xref: /freebsd/sys/fs/unionfs/union_subr.c (revision 66fd12cf4896eb08ad8e7a2627537f84ead84dd3)
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_uint8(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, LK_EXCLUSIVE, uppervp, false,
392 	    LK_EXCLUSIVE);
393 	error = insmntque1(vp, mp);
394 	if (error != 0) {
395 		unionfs_nodeget_cleanup(vp, unp);
396 		return (error);
397 	}
398 	if (lowervp != NULL && VN_IS_DOOMED(lowervp)) {
399 		vput(lowervp);
400 		unp->un_lowervp = lowervp = NULL;
401 	}
402 	if (uppervp != NULL && VN_IS_DOOMED(uppervp)) {
403 		vput(uppervp);
404 		unp->un_uppervp = uppervp = NULL;
405 		if (lowervp != NULLVP)
406 			vp->v_vnlock = lowervp->v_vnlock;
407 	}
408 	if (lowervp == NULL && uppervp == NULL) {
409 		unionfs_nodeget_cleanup(vp, unp);
410 		return (ENOENT);
411 	}
412 
413 	vn_set_state(vp, VSTATE_CONSTRUCTED);
414 
415 	if (dvp != NULLVP && vt == VDIR)
416 		*vpp = unionfs_ins_cached_vnode(unp, dvp);
417 	if (*vpp != NULLVP) {
418 		unionfs_nodeget_cleanup(vp, unp);
419 		vp = *vpp;
420 	} else {
421 		if (uppervp != NULL)
422 			VOP_UNLOCK(uppervp);
423 		if (lowervp != NULL)
424 			VOP_UNLOCK(lowervp);
425 		*vpp = vp;
426 	}
427 
428 unionfs_nodeget_out:
429 	if (lkflags & LK_TYPE_MASK)
430 		vn_lock(vp, lkflags | LK_RETRY);
431 
432 	return (0);
433 }
434 
435 /*
436  * Clean up the unionfs node.
437  */
438 void
439 unionfs_noderem(struct vnode *vp)
440 {
441 	struct unionfs_node *unp, *unp_t1, *unp_t2;
442 	struct unionfs_node_hashhead *hd;
443 	struct unionfs_node_status *unsp, *unsp_tmp;
444 	struct vnode   *lvp;
445 	struct vnode   *uvp;
446 	struct vnode   *dvp;
447 	int		count;
448 	int		writerefs;
449 
450 	/*
451 	 * The root vnode lock may be recursed during unmount, because
452 	 * it may share the same lock as the unionfs mount's covered vnode,
453 	 * which is locked across VFS_UNMOUNT().  This lock will then be
454 	 * recursively taken during the vflush() issued by unionfs_unmount().
455 	 * But we still only need to lock the unionfs lock once, because only
456 	 * one of those lock operations was taken against a unionfs vnode and
457 	 * will be undone against a unionfs vnode.
458 	 */
459 	KASSERT(vp->v_vnlock->lk_recurse == 0 || (vp->v_vflag & VV_ROOT) != 0,
460 	    ("%s: vnode %p locked recursively", __func__, vp));
461 	if (lockmgr(&vp->v_lock, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0)
462 		panic("%s: failed to acquire lock for vnode lock", __func__);
463 
464 	/*
465 	 * Use the interlock to protect the clearing of v_data to
466 	 * prevent faults in unionfs_lock().
467 	 */
468 	VI_LOCK(vp);
469 	unp = VTOUNIONFS(vp);
470 	lvp = unp->un_lowervp;
471 	uvp = unp->un_uppervp;
472 	dvp = unp->un_dvp;
473 	unp->un_lowervp = unp->un_uppervp = NULLVP;
474 	vp->v_vnlock = &(vp->v_lock);
475 	vp->v_data = NULL;
476 	vp->v_object = NULL;
477 	if (unp->un_hashtbl != NULL) {
478 		/*
479 		 * Clear out any cached child vnodes.  This should only
480 		 * be necessary during forced unmount, when the vnode may
481 		 * be reclaimed with a non-zero use count.  Otherwise the
482 		 * reference held by each child should prevent reclamation.
