xref: /freebsd/sys/fs/unionfs/union_vnops.c (revision 6e8394b8baa7d5d9153ab90de6824bcd19b3b4e1)
1 /*
2  * Copyright (c) 1992, 1993, 1994, 1995 Jan-Simon Pendry.
3  * Copyright (c) 1992, 1993, 1994, 1995
4  *	The Regents of the University of California.  All rights reserved.
5  *
6  * This code is derived from software contributed to Berkeley by
7  * Jan-Simon Pendry.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. All advertising materials mentioning features or use of this software
18  *    must display the following acknowledgement:
19  *	This product includes software developed by the University of
20  *	California, Berkeley and its contributors.
21  * 4. Neither the name of the University nor the names of its contributors
22  *    may be used to endorse or promote products derived from this software
23  *    without specific prior written permission.
24  *
25  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35  * SUCH DAMAGE.
36  *
37  *	@(#)union_vnops.c	8.32 (Berkeley) 6/23/95
38  * $Id: union_vnops.c,v 1.60 1999/01/27 22:42:08 dillon Exp $
39  */
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/proc.h>
44 #include <sys/fcntl.h>
45 #include <sys/stat.h>
46 #include <sys/kernel.h>
47 #include <sys/vnode.h>
48 #include <sys/mount.h>
49 #include <sys/namei.h>
50 #include <sys/malloc.h>
51 #include <sys/buf.h>
52 #include <sys/lock.h>
53 #include <miscfs/union/union.h>
54 
55 #define FIXUP(un, p) { \
56 	if (((un)->un_flags & UN_ULOCK) == 0) { \
57 		union_fixup(un, p); \
58 	} \
59 }
60 
61 static int	union_abortop __P((struct vop_abortop_args *ap));
62 static int	union_access __P((struct vop_access_args *ap));
63 static int	union_advlock __P((struct vop_advlock_args *ap));
64 static int	union_bmap __P((struct vop_bmap_args *ap));
65 static int	union_close __P((struct vop_close_args *ap));
66 static int	union_create __P((struct vop_create_args *ap));
67 static void	union_fixup __P((struct union_node *un, struct proc *p));
68 static int	union_fsync __P((struct vop_fsync_args *ap));
69 static int	union_getattr __P((struct vop_getattr_args *ap));
70 static int	union_inactive __P((struct vop_inactive_args *ap));
71 static int	union_ioctl __P((struct vop_ioctl_args *ap));
72 static int	union_islocked __P((struct vop_islocked_args *ap));
73 static int	union_lease __P((struct vop_lease_args *ap));
74 static int	union_link __P((struct vop_link_args *ap));
75 static int	union_lock __P((struct vop_lock_args *ap));
76 static int	union_lookup __P((struct vop_lookup_args *ap));
77 static int	union_lookup1 __P((struct vnode *udvp, struct vnode **dvpp,
78 				   struct vnode **vpp,
79 				   struct componentname *cnp));
80 static int	union_mkdir __P((struct vop_mkdir_args *ap));
81 static int	union_mknod __P((struct vop_mknod_args *ap));
82 static int	union_mmap __P((struct vop_mmap_args *ap));
83 static int	union_open __P((struct vop_open_args *ap));
84 static int	union_pathconf __P((struct vop_pathconf_args *ap));
85 static int	union_print __P((struct vop_print_args *ap));
86 static int	union_read __P((struct vop_read_args *ap));
87 static int	union_readdir __P((struct vop_readdir_args *ap));
88 static int	union_readlink __P((struct vop_readlink_args *ap));
89 static int	union_reclaim __P((struct vop_reclaim_args *ap));
90 static int	union_remove __P((struct vop_remove_args *ap));
91 static int	union_rename __P((struct vop_rename_args *ap));
92 static int	union_revoke __P((struct vop_revoke_args *ap));
93 static int	union_rmdir __P((struct vop_rmdir_args *ap));
94 static int	union_poll __P((struct vop_poll_args *ap));
95 static int	union_setattr __P((struct vop_setattr_args *ap));
96 static int	union_strategy __P((struct vop_strategy_args *ap));
97 static int	union_symlink __P((struct vop_symlink_args *ap));
98 static int	union_unlock __P((struct vop_unlock_args *ap));
99 static int	union_whiteout __P((struct vop_whiteout_args *ap));
100 static int	union_write __P((struct vop_read_args *ap));
101 
102 static void
103 union_fixup(un, p)
104 	struct union_node *un;
105 	struct proc *p;
106 {
107 
108 	vn_lock(un->un_uppervp, LK_EXCLUSIVE | LK_RETRY, p);
109 	un->un_flags |= UN_ULOCK;
110 }
111 
112 static int
113 union_lookup1(udvp, dvpp, vpp, cnp)
114 	struct vnode *udvp;
115 	struct vnode **dvpp;
116 	struct vnode **vpp;
117 	struct componentname *cnp;
118 {
119 	int error;
120 	struct proc *p = cnp->cn_proc;
121 	struct vnode *tdvp;
122 	struct vnode *dvp;
123 	struct mount *mp;
124 
125 	dvp = *dvpp;
126 
127 	/*
128 	 * If stepping up the directory tree, check for going
129 	 * back across the mount point, in which case do what
130 	 * lookup would do by stepping back down the mount
131 	 * hierarchy.
132 	 */
133 	if (cnp->cn_flags & ISDOTDOT) {
134 		while ((dvp != udvp) && (dvp->v_flag & VROOT)) {
135 			/*
136 			 * Don't do the NOCROSSMOUNT check
137 			 * at this level.  By definition,
138 			 * union fs deals with namespaces, not
139 			 * filesystems.
140 			 */
141 			tdvp = dvp;
142 			*dvpp = dvp = dvp->v_mount->mnt_vnodecovered;
143 			vput(tdvp);
144 			VREF(dvp);
145 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, p);
146 		}
147 	}
148 
149         error = VOP_LOOKUP(dvp, &tdvp, cnp);
150 	if (error)
151 		return (error);
152 
153 	/*
154 	 * The parent directory will have been unlocked, unless lookup
155 	 * found the last component.  In which case, re-lock the node
156 	 * here to allow it to be unlocked again (phew) in union_lookup.
157 	 */
158 	if (dvp != tdvp && !(cnp->cn_flags & ISLASTCN))
159 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, p);
160 
161 	dvp = tdvp;
162 
163 	/*
164 	 * Lastly check if the current node is a mount point in
165 	 * which case walk up the mount hierarchy making sure not to
166 	 * bump into the root of the mount tree (ie. dvp != udvp).
167 	 */
168 	while (dvp != udvp && (dvp->v_type == VDIR) &&
169 	       (mp = dvp->v_mountedhere)) {
170 
171 		if (vfs_busy(mp, 0, 0, p))
172 			continue;
173 
174 		error = VFS_ROOT(mp, &tdvp);
175 		vfs_unbusy(mp, p);
176 		if (error) {
177 			vput(dvp);
178 			return (error);
179 		}
180 
181 		vput(dvp);
182 		dvp = tdvp;
183 	}
184 
185 	*vpp = dvp;
186 	return (0);
187 }
188 
189 static int
190 union_lookup(ap)
191 	struct vop_lookup_args /* {
192 		struct vnodeop_desc *a_desc;
193 		struct vnode *a_dvp;
194 		struct vnode **a_vpp;
195 		struct componentname *a_cnp;
196 	} */ *ap;
197 {
198 	int error;
199 	int uerror, lerror;
200 	struct vnode *uppervp, *lowervp;
201 	struct vnode *upperdvp, *lowerdvp;
202 	struct vnode *dvp = ap->a_dvp;
203 	struct union_node *dun = VTOUNION(dvp);
204 	struct componentname *cnp = ap->a_cnp;
205 	struct proc *p = cnp->cn_proc;
206 	int lockparent = cnp->cn_flags & LOCKPARENT;
207 	struct union_mount *um = MOUNTTOUNIONMOUNT(dvp->v_mount);
208 	struct ucred *saved_cred = NULL;
209 	int iswhiteout;
210 	struct vattr va;
211 
212 
213 	/*
214 	 * Disallow write attemps to the filesystem mounted read-only.
