xref: /freebsd/sys/fs/unionfs/union_vnops.c (revision 5956d97f4b3204318ceb6aa9c77bd0bc6ea87a41)
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 	/* Ensure subsequent vnops will get a valid pathname buffer. */
352 	if (nameiop != LOOKUP && (error == 0 || error == EJUSTRETURN))
353 		cnp->cn_flags |= SAVENAME;
354 
355 	UNIONFS_INTERNAL_DEBUG("unionfs_lookup: leave (%d)\n", error);
356 
357 	return (error);
358 }
359 
360 static int
361 unionfs_create(struct vop_create_args *ap)
362 {
363 	struct unionfs_node *dunp;
364 	struct componentname *cnp;
365 	struct vnode   *udvp;
366 	struct vnode   *vp;
367 	int		error;
368 
369 	UNIONFS_INTERNAL_DEBUG("unionfs_create: enter\n");
370 
371 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
372 
373 	dunp = VTOUNIONFS(ap->a_dvp);
374 	cnp = ap->a_cnp;
375 	udvp = dunp->un_uppervp;
376 	error = EROFS;
377 
378 	if (udvp != NULLVP) {
379 		error = VOP_CREATE(udvp, &vp, cnp, ap->a_vap);
380 		if (error != 0)
381 			goto unionfs_create_abort;
382 
383 		if (vp->v_type == VSOCK)
384 			*(ap->a_vpp) = vp;
385 		else {
386 			VOP_UNLOCK(vp);
387 			error = unionfs_nodeget(ap->a_dvp->v_mount, vp, NULLVP,
388 			    ap->a_dvp, ap->a_vpp, cnp);
389 			vrele(vp);
390 		}
391 	}
392 
393 unionfs_create_abort:
394 	UNIONFS_INTERNAL_DEBUG("unionfs_create: leave (%d)\n", error);
395 
396 	return (error);
397 }
398 
399 static int
400 unionfs_whiteout(struct vop_whiteout_args *ap)
401 {
402 	struct unionfs_node *dunp;
403 	struct componentname *cnp;
404 	struct vnode   *udvp;
405 	int		error;
406 
407 	UNIONFS_INTERNAL_DEBUG("unionfs_whiteout: enter\n");
408 
409 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
410 
411 	dunp = VTOUNIONFS(ap->a_dvp);
412 	cnp = ap->a_cnp;
413 	udvp = dunp->un_uppervp;
414 	error = EOPNOTSUPP;
415 
416 	if (udvp != NULLVP) {
417 		switch (ap->a_flags) {
418 		case CREATE:
419 		case DELETE:
420 		case LOOKUP:
421 			error = VOP_WHITEOUT(udvp, cnp, ap->a_flags);
422 			break;
423 		default:
424 			error = EINVAL;
425 			break;
426 		}
427 	}
428 
429 	UNIONFS_INTERNAL_DEBUG("unionfs_whiteout: leave (%d)\n", error);
430 
431 	return (error);
432 }
433 
434 static int
435 unionfs_mknod(struct vop_mknod_args *ap)
436 {
437 	struct unionfs_node *dunp;
438 	struct componentname *cnp;
439 	struct vnode   *udvp;
440 	struct vnode   *vp;
441 	int		error;
442 
443 	UNIONFS_INTERNAL_DEBUG("unionfs_mknod: enter\n");
444 
445 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
446 
447 	dunp = VTOUNIONFS(ap->a_dvp);
448 	cnp = ap->a_cnp;
449 	udvp = dunp->un_uppervp;
450 	error = EROFS;
451 
452 	if (udvp != NULLVP) {
453 		error = VOP_MKNOD(udvp, &vp, cnp, ap->a_vap);
454 		if (error != 0)
455 			goto unionfs_mknod_abort;
456 
457 		if (vp->v_type == VSOCK)
458 			*(ap->a_vpp) = vp;
459 		else {
460 			VOP_UNLOCK(vp);
461 			error = unionfs_nodeget(ap->a_dvp->v_mount, vp, NULLVP,
462 			    ap->a_dvp, ap->a_vpp, cnp);
463 			vrele(vp);
464 		}
465 	}
466 
467 unionfs_mknod_abort:
468 	UNIONFS_INTERNAL_DEBUG("unionfs_mknod: leave (%d)\n", error);
469 
470 	return (error);
471 }
472 
473 enum unionfs_lkupgrade {
474 	UNIONFS_LKUPGRADE_SUCCESS, /* lock successfully upgraded */
475 	UNIONFS_LKUPGRADE_ALREADY, /* lock already held exclusive */
476 	UNIONFS_LKUPGRADE_DOOMED   /* lock was upgraded, but vnode reclaimed */
477 };
478 
479 static inline enum unionfs_lkupgrade
480 unionfs_upgrade_lock(struct vnode *vp)
481 {
482 	ASSERT_VOP_LOCKED(vp, __func__);
483 
484 	if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
485 		return (UNIONFS_LKUPGRADE_ALREADY);
486 
487 	if (vn_lock(vp, LK_UPGRADE) != 0) {
488 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
489 		if (VN_IS_DOOMED(vp))
490 			return (UNIONFS_LKUPGRADE_DOOMED);
491 	}
492 	return (UNIONFS_LKUPGRADE_SUCCESS);
493 }
494 
495 static inline void
496 unionfs_downgrade_lock(struct vnode *vp, enum unionfs_lkupgrade status)
497 {
498 	if (status != UNIONFS_LKUPGRADE_ALREADY)
499 		vn_lock(vp, LK_DOWNGRADE | LK_RETRY);
500 }
501 
502 static int
503 unionfs_open(struct vop_open_args *ap)
504 {
505 	struct unionfs_node *unp;
506 	struct unionfs_node_status *unsp;
507 	struct vnode   *vp;
508 	struct vnode   *uvp;
509 	struct vnode   *lvp;
510 	struct vnode   *targetvp;
511 	struct ucred   *cred;
512 	struct thread  *td;
513 	int		error;
514 	enum unionfs_lkupgrade lkstatus;
515 
516 	UNIONFS_INTERNAL_DEBUG("unionfs_open: enter\n");
517 
518 	KASSERT_UNIONFS_VNODE(ap->a_vp);
519 
520 	error = 0;
521 	vp = ap->a_vp;
522 	targetvp = NULLVP;
523 	cred = ap->a_cred;
524 	td = ap->a_td;
525 
526 	/*
527 	 * The executable loader path may call this function with vp locked
528 	 * shared.  If the vnode is reclaimed while upgrading, we can't safely
529 	 * use unp or do anything else unionfs- specific.
530 	 */
531 	lkstatus = unionfs_upgrade_lock(vp);
532 	if (lkstatus == UNIONFS_LKUPGRADE_DOOMED) {
533 		error = ENOENT;
534 		goto unionfs_open_cleanup;
535 	}
536 
537 	unp = VTOUNIONFS(vp);
538 	uvp = unp->un_uppervp;
539 	lvp = unp->un_lowervp;
540 	unionfs_get_node_status(unp, td, &unsp);
541 
542 	if (unsp->uns_lower_opencnt > 0 || unsp->uns_upper_opencnt > 0) {
543 		/* vnode is already opend. */
544 		if (unsp->uns_upper_opencnt > 0)
545 			targetvp = uvp;
546 		else
547 			targetvp = lvp;
548 
549 		if (targetvp == lvp &&
550 		    (ap->a_mode & FWRITE) && lvp->v_type == VREG)
551 			targetvp = NULLVP;
552 	}
553 	if (targetvp == NULLVP) {
554 		if (uvp == NULLVP) {
555 			if ((ap->a_mode & FWRITE) && lvp->v_type == VREG) {
556 				error = unionfs_copyfile(unp,
557 				    !(ap->a_mode & O_TRUNC), cred, td);
558 				if (error != 0)
559 					goto unionfs_open_abort;
560 				targetvp = uvp = unp->un_uppervp;
561 			} else
562 				targetvp = lvp;
563 		} else
564 			targetvp = uvp;
565 	}
566 
567 	error = VOP_OPEN(targetvp, ap->a_mode, cred, td, ap->a_fp);
568 	if (error == 0) {
569 		if (targetvp == uvp) {
570 			if (uvp->v_type == VDIR && lvp != NULLVP &&
571 			    unsp->uns_lower_opencnt <= 0) {
572 				/* open lower for readdir */
573 				error = VOP_OPEN(lvp, FREAD, cred, td, NULL);
574 				if (error != 0) {
575 					VOP_CLOSE(uvp, ap->a_mode, cred, td);
576 					goto unionfs_open_abort;
577 				}
578 				unsp->uns_node_flag |= UNS_OPENL_4_READDIR;
579 				unsp->uns_lower_opencnt++;
580 			}
581 			unsp->uns_upper_opencnt++;
582 		} else {
583 			unsp->uns_lower_opencnt++;
584 			unsp->uns_lower_openmode = ap->a_mode;
585 		}
586 		vp->v_object = targetvp->v_object;
587 	}
588 
589 unionfs_open_abort:
590 	if (error != 0)
591 		unionfs_tryrem_node_status(unp, unsp);
592 
593 unionfs_open_cleanup:
594 	unionfs_downgrade_lock(vp, lkstatus);
595 
596 	UNIONFS_INTERNAL_DEBUG("unionfs_open: leave (%d)\n", error);
597 
598 	return (error);
599 }
600 
601 static int
602 unionfs_close(struct vop_close_args *ap)
603 {
604 	struct unionfs_node *unp;
605 	struct unionfs_node_status *unsp;
606 	struct ucred   *cred;
607 	struct thread  *td;
608 	struct vnode   *vp;
609 	struct vnode   *ovp;
610 	int		error;
611 	enum unionfs_lkupgrade lkstatus;;
612 
613 	UNIONFS_INTERNAL_DEBUG("unionfs_close: enter\n");
614 
615 	KASSERT_UNIONFS_VNODE(ap->a_vp);
616 
617 	vp = ap->a_vp;
618 	cred = ap->a_cred;
619 	td = ap->a_td;
620 	error = 0;
621 
622 	/*
623 	 * If the vnode is reclaimed while upgrading, we can't safely use unp
624 	 * or do anything else unionfs- specific.
625 	 */
626 	lkstatus = unionfs_upgrade_lock(vp);
627 	if (lkstatus == UNIONFS_LKUPGRADE_DOOMED)
628 		goto unionfs_close_cleanup;
629 
630 	unp = VTOUNIONFS(vp);
631 	unionfs_get_node_status(unp, td, &unsp);
632 
633 	if (unsp->uns_lower_opencnt <= 0 && unsp->uns_upper_opencnt <= 0) {
634 #ifdef DIAGNOSTIC
635 		printf("unionfs_close: warning: open count is 0\n");
636 #endif
637 		if (unp->un_uppervp != NULLVP)
638 			ovp = unp->un_uppervp;
639 		else
640 			ovp = unp->un_lowervp;
641 	} else if (unsp->uns_upper_opencnt > 0)
642 		ovp = unp->un_uppervp;
643 	else
644 		ovp = unp->un_lowervp;
645 
646 	error = VOP_CLOSE(ovp, ap->a_fflag, cred, td);
647 
648 	if (error != 0)
649 		goto unionfs_close_abort;
650 
651 	vp->v_object = ovp->v_object;
652 
653 	if (ovp == unp->un_uppervp) {
654 		unsp->uns_upper_opencnt--;
655 		if (unsp->uns_upper_opencnt == 0) {
656 			if (unsp->uns_node_flag & UNS_OPENL_4_READDIR) {
657 				VOP_CLOSE(unp->un_lowervp, FREAD, cred, td);
658 				unsp->uns_node_flag &= ~UNS_OPENL_4_READDIR;
659 				unsp->uns_lower_opencnt--;
660 			}
661 			if (unsp->uns_lower_opencnt > 0)
662 				vp->v_object = unp->un_lowervp->v_object;
663 		}
664 	} else
665 		unsp->uns_lower_opencnt--;
666 
667 unionfs_close_abort:
668 	unionfs_tryrem_node_status(unp, unsp);
669 
670 unionfs_close_cleanup:
671 	unionfs_downgrade_lock(vp, lkstatus);
672 
673 	UNIONFS_INTERNAL_DEBUG("unionfs_close: leave (%d)\n", error);
674 
675 	return (error);
676 }
677 
678 /*
679  * Check the access mode toward shadow file/dir.