483 		 */
484 		for (count = 0; count <= UNIONFSHASHMASK; count++) {
485 			hd = unp->un_hashtbl + count;
486 			LIST_FOREACH_SAFE(unp_t1, hd, un_hash, unp_t2) {
487 				LIST_REMOVE(unp_t1, un_hash);
488 				unp_t1->un_hash.le_next = NULL;
489 				unp_t1->un_hash.le_prev = NULL;
490 			}
491 		}
492 	}
493 	VI_UNLOCK(vp);
494 
495 	writerefs = atomic_load_int(&vp->v_writecount);
496 	VNASSERT(writerefs >= 0, vp,
497 	    ("%s: write count %d, unexpected text ref", __func__, writerefs));
498 	/*
499 	 * If we were opened for write, we leased the write reference
500 	 * to the lower vnode.  If this is a reclamation due to the
501 	 * forced unmount, undo the reference now.
502 	 */
503 	if (writerefs > 0) {
504 		VNASSERT(uvp != NULL, vp,
505 		    ("%s: write reference without upper vnode", __func__));
506 		VOP_ADD_WRITECOUNT(uvp, -writerefs);
507 	}
508 	if (lvp != NULLVP)
509 		VOP_UNLOCK(lvp);
510 	if (uvp != NULLVP)
511 		VOP_UNLOCK(uvp);
512 
513 	if (dvp != NULLVP)
514 		unionfs_rem_cached_vnode(unp, dvp);
515 
516 	if (lvp != NULLVP)
517 		vrele(lvp);
518 	if (uvp != NULLVP)
519 		vrele(uvp);
520 	if (unp->un_path != NULL) {
521 		free(unp->un_path, M_UNIONFSPATH);
522 		unp->un_path = NULL;
523 		unp->un_pathlen = 0;
524 	}
525 
526 	if (unp->un_hashtbl != NULL) {
527 		hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, UNIONFSHASHMASK);
528 	}
529 
530 	LIST_FOREACH_SAFE(unsp, &(unp->un_unshead), uns_list, unsp_tmp) {
531 		LIST_REMOVE(unsp, uns_list);
532 		free(unsp, M_TEMP);
533 	}
534 	if (dvp != NULLVP) {
535 		mtx_lock(&unionfs_deferred_rele_lock);
536 		STAILQ_INSERT_TAIL(&unionfs_deferred_rele_list, unp, un_rele);
537 		mtx_unlock(&unionfs_deferred_rele_lock);
538 		taskqueue_enqueue(taskqueue_unionfs_rele,
539 		    &unionfs_deferred_rele_task);
540 	} else
541 		free(unp, M_UNIONFSNODE);
542 }
543 
544 /*
545  * Get the unionfs node status object for the vnode corresponding to unp,
546  * for the process that owns td.  Allocate a new status object if one
547  * does not already exist.
548  */
549 void
550 unionfs_get_node_status(struct unionfs_node *unp, struct thread *td,
551     struct unionfs_node_status **unspp)
552 {
553 	struct unionfs_node_status *unsp;
554 	pid_t pid;
555 
556 	pid = td->td_proc->p_pid;
557 
558 	KASSERT(NULL != unspp, ("%s: NULL status", __func__));
559 	ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), __func__);
560 
561 	LIST_FOREACH(unsp, &(unp->un_unshead), uns_list) {
562 		if (unsp->uns_pid == pid) {
563 			*unspp = unsp;
564 			return;
565 		}
566 	}
567 
568 	/* create a new unionfs node status */
569 	unsp = malloc(sizeof(struct unionfs_node_status),
570 	    M_TEMP, M_WAITOK | M_ZERO);
571 
572 	unsp->uns_pid = pid;
573 	LIST_INSERT_HEAD(&(unp->un_unshead), unsp, uns_list);
574 
575 	*unspp = unsp;
576 }
577 
578 /*
579  * Remove the unionfs node status, if you can.
580  * You need exclusive lock this vnode.
581  */
582 void
583 unionfs_tryrem_node_status(struct unionfs_node *unp,
584     struct unionfs_node_status *unsp)
585 {
586 	KASSERT(NULL != unsp, ("%s: NULL status", __func__));
587 	ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), __func__);
588 
589 	if (0 < unsp->uns_lower_opencnt || 0 < unsp->uns_upper_opencnt)
590 		return;
591 
592 	LIST_REMOVE(unsp, uns_list);
593 	free(unsp, M_TEMP);
594 }
595 
596 /*
597  * Create upper node attr.