215 	 */
216 	if ((cnp->cn_flags & ISLASTCN) && (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
217 		(cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
218 		return (EROFS);
219 
220 #ifdef notyet
221 	if (cnp->cn_namelen == 3 &&
222 			cnp->cn_nameptr[2] == '.' &&
223 			cnp->cn_nameptr[1] == '.' &&
224 			cnp->cn_nameptr[0] == '.') {
225 		dvp = *ap->a_vpp = LOWERVP(ap->a_dvp);
226 		if (dvp == NULLVP)
227 			return (ENOENT);
228 		VREF(dvp);
229 		vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, p);
230 		if (!lockparent || !(cnp->cn_flags & ISLASTCN))
231 			VOP_UNLOCK(ap->a_dvp, 0, p);
232 		return (0);
233 	}
234 #endif
235 
236 	cnp->cn_flags |= LOCKPARENT;
237 
238 	upperdvp = dun->un_uppervp;
239 	lowerdvp = dun->un_lowervp;
240 	uppervp = NULLVP;
241 	lowervp = NULLVP;
242 	iswhiteout = 0;
243 
244 	if (cnp->cn_flags & ISDOTDOT) {
245 		if (upperdvp != NULL)
246 			VREF(upperdvp);
247 		if (lowerdvp != NULL)
248 			VREF(lowerdvp);
249 	}
250 
251 	/*
252 	 * do the lookup in the upper level.
253 	 * if that level comsumes additional pathnames,
254 	 * then assume that something special is going
255 	 * on and just return that vnode.
256 	 */
257 	if (upperdvp != NULLVP) {
258 		FIXUP(dun, p);
259 		/*
260 		 * If we're doing `..' in the underlying filesystem,
261 		 * we must drop our lock on the union node before
262 		 * going up the tree in the lower file system--if we block
263 		 * on the lowervp lock, and that's held by someone else
264 		 * coming down the tree and who's waiting for our lock,
265 		 * we would be hosed.
266 		 */
267 		if (cnp->cn_flags & ISDOTDOT) {
268 			/* retain lock on underlying VP: */
269 			dun->un_flags |= UN_KLOCK;
270 			VOP_UNLOCK(dvp, 0, p);
271 		}
272 		uerror = union_lookup1(um->um_uppervp, &upperdvp,
273 					&uppervp, cnp);
274 		/*
275 		 * Disallow write attemps to the filesystem mounted read-only.
276 		 */
277 		if (uerror == EJUSTRETURN && (cnp->cn_flags & ISLASTCN) &&
278 			(dvp->v_mount->mnt_flag & MNT_RDONLY) &&
279 			(cnp->cn_nameiop == CREATE || cnp->cn_nameiop == RENAME)) {
280 			if (!lockparent)
281 				cnp->cn_flags &= ~LOCKPARENT;
282 			return (EROFS);
283 		}
284 
285 		if (cnp->cn_flags & ISDOTDOT) {
286 			if (dun->un_uppervp == upperdvp) {
287 				/*
288 				 * We got the underlying bugger back locked...
289 				 * now take back the union node lock.  Since we
290 				 * hold the uppervp lock, we can diddle union
291 				 * locking flags at will. :)
292 				 */
293 				dun->un_flags |= UN_ULOCK;
294 			}
295 			/*
296 			 * If upperdvp got swapped out, it means we did
297 			 * some mount point magic, and we do not have
298 			 * dun->un_uppervp locked currently--so we get it
299 			 * locked here (don't set the UN_ULOCK flag).
300 			 */
301 			vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY, p);
302 		}
303 
304 		/*if (uppervp == upperdvp)
305 			dun->un_flags |= UN_KLOCK;*/
306 
307 		if (cnp->cn_consume != 0) {
308 			*ap->a_vpp = uppervp;
309 			if (!lockparent)
310 				cnp->cn_flags &= ~LOCKPARENT;
311 			error = uerror;
312 			goto out;
313 		}
314 		if (uerror == ENOENT || uerror == EJUSTRETURN) {
315 			if (cnp->cn_flags & ISWHITEOUT) {
316 				iswhiteout = 1;
317 			} else if (lowerdvp != NULLVP) {
318 				lerror = VOP_GETATTR(upperdvp, &va,
319 					cnp->cn_cred, cnp->cn_proc);
320 				if (lerror == 0 && (va.va_flags & OPAQUE))
321 					iswhiteout = 1;
322 			}
323 		}
324 	} else {
325 		uerror = ENOENT;
326 	}
327 
328 	/*
329 	 * in a similar way to the upper layer, do the lookup
330 	 * in the lower layer.   this time, if there is some
331 	 * component magic going on, then vput whatever we got
332 	 * back from the upper layer and return the lower vnode
333 	 * instead.
334 	 */
335 	if (lowerdvp != NULLVP && !iswhiteout) {
336 		int nameiop;
337 
338 		vn_lock(lowerdvp, LK_EXCLUSIVE | LK_RETRY, p);
339 
340 		/*
341 		 * Only do a LOOKUP on the bottom node, since
342 		 * we won't be making changes to it anyway.
343 		 */
344 		nameiop = cnp->cn_nameiop;
345 		cnp->cn_nameiop = LOOKUP;
346 		if (um->um_op == UNMNT_BELOW) {
347 			saved_cred = cnp->cn_cred;
348 			cnp->cn_cred = um->um_cred;
349 		}
350 		/*
351 		 * We shouldn't have to worry about locking interactions
352 		 * between the lower layer and our union layer (w.r.t.
353 		 * `..' processing) because we don't futz with lowervp
354 		 * locks in the union-node instantiation code path.
355 		 */
356 		lerror = union_lookup1(um->um_lowervp, &lowerdvp,
357 				&lowervp, cnp);
358 		if (um->um_op == UNMNT_BELOW)
359 			cnp->cn_cred = saved_cred;
360 		cnp->cn_nameiop = nameiop;
361 
362 		if (lowervp != lowerdvp)
363 			VOP_UNLOCK(lowerdvp, 0, p);
364 
365 		if (cnp->cn_consume != 0 || lerror == EACCES) {
366 			if (lerror == EACCES)
367 				lowervp = NULLVP;
368 			if (uppervp != NULLVP) {
369 				if (uppervp == upperdvp)
370 					vrele(uppervp);
371 				else
372 					vput(uppervp);
373 				uppervp = NULLVP;
374 			}
375 			*ap->a_vpp = lowervp;
376 			if (!lockparent)
377 				cnp->cn_flags &= ~LOCKPARENT;
378 			error = lerror;
379 			goto out;
380 		}
381 	} else {
382 		lerror = ENOENT;
383 		if ((cnp->cn_flags & ISDOTDOT) && dun->un_pvp != NULLVP) {
384 			lowervp = LOWERVP(dun->un_pvp);
385 			if (lowervp != NULLVP) {
386 				VREF(lowervp);
387 				vn_lock(lowervp, LK_EXCLUSIVE | LK_RETRY, p);
388 				lerror = 0;
389 			}
390 		}
391 	}
392 
393 	if (!lockparent)
394 		cnp->cn_flags &= ~LOCKPARENT;
395 
396 	/*
397 	 * at this point, we have uerror and lerror indicating
398 	 * possible errors with the lookups in the upper and lower
399 	 * layers.  additionally, uppervp and lowervp are (locked)
400 	 * references to existing vnodes in the upper and lower layers.
401 	 *
402 	 * there are now three cases to consider.
403 	 * 1. if both layers returned an error, then return whatever
404 	 *    error the upper layer generated.
405 	 *
406 	 * 2. if the top layer failed and the bottom layer succeeded
407 	 *    then two subcases occur.
408 	 *    a.  the bottom vnode is not a directory, in which
409 	 *	  case just return a new union vnode referencing
410 	 *	  an empty top layer and the existing bottom layer.
411 	 *    b.  the bottom vnode is a directory, in which case
412 	 *	  create a new directory in the top-level and
413 	 *	  continue as in case 3.
414 	 *
415 	 * 3. if the top layer succeeded then return a new union
416 	 *    vnode referencing whatever the new top layer and
417 	 *    whatever the bottom layer returned.
