xref: /illumos-gate/usr/src/uts/common/fs/nfs/nfs4_srv_ns.c (revision a7123019ed2e4decef929dc2e426f3f310788fa1)
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  * Copyright 2005 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/systm.h>
29 
30 #include <nfs/nfs.h>
31 #include <nfs/export.h>
32 #include <sys/cmn_err.h>
33 
34 /*
35  * A version of VOP_FID that deals with a remote VOP_FID for nfs.
36  * If vp is an nfs node, nfs4_fid() returns EREMOTE, nfs3_fid() and nfs_fid()
37  * returns the filehandle of vp as its fid. When nfs uses fid to set the
38  * exportinfo filehandle template, a remote nfs filehandle would be too big for
39  * the fid of the exported directory. This routine remaps the value of the
40  * attribute va_nodeid of vp to be the fid of vp, so that the fid can fit.
41  *
42  * We need this fid mainly for setting up NFSv4 server namespace where an
43  * nfs filesystem is also part of it. Thus, need to be able to setup a pseudo
44  * exportinfo for an nfs node.
45  *
46  * e.g. mount an ufs filesystem on an nfs filesystem, then share the ufs
47  *      filesystem. (like exporting a local disk from a "diskless" client)
48  */
49 int
50 vop_fid_pseudo(vnode_t *vp, fid_t *fidp)
51 {
52 	struct vattr va;
53 	int error;
54 
55 	error = VOP_FID(vp, fidp);
56 
57 	/*
58 	 * XXX nfs4_fid() does nothing and returns EREMOTE.
59 	 * XXX nfs3_fid()/nfs_fid() returns nfs filehandle as its fid
60 	 * which has a bigger length than local fid.
61 	 * NFS_FHMAXDATA is the size of fhandle_t.fh_xdata[NFS_FHMAXDATA].
62 	 */
63 	if (error == EREMOTE || (error == 0 && fidp->fid_len > NFS_FHMAXDATA)) {
64 
65 		va.va_mask = AT_NODEID;
66 		error = VOP_GETATTR(vp, &va, 0, CRED());
67 		if (error)
68 			return (error);
69 
70 		fidp->fid_len = sizeof (va.va_nodeid);
71 		bcopy(&va.va_nodeid, fidp->fid_data, fidp->fid_len);
72 		return (0);
73 	}
74 
75 	return (error);
76 }
77 
78 /*
79  * Get an nfsv4 vnode of the given fid from the visible list of an
80  * nfs filesystem or get the exi_vp if it is the root node.
81  */
82 int
83 nfs4_vget_pseudo(struct exportinfo *exi, vnode_t **vpp, fid_t *fidp)
84 {
85 	fid_t exp_fid;
86 	struct exp_visible *visp;
87 	int error;
88 
89 	/* check if the given fid is in the visible list */
90 
91 	for (visp = exi->exi_visible; visp; visp = visp->vis_next) {
92 		if (EQFID(fidp, &visp->vis_fid)) {
93 			VN_HOLD(visp->vis_vp);
94 			*vpp = visp->vis_vp;
95 			return (0);
96 		}
97 	}
98 
99 	/* check if the given fid is the same as the exported node */
100 
101 	bzero(&exp_fid, sizeof (exp_fid));
102 	exp_fid.fid_len = MAXFIDSZ;
103 	error = vop_fid_pseudo(exi->exi_vp, &exp_fid);
104 	if (error)
105 		return (error);
106 
107 	if (EQFID(fidp, &exp_fid)) {
108 		VN_HOLD(exi->exi_vp);
109 		*vpp = exi->exi_vp;
110 		return (0);
111 	}
112 
113 	return (ENOENT);
114 }
115 
116 /*
117  * Create a pseudo export entry
118  *
119  * This is an export entry that's created as the
120  * side-effect of a "real" export.  As a part of
121  * a real export, the pathname to the export is
122  * checked to see if all the directory components
123  * are accessible via an NFSv4 client, i.e. are
124  * exported.  If tree_climb() finds an unexported
125  * mountpoint along the path, then it calls this
126  * function to export it.
127  *
128  * This pseudo export differs from a real export
129  * in restriction on simple. read-only access,
130  * and the addition of a "visible" list of directories.
131  * A real export may have a visible list if it is a root of
132  * a file system and at least one of its subtree resides in
133  * a different file system is shared.
