xref: /freebsd/sys/fs/unionfs/union_subr.c (revision 9768746ba83efa02837c5b9c66348db6e900208f)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1994 Jan-Simon Pendry
5  * Copyright (c) 1994
6  *	The Regents of the University of California.  All rights reserved.
7  * Copyright (c) 2005, 2006, 2012 Masanori Ozawa <ozawa@ongs.co.jp>, ONGS Inc.
8  * Copyright (c) 2006, 2012 Daichi Goto <daichi@freebsd.org>
9  *
10  * This code is derived from software contributed to Berkeley by
11  * Jan-Simon Pendry.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
16  * 1. Redistributions of source code must retain the above copyright
17  *    notice, this list of conditions and the following disclaimer.
18  * 2. Redistributions in binary form must reproduce the above copyright
19  *    notice, this list of conditions and the following disclaimer in the
20  *    documentation and/or other materials provided with the distribution.
21  * 3. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)union_subr.c	8.20 (Berkeley) 5/20/95
38  * $FreeBSD$
39  */
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/ktr.h>
45 #include <sys/lock.h>
46 #include <sys/mutex.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
49 #include <sys/namei.h>
50 #include <sys/proc.h>
51 #include <sys/vnode.h>
52 #include <sys/dirent.h>
53 #include <sys/fcntl.h>
54 #include <sys/filedesc.h>
55 #include <sys/stat.h>
56 #include <sys/sysctl.h>
57 #include <sys/taskqueue.h>
58 #include <sys/resourcevar.h>
59 
60 #include <machine/atomic.h>
61 
62 #include <security/mac/mac_framework.h>
63 
64 #include <vm/uma.h>
65 
66 #include <fs/unionfs/union.h>
67 
68 #define NUNIONFSNODECACHE 16
69 #define UNIONFSHASHMASK (NUNIONFSNODECACHE - 1)
70 
71 static MALLOC_DEFINE(M_UNIONFSHASH, "UNIONFS hash", "UNIONFS hash table");
72 MALLOC_DEFINE(M_UNIONFSNODE, "UNIONFS node", "UNIONFS vnode private part");
73 MALLOC_DEFINE(M_UNIONFSPATH, "UNIONFS path", "UNIONFS path private part");
74 
75 static struct task unionfs_deferred_rele_task;
76 static struct mtx unionfs_deferred_rele_lock;
77 static STAILQ_HEAD(, unionfs_node) unionfs_deferred_rele_list =
78     STAILQ_HEAD_INITIALIZER(unionfs_deferred_rele_list);
79 static TASKQUEUE_DEFINE_THREAD(unionfs_rele);
80 
81 unsigned int unionfs_ndeferred = 0;
82 SYSCTL_UINT(_vfs, OID_AUTO, unionfs_ndeferred, CTLFLAG_RD,
83     &unionfs_ndeferred, 0, "unionfs deferred vnode release");
84 
85 static void unionfs_deferred_rele(void *, int);
86 
87 /*
88  * Initialize
89  */
90 int
91 unionfs_init(struct vfsconf *vfsp)
92 {
93 	UNIONFSDEBUG("unionfs_init\n");	/* printed during system boot */
94 	TASK_INIT(&unionfs_deferred_rele_task, 0, unionfs_deferred_rele, NULL);
95 	mtx_init(&unionfs_deferred_rele_lock, "uniondefr", NULL, MTX_DEF);
96 	return (0);
97 }
98 
99 /*
100  * Uninitialize
101  */
102 int
103 unionfs_uninit(struct vfsconf *vfsp)
104 {
105 	taskqueue_quiesce(taskqueue_unionfs_rele);
106 	taskqueue_free(taskqueue_unionfs_rele);
107 	mtx_destroy(&unionfs_deferred_rele_lock);
108 	return (0);
109 }
110 
111 static void
112 unionfs_deferred_rele(void *arg __unused, int pending __unused)
113 {
114 	STAILQ_HEAD(, unionfs_node) local_rele_list;
115 	struct unionfs_node *unp, *tunp;
116 	unsigned int ndeferred;
117 
118 	ndeferred = 0;
119 	STAILQ_INIT(&local_rele_list);
120 	mtx_lock(&unionfs_deferred_rele_lock);
121 	STAILQ_CONCAT(&local_rele_list, &unionfs_deferred_rele_list);
122 	mtx_unlock(&unionfs_deferred_rele_lock);
123 	STAILQ_FOREACH_SAFE(unp, &local_rele_list, un_rele, tunp) {
124 		++ndeferred;
125 		MPASS(unp->un_dvp != NULL);
126 		vrele(unp->un_dvp);
127 		free(unp, M_UNIONFSNODE);
128 	}
129 
130 	/* We expect this function to be single-threaded, thus no atomic */
131 	unionfs_ndeferred += ndeferred;
132 }
133 
134 static struct unionfs_node_hashhead *
135 unionfs_get_hashhead(struct vnode *dvp, struct vnode *lookup)
136 {
137 	struct unionfs_node *unp;
138 
139 	unp = VTOUNIONFS(dvp);
140 
141 	return (&(unp->un_hashtbl[vfs_hash_index(lookup) & UNIONFSHASHMASK]));
142 }
143 
144 /*
145  * Attempt to lookup a cached unionfs vnode by upper/lower vp
146  * from dvp, with dvp's interlock held.
147  */
148 static struct vnode *
149 unionfs_get_cached_vnode_locked(struct vnode *lookup, struct vnode *dvp)
150 {
151 	struct unionfs_node *unp;
152 	struct unionfs_node_hashhead *hd;
153 	struct vnode *vp;
154 
155 	hd = unionfs_get_hashhead(dvp, lookup);
156 
157 	LIST_FOREACH(unp, hd, un_hash) {
158 		if (unp->un_uppervp == lookup ||
159 		    unp->un_lowervp == lookup) {
160 			vp = UNIONFSTOV(unp);
161 			VI_LOCK_FLAGS(vp, MTX_DUPOK);
162 			vp->v_iflag &= ~VI_OWEINACT;
163 			if (VN_IS_DOOMED(vp) ||
164 			    ((vp->v_iflag & VI_DOINGINACT) != 0)) {
165 				VI_UNLOCK(vp);
166 				vp = NULLVP;
167 			} else {
168 				vrefl(vp);
169 				VI_UNLOCK(vp);
170 			}
171 			return (vp);
172 		}
173 	}
174 
175 	return (NULLVP);
176 }
177 
178 
179 /*
180  * Get the cached vnode.