598  */
599 void
600 unionfs_create_uppervattr_core(struct unionfs_mount *ump, struct vattr *lva,
601     struct vattr *uva, struct thread *td)
602 {
603 	VATTR_NULL(uva);
604 	uva->va_type = lva->va_type;
605 	uva->va_atime = lva->va_atime;
606 	uva->va_mtime = lva->va_mtime;
607 	uva->va_ctime = lva->va_ctime;
608 
609 	switch (ump->um_copymode) {
610 	case UNIONFS_TRANSPARENT:
611 		uva->va_mode = lva->va_mode;
612 		uva->va_uid = lva->va_uid;
613 		uva->va_gid = lva->va_gid;
614 		break;
615 	case UNIONFS_MASQUERADE:
616 		if (ump->um_uid == lva->va_uid) {
617 			uva->va_mode = lva->va_mode & 077077;
618 			uva->va_mode |= (lva->va_type == VDIR ?
619 			    ump->um_udir : ump->um_ufile) & 0700;
620 			uva->va_uid = lva->va_uid;
621 			uva->va_gid = lva->va_gid;
622 		} else {
623 			uva->va_mode = (lva->va_type == VDIR ?
624 			    ump->um_udir : ump->um_ufile);
625 			uva->va_uid = ump->um_uid;
626 			uva->va_gid = ump->um_gid;
627 		}
628 		break;
629 	default:		/* UNIONFS_TRADITIONAL */
630 		uva->va_mode = 0777 & ~td->td_proc->p_pd->pd_cmask;
631 		uva->va_uid = ump->um_uid;
632 		uva->va_gid = ump->um_gid;
633 		break;
634 	}
635 }
636 
637 /*
638  * Create upper node attr.
639  */
640 int
641 unionfs_create_uppervattr(struct unionfs_mount *ump, struct vnode *lvp,
642     struct vattr *uva, struct ucred *cred, struct thread *td)
643 {
644 	struct vattr	lva;
645 	int		error;
646 
647 	if ((error = VOP_GETATTR(lvp, &lva, cred)))
648 		return (error);
649 
650 	unionfs_create_uppervattr_core(ump, &lva, uva, td);
651 
652 	return (error);
653 }
654 
655 /*
656  * relookup
657  *
658  * dvp should be locked on entry and will be locked on return.
659  *
660  * If an error is returned, *vpp will be invalid, otherwise it will hold a
661  * locked, referenced vnode. If *vpp == dvp then remember that only one
662  * LK_EXCLUSIVE lock is held.
663  */
664 int
665 unionfs_relookup(struct vnode *dvp, struct vnode **vpp,
666     struct componentname *cnp, struct componentname *cn, struct thread *td,
667     char *path, int pathlen, u_long nameiop)
668 {
669 	int error;
670 	bool refstart;
671 
672 	cn->cn_namelen = pathlen;
673 	cn->cn_pnbuf = path;
674 	cn->cn_nameiop = nameiop;
675 	cn->cn_flags = (LOCKPARENT | LOCKLEAF | ISLASTCN);
676 	cn->cn_lkflags = LK_EXCLUSIVE;
677 	cn->cn_cred = cnp->cn_cred;
678 	cn->cn_nameptr = cn->cn_pnbuf;
679 
680 	refstart = false;
681 	if (nameiop == DELETE) {
682 		cn->cn_flags |= (cnp->cn_flags & DOWHITEOUT);
683 	} else if (nameiop == RENAME) {
684 		refstart = true;
685 	} else if (nameiop == CREATE) {
686 		cn->cn_flags |= NOCACHE;
687 	}
688 
689 	vref(dvp);
690 	VOP_UNLOCK(dvp);
691 
692 	if ((error = vfs_relookup(dvp, vpp, cn, refstart))) {
693 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
694 	} else
695 		vrele(dvp);
696 
697 	KASSERT(cn->cn_pnbuf == path, ("%s: cn_pnbuf changed", __func__));
698 
699 	return (error);
700 }
701 
702 /*
703  * relookup for CREATE namei operation.
704  *
705  * dvp is unionfs vnode. dvp should be locked.
706  *
707  * If it called 'unionfs_copyfile' function by unionfs_link etc,
708  * VOP_LOOKUP information is broken.
709  * So it need relookup in order to create link etc.
710  */
711 int
712 unionfs_relookup_for_create(struct vnode *dvp, struct componentname *cnp,
713     struct thread *td)
714 {
715 	struct vnode *udvp;
716 	struct vnode *vp;
717 	struct componentname cn;
718 	int error;
719 
720 	udvp = UNIONFSVPTOUPPERVP(dvp);
721 	vp = NULLVP;
722 
723 	error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
724 	    cnp->cn_namelen, CREATE);
725 	if (error)
726 		return (error);
727 
728 	if (vp != NULLVP) {
729 		if (udvp == vp)
730 			vrele(vp);
731 		else
732 			vput(vp);
733 
734 		error = EEXIST;
735 	}
736 
737 	return (error);
738 }
739 
740 /*
741  * relookup for DELETE namei operation.