680  */
681 static int
682 unionfs_check_corrected_access(accmode_t accmode, struct vattr *va,
683     struct ucred *cred)
684 {
685 	uid_t		uid;	/* upper side vnode's uid */
686 	gid_t		gid;	/* upper side vnode's gid */
687 	u_short		vmode;	/* upper side vnode's mode */
688 	u_short		mask;
689 
690 	mask = 0;
691 	uid = va->va_uid;
692 	gid = va->va_gid;
693 	vmode = va->va_mode;
694 
695 	/* check owner */
696 	if (cred->cr_uid == uid) {
697 		if (accmode & VEXEC)
698 			mask |= S_IXUSR;
699 		if (accmode & VREAD)
700 			mask |= S_IRUSR;
701 		if (accmode & VWRITE)
702 			mask |= S_IWUSR;
703 		return ((vmode & mask) == mask ? 0 : EACCES);
704 	}
705 
706 	/* check group */
707 	if (groupmember(gid, cred)) {
708 		if (accmode & VEXEC)
709 			mask |= S_IXGRP;
710 		if (accmode & VREAD)
711 			mask |= S_IRGRP;
712 		if (accmode & VWRITE)
713 			mask |= S_IWGRP;
714 		return ((vmode & mask) == mask ? 0 : EACCES);
715 	}
716 
717 	/* check other */
718 	if (accmode & VEXEC)
719 		mask |= S_IXOTH;
720 	if (accmode & VREAD)
721 		mask |= S_IROTH;
722 	if (accmode & VWRITE)
723 		mask |= S_IWOTH;
724 
725 	return ((vmode & mask) == mask ? 0 : EACCES);
726 }
727 
728 static int
729 unionfs_access(struct vop_access_args *ap)
730 {
731 	struct unionfs_mount *ump;
732 	struct unionfs_node *unp;
733 	struct vnode   *uvp;
734 	struct vnode   *lvp;
735 	struct thread  *td;
736 	struct vattr	va;
737 	accmode_t	accmode;
738 	int		error;
739 
740 	UNIONFS_INTERNAL_DEBUG("unionfs_access: enter\n");
741 
742 	KASSERT_UNIONFS_VNODE(ap->a_vp);
743 
744 	ump = MOUNTTOUNIONFSMOUNT(ap->a_vp->v_mount);
745 	unp = VTOUNIONFS(ap->a_vp);
746 	uvp = unp->un_uppervp;
747 	lvp = unp->un_lowervp;
748 	td = ap->a_td;
749 	accmode = ap->a_accmode;
750 	error = EACCES;
751 
752 	if ((accmode & VWRITE) &&
753 	    (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)) {
754 		switch (ap->a_vp->v_type) {
755 		case VREG:
756 		case VDIR:
757 		case VLNK:
758 			return (EROFS);
759 		default:
760 			break;
761 		}
762 	}
763 
764 	if (uvp != NULLVP) {
765 		error = VOP_ACCESS(uvp, accmode, ap->a_cred, td);
766 
767 		UNIONFS_INTERNAL_DEBUG("unionfs_access: leave (%d)\n", error);
768 
769 		return (error);
770 	}
771 
772 	if (lvp != NULLVP) {
773 		if (accmode & VWRITE) {
774 			if (ump->um_uppervp->v_mount->mnt_flag & MNT_RDONLY) {
775 				switch (ap->a_vp->v_type) {
776 				case VREG:
777 				case VDIR:
778 				case VLNK:
779 					return (EROFS);
780 				default:
781 					break;
782 				}
783 			} else if (ap->a_vp->v_type == VREG ||
784 			    ap->a_vp->v_type == VDIR) {
785 				/* check shadow file/dir */
786 				if (ump->um_copymode != UNIONFS_TRANSPARENT) {
787 					error = unionfs_create_uppervattr(ump,
788 					    lvp, &va, ap->a_cred, td);
789 					if (error != 0)
790 						return (error);
791 
792 					error = unionfs_check_corrected_access(
793 					    accmode, &va, ap->a_cred);
794 					if (error != 0)
795 						return (error);
796 				}
797 			}
798 			accmode &= ~(VWRITE | VAPPEND);
799 			accmode |= VREAD; /* will copy to upper */
800 		}
801 		error = VOP_ACCESS(lvp, accmode, ap->a_cred, td);
802 	}
803 
804 	UNIONFS_INTERNAL_DEBUG("unionfs_access: leave (%d)\n", error);
805 
806 	return (error);
807 }
808 
809 static int
810 unionfs_getattr(struct vop_getattr_args *ap)
811 {
812 	struct unionfs_node *unp;
813 	struct unionfs_mount *ump;
814 	struct vnode   *uvp;
815 	struct vnode   *lvp;
816 	struct thread  *td;
817 	struct vattr	va;
818 	int		error;
819 
820 	UNIONFS_INTERNAL_DEBUG("unionfs_getattr: enter\n");
821 
822 	KASSERT_UNIONFS_VNODE(ap->a_vp);
823 
824 	unp = VTOUNIONFS(ap->a_vp);
825 	ump = MOUNTTOUNIONFSMOUNT(ap->a_vp->v_mount);
826 	uvp = unp->un_uppervp;
827 	lvp = unp->un_lowervp;
828 	td = curthread;
829 
830 	if (uvp != NULLVP) {
831 		if ((error = VOP_GETATTR(uvp, ap->a_vap, ap->a_cred)) == 0)
832 			ap->a_vap->va_fsid =
833 			    ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
834 
835 		UNIONFS_INTERNAL_DEBUG(
836 		    "unionfs_getattr: leave mode=%o, uid=%d, gid=%d (%d)\n",
837 		    ap->a_vap->va_mode, ap->a_vap->va_uid,
838 		    ap->a_vap->va_gid, error);
839 
840 		return (error);
841 	}
842 
843 	error = VOP_GETATTR(lvp, ap->a_vap, ap->a_cred);
844 
845 	if (error == 0 && !(ump->um_uppervp->v_mount->mnt_flag & MNT_RDONLY)) {
846 		/* correct the attr toward shadow file/dir. */
847 		if (ap->a_vp->v_type == VREG || ap->a_vp->v_type == VDIR) {
848 			unionfs_create_uppervattr_core(ump, ap->a_vap, &va, td);
849 			ap->a_vap->va_mode = va.va_mode;
850 			ap->a_vap->va_uid = va.va_uid;
851 			ap->a_vap->va_gid = va.va_gid;
852 		}
853 	}
854 
855 	if (error == 0)
856 		ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
857 
858 	UNIONFS_INTERNAL_DEBUG(
859 	    "unionfs_getattr: leave mode=%o, uid=%d, gid=%d (%d)\n",
860 	    ap->a_vap->va_mode, ap->a_vap->va_uid, ap->a_vap->va_gid, error);
861 
862 	return (error);
863 }
864 
865 static int
866 unionfs_setattr(struct vop_setattr_args *ap)
867 {
868 	struct unionfs_node *unp;
869 	struct vnode   *uvp;
870 	struct vnode   *lvp;
871 	struct thread  *td;
872 	struct vattr   *vap;
873 	int		error;
874 
875 	UNIONFS_INTERNAL_DEBUG("unionfs_setattr: enter\n");
876 
877 	KASSERT_UNIONFS_VNODE(ap->a_vp);
878 
879 	error = EROFS;
880 	unp = VTOUNIONFS(ap->a_vp);
881 	uvp = unp->un_uppervp;
882 	lvp = unp->un_lowervp;
883 	td = curthread;
884 	vap = ap->a_vap;
885 
886 	if ((ap->a_vp->v_mount->mnt_flag & MNT_RDONLY) &&
887 	    (vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
888 	     vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
889 	     vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL))
890 		return (EROFS);
891 
892 	if (uvp == NULLVP && lvp->v_type == VREG) {
893 		error = unionfs_copyfile(unp, (vap->va_size != 0),
894 		    ap->a_cred, td);
895 		if (error != 0)
896 			return (error);
897 		uvp = unp->un_uppervp;
898 	}
899 
900 	if (uvp != NULLVP)
901 		error = VOP_SETATTR(uvp, vap, ap->a_cred);
902 
903 	UNIONFS_INTERNAL_DEBUG("unionfs_setattr: leave (%d)\n", error);
904 
905 	return (error);
906 }
907 
908 static int
909 unionfs_read(struct vop_read_args *ap)
910 {
911 	struct unionfs_node *unp;
912 	struct vnode   *tvp;
913 	int		error;
914 
915 	/* UNIONFS_INTERNAL_DEBUG("unionfs_read: enter\n"); */
916 
917 	KASSERT_UNIONFS_VNODE(ap->a_vp);
918 
919 	unp = VTOUNIONFS(ap->a_vp);
920 	tvp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
921 
922 	error = VOP_READ(tvp, ap->a_uio, ap->a_ioflag, ap->a_cred);
923 
924 	/* UNIONFS_INTERNAL_DEBUG("unionfs_read: leave (%d)\n", error); */
925 
926 	return (error);
927 }
928 
929 static int
930 unionfs_write(struct vop_write_args *ap)
931 {
932 	struct unionfs_node *unp;
933 	struct vnode   *tvp;
934 	int		error;
935 
936 	/* UNIONFS_INTERNAL_DEBUG("unionfs_write: enter\n"); */
937 
938 	KASSERT_UNIONFS_VNODE(ap->a_vp);
939 
940 	unp = VTOUNIONFS(ap->a_vp);
941 	tvp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
942 
943 	error = VOP_WRITE(tvp, ap->a_uio, ap->a_ioflag, ap->a_cred);
944 
945 	/* UNIONFS_INTERNAL_DEBUG("unionfs_write: leave (%d)\n", error); */
946 
947 	return (error);
948 }
949 
950 static int
951 unionfs_ioctl(struct vop_ioctl_args *ap)
952 {
953 	struct unionfs_node *unp;
954 	struct unionfs_node_status *unsp;
955 	struct vnode   *ovp;
956 	int error;
957 
958 	UNIONFS_INTERNAL_DEBUG("unionfs_ioctl: enter\n");
959 
960 	KASSERT_UNIONFS_VNODE(ap->a_vp);
961 
962  	vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
963 	unp = VTOUNIONFS(ap->a_vp);
964 	unionfs_get_node_status(unp, ap->a_td, &unsp);
965 	ovp = (unsp->uns_upper_opencnt ? unp->un_uppervp : unp->un_lowervp);
966 	unionfs_tryrem_node_status(unp, unsp);
967 	VOP_UNLOCK(ap->a_vp);
968 
969 	if (ovp == NULLVP)
970 		return (EBADF);
971 
972 	error = VOP_IOCTL(ovp, ap->a_command, ap->a_data, ap->a_fflag,
973 	    ap->a_cred, ap->a_td);
974 
975 	UNIONFS_INTERNAL_DEBUG("unionfs_ioctl: leave (%d)\n", error);
976 
977 	return (error);
978 }
979 
980 static int
981 unionfs_poll(struct vop_poll_args *ap)
982 {
983 	struct unionfs_node *unp;
984 	struct unionfs_node_status *unsp;
985 	struct vnode *ovp;
986 
987 	KASSERT_UNIONFS_VNODE(ap->a_vp);
988 
989  	vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
990 	unp = VTOUNIONFS(ap->a_vp);
991 	unionfs_get_node_status(unp, ap->a_td, &unsp);
992 	ovp = (unsp->uns_upper_opencnt ? unp->un_uppervp : unp->un_lowervp);
993 	unionfs_tryrem_node_status(unp, unsp);
994 	VOP_UNLOCK(ap->a_vp);
995 
996 	if (ovp == NULLVP)
997 		return (EBADF);
998 
999 	return (VOP_POLL(ovp, ap->a_events, ap->a_cred, ap->a_td));
1000 }
1001 
1002 static int
1003 unionfs_fsync(struct vop_fsync_args *ap)
1004 {
1005 	struct unionfs_node *unp;
1006 	struct unionfs_node_status *unsp;
1007 	struct vnode *ovp;
1008 
1009 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1010 
1011 	unp = VTOUNIONFS(ap->a_vp);
1012 	unionfs_get_node_status(unp, ap->a_td, &unsp);
1013 	ovp = (unsp->uns_upper_opencnt ? unp->un_uppervp : unp->un_lowervp);
1014 	unionfs_tryrem_node_status(unp, unsp);
1015 
1016 	if (ovp == NULLVP)
1017 		return (EBADF);
1018 
1019 	return (VOP_FSYNC(ovp, ap->a_waitfor, ap->a_td));
1020 }
1021 
1022 static int
1023 unionfs_remove(struct vop_remove_args *ap)
1024 {
1025 	char	       *path;
1026 	struct unionfs_node *dunp;
1027 	struct unionfs_node *unp;
1028 	struct unionfs_mount *ump;
1029 	struct vnode   *udvp;
1030 	struct vnode   *uvp;
1031 	struct vnode   *lvp;
1032 	struct vnode   *vp;
1033 	struct componentname *cnp;
1034 	struct componentname cn;
1035 	struct thread  *td;
1036 	int		error;
1037 	int		pathlen;
1038 
1039 	UNIONFS_INTERNAL_DEBUG("unionfs_remove: enter\n");
1040 
1041 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
1042 
1043 	error = 0;
1044 	dunp = VTOUNIONFS(ap->a_dvp);
1045 	udvp = dunp->un_uppervp;
1046 	cnp = ap->a_cnp;
1047 	td = curthread;
1048 
1049 	if (ap->a_vp->v_op != &unionfs_vnodeops) {
1050 		if (ap->a_vp->v_type != VSOCK)
1051 			return (EINVAL);
1052 		ump = NULL;
1053 		vp = uvp = lvp = NULLVP;
1054 		/* search vnode */
1055 		VOP_UNLOCK(ap->a_vp);
1056 		error = unionfs_relookup(udvp, &vp, cnp, &cn, td,
1057 		    cnp->cn_nameptr, cnp->cn_namelen, DELETE);
1058 		if (error != 0 && error != ENOENT) {
1059 			vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
1060 			return (error);
1061 		}
1062 
1063 		if (error == 0 && vp == ap->a_vp) {
1064 			/* target vnode in upper */
1065 			uvp = vp;
1066 			vrele(vp);
1067 		} else {
1068 			/* target vnode in lower */
1069 			if (vp != NULLVP) {
1070 				if (udvp == vp)
1071 					vrele(vp);
1072 				else
1073 					vput(vp);
1074 			}
1075 			vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
1076 			lvp = ap->a_vp;
1077 		}
1078 		path = cnp->cn_nameptr;
1079 		pathlen = cnp->cn_namelen;
1080 	} else {
1081 		ump = MOUNTTOUNIONFSMOUNT(ap->a_vp->v_mount);
1082 		unp = VTOUNIONFS(ap->a_vp);
1083 		uvp = unp->un_uppervp;
1084 		lvp = unp->un_lowervp;
1085 		path = unp->un_path;
1086 		pathlen = unp->un_pathlen;
1087 	}
1088 
1089 	if (udvp == NULLVP)
1090 		return (EROFS);
1091 
1092 	if (uvp != NULLVP) {
1093 		/*
1094 		 * XXX: if the vnode type is VSOCK, it will create whiteout
1095 		 *      after remove.
