xref: /freebsd/sys/fs/unionfs/union_vnops.c (revision 9768746b)
1 /*-
2  * SPDX-License-Identifier: BSD-3-Clause
3  *
4  * Copyright (c) 1992, 1993, 1994, 1995 Jan-Simon Pendry.
5  * Copyright (c) 1992, 1993, 1994, 1995
6  *      The Regents of the University of California.
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  * All rights reserved.
10  *
11  * This code is derived from software contributed to Berkeley by
12  * Jan-Simon Pendry.
13  *
14  * Redistribution and use in source and binary forms, with or without
15  * modification, are permitted provided that the following conditions
16  * are met:
17  * 1. Redistributions of source code must retain the above copyright
18  *    notice, this list of conditions and the following disclaimer.
19  * 2. Redistributions in binary form must reproduce the above copyright
20  *    notice, this list of conditions and the following disclaimer in the
21  *    documentation and/or other materials provided with the distribution.
22  * 3. Neither the name of the University nor the names of its contributors
23  *    may be used to endorse or promote products derived from this software
24  *    without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
27  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
28  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
29  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
30  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
31  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
32  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
33  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
34  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
35  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36  * SUCH DAMAGE.
37  *
38  *	@(#)union_vnops.c	8.32 (Berkeley) 6/23/95
39  * $FreeBSD$
40  *
41  */
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/conf.h>
46 #include <sys/kernel.h>
47 #include <sys/lock.h>
48 #include <sys/malloc.h>
49 #include <sys/mount.h>
50 #include <sys/mutex.h>
51 #include <sys/namei.h>
52 #include <sys/sysctl.h>
53 #include <sys/vnode.h>
54 #include <sys/kdb.h>
55 #include <sys/fcntl.h>
56 #include <sys/stat.h>
57 #include <sys/dirent.h>
58 #include <sys/proc.h>
59 #include <sys/bio.h>
60 #include <sys/buf.h>
61 
62 #include <fs/unionfs/union.h>
63 
64 #include <machine/atomic.h>
65 
66 #include <vm/vm.h>
67 #include <vm/vm_extern.h>
68 #include <vm/vm_object.h>
69 #include <vm/vnode_pager.h>
70 
71 #if 0
72 #define UNIONFS_INTERNAL_DEBUG(msg, args...)    printf(msg, ## args)
73 #define UNIONFS_IDBG_RENAME
74 #else
75 #define UNIONFS_INTERNAL_DEBUG(msg, args...)
76 #endif
77 
78 #define KASSERT_UNIONFS_VNODE(vp) \
79 	VNASSERT(((vp)->v_op == &unionfs_vnodeops), vp, \
80 	    ("%s: non-unionfs vnode", __func__))
81 
82 static int
83 unionfs_lookup(struct vop_cachedlookup_args *ap)
84 {
85 	struct unionfs_node *dunp;
86 	struct vnode   *dvp, *udvp, *ldvp, *vp, *uvp, *lvp, *dtmpvp;
87 	struct vattr	va;
88 	struct componentname *cnp;
89 	struct thread  *td;
90 	u_long		nameiop;
91 	u_long		cnflags, cnflagsbk;
92 	int		iswhiteout;
93 	int		lockflag;
94 	int		error , uerror, lerror;
95 
96 	iswhiteout = 0;
97 	lockflag = 0;
98 	error = uerror = lerror = ENOENT;
99 	cnp = ap->a_cnp;
100 	nameiop = cnp->cn_nameiop;
101 	cnflags = cnp->cn_flags;
102 	dvp = ap->a_dvp;
103 	dunp = VTOUNIONFS(dvp);
104 	udvp = dunp->un_uppervp;
105 	ldvp = dunp->un_lowervp;
106 	vp = uvp = lvp = NULLVP;
107 	td = curthread;
108 	*(ap->a_vpp) = NULLVP;
109 
110 	UNIONFS_INTERNAL_DEBUG(
111 	    "unionfs_lookup: enter: nameiop=%ld, flags=%lx, path=%s\n",
112 	    nameiop, cnflags, cnp->cn_nameptr);
113 
114 	if (dvp->v_type != VDIR)
115 		return (ENOTDIR);
116 
117 	/*
118 	 * If read-only and op is not LOOKUP, will return EROFS.
119 	 */
120 	if ((cnflags & ISLASTCN) &&
121 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
122 	    LOOKUP != nameiop)
123 		return (EROFS);
124 
125 	/*
126 	 * lookup dotdot
127 	 */
128 	if (cnflags & ISDOTDOT) {
129 		if (LOOKUP != nameiop && udvp == NULLVP)
130 			return (EROFS);
131 
132 		if (udvp != NULLVP) {
133 			dtmpvp = udvp;
134 			if (ldvp != NULLVP)
135 				VOP_UNLOCK(ldvp);
136 		}
137 		else
138 			dtmpvp = ldvp;
139 
140 		error = VOP_LOOKUP(dtmpvp, &vp, cnp);
141 
142 		if (dtmpvp == udvp && ldvp != NULLVP) {
143 			VOP_UNLOCK(udvp);
144 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
145 		}
146 
147 		if (error == 0) {
148 			/*
149 			 * Exchange lock and reference from vp to
150 			 * dunp->un_dvp. vp is upper/lower vnode, but it
151 			 * will need to return the unionfs vnode.
152 			 */
153 			if (nameiop == DELETE  || nameiop == RENAME ||
154 			    (cnp->cn_lkflags & LK_TYPE_MASK))
155 				VOP_UNLOCK(vp);
156 			vrele(vp);
157 
158 			VOP_UNLOCK(dvp);
159 			*(ap->a_vpp) = dunp->un_dvp;
160 			vref(dunp->un_dvp);
161 
162 			if (nameiop == DELETE || nameiop == RENAME)
163 				vn_lock(dunp->un_dvp, LK_EXCLUSIVE | LK_RETRY);
164 			else if (cnp->cn_lkflags & LK_TYPE_MASK)
165 				vn_lock(dunp->un_dvp, cnp->cn_lkflags |
166 				    LK_RETRY);
167 
168 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
169 		} else if (error == ENOENT && (cnflags & MAKEENTRY) != 0)
170 			cache_enter(dvp, NULLVP, cnp);
171 
172 		goto unionfs_lookup_return;
173 	}
174 
175 	/*
176 	 * lookup upper layer
177 	 */
178 	if (udvp != NULLVP) {
179 		uerror = VOP_LOOKUP(udvp, &uvp, cnp);
180 
181 		if (uerror == 0) {
182 			if (udvp == uvp) {	/* is dot */
183 				vrele(uvp);
184 				*(ap->a_vpp) = dvp;
185 				vref(dvp);
186 
187 				error = uerror;
188 				goto unionfs_lookup_return;
189 			}
190 			if (nameiop == DELETE || nameiop == RENAME ||
191 			    (cnp->cn_lkflags & LK_TYPE_MASK))
192 				VOP_UNLOCK(uvp);
193 		}
194 
195 		/* check whiteout */
196 		if (uerror == ENOENT || uerror == EJUSTRETURN)
197 			if (cnp->cn_flags & ISWHITEOUT)
198 				iswhiteout = 1;	/* don't lookup lower */
199 		if (iswhiteout == 0 && ldvp != NULLVP)
200 			if (!VOP_GETATTR(udvp, &va, cnp->cn_cred) &&
201 			    (va.va_flags & OPAQUE))
202 				iswhiteout = 1;	/* don't lookup lower */
203 #if 0
204 		UNIONFS_INTERNAL_DEBUG(
205 		    "unionfs_lookup: debug: whiteout=%d, path=%s\n",
206 		    iswhiteout, cnp->cn_nameptr);
207 #endif
208 	}
209 
210 	/*
211 	 * lookup lower layer
212 	 */
213 	if (ldvp != NULLVP && !(cnflags & DOWHITEOUT) && iswhiteout == 0) {
214 		/* always op is LOOKUP */
215 		cnp->cn_nameiop = LOOKUP;
216 		cnflagsbk = cnp->cn_flags;
217 		cnp->cn_flags = cnflags;
218 
219 		lerror = VOP_LOOKUP(ldvp, &lvp, cnp);
220 
221 		cnp->cn_nameiop = nameiop;
222 		if (udvp != NULLVP && (uerror == 0 || uerror == EJUSTRETURN))
223 			cnp->cn_flags = cnflagsbk;
224 
225 		if (lerror == 0) {
226 			if (ldvp == lvp) {	/* is dot */
227 				if (uvp != NULLVP)
228 					vrele(uvp);	/* no need? */
229 				vrele(lvp);
230 				*(ap->a_vpp) = dvp;
231 				vref(dvp);
232 
233 				UNIONFS_INTERNAL_DEBUG(
234 				    "unionfs_lookup: leave (%d)\n", lerror);
235 
236 				return (lerror);
237 			}
238 			if (cnp->cn_lkflags & LK_TYPE_MASK)
239 				VOP_UNLOCK(lvp);
240 		}
241 	}
242 
243 	/*
244 	 * check lookup result
245 	 */
246 	if (uvp == NULLVP && lvp == NULLVP) {
247 		error = (udvp != NULLVP ? uerror : lerror);
248 		goto unionfs_lookup_return;
249 	}
250 
251 	/*
252 	 * check vnode type
253 	 */
254 	if (uvp != NULLVP && lvp != NULLVP && uvp->v_type != lvp->v_type) {
255 		vrele(lvp);
256 		lvp = NULLVP;
257 	}
258 
259 	/*
260 	 * check shadow dir
261 	 */
262 	if (uerror != 0 && uerror != EJUSTRETURN && udvp != NULLVP &&
263 	    lerror == 0 && lvp != NULLVP && lvp->v_type == VDIR &&
264 	    !(dvp->v_mount->mnt_flag & MNT_RDONLY) &&
265 	    (1 < cnp->cn_namelen || '.' != *(cnp->cn_nameptr))) {
266 		/* get unionfs vnode in order to create a new shadow dir. */
267 		error = unionfs_nodeget(dvp->v_mount, NULLVP, lvp, dvp, &vp,
268 		    cnp);
269 		if (error != 0)
270 			goto unionfs_lookup_cleanup;
271 
272 		if (LK_SHARED == (cnp->cn_lkflags & LK_TYPE_MASK))
273 			VOP_UNLOCK(vp);
274 		if (LK_EXCLUSIVE != VOP_ISLOCKED(vp)) {
275 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
276 			lockflag = 1;
277 		}
278 		error = unionfs_mkshadowdir(MOUNTTOUNIONFSMOUNT(dvp->v_mount),
279 		    udvp, VTOUNIONFS(vp), cnp, td);
280 		if (lockflag != 0)
281 			VOP_UNLOCK(vp);
282 		if (error != 0) {
283 			UNIONFSDEBUG(
284 			    "unionfs_lookup: Unable to create shadow dir.");
285 			if ((cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE)
286 				vput(vp);
287 			else
288 				vrele(vp);
289 			goto unionfs_lookup_cleanup;
290 		}
291 		if ((cnp->cn_lkflags & LK_TYPE_MASK) == LK_SHARED)
292 			vn_lock(vp, LK_SHARED | LK_RETRY);
293 	}
294 	/*
295 	 * get unionfs vnode.
296 	 */
297 	else {
298 		if (uvp != NULLVP)
299 			error = uerror;
300 		else
301 			error = lerror;
302 		if (error != 0)
303 			goto unionfs_lookup_cleanup;
304 		/*
305 		 * get socket vnode.
306 		 */
307 		if (uvp != NULLVP && uvp->v_type == VSOCK) {
308 			vp = uvp;
309 			vref(vp);
310 			if (cnp->cn_lkflags & LK_TYPE_MASK)
311 				vn_lock(vp, cnp->cn_lkflags | LK_RETRY);
312 		}
313 		else if (lvp != NULLVP && lvp->v_type == VSOCK) {
314 			vp = lvp;
315 			vref(vp);
316 			if (cnp->cn_lkflags & LK_TYPE_MASK)
317 				vn_lock(vp, cnp->cn_lkflags | LK_RETRY);
318 		}
319 		/*
320 		 * get unionfs vnode.