418 	 */
419 
420 	*ap->a_vpp = NULLVP;
421 
422 	/* case 1. */
423 	if ((uerror != 0) && (lerror != 0)) {
424 		error = uerror;
425 		goto out;
426 	}
427 
428 	/* case 2. */
429 	if (uerror != 0 /* && (lerror == 0) */ ) {
430 		if (lowervp->v_type == VDIR) { /* case 2b. */
431 			dun->un_flags &= ~UN_ULOCK;
432 			VOP_UNLOCK(upperdvp, 0, p);
433 			uerror = union_mkshadow(um, upperdvp, cnp, &uppervp);
434 			vn_lock(upperdvp, LK_EXCLUSIVE | LK_RETRY, p);
435 			dun->un_flags |= UN_ULOCK;
436 
437 			if (uerror) {
438 				if (lowervp != NULLVP) {
439 					vput(lowervp);
440 					lowervp = NULLVP;
441 				}
442 				error = uerror;
443 				goto out;
444 			}
445 		}
446 	}
447 
448 	if (lowervp != NULLVP)
449 		VOP_UNLOCK(lowervp, 0, p);
450 
451 	error = union_allocvp(ap->a_vpp, dvp->v_mount, dvp, upperdvp, cnp,
452 			      uppervp, lowervp, 1);
453 
454 	if (error) {
455 		if (uppervp != NULLVP)
456 			vput(uppervp);
457 		if (lowervp != NULLVP)
458 			vrele(lowervp);
459 	} else {
460 		if (*ap->a_vpp != dvp)
461 			if (!lockparent || !(cnp->cn_flags & ISLASTCN))
462 				VOP_UNLOCK(dvp, 0, p);
463 #ifdef DIAGNOSTIC
464         if (cnp->cn_namelen == 1 &&
465             cnp->cn_nameptr[0] == '.' &&
466             *ap->a_vpp != dvp) {
467             panic("union_lookup returning . (%p) not same as startdir (%p)",
468 				  ap->a_vpp, dvp);
469 		}
470 #endif
471 	}
472 
473 out:
474     if (cnp->cn_flags & ISDOTDOT) {
475         if (upperdvp != NULL)
476             vrele(upperdvp);
477         if (lowerdvp != NULL)
478             vrele(lowerdvp);
479     }
480 
481 	return (error);
482 }
483 
484 static int
485 union_create(ap)
486 	struct vop_create_args /* {
487 		struct vnode *a_dvp;
488 		struct vnode **a_vpp;
489 		struct componentname *a_cnp;
490 		struct vattr *a_vap;
491 	} */ *ap;
492 {
493 	struct union_node *dun = VTOUNION(ap->a_dvp);
494 	struct vnode *dvp = dun->un_uppervp;
495 	struct componentname *cnp = ap->a_cnp;
496 	struct proc *p = cnp->cn_proc;
497 
498 	if (dvp != NULLVP) {
499 		struct vnode *vp;
500 		struct mount *mp;
501 		int error;
502 
503 		FIXUP(dun, p);
504 
505 		dun->un_flags |= UN_KLOCK;
506 		VOP_UNLOCK(ap->a_dvp, 0, p);
507 		error = VOP_CREATE(dvp, &vp, cnp, ap->a_vap);
508 		if (error) {
509 			dun->un_flags |= UN_ULOCK;
510 			return (error);
511 		}
512 
513 		mp = ap->a_dvp->v_mount;
514 		VOP_UNLOCK(dvp, 0, p);
515 		error = union_allocvp(ap->a_vpp, mp, NULLVP, NULLVP, cnp, vp,
516 				NULLVP, 1);
517 		if (error)
518 			vput(vp);
519 		vn_lock(ap->a_dvp, LK_EXCLUSIVE| LK_RETRY, p);
520 		return (error);
521 	}
522 
523 	return (EROFS);
524 }
525 
526 static int
527 union_whiteout(ap)
528 	struct vop_whiteout_args /* {
529 		struct vnode *a_dvp;
530 		struct componentname *a_cnp;
531 		int a_flags;
532 	} */ *ap;
533 {
534 	struct union_node *un = VTOUNION(ap->a_dvp);
535 	struct componentname *cnp = ap->a_cnp;
536 	struct proc *p = cnp->cn_proc;
537 
538 	if (un->un_uppervp == NULLVP)
539 		return (EOPNOTSUPP);
540 
541 	FIXUP(un, p);
542 	return (VOP_WHITEOUT(un->un_uppervp, cnp, ap->a_flags));
543 }
544 
545 static int
546 union_mknod(ap)
547 	struct vop_mknod_args /* {
548 		struct vnode *a_dvp;
549 		struct vnode **a_vpp;
550 		struct componentname *a_cnp;
551 		struct vattr *a_vap;
552 	} */ *ap;
553 {
554 	struct union_node *dun = VTOUNION(ap->a_dvp);
555 	struct vnode *dvp = dun->un_uppervp;
556 	struct componentname *cnp = ap->a_cnp;
557 	struct proc *p = cnp->cn_proc;
558 
559 	if (dvp != NULLVP) {
560 		struct vnode *vp;
561 		struct mount *mp;
562 		int error;
563 
564 		FIXUP(dun, p);
565 
566 		dun->un_flags |= UN_KLOCK;
567 		VOP_UNLOCK(ap->a_dvp, 0, p);
568 		error = VOP_MKNOD(dvp, &vp, cnp, ap->a_vap);
569 		if (error) {
570 			dun->un_flags |= UN_ULOCK;
571 			return (error);
572 		}
573 
574 		if (vp != NULLVP) {
575 			mp = ap->a_dvp->v_mount;
576 			VOP_UNLOCK(dvp, 0, p);
577 			error = union_allocvp(ap->a_vpp, mp, NULLVP, NULLVP,
578 					cnp, vp, NULLVP, 1);
579 			if (error)
580 				vput(vp);
581 			vn_lock(ap->a_dvp, LK_EXCLUSIVE| LK_RETRY, p);
582 		} else {
583 			dun->un_flags |= UN_ULOCK;
584 		}
585 		return (error);
586 	}
587 
588 	return (EROFS);
589 }
590 
591 static int
592 union_open(ap)
593 	struct vop_open_args /* {
594 		struct vnodeop_desc *a_desc;
595 		struct vnode *a_vp;
596 		int a_mode;
597 		struct ucred *a_cred;
598 		struct proc *a_p;
599 	} */ *ap;
600 {
601 	struct union_node *un = VTOUNION(ap->a_vp);
602 	struct vnode *tvp;
603 	int mode = ap->a_mode;
604 	struct ucred *cred = ap->a_cred;
605 	struct proc *p = ap->a_p;
606 	int error;
607 
608 	/*
609 	 * If there is an existing upper vp then simply open that.
610 	 */
611 	tvp = un->un_uppervp;
612 	if (tvp == NULLVP) {
613 		/*
614 		 * If the lower vnode is being opened for writing, then
615 		 * copy the file contents to the upper vnode and open that,
616 		 * otherwise can simply open the lower vnode.
617 		 */
618 		tvp = un->un_lowervp;
619 		if ((ap->a_mode & FWRITE) && (tvp->v_type == VREG)) {
620 			error = union_copyup(un, (mode&O_TRUNC) == 0, cred, p);
621 			if (error == 0)
622 				error = VOP_OPEN(un->un_uppervp, mode, cred, p);
623 			return (error);
624 		}
625 
626 		/*
627 		 * Just open the lower vnode
628 		 */
629 		un->un_openl++;
630 		vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY, p);
631 		error = VOP_OPEN(tvp, mode, cred, p);
632 		VOP_UNLOCK(tvp, 0, p);
633 
634 		return (error);
635 	}
636 
637 	FIXUP(un, p);
638 
639 	error = VOP_OPEN(tvp, mode, cred, p);
640 
641 	return (error);
642 }
643 
644 static int
645 union_close(ap)
646 	struct vop_close_args /* {
647 		struct vnode *a_vp;
648 		int  a_fflag;
649 		struct ucred *a_cred;
650 		struct proc *a_p;
651 	} */ *ap;
652 {
653 	struct union_node *un = VTOUNION(ap->a_vp);
654 	struct vnode *vp;
655 
656 	if ((vp = un->un_uppervp) == NULLVP) {
657 #ifdef UNION_DIAGNOSTIC
658 		if (un->un_openl <= 0)
659 			panic("union: un_openl cnt");
660 #endif
661 		--un->un_openl;
662 		vp = un->un_lowervp;
663 	}
664 
665 	ap->a_vp = vp;
666 	return (VCALL(vp, VOFFSET(vop_close), ap));
667 }
668 
669 /*
670  * Check access permission on the union vnode.
671  * The access check being enforced is to check
672  * against both the underlying vnode, and any
673  * copied vnode.  This ensures that no additional
674  * file permissions are given away simply because
675  * the user caused an implicit file copy.
