xref: /freebsd/sys/fs/unionfs/union_vnops.c (revision 1edb7116)
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  */
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/conf.h>
43 #include <sys/kernel.h>
44 #include <sys/lock.h>
45 #include <sys/malloc.h>
46 #include <sys/mount.h>
47 #include <sys/mutex.h>
48 #include <sys/namei.h>
49 #include <sys/sysctl.h>
50 #include <sys/vnode.h>
51 #include <sys/kdb.h>
52 #include <sys/fcntl.h>
53 #include <sys/stat.h>
54 #include <sys/dirent.h>
55 #include <sys/proc.h>
56 #include <sys/bio.h>
57 #include <sys/buf.h>
58 
59 #include <fs/unionfs/union.h>
60 
61 #include <machine/atomic.h>
62 
63 #include <vm/vm.h>
64 #include <vm/vm_extern.h>
65 #include <vm/vm_object.h>
66 #include <vm/vnode_pager.h>
67 
68 #if 0
69 #define UNIONFS_INTERNAL_DEBUG(msg, args...)    printf(msg, ## args)
70 #define UNIONFS_IDBG_RENAME
71 #else
72 #define UNIONFS_INTERNAL_DEBUG(msg, args...)
73 #endif
74 
75 #define KASSERT_UNIONFS_VNODE(vp) \
76 	VNASSERT(((vp)->v_op == &unionfs_vnodeops), vp, \
77 	    ("%s: non-unionfs vnode", __func__))
78 
79 static int
80 unionfs_lookup(struct vop_cachedlookup_args *ap)
81 {
82 	struct unionfs_node *dunp;
83 	struct vnode   *dvp, *udvp, *ldvp, *vp, *uvp, *lvp, *dtmpvp;
84 	struct vattr	va;
85 	struct componentname *cnp;
86 	struct thread  *td;
87 	u_long		nameiop;
88 	u_long		cnflags, cnflagsbk;
89 	int		iswhiteout;
90 	int		lockflag;
91 	int		error , uerror, lerror;
92 
93 	iswhiteout = 0;
94 	lockflag = 0;
95 	error = uerror = lerror = ENOENT;
96 	cnp = ap->a_cnp;
97 	nameiop = cnp->cn_nameiop;
98 	cnflags = cnp->cn_flags;
99 	dvp = ap->a_dvp;
100 	dunp = VTOUNIONFS(dvp);
101 	udvp = dunp->un_uppervp;
102 	ldvp = dunp->un_lowervp;
103 	vp = uvp = lvp = NULLVP;
104 	td = curthread;
105 	*(ap->a_vpp) = NULLVP;
106 
107 	UNIONFS_INTERNAL_DEBUG(
108 	    "unionfs_lookup: enter: nameiop=%ld, flags=%lx, path=%s\n",
109 	    nameiop, cnflags, cnp->cn_nameptr);
110 
111 	if (dvp->v_type != VDIR)
112 		return (ENOTDIR);
113 
114 	/*
115 	 * If read-only and op is not LOOKUP, will return EROFS.
116 	 */
117 	if ((cnflags & ISLASTCN) &&
118 	    (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
119 	    LOOKUP != nameiop)
120 		return (EROFS);
121 
122 	/*
123 	 * lookup dotdot
124 	 */
125 	if (cnflags & ISDOTDOT) {
126 		if (LOOKUP != nameiop && udvp == NULLVP)
127 			return (EROFS);
128 
129 		if (udvp != NULLVP) {
130 			dtmpvp = udvp;
131 			if (ldvp != NULLVP)
132 				VOP_UNLOCK(ldvp);
133 		}
134 		else
135 			dtmpvp = ldvp;
136 
137 		error = VOP_LOOKUP(dtmpvp, &vp, cnp);
138 
139 		if (dtmpvp == udvp && ldvp != NULLVP) {
140 			VOP_UNLOCK(udvp);
141 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
142 		}
143 
144 		if (error == 0) {
145 			/*
146 			 * Exchange lock and reference from vp to
147 			 * dunp->un_dvp. vp is upper/lower vnode, but it
148 			 * will need to return the unionfs vnode.
149 			 */
150 			if (nameiop == DELETE  || nameiop == RENAME ||
151 			    (cnp->cn_lkflags & LK_TYPE_MASK))
152 				VOP_UNLOCK(vp);
153 			vrele(vp);
154 
155 			VOP_UNLOCK(dvp);
156 			*(ap->a_vpp) = dunp->un_dvp;
157 			vref(dunp->un_dvp);
158 
159 			if (nameiop == DELETE || nameiop == RENAME)
160 				vn_lock(dunp->un_dvp, LK_EXCLUSIVE | LK_RETRY);
161 			else if (cnp->cn_lkflags & LK_TYPE_MASK)
162 				vn_lock(dunp->un_dvp, cnp->cn_lkflags |
163 				    LK_RETRY);
164 
165 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
166 		} else if (error == ENOENT && (cnflags & MAKEENTRY) != 0)
167 			cache_enter(dvp, NULLVP, cnp);
168 
169 		goto unionfs_lookup_return;
170 	}
171 
172 	/*
173 	 * lookup upper layer
174 	 */
175 	if (udvp != NULLVP) {
176 		uerror = VOP_LOOKUP(udvp, &uvp, cnp);
177 
178 		if (uerror == 0) {
179 			if (udvp == uvp) {	/* is dot */
180 				vrele(uvp);
181 				*(ap->a_vpp) = dvp;
182 				vref(dvp);
183 
184 				error = uerror;
185 				goto unionfs_lookup_return;
186 			}
187 			if (nameiop == DELETE || nameiop == RENAME ||
188 			    (cnp->cn_lkflags & LK_TYPE_MASK))
189 				VOP_UNLOCK(uvp);
190 		}
191 
192 		/* check whiteout */
193 		if (uerror == ENOENT || uerror == EJUSTRETURN)
194 			if (cnp->cn_flags & ISWHITEOUT)
195 				iswhiteout = 1;	/* don't lookup lower */
196 		if (iswhiteout == 0 && ldvp != NULLVP)
197 			if (!VOP_GETATTR(udvp, &va, cnp->cn_cred) &&
198 			    (va.va_flags & OPAQUE))
199 				iswhiteout = 1;	/* don't lookup lower */
200 #if 0
201 		UNIONFS_INTERNAL_DEBUG(
202 		    "unionfs_lookup: debug: whiteout=%d, path=%s\n",
203 		    iswhiteout, cnp->cn_nameptr);
204 #endif
205 	}
206 
207 	/*
208 	 * lookup lower layer
209 	 */
210 	if (ldvp != NULLVP && !(cnflags & DOWHITEOUT) && iswhiteout == 0) {
211 		/* always op is LOOKUP */
212 		cnp->cn_nameiop = LOOKUP;
213 		cnflagsbk = cnp->cn_flags;
214 		cnp->cn_flags = cnflags;
215 
216 		lerror = VOP_LOOKUP(ldvp, &lvp, cnp);
217 
218 		cnp->cn_nameiop = nameiop;
219 		if (udvp != NULLVP && (uerror == 0 || uerror == EJUSTRETURN))
220 			cnp->cn_flags = cnflagsbk;
221 
222 		if (lerror == 0) {
223 			if (ldvp == lvp) {	/* is dot */
224 				if (uvp != NULLVP)
225 					vrele(uvp);	/* no need? */
226 				vrele(lvp);
227 				*(ap->a_vpp) = dvp;
228 				vref(dvp);
229 
230 				UNIONFS_INTERNAL_DEBUG(
231 				    "unionfs_lookup: leave (%d)\n", lerror);
232 
233 				return (lerror);
234 			}
235 			if (cnp->cn_lkflags & LK_TYPE_MASK)
236 				VOP_UNLOCK(lvp);
237 		}
238 	}
239 
240 	/*
241 	 * check lookup result
242 	 */
243 	if (uvp == NULLVP && lvp == NULLVP) {
244 		error = (udvp != NULLVP ? uerror : lerror);
245 		goto unionfs_lookup_return;
246 	}
247 
248 	/*
249 	 * check vnode type
250 	 */
251 	if (uvp != NULLVP && lvp != NULLVP && uvp->v_type != lvp->v_type) {
252 		vrele(lvp);
253 		lvp = NULLVP;
254 	}
255 
256 	/*
257 	 * check shadow dir
258 	 */
259 	if (uerror != 0 && uerror != EJUSTRETURN && udvp != NULLVP &&
260 	    lerror == 0 && lvp != NULLVP && lvp->v_type == VDIR &&
261 	    !(dvp->v_mount->mnt_flag & MNT_RDONLY) &&
262 	    (1 < cnp->cn_namelen || '.' != *(cnp->cn_nameptr))) {
263 		/* get unionfs vnode in order to create a new shadow dir. */
264 		error = unionfs_nodeget(dvp->v_mount, NULLVP, lvp, dvp, &vp,
265 		    cnp);
266 		if (error != 0)
267 			goto unionfs_lookup_cleanup;
268 
269 		if (LK_SHARED == (cnp->cn_lkflags & LK_TYPE_MASK))
270 			VOP_UNLOCK(vp);
271 		if (LK_EXCLUSIVE != VOP_ISLOCKED(vp)) {
272 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
273 			lockflag = 1;
274 		}
275 		error = unionfs_mkshadowdir(MOUNTTOUNIONFSMOUNT(dvp->v_mount),
276 		    udvp, VTOUNIONFS(vp), cnp, td);
277 		if (lockflag != 0)
278 			VOP_UNLOCK(vp);
279 		if (error != 0) {
280 			UNIONFSDEBUG(
281 			    "unionfs_lookup: Unable to create shadow dir.");
282 			if ((cnp->cn_lkflags & LK_TYPE_MASK) == LK_EXCLUSIVE)
283 				vput(vp);
284 			else
285 				vrele(vp);
286 			goto unionfs_lookup_cleanup;
287 		}
288 		if ((cnp->cn_lkflags & LK_TYPE_MASK) == LK_SHARED)
289 			vn_lock(vp, LK_SHARED | LK_RETRY);
290 	}
291 	/*
292 	 * get unionfs vnode.
293 	 */
294 	else {
295 		if (uvp != NULLVP)
296 			error = uerror;
297 		else
298 			error = lerror;
299 		if (error != 0)
300 			goto unionfs_lookup_cleanup;
301 		/*
302 		 * get socket vnode.
303 		 */
304 		if (uvp != NULLVP && uvp->v_type == VSOCK) {
305 			vp = uvp;
306 			vref(vp);
307 			if (cnp->cn_lkflags & LK_TYPE_MASK)
308 				vn_lock(vp, cnp->cn_lkflags | LK_RETRY);
309 		}
310 		else if (lvp != NULLVP && lvp->v_type == VSOCK) {
311 			vp = lvp;
312 			vref(vp);
313 			if (cnp->cn_lkflags & LK_TYPE_MASK)
314 				vn_lock(vp, cnp->cn_lkflags | LK_RETRY);
315 		}
316 		/*
317 		 * get unionfs vnode.
