1/* $NetBSD: union_vnops.c,v 1.63 2015/04/20 23:03:08 riastradh Exp $ */
2
3/*
4 * Copyright (c) 1992, 1993, 1994, 1995
5 * The Regents of the University of California. All rights reserved.
6 *
7 * This code is derived from software contributed to Berkeley by
8 * Jan-Simon Pendry.
9 *
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
12 * are met:
13 * 1. Redistributions of source code must retain the above copyright
14 * notice, this list of conditions and the following disclaimer.
15 * 2. Redistributions in binary form must reproduce the above copyright
16 * notice, this list of conditions and the following disclaimer in the
17 * documentation and/or other materials provided with the distribution.
18 * 3. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
21 *
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 * SUCH DAMAGE.
33 *
34 * @(#)union_vnops.c 8.33 (Berkeley) 7/31/95
35 */
36
37/*
38 * Copyright (c) 1992, 1993, 1994, 1995 Jan-Simon Pendry.
39 *
40 * This code is derived from software contributed to Berkeley by
41 * Jan-Simon Pendry.
42 *
43 * Redistribution and use in source and binary forms, with or without
44 * modification, are permitted provided that the following conditions
45 * are met:
46 * 1. Redistributions of source code must retain the above copyright
47 * notice, this list of conditions and the following disclaimer.
48 * 2. Redistributions in binary form must reproduce the above copyright
49 * notice, this list of conditions and the following disclaimer in the
50 * documentation and/or other materials provided with the distribution.
51 * 3. All advertising materials mentioning features or use of this software
52 * must display the following acknowledgement:
53 * This product includes software developed by the University of
54 * California, Berkeley and its contributors.
55 * 4. Neither the name of the University nor the names of its contributors
56 * may be used to endorse or promote products derived from this software
57 * without specific prior written permission.
58 *
59 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
60 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
61 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
62 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
63 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
64 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
65 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
66 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
67 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
68 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
69 * SUCH DAMAGE.
70 *
71 * @(#)union_vnops.c 8.33 (Berkeley) 7/31/95
72 */
73
74#include <sys/cdefs.h>
75__KERNEL_RCSID(0, "$NetBSD: union_vnops.c,v 1.63 2015/04/20 23:03:08 riastradh Exp $");
76
77#include <sys/param.h>
78#include <sys/systm.h>
79#include <sys/proc.h>
80#include <sys/file.h>
81#include <sys/time.h>
82#include <sys/stat.h>
83#include <sys/vnode.h>
84#include <sys/mount.h>
85#include <sys/namei.h>
86#include <sys/malloc.h>
87#include <sys/buf.h>
88#include <sys/queue.h>
89#include <sys/lock.h>
90#include <sys/kauth.h>
91
92#include <fs/union/union.h>
93#include <miscfs/genfs/genfs.h>
94#include <miscfs/specfs/specdev.h>
95
96int union_lookup(void *);
97int union_create(void *);
98int union_whiteout(void *);
99int union_mknod(void *);
100int union_open(void *);
101int union_close(void *);
102int union_access(void *);
103int union_getattr(void *);
104int union_setattr(void *);
105int union_read(void *);
106int union_write(void *);
107int union_ioctl(void *);
108int union_poll(void *);
109int union_revoke(void *);
110int union_mmap(void *);
111int union_fsync(void *);
112int union_seek(void *);
113int union_remove(void *);
114int union_link(void *);
115int union_rename(void *);
116int union_mkdir(void *);
117int union_rmdir(void *);
118int union_symlink(void *);
119int union_readdir(void *);
120int union_readlink(void *);
121int union_abortop(void *);
122int union_inactive(void *);
123int union_reclaim(void *);
124int union_lock(void *);
125int union_unlock(void *);
126int union_bmap(void *);
127int union_print(void *);
128int union_islocked(void *);
129int union_pathconf(void *);
130int union_advlock(void *);
131int union_strategy(void *);
132int union_bwrite(void *);
133int union_getpages(void *);
134int union_putpages(void *);
135int union_kqfilter(void *);
136
137static int union_lookup1(struct vnode *, struct vnode **,
138 struct vnode **, struct componentname *);
139
140
141/*
142 * Global vfs data structures
143 */
144int (**union_vnodeop_p)(void *);
145const struct vnodeopv_entry_desc union_vnodeop_entries[] = {
146 { &vop_default_desc, vn_default_error },
147 { &vop_lookup_desc, union_lookup }, /* lookup */
148 { &vop_create_desc, union_create }, /* create */
149 { &vop_whiteout_desc, union_whiteout }, /* whiteout */
150 { &vop_mknod_desc, union_mknod }, /* mknod */
151 { &vop_open_desc, union_open }, /* open */
152 { &vop_close_desc, union_close }, /* close */
153 { &vop_access_desc, union_access }, /* access */
154 { &vop_getattr_desc, union_getattr }, /* getattr */
155 { &vop_setattr_desc, union_setattr }, /* setattr */
156 { &vop_read_desc, union_read }, /* read */
157 { &vop_write_desc, union_write }, /* write */
158 { &vop_fallocate_desc, genfs_eopnotsupp }, /* fallocate */
159 { &vop_fdiscard_desc, genfs_eopnotsupp }, /* fdiscard */
160 { &vop_ioctl_desc, union_ioctl }, /* ioctl */
161 { &vop_poll_desc, union_poll }, /* select */
162 { &vop_revoke_desc, union_revoke }, /* revoke */
163 { &vop_mmap_desc, union_mmap }, /* mmap */
164 { &vop_fsync_desc, union_fsync }, /* fsync */
165 { &vop_seek_desc, union_seek }, /* seek */
166 { &vop_remove_desc, union_remove }, /* remove */
167 { &vop_link_desc, union_link }, /* link */
168 { &vop_rename_desc, union_rename }, /* rename */
169 { &vop_mkdir_desc, union_mkdir }, /* mkdir */
170 { &vop_rmdir_desc, union_rmdir }, /* rmdir */
171 { &vop_symlink_desc, union_symlink }, /* symlink */
172 { &vop_readdir_desc, union_readdir }, /* readdir */
173 { &vop_readlink_desc, union_readlink }, /* readlink */
174 { &vop_abortop_desc, union_abortop }, /* abortop */
175 { &vop_inactive_desc, union_inactive }, /* inactive */
176 { &vop_reclaim_desc, union_reclaim }, /* reclaim */
177 { &vop_lock_desc, union_lock }, /* lock */
178 { &vop_unlock_desc, union_unlock }, /* unlock */
179 { &vop_bmap_desc, union_bmap }, /* bmap */
180 { &vop_strategy_desc, union_strategy }, /* strategy */
181 { &vop_bwrite_desc, union_bwrite }, /* bwrite */
182 { &vop_print_desc, union_print }, /* print */
183 { &vop_islocked_desc, union_islocked }, /* islocked */
184 { &vop_pathconf_desc, union_pathconf }, /* pathconf */
185 { &vop_advlock_desc, union_advlock }, /* advlock */
186 { &vop_getpages_desc, union_getpages }, /* getpages */
187 { &vop_putpages_desc, union_putpages }, /* putpages */
188 { &vop_kqfilter_desc, union_kqfilter }, /* kqfilter */
189 { NULL, NULL }
190};
191const struct vnodeopv_desc union_vnodeop_opv_desc =
192 { &union_vnodeop_p, union_vnodeop_entries };
193
194#define NODE_IS_SPECIAL(vp) \
195 ((vp)->v_type == VBLK || (vp)->v_type == VCHR || \
196 (vp)->v_type == VSOCK || (vp)->v_type == VFIFO)
197
198static int
199union_lookup1(struct vnode *udvp, struct vnode **dvpp, struct vnode **vpp,
200 struct componentname *cnp)
201{
202 int error;
203 struct vnode *tdvp;
204 struct vnode *dvp;
205 struct mount *mp;
206
207 dvp = *dvpp;
208
209 /*
210 * If stepping up the directory tree, check for going
211 * back across the mount point, in which case do what
212 * lookup would do by stepping back down the mount
213 * hierarchy.