181  */
182 static struct vnode *
183 unionfs_get_cached_vnode(struct vnode *uvp, struct vnode *lvp,
184     struct vnode *dvp)
185 {
186 	struct vnode *vp;
187 
188 	vp = NULLVP;
189 	VI_LOCK(dvp);
190 	if (uvp != NULLVP)
191 		vp = unionfs_get_cached_vnode_locked(uvp, dvp);
192 	else if (lvp != NULLVP)
193 		vp = unionfs_get_cached_vnode_locked(lvp, dvp);
194 	VI_UNLOCK(dvp);
195 
196 	return (vp);
197 }
198 
199 /*
200  * Add the new vnode into cache.
201  */
202 static struct vnode *
203 unionfs_ins_cached_vnode(struct unionfs_node *uncp,
204     struct vnode *dvp)
205 {
206 	struct unionfs_node_hashhead *hd;
207 	struct vnode *vp;
208 
209 	ASSERT_VOP_ELOCKED(uncp->un_uppervp, __func__);
210 	ASSERT_VOP_ELOCKED(uncp->un_lowervp, __func__);
211 	KASSERT(uncp->un_uppervp == NULLVP || uncp->un_uppervp->v_type == VDIR,
212 	    ("%s: v_type != VDIR", __func__));
213 	KASSERT(uncp->un_lowervp == NULLVP || uncp->un_lowervp->v_type == VDIR,
214 	    ("%s: v_type != VDIR", __func__));
215 
216 	vp = NULLVP;
217 	VI_LOCK(dvp);
218 	if (uncp->un_uppervp != NULL)
219 		vp = unionfs_get_cached_vnode_locked(uncp->un_uppervp, dvp);
220 	else if (uncp->un_lowervp != NULL)
221 		vp = unionfs_get_cached_vnode_locked(uncp->un_lowervp, dvp);
222 	if (vp == NULLVP) {
223 		hd = unionfs_get_hashhead(dvp, (uncp->un_uppervp != NULLVP ?
224 		    uncp->un_uppervp : uncp->un_lowervp));
225 		LIST_INSERT_HEAD(hd, uncp, un_hash);
226 	}
227 	VI_UNLOCK(dvp);
228 
229 	return (vp);
230 }
231 
232 /*
233  * Remove the vnode.
234  */
235 static void
236 unionfs_rem_cached_vnode(struct unionfs_node *unp, struct vnode *dvp)
237 {
238 	KASSERT(unp != NULL, ("%s: null node", __func__));
239 	KASSERT(dvp != NULLVP,
240 	    ("%s: null parent vnode", __func__));
241 
242 	VI_LOCK(dvp);
243 	if (unp->un_hash.le_prev != NULL) {
244 		LIST_REMOVE(unp, un_hash);
245 		unp->un_hash.le_next = NULL;
246 		unp->un_hash.le_prev = NULL;
247 	}
248 	VI_UNLOCK(dvp);
249 }
250 
251 /*
252  * Common cleanup handling for unionfs_nodeget
253  * Upper, lower, and parent directory vnodes are expected to be referenced by
254  * the caller.  Upper and lower vnodes, if non-NULL, are also expected to be
255  * exclusively locked by the caller.
256  * This function will return with the caller's locks and references undone.
257  */
258 static void
259 unionfs_nodeget_cleanup(struct vnode *vp, struct unionfs_node *unp)
260 {
261 
262 	/*
263 	 * Lock and reset the default vnode lock; vgone() expects a locked
264 	 * vnode, and we're going to reset the vnode ops.
265 	 */
266 	lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
267 
268 	/*
269 	 * Clear out private data and reset the vnode ops to avoid use of
270 	 * unionfs vnode ops on a partially constructed vnode.
271 	 */
272 	VI_LOCK(vp);
273 	vp->v_data = NULL;
274 	vp->v_vnlock = &vp->v_lock;
275 	vp->v_op = &dead_vnodeops;
276 	VI_UNLOCK(vp);
277 	vgone(vp);
278 	vput(vp);
279 
280 	if (unp->un_dvp != NULLVP)
281 		vrele(unp->un_dvp);
282 	if (unp->un_uppervp != NULLVP)
283 		vput(unp->un_uppervp);
284 	if (unp->un_lowervp != NULLVP)
285 		vput(unp->un_lowervp);
286 	if (unp->un_hashtbl != NULL)
287 		hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, UNIONFSHASHMASK);
288 	free(unp->un_path, M_UNIONFSPATH);
289 	free(unp, M_UNIONFSNODE);
290 }
291 
292 /*
293  * Make a new or get existing unionfs node.
294  *
295  * uppervp and lowervp should be unlocked. Because if new unionfs vnode is
296  * locked, uppervp or lowervp is locked too. In order to prevent dead lock,
297  * you should not lock plurality simultaneously.
