581 		if (namelen < (socklen_t)sizeof (short)) {
582 			eprintsoline(so, ENOENT);
583 			return (ENOENT);
584 		}
585 		if (name->sa_family != family) {
586 			eprintsoline(so, EAFNOSUPPORT);
587 			return (EAFNOSUPPORT);
588 		}
589 		/* MAXPATHLEN + soun_family + nul termination */
590 		if (namelen > (socklen_t)(MAXPATHLEN + sizeof (short) + 1)) {
591 			eprintsoline(so, ENAMETOOLONG);
592 			return (ENAMETOOLONG);
593 		}
594 
595 		break;
596 
597 	default:
598 		/*
599 		 * Default is don't do any length or sa_family check
600 		 * to allow non-sockaddr style addresses.
601 		 */
602 		break;
603 	}
604 
605 	return (0);
606 }
607 
608 
609 /*
610  * Translate an AF_UNIX sockaddr_un to the transport internal name.
611  * Assumes caller has called so_addr_verify first.  The translated
612  * (internal form) address is stored in sti->sti_ux_taddr.
613  */
614 /*ARGSUSED*/
615 int
616 so_ux_addr_xlate(struct sonode *so, struct sockaddr *name,
617     socklen_t namelen, int checkaccess,
618     void **addrp, socklen_t *addrlenp)
619 {
620 	int			error;
621 	struct sockaddr_un	*soun;
622 	vnode_t			*vp;
623 	void			*addr;
624 	socklen_t		addrlen;
625 	sotpi_info_t		*sti = SOTOTPI(so);
626 
627 	dprintso(so, 1, ("so_ux_addr_xlate(%p, %p, %d, %d)\n",
628 	    (void *)so, (void *)name, namelen, checkaccess));
629 
630 	ASSERT(name != NULL);
631 	ASSERT(so->so_family == AF_UNIX);
632 	ASSERT(!sti->sti_faddr_noxlate);
633 	ASSERT(namelen >= (socklen_t)sizeof (short));
634 	ASSERT(name->sa_family == AF_UNIX);
635 	soun = (struct sockaddr_un *)name;
636 	/*
637 	 * Lookup vnode for the specified path name and verify that
638 	 * it is a socket.
639 	 */
640 	error = so_ux_lookup(so, soun, checkaccess, &vp);
641 	if (error) {
642 		eprintsoline(so, error);
643 		return (error);
644 	}
645 	/*
646 	 * Use the address of the peer vnode as the address to send
647 	 * to. We release the peer vnode here. In case it has been
648 	 * closed by the time the T_CONN_REQ or T_UNITDATA_REQ reaches the
649 	 * transport the message will get an error or be dropped.
650 	 * Note that that soua_vp is never dereferenced; it's just a
651 	 * convenient value by which we can identify the peer.
652 	 */
653 	sti->sti_ux_taddr.soua_vp = vp;
654 	sti->sti_ux_taddr.soua_magic = SOU_MAGIC_EXPLICIT;
655 	addr = &sti->sti_ux_taddr;
656 	addrlen = (socklen_t)sizeof (sti->sti_ux_taddr);
657 	dprintso(so, 1, ("ux_xlate UNIX: addrlen %d, vp %p\n",
658 	    addrlen, (void *)vp));
659 	VN_RELE(vp);
660 	*addrp = addr;
661 	*addrlenp = (socklen_t)addrlen;
662 	return (0);
663 }
664 
665 /*
666  * Esballoc free function for messages that contain SO_FILEP option.
667  * Decrement the reference count on the file pointers using closef.
668  */
669 void
670 fdbuf_free(struct fdbuf *fdbuf)
671 {
672 	int	i;
673 	struct file *fp;
674 
675 	dprint(1, ("fdbuf_free: %d fds\n", fdbuf->fd_numfd));
676 	for (i = 0; i < fdbuf->fd_numfd; i++) {
677 		/*
678 		 * We need pointer size alignment for fd_fds. On a LP64
679 		 * kernel, the required alignment is 8 bytes while
680 		 * the option headers and values are only 4 bytes
681 		 * aligned. So its safer to do a bcopy compared to
682 		 * assigning fdbuf->fd_fds[i] to fp.
683 		 */
684 		bcopy((char *)&fdbuf->fd_fds[i], (char *)&fp, sizeof (fp));
685 		dprint(1, ("fdbuf_free: [%d] = %p\n", i, (void *)fp));
686 		(void) closef(fp);
687 	}
688 	if (fdbuf->fd_ebuf != NULL)
689 		kmem_free(fdbuf->fd_ebuf, fdbuf->fd_ebuflen);
690 	kmem_free(fdbuf, fdbuf->fd_size);
691 }
692 
693 /*
694  * Allocate an esballoc'ed message for AF_UNIX file descriptor passing.
695  * Waits if memory is not available.
696  */
697 mblk_t *
698 fdbuf_allocmsg(int size, struct fdbuf *fdbuf)
699 {
700 	uchar_t	*buf;
701 	mblk_t	*mp;
702 
703 	dprint(1, ("fdbuf_allocmsg: size %d, %d fds\n", size, fdbuf->fd_numfd));
704 	buf = kmem_alloc(size, KM_SLEEP);
705 	fdbuf->fd_ebuf = (caddr_t)buf;
706 	fdbuf->fd_ebuflen = size;
707 	fdbuf->fd_frtn.free_func = fdbuf_free;
708 	fdbuf->fd_frtn.free_arg = (caddr_t)fdbuf;
709 
710 	mp = esballoc_wait(buf, size, BPRI_MED, &fdbuf->fd_frtn);
711 	mp->b_datap->db_type = M_PROTO;
712 	return (mp);
713 }
714 
715 /*
716  * Extract file descriptors from a fdbuf.
717  * Return list in rights/rightslen.
718  */
719 /*ARGSUSED*/
720 static int
721 fdbuf_extract(struct fdbuf *fdbuf, void *rights, int rightslen)
722 {
723 	int	i, fd;
724 	int	*rp;
725 	struct file *fp;
726 	int	numfd;
727 
728 	dprint(1, ("fdbuf_extract: %d fds, len %d\n",
729 	    fdbuf->fd_numfd, rightslen));
730 
731 	numfd = fdbuf->fd_numfd;
732 	ASSERT(rightslen == numfd * (int)sizeof (int));
733 
734 	/*
735 	 * Allocate a file descriptor and increment the f_count.
736 	 * The latter is needed since we always call fdbuf_free
737 	 * which performs a closef.
738 	 */
739 	rp = (int *)rights;
740 	for (i = 0; i < numfd; i++) {
741 		if ((fd = ufalloc(0)) == -1)
742 			goto cleanup;
743 		/*
744 		 * We need pointer size alignment for fd_fds. On a LP64
745 		 * kernel, the required alignment is 8 bytes while
746 		 * the option headers and values are only 4 bytes
747 		 * aligned. So its safer to do a bcopy compared to
748 		 * assigning fdbuf->fd_fds[i] to fp.
749 		 */
750 		bcopy((char *)&fdbuf->fd_fds[i], (char *)&fp, sizeof (fp));
751 		mutex_enter(&fp->f_tlock);
752 		fp->f_count++;
753 		mutex_exit(&fp->f_tlock);
754 		setf(fd, fp);
755 		*rp++ = fd;
756 		if (AU_AUDITING())
757 			audit_fdrecv(fd, fp);
758 		dprint(1, ("fdbuf_extract: [%d] = %d, %p refcnt %d\n",
759 		    i, fd, (void *)fp, fp->f_count));
760 	}
761 	return (0);
762 
763 cleanup:
764 	/*
765 	 * Undo whatever partial work the loop above has done.