214 */
215 if (cnp->cn_flags & ISDOTDOT) {
216 while ((dvp != udvp) && (dvp->v_vflag & VV_ROOT)) {
217 /*
218 * Don't do the NOCROSSMOUNT check
219 * at this level. By definition,
220 * union fs deals with namespaces, not
221 * filesystems.
222 */
223 tdvp = dvp;
224 *dvpp = dvp = dvp->v_mount->mnt_vnodecovered;
225 VOP_UNLOCK(tdvp);
226 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
227 }
228 }
229
230 error = VOP_LOOKUP(dvp, &tdvp, cnp);
231 if (error)
232 return (error);
233 if (dvp != tdvp) {
234 if (cnp->cn_flags & ISDOTDOT)
235 VOP_UNLOCK(dvp);
236 error = vn_lock(tdvp, LK_EXCLUSIVE);
237 if (cnp->cn_flags & ISDOTDOT)
238 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
239 if (error) {
240 vrele(tdvp);
241 return error;
242 }
243 dvp = tdvp;
244 }
245
246 /*
247 * Lastly check if the current node is a mount point in
248 * which case walk up the mount hierarchy making sure not to
249 * bump into the root of the mount tree (ie. dvp != udvp).
250 */
251 while (dvp != udvp && (dvp->v_type == VDIR) &&
252 (mp = dvp->v_mountedhere)) {
253 if (vfs_busy(mp, NULL))
254 continue;
255 vput(dvp);
256 error = VFS_ROOT(mp, &tdvp);
257 vfs_unbusy(mp, false, NULL);
258 if (error) {
259 return (error);
260 }
261 dvp = tdvp;
262 }
263
264 *vpp = dvp;
265 return (0);
266}
267
268int
269union_lookup(void *v)
270{
271 struct vop_lookup_v2_args /* {
272 struct vnodeop_desc *a_desc;
273 struct vnode *a_dvp;
274 struct vnode **a_vpp;
275 struct componentname *a_cnp;
276 } */ *ap = v;
277 int error;
278 int uerror, lerror;
279 struct vnode *uppervp, *lowervp;
280 struct vnode *upperdvp, *lowerdvp;
281 struct vnode *dvp = ap->a_dvp;
282 struct union_node *dun = VTOUNION(dvp);
283 struct componentname *cnp = ap->a_cnp;
284 struct union_mount *um = MOUNTTOUNIONMOUNT(dvp->v_mount);
285 kauth_cred_t saved_cred = NULL;
286 int iswhiteout;
287 struct vattr va;
288
289#ifdef notyet
290 if (cnp->cn_namelen == 3 &&
291 cnp->cn_nameptr[2] == '.' &&
292 cnp->cn_nameptr[1] == '.' &&
293 cnp->cn_nameptr[0] == '.') {
294 dvp = *ap->a_vpp = LOWERVP(ap->a_dvp);
295 if (dvp == NULLVP)
296 return (ENOENT);
297 vref(dvp);
298 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
299 return (0);
300 }
301#endif
302
303 if ((cnp->cn_flags & ISLASTCN) &&
304 (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
305 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
306 return (EROFS);
307
308start:
309 upperdvp = dun->un_uppervp;
310 lowerdvp = dun->un_lowervp;
311 uppervp = NULLVP;
312 lowervp = NULLVP;
313 iswhiteout = 0;
314
315 /*
316 * do the lookup in the upper level.
317 * if that level comsumes additional pathnames,
318 * then assume that something special is going
319 * on and just return that vnode.
320 */
321 if (upperdvp != NULLVP) {
322 uerror = union_lookup1(um->um_uppervp, &upperdvp,
323 &uppervp, cnp);
324 if (cnp->cn_consume != 0) {
325 if (uppervp != upperdvp)
326 VOP_UNLOCK(uppervp);
327 *ap->a_vpp = uppervp;
328 return (uerror);
329 }
330 if (uerror == ENOENT || uerror == EJUSTRETURN) {
331 if (cnp->cn_flags & ISWHITEOUT) {
332 iswhiteout = 1;
333 } else if (lowerdvp != NULLVP) {
334 lerror = VOP_GETATTR(upperdvp, &va,
335 cnp->cn_cred);
336 if (lerror == 0 && (va.va_flags & OPAQUE))
337 iswhiteout = 1;
338 }
339 }
340 } else {
341 uerror = ENOENT;
342 }
343
344 /*
345 * in a similar way to the upper layer, do the lookup
346 * in the lower layer. this time, if there is some
347 * component magic going on, then vput whatever we got
348 * back from the upper layer and return the lower vnode
349 * instead.
350 */
351 if (lowerdvp != NULLVP && !iswhiteout) {
352 int nameiop;
353
354 vn_lock(lowerdvp, LK_EXCLUSIVE | LK_RETRY);
355
356 /*
357 * Only do a LOOKUP on the bottom node, since
358 * we won't be making changes to it anyway.
359 */
360 nameiop = cnp->cn_nameiop;
361 cnp->cn_nameiop = LOOKUP;
362 if (um->um_op == UNMNT_BELOW) {
363 saved_cred = cnp->cn_cred;
364 cnp->cn_cred = um->um_cred;
365 }
366
367 /*
368 * we shouldn't have to worry about locking interactions
369 * between the lower layer and our union layer (w.r.t.
370 * `..' processing) because we don't futz with lowervp
371 * locks in the union-node instantiation code path.
372 */
373 lerror = union_lookup1(um->um_lowervp, &lowerdvp,
374 &lowervp, cnp);
375 if (um->um_op == UNMNT_BELOW)
376 cnp->cn_cred = saved_cred;
377 cnp->cn_nameiop = nameiop;
378
379 if (lowervp != lowerdvp)
380 VOP_UNLOCK(lowerdvp);
381
382 if (cnp->cn_consume != 0) {
383 if (uppervp != NULLVP) {
384 if (uppervp == upperdvp)
385 vrele(uppervp);
386 else
387 vput(uppervp);
388 uppervp = NULLVP;
389 }
390 *ap->a_vpp = lowervp;
391 return (lerror);
392 }
393 } else {
394 lerror = ENOENT;
395 if ((cnp->cn_flags & ISDOTDOT) && dun->un_pvp != NULLVP) {
396 lowervp = LOWERVP(dun->un_pvp);
397 if (lowervp != NULLVP) {
398 vref(lowervp);
399 vn_lock(lowervp, LK_EXCLUSIVE | LK_RETRY);
400 lerror = 0;
401 }
402 }
403 }
404
405 /*
406 * EJUSTRETURN is used by underlying filesystems to indicate that
407 * a directory modification op was started successfully.
408 * This will only happen in the upper layer, since
409 * the lower layer only does LOOKUPs.
410 * If this union is mounted read-only, bounce it now.
411 */
412
413 if ((uerror == EJUSTRETURN) && (cnp->cn_flags & ISLASTCN) &&
414 (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
415 ((cnp->cn_nameiop == CREATE) || (cnp->cn_nameiop == RENAME)))
416 uerror = EROFS;
417
418 /*
419 * at this point, we have uerror and lerror indicating
420 * possible errors with the lookups in the upper and lower
421 * layers. additionally, uppervp and lowervp are (locked)
422 * references to existing vnodes in the upper and lower layers.
423 *
424 * there are now three cases to consider.
425 * 1. if both layers returned an error, then return whatever
426 * error the upper layer generated.
427 *
428 * 2. if the top layer failed and the bottom layer succeeded
429 * then two subcases occur.
430 * a. the bottom vnode is not a directory, in which
431 * case just return a new union vnode referencing
432 * an empty top layer and the existing bottom layer.
433 * b. the bottom vnode is a directory, in which case
434 * create a new directory in the top-level and
435 * continue as in case 3.