676  */
677 static int
678 union_access(ap)
679 	struct vop_access_args /* {
680 		struct vnodeop_desc *a_desc;
681 		struct vnode *a_vp;
682 		int a_mode;
683 		struct ucred *a_cred;
684 		struct proc *a_p;
685 	} */ *ap;
686 {
687 	struct union_node *un = VTOUNION(ap->a_vp);
688 	struct proc *p = ap->a_p;
689 	int error = EACCES;
690 	struct vnode *vp;
691 	struct vnode *savedvp;
692 
693 	/*
694 	 * Disallow write attempts on filesystems mounted read-only.
695 	 */
696 	if (ap->a_mode & VWRITE && (ap->a_vp->v_mount->mnt_flag & MNT_RDONLY)) {
697 		switch (ap->a_vp->v_type) {
698 		case VREG:
699 		case VDIR:
700 		case VLNK:
701 			return (EROFS);
702 		default:
703 			break;
704 		}
705 	}
706 	if ((vp = un->un_uppervp) != NULLVP) {
707 		FIXUP(un, p);
708 		ap->a_vp = vp;
709 		return (VCALL(vp, VOFFSET(vop_access), ap));
710 	}
711 
712 	if ((vp = un->un_lowervp) != NULLVP) {
713 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
714 		savedvp = ap->a_vp;
715 		ap->a_vp = vp;
716 		error = VCALL(vp, VOFFSET(vop_access), ap);
717 		if (error == 0) {
718 			struct union_mount *um = MOUNTTOUNIONMOUNT(savedvp->v_mount);
719 
720 			if (um->um_op == UNMNT_BELOW) {
721 				ap->a_cred = um->um_cred;
722 				error = VCALL(vp, VOFFSET(vop_access), ap);
723 			}
724 		}
725 		VOP_UNLOCK(vp, 0, p);
726 		if (error)
727 			return (error);
728 	}
729 
730 	return (error);
731 }
732 
733 /*
734  * We handle getattr only to change the fsid and
735  * track object sizes
736  */
737 static int
738 union_getattr(ap)
739 	struct vop_getattr_args /* {
740 		struct vnode *a_vp;
741 		struct vattr *a_vap;
742 		struct ucred *a_cred;
743 		struct proc *a_p;
744 	} */ *ap;
745 {
746 	int error;
747 	struct union_node *un = VTOUNION(ap->a_vp);
748 	struct vnode *vp = un->un_uppervp;
749 	struct proc *p = ap->a_p;
750 	struct vattr *vap;
751 	struct vattr va;
752 
753 
754 	/*
755 	 * Some programs walk the filesystem hierarchy by counting
756 	 * links to directories to avoid stat'ing all the time.
757 	 * This means the link count on directories needs to be "correct".
758 	 * The only way to do that is to call getattr on both layers
759 	 * and fix up the link count.  The link count will not necessarily
760 	 * be accurate but will be large enough to defeat the tree walkers.
761 	 */
762 
763 	vap = ap->a_vap;
764 
765 	vp = un->un_uppervp;
766 	if (vp != NULLVP) {
767 		/*
768 		 * It's not clear whether VOP_GETATTR is to be
769 		 * called with the vnode locked or not.  stat() calls
770 		 * it with (vp) locked, and fstat calls it with
771 		 * (vp) unlocked.
772 		 * In the mean time, compensate here by checking
773 		 * the union_node's lock flag.
774 		 */
775 		if (un->un_flags & UN_LOCKED)
776 			FIXUP(un, p);
777 
778 		error = VOP_GETATTR(vp, vap, ap->a_cred, ap->a_p);
779 		if (error)
780 			return (error);
781 		union_newsize(ap->a_vp, vap->va_size, VNOVAL);
782 	}
783 
784 	if (vp == NULLVP) {
785 		vp = un->un_lowervp;
786 	} else if (vp->v_type == VDIR && un->un_lowervp != NULLVP) {
787 		vp = un->un_lowervp;
788 		vap = &va;
789 	} else {
790 		vp = NULLVP;
791 	}
792 
793 	if (vp != NULLVP) {
794 		error = VOP_GETATTR(vp, vap, ap->a_cred, ap->a_p);
795 		if (error)
796 			return (error);
797 		union_newsize(ap->a_vp, VNOVAL, vap->va_size);
798 	}
799 
800 	if ((vap != ap->a_vap) && (vap->va_type == VDIR))
801 		ap->a_vap->va_nlink += vap->va_nlink;
802 
803 	ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
804 	return (0);
805 }
806 
807 static int
808 union_setattr(ap)
809 	struct vop_setattr_args /* {
810 		struct vnode *a_vp;
811 		struct vattr *a_vap;
812 		struct ucred *a_cred;
813 		struct proc *a_p;
814 	} */ *ap;
815 {
816 	struct union_node *un = VTOUNION(ap->a_vp);
817 	struct proc *p = ap->a_p;
818 	struct vattr *vap = ap->a_vap;
819 	int error;
820 
821 	/*
822 	 * Disallow write attempts on filesystems mounted read-only.
823 	 */
824 	if ((ap->a_vp->v_mount->mnt_flag & MNT_RDONLY) &&
825 		(vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
826 		 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
827 		 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL))
828 		return (EROFS);
829 
830 	/*
831 	 * Handle case of truncating lower object to zero size,
832 	 * by creating a zero length upper object.  This is to
833 	 * handle the case of open with O_TRUNC and O_CREAT.
834 	 */
835 	if ((un->un_uppervp == NULLVP) &&
836 	    /* assert(un->un_lowervp != NULLVP) */
837 	    (un->un_lowervp->v_type == VREG)) {
838 		error = union_copyup(un, (ap->a_vap->va_size != 0),
839 						ap->a_cred, ap->a_p);
840 		if (error)
841 			return (error);
842 	}
843 
844 	/*
845 	 * Try to set attributes in upper layer,
846 	 * otherwise return read-only filesystem error.
847 	 */
848 	if (un->un_uppervp != NULLVP) {
849 		FIXUP(un, p);
850 		error = VOP_SETATTR(un->un_uppervp, ap->a_vap,
851 					ap->a_cred, ap->a_p);
852 		if ((error == 0) && (ap->a_vap->va_size != VNOVAL))
853 			union_newsize(ap->a_vp, ap->a_vap->va_size, VNOVAL);
854 	} else {
855 		error = EROFS;
856 	}
857 
858 	return (error);
859 }
860 
861 static int
862 union_read(ap)
863 	struct vop_read_args /* {
864 		struct vnode *a_vp;
865 		struct uio *a_uio;
866 		int  a_ioflag;
867 		struct ucred *a_cred;
868 	} */ *ap;
869 {
870 	int error;
871 	struct proc *p = ap->a_uio->uio_procp;
872 	struct vnode *vp = OTHERVP(ap->a_vp);
873 	int dolock = (vp == LOWERVP(ap->a_vp));
874 
875 	if (dolock)
876 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
877 	else
878 		FIXUP(VTOUNION(ap->a_vp), p);
879 	error = VOP_READ(vp, ap->a_uio, ap->a_ioflag, ap->a_cred);
880 	if (dolock)
881 		VOP_UNLOCK(vp, 0, p);
882 
883 	/*
884 	 * XXX
885 	 * perhaps the size of the underlying object has changed under
886 	 * our feet.  take advantage of the offset information present
887 	 * in the uio structure.
888 	 */
889 	if (error == 0) {
890 		struct union_node *un = VTOUNION(ap->a_vp);
891 		off_t cur = ap->a_uio->uio_offset;
892 
893 		if (vp == un->un_uppervp) {
894 			if (cur > un->un_uppersz)
895 				union_newsize(ap->a_vp, cur, VNOVAL);
896 		} else {
897 			if (cur > un->un_lowersz)
898 				union_newsize(ap->a_vp, VNOVAL, cur);
899 		}
900 	}
901 
902 	return (error);
903 }
904 
905 static int
906 union_write(ap)
907 	struct vop_read_args /* {
908 		struct vnode *a_vp;
909 		struct uio *a_uio;
910 		int  a_ioflag;
911 		struct ucred *a_cred;
912 	} */ *ap;
913 {
914 	int error;
915 	struct vnode *vp;
916 	struct union_node *un = VTOUNION(ap->a_vp);
917 	struct proc *p = ap->a_uio->uio_procp;
918 
919 	vp = UPPERVP(ap->a_vp);
920 	if (vp == NULLVP)
921 		panic("union: missing upper layer in write");
922 
923 	FIXUP(un, p);
924 	error = VOP_WRITE(vp, ap->a_uio, ap->a_ioflag, ap->a_cred);
925 
926 	/*
927 	 * the size of the underlying object may be changed by the
928 	 * write.