318 		 */
319 		else
320 			error = unionfs_nodeget(dvp->v_mount, uvp, lvp,
321 			    dvp, &vp, cnp);
322 		if (error != 0) {
323 			UNIONFSDEBUG(
324 			    "unionfs_lookup: Unable to create unionfs vnode.");
325 			goto unionfs_lookup_cleanup;
326 		}
327 		if ((nameiop == DELETE || nameiop == RENAME) &&
328 		    (cnp->cn_lkflags & LK_TYPE_MASK) == 0)
329 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
330 	}
331 
332 	*(ap->a_vpp) = vp;
333 
334 	if ((cnflags & MAKEENTRY) && vp->v_type != VSOCK)
335 		cache_enter(dvp, vp, cnp);
336 
337 unionfs_lookup_cleanup:
338 	if (uvp != NULLVP)
339 		vrele(uvp);
340 	if (lvp != NULLVP)
341 		vrele(lvp);
342 
343 	if (error == ENOENT && (cnflags & MAKEENTRY) != 0)
344 		cache_enter(dvp, NULLVP, cnp);
345 
346 unionfs_lookup_return:
347 
348 	UNIONFS_INTERNAL_DEBUG("unionfs_lookup: leave (%d)\n", error);
349 
350 	return (error);
351 }
352 
353 static int
354 unionfs_create(struct vop_create_args *ap)
355 {
356 	struct unionfs_node *dunp;
357 	struct componentname *cnp;
358 	struct vnode   *udvp;
359 	struct vnode   *vp;
360 	int		error;
361 
362 	UNIONFS_INTERNAL_DEBUG("unionfs_create: enter\n");
363 
364 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
365 
366 	dunp = VTOUNIONFS(ap->a_dvp);
367 	cnp = ap->a_cnp;
368 	udvp = dunp->un_uppervp;
369 	error = EROFS;
370 
371 	if (udvp != NULLVP) {
372 		error = VOP_CREATE(udvp, &vp, cnp, ap->a_vap);
373 		if (error != 0)
374 			goto unionfs_create_abort;
375 
376 		if (vp->v_type == VSOCK)
377 			*(ap->a_vpp) = vp;
378 		else {
379 			VOP_UNLOCK(vp);
380 			error = unionfs_nodeget(ap->a_dvp->v_mount, vp, NULLVP,
381 			    ap->a_dvp, ap->a_vpp, cnp);
382 			vrele(vp);
383 		}
384 	}
385 
386 unionfs_create_abort:
387 	UNIONFS_INTERNAL_DEBUG("unionfs_create: leave (%d)\n", error);
388 
389 	return (error);
390 }
391 
392 static int
393 unionfs_whiteout(struct vop_whiteout_args *ap)
394 {
395 	struct unionfs_node *dunp;
396 	struct componentname *cnp;
397 	struct vnode   *udvp;
398 	int		error;
399 
400 	UNIONFS_INTERNAL_DEBUG("unionfs_whiteout: enter\n");
401 
402 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
403 
404 	dunp = VTOUNIONFS(ap->a_dvp);
405 	cnp = ap->a_cnp;
406 	udvp = dunp->un_uppervp;
407 	error = EOPNOTSUPP;
408 
409 	if (udvp != NULLVP) {
410 		switch (ap->a_flags) {
411 		case CREATE:
412 		case DELETE:
413 		case LOOKUP:
414 			error = VOP_WHITEOUT(udvp, cnp, ap->a_flags);
415 			break;
416 		default:
417 			error = EINVAL;
418 			break;
419 		}
420 	}
421 
422 	UNIONFS_INTERNAL_DEBUG("unionfs_whiteout: leave (%d)\n", error);
423 
424 	return (error);
425 }
426 
427 static int
428 unionfs_mknod(struct vop_mknod_args *ap)
429 {
430 	struct unionfs_node *dunp;
431 	struct componentname *cnp;
432 	struct vnode   *udvp;
433 	struct vnode   *vp;
434 	int		error;
435 
436 	UNIONFS_INTERNAL_DEBUG("unionfs_mknod: enter\n");
437 
438 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
439 
440 	dunp = VTOUNIONFS(ap->a_dvp);
441 	cnp = ap->a_cnp;
442 	udvp = dunp->un_uppervp;
443 	error = EROFS;
444 
445 	if (udvp != NULLVP) {
446 		error = VOP_MKNOD(udvp, &vp, cnp, ap->a_vap);
447 		if (error != 0)
448 			goto unionfs_mknod_abort;
449 
450 		if (vp->v_type == VSOCK)
451 			*(ap->a_vpp) = vp;
452 		else {
453 			VOP_UNLOCK(vp);
454 			error = unionfs_nodeget(ap->a_dvp->v_mount, vp, NULLVP,
455 			    ap->a_dvp, ap->a_vpp, cnp);
456 			vrele(vp);
457 		}
458 	}
459 
460 unionfs_mknod_abort:
461 	UNIONFS_INTERNAL_DEBUG("unionfs_mknod: leave (%d)\n", error);
462 
463 	return (error);
464 }
465 
466 enum unionfs_lkupgrade {
467 	UNIONFS_LKUPGRADE_SUCCESS, /* lock successfully upgraded */
468 	UNIONFS_LKUPGRADE_ALREADY, /* lock already held exclusive */
469 	UNIONFS_LKUPGRADE_DOOMED   /* lock was upgraded, but vnode reclaimed */
470 };
471 
472 static inline enum unionfs_lkupgrade
473 unionfs_upgrade_lock(struct vnode *vp)
474 {
475 	ASSERT_VOP_LOCKED(vp, __func__);
476 
477 	if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
478 		return (UNIONFS_LKUPGRADE_ALREADY);
479 
480 	if (vn_lock(vp, LK_UPGRADE) != 0) {
481 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
482 		if (VN_IS_DOOMED(vp))
483 			return (UNIONFS_LKUPGRADE_DOOMED);
484 	}
485 	return (UNIONFS_LKUPGRADE_SUCCESS);
486 }
487 
488 static inline void
489 unionfs_downgrade_lock(struct vnode *vp, enum unionfs_lkupgrade status)
490 {
491 	if (status != UNIONFS_LKUPGRADE_ALREADY)
492 		vn_lock(vp, LK_DOWNGRADE | LK_RETRY);
493 }
494 
495 static int
496 unionfs_open(struct vop_open_args *ap)
497 {
498 	struct unionfs_node *unp;
499 	struct unionfs_node_status *unsp;
500 	struct vnode   *vp;
501 	struct vnode   *uvp;
502 	struct vnode   *lvp;
503 	struct vnode   *targetvp;
504 	struct ucred   *cred;
505 	struct thread  *td;
506 	int		error;
507 	enum unionfs_lkupgrade lkstatus;
508 
509 	UNIONFS_INTERNAL_DEBUG("unionfs_open: enter\n");
510 
511 	KASSERT_UNIONFS_VNODE(ap->a_vp);
512 
513 	error = 0;
514 	vp = ap->a_vp;
515 	targetvp = NULLVP;
516 	cred = ap->a_cred;
517 	td = ap->a_td;
518 
519 	/*
520 	 * The executable loader path may call this function with vp locked
521 	 * shared.  If the vnode is reclaimed while upgrading, we can't safely
522 	 * use unp or do anything else unionfs- specific.
523 	 */
524 	lkstatus = unionfs_upgrade_lock(vp);
525 	if (lkstatus == UNIONFS_LKUPGRADE_DOOMED) {
526 		error = ENOENT;
527 		goto unionfs_open_cleanup;
528 	}
529 
530 	unp = VTOUNIONFS(vp);
531 	uvp = unp->un_uppervp;
532 	lvp = unp->un_lowervp;
533 	unionfs_get_node_status(unp, td, &unsp);
534 
535 	if (unsp->uns_lower_opencnt > 0 || unsp->uns_upper_opencnt > 0) {
536 		/* vnode is already opend. */
537 		if (unsp->uns_upper_opencnt > 0)
538 			targetvp = uvp;
539 		else
540 			targetvp = lvp;
541 
542 		if (targetvp == lvp &&
543 		    (ap->a_mode & FWRITE) && lvp->v_type == VREG)
544 			targetvp = NULLVP;
545 	}
546 	if (targetvp == NULLVP) {
547 		if (uvp == NULLVP) {
548 			if ((ap->a_mode & FWRITE) && lvp->v_type == VREG) {
549 				error = unionfs_copyfile(unp,
550 				    !(ap->a_mode & O_TRUNC), cred, td);
551 				if (error != 0)
552 					goto unionfs_open_abort;
553 				targetvp = uvp = unp->un_uppervp;
554 			} else
555 				targetvp = lvp;
556 		} else
557 			targetvp = uvp;
558 	}
559 
560 	error = VOP_OPEN(targetvp, ap->a_mode, cred, td, ap->a_fp);
561 	if (error == 0) {
562 		if (targetvp == uvp) {
563 			if (uvp->v_type == VDIR && lvp != NULLVP &&
564 			    unsp->uns_lower_opencnt <= 0) {
565 				/* open lower for readdir */
566 				error = VOP_OPEN(lvp, FREAD, cred, td, NULL);
567 				if (error != 0) {
568 					VOP_CLOSE(uvp, ap->a_mode, cred, td);
569 					goto unionfs_open_abort;
570 				}
571 				unsp->uns_node_flag |= UNS_OPENL_4_READDIR;
572 				unsp->uns_lower_opencnt++;
573 			}
574 			unsp->uns_upper_opencnt++;
575 		} else {
576 			unsp->uns_lower_opencnt++;
577 			unsp->uns_lower_openmode = ap->a_mode;
578 		}
579 		vp->v_object = targetvp->v_object;
580 	}
581 
582 unionfs_open_abort:
583 	if (error != 0)
584 		unionfs_tryrem_node_status(unp, unsp);
585 
586 unionfs_open_cleanup:
587 	unionfs_downgrade_lock(vp, lkstatus);
588 
589 	UNIONFS_INTERNAL_DEBUG("unionfs_open: leave (%d)\n", error);
590 
591 	return (error);
592 }
593 
594 static int
595 unionfs_close(struct vop_close_args *ap)
596 {
597 	struct unionfs_node *unp;
598 	struct unionfs_node_status *unsp;
599 	struct ucred   *cred;
600 	struct thread  *td;
601 	struct vnode   *vp;
602 	struct vnode   *ovp;
603 	int		error;
604 	enum unionfs_lkupgrade lkstatus;
605 
606 	UNIONFS_INTERNAL_DEBUG("unionfs_close: enter\n");
607 
608 	KASSERT_UNIONFS_VNODE(ap->a_vp);
609 
610 	vp = ap->a_vp;
611 	cred = ap->a_cred;
612 	td = ap->a_td;
613 	error = 0;
614 
615 	/*
616 	 * If the vnode is reclaimed while upgrading, we can't safely use unp
617 	 * or do anything else unionfs- specific.
618 	 */
619 	lkstatus = unionfs_upgrade_lock(vp);
620 	if (lkstatus == UNIONFS_LKUPGRADE_DOOMED)
621 		goto unionfs_close_cleanup;
622 
623 	unp = VTOUNIONFS(vp);
624 	unionfs_get_node_status(unp, td, &unsp);
625 
626 	if (unsp->uns_lower_opencnt <= 0 && unsp->uns_upper_opencnt <= 0) {
627 #ifdef DIAGNOSTIC
628 		printf("unionfs_close: warning: open count is 0\n");
629 #endif
630 		if (unp->un_uppervp != NULLVP)
631 			ovp = unp->un_uppervp;
632 		else
633 			ovp = unp->un_lowervp;
634 	} else if (unsp->uns_upper_opencnt > 0)
635 		ovp = unp->un_uppervp;
636 	else
637 		ovp = unp->un_lowervp;
638 
639 	error = VOP_CLOSE(ovp, ap->a_fflag, cred, td);
640 
641 	if (error != 0)
642 		goto unionfs_close_abort;
643 
644 	vp->v_object = ovp->v_object;
645 
646 	if (ovp == unp->un_uppervp) {
647 		unsp->uns_upper_opencnt--;
648 		if (unsp->uns_upper_opencnt == 0) {
649 			if (unsp->uns_node_flag & UNS_OPENL_4_READDIR) {
650 				VOP_CLOSE(unp->un_lowervp, FREAD, cred, td);
651 				unsp->uns_node_flag &= ~UNS_OPENL_4_READDIR;
652 				unsp->uns_lower_opencnt--;
653 			}
654 			if (unsp->uns_lower_opencnt > 0)
655 				vp->v_object = unp->un_lowervp->v_object;
656 		}
657 	} else
658 		unsp->uns_lower_opencnt--;
659 
660 unionfs_close_abort:
661 	unionfs_tryrem_node_status(unp, unsp);
662 
663 unionfs_close_cleanup:
664 	unionfs_downgrade_lock(vp, lkstatus);
665 
666 	UNIONFS_INTERNAL_DEBUG("unionfs_close: leave (%d)\n", error);
667 
668 	return (error);
669 }
670 
671 /*
672  * Check the access mode toward shadow file/dir.