436 *
437 * 3. if the top layer succeeded then return a new union
438 * vnode referencing whatever the new top layer and
439 * whatever the bottom layer returned.
440 */
441
442 *ap->a_vpp = NULLVP;
443
444
445 /* case 1. */
446 if ((uerror != 0) && (lerror != 0)) {
447 return (uerror);
448 }
449
450 /* case 2. */
451 if (uerror != 0 /* && (lerror == 0) */ ) {
452 if (lowervp->v_type == VDIR) { /* case 2b. */
453 /*
454 * We may be racing another process to make the
455 * upper-level shadow directory. Be careful with
456 * locks/etc!
457 * If we have to create a shadow directory and want
458 * to commit the node we have to restart the lookup
459 * to get the componentname right.
460 */
461 if (upperdvp) {
462 VOP_UNLOCK(upperdvp);
463 uerror = union_mkshadow(um, upperdvp, cnp,
464 &uppervp);
465 vn_lock(upperdvp, LK_EXCLUSIVE | LK_RETRY);
466 if (uerror == 0 && cnp->cn_nameiop != LOOKUP) {
467 vrele(uppervp);
468 if (lowervp != NULLVP)
469 vput(lowervp);
470 goto start;
471 }
472 }
473 if (uerror) {
474 if (lowervp != NULLVP) {
475 vput(lowervp);
476 lowervp = NULLVP;
477 }
478 return (uerror);
479 }
480 }
481 } else { /* uerror == 0 */
482 if (uppervp != upperdvp)
483 VOP_UNLOCK(uppervp);
484 }
485
486 if (lowervp != NULLVP)
487 VOP_UNLOCK(lowervp);
488
489 error = union_allocvp(ap->a_vpp, dvp->v_mount, dvp, upperdvp, cnp,
490 uppervp, lowervp, 1);
491
492 if (error) {
493 if (uppervp != NULLVP)
494 vrele(uppervp);
495 if (lowervp != NULLVP)
496 vrele(lowervp);
497 return error;
498 }
499
500 return 0;
501}
502
503int
504union_create(void *v)
505{
506 struct vop_create_v3_args /* {
507 struct vnode *a_dvp;
508 struct vnode **a_vpp;
509 struct componentname *a_cnp;
510 struct vattr *a_vap;
511 } */ *ap = v;
512 struct union_node *un = VTOUNION(ap->a_dvp);
513 struct vnode *dvp = un->un_uppervp;
514 struct componentname *cnp = ap->a_cnp;
515
516 if (dvp != NULLVP) {
517 int error;
518 struct vnode *vp;
519 struct mount *mp;
520
521 mp = ap->a_dvp->v_mount;
522
523 vp = NULL;
524 error = VOP_CREATE(dvp, &vp, cnp, ap->a_vap);
525 if (error)
526 return (error);
527
528 error = union_allocvp(ap->a_vpp, mp, NULLVP, NULLVP, cnp, vp,
529 NULLVP, 1);
530 if (error)
531 vrele(vp);
532 return (error);
533 }
534
535 return (EROFS);
536}
537
538int
539union_whiteout(void *v)
540{
541 struct vop_whiteout_args /* {
542 struct vnode *a_dvp;
543 struct componentname *a_cnp;
544 int a_flags;
545 } */ *ap = v;
546 struct union_node *un = VTOUNION(ap->a_dvp);
547 struct componentname *cnp = ap->a_cnp;
548
549 if (un->un_uppervp == NULLVP)
550 return (EOPNOTSUPP);
551
552 return (VOP_WHITEOUT(un->un_uppervp, cnp, ap->a_flags));
553}
554
555int
556union_mknod(void *v)
557{
558 struct vop_mknod_v3_args /* {
559 struct vnode *a_dvp;
560 struct vnode **a_vpp;
561 struct componentname *a_cnp;
562 struct vattr *a_vap;
563 } */ *ap = v;
564 struct union_node *un = VTOUNION(ap->a_dvp);
565 struct vnode *dvp = un->un_uppervp;
566 struct componentname *cnp = ap->a_cnp;
567
568 if (dvp != NULLVP) {
569 int error;
570 struct vnode *vp;
571 struct mount *mp;
572
573 mp = ap->a_dvp->v_mount;
574 error = VOP_MKNOD(dvp, &vp, cnp, ap->a_vap);
575 if (error)
576 return (error);
577
578 error = union_allocvp(ap->a_vpp, mp, NULLVP, NULLVP,
579 cnp, vp, NULLVP, 1);
580 if (error)
581 vrele(vp);
582 return (error);
583 }
584
585 return (EROFS);
586}
587
588int
589union_open(void *v)
590{
591 struct vop_open_args /* {
592 struct vnodeop_desc *a_desc;
593 struct vnode *a_vp;
594 int a_mode;
595 kauth_cred_t a_cred;
596 } */ *ap = v;
597 struct union_node *un = VTOUNION(ap->a_vp);
598 struct vnode *tvp;
599 int mode = ap->a_mode;
600 kauth_cred_t cred = ap->a_cred;
601 struct lwp *l = curlwp;
602 int error;
603
604 /*
605 * If there is an existing upper vp then simply open that.
606 */
607 tvp = un->un_uppervp;
608 if (tvp == NULLVP) {
609 /*
610 * If the lower vnode is being opened for writing, then
611 * copy the file contents to the upper vnode and open that,
612 * otherwise can simply open the lower vnode.
613 */
614 tvp = un->un_lowervp;
615 if ((ap->a_mode & FWRITE) && (tvp->v_type == VREG)) {
616 error = union_copyup(un, (mode&O_TRUNC) == 0, cred, l);
617 if (error == 0)
618 error = VOP_OPEN(un->un_uppervp, mode, cred);
619 return (error);
620 }
621
622 /*
623 * Just open the lower vnode, but check for nodev mount flag
624 */
625 if ((tvp->v_type == VBLK || tvp->v_type == VCHR) &&
626 (ap->a_vp->v_mount->mnt_flag & MNT_NODEV))
627 return ENXIO;
628 un->un_openl++;
629 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
630 error = VOP_OPEN(tvp, mode, cred);
631 VOP_UNLOCK(tvp);
632
633 return (error);
634 }
635 /*
636 * Just open the upper vnode, checking for nodev mount flag first
637 */
638 if ((tvp->v_type == VBLK || tvp->v_type == VCHR) &&
639 (ap->a_vp->v_mount->mnt_flag & MNT_NODEV))
640 return ENXIO;
641
642 error = VOP_OPEN(tvp, mode, cred);
643
644 return (error);
645}
646
647int
648union_close(void *v)
649{
650 struct vop_close_args /* {
651 struct vnode *a_vp;
652 int a_fflag;
653 kauth_cred_t a_cred;
654 } */ *ap = v;
655 struct union_node *un = VTOUNION(ap->a_vp);
656 struct vnode *vp;
657 int error;
658 bool do_lock;
659
660 vp = un->un_uppervp;
661 if (vp != NULLVP) {
662 do_lock = false;
663 } else {
664 KASSERT(un->un_openl > 0);
665 --un->un_openl;
666 vp = un->un_lowervp;
667 do_lock = true;
668 }
669
670 KASSERT(vp != NULLVP);
671 ap->a_vp = vp;
672 if (do_lock)
673 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
674 error = VCALL(vp, VOFFSET(vop_close), ap);
675 if (do_lock)
676 VOP_UNLOCK(vp);
677
678 return error;
679}
680
681/*
682 * Check access permission on the union vnode.
683 * The access check being enforced is to check
684 * against both the underlying vnode, and any
685 * copied vnode. This ensures that no additional
686 * file permissions are given away simply because
687 * the user caused an implicit file copy.
688 */
689int
690union_access(void *v)
691{
692 struct vop_access_args /* {
693 struct vnodeop_desc *a_desc;
694 struct vnode *a_vp;
695 int a_mode;
696 kauth_cred_t a_cred;
697 } */ *ap = v;
698 struct vnode *vp = ap->a_vp;
699 struct union_node *un = VTOUNION(vp);
700 int error = EACCES;
701 struct union_mount *um = MOUNTTOUNIONMOUNT(vp->v_mount);
702
703 /*
704 * Disallow write attempts on read-only file systems;
705 * unless the file is a socket, fifo, or a block or
706 * character device resident on the file system.