321 		 */
322 		else
323 			error = unionfs_nodeget(dvp->v_mount, uvp, lvp,
324 			    dvp, &vp, cnp);
325 		if (error != 0) {
326 			UNIONFSDEBUG(
327 			    "unionfs_lookup: Unable to create unionfs vnode.");
328 			goto unionfs_lookup_cleanup;
329 		}
330 		if ((nameiop == DELETE || nameiop == RENAME) &&
331 		    (cnp->cn_lkflags & LK_TYPE_MASK) == 0)
332 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
333 	}
334 
335 	*(ap->a_vpp) = vp;
336 
337 	if ((cnflags & MAKEENTRY) && vp->v_type != VSOCK)
338 		cache_enter(dvp, vp, cnp);
339 
340 unionfs_lookup_cleanup:
341 	if (uvp != NULLVP)
342 		vrele(uvp);
343 	if (lvp != NULLVP)
344 		vrele(lvp);
345 
346 	if (error == ENOENT && (cnflags & MAKEENTRY) != 0)
347 		cache_enter(dvp, NULLVP, cnp);
348 
349 unionfs_lookup_return:
350 
351 	UNIONFS_INTERNAL_DEBUG("unionfs_lookup: leave (%d)\n", error);
352 
353 	return (error);
354 }
355 
356 static int
357 unionfs_create(struct vop_create_args *ap)
358 {
359 	struct unionfs_node *dunp;
360 	struct componentname *cnp;
361 	struct vnode   *udvp;
362 	struct vnode   *vp;
363 	int		error;
364 
365 	UNIONFS_INTERNAL_DEBUG("unionfs_create: enter\n");
366 
367 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
368 
369 	dunp = VTOUNIONFS(ap->a_dvp);
370 	cnp = ap->a_cnp;
371 	udvp = dunp->un_uppervp;
372 	error = EROFS;
373 
374 	if (udvp != NULLVP) {
375 		error = VOP_CREATE(udvp, &vp, cnp, ap->a_vap);
376 		if (error != 0)
377 			goto unionfs_create_abort;
378 
379 		if (vp->v_type == VSOCK)
380 			*(ap->a_vpp) = vp;
381 		else {
382 			VOP_UNLOCK(vp);
383 			error = unionfs_nodeget(ap->a_dvp->v_mount, vp, NULLVP,
384 			    ap->a_dvp, ap->a_vpp, cnp);
385 			vrele(vp);
386 		}
387 	}
388 
389 unionfs_create_abort:
390 	UNIONFS_INTERNAL_DEBUG("unionfs_create: leave (%d)\n", error);
391 
392 	return (error);
393 }
394 
395 static int
396 unionfs_whiteout(struct vop_whiteout_args *ap)
397 {
398 	struct unionfs_node *dunp;
399 	struct componentname *cnp;
400 	struct vnode   *udvp;
401 	int		error;
402 
403 	UNIONFS_INTERNAL_DEBUG("unionfs_whiteout: enter\n");
404 
405 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
406 
407 	dunp = VTOUNIONFS(ap->a_dvp);
408 	cnp = ap->a_cnp;
409 	udvp = dunp->un_uppervp;
410 	error = EOPNOTSUPP;
411 
412 	if (udvp != NULLVP) {
413 		switch (ap->a_flags) {
414 		case CREATE:
415 		case DELETE:
416 		case LOOKUP:
417 			error = VOP_WHITEOUT(udvp, cnp, ap->a_flags);
418 			break;
419 		default:
420 			error = EINVAL;
421 			break;
422 		}
423 	}
424 
425 	UNIONFS_INTERNAL_DEBUG("unionfs_whiteout: leave (%d)\n", error);
426 
427 	return (error);
428 }
429 
430 static int
431 unionfs_mknod(struct vop_mknod_args *ap)
432 {
433 	struct unionfs_node *dunp;
434 	struct componentname *cnp;
435 	struct vnode   *udvp;
436 	struct vnode   *vp;
437 	int		error;
438 
439 	UNIONFS_INTERNAL_DEBUG("unionfs_mknod: enter\n");
440 
441 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
442 
443 	dunp = VTOUNIONFS(ap->a_dvp);
444 	cnp = ap->a_cnp;
445 	udvp = dunp->un_uppervp;
446 	error = EROFS;
447 
448 	if (udvp != NULLVP) {
449 		error = VOP_MKNOD(udvp, &vp, cnp, ap->a_vap);
450 		if (error != 0)
451 			goto unionfs_mknod_abort;
452 
453 		if (vp->v_type == VSOCK)
454 			*(ap->a_vpp) = vp;
455 		else {
456 			VOP_UNLOCK(vp);
457 			error = unionfs_nodeget(ap->a_dvp->v_mount, vp, NULLVP,
458 			    ap->a_dvp, ap->a_vpp, cnp);
459 			vrele(vp);
460 		}
461 	}
462 
463 unionfs_mknod_abort:
464 	UNIONFS_INTERNAL_DEBUG("unionfs_mknod: leave (%d)\n", error);
465 
466 	return (error);
467 }
468 
469 enum unionfs_lkupgrade {
470 	UNIONFS_LKUPGRADE_SUCCESS, /* lock successfully upgraded */
471 	UNIONFS_LKUPGRADE_ALREADY, /* lock already held exclusive */
472 	UNIONFS_LKUPGRADE_DOOMED   /* lock was upgraded, but vnode reclaimed */
473 };
474 
475 static inline enum unionfs_lkupgrade
476 unionfs_upgrade_lock(struct vnode *vp)
477 {
478 	ASSERT_VOP_LOCKED(vp, __func__);
479 
480 	if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
481 		return (UNIONFS_LKUPGRADE_ALREADY);
482 
483 	if (vn_lock(vp, LK_UPGRADE) != 0) {
484 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
485 		if (VN_IS_DOOMED(vp))
486 			return (UNIONFS_LKUPGRADE_DOOMED);
487 	}
488 	return (UNIONFS_LKUPGRADE_SUCCESS);
489 }
490 
491 static inline void
492 unionfs_downgrade_lock(struct vnode *vp, enum unionfs_lkupgrade status)
493 {
494 	if (status != UNIONFS_LKUPGRADE_ALREADY)
495 		vn_lock(vp, LK_DOWNGRADE | LK_RETRY);
496 }
497 
498 static int
499 unionfs_open(struct vop_open_args *ap)
500 {
501 	struct unionfs_node *unp;
502 	struct unionfs_node_status *unsp;
503 	struct vnode   *vp;
504 	struct vnode   *uvp;
505 	struct vnode   *lvp;
506 	struct vnode   *targetvp;
507 	struct ucred   *cred;
508 	struct thread  *td;
509 	int		error;
510 	enum unionfs_lkupgrade lkstatus;
511 
512 	UNIONFS_INTERNAL_DEBUG("unionfs_open: enter\n");
513 
514 	KASSERT_UNIONFS_VNODE(ap->a_vp);
515 
516 	error = 0;
517 	vp = ap->a_vp;
518 	targetvp = NULLVP;
519 	cred = ap->a_cred;
520 	td = ap->a_td;
521 
522 	/*
523 	 * The executable loader path may call this function with vp locked
524 	 * shared.  If the vnode is reclaimed while upgrading, we can't safely
525 	 * use unp or do anything else unionfs- specific.
526 	 */
527 	lkstatus = unionfs_upgrade_lock(vp);
528 	if (lkstatus == UNIONFS_LKUPGRADE_DOOMED) {
529 		error = ENOENT;
530 		goto unionfs_open_cleanup;
531 	}
532 
533 	unp = VTOUNIONFS(vp);
534 	uvp = unp->un_uppervp;
535 	lvp = unp->un_lowervp;
536 	unionfs_get_node_status(unp, td, &unsp);
537 
538 	if (unsp->uns_lower_opencnt > 0 || unsp->uns_upper_opencnt > 0) {
539 		/* vnode is already opend. */
540 		if (unsp->uns_upper_opencnt > 0)
541 			targetvp = uvp;
542 		else
543 			targetvp = lvp;
544 
545 		if (targetvp == lvp &&
546 		    (ap->a_mode & FWRITE) && lvp->v_type == VREG)
547 			targetvp = NULLVP;
548 	}
549 	if (targetvp == NULLVP) {
550 		if (uvp == NULLVP) {
551 			if ((ap->a_mode & FWRITE) && lvp->v_type == VREG) {
552 				error = unionfs_copyfile(unp,
553 				    !(ap->a_mode & O_TRUNC), cred, td);
554 				if (error != 0)
555 					goto unionfs_open_abort;
556 				targetvp = uvp = unp->un_uppervp;
557 			} else
558 				targetvp = lvp;
559 		} else
560 			targetvp = uvp;
561 	}
562 
563 	error = VOP_OPEN(targetvp, ap->a_mode, cred, td, ap->a_fp);
564 	if (error == 0) {
565 		if (targetvp == uvp) {
566 			if (uvp->v_type == VDIR && lvp != NULLVP &&
567 			    unsp->uns_lower_opencnt <= 0) {
568 				/* open lower for readdir */
569 				error = VOP_OPEN(lvp, FREAD, cred, td, NULL);
570 				if (error != 0) {
571 					VOP_CLOSE(uvp, ap->a_mode, cred, td);
572 					goto unionfs_open_abort;
573 				}
574 				unsp->uns_node_flag |= UNS_OPENL_4_READDIR;
575 				unsp->uns_lower_opencnt++;
576 			}
577 			unsp->uns_upper_opencnt++;
578 		} else {
579 			unsp->uns_lower_opencnt++;
580 			unsp->uns_lower_openmode = ap->a_mode;
581 		}
582 		vp->v_object = targetvp->v_object;
583 	}
584 
585 unionfs_open_abort:
586 	if (error != 0)
587 		unionfs_tryrem_node_status(unp, unsp);
588 
589 unionfs_open_cleanup:
590 	unionfs_downgrade_lock(vp, lkstatus);
591 
592 	UNIONFS_INTERNAL_DEBUG("unionfs_open: leave (%d)\n", error);
593 
594 	return (error);
595 }
596 
597 static int
598 unionfs_close(struct vop_close_args *ap)
599 {
600 	struct unionfs_node *unp;
601 	struct unionfs_node_status *unsp;
602 	struct ucred   *cred;
603 	struct thread  *td;
604 	struct vnode   *vp;
605 	struct vnode   *ovp;
606 	int		error;
607 	enum unionfs_lkupgrade lkstatus;;
608 
609 	UNIONFS_INTERNAL_DEBUG("unionfs_close: enter\n");
610 
611 	KASSERT_UNIONFS_VNODE(ap->a_vp);
612 
613 	vp = ap->a_vp;
614 	cred = ap->a_cred;
615 	td = ap->a_td;
616 	error = 0;
617 
618 	/*
619 	 * If the vnode is reclaimed while upgrading, we can't safely use unp
620 	 * or do anything else unionfs- specific.
621 	 */
622 	lkstatus = unionfs_upgrade_lock(vp);
623 	if (lkstatus == UNIONFS_LKUPGRADE_DOOMED)
624 		goto unionfs_close_cleanup;
625 
626 	unp = VTOUNIONFS(vp);
627 	unionfs_get_node_status(unp, td, &unsp);
628 
629 	if (unsp->uns_lower_opencnt <= 0 && unsp->uns_upper_opencnt <= 0) {
630 #ifdef DIAGNOSTIC
631 		printf("unionfs_close: warning: open count is 0\n");
632 #endif
633 		if (unp->un_uppervp != NULLVP)
634 			ovp = unp->un_uppervp;
635 		else
636 			ovp = unp->un_lowervp;
637 	} else if (unsp->uns_upper_opencnt > 0)
638 		ovp = unp->un_uppervp;
639 	else
640 		ovp = unp->un_lowervp;
641 
642 	error = VOP_CLOSE(ovp, ap->a_fflag, cred, td);
643 
644 	if (error != 0)
645 		goto unionfs_close_abort;
646 
647 	vp->v_object = ovp->v_object;
648 
649 	if (ovp == unp->un_uppervp) {
650 		unsp->uns_upper_opencnt--;
651 		if (unsp->uns_upper_opencnt == 0) {
652 			if (unsp->uns_node_flag & UNS_OPENL_4_READDIR) {
653 				VOP_CLOSE(unp->un_lowervp, FREAD, cred, td);
654 				unsp->uns_node_flag &= ~UNS_OPENL_4_READDIR;
655 				unsp->uns_lower_opencnt--;
656 			}
657 			if (unsp->uns_lower_opencnt > 0)
658 				vp->v_object = unp->un_lowervp->v_object;
659 		}
660 	} else
661 		unsp->uns_lower_opencnt--;
662 
663 unionfs_close_abort:
664 	unionfs_tryrem_node_status(unp, unsp);
665 
666 unionfs_close_cleanup:
667 	unionfs_downgrade_lock(vp, lkstatus);
668 
669 	UNIONFS_INTERNAL_DEBUG("unionfs_close: leave (%d)\n", error);
670 
671 	return (error);
672 }
673 
674 /*
675  * Check the access mode toward shadow file/dir.