134  *
135  * A visible list is per file system. It resides in the exportinfo
136  * for the pseudo node (VROOT) and it could reside in a real export
137  * of a VROOT node.
138  */
139 int
140 pseudo_exportfs(vnode_t *vp, struct exp_visible *vis_head,
141 					struct exportdata *exdata)
142 {
143 	struct exportinfo *exi;
144 	struct exportdata *kex;
145 	fid_t fid;
146 	fsid_t fsid;
147 	int error;
148 	char *pseudo;
149 
150 	ASSERT(RW_WRITE_HELD(&exported_lock));
151 
152 	/*
153 	 * Get the vfs id
154 	 */
155 	bzero(&fid, sizeof (fid));
156 	fid.fid_len = MAXFIDSZ;
157 	error = vop_fid_pseudo(vp, &fid);
158 	if (error) {
159 		/*
160 		 * If VOP_FID returns ENOSPC then the fid supplied
161 		 * is too small.  For now we simply return EREMOTE.
162 		 */
163 		if (error == ENOSPC)
164 			error = EREMOTE;
165 		return (error);
166 	}
167 
168 	fsid = vp->v_vfsp->vfs_fsid;
169 	exi = kmem_zalloc(sizeof (*exi), KM_SLEEP);
170 	exi->exi_fsid = fsid;
171 	exi->exi_fid = fid;
172 	exi->exi_vp = vp;
173 	VN_HOLD(exi->exi_vp);
174 	exi->exi_visible = vis_head;
175 	exi->exi_count = 1;
176 	exi->exi_volatile_dev = (vfssw[vp->v_vfsp->vfs_fstype].vsw_flag &
177 				VSW_VOLATILEDEV) ? 1 : 0;
178 	mutex_init(&exi->exi_lock, NULL, MUTEX_DEFAULT, NULL);
179 
180 	/*
181 	 * Build up the template fhandle
182 	 */
183 	exi->exi_fh.fh_fsid = fsid;
184 	ASSERT(exi->exi_fid.fid_len <= sizeof (exi->exi_fh.fh_xdata));
185 	exi->exi_fh.fh_xlen = exi->exi_fid.fid_len;
186 	bcopy(exi->exi_fid.fid_data, exi->exi_fh.fh_xdata,
187 	    exi->exi_fid.fid_len);
188 	exi->exi_fh.fh_len = sizeof (exi->exi_fh.fh_data);
189 
190 	kex = &exi->exi_export;
191 	kex->ex_flags = EX_PSEUDO;
192 
193 	/* Set up a generic pathname */
194 
195 	pseudo = "(pseudo)";
196 	kex->ex_pathlen = strlen(pseudo);
197 	kex->ex_path = kmem_alloc(kex->ex_pathlen + 1, KM_SLEEP);
198 	(void) strcpy(kex->ex_path, pseudo);
199 
200 	/* Transfer the secinfo data from exdata to this new pseudo node */
201 	if (exdata)
202 		srv_secinfo_exp2pseu(&exi->exi_export, exdata);
203 
204 	/*
205 	 * Initialize auth cache lock
206 	 */
207 	rw_init(&exi->exi_cache_lock, NULL, RW_DEFAULT, NULL);
208 
209 	/*
210 	 * Insert the new entry at the front of the export list
211 	 */
212 	export_link(exi);
213 
214 	return (0);
215 }
216 
217 /*
218  * Free a list of visible directories
219  */
220 void
221 free_visible(struct exp_visible *head)
222 {
223 	struct exp_visible *visp, *next;
224 
225 	for (visp = head; visp; visp = next) {
226 		if (visp->vis_vp != NULL)
227 			VN_RELE(visp->vis_vp);
228 		next = visp->vis_next;
229 		kmem_free(visp, sizeof (*visp));
230 	}
231 }
232 
233 /*
234  * Add a list of visible directories to a pseudo exportfs.
235  *
236  * When we export a new directory we need to add a new
237  * path segment through the pseudofs to reach the new
238  * directory. This new path is reflected in a list of
239  * directories added to the "visible" list.
240  *
241  * Here there are two lists of visible fids: one hanging off the
242  * pseudo exportinfo, and the one we want to add.  It's possible
243  * that the two lists share a common path segment
244  * and have some common directories.  We need to combine
245  * the lists so there's no duplicate entries. Where a common
246  * path component is found, the vis_count field is bumped.
247  *
248  * When the addition is complete, the supplied list is freed.
