xref: /illumos-gate/usr/src/uts/common/fs/nfs/nfs4_srv_ns.c (revision 6a074c93c5dee390d8ca2377f42e55418f0a9eb3)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  *
21  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
22  * Use is subject to license terms.
23  */
24 
25 #pragma ident	"%Z%%M%	%I%	%E% SMI"
26 
27 #include <sys/systm.h>
28 
29 #include <nfs/nfs.h>
30 #include <nfs/export.h>
31 #include <sys/cmn_err.h>
32 
33 #define	PSEUDOFS_SUFFIX		" (pseudo)"
34 
35 /*
36  * A version of VOP_FID that deals with a remote VOP_FID for nfs.
37  * If vp is an nfs node, nfs4_fid() returns EREMOTE, nfs3_fid() and nfs_fid()
38  * returns the filehandle of vp as its fid. When nfs uses fid to set the
39  * exportinfo filehandle template, a remote nfs filehandle would be too big for
40  * the fid of the exported directory. This routine remaps the value of the
41  * attribute va_nodeid of vp to be the fid of vp, so that the fid can fit.
42  *
43  * We need this fid mainly for setting up NFSv4 server namespace where an
44  * nfs filesystem is also part of it. Thus, need to be able to setup a pseudo
45  * exportinfo for an nfs node.
46  *
47  * e.g. mount a filesystem on top of a nfs dir, and then share the new mount
48  *      (like exporting a local disk from a "diskless" client)
49  */
50 int
51 vop_fid_pseudo(vnode_t *vp, fid_t *fidp)
52 {
53 	struct vattr va;
54 	int error;
55 
56 	error = VOP_FID(vp, fidp);
57 
58 	/*
59 	 * XXX nfs4_fid() does nothing and returns EREMOTE.
60 	 * XXX nfs3_fid()/nfs_fid() returns nfs filehandle as its fid
61 	 * which has a bigger length than local fid.
62 	 * NFS_FH4MAXDATA is the size of
63 	 * fhandle4_t.fh_xdata[NFS_FH4MAXDATA].
64 	 *
65 	 * Note: nfs[2,3,4]_fid() only gets called for diskless clients.
66 	 */
67 	if (error == EREMOTE ||
68 	    (error == 0 && fidp->fid_len > NFS_FH4MAXDATA)) {
69 
70 		va.va_mask = AT_NODEID;
71 		error = VOP_GETATTR(vp, &va, 0, CRED());
72 		if (error)
73 			return (error);
74 
75 		fidp->fid_len = sizeof (va.va_nodeid);
76 		bcopy(&va.va_nodeid, fidp->fid_data, fidp->fid_len);
77 		return (0);
78 	}
79 
80 	return (error);
81 }
82 
83 /*
84  * Get an nfsv4 vnode of the given fid from the visible list of an
85  * nfs filesystem or get the exi_vp if it is the root node.
86  */
87 int
88 nfs4_vget_pseudo(struct exportinfo *exi, vnode_t **vpp, fid_t *fidp)
89 {
90 	fid_t exp_fid;
91 	struct exp_visible *visp;
92 	int error;
93 
94 	/* check if the given fid is in the visible list */
95 
96 	for (visp = exi->exi_visible; visp; visp = visp->vis_next) {
97 		if (EQFID(fidp, &visp->vis_fid)) {
98 			VN_HOLD(visp->vis_vp);
99 			*vpp = visp->vis_vp;
100 			return (0);
101 		}
102 	}
103 
104 	/* check if the given fid is the same as the exported node */
105 
106 	bzero(&exp_fid, sizeof (exp_fid));
107 	exp_fid.fid_len = MAXFIDSZ;
108 	error = vop_fid_pseudo(exi->exi_vp, &exp_fid);
109 	if (error)
110 		return (error);
111 
112 	if (EQFID(fidp, &exp_fid)) {
113 		VN_HOLD(exi->exi_vp);
114 		*vpp = exi->exi_vp;
115 		return (0);
116 	}
117 
118 	return (ENOENT);
119 }
120 
121 /*
122  * Create a pseudo export entry
123  *
124  * This is an export entry that's created as the
125  * side-effect of a "real" export.  As a part of
126  * a real export, the pathname to the export is
127  * checked to see if all the directory components
128  * are accessible via an NFSv4 client, i.e. are
129  * exported.  If treeclimb_export() finds an unexported
130  * mountpoint along the path, then it calls this
131  * function to export it.