1096 		 */
1097 		if (ump == NULL || ump->um_whitemode == UNIONFS_WHITE_ALWAYS ||
1098 		    lvp != NULLVP)
1099 			cnp->cn_flags |= DOWHITEOUT;
1100 		error = VOP_REMOVE(udvp, uvp, cnp);
1101 	} else if (lvp != NULLVP)
1102 		error = unionfs_mkwhiteout(udvp, cnp, td, path, pathlen);
1103 
1104 	UNIONFS_INTERNAL_DEBUG("unionfs_remove: leave (%d)\n", error);
1105 
1106 	return (error);
1107 }
1108 
1109 static int
1110 unionfs_link(struct vop_link_args *ap)
1111 {
1112 	struct unionfs_node *dunp;
1113 	struct unionfs_node *unp;
1114 	struct vnode   *udvp;
1115 	struct vnode   *uvp;
1116 	struct componentname *cnp;
1117 	struct thread  *td;
1118 	int		error;
1119 	int		needrelookup;
1120 
1121 	UNIONFS_INTERNAL_DEBUG("unionfs_link: enter\n");
1122 
1123 	KASSERT_UNIONFS_VNODE(ap->a_tdvp);
1124 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1125 
1126 	error = 0;
1127 	needrelookup = 0;
1128 	dunp = VTOUNIONFS(ap->a_tdvp);
1129 	unp = NULL;
1130 	udvp = dunp->un_uppervp;
1131 	uvp = NULLVP;
1132 	cnp = ap->a_cnp;
1133 	td = curthread;
1134 
1135 	if (udvp == NULLVP)
1136 		return (EROFS);
1137 
1138 	if (ap->a_vp->v_op != &unionfs_vnodeops)
1139 		uvp = ap->a_vp;
1140 	else {
1141 		unp = VTOUNIONFS(ap->a_vp);
1142 
1143 		if (unp->un_uppervp == NULLVP) {
1144 			if (ap->a_vp->v_type != VREG)
1145 				return (EOPNOTSUPP);
1146 
1147 			error = unionfs_copyfile(unp, 1, cnp->cn_cred, td);
1148 			if (error != 0)
1149 				return (error);
1150 			needrelookup = 1;
1151 		}
1152 		uvp = unp->un_uppervp;
1153 	}
1154 
1155 	if (needrelookup != 0)
1156 		error = unionfs_relookup_for_create(ap->a_tdvp, cnp, td);
1157 
1158 	if (error == 0)
1159 		error = VOP_LINK(udvp, uvp, cnp);
1160 
1161 	UNIONFS_INTERNAL_DEBUG("unionfs_link: leave (%d)\n", error);
1162 
1163 	return (error);
1164 }
1165 
1166 static int
1167 unionfs_rename(struct vop_rename_args *ap)
1168 {
1169 	struct vnode   *fdvp;
1170 	struct vnode   *fvp;
1171 	struct componentname *fcnp;
1172 	struct vnode   *tdvp;
1173 	struct vnode   *tvp;
1174 	struct componentname *tcnp;
1175 	struct vnode   *ltdvp;
1176 	struct vnode   *ltvp;
1177 	struct thread  *td;
1178 
1179 	/* rename target vnodes */
1180 	struct vnode   *rfdvp;
1181 	struct vnode   *rfvp;
1182 	struct vnode   *rtdvp;
1183 	struct vnode   *rtvp;
1184 
1185 	struct unionfs_mount *ump;
1186 	struct unionfs_node *unp;
1187 	int		error;
1188 	int		needrelookup;
1189 
1190 	UNIONFS_INTERNAL_DEBUG("unionfs_rename: enter\n");
1191 
1192 	error = 0;
1193 	fdvp = ap->a_fdvp;
1194 	fvp = ap->a_fvp;
1195 	fcnp = ap->a_fcnp;
1196 	tdvp = ap->a_tdvp;
1197 	tvp = ap->a_tvp;
1198 	tcnp = ap->a_tcnp;
1199 	ltdvp = NULLVP;
1200 	ltvp = NULLVP;
1201 	td = curthread;
1202 	rfdvp = fdvp;
1203 	rfvp = fvp;
1204 	rtdvp = tdvp;
1205 	rtvp = tvp;
1206 	needrelookup = 0;
1207 
1208 #ifdef DIAGNOSTIC
1209 	if (!(fcnp->cn_flags & HASBUF) || !(tcnp->cn_flags & HASBUF))
1210 		panic("unionfs_rename: no name");
1211 #endif
1212 
1213 	/* check for cross device rename */
1214 	if (fvp->v_mount != tdvp->v_mount ||
1215 	    (tvp != NULLVP && fvp->v_mount != tvp->v_mount)) {
1216 		if (fvp->v_op != &unionfs_vnodeops)
1217 			error = ENODEV;
1218 		else
1219 			error = EXDEV;
1220 		goto unionfs_rename_abort;
1221 	}
1222 
1223 	/* Renaming a file to itself has no effect. */
1224 	if (fvp == tvp)
1225 		goto unionfs_rename_abort;
1226 
1227 	/*
1228 	 * from/to vnode is unionfs node.
1229 	 */
1230 
1231 	KASSERT_UNIONFS_VNODE(fdvp);
1232 	KASSERT_UNIONFS_VNODE(fvp);
1233 	KASSERT_UNIONFS_VNODE(tdvp);
1234 	if (tvp != NULLVP)
1235 		KASSERT_UNIONFS_VNODE(tvp);
1236 
1237 	unp = VTOUNIONFS(fdvp);
1238 #ifdef UNIONFS_IDBG_RENAME
1239 	UNIONFS_INTERNAL_DEBUG("fdvp=%p, ufdvp=%p, lfdvp=%p\n",
1240 	    fdvp, unp->un_uppervp, unp->un_lowervp);
1241 #endif
1242 	if (unp->un_uppervp == NULLVP) {
1243 		error = ENODEV;
1244 		goto unionfs_rename_abort;
1245 	}
1246 	rfdvp = unp->un_uppervp;
1247 	vref(rfdvp);
1248 
1249 	unp = VTOUNIONFS(fvp);
1250 #ifdef UNIONFS_IDBG_RENAME
1251 	UNIONFS_INTERNAL_DEBUG("fvp=%p, ufvp=%p, lfvp=%p\n",
1252 	    fvp, unp->un_uppervp, unp->un_lowervp);
1253 #endif
1254 	ump = MOUNTTOUNIONFSMOUNT(fvp->v_mount);
1255 	if (unp->un_uppervp == NULLVP) {
1256 		switch (fvp->v_type) {
1257 		case VREG:
1258 			if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1259 				goto unionfs_rename_abort;
1260 			error = unionfs_copyfile(unp, 1, fcnp->cn_cred, td);
1261 			VOP_UNLOCK(fvp);
1262 			if (error != 0)
1263 				goto unionfs_rename_abort;
1264 			break;
1265 		case VDIR:
1266 			if ((error = vn_lock(fvp, LK_EXCLUSIVE)) != 0)
1267 				goto unionfs_rename_abort;
1268 			error = unionfs_mkshadowdir(ump, rfdvp, unp, fcnp, td);
1269 			VOP_UNLOCK(fvp);
1270 			if (error != 0)
1271 				goto unionfs_rename_abort;
1272 			break;
1273 		default:
1274 			error = ENODEV;
1275 			goto unionfs_rename_abort;
1276 		}
1277 
1278 		needrelookup = 1;
1279 	}
1280 
1281 	if (unp->un_lowervp != NULLVP)
1282 		fcnp->cn_flags |= DOWHITEOUT;
1283 	rfvp = unp->un_uppervp;
1284 	vref(rfvp);
1285 
1286 	unp = VTOUNIONFS(tdvp);
1287 #ifdef UNIONFS_IDBG_RENAME
1288 	UNIONFS_INTERNAL_DEBUG("tdvp=%p, utdvp=%p, ltdvp=%p\n",
1289 	    tdvp, unp->un_uppervp, unp->un_lowervp);
1290 #endif
1291 	if (unp->un_uppervp == NULLVP) {
1292 		error = ENODEV;
1293 		goto unionfs_rename_abort;
1294 	}
1295 	rtdvp = unp->un_uppervp;
1296 	ltdvp = unp->un_lowervp;
1297 	vref(rtdvp);
1298 
1299 	if (tdvp == tvp) {
1300 		rtvp = rtdvp;
1301 		vref(rtvp);
1302 	} else if (tvp != NULLVP) {
1303 		unp = VTOUNIONFS(tvp);
1304 #ifdef UNIONFS_IDBG_RENAME
1305 		UNIONFS_INTERNAL_DEBUG("tvp=%p, utvp=%p, ltvp=%p\n",
1306 		    tvp, unp->un_uppervp, unp->un_lowervp);
1307 #endif
1308 		if (unp->un_uppervp == NULLVP)
1309 			rtvp = NULLVP;
1310 		else {
1311 			if (tvp->v_type == VDIR) {
1312 				error = EINVAL;
1313 				goto unionfs_rename_abort;
1314 			}
1315 			rtvp = unp->un_uppervp;
1316 			ltvp = unp->un_lowervp;
1317 			vref(rtvp);
1318 		}
1319 	}
1320 
1321 	if (rfvp == rtvp)
1322 		goto unionfs_rename_abort;
1323 
1324 	if (needrelookup != 0) {
1325 		if ((error = vn_lock(fdvp, LK_EXCLUSIVE)) != 0)
1326 			goto unionfs_rename_abort;
1327 		error = unionfs_relookup_for_delete(fdvp, fcnp, td);
1328 		VOP_UNLOCK(fdvp);
1329 		if (error != 0)
1330 			goto unionfs_rename_abort;
1331 
1332 		/* Lock of tvp is canceled in order to avoid recursive lock. */
1333 		if (tvp != NULLVP && tvp != tdvp)
1334 			VOP_UNLOCK(tvp);
1335 		error = unionfs_relookup_for_rename(tdvp, tcnp, td);
1336 		if (tvp != NULLVP && tvp != tdvp)
1337 			vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
1338 		if (error != 0)
1339 			goto unionfs_rename_abort;
1340 	}
1341 
1342 	error = VOP_RENAME(rfdvp, rfvp, fcnp, rtdvp, rtvp, tcnp);
1343 
1344 	if (error == 0) {