298  */
299 int
300 unionfs_nodeget(struct mount *mp, struct vnode *uppervp,
301     struct vnode *lowervp, struct vnode *dvp, struct vnode **vpp,
302     struct componentname *cnp)
303 {
304 	char	       *path;
305 	struct unionfs_mount *ump;
306 	struct unionfs_node *unp;
307 	struct vnode   *vp;
308 	u_long		hashmask;
309 	int		error;
310 	int		lkflags;
311 	enum vtype	vt;
312 
313 	error = 0;
314 	ump = MOUNTTOUNIONFSMOUNT(mp);
315 	lkflags = (cnp ? cnp->cn_lkflags : 0);
316 	path = (cnp ? cnp->cn_nameptr : NULL);
317 	*vpp = NULLVP;
318 
319 	if (uppervp == NULLVP && lowervp == NULLVP)
320 		panic("%s: upper and lower is null", __func__);
321 
322 	vt = (uppervp != NULLVP ? uppervp->v_type : lowervp->v_type);
323 
324 	/* If it has no ISLASTCN flag, path check is skipped. */
325 	if (cnp && !(cnp->cn_flags & ISLASTCN))
326 		path = NULL;
327 
328 	/* check the cache */
329 	if (dvp != NULLVP && vt == VDIR) {
330 		vp = unionfs_get_cached_vnode(uppervp, lowervp, dvp);
331 		if (vp != NULLVP) {
332 			*vpp = vp;
333 			goto unionfs_nodeget_out;
334 		}
335 	}
336 
337 	unp = malloc(sizeof(struct unionfs_node),
338 	    M_UNIONFSNODE, M_WAITOK | M_ZERO);
339 
340 	error = getnewvnode("unionfs", mp, &unionfs_vnodeops, &vp);
341 	if (error != 0) {
342 		free(unp, M_UNIONFSNODE);
343 		return (error);
344 	}
345 	if (dvp != NULLVP)
346 		vref(dvp);
347 	if (uppervp != NULLVP)
348 		vref(uppervp);
349 	if (lowervp != NULLVP)
350 		vref(lowervp);
351 
352 	if (vt == VDIR) {
353 		unp->un_hashtbl = hashinit(NUNIONFSNODECACHE, M_UNIONFSHASH,
354 		    &hashmask);
355 		KASSERT(hashmask == UNIONFSHASHMASK,
356 		    ("unexpected unionfs hash mask 0x%lx", hashmask));
357 	}
358 
359 	unp->un_vnode = vp;
360 	unp->un_uppervp = uppervp;
361 	unp->un_lowervp = lowervp;
362 	unp->un_dvp = dvp;
363 	if (uppervp != NULLVP)
364 		vp->v_vnlock = uppervp->v_vnlock;
365 	else
366 		vp->v_vnlock = lowervp->v_vnlock;
367 
368 	if (path != NULL) {
369 		unp->un_path = malloc(cnp->cn_namelen + 1,
370 		    M_UNIONFSPATH, M_WAITOK | M_ZERO);
371 		bcopy(cnp->cn_nameptr, unp->un_path, cnp->cn_namelen);
372 		unp->un_path[cnp->cn_namelen] = '\0';
373 		unp->un_pathlen = cnp->cn_namelen;
374 	}
375 	vp->v_type = vt;
376 	vp->v_data = unp;
377 
378 	/*
379 	 * TODO: This is an imperfect check, as there's no guarantee that
380 	 * the underlying filesystems will always return vnode pointers
381 	 * for the root inodes that match our cached values.  To reduce
382 	 * the likelihood of failure, for example in the case where either
383 	 * vnode has been forcibly doomed, we check both pointers and set
384 	 * VV_ROOT if either matches.
385 	 */
386 	if (ump->um_uppervp == uppervp || ump->um_lowervp == lowervp)
387 		vp->v_vflag |= VV_ROOT;
388 	KASSERT(dvp != NULL || (vp->v_vflag & VV_ROOT) != 0,
389 	    ("%s: NULL dvp for non-root vp %p", __func__, vp));
390 
391 	vn_lock_pair(lowervp, false, uppervp, false);
392 	error = insmntque1(vp, mp);
393 	if (error != 0) {
394 		unionfs_nodeget_cleanup(vp, unp);
395 		return (error);
396 	}
397 	if (lowervp != NULL && VN_IS_DOOMED(lowervp)) {
398 		vput(lowervp);
399 		unp->un_lowervp = NULL;
400 	}
401 	if (uppervp != NULL && VN_IS_DOOMED(uppervp)) {
402 		vput(uppervp);
403 		unp->un_uppervp = NULL;
404 	}
405 	if (unp->un_lowervp == NULL && unp->un_uppervp == NULL) {
406 		unionfs_nodeget_cleanup(vp, unp);
407 		return (ENOENT);
408 	}
409 
410 	vn_set_state(vp, VSTATE_CONSTRUCTED);
411 
412 	if (dvp != NULLVP && vt == VDIR)
413 		*vpp = unionfs_ins_cached_vnode(unp, dvp);
414 	if (*vpp != NULLVP) {
415 		unionfs_nodeget_cleanup(vp, unp);
416 		vp = *vpp;
417 	} else {
418 		if (uppervp != NULL)
419 			VOP_UNLOCK(uppervp);
420 		if (lowervp != NULL)
421 			VOP_UNLOCK(lowervp);
422 		*vpp = vp;
423 	}
424 
425 unionfs_nodeget_out:
426 	if (lkflags & LK_TYPE_MASK)
427 		vn_lock(vp, lkflags | LK_RETRY);
428 
429 	return (0);
430 }
431 
432 /*
433  * Clean up the unionfs node.
434  */
435 void
436 unionfs_noderem(struct vnode *vp)
437 {
438 	struct unionfs_node *unp, *unp_t1, *unp_t2;
439 	struct unionfs_node_hashhead *hd;
440 	struct unionfs_node_status *unsp, *unsp_tmp;
441 	struct vnode   *lvp;
442 	struct vnode   *uvp;
443 	struct vnode   *dvp;
444 	int		count;
445 	int		writerefs;
446 
447 	/*
448 	 * The root vnode lock may be recursed during unmount, because
449 	 * it may share the same lock as the unionfs mount's covered vnode,
450 	 * which is locked across VFS_UNMOUNT().  This lock will then be
451 	 * recursively taken during the vflush() issued by unionfs_unmount().
452 	 * But we still only need to lock the unionfs lock once, because only
453 	 * one of those lock operations was taken against a unionfs vnode and
454 	 * will be undone against a unionfs vnode.
455 	 */
456 	KASSERT(vp->v_vnlock->lk_recurse == 0 || (vp->v_vflag & VV_ROOT) != 0,
457 	    ("%s: vnode %p locked recursively", __func__, vp));
458 	if (lockmgr(&vp->v_lock, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0)
459 		panic("%s: failed to acquire lock for vnode lock", __func__);
460 
461 	/*
462 	 * Use the interlock to protect the clearing of v_data to
463 	 * prevent faults in unionfs_lock().
464 	 */
465 	VI_LOCK(vp);
466 	unp = VTOUNIONFS(vp);
467 	lvp = unp->un_lowervp;
468 	uvp = unp->un_uppervp;
469 	dvp = unp->un_dvp;
470 	unp->un_lowervp = unp->un_uppervp = NULLVP;
471 	vp->v_vnlock = &(vp->v_lock);
472 	vp->v_data = NULL;
473 	vp->v_object = NULL;
474 	if (unp->un_hashtbl != NULL) {
475 		/*
476 		 * Clear out any cached child vnodes.  This should only
477 		 * be necessary during forced unmount, when the vnode may
478 		 * be reclaimed with a non-zero use count.  Otherwise the
479 		 * reference held by each child should prevent reclamation.