929 	 */
930 	if (error == 0) {
931 		off_t cur = ap->a_uio->uio_offset;
932 
933 		if (cur > un->un_uppersz)
934 			union_newsize(ap->a_vp, cur, VNOVAL);
935 	}
936 
937 	return (error);
938 }
939 
940 static int
941 union_lease(ap)
942 	struct vop_lease_args /* {
943 		struct vnode *a_vp;
944 		struct proc *a_p;
945 		struct ucred *a_cred;
946 		int a_flag;
947 	} */ *ap;
948 {
949 	register struct vnode *ovp = OTHERVP(ap->a_vp);
950 
951 	ap->a_vp = ovp;
952 	return (VCALL(ovp, VOFFSET(vop_lease), ap));
953 }
954 
955 static int
956 union_ioctl(ap)
957 	struct vop_ioctl_args /* {
958 		struct vnode *a_vp;
959 		int  a_command;
960 		caddr_t  a_data;
961 		int  a_fflag;
962 		struct ucred *a_cred;
963 		struct proc *a_p;
964 	} */ *ap;
965 {
966 	register struct vnode *ovp = OTHERVP(ap->a_vp);
967 
968 	ap->a_vp = ovp;
969 	return (VCALL(ovp, VOFFSET(vop_ioctl), ap));
970 }
971 
972 static int
973 union_poll(ap)
974 	struct vop_poll_args /* {
975 		struct vnode *a_vp;
976 		int  a_events;
977 		struct ucred *a_cred;
978 		struct proc *a_p;
979 	} */ *ap;
980 {
981 	register struct vnode *ovp = OTHERVP(ap->a_vp);
982 
983 	ap->a_vp = ovp;
984 	return (VCALL(ovp, VOFFSET(vop_poll), ap));
985 }
986 
987 static int
988 union_revoke(ap)
989 	struct vop_revoke_args /* {
990 		struct vnode *a_vp;
991 		int a_flags;
992 		struct proc *a_p;
993 	} */ *ap;
994 {
995 	struct vnode *vp = ap->a_vp;
996 
997 	if (UPPERVP(vp))
998 		VOP_REVOKE(UPPERVP(vp), ap->a_flags);
999 	if (LOWERVP(vp))
1000 		VOP_REVOKE(LOWERVP(vp), ap->a_flags);
1001 	vgone(vp);
1002 	return (0);
1003 }
1004 
1005 static int
1006 union_mmap(ap)
1007 	struct vop_mmap_args /* {
1008 		struct vnode *a_vp;
1009 		int  a_fflags;
1010 		struct ucred *a_cred;
1011 		struct proc *a_p;
1012 	} */ *ap;
1013 {
1014 	register struct vnode *ovp = OTHERVP(ap->a_vp);
1015 
1016 	ap->a_vp = ovp;
1017 	return (VCALL(ovp, VOFFSET(vop_mmap), ap));
1018 }
1019 
1020 static int
1021 union_fsync(ap)
1022 	struct vop_fsync_args /* {
1023 		struct vnode *a_vp;
1024 		struct ucred *a_cred;
1025 		int  a_waitfor;
1026 		struct proc *a_p;
1027 	} */ *ap;
1028 {
1029 	int error = 0;
1030 	struct proc *p = ap->a_p;
1031 	struct vnode *targetvp = OTHERVP(ap->a_vp);
1032 	struct union_node *un;
1033 
1034 	if (targetvp != NULLVP) {
1035 		int dolock = (targetvp == LOWERVP(ap->a_vp));
1036 
1037 		un = VTOUNION(ap->a_vp);
1038 		if (dolock)
1039 			vn_lock(targetvp, LK_EXCLUSIVE | LK_RETRY, p);
1040 		else  {
1041 			un = VTOUNION(ap->a_vp);
1042 			if ((un->un_flags & UN_ULOCK) == 0 &&
1043 			    targetvp->v_data != NULL &&
1044 			    ((struct lock *)targetvp->v_data)->lk_lockholder
1045 			        == curproc->p_pid &&
1046 			    VOP_ISLOCKED(targetvp) != 0)
1047 				return 0;   /* XXX */
1048 
1049 			FIXUP(un, p);
1050 		}
1051 
1052 		error = VOP_FSYNC(targetvp, ap->a_cred, ap->a_waitfor, p);
1053 		if (dolock)
1054 			VOP_UNLOCK(targetvp, 0, p);
1055 	}
1056 
1057 	return (error);
1058 }
1059 
1060 static int
1061 union_remove(ap)
1062 	struct vop_remove_args /* {
1063 		struct vnode *a_dvp;
1064 		struct vnode *a_vp;
1065 		struct componentname *a_cnp;
1066 	} */ *ap;
1067 {
1068 	struct union_node *dun = VTOUNION(ap->a_dvp);
1069 	struct union_node *un = VTOUNION(ap->a_vp);
1070 	struct componentname *cnp = ap->a_cnp;
1071 	struct proc *p = cnp->cn_proc;
1072 	int error;
1073 
1074 	if (dun->un_uppervp == NULLVP)
1075 		panic("union remove: null upper vnode");
1076 
1077 	if (un->un_uppervp != NULLVP) {
1078 		struct vnode *dvp = dun->un_uppervp;
1079 		struct vnode *vp = un->un_uppervp;
1080 
1081 		FIXUP(dun, p);
1082 		dun->un_flags |= UN_KLOCK;
1083 		VOP_UNLOCK(ap->a_dvp, 0, p);
1084 		FIXUP(un, p);
1085 		un->un_flags |= UN_KLOCK;
1086 		VOP_UNLOCK(ap->a_vp, 0, p);
1087 
1088 		if (union_dowhiteout(un, cnp->cn_cred, p))
1089 			cnp->cn_flags |= DOWHITEOUT;
1090 		error = VOP_REMOVE(dvp, vp, cnp);
1091 #if 0
1092 		/* XXX */
1093 		if (!error)
1094 			union_removed_upper(un);
1095 #endif
1096 		dun->un_flags |= UN_ULOCK;
1097 		un->un_flags |= UN_ULOCK;
1098 	} else {
1099 		FIXUP(dun, p);
1100 		error = union_mkwhiteout(
1101 			MOUNTTOUNIONMOUNT(UNIONTOV(dun)->v_mount),
1102 			dun->un_uppervp, ap->a_cnp, un->un_path);
1103 	}
1104 
1105 	return (error);
1106 }
1107 
1108 static int
1109 union_link(ap)
1110 	struct vop_link_args /* {
1111 		struct vnode *a_tdvp;
1112 		struct vnode *a_vp;
1113 		struct componentname *a_cnp;
1114 	} */ *ap;
1115 {
1116 	struct componentname *cnp = ap->a_cnp;
1117 	struct proc *p = cnp->cn_proc;
1118 	struct union_node *dun = VTOUNION(ap->a_tdvp);
1119 	struct vnode *vp;
1120 	struct vnode *tdvp;
1121 	int error = 0;
1122 
1123 
1124 	if (ap->a_tdvp->v_op != ap->a_vp->v_op) {
1125 		vp = ap->a_vp;
1126 	} else {
1127 		struct union_node *tun = VTOUNION(ap->a_vp);
1128 		if (tun->un_uppervp == NULLVP) {
1129 			vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY, p);
1130 			if (dun->un_uppervp == tun->un_dirvp) {
1131 				dun->un_flags &= ~UN_ULOCK;
1132 				VOP_UNLOCK(dun->un_uppervp, 0, p);
1133 			}
1134 			error = union_copyup(tun, 1, cnp->cn_cred, p);
1135 			if (dun->un_uppervp == tun->un_dirvp) {
1136 				vn_lock(dun->un_uppervp,
1137 						LK_EXCLUSIVE | LK_RETRY, p);
1138 				dun->un_flags |= UN_ULOCK;
1139 			}
1140 			VOP_UNLOCK(ap->a_vp, 0, p);
1141 		}
1142 		vp = tun->un_uppervp;
1143 	}
1144 
1145 	tdvp = dun->un_uppervp;
1146 	if (tdvp == NULLVP)
1147 		error = EROFS;
1148 
1149 	if (error)
1150 		return (error);
1151 
1152 	FIXUP(dun, p);
1153 	dun->un_flags |= UN_KLOCK;
1154 	VOP_UNLOCK(ap->a_tdvp, 0, p);
1155 
1156 	error = VOP_LINK(tdvp, vp, cnp);
1157 
1158 	dun->un_flags |= UN_ULOCK;
1159 
1160 	return (error);
1161 }
1162 
1163 static int
1164 union_rename(ap)
1165 	struct vop_rename_args  /* {
1166 		struct vnode *a_fdvp;
1167 		struct vnode *a_fvp;
1168 		struct componentname *a_fcnp;
1169 		struct vnode *a_tdvp;
1170 		struct vnode *a_tvp;
1171 		struct componentname *a_tcnp;
1172 	} */ *ap;
1173 {
1174 	int error;
1175 
1176 	struct vnode *fdvp = ap->a_fdvp;
1177 	struct vnode *fvp = ap->a_fvp;
1178 	struct vnode *tdvp = ap->a_tdvp;
1179 	struct vnode *tvp = ap->a_tvp;
1180 
1181 	if (fdvp->v_op == union_vnodeop_p) {	/* always true */
1182 		struct union_node *un = VTOUNION(fdvp);
1183 		if (un->un_uppervp == NULLVP) {
1184 			/*
1185 			 * this should never happen in normal
1186 			 * operation but might if there was
1187 			 * a problem creating the top-level shadow
1188 			 * directory.
