xref: /freebsd/sys/fs/unionfs/union_vnops.c (revision a8445737e740901f5f2c8d24c12ef7fc8b00134e)
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 1.57 1998/05/07 04:58:38 msmith 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;
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: case VDIR: case VLNK:
699 			return (EROFS);
700 		}
701 	}
702 	if ((vp = un->un_uppervp) != NULLVP) {
703 		FIXUP(un, p);
704 		ap->a_vp = vp;
705 		return (VCALL(vp, VOFFSET(vop_access), ap));
706 	}
707 
708 	if ((vp = un->un_lowervp) != NULLVP) {
709 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
710 		savedvp = ap->a_vp;
711 		ap->a_vp = vp;
712 		error = VCALL(vp, VOFFSET(vop_access), ap);
713 		if (error == 0) {
714 			struct union_mount *um = MOUNTTOUNIONMOUNT(savedvp->v_mount);
715 
716 			if (um->um_op == UNMNT_BELOW) {
717 				ap->a_cred = um->um_cred;
718 				error = VCALL(vp, VOFFSET(vop_access), ap);
719 			}
720 		}
721 		VOP_UNLOCK(vp, 0, p);
722 		if (error)
723 			return (error);
724 	}
725 
726 	return (error);
727 }
728 
729 /*
730  * We handle getattr only to change the fsid and
731  * track object sizes
732  */
733 static int
734 union_getattr(ap)
735 	struct vop_getattr_args /* {
736 		struct vnode *a_vp;
737 		struct vattr *a_vap;
738 		struct ucred *a_cred;
739 		struct proc *a_p;
740 	} */ *ap;
741 {
742 	int error;
743 	struct union_node *un = VTOUNION(ap->a_vp);
744 	struct vnode *vp = un->un_uppervp;
745 	struct proc *p = ap->a_p;
746 	struct vattr *vap;
747 	struct vattr va;
748 
749 
750 	/*
751 	 * Some programs walk the filesystem hierarchy by counting
752 	 * links to directories to avoid stat'ing all the time.
753 	 * This means the link count on directories needs to be "correct".
754 	 * The only way to do that is to call getattr on both layers
755 	 * and fix up the link count.  The link count will not necessarily
756 	 * be accurate but will be large enough to defeat the tree walkers.
757 	 */
758 
759 	vap = ap->a_vap;
760 
761 	vp = un->un_uppervp;
762 	if (vp != NULLVP) {
763 		/*
764 		 * It's not clear whether VOP_GETATTR is to be
765 		 * called with the vnode locked or not.  stat() calls
766 		 * it with (vp) locked, and fstat calls it with
767 		 * (vp) unlocked.
768 		 * In the mean time, compensate here by checking
769 		 * the union_node's lock flag.
770 		 */
771 		if (un->un_flags & UN_LOCKED)
772 			FIXUP(un, p);
773 
774 		error = VOP_GETATTR(vp, vap, ap->a_cred, ap->a_p);
775 		if (error)
776 			return (error);
777 		union_newsize(ap->a_vp, vap->va_size, VNOVAL);
778 	}
779 
780 	if (vp == NULLVP) {
781 		vp = un->un_lowervp;
782 	} else if (vp->v_type == VDIR && un->un_lowervp != NULLVP) {
783 		vp = un->un_lowervp;
784 		vap = &va;
785 	} else {
786 		vp = NULLVP;
787 	}
788 
789 	if (vp != NULLVP) {
790 		error = VOP_GETATTR(vp, vap, ap->a_cred, ap->a_p);
791 		if (error)
792 			return (error);
793 		union_newsize(ap->a_vp, VNOVAL, vap->va_size);
794 	}
795 
796 	if ((vap != ap->a_vap) && (vap->va_type == VDIR))
797 		ap->a_vap->va_nlink += vap->va_nlink;
798 
799 	ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsid.val[0];
800 	return (0);
801 }
802 
803 static int
804 union_setattr(ap)
805 	struct vop_setattr_args /* {
806 		struct vnode *a_vp;
807 		struct vattr *a_vap;
808 		struct ucred *a_cred;
809 		struct proc *a_p;
810 	} */ *ap;
811 {
812 	struct union_node *un = VTOUNION(ap->a_vp);
813 	struct proc *p = ap->a_p;
814 	struct vattr *vap = ap->a_vap;
815 	int error;
816 
817 	/*
818 	 * Disallow write attempts on filesystems mounted read-only.
819 	 */
820 	if ((ap->a_vp->v_mount->mnt_flag & MNT_RDONLY) &&
821 		(vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
822 		 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
823 		 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL))
824 		return (EROFS);
825 
826 	/*
827 	 * Handle case of truncating lower object to zero size,
828 	 * by creating a zero length upper object.  This is to
829 	 * handle the case of open with O_TRUNC and O_CREAT.
830 	 */
831 	if ((un->un_uppervp == NULLVP) &&
832 	    /* assert(un->un_lowervp != NULLVP) */
833 	    (un->un_lowervp->v_type == VREG)) {
834 		error = union_copyup(un, (ap->a_vap->va_size != 0),
835 						ap->a_cred, ap->a_p);
836 		if (error)
837 			return (error);
838 	}
839 
840 	/*
841 	 * Try to set attributes in upper layer,
842 	 * otherwise return read-only filesystem error.