673  */
674 static int
675 unionfs_check_corrected_access(accmode_t accmode, struct vattr *va,
676     struct ucred *cred)
677 {
678 	uid_t		uid;	/* upper side vnode's uid */
679 	gid_t		gid;	/* upper side vnode's gid */
680 	u_short		vmode;	/* upper side vnode's mode */
681 	u_short		mask;
682 
683 	mask = 0;
684 	uid = va->va_uid;
685 	gid = va->va_gid;
686 	vmode = va->va_mode;
687 
688 	/* check owner */
689 	if (cred->cr_uid == uid) {
690 		if (accmode & VEXEC)
691 			mask |= S_IXUSR;
692 		if (accmode & VREAD)
693 			mask |= S_IRUSR;
694 		if (accmode & VWRITE)
695 			mask |= S_IWUSR;
696 		return ((vmode & mask) == mask ? 0 : EACCES);
697 	}
698 
699 	/* check group */
700 	if (groupmember(gid, cred)) {
701 		if (accmode & VEXEC)
702 			mask |= S_IXGRP;
703 		if (accmode & VREAD)
704 			mask |= S_IRGRP;
705 		if (accmode & VWRITE)
706 			mask |= S_IWGRP;
707 		return ((vmode & mask) == mask ? 0 : EACCES);
708 	}
709 
710 	/* check other */
711 	if (accmode & VEXEC)
712 		mask |= S_IXOTH;
713 	if (accmode & VREAD)
714 		mask |= S_IROTH;
715 	if (accmode & VWRITE)
716 		mask |= S_IWOTH;
717 
718 	return ((vmode & mask) == mask ? 0 : EACCES);
719 }
720 
721 static int
722 unionfs_access(struct vop_access_args *ap)
723 {
724 	struct unionfs_mount *ump;
725 	struct unionfs_node *unp;
726 	struct vnode   *uvp;
727 	struct vnode   *lvp;
728 	struct thread  *td;
729 	struct vattr	va;
730 	accmode_t	accmode;
731 	int		error;
732 
733 	UNIONFS_INTERNAL_DEBUG("unionfs_access: enter\n");
734 
735 	KASSERT_UNIONFS_VNODE(ap->a_vp);
736 
737 	ump = MOUNTTOUNIONFSMOUNT(ap->a_vp->v_mount);
738 	unp = VTOUNIONFS(ap->a_vp);
739 	uvp = unp->un_uppervp;
740 	lvp = unp->un_lowervp;
741 	td = ap->a_td;
742 	accmode = ap->a_accmode;
743 	error = EACCES;
744 
745 	if ((accmode & VWRITE) &&
746 	    (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)) {
747 		switch (ap->a_vp->v_type) {
748 		case VREG:
749 		case VDIR:
750 		case VLNK:
751 			return (EROFS);
752 		default:
753 			break;
754 		}
755 	}
756 
757 	if (uvp != NULLVP) {
758 		error = VOP_ACCESS(uvp, accmode, ap->a_cred, td);
759 
760 		UNIONFS_INTERNAL_DEBUG("unionfs_access: leave (%d)\n", error);
761 
762 		return (error);
763 	}
764 
765 	if (lvp != NULLVP) {
766 		if (accmode & VWRITE) {
767 			if ((ump->um_uppermp->mnt_flag & MNT_RDONLY) != 0) {
768 				switch (ap->a_vp->v_type) {
769 				case VREG:
770 				case VDIR:
771 				case VLNK:
772 					return (EROFS);
773 				default:
774 					break;
775 				}
776 			} else if (ap->a_vp->v_type == VREG ||
777 			    ap->a_vp->v_type == VDIR) {
778 				/* check shadow file/dir */
779 				if (ump->um_copymode != UNIONFS_TRANSPARENT) {
780 					error = unionfs_create_uppervattr(ump,
781 					    lvp, &va, ap->a_cred, td);
782 					if (error != 0)
783 						return (error);
784 
785 					error = unionfs_check_corrected_access(
786 					    accmode, &va, ap->a_cred);
787 					if (error != 0)
788 						return (error);
789 				}
790 			}
791 			accmode &= ~(VWRITE | VAPPEND);
792 			accmode |= VREAD; /* will copy to upper */
793 		}
794 		error = VOP_ACCESS(lvp, accmode, ap->a_cred, td);
795 	}
796 
797 	UNIONFS_INTERNAL_DEBUG("unionfs_access: leave (%d)\n", error);
798 
799 	return (error);
800 }
801 
802 static int
803 unionfs_getattr(struct vop_getattr_args *ap)
804 {
805 	struct unionfs_node *unp;
806 	struct unionfs_mount *ump;
807 	struct vnode   *uvp;
808 	struct vnode   *lvp;
809 	struct thread  *td;
810 	struct vattr	va;
811 	int		error;
812 
813 	UNIONFS_INTERNAL_DEBUG("unionfs_getattr: enter\n");
814 
815 	KASSERT_UNIONFS_VNODE(ap->a_vp);
816 
817 	unp = VTOUNIONFS(ap->a_vp);
818 	ump = MOUNTTOUNIONFSMOUNT(ap->a_vp->v_mount);
819 	uvp = unp->un_uppervp;
820 	lvp = unp->un_lowervp;
821 	td = curthread;
822 
823 	if (uvp != NULLVP) {
824 		if ((error = VOP_GETATTR(uvp, ap->a_vap, ap->a_cred)) == 0)
825 			ap->a_vap->va_fsid =
826 			    ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
827 
828 		UNIONFS_INTERNAL_DEBUG(
829 		    "unionfs_getattr: leave mode=%o, uid=%d, gid=%d (%d)\n",
830 		    ap->a_vap->va_mode, ap->a_vap->va_uid,
831 		    ap->a_vap->va_gid, error);
832 
833 		return (error);
834 	}
835 
836 	error = VOP_GETATTR(lvp, ap->a_vap, ap->a_cred);
837 
838 	if (error == 0 && (ump->um_uppermp->mnt_flag & MNT_RDONLY) == 0) {
839 		/* correct the attr toward shadow file/dir. */
840 		if (ap->a_vp->v_type == VREG || ap->a_vp->v_type == VDIR) {
841 			unionfs_create_uppervattr_core(ump, ap->a_vap, &va, td);
842 			ap->a_vap->va_mode = va.va_mode;
843 			ap->a_vap->va_uid = va.va_uid;
844 			ap->a_vap->va_gid = va.va_gid;
845 		}
846 	}
847 
848 	if (error == 0)
849 		ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
850 
851 	UNIONFS_INTERNAL_DEBUG(
852 	    "unionfs_getattr: leave mode=%o, uid=%d, gid=%d (%d)\n",
853 	    ap->a_vap->va_mode, ap->a_vap->va_uid, ap->a_vap->va_gid, error);
854 
855 	return (error);
856 }
857 
858 static int
859 unionfs_setattr(struct vop_setattr_args *ap)
860 {
861 	struct unionfs_node *unp;
862 	struct vnode   *uvp;
863 	struct vnode   *lvp;
864 	struct thread  *td;
865 	struct vattr   *vap;
866 	int		error;
867 
868 	UNIONFS_INTERNAL_DEBUG("unionfs_setattr: enter\n");
869 
870 	KASSERT_UNIONFS_VNODE(ap->a_vp);
871 
872 	error = EROFS;
873 	unp = VTOUNIONFS(ap->a_vp);
874 	uvp = unp->un_uppervp;
875 	lvp = unp->un_lowervp;
876 	td = curthread;
877 	vap = ap->a_vap;
878 
879 	if ((ap->a_vp->v_mount->mnt_flag & MNT_RDONLY) &&
880 	    (vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
881 	     vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
882 	     vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL))
883 		return (EROFS);
884 
885 	if (uvp == NULLVP && lvp->v_type == VREG) {
886 		error = unionfs_copyfile(unp, (vap->va_size != 0),
887 		    ap->a_cred, td);
888 		if (error != 0)
889 			return (error);
890 		uvp = unp->un_uppervp;
891 	}
892 
893 	if (uvp != NULLVP)
894 		error = VOP_SETATTR(uvp, vap, ap->a_cred);
895 
896 	UNIONFS_INTERNAL_DEBUG("unionfs_setattr: leave (%d)\n", error);
897 
898 	return (error);
899 }
900 
901 static int
902 unionfs_read(struct vop_read_args *ap)
903 {
904 	struct unionfs_node *unp;
905 	struct vnode   *tvp;
906 	int		error;
907 
908 	/* UNIONFS_INTERNAL_DEBUG("unionfs_read: enter\n"); */
909 
910 	KASSERT_UNIONFS_VNODE(ap->a_vp);
911 
912 	unp = VTOUNIONFS(ap->a_vp);
913 	tvp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
914 
915 	error = VOP_READ(tvp, ap->a_uio, ap->a_ioflag, ap->a_cred);
916 
917 	/* UNIONFS_INTERNAL_DEBUG("unionfs_read: leave (%d)\n", error); */
918 
919 	return (error);
920 }
921 
922 static int
923 unionfs_write(struct vop_write_args *ap)
924 {
925 	struct unionfs_node *unp;
926 	struct vnode   *tvp;
927 	int		error;
928 
929 	/* UNIONFS_INTERNAL_DEBUG("unionfs_write: enter\n"); */
930 
931 	KASSERT_UNIONFS_VNODE(ap->a_vp);
932 
933 	unp = VTOUNIONFS(ap->a_vp);
934 	tvp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
935 
936 	error = VOP_WRITE(tvp, ap->a_uio, ap->a_ioflag, ap->a_cred);
937 
938 	/* UNIONFS_INTERNAL_DEBUG("unionfs_write: leave (%d)\n", error); */
939 
940 	return (error);
941 }
942 
943 static int
944 unionfs_ioctl(struct vop_ioctl_args *ap)
945 {
946 	struct unionfs_node *unp;
947 	struct unionfs_node_status *unsp;
948 	struct vnode   *ovp;
949 	int error;
950 
951 	UNIONFS_INTERNAL_DEBUG("unionfs_ioctl: enter\n");
952 
953 	KASSERT_UNIONFS_VNODE(ap->a_vp);
954 
955  	vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
956 	unp = VTOUNIONFS(ap->a_vp);
957 	unionfs_get_node_status(unp, ap->a_td, &unsp);
958 	ovp = (unsp->uns_upper_opencnt ? unp->un_uppervp : unp->un_lowervp);
959 	unionfs_tryrem_node_status(unp, unsp);
960 	VOP_UNLOCK(ap->a_vp);
961 
962 	if (ovp == NULLVP)
963 		return (EBADF);
964 
965 	error = VOP_IOCTL(ovp, ap->a_command, ap->a_data, ap->a_fflag,
966 	    ap->a_cred, ap->a_td);
967 
968 	UNIONFS_INTERNAL_DEBUG("unionfs_ioctl: leave (%d)\n", error);
969 
970 	return (error);
971 }
972 
973 static int
974 unionfs_poll(struct vop_poll_args *ap)
975 {
976 	struct unionfs_node *unp;
977 	struct unionfs_node_status *unsp;
978 	struct vnode *ovp;
979 
980 	KASSERT_UNIONFS_VNODE(ap->a_vp);
981 
982  	vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
983 	unp = VTOUNIONFS(ap->a_vp);
984 	unionfs_get_node_status(unp, ap->a_td, &unsp);
985 	ovp = (unsp->uns_upper_opencnt ? unp->un_uppervp : unp->un_lowervp);
986 	unionfs_tryrem_node_status(unp, unsp);
987 	VOP_UNLOCK(ap->a_vp);
988 
989 	if (ovp == NULLVP)
990 		return (EBADF);
991 
992 	return (VOP_POLL(ovp, ap->a_events, ap->a_cred, ap->a_td));
993 }
994 
995 static int
996 unionfs_fsync(struct vop_fsync_args *ap)
997 {
998 	struct unionfs_node *unp;
999 	struct unionfs_node_status *unsp;
1000 	struct vnode *ovp;
1001 	enum unionfs_lkupgrade lkstatus;
1002 
1003 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1004 
1005 	unp = VTOUNIONFS(ap->a_vp);
1006 	lkstatus = unionfs_upgrade_lock(ap->a_vp);
1007 	if (lkstatus == UNIONFS_LKUPGRADE_DOOMED) {
1008 		unionfs_downgrade_lock(ap->a_vp, lkstatus);
1009 		return (ENOENT);
1010 	}
1011 	unionfs_get_node_status(unp, ap->a_td, &unsp);
1012 	ovp = (unsp->uns_upper_opencnt ? unp->un_uppervp : unp->un_lowervp);
1013 	unionfs_tryrem_node_status(unp, unsp);
1014 
1015 	unionfs_downgrade_lock(ap->a_vp, lkstatus);
1016 
1017 	if (ovp == NULLVP)
1018 		return (EBADF);
1019 
1020 	return (VOP_FSYNC(ovp, ap->a_waitfor, ap->a_td));
1021 }
1022 
1023 static int
1024 unionfs_remove(struct vop_remove_args *ap)
1025 {
1026 	char	       *path;
1027 	struct unionfs_node *dunp;
1028 	struct unionfs_node *unp;
1029 	struct unionfs_mount *ump;
1030 	struct vnode   *udvp;
1031 	struct vnode   *uvp;
1032 	struct vnode   *lvp;
1033 	struct vnode   *vp;
1034 	struct componentname *cnp;
1035 	struct componentname cn;
1036 	struct thread  *td;
1037 	int		error;
1038 	int		pathlen;
1039 
1040 	UNIONFS_INTERNAL_DEBUG("unionfs_remove: enter\n");
1041 
1042 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
1043 
1044 	error = 0;
1045 	dunp = VTOUNIONFS(ap->a_dvp);
1046 	udvp = dunp->un_uppervp;
1047 	cnp = ap->a_cnp;
1048 	td = curthread;
1049 
1050 	if (ap->a_vp->v_op != &unionfs_vnodeops) {
1051 		if (ap->a_vp->v_type != VSOCK)
1052 			return (EINVAL);
1053 		ump = NULL;
1054 		vp = uvp = lvp = NULLVP;
1055 		/* search vnode */
1056 		VOP_UNLOCK(ap->a_vp);
1057 		error = unionfs_relookup(udvp, &vp, cnp, &cn, td,
1058 		    cnp->cn_nameptr, cnp->cn_namelen, DELETE);
1059 		if (error != 0 && error != ENOENT) {
1060 			vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
1061 			return (error);
1062 		}
1063 
1064 		if (error == 0 && vp == ap->a_vp) {
1065 			/* target vnode in upper */
1066 			uvp = vp;
1067 			vrele(vp);
1068 		} else {
1069 			/* target vnode in lower */
1070 			if (vp != NULLVP) {
1071 				if (udvp == vp)
1072 					vrele(vp);
1073 				else
1074 					vput(vp);
1075 			}
1076 			vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
1077 			lvp = ap->a_vp;
1078 		}
1079 		path = cnp->cn_nameptr;
1080 		pathlen = cnp->cn_namelen;
1081 	} else {
1082 		ump = MOUNTTOUNIONFSMOUNT(ap->a_vp->v_mount);
1083 		unp = VTOUNIONFS(ap->a_vp);
1084 		uvp = unp->un_uppervp;
1085 		lvp = unp->un_lowervp;
1086 		path = unp->un_path;
1087 		pathlen = unp->un_pathlen;
1088 	}
1089 
1090 	if (udvp == NULLVP)
1091 		return (EROFS);
1092 
1093 	if (uvp != NULLVP) {
1094 		/*
1095 		 * XXX: if the vnode type is VSOCK, it will create whiteout
1096 		 *      after remove.