742  *
743  * dvp is unionfs vnode. dvp should be locked.
744  */
745 int
746 unionfs_relookup_for_delete(struct vnode *dvp, struct componentname *cnp,
747     struct thread *td)
748 {
749 	struct vnode *udvp;
750 	struct vnode *vp;
751 	struct componentname cn;
752 	int error;
753 
754 	udvp = UNIONFSVPTOUPPERVP(dvp);
755 	vp = NULLVP;
756 
757 	error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
758 	    cnp->cn_namelen, DELETE);
759 	if (error)
760 		return (error);
761 
762 	if (vp == NULLVP)
763 		error = ENOENT;
764 	else {
765 		if (udvp == vp)
766 			vrele(vp);
767 		else
768 			vput(vp);
769 	}
770 
771 	return (error);
772 }
773 
774 /*
775  * relookup for RENAME namei operation.
776  *
777  * dvp is unionfs vnode. dvp should be locked.
778  */
779 int
780 unionfs_relookup_for_rename(struct vnode *dvp, struct componentname *cnp,
781     struct thread *td)
782 {
783 	struct vnode *udvp;
784 	struct vnode *vp;
785 	struct componentname cn;
786 	int error;
787 
788 	udvp = UNIONFSVPTOUPPERVP(dvp);
789 	vp = NULLVP;
790 
791 	error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
792 	    cnp->cn_namelen, RENAME);
793 	if (error)
794 		return (error);
795 
796 	if (vp != NULLVP) {
797 		if (udvp == vp)
798 			vrele(vp);
799 		else
800 			vput(vp);
801 	}
802 
803 	return (error);
804 }
805 
806 /*
807  * Update the unionfs_node.
808  *
809  * uvp is new locked upper vnode. unionfs vnode's lock will be exchanged to the
810  * uvp's lock and lower's lock will be unlocked.
811  */
812 static void
813 unionfs_node_update(struct unionfs_node *unp, struct vnode *uvp,
814     struct thread *td)
815 {
816 	struct unionfs_node_hashhead *hd;
817 	struct vnode   *vp;
818 	struct vnode   *lvp;
819 	struct vnode   *dvp;
820 	unsigned	count, lockrec;
821 
822 	vp = UNIONFSTOV(unp);
823 	lvp = unp->un_lowervp;
824 	ASSERT_VOP_ELOCKED(lvp, __func__);
825 	ASSERT_VOP_ELOCKED(uvp, __func__);
826 	dvp = unp->un_dvp;
827 
828 	VNASSERT(vp->v_writecount == 0, vp,
829 	    ("%s: non-zero writecount", __func__));
830 	/*
831 	 * Update the upper vnode's lock state to match the lower vnode,
832 	 * and then switch the unionfs vnode's lock to the upper vnode.
833 	 */
834 	lockrec = lvp->v_vnlock->lk_recurse;
835 	for (count = 0; count < lockrec; count++)
836 		vn_lock(uvp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY);
837 	VI_LOCK(vp);
838 	unp->un_uppervp = uvp;
839 	vp->v_vnlock = uvp->v_vnlock;
840 	VI_UNLOCK(vp);
841 
842 	/*
843 	 * Re-cache the unionfs vnode against the upper vnode
844 	 */
845 	if (dvp != NULLVP && vp->v_type == VDIR) {
846 		VI_LOCK(dvp);
847 		if (unp->un_hash.le_prev != NULL) {
848 			LIST_REMOVE(unp, un_hash);
849 			hd = unionfs_get_hashhead(dvp, uvp);
850 			LIST_INSERT_HEAD(hd, unp, un_hash);
851 		}
852 		VI_UNLOCK(unp->un_dvp);
853 	}
854 }
855 
856 /*
857  * Create a new shadow dir.
858  *
859  * udvp should be locked on entry and will be locked on return.
860  *
861  * If no error returned, unp will be updated.
