xref: /titanic_50/usr/src/uts/common/fs/sockfs/socksubr.c (revision 25cf1a301a396c38e8adf52c15f537b80d2483f7)
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 2006 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #pragma ident	"%Z%%M%	%I%	%E% SMI"
28 
29 #include <sys/types.h>
30 #include <sys/t_lock.h>
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/buf.h>
34 #include <sys/conf.h>
35 #include <sys/cred.h>
36 #include <sys/kmem.h>
37 #include <sys/sysmacros.h>
38 #include <sys/vfs.h>
39 #include <sys/vnode.h>
40 #include <sys/debug.h>
41 #include <sys/errno.h>
42 #include <sys/time.h>
43 #include <sys/file.h>
44 #include <sys/open.h>
45 #include <sys/user.h>
46 #include <sys/termios.h>
47 #include <sys/stream.h>
48 #include <sys/strsubr.h>
49 #include <sys/strsun.h>
50 #include <sys/esunddi.h>
51 #include <sys/flock.h>
52 #include <sys/modctl.h>
53 #include <sys/cmn_err.h>
54 #include <sys/mkdev.h>
55 #include <sys/pathname.h>
56 #include <sys/ddi.h>
57 #include <sys/stat.h>
58 #include <sys/fs/snode.h>
59 #include <sys/fs/dv_node.h>
60 #include <sys/zone.h>
61 
62 #include <sys/socket.h>
63 #include <sys/socketvar.h>
64 #include <netinet/in.h>
65 #include <sys/un.h>
66 
67 #include <sys/ucred.h>
68 
69 #include <sys/tiuser.h>
70 #define	_SUN_TPI_VERSION	2
71 #include <sys/tihdr.h>
72 
73 #include <c2/audit.h>
74 
75 #include <fs/sockfs/nl7c.h>
76 
77 /*
78  * Macros that operate on struct cmsghdr.
79  * The CMSG_VALID macro does not assume that the last option buffer is padded.
80  */
81 #define	CMSG_CONTENT(cmsg)	(&((cmsg)[1]))
82 #define	CMSG_CONTENTLEN(cmsg)	((cmsg)->cmsg_len - sizeof (struct cmsghdr))
83 #define	CMSG_VALID(cmsg, start, end)					\
84 	(ISALIGNED_cmsghdr(cmsg) &&					\
85 	((uintptr_t)(cmsg) >= (uintptr_t)(start)) &&			\
86 	((uintptr_t)(cmsg) < (uintptr_t)(end)) &&			\
87 	((ssize_t)(cmsg)->cmsg_len >= sizeof (struct cmsghdr)) &&	\
88 	((uintptr_t)(cmsg) + (cmsg)->cmsg_len <= (uintptr_t)(end)))
89 #define	SO_LOCK_WAKEUP_TIME	3000	/* Wakeup time in milliseconds */
90 
91 static struct kmem_cache *socktpi_cache, *socktpi_unix_cache;
92 
93 dev_t sockdev;	/* For fsid in getattr */
94 
95 struct sockparams *sphead;
96 krwlock_t splist_lock;
97 
98 struct socklist socklist;
99 
100 static int sockfs_update(kstat_t *, int);
101 static int sockfs_snapshot(kstat_t *, void *, int);
102 
103 extern void sendfile_init();
104 
105 extern void nl7c_init(void);
106 
107 #define	ADRSTRLEN (2 * sizeof (void *) + 1)
108 /*
109  * kernel structure for passing the sockinfo data back up to the user.
110  * the strings array allows us to convert AF_UNIX addresses into strings
111  * with a common method regardless of which n-bit kernel we're running.
112  */
113 struct k_sockinfo {
114 	struct sockinfo	ks_si;
115 	char		ks_straddr[3][ADRSTRLEN];
116 };
117 
118 /*
119  * Translate from a device pathname (e.g. "/dev/tcp") to a vnode.
120  * Returns with the vnode held.
121  */
122 static int
123 sogetvp(char *devpath, vnode_t **vpp, int uioflag)
124 {
125 	struct snode *csp;
126 	vnode_t *vp, *dvp;
127 	major_t maj;
128 	int error;
129 
130 	ASSERT(uioflag == UIO_SYSSPACE || uioflag == UIO_USERSPACE);
131 	/*
132 	 * Lookup the underlying filesystem vnode.
133 	 */
134 	error = lookupname(devpath, uioflag, FOLLOW, NULLVPP, &vp);
135 	if (error)
136 		return (error);
137 
138 	/* Check that it is the correct vnode */
139 	if (vp->v_type != VCHR) {
140 		VN_RELE(vp);
141 		return (ENOTSOCK);
142 	}
143 
144 	/*
145 	 * If devpath went through devfs, the device should already
146 	 * be configured. If devpath is a mknod file, however, we
147 	 * need to make sure the device is properly configured.
148 	 * To do this, we do something similar to spec_open()
149 	 * except that we resolve to the minor/leaf level since
150 	 * we need to return a vnode.
151 	 */
152 	csp = VTOS(VTOS(vp)->s_commonvp);
153 	if (!(csp->s_flag & SDIPSET)) {
154 		char *pathname = kmem_alloc(MAXPATHLEN, KM_SLEEP);
155 		error = ddi_dev_pathname(vp->v_rdev, S_IFCHR, pathname);
156 		if (error == 0)
157 			error = devfs_lookupname(pathname, NULLVPP, &dvp);
158 		VN_RELE(vp);
159 		kmem_free(pathname, MAXPATHLEN);
160 		if (error != 0)
161 			return (ENXIO);
162 		vp = dvp;	/* use the devfs vp */
163 	}
164 
165 	/* device is configured at this point */
166 	maj = getmajor(vp->v_rdev);
167 	if (!STREAMSTAB(maj)) {
168 		VN_RELE(vp);
169 		return (ENOSTR);
170 	}
171 
172 	*vpp = vp;
173 	return (0);
174 }
175 
176 /*
177  * Add or delete (latter if devpath is NULL) an enter to the sockparams
178  * table. If devpathlen is zero the devpath with not be kmem_freed. Otherwise
179  * this routine assumes that the caller has kmem_alloced devpath/devpathlen
180  * for this routine to consume.
181  * The zero devpathlen could be used if the kernel wants to create entries
182  * itself by calling sockconfig(1,2,3, "/dev/tcp", 0);
183  */
184 int
185 soconfig(int domain, int type, int protocol,
186     char *devpath, int devpathlen)
187 {
188 	struct sockparams **spp;
189 	struct sockparams *sp;
190 	int error = 0;
191 
192 	dprint(0, ("soconfig(%d,%d,%d,%s,%d)\n",
193 		domain, type, protocol, devpath, devpathlen));
194 
195 	/*
196 	 * Look for an existing match.
197 	 */
198 	rw_enter(&splist_lock, RW_WRITER);
199 	for (spp = &sphead; (sp = *spp) != NULL; spp = &sp->sp_next) {
200 		if (sp->sp_domain == domain &&
201 		    sp->sp_type == type &&
202 		    sp->sp_protocol == protocol) {
203 			break;
204 		}
205 	}
206 	if (devpath == NULL) {
207 		ASSERT(devpathlen == 0);
208 
209 		/* Delete existing entry */
210 		if (sp == NULL) {
211 			error = ENXIO;
212 			goto done;
213 		}
214 		/* Unlink and free existing entry */
215 		*spp = sp->sp_next;
216 		ASSERT(sp->sp_vnode);
217 		VN_RELE(sp->sp_vnode);
218 		if (sp->sp_devpathlen != 0)
219 			kmem_free(sp->sp_devpath, sp->sp_devpathlen);
220 		kmem_free(sp, sizeof (*sp));
221 	} else {
222 		vnode_t *vp;
223 
224 		/* Add new entry */
225 		if (sp != NULL) {
226 			error = EEXIST;
227 			goto done;
228 		}
229 
230 		error = sogetvp(devpath, &vp, UIO_SYSSPACE);
231 		if (error) {
232 			dprint(0, ("soconfig: vp %s failed with %d\n",
233 				devpath, error));
234 			goto done;
235 		}
236 
237 		dprint(0, ("soconfig: %s => vp %p, dev 0x%lx\n",
238 		    devpath, vp, vp->v_rdev));
239 
240 		sp = kmem_alloc(sizeof (*sp), KM_SLEEP);
241 		sp->sp_domain = domain;
242 		sp->sp_type = type;
243 		sp->sp_protocol = protocol;
244 		sp->sp_devpath = devpath;
245 		sp->sp_devpathlen = devpathlen;
246 		sp->sp_vnode = vp;
247 		sp->sp_next = NULL;
248 		*spp = sp;
249 	}
250 done:
251 	rw_exit(&splist_lock);
252 	if (error) {
253 		if (devpath != NULL)
254 			kmem_free(devpath, devpathlen);
255 #ifdef SOCK_DEBUG
256 		eprintline(error);
257 #endif /* SOCK_DEBUG */
258 	}
259 	return (error);
260 }
261 
262 /*
263  * Lookup an entry in the sockparams list based on the triple.
264  * If no entry is found and devpath is not NULL translate devpath to a
265  * vnode. Note that devpath is a pointer to a user address!
266  * Returns with the vnode held.
267  *
268  * When this routine uses devpath it does not create an entry in the sockparams
269  * list since this routine can run on behalf of any user and one user
270  * should not be able to effect the transport used by another user.
271  *
272  * In order to return the correct error this routine has to do wildcard scans
273  * of the list. The errors are (in decreasing precedence):
274  *	EAFNOSUPPORT - address family not in list
275  *	EPROTONOSUPPORT - address family supported but not protocol.
276  *	EPROTOTYPE - address family and protocol supported but not socket type.
277  */
278 vnode_t *
279 solookup(int domain, int type, int protocol, char *devpath, int *errorp)
280 {
281 	struct sockparams *sp;
282 	int error;
283 	vnode_t *vp;
284 
285 	rw_enter(&splist_lock, RW_READER);
286 	for (sp = sphead; sp != NULL; sp = sp->sp_next) {
287 		if (sp->sp_domain == domain &&
288 		    sp->sp_type == type &&
289 		    sp->sp_protocol == protocol) {
290 			break;
291 		}
292 	}
293 	if (sp == NULL) {
294 		dprint(0, ("solookup(%d,%d,%d) not found\n",
295 			domain, type, protocol));
296 		if (devpath == NULL) {
297 			/* Determine correct error code */
298 			int found = 0;
299 
300 			for (sp = sphead; sp != NULL; sp = sp->sp_next) {
301 				if (sp->sp_domain == domain && found < 1)
302 					found = 1;
303 				if (sp->sp_domain == domain &&
304 				    sp->sp_protocol == protocol && found < 2)
305 					found = 2;
306 			}
307 			rw_exit(&splist_lock);
308 			switch (found) {
309 			case 0:
310 				*errorp = EAFNOSUPPORT;
311 				break;
312 			case 1:
313 				*errorp = EPROTONOSUPPORT;
314 				break;
315 			case 2:
316 				*errorp = EPROTOTYPE;
317 				break;
318 			}
319 			return (NULL);
320 		}
321 		rw_exit(&splist_lock);
322 
323 		/*
324 		 * Return vp based on devpath.
325 		 * Do not enter into table to avoid random users
326 		 * modifying the sockparams list.