1097 		 */
1098 		if (ump == NULL || ump->um_whitemode == UNIONFS_WHITE_ALWAYS ||
1099 		    lvp != NULLVP)
1100 			cnp->cn_flags |= DOWHITEOUT;
1101 		error = VOP_REMOVE(udvp, uvp, cnp);
1102 	} else if (lvp != NULLVP)
1103 		error = unionfs_mkwhiteout(udvp, cnp, td, path, pathlen);
1104 
1105 	UNIONFS_INTERNAL_DEBUG("unionfs_remove: leave (%d)\n", error);
1106 
1107 	return (error);
1108 }
1109 
1110 static int
1111 unionfs_link(struct vop_link_args *ap)
1112 {
1113 	struct unionfs_node *dunp;
1114 	struct unionfs_node *unp;
1115 	struct vnode   *udvp;
1116 	struct vnode   *uvp;
1117 	struct componentname *cnp;
1118 	struct thread  *td;
1119 	int		error;
1120 	int		needrelookup;
1121 
1122 	UNIONFS_INTERNAL_DEBUG("unionfs_link: enter\n");
1123 
1124 	KASSERT_UNIONFS_VNODE(ap->a_tdvp);
1125 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1126 
1127 	error = 0;
1128 	needrelookup = 0;
1129 	dunp = VTOUNIONFS(ap->a_tdvp);
1130 	unp = NULL;
1131 	udvp = dunp->un_uppervp;
1132 	uvp = NULLVP;
1133 	cnp = ap->a_cnp;
1134 	td = curthread;
1135 
1136 	if (udvp == NULLVP)
1137 		return (EROFS);
1138 
1139 	if (ap->a_vp->v_op != &unionfs_vnodeops)
1140 		uvp = ap->a_vp;
1141 	else {
1142 		unp = VTOUNIONFS(ap->a_vp);
1143 
1144 		if (unp->un_uppervp == NULLVP) {
1145 			if (ap->a_vp->v_type != VREG)
1146 				return (EOPNOTSUPP);
1147 
1148 			error = unionfs_copyfile(unp, 1, cnp->cn_cred, td);
1149 			if (error != 0)
1150 				return (error);
1151 			needrelookup = 1;
1152 		}
1153 		uvp = unp->un_uppervp;
1154 	}
1155 
1156 	if (needrelookup != 0)
1157 		error = unionfs_relookup_for_create(ap->a_tdvp, cnp, td);
1158 
1159 	if (error == 0)
1160 		error = VOP_LINK(udvp, uvp, cnp);
1161 
1162 	UNIONFS_INTERNAL_DEBUG("unionfs_link: leave (%d)\n", error);
1163 
1164 	return (error);
1165 }
1166 
1167 static int
1168 unionfs_rename(struct vop_rename_args *ap)
1169 {
1170 	struct vnode   *fdvp;
1171 	struct vnode   *fvp;
1172 	struct componentname *fcnp;
1173 	struct vnode   *tdvp;
1174 	struct vnode   *tvp;
1175 	struct componentname *tcnp;
1176 	struct vnode   *ltdvp;
1177 	struct vnode   *ltvp;
1178 	struct thread  *td;
1179 
1180 	/* rename target vnodes */
1181 	struct vnode   *rfdvp;
1182 	struct vnode   *rfvp;
1183 	struct vnode   *rtdvp;
1184 	struct vnode   *rtvp;
1185 
1186 	struct unionfs_mount *ump;
1187 	struct unionfs_node *unp;
1188 	int		error;
1189 	int		needrelookup;
1190 
1191 	UNIONFS_INTERNAL_DEBUG("unionfs_rename: enter\n");
1192 
1193 	error = 0;
1194 	fdvp = ap->a_fdvp;
1195 	fvp = ap->a_fvp;
1196 	fcnp = ap->a_fcnp;
1197 	tdvp = ap->a_tdvp;
1198 	tvp = ap->a_tvp;
1199 	tcnp = ap->a_tcnp;
1200 	ltdvp = NULLVP;
1201 	ltvp = NULLVP;
1202 	td = curthread;
1203 	rfdvp = fdvp;
1204 	rfvp = fvp;
1205 	rtdvp = tdvp;
1206 	rtvp = tvp;
1207 	needrelookup = 0;
1208 
1209 	/* check for cross device rename */
1210 	if (fvp->v_mount != tdvp->v_mount ||
1211 	    (tvp != NULLVP && fvp->v_mount != tvp->v_mount)) {
1212 		if (fvp->v_op != &unionfs_vnodeops)
1213 			error = ENODEV;
1214 		else
1215 			error = EXDEV;
1216 		goto unionfs_rename_abort;
1217 	}
1218 
1219 	/* Renaming a file to itself has no effect. */
1220 	if (fvp == tvp)
1221 		goto unionfs_rename_abort;
1222 
1223 	/*
1224 	 * from/to vnode is unionfs node.
1225 	 */
1226 
1227 	KASSERT_UNIONFS_VNODE(fdvp);
1228 	KASSERT_UNIONFS_VNODE(fvp);
1229 	KASSERT_UNIONFS_VNODE(tdvp);
1230 	if (tvp != NULLVP)
1231 		KASSERT_UNIONFS_VNODE(tvp);
1232 
1233 	unp = VTOUNIONFS(fdvp);
1234 #ifdef UNIONFS_IDBG_RENAME
1235 	UNIONFS_INTERNAL_DEBUG("fdvp=%p, ufdvp=%p, lfdvp=%p\n",
1236 	    fdvp, unp->un_uppervp, unp->un_lowervp);
1237 #endif
1238 	if (unp->un_uppervp == NULLVP) {
1239 		error = ENODEV;
1240 		goto unionfs_rename_abort;
1241 	}
1242 	rfdvp = unp->un_uppervp;
1243 	vref(rfdvp);
1244 
1245 	unp = VTOUNIONFS(fvp);
1246 #ifdef UNIONFS_IDBG_RENAME
1247 	UNIONFS_INTERNAL_DEBUG("fvp=%p, ufvp=%p, lfvp=%p\n",
1248 	    fvp, unp->un_uppervp, unp->un_lowervp);
1249 #endif
1250 	ump = MOUNTTOUNIONFSMOUNT(fvp->v_mount);
1251 	if (unp->un_uppervp == NULLVP) {
1252 		switch (fvp->v_type) {
1253 		case VREG:
1254 			if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1255 				goto unionfs_rename_abort;
1256 			error = unionfs_copyfile(unp, 1, fcnp->cn_cred, td);
1257 			VOP_UNLOCK(fvp);
1258 			if (error != 0)
1259 				goto unionfs_rename_abort;
1260 			break;
1261 		case VDIR:
1262 			if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1263 				goto unionfs_rename_abort;
1264 			error = unionfs_mkshadowdir(ump, rfdvp, unp, fcnp, td);
1265 			VOP_UNLOCK(fvp);
1266 			if (error != 0)
1267 				goto unionfs_rename_abort;
1268 			break;
1269 		default:
1270 			error = ENODEV;
1271 			goto unionfs_rename_abort;
1272 		}
1273 
1274 		needrelookup = 1;
1275 	}
1276 
1277 	if (unp->un_lowervp != NULLVP)
1278 		fcnp->cn_flags |= DOWHITEOUT;
1279 	rfvp = unp->un_uppervp;
1280 	vref(rfvp);
1281 
1282 	unp = VTOUNIONFS(tdvp);
1283 #ifdef UNIONFS_IDBG_RENAME
1284 	UNIONFS_INTERNAL_DEBUG("tdvp=%p, utdvp=%p, ltdvp=%p\n",
1285 	    tdvp, unp->un_uppervp, unp->un_lowervp);
1286 #endif
1287 	if (unp->un_uppervp == NULLVP) {
1288 		error = ENODEV;
1289 		goto unionfs_rename_abort;
1290 	}
1291 	rtdvp = unp->un_uppervp;
1292 	ltdvp = unp->un_lowervp;
1293 	vref(rtdvp);
1294 
1295 	if (tdvp == tvp) {
1296 		rtvp = rtdvp;
1297 		vref(rtvp);
1298 	} else if (tvp != NULLVP) {
1299 		unp = VTOUNIONFS(tvp);
1300 #ifdef UNIONFS_IDBG_RENAME
1301 		UNIONFS_INTERNAL_DEBUG("tvp=%p, utvp=%p, ltvp=%p\n",
1302 		    tvp, unp->un_uppervp, unp->un_lowervp);
1303 #endif
1304 		if (unp->un_uppervp == NULLVP)
1305 			rtvp = NULLVP;
1306 		else {
1307 			if (tvp->v_type == VDIR) {
1308 				error = EINVAL;
1309 				goto unionfs_rename_abort;
1310 			}
1311 			rtvp = unp->un_uppervp;
1312 			ltvp = unp->un_lowervp;
1313 			vref(rtvp);
1314 		}
1315 	}
1316 
1317 	if (rfvp == rtvp)
1318 		goto unionfs_rename_abort;
1319 
1320 	if (needrelookup != 0) {
1321 		if ((error = vn_lock(fdvp, LK_EXCLUSIVE)) != 0)
1322 			goto unionfs_rename_abort;
1323 		error = unionfs_relookup_for_delete(fdvp, fcnp, td);
1324 		VOP_UNLOCK(fdvp);
1325 		if (error != 0)
1326 			goto unionfs_rename_abort;
1327 
1328 		/* Lock of tvp is canceled in order to avoid recursive lock. */
1329 		if (tvp != NULLVP && tvp != tdvp)
1330 			VOP_UNLOCK(tvp);
1331 		error = unionfs_relookup_for_rename(tdvp, tcnp, td);
1332 		if (tvp != NULLVP && tvp != tdvp)
1333 			vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
1334 		if (error != 0)
1335 			goto unionfs_rename_abort;
1336 	}
1337 
1338 	error = VOP_RENAME(rfdvp, rfvp, fcnp, rtdvp, rtvp, tcnp);
1339 
1340 	if (error == 0) {
1341 		if (rtvp != NULLVP && rtvp->v_type == VDIR)
1342 			cache_purge(tdvp);
1343 		if (fvp->v_type == VDIR && fdvp != tdvp)
1344 			cache_purge(fdvp);
1345 	}
1346 
1347 	if (ltdvp != NULLVP)
1348 		VOP_UNLOCK(ltdvp);
1349 	if (tdvp != rtdvp)
1350 		vrele(tdvp);
1351 	if (ltvp != NULLVP)
1352 		VOP_UNLOCK(ltvp);
1353 	if (tvp != rtvp && tvp != NULLVP) {
1354 		if (rtvp == NULLVP)
1355 			vput(tvp);
1356 		else
1357 			vrele(tvp);
1358 	}
1359 	if (fdvp != rfdvp)
1360 		vrele(fdvp);
1361 	if (fvp != rfvp)
1362 		vrele(fvp);
1363 
1364 	UNIONFS_INTERNAL_DEBUG("unionfs_rename: leave (%d)\n", error);
1365 
1366 	return (error);
1367 
1368 unionfs_rename_abort:
1369 	vput(tdvp);
1370 	if (tdvp != rtdvp)
1371 		vrele(rtdvp);
1372 	if (tvp != NULLVP) {
1373 		if (tdvp != tvp)
1374 			vput(tvp);
1375 		else
1376 			vrele(tvp);
1377 	}
1378 	if (tvp != rtvp && rtvp != NULLVP)
1379 		vrele(rtvp);
1380 	if (fdvp != rfdvp)
1381 		vrele(rfdvp);
1382 	if (fvp != rfvp)
1383 		vrele(rfvp);
1384 	vrele(fdvp);
1385 	vrele(fvp);
1386 
1387 	UNIONFS_INTERNAL_DEBUG("unionfs_rename: leave (%d)\n", error);
1388 
1389 	return (error);
1390 }
1391 
1392 static int
1393 unionfs_mkdir(struct vop_mkdir_args *ap)
1394 {
1395 	struct unionfs_node *dunp;
1396 	struct componentname *cnp;
1397 	struct vnode   *dvp;
1398 	struct vnode   *udvp;
1399 	struct vnode   *uvp;
1400 	struct vattr	va;
1401 	int		error;
1402 	int		lkflags;
1403 
1404 	UNIONFS_INTERNAL_DEBUG("unionfs_mkdir: enter\n");
1405 
1406 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
1407 
1408 	error = EROFS;
1409 	dvp = ap->a_dvp;
1410 	dunp = VTOUNIONFS(dvp);
1411 	cnp = ap->a_cnp;
1412 	lkflags = cnp->cn_lkflags;
1413 	udvp = dunp->un_uppervp;
1414 
1415 	if (udvp != NULLVP) {
1416 		vref(udvp);
1417 		/* check opaque */
1418 		if (!(cnp->cn_flags & ISWHITEOUT)) {
1419 			error = VOP_GETATTR(udvp, &va, cnp->cn_cred);
1420 			if (error != 0)
1421 				goto unionfs_mkdir_cleanup;
1422 			if ((va.va_flags & OPAQUE) != 0)
1423 				cnp->cn_flags |= ISWHITEOUT;
1424 		}
1425 
1426 		if ((error = VOP_MKDIR(udvp, &uvp, cnp, ap->a_vap)) == 0) {
1427 			VOP_UNLOCK(uvp);
1428 			cnp->cn_lkflags = LK_EXCLUSIVE;
1429 			/*
1430 			 * The underlying VOP_MKDIR() implementation may have
1431 			 * temporarily dropped the parent directory vnode lock.
1432 			 * Because the unionfs vnode ordinarily shares that
1433 			 * lock, this may allow the unionfs vnode to be reclaimed
1434 			 * and its lock field reset.  In that case, the unionfs
1435 			 * vnode is effectively no longer locked, and we must
1436 			 * explicitly lock it before returning in order to meet
1437 			 * the locking requirements of VOP_MKDIR().