1189 			 */
1190 			error = EXDEV;
1191 			goto bad;
1192 		}
1193 
1194 		fdvp = un->un_uppervp;
1195 		VREF(fdvp);
1196 		vrele(ap->a_fdvp);
1197 	}
1198 
1199 	if (fvp->v_op == union_vnodeop_p) {	/* always true */
1200 		struct union_node *un = VTOUNION(fvp);
1201 		if (un->un_uppervp == NULLVP) {
1202 			/* XXX: should do a copyup */
1203 			error = EXDEV;
1204 			goto bad;
1205 		}
1206 
1207 		if (un->un_lowervp != NULLVP)
1208 			ap->a_fcnp->cn_flags |= DOWHITEOUT;
1209 
1210 		fvp = un->un_uppervp;
1211 		VREF(fvp);
1212 		vrele(ap->a_fvp);
1213 	}
1214 
1215 	if (tdvp->v_op == union_vnodeop_p) {
1216 		struct union_node *un = VTOUNION(tdvp);
1217 		if (un->un_uppervp == NULLVP) {
1218 			/*
1219 			 * this should never happen in normal
1220 			 * operation but might if there was
1221 			 * a problem creating the top-level shadow
1222 			 * directory.
1223 			 */
1224 			error = EXDEV;
1225 			goto bad;
1226 		}
1227 
1228 		tdvp = un->un_uppervp;
1229 		VREF(tdvp);
1230 		un->un_flags |= UN_KLOCK;
1231 		vput(ap->a_tdvp);
1232 	}
1233 
1234 	if (tvp != NULLVP && tvp->v_op == union_vnodeop_p) {
1235 		struct union_node *un = VTOUNION(tvp);
1236 
1237 		tvp = un->un_uppervp;
1238 		if (tvp != NULLVP) {
1239 			VREF(tvp);
1240 			un->un_flags |= UN_KLOCK;
1241 		}
1242 		vput(ap->a_tvp);
1243 	}
1244 
1245 	return (VOP_RENAME(fdvp, fvp, ap->a_fcnp, tdvp, tvp, ap->a_tcnp));
1246 
1247 bad:
1248 	vrele(fdvp);
1249 	vrele(fvp);
1250 	vput(tdvp);
1251 	if (tvp != NULLVP)
1252 		vput(tvp);
1253 
1254 	return (error);
1255 }
1256 
1257 static int
1258 union_mkdir(ap)
1259 	struct vop_mkdir_args /* {
1260 		struct vnode *a_dvp;
1261 		struct vnode **a_vpp;
1262 		struct componentname *a_cnp;
1263 		struct vattr *a_vap;
1264 	} */ *ap;
1265 {
1266 	struct union_node *dun = VTOUNION(ap->a_dvp);
1267 	struct vnode *dvp = dun->un_uppervp;
1268 	struct componentname *cnp = ap->a_cnp;
1269 	struct proc *p = cnp->cn_proc;
1270 
1271 	if (dvp != NULLVP) {
1272 		struct vnode *vp;
1273 		int error;
1274 
1275 		FIXUP(dun, p);
1276 		dun->un_flags |= UN_KLOCK;
1277 		VOP_UNLOCK(ap->a_dvp, 0, p);
1278 		error = VOP_MKDIR(dvp, &vp, cnp, ap->a_vap);
1279 		if (error) {
1280 			dun->un_flags |= UN_ULOCK;
1281 			return (error);
1282 		}
1283 
1284 		VOP_UNLOCK(dvp, 0, p);
1285 		error = union_allocvp(ap->a_vpp, ap->a_dvp->v_mount, ap->a_dvp,
1286 				NULLVP, cnp, vp, NULLVP, 1);
1287 		if (error)
1288 			vput(vp);
1289 		vn_lock(ap->a_dvp, LK_EXCLUSIVE| LK_RETRY, p);
1290 
1291 		return (error);
1292 	}
1293 
1294 	return (EROFS);
1295 }
1296 
1297 static int
1298 union_rmdir(ap)
1299 	struct vop_rmdir_args /* {
1300 		struct vnode *a_dvp;
1301 		struct vnode *a_vp;
1302 		struct componentname *a_cnp;
1303 	} */ *ap;
1304 {
1305 	struct union_node *dun = VTOUNION(ap->a_dvp);
1306 	struct union_node *un = VTOUNION(ap->a_vp);
1307 	struct componentname *cnp = ap->a_cnp;
1308 	struct proc *p = cnp->cn_proc;
1309 	int error;
1310 
1311 	if (dun->un_uppervp == NULLVP)
1312 		panic("union rmdir: null upper vnode");
1313 
1314 	if (un->un_uppervp != NULLVP) {
1315 		struct vnode *dvp = dun->un_uppervp;
1316 		struct vnode *vp = un->un_uppervp;
1317 
1318 		FIXUP(dun, p);
1319 		dun->un_flags |= UN_KLOCK;
1320 		VOP_UNLOCK(ap->a_dvp, 0, p);
1321 		FIXUP(un, p);
1322 		un->un_flags |= UN_KLOCK;
1323 		VOP_UNLOCK(ap->a_vp, 0, p);
1324 
1325 		if (union_dowhiteout(un, cnp->cn_cred, p))
1326 			cnp->cn_flags |= DOWHITEOUT;
1327 		error = VOP_RMDIR(dvp, vp, ap->a_cnp);
1328 #if 0
1329 		/* XXX */
1330 		if (!error)
1331 			union_removed_upper(un);
1332 #endif
1333 		dun->un_flags |= UN_ULOCK;
1334 		un->un_flags |= UN_ULOCK;
1335 	} else {
1336 		FIXUP(dun, p);
1337 		error = union_mkwhiteout(
1338 			MOUNTTOUNIONMOUNT(UNIONTOV(dun)->v_mount),
1339 			dun->un_uppervp, ap->a_cnp, un->un_path);
1340 	}
1341 
1342 	return (error);
1343 }
1344 
1345 static int
1346 union_symlink(ap)
1347 	struct vop_symlink_args /* {
1348 		struct vnode *a_dvp;
1349 		struct vnode **a_vpp;
1350 		struct componentname *a_cnp;
1351 		struct vattr *a_vap;
1352 		char *a_target;
1353 	} */ *ap;
1354 {
1355 	struct union_node *dun = VTOUNION(ap->a_dvp);
1356 	struct vnode *dvp = dun->un_uppervp;
1357 	struct componentname *cnp = ap->a_cnp;
1358 	struct proc *p = cnp->cn_proc;
1359 
1360 	if (dvp != NULLVP) {
1361 		struct vnode *vp;
1362 		int error;
1363 
1364 		FIXUP(dun, p);
1365 		dun->un_flags |= UN_KLOCK;
1366 		VOP_UNLOCK(ap->a_dvp, 0, p);
1367 		error = VOP_SYMLINK(dvp, &vp, cnp, ap->a_vap, ap->a_target);
1368 		dun->un_flags |= UN_ULOCK;
1369 		*ap->a_vpp = NULLVP;
1370 		return (error);
1371 	}
1372 
1373 	return (EROFS);
1374 }
1375 
1376 /*
1377  * union_readdir works in concert with getdirentries and
1378  * readdir(3) to provide a list of entries in the unioned
1379  * directories.  getdirentries is responsible for walking
1380  * down the union stack.  readdir(3) is responsible for
1381  * eliminating duplicate names from the returned data stream.