766 	 */
767 	{
768 		int j;
769 
770 		rp = (int *)rights;
771 		for (j = 0; j < i; j++) {
772 			dprint(0,
773 			    ("fdbuf_extract: cleanup[%d] = %d\n", j, *rp));
774 			(void) closeandsetf(*rp++, NULL);
775 		}
776 	}
777 
778 	return (EMFILE);
779 }
780 
781 /*
782  * Insert file descriptors into an fdbuf.
783  * Returns a kmem_alloc'ed fdbuf. The fdbuf should be freed
784  * by calling fdbuf_free().
785  */
786 int
787 fdbuf_create(void *rights, int rightslen, struct fdbuf **fdbufp)
788 {
789 	int		numfd, i;
790 	int		*fds;
791 	struct file	*fp;
792 	struct fdbuf	*fdbuf;
793 	int		fdbufsize;
794 
795 	dprint(1, ("fdbuf_create: len %d\n", rightslen));
796 
797 	numfd = rightslen / (int)sizeof (int);
798 
799 	fdbufsize = (int)FDBUF_HDRSIZE + (numfd * (int)sizeof (struct file *));
800 	fdbuf = kmem_alloc(fdbufsize, KM_SLEEP);
801 	fdbuf->fd_size = fdbufsize;
802 	fdbuf->fd_numfd = 0;
803 	fdbuf->fd_ebuf = NULL;
804 	fdbuf->fd_ebuflen = 0;
805 	fds = (int *)rights;
806 	for (i = 0; i < numfd; i++) {
807 		if ((fp = getf(fds[i])) == NULL) {
808 			fdbuf_free(fdbuf);
809 			return (EBADF);
810 		}
811 		dprint(1, ("fdbuf_create: [%d] = %d, %p refcnt %d\n",
812 		    i, fds[i], (void *)fp, fp->f_count));
813 		mutex_enter(&fp->f_tlock);
814 		fp->f_count++;
815 		mutex_exit(&fp->f_tlock);
816 		/*
817 		 * The maximum alignment for fdbuf (or any option header
818 		 * and its value) it 4 bytes. On a LP64 kernel, the alignment
819 		 * is not sufficient for pointers (fd_fds in this case). Since
820 		 * we just did a kmem_alloc (we get a double word alignment),
821 		 * we don't need to do anything on the send side (we loose
822 		 * the double word alignment because fdbuf goes after an
823 		 * option header (eg T_unitdata_req) which is only 4 byte
824 		 * aligned). We take care of this when we extract the file
825 		 * descriptor in fdbuf_extract or fdbuf_free.
826 		 */
827 		fdbuf->fd_fds[i] = fp;
828 		fdbuf->fd_numfd++;
829 		releasef(fds[i]);
830 		if (AU_AUDITING())
831 			audit_fdsend(fds[i], fp, 0);
832 	}
833 	*fdbufp = fdbuf;
834 	return (0);
835 }
836 
837 static int
838 fdbuf_optlen(int rightslen)
839 {
840 	int numfd;
841 
842 	numfd = rightslen / (int)sizeof (int);
843 
844 	return ((int)FDBUF_HDRSIZE + (numfd * (int)sizeof (struct file *)));
845 }
846 
847 static t_uscalar_t
848 fdbuf_cmsglen(int fdbuflen)
849 {
850 	return (t_uscalar_t)((fdbuflen - FDBUF_HDRSIZE) /
851 	    (int)sizeof (struct file *) * (int)sizeof (int));
852 }
853 
854 
855 /*
856  * Return non-zero if the mblk and fdbuf are consistent.
857  */
858 static int
859 fdbuf_verify(mblk_t *mp, struct fdbuf *fdbuf, int fdbuflen)
860 {
861 	if (fdbuflen >= FDBUF_HDRSIZE &&
862 	    fdbuflen == fdbuf->fd_size) {
863 		frtn_t *frp = mp->b_datap->db_frtnp;
864 		/*
865 		 * Check that the SO_FILEP portion of the
866 		 * message has not been modified by
867 		 * the loopback transport. The sending sockfs generates
868 		 * a message that is esballoc'ed with the free function
869 		 * being fdbuf_free() and where free_arg contains the
870 		 * identical information as the SO_FILEP content.
871 		 *
872 		 * If any of these constraints are not satisfied we
873 		 * silently ignore the option.
874 		 */
875 		ASSERT(mp);
876 		if (frp != NULL &&
877 		    frp->free_func == fdbuf_free &&
878 		    frp->free_arg != NULL &&
879 		    bcmp(frp->free_arg, fdbuf, fdbuflen) == 0) {
880 			dprint(1, ("fdbuf_verify: fdbuf %p len %d\n",
881 			    (void *)fdbuf, fdbuflen));
882 			return (1);
883 		} else {
884 			zcmn_err(getzoneid(), CE_WARN,
885 			    "sockfs: mismatched fdbuf content (%p)",
886 			    (void *)mp);
887 			return (0);
888 		}
889 	} else {
890 		zcmn_err(getzoneid(), CE_WARN,
891 		    "sockfs: mismatched fdbuf len %d, %d\n",
892 		    fdbuflen, fdbuf->fd_size);
893 		return (0);
894 	}
895 }
896 
897 /*
898  * When the file descriptors returned by sorecvmsg can not be passed
899  * to the application this routine will cleanup the references on
900  * the files. Start at startoff bytes into the buffer.
901  */
902 static void
903 close_fds(void *fdbuf, int fdbuflen, int startoff)
904 {
905 	int *fds = (int *)fdbuf;
906 	int numfd = fdbuflen / (int)sizeof (int);
907 	int i;
908 
909 	dprint(1, ("close_fds(%p, %d, %d)\n", fdbuf, fdbuflen, startoff));
910 
911 	for (i = 0; i < numfd; i++) {
912 		if (startoff < 0)
913 			startoff = 0;
914 		if (startoff < (int)sizeof (int)) {
915 			/*
916 			 * This file descriptor is partially or fully after
917 			 * the offset
918 			 */
919 			dprint(0,
920 			    ("close_fds: cleanup[%d] = %d\n", i, fds[i]));
921 			(void) closeandsetf(fds[i], NULL);
922 		}
923 		startoff -= (int)sizeof (int);
924 	}
925 }
926 
927 /*
928  * Close all file descriptors contained in the control part starting at
929  * the startoffset.
930  */
931 void
932 so_closefds(void *control, t_uscalar_t controllen, int oldflg,
933     int startoff)
934 {
935 	struct cmsghdr *cmsg;
936 
937 	if (control == NULL)
938 		return;
939 
940 	if (oldflg) {
941 		close_fds(control, controllen, startoff);
942 		return;
943 	}
944 	/* Scan control part for file descriptors. */
945 	for (cmsg = (struct cmsghdr *)control;
946 	    CMSG_VALID(cmsg, control, (uintptr_t)control + controllen);
947 	    cmsg = CMSG_NEXT(cmsg)) {
948 		if (cmsg->cmsg_level == SOL_SOCKET &&
949 		    cmsg->cmsg_type == SCM_RIGHTS) {
950 			close_fds(CMSG_CONTENT(cmsg),
951 			    (int)CMSG_CONTENTLEN(cmsg),
952 			    startoff - (int)sizeof (struct cmsghdr));
953 		}
954 		startoff -= ROUNDUP_cmsglen(cmsg->cmsg_len);
955 	}
956 }
957 
958 /*
959  * Handle truncation of a cmsg when the receive buffer is not big enough.
