xref: /titanic_41/usr/src/uts/common/fs/sockfs/socksubr.c (revision 749f21d359d8fbd020c974a1a5227316221bfc9c)
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, Version 1.0 only
6  * (the "License").  You may not use this file except in compliance
7  * with the License.
8  *
9  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
10  * or http://www.opensolaris.org/os/licensing.
11  * See the License for the specific language governing permissions
12  * and limitations under the License.
13  *
14  * When distributing Covered Code, include this CDDL HEADER in each
15  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
16  * If applicable, add the following below this CDDL HEADER, with the
17  * fields enclosed by brackets "[]" replaced with your own identifying
18  * information: Portions Copyright [yyyy] [name of copyright owner]
19  *
20  * CDDL HEADER END
21  */
22 /*
23  * Copyright 2005 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 		cmn_err(CE_WARN, "sockinit: bad vfs ops template");
688 		return (error);
689 	}
690 
691 	error = vn_make_ops(name, socktpi_vnodeops_template, &socktpi_vnodeops);
692 	if (error != 0) {
693 		err_str = "sockinit: bad sock vnode ops template";
694 		/* vn_make_ops() does not reset socktpi_vnodeops on failure. */
695 		socktpi_vnodeops = NULL;
696 		goto failure;
697 	}
698 
699 	error = vn_make_ops("nca", socknca_vnodeops_template,
700 	    &socknca_vnodeops);
701 	if (error != 0) {
702 		err_str = "sockinit: bad nca vnode ops template";
703 		socknca_vnodeops = NULL;
704 		goto failure;
705 	}
706 
707 	error = sosctp_init();
708 	if (error != 0) {
709 		err_str = NULL;
710 		goto failure;
711 	}
712 
713 	/*
714 	 * Create sonode caches.  We create a special one for AF_UNIX so
715 	 * that we can track them for netstat(1m).
716 	 */
717 	socktpi_cache = kmem_cache_create("socktpi_cache",
718 	    sizeof (struct sonode), 0, socktpi_constructor,
719 	    socktpi_destructor, NULL, NULL, NULL, 0);
720 
721 	socktpi_unix_cache = kmem_cache_create("socktpi_unix_cache",
722 	    sizeof (struct sonode), 0, socktpi_unix_constructor,
723 	    socktpi_unix_destructor, NULL, NULL, NULL, 0);
724 
725 	/*
726 	 * Build initial list mapping socket parameters to vnode.
727 	 */
728 	rw_init(&splist_lock, NULL, RW_DEFAULT, NULL);
729 
730 	/*
731 	 * If sockets are needed before init runs /sbin/soconfig
732 	 * it is possible to preload the sockparams list here using
733 	 * calls like:
734 	 *	sockconfig(1,2,3, "/dev/tcp", 0);
735 	 */
736 
737 	/*
738 	 * Create a unique dev_t for use in so_fsid.
739 	 */
740 
741 	if ((dev = getudev()) == (major_t)-1)
742 		dev = 0;
743 	sockdev = makedevice(dev, 0);
744 
745 	mutex_init(&socklist.sl_lock, NULL, MUTEX_DEFAULT, NULL);
746 	sonca_init();
747 	sendfile_init();
748 	nl7c_init();
749 
750 	return (0);
751 
752 failure:
753 	(void) vfs_freevfsops_by_type(fstype);
754 	if (socktpi_vnodeops != NULL)
755 		vn_freevnodeops(socktpi_vnodeops);
756 	if (socknca_vnodeops != NULL)
757 		vn_freevnodeops(socknca_vnodeops);
758 	if (err_str != NULL)
759 		cmn_err(CE_WARN, err_str);
760 	return (error);
761 }
762 
763 /*
764  * Caller must hold the mutex. Used to set SOLOCKED.
765  */
766 void
767 so_lock_single(struct sonode *so)
768 {
769 	ASSERT(MUTEX_HELD(&so->so_lock));
770 
771 	while (so->so_flag & (SOLOCKED | SOASYNC_UNBIND)) {
772 		so->so_flag |= SOWANT;
773 		cv_wait_stop(&so->so_want_cv, &so->so_lock,
774 			SO_LOCK_WAKEUP_TIME);
775 	}
776 	so->so_flag |= SOLOCKED;
777 }
778 
779 /*
780  * Caller must hold the mutex and pass in SOLOCKED or SOASYNC_UNBIND.
781  * Used to clear SOLOCKED or SOASYNC_UNBIND.
782  */
783 void
784 so_unlock_single(struct sonode *so, int flag)
785 {
786 	ASSERT(MUTEX_HELD(&so->so_lock));
787 	ASSERT(flag & (SOLOCKED|SOASYNC_UNBIND));
788 	ASSERT((flag & ~(SOLOCKED|SOASYNC_UNBIND)) == 0);
789 	ASSERT(so->so_flag & flag);
790 
791 	/*
792 	 * Process the T_DISCON_IND on so_discon_ind_mp.
793 	 *
794 	 * Call to so_drain_discon_ind will result in so_lock
795 	 * being dropped and re-acquired later.
796 	 */
797 	if (so->so_discon_ind_mp != NULL)
798 		so_drain_discon_ind(so);
799 
800 	if (so->so_flag & SOWANT)
801 		cv_broadcast(&so->so_want_cv);
802 	so->so_flag &= ~(SOWANT|flag);
803 }
804 
805 /*
806  * Caller must hold the mutex. Used to set SOREADLOCKED.
807  * If the caller wants nonblocking behavior it should set fmode.
808  */
809 int
810 so_lock_read(struct sonode *so, int fmode)
811 {
812 	ASSERT(MUTEX_HELD(&so->so_lock));
813 
814 	while (so->so_flag & SOREADLOCKED) {
815 		if (fmode & (FNDELAY|FNONBLOCK))
816 			return (EWOULDBLOCK);
817 		so->so_flag |= SOWANT;
818 		cv_wait_stop(&so->so_want_cv, &so->so_lock,
819 			SO_LOCK_WAKEUP_TIME);
820 	}
821 	so->so_flag |= SOREADLOCKED;
822 	return (0);
823 }
824 
825 /*
826  * Like so_lock_read above but allows signals.
827  */
828 int
829 so_lock_read_intr(struct sonode *so, int fmode)
830 {
831 	ASSERT(MUTEX_HELD(&so->so_lock));
832 
833 	while (so->so_flag & SOREADLOCKED) {
834 		if (fmode & (FNDELAY|FNONBLOCK))
835 			return (EWOULDBLOCK);
836 		so->so_flag |= SOWANT;
837 		if (!cv_wait_sig(&so->so_want_cv, &so->so_lock))
838 			return (EINTR);
839 	}
840 	so->so_flag |= SOREADLOCKED;
841 	return (0);
842 }
843 
844 /*
845  * Caller must hold the mutex. Used to clear SOREADLOCKED,
846  * set in so_lock_read() or so_lock_read_intr().
847  */
848 void
849 so_unlock_read(struct sonode *so)
850 {
851 	ASSERT(MUTEX_HELD(&so->so_lock));
852 	ASSERT(so->so_flag & SOREADLOCKED);
853 
854 	if (so->so_flag & SOWANT)
855 		cv_broadcast(&so->so_want_cv);
856 	so->so_flag &= ~(SOWANT|SOREADLOCKED);
857 }
858 
859 /*
860  * Verify that the specified offset falls within the mblk and
861  * that the resulting pointer is aligned.
862  * Returns NULL if not.
863  */
864 void *
865 sogetoff(mblk_t *mp, t_uscalar_t offset,
866     t_uscalar_t length, uint_t align_size)
867 {
868 	uintptr_t ptr1, ptr2;
869 
870 	ASSERT(mp && mp->b_wptr >= mp->b_rptr);
871 	ptr1 = (uintptr_t)mp->b_rptr + offset;
872 	ptr2 = (uintptr_t)ptr1 + length;
873 	if (ptr1 < (uintptr_t)mp->b_rptr || ptr2 > (uintptr_t)mp->b_wptr) {
874 		eprintline(0);
875 		return (NULL);
876 	}
877 	if ((ptr1 & (align_size - 1)) != 0) {
878 		eprintline(0);
879 		return (NULL);
880 	}
881 	return ((void *)ptr1);
882 }
883 
884 /*
885  * Return the AF_UNIX underlying filesystem vnode matching a given name.
886  * Makes sure the sending and the destination sonodes are compatible.