480 		 */
481 		for (count = 0; count <= UNIONFSHASHMASK; count++) {
482 			hd = unp->un_hashtbl + count;
483 			LIST_FOREACH_SAFE(unp_t1, hd, un_hash, unp_t2) {
484 				LIST_REMOVE(unp_t1, un_hash);
485 				unp_t1->un_hash.le_next = NULL;
486 				unp_t1->un_hash.le_prev = NULL;
487 			}
488 		}
489 	}
490 	VI_UNLOCK(vp);
491 
492 	writerefs = atomic_load_int(&vp->v_writecount);
493 	VNASSERT(writerefs >= 0, vp,
494 	    ("%s: write count %d, unexpected text ref", __func__, writerefs));
495 	/*
496 	 * If we were opened for write, we leased the write reference
497 	 * to the lower vnode.  If this is a reclamation due to the
498 	 * forced unmount, undo the reference now.
499 	 */
500 	if (writerefs > 0) {
501 		VNASSERT(uvp != NULL, vp,
502 		    ("%s: write reference without upper vnode", __func__));
503 		VOP_ADD_WRITECOUNT(uvp, -writerefs);
504 	}
505 	if (lvp != NULLVP)
506 		VOP_UNLOCK(lvp);
507 	if (uvp != NULLVP)
508 		VOP_UNLOCK(uvp);
509 
510 	if (dvp != NULLVP)
511 		unionfs_rem_cached_vnode(unp, dvp);
512 
513 	if (lvp != NULLVP)
514 		vrele(lvp);
515 	if (uvp != NULLVP)
516 		vrele(uvp);
517 	if (unp->un_path != NULL) {
518 		free(unp->un_path, M_UNIONFSPATH);
519 		unp->un_path = NULL;
520 		unp->un_pathlen = 0;
521 	}
522 
523 	if (unp->un_hashtbl != NULL) {
524 		hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, UNIONFSHASHMASK);
525 	}
526 
527 	LIST_FOREACH_SAFE(unsp, &(unp->un_unshead), uns_list, unsp_tmp) {
528 		LIST_REMOVE(unsp, uns_list);
529 		free(unsp, M_TEMP);
530 	}
531 	if (dvp != NULLVP) {
532 		mtx_lock(&unionfs_deferred_rele_lock);
533 		STAILQ_INSERT_TAIL(&unionfs_deferred_rele_list, unp, un_rele);
534 		mtx_unlock(&unionfs_deferred_rele_lock);
535 		taskqueue_enqueue(taskqueue_unionfs_rele,
536 		    &unionfs_deferred_rele_task);
537 	} else
538 		free(unp, M_UNIONFSNODE);
539 }
540 
541 /*
542  * Get the unionfs node status object for the vnode corresponding to unp,
543  * for the process that owns td.  Allocate a new status object if one
544  * does not already exist.
545  */
546 void
547 unionfs_get_node_status(struct unionfs_node *unp, struct thread *td,
548     struct unionfs_node_status **unspp)
549 {
550 	struct unionfs_node_status *unsp;
551 	pid_t pid;
552 
553 	pid = td->td_proc->p_pid;
554 
555 	KASSERT(NULL != unspp, ("%s: NULL status", __func__));
556 	ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), __func__);
557 
558 	LIST_FOREACH(unsp, &(unp->un_unshead), uns_list) {
559 		if (unsp->uns_pid == pid) {
560 			*unspp = unsp;
561 			return;
562 		}
563 	}
564 
565 	/* create a new unionfs node status */
566 	unsp = malloc(sizeof(struct unionfs_node_status),
567 	    M_TEMP, M_WAITOK | M_ZERO);
568 
569 	unsp->uns_pid = pid;
570 	LIST_INSERT_HEAD(&(unp->un_unshead), unsp, uns_list);
571 
572 	*unspp = unsp;
573 }
574 
575 /*
576  * Remove the unionfs node status, if you can.
577  * You need exclusive lock this vnode.
578  */
579 void
580 unionfs_tryrem_node_status(struct unionfs_node *unp,
581     struct unionfs_node_status *unsp)
582 {
583 	KASSERT(NULL != unsp, ("%s: NULL status", __func__));
584 	ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), __func__);
585 
586 	if (0 < unsp->uns_lower_opencnt || 0 < unsp->uns_upper_opencnt)
587 		return;
588 
589 	LIST_REMOVE(unsp, uns_list);
590 	free(unsp, M_TEMP);
591 }
592 
593 /*
594  * Create upper node attr.
595  */
596 void
597 unionfs_create_uppervattr_core(struct unionfs_mount *ump, struct vattr *lva,
598     struct vattr *uva, struct thread *td)
599 {
600 	VATTR_NULL(uva);
601 	uva->va_type = lva->va_type;
602 	uva->va_atime = lva->va_atime;
603 	uva->va_mtime = lva->va_mtime;
604 	uva->va_ctime = lva->va_ctime;
605 
606 	switch (ump->um_copymode) {
607 	case UNIONFS_TRANSPARENT:
608 		uva->va_mode = lva->va_mode;
609 		uva->va_uid = lva->va_uid;
610 		uva->va_gid = lva->va_gid;
611 		break;
612 	case UNIONFS_MASQUERADE:
613 		if (ump->um_uid == lva->va_uid) {
614 			uva->va_mode = lva->va_mode & 077077;
615 			uva->va_mode |= (lva->va_type == VDIR ?
616 			    ump->um_udir : ump->um_ufile) & 0700;
617 			uva->va_uid = lva->va_uid;
618 			uva->va_gid = lva->va_gid;
619 		} else {
620 			uva->va_mode = (lva->va_type == VDIR ?
621 			    ump->um_udir : ump->um_ufile);
622 			uva->va_uid = ump->um_uid;
623 			uva->va_gid = ump->um_gid;
624 		}
625 		break;
626 	default:		/* UNIONFS_TRADITIONAL */
627 		uva->va_mode = 0777 & ~td->td_proc->p_pd->pd_cmask;
628 		uva->va_uid = ump->um_uid;
629 		uva->va_gid = ump->um_gid;
630 		break;
631 	}
632 }
633 
634 /*
635  * Create upper node attr.