1345 		if (rtvp != NULLVP && rtvp->v_type == VDIR)
1346 			cache_purge(tdvp);
1347 		if (fvp->v_type == VDIR && fdvp != tdvp)
1348 			cache_purge(fdvp);
1349 	}
1350 
1351 	if (ltdvp != NULLVP)
1352 		VOP_UNLOCK(ltdvp);
1353 	if (tdvp != rtdvp)
1354 		vrele(tdvp);
1355 	if (ltvp != NULLVP)
1356 		VOP_UNLOCK(ltvp);
1357 	if (tvp != rtvp && tvp != NULLVP) {
1358 		if (rtvp == NULLVP)
1359 			vput(tvp);
1360 		else
1361 			vrele(tvp);
1362 	}
1363 	if (fdvp != rfdvp)
1364 		vrele(fdvp);
1365 	if (fvp != rfvp)
1366 		vrele(fvp);
1367 
1368 	UNIONFS_INTERNAL_DEBUG("unionfs_rename: leave (%d)\n", error);
1369 
1370 	return (error);
1371 
1372 unionfs_rename_abort:
1373 	vput(tdvp);
1374 	if (tdvp != rtdvp)
1375 		vrele(rtdvp);
1376 	if (tvp != NULLVP) {
1377 		if (tdvp != tvp)
1378 			vput(tvp);
1379 		else
1380 			vrele(tvp);
1381 	}
1382 	if (tvp != rtvp && rtvp != NULLVP)
1383 		vrele(rtvp);
1384 	if (fdvp != rfdvp)
1385 		vrele(rfdvp);
1386 	if (fvp != rfvp)
1387 		vrele(rfvp);
1388 	vrele(fdvp);
1389 	vrele(fvp);
1390 
1391 	UNIONFS_INTERNAL_DEBUG("unionfs_rename: leave (%d)\n", error);
1392 
1393 	return (error);
1394 }
1395 
1396 static int
1397 unionfs_mkdir(struct vop_mkdir_args *ap)
1398 {
1399 	struct unionfs_node *dunp;
1400 	struct componentname *cnp;
1401 	struct vnode   *udvp;
1402 	struct vnode   *uvp;
1403 	struct vattr	va;
1404 	int		error;
1405 	int		lkflags;
1406 
1407 	UNIONFS_INTERNAL_DEBUG("unionfs_mkdir: enter\n");
1408 
1409 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
1410 
1411 	error = EROFS;
1412 	dunp = VTOUNIONFS(ap->a_dvp);
1413 	cnp = ap->a_cnp;
1414 	lkflags = cnp->cn_lkflags;
1415 	udvp = dunp->un_uppervp;
1416 
1417 	if (udvp != NULLVP) {
1418 		/* check opaque */
1419 		if (!(cnp->cn_flags & ISWHITEOUT)) {
1420 			error = VOP_GETATTR(udvp, &va, cnp->cn_cred);
1421 			if (error != 0)
1422 				return (error);
1423 			if ((va.va_flags & OPAQUE) != 0)
1424 				cnp->cn_flags |= ISWHITEOUT;
1425 		}
1426 
1427 		if ((error = VOP_MKDIR(udvp, &uvp, cnp, ap->a_vap)) == 0) {
1428 			VOP_UNLOCK(uvp);
1429 			cnp->cn_lkflags = LK_EXCLUSIVE;
1430 			error = unionfs_nodeget(ap->a_dvp->v_mount, uvp, NULLVP,
1431 			    ap->a_dvp, ap->a_vpp, cnp);
1432 			cnp->cn_lkflags = lkflags;
1433 			vrele(uvp);
1434 		}
1435 	}
1436 
1437 	UNIONFS_INTERNAL_DEBUG("unionfs_mkdir: leave (%d)\n", error);
1438 
1439 	return (error);
1440 }
1441 
1442 static int
1443 unionfs_rmdir(struct vop_rmdir_args *ap)
1444 {
1445 	struct unionfs_node *dunp;
1446 	struct unionfs_node *unp;
1447 	struct unionfs_mount *ump;
1448 	struct componentname *cnp;
1449 	struct thread  *td;
1450 	struct vnode   *udvp;
1451 	struct vnode   *uvp;
1452 	struct vnode   *lvp;
1453 	int		error;
1454 
1455 	UNIONFS_INTERNAL_DEBUG("unionfs_rmdir: enter\n");
1456 
1457 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
1458 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1459 
1460 	error = 0;
1461 	dunp = VTOUNIONFS(ap->a_dvp);
1462 	unp = VTOUNIONFS(ap->a_vp);
1463 	cnp = ap->a_cnp;
1464 	td = curthread;
1465 	udvp = dunp->un_uppervp;
1466 	uvp = unp->un_uppervp;
1467 	lvp = unp->un_lowervp;
1468 
1469 	if (udvp == NULLVP)
1470 		return (EROFS);
1471 
1472 	if (udvp == uvp)
1473 		return (EOPNOTSUPP);
1474 
1475 	if (uvp != NULLVP) {
1476 		if (lvp != NULLVP) {
1477 			error = unionfs_check_rmdir(ap->a_vp, cnp->cn_cred, td);
1478 			if (error != 0)
1479 				return (error);
1480 		}
1481 		ump = MOUNTTOUNIONFSMOUNT(ap->a_vp->v_mount);
1482 		if (ump->um_whitemode == UNIONFS_WHITE_ALWAYS || lvp != NULLVP)
1483 			cnp->cn_flags |= DOWHITEOUT;
1484 		/*
1485 		 * The relookup path will need to relock the parent dvp and
1486 		 * possibly the vp as well.  Locking is expected to be done
1487 		 * in parent->child order; drop the lock on vp to avoid LOR
1488 		 * and potential recursion on vp's lock.
1489 		 * vp is expected to remain referenced during VOP_RMDIR(),
1490 		 * so vref/vrele should not be necessary here.
1491 		 */
1492 		VOP_UNLOCK(ap->a_vp);
1493 		VNPASS(vrefcnt(ap->a_vp) > 0, ap->a_vp);
1494 		error = unionfs_relookup_for_delete(ap->a_dvp, cnp, td);
1495 		vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
1496 		/*
1497 		 * VOP_RMDIR is dispatched against udvp, so if uvp became
1498 		 * doomed while the lock was dropped above the target
1499 		 * filesystem may not be able to cope.
1500 		 */
1501 		if (error == 0 && VN_IS_DOOMED(uvp))
1502 			error = ENOENT;
1503 		if (error == 0)
1504 			error = VOP_RMDIR(udvp, uvp, cnp);
1505 	} else if (lvp != NULLVP)
1506 		error = unionfs_mkwhiteout(udvp, cnp, td,
1507 		    unp->un_path, unp->un_pathlen);
1508 
1509 	if (error == 0) {
1510 		cache_purge(ap->a_dvp);
1511 		cache_purge(ap->a_vp);
1512 	}
1513 
1514 	UNIONFS_INTERNAL_DEBUG("unionfs_rmdir: leave (%d)\n", error);
1515 
1516 	return (error);
1517 }
1518 
1519 static int
1520 unionfs_symlink(struct vop_symlink_args *ap)
1521 {
1522 	struct unionfs_node *dunp;
1523 	struct componentname *cnp;
1524 	struct vnode   *udvp;
1525 	struct vnode   *uvp;
1526 	int		error;
1527 	int		lkflags;
1528 
1529 	UNIONFS_INTERNAL_DEBUG("unionfs_symlink: enter\n");
1530 
1531 	KASSERT_UNIONFS_VNODE(ap->a_dvp);
1532 
1533 	error = EROFS;
1534 	dunp = VTOUNIONFS(ap->a_dvp);
1535 	cnp = ap->a_cnp;
1536 	lkflags = cnp->cn_lkflags;
1537 	udvp = dunp->un_uppervp;
1538 
1539 	if (udvp != NULLVP) {
1540 		error = VOP_SYMLINK(udvp, &uvp, cnp, ap->a_vap, ap->a_target);
1541 		if (error == 0) {
1542 			VOP_UNLOCK(uvp);
1543 			cnp->cn_lkflags = LK_EXCLUSIVE;
1544 			error = unionfs_nodeget(ap->a_dvp->v_mount, uvp, NULLVP,
1545 			    ap->a_dvp, ap->a_vpp, cnp);
1546 			cnp->cn_lkflags = lkflags;
1547 			vrele(uvp);
1548 		}
1549 	}
1550 
1551 	UNIONFS_INTERNAL_DEBUG("unionfs_symlink: leave (%d)\n", error);
1552 
1553 	return (error);
1554 }
1555 
1556 static int
1557 unionfs_readdir(struct vop_readdir_args *ap)
1558 {
1559 	struct unionfs_node *unp;
1560 	struct unionfs_node_status *unsp;
1561 	struct uio     *uio;
1562 	struct vnode   *vp;
1563 	struct vnode   *uvp;
1564 	struct vnode   *lvp;
1565 	struct thread  *td;
1566 	struct vattr    va;
1567 
1568 	uint64_t	*cookies_bk;
1569 	int		error;
1570 	int		eofflag;
1571 	int		ncookies_bk;
1572 	int		uio_offset_bk;
1573 	enum unionfs_lkupgrade lkstatus;
1574 
1575 	UNIONFS_INTERNAL_DEBUG("unionfs_readdir: enter\n");
1576 
1577 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1578 
1579 	error = 0;
1580 	eofflag = 0;
1581 	uio_offset_bk = 0;
1582 	uio = ap->a_uio;
1583 	uvp = NULLVP;
1584 	lvp = NULLVP;
1585 	td = uio->uio_td;
1586 	ncookies_bk = 0;
1587 	cookies_bk = NULL;
1588 
1589 	vp = ap->a_vp;
1590 	if (vp->v_type != VDIR)
1591 		return (ENOTDIR);
1592 
1593 	/*
1594 	 * If the vnode is reclaimed while upgrading, we can't safely use unp
1595 	 * or do anything else unionfs- specific.