887  * The vnode is returned held.
888  *
889  * The underlying filesystem VSOCK vnode has a v_stream pointer that
890  * references the actual stream head (hence indirectly the actual sonode).
891  */
892 static int
893 so_ux_lookup(struct sonode *so, struct sockaddr_un *soun, int checkaccess,
894 		vnode_t **vpp)
895 {
896 	vnode_t		*vp;	/* Underlying filesystem vnode */
897 	vnode_t		*svp;	/* sockfs vnode */
898 	struct sonode	*so2;
899 	int		error;
900 
901 	dprintso(so, 1, ("so_ux_lookup(%p) name <%s>\n",
902 		so, soun->sun_path));
903 
904 	error = lookupname(soun->sun_path, UIO_SYSSPACE, FOLLOW, NULLVPP, &vp);
905 	if (error) {
906 		eprintsoline(so, error);
907 		return (error);
908 	}
909 	if (vp->v_type != VSOCK) {
910 		error = ENOTSOCK;
911 		eprintsoline(so, error);
912 		goto done2;
913 	}
914 
915 	if (checkaccess) {
916 		/*
917 		 * Check that we have permissions to access the destination
918 		 * vnode. This check is not done in BSD but it is required
919 		 * by X/Open.
920 		 */
921 		if (error = VOP_ACCESS(vp, VREAD|VWRITE, 0, CRED())) {
922 			eprintsoline(so, error);
923 			goto done2;
924 		}
925 	}
926 
927 	/*
928 	 * Check if the remote socket has been closed.
929 	 *
930 	 * Synchronize with vn_rele_stream by holding v_lock while traversing
931 	 * v_stream->sd_vnode.
932 	 */
933 	mutex_enter(&vp->v_lock);
934 	if (vp->v_stream == NULL) {
935 		mutex_exit(&vp->v_lock);
936 		if (so->so_type == SOCK_DGRAM)
937 			error = EDESTADDRREQ;
938 		else
939 			error = ECONNREFUSED;
940 
941 		eprintsoline(so, error);
942 		goto done2;
943 	}
944 	ASSERT(vp->v_stream->sd_vnode);
945 	svp = vp->v_stream->sd_vnode;
946 	/*
947 	 * holding v_lock on underlying filesystem vnode and acquiring
948 	 * it on sockfs vnode. Assumes that no code ever attempts to
949 	 * acquire these locks in the reverse order.
950 	 */
951 	VN_HOLD(svp);
952 	mutex_exit(&vp->v_lock);
953 
954 	if (svp->v_type != VSOCK) {
955 		error = ENOTSOCK;
956 		eprintsoline(so, error);
957 		goto done;
958 	}
959 
960 	so2 = VTOSO(svp);
961 
962 	if (so->so_type != so2->so_type) {
963 		error = EPROTOTYPE;
964 		eprintsoline(so, error);
965 		goto done;
966 	}
967 
968 	VN_RELE(svp);
969 	*vpp = vp;
970 	return (0);
971 
972 done:
973 	VN_RELE(svp);
974 done2:
975 	VN_RELE(vp);
976 	return (error);
977 }
978 
979 /*
980  * Verify peer address for connect and sendto/sendmsg.
981  * Since sendto/sendmsg would not get synchronous errors from the transport
982  * provider we have to do these ugly checks in the socket layer to
983  * preserve compatibility with SunOS 4.X.
984  */
985 int
986 so_addr_verify(struct sonode *so, const struct sockaddr *name,
987     socklen_t namelen)
988 {
989 	int		family;
990 
991 	dprintso(so, 1, ("so_addr_verify(%p, %p, %d)\n", so, name, namelen));
992 
993 	ASSERT(name != NULL);
994 
995 	family = so->so_family;
996 	switch (family) {
997 	case AF_INET:
998 		if (name->sa_family != family) {
999 			eprintsoline(so, EAFNOSUPPORT);
1000 			return (EAFNOSUPPORT);
1001 		}
1002 		if (namelen != (socklen_t)sizeof (struct sockaddr_in)) {
1003 			eprintsoline(so, EINVAL);
1004 			return (EINVAL);
1005 		}
1006 		break;
1007 	case AF_INET6: {
1008 #ifdef DEBUG
1009 		struct sockaddr_in6 *sin6;
1010 #endif /* DEBUG */
1011 
1012 		if (name->sa_family != family) {
1013 			eprintsoline(so, EAFNOSUPPORT);
1014 			return (EAFNOSUPPORT);
1015 		}
1016 		if (namelen != (socklen_t)sizeof (struct sockaddr_in6)) {
1017 			eprintsoline(so, EINVAL);
1018 			return (EINVAL);
1019 		}
1020 #ifdef DEBUG
1021 		/* Verify that apps don't forget to clear sin6_scope_id etc */
1022 		sin6 = (struct sockaddr_in6 *)name;
1023 		if (sin6->sin6_scope_id != 0 &&
1024 		    !IN6_IS_ADDR_LINKSCOPE(&sin6->sin6_addr)) {
1025 			cmn_err(CE_WARN,
1026 			    "connect/send* with uninitialized sin6_scope_id "
1027 			    "(%d) on socket. Pid = %d\n",
1028 			    (int)sin6->sin6_scope_id, (int)curproc->p_pid);
1029 		}
1030 #endif /* DEBUG */
1031 		break;
1032 	}
1033 	case AF_UNIX:
1034 		if (so->so_state & SS_FADDR_NOXLATE) {
1035 			return (0);
1036 		}
1037 		if (namelen < (socklen_t)sizeof (short)) {
1038 			eprintsoline(so, ENOENT);
1039 			return (ENOENT);
1040 		}
1041 		if (name->sa_family != family) {
1042 			eprintsoline(so, EAFNOSUPPORT);
1043 			return (EAFNOSUPPORT);
1044 		}
1045 		/* MAXPATHLEN + soun_family + nul termination */
1046 		if (namelen > (socklen_t)(MAXPATHLEN + sizeof (short) + 1)) {
1047 			eprintsoline(so, ENAMETOOLONG);
1048 			return (ENAMETOOLONG);
1049 		}
1050 
1051 		break;
1052 
1053 	default:
1054 		/*
1055 		 * Default is don't do any length or sa_family check
1056 		 * to allow non-sockaddr style addresses.
1057 		 */
1058 		break;
1059 	}
1060 
1061 	return (0);
1062 }
1063 
1064 
1065 /*
1066  * Translate an AF_UNIX sockaddr_un to the transport internal name.
1067  * Assumes caller has called so_addr_verify first.
1068  */
1069 /*ARGSUSED*/
1070 int
1071 so_ux_addr_xlate(struct sonode *so, struct sockaddr *name,
1072     socklen_t namelen, int checkaccess,
1073     void **addrp, socklen_t *addrlenp)
1074 {
1075 	int			error;
1076 	struct sockaddr_un	*soun;
1077 	vnode_t			*vp;
1078 	void			*addr;
1079 	socklen_t		addrlen;
1080 
1081 	dprintso(so, 1, ("so_ux_addr_xlate(%p, %p, %d, %d)\n",
1082 			so, name, namelen, checkaccess));
1083 
1084 	ASSERT(name != NULL);
1085 	ASSERT(so->so_family == AF_UNIX);
1086 	ASSERT(!(so->so_state & SS_FADDR_NOXLATE));
1087 	ASSERT(namelen >= (socklen_t)sizeof (short));
1088 	ASSERT(name->sa_family == AF_UNIX);
1089 	soun = (struct sockaddr_un *)name;
1090 	/*
1091 	 * Lookup vnode for the specified path name and verify that
1092 	 * it is a socket.
1093 	 */
1094 	error = so_ux_lookup(so, soun, checkaccess, &vp);
1095 	if (error) {
1096 		eprintsoline(so, error);
1097 		return (error);
1098 	}
1099 	/*
1100 	 * Use the address of the peer vnode as the address to send
1101 	 * to. We release the peer vnode here. In case it has been
1102 	 * closed by the time the T_CONN_REQ or T_UNIDATA_REQ reaches the
1103 	 * transport the message will get an error or be dropped.