636  */
637 int
638 unionfs_create_uppervattr(struct unionfs_mount *ump, struct vnode *lvp,
639     struct vattr *uva, struct ucred *cred, struct thread *td)
640 {
641 	struct vattr	lva;
642 	int		error;
643 
644 	if ((error = VOP_GETATTR(lvp, &lva, cred)))
645 		return (error);
646 
647 	unionfs_create_uppervattr_core(ump, &lva, uva, td);
648 
649 	return (error);
650 }
651 
652 /*
653  * relookup
654  *
655  * dvp should be locked on entry and will be locked on return.
656  *
657  * If an error is returned, *vpp will be invalid, otherwise it will hold a
658  * locked, referenced vnode. If *vpp == dvp then remember that only one
659  * LK_EXCLUSIVE lock is held.
660  */
661 int
662 unionfs_relookup(struct vnode *dvp, struct vnode **vpp,
663     struct componentname *cnp, struct componentname *cn, struct thread *td,
664     char *path, int pathlen, u_long nameiop)
665 {
666 	int error;
667 	bool refstart;
668 
669 	cn->cn_namelen = pathlen;
670 	cn->cn_pnbuf = path;
671 	cn->cn_nameiop = nameiop;
672 	cn->cn_flags = (LOCKPARENT | LOCKLEAF | ISLASTCN);
673 	cn->cn_lkflags = LK_EXCLUSIVE;
674 	cn->cn_cred = cnp->cn_cred;
675 	cn->cn_nameptr = cn->cn_pnbuf;
676 
677 	refstart = false;
678 	if (nameiop == DELETE) {
679 		cn->cn_flags |= (cnp->cn_flags & DOWHITEOUT);
680 	} else if (nameiop == RENAME) {
681 		refstart = true;
682 	} else if (nameiop == CREATE) {
683 		cn->cn_flags |= NOCACHE;
684 	}
685 
686 	vref(dvp);
687 	VOP_UNLOCK(dvp);
688 
689 	if ((error = vfs_relookup(dvp, vpp, cn, refstart))) {
690 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
691 	} else
692 		vrele(dvp);
693 
694 	KASSERT(cn->cn_pnbuf == path, ("%s: cn_pnbuf changed", __func__));
695 
696 	return (error);
697 }
698 
699 /*
700  * relookup for CREATE namei operation.
701  *
702  * dvp is unionfs vnode. dvp should be locked.
703  *
704  * If it called 'unionfs_copyfile' function by unionfs_link etc,
705  * VOP_LOOKUP information is broken.
706  * So it need relookup in order to create link etc.
707  */
708 int
709 unionfs_relookup_for_create(struct vnode *dvp, struct componentname *cnp,
710     struct thread *td)
711 {
712 	struct vnode *udvp;
713 	struct vnode *vp;
714 	struct componentname cn;
715 	int error;
716 
717 	udvp = UNIONFSVPTOUPPERVP(dvp);
718 	vp = NULLVP;
719 
720 	error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
721 	    cnp->cn_namelen, CREATE);
722 	if (error)
723 		return (error);
724 
725 	if (vp != NULLVP) {
726 		if (udvp == vp)
727 			vrele(vp);
728 		else
729 			vput(vp);
730 
731 		error = EEXIST;
732 	}
733 
734 	return (error);
735 }
736 
737 /*
738  * relookup for DELETE namei operation.
739  *
740  * dvp is unionfs vnode. dvp should be locked.
741  */
742 int
743 unionfs_relookup_for_delete(struct vnode *dvp, struct componentname *cnp,
744     struct thread *td)
745 {
746 	struct vnode *udvp;
747 	struct vnode *vp;
748 	struct componentname cn;
749 	int error;
750 
751 	udvp = UNIONFSVPTOUPPERVP(dvp);
752 	vp = NULLVP;
753 
754 	error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
755 	    cnp->cn_namelen, DELETE);
756 	if (error)
757 		return (error);
758 
759 	if (vp == NULLVP)
760 		error = ENOENT;
761 	else {
762 		if (udvp == vp)
763 			vrele(vp);
764 		else
765 			vput(vp);
766 	}
767 
768 	return (error);
769 }
770 
771 /*
772  * relookup for RENAME namei operation.
773  *
774  * dvp is unionfs vnode. dvp should be locked.
775  */
776 int
777 unionfs_relookup_for_rename(struct vnode *dvp, struct componentname *cnp,
778     struct thread *td)
779 {
780 	struct vnode *udvp;
781 	struct vnode *vp;
782 	struct componentname cn;
783 	int error;
784 
785 	udvp = UNIONFSVPTOUPPERVP(dvp);
786 	vp = NULLVP;
787 
788 	error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
789 	    cnp->cn_namelen, RENAME);
790 	if (error)
791 		return (error);
792 
793 	if (vp != NULLVP) {
794 		if (udvp == vp)
795 			vrele(vp);
796 		else
797 			vput(vp);
798 	}
799 
800 	return (error);
801 }
802 
803 /*
804  * Update the unionfs_node.
805  *
806  * uvp is new locked upper vnode. unionfs vnode's lock will be exchanged to the
807  * uvp's lock and lower's lock will be unlocked.
808  */
809 static void
810 unionfs_node_update(struct unionfs_node *unp, struct vnode *uvp,
811     struct thread *td)
812 {
813 	struct unionfs_node_hashhead *hd;
814 	struct vnode   *vp;
815 	struct vnode   *lvp;
816 	struct vnode   *dvp;
817 	unsigned	count, lockrec;
818 
819 	vp = UNIONFSTOV(unp);
820 	lvp = unp->un_lowervp;
821 	ASSERT_VOP_ELOCKED(lvp, __func__);
822 	ASSERT_VOP_ELOCKED(uvp, __func__);
823 	dvp = unp->un_dvp;
824 
825 	VNASSERT(vp->v_writecount == 0, vp,
826 	    ("%s: non-zero writecount", __func__));
827 	/*
828 	 * Update the upper vnode's lock state to match the lower vnode,
829 	 * and then switch the unionfs vnode's lock to the upper vnode.