1596 	 */
1597 	lkstatus = unionfs_upgrade_lock(vp);
1598 	if (lkstatus == UNIONFS_LKUPGRADE_DOOMED)
1599 		error = EBADF;
1600 	if (error == 0) {
1601 		unp = VTOUNIONFS(vp);
1602 		uvp = unp->un_uppervp;
1603 		lvp = unp->un_lowervp;
1604 		/* check the open count. unionfs needs open before readdir. */
1605 		unionfs_get_node_status(unp, td, &unsp);
1606 		if ((uvp != NULLVP && unsp->uns_upper_opencnt <= 0) ||
1607 			(lvp != NULLVP && unsp->uns_lower_opencnt <= 0)) {
1608 			unionfs_tryrem_node_status(unp, unsp);
1609 			error = EBADF;
1610 		}
1611 	}
1612 	unionfs_downgrade_lock(vp, lkstatus);
1613 	if (error != 0)
1614 		goto unionfs_readdir_exit;
1615 
1616 	/* check opaque */
1617 	if (uvp != NULLVP && lvp != NULLVP) {
1618 		if ((error = VOP_GETATTR(uvp, &va, ap->a_cred)) != 0)
1619 			goto unionfs_readdir_exit;
1620 		if (va.va_flags & OPAQUE)
1621 			lvp = NULLVP;
1622 	}
1623 
1624 	/* upper only */
1625 	if (uvp != NULLVP && lvp == NULLVP) {
1626 		error = VOP_READDIR(uvp, uio, ap->a_cred, ap->a_eofflag,
1627 		    ap->a_ncookies, ap->a_cookies);
1628 		unsp->uns_readdir_status = 0;
1629 
1630 		goto unionfs_readdir_exit;
1631 	}
1632 
1633 	/* lower only */
1634 	if (uvp == NULLVP && lvp != NULLVP) {
1635 		error = VOP_READDIR(lvp, uio, ap->a_cred, ap->a_eofflag,
1636 		    ap->a_ncookies, ap->a_cookies);
1637 		unsp->uns_readdir_status = 2;
1638 
1639 		goto unionfs_readdir_exit;
1640 	}
1641 
1642 	/*
1643 	 * readdir upper and lower
1644 	 */
1645 	KASSERT(uvp != NULLVP, ("unionfs_readdir: null upper vp"));
1646 	KASSERT(lvp != NULLVP, ("unionfs_readdir: null lower vp"));
1647 	if (uio->uio_offset == 0)
1648 		unsp->uns_readdir_status = 0;
1649 
1650 	if (unsp->uns_readdir_status == 0) {
1651 		/* read upper */
1652 		error = VOP_READDIR(uvp, uio, ap->a_cred, &eofflag,
1653 				    ap->a_ncookies, ap->a_cookies);
1654 
1655 		if (error != 0 || eofflag == 0)
1656 			goto unionfs_readdir_exit;
1657 		unsp->uns_readdir_status = 1;
1658 
1659 		/*
1660 		 * UFS(and other FS) needs size of uio_resid larger than
1661 		 * DIRBLKSIZ.
1662 		 * size of DIRBLKSIZ equals DEV_BSIZE.
1663 		 * (see: ufs/ufs/ufs_vnops.c ufs_readdir func , ufs/ufs/dir.h)
1664 		 */
1665 		if (uio->uio_resid <= (uio->uio_resid & (DEV_BSIZE -1)))
1666 			goto unionfs_readdir_exit;
1667 
1668 		/*
1669 		 * Backup cookies.
1670 		 * It prepares to readdir in lower.
1671 		 */
1672 		if (ap->a_ncookies != NULL) {
1673 			ncookies_bk = *(ap->a_ncookies);
1674 			*(ap->a_ncookies) = 0;
1675 		}
1676 		if (ap->a_cookies != NULL) {
1677 			cookies_bk = *(ap->a_cookies);
1678 			*(ap->a_cookies) = NULL;
1679 		}
1680 	}
1681 
1682 	/* initialize for readdir in lower */
1683 	if (unsp->uns_readdir_status == 1) {
1684 		unsp->uns_readdir_status = 2;
1685 		/*
1686 		 * Backup uio_offset. See the comment after the
1687 		 * VOP_READDIR call on the lower layer.
1688 		 */
1689 		uio_offset_bk = uio->uio_offset;
1690 		uio->uio_offset = 0;
1691 	}
1692 
1693 	if (lvp == NULLVP) {
1694 		error = EBADF;
1695 		goto unionfs_readdir_exit;
1696 	}
1697 	/* read lower */
1698 	error = VOP_READDIR(lvp, uio, ap->a_cred, ap->a_eofflag,
1699 			    ap->a_ncookies, ap->a_cookies);
1700 
1701 	/*
1702 	 * We can't return an uio_offset of 0: this would trigger an
1703 	 * infinite loop, because the next call to unionfs_readdir would
1704 	 * always restart with the upper layer (uio_offset == 0) and
1705 	 * always return some data.
1706 	 *
1707 	 * This happens when the lower layer root directory is removed.
1708 	 * (A root directory deleting of unionfs should not be permitted.
1709 	 *  But current VFS can not do it.)
1710 	 */
1711 	if (uio->uio_offset == 0)
1712 		uio->uio_offset = uio_offset_bk;
1713 
1714 	if (cookies_bk != NULL) {
1715 		/* merge cookies */
1716 		int		size;
1717 		uint64_t         *newcookies, *pos;
1718 
1719 		size = *(ap->a_ncookies) + ncookies_bk;
1720 		newcookies = (uint64_t *) malloc(size * sizeof(*newcookies),
1721 		    M_TEMP, M_WAITOK);
1722 		pos = newcookies;
1723 
1724 		memcpy(pos, cookies_bk, ncookies_bk * sizeof(*newcookies));
1725 		pos += ncookies_bk;
1726 		memcpy(pos, *(ap->a_cookies),
1727 		    *(ap->a_ncookies) * sizeof(*newcookies));
1728 		free(cookies_bk, M_TEMP);
1729 		free(*(ap->a_cookies), M_TEMP);
1730 		*(ap->a_ncookies) = size;
1731 		*(ap->a_cookies) = newcookies;
1732 	}
1733 
1734 unionfs_readdir_exit:
1735 	if (error != 0 && ap->a_eofflag != NULL)
1736 		*(ap->a_eofflag) = 1;
1737 
1738 	UNIONFS_INTERNAL_DEBUG("unionfs_readdir: leave (%d)\n", error);
1739 
1740 	return (error);
1741 }
1742 
1743 static int
1744 unionfs_readlink(struct vop_readlink_args *ap)
1745 {
1746 	struct unionfs_node *unp;
1747 	struct vnode   *vp;
1748 	int error;
1749 
1750 	UNIONFS_INTERNAL_DEBUG("unionfs_readlink: enter\n");
1751 
1752 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1753 
1754 	unp = VTOUNIONFS(ap->a_vp);
1755 	vp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
1756 
1757 	error = VOP_READLINK(vp, ap->a_uio, ap->a_cred);
1758 
1759 	UNIONFS_INTERNAL_DEBUG("unionfs_readlink: leave (%d)\n", error);
1760 
1761 	return (error);
1762 }
1763 
1764 static int
1765 unionfs_getwritemount(struct vop_getwritemount_args *ap)
1766 {
1767 	struct unionfs_node *unp;
1768 	struct vnode   *uvp;
1769 	struct vnode   *vp, *ovp;
1770 	int		error;
1771 
1772 	UNIONFS_INTERNAL_DEBUG("unionfs_getwritemount: enter\n");
1773 
1774 	error = 0;
1775 	vp = ap->a_vp;
1776 	uvp = NULLVP;
1777 
1778 	VI_LOCK(vp);
1779 	unp = VTOUNIONFS(vp);
1780 	if (unp != NULL)
1781 		uvp = unp->un_uppervp;
1782 
1783 	/*
1784 	 * If our node has no upper vnode, check the parent directory.
1785 	 * We may be initiating a write operation that will produce a
1786 	 * new upper vnode through CoW.
1787 	 */
1788 	if (uvp == NULLVP && unp != NULL) {
1789 		ovp = vp;
1790 		vp = unp->un_dvp;
1791 		/*
1792 		 * Only the root vnode should have an empty parent, but it
1793 		 * should not have an empty uppervp, so we shouldn't get here.
1794 		 */
1795 		VNASSERT(vp != NULL, ovp, ("%s: NULL parent vnode", __func__));
1796 		VI_UNLOCK(ovp);
1797 		VI_LOCK(vp);
1798 		unp = VTOUNIONFS(vp);
1799 		if (unp != NULL)
1800 			uvp = unp->un_uppervp;
1801 		if (uvp == NULLVP)
1802 			error = EACCES;
1803 	}
1804 
1805 	if (uvp != NULLVP) {
1806 		vholdnz(uvp);
1807 		VI_UNLOCK(vp);
1808 		error = VOP_GETWRITEMOUNT(uvp, ap->a_mpp);
1809 		vdrop(uvp);
1810 	} else {
1811 		VI_UNLOCK(vp);
1812 		*(ap->a_mpp) = NULL;
1813 	}
1814 
1815 	UNIONFS_INTERNAL_DEBUG("unionfs_getwritemount: leave (%d)\n", error);
1816 
1817 	return (error);
1818 }
1819 
1820 static int
1821 unionfs_inactive(struct vop_inactive_args *ap)
1822 {
1823 	ap->a_vp->v_object = NULL;
1824 	vrecycle(ap->a_vp);
1825 	return (0);
1826 }
1827 
1828 static int
1829 unionfs_reclaim(struct vop_reclaim_args *ap)
1830 {
1831 	/* UNIONFS_INTERNAL_DEBUG("unionfs_reclaim: enter\n"); */
1832 
1833 	unionfs_noderem(ap->a_vp);
1834 
1835 	/* UNIONFS_INTERNAL_DEBUG("unionfs_reclaim: leave\n"); */
1836 
1837 	return (0);
1838 }
1839 
1840 static int
1841 unionfs_print(struct vop_print_args *ap)
1842 {
1843 	struct unionfs_node *unp;
1844 	/* struct unionfs_node_status *unsp; */
1845 
1846 	unp = VTOUNIONFS(ap->a_vp);
1847 	/* unionfs_get_node_status(unp, curthread, &unsp); */
1848 
1849 	printf("unionfs_vp=%p, uppervp=%p, lowervp=%p\n",
1850 	    ap->a_vp, unp->un_uppervp, unp->un_lowervp);
1851 	/*
1852 	printf("unionfs opencnt: uppervp=%d, lowervp=%d\n",
1853 	    unsp->uns_upper_opencnt, unsp->uns_lower_opencnt);
1854 	*/
1855 
1856 	if (unp->un_uppervp != NULLVP)
1857 		vn_printf(unp->un_uppervp, "unionfs: upper ");
1858 	if (unp->un_lowervp != NULLVP)
1859 		vn_printf(unp->un_lowervp, "unionfs: lower ");
1860 
1861 	return (0);
1862 }
1863 
1864 static int
1865 unionfs_islocked(struct vop_islocked_args *ap)
1866 {
1867 	struct unionfs_node *unp;
1868 
1869 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1870 
1871 	unp = VTOUNIONFS(ap->a_vp);
1872 	if (unp == NULL)
1873 		return (vop_stdislocked(ap));
1874 
1875 	if (unp->un_uppervp != NULLVP)
1876 		return (VOP_ISLOCKED(unp->un_uppervp));
1877 	if (unp->un_lowervp != NULLVP)
1878 		return (VOP_ISLOCKED(unp->un_lowervp));
1879 	return (vop_stdislocked(ap));
1880 }
1881 
1882 static int
1883 unionfs_get_llt_revlock(struct vnode *vp, int flags)
1884 {
1885 	int revlock;
1886 
1887 	revlock = 0;
1888 
1889 	switch (flags & LK_TYPE_MASK) {
1890 	case LK_SHARED:
1891 		if (VOP_ISLOCKED(vp) == LK_EXCLUSIVE)
1892 			revlock = LK_UPGRADE;
1893 		else
1894 			revlock = LK_RELEASE;
1895 		break;
1896 	case LK_EXCLUSIVE:
1897 	case LK_UPGRADE:
1898 		revlock = LK_RELEASE;
1899 		break;
1900 	case LK_DOWNGRADE:
1901 		revlock = LK_UPGRADE;
1902 		break;
1903 	default:
1904 		break;
1905 	}
1906 
1907 	return (revlock);
1908 }
1909 
1910 /*
1911  * The state of an acquired lock is adjusted similarly to
1912  * the time of error generating.