1438 			 */
1439 			if (__predict_false(VTOUNIONFS(dvp) == NULL)) {
1440 				error = ENOENT;
1441 				goto unionfs_mkdir_cleanup;
1442 			}
1443 			error = unionfs_nodeget(dvp->v_mount, uvp, NULLVP,
1444 			    dvp, ap->a_vpp, cnp);
1445 			cnp->cn_lkflags = lkflags;
1446 			vrele(uvp);
1447 		}
1448 	}
1449 
1450 unionfs_mkdir_cleanup:
1451 
1452 	if (__predict_false(VTOUNIONFS(dvp) == NULL)) {
1453 		vput(udvp);
1454 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
1455 	} else if (udvp != NULLVP)
1456 		vrele(udvp);
1457 
1458 	UNIONFS_INTERNAL_DEBUG("unionfs_mkdir: leave (%d)\n", error);
1459 
1460 	return (error);
1461 }
1462 
1463 static int
1464 unionfs_rmdir(struct vop_rmdir_args *ap)
1465 {
1466 	struct unionfs_node *dunp;
1467 	struct unionfs_node *unp;
1468 	struct unionfs_mount *ump;
1469 	struct componentname *cnp;
1470 	struct thread  *td;
1471 	struct vnode   *udvp;
1472 	struct vnode   *uvp;
1473 	struct vnode   *lvp;
1474 	int		error;
1475 
1476 	UNIONFS_INTERNAL_DEBUG("unionfs_rmdir: enter\n");
1477 
1478 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
1479 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1480 
1481 	error = 0;
1482 	dunp = VTOUNIONFS(ap->a_dvp);
1483 	unp = VTOUNIONFS(ap->a_vp);
1484 	cnp = ap->a_cnp;
1485 	td = curthread;
1486 	udvp = dunp->un_uppervp;
1487 	uvp = unp->un_uppervp;
1488 	lvp = unp->un_lowervp;
1489 
1490 	if (udvp == NULLVP)
1491 		return (EROFS);
1492 
1493 	if (udvp == uvp)
1494 		return (EOPNOTSUPP);
1495 
1496 	if (uvp != NULLVP) {
1497 		if (lvp != NULLVP) {
1498 			error = unionfs_check_rmdir(ap->a_vp, cnp->cn_cred, td);
1499 			if (error != 0)
1500 				return (error);
1501 		}
1502 		ump = MOUNTTOUNIONFSMOUNT(ap->a_vp->v_mount);
1503 		if (ump->um_whitemode == UNIONFS_WHITE_ALWAYS || lvp != NULLVP)
1504 			cnp->cn_flags |= DOWHITEOUT;
1505 		/*
1506 		 * The relookup path will need to relock the parent dvp and
1507 		 * possibly the vp as well.  Locking is expected to be done
1508 		 * in parent->child order; drop the lock on vp to avoid LOR
1509 		 * and potential recursion on vp's lock.
1510 		 * vp is expected to remain referenced during VOP_RMDIR(),
1511 		 * so vref/vrele should not be necessary here.
1512 		 */
1513 		VOP_UNLOCK(ap->a_vp);
1514 		VNPASS(vrefcnt(ap->a_vp) > 0, ap->a_vp);
1515 		error = unionfs_relookup_for_delete(ap->a_dvp, cnp, td);
1516 		vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
1517 		/*
1518 		 * VOP_RMDIR is dispatched against udvp, so if uvp became
1519 		 * doomed while the lock was dropped above the target
1520 		 * filesystem may not be able to cope.
1521 		 */
1522 		if (error == 0 && VN_IS_DOOMED(uvp))
1523 			error = ENOENT;
1524 		if (error == 0)
1525 			error = VOP_RMDIR(udvp, uvp, cnp);
1526 	} else if (lvp != NULLVP)
1527 		error = unionfs_mkwhiteout(udvp, cnp, td,
1528 		    unp->un_path, unp->un_pathlen);
1529 
1530 	if (error == 0) {
1531 		cache_purge(ap->a_dvp);
1532 		cache_purge(ap->a_vp);
1533 	}
1534 
1535 	UNIONFS_INTERNAL_DEBUG("unionfs_rmdir: leave (%d)\n", error);
1536 
1537 	return (error);
1538 }
1539 
1540 static int
1541 unionfs_symlink(struct vop_symlink_args *ap)
1542 {
1543 	struct unionfs_node *dunp;
1544 	struct componentname *cnp;
1545 	struct vnode   *udvp;
1546 	struct vnode   *uvp;
1547 	int		error;
1548 	int		lkflags;
1549 
1550 	UNIONFS_INTERNAL_DEBUG("unionfs_symlink: enter\n");
1551 
1552 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
1553 
1554 	error = EROFS;
1555 	dunp = VTOUNIONFS(ap->a_dvp);
1556 	cnp = ap->a_cnp;
1557 	lkflags = cnp->cn_lkflags;
1558 	udvp = dunp->un_uppervp;
1559 
1560 	if (udvp != NULLVP) {
1561 		error = VOP_SYMLINK(udvp, &uvp, cnp, ap->a_vap, ap->a_target);
1562 		if (error == 0) {
1563 			VOP_UNLOCK(uvp);
1564 			cnp->cn_lkflags = LK_EXCLUSIVE;
1565 			error = unionfs_nodeget(ap->a_dvp->v_mount, uvp, NULLVP,
1566 			    ap->a_dvp, ap->a_vpp, cnp);
1567 			cnp->cn_lkflags = lkflags;
1568 			vrele(uvp);
1569 		}
1570 	}
1571 
1572 	UNIONFS_INTERNAL_DEBUG("unionfs_symlink: leave (%d)\n", error);
1573 
1574 	return (error);
1575 }
1576 
1577 static int
1578 unionfs_readdir(struct vop_readdir_args *ap)
1579 {
1580 	struct unionfs_node *unp;
1581 	struct unionfs_node_status *unsp;
1582 	struct uio     *uio;
1583 	struct vnode   *vp;
1584 	struct vnode   *uvp;
1585 	struct vnode   *lvp;
1586 	struct thread  *td;
1587 	struct vattr    va;
1588 
1589 	uint64_t	*cookies_bk;
1590 	int		error;
1591 	int		eofflag;
1592 	int		ncookies_bk;
1593 	int		uio_offset_bk;
1594 	enum unionfs_lkupgrade lkstatus;
1595 
1596 	UNIONFS_INTERNAL_DEBUG("unionfs_readdir: enter\n");
1597 
1598 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1599 
1600 	error = 0;
1601 	eofflag = 0;
1602 	uio_offset_bk = 0;
1603 	uio = ap->a_uio;
1604 	uvp = NULLVP;
1605 	lvp = NULLVP;
1606 	td = uio->uio_td;
1607 	ncookies_bk = 0;
1608 	cookies_bk = NULL;
1609 
1610 	vp = ap->a_vp;
1611 	if (vp->v_type != VDIR)
1612 		return (ENOTDIR);
1613 
1614 	/*
1615 	 * If the vnode is reclaimed while upgrading, we can't safely use unp
1616 	 * or do anything else unionfs- specific.
1617 	 */
1618 	lkstatus = unionfs_upgrade_lock(vp);
1619 	if (lkstatus == UNIONFS_LKUPGRADE_DOOMED)
1620 		error = EBADF;
1621 	if (error == 0) {
1622 		unp = VTOUNIONFS(vp);
1623 		uvp = unp->un_uppervp;
1624 		lvp = unp->un_lowervp;
1625 		/* check the open count. unionfs needs open before readdir. */
1626 		unionfs_get_node_status(unp, td, &unsp);
1627 		if ((uvp != NULLVP && unsp->uns_upper_opencnt <= 0) ||
1628 			(lvp != NULLVP && unsp->uns_lower_opencnt <= 0)) {
1629 			unionfs_tryrem_node_status(unp, unsp);
1630 			error = EBADF;
1631 		}
1632 	}
1633 	unionfs_downgrade_lock(vp, lkstatus);
1634 	if (error != 0)
1635 		goto unionfs_readdir_exit;
1636 
1637 	/* check opaque */
1638 	if (uvp != NULLVP && lvp != NULLVP) {
1639 		if ((error = VOP_GETATTR(uvp, &va, ap->a_cred)) != 0)
1640 			goto unionfs_readdir_exit;
1641 		if (va.va_flags & OPAQUE)
1642 			lvp = NULLVP;
1643 	}
1644 
1645 	/* upper only */
1646 	if (uvp != NULLVP && lvp == NULLVP) {
1647 		error = VOP_READDIR(uvp, uio, ap->a_cred, ap->a_eofflag,
1648 		    ap->a_ncookies, ap->a_cookies);
1649 		unsp->uns_readdir_status = 0;
1650 
1651 		goto unionfs_readdir_exit;
1652 	}
1653 
1654 	/* lower only */
1655 	if (uvp == NULLVP && lvp != NULLVP) {
1656 		error = VOP_READDIR(lvp, uio, ap->a_cred, ap->a_eofflag,
1657 		    ap->a_ncookies, ap->a_cookies);
1658 		unsp->uns_readdir_status = 2;
1659 
1660 		goto unionfs_readdir_exit;
1661 	}
1662 
1663 	/*
1664 	 * readdir upper and lower
1665 	 */
1666 	KASSERT(uvp != NULLVP, ("unionfs_readdir: null upper vp"));
1667 	KASSERT(lvp != NULLVP, ("unionfs_readdir: null lower vp"));
1668 	if (uio->uio_offset == 0)
1669 		unsp->uns_readdir_status = 0;
1670 
1671 	if (unsp->uns_readdir_status == 0) {
1672 		/* read upper */
1673 		error = VOP_READDIR(uvp, uio, ap->a_cred, &eofflag,
1674 				    ap->a_ncookies, ap->a_cookies);
1675 
1676 		if (error != 0 || eofflag == 0)
1677 			goto unionfs_readdir_exit;
1678 		unsp->uns_readdir_status = 1;
1679 
1680 		/*
1681 		 * UFS(and other FS) needs size of uio_resid larger than
1682 		 * DIRBLKSIZ.
1683 		 * size of DIRBLKSIZ equals DEV_BSIZE.
1684 		 * (see: ufs/ufs/ufs_vnops.c ufs_readdir func , ufs/ufs/dir.h)
1685 		 */
1686 		if (uio->uio_resid <= (uio->uio_resid & (DEV_BSIZE -1)))
1687 			goto unionfs_readdir_exit;
1688 
1689 		/*
1690 		 * Backup cookies.
1691 		 * It prepares to readdir in lower.
1692 		 */
1693 		if (ap->a_ncookies != NULL) {
1694 			ncookies_bk = *(ap->a_ncookies);
1695 			*(ap->a_ncookies) = 0;
1696 		}
1697 		if (ap->a_cookies != NULL) {
1698 			cookies_bk = *(ap->a_cookies);
1699 			*(ap->a_cookies) = NULL;
1700 		}
1701 	}
1702 
1703 	/* initialize for readdir in lower */
1704 	if (unsp->uns_readdir_status == 1) {
1705 		unsp->uns_readdir_status = 2;
1706 		/*
1707 		 * Backup uio_offset. See the comment after the
1708 		 * VOP_READDIR call on the lower layer.
1709 		 */
1710 		uio_offset_bk = uio->uio_offset;
1711 		uio->uio_offset = 0;
1712 	}
1713 
1714 	if (lvp == NULLVP) {
1715 		error = EBADF;
1716 		goto unionfs_readdir_exit;
1717 	}
1718 	/* read lower */
1719 	error = VOP_READDIR(lvp, uio, ap->a_cred, ap->a_eofflag,
1720 			    ap->a_ncookies, ap->a_cookies);
1721 
1722 	/*
1723 	 * We can't return an uio_offset of 0: this would trigger an
1724 	 * infinite loop, because the next call to unionfs_readdir would
1725 	 * always restart with the upper layer (uio_offset == 0) and
1726 	 * always return some data.
1727 	 *
1728 	 * This happens when the lower layer root directory is removed.
1729 	 * (A root directory deleting of unionfs should not be permitted.
1730 	 *  But current VFS can not do it.)
1731 	 */
1732 	if (uio->uio_offset == 0)
1733 		uio->uio_offset = uio_offset_bk;
1734 
1735 	if (cookies_bk != NULL) {
1736 		/* merge cookies */
1737 		int		size;
1738 		uint64_t         *newcookies, *pos;
1739 
1740 		size = *(ap->a_ncookies) + ncookies_bk;
1741 		newcookies = (uint64_t *) malloc(size * sizeof(*newcookies),
1742 		    M_TEMP, M_WAITOK);
1743 		pos = newcookies;
1744 
1745 		memcpy(pos, cookies_bk, ncookies_bk * sizeof(*newcookies));
1746 		pos += ncookies_bk;
1747 		memcpy(pos, *(ap->a_cookies),
1748 		    *(ap->a_ncookies) * sizeof(*newcookies));
1749 		free(cookies_bk, M_TEMP);
1750 		free(*(ap->a_cookies), M_TEMP);
1751 		*(ap->a_ncookies) = size;
1752 		*(ap->a_cookies) = newcookies;
1753 	}
1754 
1755 unionfs_readdir_exit:
1756 	if (error != 0 && ap->a_eofflag != NULL)
1757 		*(ap->a_eofflag) = 1;
1758 
1759 	UNIONFS_INTERNAL_DEBUG("unionfs_readdir: leave (%d)\n", error);
1760 
1761 	return (error);
1762 }
1763 
1764 static int
1765 unionfs_readlink(struct vop_readlink_args *ap)
1766 {
1767 	struct unionfs_node *unp;
1768 	struct vnode   *vp;
1769 	int error;
1770 
1771 	UNIONFS_INTERNAL_DEBUG("unionfs_readlink: enter\n");
1772 
1773 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1774 
1775 	unp = VTOUNIONFS(ap->a_vp);
1776 	vp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
1777 
1778 	error = VOP_READLINK(vp, ap->a_uio, ap->a_cred);
1779 
1780 	UNIONFS_INTERNAL_DEBUG("unionfs_readlink: leave (%d)\n", error);
1781 
1782 	return (error);
1783 }
1784 
1785 static int
1786 unionfs_getwritemount(struct vop_getwritemount_args *ap)
1787 {
1788 	struct unionfs_node *unp;
1789 	struct vnode   *uvp;
1790 	struct vnode   *vp, *ovp;
1791 	int		error;
1792 
1793 	UNIONFS_INTERNAL_DEBUG("unionfs_getwritemount: enter\n");
1794 
1795 	error = 0;
1796 	vp = ap->a_vp;
1797 	uvp = NULLVP;
1798 
1799 	VI_LOCK(vp);
1800 	unp = VTOUNIONFS(vp);
1801 	if (unp != NULL)
1802 		uvp = unp->un_uppervp;
1803 
1804 	/*
1805 	 * If our node has no upper vnode, check the parent directory.