862  */
863 int
864 unionfs_mkshadowdir(struct unionfs_mount *ump, struct vnode *udvp,
865     struct unionfs_node *unp, struct componentname *cnp, struct thread *td)
866 {
867 	struct vnode   *lvp;
868 	struct vnode   *uvp;
869 	struct vattr	va;
870 	struct vattr	lva;
871 	struct nameidata nd;
872 	struct mount   *mp;
873 	struct ucred   *cred;
874 	struct ucred   *credbk;
875 	struct uidinfo *rootinfo;
876 	int		error;
877 
878 	if (unp->un_uppervp != NULLVP)
879 		return (EEXIST);
880 
881 	lvp = unp->un_lowervp;
882 	uvp = NULLVP;
883 	credbk = cnp->cn_cred;
884 
885 	/* Authority change to root */
886 	rootinfo = uifind((uid_t)0);
887 	cred = crdup(cnp->cn_cred);
888 	/*
889 	 * The calls to chgproccnt() are needed to compensate for change_ruid()
890 	 * calling chgproccnt().
891 	 */
892 	chgproccnt(cred->cr_ruidinfo, 1, 0);
893 	change_euid(cred, rootinfo);
894 	change_ruid(cred, rootinfo);
895 	change_svuid(cred, (uid_t)0);
896 	uifree(rootinfo);
897 	cnp->cn_cred = cred;
898 
899 	memset(&nd.ni_cnd, 0, sizeof(struct componentname));
900 	NDPREINIT(&nd);
901 
902 	if ((error = VOP_GETATTR(lvp, &lva, cnp->cn_cred)))
903 		goto unionfs_mkshadowdir_abort;
904 
905 	if ((error = unionfs_relookup(udvp, &uvp, cnp, &nd.ni_cnd, td,
906 	    cnp->cn_nameptr, cnp->cn_namelen, CREATE)))
907 		goto unionfs_mkshadowdir_abort;
908 	if (uvp != NULLVP) {
909 		if (udvp == uvp)
910 			vrele(uvp);
911 		else
912 			vput(uvp);
913 
914 		error = EEXIST;
915 		goto unionfs_mkshadowdir_abort;
916 	}
917 
918 	if ((error = vn_start_write(udvp, &mp, V_WAIT | V_PCATCH)))
919 		goto unionfs_mkshadowdir_abort;
920 	unionfs_create_uppervattr_core(ump, &lva, &va, td);
921 
922 	error = VOP_MKDIR(udvp, &uvp, &nd.ni_cnd, &va);
923 
924 	if (!error) {
925 		unionfs_node_update(unp, uvp, td);
926 
927 		/*
928 		 * XXX The bug which cannot set uid/gid was corrected.
929 		 * Ignore errors.
930 		 */
931 		va.va_type = VNON;
932 		VOP_SETATTR(uvp, &va, nd.ni_cnd.cn_cred);
933 	}
934 	vn_finished_write(mp);
935 
936 unionfs_mkshadowdir_abort:
937 	cnp->cn_cred = credbk;
938 	chgproccnt(cred->cr_ruidinfo, -1, 0);
939 	crfree(cred);
940 
941 	return (error);
942 }
943 
944 /*
945  * Create a new whiteout.
946  *
947  * dvp should be locked on entry and will be locked on return.
948  */
949 int
950 unionfs_mkwhiteout(struct vnode *dvp, struct componentname *cnp,
951     struct thread *td, char *path, int pathlen)
952 {
953 	struct vnode   *wvp;
954 	struct nameidata nd;
955 	struct mount   *mp;
956 	int		error;
957 
958 	wvp = NULLVP;
959 	NDPREINIT(&nd);
960 	if ((error = unionfs_relookup(dvp, &wvp, cnp, &nd.ni_cnd, td, path,
961 	    pathlen, CREATE))) {
962 		return (error);
963 	}
964 	if (wvp != NULLVP) {
965 		if (dvp == wvp)
966 			vrele(wvp);
967 		else
968 			vput(wvp);
969 
970 		return (EEXIST);
971 	}
972 
973 	if ((error = vn_start_write(dvp, &mp, V_WAIT | V_PCATCH)))
974 		goto unionfs_mkwhiteout_free_out;
975 	error = VOP_WHITEOUT(dvp, &nd.ni_cnd, CREATE);
976 
977 	vn_finished_write(mp);
978 
979 unionfs_mkwhiteout_free_out:
980 	return (error);
981 }
982 
983 /*
984  * Create a new vnode for create a new shadow file.
985  *
986  * If an error is returned, *vpp will be invalid, otherwise it will hold a
987  * locked, referenced and opened vnode.
988  *
989  * unp is never updated.