249  */
250 
251 static void
252 more_visible(struct exportinfo *exi, struct exp_visible *vis_head)
253 {
254 	struct exp_visible *vp1, *vp2;
255 	struct exp_visible *tail, *new;
256 	int found;
257 
258 	/*
259 	 * If exportinfo doesn't already have a visible
260 	 * list just assign the entire supplied list.
261 	 */
262 	if (exi->exi_visible == NULL) {
263 		exi->exi_visible = vis_head;
264 		return;
265 	}
266 
267 	/*
268 	 * The outer loop traverses the supplied list.
269 	 */
270 	for (vp1 = vis_head; vp1; vp1 = vp1->vis_next) {
271 
272 		/*
273 		 * Given an element from the list to be added,
274 		 * search the exportinfo visible list looking for a match.
275 		 * If a match is found, increment the reference count.
276 		 */
277 		found = 0;
278 
279 		for (vp2 = exi->exi_visible; vp2; vp2 = vp2->vis_next) {
280 
281 			tail = vp2;
282 
283 			if (EQFID(&vp1->vis_fid, &vp2->vis_fid)) {
284 				found = 1;
285 				vp2->vis_count++;
286 				VN_RELE(vp1->vis_vp);
287 				vp1->vis_vp = NULL;
288 
289 				/*
290 				 * If the visible struct we want to add
291 				 * (vp1) has vis_exported set to 1, then
292 				 * the matching visible struct we just found
293 				 * must also have it's vis_exported field
294 				 * set to 1.
295 				 *
296 				 * For example, if /export/home was shared
297 				 * (and a UFS mountpoint), then "export" and
298 				 * "home" would each have visible structs in
299 				 * the root pseudo exportinfo. The vis_exported
300 				 * for home would be 1, and vis_exported for
301 				 * export would be 0.  Now, if /export was
302 				 * also shared, more_visible would find the
303 				 * existing visible struct for export, and
304 				 * see that vis_exported was 0.  The code
305 				 * below will set it to 1.
306 				 *
307 				 * vp1 is from vis list passed in (vis_head)
308 				 * vp2 is from vis list on pseudo exportinfo
309 				 */
310 				if (vp1->vis_exported && !vp2->vis_exported)
311 					vp2->vis_exported = 1;
312 				break;
313 			}
314 		}
315 
316 		/* If not found - add to the end of the list */
317 		if (! found) {
318 			new = kmem_zalloc(sizeof (*new), KM_SLEEP);
319 			*new = *vp1;
320 			tail->vis_next = new;
321 			new->vis_next = NULL;
322 			vp1->vis_vp = NULL;
323 		}
324 	}
325 
326 	/*
327 	 * Throw away the path list. vis_vp pointers in vis_head list
328 	 * are either VN_RELEed or reassigned, and are set to NULL.
329 	 * There is no need to VN_RELE in free_visible for this vis_head.
330 	 */
331 	free_visible(vis_head);
332 }
333 
334 /*
335  * Remove a list of visible directories from the pseudo exportfs.
336  *
337  * When we unexport a directory, we have to remove path
338  * components from the visible list in the pseudo exportfs
339  * entry.  The supplied visible list contains the fids of the path
340  * to the unexported directory.  The visible list of the export
341  * is checked against this list any matching fids have their
342  * reference count decremented.  If a reference count drops to
343  * zero, then it means no paths now use this directory, so its
344  * fid can be removed from the visible list.
345  *
346  * When the last path is removed, the visible list will be null.
347  */
348 static void
349 less_visible(struct exportinfo *exi, struct exp_visible *vis_head)
350 {
351 	struct exp_visible *vp1, *vp2;
352 	struct exp_visible *prev, *next;
353 
354 	/*
355 	 * The outer loop traverses the supplied list.
356 	 */
357 	for (vp1 = vis_head; vp1; vp1 = vp1->vis_next) {
358 
359 		/*
360 		 * Given an element from the list to be removed,
361 		 * search the exportinfo list looking for a match.
362 		 * If a match is found, decrement the reference
363 		 * count and drop the element if the count drops
364 		 * to zero.
365 		 */
366 		for (vp2 = exi->exi_visible, prev = NULL; vp2; vp2 = next) {
367 
368 			next = vp2->vis_next;
369 
370 			if (EQFID(&vp1->vis_fid, &vp2->vis_fid)) {
371 
372 				/*
373 				 * Decrement the ref count.