327 		 */
328 		error = sogetvp(devpath, &vp, UIO_USERSPACE);
329 		if (error) {
330 			dprint(0, ("solookup: vp %p failed with %d\n",
331 				devpath, error));
332 			*errorp = EPROTONOSUPPORT;
333 			return (NULL);
334 		}
335 		dprint(0, ("solookup: %p => vp %p, dev 0x%lx\n",
336 		    devpath, vp, vp->v_rdev));
337 
338 		return (vp);
339 	}
340 	dprint(0, ("solookup(%d,%d,%d) vp %p devpath %s\n",
341 		domain, type, protocol, sp->sp_vnode, sp->sp_devpath));
342 
343 	vp = sp->sp_vnode;
344 	VN_HOLD(vp);
345 	rw_exit(&splist_lock);
346 	return (vp);
347 }
348 
349 /*
350  * Return a socket vnode.
351  *
352  * Assumes that the caller is "passing" an VN_HOLD for accessvp i.e.
353  * when the socket is freed a VN_RELE will take place.
354  *
355  * Note that sockets assume that the driver will clone (either itself
356  * or by using the clone driver) i.e. a socket() call will always
357  * result in a new vnode being created.
358  */
359 struct vnode *
360 makesockvp(struct vnode *accessvp, int domain, int type, int protocol)
361 {
362 	kmem_cache_t *cp;
363 	struct sonode *so;
364 	struct vnode *vp;
365 	time_t now;
366 	dev_t dev;
367 
368 	cp = (domain == AF_UNIX) ? socktpi_unix_cache : socktpi_cache;
369 	so = kmem_cache_alloc(cp, KM_SLEEP);
370 	so->so_cache = cp;
371 	so->so_obj = so;
372 	vp = SOTOV(so);
373 	now = gethrestime_sec();
374 
375 	so->so_flag	= 0;
376 	ASSERT(so->so_accessvp == NULL);
377 	so->so_accessvp	= accessvp;
378 	dev = accessvp->v_rdev;
379 
380 	/*
381 	 * Record in so_flag that it is a clone.
382 	 */
383 	if (getmajor(dev) == clone_major) {
384 		so->so_flag |= SOCLONE;
385 	}
386 	so->so_dev = dev;
387 
388 	so->so_state	= 0;
389 	so->so_mode	= 0;
390 
391 	so->so_fsid	= sockdev;
392 	so->so_atime	= now;
393 	so->so_mtime	= now;
394 	so->so_ctime	= now;		/* Never modified */
395 	so->so_count	= 0;
396 
397 	so->so_family	= (short)domain;
398 	so->so_type	= (short)type;
399 	so->so_protocol	= (short)protocol;
400 	so->so_pushcnt	= 0;
401 
402 	so->so_options	= 0;
403 	so->so_linger.l_onoff	= 0;
404 	so->so_linger.l_linger = 0;
405 	so->so_sndbuf	= 0;
406 	so->so_rcvbuf	= 0;
407 	so->so_sndlowat	= 0;
408 	so->so_rcvlowat	= 0;
409 #ifdef notyet
410 	so->so_sndtimeo	= 0;
411 	so->so_rcvtimeo	= 0;
412 #endif /* notyet */
413 	so->so_error	= 0;
414 	so->so_delayed_error = 0;
415 
416 	ASSERT(so->so_oobmsg == NULL);
417 	so->so_oobcnt	= 0;
418 	so->so_oobsigcnt = 0;
419 	so->so_pgrp	= 0;
420 	so->so_provinfo = NULL;
421 
422 	ASSERT(so->so_laddr_sa == NULL && so->so_faddr_sa == NULL);
423 	so->so_laddr_len = so->so_faddr_len = 0;
424 	so->so_laddr_maxlen = so->so_faddr_maxlen = 0;
425 	so->so_eaddr_mp = NULL;
426 	so->so_priv = NULL;
427 
428 	so->so_peercred = NULL;
429 
430 	ASSERT(so->so_ack_mp == NULL);
431 	ASSERT(so->so_conn_ind_head == NULL);
432 	ASSERT(so->so_conn_ind_tail == NULL);
433 	ASSERT(so->so_ux_bound_vp == NULL);
434 	ASSERT(so->so_unbind_mp == NULL);
435 
436 	vn_reinit(vp);
437 	vp->v_vfsp	= rootvfs;
438 	vp->v_type	= VSOCK;
439 	vp->v_rdev	= so->so_dev;
440 	vn_exists(vp);
441 
442 	return (vp);
443 }
444 
445 void
446 sockfree(struct sonode *so)
447 {
448 	mblk_t *mp;
449 	vnode_t *vp;
450 
451 	ASSERT(so->so_count == 0);
452 	ASSERT(so->so_accessvp);
453 	ASSERT(so->so_discon_ind_mp == NULL);
454 
455 	vp = so->so_accessvp;
456 	VN_RELE(vp);
457 
458 	/*
459 	 * Protect so->so_[lf]addr_sa so that sockfs_snapshot() can safely
460 	 * indirect them.  It also uses so_accessvp as a validity test.
461 	 */
462 	mutex_enter(&so->so_lock);
463 
464 	so->so_accessvp = NULL;
465 
466 	if (so->so_laddr_sa) {
467 		ASSERT((caddr_t)so->so_faddr_sa ==
468 		    (caddr_t)so->so_laddr_sa + so->so_laddr_maxlen);
469 		ASSERT(so->so_faddr_maxlen == so->so_laddr_maxlen);
470 		so->so_state &= ~(SS_LADDR_VALID | SS_FADDR_VALID);
471 		kmem_free(so->so_laddr_sa, so->so_laddr_maxlen * 2);
472 		so->so_laddr_sa = NULL;
473 		so->so_laddr_len = so->so_laddr_maxlen = 0;
474 		so->so_faddr_sa = NULL;
475 		so->so_faddr_len = so->so_faddr_maxlen = 0;
476 	}
477 
478 	mutex_exit(&so->so_lock);
479 
480 	if ((mp = so->so_eaddr_mp) != NULL) {
481 		freemsg(mp);
482 		so->so_eaddr_mp = NULL;
483 		so->so_delayed_error = 0;
484 	}
485 	if ((mp = so->so_ack_mp) != NULL) {
486 		freemsg(mp);
487 		so->so_ack_mp = NULL;
488 	}
489 	if ((mp = so->so_conn_ind_head) != NULL) {
490 		mblk_t *mp1;
491 
492 		while (mp) {
493 			mp1 = mp->b_next;
494 			mp->b_next = NULL;
495 			freemsg(mp);
496 			mp = mp1;
497 		}
498 		so->so_conn_ind_head = so->so_conn_ind_tail = NULL;
499 		so->so_state &= ~SS_HASCONNIND;
500 	}
501 #ifdef DEBUG
502 	mutex_enter(&so->so_lock);
503 	ASSERT(so_verify_oobstate(so));
504 	mutex_exit(&so->so_lock);
505 #endif /* DEBUG */
506 	if ((mp = so->so_oobmsg) != NULL) {
507 		freemsg(mp);
508 		so->so_oobmsg = NULL;
509 		so->so_state &= ~(SS_OOBPEND|SS_HAVEOOBDATA|SS_HADOOBDATA);
510 	}
511 
512 	if ((mp = so->so_nl7c_rcv_mp) != NULL) {
513 		so->so_nl7c_rcv_mp = NULL;
514 		freemsg(mp);
515 	}
516 	so->so_nl7c_rcv_rval = 0;
517 	if (so->so_nl7c_uri != NULL) {
518 		nl7c_urifree(so);
519 	}
520 	so->so_nl7c_flags = 0;
521 
522 	ASSERT(so->so_ux_bound_vp == NULL);
523 	if ((mp = so->so_unbind_mp) != NULL) {
524 		freemsg(mp);
525 		so->so_unbind_mp = NULL;
526 	}
527 	vn_invalid(SOTOV(so));
528 
529 	if (so->so_peercred != NULL)
530 		crfree(so->so_peercred);
531 
532 	kmem_cache_free(so->so_cache, so->so_obj);
533 }
534 
535 /*
536  * Update the accessed, updated, or changed times in an sonode
537  * with the current time.
538  *
539  * Note that both SunOS 4.X and 4.4BSD sockets do not present reasonable
540  * attributes in a fstat call. (They return the current time and 0 for
541  * all timestamps, respectively.) We maintain the current timestamps
542  * here primarily so that should sockmod be popped the resulting
543  * file descriptor will behave like a stream w.r.t. the timestamps.
544  */
545 void
546 so_update_attrs(struct sonode *so, int flag)
547 {
548 	time_t now = gethrestime_sec();
549 
550 	mutex_enter(&so->so_lock);
551 	so->so_flag |= flag;
552 	if (flag & SOACC)
553 		so->so_atime = now;
554 	if (flag & SOMOD)
555 		so->so_mtime = now;
556 	mutex_exit(&so->so_lock);
557 }
558 
559 /*ARGSUSED*/
560 static int
561 socktpi_constructor(void *buf, void *cdrarg, int kmflags)
562 {
563 	struct sonode *so = buf;
564 	struct vnode *vp;
565 
566 	so->so_nl7c_flags	= 0;
567 	so->so_nl7c_uri		= NULL;
568 	so->so_nl7c_rcv_mp	= NULL;
569 
570 	so->so_oobmsg		= NULL;
571 	so->so_ack_mp		= NULL;
572 	so->so_conn_ind_head	= NULL;
573 	so->so_conn_ind_tail	= NULL;
574 	so->so_discon_ind_mp	= NULL;
575 	so->so_ux_bound_vp	= NULL;
576 	so->so_unbind_mp	= NULL;
577 	so->so_accessvp		= NULL;
578 	so->so_laddr_sa		= NULL;
579 	so->so_faddr_sa		= NULL;
580 	so->so_ops		= &sotpi_sonodeops;
581 
582 	vp = vn_alloc(KM_SLEEP);
583 	so->so_vnode = vp;
584 
585 	vn_setops(vp, socktpi_vnodeops);
586 	vp->v_data = (caddr_t)so;
587 
588 	mutex_init(&so->so_lock, NULL, MUTEX_DEFAULT, NULL);
589 	mutex_init(&so->so_plumb_lock, NULL, MUTEX_DEFAULT, NULL);
590 	cv_init(&so->so_state_cv, NULL, CV_DEFAULT, NULL);
591 	cv_init(&so->so_ack_cv, NULL, CV_DEFAULT, NULL);
592 	cv_init(&so->so_connind_cv, NULL, CV_DEFAULT, NULL);
593 	cv_init(&so->so_want_cv, NULL, CV_DEFAULT, NULL);
594 
595 	return (0);
596 }
597 
598 /*ARGSUSED1*/
599 static void
600 socktpi_destructor(void *buf, void *cdrarg)
601 {
602 	struct sonode *so = buf;
603 	struct vnode *vp = SOTOV(so);
604 
605 	ASSERT(so->so_nl7c_flags == 0);
606 	ASSERT(so->so_nl7c_uri == NULL);
607 	ASSERT(so->so_nl7c_rcv_mp == NULL);
608 
609 	ASSERT(so->so_oobmsg == NULL);
610 	ASSERT(so->so_ack_mp == NULL);
611 	ASSERT(so->so_conn_ind_head == NULL);
612 	ASSERT(so->so_conn_ind_tail == NULL);
613 	ASSERT(so->so_discon_ind_mp == NULL);
614 	ASSERT(so->so_ux_bound_vp == NULL);
615 	ASSERT(so->so_unbind_mp == NULL);
616 	ASSERT(so->so_ops == &sotpi_sonodeops);
617 
618 	ASSERT(vn_matchops(vp, socktpi_vnodeops));
619 	ASSERT(vp->v_data == (caddr_t)so);
620 
621 	vn_free(vp);
622 
623 	mutex_destroy(&so->so_lock);
624 	mutex_destroy(&so->so_plumb_lock);
625 	cv_destroy(&so->so_state_cv);
626 	cv_destroy(&so->so_ack_cv);
627 	cv_destroy(&so->so_connind_cv);
628 	cv_destroy(&so->so_want_cv);
629 }
630 
631 static int
632 socktpi_unix_constructor(void *buf, void *cdrarg, int kmflags)
633 {
634 	int retval;
635 
636 	if ((retval = socktpi_constructor(buf, cdrarg, kmflags)) == 0) {
637 		struct sonode *so = (struct sonode *)buf;
638 
639 		mutex_enter(&socklist.sl_lock);
640 
641 		so->so_next = socklist.sl_list;
642 		so->so_prev = NULL;
643 		if (so->so_next != NULL)
644 			so->so_next->so_prev = so;
645 		socklist.sl_list = so;
646 
647 		mutex_exit(&socklist.sl_lock);
648 
649 	}
650 	return (retval);
651 }
652 
653 static void
654 socktpi_unix_destructor(void *buf, void *cdrarg)
655 {
656 	struct sonode	*so	= (struct sonode *)buf;
657 
658 	mutex_enter(&socklist.sl_lock);
659 
660 	if (so->so_next != NULL)
661 		so->so_next->so_prev = so->so_prev;
662 	if (so->so_prev != NULL)
663 		so->so_prev->so_next = so->so_next;
664 	else
665 		socklist.sl_list = so->so_next;
666 
667 	mutex_exit(&socklist.sl_lock);
668 
669 	socktpi_destructor(buf, cdrarg);
670 }
671 
672 /*
673  * Init function called when sockfs is loaded.