1913  * flags: LK_RELEASE or LK_UPGRADE
1914  */
1915 static void
1916 unionfs_revlock(struct vnode *vp, int flags)
1917 {
1918 	if (flags & LK_RELEASE)
1919 		VOP_UNLOCK_FLAGS(vp, flags);
1920 	else {
1921 		/* UPGRADE */
1922 		if (vn_lock(vp, flags) != 0)
1923 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1924 	}
1925 }
1926 
1927 static int
1928 unionfs_lock(struct vop_lock1_args *ap)
1929 {
1930 	struct unionfs_node *unp;
1931 	struct vnode   *vp;
1932 	struct vnode   *uvp;
1933 	struct vnode   *lvp;
1934 	int		error;
1935 	int		flags;
1936 	int		revlock;
1937 	int		interlock;
1938 	int		uhold;
1939 
1940 	/*
1941 	 * TODO: rework the unionfs locking scheme.
1942 	 * It's not guaranteed to be safe to blindly lock two vnodes on
1943 	 * different mounts as is done here.  Further, the entanglement
1944 	 * of locking both vnodes with the various options that can be
1945 	 * passed to VOP_LOCK() makes this code hard to reason about.
1946 	 * Instead, consider locking only the upper vnode, or the lower
1947 	 * vnode is the upper is not present, and taking separate measures
1948 	 * to lock both vnodes in the few cases when that is needed.
1949 	 */
1950 	error = 0;
1951 	interlock = 1;
1952 	uhold = 0;
1953 	flags = ap->a_flags;
1954 	vp = ap->a_vp;
1955 
1956 	if (LK_RELEASE == (flags & LK_TYPE_MASK) || !(flags & LK_TYPE_MASK))
1957 		return (VOP_UNLOCK_FLAGS(vp, flags | LK_RELEASE));
1958 
1959 	if ((flags & LK_INTERLOCK) == 0)
1960 		VI_LOCK(vp);
1961 
1962 	unp = VTOUNIONFS(vp);
1963 	if (unp == NULL)
1964 		goto unionfs_lock_null_vnode;
1965 
1966 	KASSERT_UNIONFS_VNODE(ap->a_vp);
1967 
1968 	lvp = unp->un_lowervp;
1969 	uvp = unp->un_uppervp;
1970 
1971 	if ((revlock = unionfs_get_llt_revlock(vp, flags)) == 0)
1972 		panic("unknown lock type: 0x%x", flags & LK_TYPE_MASK);
1973 
1974 	/*
1975 	 * During unmount, the root vnode lock may be taken recursively,
1976 	 * because it may share the same v_vnlock field as the vnode covered by
1977 	 * the unionfs mount.  The covered vnode is locked across VFS_UNMOUNT(),
1978 	 * and the same lock may be taken recursively here during vflush()
1979 	 * issued by unionfs_unmount().
1980 	 */
1981 	if ((flags & LK_TYPE_MASK) == LK_EXCLUSIVE &&
1982 	    (vp->v_vflag & VV_ROOT) != 0)
1983 		flags |= LK_CANRECURSE;
1984 
1985 	if (lvp != NULLVP) {
1986 		if (uvp != NULLVP && flags & LK_UPGRADE) {
1987 			/*
1988 			 * Share Lock is once released and a deadlock is
1989 			 * avoided.
1990 			 */
1991 			vholdnz(uvp);
1992 			uhold = 1;
1993 			VOP_UNLOCK(uvp);
1994 			unp = VTOUNIONFS(vp);
1995 			if (unp == NULL) {
1996 				/* vnode is released. */
1997 				VI_UNLOCK(vp);
1998 				VOP_UNLOCK(lvp);
1999 				vdrop(uvp);
2000 				return (EBUSY);
2001 			}
2002 		}
2003 		VI_LOCK_FLAGS(lvp, MTX_DUPOK);
2004 		flags |= LK_INTERLOCK;
2005 		vholdl(lvp);
2006 
2007 		VI_UNLOCK(vp);
2008 		ap->a_flags &= ~LK_INTERLOCK;
2009 
2010 		error = VOP_LOCK(lvp, flags);
2011 
2012 		VI_LOCK(vp);
2013 		unp = VTOUNIONFS(vp);
2014 		if (unp == NULL) {
2015 			/* vnode is released. */
2016 			VI_UNLOCK(vp);
2017 			if (error == 0)
2018 				VOP_UNLOCK(lvp);
2019 			vdrop(lvp);
2020 			if (uhold != 0)
2021 				vdrop(uvp);
2022 			return (vop_stdlock(ap));
2023 		}
2024 	}
2025 
2026 	if (error == 0 && uvp != NULLVP) {
2027 		if (uhold && flags & LK_UPGRADE) {
2028 			flags &= ~LK_TYPE_MASK;
2029 			flags |= LK_EXCLUSIVE;
2030 		}
2031 		VI_LOCK_FLAGS(uvp, MTX_DUPOK);
2032 		flags |= LK_INTERLOCK;
2033 		if (uhold == 0) {
2034 			vholdl(uvp);
2035 			uhold = 1;
2036 		}
2037 
2038 		VI_UNLOCK(vp);
2039 		ap->a_flags &= ~LK_INTERLOCK;
2040 
2041 		error = VOP_LOCK(uvp, flags);
2042 
2043 		VI_LOCK(vp);
2044 		unp = VTOUNIONFS(vp);
2045 		if (unp == NULL) {
2046 			/* vnode is released. */
2047 			VI_UNLOCK(vp);
2048 			if (error == 0)
2049 				VOP_UNLOCK(uvp);
2050 			vdrop(uvp);
2051 			if (lvp != NULLVP) {
2052 				VOP_UNLOCK(lvp);
2053 				vdrop(lvp);
2054 			}
2055 			return (vop_stdlock(ap));
2056 		}
2057 		if (error != 0 && lvp != NULLVP) {
2058 			/* rollback */
2059 			VI_UNLOCK(vp);
2060 			unionfs_revlock(lvp, revlock);
2061 			interlock = 0;
2062 		}
2063 	}
2064 
2065 	if (interlock)
2066 		VI_UNLOCK(vp);
2067 	if (lvp != NULLVP)
2068 		vdrop(lvp);
2069 	if (uhold != 0)
2070 		vdrop(uvp);
2071 
2072 	return (error);
2073 
2074 unionfs_lock_null_vnode:
2075 	ap->a_flags |= LK_INTERLOCK;
2076 	return (vop_stdlock(ap));
2077 }
2078 
2079 static int
2080 unionfs_unlock(struct vop_unlock_args *ap)
2081 {
2082 	struct vnode   *vp;
2083 	struct vnode   *lvp;
2084 	struct vnode   *uvp;
2085 	struct unionfs_node *unp;
2086 	int		error;
2087 	int		uhold;
2088 
2089 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2090 
2091 	error = 0;
2092 	uhold = 0;
2093 	vp = ap->a_vp;
2094 
2095 	unp = VTOUNIONFS(vp);
2096 	if (unp == NULL)
2097 		goto unionfs_unlock_null_vnode;
2098 	lvp = unp->un_lowervp;
2099 	uvp = unp->un_uppervp;
2100 
2101 	if (lvp != NULLVP) {
2102 		vholdnz(lvp);
2103 		error = VOP_UNLOCK(lvp);
2104 	}
2105 
2106 	if (error == 0 && uvp != NULLVP) {
2107 		vholdnz(uvp);
2108 		uhold = 1;
2109 		error = VOP_UNLOCK(uvp);
2110 	}
2111 
2112 	if (lvp != NULLVP)
2113 		vdrop(lvp);
2114 	if (uhold != 0)
2115 		vdrop(uvp);
2116 
2117 	return error;
2118 
2119 unionfs_unlock_null_vnode:
2120 	return (vop_stdunlock(ap));
2121 }
2122 
2123 static int
2124 unionfs_pathconf(struct vop_pathconf_args *ap)
2125 {
2126 	struct unionfs_node *unp;
2127 	struct vnode   *vp;
2128 
2129 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2130 
2131 	unp = VTOUNIONFS(ap->a_vp);
2132 	vp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
2133 
2134 	return (VOP_PATHCONF(vp, ap->a_name, ap->a_retval));
2135 }
2136 
2137 static int
2138 unionfs_advlock(struct vop_advlock_args *ap)
2139 {
2140 	struct unionfs_node *unp;
2141 	struct unionfs_node_status *unsp;
2142 	struct vnode   *vp;
2143 	struct vnode   *uvp;
2144 	struct thread  *td;
2145 	int error;
2146 
2147 	UNIONFS_INTERNAL_DEBUG("unionfs_advlock: enter\n");
2148 
2149 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2150 
2151 	vp = ap->a_vp;
2152 	td = curthread;
2153 
2154 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2155 
2156 	unp = VTOUNIONFS(ap->a_vp);
2157 	uvp = unp->un_uppervp;
2158 
2159 	if (uvp == NULLVP) {
2160 		error = unionfs_copyfile(unp, 1, td->td_ucred, td);
2161 		if (error != 0)
2162 			goto unionfs_advlock_abort;
2163 		uvp = unp->un_uppervp;
2164 
2165 		unionfs_get_node_status(unp, td, &unsp);
2166 		if (unsp->uns_lower_opencnt > 0) {
2167 			/* try reopen the vnode */
2168 			error = VOP_OPEN(uvp, unsp->uns_lower_openmode,
2169 				td->td_ucred, td, NULL);
2170 			if (error)
2171 				goto unionfs_advlock_abort;
2172 			unsp->uns_upper_opencnt++;
2173 			VOP_CLOSE(unp->un_lowervp, unsp->uns_lower_openmode,
2174 			    td->td_ucred, td);
2175 			unsp->uns_lower_opencnt--;
2176 		} else
2177 			unionfs_tryrem_node_status(unp, unsp);
2178 	}
2179 
2180 	VOP_UNLOCK(vp);
2181 
2182 	error = VOP_ADVLOCK(uvp, ap->a_id, ap->a_op, ap->a_fl, ap->a_flags);
2183 
2184 	UNIONFS_INTERNAL_DEBUG("unionfs_advlock: leave (%d)\n", error);
2185 
2186 	return error;
2187 
2188 unionfs_advlock_abort:
2189 	VOP_UNLOCK(vp);
2190 
2191 	UNIONFS_INTERNAL_DEBUG("unionfs_advlock: leave (%d)\n", error);
2192 
2193 	return error;
2194 }
2195 
2196 static int
2197 unionfs_strategy(struct vop_strategy_args *ap)
2198 {
2199 	struct unionfs_node *unp;
2200 	struct vnode   *vp;
2201 
2202 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2203 
2204 	unp = VTOUNIONFS(ap->a_vp);
2205 	vp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
2206 
2207 #ifdef DIAGNOSTIC
2208 	if (vp == NULLVP)
2209 		panic("unionfs_strategy: nullvp");
2210 
2211 	if (ap->a_bp->b_iocmd == BIO_WRITE && vp == unp->un_lowervp)
2212 		panic("unionfs_strategy: writing to lowervp");
2213 #endif
2214 
2215 	return (VOP_STRATEGY(vp, ap->a_bp));
2216 }
2217 
2218 static int
2219 unionfs_getacl(struct vop_getacl_args *ap)
2220 {
2221 	struct unionfs_node *unp;
2222 	struct vnode   *vp;
2223 	int		error;
2224 
2225 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2226 
2227 	unp = VTOUNIONFS(ap->a_vp);
2228 	vp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
2229 
2230 	UNIONFS_INTERNAL_DEBUG("unionfs_getacl: enter\n");
2231 
2232 	error = VOP_GETACL(vp, ap->a_type, ap->a_aclp, ap->a_cred, ap->a_td);
2233 
2234 	UNIONFS_INTERNAL_DEBUG("unionfs_getacl: leave (%d)\n", error);
2235 
2236 	return (error);
2237 }
2238 
2239 static int
2240 unionfs_setacl(struct vop_setacl_args *ap)
2241 {
2242 	struct unionfs_node *unp;
2243 	struct vnode   *uvp;
2244 	struct vnode   *lvp;
2245 	struct thread  *td;
2246 	int		error;
2247 
2248 	UNIONFS_INTERNAL_DEBUG("unionfs_setacl: enter\n");
2249 
2250 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2251 
2252 	error = EROFS;
2253 	unp = VTOUNIONFS(ap->a_vp);
2254 	uvp = unp->un_uppervp;
2255 	lvp = unp->un_lowervp;
2256 	td = ap->a_td;
2257 
2258 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
2259 		return (EROFS);
2260 
2261 	if (uvp == NULLVP && lvp->v_type == VREG) {
2262 		if ((error = unionfs_copyfile(unp, 1, ap->a_cred, td)) != 0)
2263 			return (error);
2264 		uvp = unp->un_uppervp;
2265 	}
2266 
2267 	if (uvp != NULLVP)
2268 		error = VOP_SETACL(uvp, ap->a_type, ap->a_aclp, ap->a_cred, td);
2269 
2270 	UNIONFS_INTERNAL_DEBUG("unionfs_setacl: leave (%d)\n", error);
2271 