132  *
133  * This pseudo export differs from a real export in that
134  * it only allows read-only access.  A "visible" list of
135  * directories is added to filter lookup and readdir results
136  * to only contain dirnames which lead to descendant shares.
137  *
138  * A visible list has a per-file-system scope.  Any exportinfo
139  * struct (real or pseudo) can have a visible list as long as
140  * a) its export root is VROOT
141  * b) a descendant of the export root is shared
142  */
143 int
144 pseudo_exportfs(vnode_t *vp, struct exp_visible *vis_head,
145 					struct exportdata *exdata)
146 {
147 	struct exportinfo *exi;
148 	struct exportdata *kex;
149 	fid_t fid;
150 	fsid_t fsid;
151 	int error, vpathlen;
152 
153 	ASSERT(RW_WRITE_HELD(&exported_lock));
154 
155 	/*
156 	 * Get the vfs id
157 	 */
158 	bzero(&fid, sizeof (fid));
159 	fid.fid_len = MAXFIDSZ;
160 	error = vop_fid_pseudo(vp, &fid);
161 	if (error) {
162 		/*
163 		 * If VOP_FID returns ENOSPC then the fid supplied
164 		 * is too small.  For now we simply return EREMOTE.
165 		 */
166 		if (error == ENOSPC)
167 			error = EREMOTE;
168 		return (error);
169 	}
170 
171 	fsid = vp->v_vfsp->vfs_fsid;
172 	exi = kmem_zalloc(sizeof (*exi), KM_SLEEP);
173 	exi->exi_fsid = fsid;
174 	exi->exi_fid = fid;
175 	exi->exi_vp = vp;
176 	VN_HOLD(exi->exi_vp);
177 	exi->exi_visible = vis_head;
178 	exi->exi_count = 1;
179 	exi->exi_volatile_dev = (vfssw[vp->v_vfsp->vfs_fstype].vsw_flag &
180 	    VSW_VOLATILEDEV) ? 1 : 0;
181 	mutex_init(&exi->exi_lock, NULL, MUTEX_DEFAULT, NULL);
182 
183 	/*
184 	 * Build up the template fhandle
185 	 */
186 	exi->exi_fh.fh_fsid = fsid;
187 	ASSERT(exi->exi_fid.fid_len <= sizeof (exi->exi_fh.fh_xdata));
188 	exi->exi_fh.fh_xlen = exi->exi_fid.fid_len;
189 	bcopy(exi->exi_fid.fid_data, exi->exi_fh.fh_xdata,
190 	    exi->exi_fid.fid_len);
191 	exi->exi_fh.fh_len = sizeof (exi->exi_fh.fh_data);
192 
193 	kex = &exi->exi_export;
194 	kex->ex_flags = EX_PSEUDO;
195 
196 	vpathlen = vp->v_path ? strlen(vp->v_path) : 0;
197 	kex->ex_pathlen = vpathlen + strlen(PSEUDOFS_SUFFIX);
198 	kex->ex_path = kmem_alloc(kex->ex_pathlen + 1, KM_SLEEP);
199 
200 	if (vpathlen)
201 		(void) strcpy(kex->ex_path, vp->v_path);
202 	(void) strcpy(kex->ex_path + vpathlen, PSEUDOFS_SUFFIX);
203 
204 	/* Transfer the secinfo data from exdata to this new pseudo node */
205 	if (exdata)
206 		srv_secinfo_exp2pseu(&exi->exi_export, exdata);
207 
208 	/*
209 	 * Initialize auth cache lock
210 	 */
211 	rw_init(&exi->exi_cache_lock, NULL, RW_DEFAULT, NULL);
212 
213 	/*
214 	 * Insert the new entry at the front of the export list
215 	 */
216 	export_link(exi);
217 
218 	return (0);
219 }
220 
221 /*
222  * Free a list of visible directories
223  */
224 void
225 free_visible(struct exp_visible *head)
226 {
227 	struct exp_visible *visp, *next;
228 
229 	for (visp = head; visp; visp = next) {
230 		if (visp->vis_vp != NULL)
231 			VN_RELE(visp->vis_vp);
232 
233 		next = visp->vis_next;
234 		srv_secinfo_list_free(visp->vis_secinfo, visp->vis_seccnt);
235 		kmem_free(visp, sizeof (*visp));
236 	}
237 }
238 
239 /*
240  * Add a list of visible directories to a pseudo exportfs.
241  *
242  * When we export a new directory we need to add a new
243  * path segment through the pseudofs to reach the new
244  * directory. This new path is reflected in a list of
245  * directories added to the "visible" list.