674  */
675 int
676 sockinit(int fstype, char *name)
677 {
678 	static const fs_operation_def_t sock_vfsops_template[] = {
679 		NULL, NULL
680 	};
681 	int error;
682 	major_t dev;
683 	char *err_str;
684 
685 	error = vfs_setfsops(fstype, sock_vfsops_template, NULL);
686 	if (error != 0) {
687 		zcmn_err(GLOBAL_ZONEID, CE_WARN,
688 		    "sockinit: bad vfs ops template");
689 		return (error);
690 	}
691 
692 	error = vn_make_ops(name, socktpi_vnodeops_template, &socktpi_vnodeops);
693 	if (error != 0) {
694 		err_str = "sockinit: bad sock vnode ops template";
695 		/* vn_make_ops() does not reset socktpi_vnodeops on failure. */
696 		socktpi_vnodeops = NULL;
697 		goto failure;
698 	}
699 
700 	error = sosctp_init();
701 	if (error != 0) {
702 		err_str = NULL;
703 		goto failure;
704 	}
705 
706 	/*
707 	 * Create sonode caches.  We create a special one for AF_UNIX so
708 	 * that we can track them for netstat(1m).
709 	 */
710 	socktpi_cache = kmem_cache_create("socktpi_cache",
711 	    sizeof (struct sonode), 0, socktpi_constructor,
712 	    socktpi_destructor, NULL, NULL, NULL, 0);
713 
714 	socktpi_unix_cache = kmem_cache_create("socktpi_unix_cache",
715 	    sizeof (struct sonode), 0, socktpi_unix_constructor,
716 	    socktpi_unix_destructor, NULL, NULL, NULL, 0);
717 
718 	/*
719 	 * Build initial list mapping socket parameters to vnode.
720 	 */
721 	rw_init(&splist_lock, NULL, RW_DEFAULT, NULL);
722 
723 	/*
724 	 * If sockets are needed before init runs /sbin/soconfig
725 	 * it is possible to preload the sockparams list here using
726 	 * calls like:
727 	 *	sockconfig(1,2,3, "/dev/tcp", 0);
728 	 */
729 
730 	/*
731 	 * Create a unique dev_t for use in so_fsid.
732 	 */
733 
734 	if ((dev = getudev()) == (major_t)-1)
735 		dev = 0;
736 	sockdev = makedevice(dev, 0);
737 
738 	mutex_init(&socklist.sl_lock, NULL, MUTEX_DEFAULT, NULL);
739 	sendfile_init();
740 	nl7c_init();
741 
742 	return (0);
743 
744 failure:
745 	(void) vfs_freevfsops_by_type(fstype);
746 	if (socktpi_vnodeops != NULL)
747 		vn_freevnodeops(socktpi_vnodeops);
748 	if (err_str != NULL)
749 		zcmn_err(GLOBAL_ZONEID, CE_WARN, err_str);
750 	return (error);
751 }
752 
753 /*
754  * Caller must hold the mutex. Used to set SOLOCKED.
755  */
756 void
757 so_lock_single(struct sonode *so)
758 {
759 	ASSERT(MUTEX_HELD(&so->so_lock));
760 
761 	while (so->so_flag & (SOLOCKED | SOASYNC_UNBIND)) {
762 		so->so_flag |= SOWANT;
763 		cv_wait_stop(&so->so_want_cv, &so->so_lock,
764 			SO_LOCK_WAKEUP_TIME);
765 	}
766 	so->so_flag |= SOLOCKED;
767 }
768 
769 /*
770  * Caller must hold the mutex and pass in SOLOCKED or SOASYNC_UNBIND.
771  * Used to clear SOLOCKED or SOASYNC_UNBIND.
772  */
773 void
774 so_unlock_single(struct sonode *so, int flag)
775 {
776 	ASSERT(MUTEX_HELD(&so->so_lock));
777 	ASSERT(flag & (SOLOCKED|SOASYNC_UNBIND));
778 	ASSERT((flag & ~(SOLOCKED|SOASYNC_UNBIND)) == 0);
779 	ASSERT(so->so_flag & flag);
780 
781 	/*
782 	 * Process the T_DISCON_IND on so_discon_ind_mp.
783 	 *
784 	 * Call to so_drain_discon_ind will result in so_lock
785 	 * being dropped and re-acquired later.
786 	 */
787 	if (so->so_discon_ind_mp != NULL)
788 		so_drain_discon_ind(so);
789 
790 	if (so->so_flag & SOWANT)
791 		cv_broadcast(&so->so_want_cv);
792 	so->so_flag &= ~(SOWANT|flag);
793 }
794 
795 /*
796  * Caller must hold the mutex. Used to set SOREADLOCKED.
797  * If the caller wants nonblocking behavior it should set fmode.
798  */
799 int
800 so_lock_read(struct sonode *so, int fmode)
801 {
802 	ASSERT(MUTEX_HELD(&so->so_lock));
803 
804 	while (so->so_flag & SOREADLOCKED) {
805 		if (fmode & (FNDELAY|FNONBLOCK))
806 			return (EWOULDBLOCK);
807 		so->so_flag |= SOWANT;
808 		cv_wait_stop(&so->so_want_cv, &so->so_lock,
809 			SO_LOCK_WAKEUP_TIME);
810 	}
811 	so->so_flag |= SOREADLOCKED;
812 	return (0);
813 }
814 
815 /*
816  * Like so_lock_read above but allows signals.
817  */
818 int
819 so_lock_read_intr(struct sonode *so, int fmode)
820 {
821 	ASSERT(MUTEX_HELD(&so->so_lock));
822 
823 	while (so->so_flag & SOREADLOCKED) {
824 		if (fmode & (FNDELAY|FNONBLOCK))
825 			return (EWOULDBLOCK);
826 		so->so_flag |= SOWANT;
827 		if (!cv_wait_sig(&so->so_want_cv, &so->so_lock))
828 			return (EINTR);
829 	}
830 	so->so_flag |= SOREADLOCKED;
831 	return (0);
832 }
833 
834 /*
835  * Caller must hold the mutex. Used to clear SOREADLOCKED,
836  * set in so_lock_read() or so_lock_read_intr().
837  */
838 void
839 so_unlock_read(struct sonode *so)
840 {
841 	ASSERT(MUTEX_HELD(&so->so_lock));
842 	ASSERT(so->so_flag & SOREADLOCKED);
843 
844 	if (so->so_flag & SOWANT)
845 		cv_broadcast(&so->so_want_cv);
846 	so->so_flag &= ~(SOWANT|SOREADLOCKED);
847 }
848 
849 /*
850  * Verify that the specified offset falls within the mblk and
851  * that the resulting pointer is aligned.
852  * Returns NULL if not.
853  */
854 void *
855 sogetoff(mblk_t *mp, t_uscalar_t offset,
856     t_uscalar_t length, uint_t align_size)
857 {
858 	uintptr_t ptr1, ptr2;
859 
860 	ASSERT(mp && mp->b_wptr >= mp->b_rptr);
861 	ptr1 = (uintptr_t)mp->b_rptr + offset;
862 	ptr2 = (uintptr_t)ptr1 + length;
863 	if (ptr1 < (uintptr_t)mp->b_rptr || ptr2 > (uintptr_t)mp->b_wptr) {
864 		eprintline(0);
865 		return (NULL);
866 	}
867 	if ((ptr1 & (align_size - 1)) != 0) {
868 		eprintline(0);
869 		return (NULL);
870 	}
871 	return ((void *)ptr1);
872 }
873 
874 /*
875  * Return the AF_UNIX underlying filesystem vnode matching a given name.
876  * Makes sure the sending and the destination sonodes are compatible.
877  * The vnode is returned held.
878  *
879  * The underlying filesystem VSOCK vnode has a v_stream pointer that
880  * references the actual stream head (hence indirectly the actual sonode).
881  */
882 static int
883 so_ux_lookup(struct sonode *so, struct sockaddr_un *soun, int checkaccess,
884 		vnode_t **vpp)
885 {
886 	vnode_t		*vp;	/* Underlying filesystem vnode */
887 	vnode_t		*svp;	/* sockfs vnode */
888 	struct sonode	*so2;
889 	int		error;
890 
891 	dprintso(so, 1, ("so_ux_lookup(%p) name <%s>\n",
892 		so, soun->sun_path));
893 
894 	error = lookupname(soun->sun_path, UIO_SYSSPACE, FOLLOW, NULLVPP, &vp);
895 	if (error) {
896 		eprintsoline(so, error);
897 		return (error);
898 	}
899 	if (vp->v_type != VSOCK) {
900 		error = ENOTSOCK;
901 		eprintsoline(so, error);
902 		goto done2;
903 	}
904 
905 	if (checkaccess) {
906 		/*
907 		 * Check that we have permissions to access the destination
908 		 * vnode. This check is not done in BSD but it is required
909 		 * by X/Open.
910 		 */
911 		if (error = VOP_ACCESS(vp, VREAD|VWRITE, 0, CRED())) {
912 			eprintsoline(so, error);
913 			goto done2;
914 		}
915 	}
916 
917 	/*
918 	 * Check if the remote socket has been closed.
919 	 *
920 	 * Synchronize with vn_rele_stream by holding v_lock while traversing
921 	 * v_stream->sd_vnode.
922 	 */
923 	mutex_enter(&vp->v_lock);
924 	if (vp->v_stream == NULL) {
925 		mutex_exit(&vp->v_lock);
926 		if (so->so_type == SOCK_DGRAM)
927 			error = EDESTADDRREQ;
928 		else
929 			error = ECONNREFUSED;
930 
931 		eprintsoline(so, error);
932 		goto done2;
933 	}
934 	ASSERT(vp->v_stream->sd_vnode);
935 	svp = vp->v_stream->sd_vnode;
936 	/*
937 	 * holding v_lock on underlying filesystem vnode and acquiring
938 	 * it on sockfs vnode. Assumes that no code ever attempts to
939 	 * acquire these locks in the reverse order.
