union_vnops.c revision 1.21 1 /* $NetBSD: union_vnops.c,v 1.21 2007/07/29 13:12:42 pooka 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 1.21 2007/07/29 13:12:42 pooka 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
95 int union_lookup(void *);
96 int union_create(void *);
97 int union_whiteout(void *);
98 int union_mknod(void *);
99 int union_open(void *);
100 int union_close(void *);
101 int union_access(void *);
102 int union_getattr(void *);
103 int union_setattr(void *);
104 int union_read(void *);
105 int union_write(void *);
106 int union_lease(void *);
107 int union_ioctl(void *);
108 int union_poll(void *);
109 int union_revoke(void *);
110 int union_mmap(void *);
111 int union_fsync(void *);
112 int union_seek(void *);
113 int union_remove(void *);
114 int union_link(void *);
115 int union_rename(void *);
116 int union_mkdir(void *);
117 int union_rmdir(void *);
118 int union_symlink(void *);
119 int union_readdir(void *);
120 int union_readlink(void *);
121 int union_abortop(void *);
122 int union_inactive(void *);
123 int union_reclaim(void *);
124 int union_lock(void *);
125 int union_unlock(void *);
126 int union_bmap(void *);
127 int union_print(void *);
128 int union_islocked(void *);
129 int union_pathconf(void *);
130 int union_advlock(void *);
131 int union_strategy(void *);
132 int union_getpages(void *);
133 int union_putpages(void *);
134 int union_kqfilter(void *);
135
136 static void union_fixup(struct union_node *);
137 static int union_lookup1(struct vnode *, struct vnode **,
138 struct vnode **, struct componentname *);
139
140
141 /*
142 * Global vfs data structures
143 */
144 int (**union_vnodeop_p)(void *);
145 const 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_lease_desc, union_lease }, /* lease */
159 { &vop_ioctl_desc, union_ioctl }, /* ioctl */
160 { &vop_poll_desc, union_poll }, /* select */
161 { &vop_revoke_desc, union_revoke }, /* revoke */
162 { &vop_mmap_desc, union_mmap }, /* mmap */
163 { &vop_fsync_desc, union_fsync }, /* fsync */
164 { &vop_seek_desc, union_seek }, /* seek */
165 { &vop_remove_desc, union_remove }, /* remove */
166 { &vop_link_desc, union_link }, /* link */
167 { &vop_rename_desc, union_rename }, /* rename */
168 { &vop_mkdir_desc, union_mkdir }, /* mkdir */
169 { &vop_rmdir_desc, union_rmdir }, /* rmdir */
170 { &vop_symlink_desc, union_symlink }, /* symlink */
171 { &vop_readdir_desc, union_readdir }, /* readdir */
172 { &vop_readlink_desc, union_readlink }, /* readlink */
173 { &vop_abortop_desc, union_abortop }, /* abortop */
174 { &vop_inactive_desc, union_inactive }, /* inactive */
175 { &vop_reclaim_desc, union_reclaim }, /* reclaim */
176 { &vop_lock_desc, union_lock }, /* lock */
177 { &vop_unlock_desc, union_unlock }, /* unlock */
178 { &vop_bmap_desc, union_bmap }, /* bmap */
179 { &vop_strategy_desc, union_strategy }, /* strategy */
180 { &vop_print_desc, union_print }, /* print */
181 { &vop_islocked_desc, union_islocked }, /* islocked */
182 { &vop_pathconf_desc, union_pathconf }, /* pathconf */
183 { &vop_advlock_desc, union_advlock }, /* advlock */
184 { &vop_getpages_desc, union_getpages }, /* getpages */
185 { &vop_putpages_desc, union_putpages }, /* putpages */
186 { &vop_kqfilter_desc, union_kqfilter }, /* kqfilter */
187 #ifdef notdef
188 { &vop_bwrite_desc, union_bwrite }, /* bwrite */
189 #endif
190 { NULL, NULL }
191 };
192 const struct vnodeopv_desc union_vnodeop_opv_desc =
193 { &union_vnodeop_p, union_vnodeop_entries };
194
195 #define FIXUP(un) { \
196 if (((un)->un_flags & UN_ULOCK) == 0) { \
197 union_fixup(un); \
198 } \
199 }
200
201 static void
202 union_fixup(un)
203 struct union_node *un;
204 {
205
206 vn_lock(un->un_uppervp, LK_EXCLUSIVE | LK_RETRY);
207 un->un_flags |= UN_ULOCK;
208 }
209
210 static int
211 union_lookup1(udvp, dvpp, vpp, cnp)
212 struct vnode *udvp;
213 struct vnode **dvpp;
214 struct vnode **vpp;
215 struct componentname *cnp;
216 {
217 int error;
218 struct vnode *tdvp;
219 struct vnode *dvp;
220 struct mount *mp;
221
222 dvp = *dvpp;
223
224 /*
225 * If stepping up the directory tree, check for going
226 * back across the mount point, in which case do what
227 * lookup would do by stepping back down the mount
228 * hierarchy.