2272 	return (error);
2273 }
2274 
2275 static int
2276 unionfs_aclcheck(struct vop_aclcheck_args *ap)
2277 {
2278 	struct unionfs_node *unp;
2279 	struct vnode   *vp;
2280 	int		error;
2281 
2282 	UNIONFS_INTERNAL_DEBUG("unionfs_aclcheck: enter\n");
2283 
2284 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2285 
2286 	unp = VTOUNIONFS(ap->a_vp);
2287 	vp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
2288 
2289 	error = VOP_ACLCHECK(vp, ap->a_type, ap->a_aclp, ap->a_cred, ap->a_td);
2290 
2291 	UNIONFS_INTERNAL_DEBUG("unionfs_aclcheck: leave (%d)\n", error);
2292 
2293 	return (error);
2294 }
2295 
2296 static int
2297 unionfs_openextattr(struct vop_openextattr_args *ap)
2298 {
2299 	struct unionfs_node *unp;
2300 	struct vnode   *vp;
2301 	struct vnode   *tvp;
2302 	int		error;
2303 
2304 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2305 
2306 	vp = ap->a_vp;
2307 	unp = VTOUNIONFS(vp);
2308 	tvp = (unp->un_uppervp != NULLVP ? unp->un_uppervp : unp->un_lowervp);
2309 
2310 	if ((tvp == unp->un_uppervp && (unp->un_flag & UNIONFS_OPENEXTU)) ||
2311 	    (tvp == unp->un_lowervp && (unp->un_flag & UNIONFS_OPENEXTL)))
2312 		return (EBUSY);
2313 
2314 	error = VOP_OPENEXTATTR(tvp, ap->a_cred, ap->a_td);
2315 
2316 	if (error == 0) {
2317 		if (vn_lock(vp, LK_UPGRADE) != 0)
2318 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2319 		if (!VN_IS_DOOMED(vp)) {
2320 			if (tvp == unp->un_uppervp)
2321 				unp->un_flag |= UNIONFS_OPENEXTU;
2322 			else
2323 				unp->un_flag |= UNIONFS_OPENEXTL;
2324 		}
2325 		vn_lock(vp, LK_DOWNGRADE | LK_RETRY);
2326 	}
2327 
2328 	return (error);
2329 }
2330 
2331 static int
2332 unionfs_closeextattr(struct vop_closeextattr_args *ap)
2333 {
2334 	struct unionfs_node *unp;
2335 	struct vnode   *vp;
2336 	struct vnode   *tvp;
2337 	int		error;
2338 
2339 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2340 
2341 	vp = ap->a_vp;
2342 	unp = VTOUNIONFS(vp);
2343 	tvp = NULLVP;
2344 
2345 	if (unp->un_flag & UNIONFS_OPENEXTU)
2346 		tvp = unp->un_uppervp;
2347 	else if (unp->un_flag & UNIONFS_OPENEXTL)
2348 		tvp = unp->un_lowervp;
2349 
2350 	if (tvp == NULLVP)
2351 		return (EOPNOTSUPP);
2352 
2353 	error = VOP_CLOSEEXTATTR(tvp, ap->a_commit, ap->a_cred, ap->a_td);
2354 
2355 	if (error == 0) {
2356 		if (vn_lock(vp, LK_UPGRADE) != 0)
2357 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2358 		if (!VN_IS_DOOMED(vp)) {
2359 			if (tvp == unp->un_uppervp)
2360 				unp->un_flag &= ~UNIONFS_OPENEXTU;
2361 			else
2362 				unp->un_flag &= ~UNIONFS_OPENEXTL;
2363 		}
2364 		vn_lock(vp, LK_DOWNGRADE | LK_RETRY);
2365 	}
2366 
2367 	return (error);
2368 }
2369 
2370 static int
2371 unionfs_getextattr(struct vop_getextattr_args *ap)
2372 {
2373 	struct unionfs_node *unp;
2374 	struct vnode   *vp;
2375 
2376 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2377 
2378 	unp = VTOUNIONFS(ap->a_vp);
2379 	vp = NULLVP;
2380 
2381 	if (unp->un_flag & UNIONFS_OPENEXTU)
2382 		vp = unp->un_uppervp;
2383 	else if (unp->un_flag & UNIONFS_OPENEXTL)
2384 		vp = unp->un_lowervp;
2385 
2386 	if (vp == NULLVP)
2387 		return (EOPNOTSUPP);
2388 
2389 	return (VOP_GETEXTATTR(vp, ap->a_attrnamespace, ap->a_name,
2390 	    ap->a_uio, ap->a_size, ap->a_cred, ap->a_td));
2391 }
2392 
2393 static int
2394 unionfs_setextattr(struct vop_setextattr_args *ap)
2395 {
2396 	struct unionfs_node *unp;
2397 	struct vnode   *uvp;
2398 	struct vnode   *lvp;
2399 	struct vnode   *ovp;
2400 	struct ucred   *cred;
2401 	struct thread  *td;
2402 	int		error;
2403 
2404 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2405 
2406 	error = EROFS;
2407 	unp = VTOUNIONFS(ap->a_vp);
2408 	uvp = unp->un_uppervp;
2409 	lvp = unp->un_lowervp;
2410 	ovp = NULLVP;
2411 	cred = ap->a_cred;
2412 	td = ap->a_td;
2413 
2414 	UNIONFS_INTERNAL_DEBUG("unionfs_setextattr: enter (un_flag=%x)\n",
2415 	    unp->un_flag);
2416 
2417 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
2418 		return (EROFS);
2419 
2420 	if (unp->un_flag & UNIONFS_OPENEXTU)
2421 		ovp = unp->un_uppervp;
2422 	else if (unp->un_flag & UNIONFS_OPENEXTL)
2423 		ovp = unp->un_lowervp;
2424 
2425 	if (ovp == NULLVP)
2426 		return (EOPNOTSUPP);
2427 
2428 	if (ovp == lvp && lvp->v_type == VREG) {
2429 		VOP_CLOSEEXTATTR(lvp, 0, cred, td);
2430 		if (uvp == NULLVP &&
2431 		    (error = unionfs_copyfile(unp, 1, cred, td)) != 0) {
2432 unionfs_setextattr_reopen:
2433 			if ((unp->un_flag & UNIONFS_OPENEXTL) &&
2434 			    VOP_OPENEXTATTR(lvp, cred, td)) {
2435 #ifdef DIAGNOSTIC
2436 				panic("unionfs: VOP_OPENEXTATTR failed");
2437 #endif
2438 				unp->un_flag &= ~UNIONFS_OPENEXTL;
2439 			}
2440 			goto unionfs_setextattr_abort;
2441 		}
2442 		uvp = unp->un_uppervp;
2443 		if ((error = VOP_OPENEXTATTR(uvp, cred, td)) != 0)
2444 			goto unionfs_setextattr_reopen;
2445 		unp->un_flag &= ~UNIONFS_OPENEXTL;
2446 		unp->un_flag |= UNIONFS_OPENEXTU;
2447 		ovp = uvp;
2448 	}
2449 
2450 	if (ovp == uvp)
2451 		error = VOP_SETEXTATTR(ovp, ap->a_attrnamespace, ap->a_name,
2452 		    ap->a_uio, cred, td);
2453 
2454 unionfs_setextattr_abort:
2455 	UNIONFS_INTERNAL_DEBUG("unionfs_setextattr: leave (%d)\n", error);
2456 
2457 	return (error);
2458 }
2459 
2460 static int
2461 unionfs_listextattr(struct vop_listextattr_args *ap)
2462 {
2463 	struct unionfs_node *unp;
2464 	struct vnode *vp;
2465 
2466 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2467 
2468 	unp = VTOUNIONFS(ap->a_vp);
2469 	vp = NULLVP;
2470 
2471 	if (unp->un_flag & UNIONFS_OPENEXTU)
2472 		vp = unp->un_uppervp;
2473 	else if (unp->un_flag & UNIONFS_OPENEXTL)
2474 		vp = unp->un_lowervp;
2475 
2476 	if (vp == NULLVP)
2477 		return (EOPNOTSUPP);
2478 
2479 	return (VOP_LISTEXTATTR(vp, ap->a_attrnamespace, ap->a_uio,
2480 	    ap->a_size, ap->a_cred, ap->a_td));
2481 }
2482 
2483 static int
2484 unionfs_deleteextattr(struct vop_deleteextattr_args *ap)
2485 {
2486 	struct unionfs_node *unp;
2487 	struct vnode   *uvp;
2488 	struct vnode   *lvp;
2489 	struct vnode   *ovp;
2490 	struct ucred   *cred;
2491 	struct thread  *td;
2492 	int		error;
2493 
2494 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2495 
2496 	error = EROFS;
2497 	unp = VTOUNIONFS(ap->a_vp);
2498 	uvp = unp->un_uppervp;
2499 	lvp = unp->un_lowervp;
2500 	ovp = NULLVP;
2501 	cred = ap->a_cred;
2502 	td = ap->a_td;
2503 
2504 	UNIONFS_INTERNAL_DEBUG("unionfs_deleteextattr: enter (un_flag=%x)\n",
2505 	    unp->un_flag);
2506 
2507 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
2508 		return (EROFS);
2509 
2510 	if (unp->un_flag & UNIONFS_OPENEXTU)
2511 		ovp = unp->un_uppervp;
2512 	else if (unp->un_flag & UNIONFS_OPENEXTL)
2513 		ovp = unp->un_lowervp;
2514 
2515 	if (ovp == NULLVP)
2516 		return (EOPNOTSUPP);
2517 
2518 	if (ovp == lvp && lvp->v_type == VREG) {
2519 		VOP_CLOSEEXTATTR(lvp, 0, cred, td);
2520 		if (uvp == NULLVP &&
2521 		    (error = unionfs_copyfile(unp, 1, cred, td)) != 0) {
2522 unionfs_deleteextattr_reopen:
2523 			if ((unp->un_flag & UNIONFS_OPENEXTL) &&
2524 			    VOP_OPENEXTATTR(lvp, cred, td)) {
2525 #ifdef DIAGNOSTIC
2526 				panic("unionfs: VOP_OPENEXTATTR failed");
2527 #endif
2528 				unp->un_flag &= ~UNIONFS_OPENEXTL;
2529 			}
2530 			goto unionfs_deleteextattr_abort;
2531 		}
2532 		uvp = unp->un_uppervp;
2533 		if ((error = VOP_OPENEXTATTR(uvp, cred, td)) != 0)
2534 			goto unionfs_deleteextattr_reopen;
2535 		unp->un_flag &= ~UNIONFS_OPENEXTL;
2536 		unp->un_flag |= UNIONFS_OPENEXTU;
2537 		ovp = uvp;
2538 	}
2539 
2540 	if (ovp == uvp)
2541 		error = VOP_DELETEEXTATTR(ovp, ap->a_attrnamespace, ap->a_name,
2542 		    ap->a_cred, ap->a_td);
2543 
2544 unionfs_deleteextattr_abort:
2545 	UNIONFS_INTERNAL_DEBUG("unionfs_deleteextattr: leave (%d)\n", error);
2546 
2547 	return (error);
2548 }
2549 
2550 static int
2551 unionfs_setlabel(struct vop_setlabel_args *ap)
2552 {
2553 	struct unionfs_node *unp;
2554 	struct vnode   *uvp;
2555 	struct vnode   *lvp;
2556 	struct thread  *td;
2557 	int		error;
2558 
2559 	UNIONFS_INTERNAL_DEBUG("unionfs_setlabel: enter\n");
2560 
2561 	KASSERT_UNIONFS_VNODE(ap->a_vp);
2562 
2563 	error = EROFS;
2564 	unp = VTOUNIONFS(ap->a_vp);
2565 	uvp = unp->un_uppervp;
2566 	lvp = unp->un_lowervp;
2567 	td = ap->a_td;
2568 
2569 	if (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)
2570 		return (EROFS);
2571 
2572 	if (uvp == NULLVP && lvp->v_type == VREG) {
2573 		if ((error = unionfs_copyfile(unp, 1, ap->a_cred, td)) != 0)
2574 			return (error);
2575 		uvp = unp->un_uppervp;
2576 	}
2577 
2578 	if (uvp != NULLVP)
2579 		error = VOP_SETLABEL(uvp, ap->a_label, ap->a_cred, td);
2580 
2581 	UNIONFS_INTERNAL_DEBUG("unionfs_setlabel: leave (%d)\n", error);
2582 
2583 	return (error);
2584 }
2585 
2586 static int
2587 unionfs_vptofh(struct vop_vptofh_args *ap)
2588 {
2589 	return (EOPNOTSUPP);
2590 }
2591 
2592 static int
2593 unionfs_add_writecount(struct vop_add_writecount_args *ap)
2594 {
2595 	struct vnode *tvp, *vp;
2596 	struct unionfs_node *unp;
2597 	int error, writerefs __diagused;
2598 
2599 	vp = ap->a_vp;
2600 	unp = VTOUNIONFS(vp);
2601 	tvp = unp->un_uppervp;
2602 	KASSERT(tvp != NULL,
2603 	    ("%s: adding write ref without upper vnode", __func__));
2604 	error = VOP_ADD_WRITECOUNT(tvp, ap->a_inc);
2605 	if (error != 0)
2606 		return (error);
2607 	/*
2608 	 * We need to track the write refs we've passed to the underlying
2609 	 * vnodes so that we can undo them in case we are forcibly unmounted.