940 	 */
941 	VN_HOLD(svp);
942 	mutex_exit(&vp->v_lock);
943 
944 	if (svp->v_type != VSOCK) {
945 		error = ENOTSOCK;
946 		eprintsoline(so, error);
947 		goto done;
948 	}
949 
950 	so2 = VTOSO(svp);
951 
952 	if (so->so_type != so2->so_type) {
953 		error = EPROTOTYPE;
954 		eprintsoline(so, error);
955 		goto done;
956 	}
957 
958 	VN_RELE(svp);
959 	*vpp = vp;
960 	return (0);
961 
962 done:
963 	VN_RELE(svp);
964 done2:
965 	VN_RELE(vp);
966 	return (error);
967 }
968 
969 /*
970  * Verify peer address for connect and sendto/sendmsg.
971  * Since sendto/sendmsg would not get synchronous errors from the transport
972  * provider we have to do these ugly checks in the socket layer to
973  * preserve compatibility with SunOS 4.X.
974  */
975 int
976 so_addr_verify(struct sonode *so, const struct sockaddr *name,
977     socklen_t namelen)
978 {
979 	int		family;
980 
981 	dprintso(so, 1, ("so_addr_verify(%p, %p, %d)\n", so, name, namelen));
982 
983 	ASSERT(name != NULL);
984 
985 	family = so->so_family;
986 	switch (family) {
987 	case AF_INET:
988 		if (name->sa_family != family) {
989 			eprintsoline(so, EAFNOSUPPORT);
990 			return (EAFNOSUPPORT);
991 		}
992 		if (namelen != (socklen_t)sizeof (struct sockaddr_in)) {
993 			eprintsoline(so, EINVAL);
994 			return (EINVAL);
995 		}
996 		break;
997 	case AF_INET6: {
998 #ifdef DEBUG
999 		struct sockaddr_in6 *sin6;
1000 #endif /* DEBUG */
1001 
1002 		if (name->sa_family != family) {
1003 			eprintsoline(so, EAFNOSUPPORT);
1004 			return (EAFNOSUPPORT);
1005 		}
1006 		if (namelen != (socklen_t)sizeof (struct sockaddr_in6)) {
1007 			eprintsoline(so, EINVAL);
1008 			return (EINVAL);
1009 		}
1010 #ifdef DEBUG
1011 		/* Verify that apps don't forget to clear sin6_scope_id etc */
1012 		sin6 = (struct sockaddr_in6 *)name;
1013 		if (sin6->sin6_scope_id != 0 &&
1014 		    !IN6_IS_ADDR_LINKSCOPE(&sin6->sin6_addr)) {
1015 			zcmn_err(getzoneid(), CE_WARN,
1016 			    "connect/send* with uninitialized sin6_scope_id "
1017 			    "(%d) on socket. Pid = %d\n",
1018 			    (int)sin6->sin6_scope_id, (int)curproc->p_pid);
1019 		}
1020 #endif /* DEBUG */
1021 		break;
1022 	}
1023 	case AF_UNIX:
1024 		if (so->so_state & SS_FADDR_NOXLATE) {
1025 			return (0);
1026 		}
1027 		if (namelen < (socklen_t)sizeof (short)) {
1028 			eprintsoline(so, ENOENT);
1029 			return (ENOENT);
1030 		}
1031 		if (name->sa_family != family) {
1032 			eprintsoline(so, EAFNOSUPPORT);
1033 			return (EAFNOSUPPORT);
1034 		}
1035 		/* MAXPATHLEN + soun_family + nul termination */
1036 		if (namelen > (socklen_t)(MAXPATHLEN + sizeof (short) + 1)) {
1037 			eprintsoline(so, ENAMETOOLONG);
1038 			return (ENAMETOOLONG);
1039 		}
1040 
1041 		break;
1042 
1043 	default:
1044 		/*
1045 		 * Default is don't do any length or sa_family check
1046 		 * to allow non-sockaddr style addresses.
1047 		 */
1048 		break;
1049 	}
1050 
1051 	return (0);
1052 }
1053 
1054 
1055 /*
1056  * Translate an AF_UNIX sockaddr_un to the transport internal name.
1057  * Assumes caller has called so_addr_verify first.
1058  */
1059 /*ARGSUSED*/
1060 int
1061 so_ux_addr_xlate(struct sonode *so, struct sockaddr *name,
1062     socklen_t namelen, int checkaccess,
1063     void **addrp, socklen_t *addrlenp)
1064 {
1065 	int			error;
1066 	struct sockaddr_un	*soun;
1067 	vnode_t			*vp;
1068 	void			*addr;
1069 	socklen_t		addrlen;
1070 
1071 	dprintso(so, 1, ("so_ux_addr_xlate(%p, %p, %d, %d)\n",
1072 			so, name, namelen, checkaccess));
1073 
1074 	ASSERT(name != NULL);
1075 	ASSERT(so->so_family == AF_UNIX);
1076 	ASSERT(!(so->so_state & SS_FADDR_NOXLATE));
1077 	ASSERT(namelen >= (socklen_t)sizeof (short));
1078 	ASSERT(name->sa_family == AF_UNIX);
1079 	soun = (struct sockaddr_un *)name;
1080 	/*
1081 	 * Lookup vnode for the specified path name and verify that
1082 	 * it is a socket.
1083 	 */
1084 	error = so_ux_lookup(so, soun, checkaccess, &vp);
1085 	if (error) {
1086 		eprintsoline(so, error);
1087 		return (error);
1088 	}
1089 	/*
1090 	 * Use the address of the peer vnode as the address to send
1091 	 * to. We release the peer vnode here. In case it has been
1092 	 * closed by the time the T_CONN_REQ or T_UNIDATA_REQ reaches the
1093 	 * transport the message will get an error or be dropped.
1094 	 */
1095 	so->so_ux_faddr.soua_vp = vp;
1096 	so->so_ux_faddr.soua_magic = SOU_MAGIC_EXPLICIT;
1097 	addr = &so->so_ux_faddr;
1098 	addrlen = (socklen_t)sizeof (so->so_ux_faddr);
1099 	dprintso(so, 1, ("ux_xlate UNIX: addrlen %d, vp %p\n",
1100 				addrlen, vp));
1101 	VN_RELE(vp);
1102 	*addrp = addr;
1103 	*addrlenp = (socklen_t)addrlen;
1104 	return (0);
1105 }
1106 
1107 /*
1108  * Esballoc free function for messages that contain SO_FILEP option.
1109  * Decrement the reference count on the file pointers using closef.
1110  */
1111 void
1112 fdbuf_free(struct fdbuf *fdbuf)
1113 {
1114 	int	i;
1115 	struct file *fp;
1116 
1117 	dprint(1, ("fdbuf_free: %d fds\n", fdbuf->fd_numfd));
1118 	for (i = 0; i < fdbuf->fd_numfd; i++) {
1119 		/*
1120 		 * We need pointer size alignment for fd_fds. On a LP64
1121 		 * kernel, the required alignment is 8 bytes while
1122 		 * the option headers and values are only 4 bytes
1123 		 * aligned. So its safer to do a bcopy compared to
1124 		 * assigning fdbuf->fd_fds[i] to fp.
1125 		 */
1126 		bcopy((char *)&fdbuf->fd_fds[i], (char *)&fp, sizeof (fp));
1127 		dprint(1, ("fdbuf_free: [%d] = %p\n", i, fp));
1128 		(void) closef(fp);
1129 	}
1130 	if (fdbuf->fd_ebuf != NULL)
1131 		kmem_free(fdbuf->fd_ebuf, fdbuf->fd_ebuflen);
1132 	kmem_free(fdbuf, fdbuf->fd_size);
1133 }
1134 
1135 /*
1136  * Allocate an esballoc'ed message for AF_UNIX file descriptor passing.
1137  * Waits if memory is not available.
1138  */
1139 mblk_t *
1140 fdbuf_allocmsg(int size, struct fdbuf *fdbuf)
1141 {
1142 	uchar_t	*buf;
1143 	mblk_t	*mp;
1144 
1145 	dprint(1, ("fdbuf_allocmsg: size %d, %d fds\n", size, fdbuf->fd_numfd));
1146 	buf = kmem_alloc(size, KM_SLEEP);
1147 	fdbuf->fd_ebuf = (caddr_t)buf;
1148 	fdbuf->fd_ebuflen = size;
1149 	fdbuf->fd_frtn.free_func = fdbuf_free;
1150 	fdbuf->fd_frtn.free_arg = (caddr_t)fdbuf;
1151 
1152 	mp = esballoc_wait(buf, size, BPRI_MED, &fdbuf->fd_frtn);
1153 	mp->b_datap->db_type = M_PROTO;
1154 	return (mp);
1155 }
1156 
1157 /*
1158  * Extract file descriptors from a fdbuf.
1159  * Return list in rights/rightslen.
1160  */
1161 /*ARGSUSED*/
1162 static int
1163 fdbuf_extract(struct fdbuf *fdbuf, void *rights, int rightslen)
1164 {
1165 	int	i, fd;
1166 	int	*rp;
1167 	struct file *fp;
1168 	int	numfd;
1169 
1170 	dprint(1, ("fdbuf_extract: %d fds, len %d\n",
1171 		fdbuf->fd_numfd, rightslen));
1172 
1173 	numfd = fdbuf->fd_numfd;
1174 	ASSERT(rightslen == numfd * (int)sizeof (int));
1175 
1176 	/*
1177 	 * Allocate a file descriptor and increment the f_count.
1178 	 * The latter is needed since we always call fdbuf_free
1179 	 * which performs a closef.
1180 	 */
1181 	rp = (int *)rights;
1182 	for (i = 0; i < numfd; i++) {
1183 		if ((fd = ufalloc(0)) == -1)
1184 			goto cleanup;
1185 		/*
1186 		 * We need pointer size alignment for fd_fds. On a LP64
1187 		 * kernel, the required alignment is 8 bytes while
1188 		 * the option headers and values are only 4 bytes
1189 		 * aligned. So its safer to do a bcopy compared to
1190 		 * assigning fdbuf->fd_fds[i] to fp.
1191 		 */
1192 		bcopy((char *)&fdbuf->fd_fds[i], (char *)&fp, sizeof (fp));
1193 		mutex_enter(&fp->f_tlock);
1194 		fp->f_count++;
1195 		mutex_exit(&fp->f_tlock);
1196 		setf(fd, fp);
1197 		*rp++ = fd;
1198 #ifdef C2_AUDIT
1199 		if (audit_active)
1200 			audit_fdrecv(fd, fp);
1201 #endif
1202 		dprint(1, ("fdbuf_extract: [%d] = %d, %p refcnt %d\n",
1203 			i, fd, fp, fp->f_count));
1204 	}
1205 	return (0);
1206 
1207 cleanup:
1208 	/*
1209 	 * Undo whatever partial work the loop above has done.
1210 	 */
1211 	{
1212 		int j;
1213 
1214 		rp = (int *)rights;
1215 		for (j = 0; j < i; j++) {
1216 			dprint(0,
1217 			    ("fdbuf_extract: cleanup[%d] = %d\n", j, *rp));
1218 			(void) closeandsetf(*rp++, NULL);
1219 		}
1220 	}
1221 
1222 	return (EMFILE);
1223 }
1224 
1225 /*
1226  * Insert file descriptors into an fdbuf.
1227  * Returns a kmem_alloc'ed fdbuf. The fdbuf should be freed
1228  * by calling fdbuf_free().