374 				 * Remove the entry if it's zero.
375 				 */
376 				if (--vp2->vis_count <= 0) {
377 					if (prev == NULL)
378 						exi->exi_visible = next;
379 					else
380 						prev->vis_next = next;
381 
382 					VN_RELE(vp2->vis_vp);
383 					kmem_free(vp2, sizeof (*vp1));
384 				} else {
385 					/*
386 					 * If we're here, then the vp2 will
387 					 * remain in the vis list.  If the
388 					 * vis entry corresponds to the object
389 					 * being unshared, then vis_exported
390 					 * needs to be set to 0.
391 					 *
392 					 * vp1 is a node from caller's list
393 					 * vp2 is node from exportinfo's list
394 					 *
395 					 * Only 1 node in the caller's list
396 					 * will have vis_exported set to 1,
397 					 * and it corresponds to the obj being
398 					 * unshared.  It should always be the
399 					 * last element of the caller's list.
400 					 */
401 					if (vp1->vis_exported &&
402 					    vp2->vis_exported) {
403 						vp2->vis_exported = 0;
404 					}
405 				}
406 
407 				break;
408 			}
409 
410 			prev = vp2;
411 		}
412 	}
413 
414 	free_visible(vis_head);
415 }
416 
417 /*
418  * This function checks the path to a new export to
419  * check whether all the pathname components are
420  * exported. It works by climbing the file tree one
421  * component at a time via "..", crossing mountpoints
422  * if necessary until an export entry is found, or the
423  * system root is reached.
424  *
425  * If an unexported mountpoint is found, then
426  * a new pseudo export is added and the pathname from
427  * the mountpoint down to the export is added to the
428  * visible list for the new pseudo export.  If an existing
429  * pseudo export is found, then the pathname is added
430  * to its visible list.
431  *
432  * Note that there's some tests for exportdir.
433  * The exportinfo entry that's passed as a parameter
434  * is that of the real export and exportdir is set
435  * for this case.
436  *
437  * Here is an example of a possible setup:
438  *
439  * () - a new fs; fs mount point
440  * EXPORT - a real exported node
441  * PSEUDO - a pseudo node
442  * vis - visible list
443  * f# - security flavor#
444  * (f#) - security flavor# propagated from its decendents
445  * "" - covered vnode
446  *
447  *
448  *                 /
449  *                 |
450  *                 (a) PSEUDO (f1,f2)
451  *                 |   vis: b,b,"c","n"
452  *                 |
453  *                 b
454  *        ---------|------------------
455  *        |                          |
456  *        (c) EXPORT,f1(f2)          (n) PSEUDO (f1,f2)
457  *        |   vis: "e","d"           |   vis: m,m,,p,q,"o"
458  *        |                          |
459  *  ------------------          -------------------
460  *  |        |        |         |                  |
461  *  (d)      (e)      f         m EXPORT,f1(f2)    p
462  *  EXPORT   EXPORT             |                  |
463  *  f1       f2                 |                  |
464  *           |                  |                  |
465  *           j                 (o) EXPORT,f2       q EXPORT f2
466  *
467  */
468 int
469 treeclimb_export(struct exportinfo *exip)
470 {
471 	vnode_t *dvp, *vp;
472 	fid_t fid;
473 	int error;
474 	int exportdir;
475 	struct exportinfo *exi = NULL;
476 	struct exp_visible *visp;
477 	struct exp_visible *vis_head = NULL;
478 	struct vattr va;
479 
480 	ASSERT(RW_WRITE_HELD(&exported_lock));
481 
482 	vp = exip->exi_vp;
483 	VN_HOLD(vp);
484 	exportdir = 1;
485 
486 	for (;;) {
487 
488 		bzero(&fid, sizeof (fid));
489 		fid.fid_len = MAXFIDSZ;
490 		error = vop_fid_pseudo(vp, &fid);
491 		if (error)
492 			break;
493 
494 		if (! exportdir) {
495 			/*
496 			 * Check if this exportroot is a VROOT dir.  If so,
497 			 * then attach the pseudonodes.  If not, then
498 			 * continue .. traversal until we hit a VROOT
499 			 * export (pseudo or real).
500 			 */
501 			exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
502 			if (exi != NULL && vp->v_flag & VROOT) {
503 				/*
504 				 * Found an export info
505 				 *
506 				 * Extend the list of visible
507 				 * directories whether it's a pseudo
508 				 * or a real export.