707 */
708 if (ap->a_mode & VWRITE) {
709 switch (vp->v_type) {
710 case VDIR:
711 case VLNK:
712 case VREG:
713 if (vp->v_mount->mnt_flag & MNT_RDONLY)
714 return (EROFS);
715 break;
716 case VBAD:
717 case VBLK:
718 case VCHR:
719 case VSOCK:
720 case VFIFO:
721 case VNON:
722 default:
723 break;
724 }
725 }
726
727
728 if ((vp = un->un_uppervp) != NULLVP) {
729 ap->a_vp = vp;
730 return (VCALL(vp, VOFFSET(vop_access), ap));
731 }
732
733 if ((vp = un->un_lowervp) != NULLVP) {
734 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
735 ap->a_vp = vp;
736 error = VCALL(vp, VOFFSET(vop_access), ap);
737 if (error == 0) {
738 if (um->um_op == UNMNT_BELOW) {
739 ap->a_cred = um->um_cred;
740 error = VCALL(vp, VOFFSET(vop_access), ap);
741 }
742 }
743 VOP_UNLOCK(vp);
744 if (error)
745 return (error);
746 }
747
748 return (error);
749}
750
751/*
752 * We handle getattr only to change the fsid and
753 * track object sizes
754 */
755int
756union_getattr(void *v)
757{
758 struct vop_getattr_args /* {
759 struct vnode *a_vp;
760 struct vattr *a_vap;
761 kauth_cred_t a_cred;
762 } */ *ap = v;
763 int error;
764 struct union_node *un = VTOUNION(ap->a_vp);
765 struct vnode *vp = un->un_uppervp;
766 struct vattr *vap;
767 struct vattr va;
768
769
770 /*
771 * Some programs walk the filesystem hierarchy by counting
772 * links to directories to avoid stat'ing all the time.
773 * This means the link count on directories needs to be "correct".
774 * The only way to do that is to call getattr on both layers
775 * and fix up the link count. The link count will not necessarily
776 * be accurate but will be large enough to defeat the tree walkers.
777 *
778 * To make life more interesting, some filesystems don't keep
779 * track of link counts in the expected way, and return a
780 * link count of `1' for those directories; if either of the
781 * component directories returns a link count of `1', we return a 1.
782 */
783
784 vap = ap->a_vap;
785
786 vp = un->un_uppervp;
787 if (vp != NULLVP) {
788 error = VOP_GETATTR(vp, vap, ap->a_cred);
789 if (error)
790 return (error);
791 mutex_enter(&un->un_lock);
792 union_newsize(ap->a_vp, vap->va_size, VNOVAL);
793 }
794
795 if (vp == NULLVP) {
796 vp = un->un_lowervp;
797 } else if (vp->v_type == VDIR) {
798 vp = un->un_lowervp;
799 if (vp != NULLVP)
800 vap = &va;
801 } else {
802 vp = NULLVP;
803 }
804
805 if (vp != NULLVP) {
806 if (vp == un->un_lowervp)
807 vn_lock(vp, LK_SHARED | LK_RETRY);
808 error = VOP_GETATTR(vp, vap, ap->a_cred);
809 if (vp == un->un_lowervp)
810 VOP_UNLOCK(vp);
811 if (error)
812 return (error);
813 mutex_enter(&un->un_lock);
814 union_newsize(ap->a_vp, VNOVAL, vap->va_size);
815 }
816
817 if ((vap != ap->a_vap) && (vap->va_type == VDIR)) {
818 /*
819 * Link count manipulation:
820 * - If both return "2", return 2 (no subdirs)
821 * - If one or the other return "1", return "1" (ENOCLUE)
822 */
823 if ((ap->a_vap->va_nlink == 2) &&
824 (vap->va_nlink == 2))
825 ;
826 else if (ap->a_vap->va_nlink != 1) {
827 if (vap->va_nlink == 1)
828 ap->a_vap->va_nlink = 1;
829 else
830 ap->a_vap->va_nlink += vap->va_nlink;
831 }
832 }
833 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
834 return (0);
835}
836
837int
838union_setattr(void *v)
839{
840 struct vop_setattr_args /* {
841 struct vnode *a_vp;
842 struct vattr *a_vap;
843 kauth_cred_t a_cred;
844 } */ *ap = v;
845 struct vattr *vap = ap->a_vap;
846 struct vnode *vp = ap->a_vp;
847 struct union_node *un = VTOUNION(vp);
848 bool size_only; /* All but va_size are VNOVAL. */
849 int error;
850
851 size_only = (vap->va_flags == VNOVAL && vap->va_uid == (uid_t)VNOVAL &&
852 vap->va_gid == (gid_t)VNOVAL && vap->va_atime.tv_sec == VNOVAL &&
853 vap->va_mtime.tv_sec == VNOVAL && vap->va_mode == (mode_t)VNOVAL);
854
855 if (!size_only && (vp->v_mount->mnt_flag & MNT_RDONLY))
856 return (EROFS);
857 if (vap->va_size != VNOVAL) {
858 switch (vp->v_type) {
859 case VDIR:
860 return (EISDIR);
861 case VCHR:
862 case VBLK:
863 case VSOCK:
864 case VFIFO:
865 break;
866 case VREG:
867 case VLNK:
868 default:
869 /*
870 * Disallow write attempts if the filesystem is
871 * mounted read-only.
872 */
873 if (vp->v_mount->mnt_flag & MNT_RDONLY)
874 return (EROFS);
875 }
876 }
877
878 /*
879 * Handle case of truncating lower object to zero size,
880 * by creating a zero length upper object. This is to
881 * handle the case of open with O_TRUNC and O_CREAT.
882 */
883 if ((un->un_uppervp == NULLVP) &&
884 /* assert(un->un_lowervp != NULLVP) */
885 (un->un_lowervp->v_type == VREG)) {
886 error = union_copyup(un, (vap->va_size != 0),
887 ap->a_cred, curlwp);
888 if (error)
889 return (error);
890 }
891
892 /*
893 * Try to set attributes in upper layer, ignore size change to zero
894 * for devices to handle O_TRUNC and return read-only filesystem error
895 * otherwise.
896 */
897 if (un->un_uppervp != NULLVP) {
898 error = VOP_SETATTR(un->un_uppervp, vap, ap->a_cred);
899 if ((error == 0) && (vap->va_size != VNOVAL)) {
900 mutex_enter(&un->un_lock);
901 union_newsize(ap->a_vp, vap->va_size, VNOVAL);
902 }
903 } else {
904 KASSERT(un->un_lowervp != NULLVP);
905 if (NODE_IS_SPECIAL(un->un_lowervp)) {
906 if (size_only &&
907 (vap->va_size == 0 || vap->va_size == VNOVAL))
908 error = 0;
909 else
910 error = EROFS;
911 } else {
912 error = EROFS;
913 }
914 }
915
916 return (error);
917}
918
919int
920union_read(void *v)
921{
922 struct vop_read_args /* {
923 struct vnode *a_vp;
924 struct uio *a_uio;
925 int a_ioflag;
926 kauth_cred_t a_cred;
927 } */ *ap = v;
928 int error;
929 struct vnode *vp = OTHERVP(ap->a_vp);
930 int dolock = (vp == LOWERVP(ap->a_vp));
931
932 if (dolock)
933 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
934 error = VOP_READ(vp, ap->a_uio, ap->a_ioflag, ap->a_cred);
935 if (dolock)
936 VOP_UNLOCK(vp);
937
938 /*
939 * XXX
940 * perhaps the size of the underlying object has changed under
941 * our feet. take advantage of the offset information present
942 * in the uio structure.