676  */
677 static int
678 unionfs_check_corrected_access(accmode_t accmode, struct vattr *va,
679     struct ucred *cred)
680 {
681 	uid_t		uid;	/* upper side vnode's uid */
682 	gid_t		gid;	/* upper side vnode's gid */
683 	u_short		vmode;	/* upper side vnode's mode */
684 	u_short		mask;
685 
686 	mask = 0;
687 	uid = va->va_uid;
688 	gid = va->va_gid;
689 	vmode = va->va_mode;
690 
691 	/* check owner */
692 	if (cred->cr_uid == uid) {
693 		if (accmode & VEXEC)
694 			mask |= S_IXUSR;
695 		if (accmode & VREAD)
696 			mask |= S_IRUSR;
697 		if (accmode & VWRITE)
698 			mask |= S_IWUSR;
699 		return ((vmode & mask) == mask ? 0 : EACCES);
700 	}
701 
702 	/* check group */
703 	if (groupmember(gid, cred)) {
704 		if (accmode & VEXEC)
705 			mask |= S_IXGRP;
706 		if (accmode & VREAD)
707 			mask |= S_IRGRP;
708 		if (accmode & VWRITE)
709 			mask |= S_IWGRP;
710 		return ((vmode & mask) == mask ? 0 : EACCES);
711 	}
712 
713 	/* check other */
714 	if (accmode & VEXEC)
715 		mask |= S_IXOTH;
716 	if (accmode & VREAD)
717 		mask |= S_IROTH;
718 	if (accmode & VWRITE)
719 		mask |= S_IWOTH;
720 
721 	return ((vmode & mask) == mask ? 0 : EACCES);
722 }
723 
724 static int
725 unionfs_access(struct vop_access_args *ap)
726 {
727 	struct unionfs_mount *ump;
728 	struct unionfs_node *unp;
729 	struct vnode   *uvp;
730 	struct vnode   *lvp;
731 	struct thread  *td;
732 	struct vattr	va;
733 	accmode_t	accmode;
734 	int		error;
735 
736 	UNIONFS_INTERNAL_DEBUG("unionfs_access: enter\n");
737 
738 	KASSERT_UNIONFS_VNODE(ap->a_vp);
739 
740 	ump = MOUNTTOUNIONFSMOUNT(ap->a_vp->v_mount);
741 	unp = VTOUNIONFS(ap->a_vp);
742 	uvp = unp->un_uppervp;
743 	lvp = unp->un_lowervp;
744 	td = ap->a_td;
745 	accmode = ap->a_accmode;
746 	error = EACCES;
747 
748 	if ((accmode & VWRITE) &&
749 	    (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)) {
750 		switch (ap->a_vp->v_type) {
751 		case VREG:
752 		case VDIR:
753 		case VLNK:
754 			return (EROFS);
755 		default:
756 			break;
757 		}
758 	}
759 
760 	if (uvp != NULLVP) {
761 		error = VOP_ACCESS(uvp, accmode, ap->a_cred, td);
762 
763 		UNIONFS_INTERNAL_DEBUG("unionfs_access: leave (%d)\n", error);
764 
765 		return (error);
766 	}
767 
768 	if (lvp != NULLVP) {
769 		if (accmode & VWRITE) {
770 			if (ump->um_uppervp->v_mount->mnt_flag & MNT_RDONLY) {
771 				switch (ap->a_vp->v_type) {
772 				case VREG:
773 				case VDIR:
774 				case VLNK:
775 					return (EROFS);
776 				default:
777 					break;
778 				}
779 			} else if (ap->a_vp->v_type == VREG ||
780 			    ap->a_vp->v_type == VDIR) {
781 				/* check shadow file/dir */
782 				if (ump->um_copymode != UNIONFS_TRANSPARENT) {
783 					error = unionfs_create_uppervattr(ump,
784 					    lvp, &va, ap->a_cred, td);
785 					if (error != 0)
786 						return (error);
787 
788 					error = unionfs_check_corrected_access(
789 					    accmode, &va, ap->a_cred);
790 					if (error != 0)
791 						return (error);
792 				}
793 			}
794 			accmode &= ~(VWRITE | VAPPEND);
795 			accmode |= VREAD; /* will copy to upper */
796 		}
797 		error = VOP_ACCESS(lvp, accmode, ap->a_cred, td);
798 	}
799 
800 	UNIONFS_INTERNAL_DEBUG("unionfs_access: leave (%d)\n", error);
801 
802 	return (error);
803 }
804 
805 static int
806 unionfs_getattr(struct vop_getattr_args *ap)
807 {
808 	struct unionfs_node *unp;
809 	struct unionfs_mount *ump;
810 	struct vnode   *uvp;
811 	struct vnode   *lvp;
812 	struct thread  *td;
813 	struct vattr	va;
814 	int		error;
815 
816 	UNIONFS_INTERNAL_DEBUG("unionfs_getattr: enter\n");
817 
818 	KASSERT_UNIONFS_VNODE(ap->a_vp);
819 
820 	unp = VTOUNIONFS(ap->a_vp);
821 	ump = MOUNTTOUNIONFSMOUNT(ap->a_vp->v_mount);
822 	uvp = unp->un_uppervp;
823 	lvp = unp->un_lowervp;
824 	td = curthread;
825 
826 	if (uvp != NULLVP) {
827 		if ((error = VOP_GETATTR(uvp, ap->a_vap, ap->a_cred)) == 0)
828 			ap->a_vap->va_fsid =
829 			    ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
830 
831 		UNIONFS_INTERNAL_DEBUG(
832 		    "unionfs_getattr: leave mode=%o, uid=%d, gid=%d (%d)\n",
833 		    ap->a_vap->va_mode, ap->a_vap->va_uid,
834 		    ap->a_vap->va_gid, error);
835 
836 		return (error);
837 	}
838 
839 	error = VOP_GETATTR(lvp, ap->a_vap, ap->a_cred);
840 
841 	if (error == 0 && !(ump->um_uppervp->v_mount->mnt_flag & MNT_RDONLY)) {
842 		/* correct the attr toward shadow file/dir. */
843 		if (ap->a_vp->v_type == VREG || ap->a_vp->v_type == VDIR) {
844 			unionfs_create_uppervattr_core(ump, ap->a_vap, &va, td);
845 			ap->a_vap->va_mode = va.va_mode;
846 			ap->a_vap->va_uid = va.va_uid;
847 			ap->a_vap->va_gid = va.va_gid;
848 		}
849 	}
850 
851 	if (error == 0)
852 		ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
853 
854 	UNIONFS_INTERNAL_DEBUG(
855 	    "unionfs_getattr: leave mode=%o, uid=%d, gid=%d (%d)\n",
856 	    ap->a_vap->va_mode, ap->a_vap->va_uid, ap->a_vap->va_gid, error);
857 
858 	return (error);
859 }
860 
861 static int
862 unionfs_setattr(struct vop_setattr_args *ap)
863 {
864 	struct unionfs_node *unp;
865 	struct vnode   *uvp;
866 	struct vnode   *lvp;
867 	struct thread  *td;
868 	struct vattr   *vap;
869 	int		error;
870 
871 	UNIONFS_INTERNAL_DEBUG("unionfs_setattr: enter\n");
872 
873 	KASSERT_UNIONFS_VNODE(ap->a_vp);
874 
875 	error = EROFS;
876 	unp = VTOUNIONFS(ap->a_vp);
877 	uvp = unp->un_uppervp;
878 	lvp = unp->un_lowervp;
879 	td = curthread;
880 	vap = ap->a_vap;
881 
882 	if ((ap->a_vp->v_mount->mnt_flag & MNT_RDONLY) &&
883 	    (vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
884 	     vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
885 	     vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL))
886 		return (EROFS);
887 
888 	if (uvp == NULLVP && lvp->v_type == VREG) {
889 		error = unionfs_copyfile(unp, (vap->va_size != 0),
890 		    ap->a_cred, td);
891 		if (error != 0)
892 			return (error);
893 		uvp = unp->un_uppervp;
894 	}
895 
896 	if (uvp != NULLVP)
897 		error = VOP_SETATTR(uvp, vap, ap->a_cred);
898 
899 	UNIONFS_INTERNAL_DEBUG("unionfs_setattr: leave (%d)\n", error);
900 
901 	return (error);
902 }
903 
904 static int
905 unionfs_read(struct vop_read_args *ap)
906 {
907 	struct unionfs_node *unp;
908 	struct vnode   *tvp;
909 	int		error;
910 
911 	/* UNIONFS_INTERNAL_DEBUG("unionfs_read: enter\n"); */
912 
913 	KASSERT_UNIONFS_VNODE(ap->a_vp);
914 
915 	unp = VTOUNIONFS(ap->a_vp);
916 	tvp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
917 
918 	error = VOP_READ(tvp, ap->a_uio, ap->a_ioflag, ap->a_cred);
919 
920 	/* UNIONFS_INTERNAL_DEBUG("unionfs_read: leave (%d)\n", error); */
921 
922 	return (error);
923 }
924 
925 static int
926 unionfs_write(struct vop_write_args *ap)
927 {
928 	struct unionfs_node *unp;
929 	struct vnode   *tvp;
930 	int		error;
931 
932 	/* UNIONFS_INTERNAL_DEBUG("unionfs_write: enter\n"); */
933 
934 	KASSERT_UNIONFS_VNODE(ap->a_vp);
935 
936 	unp = VTOUNIONFS(ap->a_vp);
937 	tvp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
938 
939 	error = VOP_WRITE(tvp, ap->a_uio, ap->a_ioflag, ap->a_cred);
940 
941 	/* UNIONFS_INTERNAL_DEBUG("unionfs_write: leave (%d)\n", error); */
942 
943 	return (error);
944 }
945 
946 static int
947 unionfs_ioctl(struct vop_ioctl_args *ap)
948 {
949 	struct unionfs_node *unp;
950 	struct unionfs_node_status *unsp;
951 	struct vnode   *ovp;
952 	int error;
953 
954 	UNIONFS_INTERNAL_DEBUG("unionfs_ioctl: enter\n");
955 
956 	KASSERT_UNIONFS_VNODE(ap->a_vp);
957 
958  	vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
959 	unp = VTOUNIONFS(ap->a_vp);
960 	unionfs_get_node_status(unp, ap->a_td, &unsp);
961 	ovp = (unsp->uns_upper_opencnt ? unp->un_uppervp : unp->un_lowervp);
962 	unionfs_tryrem_node_status(unp, unsp);
963 	VOP_UNLOCK(ap->a_vp);
964 
965 	if (ovp == NULLVP)
966 		return (EBADF);
967 
968 	error = VOP_IOCTL(ovp, ap->a_command, ap->a_data, ap->a_fflag,
969 	    ap->a_cred, ap->a_td);
970 
971 	UNIONFS_INTERNAL_DEBUG("unionfs_ioctl: leave (%d)\n", error);
972 
973 	return (error);
974 }
975 
976 static int
977 unionfs_poll(struct vop_poll_args *ap)
978 {
979 	struct unionfs_node *unp;
980 	struct unionfs_node_status *unsp;
981 	struct vnode *ovp;
982 
983 	KASSERT_UNIONFS_VNODE(ap->a_vp);
984 
985  	vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
986 	unp = VTOUNIONFS(ap->a_vp);
987 	unionfs_get_node_status(unp, ap->a_td, &unsp);
988 	ovp = (unsp->uns_upper_opencnt ? unp->un_uppervp : unp->un_lowervp);
989 	unionfs_tryrem_node_status(unp, unsp);
990 	VOP_UNLOCK(ap->a_vp);
991 
992 	if (ovp == NULLVP)
993 		return (EBADF);
994 
995 	return (VOP_POLL(ovp, ap->a_events, ap->a_cred, ap->a_td));
996 }
997 
998 static int
999 unionfs_fsync(struct vop_fsync_args *ap)
1000 {
1001 	struct unionfs_node *unp;
1002 	struct unionfs_node_status *unsp;
1003 	struct vnode *ovp;
1004 
1005 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1006 
1007 	unp = VTOUNIONFS(ap->a_vp);
1008 	unionfs_get_node_status(unp, ap->a_td, &unsp);
1009 	ovp = (unsp->uns_upper_opencnt ? unp->un_uppervp : unp->un_lowervp);
1010 	unionfs_tryrem_node_status(unp, unsp);
1011 
1012 	if (ovp == NULLVP)
1013 		return (EBADF);
1014 
1015 	return (VOP_FSYNC(ovp, ap->a_waitfor, ap->a_td));
1016 }
1017 
1018 static int
1019 unionfs_remove(struct vop_remove_args *ap)
1020 {
1021 	char	       *path;
1022 	struct unionfs_node *dunp;
1023 	struct unionfs_node *unp;
1024 	struct unionfs_mount *ump;
1025 	struct vnode   *udvp;
1026 	struct vnode   *uvp;
1027 	struct vnode   *lvp;
1028 	struct vnode   *vp;
1029 	struct componentname *cnp;
1030 	struct componentname cn;
1031 	struct thread  *td;
1032 	int		error;
1033 	int		pathlen;
1034 
1035 	UNIONFS_INTERNAL_DEBUG("unionfs_remove: enter\n");
1036 
1037 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
1038 
1039 	error = 0;
1040 	dunp = VTOUNIONFS(ap->a_dvp);
1041 	udvp = dunp->un_uppervp;
1042 	cnp = ap->a_cnp;
1043 	td = curthread;
1044 
1045 	if (ap->a_vp->v_op != &unionfs_vnodeops) {
1046 		if (ap->a_vp->v_type != VSOCK)
1047 			return (EINVAL);
1048 		ump = NULL;
1049 		vp = uvp = lvp = NULLVP;
1050 		/* search vnode */
1051 		VOP_UNLOCK(ap->a_vp);
1052 		error = unionfs_relookup(udvp, &vp, cnp, &cn, td,
1053 		    cnp->cn_nameptr, cnp->cn_namelen, DELETE);
1054 		if (error != 0 && error != ENOENT) {
1055 			vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
1056 			return (error);
1057 		}
1058 
1059 		if (error == 0 && vp == ap->a_vp) {
1060 			/* target vnode in upper */
1061 			uvp = vp;
1062 			vrele(vp);
1063 		} else {
1064 			/* target vnode in lower */
1065 			if (vp != NULLVP) {
1066 				if (udvp == vp)
1067 					vrele(vp);
1068 				else
1069 					vput(vp);
1070 			}
1071 			vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
1072 			lvp = ap->a_vp;
1073 		}
1074 		path = cnp->cn_nameptr;
1075 		pathlen = cnp->cn_namelen;
1076 	} else {
1077 		ump = MOUNTTOUNIONFSMOUNT(ap->a_vp->v_mount);
1078 		unp = VTOUNIONFS(ap->a_vp);
1079 		uvp = unp->un_uppervp;
1080 		lvp = unp->un_lowervp;
1081 		path = unp->un_path;
1082 		pathlen = unp->un_pathlen;
1083 	}
1084 
1085 	if (udvp == NULLVP)
1086 		return (EROFS);
1087 
1088 	if (uvp != NULLVP) {
1089 		/*
1090 		 * XXX: if the vnode type is VSOCK, it will create whiteout
1091 		 *      after remove.