843 	 */
844 	if (un->un_uppervp != NULLVP) {
845 		FIXUP(un, p);
846 		error = VOP_SETATTR(un->un_uppervp, ap->a_vap,
847 					ap->a_cred, ap->a_p);
848 		if ((error == 0) && (ap->a_vap->va_size != VNOVAL))
849 			union_newsize(ap->a_vp, ap->a_vap->va_size, VNOVAL);
850 	} else {
851 		error = EROFS;
852 	}
853 
854 	return (error);
855 }
856 
857 static int
858 union_read(ap)
859 	struct vop_read_args /* {
860 		struct vnode *a_vp;
861 		struct uio *a_uio;
862 		int  a_ioflag;
863 		struct ucred *a_cred;
864 	} */ *ap;
865 {
866 	int error;
867 	struct proc *p = ap->a_uio->uio_procp;
868 	struct vnode *vp = OTHERVP(ap->a_vp);
869 	int dolock = (vp == LOWERVP(ap->a_vp));
870 
871 	if (dolock)
872 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
873 	else
874 		FIXUP(VTOUNION(ap->a_vp), p);
875 	error = VOP_READ(vp, ap->a_uio, ap->a_ioflag, ap->a_cred);
876 	if (dolock)
877 		VOP_UNLOCK(vp, 0, p);
878 
879 	/*
880 	 * XXX
881 	 * perhaps the size of the underlying object has changed under
882 	 * our feet.  take advantage of the offset information present
883 	 * in the uio structure.
884 	 */
885 	if (error == 0) {
886 		struct union_node *un = VTOUNION(ap->a_vp);
887 		off_t cur = ap->a_uio->uio_offset;
888 
889 		if (vp == un->un_uppervp) {
890 			if (cur > un->un_uppersz)
891 				union_newsize(ap->a_vp, cur, VNOVAL);
892 		} else {
893 			if (cur > un->un_lowersz)
894 				union_newsize(ap->a_vp, VNOVAL, cur);
895 		}
896 	}
897 
898 	return (error);
899 }
900 
901 static int
902 union_write(ap)
903 	struct vop_read_args /* {
904 		struct vnode *a_vp;
905 		struct uio *a_uio;
906 		int  a_ioflag;
907 		struct ucred *a_cred;
908 	} */ *ap;
909 {
910 	int error;
911 	struct vnode *vp;
912 	struct union_node *un = VTOUNION(ap->a_vp);
913 	struct proc *p = ap->a_uio->uio_procp;
914 
915 	vp = UPPERVP(ap->a_vp);
916 	if (vp == NULLVP)
917 		panic("union: missing upper layer in write");
918 
919 	FIXUP(un, p);
920 	error = VOP_WRITE(vp, ap->a_uio, ap->a_ioflag, ap->a_cred);
921 
922 	/*
923 	 * the size of the underlying object may be changed by the
924 	 * write.
925 	 */
926 	if (error == 0) {
927 		off_t cur = ap->a_uio->uio_offset;
928 
929 		if (cur > un->un_uppersz)
930 			union_newsize(ap->a_vp, cur, VNOVAL);
931 	}
932 
933 	return (error);
934 }
935 
936 static int
937 union_lease(ap)
938 	struct vop_lease_args /* {
939 		struct vnode *a_vp;
940 		struct proc *a_p;
941 		struct ucred *a_cred;
942 		int a_flag;
943 	} */ *ap;
944 {
945 	register struct vnode *ovp = OTHERVP(ap->a_vp);
946 
947 	ap->a_vp = ovp;
948 	return (VCALL(ovp, VOFFSET(vop_lease), ap));
949 }
950 
951 static int
952 union_ioctl(ap)
953 	struct vop_ioctl_args /* {
954 		struct vnode *a_vp;
955 		int  a_command;
956 		caddr_t  a_data;
957 		int  a_fflag;
958 		struct ucred *a_cred;
959 		struct proc *a_p;
960 	} */ *ap;
961 {
962 	register struct vnode *ovp = OTHERVP(ap->a_vp);
963 
964 	ap->a_vp = ovp;
965 	return (VCALL(ovp, VOFFSET(vop_ioctl), ap));
966 }
967 
968 static int
969 union_poll(ap)
970 	struct vop_poll_args /* {
971 		struct vnode *a_vp;
972 		int  a_events;
973 		struct ucred *a_cred;
974 		struct proc *a_p;
975 	} */ *ap;
976 {
977 	register struct vnode *ovp = OTHERVP(ap->a_vp);
978 
979 	ap->a_vp = ovp;
980 	return (VCALL(ovp, VOFFSET(vop_poll), ap));
981 }
982 
983 static int
984 union_revoke(ap)
985 	struct vop_revoke_args /* {
986 		struct vnode *a_vp;