509 				 */
510 				more_visible(exi, vis_head);
511 				vis_head = NULL;
512 				break;	/* and climb no further */
513 			}
514 		}
515 
516 		/*
517 		 * If at the root of the filesystem, need
518 		 * to traverse across the mountpoint
519 		 * and continue the climb on the mounted-on
520 		 * filesystem.
521 		 */
522 		if (vp->v_flag & VROOT) {
523 
524 			if (! exportdir) {
525 				/*
526 				 * Found the root directory of a filesystem
527 				 * that isn't exported.  Need to export
528 				 * this as a pseudo export so that an NFS v4
529 				 * client can do lookups in it.
530 				 */
531 				error = pseudo_exportfs(vp, vis_head, NULL);
532 				if (error)
533 					break;
534 				vis_head = NULL;
535 			}
536 
537 			if (VN_CMP(vp, rootdir)) {
538 				/* at system root */
539 				break;
540 			}
541 
542 			vp = untraverse(vp);
543 			exportdir = 0;
544 			continue;
545 		}
546 
547 		/*
548 		 * Do a getattr to obtain the nodeid (inode num)
549 		 * for this vnode.
550 		 */
551 		va.va_mask = AT_NODEID;
552 		error = VOP_GETATTR(vp, &va, 0, CRED());
553 		if (error)
554 			break;
555 
556 		/*
557 		 *  Add this directory fid to visible list
558 		 */
559 		visp = kmem_alloc(sizeof (*visp), KM_SLEEP);
560 		VN_HOLD(vp);
561 		visp->vis_vp = vp;
562 		visp->vis_fid = fid;		/* structure copy */
563 		visp->vis_ino = va.va_nodeid;
564 		visp->vis_count = 1;
565 		visp->vis_exported = exportdir;
566 		visp->vis_next = vis_head;
567 		vis_head = visp;
568 
569 		/*
570 		 * Now, do a ".." to find parent dir of vp.
571 		 */
572 		error = VOP_LOOKUP(vp, "..", &dvp, NULL, 0, NULL, CRED());
573 
574 		if (error == ENOTDIR && exportdir) {
575 			dvp = exip->exi_dvp;
576 			ASSERT(dvp != NULL);
577 			VN_HOLD(dvp);
578 			error = 0;
579 		}
580 
581 		if (error)
582 			break;
583 
584 		exportdir = 0;
585 		VN_RELE(vp);
586 		vp = dvp;
587 	}
588 
589 	VN_RELE(vp);
590 	return (error);
591 }
592 
593 /*
594  * Walk up the tree looking for pseudo export entries.
595  *
596  * If a pseudo export is found, remove the path we've
597  * climbed from its visible list. If the visible list
598  * still has entries after the removal, then we can stop.
599  * If it becomes null, then remove the pseudo export entry
600  * and carry on up the tree to see if there's any more.
601  */
602 int
603 treeclimb_unexport(struct exportinfo *exip)
604 {
605 	vnode_t *dvp, *vp;
606 	fid_t fid;
607 	int error = 0;
608 	int exportdir;
609 	struct exportinfo *exi = NULL;
610 	struct exp_visible *vis_head = NULL, *visp;
611 
612 	ASSERT(RW_WRITE_HELD(&exported_lock));
613 
614 	exportdir = 1;
615 	vp = exip->exi_vp;
616 	VN_HOLD(vp);
617 
618 	for (;;) {
619 
620 		bzero(&fid, sizeof (fid));
621 		fid.fid_len = MAXFIDSZ;
622 		error = vop_fid_pseudo(vp, &fid);
623 		if (error)
624 			break;
625 
626 		if (! exportdir) {
627 
628 			/*
629 			 * We need to use checkexport4() here because it
630 			 * doesn't acquire exported_lock and it doesn't
631 			 * manipulate exi_count.
632 			 *
633 			 * Remove directories from the visible
634 			 * list that are unique to the path
635 			 * for this export.  (Only VROOT exportinfos
636 			 * have can have visible entries).
637 			 */
638 			exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
639 			if (exi != NULL && (vp->v_flag & VROOT)) {
640 
641 				less_visible(exi, vis_head);
642 				vis_head = NULL;
643 
644 				/*
645 				 * If the visible list has entries
646 				 * or if it's a real export, then
647 				 * there's no need to keep climbing.