990  */
991 static int
992 unionfs_vn_create_on_upper(struct vnode **vpp, struct vnode *udvp,
993     struct unionfs_node *unp, struct vattr *uvap, struct thread *td)
994 {
995 	struct unionfs_mount *ump;
996 	struct vnode   *vp;
997 	struct vnode   *lvp;
998 	struct ucred   *cred;
999 	struct vattr	lva;
1000 	struct nameidata nd;
1001 	int		fmode;
1002 	int		error;
1003 
1004 	ump = MOUNTTOUNIONFSMOUNT(UNIONFSTOV(unp)->v_mount);
1005 	vp = NULLVP;
1006 	lvp = unp->un_lowervp;
1007 	cred = td->td_ucred;
1008 	fmode = FFLAGS(O_WRONLY | O_CREAT | O_TRUNC | O_EXCL);
1009 	error = 0;
1010 
1011 	if ((error = VOP_GETATTR(lvp, &lva, cred)) != 0)
1012 		return (error);
1013 	unionfs_create_uppervattr_core(ump, &lva, uvap, td);
1014 
1015 	if (unp->un_path == NULL)
1016 		panic("%s: NULL un_path", __func__);
1017 
1018 	nd.ni_cnd.cn_namelen = unp->un_pathlen;
1019 	nd.ni_cnd.cn_pnbuf = unp->un_path;
1020 	nd.ni_cnd.cn_nameiop = CREATE;
1021 	nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | ISLASTCN;
1022 	nd.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
1023 	nd.ni_cnd.cn_cred = cred;
1024 	nd.ni_cnd.cn_nameptr = nd.ni_cnd.cn_pnbuf;
1025 	NDPREINIT(&nd);
1026 
1027 	vref(udvp);
1028 	if ((error = vfs_relookup(udvp, &vp, &nd.ni_cnd, false)) != 0)
1029 		goto unionfs_vn_create_on_upper_free_out2;
1030 	vrele(udvp);
1031 
1032 	if (vp != NULLVP) {
1033 		if (vp == udvp)
1034 			vrele(vp);
1035 		else
1036 			vput(vp);
1037 		error = EEXIST;
1038 		goto unionfs_vn_create_on_upper_free_out1;
1039 	}
1040 
1041 	if ((error = VOP_CREATE(udvp, &vp, &nd.ni_cnd, uvap)) != 0)
1042 		goto unionfs_vn_create_on_upper_free_out1;
1043 
1044 	if ((error = VOP_OPEN(vp, fmode, cred, td, NULL)) != 0) {
1045 		vput(vp);
1046 		goto unionfs_vn_create_on_upper_free_out1;
1047 	}
1048 	error = VOP_ADD_WRITECOUNT(vp, 1);
1049 	CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
1050 	    __func__, vp, vp->v_writecount);
1051 	if (error == 0) {
1052 		*vpp = vp;
1053 	} else {
1054 		VOP_CLOSE(vp, fmode, cred, td);
1055 	}
1056 
1057 unionfs_vn_create_on_upper_free_out1:
1058 	VOP_UNLOCK(udvp);
1059 
1060 unionfs_vn_create_on_upper_free_out2:
1061 	KASSERT(nd.ni_cnd.cn_pnbuf == unp->un_path,
1062 	    ("%s: cn_pnbuf changed", __func__));
1063 
1064 	return (error);
1065 }
1066 
1067 /*
1068  * Copy from lvp to uvp.
1069  *
1070  * lvp and uvp should be locked and opened on entry and will be locked and
1071  * opened on return.