830 	 */
831 	lockrec = lvp->v_vnlock->lk_recurse;
832 	for (count = 0; count < lockrec; count++)
833 		vn_lock(uvp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY);
834 	VI_LOCK(vp);
835 	unp->un_uppervp = uvp;
836 	vp->v_vnlock = uvp->v_vnlock;
837 	VI_UNLOCK(vp);
838 
839 	/*
840 	 * Re-cache the unionfs vnode against the upper vnode
841 	 */
842 	if (dvp != NULLVP && vp->v_type == VDIR) {
843 		VI_LOCK(dvp);
844 		if (unp->un_hash.le_prev != NULL) {
845 			LIST_REMOVE(unp, un_hash);
846 			hd = unionfs_get_hashhead(dvp, uvp);
847 			LIST_INSERT_HEAD(hd, unp, un_hash);
848 		}
849 		VI_UNLOCK(unp->un_dvp);
850 	}
851 }
852 
853 /*
854  * Create a new shadow dir.
855  *
856  * udvp should be locked on entry and will be locked on return.
857  *
858  * If no error returned, unp will be updated.
859  */
860 int
861 unionfs_mkshadowdir(struct unionfs_mount *ump, struct vnode *udvp,
862     struct unionfs_node *unp, struct componentname *cnp, struct thread *td)
863 {
864 	struct vnode   *lvp;
865 	struct vnode   *uvp;
866 	struct vattr	va;
867 	struct vattr	lva;
868 	struct nameidata nd;
869 	struct mount   *mp;
870 	struct ucred   *cred;
871 	struct ucred   *credbk;
872 	struct uidinfo *rootinfo;
873 	int		error;
874 
875 	if (unp->un_uppervp != NULLVP)
876 		return (EEXIST);
877 
878 	lvp = unp->un_lowervp;
879 	uvp = NULLVP;
880 	credbk = cnp->cn_cred;
881 
882 	/* Authority change to root */
883 	rootinfo = uifind((uid_t)0);
884 	cred = crdup(cnp->cn_cred);
885 	/*
886 	 * The calls to chgproccnt() are needed to compensate for change_ruid()
887 	 * calling chgproccnt().
888 	 */
889 	chgproccnt(cred->cr_ruidinfo, 1, 0);
890 	change_euid(cred, rootinfo);
891 	change_ruid(cred, rootinfo);
892 	change_svuid(cred, (uid_t)0);
893 	uifree(rootinfo);
894 	cnp->cn_cred = cred;
895 
896 	memset(&nd.ni_cnd, 0, sizeof(struct componentname));
897 	NDPREINIT(&nd);
898 
899 	if ((error = VOP_GETATTR(lvp, &lva, cnp->cn_cred)))
900 		goto unionfs_mkshadowdir_abort;
901 
902 	if ((error = unionfs_relookup(udvp, &uvp, cnp, &nd.ni_cnd, td,
903 	    cnp->cn_nameptr, cnp->cn_namelen, CREATE)))
904 		goto unionfs_mkshadowdir_abort;
905 	if (uvp != NULLVP) {
906 		if (udvp == uvp)
907 			vrele(uvp);
908 		else
909 			vput(uvp);
910 
911 		error = EEXIST;
912 		goto unionfs_mkshadowdir_abort;
913 	}
914 
915 	if ((error = vn_start_write(udvp, &mp, V_WAIT | V_PCATCH)))
916 		goto unionfs_mkshadowdir_abort;
917 	unionfs_create_uppervattr_core(ump, &lva, &va, td);
918 
919 	error = VOP_MKDIR(udvp, &uvp, &nd.ni_cnd, &va);
920 
921 	if (!error) {
922 		unionfs_node_update(unp, uvp, td);
923 
924 		/*
925 		 * XXX The bug which cannot set uid/gid was corrected.
926 		 * Ignore errors.
927 		 */
928 		va.va_type = VNON;
929 		VOP_SETATTR(uvp, &va, nd.ni_cnd.cn_cred);
930 	}
931 	vn_finished_write(mp);
932 
933 unionfs_mkshadowdir_abort:
934 	cnp->cn_cred = credbk;
935 	chgproccnt(cred->cr_ruidinfo, -1, 0);
936 	crfree(cred);
937 
938 	return (error);
939 }
940 
941 /*
942  * Create a new whiteout.
943  *
944  * dvp should be locked on entry and will be locked on return.
945  */
946 int
947 unionfs_mkwhiteout(struct vnode *dvp, struct componentname *cnp,
948     struct thread *td, char *path, int pathlen)
949 {
950 	struct vnode   *wvp;
951 	struct nameidata nd;
952 	struct mount   *mp;
953 	int		error;
954 
955 	wvp = NULLVP;
956 	NDPREINIT(&nd);
957 	if ((error = unionfs_relookup(dvp, &wvp, cnp, &nd.ni_cnd, td, path,
958 	    pathlen, CREATE))) {
959 		return (error);
960 	}
961 	if (wvp != NULLVP) {
962 		if (dvp == wvp)
963 			vrele(wvp);
964 		else
965 			vput(wvp);
966 
967 		return (EEXIST);
968 	}
969 
970 	if ((error = vn_start_write(dvp, &mp, V_WAIT | V_PCATCH)))
971 		goto unionfs_mkwhiteout_free_out;
972 	error = VOP_WHITEOUT(dvp, &nd.ni_cnd, CREATE);
973 
974 	vn_finished_write(mp);
975 
976 unionfs_mkwhiteout_free_out:
977 	return (error);
978 }
979 
980 /*
981  * Create a new vnode for create a new shadow file.
982  *
983  * If an error is returned, *vpp will be invalid, otherwise it will hold a
984  * locked, referenced and opened vnode.
985  *
986  * unp is never updated.