1229  */
1230 int
1231 fdbuf_create(void *rights, int rightslen, struct fdbuf **fdbufp)
1232 {
1233 	int		numfd, i;
1234 	int		*fds;
1235 	struct file	*fp;
1236 	struct fdbuf	*fdbuf;
1237 	int		fdbufsize;
1238 
1239 	dprint(1, ("fdbuf_create: len %d\n", rightslen));
1240 
1241 	numfd = rightslen / (int)sizeof (int);
1242 
1243 	fdbufsize = (int)FDBUF_HDRSIZE + (numfd * (int)sizeof (struct file *));
1244 	fdbuf = kmem_alloc(fdbufsize, KM_SLEEP);
1245 	fdbuf->fd_size = fdbufsize;
1246 	fdbuf->fd_numfd = 0;
1247 	fdbuf->fd_ebuf = NULL;
1248 	fdbuf->fd_ebuflen = 0;
1249 	fds = (int *)rights;
1250 	for (i = 0; i < numfd; i++) {
1251 		if ((fp = getf(fds[i])) == NULL) {
1252 			fdbuf_free(fdbuf);
1253 			return (EBADF);
1254 		}
1255 		dprint(1, ("fdbuf_create: [%d] = %d, %p refcnt %d\n",
1256 			i, fds[i], fp, fp->f_count));
1257 		mutex_enter(&fp->f_tlock);
1258 		fp->f_count++;
1259 		mutex_exit(&fp->f_tlock);
1260 		/*
1261 		 * The maximum alignment for fdbuf (or any option header
1262 		 * and its value) it 4 bytes. On a LP64 kernel, the alignment
1263 		 * is not sufficient for pointers (fd_fds in this case). Since
1264 		 * we just did a kmem_alloc (we get a double word alignment),
1265 		 * we don't need to do anything on the send side (we loose
1266 		 * the double word alignment because fdbuf goes after an
1267 		 * option header (eg T_unitdata_req) which is only 4 byte
1268 		 * aligned). We take care of this when we extract the file
1269 		 * descriptor in fdbuf_extract or fdbuf_free.
1270 		 */
1271 		fdbuf->fd_fds[i] = fp;
1272 		fdbuf->fd_numfd++;
1273 		releasef(fds[i]);
1274 #ifdef C2_AUDIT
1275 		if (audit_active)
1276 			audit_fdsend(fds[i], fp, 0);
1277 #endif
1278 	}
1279 	*fdbufp = fdbuf;
1280 	return (0);
1281 }
1282 
1283 static int
1284 fdbuf_optlen(int rightslen)
1285 {
1286 	int numfd;
1287 
1288 	numfd = rightslen / (int)sizeof (int);
1289 
1290 	return ((int)FDBUF_HDRSIZE + (numfd * (int)sizeof (struct file *)));
1291 }
1292 
1293 static t_uscalar_t
1294 fdbuf_cmsglen(int fdbuflen)
1295 {
1296 	return (t_uscalar_t)((fdbuflen - FDBUF_HDRSIZE) /
1297 	    (int)sizeof (struct file *) * (int)sizeof (int));
1298 }
1299 
1300 
1301 /*
1302  * Return non-zero if the mblk and fdbuf are consistent.
1303  */
1304 static int
1305 fdbuf_verify(mblk_t *mp, struct fdbuf *fdbuf, int fdbuflen)
1306 {
1307 	if (fdbuflen >= FDBUF_HDRSIZE &&
1308 	    fdbuflen == fdbuf->fd_size) {
1309 		frtn_t *frp = mp->b_datap->db_frtnp;
1310 		/*
1311 		 * Check that the SO_FILEP portion of the
1312 		 * message has not been modified by
1313 		 * the loopback transport. The sending sockfs generates
1314 		 * a message that is esballoc'ed with the free function
1315 		 * being fdbuf_free() and where free_arg contains the
1316 		 * identical information as the SO_FILEP content.
1317 		 *
1318 		 * If any of these constraints are not satisfied we
1319 		 * silently ignore the option.
1320 		 */
1321 		ASSERT(mp);
1322 		if (frp != NULL &&
1323 		    frp->free_func == fdbuf_free &&
1324 		    frp->free_arg != NULL &&
1325 		    bcmp(frp->free_arg, fdbuf, fdbuflen) == 0) {
1326 			dprint(1, ("fdbuf_verify: fdbuf %p len %d\n",
1327 				fdbuf, fdbuflen));
1328 			return (1);
1329 		} else {
1330 			zcmn_err(getzoneid(), CE_WARN,
1331 			    "sockfs: mismatched fdbuf content (%p)",
1332 			    (void *)mp);
1333 			return (0);
1334 		}
1335 	} else {
1336 		zcmn_err(getzoneid(), CE_WARN,
1337 		    "sockfs: mismatched fdbuf len %d, %d\n",
1338 		    fdbuflen, fdbuf->fd_size);
1339 		return (0);
1340 	}
1341 }
1342 
1343 /*
1344  * When the file descriptors returned by sorecvmsg can not be passed
1345  * to the application this routine will cleanup the references on
1346  * the files. Start at startoff bytes into the buffer.
1347  */
1348 static void
1349 close_fds(void *fdbuf, int fdbuflen, int startoff)
1350 {
1351 	int *fds = (int *)fdbuf;
1352 	int numfd = fdbuflen / (int)sizeof (int);
1353 	int i;
1354 
1355 	dprint(1, ("close_fds(%p, %d, %d)\n", fdbuf, fdbuflen, startoff));
1356 
1357 	for (i = 0; i < numfd; i++) {
1358 		if (startoff < 0)
1359 			startoff = 0;
1360 		if (startoff < (int)sizeof (int)) {
1361 			/*
1362 			 * This file descriptor is partially or fully after
1363 			 * the offset
1364 			 */
1365 			dprint(0,
1366 			    ("close_fds: cleanup[%d] = %d\n", i, fds[i]));
1367 			(void) closeandsetf(fds[i], NULL);
1368 		}
1369 		startoff -= (int)sizeof (int);
1370 	}
1371 }
1372 
1373 /*
1374  * Close all file descriptors contained in the control part starting at
1375  * the startoffset.
1376  */
1377 void
1378 so_closefds(void *control, t_uscalar_t controllen, int oldflg,
1379     int startoff)
1380 {
1381 	struct cmsghdr *cmsg;
1382 
1383 	if (control == NULL)
1384 		return;
1385 
1386 	if (oldflg) {
1387 		close_fds(control, controllen, startoff);
1388 		return;
1389 	}
1390 	/* Scan control part for file descriptors. */
1391 	for (cmsg = (struct cmsghdr *)control;
1392 	    CMSG_VALID(cmsg, control, (uintptr_t)control + controllen);
1393 	    cmsg = CMSG_NEXT(cmsg)) {
1394 		if (cmsg->cmsg_level == SOL_SOCKET &&
1395 		    cmsg->cmsg_type == SCM_RIGHTS) {
1396 			close_fds(CMSG_CONTENT(cmsg),
1397 			    (int)CMSG_CONTENTLEN(cmsg),
1398 			    startoff - (int)sizeof (struct cmsghdr));
1399 		}
1400 		startoff -= cmsg->cmsg_len;
1401 	}
1402 }
1403 
1404 /*
1405  * Returns a pointer/length for the file descriptors contained
1406  * in the control buffer. Returns with *fdlenp == -1 if there are no
1407  * file descriptor options present. This is different than there being
1408  * a zero-length file descriptor option.
1409  * Fail if there are multiple SCM_RIGHT cmsgs.
1410  */
1411 int
1412 so_getfdopt(void *control, t_uscalar_t controllen, int oldflg,
1413     void **fdsp, int *fdlenp)
1414 {
1415 	struct cmsghdr *cmsg;
1416 	void *fds;
1417 	int fdlen;
1418 
1419 	if (control == NULL) {
1420 		*fdsp = NULL;
1421 		*fdlenp = -1;
1422 		return (0);
1423 	}
1424 
1425 	if (oldflg) {
1426 		*fdsp = control;
1427 		if (controllen == 0)
1428 			*fdlenp = -1;
1429 		else
1430 			*fdlenp = controllen;
1431 		dprint(1, ("so_getfdopt: old %d\n", *fdlenp));
1432 		return (0);
1433 	}
1434 
1435 	fds = NULL;
1436 	fdlen = 0;
1437 
1438 	for (cmsg = (struct cmsghdr *)control;
1439 	    CMSG_VALID(cmsg, control, (uintptr_t)control + controllen);
1440 	    cmsg = CMSG_NEXT(cmsg)) {
1441 		if (cmsg->cmsg_level == SOL_SOCKET &&
1442 		    cmsg->cmsg_type == SCM_RIGHTS) {
1443 			if (fds != NULL)
1444 				return (EINVAL);
1445 			fds = CMSG_CONTENT(cmsg);
1446 			fdlen = (int)CMSG_CONTENTLEN(cmsg);
1447 			dprint(1, ("so_getfdopt: new %lu\n",
1448 				(size_t)CMSG_CONTENTLEN(cmsg)));
1449 		}
1450 	}
1451 	if (fds == NULL) {
1452 		dprint(1, ("so_getfdopt: NONE\n"));
1453 		*fdlenp = -1;
1454 	} else
1455 		*fdlenp = fdlen;
1456 	*fdsp = fds;
1457 	return (0);
1458 }
1459 
1460 /*
1461  * Return the length of the options including any file descriptor options.
1462  */
1463 t_uscalar_t
1464 so_optlen(void *control, t_uscalar_t controllen, int oldflg)
1465 {
1466 	struct cmsghdr *cmsg;
1467 	t_uscalar_t optlen = 0;
1468 	t_uscalar_t len;
1469 
1470 	if (control == NULL)
1471 		return (0);
1472 
1473 	if (oldflg)
1474 		return ((t_uscalar_t)(sizeof (struct T_opthdr) +
1475 		    fdbuf_optlen(controllen)));
1476 
1477 	for (cmsg = (struct cmsghdr *)control;
1478 	    CMSG_VALID(cmsg, control, (uintptr_t)control + controllen);
1479 	    cmsg = CMSG_NEXT(cmsg)) {
1480 		if (cmsg->cmsg_level == SOL_SOCKET &&
1481 		    cmsg->cmsg_type == SCM_RIGHTS) {
1482 			len = fdbuf_optlen((int)CMSG_CONTENTLEN(cmsg));
1483 		} else {
1484 			len = (t_uscalar_t)CMSG_CONTENTLEN(cmsg);
1485 		}
1486 		optlen += (t_uscalar_t)(_TPI_ALIGN_TOPT(len) +
1487 		    sizeof (struct T_opthdr));
1488 	}
1489 	dprint(1, ("so_optlen: controllen %d, flg %d -> optlen %d\n",
1490 		controllen, oldflg, optlen));
1491 	return (optlen);
1492 }
1493 
1494 /*
1495  * Copy options from control to the mblk. Skip any file descriptor options.
1496  */
1497 void
1498 so_cmsg2opt(void *control, t_uscalar_t controllen, int oldflg, mblk_t *mp)
1499 {
1500 	struct T_opthdr toh;
1501 	struct cmsghdr *cmsg;
1502 
1503 	if (control == NULL)
1504 		return;
1505 
1506 	if (oldflg) {
1507 		/* No real options - caller has handled file descriptors */
1508 		return;
1509 	}
1510 	for (cmsg = (struct cmsghdr *)control;
1511 	    CMSG_VALID(cmsg, control, (uintptr_t)control + controllen);
1512 	    cmsg = CMSG_NEXT(cmsg)) {
1513 		/*
1514 		 * Note: The caller handles file descriptors prior
1515 		 * to calling this function.