648 				 */
649 				if (exi->exi_visible || ! PSEUDO(exi))
650 					break;
651 
652 				/*
653 				 * Otherwise, we have a pseudo export
654 				 * with an empty list (no exports below
655 				 * it) so we must remove and continue
656 				 * the climb to remove its name from
657 				 * the parent export.
658 				 */
659 				error = export_unlink(&vp->v_vfsp->vfs_fsid,
660 						&fid, vp, NULL);
661 				if (error)
662 					break;
663 
664 				exi_rele(exi);
665 			}
666 		}
667 
668 		/*
669 		 * If at the root of the filesystem, need
670 		 * to traverse across the mountpoint
671 		 * and continue the climb on the mounted-on
672 		 * filesystem.
673 		 */
674 		if (vp->v_flag & VROOT) {
675 			if (VN_CMP(vp, rootdir)) {
676 				/* at system root */
677 				break;
678 			}
679 			vp = untraverse(vp);
680 			exportdir = 0;
681 			continue;
682 		}
683 
684 		/*
685 		 *  Add this directory fid to path list
686 		 */
687 		visp = kmem_alloc(sizeof (*visp), KM_SLEEP);
688 		VN_HOLD(vp);
689 		visp->vis_vp = vp;
690 		visp->vis_fid = fid;		/* structure copy */
691 		visp->vis_ino = 0;
692 		visp->vis_count = 1;
693 		visp->vis_exported = exportdir;
694 		visp->vis_next = vis_head;
695 		vis_head = visp;
696 
697 		/*
698 		 * Do a ".." to find parent dir of vp.
699 		 */
700 		error = VOP_LOOKUP(vp, "..", &dvp, NULL, 0, NULL, CRED());
701 
702 		if (error == ENOTDIR && exportdir) {
703 			dvp = exip->exi_dvp;
704 			ASSERT(dvp != NULL);
705 			VN_HOLD(dvp);
706 			error = 0;
707 		}
708 		if (error)
709 			break;
710 
711 		exportdir = 0;
712 		VN_RELE(vp);
713 		vp = dvp;
714 	}
715 
716 	VN_RELE(vp);
717 	return (error);
718 }
719 
720 
721 /*
722  * Traverse backward across mountpoint from the
723  * root vnode of a filesystem to its mounted-on
724  * vnode.
725  */
726 vnode_t *
727 untraverse(vnode_t *vp)
728 {
729 	vnode_t *tvp, *nextvp;
730 
731 	tvp = vp;
732 	for (;;) {
733 		if (! (tvp->v_flag & VROOT))
734 			break;
735 
736 		/* lock vfs to prevent unmount of this vfs */
737 		vfs_lock_wait(tvp->v_vfsp);
738 
739 		if ((nextvp = tvp->v_vfsp->vfs_vnodecovered) == NULL) {
740 			vfs_unlock(tvp->v_vfsp);
741 			break;
742 		}
743 
744 		/*
745 		 * Hold nextvp to prevent unmount.  After unlock vfs and
746 		 * rele tvp, any number of overlays could be unmounted.
747 		 * Putting a hold on vfs_vnodecovered will only allow
748 		 * tvp's vfs to be unmounted. Of course if caller placed
749 		 * extra hold on vp before calling untraverse, the following
750 		 * hold would not be needed.  Since prev actions of caller
751 		 * are unknown, we need to hold here just to be safe.
752 		 */
753 		VN_HOLD(nextvp);
754 		vfs_unlock(tvp->v_vfsp);
755 		VN_RELE(tvp);
756 		tvp = nextvp;
757 	}
758 
759 	return (tvp);
760 }
761 
762 /*
763  * Given an exportinfo, climb up to find the exportinfo for the VROOT
764  * of the filesystem.
765  *
766  * e.g.         /
767  *              |
768  *              a (VROOT) pseudo-exportinfo
769  *		|
770  *		b
771  *		|
772  *		c  #share /a/b/c
773  *		|
774  *		d
775  *
776  * where c is in the same filesystem as a.
777  * So, get_root_export(*exportinfo_for_c) returns exportinfo_for_a
778  *
779  * If d is shared, then c will be put into a's visible list.
780  * Note: visible list is per filesystem and is attached to the
781  * VROOT exportinfo.