987 		int a_flags;
988 		struct proc *a_p;
989 	} */ *ap;
990 {
991 	struct vnode *vp = ap->a_vp;
992 
993 	if (UPPERVP(vp))
994 		VOP_REVOKE(UPPERVP(vp), ap->a_flags);
995 	if (LOWERVP(vp))
996 		VOP_REVOKE(LOWERVP(vp), ap->a_flags);
997 	vgone(vp);
998 	return (0);
999 }
1000 
1001 static int
1002 union_mmap(ap)
1003 	struct vop_mmap_args /* {
1004 		struct vnode *a_vp;
1005 		int  a_fflags;
1006 		struct ucred *a_cred;
1007 		struct proc *a_p;
1008 	} */ *ap;
1009 {
1010 	register struct vnode *ovp = OTHERVP(ap->a_vp);
1011 
1012 	ap->a_vp = ovp;
1013 	return (VCALL(ovp, VOFFSET(vop_mmap), ap));
1014 }
1015 
1016 static int
1017 union_fsync(ap)
1018 	struct vop_fsync_args /* {
1019 		struct vnode *a_vp;
1020 		struct ucred *a_cred;
1021 		int  a_waitfor;
1022 		struct proc *a_p;
1023 	} */ *ap;
1024 {
1025 	int error = 0;
1026 	struct proc *p = ap->a_p;
1027 	struct vnode *targetvp = OTHERVP(ap->a_vp);
1028 	struct union_node *un;
1029 
1030 	if (targetvp != NULLVP) {
1031 		int dolock = (targetvp == LOWERVP(ap->a_vp));
1032 
1033 		un = VTOUNION(ap->a_vp);
1034 		if (dolock)
1035 			vn_lock(targetvp, LK_EXCLUSIVE | LK_RETRY, p);
1036 		else  {
1037 			un = VTOUNION(ap->a_vp);
1038 			if ((un->un_flags & UN_ULOCK) == 0 &&
1039 			    targetvp->v_data != NULL &&
1040 			    ((struct lock *)targetvp->v_data)->lk_lockholder
1041 			        == curproc->p_pid &&
1042 			    VOP_ISLOCKED(targetvp) != 0)
1043 				return 0;   /* XXX */
1044 
1045 			FIXUP(un, p);
1046 		}
1047 
1048 		error = VOP_FSYNC(targetvp, ap->a_cred, ap->a_waitfor, p);
1049 		if (dolock)
1050 			VOP_UNLOCK(targetvp, 0, p);
1051 	}
1052 
1053 	return (error);
1054 }
1055 
1056 static int
1057 union_remove(ap)
1058 	struct vop_remove_args /* {
1059 		struct vnode *a_dvp;
1060 		struct vnode *a_vp;
1061 		struct componentname *a_cnp;
1062 	} */ *ap;
1063 {
1064 	struct union_node *dun = VTOUNION(ap->a_dvp);
1065 	struct union_node *un = VTOUNION(ap->a_vp);
1066 	struct componentname *cnp = ap->a_cnp;
1067 	struct proc *p = cnp->cn_proc;
1068 	int error;
1069 
1070 	if (dun->un_uppervp == NULLVP)
1071 		panic("union remove: null upper vnode");
1072 
1073 	if (un->un_uppervp != NULLVP) {
1074 		struct vnode *dvp = dun->un_uppervp;
1075 		struct vnode *vp = un->un_uppervp;
1076 
1077 		FIXUP(dun, p);
1078 		dun->un_flags |= UN_KLOCK;
1079 		VOP_UNLOCK(ap->a_dvp, 0, p);
1080 		FIXUP(un, p);
1081 		un->un_flags |= UN_KLOCK;
1082 		VOP_UNLOCK(ap->a_vp, 0, p);
1083 
1084 		if (union_dowhiteout(un, cnp->cn_cred, p))
1085 			cnp->cn_flags |= DOWHITEOUT;
1086 		error = VOP_REMOVE(dvp, vp, cnp);
1087 #if 0
1088 		/* XXX */
1089 		if (!error)
1090 			union_removed_upper(un);
1091 #endif
1092 		dun->un_flags |= UN_ULOCK;
1093 		un->un_flags |= UN_ULOCK;
1094 	} else {
1095 		FIXUP(dun, p);
1096 		error = union_mkwhiteout(
1097 			MOUNTTOUNIONMOUNT(UNIONTOV(dun)->v_mount),
1098 			dun->un_uppervp, ap->a_cnp, un->un_path);
1099 	}
1100 
1101 	return (error);
1102 }
1103 
1104 static int
1105 union_link(ap)
1106 	struct vop_link_args /* {
1107 		struct vnode *a_tdvp;
1108 		struct vnode *a_vp;
1109 		struct componentname *a_cnp;
1110 	} */ *ap;
1111 {
1112 	struct componentname *cnp = ap->a_cnp;
1113 	struct proc *p = cnp->cn_proc;
1114 	struct union_node *dun = VTOUNION(ap->a_tdvp);
1115 	struct vnode *vp;
1116 	struct vnode *tdvp;
1117 	int error = 0;
1118 
1119 