1382  */
1383 static int
1384 union_readdir(ap)
1385 	struct vop_readdir_args /* {
1386 		struct vnode *a_vp;
1387 		struct uio *a_uio;
1388 		struct ucred *a_cred;
1389 		int *a_eofflag;
1390 		u_long *a_cookies;
1391 		int a_ncookies;
1392 	} */ *ap;
1393 {
1394 	struct union_node *un = VTOUNION(ap->a_vp);
1395 	struct vnode *uvp = un->un_uppervp;
1396 	struct proc *p = ap->a_uio->uio_procp;
1397 
1398 	if (uvp == NULLVP)
1399 		return (0);
1400 
1401 	FIXUP(un, p);
1402 	ap->a_vp = uvp;
1403 	return (VCALL(uvp, VOFFSET(vop_readdir), ap));
1404 }
1405 
1406 static int
1407 union_readlink(ap)
1408 	struct vop_readlink_args /* {
1409 		struct vnode *a_vp;
1410 		struct uio *a_uio;
1411 		struct ucred *a_cred;
1412 	} */ *ap;
1413 {
1414 	int error;
1415 	struct uio *uio = ap->a_uio;
1416 	struct proc *p = uio->uio_procp;
1417 	struct vnode *vp = OTHERVP(ap->a_vp);
1418 	int dolock = (vp == LOWERVP(ap->a_vp));
1419 
1420 	if (dolock)
1421 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
1422 	else
1423 		FIXUP(VTOUNION(ap->a_vp), p);
1424 	ap->a_vp = vp;
1425 	error = VCALL(vp, VOFFSET(vop_readlink), ap);
1426 	if (dolock)
1427 		VOP_UNLOCK(vp, 0, p);
1428 
1429 	return (error);
1430 }
1431 
1432 static int
1433 union_abortop(ap)
1434 	struct vop_abortop_args /* {
1435 		struct vnode *a_dvp;
1436 		struct componentname *a_cnp;
1437 	} */ *ap;
1438 {
1439 	int error;
1440 	struct componentname *cnp = ap->a_cnp;
1441 	struct proc *p = cnp->cn_proc;
1442 	struct vnode *vp = OTHERVP(ap->a_dvp);
1443 	struct union_node *un = VTOUNION(ap->a_dvp);
1444 	int islocked = un->un_flags & UN_LOCKED;
1445 	int dolock = (vp == LOWERVP(ap->a_dvp));
1446 
1447 	if (islocked) {
1448 		if (dolock)
1449 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
1450 		else
1451 			FIXUP(VTOUNION(ap->a_dvp), p);
1452 	}
1453 	ap->a_dvp = vp;
1454 	error = VCALL(vp, VOFFSET(vop_abortop), ap);
1455 	if (islocked && dolock)
1456 		VOP_UNLOCK(vp, 0, p);
1457 
1458 	return (error);
1459 }
1460 
1461 static int
1462 union_inactive(ap)
1463 	struct vop_inactive_args /* {
1464 		struct vnode *a_vp;
1465 		struct proc *a_p;
1466 	} */ *ap;
1467 {
1468 	struct vnode *vp = ap->a_vp;
1469 	struct proc *p = ap->a_p;
1470 	struct union_node *un = VTOUNION(vp);
1471 	struct vnode **vpp;
1472 
1473 	/*
1474 	 * Do nothing (and _don't_ bypass).
1475 	 * Wait to vrele lowervp until reclaim,
1476 	 * so that until then our union_node is in the
1477 	 * cache and reusable.
1478 	 *
1479 	 * NEEDSWORK: Someday, consider inactive'ing
1480 	 * the lowervp and then trying to reactivate it
1481 	 * with capabilities (v_id)
1482 	 * like they do in the name lookup cache code.
1483 	 * That's too much work for now.
1484 	 */
1485 
1486 	if (un->un_dircache != 0) {
1487 		for (vpp = un->un_dircache; *vpp != NULLVP; vpp++)
1488 			vrele(*vpp);
1489 		free(un->un_dircache, M_TEMP);
1490 		un->un_dircache = 0;
1491 	}
1492 
1493 	VOP_UNLOCK(vp, 0, p);
1494 
1495 	if ((un->un_flags & UN_CACHED) == 0)
1496 		vgone(vp);
1497 
1498 	return (0);
1499 }
1500 
1501 static int
1502 union_reclaim(ap)
1503 	struct vop_reclaim_args /* {
1504 		struct vnode *a_vp;
1505 	} */ *ap;
1506 {
1507 
1508 	union_freevp(ap->a_vp);
1509 
1510 	return (0);
1511 }
1512 
1513 static int
1514 union_lock(ap)
1515 	struct vop_lock_args *ap;
1516 {
1517 	struct vnode *vp = ap->a_vp;
1518 	struct proc *p = ap->a_p;
1519 	int flags = ap->a_flags;
1520 	struct union_node *un;
1521 	int error;
1522 
1523 	vop_nolock(ap);
1524 	/*
1525 	 * Need to do real lockmgr-style locking here.
1526 	 * in the mean time, draining won't work quite right,
1527 	 * which could lead to a few race conditions.
1528 	 * the following test was here, but is not quite right, we
1529 	 * still need to take the lock:
1530 	if ((flags & LK_TYPE_MASK) == LK_DRAIN)
1531 		return (0);
1532 	 */
1533 	flags &= ~LK_INTERLOCK;
1534 
1535 start:
1536 	un = VTOUNION(vp);
1537 
1538 	if (un->un_uppervp != NULLVP) {
1539 		if (((un->un_flags & UN_ULOCK) == 0) &&
1540 		    (vp->v_usecount != 0)) {
1541 			error = vn_lock(un->un_uppervp, flags, p);
1542 			if (error)
1543 				return (error);
1544 			un->un_flags |= UN_ULOCK;
1545 		}
1546 #ifdef DIAGNOSTIC
1547 		if (un->un_flags & UN_KLOCK) {
1548 			vprint("dangling upper lock", vp);
1549 			panic("union: dangling upper lock");
1550 		}
1551 #endif
1552 	}
1553 
1554 	if (un->un_flags & UN_LOCKED) {
1555 #ifdef DIAGNOSTIC
1556 		if (curproc && un->un_pid == curproc->p_pid &&
1557 			    un->un_pid > -1 && curproc->p_pid > -1)
1558 			panic("union: locking against myself");
1559 #endif
1560 		un->un_flags |= UN_WANT;
1561 		tsleep((caddr_t)&un->un_flags, PINOD, "unionlk2", 0);
1562 		goto start;
1563 	}
1564 
1565 #ifdef DIAGNOSTIC
1566 	if (curproc)
1567 		un->un_pid = curproc->p_pid;
1568 	else
1569 		un->un_pid = -1;
1570 #endif
1571 
1572 	un->un_flags |= UN_LOCKED;
1573 	return (0);
1574 }
1575 
1576 /*
1577  * When operations want to vput() a union node yet retain a lock on
1578  * the upper vnode (say, to do some further operations like link(),
1579  * mkdir(), ...), they set UN_KLOCK on the union node, then call
1580  * vput() which calls VOP_UNLOCK() and comes here.  union_unlock()
1581  * unlocks the union node (leaving the upper vnode alone), clears the
1582  * KLOCK flag, and then returns to vput().  The caller then does whatever
1583  * is left to do with the upper vnode, and ensures that it gets unlocked.
1584  *
1585  * If UN_KLOCK isn't set, then the upper vnode is unlocked here.