246  *
247  * Here there are two lists of visible fids: one hanging off the
248  * pseudo exportinfo, and the one we want to add.  It's possible
249  * that the two lists share a common path segment
250  * and have some common directories.  We need to combine
251  * the lists so there's no duplicate entries. Where a common
252  * path component is found, the vis_count field is bumped.
253  *
254  * When the addition is complete, the supplied list is freed.
255  */
256 
257 static void
258 more_visible(struct exportinfo *exi, struct exp_visible *vis_head)
259 {
260 	struct exp_visible *vp1, *vp2;
261 	struct exp_visible *tail, *new;
262 	int found;
263 
264 	/*
265 	 * If exportinfo doesn't already have a visible
266 	 * list just assign the entire supplied list.
267 	 */
268 	if (exi->exi_visible == NULL) {
269 		exi->exi_visible = vis_head;
270 		return;
271 	}
272 
273 	/*
274 	 * The outer loop traverses the supplied list.
275 	 */
276 	for (vp1 = vis_head; vp1; vp1 = vp1->vis_next) {
277 
278 		/*
279 		 * Given an element from the list to be added,
280 		 * search the exportinfo visible list looking for a match.
281 		 * If a match is found, increment the reference count.
282 		 */
283 		found = 0;
284 
285 		for (vp2 = exi->exi_visible; vp2; vp2 = vp2->vis_next) {
286 
287 			tail = vp2;
288 
289 			if (EQFID(&vp1->vis_fid, &vp2->vis_fid)) {
290 				found = 1;
291 				vp2->vis_count++;
292 				VN_RELE(vp1->vis_vp);
293 				vp1->vis_vp = NULL;
294 
295 				/*
296 				 * If the visible struct we want to add
297 				 * (vp1) has vis_exported set to 1, then
298 				 * the matching visible struct we just found
299 				 * must also have it's vis_exported field
300 				 * set to 1.
301 				 *
302 				 * For example, if /export/home was shared
303 				 * (and a mountpoint), then "export" and
304 				 * "home" would each have visible structs in
305 				 * the root pseudo exportinfo. The vis_exported
306 				 * for home would be 1, and vis_exported for
307 				 * export would be 0.  Now, if /export was
308 				 * also shared, more_visible would find the
309 				 * existing visible struct for export, and
310 				 * see that vis_exported was 0.  The code
311 				 * below will set it to 1.
312 				 *
313 				 * vp1 is from vis list passed in (vis_head)
314 				 * vp2 is from vis list on pseudo exportinfo
315 				 */
316 				if (vp1->vis_exported && !vp2->vis_exported)
317 					vp2->vis_exported = 1;
318 				break;
319 			}
320 		}
321 
322 		/* If not found - add to the end of the list */
323 		if (! found) {
324 			new = kmem_zalloc(sizeof (*new), KM_SLEEP);
325 			*new = *vp1;
326 			tail->vis_next = new;
327 			new->vis_next = NULL;
328 			vp1->vis_vp = NULL;
329 		}
330 	}
331 
332 	/*
333 	 * Throw away the path list. vis_vp pointers in vis_head list
334 	 * are either VN_RELEed or reassigned, and are set to NULL.
335 	 * There is no need to VN_RELE in free_visible for this vis_head.
336 	 */
337 	free_visible(vis_head);
338 }
339 
340 /*
341  * Remove a list of visible directories from the pseudo exportfs.
342  *
343  * When we unexport a directory, we have to remove path
344  * components from the visible list in the pseudo exportfs
345  * entry.  The supplied visible list contains the fids of the path
346  * to the unexported directory.  The visible list of the export
347  * is checked against this list any matching fids have their
348  * reference count decremented.  If a reference count drops to
349  * zero, then it means no paths now use this directory, so its
350  * fid can be removed from the visible list.
351  *
352  * When the last path is removed, the visible list will be null.
353  */
354 static void
355 less_visible(struct exportinfo *exi, struct exp_visible *vis_head)
356 {
357 	struct exp_visible *vp1, *vp2;
358 	struct exp_visible *prev, *next;
359 
360 	/*
361 	 * The outer loop traverses the supplied list.
362 	 */
363 	for (vp1 = vis_head; vp1; vp1 = vp1->vis_next) {
364 
365 		/*
366 		 * Given an element from the list to be removed,
367 		 * search the exportinfo list looking for a match.
368 		 * If a match is found, decrement the reference
369 		 * count and drop the element if the count drops
370 		 * to zero.