1806 	 * We may be initiating a write operation that will produce a
1807 	 * new upper vnode through CoW.
1808 	 */
1809 	if (uvp == NULLVP && unp != NULL) {
1810 		ovp = vp;
1811 		vp = unp->un_dvp;
1812 		/*
1813 		 * Only the root vnode should have an empty parent, but it
1814 		 * should not have an empty uppervp, so we shouldn't get here.
1815 		 */
1816 		VNASSERT(vp != NULL, ovp, ("%s: NULL parent vnode", __func__));
1817 		VI_UNLOCK(ovp);
1818 		VI_LOCK(vp);
1819 		unp = VTOUNIONFS(vp);
1820 		if (unp != NULL)
1821 			uvp = unp->un_uppervp;
1822 		if (uvp == NULLVP)
1823 			error = EACCES;
1824 	}
1825 
1826 	if (uvp != NULLVP) {
1827 		vholdnz(uvp);
1828 		VI_UNLOCK(vp);
1829 		error = VOP_GETWRITEMOUNT(uvp, ap->a_mpp);
1830 		vdrop(uvp);
1831 	} else {
1832 		VI_UNLOCK(vp);
1833 		*(ap->a_mpp) = NULL;
1834 	}
1835 
1836 	UNIONFS_INTERNAL_DEBUG("unionfs_getwritemount: leave (%d)\n", error);
1837 
1838 	return (error);
1839 }
1840 
1841 static int
1842 unionfs_inactive(struct vop_inactive_args *ap)
1843 {
1844 	ap->a_vp->v_object = NULL;
1845 	vrecycle(ap->a_vp);
1846 	return (0);
1847 }
1848 
1849 static int
1850 unionfs_reclaim(struct vop_reclaim_args *ap)
1851 {
1852 	/* UNIONFS_INTERNAL_DEBUG("unionfs_reclaim: enter\n"); */
1853 
1854 	unionfs_noderem(ap->a_vp);
1855 
1856 	/* UNIONFS_INTERNAL_DEBUG("unionfs_reclaim: leave\n"); */
1857 
1858 	return (0);
1859 }
1860 
1861 static int
1862 unionfs_print(struct vop_print_args *ap)
1863 {
1864 	struct unionfs_node *unp;
1865 	/* struct unionfs_node_status *unsp; */
1866 
1867 	unp = VTOUNIONFS(ap->a_vp);
1868 	/* unionfs_get_node_status(unp, curthread, &unsp); */
1869 
1870 	printf("unionfs_vp=%p, uppervp=%p, lowervp=%p\n",
1871 	    ap->a_vp, unp->un_uppervp, unp->un_lowervp);
1872 	/*
1873 	printf("unionfs opencnt: uppervp=%d, lowervp=%d\n",
1874 	    unsp->uns_upper_opencnt, unsp->uns_lower_opencnt);
1875 	*/
1876 
1877 	if (unp->un_uppervp != NULLVP)
1878 		vn_printf(unp->un_uppervp, "unionfs: upper ");
1879 	if (unp->un_lowervp != NULLVP)
1880 		vn_printf(unp->un_lowervp, "unionfs: lower ");
1881 
1882 	return (0);
1883 }
1884 
1885 static int
1886 unionfs_get_llt_revlock(struct vnode *vp, int flags)
1887 {
1888 	int revlock;
1889 
1890 	revlock = 0;
1891 
1892 	switch (flags & LK_TYPE_MASK) {
1893 	case LK_SHARED:
1894 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
1895 			revlock = LK_UPGRADE;
1896 		else
1897 			revlock = LK_RELEASE;
1898 		break;
1899 	case LK_EXCLUSIVE:
1900 	case LK_UPGRADE:
1901 		revlock = LK_RELEASE;
1902 		break;
1903 	case LK_DOWNGRADE:
1904 		revlock = LK_UPGRADE;
1905 		break;
1906 	default:
1907 		break;
1908 	}
1909 
1910 	return (revlock);
1911 }
1912 
1913 /*
1914  * The state of an acquired lock is adjusted similarly to
1915  * the time of error generating.
1916  * flags: LK_RELEASE or LK_UPGRADE
1917  */
1918 static void
1919 unionfs_revlock(struct vnode *vp, int flags)
1920 {
1921 	if (flags & LK_RELEASE)
1922 		VOP_UNLOCK_FLAGS(vp, flags);
1923 	else {
1924 		/* UPGRADE */
1925 		if (vn_lock(vp, flags) != 0)
1926 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1927 	}
1928 }
1929 
1930 static int
1931 unionfs_lock(struct vop_lock1_args *ap)
1932 {
1933 	struct unionfs_node *unp;
1934 	struct vnode   *vp;
1935 	struct vnode   *uvp;
1936 	struct vnode   *lvp;
1937 	int		error;
1938 	int		flags;
1939 	int		revlock;
1940 	int		interlock;
1941 	int		uhold;
1942 
1943 	/*
1944 	 * TODO: rework the unionfs locking scheme.
1945 	 * It's not guaranteed to be safe to blindly lock two vnodes on
1946 	 * different mounts as is done here.  Further, the entanglement
1947 	 * of locking both vnodes with the various options that can be
1948 	 * passed to VOP_LOCK() makes this code hard to reason about.
1949 	 * Instead, consider locking only the upper vnode, or the lower
1950 	 * vnode is the upper is not present, and taking separate measures
1951 	 * to lock both vnodes in the few cases when that is needed.
1952 	 */
1953 	error = 0;
1954 	interlock = 1;
1955 	uhold = 0;
1956 	flags = ap->a_flags;
1957 	vp = ap->a_vp;
1958 
1959 	if (LK_RELEASE == (flags & LK_TYPE_MASK) || !(flags & LK_TYPE_MASK))
1960 		return (VOP_UNLOCK_FLAGS(vp, flags | LK_RELEASE));
1961 
1962 	if ((flags & LK_INTERLOCK) == 0)
1963 		VI_LOCK(vp);
1964 
1965 	unp = VTOUNIONFS(vp);
1966 	if (unp == NULL)
1967 		goto unionfs_lock_null_vnode;
1968 
1969 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1970 
1971 	lvp = unp->un_lowervp;
1972 	uvp = unp->un_uppervp;
1973 
1974 	if ((revlock = unionfs_get_llt_revlock(vp, flags)) == 0)
1975 		panic("unknown lock type: 0x%x", flags & LK_TYPE_MASK);
1976 
1977 	/*
1978 	 * During unmount, the root vnode lock may be taken recursively,
1979 	 * because it may share the same v_vnlock field as the vnode covered by
1980 	 * the unionfs mount.  The covered vnode is locked across VFS_UNMOUNT(),
1981 	 * and the same lock may be taken recursively here during vflush()
1982 	 * issued by unionfs_unmount().
1983 	 */
1984 	if ((flags & LK_TYPE_MASK) == LK_EXCLUSIVE &&
1985 	    (vp->v_vflag & VV_ROOT) != 0)
1986 		flags |= LK_CANRECURSE;
1987 
1988 	if (lvp != NULLVP) {
1989 		if (uvp != NULLVP && flags & LK_UPGRADE) {
1990 			/*
1991 			 * Share Lock is once released and a deadlock is
1992 			 * avoided.
1993 			 */
1994 			vholdnz(uvp);
1995 			uhold = 1;
1996 			VOP_UNLOCK(uvp);
1997 		}
1998 		VI_LOCK_FLAGS(lvp, MTX_DUPOK);
1999 		flags |= LK_INTERLOCK;
2000 		vholdl(lvp);
2001 
2002 		VI_UNLOCK(vp);
2003 		ap->a_flags &= ~LK_INTERLOCK;
2004 
2005 		error = VOP_LOCK(lvp, flags);
2006 
2007 		VI_LOCK(vp);
2008 		unp = VTOUNIONFS(vp);
2009 		if (unp == NULL) {
2010 			/* vnode is released. */
2011 			VI_UNLOCK(vp);
2012 			if (error == 0)
2013 				VOP_UNLOCK(lvp);
2014 			vdrop(lvp);
2015 			if (uhold != 0)
2016 				vdrop(uvp);
2017 			goto unionfs_lock_fallback;
2018 		}
2019 	}
2020 
2021 	if (error == 0 && uvp != NULLVP) {
2022 		if (uhold && flags & LK_UPGRADE) {
2023 			flags &= ~LK_TYPE_MASK;
2024 			flags |= LK_EXCLUSIVE;
2025 		}
2026 		VI_LOCK_FLAGS(uvp, MTX_DUPOK);
2027 		flags |= LK_INTERLOCK;
2028 		if (uhold == 0) {
2029 			vholdl(uvp);
2030 			uhold = 1;
2031 		}
2032 
2033 		VI_UNLOCK(vp);
2034 		ap->a_flags &= ~LK_INTERLOCK;
2035 
2036 		error = VOP_LOCK(uvp, flags);
2037 
2038 		VI_LOCK(vp);
2039 		unp = VTOUNIONFS(vp);
2040 		if (unp == NULL) {
2041 			/* vnode is released. */
2042 			VI_UNLOCK(vp);
2043 			if (error == 0)
2044 				VOP_UNLOCK(uvp);
2045 			vdrop(uvp);
2046 			if (lvp != NULLVP) {
2047 				VOP_UNLOCK(lvp);
2048 				vdrop(lvp);
2049 			}
2050 			goto unionfs_lock_fallback;
2051 		}
2052 		if (error != 0 && lvp != NULLVP) {
2053 			/* rollback */
2054 			VI_UNLOCK(vp);
2055 			unionfs_revlock(lvp, revlock);
2056 			interlock = 0;
2057 		}
2058 	}
2059 
2060 	if (interlock)
2061 		VI_UNLOCK(vp);
2062 	if (lvp != NULLVP)
2063 		vdrop(lvp);
2064 	if (uhold != 0)
2065 		vdrop(uvp);
2066 
2067 	return (error);
2068 
2069 unionfs_lock_null_vnode:
2070 	ap->a_flags |= LK_INTERLOCK;
2071 	return (vop_stdlock(ap));
2072 
2073 unionfs_lock_fallback:
2074 	/*
2075 	 * If we reach this point, we've discovered the unionfs vnode
2076 	 * has been reclaimed while the upper/lower vnode locks were
2077 	 * temporarily dropped.  Such temporary droppage may happen
2078 	 * during the course of an LK_UPGRADE operation itself, and in
2079 	 * that case LK_UPGRADE must be cleared as the unionfs vnode's
2080 	 * lock has been reset to point to the standard v_lock field,
2081 	 * which has not previously been held.