1104 	 */
1105 	so->so_ux_faddr.soua_vp = vp;
1106 	so->so_ux_faddr.soua_magic = SOU_MAGIC_EXPLICIT;
1107 	addr = &so->so_ux_faddr;
1108 	addrlen = (socklen_t)sizeof (so->so_ux_faddr);
1109 	dprintso(so, 1, ("ux_xlate UNIX: addrlen %d, vp %p\n",
1110 				addrlen, vp));
1111 	VN_RELE(vp);
1112 	*addrp = addr;
1113 	*addrlenp = (socklen_t)addrlen;
1114 	return (0);
1115 }
1116 
1117 /*
1118  * Esballoc free function for messages that contain SO_FILEP option.
1119  * Decrement the reference count on the file pointers using closef.
1120  */
1121 void
1122 fdbuf_free(struct fdbuf *fdbuf)
1123 {
1124 	int	i;
1125 	struct file *fp;
1126 
1127 	dprint(1, ("fdbuf_free: %d fds\n", fdbuf->fd_numfd));
1128 	for (i = 0; i < fdbuf->fd_numfd; i++) {
1129 		/*
1130 		 * We need pointer size alignment for fd_fds. On a LP64
1131 		 * kernel, the required alignment is 8 bytes while
1132 		 * the option headers and values are only 4 bytes
1133 		 * aligned. So its safer to do a bcopy compared to
1134 		 * assigning fdbuf->fd_fds[i] to fp.
1135 		 */
1136 		bcopy((char *)&fdbuf->fd_fds[i], (char *)&fp, sizeof (fp));
1137 		dprint(1, ("fdbuf_free: [%d] = %p\n", i, fp));
1138 		(void) closef(fp);
1139 	}
1140 	if (fdbuf->fd_ebuf != NULL)
1141 		kmem_free(fdbuf->fd_ebuf, fdbuf->fd_ebuflen);
1142 	kmem_free(fdbuf, fdbuf->fd_size);
1143 }
1144 
1145 /*
1146  * Allocate an esballoc'ed message for AF_UNIX file descriptor passing.
1147  * Waits if memory is not available.
1148  */
1149 mblk_t *
1150 fdbuf_allocmsg(int size, struct fdbuf *fdbuf)
1151 {
1152 	uchar_t	*buf;
1153 	mblk_t	*mp;
1154 
1155 	dprint(1, ("fdbuf_allocmsg: size %d, %d fds\n", size, fdbuf->fd_numfd));
1156 	buf = kmem_alloc(size, KM_SLEEP);
1157 	fdbuf->fd_ebuf = (caddr_t)buf;
1158 	fdbuf->fd_ebuflen = size;
1159 	fdbuf->fd_frtn.free_func = fdbuf_free;
1160 	fdbuf->fd_frtn.free_arg = (caddr_t)fdbuf;
1161 
1162 	mp = esballoc_wait(buf, size, BPRI_MED, &fdbuf->fd_frtn);
1163 	mp->b_datap->db_type = M_PROTO;
1164 	return (mp);
1165 }
1166 
1167 /*
1168  * Extract file descriptors from a fdbuf.
1169  * Return list in rights/rightslen.
1170  */
1171 /*ARGSUSED*/
1172 static int
1173 fdbuf_extract(struct fdbuf *fdbuf, void *rights, int rightslen)
1174 {
1175 	int	i, fd;
1176 	int	*rp;
1177 	struct file *fp;
1178 	int	numfd;
1179 
1180 	dprint(1, ("fdbuf_extract: %d fds, len %d\n",
1181 		fdbuf->fd_numfd, rightslen));
1182 
1183 	numfd = fdbuf->fd_numfd;
1184 	ASSERT(rightslen == numfd * (int)sizeof (int));
1185 
1186 	/*
1187 	 * Allocate a file descriptor and increment the f_count.
1188 	 * The latter is needed since we always call fdbuf_free
1189 	 * which performs a closef.
1190 	 */
1191 	rp = (int *)rights;
1192 	for (i = 0; i < numfd; i++) {
1193 		if ((fd = ufalloc(0)) == -1)
1194 			goto cleanup;
1195 		/*
1196 		 * We need pointer size alignment for fd_fds. On a LP64
1197 		 * kernel, the required alignment is 8 bytes while
1198 		 * the option headers and values are only 4 bytes
1199 		 * aligned. So its safer to do a bcopy compared to
1200 		 * assigning fdbuf->fd_fds[i] to fp.
1201 		 */
1202 		bcopy((char *)&fdbuf->fd_fds[i], (char *)&fp, sizeof (fp));
1203 		mutex_enter(&fp->f_tlock);
1204 		fp->f_count++;
1205 		mutex_exit(&fp->f_tlock);
1206 		setf(fd, fp);
1207 		*rp++ = fd;
1208 #ifdef C2_AUDIT
1209 		if (audit_active)
1210 			audit_fdrecv(fd, fp);
1211 #endif
1212 		dprint(1, ("fdbuf_extract: [%d] = %d, %p refcnt %d\n",
1213 			i, fd, fp, fp->f_count));
1214 	}
1215 	return (0);
1216 
1217 cleanup:
1218 	/*
1219 	 * Undo whatever partial work the loop above has done.
1220 	 */
1221 	{
1222 		int j;
1223 
1224 		rp = (int *)rights;
1225 		for (j = 0; j < i; j++) {
1226 			dprint(0,
1227 			    ("fdbuf_extract: cleanup[%d] = %d\n", j, *rp));
1228 			(void) closeandsetf(*rp++, NULL);
1229 		}
1230 	}
1231 
1232 	return (EMFILE);
1233 }
1234 
1235 /*
1236  * Insert file descriptors into an fdbuf.
1237  * Returns a kmem_alloc'ed fdbuf. The fdbuf should be freed
1238  * by calling fdbuf_free().
1239  */
1240 int
1241 fdbuf_create(void *rights, int rightslen, struct fdbuf **fdbufp)
1242 {
1243 	int		numfd, i;
1244 	int		*fds;
1245 	struct file	*fp;
1246 	struct fdbuf	*fdbuf;
1247 	int		fdbufsize;
1248 
1249 	dprint(1, ("fdbuf_create: len %d\n", rightslen));
1250 
1251 	numfd = rightslen / (int)sizeof (int);
1252 
1253 	fdbufsize = (int)FDBUF_HDRSIZE + (numfd * (int)sizeof (struct file *));
1254 	fdbuf = kmem_alloc(fdbufsize, KM_SLEEP);
1255 	fdbuf->fd_size = fdbufsize;
1256 	fdbuf->fd_numfd = 0;
1257 	fdbuf->fd_ebuf = NULL;
1258 	fdbuf->fd_ebuflen = 0;
1259 	fds = (int *)rights;
1260 	for (i = 0; i < numfd; i++) {
1261 		if ((fp = getf(fds[i])) == NULL) {
1262 			fdbuf_free(fdbuf);
1263 			return (EBADF);
1264 		}
1265 		dprint(1, ("fdbuf_create: [%d] = %d, %p refcnt %d\n",
1266 			i, fds[i], fp, fp->f_count));
1267 		mutex_enter(&fp->f_tlock);
1268 		fp->f_count++;
1269 		mutex_exit(&fp->f_tlock);
1270 		/*
1271 		 * The maximum alignment for fdbuf (or any option header
1272 		 * and its value) it 4 bytes. On a LP64 kernel, the alignment
1273 		 * is not sufficient for pointers (fd_fds in this case). Since
1274 		 * we just did a kmem_alloc (we get a double word alignment),
1275 		 * we don't need to do anything on the send side (we loose
1276 		 * the double word alignment because fdbuf goes after an
1277 		 * option header (eg T_unitdata_req) which is only 4 byte
1278 		 * aligned). We take care of this when we extract the file
1279 		 * descriptor in fdbuf_extract or fdbuf_free.
1280 		 */
1281 		fdbuf->fd_fds[i] = fp;
1282 		fdbuf->fd_numfd++;
1283 		releasef(fds[i]);
1284 #ifdef C2_AUDIT
1285 		if (audit_active)
1286 			audit_fdsend(fds[i], fp, 0);
1287 #endif
1288 	}
1289 	*fdbufp = fdbuf;
1290 	return (0);
1291 }
1292 
1293 static int
1294 fdbuf_optlen(int rightslen)
1295 {
1296 	int numfd;
1297 
1298 	numfd = rightslen / (int)sizeof (int);
1299 
1300 	return ((int)FDBUF_HDRSIZE + (numfd * (int)sizeof (struct file *)));
1301 }
1302 
1303 static t_uscalar_t
1304 fdbuf_cmsglen(int fdbuflen)
1305 {
1306 	return (t_uscalar_t)((fdbuflen - FDBUF_HDRSIZE) /
1307 	    (int)sizeof (struct file *) * (int)sizeof (int));
1308 }
1309 
1310 
1311 /*
1312  * Return non-zero if the mblk and fdbuf are consistent.