1120 	if (ap->a_tdvp->v_op != ap->a_vp->v_op) {
1121 		vp = ap->a_vp;
1122 	} else {
1123 		struct union_node *tun = VTOUNION(ap->a_vp);
1124 		if (tun->un_uppervp == NULLVP) {
1125 			vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY, p);
1126 			if (dun->un_uppervp == tun->un_dirvp) {
1127 				dun->un_flags &= ~UN_ULOCK;
1128 				VOP_UNLOCK(dun->un_uppervp, 0, p);
1129 			}
1130 			error = union_copyup(tun, 1, cnp->cn_cred, p);
1131 			if (dun->un_uppervp == tun->un_dirvp) {
1132 				vn_lock(dun->un_uppervp,
1133 						LK_EXCLUSIVE | LK_RETRY, p);
1134 				dun->un_flags |= UN_ULOCK;
1135 			}
1136 			VOP_UNLOCK(ap->a_vp, 0, p);
1137 		}
1138 		vp = tun->un_uppervp;
1139 	}
1140 
1141 	tdvp = dun->un_uppervp;
1142 	if (tdvp == NULLVP)
1143 		error = EROFS;
1144 
1145 	if (error)
1146 		return (error);
1147 
1148 	FIXUP(dun, p);
1149 	dun->un_flags |= UN_KLOCK;
1150 	VOP_UNLOCK(ap->a_tdvp, 0, p);
1151 
1152 	error = VOP_LINK(tdvp, vp, cnp);
1153 
1154 	dun->un_flags |= UN_ULOCK;
1155 
1156 	return (error);
1157 }
1158 
1159 static int
1160 union_rename(ap)
1161 	struct vop_rename_args  /* {
1162 		struct vnode *a_fdvp;
1163 		struct vnode *a_fvp;
1164 		struct componentname *a_fcnp;
1165 		struct vnode *a_tdvp;
1166 		struct vnode *a_tvp;
1167 		struct componentname *a_tcnp;
1168 	} */ *ap;
1169 {
1170 	int error;
1171 
1172 	struct vnode *fdvp = ap->a_fdvp;
1173 	struct vnode *fvp = ap->a_fvp;
1174 	struct vnode *tdvp = ap->a_tdvp;
1175 	struct vnode *tvp = ap->a_tvp;
1176 
1177 	if (fdvp->v_op == union_vnodeop_p) {	/* always true */
1178 		struct union_node *un = VTOUNION(fdvp);
1179 		if (un->un_uppervp == NULLVP) {
1180 			/*
1181 			 * this should never happen in normal
1182 			 * operation but might if there was
1183 			 * a problem creating the top-level shadow
1184 			 * directory.
1185 			 */
1186 			error = EXDEV;
1187 			goto bad;
1188 		}
1189 
1190 		fdvp = un->un_uppervp;
1191 		VREF(fdvp);
1192 		vrele(ap->a_fdvp);
1193 	}
1194 
1195 	if (fvp->v_op == union_vnodeop_p) {	/* always true */
1196 		struct union_node *un = VTOUNION(fvp);
1197 		if (un->un_uppervp == NULLVP) {
1198 			/* XXX: should do a copyup */
1199 			error = EXDEV;
1200 			goto bad;
1201 		}
1202 
1203 		if (un->un_lowervp != NULLVP)
1204 			ap->a_fcnp->cn_flags |= DOWHITEOUT;
1205 
1206 		fvp = un->un_uppervp;
1207 		VREF(fvp);
1208 		vrele(ap->a_fvp);
1209 	}
1210 
1211 	if (tdvp->v_op == union_vnodeop_p) {
1212 		struct union_node *un = VTOUNION(tdvp);
1213 		if (un->un_uppervp == NULLVP) {
1214 			/*
1215 			 * this should never happen in normal
1216 			 * operation but might if there was
1217 			 * a problem creating the top-level shadow
1218 			 * directory.
1219 			 */
1220 			error = EXDEV;
1221 			goto bad;
1222 		}
1223 
1224 		tdvp = un->un_uppervp;
1225 		VREF(tdvp);
1226 		un->un_flags |= UN_KLOCK;
1227 		vput(ap->a_tdvp);
1228 	}
1229 
1230 	if (tvp != NULLVP && tvp->v_op == union_vnodeop_p) {
1231 		struct union_node *un = VTOUNION(tvp);
1232 
1233 		tvp = un->un_uppervp;
1234 		if (tvp != NULLVP) {
1235 			VREF(tvp);
1236 			un->un_flags |= UN_KLOCK;
1237 		}
1238 		vput(ap->a_tvp);
1239 	}
1240 
1241 	return (VOP_RENAME(fdvp, fvp, ap->a_fcnp, tdvp, tvp, ap->a_tcnp));
1242 
1243 bad:
1244 	vrele(fdvp);
1245 	vrele(fvp);
1246 	vput(tdvp);
1247 	if (tvp != NULLVP)
1248 		vput(tvp);
1249 
1250 	return (error);
1251 }
1252 
1253 static int
1254 union_mkdir(ap)