943 */
944 if (error == 0) {
945 struct union_node *un = VTOUNION(ap->a_vp);
946 off_t cur = ap->a_uio->uio_offset;
947 off_t usz = VNOVAL, lsz = VNOVAL;
948
949 mutex_enter(&un->un_lock);
950 if (vp == un->un_uppervp) {
951 if (cur > un->un_uppersz)
952 usz = cur;
953 } else {
954 if (cur > un->un_lowersz)
955 lsz = cur;
956 }
957
958 if (usz != VNOVAL || lsz != VNOVAL)
959 union_newsize(ap->a_vp, usz, lsz);
960 else
961 mutex_exit(&un->un_lock);
962 }
963
964 return (error);
965}
966
967int
968union_write(void *v)
969{
970 struct vop_read_args /* {
971 struct vnode *a_vp;
972 struct uio *a_uio;
973 int a_ioflag;
974 kauth_cred_t a_cred;
975 } */ *ap = v;
976 int error;
977 struct vnode *vp;
978 struct union_node *un = VTOUNION(ap->a_vp);
979
980 vp = UPPERVP(ap->a_vp);
981 if (vp == NULLVP) {
982 vp = LOWERVP(ap->a_vp);
983 if (NODE_IS_SPECIAL(vp)) {
984 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
985 error = VOP_WRITE(vp, ap->a_uio, ap->a_ioflag,
986 ap->a_cred);
987 VOP_UNLOCK(vp);
988 return error;
989 }
990 panic("union: missing upper layer in write");
991 }
992
993 error = VOP_WRITE(vp, ap->a_uio, ap->a_ioflag, ap->a_cred);
994
995 /*
996 * the size of the underlying object may be changed by the
997 * write.
998 */
999 if (error == 0) {
1000 off_t cur = ap->a_uio->uio_offset;
1001
1002 mutex_enter(&un->un_lock);
1003 if (cur > un->un_uppersz)
1004 union_newsize(ap->a_vp, cur, VNOVAL);
1005 else
1006 mutex_exit(&un->un_lock);
1007 }
1008
1009 return (error);
1010}
1011
1012int
1013union_ioctl(void *v)
1014{
1015 struct vop_ioctl_args /* {
1016 struct vnode *a_vp;
1017 int a_command;
1018 void *a_data;
1019 int a_fflag;
1020 kauth_cred_t a_cred;
1021 } */ *ap = v;
1022 struct vnode *ovp = OTHERVP(ap->a_vp);
1023
1024 ap->a_vp = ovp;
1025 return (VCALL(ovp, VOFFSET(vop_ioctl), ap));
1026}
1027
1028int
1029union_poll(void *v)
1030{
1031 struct vop_poll_args /* {
1032 struct vnode *a_vp;
1033 int a_events;
1034 } */ *ap = v;
1035 struct vnode *ovp = OTHERVP(ap->a_vp);
1036
1037 ap->a_vp = ovp;
1038 return (VCALL(ovp, VOFFSET(vop_poll), ap));
1039}
1040
1041int
1042union_revoke(void *v)
1043{
1044 struct vop_revoke_args /* {
1045 struct vnode *a_vp;
1046 int a_flags;
1047 struct proc *a_p;
1048 } */ *ap = v;
1049 struct vnode *vp = ap->a_vp;
1050
1051 if (UPPERVP(vp))
1052 VOP_REVOKE(UPPERVP(vp), ap->a_flags);
1053 if (LOWERVP(vp))
1054 VOP_REVOKE(LOWERVP(vp), ap->a_flags);
1055 vgone(vp); /* XXXAD?? */
1056 return (0);
1057}
1058
1059int
1060union_mmap(void *v)
1061{
1062 struct vop_mmap_args /* {
1063 struct vnode *a_vp;
1064 vm_prot_t a_prot;
1065 kauth_cred_t a_cred;
1066 } */ *ap = v;
1067 struct vnode *ovp = OTHERVP(ap->a_vp);
1068
1069 ap->a_vp = ovp;
1070 return (VCALL(ovp, VOFFSET(vop_mmap), ap));
1071}
1072
1073int
1074union_fsync(void *v)
1075{
1076 struct vop_fsync_args /* {
1077 struct vnode *a_vp;
1078 kauth_cred_t a_cred;
1079 int a_flags;
1080 off_t offhi;
1081 off_t offlo;
1082 } */ *ap = v;
1083 int error = 0;
1084 struct vnode *targetvp;
1085
1086 /*
1087 * If vinvalbuf is calling us, it's a "shallow fsync" -- don't
1088 * bother syncing the underlying vnodes, since (a) they'll be
1089 * fsync'ed when reclaimed and (b) we could deadlock if
1090 * they're locked; otherwise, pass it through to the
1091 * underlying layer.
1092 */
1093 if (ap->a_vp->v_type == VBLK || ap->a_vp->v_type == VCHR) {
1094 error = spec_fsync(v);
1095 if (error)
1096 return error;
1097 }
1098
1099 if (ap->a_flags & FSYNC_RECLAIM)
1100 return 0;
1101
1102 targetvp = OTHERVP(ap->a_vp);
1103 if (targetvp != NULLVP) {
1104 int dolock = (targetvp == LOWERVP(ap->a_vp));
1105
1106 if (dolock)
1107 vn_lock(targetvp, LK_EXCLUSIVE | LK_RETRY);
1108 error = VOP_FSYNC(targetvp, ap->a_cred, ap->a_flags,
1109 ap->a_offlo, ap->a_offhi);
1110 if (dolock)
1111 VOP_UNLOCK(targetvp);
1112 }
1113
1114 return (error);
1115}
1116
1117int
1118union_seek(void *v)
1119{
1120 struct vop_seek_args /* {
1121 struct vnode *a_vp;
1122 off_t a_oldoff;
1123 off_t a_newoff;
1124 kauth_cred_t a_cred;
1125 } */ *ap = v;
1126 struct vnode *ovp = OTHERVP(ap->a_vp);
1127
1128 ap->a_vp = ovp;
1129 return (VCALL(ovp, VOFFSET(vop_seek), ap));
1130}
1131
1132int
1133union_remove(void *v)
1134{
1135 struct vop_remove_args /* {
1136 struct vnode *a_dvp;
1137 struct vnode *a_vp;
1138 struct componentname *a_cnp;
1139 } */ *ap = v;
1140 int error;
1141 struct union_node *dun = VTOUNION(ap->a_dvp);
1142 struct union_node *un = VTOUNION(ap->a_vp);
1143 struct componentname *cnp = ap->a_cnp;
1144
1145 if (dun->un_uppervp == NULLVP)
1146 panic("union remove: null upper vnode");
1147
1148 if (un->un_uppervp != NULLVP) {
1149 struct vnode *dvp = dun->un_uppervp;
1150 struct vnode *vp = un->un_uppervp;
1151
1152 /*
1153 * Account for VOP_REMOVE to vrele dvp and vp.
1154 * Note: VOP_REMOVE will unlock dvp and vp.
1155 */
1156 vref(dvp);
1157 vref(vp);
1158 if (union_dowhiteout(un, cnp->cn_cred))
1159 cnp->cn_flags |= DOWHITEOUT;
1160 error = VOP_REMOVE(dvp, vp, cnp);
1161 if (!error)
1162 union_removed_upper(un);
1163 vrele(ap->a_dvp);
1164 vrele(ap->a_vp);
1165 } else {
1166 error = union_mkwhiteout(
1167 MOUNTTOUNIONMOUNT(UNIONTOV(dun)->v_mount),
1168 dun->un_uppervp, ap->a_cnp, un);
1169 vput(ap->a_dvp);
1170 vput(ap->a_vp);
1171 }
1172
1173 return (error);
1174}
1175
1176int
1177union_link(void *v)
1178{
1179 struct vop_link_v2_args /* {
1180 struct vnode *a_dvp;
1181 struct vnode *a_vp;
1182 struct componentname *a_cnp;
1183 } */ *ap = v;
1184 int error = 0;
1185 struct componentname *cnp = ap->a_cnp;
1186 struct union_node *dun;
1187 struct vnode *vp;
1188 struct vnode *dvp;
1189
1190 dun = VTOUNION(ap->a_dvp);
1191
1192 KASSERT((ap->a_cnp->cn_flags & LOCKPARENT) != 0);
1193
1194 if (ap->a_dvp->v_op != ap->a_vp->v_op) {
1195 vp = ap->a_vp;
1196 } else {
1197 struct union_node *un = VTOUNION(ap->a_vp);
1198 if (un->un_uppervp == NULLVP) {
1199 const bool droplock = (dun->un_uppervp == un->un_dirvp);
1200
1201 /*
1202 * Needs to be copied before we can link it.