1313  */
1314 static int
1315 fdbuf_verify(mblk_t *mp, struct fdbuf *fdbuf, int fdbuflen)
1316 {
1317 	if (fdbuflen >= FDBUF_HDRSIZE &&
1318 	    fdbuflen == fdbuf->fd_size) {
1319 		frtn_t *frp = mp->b_datap->db_frtnp;
1320 		/*
1321 		 * Check that the SO_FILEP portion of the
1322 		 * message has not been modified by
1323 		 * the loopback transport. The sending sockfs generates
1324 		 * a message that is esballoc'ed with the free function
1325 		 * being fdbuf_free() and where free_arg contains the
1326 		 * identical information as the SO_FILEP content.
1327 		 *
1328 		 * If any of these constraints are not satisfied we
1329 		 * silently ignore the option.
1330 		 */
1331 		ASSERT(mp);
1332 		if (frp != NULL &&
1333 		    frp->free_func == fdbuf_free &&
1334 		    frp->free_arg != NULL &&
1335 		    bcmp(frp->free_arg, fdbuf, fdbuflen) == 0) {
1336 			dprint(1, ("fdbuf_verify: fdbuf %p len %d\n",
1337 				fdbuf, fdbuflen));
1338 			return (1);
1339 		} else {
1340 			cmn_err(CE_WARN,
1341 			    "sockfs: mismatched fdbuf content (%p)",
1342 			    (void *)mp);
1343 			return (0);
1344 		}
1345 	} else {
1346 		cmn_err(CE_WARN,
1347 		    "sockfs: mismatched fdbuf len %d, %d\n",
1348 		    fdbuflen, fdbuf->fd_size);
1349 		return (0);
1350 	}
1351 }
1352 
1353 /*
1354  * When the file descriptors returned by sorecvmsg can not be passed
1355  * to the application this routine will cleanup the references on
1356  * the files. Start at startoff bytes into the buffer.
1357  */
1358 static void
1359 close_fds(void *fdbuf, int fdbuflen, int startoff)
1360 {
1361 	int *fds = (int *)fdbuf;
1362 	int numfd = fdbuflen / (int)sizeof (int);
1363 	int i;
1364 
1365 	dprint(1, ("close_fds(%p, %d, %d)\n", fdbuf, fdbuflen, startoff));
1366 
1367 	for (i = 0; i < numfd; i++) {
1368 		if (startoff < 0)
1369 			startoff = 0;
1370 		if (startoff < (int)sizeof (int)) {
1371 			/*
1372 			 * This file descriptor is partially or fully after
1373 			 * the offset
1374 			 */
1375 			dprint(0,
1376 			    ("close_fds: cleanup[%d] = %d\n", i, fds[i]));
1377 			(void) closeandsetf(fds[i], NULL);
1378 		}
1379 		startoff -= (int)sizeof (int);
1380 	}
1381 }
1382 
1383 /*
1384  * Close all file descriptors contained in the control part starting at
1385  * the startoffset.
1386  */
1387 void
1388 so_closefds(void *control, t_uscalar_t controllen, int oldflg,
1389     int startoff)
1390 {
1391 	struct cmsghdr *cmsg;
1392 
1393 	if (control == NULL)
1394 		return;
1395 
1396 	if (oldflg) {
1397 		close_fds(control, controllen, startoff);
1398 		return;
1399 	}
1400 	/* Scan control part for file descriptors. */
1401 	for (cmsg = (struct cmsghdr *)control;
1402 	    CMSG_VALID(cmsg, control, (uintptr_t)control + controllen);
1403 	    cmsg = CMSG_NEXT(cmsg)) {
1404 		if (cmsg->cmsg_level == SOL_SOCKET &&
1405 		    cmsg->cmsg_type == SCM_RIGHTS) {
1406 			close_fds(CMSG_CONTENT(cmsg),
1407 			    (int)CMSG_CONTENTLEN(cmsg),
1408 			    startoff - (int)sizeof (struct cmsghdr));
1409 		}
1410 		startoff -= cmsg->cmsg_len;
1411 	}
1412 }
1413 
1414 /*
1415  * Returns a pointer/length for the file descriptors contained
1416  * in the control buffer. Returns with *fdlenp == -1 if there are no
1417  * file descriptor options present. This is different than there being
1418  * a zero-length file descriptor option.
1419  * Fail if there are multiple SCM_RIGHT cmsgs.
1420  */
1421 int
1422 so_getfdopt(void *control, t_uscalar_t controllen, int oldflg,
1423     void **fdsp, int *fdlenp)
1424 {
1425 	struct cmsghdr *cmsg;
1426 	void *fds;
1427 	int fdlen;
1428 
1429 	if (control == NULL) {
1430 		*fdsp = NULL;
1431 		*fdlenp = -1;
1432 		return (0);
1433 	}
1434 
1435 	if (oldflg) {
1436 		*fdsp = control;
1437 		if (controllen == 0)
1438 			*fdlenp = -1;
1439 		else
1440 			*fdlenp = controllen;
1441 		dprint(1, ("so_getfdopt: old %d\n", *fdlenp));
1442 		return (0);
1443 	}
1444 
1445 	fds = NULL;
1446 	fdlen = 0;
1447 
1448 	for (cmsg = (struct cmsghdr *)control;
1449 	    CMSG_VALID(cmsg, control, (uintptr_t)control + controllen);
1450 	    cmsg = CMSG_NEXT(cmsg)) {
1451 		if (cmsg->cmsg_level == SOL_SOCKET &&
1452 		    cmsg->cmsg_type == SCM_RIGHTS) {
1453 			if (fds != NULL)
1454 				return (EINVAL);
1455 			fds = CMSG_CONTENT(cmsg);
1456 			fdlen = (int)CMSG_CONTENTLEN(cmsg);
1457 			dprint(1, ("so_getfdopt: new %lu\n",
1458 				(size_t)CMSG_CONTENTLEN(cmsg)));
1459 		}
1460 	}
1461 	if (fds == NULL) {
1462 		dprint(1, ("so_getfdopt: NONE\n"));
1463 		*fdlenp = -1;
1464 	} else
1465 		*fdlenp = fdlen;
1466 	*fdsp = fds;
1467 	return (0);
1468 }
1469 
1470 /*
1471  * Return the length of the options including any file descriptor options.
1472  */
1473 t_uscalar_t
1474 so_optlen(void *control, t_uscalar_t controllen, int oldflg)
1475 {
1476 	struct cmsghdr *cmsg;
1477 	t_uscalar_t optlen = 0;
1478 	t_uscalar_t len;
1479 
1480 	if (control == NULL)
1481 		return (0);
1482 
1483 	if (oldflg)
1484 		return ((t_uscalar_t)(sizeof (struct T_opthdr) +
1485 		    fdbuf_optlen(controllen)));
1486 
1487 	for (cmsg = (struct cmsghdr *)control;
1488 	    CMSG_VALID(cmsg, control, (uintptr_t)control + controllen);
1489 	    cmsg = CMSG_NEXT(cmsg)) {
1490 		if (cmsg->cmsg_level == SOL_SOCKET &&
1491 		    cmsg->cmsg_type == SCM_RIGHTS) {
1492 			len = fdbuf_optlen((int)CMSG_CONTENTLEN(cmsg));
1493 		} else {
1494 			len = (t_uscalar_t)CMSG_CONTENTLEN(cmsg);
1495 		}
1496 		optlen += (t_uscalar_t)(_TPI_ALIGN_TOPT(len) +
1497 		    sizeof (struct T_opthdr));
1498 	}
1499 	dprint(1, ("so_optlen: controllen %d, flg %d -> optlen %d\n",
1500 		controllen, oldflg, optlen));
1501 	return (optlen);
1502 }
1503 
1504 /*
1505  * Copy options from control to the mblk. Skip any file descriptor options.