371 		 */
372 		for (vp2 = exi->exi_visible, prev = NULL; vp2; vp2 = next) {
373 
374 			next = vp2->vis_next;
375 
376 			if (EQFID(&vp1->vis_fid, &vp2->vis_fid)) {
377 
378 				/*
379 				 * Decrement the ref count.
380 				 * Remove the entry if it's zero.
381 				 */
382 				if (--vp2->vis_count <= 0) {
383 					if (prev == NULL)
384 						exi->exi_visible = next;
385 					else
386 						prev->vis_next = next;
387 
388 					VN_RELE(vp2->vis_vp);
389 					srv_secinfo_list_free(vp2->vis_secinfo,
390 					    vp2->vis_seccnt);
391 					kmem_free(vp2, sizeof (*vp1));
392 				} else {
393 					/*
394 					 * If we're here, then the vp2 will
395 					 * remain in the vis list.  If the
396 					 * vis entry corresponds to the object
397 					 * being unshared, then vis_exported
398 					 * needs to be set to 0.
399 					 *
400 					 * vp1 is a node from caller's list
401 					 * vp2 is node from exportinfo's list
402 					 *
403 					 * Only 1 node in the caller's list
404 					 * will have vis_exported set to 1,
405 					 * and it corresponds to the obj being
406 					 * unshared.  It should always be the
407 					 * last element of the caller's list.
408 					 */
409 					if (vp1->vis_exported &&
410 					    vp2->vis_exported) {
411 						vp2->vis_exported = 0;
412 					}
413 				}
414 
415 				break;
416 			}
417 
418 			prev = vp2;
419 		}
420 	}
421 
422 	free_visible(vis_head);
423 }
424 
425 /*
426  * This function checks the path to a new export to
427  * check whether all the pathname components are
428  * exported. It works by climbing the file tree one
429  * component at a time via "..", crossing mountpoints
430  * if necessary until an export entry is found, or the
431  * system root is reached.
432  *
433  * If an unexported mountpoint is found, then
434  * a new pseudo export is added and the pathname from
435  * the mountpoint down to the export is added to the
436  * visible list for the new pseudo export.  If an existing
437  * pseudo export is found, then the pathname is added
438  * to its visible list.
439  *
440  * Note that there's some tests for exportdir.
441  * The exportinfo entry that's passed as a parameter
442  * is that of the real export and exportdir is set
443  * for this case.
444  *
445  * Here is an example of a possible setup:
446  *
447  * () - a new fs; fs mount point
448  * EXPORT - a real exported node
449  * PSEUDO - a pseudo node
450  * vis - visible list
451  * f# - security flavor#
452  * (f#) - security flavor# propagated from its decendents
453  * "" - covered vnode
454  *
455  *
456  *                 /
457  *                 |
458  *                 (a) PSEUDO (f1,f2)
459  *                 |   vis: b,b,"c","n"
460  *                 |
461  *                 b
462  *        ---------|------------------
463  *        |                          |
464  *        (c) EXPORT,f1(f2)          (n) PSEUDO (f1,f2)
465  *        |   vis: "e","d"           |   vis: m,m,,p,q,"o"
466  *        |                          |
467  *  ------------------          -------------------
468  *  |        |        |         |                  |
469  *  (d)      (e)      f         m EXPORT,f1(f2)    p
470  *  EXPORT   EXPORT             |                  |
471  *  f1       f2                 |                  |
472  *           |                  |                  |
473  *           j                 (o) EXPORT,f2       q EXPORT f2
474  *
475  */
476 int
477 treeclimb_export(struct exportinfo *exip)
478 {
479 	vnode_t *dvp, *vp;
480 	fid_t fid;
481 	int error;
482 	int exportdir;
483 	struct exportinfo *exi = NULL;
484 	struct exp_visible *visp;
485 	struct exp_visible *vis_head = NULL;
486 	struct vattr va;
487 
488 	ASSERT(RW_WRITE_HELD(&exported_lock));
489 
490 	vp = exip->exi_vp;
491 	VN_HOLD(vp);
492 	exportdir = 1;
493 
494 	for (;;) {
495 
496 		bzero(&fid, sizeof (fid));
497 		fid.fid_len = MAXFIDSZ;
498 		error = vop_fid_pseudo(vp, &fid);
499 		if (error)
500 			break;
501 
502 		if (! exportdir) {
503 			/*
504 			 * Check if this exportroot is a VROOT dir.  If so,
505 			 * then attach the pseudonodes.  If not, then
506 			 * continue .. traversal until we hit a VROOT
507 			 * export (pseudo or real).