782  */
783 struct exportinfo *
784 get_root_export(struct exportinfo *exip)
785 {
786 	vnode_t *dvp, *vp;
787 	fid_t fid;
788 	struct exportinfo *exi = exip;
789 	int error;
790 
791 	vp = exi->exi_vp;
792 	VN_HOLD(vp);
793 
794 	for (;;) {
795 
796 		if (vp->v_flag & VROOT) {
797 			ASSERT(exi != NULL);
798 			break;
799 		}
800 
801 		/*
802 		 * Now, do a ".." to find parent dir of vp.
803 		 */
804 		error = VOP_LOOKUP(vp, "..", &dvp, NULL, 0, NULL, CRED());
805 
806 		if (error) {
807 			exi = NULL;
808 			break;
809 		}
810 
811 		VN_RELE(vp);
812 		vp = dvp;
813 
814 		bzero(&fid, sizeof (fid));
815 		fid.fid_len = MAXFIDSZ;
816 		error = vop_fid_pseudo(vp, &fid);
817 		if (error) {
818 			exi = NULL;
819 			break;
820 		}
821 
822 		exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
823 	}
824 
825 	VN_RELE(vp);
826 	return (exi);
827 }
828 
829 /*
830  * Return true if the supplied vnode has a sub-directory exported.
831  */
832 int
833 has_visible(struct exportinfo *exi, vnode_t *vp)
834 {
835 	struct exp_visible *visp;
836 	fid_t fid;
837 	bool_t vp_is_exported;
838 
839 	vp_is_exported = VN_CMP(vp,  exi->exi_vp);
840 
841 	/*
842 	 * An exported root vnode has a sub-dir shared if it has a visible list.
843 	 * i.e. if it does not have a visible list, then there is no node in
844 	 * this filesystem leads to any other shared node.
845 	 */
846 	if (vp_is_exported && (vp->v_flag & VROOT))
847 		return (exi->exi_visible ? 1 : 0);
848 
849 	/*
850 	 * Only the exportinfo of a fs root node may have a visible list.
851 	 * Either it is a pseudo root node, or a real exported root node.
852 	 */
853 	if ((exi = get_root_export(exi)) == NULL) {
854 		return (0);
855 	}
856 
857 	if (!exi->exi_visible)
858 		return (0);
859 
860 	/* Get the fid of the vnode */
861 	bzero(&fid, sizeof (fid));
862 	fid.fid_len = MAXFIDSZ;
863 	if (vop_fid_pseudo(vp, &fid) != 0) {
864 		return (0);
865 	}
866 
867 	/*
868 	 * See if vp is in the visible list of the root node exportinfo.
869 	 */
870 	for (visp = exi->exi_visible; visp; visp = visp->vis_next) {
871 		if (EQFID(&fid, &visp->vis_fid)) {
872 			/*
873 			 * If vp is an exported non-root node with only 1 path
874 			 * count (for itself), it indicates no sub-dir shared
875 			 * using this vp as a path.
876 			 */
877 			if (vp_is_exported && visp->vis_count < 2)
878 				break;
879 
880 			return (1);
881 		}
882 	}
883 
884 	return (0);
885 }
886 
887 /*
888  * Returns true if the supplied vnode is visible
889  * in this export.  If vnode is visible, return
890  * vis_exported in expseudo.
891  */
892 int
893 nfs_visible(struct exportinfo *exi, vnode_t *vp, int *expseudo)
894 {
895 	struct exp_visible *visp;
896 	fid_t fid;
897 
898 	/*
899 	 * First check to see if vp is export root.
900 	 *
901 	 * A pseudo export root can never be exported
902 	 * (it would be a real export then); however,
903 	 * it is always visible.  If a pseudo root object
904 	 * was exported by server admin, then the entire
905 	 * pseudo exportinfo (and all visible entries) would
906 	 * be destroyed.  A pseudo exportinfo only exists
907 	 * to provide access to real (descendant) export(s).
908 	 *
909 	 * Previously, rootdir was special cased here; however,
910 	 * the export root special case handles the rootdir
911 	 * case also.
912 	 */
913 	if (VN_CMP(vp, exi->exi_vp)) {
914 		*expseudo = 0;
915 		return (1);
916 	}
917 
918 	/*
919 	 * Only a PSEUDO node has a visible list or an exported VROOT
920 	 * node may have a visible list.