229 */
230 if (cnp->cn_flags & ISDOTDOT) {
231 while ((dvp != udvp) && (dvp->v_flag & VROOT)) {
232 /*
233 * Don't do the NOCROSSMOUNT check
234 * at this level. By definition,
235 * union fs deals with namespaces, not
236 * filesystems.
237 */
238 tdvp = dvp;
239 *dvpp = dvp = dvp->v_mount->mnt_vnodecovered;
240 VOP_UNLOCK(tdvp, 0);
241 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
242 }
243 }
244
245 error = VOP_LOOKUP(dvp, &tdvp, cnp);
246 if (error)
247 return (error);
248
249 dvp = tdvp;
250
251 /*
252 * Lastly check if the current node is a mount point in
253 * which case walk up the mount hierarchy making sure not to
254 * bump into the root of the mount tree (ie. dvp != udvp).
255 */
256 while (dvp != udvp && (dvp->v_type == VDIR) &&
257 (mp = dvp->v_mountedhere)) {
258
259 if (vfs_busy(mp, 0, 0))
260 continue;
261
262 vput(dvp);
263 error = VFS_ROOT(mp, &tdvp);
264 vfs_unbusy(mp);
265 if (error) {
266 return (error);
267 }
268 dvp = tdvp;
269 }
270
271 *vpp = dvp;
272 return (0);
273 }
274
275 int
276 union_lookup(v)
277 void *v;
278 {
279 struct vop_lookup_args /* {
280 struct vnodeop_desc *a_desc;
281 struct vnode *a_dvp;
282 struct vnode **a_vpp;
283 struct componentname *a_cnp;
284 } */ *ap = v;
285 int error;
286 int uerror, lerror;
287 struct vnode *uppervp, *lowervp;
288 struct vnode *upperdvp, *lowerdvp;
289 struct vnode *dvp = ap->a_dvp;
290 struct union_node *dun = VTOUNION(dvp);
291 struct componentname *cnp = ap->a_cnp;
292 struct union_mount *um = MOUNTTOUNIONMOUNT(dvp->v_mount);
293 kauth_cred_t saved_cred = NULL;
294 int iswhiteout;
295 struct vattr va;
296
297 #ifdef notyet
298 if (cnp->cn_namelen == 3 &&
299 cnp->cn_nameptr[2] == '.' &&
300 cnp->cn_nameptr[1] == '.' &&
301 cnp->cn_nameptr[0] == '.') {
302 dvp = *ap->a_vpp = LOWERVP(ap->a_dvp);
303 if (dvp == NULLVP)
304 return (ENOENT);
305 VREF(dvp);
306 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
307 return (0);
308 }
309 #endif
310
311 if ((cnp->cn_flags & ISLASTCN) &&
312 (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
313 (cnp->cn_nameiop == DELETE || cnp->cn_nameiop == RENAME))
314 return (EROFS);
315
316 upperdvp = dun->un_uppervp;
317 lowerdvp = dun->un_lowervp;
318 uppervp = NULLVP;
319 lowervp = NULLVP;
320 iswhiteout = 0;
321
322 /*
323 * do the lookup in the upper level.
324 * if that level comsumes additional pathnames,
325 * then assume that something special is going
326 * on and just return that vnode.
327 */
328 if (upperdvp != NULLVP) {
329 FIXUP(dun);
330 /*
331 * If we're doing `..' in the underlying filesystem,
332 * we must drop our lock on the union node before
333 * going up the tree in the lower file system--if we block
334 * on the lowervp lock, and that's held by someone else
335 * coming down the tree and who's waiting for our lock,
336 * we would be hosed.
337 */
338 if (cnp->cn_flags & ISDOTDOT) {
339 /* retain lock on underlying VP */
340 dun->un_flags |= UN_KLOCK;
341 VOP_UNLOCK(dvp, 0);
342 }
343 uerror = union_lookup1(um->um_uppervp, &upperdvp,
344 &uppervp, cnp);
345
346 if (cnp->cn_flags & ISDOTDOT) {
347 if (dun->un_uppervp == upperdvp) {
348 /*
349 * we got the underlying bugger back locked...
350 * now take back the union node lock. Since we
351 * hold the uppervp lock, we can diddle union
352 * locking flags at will. :)
353 */
354 dun->un_flags |= UN_ULOCK;
355 }
356 /*
357 * if upperdvp got swapped out, it means we did
358 * some mount point magic, and we do not have
359 * dun->un_uppervp locked currently--so we get it
360 * locked here (don't set the UN_ULOCK flag).