508 			 */
509 			exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
510 			if (exi != NULL && vp->v_flag & VROOT) {
511 				/*
512 				 * Found an export info
513 				 *
514 				 * Extend the list of visible
515 				 * directories whether it's a pseudo
516 				 * or a real export.
517 				 */
518 				more_visible(exi, vis_head);
519 				vis_head = NULL;
520 				break;	/* and climb no further */
521 			}
522 		}
523 
524 		/*
525 		 * If at the root of the filesystem, need
526 		 * to traverse across the mountpoint
527 		 * and continue the climb on the mounted-on
528 		 * filesystem.
529 		 */
530 		if (vp->v_flag & VROOT) {
531 
532 			if (! exportdir) {
533 				/*
534 				 * Found the root directory of a filesystem
535 				 * that isn't exported.  Need to export
536 				 * this as a pseudo export so that an NFS v4
537 				 * client can do lookups in it.
538 				 */
539 				error = pseudo_exportfs(vp, vis_head, NULL);
540 				if (error)
541 					break;
542 				vis_head = NULL;
543 			}
544 
545 			if (VN_CMP(vp, rootdir)) {
546 				/* at system root */
547 				break;
548 			}
549 
550 			vp = untraverse(vp);
551 			exportdir = 0;
552 			continue;
553 		}
554 
555 		/*
556 		 * Do a getattr to obtain the nodeid (inode num)
557 		 * for this vnode.
558 		 */
559 		va.va_mask = AT_NODEID;
560 		error = VOP_GETATTR(vp, &va, 0, CRED());
561 		if (error)
562 			break;
563 
564 		/*
565 		 *  Add this directory fid to visible list
566 		 */
567 		visp = kmem_alloc(sizeof (*visp), KM_SLEEP);
568 		VN_HOLD(vp);
569 		visp->vis_vp = vp;
570 		visp->vis_fid = fid;		/* structure copy */
571 		visp->vis_ino = va.va_nodeid;
572 		visp->vis_count = 1;
573 		visp->vis_exported = exportdir;
574 		visp->vis_secinfo = NULL;
575 		visp->vis_seccnt = 0;
576 		visp->vis_next = vis_head;
577 		vis_head = visp;
578 
579 		/*
580 		 * Now, do a ".." to find parent dir of vp.
581 		 */
582 		error = VOP_LOOKUP(vp, "..", &dvp, NULL, 0, NULL, CRED());
583 
584 		if (error == ENOTDIR && exportdir) {
585 			dvp = exip->exi_dvp;
586 			ASSERT(dvp != NULL);
587 			VN_HOLD(dvp);
588 			error = 0;
589 		}
590 
591 		if (error)
592 			break;
593 
594 		exportdir = 0;
595 		VN_RELE(vp);
596 		vp = dvp;
597 	}
598 
599 	VN_RELE(vp);
600 	return (error);
601 }
602 
603 /*
604  * Walk up the tree looking for pseudo export entries.
605  *
606  * If a pseudo export is found, remove the path we've
607  * climbed from its visible list. If the visible list
608  * still has entries after the removal, then we can stop.
609  * If it becomes null, then remove the pseudo export entry
610  * and carry on up the tree to see if there's any more.
611  */
612 int
613 treeclimb_unexport(struct exportinfo *exip)
614 {
615 	vnode_t *dvp, *vp;
616 	fid_t fid;
617 	int error = 0;
618 	int exportdir;
619 	struct exportinfo *exi = NULL;
620 	struct exp_visible *vis_head = NULL, *visp;
621 
622 	ASSERT(RW_WRITE_HELD(&exported_lock));
623 
624 	exportdir = 1;
625 	vp = exip->exi_vp;
626 	VN_HOLD(vp);
627 
628 	for (;;) {
629 
630 		bzero(&fid, sizeof (fid));
631 		fid.fid_len = MAXFIDSZ;
632 		error = vop_fid_pseudo(vp, &fid);
633 		if (error)
634 			break;
635 
636 		if (! exportdir) {
637 
638 			/*
639 			 * We need to use checkexport4() here because it
640 			 * doesn't acquire exported_lock and it doesn't
641 			 * manipulate exi_count.
642 			 *
643 			 * Remove directories from the visible
644 			 * list that are unique to the path
645 			 * for this export.  (Only VROOT exportinfos
646 			 * have can have visible entries).