1506  */
1507 void
1508 so_cmsg2opt(void *control, t_uscalar_t controllen, int oldflg, mblk_t *mp)
1509 {
1510 	struct T_opthdr toh;
1511 	struct cmsghdr *cmsg;
1512 
1513 	if (control == NULL)
1514 		return;
1515 
1516 	if (oldflg) {
1517 		/* No real options - caller has handled file descriptors */
1518 		return;
1519 	}
1520 	for (cmsg = (struct cmsghdr *)control;
1521 	    CMSG_VALID(cmsg, control, (uintptr_t)control + controllen);
1522 	    cmsg = CMSG_NEXT(cmsg)) {
1523 		/*
1524 		 * Note: The caller handles file descriptors prior
1525 		 * to calling this function.
1526 		 */
1527 		t_uscalar_t len;
1528 
1529 		if (cmsg->cmsg_level == SOL_SOCKET &&
1530 		    cmsg->cmsg_type == SCM_RIGHTS)
1531 			continue;
1532 
1533 		len = (t_uscalar_t)CMSG_CONTENTLEN(cmsg);
1534 		toh.level = cmsg->cmsg_level;
1535 		toh.name = cmsg->cmsg_type;
1536 		toh.len = len + (t_uscalar_t)sizeof (struct T_opthdr);
1537 		toh.status = 0;
1538 
1539 		soappendmsg(mp, &toh, sizeof (toh));
1540 		soappendmsg(mp, CMSG_CONTENT(cmsg), len);
1541 		mp->b_wptr += _TPI_ALIGN_TOPT(len) - len;
1542 		ASSERT(mp->b_wptr <= mp->b_datap->db_lim);
1543 	}
1544 }
1545 
1546 /*
1547  * Return the length of the control message derived from the options.
1548  * Exclude SO_SRCADDR and SO_UNIX_CLOSE options. Include SO_FILEP.
1549  * When oldflg is set only include SO_FILEP.
1550  */
1551 t_uscalar_t
1552 so_cmsglen(mblk_t *mp, void *opt, t_uscalar_t optlen, int oldflg)
1553 {
1554 	t_uscalar_t cmsglen = 0;
1555 	struct T_opthdr *tohp;
1556 	t_uscalar_t len;
1557 	t_uscalar_t last_roundup = 0;
1558 
1559 	ASSERT(__TPI_TOPT_ISALIGNED(opt));
1560 
1561 	for (tohp = (struct T_opthdr *)opt;
1562 	    tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen);
1563 	    tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) {
1564 		dprint(1, ("so_cmsglen: level 0x%x, name %d, len %d\n",
1565 			tohp->level, tohp->name, tohp->len));
1566 		if (tohp->level == SOL_SOCKET &&
1567 		    (tohp->name == SO_SRCADDR ||
1568 		    tohp->name == SO_UNIX_CLOSE)) {
1569 			continue;
1570 		}
1571 		if (tohp->level == SOL_SOCKET && tohp->name == SO_FILEP) {
1572 			struct fdbuf *fdbuf;
1573 			int fdbuflen;
1574 
1575 			fdbuf = (struct fdbuf *)_TPI_TOPT_DATA(tohp);
1576 			fdbuflen = (int)_TPI_TOPT_DATALEN(tohp);
1577 
1578 			if (!fdbuf_verify(mp, fdbuf, fdbuflen))
1579 				continue;
1580 			if (oldflg) {
1581 				cmsglen += fdbuf_cmsglen(fdbuflen);
1582 				continue;
1583 			}
1584 			len = fdbuf_cmsglen(fdbuflen);
1585 		} else {
1586 			if (oldflg)
1587 				continue;
1588 			len = (t_uscalar_t)_TPI_TOPT_DATALEN(tohp);
1589 		}
1590 		/*
1591 		 * Exlucde roundup for last option to not set
1592 		 * MSG_CTRUNC when the cmsg fits but the padding doesn't fit.
1593 		 */
1594 		last_roundup = (t_uscalar_t)
1595 		    (ROUNDUP_cmsglen(len + (int)sizeof (struct cmsghdr)) -
1596 		    (len + (int)sizeof (struct cmsghdr)));
1597 		cmsglen += (t_uscalar_t)(len + (int)sizeof (struct cmsghdr)) +
1598 		    last_roundup;
1599 	}
1600 	cmsglen -= last_roundup;
1601 	dprint(1, ("so_cmsglen: optlen %d, flg %d -> cmsglen %d\n",
1602 		optlen, oldflg, cmsglen));
1603 	return (cmsglen);
1604 }
1605 
1606 /*
1607  * Copy options from options to the control. Convert SO_FILEP to
1608  * file descriptors.
1609  * Returns errno or zero.
1610  */
1611 int
1612 so_opt2cmsg(mblk_t *mp, void *opt, t_uscalar_t optlen, int oldflg,
1613     void *control, t_uscalar_t controllen)
1614 {
1615 	struct T_opthdr *tohp;
1616 	struct cmsghdr *cmsg;
1617 	struct fdbuf *fdbuf;
1618 	int fdbuflen;
1619 	int error;
1620 
1621 	cmsg = (struct cmsghdr *)control;
1622 
1623 	ASSERT(__TPI_TOPT_ISALIGNED(opt));
1624 
1625 	for (tohp = (struct T_opthdr *)opt;
1626 	    tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen);
1627 	    tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) {
1628 		dprint(1, ("so_opt2cmsg: level 0x%x, name %d, len %d\n",
1629 			tohp->level, tohp->name, tohp->len));
1630 
1631 		if (tohp->level == SOL_SOCKET &&
1632 		    (tohp->name == SO_SRCADDR ||
1633 		    tohp->name == SO_UNIX_CLOSE)) {
1634 			continue;
1635 		}
1636 		ASSERT((uintptr_t)cmsg <= (uintptr_t)control + controllen);
1637 		if (tohp->level == SOL_SOCKET && tohp->name == SO_FILEP) {
1638 			fdbuf = (struct fdbuf *)_TPI_TOPT_DATA(tohp);
1639 			fdbuflen = (int)_TPI_TOPT_DATALEN(tohp);
1640 
1641 			if (!fdbuf_verify(mp, fdbuf, fdbuflen))
1642 				return (EPROTO);
1643 			if (oldflg) {
1644 				error = fdbuf_extract(fdbuf, control,
1645 				    (int)controllen);
1646 				if (error != 0)
1647 					return (error);
1648 				continue;
1649 			} else {
1650 				int fdlen;
1651 
1652 				fdlen = (int)fdbuf_cmsglen(
1653 				    (int)_TPI_TOPT_DATALEN(tohp));
1654 
1655 				cmsg->cmsg_level = tohp->level;
1656 				cmsg->cmsg_type = SCM_RIGHTS;
1657 				cmsg->cmsg_len = (socklen_t)(fdlen +
1658 					sizeof (struct cmsghdr));
1659 
1660 				error = fdbuf_extract(fdbuf,
1661 						CMSG_CONTENT(cmsg), fdlen);
1662 				if (error != 0)
1663 					return (error);
1664 			}
1665 		} else {
1666 			if (oldflg)
1667 				continue;
1668 
1669 			cmsg->cmsg_level = tohp->level;
1670 			cmsg->cmsg_type = tohp->name;
1671 			cmsg->cmsg_len = (socklen_t)(_TPI_TOPT_DATALEN(tohp) +
1672 			    sizeof (struct cmsghdr));
1673 
1674 			/* copy content to control data part */
1675 			bcopy(&tohp[1], CMSG_CONTENT(cmsg),
1676 				CMSG_CONTENTLEN(cmsg));
1677 		}
1678 		/* move to next CMSG structure! */
1679 		cmsg = CMSG_NEXT(cmsg);
1680 	}
1681 	return (0);
1682 }
1683 
1684 /*
1685  * Extract the SO_SRCADDR option value if present.
1686  */
1687 void
1688 so_getopt_srcaddr(void *opt, t_uscalar_t optlen, void **srcp,
1689     t_uscalar_t *srclenp)
1690 {
1691 	struct T_opthdr		*tohp;
1692 
1693 	ASSERT(__TPI_TOPT_ISALIGNED(opt));
1694 
1695 	ASSERT(srcp != NULL && srclenp != NULL);
1696 	*srcp = NULL;
1697 	*srclenp = 0;
1698 
1699 	for (tohp = (struct T_opthdr *)opt;
1700 	    tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen);
1701 	    tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) {
1702 		dprint(1, ("so_getopt_srcaddr: level 0x%x, name %d, len %d\n",
1703 			tohp->level, tohp->name, tohp->len));
1704 		if (tohp->level == SOL_SOCKET &&
1705 		    tohp->name == SO_SRCADDR) {
1706 			*srcp = _TPI_TOPT_DATA(tohp);
1707 			*srclenp = (t_uscalar_t)_TPI_TOPT_DATALEN(tohp);
1708 		}
1709 	}
1710 }
1711 
1712 /*
1713  * Verify if the SO_UNIX_CLOSE option is present.