987  */
988 static int
989 unionfs_vn_create_on_upper(struct vnode **vpp, struct vnode *udvp,
990     struct unionfs_node *unp, struct vattr *uvap, struct thread *td)
991 {
992 	struct unionfs_mount *ump;
993 	struct vnode   *vp;
994 	struct vnode   *lvp;
995 	struct ucred   *cred;
996 	struct vattr	lva;
997 	struct nameidata nd;
998 	int		fmode;
999 	int		error;
1000 
1001 	ump = MOUNTTOUNIONFSMOUNT(UNIONFSTOV(unp)->v_mount);
1002 	vp = NULLVP;
1003 	lvp = unp->un_lowervp;
1004 	cred = td->td_ucred;
1005 	fmode = FFLAGS(O_WRONLY | O_CREAT | O_TRUNC | O_EXCL);
1006 	error = 0;
1007 
1008 	if ((error = VOP_GETATTR(lvp, &lva, cred)) != 0)
1009 		return (error);
1010 	unionfs_create_uppervattr_core(ump, &lva, uvap, td);
1011 
1012 	if (unp->un_path == NULL)
1013 		panic("%s: NULL un_path", __func__);
1014 
1015 	nd.ni_cnd.cn_namelen = unp->un_pathlen;
1016 	nd.ni_cnd.cn_pnbuf = unp->un_path;
1017 	nd.ni_cnd.cn_nameiop = CREATE;
1018 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | ISLASTCN;
1019 	nd.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
1020 	nd.ni_cnd.cn_cred = cred;
1021 	nd.ni_cnd.cn_nameptr = nd.ni_cnd.cn_pnbuf;
1022 	NDPREINIT(&nd);
1023 
1024 	vref(udvp);
1025 	if ((error = vfs_relookup(udvp, &vp, &nd.ni_cnd, false)) != 0)
1026 		goto unionfs_vn_create_on_upper_free_out2;
1027 	vrele(udvp);
1028 
1029 	if (vp != NULLVP) {
1030 		if (vp == udvp)
1031 			vrele(vp);
1032 		else
1033 			vput(vp);
1034 		error = EEXIST;
1035 		goto unionfs_vn_create_on_upper_free_out1;
1036 	}
1037 
1038 	if ((error = VOP_CREATE(udvp, &vp, &nd.ni_cnd, uvap)) != 0)
1039 		goto unionfs_vn_create_on_upper_free_out1;
1040 
1041 	if ((error = VOP_OPEN(vp, fmode, cred, td, NULL)) != 0) {
1042 		vput(vp);
1043 		goto unionfs_vn_create_on_upper_free_out1;
1044 	}
1045 	error = VOP_ADD_WRITECOUNT(vp, 1);
1046 	CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
1047 	    __func__, vp, vp->v_writecount);
1048 	if (error == 0) {
1049 		*vpp = vp;
1050 	} else {
1051 		VOP_CLOSE(vp, fmode, cred, td);
1052 	}
1053 
1054 unionfs_vn_create_on_upper_free_out1:
1055 	VOP_UNLOCK(udvp);
1056 
1057 unionfs_vn_create_on_upper_free_out2:
1058 	KASSERT(nd.ni_cnd.cn_pnbuf == unp->un_path,
1059 	    ("%s: cn_pnbuf changed", __func__));
1060 
1061 	return (error);
1062 }
1063 
1064 /*
1065  * Copy from lvp to uvp.
1066  *
1067  * lvp and uvp should be locked and opened on entry and will be locked and
1068  * opened on return.
1069  */
1070 static int
1071 unionfs_copyfile_core(struct vnode *lvp, struct vnode *uvp,
1072     struct ucred *cred, struct thread *td)
1073 {
1074 	char           *buf;
1075 	struct uio	uio;
1076 	struct iovec	iov;
1077 	off_t		offset;
1078 	int		count;
1079 	int		error;
1080 	int		bufoffset;
1081 
1082 	error = 0;
1083 	memset(&uio, 0, sizeof(uio));
1084 
1085 	uio.uio_td = td;
1086 	uio.uio_segflg = UIO_SYSSPACE;
1087 	uio.uio_offset = 0;
1088 
1089 	buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK);
1090 
1091 	while (error == 0) {
1092 		offset = uio.uio_offset;
1093 
1094 		uio.uio_iov = &iov;
1095 		uio.uio_iovcnt = 1;
1096 		iov.iov_base = buf;
1097 		iov.iov_len = MAXBSIZE;
1098 		uio.uio_resid = iov.iov_len;
1099 		uio.uio_rw = UIO_READ;
1100 
1101 		if ((error = VOP_READ(lvp, &uio, 0, cred)) != 0)
1102 			break;
1103 		if ((count = MAXBSIZE - uio.uio_resid) == 0)
1104 			break;
1105 
1106 		bufoffset = 0;
1107 		while (bufoffset < count) {
1108 			uio.uio_iov = &iov;
1109 			uio.uio_iovcnt = 1;
1110 			iov.iov_base = buf + bufoffset;
1111 			iov.iov_len = count - bufoffset;
1112 			uio.uio_offset = offset + bufoffset;
1113 			uio.uio_resid = iov.iov_len;
1114 			uio.uio_rw = UIO_WRITE;
1115 
1116 			if ((error = VOP_WRITE(uvp, &uio, 0, cred)) != 0)
1117 				break;
1118 
1119 			bufoffset += (count - bufoffset) - uio.uio_resid;
1120 		}
1121 
1122 		uio.uio_offset = offset + bufoffset;
1123 	}
1124 
1125 	free(buf, M_TEMP);
1126 
1127 	return (error);
1128 }
1129 
1130 /*
1131  * Copy file from lower to upper.
1132  *
1133  * If you need copy of the contents, set 1 to docopy. Otherwise, set 0 to
1134  * docopy.
1135  *
1136  * If no error returned, unp will be updated.