1516 		 */
1517 		t_uscalar_t len;
1518 
1519 		if (cmsg->cmsg_level == SOL_SOCKET &&
1520 		    cmsg->cmsg_type == SCM_RIGHTS)
1521 			continue;
1522 
1523 		len = (t_uscalar_t)CMSG_CONTENTLEN(cmsg);
1524 		toh.level = cmsg->cmsg_level;
1525 		toh.name = cmsg->cmsg_type;
1526 		toh.len = len + (t_uscalar_t)sizeof (struct T_opthdr);
1527 		toh.status = 0;
1528 
1529 		soappendmsg(mp, &toh, sizeof (toh));
1530 		soappendmsg(mp, CMSG_CONTENT(cmsg), len);
1531 		mp->b_wptr += _TPI_ALIGN_TOPT(len) - len;
1532 		ASSERT(mp->b_wptr <= mp->b_datap->db_lim);
1533 	}
1534 }
1535 
1536 /*
1537  * Return the length of the control message derived from the options.
1538  * Exclude SO_SRCADDR and SO_UNIX_CLOSE options. Include SO_FILEP.
1539  * When oldflg is set only include SO_FILEP.
1540  */
1541 t_uscalar_t
1542 so_cmsglen(mblk_t *mp, void *opt, t_uscalar_t optlen, int oldflg)
1543 {
1544 	t_uscalar_t cmsglen = 0;
1545 	struct T_opthdr *tohp;
1546 	t_uscalar_t len;
1547 	t_uscalar_t last_roundup = 0;
1548 
1549 	ASSERT(__TPI_TOPT_ISALIGNED(opt));
1550 
1551 	for (tohp = (struct T_opthdr *)opt;
1552 	    tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen);
1553 	    tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) {
1554 		dprint(1, ("so_cmsglen: level 0x%x, name %d, len %d\n",
1555 			tohp->level, tohp->name, tohp->len));
1556 		if (tohp->level == SOL_SOCKET &&
1557 		    (tohp->name == SO_SRCADDR ||
1558 		    tohp->name == SO_UNIX_CLOSE)) {
1559 			continue;
1560 		}
1561 		if (tohp->level == SOL_SOCKET && tohp->name == SO_FILEP) {
1562 			struct fdbuf *fdbuf;
1563 			int fdbuflen;
1564 
1565 			fdbuf = (struct fdbuf *)_TPI_TOPT_DATA(tohp);
1566 			fdbuflen = (int)_TPI_TOPT_DATALEN(tohp);
1567 
1568 			if (!fdbuf_verify(mp, fdbuf, fdbuflen))
1569 				continue;
1570 			if (oldflg) {
1571 				cmsglen += fdbuf_cmsglen(fdbuflen);
1572 				continue;
1573 			}
1574 			len = fdbuf_cmsglen(fdbuflen);
1575 		} else {
1576 			if (oldflg)
1577 				continue;
1578 			len = (t_uscalar_t)_TPI_TOPT_DATALEN(tohp);
1579 		}
1580 		/*
1581 		 * Exlucde roundup for last option to not set
1582 		 * MSG_CTRUNC when the cmsg fits but the padding doesn't fit.
1583 		 */
1584 		last_roundup = (t_uscalar_t)
1585 		    (ROUNDUP_cmsglen(len + (int)sizeof (struct cmsghdr)) -
1586 		    (len + (int)sizeof (struct cmsghdr)));
1587 		cmsglen += (t_uscalar_t)(len + (int)sizeof (struct cmsghdr)) +
1588 		    last_roundup;
1589 	}
1590 	cmsglen -= last_roundup;
1591 	dprint(1, ("so_cmsglen: optlen %d, flg %d -> cmsglen %d\n",
1592 		optlen, oldflg, cmsglen));
1593 	return (cmsglen);
1594 }
1595 
1596 /*
1597  * Copy options from options to the control. Convert SO_FILEP to
1598  * file descriptors.
1599  * Returns errno or zero.
1600  */
1601 int
1602 so_opt2cmsg(mblk_t *mp, void *opt, t_uscalar_t optlen, int oldflg,
1603     void *control, t_uscalar_t controllen)
1604 {
1605 	struct T_opthdr *tohp;
1606 	struct cmsghdr *cmsg;
1607 	struct fdbuf *fdbuf;
1608 	int fdbuflen;
1609 	int error;
1610 
1611 	cmsg = (struct cmsghdr *)control;
1612 
1613 	ASSERT(__TPI_TOPT_ISALIGNED(opt));
1614 
1615 	for (tohp = (struct T_opthdr *)opt;
1616 	    tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen);
1617 	    tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) {
1618 		dprint(1, ("so_opt2cmsg: level 0x%x, name %d, len %d\n",
1619 			tohp->level, tohp->name, tohp->len));
1620 
1621 		if (tohp->level == SOL_SOCKET &&
1622 		    (tohp->name == SO_SRCADDR ||
1623 		    tohp->name == SO_UNIX_CLOSE)) {
1624 			continue;
1625 		}
1626 		ASSERT((uintptr_t)cmsg <= (uintptr_t)control + controllen);
1627 		if (tohp->level == SOL_SOCKET && tohp->name == SO_FILEP) {
1628 			fdbuf = (struct fdbuf *)_TPI_TOPT_DATA(tohp);
1629 			fdbuflen = (int)_TPI_TOPT_DATALEN(tohp);
1630 
1631 			if (!fdbuf_verify(mp, fdbuf, fdbuflen))
1632 				return (EPROTO);
1633 			if (oldflg) {
1634 				error = fdbuf_extract(fdbuf, control,
1635 				    (int)controllen);
1636 				if (error != 0)
1637 					return (error);
1638 				continue;
1639 			} else {
1640 				int fdlen;
1641 
1642 				fdlen = (int)fdbuf_cmsglen(
1643 				    (int)_TPI_TOPT_DATALEN(tohp));
1644 
1645 				cmsg->cmsg_level = tohp->level;
1646 				cmsg->cmsg_type = SCM_RIGHTS;
1647 				cmsg->cmsg_len = (socklen_t)(fdlen +
1648 					sizeof (struct cmsghdr));
1649 
1650 				error = fdbuf_extract(fdbuf,
1651 						CMSG_CONTENT(cmsg), fdlen);
1652 				if (error != 0)
1653 					return (error);
1654 			}
1655 		} else if (tohp->level == SOL_SOCKET &&
1656 		    tohp->name == SCM_TIMESTAMP) {
1657 			timestruc_t *timestamp;
1658 
1659 			if (oldflg)
1660 				continue;
1661 
1662 			cmsg->cmsg_level = tohp->level;
1663 			cmsg->cmsg_type = tohp->name;
1664 
1665 			timestamp =
1666 			    (timestruc_t *)P2ROUNDUP((intptr_t)&tohp[1],
1667 			    sizeof (intptr_t));
1668 
1669 			if (get_udatamodel() == DATAMODEL_NATIVE) {
1670 				struct timeval *time_native;
1671 
1672 				cmsg->cmsg_len = sizeof (struct timeval) +
1673 				    sizeof (struct cmsghdr);
1674 				time_native =
1675 				    (struct timeval *)CMSG_CONTENT(cmsg);
1676 				time_native->tv_sec = timestamp->tv_sec;
1677 				time_native->tv_usec =
1678 				    timestamp->tv_nsec / (NANOSEC / MICROSEC);
1679 			} else {
1680 				struct timeval32 *time32;
1681 
1682 				cmsg->cmsg_len = sizeof (struct timeval32) +
1683 				    sizeof (struct cmsghdr);
1684 				time32 = (struct timeval32 *)CMSG_CONTENT(cmsg);
1685 				time32->tv_sec = (time32_t)timestamp->tv_sec;
1686 				time32->tv_usec =
1687 				    (int32_t)(timestamp->tv_nsec /
1688 				    (NANOSEC / MICROSEC));
1689 			}
1690 
1691 		} else {
1692 			if (oldflg)
1693 				continue;
1694 
1695 			cmsg->cmsg_level = tohp->level;
1696 			cmsg->cmsg_type = tohp->name;
1697 			cmsg->cmsg_len = (socklen_t)(_TPI_TOPT_DATALEN(tohp) +
1698 			    sizeof (struct cmsghdr));
1699 
1700 			/* copy content to control data part */
1701 			bcopy(&tohp[1], CMSG_CONTENT(cmsg),
1702 				CMSG_CONTENTLEN(cmsg));
1703 		}
1704 		/* move to next CMSG structure! */
1705 		cmsg = CMSG_NEXT(cmsg);
1706 	}
1707 	return (0);
1708 }
1709 
1710 /*
1711  * Extract the SO_SRCADDR option value if present.
1712  */
1713 void
1714 so_getopt_srcaddr(void *opt, t_uscalar_t optlen, void **srcp,
1715     t_uscalar_t *srclenp)
1716 {
1717 	struct T_opthdr		*tohp;
1718 
1719 	ASSERT(__TPI_TOPT_ISALIGNED(opt));
1720 
1721 	ASSERT(srcp != NULL && srclenp != NULL);
1722 	*srcp = NULL;
1723 	*srclenp = 0;
1724 
1725 	for (tohp = (struct T_opthdr *)opt;
1726 	    tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen);
1727 	    tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) {
1728 		dprint(1, ("so_getopt_srcaddr: level 0x%x, name %d, len %d\n",
1729 			tohp->level, tohp->name, tohp->len));
1730 		if (tohp->level == SOL_SOCKET &&
1731 		    tohp->name == SO_SRCADDR) {
1732 			*srcp = _TPI_TOPT_DATA(tohp);
1733 			*srclenp = (t_uscalar_t)_TPI_TOPT_DATALEN(tohp);
1734 		}
1735 	}
1736 }
1737 
1738 /*
1739  * Verify if the SO_UNIX_CLOSE option is present.
1740  */
1741 int
1742 so_getopt_unix_close(void *opt, t_uscalar_t optlen)
1743 {
1744 	struct T_opthdr		*tohp;
1745 
1746 	ASSERT(__TPI_TOPT_ISALIGNED(opt));
1747 
1748 	for (tohp = (struct T_opthdr *)opt;
1749 	    tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen);
1750 	    tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) {
1751 		dprint(1,
1752 			("so_getopt_unix_close: level 0x%x, name %d, len %d\n",
1753 			tohp->level, tohp->name, tohp->len));
1754 		if (tohp->level == SOL_SOCKET &&
1755 		    tohp->name == SO_UNIX_CLOSE)
1756 			return (1);
1757 	}
1758 	return (0);
1759 }
1760 
1761 /*
1762  * Allocate an M_PROTO message.
1763  *
1764  * If allocation fails the behavior depends on sleepflg:
1765  *	_ALLOC_NOSLEEP	fail immediately
1766  *	_ALLOC_INTR	sleep for memory until a signal is caught
1767  *	_ALLOC_SLEEP	sleep forever. Don't return NULL.