1203 */
1204 vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
1205 if (droplock)
1206 VOP_UNLOCK(dun->un_uppervp);
1207 error = union_copyup(un, 1, cnp->cn_cred, curlwp);
1208 if (droplock) {
1209 vn_lock(dun->un_uppervp,
1210 LK_EXCLUSIVE | LK_RETRY);
1211 /*
1212 * During copyup, we dropped the lock on the
1213 * dir and invalidated any saved namei lookup
1214 * state for the directory we'll be entering
1215 * the link in. We need to re-run the lookup
1216 * in that directory to reset any state needed
1217 * for VOP_LINK.
1218 * Call relookup on the union-layer to reset
1219 * the state.
1220 */
1221 vp = NULLVP;
1222 if (dun->un_uppervp == NULLVP)
1223 panic("union: null upperdvp?");
1224 error = relookup(ap->a_dvp, &vp, ap->a_cnp, 0);
1225 if (error) {
1226 VOP_UNLOCK(ap->a_vp);
1227 return EROFS; /* ? */
1228 }
1229 if (vp != NULLVP) {
1230 /*
1231 * The name we want to create has
1232 * mysteriously appeared (a race?)
1233 */
1234 error = EEXIST;
1235 VOP_UNLOCK(ap->a_vp);
1236 vput(vp);
1237 return (error);
1238 }
1239 }
1240 VOP_UNLOCK(ap->a_vp);
1241 }
1242 vp = un->un_uppervp;
1243 }
1244
1245 dvp = dun->un_uppervp;
1246 if (dvp == NULLVP)
1247 error = EROFS;
1248
1249 if (error)
1250 return (error);
1251
1252 return VOP_LINK(dvp, vp, cnp);
1253}
1254
1255int
1256union_rename(void *v)
1257{
1258 struct vop_rename_args /* {
1259 struct vnode *a_fdvp;
1260 struct vnode *a_fvp;
1261 struct componentname *a_fcnp;
1262 struct vnode *a_tdvp;
1263 struct vnode *a_tvp;
1264 struct componentname *a_tcnp;
1265 } */ *ap = v;
1266 int error;
1267
1268 struct vnode *fdvp = ap->a_fdvp;
1269 struct vnode *fvp = ap->a_fvp;
1270 struct vnode *tdvp = ap->a_tdvp;
1271 struct vnode *tvp = ap->a_tvp;
1272
1273 /*
1274 * Account for VOP_RENAME to vrele all nodes.
1275 * Note: VOP_RENAME will unlock tdvp.
1276 */
1277
1278 if (fdvp->v_op == union_vnodeop_p) { /* always true */
1279 struct union_node *un = VTOUNION(fdvp);
1280 if (un->un_uppervp == NULLVP) {
1281 /*
1282 * this should never happen in normal
1283 * operation but might if there was
1284 * a problem creating the top-level shadow
1285 * directory.
1286 */
1287 error = EXDEV;
1288 goto bad;
1289 }
1290
1291 fdvp = un->un_uppervp;
1292 vref(fdvp);
1293 }
1294
1295 if (fvp->v_op == union_vnodeop_p) { /* always true */
1296 struct union_node *un = VTOUNION(fvp);
1297 if (un->un_uppervp == NULLVP) {
1298 /* XXX: should do a copyup */
1299 error = EXDEV;
1300 goto bad;
1301 }
1302
1303 if (un->un_lowervp != NULLVP)
1304 ap->a_fcnp->cn_flags |= DOWHITEOUT;
1305
1306 fvp = un->un_uppervp;
1307 vref(fvp);
1308 }
1309
1310 if (tdvp->v_op == union_vnodeop_p) {
1311 struct union_node *un = VTOUNION(tdvp);
1312 if (un->un_uppervp == NULLVP) {
1313 /*
1314 * this should never happen in normal
1315 * operation but might if there was
1316 * a problem creating the top-level shadow
1317 * directory.
1318 */
1319 error = EXDEV;
1320 goto bad;
1321 }
1322
1323 tdvp = un->un_uppervp;
1324 vref(tdvp);
1325 }
1326
1327 if (tvp != NULLVP && tvp->v_op == union_vnodeop_p) {
1328 struct union_node *un = VTOUNION(tvp);
1329
1330 tvp = un->un_uppervp;
1331 if (tvp != NULLVP) {
1332 vref(tvp);
1333 }
1334 }
1335
1336 error = VOP_RENAME(fdvp, fvp, ap->a_fcnp, tdvp, tvp, ap->a_tcnp);
1337 goto out;
1338
1339bad:
1340 vput(tdvp);
1341 if (tvp != NULLVP)
1342 vput(tvp);
1343 vrele(fdvp);
1344 vrele(fvp);
1345
1346out:
1347 if (fdvp != ap->a_fdvp) {
1348 vrele(ap->a_fdvp);
1349 }
1350 if (fvp != ap->a_fvp) {
1351 vrele(ap->a_fvp);
1352 }
1353 if (tdvp != ap->a_tdvp) {
1354 vrele(ap->a_tdvp);
1355 }
1356 if (tvp != ap->a_tvp) {
1357 vrele(ap->a_tvp);
1358 }
1359 return (error);
1360}
1361
1362int
1363union_mkdir(void *v)
1364{
1365 struct vop_mkdir_v3_args /* {
1366 struct vnode *a_dvp;
1367 struct vnode **a_vpp;
1368 struct componentname *a_cnp;
1369 struct vattr *a_vap;
1370 } */ *ap = v;
1371 struct union_node *un = VTOUNION(ap->a_dvp);
1372 struct vnode *dvp = un->un_uppervp;
1373 struct componentname *cnp = ap->a_cnp;
1374
1375 if (dvp != NULLVP) {
1376 int error;
1377 struct vnode *vp;
1378
1379 vp = NULL;
1380 error = VOP_MKDIR(dvp, &vp, cnp, ap->a_vap);
1381 if (error) {
1382 vrele(ap->a_dvp);
1383 return (error);
1384 }
1385
1386 error = union_allocvp(ap->a_vpp, ap->a_dvp->v_mount, ap->a_dvp,
1387 NULLVP, cnp, vp, NULLVP, 1);
1388 if (error)
1389 vrele(vp);
1390 return (error);
1391 }
1392
1393 return (EROFS);
1394}
1395
1396int
1397union_rmdir(void *v)
1398{
1399 struct vop_rmdir_args /* {
1400 struct vnode *a_dvp;
1401 struct vnode *a_vp;
1402 struct componentname *a_cnp;
1403 } */ *ap = v;
1404 int error;
1405 struct union_node *dun = VTOUNION(ap->a_dvp);
1406 struct union_node *un = VTOUNION(ap->a_vp);
1407 struct componentname *cnp = ap->a_cnp;
1408
1409 if (dun->un_uppervp == NULLVP)
1410 panic("union rmdir: null upper vnode");
1411
1412 error = union_check_rmdir(un, cnp->cn_cred);
1413 if (error) {
1414 vput(ap->a_dvp);
1415 vput(ap->a_vp);
1416 return error;
1417 }
1418
1419 if (un->un_uppervp != NULLVP) {
1420 struct vnode *dvp = dun->un_uppervp;
1421 struct vnode *vp = un->un_uppervp;
1422
1423 /*
1424 * Account for VOP_RMDIR to vrele dvp and vp.