2082 	 */
2083 	if (flags & LK_UPGRADE) {
2084 		ap->a_flags &= ~LK_TYPE_MASK;
2085 		ap->a_flags |= LK_EXCLUSIVE;
2086 	}
2087 	return (vop_stdlock(ap));
2088 }
2089 
2090 static int
2091 unionfs_unlock(struct vop_unlock_args *ap)
2092 {
2093 	struct vnode   *vp;
2094 	struct vnode   *lvp;
2095 	struct vnode   *uvp;
2096 	struct unionfs_node *unp;
2097 	int		error;
2098 	int		uhold;
2099 
2100 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2101 
2102 	error = 0;
2103 	uhold = 0;
2104 	vp = ap->a_vp;
2105 
2106 	unp = VTOUNIONFS(vp);
2107 	if (unp == NULL)
2108 		goto unionfs_unlock_null_vnode;
2109 	lvp = unp->un_lowervp;
2110 	uvp = unp->un_uppervp;
2111 
2112 	if (lvp != NULLVP) {
2113 		vholdnz(lvp);
2114 		error = VOP_UNLOCK(lvp);
2115 	}
2116 
2117 	if (error == 0 && uvp != NULLVP) {
2118 		vholdnz(uvp);
2119 		uhold = 1;
2120 		error = VOP_UNLOCK(uvp);
2121 	}
2122 
2123 	if (lvp != NULLVP)
2124 		vdrop(lvp);
2125 	if (uhold != 0)
2126 		vdrop(uvp);
2127 
2128 	return error;
2129 
2130 unionfs_unlock_null_vnode:
2131 	return (vop_stdunlock(ap));
2132 }
2133 
2134 static int
2135 unionfs_pathconf(struct vop_pathconf_args *ap)
2136 {
2137 	struct unionfs_node *unp;
2138 	struct vnode   *vp;
2139 
2140 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2141 
2142 	unp = VTOUNIONFS(ap->a_vp);
2143 	vp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
2144 
2145 	return (VOP_PATHCONF(vp, ap->a_name, ap->a_retval));
2146 }
2147 
2148 static int
2149 unionfs_advlock(struct vop_advlock_args *ap)
2150 {
2151 	struct unionfs_node *unp;
2152 	struct unionfs_node_status *unsp;
2153 	struct vnode   *vp;
2154 	struct vnode   *uvp;
2155 	struct thread  *td;
2156 	int error;
2157 
2158 	UNIONFS_INTERNAL_DEBUG("unionfs_advlock: enter\n");
2159 
2160 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2161 
2162 	vp = ap->a_vp;
2163 	td = curthread;
2164 
2165 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2166 
2167 	unp = VTOUNIONFS(ap->a_vp);
2168 	uvp = unp->un_uppervp;
2169 
2170 	if (uvp == NULLVP) {
2171 		error = unionfs_copyfile(unp, 1, td->td_ucred, td);
2172 		if (error != 0)
2173 			goto unionfs_advlock_abort;
2174 		uvp = unp->un_uppervp;
2175 
2176 		unionfs_get_node_status(unp, td, &unsp);
2177 		if (unsp->uns_lower_opencnt > 0) {
2178 			/* try reopen the vnode */
2179 			error = VOP_OPEN(uvp, unsp->uns_lower_openmode,
2180 				td->td_ucred, td, NULL);
2181 			if (error)
2182 				goto unionfs_advlock_abort;
2183 			unsp->uns_upper_opencnt++;
2184 			VOP_CLOSE(unp->un_lowervp, unsp->uns_lower_openmode,
2185 			    td->td_ucred, td);
2186 			unsp->uns_lower_opencnt--;
2187 		} else
2188 			unionfs_tryrem_node_status(unp, unsp);
2189 	}
2190 
2191 	VOP_UNLOCK(vp);
2192 
2193 	error = VOP_ADVLOCK(uvp, ap->a_id, ap->a_op, ap->a_fl, ap->a_flags);
2194 
2195 	UNIONFS_INTERNAL_DEBUG("unionfs_advlock: leave (%d)\n", error);
2196 
2197 	return error;
2198 
2199 unionfs_advlock_abort:
2200 	VOP_UNLOCK(vp);
2201 
2202 	UNIONFS_INTERNAL_DEBUG("unionfs_advlock: leave (%d)\n", error);
2203 
2204 	return error;
2205 }
2206 
2207 static int
2208 unionfs_strategy(struct vop_strategy_args *ap)
2209 {
2210 	struct unionfs_node *unp;
2211 	struct vnode   *vp;
2212 
2213 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2214 
2215 	unp = VTOUNIONFS(ap->a_vp);
2216 	vp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
2217 
2218 #ifdef DIAGNOSTIC
2219 	if (vp == NULLVP)
2220 		panic("unionfs_strategy: nullvp");
2221 
2222 	if (ap->a_bp->b_iocmd == BIO_WRITE && vp == unp->un_lowervp)
2223 		panic("unionfs_strategy: writing to lowervp");
2224 #endif
2225 
2226 	return (VOP_STRATEGY(vp, ap->a_bp));
2227 }
2228 
2229 static int
2230 unionfs_getacl(struct vop_getacl_args *ap)
2231 {
2232 	struct unionfs_node *unp;
2233 	struct vnode   *vp;
2234 	int		error;
2235 
2236 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2237 
2238 	unp = VTOUNIONFS(ap->a_vp);
2239 	vp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
2240 
2241 	UNIONFS_INTERNAL_DEBUG("unionfs_getacl: enter\n");
2242 
2243 	error = VOP_GETACL(vp, ap->a_type, ap->a_aclp, ap->a_cred, ap->a_td);
2244 
2245 	UNIONFS_INTERNAL_DEBUG("unionfs_getacl: leave (%d)\n", error);
2246 
2247 	return (error);
2248 }
2249 
2250 static int
2251 unionfs_setacl(struct vop_setacl_args *ap)
2252 {
2253 	struct unionfs_node *unp;
2254 	struct vnode   *uvp;
2255 	struct vnode   *lvp;
2256 	struct thread  *td;
2257 	int		error;
2258 
2259 	UNIONFS_INTERNAL_DEBUG("unionfs_setacl: enter\n");
2260 
2261 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2262 
2263 	error = EROFS;
2264 	unp = VTOUNIONFS(ap->a_vp);
2265 	uvp = unp->un_uppervp;
2266 	lvp = unp->un_lowervp;
2267 	td = ap->a_td;
2268 
2269 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
2270 		return (EROFS);
2271 
2272 	if (uvp == NULLVP && lvp->v_type == VREG) {
2273 		if ((error = unionfs_copyfile(unp, 1, ap->a_cred, td)) != 0)
2274 			return (error);
2275 		uvp = unp->un_uppervp;
2276 	}
2277 
2278 	if (uvp != NULLVP)
2279 		error = VOP_SETACL(uvp, ap->a_type, ap->a_aclp, ap->a_cred, td);
2280 
2281 	UNIONFS_INTERNAL_DEBUG("unionfs_setacl: leave (%d)\n", error);
2282 
2283 	return (error);
2284 }
2285 
2286 static int
2287 unionfs_aclcheck(struct vop_aclcheck_args *ap)
2288 {
2289 	struct unionfs_node *unp;
2290 	struct vnode   *vp;
2291 	int		error;
2292 
2293 	UNIONFS_INTERNAL_DEBUG("unionfs_aclcheck: enter\n");
2294 
2295 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2296 
2297 	unp = VTOUNIONFS(ap->a_vp);
2298 	vp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
2299 
2300 	error = VOP_ACLCHECK(vp, ap->a_type, ap->a_aclp, ap->a_cred, ap->a_td);
2301 
2302 	UNIONFS_INTERNAL_DEBUG("unionfs_aclcheck: leave (%d)\n", error);
2303 
2304 	return (error);
2305 }
2306 
2307 static int
2308 unionfs_openextattr(struct vop_openextattr_args *ap)
2309 {
2310 	struct unionfs_node *unp;
2311 	struct vnode   *vp;
2312 	struct vnode   *tvp;
2313 	int		error;
2314 
2315 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2316 
2317 	vp = ap->a_vp;
2318 	unp = VTOUNIONFS(vp);
2319 	tvp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
2320 
2321 	if ((tvp == unp->un_uppervp && (unp->un_flag & UNIONFS_OPENEXTU)) ||
2322 	    (tvp == unp->un_lowervp && (unp->un_flag & UNIONFS_OPENEXTL)))
2323 		return (EBUSY);
2324 
2325 	error = VOP_OPENEXTATTR(tvp, ap->a_cred, ap->a_td);
2326 
2327 	if (error == 0) {
2328 		if (vn_lock(vp, LK_UPGRADE) != 0)
2329 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2330 		if (!VN_IS_DOOMED(vp)) {
2331 			if (tvp == unp->un_uppervp)
2332 				unp->un_flag |= UNIONFS_OPENEXTU;
2333 			else
2334 				unp->un_flag |= UNIONFS_OPENEXTL;
2335 		}
2336 		vn_lock(vp, LK_DOWNGRADE | LK_RETRY);
2337 	}
2338 
2339 	return (error);
2340 }
2341 
2342 static int
2343 unionfs_closeextattr(struct vop_closeextattr_args *ap)
2344 {
2345 	struct unionfs_node *unp;
2346 	struct vnode   *vp;
2347 	struct vnode   *tvp;
2348 	int		error;
2349 
2350 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2351 
2352 	vp = ap->a_vp;
2353 	unp = VTOUNIONFS(vp);
2354 	tvp = NULLVP;
2355 
2356 	if (unp->un_flag & UNIONFS_OPENEXTU)
2357 		tvp = unp->un_uppervp;
2358 	else if (unp->un_flag & UNIONFS_OPENEXTL)
2359 		tvp = unp->un_lowervp;
2360 
2361 	if (tvp == NULLVP)
2362 		return (EOPNOTSUPP);
2363 
2364 	error = VOP_CLOSEEXTATTR(tvp, ap->a_commit, ap->a_cred, ap->a_td);
2365 
2366 	if (error == 0) {
2367 		if (vn_lock(vp, LK_UPGRADE) != 0)
2368 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2369 		if (!VN_IS_DOOMED(vp)) {
2370 			if (tvp == unp->un_uppervp)
2371 				unp->un_flag &= ~UNIONFS_OPENEXTU;
2372 			else
2373 				unp->un_flag &= ~UNIONFS_OPENEXTL;
2374 		}
2375 		vn_lock(vp, LK_DOWNGRADE | LK_RETRY);
2376 	}
2377 
2378 	return (error);
2379 }
2380 
2381 static int
2382 unionfs_getextattr(struct vop_getextattr_args *ap)
2383 {
2384 	struct unionfs_node *unp;
2385 	struct vnode   *vp;
2386 
2387 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2388 
2389 	unp = VTOUNIONFS(ap->a_vp);
2390 	vp = NULLVP;
2391 
2392 	if (unp->un_flag & UNIONFS_OPENEXTU)
2393 		vp = unp->un_uppervp;
2394 	else if (unp->un_flag & UNIONFS_OPENEXTL)
2395 		vp = unp->un_lowervp;
2396 
2397 	if (vp == NULLVP)
2398 		return (EOPNOTSUPP);
2399 
2400 	return (VOP_GETEXTATTR(vp, ap->a_attrnamespace, ap->a_name,
2401 	    ap->a_uio, ap->a_size, ap->a_cred, ap->a_td));
2402 }
2403 
2404 static int
2405 unionfs_setextattr(struct vop_setextattr_args *ap)
2406 {
2407 	struct unionfs_node *unp;
2408 	struct vnode   *uvp;
2409 	struct vnode   *lvp;
2410 	struct vnode   *ovp;
2411 	struct ucred   *cred;
2412 	struct thread  *td;
2413 	int		error;
2414 
2415 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2416 
2417 	error = EROFS;
2418 	unp = VTOUNIONFS(ap->a_vp);
2419 	uvp = unp->un_uppervp;
2420 	lvp = unp->un_lowervp;
2421 	ovp = NULLVP;
2422 	cred = ap->a_cred;
2423 	td = ap->a_td;
2424 
2425 	UNIONFS_INTERNAL_DEBUG("unionfs_setextattr: enter (un_flag=%x)\n",
2426 	    unp->un_flag);
2427 
2428 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
2429 		return (EROFS);
2430 
2431 	if (unp->un_flag & UNIONFS_OPENEXTU)
2432 		ovp = unp->un_uppervp;
2433 	else if (unp->un_flag & UNIONFS_OPENEXTL)
2434 		ovp = unp->un_lowervp;
2435 
2436 	if (ovp == NULLVP)
2437 		return (EOPNOTSUPP);
2438 
2439 	if (ovp == lvp && lvp->v_type == VREG) {
2440 		VOP_CLOSEEXTATTR(lvp, 0, cred, td);
2441 		if (uvp == NULLVP &&
2442 		    (error = unionfs_copyfile(unp, 1, cred, td)) != 0) {
2443 unionfs_setextattr_reopen:
2444 			if ((unp->un_flag & UNIONFS_OPENEXTL) &&
2445 			    VOP_OPENEXTATTR(lvp, cred, td)) {
2446 #ifdef DIAGNOSTIC
2447 				panic("unionfs: VOP_OPENEXTATTR failed");
2448 #endif
2449 				unp->un_flag &= ~UNIONFS_OPENEXTL;
2450 			}
2451 			goto unionfs_setextattr_abort;
2452 		}
2453 		uvp = unp->un_uppervp;
2454 		if ((error = VOP_OPENEXTATTR(uvp, cred, td)) != 0)
2455 			goto unionfs_setextattr_reopen;
2456 		unp->un_flag &= ~UNIONFS_OPENEXTL;
2457 		unp->un_flag |= UNIONFS_OPENEXTU;
2458 		ovp = uvp;
2459 	}
2460 
2461 	if (ovp == uvp)
2462 		error = VOP_SETEXTATTR(ovp, ap->a_attrnamespace, ap->a_name,
2463 		    ap->a_uio, cred, td);
2464 
2465 unionfs_setextattr_abort:
2466 	UNIONFS_INTERNAL_DEBUG("unionfs_setextattr: leave (%d)\n", error);
2467 
2468 	return (error);
2469 }
2470 
2471 static int
2472 unionfs_listextattr(struct vop_listextattr_args *ap)
2473 {
2474 	struct unionfs_node *unp;
2475 	struct vnode *vp;
2476 
2477 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2478 
2479 	unp = VTOUNIONFS(ap->a_vp);
2480 	vp = NULLVP;
2481 
2482 	if (unp->un_flag & UNIONFS_OPENEXTU)
2483 		vp = unp->un_uppervp;
2484 	else if (unp->un_flag & UNIONFS_OPENEXTL)
2485 		vp = unp->un_lowervp;
2486 
2487 	if (vp == NULLVP)
2488 		return (EOPNOTSUPP);
2489 
2490 	return (VOP_LISTEXTATTR(vp, ap->a_attrnamespace, ap->a_uio,
2491 	    ap->a_size, ap->a_cred, ap->a_td));
2492 }
2493 
2494 static int
2495 unionfs_deleteextattr(struct vop_deleteextattr_args *ap)
2496 {
2497 	struct unionfs_node *unp;
2498 	struct vnode   *uvp;
2499 	struct vnode   *lvp;
2500 	struct vnode   *ovp;
2501 	struct ucred   *cred;
2502 	struct thread  *td;
2503 	int		error;
2504 
2505 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2506 
2507 	error = EROFS;
2508 	unp = VTOUNIONFS(ap->a_vp);
2509 	uvp = unp->un_uppervp;
2510 	lvp = unp->un_lowervp;
2511 	ovp = NULLVP;
2512 	cred = ap->a_cred;
2513 	td = ap->a_td;
2514 
2515 	UNIONFS_INTERNAL_DEBUG("unionfs_deleteextattr: enter (un_flag=%x)\n",
2516 	    unp->un_flag);
2517 
2518 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
2519 		return (EROFS);
2520 
2521 	if (unp->un_flag & UNIONFS_OPENEXTU)
2522 		ovp = unp->un_uppervp;
2523 	else if (unp->un_flag & UNIONFS_OPENEXTL)
2524 		ovp = unp->un_lowervp;
2525 
2526 	if (ovp == NULLVP)
2527 		return (EOPNOTSUPP);
2528 
2529 	if (ovp == lvp && lvp->v_type == VREG) {
2530 		VOP_CLOSEEXTATTR(lvp, 0, cred, td);
2531 		if (uvp == NULLVP &&
2532 		    (error = unionfs_copyfile(unp, 1, cred, td)) != 0) {
2533 unionfs_deleteextattr_reopen:
2534 			if ((unp->un_flag & UNIONFS_OPENEXTL) &&
2535 			    VOP_OPENEXTATTR(lvp, cred, td)) {
2536 #ifdef DIAGNOSTIC
2537 				panic("unionfs: VOP_OPENEXTATTR failed");
2538 #endif
2539 				unp->un_flag &= ~UNIONFS_OPENEXTL;
2540 			}
2541 			goto unionfs_deleteextattr_abort;
2542 		}
2543 		uvp = unp->un_uppervp;
2544 		if ((error = VOP_OPENEXTATTR(uvp, cred, td)) != 0)
2545 			goto unionfs_deleteextattr_reopen;
2546 		unp->un_flag &= ~UNIONFS_OPENEXTL;
2547 		unp->un_flag |= UNIONFS_OPENEXTU;
2548 		ovp = uvp;
2549 	}
2550 
2551 	if (ovp == uvp)
2552 		error = VOP_DELETEEXTATTR(ovp, ap->a_attrnamespace, ap->a_name,
2553 		    ap->a_cred, ap->a_td);
2554 
2555 unionfs_deleteextattr_abort:
2556 	UNIONFS_INTERNAL_DEBUG("unionfs_deleteextattr: leave (%d)\n", error);
2557 
2558 	return (error);
2559 }
2560 
2561 static int
2562 unionfs_setlabel(struct vop_setlabel_args *ap)
2563 {
2564 	struct unionfs_node *unp;
2565 	struct vnode   *uvp;
2566 	struct vnode   *lvp;
2567 	struct thread  *td;
2568 	int		error;
2569 
2570 	UNIONFS_INTERNAL_DEBUG("unionfs_setlabel: enter\n");
2571 
2572 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2573 
2574 	error = EROFS;
2575 	unp = VTOUNIONFS(ap->a_vp);
2576 	uvp = unp->un_uppervp;
2577 	lvp = unp->un_lowervp;
2578 	td = ap->a_td;
2579 
2580 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
2581 		return (EROFS);
2582 
2583 	if (uvp == NULLVP && lvp->v_type == VREG) {
2584 		if ((error = unionfs_copyfile(unp, 1, ap->a_cred, td)) != 0)
2585 			return (error);
2586 		uvp = unp->un_uppervp;
2587 	}
2588 
2589 	if (uvp != NULLVP)
2590 		error = VOP_SETLABEL(uvp, ap->a_label, ap->a_cred, td);
2591 
2592 	UNIONFS_INTERNAL_DEBUG("unionfs_setlabel: leave (%d)\n", error);
2593 
2594 	return (error);
2595 }
2596 
2597 static int
2598 unionfs_vptofh(struct vop_vptofh_args *ap)
2599 {
2600 	return (EOPNOTSUPP);
2601 }
2602 
2603 static int
2604 unionfs_add_writecount(struct vop_add_writecount_args *ap)
2605 {
2606 	struct vnode *tvp, *vp;
2607 	struct unionfs_node *unp;
2608 	int error, writerefs __diagused;
2609 
2610 	vp = ap->a_vp;
2611 	unp = VTOUNIONFS(vp);
2612 	tvp = unp->un_uppervp;
2613 	KASSERT(tvp != NULL,
2614 	    ("%s: adding write ref without upper vnode", __func__));
2615 	error = VOP_ADD_WRITECOUNT(tvp, ap->a_inc);
2616 	if (error != 0)
2617 		return (error);
2618 	/*
2619 	 * We need to track the write refs we've passed to the underlying
2620 	 * vnodes so that we can undo them in case we are forcibly unmounted.