647 			 */
648 			exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
649 			if (exi != NULL && (vp->v_flag & VROOT)) {
650 
651 				less_visible(exi, vis_head);
652 				vis_head = NULL;
653 
654 				/*
655 				 * If the visible list has entries
656 				 * or if it's a real export, then
657 				 * there's no need to keep climbing.
658 				 */
659 				if (exi->exi_visible || ! PSEUDO(exi))
660 					break;
661 
662 				/*
663 				 * Otherwise, we have a pseudo export
664 				 * with an empty list (no exports below
665 				 * it) so we must remove and continue
666 				 * the climb to remove its name from
667 				 * the parent export.
668 				 */
669 				error = export_unlink(&vp->v_vfsp->vfs_fsid,
670 				    &fid, vp, NULL);
671 				if (error)
672 					break;
673 
674 				exi_rele(exi);
675 			}
676 		}
677 
678 		/*
679 		 * If at the root of the filesystem, need
680 		 * to traverse across the mountpoint
681 		 * and continue the climb on the mounted-on
682 		 * filesystem.
683 		 */
684 		if (vp->v_flag & VROOT) {
685 			if (VN_CMP(vp, rootdir)) {
686 				/* at system root */
687 				break;
688 			}
689 			vp = untraverse(vp);
690 			exportdir = 0;
691 			continue;
692 		}
693 
694 		/*
695 		 *  Add this directory fid to path list
696 		 */
697 		visp = kmem_alloc(sizeof (*visp), KM_SLEEP);
698 		VN_HOLD(vp);
699 		visp->vis_vp = vp;
700 		visp->vis_fid = fid;		/* structure copy */
701 		visp->vis_ino = 0;
702 		visp->vis_count = 1;
703 		visp->vis_exported = exportdir;
704 		visp->vis_secinfo = NULL;
705 		visp->vis_seccnt = 0;
706 		visp->vis_next = vis_head;
707 		vis_head = visp;
708 
709 		/*
710 		 * Do a ".." to find parent dir of vp.
711 		 */
712 		error = VOP_LOOKUP(vp, "..", &dvp, NULL, 0, NULL, CRED());
713 
714 		if (error == ENOTDIR && exportdir) {
715 			dvp = exip->exi_dvp;
716 			ASSERT(dvp != NULL);
717 			VN_HOLD(dvp);
718 			error = 0;
719 		}
720 		if (error)
721 			break;
722 
723 		exportdir = 0;
724 		VN_RELE(vp);
725 		vp = dvp;
726 	}
727 
728 	VN_RELE(vp);
729 	return (error);
730 }
731 
732 
733 /*
734  * Traverse backward across mountpoint from the
735  * root vnode of a filesystem to its mounted-on
736  * vnode.
737  */
738 vnode_t *
739 untraverse(vnode_t *vp)
740 {
741 	vnode_t *tvp, *nextvp;
742 
743 	tvp = vp;
744 	for (;;) {
745 		if (! (tvp->v_flag & VROOT))
746 			break;
747 
748 		/* lock vfs to prevent unmount of this vfs */
749 		vfs_lock_wait(tvp->v_vfsp);
750 
751 		if ((nextvp = tvp->v_vfsp->vfs_vnodecovered) == NULL) {
752 			vfs_unlock(tvp->v_vfsp);
753 			break;
754 		}
755 
756 		/*
757 		 * Hold nextvp to prevent unmount.  After unlock vfs and
758 		 * rele tvp, any number of overlays could be unmounted.
759 		 * Putting a hold on vfs_vnodecovered will only allow
760 		 * tvp's vfs to be unmounted. Of course if caller placed
761 		 * extra hold on vp before calling untraverse, the following
762 		 * hold would not be needed.  Since prev actions of caller
763 		 * are unknown, we need to hold here just to be safe.
764 		 */
765 		VN_HOLD(nextvp);
766 		vfs_unlock(tvp->v_vfsp);
767 		VN_RELE(tvp);
768 		tvp = nextvp;
769 	}
770 
771 	return (tvp);
772 }
773 
774 /*
775  * Given an exportinfo, climb up to find the exportinfo for the VROOT
776  * of the filesystem.
777  *
778  * e.g.         /
779  *              |
780  *              a (VROOT) pseudo-exportinfo
781  *		|
782  *		b
783  *		|
784  *		c  #share /a/b/c
785  *		|
786  *		d
787  *
788  * where c is in the same filesystem as a.
789  * So, get_root_export(*exportinfo_for_c) returns exportinfo_for_a
790  *
791  * If d is shared, then c will be put into a's visible list.