960  * Adjust the cmsg_len header field in the last cmsg that will be included in
961  * the buffer to reflect the number of bytes included.
962  */
963 void
964 so_truncatecmsg(void *control, t_uscalar_t controllen, uint_t maxlen)
965 {
966 	struct cmsghdr *cmsg;
967 	uint_t len = 0;
968 
969 	if (control == NULL)
970 		return;
971 
972 	for (cmsg = control;
973 	    CMSG_VALID(cmsg, control, (uintptr_t)control + controllen);
974 	    cmsg = CMSG_NEXT(cmsg)) {
975 
976 		len += ROUNDUP_cmsglen(cmsg->cmsg_len);
977 
978 		if (len > maxlen) {
979 			/*
980 			 * This cmsg is the last one that will be included in
981 			 * the truncated buffer.
982 			 */
983 			socklen_t diff = len - maxlen;
984 
985 			if (diff < CMSG_CONTENTLEN(cmsg)) {
986 				dprint(1, ("so_truncatecmsg: %d -> %d\n",
987 				    cmsg->cmsg_len, cmsg->cmsg_len - diff));
988 				cmsg->cmsg_len -= diff;
989 			} else {
990 				cmsg->cmsg_len = sizeof (struct cmsghdr);
991 			}
992 			break;
993 		}
994 	}
995 }
996 
997 /*
998  * Returns a pointer/length for the file descriptors contained
999  * in the control buffer. Returns with *fdlenp == -1 if there are no
1000  * file descriptor options present. This is different than there being
1001  * a zero-length file descriptor option.
1002  * Fail if there are multiple SCM_RIGHT cmsgs.
1003  */
1004 int
1005 so_getfdopt(void *control, t_uscalar_t controllen, int oldflg,
1006     void **fdsp, int *fdlenp)
1007 {
1008 	struct cmsghdr *cmsg;
1009 	void *fds;
1010 	int fdlen;
1011 
1012 	if (control == NULL) {
1013 		*fdsp = NULL;
1014 		*fdlenp = -1;
1015 		return (0);
1016 	}
1017 
1018 	if (oldflg) {
1019 		*fdsp = control;
1020 		if (controllen == 0)
1021 			*fdlenp = -1;
1022 		else
1023 			*fdlenp = controllen;
1024 		dprint(1, ("so_getfdopt: old %d\n", *fdlenp));
1025 		return (0);
1026 	}
1027 
1028 	fds = NULL;
1029 	fdlen = 0;
1030 
1031 	for (cmsg = (struct cmsghdr *)control;
1032 	    CMSG_VALID(cmsg, control, (uintptr_t)control + controllen);
1033 	    cmsg = CMSG_NEXT(cmsg)) {
1034 		if (cmsg->cmsg_level == SOL_SOCKET &&
1035 		    cmsg->cmsg_type == SCM_RIGHTS) {
1036 			if (fds != NULL)
1037 				return (EINVAL);
1038 			fds = CMSG_CONTENT(cmsg);
1039 			fdlen = (int)CMSG_CONTENTLEN(cmsg);
1040 			dprint(1, ("so_getfdopt: new %lu\n",
1041 			    (size_t)CMSG_CONTENTLEN(cmsg)));
1042 		}
1043 	}
1044 	if (fds == NULL) {
1045 		dprint(1, ("so_getfdopt: NONE\n"));
1046 		*fdlenp = -1;
1047 	} else
1048 		*fdlenp = fdlen;
1049 	*fdsp = fds;
1050 	return (0);
1051 }
1052 
1053 /*
1054  * Return the length of the options including any file descriptor options.
1055  */
1056 t_uscalar_t
1057 so_optlen(void *control, t_uscalar_t controllen, int oldflg)
1058 {
1059 	struct cmsghdr *cmsg;
1060 	t_uscalar_t optlen = 0;
1061 	t_uscalar_t len;
1062 
1063 	if (control == NULL)
1064 		return (0);
1065 
1066 	if (oldflg)
1067 		return ((t_uscalar_t)(sizeof (struct T_opthdr) +
1068 		    fdbuf_optlen(controllen)));
1069 
1070 	for (cmsg = (struct cmsghdr *)control;
1071 	    CMSG_VALID(cmsg, control, (uintptr_t)control + controllen);
1072 	    cmsg = CMSG_NEXT(cmsg)) {
1073 		if (cmsg->cmsg_level == SOL_SOCKET &&
1074 		    cmsg->cmsg_type == SCM_RIGHTS) {
1075 			len = fdbuf_optlen((int)CMSG_CONTENTLEN(cmsg));
1076 		} else {
1077 			len = (t_uscalar_t)CMSG_CONTENTLEN(cmsg);
1078 		}
1079 		optlen += (t_uscalar_t)(_TPI_ALIGN_TOPT(len) +
1080 		    sizeof (struct T_opthdr));
1081 	}
1082 	dprint(1, ("so_optlen: controllen %d, flg %d -> optlen %d\n",
1083 	    controllen, oldflg, optlen));
1084 	return (optlen);
1085 }
1086 
1087 /*
1088  * Copy options from control to the mblk. Skip any file descriptor options.
1089  */
1090 void
1091 so_cmsg2opt(void *control, t_uscalar_t controllen, int oldflg, mblk_t *mp)
1092 {
1093 	struct T_opthdr toh;
1094 	struct cmsghdr *cmsg;
1095 
1096 	if (control == NULL)
1097 		return;
1098 
1099 	if (oldflg) {
1100 		/* No real options - caller has handled file descriptors */
1101 		return;
1102 	}
1103 	for (cmsg = (struct cmsghdr *)control;
1104 	    CMSG_VALID(cmsg, control, (uintptr_t)control + controllen);
1105 	    cmsg = CMSG_NEXT(cmsg)) {
1106 		/*
1107 		 * Note: The caller handles file descriptors prior
1108 		 * to calling this function.
1109 		 */
1110 		t_uscalar_t len;
1111 
1112 		if (cmsg->cmsg_level == SOL_SOCKET &&
1113 		    cmsg->cmsg_type == SCM_RIGHTS)
1114 			continue;
1115 
1116 		len = (t_uscalar_t)CMSG_CONTENTLEN(cmsg);
1117 		toh.level = cmsg->cmsg_level;
1118 		toh.name = cmsg->cmsg_type;
1119 		toh.len = len + (t_uscalar_t)sizeof (struct T_opthdr);
1120 		toh.status = 0;
1121 
1122 		soappendmsg(mp, &toh, sizeof (toh));
1123 		soappendmsg(mp, CMSG_CONTENT(cmsg), len);
1124 		mp->b_wptr += _TPI_ALIGN_TOPT(len) - len;
1125 		ASSERT(mp->b_wptr <= mp->b_datap->db_lim);
1126 	}
1127 }
1128 
1129 /*
1130  * Return the length of the control message derived from the options.
1131  * Exclude SO_SRCADDR and SO_UNIX_CLOSE options. Include SO_FILEP.
1132  * When oldflg is set only include SO_FILEP.
1133  * so_opt2cmsg and so_cmsglen are inter-related since so_cmsglen
1134  * allocates the space that so_opt2cmsg fills. If one changes, the other should
1135  * also be checked for any possible impacts.