1092 		 */
1093 		if (ump == NULL || ump->um_whitemode == UNIONFS_WHITE_ALWAYS ||
1094 		    lvp != NULLVP)
1095 			cnp->cn_flags |= DOWHITEOUT;
1096 		error = VOP_REMOVE(udvp, uvp, cnp);
1097 	} else if (lvp != NULLVP)
1098 		error = unionfs_mkwhiteout(udvp, cnp, td, path, pathlen);
1099 
1100 	UNIONFS_INTERNAL_DEBUG("unionfs_remove: leave (%d)\n", error);
1101 
1102 	return (error);
1103 }
1104 
1105 static int
1106 unionfs_link(struct vop_link_args *ap)
1107 {
1108 	struct unionfs_node *dunp;
1109 	struct unionfs_node *unp;
1110 	struct vnode   *udvp;
1111 	struct vnode   *uvp;
1112 	struct componentname *cnp;
1113 	struct thread  *td;
1114 	int		error;
1115 	int		needrelookup;
1116 
1117 	UNIONFS_INTERNAL_DEBUG("unionfs_link: enter\n");
1118 
1119 	KASSERT_UNIONFS_VNODE(ap->a_tdvp);
1120 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1121 
1122 	error = 0;
1123 	needrelookup = 0;
1124 	dunp = VTOUNIONFS(ap->a_tdvp);
1125 	unp = NULL;
1126 	udvp = dunp->un_uppervp;
1127 	uvp = NULLVP;
1128 	cnp = ap->a_cnp;
1129 	td = curthread;
1130 
1131 	if (udvp == NULLVP)
1132 		return (EROFS);
1133 
1134 	if (ap->a_vp->v_op != &unionfs_vnodeops)
1135 		uvp = ap->a_vp;
1136 	else {
1137 		unp = VTOUNIONFS(ap->a_vp);
1138 
1139 		if (unp->un_uppervp == NULLVP) {
1140 			if (ap->a_vp->v_type != VREG)
1141 				return (EOPNOTSUPP);
1142 
1143 			error = unionfs_copyfile(unp, 1, cnp->cn_cred, td);
1144 			if (error != 0)
1145 				return (error);
1146 			needrelookup = 1;
1147 		}
1148 		uvp = unp->un_uppervp;
1149 	}
1150 
1151 	if (needrelookup != 0)
1152 		error = unionfs_relookup_for_create(ap->a_tdvp, cnp, td);
1153 
1154 	if (error == 0)
1155 		error = VOP_LINK(udvp, uvp, cnp);
1156 
1157 	UNIONFS_INTERNAL_DEBUG("unionfs_link: leave (%d)\n", error);
1158 
1159 	return (error);
1160 }
1161 
1162 static int
1163 unionfs_rename(struct vop_rename_args *ap)
1164 {
1165 	struct vnode   *fdvp;
1166 	struct vnode   *fvp;
1167 	struct componentname *fcnp;
1168 	struct vnode   *tdvp;
1169 	struct vnode   *tvp;
1170 	struct componentname *tcnp;
1171 	struct vnode   *ltdvp;
1172 	struct vnode   *ltvp;
1173 	struct thread  *td;
1174 
1175 	/* rename target vnodes */
1176 	struct vnode   *rfdvp;
1177 	struct vnode   *rfvp;
1178 	struct vnode   *rtdvp;
1179 	struct vnode   *rtvp;
1180 
1181 	struct unionfs_mount *ump;
1182 	struct unionfs_node *unp;
1183 	int		error;
1184 	int		needrelookup;
1185 
1186 	UNIONFS_INTERNAL_DEBUG("unionfs_rename: enter\n");
1187 
1188 	error = 0;
1189 	fdvp = ap->a_fdvp;
1190 	fvp = ap->a_fvp;
1191 	fcnp = ap->a_fcnp;
1192 	tdvp = ap->a_tdvp;
1193 	tvp = ap->a_tvp;
1194 	tcnp = ap->a_tcnp;
1195 	ltdvp = NULLVP;
1196 	ltvp = NULLVP;
1197 	td = curthread;
1198 	rfdvp = fdvp;
1199 	rfvp = fvp;
1200 	rtdvp = tdvp;
1201 	rtvp = tvp;
1202 	needrelookup = 0;
1203 
1204 	/* check for cross device rename */
1205 	if (fvp->v_mount != tdvp->v_mount ||
1206 	    (tvp != NULLVP && fvp->v_mount != tvp->v_mount)) {
1207 		if (fvp->v_op != &unionfs_vnodeops)
1208 			error = ENODEV;
1209 		else
1210 			error = EXDEV;
1211 		goto unionfs_rename_abort;
1212 	}
1213 
1214 	/* Renaming a file to itself has no effect. */
1215 	if (fvp == tvp)
1216 		goto unionfs_rename_abort;
1217 
1218 	/*
1219 	 * from/to vnode is unionfs node.
1220 	 */
1221 
1222 	KASSERT_UNIONFS_VNODE(fdvp);
1223 	KASSERT_UNIONFS_VNODE(fvp);
1224 	KASSERT_UNIONFS_VNODE(tdvp);
1225 	if (tvp != NULLVP)
1226 		KASSERT_UNIONFS_VNODE(tvp);
1227 
1228 	unp = VTOUNIONFS(fdvp);
1229 #ifdef UNIONFS_IDBG_RENAME
1230 	UNIONFS_INTERNAL_DEBUG("fdvp=%p, ufdvp=%p, lfdvp=%p\n",
1231 	    fdvp, unp->un_uppervp, unp->un_lowervp);
1232 #endif
1233 	if (unp->un_uppervp == NULLVP) {
1234 		error = ENODEV;
1235 		goto unionfs_rename_abort;
1236 	}
1237 	rfdvp = unp->un_uppervp;
1238 	vref(rfdvp);
1239 
1240 	unp = VTOUNIONFS(fvp);
1241 #ifdef UNIONFS_IDBG_RENAME
1242 	UNIONFS_INTERNAL_DEBUG("fvp=%p, ufvp=%p, lfvp=%p\n",
1243 	    fvp, unp->un_uppervp, unp->un_lowervp);
1244 #endif
1245 	ump = MOUNTTOUNIONFSMOUNT(fvp->v_mount);
1246 	if (unp->un_uppervp == NULLVP) {
1247 		switch (fvp->v_type) {
1248 		case VREG:
1249 			if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1250 				goto unionfs_rename_abort;
1251 			error = unionfs_copyfile(unp, 1, fcnp->cn_cred, td);
1252 			VOP_UNLOCK(fvp);
1253 			if (error != 0)
1254 				goto unionfs_rename_abort;
1255 			break;
1256 		case VDIR:
1257 			if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1258 				goto unionfs_rename_abort;
1259 			error = unionfs_mkshadowdir(ump, rfdvp, unp, fcnp, td);
1260 			VOP_UNLOCK(fvp);
1261 			if (error != 0)
1262 				goto unionfs_rename_abort;
1263 			break;
1264 		default:
1265 			error = ENODEV;
1266 			goto unionfs_rename_abort;
1267 		}
1268 
1269 		needrelookup = 1;
1270 	}
1271 
1272 	if (unp->un_lowervp != NULLVP)
1273 		fcnp->cn_flags |= DOWHITEOUT;
1274 	rfvp = unp->un_uppervp;
1275 	vref(rfvp);
1276 
1277 	unp = VTOUNIONFS(tdvp);
1278 #ifdef UNIONFS_IDBG_RENAME
1279 	UNIONFS_INTERNAL_DEBUG("tdvp=%p, utdvp=%p, ltdvp=%p\n",
1280 	    tdvp, unp->un_uppervp, unp->un_lowervp);
1281 #endif
1282 	if (unp->un_uppervp == NULLVP) {
1283 		error = ENODEV;
1284 		goto unionfs_rename_abort;
1285 	}
1286 	rtdvp = unp->un_uppervp;
1287 	ltdvp = unp->un_lowervp;
1288 	vref(rtdvp);
1289 
1290 	if (tdvp == tvp) {
1291 		rtvp = rtdvp;
1292 		vref(rtvp);
1293 	} else if (tvp != NULLVP) {
1294 		unp = VTOUNIONFS(tvp);
1295 #ifdef UNIONFS_IDBG_RENAME
1296 		UNIONFS_INTERNAL_DEBUG("tvp=%p, utvp=%p, ltvp=%p\n",
1297 		    tvp, unp->un_uppervp, unp->un_lowervp);
1298 #endif
1299 		if (unp->un_uppervp == NULLVP)
1300 			rtvp = NULLVP;
1301 		else {
1302 			if (tvp->v_type == VDIR) {
1303 				error = EINVAL;
1304 				goto unionfs_rename_abort;
1305 			}
1306 			rtvp = unp->un_uppervp;
1307 			ltvp = unp->un_lowervp;
1308 			vref(rtvp);
1309 		}
1310 	}
1311 
1312 	if (rfvp == rtvp)
1313 		goto unionfs_rename_abort;
1314 
1315 	if (needrelookup != 0) {
1316 		if ((error = vn_lock(fdvp, LK_EXCLUSIVE)) != 0)
1317 			goto unionfs_rename_abort;
1318 		error = unionfs_relookup_for_delete(fdvp, fcnp, td);
1319 		VOP_UNLOCK(fdvp);
1320 		if (error != 0)
1321 			goto unionfs_rename_abort;
1322 
1323 		/* Lock of tvp is canceled in order to avoid recursive lock. */
1324 		if (tvp != NULLVP && tvp != tdvp)
1325 			VOP_UNLOCK(tvp);
1326 		error = unionfs_relookup_for_rename(tdvp, tcnp, td);
1327 		if (tvp != NULLVP && tvp != tdvp)
1328 			vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
1329 		if (error != 0)
1330 			goto unionfs_rename_abort;
1331 	}
1332 
1333 	error = VOP_RENAME(rfdvp, rfvp, fcnp, rtdvp, rtvp, tcnp);
1334 
1335 	if (error == 0) {
1336 		if (rtvp != NULLVP && rtvp->v_type == VDIR)
1337 			cache_purge(tdvp);
1338 		if (fvp->v_type == VDIR && fdvp != tdvp)
1339 			cache_purge(fdvp);
1340 	}
1341 
1342 	if (ltdvp != NULLVP)
1343 		VOP_UNLOCK(ltdvp);
1344 	if (tdvp != rtdvp)
1345 		vrele(tdvp);
1346 	if (ltvp != NULLVP)
1347 		VOP_UNLOCK(ltvp);
1348 	if (tvp != rtvp && tvp != NULLVP) {
1349 		if (rtvp == NULLVP)
1350 			vput(tvp);
1351 		else
1352 			vrele(tvp);
1353 	}
1354 	if (fdvp != rfdvp)
1355 		vrele(fdvp);
1356 	if (fvp != rfvp)
1357 		vrele(fvp);
1358 
1359 	UNIONFS_INTERNAL_DEBUG("unionfs_rename: leave (%d)\n", error);
1360 
1361 	return (error);
1362 
1363 unionfs_rename_abort:
1364 	vput(tdvp);
1365 	if (tdvp != rtdvp)
1366 		vrele(rtdvp);
1367 	if (tvp != NULLVP) {
1368 		if (tdvp != tvp)
1369 			vput(tvp);
1370 		else
1371 			vrele(tvp);
1372 	}
1373 	if (tvp != rtvp && rtvp != NULLVP)
1374 		vrele(rtvp);
1375 	if (fdvp != rfdvp)
1376 		vrele(rfdvp);
1377 	if (fvp != rfvp)
1378 		vrele(rfvp);
1379 	vrele(fdvp);
1380 	vrele(fvp);
1381 
1382 	UNIONFS_INTERNAL_DEBUG("unionfs_rename: leave (%d)\n", error);
1383 
1384 	return (error);
1385 }
1386 
1387 static int
1388 unionfs_mkdir(struct vop_mkdir_args *ap)
1389 {
1390 	struct unionfs_node *dunp;
1391 	struct componentname *cnp;
1392 	struct vnode   *udvp;
1393 	struct vnode   *uvp;
1394 	struct vattr	va;
1395 	int		error;
1396 	int		lkflags;
1397 
1398 	UNIONFS_INTERNAL_DEBUG("unionfs_mkdir: enter\n");
1399 
1400 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
1401 
1402 	error = EROFS;
1403 	dunp = VTOUNIONFS(ap->a_dvp);
1404 	cnp = ap->a_cnp;
1405 	lkflags = cnp->cn_lkflags;
1406 	udvp = dunp->un_uppervp;
1407 
1408 	if (udvp != NULLVP) {
1409 		/* check opaque */
1410 		if (!(cnp->cn_flags & ISWHITEOUT)) {
1411 			error = VOP_GETATTR(udvp, &va, cnp->cn_cred);
1412 			if (error != 0)
1413 				return (error);
1414 			if ((va.va_flags & OPAQUE) != 0)
1415 				cnp->cn_flags |= ISWHITEOUT;
1416 		}
1417 
1418 		if ((error = VOP_MKDIR(udvp, &uvp, cnp, ap->a_vap)) == 0) {
1419 			VOP_UNLOCK(uvp);
1420 			cnp->cn_lkflags = LK_EXCLUSIVE;
1421 			error = unionfs_nodeget(ap->a_dvp->v_mount, uvp, NULLVP,
1422 			    ap->a_dvp, ap->a_vpp, cnp);
1423 			cnp->cn_lkflags = lkflags;
1424 			vrele(uvp);
1425 		}
1426 	}
1427 
1428 	UNIONFS_INTERNAL_DEBUG("unionfs_mkdir: leave (%d)\n", error);
1429 
1430 	return (error);
1431 }
1432 
1433 static int
1434 unionfs_rmdir(struct vop_rmdir_args *ap)
1435 {
1436 	struct unionfs_node *dunp;
1437 	struct unionfs_node *unp;
1438 	struct unionfs_mount *ump;
1439 	struct componentname *cnp;
1440 	struct thread  *td;
1441 	struct vnode   *udvp;
1442 	struct vnode   *uvp;
1443 	struct vnode   *lvp;
1444 	int		error;
1445 
1446 	UNIONFS_INTERNAL_DEBUG("unionfs_rmdir: enter\n");
1447 
1448 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
1449 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1450 
1451 	error = 0;
1452 	dunp = VTOUNIONFS(ap->a_dvp);
1453 	unp = VTOUNIONFS(ap->a_vp);
1454 	cnp = ap->a_cnp;
1455 	td = curthread;
1456 	udvp = dunp->un_uppervp;
1457 	uvp = unp->un_uppervp;
1458 	lvp = unp->un_lowervp;
1459 
1460 	if (udvp == NULLVP)
1461 		return (EROFS);
1462 
1463 	if (udvp == uvp)
1464 		return (EOPNOTSUPP);
1465 
1466 	if (uvp != NULLVP) {
1467 		if (lvp != NULLVP) {
1468 			error = unionfs_check_rmdir(ap->a_vp, cnp->cn_cred, td);
1469 			if (error != 0)
1470 				return (error);
1471 		}
1472 		ump = MOUNTTOUNIONFSMOUNT(ap->a_vp->v_mount);
1473 		if (ump->um_whitemode == UNIONFS_WHITE_ALWAYS || lvp != NULLVP)
1474 			cnp->cn_flags |= DOWHITEOUT;
1475 		/*
1476 		 * The relookup path will need to relock the parent dvp and
1477 		 * possibly the vp as well.  Locking is expected to be done
1478 		 * in parent->child order; drop the lock on vp to avoid LOR
1479 		 * and potential recursion on vp's lock.