361 */
362 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
363 }
364 if (cnp->cn_consume != 0) {
365 *ap->a_vpp = uppervp;
366 return (uerror);
367 }
368 if (uerror == ENOENT || uerror == EJUSTRETURN) {
369 if (cnp->cn_flags & ISWHITEOUT) {
370 iswhiteout = 1;
371 } else if (lowerdvp != NULLVP) {
372 lerror = VOP_GETATTR(upperdvp, &va,
373 cnp->cn_cred, cnp->cn_lwp);
374 if (lerror == 0 && (va.va_flags & OPAQUE))
375 iswhiteout = 1;
376 }
377 }
378 } else {
379 uerror = ENOENT;
380 }
381
382 /*
383 * in a similar way to the upper layer, do the lookup
384 * in the lower layer. this time, if there is some
385 * component magic going on, then vput whatever we got
386 * back from the upper layer and return the lower vnode
387 * instead.
388 */
389 if (lowerdvp != NULLVP && !iswhiteout) {
390 int nameiop;
391
392 vn_lock(lowerdvp, LK_EXCLUSIVE | LK_RETRY);
393
394 /*
395 * Only do a LOOKUP on the bottom node, since
396 * we won't be making changes to it anyway.
397 */
398 nameiop = cnp->cn_nameiop;
399 cnp->cn_nameiop = LOOKUP;
400 if (um->um_op == UNMNT_BELOW) {
401 saved_cred = cnp->cn_cred;
402 cnp->cn_cred = um->um_cred;
403 }
404
405 /*
406 * we shouldn't have to worry about locking interactions
407 * between the lower layer and our union layer (w.r.t.
408 * `..' processing) because we don't futz with lowervp
409 * locks in the union-node instantiation code path.
410 */
411 lerror = union_lookup1(um->um_lowervp, &lowerdvp,
412 &lowervp, cnp);
413 if (um->um_op == UNMNT_BELOW)
414 cnp->cn_cred = saved_cred;
415 cnp->cn_nameiop = nameiop;
416
417 if (lowervp != lowerdvp)
418 VOP_UNLOCK(lowerdvp, 0);
419
420 if (cnp->cn_consume != 0) {
421 if (uppervp != NULLVP) {
422 if (uppervp == upperdvp)
423 vrele(uppervp);
424 else
425 vput(uppervp);
426 uppervp = NULLVP;
427 }
428 *ap->a_vpp = lowervp;
429 return (lerror);
430 }
431 } else {
432 lerror = ENOENT;
433 if ((cnp->cn_flags & ISDOTDOT) && dun->un_pvp != NULLVP) {
434 lowervp = LOWERVP(dun->un_pvp);
435 if (lowervp != NULLVP) {
436 VREF(lowervp);
437 vn_lock(lowervp, LK_EXCLUSIVE | LK_RETRY);
438 lerror = 0;
439 }
440 }
441 }
442
443 /*
444 * EJUSTRETURN is used by underlying filesystems to indicate that
445 * a directory modification op was started successfully.
446 * This will only happen in the upper layer, since
447 * the lower layer only does LOOKUPs.
448 * If this union is mounted read-only, bounce it now.
449 */
450
451 if ((uerror == EJUSTRETURN) && (cnp->cn_flags & ISLASTCN) &&
452 (dvp->v_mount->mnt_flag & MNT_RDONLY) &&
453 ((cnp->cn_nameiop == CREATE) || (cnp->cn_nameiop == RENAME)))
454 uerror = EROFS;
455
456 /*
457 * at this point, we have uerror and lerror indicating
458 * possible errors with the lookups in the upper and lower
459 * layers. additionally, uppervp and lowervp are (locked)
460 * references to existing vnodes in the upper and lower layers.
461 *
462 * there are now three cases to consider.
463 * 1. if both layers returned an error, then return whatever
464 * error the upper layer generated.
465 *
466 * 2. if the top layer failed and the bottom layer succeeded
467 * then two subcases occur.
468 * a. the bottom vnode is not a directory, in which
469 * case just return a new union vnode referencing
470 * an empty top layer and the existing bottom layer.
471 * b. the bottom vnode is a directory, in which case
472 * create a new directory in the top-level and
473 * continue as in case 3.
474 *
475 * 3. if the top layer succeeded then return a new union
476 * vnode referencing whatever the new top layer and
477 * whatever the bottom layer returned.
478 */
479
480 *ap->a_vpp = NULLVP;
481
482
483 /* case 1. */
484 if ((uerror != 0) && (lerror != 0)) {
485 return (uerror);
486 }
487
488 /* case 2. */
489 if (uerror != 0 /* && (lerror == 0) */ ) {
490 if (lowervp->v_type == VDIR) { /* case 2b. */
491 /*
492 * We may be racing another process to make the
493 * upper-level shadow directory. Be careful with
494 * locks/etc!