1072  */
1073 static int
1074 unionfs_copyfile_core(struct vnode *lvp, struct vnode *uvp,
1075     struct ucred *cred, struct thread *td)
1076 {
1077 	char           *buf;
1078 	struct uio	uio;
1079 	struct iovec	iov;
1080 	off_t		offset;
1081 	int		count;
1082 	int		error;
1083 	int		bufoffset;
1084 
1085 	error = 0;
1086 	memset(&uio, 0, sizeof(uio));
1087 
1088 	uio.uio_td = td;
1089 	uio.uio_segflg = UIO_SYSSPACE;
1090 	uio.uio_offset = 0;
1091 
1092 	buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK);
1093 
1094 	while (error == 0) {
1095 		offset = uio.uio_offset;
1096 
1097 		uio.uio_iov = &iov;
1098 		uio.uio_iovcnt = 1;
1099 		iov.iov_base = buf;
1100 		iov.iov_len = MAXBSIZE;
1101 		uio.uio_resid = iov.iov_len;
1102 		uio.uio_rw = UIO_READ;
1103 
1104 		if ((error = VOP_READ(lvp, &uio, 0, cred)) != 0)
1105 			break;
1106 		if ((count = MAXBSIZE - uio.uio_resid) == 0)
1107 			break;
1108 
1109 		bufoffset = 0;
1110 		while (bufoffset < count) {
1111 			uio.uio_iov = &iov;
1112 			uio.uio_iovcnt = 1;
1113 			iov.iov_base = buf + bufoffset;
1114 			iov.iov_len = count - bufoffset;
1115 			uio.uio_offset = offset + bufoffset;
1116 			uio.uio_resid = iov.iov_len;
1117 			uio.uio_rw = UIO_WRITE;
1118 
1119 			if ((error = VOP_WRITE(uvp, &uio, 0, cred)) != 0)
1120 				break;
1121 
1122 			bufoffset += (count - bufoffset) - uio.uio_resid;
1123 		}
1124 
1125 		uio.uio_offset = offset + bufoffset;
1126 	}
1127 
1128 	free(buf, M_TEMP);
1129 
1130 	return (error);
1131 }
1132 
1133 /*
1134  * Copy file from lower to upper.
1135  *
1136  * If you need copy of the contents, set 1 to docopy. Otherwise, set 0 to
1137  * docopy.
1138  *
1139  * If no error returned, unp will be updated.
1140  */
1141 int
1142 unionfs_copyfile(struct unionfs_node *unp, int docopy, struct ucred *cred,
1143     struct thread *td)
1144 {
1145 	struct mount   *mp;
1146 	struct vnode   *udvp;
1147 	struct vnode   *lvp;
1148 	struct vnode   *uvp;
1149 	struct vattr	uva;
1150 	int		error;
1151 
1152 	lvp = unp->un_lowervp;
1153 	uvp = NULLVP;
1154 
1155 	if ((UNIONFSTOV(unp)->v_mount->mnt_flag & MNT_RDONLY))
1156 		return (EROFS);
1157 	if (unp->un_dvp == NULLVP)
1158 		return (EINVAL);
1159 	if (unp->un_uppervp != NULLVP)
1160 		return (EEXIST);
1161 	udvp = VTOUNIONFS(unp->un_dvp)->un_uppervp;
1162 	if (udvp == NULLVP)
1163 		return (EROFS);
1164 	if ((udvp->v_mount->mnt_flag & MNT_RDONLY))
1165 		return (EROFS);
1166 
1167 	error = VOP_ACCESS(lvp, VREAD, cred, td);
1168 	if (error != 0)
1169 		return (error);
1170 
1171 	if ((error = vn_start_write(udvp, &mp, V_WAIT | V_PCATCH)) != 0)
1172 		return (error);
1173 	error = unionfs_vn_create_on_upper(&uvp, udvp, unp, &uva, td);
1174 	if (error != 0) {
1175 		vn_finished_write(mp);
1176 		return (error);
1177 	}
1178 
1179 	if (docopy != 0) {
1180 		error = VOP_OPEN(lvp, FREAD, cred, td, NULL);
1181 		if (error == 0) {
1182 			error = unionfs_copyfile_core(lvp, uvp, cred, td);
1183 			VOP_CLOSE(lvp, FREAD, cred, td);
1184 		}
1185 	}
1186 	VOP_CLOSE(uvp, FWRITE, cred, td);
1187 	VOP_ADD_WRITECOUNT_CHECKED(uvp, -1);
1188 	CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
1189 	    __func__, uvp, uvp->v_writecount);
1190 
1191 	vn_finished_write(mp);
1192 
1193 	if (error == 0) {
1194 		/* Reset the attributes. Ignore errors. */
1195 		uva.va_type = VNON;
1196 		VOP_SETATTR(uvp, &uva, cred);
1197 	}
1198 
1199 	unionfs_node_update(unp, uvp, td);
1200 
1201 	return (error);
1202 }
1203 
1204 /*
1205  * It checks whether vp can rmdir. (check empty)
1206  *
1207  * vp is unionfs vnode.
1208  * vp should be locked.
1209  */
1210 int
1211 unionfs_check_rmdir(struct vnode *vp, struct ucred *cred, struct thread *td)
1212 {
1213 	struct vnode   *uvp;
1214 	struct vnode   *lvp;
1215 	struct vnode   *tvp;
1216 	struct dirent  *dp;
1217 	struct dirent  *edp;
1218 	struct componentname cn;
1219 	struct iovec	iov;
1220 	struct uio	uio;
1221 	struct vattr	va;
1222 	int		error;
1223 	int		eofflag;
1224 	int		lookuperr;
1225 
1226 	/*
1227 	 * The size of buf needs to be larger than DIRBLKSIZ.