792  * Note: visible list is per filesystem and is attached to the
793  * VROOT exportinfo.
794  */
795 struct exportinfo *
796 get_root_export(struct exportinfo *exip)
797 {
798 	vnode_t *dvp, *vp;
799 	fid_t fid;
800 	struct exportinfo *exi = exip;
801 	int error;
802 
803 	vp = exi->exi_vp;
804 	VN_HOLD(vp);
805 
806 	for (;;) {
807 
808 		if (vp->v_flag & VROOT) {
809 			ASSERT(exi != NULL);
810 			break;
811 		}
812 
813 		/*
814 		 * Now, do a ".." to find parent dir of vp.
815 		 */
816 		error = VOP_LOOKUP(vp, "..", &dvp, NULL, 0, NULL, CRED());
817 
818 		if (error) {
819 			exi = NULL;
820 			break;
821 		}
822 
823 		VN_RELE(vp);
824 		vp = dvp;
825 
826 		bzero(&fid, sizeof (fid));
827 		fid.fid_len = MAXFIDSZ;
828 		error = vop_fid_pseudo(vp, &fid);
829 		if (error) {
830 			exi = NULL;
831 			break;
832 		}
833 
834 		exi = checkexport4(&vp->v_vfsp->vfs_fsid, &fid, vp);
835 	}
836 
837 	VN_RELE(vp);
838 	return (exi);
839 }
840 
841 /*
842  * Return true if the supplied vnode has a sub-directory exported.
843  */
844 int
845 has_visible(struct exportinfo *exi, vnode_t *vp)
846 {
847 	struct exp_visible *visp;
848 	fid_t fid;
849 	bool_t vp_is_exported;
850 
851 	vp_is_exported = VN_CMP(vp,  exi->exi_vp);
852 
853 	/*
854 	 * An exported root vnode has a sub-dir shared if it has a visible list.
855 	 * i.e. if it does not have a visible list, then there is no node in
856 	 * this filesystem leads to any other shared node.
857 	 */
858 	if (vp_is_exported && (vp->v_flag & VROOT))
859 		return (exi->exi_visible ? 1 : 0);
860 
861 	/*
862 	 * Only the exportinfo of a fs root node may have a visible list.
863 	 * Either it is a pseudo root node, or a real exported root node.
864 	 */
865 	if ((exi = get_root_export(exi)) == NULL) {
866 		return (0);
867 	}
868 
869 	if (!exi->exi_visible)
870 		return (0);
871 
872 	/* Get the fid of the vnode */
873 	bzero(&fid, sizeof (fid));
874 	fid.fid_len = MAXFIDSZ;
875 	if (vop_fid_pseudo(vp, &fid) != 0) {
876 		return (0);
877 	}
878 
879 	/*
880 	 * See if vp is in the visible list of the root node exportinfo.
881 	 */
882 	for (visp = exi->exi_visible; visp; visp = visp->vis_next) {
883 		if (EQFID(&fid, &visp->vis_fid)) {
884 			/*
885 			 * If vp is an exported non-root node with only 1 path
886 			 * count (for itself), it indicates no sub-dir shared
887 			 * using this vp as a path.
888 			 */
889 			if (vp_is_exported && visp->vis_count < 2)
890 				break;
891 
892 			return (1);
893 		}
894 	}
895 
896 	return (0);
897 }
898 
899 /*
900  * Returns true if the supplied vnode is visible
901  * in this export.  If vnode is visible, return
902  * vis_exported in expseudo.
903  */
904 int
905 nfs_visible(struct exportinfo *exi, vnode_t *vp, int *expseudo)
906 {
907 	struct exp_visible *visp;
908 	fid_t fid;
909 
910 	/*
911 	 * First check to see if vp is export root.
912 	 *
913 	 * A pseudo export root can never be exported
914 	 * (it would be a real export then); however,
915 	 * it is always visible.  If a pseudo root object
916 	 * was exported by server admin, then the entire
917 	 * pseudo exportinfo (and all visible entries) would
918 	 * be destroyed.  A pseudo exportinfo only exists
919 	 * to provide access to real (descendant) export(s).
920 	 *
921 	 * Previously, rootdir was special cased here; however,
922 	 * the export root special case handles the rootdir
923 	 * case also.
924 	 */
925 	if (VN_CMP(vp, exi->exi_vp)) {
926 		*expseudo = 0;
927 		return (1);
928 	}
929 
930 	/*
931 	 * Only a PSEUDO node has a visible list or an exported VROOT
932 	 * node may have a visible list.