1714  */
1715 int
1716 so_getopt_unix_close(void *opt, t_uscalar_t optlen)
1717 {
1718 	struct T_opthdr		*tohp;
1719 
1720 	ASSERT(__TPI_TOPT_ISALIGNED(opt));
1721 
1722 	for (tohp = (struct T_opthdr *)opt;
1723 	    tohp && _TPI_TOPT_VALID(tohp, opt, (uintptr_t)opt + optlen);
1724 	    tohp = _TPI_TOPT_NEXTHDR(opt, optlen, tohp)) {
1725 		dprint(1,
1726 			("so_getopt_unix_close: level 0x%x, name %d, len %d\n",
1727 			tohp->level, tohp->name, tohp->len));
1728 		if (tohp->level == SOL_SOCKET &&
1729 		    tohp->name == SO_UNIX_CLOSE)
1730 			return (1);
1731 	}
1732 	return (0);
1733 }
1734 
1735 /*
1736  * Allocate an M_PROTO message.
1737  *
1738  * If allocation fails the behavior depends on sleepflg:
1739  *	_ALLOC_NOSLEEP	fail immediately
1740  *	_ALLOC_INTR	sleep for memory until a signal is caught
1741  *	_ALLOC_SLEEP	sleep forever. Don't return NULL.
1742  */
1743 mblk_t *
1744 soallocproto(size_t size, int sleepflg)
1745 {
1746 	mblk_t	*mp;
1747 
1748 	/* Round up size for reuse */
1749 	size = MAX(size, 64);
1750 	mp = allocb(size, BPRI_MED);
1751 	if (mp == NULL) {
1752 		int error;	/* Dummy - error not returned to caller */
1753 
1754 		switch (sleepflg) {
1755 		case _ALLOC_SLEEP:
1756 			mp = allocb_wait(size, BPRI_MED, STR_NOSIG, &error);
1757 			ASSERT(mp);
1758 			break;
1759 		case _ALLOC_INTR:
1760 			mp = allocb_wait(size, BPRI_MED, 0, &error);
1761 			if (mp == NULL) {
1762 				/* Caught signal while sleeping for memory */
1763 				eprintline(ENOBUFS);
1764 				return (NULL);
1765 			}
1766 			break;
1767 		case _ALLOC_NOSLEEP:
1768 		default:
1769 			eprintline(ENOBUFS);
1770 			return (NULL);
1771 		}
1772 	}
1773 	DB_TYPE(mp) = M_PROTO;
1774 	return (mp);
1775 }
1776 
1777 /*
1778  * Allocate an M_PROTO message with a single component.
1779  * len is the length of buf. size is the amount to allocate.
1780  *
1781  * buf can be NULL with a non-zero len.
1782  * This results in a bzero'ed chunk being placed the message.
1783  */
1784 mblk_t *
1785 soallocproto1(const void *buf, ssize_t len, ssize_t size, int sleepflg)
1786 {
1787 	mblk_t	*mp;
1788 
1789 	if (size == 0)
1790 		size = len;
1791 
1792 	ASSERT(size >= len);
1793 	/* Round up size for reuse */
1794 	size = MAX(size, 64);
1795 	mp = soallocproto(size, sleepflg);
1796 	if (mp == NULL)
1797 		return (NULL);
1798 	mp->b_datap->db_type = M_PROTO;
1799 	if (len != 0) {
1800 		if (buf != NULL)
1801 			bcopy(buf, mp->b_wptr, len);
1802 		else
1803 			bzero(mp->b_wptr, len);
1804 		mp->b_wptr += len;
1805 	}
1806 	return (mp);
1807 }
1808 
1809 /*
1810  * Append buf/len to mp.
1811  * The caller has to ensure that there is enough room in the mblk.
1812  *
1813  * buf can be NULL with a non-zero len.
1814  * This results in a bzero'ed chunk being placed the message.
1815  */
1816 void
1817 soappendmsg(mblk_t *mp, const void *buf, ssize_t len)
1818 {
1819 	ASSERT(mp);
1820 
1821 	if (len != 0) {
1822 		/* Assert for room left */
1823 		ASSERT(mp->b_datap->db_lim - mp->b_wptr >= len);
1824 		if (buf != NULL)
1825 			bcopy(buf, mp->b_wptr, len);
1826 		else
1827 			bzero(mp->b_wptr, len);
1828 	}
1829 	mp->b_wptr += len;
1830 }
1831 
1832 /*
1833  * Create a message using two kernel buffers.
1834  * If size is set that will determine the allocation size (e.g. for future
1835  * soappendmsg calls). If size is zero it is derived from the buffer
1836  * lengths.
1837  */
1838 mblk_t *
1839 soallocproto2(const void *buf1, ssize_t len1, const void *buf2, ssize_t len2,
1840     ssize_t size, int sleepflg)
1841 {
1842 	mblk_t *mp;
1843 
1844 	if (size == 0)
1845 		size = len1 + len2;
1846 	ASSERT(size >= len1 + len2);
1847 
1848 	mp = soallocproto1(buf1, len1, size, sleepflg);
1849 	if (mp)
1850 		soappendmsg(mp, buf2, len2);
1851 	return (mp);
1852 }
1853 
1854 /*
1855  * Create a message using three kernel buffers.
1856  * If size is set that will determine the allocation size (for future
1857  * soappendmsg calls). If size is zero it is derived from the buffer
1858  * lengths.
1859  */
1860 mblk_t *
1861 soallocproto3(const void *buf1, ssize_t len1, const void *buf2, ssize_t len2,
1862     const void *buf3, ssize_t len3, ssize_t size, int sleepflg)
1863 {
1864 	mblk_t *mp;
1865 
1866 	if (size == 0)
1867 		size = len1 + len2 +len3;
1868 	ASSERT(size >= len1 + len2 + len3);
1869 
1870 	mp = soallocproto1(buf1, len1, size, sleepflg);
1871 	if (mp != NULL) {
1872 		soappendmsg(mp, buf2, len2);
1873 		soappendmsg(mp, buf3, len3);
1874 	}
1875 	return (mp);
1876 }
1877 
1878 #ifdef DEBUG
1879 char *
1880 pr_state(uint_t state, uint_t mode)
1881 {
1882 	static char buf[1024];
1883 
1884 	buf[0] = 0;
1885 	if (state & SS_ISCONNECTED)
1886 		strcat(buf, "ISCONNECTED ");
1887 	if (state & SS_ISCONNECTING)
1888 		strcat(buf, "ISCONNECTING ");
1889 	if (state & SS_ISDISCONNECTING)
1890 		strcat(buf, "ISDISCONNECTING ");
1891 	if (state & SS_CANTSENDMORE)
1892 		strcat(buf, "CANTSENDMORE ");
1893 
1894 	if (state & SS_CANTRCVMORE)
1895 		strcat(buf, "CANTRCVMORE ");
1896 	if (state & SS_ISBOUND)
1897 		strcat(buf, "ISBOUND ");
1898 	if (state & SS_NDELAY)
1899 		strcat(buf, "NDELAY ");
1900 	if (state & SS_NONBLOCK)
1901 		strcat(buf, "NONBLOCK ");
1902 
1903 	if (state & SS_ASYNC)
1904 		strcat(buf, "ASYNC ");
1905 	if (state & SS_ACCEPTCONN)
1906 		strcat(buf, "ACCEPTCONN ");
1907 	if (state & SS_HASCONNIND)
1908 		strcat(buf, "HASCONNIND ");
1909 	if (state & SS_SAVEDEOR)
1910 		strcat(buf, "SAVEDEOR ");
1911 
1912 	if (state & SS_RCVATMARK)
1913 		strcat(buf, "RCVATMARK ");
1914 	if (state & SS_OOBPEND)
1915 		strcat(buf, "OOBPEND ");
1916 	if (state & SS_HAVEOOBDATA)
1917 		strcat(buf, "HAVEOOBDATA ");
1918 	if (state & SS_HADOOBDATA)
1919 		strcat(buf, "HADOOBDATA ");
1920 
1921 	if (state & SS_FADDR_NOXLATE)
1922 		strcat(buf, "FADDR_NOXLATE ");