1136  */
1137 t_uscalar_t
1138 so_cmsglen(mblk_t *mp, void *opt, t_uscalar_t optlen, int oldflg)
1139 {
1140 	t_uscalar_t cmsglen = 0;
1141 	struct T_opthdr *tohp;
1142 	t_uscalar_t len;
1143 	t_uscalar_t last_roundup = 0;
1144 
1145 	ASSERT(__TPI_TOPT_ISALIGNED(opt));
1146 
1147 	for (tohp = (struct T_opthdr *)opt;
1148 	    tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen);
1149 	    tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) {
1150 		dprint(1, ("so_cmsglen: level 0x%x, name %d, len %d\n",
1151 		    tohp->level, tohp->name, tohp->len));
1152 		if (tohp->level == SOL_SOCKET &&
1153 		    (tohp->name == SO_SRCADDR ||
1154 		    tohp->name == SO_UNIX_CLOSE)) {
1155 			continue;
1156 		}
1157 		if (tohp->level == SOL_SOCKET && tohp->name == SO_FILEP) {
1158 			struct fdbuf *fdbuf;
1159 			int fdbuflen;
1160 
1161 			fdbuf = (struct fdbuf *)_TPI_TOPT_DATA(tohp);
1162 			fdbuflen = (int)_TPI_TOPT_DATALEN(tohp);
1163 
1164 			if (!fdbuf_verify(mp, fdbuf, fdbuflen))
1165 				continue;
1166 			if (oldflg) {
1167 				cmsglen += fdbuf_cmsglen(fdbuflen);
1168 				continue;
1169 			}
1170 			len = fdbuf_cmsglen(fdbuflen);
1171 		} else if (tohp->level == SOL_SOCKET &&
1172 		    tohp->name == SCM_TIMESTAMP) {
1173 			if (oldflg)
1174 				continue;
1175 
1176 			if (get_udatamodel() == DATAMODEL_NATIVE) {
1177 				len = sizeof (struct timeval);
1178 			} else {
1179 				len = sizeof (struct timeval32);
1180 			}
1181 		} else {
1182 			if (oldflg)
1183 				continue;
1184 			len = (t_uscalar_t)_TPI_TOPT_DATALEN(tohp);
1185 		}
1186 		/*
1187 		 * Exclude roundup for last option to not set
1188 		 * MSG_CTRUNC when the cmsg fits but the padding doesn't fit.
1189 		 */
1190 		last_roundup = (t_uscalar_t)
1191 		    (ROUNDUP_cmsglen(len + (int)sizeof (struct cmsghdr)) -
1192 		    (len + (int)sizeof (struct cmsghdr)));
1193 		cmsglen += (t_uscalar_t)(len + (int)sizeof (struct cmsghdr)) +
1194 		    last_roundup;
1195 	}
1196 	cmsglen -= last_roundup;
1197 	dprint(1, ("so_cmsglen: optlen %d, flg %d -> cmsglen %d\n",
1198 	    optlen, oldflg, cmsglen));
1199 	return (cmsglen);
1200 }
1201 
1202 /*
1203  * Copy options from options to the control. Convert SO_FILEP to
1204  * file descriptors.
1205  * Returns errno or zero.
1206  * so_opt2cmsg and so_cmsglen are inter-related since so_cmsglen
1207  * allocates the space that so_opt2cmsg fills. If one changes, the other should
1208  * also be checked for any possible impacts.
1209  */
1210 int
1211 so_opt2cmsg(mblk_t *mp, void *opt, t_uscalar_t optlen, int oldflg,
1212     void *control, t_uscalar_t controllen)
1213 {
1214 	struct T_opthdr *tohp;
1215 	struct cmsghdr *cmsg;
1216 	struct fdbuf *fdbuf;
1217 	int fdbuflen;
1218 	int error;
1219 #if defined(DEBUG) || defined(__lint)
1220 	struct cmsghdr *cend = (struct cmsghdr *)
1221 	    (((uint8_t *)control) + ROUNDUP_cmsglen(controllen));
1222 #endif
1223 	cmsg = (struct cmsghdr *)control;
1224 
1225 	ASSERT(__TPI_TOPT_ISALIGNED(opt));
1226 
1227 	for (tohp = (struct T_opthdr *)opt;
1228 	    tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen);
1229 	    tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) {
1230 		dprint(1, ("so_opt2cmsg: level 0x%x, name %d, len %d\n",
1231 		    tohp->level, tohp->name, tohp->len));
1232 
1233 		if (tohp->level == SOL_SOCKET &&
1234 		    (tohp->name == SO_SRCADDR ||
1235 		    tohp->name == SO_UNIX_CLOSE)) {
1236 			continue;
1237 		}
1238 		ASSERT((uintptr_t)cmsg <= (uintptr_t)control + controllen);
1239 		if (tohp->level == SOL_SOCKET && tohp->name == SO_FILEP) {
1240 			fdbuf = (struct fdbuf *)_TPI_TOPT_DATA(tohp);
1241 			fdbuflen = (int)_TPI_TOPT_DATALEN(tohp);
1242 
1243 			if (!fdbuf_verify(mp, fdbuf, fdbuflen))
1244 				return (EPROTO);
1245 			if (oldflg) {
1246 				error = fdbuf_extract(fdbuf, control,
1247 				    (int)controllen);
1248 				if (error != 0)
1249 					return (error);
1250 				continue;
1251 			} else {
1252 				int fdlen;
1253 
1254 				fdlen = (int)fdbuf_cmsglen(
1255 				    (int)_TPI_TOPT_DATALEN(tohp));
1256 
1257 				cmsg->cmsg_level = tohp->level;
1258 				cmsg->cmsg_type = SCM_RIGHTS;
1259 				cmsg->cmsg_len = (socklen_t)(fdlen +
1260 				    sizeof (struct cmsghdr));
1261 
1262 				error = fdbuf_extract(fdbuf,
1263 				    CMSG_CONTENT(cmsg), fdlen);
1264 				if (error != 0)
1265 					return (error);
1266 			}
1267 		} else if (tohp->level == SOL_SOCKET &&
1268 		    tohp->name == SCM_TIMESTAMP) {
1269 			timestruc_t *timestamp;
1270 
1271 			if (oldflg)
1272 				continue;
1273 
1274 			cmsg->cmsg_level = tohp->level;
1275 			cmsg->cmsg_type = tohp->name;
1276 
1277 			timestamp =
1278 			    (timestruc_t *)P2ROUNDUP((intptr_t)&tohp[1],
1279 			    sizeof (intptr_t));
1280 
1281 			if (get_udatamodel() == DATAMODEL_NATIVE) {
1282 				struct timeval tv;
1283 
1284 				cmsg->cmsg_len = sizeof (struct timeval) +
1285 				    sizeof (struct cmsghdr);
1286 				tv.tv_sec = timestamp->tv_sec;
1287 				tv.tv_usec = timestamp->tv_nsec /
1288 				    (NANOSEC / MICROSEC);
1289 				/*
1290 				 * on LP64 systems, the struct timeval in
1291 				 * the destination will not be 8-byte aligned,
1292 				 * so use bcopy to avoid alignment trouble
1293 				 */
1294 				bcopy(&tv, CMSG_CONTENT(cmsg), sizeof (tv));
1295 			} else {
1296 				struct timeval32 *time32;
1297 
1298 				cmsg->cmsg_len = sizeof (struct timeval32) +
1299 				    sizeof (struct cmsghdr);
1300 				time32 = (struct timeval32 *)CMSG_CONTENT(cmsg);
1301 				time32->tv_sec = (time32_t)timestamp->tv_sec;
1302 				time32->tv_usec =
1303 				    (int32_t)(timestamp->tv_nsec /
1304 				    (NANOSEC / MICROSEC));
1305 			}
1306 
1307 		} else {
1308 			if (oldflg)
1309 				continue;
1310 
1311 			cmsg->cmsg_level = tohp->level;
1312 			cmsg->cmsg_type = tohp->name;
1313 			cmsg->cmsg_len = (socklen_t)sizeof (struct cmsghdr);
1314 			if (tohp->level == IPPROTO_IP &&
1315 			    (tohp->name == IP_RECVTOS ||
1316 			    tohp->name == IP_RECVTTL)) {
1317 				/*
1318 				 * The data for these is a uint8_t but, in
1319 				 * order to maintain alignment for any
1320 				 * following TPI primitives in the message,
1321 				 * there will be some trailing padding bytes
1322 				 * which are included in the TPI_TOPT_DATALEN.