1137  */
1138 int
1139 unionfs_copyfile(struct unionfs_node *unp, int docopy, struct ucred *cred,
1140     struct thread *td)
1141 {
1142 	struct mount   *mp;
1143 	struct vnode   *udvp;
1144 	struct vnode   *lvp;
1145 	struct vnode   *uvp;
1146 	struct vattr	uva;
1147 	int		error;
1148 
1149 	lvp = unp->un_lowervp;
1150 	uvp = NULLVP;
1151 
1152 	if ((UNIONFSTOV(unp)->v_mount->mnt_flag & MNT_RDONLY))
1153 		return (EROFS);
1154 	if (unp->un_dvp == NULLVP)
1155 		return (EINVAL);
1156 	if (unp->un_uppervp != NULLVP)
1157 		return (EEXIST);
1158 	udvp = VTOUNIONFS(unp->un_dvp)->un_uppervp;
1159 	if (udvp == NULLVP)
1160 		return (EROFS);
1161 	if ((udvp->v_mount->mnt_flag & MNT_RDONLY))
1162 		return (EROFS);
1163 
1164 	error = VOP_ACCESS(lvp, VREAD, cred, td);
1165 	if (error != 0)
1166 		return (error);
1167 
1168 	if ((error = vn_start_write(udvp, &mp, V_WAIT | V_PCATCH)) != 0)
1169 		return (error);
1170 	error = unionfs_vn_create_on_upper(&uvp, udvp, unp, &uva, td);
1171 	if (error != 0) {
1172 		vn_finished_write(mp);
1173 		return (error);
1174 	}
1175 
1176 	if (docopy != 0) {
1177 		error = VOP_OPEN(lvp, FREAD, cred, td, NULL);
1178 		if (error == 0) {
1179 			error = unionfs_copyfile_core(lvp, uvp, cred, td);
1180 			VOP_CLOSE(lvp, FREAD, cred, td);
1181 		}
1182 	}
1183 	VOP_CLOSE(uvp, FWRITE, cred, td);
1184 	VOP_ADD_WRITECOUNT_CHECKED(uvp, -1);
1185 	CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
1186 	    __func__, uvp, uvp->v_writecount);
1187 
1188 	vn_finished_write(mp);
1189 
1190 	if (error == 0) {
1191 		/* Reset the attributes. Ignore errors. */
1192 		uva.va_type = VNON;
1193 		VOP_SETATTR(uvp, &uva, cred);
1194 	}
1195 
1196 	unionfs_node_update(unp, uvp, td);
1197 
1198 	return (error);
1199 }
1200 
1201 /*
1202  * It checks whether vp can rmdir. (check empty)
1203  *
1204  * vp is unionfs vnode.
1205  * vp should be locked.
1206  */
1207 int
1208 unionfs_check_rmdir(struct vnode *vp, struct ucred *cred, struct thread *td)
1209 {
1210 	struct vnode   *uvp;
1211 	struct vnode   *lvp;
1212 	struct vnode   *tvp;
1213 	struct dirent  *dp;
1214 	struct dirent  *edp;
1215 	struct componentname cn;
1216 	struct iovec	iov;
1217 	struct uio	uio;
1218 	struct vattr	va;
1219 	int		error;
1220 	int		eofflag;
1221 	int		lookuperr;
1222 
1223 	/*
1224 	 * The size of buf needs to be larger than DIRBLKSIZ.
1225 	 */
1226 	char		buf[256 * 6];
1227 
1228 	ASSERT_VOP_ELOCKED(vp, __func__);
1229 
1230 	eofflag = 0;
1231 	uvp = UNIONFSVPTOUPPERVP(vp);
1232 	lvp = UNIONFSVPTOLOWERVP(vp);
1233 
1234 	/* check opaque */
1235 	if ((error = VOP_GETATTR(uvp, &va, cred)) != 0)
1236 		return (error);
1237 	if (va.va_flags & OPAQUE)
1238 		return (0);
1239 
1240 	/* open vnode */
1241 #ifdef MAC
1242 	if ((error = mac_vnode_check_open(cred, vp, VEXEC|VREAD)) != 0)
1243 		return (error);
1244 #endif
1245 	if ((error = VOP_ACCESS(vp, VEXEC|VREAD, cred, td)) != 0)
1246 		return (error);
1247 	if ((error = VOP_OPEN(vp, FREAD, cred, td, NULL)) != 0)
1248 		return (error);
1249 
1250 	uio.uio_rw = UIO_READ;
1251 	uio.uio_segflg = UIO_SYSSPACE;
1252 	uio.uio_td = td;
1253 	uio.uio_offset = 0;
1254 
1255 #ifdef MAC
1256 	error = mac_vnode_check_readdir(td->td_ucred, lvp);
1257 #endif
1258 	while (!error && !eofflag) {
1259 		iov.iov_base = buf;
1260 		iov.iov_len = sizeof(buf);
1261 		uio.uio_iov = &iov;
1262 		uio.uio_iovcnt = 1;
1263 		uio.uio_resid = iov.iov_len;
1264 
1265 		error = VOP_READDIR(lvp, &uio, cred, &eofflag, NULL, NULL);
1266 		if (error != 0)
1267 			break;
1268 		KASSERT(eofflag != 0 || uio.uio_resid < sizeof(buf),
1269 		    ("%s: empty read from lower FS", __func__));
1270 
1271 		edp = (struct dirent*)&buf[sizeof(buf) - uio.uio_resid];
1272 		for (dp = (struct dirent*)buf; !error && dp < edp;
1273 		     dp = (struct dirent*)((caddr_t)dp + dp->d_reclen)) {
1274 			if (dp->d_type == DT_WHT || dp->d_fileno == 0 ||
1275 			    (dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1276 			    (dp->d_namlen == 2 && !bcmp(dp->d_name, "..", 2)))
1277 				continue;
1278 
1279 			cn.cn_namelen = dp->d_namlen;
1280 			cn.cn_pnbuf = NULL;
1281 			cn.cn_nameptr = dp->d_name;
1282 			cn.cn_nameiop = LOOKUP;
1283 			cn.cn_flags = LOCKPARENT | LOCKLEAF | RDONLY | ISLASTCN;
1284 			cn.cn_lkflags = LK_EXCLUSIVE;
1285 			cn.cn_cred = cred;
1286 
1287 			/*
1288 			 * check entry in lower.
1289 			 * Sometimes, readdir function returns
1290 			 * wrong entry.
1291 			 */
1292 			lookuperr = VOP_LOOKUP(lvp, &tvp, &cn);
1293 
1294 			if (!lookuperr)
1295 				vput(tvp);
1296 			else
1297 				continue; /* skip entry */
1298 
1299 			/*
1300 			 * check entry
1301 			 * If it has no exist/whiteout entry in upper,
1302 			 * directory is not empty.
1303 			 */
1304 			cn.cn_flags = LOCKPARENT | LOCKLEAF | RDONLY | ISLASTCN;
1305 			lookuperr = VOP_LOOKUP(uvp, &tvp, &cn);
1306 
1307 			if (!lookuperr)
1308 				vput(tvp);
1309 
1310 			/* ignore exist or whiteout entry */
1311 			if (!lookuperr ||
1312 			    (lookuperr == ENOENT && (cn.cn_flags & ISWHITEOUT)))
1313 				continue;
1314 
1315 			error = ENOTEMPTY;
1316 		}
1317 	}
1318 
1319 	/* close vnode */
1320 	VOP_CLOSE(vp, FREAD, cred, td);
1321 
1322 	return (error);
1323 }
1324 
1325