1480 		 * vp is expected to remain referenced during VOP_RMDIR(),
1481 		 * so vref/vrele should not be necessary here.
1482 		 */
1483 		VOP_UNLOCK(ap->a_vp);
1484 		VNPASS(vrefcnt(ap->a_vp) > 0, ap->a_vp);
1485 		error = unionfs_relookup_for_delete(ap->a_dvp, cnp, td);
1486 		vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
1487 		/*
1488 		 * VOP_RMDIR is dispatched against udvp, so if uvp became
1489 		 * doomed while the lock was dropped above the target
1490 		 * filesystem may not be able to cope.
1491 		 */
1492 		if (error == 0 && VN_IS_DOOMED(uvp))
1493 			error = ENOENT;
1494 		if (error == 0)
1495 			error = VOP_RMDIR(udvp, uvp, cnp);
1496 	} else if (lvp != NULLVP)
1497 		error = unionfs_mkwhiteout(udvp, cnp, td,
1498 		    unp->un_path, unp->un_pathlen);
1499 
1500 	if (error == 0) {
1501 		cache_purge(ap->a_dvp);
1502 		cache_purge(ap->a_vp);
1503 	}
1504 
1505 	UNIONFS_INTERNAL_DEBUG("unionfs_rmdir: leave (%d)\n", error);
1506 
1507 	return (error);
1508 }
1509 
1510 static int
1511 unionfs_symlink(struct vop_symlink_args *ap)
1512 {
1513 	struct unionfs_node *dunp;
1514 	struct componentname *cnp;
1515 	struct vnode   *udvp;
1516 	struct vnode   *uvp;
1517 	int		error;
1518 	int		lkflags;
1519 
1520 	UNIONFS_INTERNAL_DEBUG("unionfs_symlink: enter\n");
1521 
1522 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
1523 
1524 	error = EROFS;
1525 	dunp = VTOUNIONFS(ap->a_dvp);
1526 	cnp = ap->a_cnp;
1527 	lkflags = cnp->cn_lkflags;
1528 	udvp = dunp->un_uppervp;
1529 
1530 	if (udvp != NULLVP) {
1531 		error = VOP_SYMLINK(udvp, &uvp, cnp, ap->a_vap, ap->a_target);
1532 		if (error == 0) {
1533 			VOP_UNLOCK(uvp);
1534 			cnp->cn_lkflags = LK_EXCLUSIVE;
1535 			error = unionfs_nodeget(ap->a_dvp->v_mount, uvp, NULLVP,
1536 			    ap->a_dvp, ap->a_vpp, cnp);
1537 			cnp->cn_lkflags = lkflags;
1538 			vrele(uvp);
1539 		}
1540 	}
1541 
1542 	UNIONFS_INTERNAL_DEBUG("unionfs_symlink: leave (%d)\n", error);
1543 
1544 	return (error);
1545 }
1546 
1547 static int
1548 unionfs_readdir(struct vop_readdir_args *ap)
1549 {
1550 	struct unionfs_node *unp;
1551 	struct unionfs_node_status *unsp;
1552 	struct uio     *uio;
1553 	struct vnode   *vp;
1554 	struct vnode   *uvp;
1555 	struct vnode   *lvp;
1556 	struct thread  *td;
1557 	struct vattr    va;
1558 
1559 	uint64_t	*cookies_bk;
1560 	int		error;
1561 	int		eofflag;
1562 	int		ncookies_bk;
1563 	int		uio_offset_bk;
1564 	enum unionfs_lkupgrade lkstatus;
1565 
1566 	UNIONFS_INTERNAL_DEBUG("unionfs_readdir: enter\n");
1567 
1568 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1569 
1570 	error = 0;
1571 	eofflag = 0;
1572 	uio_offset_bk = 0;
1573 	uio = ap->a_uio;
1574 	uvp = NULLVP;
1575 	lvp = NULLVP;
1576 	td = uio->uio_td;
1577 	ncookies_bk = 0;
1578 	cookies_bk = NULL;
1579 
1580 	vp = ap->a_vp;
1581 	if (vp->v_type != VDIR)
1582 		return (ENOTDIR);
1583 
1584 	/*
1585 	 * If the vnode is reclaimed while upgrading, we can't safely use unp
1586 	 * or do anything else unionfs- specific.
1587 	 */
1588 	lkstatus = unionfs_upgrade_lock(vp);
1589 	if (lkstatus == UNIONFS_LKUPGRADE_DOOMED)
1590 		error = EBADF;
1591 	if (error == 0) {
1592 		unp = VTOUNIONFS(vp);
1593 		uvp = unp->un_uppervp;
1594 		lvp = unp->un_lowervp;
1595 		/* check the open count. unionfs needs open before readdir. */
1596 		unionfs_get_node_status(unp, td, &unsp);
1597 		if ((uvp != NULLVP && unsp->uns_upper_opencnt <= 0) ||
1598 			(lvp != NULLVP && unsp->uns_lower_opencnt <= 0)) {
1599 			unionfs_tryrem_node_status(unp, unsp);
1600 			error = EBADF;
1601 		}
1602 	}
1603 	unionfs_downgrade_lock(vp, lkstatus);
1604 	if (error != 0)
1605 		goto unionfs_readdir_exit;
1606 
1607 	/* check opaque */
1608 	if (uvp != NULLVP && lvp != NULLVP) {
1609 		if ((error = VOP_GETATTR(uvp, &va, ap->a_cred)) != 0)
1610 			goto unionfs_readdir_exit;
1611 		if (va.va_flags & OPAQUE)
1612 			lvp = NULLVP;
1613 	}
1614 
1615 	/* upper only */
1616 	if (uvp != NULLVP && lvp == NULLVP) {
1617 		error = VOP_READDIR(uvp, uio, ap->a_cred, ap->a_eofflag,
1618 		    ap->a_ncookies, ap->a_cookies);
1619 		unsp->uns_readdir_status = 0;
1620 
1621 		goto unionfs_readdir_exit;
1622 	}
1623 
1624 	/* lower only */
1625 	if (uvp == NULLVP && lvp != NULLVP) {
1626 		error = VOP_READDIR(lvp, uio, ap->a_cred, ap->a_eofflag,
1627 		    ap->a_ncookies, ap->a_cookies);
1628 		unsp->uns_readdir_status = 2;
1629 
1630 		goto unionfs_readdir_exit;
1631 	}
1632 
1633 	/*
1634 	 * readdir upper and lower
1635 	 */
1636 	KASSERT(uvp != NULLVP, ("unionfs_readdir: null upper vp"));
1637 	KASSERT(lvp != NULLVP, ("unionfs_readdir: null lower vp"));
1638 	if (uio->uio_offset == 0)
1639 		unsp->uns_readdir_status = 0;
1640 
1641 	if (unsp->uns_readdir_status == 0) {
1642 		/* read upper */
1643 		error = VOP_READDIR(uvp, uio, ap->a_cred, &eofflag,
1644 				    ap->a_ncookies, ap->a_cookies);
1645 
1646 		if (error != 0 || eofflag == 0)
1647 			goto unionfs_readdir_exit;
1648 		unsp->uns_readdir_status = 1;
1649 
1650 		/*
1651 		 * UFS(and other FS) needs size of uio_resid larger than
1652 		 * DIRBLKSIZ.
1653 		 * size of DIRBLKSIZ equals DEV_BSIZE.
1654 		 * (see: ufs/ufs/ufs_vnops.c ufs_readdir func , ufs/ufs/dir.h)
1655 		 */
1656 		if (uio->uio_resid <= (uio->uio_resid & (DEV_BSIZE -1)))
1657 			goto unionfs_readdir_exit;
1658 
1659 		/*
1660 		 * Backup cookies.
1661 		 * It prepares to readdir in lower.
1662 		 */
1663 		if (ap->a_ncookies != NULL) {
1664 			ncookies_bk = *(ap->a_ncookies);
1665 			*(ap->a_ncookies) = 0;
1666 		}
1667 		if (ap->a_cookies != NULL) {
1668 			cookies_bk = *(ap->a_cookies);
1669 			*(ap->a_cookies) = NULL;
1670 		}
1671 	}
1672 
1673 	/* initialize for readdir in lower */
1674 	if (unsp->uns_readdir_status == 1) {
1675 		unsp->uns_readdir_status = 2;
1676 		/*
1677 		 * Backup uio_offset. See the comment after the
1678 		 * VOP_READDIR call on the lower layer.
1679 		 */
1680 		uio_offset_bk = uio->uio_offset;
1681 		uio->uio_offset = 0;
1682 	}
1683 
1684 	if (lvp == NULLVP) {
1685 		error = EBADF;
1686 		goto unionfs_readdir_exit;
1687 	}
1688 	/* read lower */
1689 	error = VOP_READDIR(lvp, uio, ap->a_cred, ap->a_eofflag,
1690 			    ap->a_ncookies, ap->a_cookies);
1691 
1692 	/*
1693 	 * We can't return an uio_offset of 0: this would trigger an
1694 	 * infinite loop, because the next call to unionfs_readdir would
1695 	 * always restart with the upper layer (uio_offset == 0) and
1696 	 * always return some data.
1697 	 *
1698 	 * This happens when the lower layer root directory is removed.
1699 	 * (A root directory deleting of unionfs should not be permitted.
1700 	 *  But current VFS can not do it.)
1701 	 */
1702 	if (uio->uio_offset == 0)
1703 		uio->uio_offset = uio_offset_bk;
1704 
1705 	if (cookies_bk != NULL) {
1706 		/* merge cookies */
1707 		int		size;
1708 		uint64_t         *newcookies, *pos;
1709 
1710 		size = *(ap->a_ncookies) + ncookies_bk;
1711 		newcookies = (uint64_t *) malloc(size * sizeof(*newcookies),
1712 		    M_TEMP, M_WAITOK);
1713 		pos = newcookies;
1714 
1715 		memcpy(pos, cookies_bk, ncookies_bk * sizeof(*newcookies));
1716 		pos += ncookies_bk;
1717 		memcpy(pos, *(ap->a_cookies),
1718 		    *(ap->a_ncookies) * sizeof(*newcookies));
1719 		free(cookies_bk, M_TEMP);
1720 		free(*(ap->a_cookies), M_TEMP);
1721 		*(ap->a_ncookies) = size;
1722 		*(ap->a_cookies) = newcookies;
1723 	}
1724 
1725 unionfs_readdir_exit:
1726 	if (error != 0 && ap->a_eofflag != NULL)
1727 		*(ap->a_eofflag) = 1;
1728 
1729 	UNIONFS_INTERNAL_DEBUG("unionfs_readdir: leave (%d)\n", error);
1730 
1731 	return (error);
1732 }
1733 
1734 static int
1735 unionfs_readlink(struct vop_readlink_args *ap)
1736 {
1737 	struct unionfs_node *unp;
1738 	struct vnode   *vp;
1739 	int error;
1740 
1741 	UNIONFS_INTERNAL_DEBUG("unionfs_readlink: enter\n");
1742 
1743 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1744 
1745 	unp = VTOUNIONFS(ap->a_vp);
1746 	vp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
1747 
1748 	error = VOP_READLINK(vp, ap->a_uio, ap->a_cred);
1749 
1750 	UNIONFS_INTERNAL_DEBUG("unionfs_readlink: leave (%d)\n", error);
1751 
1752 	return (error);
1753 }
1754 
1755 static int
1756 unionfs_getwritemount(struct vop_getwritemount_args *ap)
1757 {
1758 	struct unionfs_node *unp;
1759 	struct vnode   *uvp;
1760 	struct vnode   *vp, *ovp;
1761 	int		error;
1762 
1763 	UNIONFS_INTERNAL_DEBUG("unionfs_getwritemount: enter\n");
1764 
1765 	error = 0;
1766 	vp = ap->a_vp;
1767 	uvp = NULLVP;
1768 
1769 	VI_LOCK(vp);
1770 	unp = VTOUNIONFS(vp);
1771 	if (unp != NULL)
1772 		uvp = unp->un_uppervp;
1773 
1774 	/*
1775 	 * If our node has no upper vnode, check the parent directory.
1776 	 * We may be initiating a write operation that will produce a
1777 	 * new upper vnode through CoW.
1778 	 */
1779 	if (uvp == NULLVP && unp != NULL) {
1780 		ovp = vp;
1781 		vp = unp->un_dvp;
1782 		/*
1783 		 * Only the root vnode should have an empty parent, but it
1784 		 * should not have an empty uppervp, so we shouldn't get here.