1323 				 * For these types, we set the cmsg_len
1324 				 * explicitly to the correct value.
1325 				 */
1326 				cmsg->cmsg_len += (socklen_t)sizeof (uint8_t);
1327 			} else {
1328 				cmsg->cmsg_len +=
1329 				    (socklen_t)(_TPI_TOPT_DATALEN(tohp));
1330 			}
1331 
1332 			/* copy content to control data part */
1333 			bcopy(&tohp[1], CMSG_CONTENT(cmsg),
1334 			    CMSG_CONTENTLEN(cmsg));
1335 		}
1336 		/* move to next CMSG structure! */
1337 		cmsg = CMSG_NEXT(cmsg);
1338 	}
1339 	dprint(1, ("so_opt2cmsg: buf %p len %d; cend %p; final cmsg %p\n",
1340 	    control, controllen, (void *)cend, (void *)cmsg));
1341 	ASSERT(cmsg <= cend);
1342 	return (0);
1343 }
1344 
1345 /*
1346  * Extract the SO_SRCADDR option value if present.
1347  */
1348 void
1349 so_getopt_srcaddr(void *opt, t_uscalar_t optlen, void **srcp,
1350     t_uscalar_t *srclenp)
1351 {
1352 	struct T_opthdr		*tohp;
1353 
1354 	ASSERT(__TPI_TOPT_ISALIGNED(opt));
1355 
1356 	ASSERT(srcp != NULL && srclenp != NULL);
1357 	*srcp = NULL;
1358 	*srclenp = 0;
1359 
1360 	for (tohp = (struct T_opthdr *)opt;
1361 	    tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen);
1362 	    tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) {
1363 		dprint(1, ("so_getopt_srcaddr: level 0x%x, name %d, len %d\n",
1364 		    tohp->level, tohp->name, tohp->len));
1365 		if (tohp->level == SOL_SOCKET &&
1366 		    tohp->name == SO_SRCADDR) {
1367 			*srcp = _TPI_TOPT_DATA(tohp);
1368 			*srclenp = (t_uscalar_t)_TPI_TOPT_DATALEN(tohp);
1369 		}
1370 	}
1371 }
1372 
1373 /*
1374  * Verify if the SO_UNIX_CLOSE option is present.
1375  */
1376 int
1377 so_getopt_unix_close(void *opt, t_uscalar_t optlen)
1378 {
1379 	struct T_opthdr		*tohp;
1380 
1381 	ASSERT(__TPI_TOPT_ISALIGNED(opt));
1382 
1383 	for (tohp = (struct T_opthdr *)opt;
1384 	    tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen);
1385 	    tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) {
1386 		dprint(1,
1387 		    ("so_getopt_unix_close: level 0x%x, name %d, len %d\n",
1388 		    tohp->level, tohp->name, tohp->len));
1389 		if (tohp->level == SOL_SOCKET &&
1390 		    tohp->name == SO_UNIX_CLOSE)
1391 			return (1);
1392 	}
1393 	return (0);
1394 }
1395 
1396 /*
1397  * Allocate an M_PROTO message.
1398  *
1399  * If allocation fails the behavior depends on sleepflg:
1400  *	_ALLOC_NOSLEEP	fail immediately
1401  *	_ALLOC_INTR	sleep for memory until a signal is caught
1402  *	_ALLOC_SLEEP	sleep forever. Don't return NULL.
1403  */
1404 mblk_t *
1405 soallocproto(size_t size, int sleepflg, cred_t *cr)
1406 {
1407 	mblk_t	*mp;
1408 
1409 	/* Round up size for reuse */
1410 	size = MAX(size, 64);
1411 	if (cr != NULL)
1412 		mp = allocb_cred(size, cr, curproc->p_pid);
1413 	else
1414 		mp = allocb(size, BPRI_MED);
1415 
1416 	if (mp == NULL) {
1417 		int error;	/* Dummy - error not returned to caller */
1418 
1419 		switch (sleepflg) {
1420 		case _ALLOC_SLEEP:
1421 			if (cr != NULL) {
1422 				mp = allocb_cred_wait(size, STR_NOSIG, &error,
1423 				    cr, curproc->p_pid);
1424 			} else {
1425 				mp = allocb_wait(size, BPRI_MED, STR_NOSIG,
1426 				    &error);
1427 			}
1428 			ASSERT(mp);
1429 			break;
1430 		case _ALLOC_INTR:
1431 			if (cr != NULL) {
1432 				mp = allocb_cred_wait(size, 0, &error, cr,
1433 				    curproc->p_pid);
1434 			} else {
1435 				mp = allocb_wait(size, BPRI_MED, 0, &error);
1436 			}
1437 			if (mp == NULL) {
1438 				/* Caught signal while sleeping for memory */
1439 				eprintline(ENOBUFS);
1440 				return (NULL);
1441 			}
1442 			break;
1443 		case _ALLOC_NOSLEEP:
1444 		default:
1445 			eprintline(ENOBUFS);
1446 			return (NULL);
1447 		}
1448 	}
1449 	DB_TYPE(mp) = M_PROTO;
1450 	return (mp);
1451 }
1452 
1453 /*
1454  * Allocate an M_PROTO message with a single component.
1455  * len is the length of buf. size is the amount to allocate.
1456  *
1457  * buf can be NULL with a non-zero len.
1458  * This results in a bzero'ed chunk being placed the message.
1459  */
1460 mblk_t *
1461 soallocproto1(const void *buf, ssize_t len, ssize_t size, int sleepflg,
1462     cred_t *cr)
1463 {
1464 	mblk_t	*mp;
1465 
1466 	if (size == 0)
1467 		size = len;
1468 
1469 	ASSERT(size >= len);
1470 	/* Round up size for reuse */
1471 	size = MAX(size, 64);
1472 	mp = soallocproto(size, sleepflg, cr);
1473 	if (mp == NULL)
1474 		return (NULL);
1475 	mp->b_datap->db_type = M_PROTO;
1476 	if (len != 0) {
1477 		if (buf != NULL)
1478 			bcopy(buf, mp->b_wptr, len);
1479 		else
1480 			bzero(mp->b_wptr, len);
1481 		mp->b_wptr += len;
1482 	}
1483 	return (mp);
1484 }
1485 
1486 /*
1487  * Append buf/len to mp.
1488  * The caller has to ensure that there is enough room in the mblk.
1489  *
1490  * buf can be NULL with a non-zero len.
1491  * This results in a bzero'ed chunk being placed the message.