921 	 */
922 	if (! PSEUDO(exi) && (exi = get_root_export(exi)) == NULL) {
923 		*expseudo = 0;
924 		return (0);
925 	}
926 
927 	/* Get the fid of the vnode */
928 
929 	bzero(&fid, sizeof (fid));
930 	fid.fid_len = MAXFIDSZ;
931 	if (vop_fid_pseudo(vp, &fid) != 0) {
932 		*expseudo = 0;
933 		return (0);
934 	}
935 
936 	/*
937 	 * We can't trust VN_CMP() above because of LOFS.
938 	 * Even though VOP_CMP will do the right thing for LOFS
939 	 * objects, VN_CMP will short circuit out early when the
940 	 * vnode ops ptrs are different.  Just in case we're dealing
941 	 * with LOFS, compare exi_fid/fsid here.
942 	 *
943 	 * expseudo is not set because this is not an export
944 	 */
945 	if (EQFID(&exi->exi_fid, &fid) &&
946 	    EQFSID(&exi->exi_fsid, &vp->v_vfsp->vfs_fsid)) {
947 		*expseudo = 0;
948 		return (1);
949 	}
950 
951 
952 	/* See if it matches any fid in the visible list */
953 
954 	for (visp = exi->exi_visible; visp; visp = visp->vis_next) {
955 		if (EQFID(&fid, &visp->vis_fid)) {
956 			*expseudo = visp->vis_exported;
957 			return (1);
958 		}
959 	}
960 
961 	*expseudo = 0;
962 
963 	return (0);
964 }
965 
966 /*
967  * Returns true if the supplied vnode is the
968  * directory of an export point.
969  */
970 int
971 nfs_exported(struct exportinfo *exi, vnode_t *vp)
972 {
973 	struct exp_visible *visp;
974 	fid_t fid;
975 
976 	/*
977 	 * First check to see if vp is the export root
978 	 * This check required for the case of lookup ..
979 	 * where .. is a V_ROOT vnode and a pseudo exportroot.
980 	 * Pseudo export root objects do not have an entry
981 	 * in the visible list even though every V_ROOT
982 	 * pseudonode is visible.  It is safe to compare
983 	 * vp here because pseudo_exportfs put a hold on
984 	 * it when exi_vp was initialized.
985 	 *
986 	 * Note: VN_CMP() won't match for LOFS shares, but they're
987 	 * handled below w/EQFID/EQFSID.
988 	 */
989 	if (VN_CMP(vp, exi->exi_vp))
990 		return (1);
991 
992 	/* Get the fid of the vnode */
993 
994 	bzero(&fid, sizeof (fid));
995 	fid.fid_len = MAXFIDSZ;
996 	if (vop_fid_pseudo(vp, &fid) != 0)
997 		return (0);
998 
999 	if (EQFID(&fid, &exi->exi_fid) &&
1000 	    EQFSID(&vp->v_vfsp->vfs_fsid, &exi->exi_fsid)) {
1001 		return (1);
1002 	}
1003 
1004 	/* See if it matches any fid in the visible list */
1005 
1006 	for (visp = exi->exi_visible; visp; visp = visp->vis_next) {
1007 		if (EQFID(&fid, &visp->vis_fid))
1008 			return (visp->vis_exported);
1009 	}
1010 
1011 	return (0);
1012 }
1013 
1014 /*
1015  * Returns true if the supplied inode is visible
1016  * in this export.  This function is used by
1017  * readdir which uses inode numbers from the
1018  * directory.
1019  *
1020  * NOTE: this code does not match inode number for ".",
1021  * but it isn't required because NFS4 server rddir
1022  * skips . and .. entries.
1023  */
1024 int
1025 nfs_visible_inode(struct exportinfo *exi, ino64_t ino, int *expseudo)
1026 {
1027 	struct exp_visible *visp;
1028 
1029 	/*
1030 	 * Only a PSEUDO node has a visible list or an exported VROOT
1031 	 * node may have a visible list.
1032 	 */
1033 	if (! PSEUDO(exi) && (exi = get_root_export(exi)) == NULL) {
1034 		*expseudo = 0;
1035 		return (0);
1036 	}
1037 
1038 	for (visp = exi->exi_visible; visp; visp = visp->vis_next)
1039 		if ((u_longlong_t)ino == visp->vis_ino) {
1040 			*expseudo = visp->vis_exported;
1041 			return (1);
1042 		}
1043 
1044 	*expseudo = 0;
1045 	return (0);
1046 }
1047