1785 		 */
1786 		VNASSERT(vp != NULL, ovp, ("%s: NULL parent vnode", __func__));
1787 		VI_UNLOCK(ovp);
1788 		VI_LOCK(vp);
1789 		unp = VTOUNIONFS(vp);
1790 		if (unp != NULL)
1791 			uvp = unp->un_uppervp;
1792 		if (uvp == NULLVP)
1793 			error = EACCES;
1794 	}
1795 
1796 	if (uvp != NULLVP) {
1797 		vholdnz(uvp);
1798 		VI_UNLOCK(vp);
1799 		error = VOP_GETWRITEMOUNT(uvp, ap->a_mpp);
1800 		vdrop(uvp);
1801 	} else {
1802 		VI_UNLOCK(vp);
1803 		*(ap->a_mpp) = NULL;
1804 	}
1805 
1806 	UNIONFS_INTERNAL_DEBUG("unionfs_getwritemount: leave (%d)\n", error);
1807 
1808 	return (error);
1809 }
1810 
1811 static int
1812 unionfs_inactive(struct vop_inactive_args *ap)
1813 {
1814 	ap->a_vp->v_object = NULL;
1815 	vrecycle(ap->a_vp);
1816 	return (0);
1817 }
1818 
1819 static int
1820 unionfs_reclaim(struct vop_reclaim_args *ap)
1821 {
1822 	/* UNIONFS_INTERNAL_DEBUG("unionfs_reclaim: enter\n"); */
1823 
1824 	unionfs_noderem(ap->a_vp);
1825 
1826 	/* UNIONFS_INTERNAL_DEBUG("unionfs_reclaim: leave\n"); */
1827 
1828 	return (0);
1829 }
1830 
1831 static int
1832 unionfs_print(struct vop_print_args *ap)
1833 {
1834 	struct unionfs_node *unp;
1835 	/* struct unionfs_node_status *unsp; */
1836 
1837 	unp = VTOUNIONFS(ap->a_vp);
1838 	/* unionfs_get_node_status(unp, curthread, &unsp); */
1839 
1840 	printf("unionfs_vp=%p, uppervp=%p, lowervp=%p\n",
1841 	    ap->a_vp, unp->un_uppervp, unp->un_lowervp);
1842 	/*
1843 	printf("unionfs opencnt: uppervp=%d, lowervp=%d\n",
1844 	    unsp->uns_upper_opencnt, unsp->uns_lower_opencnt);
1845 	*/
1846 
1847 	if (unp->un_uppervp != NULLVP)
1848 		vn_printf(unp->un_uppervp, "unionfs: upper ");
1849 	if (unp->un_lowervp != NULLVP)
1850 		vn_printf(unp->un_lowervp, "unionfs: lower ");
1851 
1852 	return (0);
1853 }
1854 
1855 static int
1856 unionfs_islocked(struct vop_islocked_args *ap)
1857 {
1858 	struct unionfs_node *unp;
1859 
1860 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1861 
1862 	unp = VTOUNIONFS(ap->a_vp);
1863 	if (unp == NULL)
1864 		return (vop_stdislocked(ap));
1865 
1866 	if (unp->un_uppervp != NULLVP)
1867 		return (VOP_ISLOCKED(unp->un_uppervp));
1868 	if (unp->un_lowervp != NULLVP)
1869 		return (VOP_ISLOCKED(unp->un_lowervp));
1870 	return (vop_stdislocked(ap));
1871 }
1872 
1873 static int
1874 unionfs_get_llt_revlock(struct vnode *vp, int flags)
1875 {
1876 	int revlock;
1877 
1878 	revlock = 0;
1879 
1880 	switch (flags & LK_TYPE_MASK) {
1881 	case LK_SHARED:
1882 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
1883 			revlock = LK_UPGRADE;
1884 		else
1885 			revlock = LK_RELEASE;
1886 		break;
1887 	case LK_EXCLUSIVE:
1888 	case LK_UPGRADE:
1889 		revlock = LK_RELEASE;
1890 		break;
1891 	case LK_DOWNGRADE:
1892 		revlock = LK_UPGRADE;
1893 		break;
1894 	default:
1895 		break;
1896 	}
1897 
1898 	return (revlock);
1899 }
1900 
1901 /*
1902  * The state of an acquired lock is adjusted similarly to
1903  * the time of error generating.
1904  * flags: LK_RELEASE or LK_UPGRADE
1905  */
1906 static void
1907 unionfs_revlock(struct vnode *vp, int flags)
1908 {
1909 	if (flags & LK_RELEASE)
1910 		VOP_UNLOCK_FLAGS(vp, flags);
1911 	else {
1912 		/* UPGRADE */
1913 		if (vn_lock(vp, flags) != 0)
1914 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1915 	}
1916 }
1917 
1918 static int
1919 unionfs_lock(struct vop_lock1_args *ap)
1920 {
1921 	struct unionfs_node *unp;
1922 	struct vnode   *vp;
1923 	struct vnode   *uvp;
1924 	struct vnode   *lvp;
1925 	int		error;
1926 	int		flags;
1927 	int		revlock;
1928 	int		interlock;
1929 	int		uhold;
1930 
1931 	/*
1932 	 * TODO: rework the unionfs locking scheme.
1933 	 * It's not guaranteed to be safe to blindly lock two vnodes on
1934 	 * different mounts as is done here.  Further, the entanglement
1935 	 * of locking both vnodes with the various options that can be
1936 	 * passed to VOP_LOCK() makes this code hard to reason about.
1937 	 * Instead, consider locking only the upper vnode, or the lower
1938 	 * vnode is the upper is not present, and taking separate measures
1939 	 * to lock both vnodes in the few cases when that is needed.
1940 	 */
1941 	error = 0;
1942 	interlock = 1;
1943 	uhold = 0;
1944 	flags = ap->a_flags;
1945 	vp = ap->a_vp;
1946 
1947 	if (LK_RELEASE == (flags & LK_TYPE_MASK) || !(flags & LK_TYPE_MASK))
1948 		return (VOP_UNLOCK_FLAGS(vp, flags | LK_RELEASE));
1949 
1950 	if ((flags & LK_INTERLOCK) == 0)
1951 		VI_LOCK(vp);
1952 
1953 	unp = VTOUNIONFS(vp);
1954 	if (unp == NULL)
1955 		goto unionfs_lock_null_vnode;
1956 
1957 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1958 
1959 	lvp = unp->un_lowervp;
1960 	uvp = unp->un_uppervp;
1961 
1962 	if ((revlock = unionfs_get_llt_revlock(vp, flags)) == 0)
1963 		panic("unknown lock type: 0x%x", flags & LK_TYPE_MASK);
1964 
1965 	/*
1966 	 * During unmount, the root vnode lock may be taken recursively,
1967 	 * because it may share the same v_vnlock field as the vnode covered by
1968 	 * the unionfs mount.  The covered vnode is locked across VFS_UNMOUNT(),
1969 	 * and the same lock may be taken recursively here during vflush()
1970 	 * issued by unionfs_unmount().
1971 	 */
1972 	if ((flags & LK_TYPE_MASK) == LK_EXCLUSIVE &&
1973 	    (vp->v_vflag & VV_ROOT) != 0)
1974 		flags |= LK_CANRECURSE;
1975 
1976 	if (lvp != NULLVP) {
1977 		if (uvp != NULLVP && flags & LK_UPGRADE) {
1978 			/*
1979 			 * Share Lock is once released and a deadlock is
1980 			 * avoided.
1981 			 */
1982 			vholdnz(uvp);
1983 			uhold = 1;
1984 			VOP_UNLOCK(uvp);
1985 			unp = VTOUNIONFS(vp);
1986 			if (unp == NULL) {
1987 				/* vnode is released. */
1988 				VI_UNLOCK(vp);
1989 				VOP_UNLOCK(lvp);
1990 				vdrop(uvp);
1991 				return (EBUSY);
1992 			}
1993 		}
1994 		VI_LOCK_FLAGS(lvp, MTX_DUPOK);
1995 		flags |= LK_INTERLOCK;
1996 		vholdl(lvp);
1997 
1998 		VI_UNLOCK(vp);
1999 		ap->a_flags &= ~LK_INTERLOCK;
2000 
2001 		error = VOP_LOCK(lvp, flags);
2002 
2003 		VI_LOCK(vp);
2004 		unp = VTOUNIONFS(vp);
2005 		if (unp == NULL) {
2006 			/* vnode is released. */
2007 			VI_UNLOCK(vp);
2008 			if (error == 0)
2009 				VOP_UNLOCK(lvp);
2010 			vdrop(lvp);
2011 			if (uhold != 0)
2012 				vdrop(uvp);
2013 			return (vop_stdlock(ap));
2014 		}
2015 	}
2016 
2017 	if (error == 0 && uvp != NULLVP) {
2018 		if (uhold && flags & LK_UPGRADE) {
2019 			flags &= ~LK_TYPE_MASK;
2020 			flags |= LK_EXCLUSIVE;
2021 		}
2022 		VI_LOCK_FLAGS(uvp, MTX_DUPOK);
2023 		flags |= LK_INTERLOCK;
2024 		if (uhold == 0) {
2025 			vholdl(uvp);
2026 			uhold = 1;
2027 		}
2028 
2029 		VI_UNLOCK(vp);
2030 		ap->a_flags &= ~LK_INTERLOCK;
2031 
2032 		error = VOP_LOCK(uvp, flags);
2033 
2034 		VI_LOCK(vp);
2035 		unp = VTOUNIONFS(vp);
2036 		if (unp == NULL) {
2037 			/* vnode is released. */
2038 			VI_UNLOCK(vp);
2039 			if (error == 0)
2040 				VOP_UNLOCK(uvp);
2041 			vdrop(uvp);
2042 			if (lvp != NULLVP) {
2043 				VOP_UNLOCK(lvp);
2044 				vdrop(lvp);
2045 			}
2046 			return (vop_stdlock(ap));
2047 		}
2048 		if (error != 0 && lvp != NULLVP) {
2049 			/* rollback */
2050 			VI_UNLOCK(vp);
2051 			unionfs_revlock(lvp, revlock);
2052 			interlock = 0;
2053 		}
2054 	}
2055 
2056 	if (interlock)
2057 		VI_UNLOCK(vp);
2058 	if (lvp != NULLVP)
2059 		vdrop(lvp);
2060 	if (uhold != 0)
2061 		vdrop(uvp);
2062 
2063 	return (error);
2064 
2065 unionfs_lock_null_vnode:
2066 	ap->a_flags |= LK_INTERLOCK;
2067 	return (vop_stdlock(ap));
2068 }
2069 
2070 static int
2071 unionfs_unlock(struct vop_unlock_args *ap)
2072 {
2073 	struct vnode   *vp;
2074 	struct vnode   *lvp;
2075 	struct vnode   *uvp;
2076 	struct unionfs_node *unp;
2077 	int		error;
2078 	int		uhold;
2079 
2080 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2081 
2082 	error = 0;
2083 	uhold = 0;
2084 	vp = ap->a_vp;
2085 
2086 	unp = VTOUNIONFS(vp);
2087 	if (unp == NULL)
2088 		goto unionfs_unlock_null_vnode;
2089 	lvp = unp->un_lowervp;
2090 	uvp = unp->un_uppervp;
2091 
2092 	if (lvp != NULLVP) {
2093 		vholdnz(lvp);
2094 		error = VOP_UNLOCK(lvp);
2095 	}
2096 
2097 	if (error == 0 && uvp != NULLVP) {
2098 		vholdnz(uvp);
2099 		uhold = 1;
2100 		error = VOP_UNLOCK(uvp);
2101 	}
2102 
2103 	if (lvp != NULLVP)
2104 		vdrop(lvp);
2105 	if (uhold != 0)
2106 		vdrop(uvp);
2107 
2108 	return error;
2109 
2110 unionfs_unlock_null_vnode:
2111 	return (vop_stdunlock(ap));
2112 }
2113 
2114 static int
2115 unionfs_pathconf(struct vop_pathconf_args *ap)
2116 {
2117 	struct unionfs_node *unp;
2118 	struct vnode   *vp;
2119 
2120 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2121 
2122 	unp = VTOUNIONFS(ap->a_vp);
2123 	vp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
2124 
2125 	return (VOP_PATHCONF(vp, ap->a_name, ap->a_retval));
2126 }
2127 
2128 static int
2129 unionfs_advlock(struct vop_advlock_args *ap)
2130 {
2131 	struct unionfs_node *unp;
2132 	struct unionfs_node_status *unsp;
2133 	struct vnode   *vp;
2134 	struct vnode   *uvp;
2135 	struct thread  *td;
2136 	int error;
2137 
2138 	UNIONFS_INTERNAL_DEBUG("unionfs_advlock: enter\n");
2139 
2140 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2141 
2142 	vp = ap->a_vp;
2143 	td = curthread;
2144 
2145 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2146 
2147 	unp = VTOUNIONFS(ap->a_vp);
2148 	uvp = unp->un_uppervp;
2149 
2150 	if (uvp == NULLVP) {
2151 		error = unionfs_copyfile(unp, 1, td->td_ucred, td);
2152 		if (error != 0)
2153 			goto unionfs_advlock_abort;
2154 		uvp = unp->un_uppervp;
2155 
2156 		unionfs_get_node_status(unp, td, &unsp);
2157 		if (unsp->uns_lower_opencnt > 0) {
2158 			/* try reopen the vnode */
2159 			error = VOP_OPEN(uvp, unsp->uns_lower_openmode,
2160 				td->td_ucred, td, NULL);
2161 			if (error)
2162 				goto unionfs_advlock_abort;
2163 			unsp->uns_upper_opencnt++;
2164 			VOP_CLOSE(unp->un_lowervp, unsp->uns_lower_openmode,
2165 			    td->td_ucred, td);
2166 			unsp->uns_lower_opencnt--;
2167 		} else
2168 			unionfs_tryrem_node_status(unp, unsp);
2169 	}
2170 
2171 	VOP_UNLOCK(vp);
2172 
2173 	error = VOP_ADVLOCK(uvp, ap->a_id, ap->a_op, ap->a_fl, ap->a_flags);
2174 
2175 	UNIONFS_INTERNAL_DEBUG("unionfs_advlock: leave (%d)\n", error);
2176 
2177 	return error;
2178 
2179 unionfs_advlock_abort:
2180 	VOP_UNLOCK(vp);
2181 
2182 	UNIONFS_INTERNAL_DEBUG("unionfs_advlock: leave (%d)\n", error);
2183 
2184 	return error;
2185 }
2186 
2187 static int
2188 unionfs_strategy(struct vop_strategy_args *ap)
2189 {
2190 	struct unionfs_node *unp;
2191 	struct vnode   *vp;
2192 
2193 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2194 
2195 	unp = VTOUNIONFS(ap->a_vp);
2196 	vp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
2197 
2198 #ifdef DIAGNOSTIC
2199 	if (vp == NULLVP)
2200 		panic("unionfs_strategy: nullvp");
2201 
2202 	if (ap->a_bp->b_iocmd == BIO_WRITE && vp == unp->un_lowervp)
2203 		panic("unionfs_strategy: writing to lowervp");
2204 #endif
2205 
2206 	return (VOP_STRATEGY(vp, ap->a_bp));
2207 }
2208 
2209 static int