1492  */
1493 void
1494 soappendmsg(mblk_t *mp, const void *buf, ssize_t len)
1495 {
1496 	ASSERT(mp);
1497 
1498 	if (len != 0) {
1499 		/* Assert for room left */
1500 		ASSERT(mp->b_datap->db_lim - mp->b_wptr >= len);
1501 		if (buf != NULL)
1502 			bcopy(buf, mp->b_wptr, len);
1503 		else
1504 			bzero(mp->b_wptr, len);
1505 	}
1506 	mp->b_wptr += len;
1507 }
1508 
1509 /*
1510  * Create a message using two kernel buffers.
1511  * If size is set that will determine the allocation size (e.g. for future
1512  * soappendmsg calls). If size is zero it is derived from the buffer
1513  * lengths.
1514  */
1515 mblk_t *
1516 soallocproto2(const void *buf1, ssize_t len1, const void *buf2, ssize_t len2,
1517     ssize_t size, int sleepflg, cred_t *cr)
1518 {
1519 	mblk_t *mp;
1520 
1521 	if (size == 0)
1522 		size = len1 + len2;
1523 	ASSERT(size >= len1 + len2);
1524 
1525 	mp = soallocproto1(buf1, len1, size, sleepflg, cr);
1526 	if (mp)
1527 		soappendmsg(mp, buf2, len2);
1528 	return (mp);
1529 }
1530 
1531 /*
1532  * Create a message using three kernel buffers.
1533  * If size is set that will determine the allocation size (for future
1534  * soappendmsg calls). If size is zero it is derived from the buffer
1535  * lengths.
1536  */
1537 mblk_t *
1538 soallocproto3(const void *buf1, ssize_t len1, const void *buf2, ssize_t len2,
1539     const void *buf3, ssize_t len3, ssize_t size, int sleepflg, cred_t *cr)
1540 {
1541 	mblk_t *mp;
1542 
1543 	if (size == 0)
1544 		size = len1 + len2 +len3;
1545 	ASSERT(size >= len1 + len2 + len3);
1546 
1547 	mp = soallocproto1(buf1, len1, size, sleepflg, cr);
1548 	if (mp != NULL) {
1549 		soappendmsg(mp, buf2, len2);
1550 		soappendmsg(mp, buf3, len3);
1551 	}
1552 	return (mp);
1553 }
1554 
1555 #ifdef DEBUG
1556 char *
1557 pr_state(uint_t state, uint_t mode)
1558 {
1559 	static char buf[1024];
1560 
1561 	buf[0] = 0;
1562 	if (state & SS_ISCONNECTED)
1563 		(void) strcat(buf, "ISCONNECTED ");
1564 	if (state & SS_ISCONNECTING)
1565 		(void) strcat(buf, "ISCONNECTING ");
1566 	if (state & SS_ISDISCONNECTING)
1567 		(void) strcat(buf, "ISDISCONNECTING ");
1568 	if (state & SS_CANTSENDMORE)
1569 		(void) strcat(buf, "CANTSENDMORE ");
1570 
1571 	if (state & SS_CANTRCVMORE)
1572 		(void) strcat(buf, "CANTRCVMORE ");
1573 	if (state & SS_ISBOUND)
1574 		(void) strcat(buf, "ISBOUND ");
1575 	if (state & SS_NDELAY)
1576 		(void) strcat(buf, "NDELAY ");
1577 	if (state & SS_NONBLOCK)
1578 		(void) strcat(buf, "NONBLOCK ");
1579 
1580 	if (state & SS_ASYNC)
1581 		(void) strcat(buf, "ASYNC ");
1582 	if (state & SS_ACCEPTCONN)
1583 		(void) strcat(buf, "ACCEPTCONN ");
1584 	if (state & SS_SAVEDEOR)
1585 		(void) strcat(buf, "SAVEDEOR ");
1586 
1587 	if (state & SS_RCVATMARK)
1588 		(void) strcat(buf, "RCVATMARK ");
1589 	if (state & SS_OOBPEND)
1590 		(void) strcat(buf, "OOBPEND ");
1591 	if (state & SS_HAVEOOBDATA)
1592 		(void) strcat(buf, "HAVEOOBDATA ");
1593 	if (state & SS_HADOOBDATA)
1594 		(void) strcat(buf, "HADOOBDATA ");
1595 
1596 	if (mode & SM_PRIV)
1597 		(void) strcat(buf, "PRIV ");
1598 	if (mode & SM_ATOMIC)
1599 		(void) strcat(buf, "ATOMIC ");
1600 	if (mode & SM_ADDR)
1601 		(void) strcat(buf, "ADDR ");
1602 	if (mode & SM_CONNREQUIRED)
1603 		(void) strcat(buf, "CONNREQUIRED ");
1604 
1605 	if (mode & SM_FDPASSING)
1606 		(void) strcat(buf, "FDPASSING ");
1607 	if (mode & SM_EXDATA)
1608 		(void) strcat(buf, "EXDATA ");
1609 	if (mode & SM_OPTDATA)
1610 		(void) strcat(buf, "OPTDATA ");
1611 	if (mode & SM_BYTESTREAM)
1612 		(void) strcat(buf, "BYTESTREAM ");
1613 	return (buf);
1614 }
1615 
1616 char *
1617 pr_addr(int family, struct sockaddr *addr, t_uscalar_t addrlen)
1618 {
1619 	static char buf[1024];
1620 
1621 	if (addr == NULL || addrlen == 0) {
1622 		(void) sprintf(buf, "(len %d) %p", addrlen, (void *)addr);
1623 		return (buf);
1624 	}
1625 	switch (family) {
1626 	case AF_INET: {
1627 		struct sockaddr_in sin;
1628 
1629 		bcopy(addr, &sin, sizeof (sin));
1630 
1631 		(void) sprintf(buf, "(len %d) %x/%d",
1632 		    addrlen, ntohl(sin.sin_addr.s_addr), ntohs(sin.sin_port));
1633 		break;
1634 	}
1635 	case AF_INET6: {
1636 		struct sockaddr_in6 sin6;
1637 		uint16_t *piece = (uint16_t *)&sin6.sin6_addr;
1638 
1639 		bcopy((char *)addr, (char *)&sin6, sizeof (sin6));
1640 		(void) sprintf(buf, "(len %d) %x:%x:%x:%x:%x:%x:%x:%x/%d",
1641 		    addrlen,
1642 		    ntohs(piece[0]), ntohs(piece[1]),
1643 		    ntohs(piece[2]), ntohs(piece[3]),
1644 		    ntohs(piece[4]), ntohs(piece[5]),
1645 		    ntohs(piece[6]), ntohs(piece[7]),
1646 		    ntohs(sin6.sin6_port));
1647 		break;
1648 	}
1649 	case AF_UNIX: {
1650 		struct sockaddr_un *soun = (struct sockaddr_un *)addr;
1651 
1652 		(void) sprintf(buf, "(len %d) %s", addrlen,
1653 		    (soun == NULL) ? "(none)" : soun->sun_path);
1654 		break;
1655 	}
1656 	default:
1657 		(void) sprintf(buf, "(unknown af %d)", family);
1658 		break;
1659 	}
1660 	return (buf);
1661 }
1662 
1663 /* The logical equivalence operator (a if-and-only-if b) */
1664 #define	EQUIVALENT(a, b)	(((a) && (b)) || (!(a) && (!(b))))
1665 
1666 /*
1667  * Verify limitations and invariants on oob state.