1425 * Note: VOP_RMDIR will unlock dvp and vp.
1426 */
1427 vref(dvp);
1428 vref(vp);
1429 if (union_dowhiteout(un, cnp->cn_cred))
1430 cnp->cn_flags |= DOWHITEOUT;
1431 error = VOP_RMDIR(dvp, vp, ap->a_cnp);
1432 if (!error)
1433 union_removed_upper(un);
1434 vrele(ap->a_dvp);
1435 vrele(ap->a_vp);
1436 } else {
1437 error = union_mkwhiteout(
1438 MOUNTTOUNIONMOUNT(UNIONTOV(dun)->v_mount),
1439 dun->un_uppervp, ap->a_cnp, un);
1440 vput(ap->a_dvp);
1441 vput(ap->a_vp);
1442 }
1443
1444 return (error);
1445}
1446
1447int
1448union_symlink(void *v)
1449{
1450 struct vop_symlink_v3_args /* {
1451 struct vnode *a_dvp;
1452 struct vnode **a_vpp;
1453 struct componentname *a_cnp;
1454 struct vattr *a_vap;
1455 char *a_target;
1456 } */ *ap = v;
1457 struct union_node *un = VTOUNION(ap->a_dvp);
1458 struct vnode *dvp = un->un_uppervp;
1459 struct componentname *cnp = ap->a_cnp;
1460
1461 if (dvp != NULLVP) {
1462 int error;
1463
1464 error = VOP_SYMLINK(dvp, ap->a_vpp, cnp, ap->a_vap,
1465 ap->a_target);
1466 return (error);
1467 }
1468
1469 return (EROFS);
1470}
1471
1472/*
1473 * union_readdir works in concert with getdirentries and
1474 * readdir(3) to provide a list of entries in the unioned
1475 * directories. getdirentries is responsible for walking
1476 * down the union stack. readdir(3) is responsible for
1477 * eliminating duplicate names from the returned data stream.
1478 */
1479int
1480union_readdir(void *v)
1481{
1482 struct vop_readdir_args /* {
1483 struct vnodeop_desc *a_desc;
1484 struct vnode *a_vp;
1485 struct uio *a_uio;
1486 kauth_cred_t a_cred;
1487 int *a_eofflag;
1488 u_long *a_cookies;
1489 int a_ncookies;
1490 } */ *ap = v;
1491 struct union_node *un = VTOUNION(ap->a_vp);
1492 struct vnode *uvp = un->un_uppervp;
1493
1494 if (uvp == NULLVP)
1495 return (0);
1496
1497 ap->a_vp = uvp;
1498 return (VCALL(uvp, VOFFSET(vop_readdir), ap));
1499}
1500
1501int
1502union_readlink(void *v)
1503{
1504 struct vop_readlink_args /* {
1505 struct vnode *a_vp;
1506 struct uio *a_uio;
1507 kauth_cred_t a_cred;
1508 } */ *ap = v;
1509 int error;
1510 struct vnode *vp = OTHERVP(ap->a_vp);
1511 int dolock = (vp == LOWERVP(ap->a_vp));
1512
1513 if (dolock)
1514 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1515 ap->a_vp = vp;
1516 error = VCALL(vp, VOFFSET(vop_readlink), ap);
1517 if (dolock)
1518 VOP_UNLOCK(vp);
1519
1520 return (error);
1521}
1522
1523int
1524union_abortop(void *v)
1525{
1526 struct vop_abortop_args /* {
1527 struct vnode *a_dvp;
1528 struct componentname *a_cnp;
1529 } */ *ap = v;
1530
1531 KASSERT(UPPERVP(ap->a_dvp) != NULL);
1532
1533 ap->a_dvp = UPPERVP(ap->a_dvp);
1534 return VCALL(ap->a_dvp, VOFFSET(vop_abortop), ap);
1535}
1536
1537int
1538union_inactive(void *v)
1539{
1540 struct vop_inactive_args /* {
1541 const struct vnodeop_desc *a_desc;
1542 struct vnode *a_vp;
1543 bool *a_recycle;
1544 } */ *ap = v;
1545 struct vnode *vp = ap->a_vp;
1546 struct union_node *un = VTOUNION(vp);
1547 struct vnode **vpp;
1548
1549 /*
1550 * Do nothing (and _don't_ bypass).
1551 * Wait to vrele lowervp until reclaim,
1552 * so that until then our union_node is in the
1553 * cache and reusable.
1554 *
1555 * NEEDSWORK: Someday, consider inactive'ing
1556 * the lowervp and then trying to reactivate it
1557 * with capabilities (v_id)
1558 * like they do in the name lookup cache code.
1559 * That's too much work for now.
1560 */
1561
1562 if (un->un_dircache != 0) {
1563 for (vpp = un->un_dircache; *vpp != NULLVP; vpp++)
1564 vrele(*vpp);
1565 free(un->un_dircache, M_TEMP);
1566 un->un_dircache = 0;
1567 }
1568
1569 *ap->a_recycle = ((un->un_cflags & UN_CACHED) == 0);
1570 VOP_UNLOCK(vp);
1571
1572 return (0);
1573}
1574
1575int
1576union_reclaim(void *v)
1577{
1578 struct vop_reclaim_args /* {
1579 struct vnode *a_vp;
1580 } */ *ap = v;
1581
1582 union_freevp(ap->a_vp);
1583
1584 return (0);
1585}
1586
1587static int
1588union_lock1(struct vnode *vp, struct vnode *lockvp, int flags)
1589{
1590 struct vop_lock_args ap;
1591
1592 if (lockvp == vp) {
1593 ap.a_vp = vp;
1594 ap.a_flags = flags;
1595 return genfs_lock(&ap);
1596 } else
1597 return VOP_LOCK(lockvp, flags);
1598}
1599
1600static int
1601union_unlock1(struct vnode *vp, struct vnode *lockvp)
1602{
1603 struct vop_unlock_args ap;
1604
1605 if (lockvp == vp) {
1606 ap.a_vp = vp;
1607 return genfs_unlock(&ap);
1608 } else
1609 return VOP_UNLOCK(lockvp);
1610}
1611
1612int
1613union_lock(void *v)
1614{
1615 struct vop_lock_args /* {
1616 struct vnode *a_vp;
1617 int a_flags;
1618 } */ *ap = v;
1619 struct vnode *vp = ap->a_vp, *lockvp;
1620 struct union_node *un = VTOUNION(vp);
1621 int flags = ap->a_flags;
1622 int error;
1623
1624 if ((flags & LK_NOWAIT) != 0) {
1625 if (!mutex_tryenter(&un->un_lock))
1626 return EBUSY;
1627 lockvp = LOCKVP(vp);
1628 error = union_lock1(vp, lockvp, flags);
1629 mutex_exit(&un->un_lock);
1630 if (error)
1631 return error;
1632 if (mutex_tryenter(vp->v_interlock)) {
1633 error = vdead_check(vp, VDEAD_NOWAIT);
1634 mutex_exit(vp->v_interlock);
1635 } else
1636 error = EBUSY;
1637 if (error)
1638 union_unlock1(vp, lockvp);
1639 return error;
1640 }
1641
1642 mutex_enter(&un->un_lock);
1643 for (;;) {
1644 lockvp = LOCKVP(vp);
1645 mutex_exit(&un->un_lock);
1646 error = union_lock1(vp, lockvp, flags);
1647 if (error != 0)
1648 return error;
1649 mutex_enter(&un->un_lock);
1650 if (lockvp == LOCKVP(vp))
1651 break;
1652 union_unlock1(vp, lockvp);
1653 }
1654 mutex_exit(&un->un_lock);
1655
1656 mutex_enter(vp->v_interlock);
1657 error = vdead_check(vp, VDEAD_NOWAIT);