2210 unionfs_getacl(struct vop_getacl_args *ap)
2211 {
2212 	struct unionfs_node *unp;
2213 	struct vnode   *vp;
2214 	int		error;
2215 
2216 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2217 
2218 	unp = VTOUNIONFS(ap->a_vp);
2219 	vp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
2220 
2221 	UNIONFS_INTERNAL_DEBUG("unionfs_getacl: enter\n");
2222 
2223 	error = VOP_GETACL(vp, ap->a_type, ap->a_aclp, ap->a_cred, ap->a_td);
2224 
2225 	UNIONFS_INTERNAL_DEBUG("unionfs_getacl: leave (%d)\n", error);
2226 
2227 	return (error);
2228 }
2229 
2230 static int
2231 unionfs_setacl(struct vop_setacl_args *ap)
2232 {
2233 	struct unionfs_node *unp;
2234 	struct vnode   *uvp;
2235 	struct vnode   *lvp;
2236 	struct thread  *td;
2237 	int		error;
2238 
2239 	UNIONFS_INTERNAL_DEBUG("unionfs_setacl: enter\n");
2240 
2241 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2242 
2243 	error = EROFS;
2244 	unp = VTOUNIONFS(ap->a_vp);
2245 	uvp = unp->un_uppervp;
2246 	lvp = unp->un_lowervp;
2247 	td = ap->a_td;
2248 
2249 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
2250 		return (EROFS);
2251 
2252 	if (uvp == NULLVP && lvp->v_type == VREG) {
2253 		if ((error = unionfs_copyfile(unp, 1, ap->a_cred, td)) != 0)
2254 			return (error);
2255 		uvp = unp->un_uppervp;
2256 	}
2257 
2258 	if (uvp != NULLVP)
2259 		error = VOP_SETACL(uvp, ap->a_type, ap->a_aclp, ap->a_cred, td);
2260 
2261 	UNIONFS_INTERNAL_DEBUG("unionfs_setacl: leave (%d)\n", error);
2262 
2263 	return (error);
2264 }
2265 
2266 static int
2267 unionfs_aclcheck(struct vop_aclcheck_args *ap)
2268 {
2269 	struct unionfs_node *unp;
2270 	struct vnode   *vp;
2271 	int		error;
2272 
2273 	UNIONFS_INTERNAL_DEBUG("unionfs_aclcheck: enter\n");
2274 
2275 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2276 
2277 	unp = VTOUNIONFS(ap->a_vp);
2278 	vp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
2279 
2280 	error = VOP_ACLCHECK(vp, ap->a_type, ap->a_aclp, ap->a_cred, ap->a_td);
2281 
2282 	UNIONFS_INTERNAL_DEBUG("unionfs_aclcheck: leave (%d)\n", error);
2283 
2284 	return (error);
2285 }
2286 
2287 static int
2288 unionfs_openextattr(struct vop_openextattr_args *ap)
2289 {
2290 	struct unionfs_node *unp;
2291 	struct vnode   *vp;
2292 	struct vnode   *tvp;
2293 	int		error;
2294 
2295 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2296 
2297 	vp = ap->a_vp;
2298 	unp = VTOUNIONFS(vp);
2299 	tvp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
2300 
2301 	if ((tvp == unp->un_uppervp && (unp->un_flag & UNIONFS_OPENEXTU)) ||
2302 	    (tvp == unp->un_lowervp && (unp->un_flag & UNIONFS_OPENEXTL)))
2303 		return (EBUSY);
2304 
2305 	error = VOP_OPENEXTATTR(tvp, ap->a_cred, ap->a_td);
2306 
2307 	if (error == 0) {
2308 		if (vn_lock(vp, LK_UPGRADE) != 0)
2309 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2310 		if (!VN_IS_DOOMED(vp)) {
2311 			if (tvp == unp->un_uppervp)
2312 				unp->un_flag |= UNIONFS_OPENEXTU;
2313 			else
2314 				unp->un_flag |= UNIONFS_OPENEXTL;
2315 		}
2316 		vn_lock(vp, LK_DOWNGRADE | LK_RETRY);
2317 	}
2318 
2319 	return (error);
2320 }
2321 
2322 static int
2323 unionfs_closeextattr(struct vop_closeextattr_args *ap)
2324 {
2325 	struct unionfs_node *unp;
2326 	struct vnode   *vp;
2327 	struct vnode   *tvp;
2328 	int		error;
2329 
2330 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2331 
2332 	vp = ap->a_vp;
2333 	unp = VTOUNIONFS(vp);
2334 	tvp = NULLVP;
2335 
2336 	if (unp->un_flag & UNIONFS_OPENEXTU)
2337 		tvp = unp->un_uppervp;
2338 	else if (unp->un_flag & UNIONFS_OPENEXTL)
2339 		tvp = unp->un_lowervp;
2340 
2341 	if (tvp == NULLVP)
2342 		return (EOPNOTSUPP);
2343 
2344 	error = VOP_CLOSEEXTATTR(tvp, ap->a_commit, ap->a_cred, ap->a_td);
2345 
2346 	if (error == 0) {
2347 		if (vn_lock(vp, LK_UPGRADE) != 0)
2348 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2349 		if (!VN_IS_DOOMED(vp)) {
2350 			if (tvp == unp->un_uppervp)
2351 				unp->un_flag &= ~UNIONFS_OPENEXTU;
2352 			else
2353 				unp->un_flag &= ~UNIONFS_OPENEXTL;
2354 		}
2355 		vn_lock(vp, LK_DOWNGRADE | LK_RETRY);
2356 	}
2357 
2358 	return (error);
2359 }
2360 
2361 static int
2362 unionfs_getextattr(struct vop_getextattr_args *ap)
2363 {
2364 	struct unionfs_node *unp;
2365 	struct vnode   *vp;
2366 
2367 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2368 
2369 	unp = VTOUNIONFS(ap->a_vp);
2370 	vp = NULLVP;
2371 
2372 	if (unp->un_flag & UNIONFS_OPENEXTU)
2373 		vp = unp->un_uppervp;
2374 	else if (unp->un_flag & UNIONFS_OPENEXTL)
2375 		vp = unp->un_lowervp;
2376 
2377 	if (vp == NULLVP)
2378 		return (EOPNOTSUPP);
2379 
2380 	return (VOP_GETEXTATTR(vp, ap->a_attrnamespace, ap->a_name,
2381 	    ap->a_uio, ap->a_size, ap->a_cred, ap->a_td));
2382 }
2383 
2384 static int
2385 unionfs_setextattr(struct vop_setextattr_args *ap)
2386 {
2387 	struct unionfs_node *unp;
2388 	struct vnode   *uvp;
2389 	struct vnode   *lvp;
2390 	struct vnode   *ovp;
2391 	struct ucred   *cred;
2392 	struct thread  *td;
2393 	int		error;
2394 
2395 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2396 
2397 	error = EROFS;
2398 	unp = VTOUNIONFS(ap->a_vp);
2399 	uvp = unp->un_uppervp;
2400 	lvp = unp->un_lowervp;
2401 	ovp = NULLVP;
2402 	cred = ap->a_cred;
2403 	td = ap->a_td;
2404 
2405 	UNIONFS_INTERNAL_DEBUG("unionfs_setextattr: enter (un_flag=%x)\n",
2406 	    unp->un_flag);
2407 
2408 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
2409 		return (EROFS);
2410 
2411 	if (unp->un_flag & UNIONFS_OPENEXTU)
2412 		ovp = unp->un_uppervp;
2413 	else if (unp->un_flag & UNIONFS_OPENEXTL)
2414 		ovp = unp->un_lowervp;
2415 
2416 	if (ovp == NULLVP)
2417 		return (EOPNOTSUPP);
2418 
2419 	if (ovp == lvp && lvp->v_type == VREG) {
2420 		VOP_CLOSEEXTATTR(lvp, 0, cred, td);
2421 		if (uvp == NULLVP &&
2422 		    (error = unionfs_copyfile(unp, 1, cred, td)) != 0) {
2423 unionfs_setextattr_reopen:
2424 			if ((unp->un_flag & UNIONFS_OPENEXTL) &&
2425 			    VOP_OPENEXTATTR(lvp, cred, td)) {
2426 #ifdef DIAGNOSTIC
2427 				panic("unionfs: VOP_OPENEXTATTR failed");
2428 #endif
2429 				unp->un_flag &= ~UNIONFS_OPENEXTL;
2430 			}
2431 			goto unionfs_setextattr_abort;
2432 		}
2433 		uvp = unp->un_uppervp;
2434 		if ((error = VOP_OPENEXTATTR(uvp, cred, td)) != 0)
2435 			goto unionfs_setextattr_reopen;
2436 		unp->un_flag &= ~UNIONFS_OPENEXTL;
2437 		unp->un_flag |= UNIONFS_OPENEXTU;
2438 		ovp = uvp;
2439 	}
2440 
2441 	if (ovp == uvp)
2442 		error = VOP_SETEXTATTR(ovp, ap->a_attrnamespace, ap->a_name,
2443 		    ap->a_uio, cred, td);
2444 
2445 unionfs_setextattr_abort:
2446 	UNIONFS_INTERNAL_DEBUG("unionfs_setextattr: leave (%d)\n", error);
2447 
2448 	return (error);
2449 }
2450 
2451 static int
2452 unionfs_listextattr(struct vop_listextattr_args *ap)
2453 {
2454 	struct unionfs_node *unp;
2455 	struct vnode *vp;
2456 
2457 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2458 
2459 	unp = VTOUNIONFS(ap->a_vp);
2460 	vp = NULLVP;
2461 
2462 	if (unp->un_flag & UNIONFS_OPENEXTU)
2463 		vp = unp->un_uppervp;
2464 	else if (unp->un_flag & UNIONFS_OPENEXTL)
2465 		vp = unp->un_lowervp;
2466 
2467 	if (vp == NULLVP)
2468 		return (EOPNOTSUPP);
2469 
2470 	return (VOP_LISTEXTATTR(vp, ap->a_attrnamespace, ap->a_uio,
2471 	    ap->a_size, ap->a_cred, ap->a_td));
2472 }
2473 
2474 static int
2475 unionfs_deleteextattr(struct vop_deleteextattr_args *ap)
2476 {
2477 	struct unionfs_node *unp;
2478 	struct vnode   *uvp;
2479 	struct vnode   *lvp;
2480 	struct vnode   *ovp;
2481 	struct ucred   *cred;
2482 	struct thread  *td;
2483 	int		error;
2484 
2485 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2486 
2487 	error = EROFS;
2488 	unp = VTOUNIONFS(ap->a_vp);
2489 	uvp = unp->un_uppervp;
2490 	lvp = unp->un_lowervp;
2491 	ovp = NULLVP;
2492 	cred = ap->a_cred;
2493 	td = ap->a_td;
2494 
2495 	UNIONFS_INTERNAL_DEBUG("unionfs_deleteextattr: enter (un_flag=%x)\n",
2496 	    unp->un_flag);
2497 
2498 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
2499 		return (EROFS);
2500 
2501 	if (unp->un_flag & UNIONFS_OPENEXTU)
2502 		ovp = unp->un_uppervp;
2503 	else if (unp->un_flag & UNIONFS_OPENEXTL)
2504 		ovp = unp->un_lowervp;
2505 
2506 	if (ovp == NULLVP)
2507 		return (EOPNOTSUPP);
2508 
2509 	if (ovp == lvp && lvp->v_type == VREG) {
2510 		VOP_CLOSEEXTATTR(lvp, 0, cred, td);
2511 		if (uvp == NULLVP &&
2512 		    (error = unionfs_copyfile(unp, 1, cred, td)) != 0) {
2513 unionfs_deleteextattr_reopen:
2514 			if ((unp->un_flag & UNIONFS_OPENEXTL) &&
2515 			    VOP_OPENEXTATTR(lvp, cred, td)) {
2516 #ifdef DIAGNOSTIC
2517 				panic("unionfs: VOP_OPENEXTATTR failed");
2518 #endif
2519 				unp->un_flag &= ~UNIONFS_OPENEXTL;
2520 			}
2521 			goto unionfs_deleteextattr_abort;
2522 		}
2523 		uvp = unp->un_uppervp;
2524 		if ((error = VOP_OPENEXTATTR(uvp, cred, td)) != 0)
2525 			goto unionfs_deleteextattr_reopen;
2526 		unp->un_flag &= ~UNIONFS_OPENEXTL;
2527 		unp->un_flag |= UNIONFS_OPENEXTU;
2528 		ovp = uvp;
2529 	}
2530 
2531 	if (ovp == uvp)
2532 		error = VOP_DELETEEXTATTR(ovp, ap->a_attrnamespace, ap->a_name,
2533 		    ap->a_cred, ap->a_td);
2534 
2535 unionfs_deleteextattr_abort:
2536 	UNIONFS_INTERNAL_DEBUG("unionfs_deleteextattr: leave (%d)\n", error);
2537 
2538 	return (error);
2539 }
2540 
2541 static int
2542 unionfs_setlabel(struct vop_setlabel_args *ap)
2543 {
2544 	struct unionfs_node *unp;
2545 	struct vnode   *uvp;
2546 	struct vnode   *lvp;
2547 	struct thread  *td;
2548 	int		error;
2549 
2550 	UNIONFS_INTERNAL_DEBUG("unionfs_setlabel: enter\n");
2551 
2552 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2553 
2554 	error = EROFS;
2555 	unp = VTOUNIONFS(ap->a_vp);
2556 	uvp = unp->un_uppervp;
2557 	lvp = unp->un_lowervp;
2558 	td = ap->a_td;
2559 
2560 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
2561 		return (EROFS);
2562 
2563 	if (uvp == NULLVP && lvp->v_type == VREG) {
2564 		if ((error = unionfs_copyfile(unp, 1, ap->a_cred, td)) != 0)
2565 			return (error);
2566 		uvp = unp->un_uppervp;
2567 	}
2568 
2569 	if (uvp != NULLVP)
2570 		error = VOP_SETLABEL(uvp, ap->a_label, ap->a_cred, td);
2571 
2572 	UNIONFS_INTERNAL_DEBUG("unionfs_setlabel: leave (%d)\n", error);
2573 
2574 	return (error);
2575 }
2576 
2577 static int
2578 unionfs_vptofh(struct vop_vptofh_args *ap)
2579 {
2580 	return (EOPNOTSUPP);
2581 }
2582 
2583 static int
2584 unionfs_add_writecount(struct vop_add_writecount_args *ap)
2585 {
2586 	struct vnode *tvp, *vp;
2587 	struct unionfs_node *unp;
2588 	int error, writerefs __diagused;
2589 
2590 	vp = ap->a_vp;
2591 	unp = VTOUNIONFS(vp);
2592 	tvp = unp->un_uppervp;
2593 	KASSERT(tvp != NULL,
2594 	    ("%s: adding write ref without upper vnode", __func__));
2595 	error = VOP_ADD_WRITECOUNT(tvp, ap->a_inc);
2596 	if (error != 0)
2597 		return (error);
2598 	/*
2599 	 * We need to track the write refs we've passed to the underlying
2600 	 * vnodes so that we can undo them in case we are forcibly unmounted.