2621 	 */
2622 	writerefs = atomic_fetchadd_int(&vp->v_writecount, ap->a_inc);
2623 	/* text refs are bypassed to lowervp */
2624 	VNASSERT(writerefs >= 0, vp,
2625 	    ("%s: invalid write count %d", __func__, writerefs));
2626 	VNASSERT(writerefs + ap->a_inc >= 0, vp,
2627 	    ("%s: invalid write count inc %d + %d", __func__,
2628 	    writerefs, ap->a_inc));
2629 	return (0);
2630 }
2631 
2632 static int
2633 unionfs_vput_pair(struct vop_vput_pair_args *ap)
2634 {
2635 	struct mount *mp;
2636 	struct vnode *dvp, *vp, **vpp, *lvp, *ldvp, *uvp, *udvp, *tempvp;
2637 	struct unionfs_node *dunp, *unp;
2638 	int error, res;
2639 
2640 	dvp = ap->a_dvp;
2641 	vpp = ap->a_vpp;
2642 	vp = NULLVP;
2643 	lvp = NULLVP;
2644 	uvp = NULLVP;
2645 	unp = NULL;
2646 
2647 	dunp = VTOUNIONFS(dvp);
2648 	udvp = dunp->un_uppervp;
2649 	ldvp = dunp->un_lowervp;
2650 
2651 	/*
2652 	 * Underlying vnodes should be locked because the encompassing unionfs
2653 	 * node is locked, but will not be referenced, as the reference will
2654 	 * only be on the unionfs node.  Reference them now so that the vput()s
2655 	 * performed by VOP_VPUT_PAIR() will have a reference to drop.
2656 	 */
2657 	if (udvp != NULLVP)
2658 		vref(udvp);
2659 	if (ldvp != NULLVP)
2660 		vref(ldvp);
2661 
2662 	if (vpp != NULL)
2663 		vp = *vpp;
2664 
2665 	if (vp != NULLVP) {
2666 		unp = VTOUNIONFS(vp);
2667 		uvp = unp->un_uppervp;
2668 		lvp = unp->un_lowervp;
2669 		if (uvp != NULLVP)
2670 			vref(uvp);
2671 		if (lvp != NULLVP)
2672 			vref(lvp);
2673 
2674 		/*
2675 		 * If we're being asked to return a locked child vnode, then
2676 		 * we may need to create a replacement vnode in case the
2677 		 * original is reclaimed while the lock is dropped.  In that
2678 		 * case we'll need to ensure the mount and the underlying
2679 		 * vnodes aren't also recycled during that window.
2680 		 */
2681 		if (!ap->a_unlock_vp) {
2682 			vhold(vp);
2683 			if (uvp != NULLVP)
2684 				vhold(uvp);
2685 			if (lvp != NULLVP)
2686 				vhold(lvp);
2687 			mp = vp->v_mount;
2688 			vfs_ref(mp);
2689 		}
2690 	}
2691 
2692 	/*
2693 	 * TODO: Because unionfs_lock() locks both the lower and upper vnodes
2694 	 * (if available), we must also call VOP_VPUT_PAIR() on both the lower
2695 	 * and upper parent/child pairs.  If unionfs_lock() is reworked to lock
2696 	 * only a single vnode, this code will need to change to also only
2697 	 * operate on one vnode pair.
2698 	 */
2699 	ASSERT_VOP_LOCKED(ldvp, __func__);
2700 	ASSERT_VOP_LOCKED(udvp, __func__);
2701 	ASSERT_VOP_LOCKED(lvp, __func__);
2702 	ASSERT_VOP_LOCKED(uvp, __func__);
2703 
2704 	KASSERT(lvp == NULLVP || ldvp != NULLVP,
2705 	    ("%s: NULL ldvp with non-NULL lvp", __func__));
2706 	if (ldvp != NULLVP)
2707 		res = VOP_VPUT_PAIR(ldvp, lvp != NULLVP ? &lvp : NULL, true);
2708 	KASSERT(uvp == NULLVP || udvp != NULLVP,
2709 	    ("%s: NULL udvp with non-NULL uvp", __func__));
2710 	if (udvp != NULLVP)
2711 		res = VOP_VPUT_PAIR(udvp, uvp != NULLVP ? &uvp : NULL, true);
2712 
2713 	ASSERT_VOP_UNLOCKED(ldvp, __func__);
2714 	ASSERT_VOP_UNLOCKED(udvp, __func__);
2715 	ASSERT_VOP_UNLOCKED(lvp, __func__);
2716 	ASSERT_VOP_UNLOCKED(uvp, __func__);
2717 
2718 	/*
2719 	 * VOP_VPUT_PAIR() dropped the references we added to the underlying
2720 	 * vnodes, now drop the caller's reference to the unionfs vnodes.
2721 	 */
2722 	if (vp != NULLVP && ap->a_unlock_vp)
2723 		vrele(vp);
2724 	vrele(dvp);
2725 
2726 	if (vp == NULLVP || ap->a_unlock_vp)
2727 		return (res);
2728 
2729 	/*
2730 	 * We're being asked to return a locked vnode.  At this point, the
2731 	 * underlying vnodes have been unlocked, so vp may have been reclaimed.
2732 	 */
2733 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2734 	if (vp->v_data == NULL && vfs_busy(mp, MBF_NOWAIT) == 0) {
2735 		vput(vp);
2736 		error = unionfs_nodeget(mp, uvp, lvp, dvp, &tempvp, NULL);
2737 		if (error == 0) {
2738 			vn_lock(tempvp, LK_EXCLUSIVE | LK_RETRY);
2739 			*vpp = tempvp;
2740 		} else
2741 			vget(vp, LK_EXCLUSIVE | LK_RETRY);
2742 		vfs_unbusy(mp);
2743 	}
2744 	if (lvp != NULLVP)
2745 		vdrop(lvp);
2746 	if (uvp != NULLVP)
2747 		vdrop(uvp);
2748 	vdrop(vp);
2749 	vfs_rel(mp);
2750 
2751 	return (res);
2752 }
2753 
2754 static int
2755 unionfs_set_text(struct vop_set_text_args *ap)
2756 {
2757 	struct vnode *tvp;
2758 	struct unionfs_node *unp;
2759 	int error;
2760 
2761 	/*
2762 	 * We assume text refs are managed against lvp/uvp through the
2763 	 * executable mapping backed by its VM object.  We therefore don't
2764 	 * need to track leased text refs in the case of a forcible unmount.
2765 	 */
2766 	unp = VTOUNIONFS(ap->a_vp);
2767 	ASSERT_VOP_LOCKED(ap->a_vp, __func__);
2768 	tvp = unp->un_uppervp != NULL ? unp->un_uppervp : unp->un_lowervp;
2769 	error = VOP_SET_TEXT(tvp);
2770 	return (error);
2771 }
2772 
2773 static int
2774 unionfs_unset_text(struct vop_unset_text_args *ap)
2775 {
2776 	struct vnode *tvp;
2777 	struct unionfs_node *unp;
2778 
2779 	ASSERT_VOP_LOCKED(ap->a_vp, __func__);
2780 	unp = VTOUNIONFS(ap->a_vp);
2781 	tvp = unp->un_uppervp != NULL ? unp->un_uppervp : unp->un_lowervp;
2782 	VOP_UNSET_TEXT_CHECKED(tvp);
2783 	return (0);
2784 }
2785 
2786 struct vop_vector unionfs_vnodeops = {
2787 	.vop_default =		&default_vnodeops,
2788 
2789 	.vop_access =		unionfs_access,
2790 	.vop_aclcheck =		unionfs_aclcheck,
2791 	.vop_advlock =		unionfs_advlock,
2792 	.vop_bmap =		VOP_EOPNOTSUPP,
2793 	.vop_cachedlookup =	unionfs_lookup,
2794 	.vop_close =		unionfs_close,
2795 	.vop_closeextattr =	unionfs_closeextattr,
2796 	.vop_create =		unionfs_create,
2797 	.vop_deleteextattr =	unionfs_deleteextattr,
2798 	.vop_fsync =		unionfs_fsync,
2799 	.vop_getacl =		unionfs_getacl,
2800 	.vop_getattr =		unionfs_getattr,
2801 	.vop_getextattr =	unionfs_getextattr,
2802 	.vop_getwritemount =	unionfs_getwritemount,
2803 	.vop_inactive =		unionfs_inactive,
2804 	.vop_need_inactive =	vop_stdneed_inactive,
2805 	.vop_islocked =		vop_stdislocked,
2806 	.vop_ioctl =		unionfs_ioctl,
2807 	.vop_link =		unionfs_link,
2808 	.vop_listextattr =	unionfs_listextattr,
2809 	.vop_lock1 =		unionfs_lock,
2810 	.vop_lookup =		vfs_cache_lookup,
2811 	.vop_mkdir =		unionfs_mkdir,
2812 	.vop_mknod =		unionfs_mknod,
2813 	.vop_open =		unionfs_open,
2814 	.vop_openextattr =	unionfs_openextattr,
2815 	.vop_pathconf =		unionfs_pathconf,
2816 	.vop_poll =		unionfs_poll,
2817 	.vop_print =		unionfs_print,
2818 	.vop_read =		unionfs_read,
2819 	.vop_readdir =		unionfs_readdir,
2820 	.vop_readlink =		unionfs_readlink,
2821 	.vop_reclaim =		unionfs_reclaim,
2822 	.vop_remove =		unionfs_remove,
2823 	.vop_rename =		unionfs_rename,
2824 	.vop_rmdir =		unionfs_rmdir,
2825 	.vop_setacl =		unionfs_setacl,
2826 	.vop_setattr =		unionfs_setattr,
2827 	.vop_setextattr =	unionfs_setextattr,
2828 	.vop_setlabel =		unionfs_setlabel,
2829 	.vop_strategy =		unionfs_strategy,
2830 	.vop_symlink =		unionfs_symlink,
2831 	.vop_unlock =		unionfs_unlock,
2832 	.vop_whiteout =		unionfs_whiteout,
2833 	.vop_write =		unionfs_write,
2834 	.vop_vptofh =		unionfs_vptofh,
2835 	.vop_add_writecount =	unionfs_add_writecount,
2836 	.vop_vput_pair =	unionfs_vput_pair,
2837 	.vop_set_text =		unionfs_set_text,
2838 	.vop_unset_text = 	unionfs_unset_text,
2839 };
2840 VFS_VOP_VECTOR_REGISTER(unionfs_vnodeops);
2841