1586  */
1587 static int
1588 union_unlock(ap)
1589 	struct vop_unlock_args /* {
1590 		struct vnode *a_vp;
1591 		int a_flags;
1592 		struct proc *a_p;
1593 	} */ *ap;
1594 {
1595 	struct union_node *un = VTOUNION(ap->a_vp);
1596 	struct proc *p = ap->a_p;
1597 
1598 #ifdef DIAGNOSTIC
1599 	if ((un->un_flags & UN_LOCKED) == 0)
1600 		panic("union: unlock unlocked node");
1601 	if (curproc && un->un_pid != curproc->p_pid &&
1602 			curproc->p_pid > -1 && un->un_pid > -1)
1603 		panic("union: unlocking other process's union node");
1604 #endif
1605 
1606 	un->un_flags &= ~UN_LOCKED;
1607 
1608 	if ((un->un_flags & (UN_ULOCK|UN_KLOCK)) == UN_ULOCK)
1609 		VOP_UNLOCK(un->un_uppervp, 0, p);
1610 
1611 	un->un_flags &= ~(UN_ULOCK|UN_KLOCK);
1612 
1613 	if (un->un_flags & UN_WANT) {
1614 		un->un_flags &= ~UN_WANT;
1615 		wakeup((caddr_t) &un->un_flags);
1616 	}
1617 
1618 #ifdef DIAGNOSTIC
1619 	un->un_pid = 0;
1620 #endif
1621 	vop_nounlock(ap);
1622 
1623 	return (0);
1624 }
1625 
1626 static int
1627 union_bmap(ap)
1628 	struct vop_bmap_args /* {
1629 		struct vnode *a_vp;
1630 		daddr_t  a_bn;
1631 		struct vnode **a_vpp;
1632 		daddr_t *a_bnp;
1633 		int *a_runp;
1634 		int *a_runb;
1635 	} */ *ap;
1636 {
1637 	int error;
1638 	struct proc *p = curproc;		/* XXX */
1639 	struct vnode *vp = OTHERVP(ap->a_vp);
1640 	int dolock = (vp == LOWERVP(ap->a_vp));
1641 
1642 	if (dolock)
1643 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
1644 	else
1645 		FIXUP(VTOUNION(ap->a_vp), p);
1646 	ap->a_vp = vp;
1647 	error = VCALL(vp, VOFFSET(vop_bmap), ap);
1648 	if (dolock)
1649 		VOP_UNLOCK(vp, 0, p);
1650 
1651 	return (error);
1652 }
1653 
1654 static int
1655 union_print(ap)
1656 	struct vop_print_args /* {
1657 		struct vnode *a_vp;
1658 	} */ *ap;
1659 {
1660 	struct vnode *vp = ap->a_vp;
1661 
1662 	printf("\ttag VT_UNION, vp=%p, uppervp=%p, lowervp=%p\n",
1663 			vp, UPPERVP(vp), LOWERVP(vp));
1664 	if (UPPERVP(vp) != NULLVP)
1665 		vprint("union: upper", UPPERVP(vp));
1666 	if (LOWERVP(vp) != NULLVP)
1667 		vprint("union: lower", LOWERVP(vp));
1668 
1669 	return (0);
1670 }
1671 
1672 static int
1673 union_islocked(ap)
1674 	struct vop_islocked_args /* {
1675 		struct vnode *a_vp;
1676 	} */ *ap;
1677 {
1678 
1679 	return ((VTOUNION(ap->a_vp)->un_flags & UN_LOCKED) ? 1 : 0);
1680 }
1681 
1682 static int
1683 union_pathconf(ap)
1684 	struct vop_pathconf_args /* {
1685 		struct vnode *a_vp;
1686 		int a_name;
1687 		int *a_retval;
1688 	} */ *ap;
1689 {
1690 	int error;
1691 	struct proc *p = curproc;		/* XXX */
1692 	struct vnode *vp = OTHERVP(ap->a_vp);
1693 	int dolock = (vp == LOWERVP(ap->a_vp));
1694 
1695 	if (dolock)
1696 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
1697 	else
1698 		FIXUP(VTOUNION(ap->a_vp), p);
1699 	ap->a_vp = vp;
1700 	error = VCALL(vp, VOFFSET(vop_pathconf), ap);
1701 	if (dolock)
1702 		VOP_UNLOCK(vp, 0, p);
1703 
1704 	return (error);
1705 }
1706 
1707 static int
1708 union_advlock(ap)
1709 	struct vop_advlock_args /* {
1710 		struct vnode *a_vp;
1711 		caddr_t  a_id;
1712 		int  a_op;
1713 		struct flock *a_fl;
1714 		int  a_flags;
1715 	} */ *ap;
1716 {
1717 	register struct vnode *ovp = OTHERVP(ap->a_vp);
1718 
1719 	ap->a_vp = ovp;
1720 	return (VCALL(ovp, VOFFSET(vop_advlock), ap));
1721 }
1722 
1723 
1724 /*
1725  * XXX - vop_strategy must be hand coded because it has no
1726  * vnode in its arguments.
1727  * This goes away with a merged VM/buffer cache.
1728  */
1729 static int
1730 union_strategy(ap)
1731 	struct vop_strategy_args /* {
1732 		struct vnode *a_vp;
1733 		struct buf *a_bp;
1734 	} */ *ap;
1735 {
1736 	struct buf *bp = ap->a_bp;
1737 	struct vnode *othervp = OTHERVP(bp->b_vp);
1738 
1739 #ifdef DIAGNOSTIC
1740 	if (othervp == NULLVP)
1741 		panic("union_strategy: nil vp");
1742 	if (((bp->b_flags & B_READ) == 0) &&
1743 	    (othervp == LOWERVP(bp->b_vp)))
1744 		panic("union_strategy: writing to lowervp");
1745 #endif
1746 
1747 	return (VOP_STRATEGY(othervp, bp));
1748 }
1749 
1750 /*
1751  * Global vfs data structures
1752  */
1753 vop_t **union_vnodeop_p;
1754 static struct vnodeopv_entry_desc union_vnodeop_entries[] = {
1755 	{ &vop_default_desc,		(vop_t *) vop_defaultop },
1756 	{ &vop_abortop_desc,		(vop_t *) union_abortop },
1757 	{ &vop_access_desc,		(vop_t *) union_access },
1758 	{ &vop_advlock_desc,		(vop_t *) union_advlock },
1759 	{ &vop_bmap_desc,		(vop_t *) union_bmap },
1760 	{ &vop_close_desc,		(vop_t *) union_close },
1761 	{ &vop_create_desc,		(vop_t *) union_create },
1762 	{ &vop_fsync_desc,		(vop_t *) union_fsync },
1763 	{ &vop_getattr_desc,		(vop_t *) union_getattr },
1764 	{ &vop_inactive_desc,		(vop_t *) union_inactive },
1765 	{ &vop_ioctl_desc,		(vop_t *) union_ioctl },
1766 	{ &vop_islocked_desc,		(vop_t *) union_islocked },
1767 	{ &vop_lease_desc,		(vop_t *) union_lease },
1768 	{ &vop_link_desc,		(vop_t *) union_link },
1769 	{ &vop_lock_desc,		(vop_t *) union_lock },
1770 	{ &vop_lookup_desc,		(vop_t *) union_lookup },
1771 	{ &vop_mkdir_desc,		(vop_t *) union_mkdir },
1772 	{ &vop_mknod_desc,		(vop_t *) union_mknod },
1773 	{ &vop_mmap_desc,		(vop_t *) union_mmap },
1774 	{ &vop_open_desc,		(vop_t *) union_open },
1775 	{ &vop_pathconf_desc,		(vop_t *) union_pathconf },
1776 	{ &vop_poll_desc,		(vop_t *) union_poll },
1777 	{ &vop_print_desc,		(vop_t *) union_print },
1778 	{ &vop_read_desc,		(vop_t *) union_read },
1779 	{ &vop_readdir_desc,		(vop_t *) union_readdir },
1780 	{ &vop_readlink_desc,		(vop_t *) union_readlink },
1781 	{ &vop_reclaim_desc,		(vop_t *) union_reclaim },
1782 	{ &vop_remove_desc,		(vop_t *) union_remove },
1783 	{ &vop_rename_desc,		(vop_t *) union_rename },
1784 	{ &vop_revoke_desc,		(vop_t *) union_revoke },
1785 	{ &vop_rmdir_desc,		(vop_t *) union_rmdir },
1786 	{ &vop_setattr_desc,		(vop_t *) union_setattr },
1787 	{ &vop_strategy_desc,		(vop_t *) union_strategy },
1788 	{ &vop_symlink_desc,		(vop_t *) union_symlink },
1789 	{ &vop_unlock_desc,		(vop_t *) union_unlock },
1790 	{ &vop_whiteout_desc,		(vop_t *) union_whiteout },
1791 	{ &vop_write_desc,		(vop_t *) union_write },
1792 	{ NULL, NULL }
1793 };
1794 static struct vnodeopv_desc union_vnodeop_opv_desc =
1795 	{ &union_vnodeop_p, union_vnodeop_entries };
1796 
1797 VNODEOP_SET(union_vnodeop_opv_desc);
1798