2601 	 */
2602 	writerefs = atomic_fetchadd_int(&vp->v_writecount, ap->a_inc);
2603 	/* text refs are bypassed to lowervp */
2604 	VNASSERT(writerefs >= 0, vp,
2605 	    ("%s: invalid write count %d", __func__, writerefs));
2606 	VNASSERT(writerefs + ap->a_inc >= 0, vp,
2607 	    ("%s: invalid write count inc %d + %d", __func__,
2608 	    writerefs, ap->a_inc));
2609 	return (0);
2610 }
2611 
2612 static int
2613 unionfs_vput_pair(struct vop_vput_pair_args *ap)
2614 {
2615 	struct mount *mp;
2616 	struct vnode *dvp, *vp, **vpp, *lvp, *ldvp, *uvp, *udvp, *tempvp;
2617 	struct unionfs_node *dunp, *unp;
2618 	int error, res;
2619 
2620 	dvp = ap->a_dvp;
2621 	vpp = ap->a_vpp;
2622 	vp = NULLVP;
2623 	lvp = NULLVP;
2624 	uvp = NULLVP;
2625 	unp = NULL;
2626 
2627 	dunp = VTOUNIONFS(dvp);
2628 	udvp = dunp->un_uppervp;
2629 	ldvp = dunp->un_lowervp;
2630 
2631 	/*
2632 	 * Underlying vnodes should be locked because the encompassing unionfs
2633 	 * node is locked, but will not be referenced, as the reference will
2634 	 * only be on the unionfs node.  Reference them now so that the vput()s
2635 	 * performed by VOP_VPUT_PAIR() will have a reference to drop.
2636 	 */
2637 	if (udvp != NULLVP)
2638 		vref(udvp);
2639 	if (ldvp != NULLVP)
2640 		vref(ldvp);
2641 
2642 	if (vpp != NULL)
2643 		vp = *vpp;
2644 
2645 	if (vp != NULLVP) {
2646 		unp = VTOUNIONFS(vp);
2647 		uvp = unp->un_uppervp;
2648 		lvp = unp->un_lowervp;
2649 		if (uvp != NULLVP)
2650 			vref(uvp);
2651 		if (lvp != NULLVP)
2652 			vref(lvp);
2653 
2654 		/*
2655 		 * If we're being asked to return a locked child vnode, then
2656 		 * we may need to create a replacement vnode in case the
2657 		 * original is reclaimed while the lock is dropped.  In that
2658 		 * case we'll need to ensure the mount and the underlying
2659 		 * vnodes aren't also recycled during that window.
2660 		 */
2661 		if (!ap->a_unlock_vp) {
2662 			vhold(vp);
2663 			if (uvp != NULLVP)
2664 				vhold(uvp);
2665 			if (lvp != NULLVP)
2666 				vhold(lvp);
2667 			mp = vp->v_mount;
2668 			vfs_ref(mp);
2669 		}
2670 	}
2671 
2672 	/*
2673 	 * TODO: Because unionfs_lock() locks both the lower and upper vnodes
2674 	 * (if available), we must also call VOP_VPUT_PAIR() on both the lower
2675 	 * and upper parent/child pairs.  If unionfs_lock() is reworked to lock
2676 	 * only a single vnode, this code will need to change to also only
2677 	 * operate on one vnode pair.
2678 	 */
2679 	ASSERT_VOP_LOCKED(ldvp, __func__);
2680 	ASSERT_VOP_LOCKED(udvp, __func__);
2681 	ASSERT_VOP_LOCKED(lvp, __func__);
2682 	ASSERT_VOP_LOCKED(uvp, __func__);
2683 
2684 	KASSERT(lvp == NULLVP || ldvp != NULLVP,
2685 	    ("%s: NULL ldvp with non-NULL lvp", __func__));
2686 	if (ldvp != NULLVP)
2687 		res = VOP_VPUT_PAIR(ldvp, lvp != NULLVP ? &lvp : NULL, true);
2688 	KASSERT(uvp == NULLVP || udvp != NULLVP,
2689 	    ("%s: NULL udvp with non-NULL uvp", __func__));
2690 	if (udvp != NULLVP)
2691 		res = VOP_VPUT_PAIR(udvp, uvp != NULLVP ? &uvp : NULL, true);
2692 
2693 	ASSERT_VOP_UNLOCKED(ldvp, __func__);
2694 	ASSERT_VOP_UNLOCKED(udvp, __func__);
2695 	ASSERT_VOP_UNLOCKED(lvp, __func__);
2696 	ASSERT_VOP_UNLOCKED(uvp, __func__);
2697 
2698 	/*
2699 	 * VOP_VPUT_PAIR() dropped the references we added to the underlying
2700 	 * vnodes, now drop the caller's reference to the unionfs vnodes.
2701 	 */
2702 	if (vp != NULLVP && ap->a_unlock_vp)
2703 		vrele(vp);
2704 	vrele(dvp);
2705 
2706 	if (vp == NULLVP || ap->a_unlock_vp)
2707 		return (res);
2708 
2709 	/*
2710 	 * We're being asked to return a locked vnode.  At this point, the
2711 	 * underlying vnodes have been unlocked, so vp may have been reclaimed.
2712 	 */
2713 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2714 	if (vp->v_data == NULL && vfs_busy(mp, MBF_NOWAIT) == 0) {
2715 		vput(vp);
2716 		error = unionfs_nodeget(mp, uvp, lvp, dvp, &tempvp, NULL);
2717 		if (error == 0) {
2718 			vn_lock(tempvp, LK_EXCLUSIVE | LK_RETRY);
2719 			*vpp = tempvp;
2720 		} else
2721 			vget(vp, LK_EXCLUSIVE | LK_RETRY);
2722 		vfs_unbusy(mp);
2723 	}
2724 	if (lvp != NULLVP)
2725 		vdrop(lvp);
2726 	if (uvp != NULLVP)
2727 		vdrop(uvp);
2728 	vdrop(vp);
2729 	vfs_rel(mp);
2730 
2731 	return (res);
2732 }
2733 
2734 static int
2735 unionfs_set_text(struct vop_set_text_args *ap)
2736 {
2737 	struct vnode *tvp;
2738 	struct unionfs_node *unp;
2739 	int error;
2740 
2741 	/*
2742 	 * We assume text refs are managed against lvp/uvp through the
2743 	 * executable mapping backed by its VM object.  We therefore don't
2744 	 * need to track leased text refs in the case of a forcible unmount.
2745 	 */
2746 	unp = VTOUNIONFS(ap->a_vp);
2747 	ASSERT_VOP_LOCKED(ap->a_vp, __func__);
2748 	tvp = unp->un_uppervp != NULL ? unp->un_uppervp : unp->un_lowervp;
2749 	error = VOP_SET_TEXT(tvp);
2750 	return (error);
2751 }
2752 
2753 static int
2754 unionfs_unset_text(struct vop_unset_text_args *ap)
2755 {
2756 	struct vnode *tvp;
2757 	struct unionfs_node *unp;
2758 
2759 	ASSERT_VOP_LOCKED(ap->a_vp, __func__);
2760 	unp = VTOUNIONFS(ap->a_vp);
2761 	tvp = unp->un_uppervp != NULL ? unp->un_uppervp : unp->un_lowervp;
2762 	VOP_UNSET_TEXT_CHECKED(tvp);
2763 	return (0);
2764 }
2765 
2766 struct vop_vector unionfs_vnodeops = {
2767 	.vop_default =		&default_vnodeops,
2768 
2769 	.vop_access =		unionfs_access,
2770 	.vop_aclcheck =		unionfs_aclcheck,
2771 	.vop_advlock =		unionfs_advlock,
2772 	.vop_bmap =		VOP_EOPNOTSUPP,
2773 	.vop_cachedlookup =	unionfs_lookup,
2774 	.vop_close =		unionfs_close,
2775 	.vop_closeextattr =	unionfs_closeextattr,
2776 	.vop_create =		unionfs_create,
2777 	.vop_deleteextattr =	unionfs_deleteextattr,
2778 	.vop_fsync =		unionfs_fsync,
2779 	.vop_getacl =		unionfs_getacl,
2780 	.vop_getattr =		unionfs_getattr,
2781 	.vop_getextattr =	unionfs_getextattr,
2782 	.vop_getwritemount =	unionfs_getwritemount,
2783 	.vop_inactive =		unionfs_inactive,
2784 	.vop_need_inactive =	vop_stdneed_inactive,
2785 	.vop_islocked =		unionfs_islocked,
2786 	.vop_ioctl =		unionfs_ioctl,
2787 	.vop_link =		unionfs_link,
2788 	.vop_listextattr =	unionfs_listextattr,
2789 	.vop_lock1 =		unionfs_lock,
2790 	.vop_lookup =		vfs_cache_lookup,
2791 	.vop_mkdir =		unionfs_mkdir,
2792 	.vop_mknod =		unionfs_mknod,
2793 	.vop_open =		unionfs_open,
2794 	.vop_openextattr =	unionfs_openextattr,
2795 	.vop_pathconf =		unionfs_pathconf,
2796 	.vop_poll =		unionfs_poll,
2797 	.vop_print =		unionfs_print,
2798 	.vop_read =		unionfs_read,
2799 	.vop_readdir =		unionfs_readdir,
2800 	.vop_readlink =		unionfs_readlink,
2801 	.vop_reclaim =		unionfs_reclaim,
2802 	.vop_remove =		unionfs_remove,
2803 	.vop_rename =		unionfs_rename,
2804 	.vop_rmdir =		unionfs_rmdir,
2805 	.vop_setacl =		unionfs_setacl,
2806 	.vop_setattr =		unionfs_setattr,
2807 	.vop_setextattr =	unionfs_setextattr,
2808 	.vop_setlabel =		unionfs_setlabel,
2809 	.vop_strategy =		unionfs_strategy,
2810 	.vop_symlink =		unionfs_symlink,
2811 	.vop_unlock =		unionfs_unlock,
2812 	.vop_whiteout =		unionfs_whiteout,
2813 	.vop_write =		unionfs_write,
2814 	.vop_vptofh =		unionfs_vptofh,
2815 	.vop_add_writecount =	unionfs_add_writecount,
2816 	.vop_vput_pair =	unionfs_vput_pair,
2817 	.vop_set_text =		unionfs_set_text,
2818 	.vop_unset_text = 	unionfs_unset_text,
2819 };
2820 VFS_VOP_VECTOR_REGISTER(unionfs_vnodeops);
2821