1768  */
1769 mblk_t *
1770 soallocproto(size_t size, int sleepflg)
1771 {
1772 	mblk_t	*mp;
1773 
1774 	/* Round up size for reuse */
1775 	size = MAX(size, 64);
1776 	mp = allocb(size, BPRI_MED);
1777 	if (mp == NULL) {
1778 		int error;	/* Dummy - error not returned to caller */
1779 
1780 		switch (sleepflg) {
1781 		case _ALLOC_SLEEP:
1782 			mp = allocb_wait(size, BPRI_MED, STR_NOSIG, &error);
1783 			ASSERT(mp);
1784 			break;
1785 		case _ALLOC_INTR:
1786 			mp = allocb_wait(size, BPRI_MED, 0, &error);
1787 			if (mp == NULL) {
1788 				/* Caught signal while sleeping for memory */
1789 				eprintline(ENOBUFS);
1790 				return (NULL);
1791 			}
1792 			break;
1793 		case _ALLOC_NOSLEEP:
1794 		default:
1795 			eprintline(ENOBUFS);
1796 			return (NULL);
1797 		}
1798 	}
1799 	DB_TYPE(mp) = M_PROTO;
1800 	return (mp);
1801 }
1802 
1803 /*
1804  * Allocate an M_PROTO message with a single component.
1805  * len is the length of buf. size is the amount to allocate.
1806  *
1807  * buf can be NULL with a non-zero len.
1808  * This results in a bzero'ed chunk being placed the message.
1809  */
1810 mblk_t *
1811 soallocproto1(const void *buf, ssize_t len, ssize_t size, int sleepflg)
1812 {
1813 	mblk_t	*mp;
1814 
1815 	if (size == 0)
1816 		size = len;
1817 
1818 	ASSERT(size >= len);
1819 	/* Round up size for reuse */
1820 	size = MAX(size, 64);
1821 	mp = soallocproto(size, sleepflg);
1822 	if (mp == NULL)
1823 		return (NULL);
1824 	mp->b_datap->db_type = M_PROTO;
1825 	if (len != 0) {
1826 		if (buf != NULL)
1827 			bcopy(buf, mp->b_wptr, len);
1828 		else
1829 			bzero(mp->b_wptr, len);
1830 		mp->b_wptr += len;
1831 	}
1832 	return (mp);
1833 }
1834 
1835 /*
1836  * Append buf/len to mp.
1837  * The caller has to ensure that there is enough room in the mblk.
1838  *
1839  * buf can be NULL with a non-zero len.
1840  * This results in a bzero'ed chunk being placed the message.
1841  */
1842 void
1843 soappendmsg(mblk_t *mp, const void *buf, ssize_t len)
1844 {
1845 	ASSERT(mp);
1846 
1847 	if (len != 0) {
1848 		/* Assert for room left */
1849 		ASSERT(mp->b_datap->db_lim - mp->b_wptr >= len);
1850 		if (buf != NULL)
1851 			bcopy(buf, mp->b_wptr, len);
1852 		else
1853 			bzero(mp->b_wptr, len);
1854 	}
1855 	mp->b_wptr += len;
1856 }
1857 
1858 /*
1859  * Create a message using two kernel buffers.
1860  * If size is set that will determine the allocation size (e.g. for future
1861  * soappendmsg calls). If size is zero it is derived from the buffer
1862  * lengths.
1863  */
1864 mblk_t *
1865 soallocproto2(const void *buf1, ssize_t len1, const void *buf2, ssize_t len2,
1866     ssize_t size, int sleepflg)
1867 {
1868 	mblk_t *mp;
1869 
1870 	if (size == 0)
1871 		size = len1 + len2;
1872 	ASSERT(size >= len1 + len2);
1873 
1874 	mp = soallocproto1(buf1, len1, size, sleepflg);
1875 	if (mp)
1876 		soappendmsg(mp, buf2, len2);
1877 	return (mp);
1878 }
1879 
1880 /*
1881  * Create a message using three kernel buffers.
1882  * If size is set that will determine the allocation size (for future
1883  * soappendmsg calls). If size is zero it is derived from the buffer
1884  * lengths.
1885  */
1886 mblk_t *
1887 soallocproto3(const void *buf1, ssize_t len1, const void *buf2, ssize_t len2,
1888     const void *buf3, ssize_t len3, ssize_t size, int sleepflg)
1889 {
1890 	mblk_t *mp;
1891 
1892 	if (size == 0)
1893 		size = len1 + len2 +len3;
1894 	ASSERT(size >= len1 + len2 + len3);
1895 
1896 	mp = soallocproto1(buf1, len1, size, sleepflg);
1897 	if (mp != NULL) {
1898 		soappendmsg(mp, buf2, len2);
1899 		soappendmsg(mp, buf3, len3);
1900 	}
1901 	return (mp);
1902 }
1903 
1904 #ifdef DEBUG
1905 char *
1906 pr_state(uint_t state, uint_t mode)
1907 {
1908 	static char buf[1024];
1909 
1910 	buf[0] = 0;
1911 	if (state & SS_ISCONNECTED)
1912 		strcat(buf, "ISCONNECTED ");
1913 	if (state & SS_ISCONNECTING)
1914 		strcat(buf, "ISCONNECTING ");
1915 	if (state & SS_ISDISCONNECTING)
1916 		strcat(buf, "ISDISCONNECTING ");
1917 	if (state & SS_CANTSENDMORE)
1918 		strcat(buf, "CANTSENDMORE ");
1919 
1920 	if (state & SS_CANTRCVMORE)
1921 		strcat(buf, "CANTRCVMORE ");
1922 	if (state & SS_ISBOUND)
1923 		strcat(buf, "ISBOUND ");
1924 	if (state & SS_NDELAY)
1925 		strcat(buf, "NDELAY ");
1926 	if (state & SS_NONBLOCK)
1927 		strcat(buf, "NONBLOCK ");
1928 
1929 	if (state & SS_ASYNC)
1930 		strcat(buf, "ASYNC ");
1931 	if (state & SS_ACCEPTCONN)
1932 		strcat(buf, "ACCEPTCONN ");
1933 	if (state & SS_HASCONNIND)
1934 		strcat(buf, "HASCONNIND ");
1935 	if (state & SS_SAVEDEOR)
1936 		strcat(buf, "SAVEDEOR ");
1937 
1938 	if (state & SS_RCVATMARK)
1939 		strcat(buf, "RCVATMARK ");
1940 	if (state & SS_OOBPEND)
1941 		strcat(buf, "OOBPEND ");
1942 	if (state & SS_HAVEOOBDATA)
1943 		strcat(buf, "HAVEOOBDATA ");
1944 	if (state & SS_HADOOBDATA)
1945 		strcat(buf, "HADOOBDATA ");
1946 
1947 	if (state & SS_FADDR_NOXLATE)
1948 		strcat(buf, "FADDR_NOXLATE ");
1949 
1950 	if (mode & SM_PRIV)
1951 		strcat(buf, "PRIV ");
1952 	if (mode & SM_ATOMIC)
1953 		strcat(buf, "ATOMIC ");
1954 	if (mode & SM_ADDR)
1955 		strcat(buf, "ADDR ");
1956 	if (mode & SM_CONNREQUIRED)
1957 		strcat(buf, "CONNREQUIRED ");
1958 
1959 	if (mode & SM_FDPASSING)
1960 		strcat(buf, "FDPASSING ");
1961 	if (mode & SM_EXDATA)
1962 		strcat(buf, "EXDATA ");
1963 	if (mode & SM_OPTDATA)
1964 		strcat(buf, "OPTDATA ");
1965 	if (mode & SM_BYTESTREAM)
1966 		strcat(buf, "BYTESTREAM ");
1967 	return (buf);
1968 }
1969 
1970 char *
1971 pr_addr(int family, struct sockaddr *addr, t_uscalar_t addrlen)
1972 {
1973 	static char buf[1024];
1974 
1975 	if (addr == NULL || addrlen == 0) {
1976 		sprintf(buf, "(len %d) %p", addrlen, addr);
1977 		return (buf);
1978 	}
1979 	switch (family) {
1980 	case AF_INET: {
1981 		struct sockaddr_in sin;
1982 
1983 		bcopy(addr, &sin, sizeof (sin));
1984 
1985 		(void) sprintf(buf, "(len %d) %x/%d",
1986 			addrlen, ntohl(sin.sin_addr.s_addr),
1987 			ntohs(sin.sin_port));
1988 		break;
1989 	}
1990 	case AF_INET6: {
1991 		struct sockaddr_in6 sin6;
1992 		uint16_t *piece = (uint16_t *)&sin6.sin6_addr;
1993 
1994 		bcopy((char *)addr, (char *)&sin6, sizeof (sin6));
1995 		sprintf(buf, "(len %d) %x:%x:%x:%x:%x:%x:%x:%x/%d",
1996 		    addrlen,
1997 		    ntohs(piece[0]), ntohs(piece[1]),
1998 		    ntohs(piece[2]), ntohs(piece[3]),
1999 		    ntohs(piece[4]), ntohs(piece[5]),
2000 		    ntohs(piece[6]), ntohs(piece[7]),
2001 		    ntohs(sin6.sin6_port));
2002 		break;
2003 	}
2004 	case AF_UNIX: {
2005 		struct sockaddr_un *soun = (struct sockaddr_un *)addr;
2006 
2007 		(void) sprintf(buf, "(len %d) %s",
2008 			addrlen,
2009 			(soun == NULL) ? "(none)" : soun->sun_path);
2010 		break;
2011 	}
2012 	default:
2013 		(void) sprintf(buf, "(unknown af %d)", family);
2014 		break;
2015 	}
2016 	return (buf);
2017 }
2018 
2019 /* The logical equivalence operator (a if-and-only-if b) */
2020 #define	EQUIV(a, b)	(((a) && (b)) || (!(a) && (!(b))))
2021 
2022 /*
2023  * Verify limitations and invariants on oob state.