1255 	struct vop_mkdir_args /* {
1256 		struct vnode *a_dvp;
1257 		struct vnode **a_vpp;
1258 		struct componentname *a_cnp;
1259 		struct vattr *a_vap;
1260 	} */ *ap;
1261 {
1262 	struct union_node *dun = VTOUNION(ap->a_dvp);
1263 	struct vnode *dvp = dun->un_uppervp;
1264 	struct componentname *cnp = ap->a_cnp;
1265 	struct proc *p = cnp->cn_proc;
1266 
1267 	if (dvp != NULLVP) {
1268 		struct vnode *vp;
1269 		int error;
1270 
1271 		FIXUP(dun, p);
1272 		dun->un_flags |= UN_KLOCK;
1273 		VOP_UNLOCK(ap->a_dvp, 0, p);
1274 		error = VOP_MKDIR(dvp, &vp, cnp, ap->a_vap);
1275 		if (error) {
1276 			dun->un_flags |= UN_ULOCK;
1277 			return (error);
1278 		}
1279 
1280 		VOP_UNLOCK(dvp, 0, p);
1281 		error = union_allocvp(ap->a_vpp, ap->a_dvp->v_mount, ap->a_dvp,
1282 				NULLVP, cnp, vp, NULLVP, 1);
1283 		if (error)
1284 			vput(vp);
1285 		vn_lock(ap->a_dvp, LK_EXCLUSIVE| LK_RETRY, p);
1286 
1287 		return (error);
1288 	}
1289 
1290 	return (EROFS);
1291 }
1292 
1293 static int
1294 union_rmdir(ap)
1295 	struct vop_rmdir_args /* {
1296 		struct vnode *a_dvp;
1297 		struct vnode *a_vp;
1298 		struct componentname *a_cnp;
1299 	} */ *ap;
1300 {
1301 	struct union_node *dun = VTOUNION(ap->a_dvp);
1302 	struct union_node *un = VTOUNION(ap->a_vp);
1303 	struct componentname *cnp = ap->a_cnp;
1304 	struct proc *p = cnp->cn_proc;
1305 	int error;
1306 
1307 	if (dun->un_uppervp == NULLVP)
1308 		panic("union rmdir: null upper vnode");
1309 
1310 	if (un->un_uppervp != NULLVP) {
1311 		struct vnode *dvp = dun->un_uppervp;
1312 		struct vnode *vp = un->un_uppervp;
1313 
1314 		FIXUP(dun, p);
1315 		dun->un_flags |= UN_KLOCK;
1316 		VOP_UNLOCK(ap->a_dvp, 0, p);
1317 		FIXUP(un, p);
1318 		un->un_flags |= UN_KLOCK;
1319 		VOP_UNLOCK(ap->a_vp, 0, p);
1320 
1321 		if (union_dowhiteout(un, cnp->cn_cred, p))
1322 			cnp->cn_flags |= DOWHITEOUT;
1323 		error = VOP_RMDIR(dvp, vp, ap->a_cnp);
1324 #if 0
1325 		/* XXX */
1326 		if (!error)
1327 			union_removed_upper(un);
1328 #endif
1329 		dun->un_flags |= UN_ULOCK;
1330 		un->un_flags |= UN_ULOCK;
1331 	} else {
1332 		FIXUP(dun, p);
1333 		error = union_mkwhiteout(
1334 			MOUNTTOUNIONMOUNT(UNIONTOV(dun)->v_mount),
1335 			dun->un_uppervp, ap->a_cnp, un->un_path);
1336 	}
1337 
1338 	return (error);
1339 }
1340 
1341 static int
1342 union_symlink(ap)
1343 	struct vop_symlink_args /* {
1344 		struct vnode *a_dvp;
1345 		struct vnode **a_vpp;
1346 		struct componentname *a_cnp;
1347 		struct vattr *a_vap;
1348 		char *a_target;
1349 	} */ *ap;
1350 {
1351 	struct union_node *dun = VTOUNION(ap->a_dvp);
1352 	struct vnode *dvp = dun->un_uppervp;
1353 	struct componentname *cnp = ap->a_cnp;
1354 	struct proc *p = cnp->cn_proc;
1355 
1356 	if (dvp != NULLVP) {
1357 		struct vnode *vp;
1358 		int error;
1359 
1360 		FIXUP(dun, p);
1361 		dun->un_flags |= UN_KLOCK;
1362 		VOP_UNLOCK(ap->a_dvp, 0, p);
1363 		error = VOP_SYMLINK(dvp, &vp, cnp, ap->a_vap, ap->a_target);
1364 		dun->un_flags |= UN_ULOCK;
1365 		*ap->a_vpp = NULLVP;
1366 		return (error);
1367 	}
1368 
1369 	return (EROFS);
1370 }
1371 
1372 /*
1373  * union_readdir works in concert with getdirentries and
1374  * readdir(3) to provide a list of entries in the unioned
1375  * directories.  getdirentries is responsible for walking
1376  * down the union stack.  readdir(3) is responsible for
1377  * eliminating duplicate names from the returned data stream.