1923 
1924 	if (mode & SM_PRIV)
1925 		strcat(buf, "PRIV ");
1926 	if (mode & SM_ATOMIC)
1927 		strcat(buf, "ATOMIC ");
1928 	if (mode & SM_ADDR)
1929 		strcat(buf, "ADDR ");
1930 	if (mode & SM_CONNREQUIRED)
1931 		strcat(buf, "CONNREQUIRED ");
1932 
1933 	if (mode & SM_FDPASSING)
1934 		strcat(buf, "FDPASSING ");
1935 	if (mode & SM_EXDATA)
1936 		strcat(buf, "EXDATA ");
1937 	if (mode & SM_OPTDATA)
1938 		strcat(buf, "OPTDATA ");
1939 	if (mode & SM_BYTESTREAM)
1940 		strcat(buf, "BYTESTREAM ");
1941 	return (buf);
1942 }
1943 
1944 char *
1945 pr_addr(int family, struct sockaddr *addr, t_uscalar_t addrlen)
1946 {
1947 	static char buf[1024];
1948 
1949 	if (addr == NULL || addrlen == 0) {
1950 		sprintf(buf, "(len %d) %p", addrlen, addr);
1951 		return (buf);
1952 	}
1953 	switch (family) {
1954 	case AF_INET: {
1955 		struct sockaddr_in sin;
1956 
1957 		bcopy(addr, &sin, sizeof (sin));
1958 
1959 		(void) sprintf(buf, "(len %d) %x/%d",
1960 			addrlen, ntohl(sin.sin_addr.s_addr),
1961 			ntohs(sin.sin_port));
1962 		break;
1963 	}
1964 	case AF_INET6: {
1965 		struct sockaddr_in6 sin6;
1966 		uint16_t *piece = (uint16_t *)&sin6.sin6_addr;
1967 
1968 		bcopy((char *)addr, (char *)&sin6, sizeof (sin6));
1969 		sprintf(buf, "(len %d) %x:%x:%x:%x:%x:%x:%x:%x/%d",
1970 		    addrlen,
1971 		    ntohs(piece[0]), ntohs(piece[1]),
1972 		    ntohs(piece[2]), ntohs(piece[3]),
1973 		    ntohs(piece[4]), ntohs(piece[5]),
1974 		    ntohs(piece[6]), ntohs(piece[7]),
1975 		    ntohs(sin6.sin6_port));
1976 		break;
1977 	}
1978 	case AF_UNIX: {
1979 		struct sockaddr_un *soun = (struct sockaddr_un *)addr;
1980 
1981 		(void) sprintf(buf, "(len %d) %s",
1982 			addrlen,
1983 			(soun == NULL) ? "(none)" : soun->sun_path);
1984 		break;
1985 	}
1986 	default:
1987 		(void) sprintf(buf, "(unknown af %d)", family);
1988 		break;
1989 	}
1990 	return (buf);
1991 }
1992 
1993 /* The logical equivalence operator (a if-and-only-if b) */
1994 #define	EQUIV(a, b)	(((a) && (b)) || (!(a) && (!(b))))
1995 
1996 /*
1997  * Verify limitations and invariants on oob state.
1998  * Return 1 if OK, otherwise 0 so that it can be used as
1999  *	ASSERT(verify_oobstate(so));
2000  */
2001 int
2002 so_verify_oobstate(struct sonode *so)
2003 {
2004 	ASSERT(MUTEX_HELD(&so->so_lock));
2005 
2006 	/*
2007 	 * The possible state combinations are:
2008 	 *	0
2009 	 *	SS_OOBPEND
2010 	 *	SS_OOBPEND|SS_HAVEOOBDATA
2011 	 *	SS_OOBPEND|SS_HADOOBDATA
2012 	 *	SS_HADOOBDATA
2013 	 */
2014 	switch (so->so_state & (SS_OOBPEND|SS_HAVEOOBDATA|SS_HADOOBDATA)) {
2015 	case 0:
2016 	case SS_OOBPEND:
2017 	case SS_OOBPEND|SS_HAVEOOBDATA:
2018 	case SS_OOBPEND|SS_HADOOBDATA:
2019 	case SS_HADOOBDATA:
2020 		break;
2021 	default:
2022 		printf("Bad oob state 1 (%p): counts %d/%d state %s\n",
2023 			so, so->so_oobsigcnt,
2024 			so->so_oobcnt, pr_state(so->so_state, so->so_mode));
2025 		return (0);
2026 	}
2027 
2028 	/* SS_RCVATMARK should only be set when SS_OOBPEND is set */
2029 	if ((so->so_state & (SS_RCVATMARK|SS_OOBPEND)) == SS_RCVATMARK) {
2030 		printf("Bad oob state 2 (%p): counts %d/%d state %s\n",
2031 			so, so->so_oobsigcnt,
2032 			so->so_oobcnt, pr_state(so->so_state, so->so_mode));
2033 		return (0);
2034 	}
2035 
2036 	/*
2037 	 * (so_oobsigcnt != 0 or SS_RCVATMARK) iff SS_OOBPEND
2038 	 */
2039 	if (!EQUIV((so->so_oobsigcnt != 0) || (so->so_state & SS_RCVATMARK),
2040 		so->so_state & SS_OOBPEND)) {
2041 		printf("Bad oob state 3 (%p): counts %d/%d state %s\n",
2042 			so, so->so_oobsigcnt,
2043 			so->so_oobcnt, pr_state(so->so_state, so->so_mode));
2044 		return (0);
2045 	}
2046 
2047 	/*
2048 	 * Unless SO_OOBINLINE we have so_oobmsg != NULL iff SS_HAVEOOBDATA
2049 	 */
2050 	if (!(so->so_options & SO_OOBINLINE) &&
2051 	    !EQUIV(so->so_oobmsg != NULL, so->so_state & SS_HAVEOOBDATA)) {
2052 		printf("Bad oob state 4 (%p): counts %d/%d state %s\n",
2053 			so, so->so_oobsigcnt,
2054 			so->so_oobcnt, pr_state(so->so_state, so->so_mode));
2055 		return (0);
2056 	}
2057 	if (so->so_oobsigcnt < so->so_oobcnt) {
2058 		printf("Bad oob state 5 (%p): counts %d/%d state %s\n",
2059 			so, so->so_oobsigcnt,
2060 			so->so_oobcnt, pr_state(so->so_state, so->so_mode));
2061 		return (0);
2062 	}
2063 	return (1);
2064 }
2065 #undef	EQUIV
2066 
2067 #endif /* DEBUG */
2068 
2069 /* initialize sockfs zone specific kstat related items			*/
2070 void *
2071 sock_kstat_init(zoneid_t zoneid)
2072 {
2073 	kstat_t	*ksp;
2074 
2075 	ksp = kstat_create_zone("sockfs", 0, "sock_unix_list", "misc",
2076 	    KSTAT_TYPE_RAW, 0, KSTAT_FLAG_VAR_SIZE|KSTAT_FLAG_VIRTUAL, zoneid);
2077 
2078 	if (ksp != NULL) {
2079 		ksp->ks_update = sockfs_update;
2080 		ksp->ks_snapshot = sockfs_snapshot;
2081 		ksp->ks_lock = &socklist.sl_lock;
2082 		ksp->ks_private = (void *)(uintptr_t)zoneid;
2083 		kstat_install(ksp);
2084 	}
2085 
2086 	return (ksp);
2087 }
2088 
2089 /* tear down sockfs zone specific kstat related items			*/
2090 /*ARGSUSED*/
2091 void
2092 sock_kstat_fini(zoneid_t zoneid, void *arg)
2093 {
2094 	kstat_t *ksp = (kstat_t *)arg;
2095 
2096 	if (ksp != NULL) {
2097 		ASSERT(zoneid == (zoneid_t)(uintptr_t)ksp->ks_private);
2098 		kstat_delete(ksp);
2099 	}
2100 }
2101 
2102 /*
2103  * Zones:
2104  * Note that nactive is going to be different for each zone.
2105  * This means we require kstat to call sockfs_update and then sockfs_snapshot
2106  * for the same zone, or sockfs_snapshot will be taken into the wrong size
2107  * buffer. This is safe, but if the buffer is too small, user will not be
2108  * given details of all sockets. However, as this kstat has a ks_lock, kstat
2109  * driver will keep it locked between the update and the snapshot, so no
2110  * other process (zone) can currently get inbetween resulting in a wrong size
2111  * buffer allocation.