1658 if (error) {
1659 union_unlock1(vp, lockvp);
1660 error = vdead_check(vp, 0);
1661 KASSERT(error == ENOENT);
1662 }
1663 mutex_exit(vp->v_interlock);
1664 return error;
1665}
1666
1667int
1668union_unlock(void *v)
1669{
1670 struct vop_unlock_args /* {
1671 struct vnode *a_vp;
1672 int a_flags;
1673 } */ *ap = v;
1674 struct vnode *vp = ap->a_vp, *lockvp;
1675
1676 lockvp = LOCKVP(vp);
1677 union_unlock1(vp, lockvp);
1678
1679 return 0;
1680}
1681
1682int
1683union_bmap(void *v)
1684{
1685 struct vop_bmap_args /* {
1686 struct vnode *a_vp;
1687 daddr_t a_bn;
1688 struct vnode **a_vpp;
1689 daddr_t *a_bnp;
1690 int *a_runp;
1691 } */ *ap = v;
1692 int error;
1693 struct vnode *vp = OTHERVP(ap->a_vp);
1694 int dolock = (vp == LOWERVP(ap->a_vp));
1695
1696 if (dolock)
1697 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1698 ap->a_vp = vp;
1699 error = VCALL(vp, VOFFSET(vop_bmap), ap);
1700 if (dolock)
1701 VOP_UNLOCK(vp);
1702
1703 return (error);
1704}
1705
1706int
1707union_print(void *v)
1708{
1709 struct vop_print_args /* {
1710 struct vnode *a_vp;
1711 } */ *ap = v;
1712 struct vnode *vp = ap->a_vp;
1713
1714 printf("\ttag VT_UNION, vp=%p, uppervp=%p, lowervp=%p\n",
1715 vp, UPPERVP(vp), LOWERVP(vp));
1716 if (UPPERVP(vp) != NULLVP)
1717 vprint("union: upper", UPPERVP(vp));
1718 if (LOWERVP(vp) != NULLVP)
1719 vprint("union: lower", LOWERVP(vp));
1720 if (VTOUNION(vp)->un_dircache) {
1721 struct vnode **vpp;
1722 for (vpp = VTOUNION(vp)->un_dircache; *vpp != NULLVP; vpp++)
1723 vprint("dircache:", *vpp);
1724 }
1725
1726 return (0);
1727}
1728
1729int
1730union_islocked(void *v)
1731{
1732 struct vop_islocked_args /* {
1733 struct vnode *a_vp;
1734 } */ *ap = v;
1735 struct vnode *vp;
1736 struct union_node *un;
1737
1738 un = VTOUNION(ap->a_vp);
1739 mutex_enter(&un->un_lock);
1740 vp = LOCKVP(ap->a_vp);
1741 mutex_exit(&un->un_lock);
1742
1743 if (vp == ap->a_vp)
1744 return genfs_islocked(ap);
1745 else
1746 return VOP_ISLOCKED(vp);
1747}
1748
1749int
1750union_pathconf(void *v)
1751{
1752 struct vop_pathconf_args /* {
1753 struct vnode *a_vp;
1754 int a_name;
1755 int *a_retval;
1756 } */ *ap = v;
1757 int error;
1758 struct vnode *vp = OTHERVP(ap->a_vp);
1759 int dolock = (vp == LOWERVP(ap->a_vp));
1760
1761 if (dolock)
1762 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1763 ap->a_vp = vp;
1764 error = VCALL(vp, VOFFSET(vop_pathconf), ap);
1765 if (dolock)
1766 VOP_UNLOCK(vp);
1767
1768 return (error);
1769}
1770
1771int
1772union_advlock(void *v)
1773{
1774 struct vop_advlock_args /* {
1775 struct vnode *a_vp;
1776 void *a_id;
1777 int a_op;
1778 struct flock *a_fl;
1779 int a_flags;
1780 } */ *ap = v;
1781 struct vnode *ovp = OTHERVP(ap->a_vp);
1782
1783 ap->a_vp = ovp;
1784 return (VCALL(ovp, VOFFSET(vop_advlock), ap));
1785}
1786
1787int
1788union_strategy(void *v)
1789{
1790 struct vop_strategy_args /* {
1791 struct vnode *a_vp;
1792 struct buf *a_bp;
1793 } */ *ap = v;
1794 struct vnode *ovp = OTHERVP(ap->a_vp);
1795 struct buf *bp = ap->a_bp;
1796
1797 KASSERT(ovp != NULLVP);
1798 if (!NODE_IS_SPECIAL(ovp))
1799 KASSERT((bp->b_flags & B_READ) || ovp != LOWERVP(bp->b_vp));
1800
1801 return (VOP_STRATEGY(ovp, bp));
1802}
1803
1804int
1805union_bwrite(void *v)
1806{
1807 struct vop_bwrite_args /* {
1808 struct vnode *a_vp;
1809 struct buf *a_bp;
1810 } */ *ap = v;
1811 struct vnode *ovp = OTHERVP(ap->a_vp);
1812 struct buf *bp = ap->a_bp;
1813
1814 KASSERT(ovp != NULLVP);
1815 if (!NODE_IS_SPECIAL(ovp))
1816 KASSERT((bp->b_flags & B_READ) || ovp != LOWERVP(bp->b_vp));
1817
1818 return (VOP_BWRITE(ovp, bp));
1819}
1820
1821int
1822union_getpages(void *v)
1823{
1824 struct vop_getpages_args /* {
1825 struct vnode *a_vp;
1826 voff_t a_offset;
1827 struct vm_page **a_m;
1828 int *a_count;
1829 int a_centeridx;
1830 vm_prot_t a_access_type;
1831 int a_advice;
1832 int a_flags;
1833 } */ *ap = v;
1834 struct vnode *vp = ap->a_vp;
1835
1836 KASSERT(mutex_owned(vp->v_interlock));
1837
1838 if (ap->a_flags & PGO_LOCKED) {
1839 return EBUSY;
1840 }
1841 ap->a_vp = OTHERVP(vp);
1842 KASSERT(vp->v_interlock == ap->a_vp->v_interlock);
1843
1844 /* Just pass the request on to the underlying layer. */
1845 return VCALL(ap->a_vp, VOFFSET(vop_getpages), ap);
1846}
1847
1848int
1849union_putpages(void *v)
1850{
1851 struct vop_putpages_args /* {
1852 struct vnode *a_vp;
1853 voff_t a_offlo;
1854 voff_t a_offhi;
1855 int a_flags;
1856 } */ *ap = v;
1857 struct vnode *vp = ap->a_vp;
1858
1859 KASSERT(mutex_owned(vp->v_interlock));
1860
1861 ap->a_vp = OTHERVP(vp);
1862 KASSERT(vp->v_interlock == ap->a_vp->v_interlock);
1863
1864 if (ap->a_flags & PGO_RECLAIM) {
1865 mutex_exit(vp->v_interlock);
1866 return 0;
1867 }
1868
1869 /* Just pass the request on to the underlying layer. */
1870 return VCALL(ap->a_vp, VOFFSET(vop_putpages), ap);
1871}
1872
1873int
1874union_kqfilter(void *v)
1875{
1876 struct vop_kqfilter_args /* {
1877 struct vnode *a_vp;
1878 struct knote *a_kn;
1879 } */ *ap = v;
1880 int error;
1881
1882 /*
1883 * We watch either the upper layer file (if it already exists),
1884 * or the lower layer one. If there is lower layer file only
1885 * at this moment, we will keep watching that lower layer file
1886 * even if upper layer file would be created later on.
1887 */
1888 if (UPPERVP(ap->a_vp))
1889 error = VOP_KQFILTER(UPPERVP(ap->a_vp), ap->a_kn);
1890 else if (LOWERVP(ap->a_vp))
1891 error = VOP_KQFILTER(LOWERVP(ap->a_vp), ap->a_kn);
1892 else {
1893 /* panic? */
1894 error = EOPNOTSUPP;
1895 }
1896
1897 return (error);
1898}
1899