1378  */
1379 static int
1380 union_readdir(ap)
1381 	struct vop_readdir_args /* {
1382 		struct vnode *a_vp;
1383 		struct uio *a_uio;
1384 		struct ucred *a_cred;
1385 		int *a_eofflag;
1386 		u_long *a_cookies;
1387 		int a_ncookies;
1388 	} */ *ap;
1389 {
1390 	struct union_node *un = VTOUNION(ap->a_vp);
1391 	struct vnode *uvp = un->un_uppervp;
1392 	struct proc *p = ap->a_uio->uio_procp;
1393 
1394 	if (uvp == NULLVP)
1395 		return (0);
1396 
1397 	FIXUP(un, p);
1398 	ap->a_vp = uvp;
1399 	return (VCALL(uvp, VOFFSET(vop_readdir), ap));
1400 }
1401 
1402 static int
1403 union_readlink(ap)
1404 	struct vop_readlink_args /* {
1405 		struct vnode *a_vp;
1406 		struct uio *a_uio;
1407 		struct ucred *a_cred;
1408 	} */ *ap;
1409 {
1410 	int error;
1411 	struct uio *uio = ap->a_uio;
1412 	struct proc *p = uio->uio_procp;
1413 	struct vnode *vp = OTHERVP(ap->a_vp);
1414 	int dolock = (vp == LOWERVP(ap->a_vp));
1415 
1416 	if (dolock)
1417 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
1418 	else
1419 		FIXUP(VTOUNION(ap->a_vp), p);
1420 	ap->a_vp = vp;
1421 	error = VCALL(vp, VOFFSET(vop_readlink), ap);
1422 	if (dolock)
1423 		VOP_UNLOCK(vp, 0, p);
1424 
1425 	return (error);
1426 }
1427 
1428 static int
1429 union_abortop(ap)
1430 	struct vop_abortop_args /* {
1431 		struct vnode *a_dvp;
1432 		struct componentname *a_cnp;
1433 	} */ *ap;
1434 {
1435 	int error;
1436 	struct componentname *cnp = ap->a_cnp;
1437 	struct proc *p = cnp->cn_proc;
1438 	struct vnode *vp = OTHERVP(ap->a_dvp);
1439 	struct union_node *un = VTOUNION(ap->a_dvp);
1440 	int islocked = un->un_flags & UN_LOCKED;
1441 	int dolock = (vp == LOWERVP(ap->a_dvp));
1442 
1443 	if (islocked) {
1444 		if (dolock)
1445 			vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
1446 		else
1447 			FIXUP(VTOUNION(ap->a_dvp), p);
1448 	}
1449 	ap->a_dvp = vp;
1450 	error = VCALL(vp, VOFFSET(vop_abortop), ap);
1451 	if (islocked && dolock)
1452 		VOP_UNLOCK(vp, 0, p);
1453 
1454 	return (error);
1455 }
1456 
1457 static int
1458 union_inactive(ap)
1459 	struct vop_inactive_args /* {
1460 		struct vnode *a_vp;
1461 		struct proc *a_p;
1462 	} */ *ap;
1463 {
1464 	struct vnode *vp = ap->a_vp;
1465 	struct proc *p = ap->a_p;
1466 	struct union_node *un = VTOUNION(vp);
1467 	struct vnode **vpp;
1468 
1469 	/*
1470 	 * Do nothing (and _don't_ bypass).
1471 	 * Wait to vrele lowervp until reclaim,
1472 	 * so that until then our union_node is in the
1473 	 * cache and reusable.
1474 	 *
1475 	 * NEEDSWORK: Someday, consider inactive'ing
1476 	 * the lowervp and then trying to reactivate it
1477 	 * with capabilities (v_id)
1478 	 * like they do in the name lookup cache code.
1479 	 * That's too much work for now.
1480 	 */
1481 
1482 	if (un->un_dircache != 0) {
1483 		for (vpp = un->un_dircache; *vpp != NULLVP; vpp++)
1484 			vrele(*vpp);
1485 		free(un->un_dircache, M_TEMP);
1486 		un->un_dircache = 0;
1487 	}
1488 
1489 	VOP_UNLOCK(vp, 0, p);
1490 
1491 	if ((un->un_flags & UN_CACHED) == 0)
1492 		vgone(vp);
1493 
1494 	return (0);
1495 }
1496 
1497 static int
1498 union_reclaim(ap)
1499 	struct vop_reclaim_args /* {
1500 		struct vnode *a_vp;
1501 	} */ *ap;
1502 {
1503 
1504 	union_freevp(ap->a_vp);
1505 
1506 	return (0);
1507 }
1508 
1509 static int
1510 union_lock(ap)
1511 	struct vop_lock_args *ap;
1512 {
1513 	struct vnode *vp = ap->a_vp;
1514 	struct proc *p = ap->a_p;
1515 	int flags = ap->a_flags;
1516 	struct union_node *un;
1517 	int error;
1518 
1519 	vop_nolock(ap);
1520 	/*
1521 	 * Need to do real lockmgr-style locking here.
1522 	 * in the mean time, draining won't work quite right,
1523 	 * which could lead to a few race conditions.
1524 	 * the following test was here, but is not quite right, we
1525 	 * still need to take the lock:
1526 	if ((flags & LK_TYPE_MASK) == LK_DRAIN)
1527 		return (0);
1528 	 */
1529 	flags &= ~LK_INTERLOCK;
1530 
1531 start:
1532 	un = VTOUNION(vp);
1533 
1534 	if (un->un_uppervp != NULLVP) {
1535 		if (((un->un_flags & UN_ULOCK) == 0) &&
1536 		    (vp->v_usecount != 0)) {
1537 			error = vn_lock(un->un_uppervp, flags, p);
1538 			if (error)
1539 				return (error);
1540 			un->un_flags |= UN_ULOCK;
1541 		}
1542 #ifdef DIAGNOSTIC
1543 		if (un->un_flags & UN_KLOCK) {
1544 			vprint("dangling upper lock", vp);
1545 			panic("union: dangling upper lock");
1546 		}
1547 #endif
1548 	}
1549 
1550 	if (un->un_flags & UN_LOCKED) {
1551 #ifdef DIAGNOSTIC
1552 		if (curproc && un->un_pid == curproc->p_pid &&
1553 			    un->un_pid > -1 && curproc->p_pid > -1)
1554 			panic("union: locking against myself");
1555 #endif
1556 		un->un_flags |= UN_WANT;
1557 		tsleep((caddr_t)&un->un_flags, PINOD, "unionlk2", 0);
1558 		goto start;
1559 	}
1560 
1561 #ifdef DIAGNOSTIC
1562 	if (curproc)
1563 		un->un_pid = curproc->p_pid;
1564 	else
1565 		un->un_pid = -1;
1566 #endif
1567 
1568 	un->un_flags |= UN_LOCKED;
1569 	return (0);
1570 }
1571 
1572 /*
1573  * When operations want to vput() a union node yet retain a lock on
1574  * the upper vnode (say, to do some further operations like link(),
1575  * mkdir(), ...), they set UN_KLOCK on the union node, then call
1576  * vput() which calls VOP_UNLOCK() and comes here.  union_unlock()
1577  * unlocks the union node (leaving the upper vnode alone), clears the
1578  * KLOCK flag, and then returns to vput().  The caller then does whatever
1579  * is left to do with the upper vnode, and ensures that it gets unlocked.