933 	 */
934 	if (! PSEUDO(exi) && (exi = get_root_export(exi)) == NULL) {
935 		*expseudo = 0;
936 		return (0);
937 	}
938 
939 	/* Get the fid of the vnode */
940 
941 	bzero(&fid, sizeof (fid));
942 	fid.fid_len = MAXFIDSZ;
943 	if (vop_fid_pseudo(vp, &fid) != 0) {
944 		*expseudo = 0;
945 		return (0);
946 	}
947 
948 	/*
949 	 * We can't trust VN_CMP() above because of LOFS.
950 	 * Even though VOP_CMP will do the right thing for LOFS
951 	 * objects, VN_CMP will short circuit out early when the
952 	 * vnode ops ptrs are different.  Just in case we're dealing
953 	 * with LOFS, compare exi_fid/fsid here.
954 	 *
955 	 * expseudo is not set because this is not an export
956 	 */
957 	if (EQFID(&exi->exi_fid, &fid) &&
958 	    EQFSID(&exi->exi_fsid, &vp->v_vfsp->vfs_fsid)) {
959 		*expseudo = 0;
960 		return (1);
961 	}
962 
963 
964 	/* See if it matches any fid in the visible list */
965 
966 	for (visp = exi->exi_visible; visp; visp = visp->vis_next) {
967 		if (EQFID(&fid, &visp->vis_fid)) {
968 			*expseudo = visp->vis_exported;
969 			return (1);
970 		}
971 	}
972 
973 	*expseudo = 0;
974 
975 	return (0);
976 }
977 
978 /*
979  * Returns true if the supplied vnode is the
980  * directory of an export point.
981  */
982 int
983 nfs_exported(struct exportinfo *exi, vnode_t *vp)
984 {
985 	struct exp_visible *visp;
986 	fid_t fid;
987 
988 	/*
989 	 * First check to see if vp is the export root
990 	 * This check required for the case of lookup ..
991 	 * where .. is a V_ROOT vnode and a pseudo exportroot.
992 	 * Pseudo export root objects do not have an entry
993 	 * in the visible list even though every V_ROOT
994 	 * pseudonode is visible.  It is safe to compare
995 	 * vp here because pseudo_exportfs put a hold on
996 	 * it when exi_vp was initialized.
997 	 *
998 	 * Note: VN_CMP() won't match for LOFS shares, but they're
999 	 * handled below w/EQFID/EQFSID.
1000 	 */
1001 	if (VN_CMP(vp, exi->exi_vp))
1002 		return (1);
1003 
1004 	/* Get the fid of the vnode */
1005 
1006 	bzero(&fid, sizeof (fid));
1007 	fid.fid_len = MAXFIDSZ;
1008 	if (vop_fid_pseudo(vp, &fid) != 0)
1009 		return (0);
1010 
1011 	if (EQFID(&fid, &exi->exi_fid) &&
1012 	    EQFSID(&vp->v_vfsp->vfs_fsid, &exi->exi_fsid)) {
1013 		return (1);
1014 	}
1015 
1016 	/* See if it matches any fid in the visible list */
1017 
1018 	for (visp = exi->exi_visible; visp; visp = visp->vis_next) {
1019 		if (EQFID(&fid, &visp->vis_fid))
1020 			return (visp->vis_exported);
1021 	}
1022 
1023 	return (0);
1024 }
1025 
1026 /*
1027  * Returns true if the supplied inode is visible
1028  * in this export.  This function is used by
1029  * readdir which uses inode numbers from the
1030  * directory.
1031  *
1032  * NOTE: this code does not match inode number for ".",
1033  * but it isn't required because NFS4 server rddir
1034  * skips . and .. entries.
1035  */
1036 int
1037 nfs_visible_inode(struct exportinfo *exi, ino64_t ino, int *expseudo)
1038 {
1039 	struct exp_visible *visp;
1040 
1041 	/*
1042 	 * Only a PSEUDO node has a visible list or an exported VROOT
1043 	 * node may have a visible list.
1044 	 */
1045 	if (! PSEUDO(exi) && (exi = get_root_export(exi)) == NULL) {
1046 		*expseudo = 0;
1047 		return (0);
1048 	}
1049 
1050 	for (visp = exi->exi_visible; visp; visp = visp->vis_next)
1051 		if ((u_longlong_t)ino == visp->vis_ino) {
1052 			*expseudo = visp->vis_exported;
1053 			return (1);
1054 		}
1055 
1056 	*expseudo = 0;
1057 	return (0);
1058 }
1059