2610 	 */
2611 	writerefs = atomic_fetchadd_int(&vp->v_writecount, ap->a_inc);
2612 	/* text refs are bypassed to lowervp */
2613 	VNASSERT(writerefs >= 0, vp,
2614 	    ("%s: invalid write count %d", __func__, writerefs));
2615 	VNASSERT(writerefs + ap->a_inc >= 0, vp,
2616 	    ("%s: invalid write count inc %d + %d", __func__,
2617 	    writerefs, ap->a_inc));
2618 	return (0);
2619 }
2620 
2621 static int
2622 unionfs_vput_pair(struct vop_vput_pair_args *ap)
2623 {
2624 	struct mount *mp;
2625 	struct vnode *dvp, *vp, **vpp, *lvp, *ldvp, *uvp, *udvp, *tempvp;
2626 	struct unionfs_node *dunp, *unp;
2627 	int error, res;
2628 
2629 	dvp = ap->a_dvp;
2630 	vpp = ap->a_vpp;
2631 	vp = NULLVP;
2632 	lvp = NULLVP;
2633 	uvp = NULLVP;
2634 	unp = NULL;
2635 
2636 	dunp = VTOUNIONFS(dvp);
2637 	udvp = dunp->un_uppervp;
2638 	ldvp = dunp->un_lowervp;
2639 
2640 	/*
2641 	 * Underlying vnodes should be locked because the encompassing unionfs
2642 	 * node is locked, but will not be referenced, as the reference will
2643 	 * only be on the unionfs node.  Reference them now so that the vput()s
2644 	 * performed by VOP_VPUT_PAIR() will have a reference to drop.
2645 	 */
2646 	if (udvp != NULLVP)
2647 		vref(udvp);
2648 	if (ldvp != NULLVP)
2649 		vref(ldvp);
2650 
2651 	if (vpp != NULL)
2652 		vp = *vpp;
2653 
2654 	if (vp != NULLVP) {
2655 		unp = VTOUNIONFS(vp);
2656 		uvp = unp->un_uppervp;
2657 		lvp = unp->un_lowervp;
2658 		if (uvp != NULLVP)
2659 			vref(uvp);
2660 		if (lvp != NULLVP)
2661 			vref(lvp);
2662 
2663 		/*
2664 		 * If we're being asked to return a locked child vnode, then
2665 		 * we may need to create a replacement vnode in case the
2666 		 * original is reclaimed while the lock is dropped.  In that
2667 		 * case we'll need to ensure the mount and the underlying
2668 		 * vnodes aren't also recycled during that window.
2669 		 */
2670 		if (!ap->a_unlock_vp) {
2671 			vhold(vp);
2672 			if (uvp != NULLVP)
2673 				vhold(uvp);
2674 			if (lvp != NULLVP)
2675 				vhold(lvp);
2676 			mp = vp->v_mount;
2677 			vfs_ref(mp);
2678 		}
2679 	}
2680 
2681 	/*
2682 	 * TODO: Because unionfs_lock() locks both the lower and upper vnodes
2683 	 * (if available), we must also call VOP_VPUT_PAIR() on both the lower
2684 	 * and upper parent/child pairs.  If unionfs_lock() is reworked to lock
2685 	 * only a single vnode, this code will need to change to also only
2686 	 * operate on one vnode pair.
2687 	 */
2688 	ASSERT_VOP_LOCKED(ldvp, __func__);
2689 	ASSERT_VOP_LOCKED(udvp, __func__);
2690 	ASSERT_VOP_LOCKED(lvp, __func__);
2691 	ASSERT_VOP_LOCKED(uvp, __func__);
2692 
2693 	KASSERT(lvp == NULLVP || ldvp != NULLVP,
2694 	    ("%s: NULL ldvp with non-NULL lvp", __func__));
2695 	if (ldvp != NULLVP)
2696 		res = VOP_VPUT_PAIR(ldvp, lvp != NULLVP ? &lvp : NULL, true);
2697 	KASSERT(uvp == NULLVP || udvp != NULLVP,
2698 	    ("%s: NULL udvp with non-NULL uvp", __func__));
2699 	if (udvp != NULLVP)
2700 		res = VOP_VPUT_PAIR(udvp, uvp != NULLVP ? &uvp : NULL, true);
2701 
2702 	ASSERT_VOP_UNLOCKED(ldvp, __func__);
2703 	ASSERT_VOP_UNLOCKED(udvp, __func__);
2704 	ASSERT_VOP_UNLOCKED(lvp, __func__);
2705 	ASSERT_VOP_UNLOCKED(uvp, __func__);
2706 
2707 	/*
2708 	 * VOP_VPUT_PAIR() dropped the references we added to the underlying
2709 	 * vnodes, now drop the caller's reference to the unionfs vnodes.
2710 	 */
2711 	if (vp != NULLVP && ap->a_unlock_vp)
2712 		vrele(vp);
2713 	vrele(dvp);
2714 
2715 	if (vp == NULLVP || ap->a_unlock_vp)
2716 		return (res);
2717 
2718 	/*
2719 	 * We're being asked to return a locked vnode.  At this point, the
2720 	 * underlying vnodes have been unlocked, so vp may have been reclaimed.
2721 	 */
2722 	vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
2723 	if (vp->v_data == NULL && vfs_busy(mp, MBF_NOWAIT) == 0) {
2724 		vput(vp);
2725 		error = unionfs_nodeget(mp, uvp, lvp, dvp, &tempvp, NULL);
2726 		if (error == 0) {
2727 			vn_lock(tempvp, LK_EXCLUSIVE | LK_RETRY);
2728 			*vpp = tempvp;
2729 		} else
2730 			vget(vp, LK_EXCLUSIVE | LK_RETRY);
2731 		vfs_unbusy(mp);
2732 	}
2733 	if (lvp != NULLVP)
2734 		vdrop(lvp);
2735 	if (uvp != NULLVP)
2736 		vdrop(uvp);
2737 	vdrop(vp);
2738 	vfs_rel(mp);
2739 
2740 	return (res);
2741 }
2742 
2743 static int
2744 unionfs_set_text(struct vop_set_text_args *ap)
2745 {
2746 	struct vnode *tvp;
2747 	struct unionfs_node *unp;
2748 	int error;
2749 
2750 	/*
2751 	 * We assume text refs are managed against lvp/uvp through the
2752 	 * executable mapping backed by its VM object.  We therefore don't
2753 	 * need to track leased text refs in the case of a forcible unmount.
2754 	 */
2755 	unp = VTOUNIONFS(ap->a_vp);
2756 	ASSERT_VOP_LOCKED(ap->a_vp, __func__);
2757 	tvp = unp->un_uppervp != NULL ? unp->un_uppervp : unp->un_lowervp;
2758 	error = VOP_SET_TEXT(tvp);
2759 	return (error);
2760 }
2761 
2762 static int
2763 unionfs_unset_text(struct vop_unset_text_args *ap)
2764 {
2765 	struct vnode *tvp;
2766 	struct unionfs_node *unp;
2767 
2768 	ASSERT_VOP_LOCKED(ap->a_vp, __func__);
2769 	unp = VTOUNIONFS(ap->a_vp);
2770 	tvp = unp->un_uppervp != NULL ? unp->un_uppervp : unp->un_lowervp;
2771 	VOP_UNSET_TEXT_CHECKED(tvp);
2772 	return (0);
2773 }
2774 
2775 struct vop_vector unionfs_vnodeops = {
2776 	.vop_default =		&default_vnodeops,
2777 
2778 	.vop_access =		unionfs_access,
2779 	.vop_aclcheck =		unionfs_aclcheck,
2780 	.vop_advlock =		unionfs_advlock,
2781 	.vop_bmap =		VOP_EOPNOTSUPP,
2782 	.vop_cachedlookup =	unionfs_lookup,
2783 	.vop_close =		unionfs_close,
2784 	.vop_closeextattr =	unionfs_closeextattr,
2785 	.vop_create =		unionfs_create,
2786 	.vop_deleteextattr =	unionfs_deleteextattr,
2787 	.vop_fsync =		unionfs_fsync,
2788 	.vop_getacl =		unionfs_getacl,
2789 	.vop_getattr =		unionfs_getattr,
2790 	.vop_getextattr =	unionfs_getextattr,
2791 	.vop_getwritemount =	unionfs_getwritemount,
2792 	.vop_inactive =		unionfs_inactive,
2793 	.vop_need_inactive =	vop_stdneed_inactive,
2794 	.vop_islocked =		unionfs_islocked,
2795 	.vop_ioctl =		unionfs_ioctl,
2796 	.vop_link =		unionfs_link,
2797 	.vop_listextattr =	unionfs_listextattr,
2798 	.vop_lock1 =		unionfs_lock,
2799 	.vop_lookup =		vfs_cache_lookup,
2800 	.vop_mkdir =		unionfs_mkdir,
2801 	.vop_mknod =		unionfs_mknod,
2802 	.vop_open =		unionfs_open,
2803 	.vop_openextattr =	unionfs_openextattr,
2804 	.vop_pathconf =		unionfs_pathconf,
2805 	.vop_poll =		unionfs_poll,
2806 	.vop_print =		unionfs_print,
2807 	.vop_read =		unionfs_read,
2808 	.vop_readdir =		unionfs_readdir,
2809 	.vop_readlink =		unionfs_readlink,
2810 	.vop_reclaim =		unionfs_reclaim,
2811 	.vop_remove =		unionfs_remove,
2812 	.vop_rename =		unionfs_rename,
2813 	.vop_rmdir =		unionfs_rmdir,
2814 	.vop_setacl =		unionfs_setacl,
2815 	.vop_setattr =		unionfs_setattr,
2816 	.vop_setextattr =	unionfs_setextattr,
2817 	.vop_setlabel =		unionfs_setlabel,
2818 	.vop_strategy =		unionfs_strategy,
2819 	.vop_symlink =		unionfs_symlink,
2820 	.vop_unlock =		unionfs_unlock,
2821 	.vop_whiteout =		unionfs_whiteout,
2822 	.vop_write =		unionfs_write,
2823 	.vop_vptofh =		unionfs_vptofh,
2824 	.vop_add_writecount =	unionfs_add_writecount,
2825 	.vop_vput_pair =	unionfs_vput_pair,
2826 	.vop_set_text =		unionfs_set_text,
2827 	.vop_unset_text = 	unionfs_unset_text,
2828 };
2829 VFS_VOP_VECTOR_REGISTER(unionfs_vnodeops);
2830