1580  *
1581  * If UN_KLOCK isn't set, then the upper vnode is unlocked here.
1582  */
1583 static int
1584 union_unlock(ap)
1585 	struct vop_unlock_args /* {
1586 		struct vnode *a_vp;
1587 		int a_flags;
1588 		struct proc *a_p;
1589 	} */ *ap;
1590 {
1591 	struct union_node *un = VTOUNION(ap->a_vp);
1592 	struct proc *p = ap->a_p;
1593 
1594 #ifdef DIAGNOSTIC
1595 	if ((un->un_flags & UN_LOCKED) == 0)
1596 		panic("union: unlock unlocked node");
1597 	if (curproc && un->un_pid != curproc->p_pid &&
1598 			curproc->p_pid > -1 && un->un_pid > -1)
1599 		panic("union: unlocking other process's union node");
1600 #endif
1601 
1602 	un->un_flags &= ~UN_LOCKED;
1603 
1604 	if ((un->un_flags & (UN_ULOCK|UN_KLOCK)) == UN_ULOCK)
1605 		VOP_UNLOCK(un->un_uppervp, 0, p);
1606 
1607 	un->un_flags &= ~(UN_ULOCK|UN_KLOCK);
1608 
1609 	if (un->un_flags & UN_WANT) {
1610 		un->un_flags &= ~UN_WANT;
1611 		wakeup((caddr_t) &un->un_flags);
1612 	}
1613 
1614 #ifdef DIAGNOSTIC
1615 	un->un_pid = 0;
1616 #endif
1617 	vop_nounlock(ap);
1618 
1619 	return (0);
1620 }
1621 
1622 static int
1623 union_bmap(ap)
1624 	struct vop_bmap_args /* {
1625 		struct vnode *a_vp;
1626 		daddr_t  a_bn;
1627 		struct vnode **a_vpp;
1628 		daddr_t *a_bnp;
1629 		int *a_runp;
1630 		int *a_runb;
1631 	} */ *ap;
1632 {
1633 	int error;
1634 	struct proc *p = curproc;		/* XXX */
1635 	struct vnode *vp = OTHERVP(ap->a_vp);
1636 	int dolock = (vp == LOWERVP(ap->a_vp));
1637 
1638 	if (dolock)
1639 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
1640 	else
1641 		FIXUP(VTOUNION(ap->a_vp), p);
1642 	ap->a_vp = vp;
1643 	error = VCALL(vp, VOFFSET(vop_bmap), ap);
1644 	if (dolock)
1645 		VOP_UNLOCK(vp, 0, p);
1646 
1647 	return (error);
1648 }
1649 
1650 static int
1651 union_print(ap)
1652 	struct vop_print_args /* {
1653 		struct vnode *a_vp;
1654 	} */ *ap;
1655 {
1656 	struct vnode *vp = ap->a_vp;
1657 
1658 	printf("\ttag VT_UNION, vp=%p, uppervp=%p, lowervp=%p\n",
1659 			vp, UPPERVP(vp), LOWERVP(vp));
1660 	if (UPPERVP(vp) != NULLVP)
1661 		vprint("union: upper", UPPERVP(vp));
1662 	if (LOWERVP(vp) != NULLVP)
1663 		vprint("union: lower", LOWERVP(vp));
1664 
1665 	return (0);
1666 }
1667 
1668 static int
1669 union_islocked(ap)
1670 	struct vop_islocked_args /* {
1671 		struct vnode *a_vp;
1672 	} */ *ap;
1673 {
1674 
1675 	return ((VTOUNION(ap->a_vp)->un_flags & UN_LOCKED) ? 1 : 0);
1676 }
1677 
1678 static int
1679 union_pathconf(ap)
1680 	struct vop_pathconf_args /* {
1681 		struct vnode *a_vp;
1682 		int a_name;
1683 		int *a_retval;
1684 	} */ *ap;
1685 {
1686 	int error;
1687 	struct proc *p = curproc;		/* XXX */
1688 	struct vnode *vp = OTHERVP(ap->a_vp);
1689 	int dolock = (vp == LOWERVP(ap->a_vp));
1690 
1691 	if (dolock)
1692 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY, p);
1693 	else
1694 		FIXUP(VTOUNION(ap->a_vp), p);
1695 	ap->a_vp = vp;
1696 	error = VCALL(vp, VOFFSET(vop_pathconf), ap);
1697 	if (dolock)
1698 		VOP_UNLOCK(vp, 0, p);
1699 
1700 	return (error);
1701 }
1702 
1703 static int
1704 union_advlock(ap)
1705 	struct vop_advlock_args /* {
1706 		struct vnode *a_vp;
1707 		caddr_t  a_id;
1708 		int  a_op;
1709 		struct flock *a_fl;
1710 		int  a_flags;
1711 	} */ *ap;
1712 {
1713 	register struct vnode *ovp = OTHERVP(ap->a_vp);
1714 
1715 	ap->a_vp = ovp;
1716 	return (VCALL(ovp, VOFFSET(vop_advlock), ap));
1717 }
1718 
1719 
1720 /*
1721  * XXX - vop_strategy must be hand coded because it has no
1722  * vnode in its arguments.