1228 	 */
1229 	char		buf[256 * 6];
1230 
1231 	ASSERT_VOP_ELOCKED(vp, __func__);
1232 
1233 	eofflag = 0;
1234 	uvp = UNIONFSVPTOUPPERVP(vp);
1235 	lvp = UNIONFSVPTOLOWERVP(vp);
1236 
1237 	/* check opaque */
1238 	if ((error = VOP_GETATTR(uvp, &va, cred)) != 0)
1239 		return (error);
1240 	if (va.va_flags & OPAQUE)
1241 		return (0);
1242 
1243 	/* open vnode */
1244 #ifdef MAC
1245 	if ((error = mac_vnode_check_open(cred, vp, VEXEC|VREAD)) != 0)
1246 		return (error);
1247 #endif
1248 	if ((error = VOP_ACCESS(vp, VEXEC|VREAD, cred, td)) != 0)
1249 		return (error);
1250 	if ((error = VOP_OPEN(vp, FREAD, cred, td, NULL)) != 0)
1251 		return (error);
1252 
1253 	uio.uio_rw = UIO_READ;
1254 	uio.uio_segflg = UIO_SYSSPACE;
1255 	uio.uio_td = td;
1256 	uio.uio_offset = 0;
1257 
1258 #ifdef MAC
1259 	error = mac_vnode_check_readdir(td->td_ucred, lvp);
1260 #endif
1261 	while (!error && !eofflag) {
1262 		iov.iov_base = buf;
1263 		iov.iov_len = sizeof(buf);
1264 		uio.uio_iov = &iov;
1265 		uio.uio_iovcnt = 1;
1266 		uio.uio_resid = iov.iov_len;
1267 
1268 		error = VOP_READDIR(lvp, &uio, cred, &eofflag, NULL, NULL);
1269 		if (error != 0)
1270 			break;
1271 		KASSERT(eofflag != 0 || uio.uio_resid < sizeof(buf),
1272 		    ("%s: empty read from lower FS", __func__));
1273 
1274 		edp = (struct dirent*)&buf[sizeof(buf) - uio.uio_resid];
1275 		for (dp = (struct dirent*)buf; !error && dp < edp;
1276 		     dp = (struct dirent*)((caddr_t)dp + dp->d_reclen)) {
1277 			if (dp->d_type == DT_WHT || dp->d_fileno == 0 ||
1278 			    (dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1279 			    (dp->d_namlen == 2 && !bcmp(dp->d_name, "..", 2)))
1280 				continue;
1281 
1282 			cn.cn_namelen = dp->d_namlen;
1283 			cn.cn_pnbuf = NULL;
1284 			cn.cn_nameptr = dp->d_name;
1285 			cn.cn_nameiop = LOOKUP;
1286 			cn.cn_flags = LOCKPARENT | LOCKLEAF | RDONLY | ISLASTCN;
1287 			cn.cn_lkflags = LK_EXCLUSIVE;
1288 			cn.cn_cred = cred;
1289 
1290 			/*
1291 			 * check entry in lower.
1292 			 * Sometimes, readdir function returns
1293 			 * wrong entry.
1294 			 */
1295 			lookuperr = VOP_LOOKUP(lvp, &tvp, &cn);
1296 
1297 			if (!lookuperr)
1298 				vput(tvp);
1299 			else
1300 				continue; /* skip entry */
1301 
1302 			/*
1303 			 * check entry
1304 			 * If it has no exist/whiteout entry in upper,
1305 			 * directory is not empty.
1306 			 */
1307 			cn.cn_flags = LOCKPARENT | LOCKLEAF | RDONLY | ISLASTCN;
1308 			lookuperr = VOP_LOOKUP(uvp, &tvp, &cn);
1309 
1310 			if (!lookuperr)
1311 				vput(tvp);
1312 
1313 			/* ignore exist or whiteout entry */
1314 			if (!lookuperr ||
1315 			    (lookuperr == ENOENT && (cn.cn_flags & ISWHITEOUT)))
1316 				continue;
1317 
1318 			error = ENOTEMPTY;
1319 		}
1320 	}
1321 
1322 	/* close vnode */
1323 	VOP_CLOSE(vp, FREAD, cred, td);
1324 
1325 	return (error);
1326 }
1327 
1328