2112  */
2113 static int
2114 sockfs_update(kstat_t *ksp, int rw)
2115 {
2116 	uint_t	nactive = 0;		/* # of active AF_UNIX sockets	*/
2117 	struct sonode	*so;		/* current sonode on socklist	*/
2118 	zoneid_t	myzoneid = (zoneid_t)(uintptr_t)ksp->ks_private;
2119 
2120 	ASSERT((zoneid_t)(uintptr_t)ksp->ks_private == getzoneid());
2121 
2122 	if (rw == KSTAT_WRITE) {	/* bounce all writes		*/
2123 		return (EACCES);
2124 	}
2125 
2126 	for (so = socklist.sl_list; so != NULL; so = so->so_next) {
2127 		if (so->so_accessvp != NULL && so->so_zoneid == myzoneid) {
2128 			nactive++;
2129 		}
2130 	}
2131 	ksp->ks_ndata = nactive;
2132 	ksp->ks_data_size = nactive * sizeof (struct k_sockinfo);
2133 
2134 	return (0);
2135 }
2136 
2137 static int
2138 sockfs_snapshot(kstat_t *ksp, void *buf, int rw)
2139 {
2140 	int			ns;	/* # of sonodes we've copied	*/
2141 	struct sonode		*so;	/* current sonode on socklist	*/
2142 	struct k_sockinfo	*pksi;	/* where we put sockinfo data	*/
2143 	t_uscalar_t		sn_len;	/* soa_len			*/
2144 	zoneid_t		myzoneid = (zoneid_t)(uintptr_t)ksp->ks_private;
2145 
2146 	ASSERT((zoneid_t)(uintptr_t)ksp->ks_private == getzoneid());
2147 
2148 	ksp->ks_snaptime = gethrtime();
2149 
2150 	if (rw == KSTAT_WRITE) {	/* bounce all writes		*/
2151 		return (EACCES);
2152 	}
2153 
2154 	/*
2155 	 * for each sonode on the socklist, we massage the important
2156 	 * info into buf, in k_sockinfo format.
2157 	 */
2158 	pksi = (struct k_sockinfo *)buf;
2159 	for (ns = 0, so = socklist.sl_list; so != NULL; so = so->so_next) {
2160 		/* only stuff active sonodes and the same zone:		*/
2161 		if (so->so_accessvp == NULL || so->so_zoneid != myzoneid) {
2162 			continue;
2163 		}
2164 
2165 		/*
2166 		 * If the sonode was activated between the update and the
2167 		 * snapshot, we're done - as this is only a snapshot.
2168 		 */
2169 		if ((caddr_t)(pksi) >= (caddr_t)buf + ksp->ks_data_size) {
2170 			break;
2171 		}
2172 
2173 		/* copy important info into buf:			*/
2174 		pksi->ks_si.si_size = sizeof (struct k_sockinfo);
2175 		pksi->ks_si.si_family = so->so_family;
2176 		pksi->ks_si.si_type = so->so_type;
2177 		pksi->ks_si.si_flag = so->so_flag;
2178 		pksi->ks_si.si_state = so->so_state;
2179 		pksi->ks_si.si_serv_type = so->so_serv_type;
2180 		pksi->ks_si.si_ux_laddr_sou_magic = so->so_ux_laddr.soua_magic;
2181 		pksi->ks_si.si_ux_faddr_sou_magic = so->so_ux_faddr.soua_magic;
2182 		pksi->ks_si.si_laddr_soa_len = so->so_laddr.soa_len;
2183 		pksi->ks_si.si_faddr_soa_len = so->so_faddr.soa_len;
2184 		pksi->ks_si.si_szoneid = so->so_zoneid;
2185 
2186 		mutex_enter(&so->so_lock);
2187 
2188 		if (so->so_laddr_sa != NULL) {
2189 			ASSERT(so->so_laddr_sa->sa_data != NULL);
2190 			sn_len = so->so_laddr_len;
2191 			ASSERT(sn_len <= sizeof (short) +
2192 			    sizeof (pksi->ks_si.si_laddr_sun_path));
2193 
2194 			pksi->ks_si.si_laddr_family =
2195 				so->so_laddr_sa->sa_family;
2196 			if (sn_len != 0) {
2197 				/* AF_UNIX socket names are NULL terminated */
2198 				(void) strncpy(pksi->ks_si.si_laddr_sun_path,
2199 				    so->so_laddr_sa->sa_data,
2200 				    sizeof (pksi->ks_si.si_laddr_sun_path));
2201 				sn_len = strlen(pksi->ks_si.si_laddr_sun_path);
2202 			}
2203 			pksi->ks_si.si_laddr_sun_path[sn_len] = 0;
2204 		}
2205 
2206 		if (so->so_faddr_sa != NULL) {
2207 			ASSERT(so->so_faddr_sa->sa_data != NULL);
2208 			sn_len = so->so_faddr_len;
2209 			ASSERT(sn_len <= sizeof (short) +
2210 			    sizeof (pksi->ks_si.si_faddr_sun_path));
2211 
2212 			pksi->ks_si.si_faddr_family =
2213 			    so->so_faddr_sa->sa_family;
2214 			if (sn_len != 0) {
2215 				(void) strncpy(pksi->ks_si.si_faddr_sun_path,
2216 				    so->so_faddr_sa->sa_data,
2217 				    sizeof (pksi->ks_si.si_faddr_sun_path));
2218 				sn_len = strlen(pksi->ks_si.si_faddr_sun_path);
2219 			}
2220 			pksi->ks_si.si_faddr_sun_path[sn_len] = 0;
2221 		}
2222 
2223 		mutex_exit(&so->so_lock);
2224 
2225 		(void) sprintf(pksi->ks_straddr[0], "%p", (void *)so);
2226 		(void) sprintf(pksi->ks_straddr[1], "%p",
2227 		    (void *)so->so_ux_laddr.soua_vp);
2228 		(void) sprintf(pksi->ks_straddr[2], "%p",
2229 		    (void *)so->so_ux_faddr.soua_vp);
2230 
2231 		ns++;
2232 		pksi++;
2233 	}
2234 
2235 	ksp->ks_ndata = ns;
2236 	return (0);
2237 }
2238 
2239 ssize_t
2240 soreadfile(file_t *fp, uchar_t *buf, u_offset_t fileoff, int *err, size_t size)
2241 {
2242 	struct uio auio;
2243 	struct iovec aiov[MSG_MAXIOVLEN];
2244 	register vnode_t *vp;
2245 	int ioflag, rwflag;
2246 	ssize_t cnt;
2247 	int error = 0;
2248 	int iovcnt = 0;
2249 	short fflag;
2250 
2251 	vp = fp->f_vnode;
2252 	fflag = fp->f_flag;
2253 
2254 	rwflag = 0;
2255 	aiov[0].iov_base = (caddr_t)buf;
2256 	aiov[0].iov_len = size;
2257 	iovcnt = 1;
2258 	cnt = (ssize_t)size;
2259 	(void) VOP_RWLOCK(vp, rwflag, NULL);
2260 
2261 	auio.uio_loffset = fileoff;
2262 	auio.uio_iov = aiov;
2263 	auio.uio_iovcnt = iovcnt;
2264 	auio.uio_resid = cnt;
2265 	auio.uio_segflg = UIO_SYSSPACE;
2266 	auio.uio_llimit = MAXOFFSET_T;
2267 	auio.uio_fmode = fflag;
2268 	auio.uio_extflg = UIO_COPY_CACHED;
2269 
2270 	ioflag = auio.uio_fmode & (FAPPEND|FSYNC|FDSYNC|FRSYNC);
2271 
2272 	/* If read sync is not asked for, filter sync flags */
2273 	if ((ioflag & FRSYNC) == 0)
2274 		ioflag &= ~(FSYNC|FDSYNC);
2275 	error = VOP_READ(vp, &auio, ioflag, fp->f_cred, NULL);
2276 	cnt -= auio.uio_resid;
2277 
2278 	VOP_RWUNLOCK(vp, rwflag, NULL);
2279 
2280 	if (error == EINTR && cnt != 0)
2281 		error = 0;
2282 out:
2283 	if (error != 0) {
2284 		*err = error;
2285 		return (0);
2286 	} else {
2287 		*err = 0;
2288 		return (cnt);
2289 	}
2290 }
2291