1668  * Return 1 if OK, otherwise 0 so that it can be used as
1669  *	ASSERT(verify_oobstate(so));
1670  */
1671 int
1672 so_verify_oobstate(struct sonode *so)
1673 {
1674 	boolean_t havemark;
1675 
1676 	ASSERT(MUTEX_HELD(&so->so_lock));
1677 
1678 	/*
1679 	 * The possible state combinations are:
1680 	 *	0
1681 	 *	SS_OOBPEND
1682 	 *	SS_OOBPEND|SS_HAVEOOBDATA
1683 	 *	SS_OOBPEND|SS_HADOOBDATA
1684 	 *	SS_HADOOBDATA
1685 	 */
1686 	switch (so->so_state & (SS_OOBPEND|SS_HAVEOOBDATA|SS_HADOOBDATA)) {
1687 	case 0:
1688 	case SS_OOBPEND:
1689 	case SS_OOBPEND|SS_HAVEOOBDATA:
1690 	case SS_OOBPEND|SS_HADOOBDATA:
1691 	case SS_HADOOBDATA:
1692 		break;
1693 	default:
1694 		printf("Bad oob state 1 (%p): state %s\n",
1695 		    (void *)so, pr_state(so->so_state, so->so_mode));
1696 		return (0);
1697 	}
1698 
1699 	/* SS_RCVATMARK should only be set when SS_OOBPEND is set */
1700 	if ((so->so_state & (SS_RCVATMARK|SS_OOBPEND)) == SS_RCVATMARK) {
1701 		printf("Bad oob state 2 (%p): state %s\n",
1702 		    (void *)so, pr_state(so->so_state, so->so_mode));
1703 		return (0);
1704 	}
1705 
1706 	/*
1707 	 * (havemark != 0 or SS_RCVATMARK) iff SS_OOBPEND
1708 	 * For TPI, the presence of a "mark" is indicated by sti_oobsigcnt.
1709 	 */
1710 	havemark = (SOCK_IS_NONSTR(so)) ? so->so_oobmark > 0 :
1711 	    SOTOTPI(so)->sti_oobsigcnt > 0;
1712 
1713 	if (!EQUIVALENT(havemark || (so->so_state & SS_RCVATMARK),
1714 	    so->so_state & SS_OOBPEND)) {
1715 		printf("Bad oob state 3 (%p): state %s\n",
1716 		    (void *)so, pr_state(so->so_state, so->so_mode));
1717 		return (0);
1718 	}
1719 
1720 	/*
1721 	 * Unless SO_OOBINLINE we have so_oobmsg != NULL iff SS_HAVEOOBDATA
1722 	 */
1723 	if (!(so->so_options & SO_OOBINLINE) &&
1724 	    !EQUIVALENT(so->so_oobmsg != NULL, so->so_state & SS_HAVEOOBDATA)) {
1725 		printf("Bad oob state 4 (%p): state %s\n",
1726 		    (void *)so, pr_state(so->so_state, so->so_mode));
1727 		return (0);
1728 	}
1729 
1730 	if (!SOCK_IS_NONSTR(so) &&
1731 	    SOTOTPI(so)->sti_oobsigcnt < SOTOTPI(so)->sti_oobcnt) {
1732 		printf("Bad oob state 5 (%p): counts %d/%d state %s\n",
1733 		    (void *)so, SOTOTPI(so)->sti_oobsigcnt,
1734 		    SOTOTPI(so)->sti_oobcnt,
1735 		    pr_state(so->so_state, so->so_mode));
1736 		return (0);
1737 	}
1738 
1739 	return (1);
1740 }
1741 #undef	EQUIVALENT
1742 #endif /* DEBUG */
1743 
1744 /* initialize sockfs zone specific kstat related items			*/
1745 void *
1746 sock_kstat_init(zoneid_t zoneid)
1747 {
1748 	kstat_t	*ksp;
1749 
1750 	ksp = kstat_create_zone("sockfs", 0, "sock_unix_list", "misc",
1751 	    KSTAT_TYPE_RAW, 0, KSTAT_FLAG_VAR_SIZE|KSTAT_FLAG_VIRTUAL, zoneid);
1752 
1753 	if (ksp != NULL) {
1754 		ksp->ks_update = sockfs_update;
1755 		ksp->ks_snapshot = sockfs_snapshot;
1756 		ksp->ks_lock = &socklist.sl_lock;
1757 		ksp->ks_private = (void *)(uintptr_t)zoneid;
1758 		kstat_install(ksp);
1759 	}
1760 
1761 	return (ksp);
1762 }
1763 
1764 /* tear down sockfs zone specific kstat related items			*/
1765 /*ARGSUSED*/
1766 void
1767 sock_kstat_fini(zoneid_t zoneid, void *arg)
1768 {
1769 	kstat_t *ksp = (kstat_t *)arg;
1770 
1771 	if (ksp != NULL) {
1772 		ASSERT(zoneid == (zoneid_t)(uintptr_t)ksp->ks_private);
1773 		kstat_delete(ksp);
1774 	}
1775 }
1776 
1777 /*
1778  * Zones:
1779  * Note that nactive is going to be different for each zone.
1780  * This means we require kstat to call sockfs_update and then sockfs_snapshot
1781  * for the same zone, or sockfs_snapshot will be taken into the wrong size
1782  * buffer. This is safe, but if the buffer is too small, user will not be
1783  * given details of all sockets. However, as this kstat has a ks_lock, kstat
1784  * driver will keep it locked between the update and the snapshot, so no
1785  * other process (zone) can currently get inbetween resulting in a wrong size
1786  * buffer allocation.
1787  */
1788 static int
1789 sockfs_update(kstat_t *ksp, int rw)
1790 {
1791 	uint_t	nactive = 0;		/* # of active AF_UNIX sockets	*/
1792 	struct sonode	*so;		/* current sonode on socklist	*/
1793 	zoneid_t	myzoneid = (zoneid_t)(uintptr_t)ksp->ks_private;
1794 
1795 	ASSERT((zoneid_t)(uintptr_t)ksp->ks_private == getzoneid());
1796 
1797 	if (rw == KSTAT_WRITE) {	/* bounce all writes		*/
1798 		return (EACCES);
1799 	}
1800 
1801 	for (so = socklist.sl_list; so != NULL; so = SOTOTPI(so)->sti_next_so) {
1802 		if (so->so_count != 0 && so->so_zoneid == myzoneid) {
1803 			nactive++;
1804 		}
1805 	}
1806 	ksp->ks_ndata = nactive;
1807 	ksp->ks_data_size = nactive * sizeof (struct sockinfo);
1808 
1809 	return (0);
1810 }
1811 
1812 static int
1813 sockfs_snapshot(kstat_t *ksp, void *buf, int rw)
1814 {
1815 	int			ns;	/* # of sonodes we've copied	*/
1816 	struct sonode		*so;	/* current sonode on socklist	*/
1817 	struct sockinfo		*psi;	/* where we put sockinfo data	*/
1818 	t_uscalar_t		sn_len;	/* soa_len			*/
1819 	zoneid_t		myzoneid = (zoneid_t)(uintptr_t)ksp->ks_private;
1820 	sotpi_info_t		*sti;
1821 
1822 	ASSERT((zoneid_t)(uintptr_t)ksp->ks_private == getzoneid());
1823 
1824 	ksp->ks_snaptime = gethrtime();
1825 
1826 	if (rw == KSTAT_WRITE) {	/* bounce all writes		*/
1827 		return (EACCES);
1828 	}
1829 
1830 	/*
1831 	 * For each sonode on the socklist, we massage the important
1832 	 * info into buf, in sockinfo format.