1723  * This goes away with a merged VM/buffer cache.
1724  */
1725 static int
1726 union_strategy(ap)
1727 	struct vop_strategy_args /* {
1728 		struct vnode *a_vp;
1729 		struct buf *a_bp;
1730 	} */ *ap;
1731 {
1732 	struct buf *bp = ap->a_bp;
1733 	int error;
1734 	struct vnode *savedvp;
1735 
1736 	savedvp = bp->b_vp;
1737 	bp->b_vp = OTHERVP(bp->b_vp);
1738 
1739 #ifdef DIAGNOSTIC
1740 	if (bp->b_vp == NULLVP)
1741 		panic("union_strategy: nil vp");
1742 	if (((bp->b_flags & B_READ) == 0) &&
1743 	    (bp->b_vp == LOWERVP(savedvp)))
1744 		panic("union_strategy: writing to lowervp");
1745 #endif
1746 
1747 	error = VOP_STRATEGY(bp->b_vp, bp);
1748 	bp->b_vp = savedvp;
1749 
1750 	return (error);
1751 }
1752 
1753 /*
1754  * Global vfs data structures
1755  */
1756 vop_t **union_vnodeop_p;
1757 static struct vnodeopv_entry_desc union_vnodeop_entries[] = {
1758 	{ &vop_default_desc,		(vop_t *) vop_defaultop },
1759 	{ &vop_abortop_desc,		(vop_t *) union_abortop },
1760 	{ &vop_access_desc,		(vop_t *) union_access },
1761 	{ &vop_advlock_desc,		(vop_t *) union_advlock },
1762 	{ &vop_bmap_desc,		(vop_t *) union_bmap },
1763 	{ &vop_close_desc,		(vop_t *) union_close },
1764 	{ &vop_create_desc,		(vop_t *) union_create },
1765 	{ &vop_fsync_desc,		(vop_t *) union_fsync },
1766 	{ &vop_getattr_desc,		(vop_t *) union_getattr },
1767 	{ &vop_inactive_desc,		(vop_t *) union_inactive },
1768 	{ &vop_ioctl_desc,		(vop_t *) union_ioctl },
1769 	{ &vop_islocked_desc,		(vop_t *) union_islocked },
1770 	{ &vop_lease_desc,		(vop_t *) union_lease },
1771 	{ &vop_link_desc,		(vop_t *) union_link },
1772 	{ &vop_lock_desc,		(vop_t *) union_lock },
1773 	{ &vop_lookup_desc,		(vop_t *) union_lookup },
1774 	{ &vop_mkdir_desc,		(vop_t *) union_mkdir },
1775 	{ &vop_mknod_desc,		(vop_t *) union_mknod },
1776 	{ &vop_mmap_desc,		(vop_t *) union_mmap },
1777 	{ &vop_open_desc,		(vop_t *) union_open },
1778 	{ &vop_pathconf_desc,		(vop_t *) union_pathconf },
1779 	{ &vop_poll_desc,		(vop_t *) union_poll },
1780 	{ &vop_print_desc,		(vop_t *) union_print },
1781 	{ &vop_read_desc,		(vop_t *) union_read },
1782 	{ &vop_readdir_desc,		(vop_t *) union_readdir },
1783 	{ &vop_readlink_desc,		(vop_t *) union_readlink },
1784 	{ &vop_reclaim_desc,		(vop_t *) union_reclaim },
1785 	{ &vop_remove_desc,		(vop_t *) union_remove },
1786 	{ &vop_rename_desc,		(vop_t *) union_rename },
1787 	{ &vop_revoke_desc,		(vop_t *) union_revoke },
1788 	{ &vop_rmdir_desc,		(vop_t *) union_rmdir },
1789 	{ &vop_setattr_desc,		(vop_t *) union_setattr },
1790 	{ &vop_strategy_desc,		(vop_t *) union_strategy },
1791 	{ &vop_symlink_desc,		(vop_t *) union_symlink },
1792 	{ &vop_unlock_desc,		(vop_t *) union_unlock },
1793 	{ &vop_whiteout_desc,		(vop_t *) union_whiteout },
1794 	{ &vop_write_desc,		(vop_t *) union_write },
1795 	{ NULL, NULL }
1796 };
1797 static struct vnodeopv_desc union_vnodeop_opv_desc =
1798 	{ &union_vnodeop_p, union_vnodeop_entries };
1799 
1800 VNODEOP_SET(union_vnodeop_opv_desc);
1801