2024  * Return 1 if OK, otherwise 0 so that it can be used as
2025  *	ASSERT(verify_oobstate(so));
2026  */
2027 int
2028 so_verify_oobstate(struct sonode *so)
2029 {
2030 	ASSERT(MUTEX_HELD(&so->so_lock));
2031 
2032 	/*
2033 	 * The possible state combinations are:
2034 	 *	0
2035 	 *	SS_OOBPEND
2036 	 *	SS_OOBPEND|SS_HAVEOOBDATA
2037 	 *	SS_OOBPEND|SS_HADOOBDATA
2038 	 *	SS_HADOOBDATA
2039 	 */
2040 	switch (so->so_state & (SS_OOBPEND|SS_HAVEOOBDATA|SS_HADOOBDATA)) {
2041 	case 0:
2042 	case SS_OOBPEND:
2043 	case SS_OOBPEND|SS_HAVEOOBDATA:
2044 	case SS_OOBPEND|SS_HADOOBDATA:
2045 	case SS_HADOOBDATA:
2046 		break;
2047 	default:
2048 		printf("Bad oob state 1 (%p): counts %d/%d state %s\n",
2049 			so, so->so_oobsigcnt,
2050 			so->so_oobcnt, pr_state(so->so_state, so->so_mode));
2051 		return (0);
2052 	}
2053 
2054 	/* SS_RCVATMARK should only be set when SS_OOBPEND is set */
2055 	if ((so->so_state & (SS_RCVATMARK|SS_OOBPEND)) == SS_RCVATMARK) {
2056 		printf("Bad oob state 2 (%p): counts %d/%d state %s\n",
2057 			so, so->so_oobsigcnt,
2058 			so->so_oobcnt, pr_state(so->so_state, so->so_mode));
2059 		return (0);
2060 	}
2061 
2062 	/*
2063 	 * (so_oobsigcnt != 0 or SS_RCVATMARK) iff SS_OOBPEND
2064 	 */
2065 	if (!EQUIV((so->so_oobsigcnt != 0) || (so->so_state & SS_RCVATMARK),
2066 		so->so_state & SS_OOBPEND)) {
2067 		printf("Bad oob state 3 (%p): counts %d/%d state %s\n",
2068 			so, so->so_oobsigcnt,
2069 			so->so_oobcnt, pr_state(so->so_state, so->so_mode));
2070 		return (0);
2071 	}
2072 
2073 	/*
2074 	 * Unless SO_OOBINLINE we have so_oobmsg != NULL iff SS_HAVEOOBDATA
2075 	 */
2076 	if (!(so->so_options & SO_OOBINLINE) &&
2077 	    !EQUIV(so->so_oobmsg != NULL, so->so_state & SS_HAVEOOBDATA)) {
2078 		printf("Bad oob state 4 (%p): counts %d/%d state %s\n",
2079 			so, so->so_oobsigcnt,
2080 			so->so_oobcnt, pr_state(so->so_state, so->so_mode));
2081 		return (0);
2082 	}
2083 	if (so->so_oobsigcnt < so->so_oobcnt) {
2084 		printf("Bad oob state 5 (%p): counts %d/%d state %s\n",
2085 			so, so->so_oobsigcnt,
2086 			so->so_oobcnt, pr_state(so->so_state, so->so_mode));
2087 		return (0);
2088 	}
2089 	return (1);
2090 }
2091 #undef	EQUIV
2092 
2093 #endif /* DEBUG */
2094 
2095 /* initialize sockfs zone specific kstat related items			*/
2096 void *
2097 sock_kstat_init(zoneid_t zoneid)
2098 {
2099 	kstat_t	*ksp;
2100 
2101 	ksp = kstat_create_zone("sockfs", 0, "sock_unix_list", "misc",
2102 	    KSTAT_TYPE_RAW, 0, KSTAT_FLAG_VAR_SIZE|KSTAT_FLAG_VIRTUAL, zoneid);
2103 
2104 	if (ksp != NULL) {
2105 		ksp->ks_update = sockfs_update;
2106 		ksp->ks_snapshot = sockfs_snapshot;
2107 		ksp->ks_lock = &socklist.sl_lock;
2108 		ksp->ks_private = (void *)(uintptr_t)zoneid;
2109 		kstat_install(ksp);
2110 	}
2111 
2112 	return (ksp);
2113 }
2114 
2115 /* tear down sockfs zone specific kstat related items			*/
2116 /*ARGSUSED*/
2117 void
2118 sock_kstat_fini(zoneid_t zoneid, void *arg)
2119 {
2120 	kstat_t *ksp = (kstat_t *)arg;
2121 
2122 	if (ksp != NULL) {
2123 		ASSERT(zoneid == (zoneid_t)(uintptr_t)ksp->ks_private);
2124 		kstat_delete(ksp);
2125 	}
2126 }
2127 
2128 /*
2129  * Zones:
2130  * Note that nactive is going to be different for each zone.
2131  * This means we require kstat to call sockfs_update and then sockfs_snapshot
2132  * for the same zone, or sockfs_snapshot will be taken into the wrong size
2133  * buffer. This is safe, but if the buffer is too small, user will not be
2134  * given details of all sockets. However, as this kstat has a ks_lock, kstat
2135  * driver will keep it locked between the update and the snapshot, so no
2136  * other process (zone) can currently get inbetween resulting in a wrong size
2137  * buffer allocation.
2138  */
2139 static int
2140 sockfs_update(kstat_t *ksp, int rw)
2141 {
2142 	uint_t	nactive = 0;		/* # of active AF_UNIX sockets	*/
2143 	struct sonode	*so;		/* current sonode on socklist	*/
2144 	zoneid_t	myzoneid = (zoneid_t)(uintptr_t)ksp->ks_private;
2145 
2146 	ASSERT((zoneid_t)(uintptr_t)ksp->ks_private == getzoneid());
2147 
2148 	if (rw == KSTAT_WRITE) {	/* bounce all writes		*/
2149 		return (EACCES);
2150 	}
2151 
2152 	for (so = socklist.sl_list; so != NULL; so = so->so_next) {
2153 		if (so->so_accessvp != NULL && so->so_zoneid == myzoneid) {
2154 			nactive++;
2155 		}
2156 	}
2157 	ksp->ks_ndata = nactive;
2158 	ksp->ks_data_size = nactive * sizeof (struct k_sockinfo);
2159 
2160 	return (0);
2161 }
2162 
2163 static int
2164 sockfs_snapshot(kstat_t *ksp, void *buf, int rw)
2165 {
2166 	int			ns;	/* # of sonodes we've copied	*/
2167 	struct sonode		*so;	/* current sonode on socklist	*/
2168 	struct k_sockinfo	*pksi;	/* where we put sockinfo data	*/
2169 	t_uscalar_t		sn_len;	/* soa_len			*/
2170 	zoneid_t		myzoneid = (zoneid_t)(uintptr_t)ksp->ks_private;
2171 
2172 	ASSERT((zoneid_t)(uintptr_t)ksp->ks_private == getzoneid());
2173 
2174 	ksp->ks_snaptime = gethrtime();
2175 
2176 	if (rw == KSTAT_WRITE) {	/* bounce all writes		*/
2177 		return (EACCES);
2178 	}
2179 
2180 	/*
2181 	 * for each sonode on the socklist, we massage the important
2182 	 * info into buf, in k_sockinfo format.
2183 	 */
2184 	pksi = (struct k_sockinfo *)buf;
2185 	for (ns = 0, so = socklist.sl_list; so != NULL; so = so->so_next) {
2186 		/* only stuff active sonodes and the same zone:		*/
2187 		if (so->so_accessvp == NULL || so->so_zoneid != myzoneid) {
2188 			continue;
2189 		}
2190 
2191 		/*
2192 		 * If the sonode was activated between the update and the
2193 		 * snapshot, we're done - as this is only a snapshot.
2194 		 */
2195 		if ((caddr_t)(pksi) >= (caddr_t)buf + ksp->ks_data_size) {
2196 			break;
2197 		}
2198 
2199 		/* copy important info into buf:			*/
2200 		pksi->ks_si.si_size = sizeof (struct k_sockinfo);
2201 		pksi->ks_si.si_family = so->so_family;
2202 		pksi->ks_si.si_type = so->so_type;
2203 		pksi->ks_si.si_flag = so->so_flag;
2204 		pksi->ks_si.si_state = so->so_state;
2205 		pksi->ks_si.si_serv_type = so->so_serv_type;
2206 		pksi->ks_si.si_ux_laddr_sou_magic = so->so_ux_laddr.soua_magic;
2207 		pksi->ks_si.si_ux_faddr_sou_magic = so->so_ux_faddr.soua_magic;
2208 		pksi->ks_si.si_laddr_soa_len = so->so_laddr.soa_len;
2209 		pksi->ks_si.si_faddr_soa_len = so->so_faddr.soa_len;
2210 		pksi->ks_si.si_szoneid = so->so_zoneid;
2211 
2212 		mutex_enter(&so->so_lock);
2213 
2214 		if (so->so_laddr_sa != NULL) {
2215 			ASSERT(so->so_laddr_sa->sa_data != NULL);
2216 			sn_len = so->so_laddr_len;
2217 			ASSERT(sn_len <= sizeof (short) +
2218 			    sizeof (pksi->ks_si.si_laddr_sun_path));
2219 
2220 			pksi->ks_si.si_laddr_family =
2221 				so->so_laddr_sa->sa_family;
2222 			if (sn_len != 0) {
2223 				/* AF_UNIX socket names are NULL terminated */
2224 				(void) strncpy(pksi->ks_si.si_laddr_sun_path,
2225 				    so->so_laddr_sa->sa_data,
2226 				    sizeof (pksi->ks_si.si_laddr_sun_path));
2227 				sn_len = strlen(pksi->ks_si.si_laddr_sun_path);
2228 			}
2229 			pksi->ks_si.si_laddr_sun_path[sn_len] = 0;
2230 		}
2231 
2232 		if (so->so_faddr_sa != NULL) {
2233 			ASSERT(so->so_faddr_sa->sa_data != NULL);
2234 			sn_len = so->so_faddr_len;
2235 			ASSERT(sn_len <= sizeof (short) +
2236 			    sizeof (pksi->ks_si.si_faddr_sun_path));
2237 
2238 			pksi->ks_si.si_faddr_family =
2239 			    so->so_faddr_sa->sa_family;
2240 			if (sn_len != 0) {
2241 				(void) strncpy(pksi->ks_si.si_faddr_sun_path,
2242 				    so->so_faddr_sa->sa_data,
2243 				    sizeof (pksi->ks_si.si_faddr_sun_path));
2244 				sn_len = strlen(pksi->ks_si.si_faddr_sun_path);
2245 			}
2246 			pksi->ks_si.si_faddr_sun_path[sn_len] = 0;
2247 		}
2248 
2249 		mutex_exit(&so->so_lock);
2250 
2251 		(void) sprintf(pksi->ks_straddr[0], "%p", (void *)so);
2252 		(void) sprintf(pksi->ks_straddr[1], "%p",
2253 		    (void *)so->so_ux_laddr.soua_vp);
2254 		(void) sprintf(pksi->ks_straddr[2], "%p",
2255 		    (void *)so->so_ux_faddr.soua_vp);
2256 
2257 		ns++;
2258 		pksi++;
2259 	}
2260 
2261 	ksp->ks_ndata = ns;
2262 	return (0);
2263 }
2264 
2265 ssize_t
2266 soreadfile(file_t *fp, uchar_t *buf, u_offset_t fileoff, int *err, size_t size)
2267 {
2268 	struct uio auio;
2269 	struct iovec aiov[MSG_MAXIOVLEN];
2270 	register vnode_t *vp;
2271 	int ioflag, rwflag;
2272 	ssize_t cnt;
2273 	int error = 0;
2274 	int iovcnt = 0;
2275 	short fflag;
2276 
2277 	vp = fp->f_vnode;
2278 	fflag = fp->f_flag;
2279 
2280 	rwflag = 0;
2281 	aiov[0].iov_base = (caddr_t)buf;
2282 	aiov[0].iov_len = size;
2283 	iovcnt = 1;
2284 	cnt = (ssize_t)size;
2285 	(void) VOP_RWLOCK(vp, rwflag, NULL);
2286 
2287 	auio.uio_loffset = fileoff;
2288 	auio.uio_iov = aiov;
2289 	auio.uio_iovcnt = iovcnt;
2290 	auio.uio_resid = cnt;
2291 	auio.uio_segflg = UIO_SYSSPACE;
2292 	auio.uio_llimit = MAXOFFSET_T;
2293 	auio.uio_fmode = fflag;
2294 	auio.uio_extflg = UIO_COPY_CACHED;
2295 
2296 	ioflag = auio.uio_fmode & (FAPPEND|FSYNC|FDSYNC|FRSYNC);
2297 
2298 	/* If read sync is not asked for, filter sync flags */
2299 	if ((ioflag & FRSYNC) == 0)
2300 		ioflag &= ~(FSYNC|FDSYNC);
2301 	error = VOP_READ(vp, &auio, ioflag, fp->f_cred, NULL);
2302 	cnt -= auio.uio_resid;
2303 
2304 	VOP_RWUNLOCK(vp, rwflag, NULL);
2305 
2306 	if (error == EINTR && cnt != 0)
2307 		error = 0;
2308 out:
2309 	if (error != 0) {
2310 		*err = error;
2311 		return (0);
2312 	} else {
2313 		*err = 0;
2314 		return (cnt);
2315 	}
2316 }
2317