1833 	 */
1834 	psi = (struct sockinfo *)buf;
1835 	ns = 0;
1836 	for (so = socklist.sl_list; so != NULL; so = SOTOTPI(so)->sti_next_so) {
1837 		vattr_t attr;
1838 
1839 		/* only stuff active sonodes and the same zone:		*/
1840 		if (so->so_count == 0 || so->so_zoneid != myzoneid) {
1841 			continue;
1842 		}
1843 
1844 		/*
1845 		 * If the sonode was activated between the update and the
1846 		 * snapshot, we're done - as this is only a snapshot.
1847 		 */
1848 		if ((caddr_t)(psi) >= (caddr_t)buf + ksp->ks_data_size) {
1849 			break;
1850 		}
1851 
1852 		sti = SOTOTPI(so);
1853 		/* copy important info into buf:			*/
1854 		psi->si_size = sizeof (struct sockinfo);
1855 		psi->si_family = so->so_family;
1856 		psi->si_type = so->so_type;
1857 		psi->si_flag = so->so_flag;
1858 		psi->si_state = so->so_state;
1859 		psi->si_serv_type = sti->sti_serv_type;
1860 		psi->si_ux_laddr_sou_magic = sti->sti_ux_laddr.soua_magic;
1861 		psi->si_ux_faddr_sou_magic = sti->sti_ux_faddr.soua_magic;
1862 		psi->si_laddr_soa_len = sti->sti_laddr.soa_len;
1863 		psi->si_faddr_soa_len = sti->sti_faddr.soa_len;
1864 		psi->si_szoneid = so->so_zoneid;
1865 		psi->si_faddr_noxlate = sti->sti_faddr_noxlate;
1866 
1867 		/*
1868 		 * Grab the inode, if possible.
1869 		 * This must be done before entering so_lock as VOP_GETATTR
1870 		 * will acquire it.
1871 		 */
1872 		if (so->so_vnode == NULL ||
1873 		    VOP_GETATTR(so->so_vnode, &attr, 0, CRED(), NULL) != 0)
1874 			attr.va_nodeid = 0;
1875 
1876 		psi->si_inode = attr.va_nodeid;
1877 
1878 		mutex_enter(&so->so_lock);
1879 
1880 		if (sti->sti_laddr_sa != NULL) {
1881 			ASSERT(sti->sti_laddr_sa->sa_data != NULL);
1882 			sn_len = sti->sti_laddr_len;
1883 			ASSERT(sn_len <= sizeof (short) +
1884 			    sizeof (psi->si_laddr_sun_path));
1885 
1886 			psi->si_laddr_family =
1887 			    sti->sti_laddr_sa->sa_family;
1888 			if (sn_len != 0) {
1889 				/* AF_UNIX socket names are NULL terminated */
1890 				(void) strncpy(psi->si_laddr_sun_path,
1891 				    sti->sti_laddr_sa->sa_data,
1892 				    sizeof (psi->si_laddr_sun_path));
1893 				sn_len = strlen(psi->si_laddr_sun_path);
1894 			}
1895 			psi->si_laddr_sun_path[sn_len] = 0;
1896 		}
1897 
1898 		if (sti->sti_faddr_sa != NULL) {
1899 			ASSERT(sti->sti_faddr_sa->sa_data != NULL);
1900 			sn_len = sti->sti_faddr_len;
1901 			ASSERT(sn_len <= sizeof (short) +
1902 			    sizeof (psi->si_faddr_sun_path));
1903 
1904 			psi->si_faddr_family =
1905 			    sti->sti_faddr_sa->sa_family;
1906 			if (sn_len != 0) {
1907 				(void) strncpy(psi->si_faddr_sun_path,
1908 				    sti->sti_faddr_sa->sa_data,
1909 				    sizeof (psi->si_faddr_sun_path));
1910 				sn_len = strlen(psi->si_faddr_sun_path);
1911 			}
1912 			psi->si_faddr_sun_path[sn_len] = 0;
1913 		}
1914 
1915 		mutex_exit(&so->so_lock);
1916 
1917 		(void) snprintf(psi->si_son_straddr,
1918 		    sizeof (psi->si_son_straddr), "%p", (void *)so);
1919 		(void) snprintf(psi->si_lvn_straddr,
1920 		    sizeof (psi->si_lvn_straddr), "%p",
1921 		    (void *)sti->sti_ux_laddr.soua_vp);
1922 		(void) snprintf(psi->si_fvn_straddr,
1923 		    sizeof (psi->si_fvn_straddr), "%p",
1924 		    (void *)sti->sti_ux_faddr.soua_vp);
1925 
1926 		ns++;
1927 		psi++;
1928 	}
1929 
1930 	ksp->ks_ndata = ns;
1931 	return (0);
1932 }
1933 
1934 ssize_t
1935 soreadfile(file_t *fp, uchar_t *buf, u_offset_t fileoff, int *err, size_t size)
1936 {
1937 	struct uio auio;
1938 	struct iovec aiov[1];
1939 	register vnode_t *vp;
1940 	int ioflag, rwflag;
1941 	ssize_t cnt;
1942 	int error = 0;
1943 	int iovcnt = 0;
1944 	short fflag;
1945 
1946 	vp = fp->f_vnode;
1947 	fflag = fp->f_flag;
1948 
1949 	rwflag = 0;
1950 	aiov[0].iov_base = (caddr_t)buf;
1951 	aiov[0].iov_len = size;
1952 	iovcnt = 1;
1953 	cnt = (ssize_t)size;
1954 	(void) VOP_RWLOCK(vp, rwflag, NULL);
1955 
1956 	auio.uio_loffset = fileoff;
1957 	auio.uio_iov = aiov;
1958 	auio.uio_iovcnt = iovcnt;
1959 	auio.uio_resid = cnt;
1960 	auio.uio_segflg = UIO_SYSSPACE;
1961 	auio.uio_llimit = MAXOFFSET_T;
1962 	auio.uio_fmode = fflag;
1963 	auio.uio_extflg = UIO_COPY_CACHED;
1964 
1965 	ioflag = auio.uio_fmode & (FAPPEND|FSYNC|FDSYNC|FRSYNC);
1966 
1967 	/* If read sync is not asked for, filter sync flags */
1968 	if ((ioflag & FRSYNC) == 0)
1969 		ioflag &= ~(FSYNC|FDSYNC);
1970 	error = VOP_READ(vp, &auio, ioflag, fp->f_cred, NULL);
1971 	cnt -= auio.uio_resid;
1972 
1973 	VOP_RWUNLOCK(vp, rwflag, NULL);
1974 
1975 	if (error == EINTR && cnt != 0)
1976 		error = 0;
1977 out:
1978 	if (error != 0) {
1979 		*err = error;
1980 		return (0);
1981 	} else {
1982 		*err = 0;
1983 		return (cnt);
1984 	}
1985 }
1986 
1987 int
1988 so_copyin(const void *from, void *to, size_t size, int fromkernel)
1989 {
1990 	if (fromkernel) {
1991 		bcopy(from, to, size);
1992 		return (0);
1993 	}
1994 	return (xcopyin(from, to, size));
1995 }
1996 
1997 int
1998 so_copyout(const void *from, void *to, size_t size, int tokernel)
1999 {
2000 	if (tokernel) {
2001 		bcopy(from, to, size);
2002 		return (0);
2003 	}
2004 	return (xcopyout(from, to, size));
2005 }
2006