495 */
496 if (upperdvp) {
497 dun->un_flags &= ~UN_ULOCK;
498 VOP_UNLOCK(upperdvp, 0);
499 uerror = union_mkshadow(um, upperdvp, cnp,
500 &uppervp);
501 vn_lock(upperdvp, LK_EXCLUSIVE | LK_RETRY);
502 dun->un_flags |= UN_ULOCK;
503 }
504 if (uerror) {
505 if (lowervp != NULLVP) {
506 vput(lowervp);
507 lowervp = NULLVP;
508 }
509 return (uerror);
510 }
511 }
512 }
513
514 if (lowervp != NULLVP)
515 VOP_UNLOCK(lowervp, 0);
516
517 error = union_allocvp(ap->a_vpp, dvp->v_mount, dvp, upperdvp, cnp,
518 uppervp, lowervp, 1);
519
520 if (error) {
521 if (uppervp != NULLVP)
522 vput(uppervp);
523 if (lowervp != NULLVP)
524 vrele(lowervp);
525 }
526
527 return (error);
528 }
529
530 int
531 union_create(v)
532 void *v;
533 {
534 struct vop_create_args /* {
535 struct vnode *a_dvp;
536 struct vnode **a_vpp;
537 struct componentname *a_cnp;
538 struct vattr *a_vap;
539 } */ *ap = v;
540 struct union_node *un = VTOUNION(ap->a_dvp);
541 struct vnode *dvp = un->un_uppervp;
542 struct componentname *cnp = ap->a_cnp;
543
544 if (dvp != NULLVP) {
545 int error;
546 struct vnode *vp;
547 struct mount *mp;
548
549 FIXUP(un);
550
551 VREF(dvp);
552 un->un_flags |= UN_KLOCK;
553 mp = ap->a_dvp->v_mount;
554 vput(ap->a_dvp);
555 error = VOP_CREATE(dvp, &vp, cnp, ap->a_vap);
556 if (error)
557 return (error);
558
559 error = union_allocvp(ap->a_vpp, mp, NULLVP, NULLVP, cnp, vp,
560 NULLVP, 1);
561 if (error)
562 vput(vp);
563 return (error);
564 }
565
566 vput(ap->a_dvp);
567 return (EROFS);
568 }
569
570 int
571 union_whiteout(v)
572 void *v;
573 {
574 struct vop_whiteout_args /* {
575 struct vnode *a_dvp;
576 struct componentname *a_cnp;
577 int a_flags;
578 } */ *ap = v;
579 struct union_node *un = VTOUNION(ap->a_dvp);
580 struct componentname *cnp = ap->a_cnp;
581
582 if (un->un_uppervp == NULLVP)
583 return (EOPNOTSUPP);
584
585 FIXUP(un);
586 return (VOP_WHITEOUT(un->un_uppervp, cnp, ap->a_flags));
587 }
588
589 int
590 union_mknod(v)
591 void *v;
592 {
593 struct vop_mknod_args /* {
594 struct vnode *a_dvp;
595 struct vnode **a_vpp;
596 struct componentname *a_cnp;
597 struct vattr *a_vap;
598 } */ *ap = v;
599 struct union_node *un = VTOUNION(ap->a_dvp);
600 struct vnode *dvp = un->un_uppervp;
601 struct componentname *cnp = ap->a_cnp;
602
603 if (dvp != NULLVP) {
604 int error;
605 struct vnode *vp;
606 struct mount *mp;
607
608 FIXUP(un);
609
610 VREF(dvp);
611 un->un_flags |= UN_KLOCK;
612 mp = ap->a_dvp->v_mount;
613 vput(ap->a_dvp);
614 error = VOP_MKNOD(dvp, &vp, cnp, ap->a_vap);
615 if (error)
616 return (error);
617
618 error = union_allocvp(ap->a_vpp, mp, NULLVP, NULLVP,
619 cnp, vp, NULLVP, 1);
620 if (error)
621 vput(vp);
622 return (error);
623 }
624
625 vput(ap->a_dvp);
626 return (EROFS);
627 }
628
629 int
630 union_open(v)
631 void *v;
632 {
633 struct vop_open_args /* {
634 struct vnodeop_desc *a_desc;
635 struct vnode *a_vp;
636 int a_mode;
637 kauth_cred_t a_cred;
638 struct lwp *a_l;
639 } */ *ap = v;
640 struct union_node *un = VTOUNION(ap->a_vp);
641 struct vnode *tvp;
642 int mode = ap->a_mode;
643 kauth_cred_t cred = ap->a_cred;
644 struct lwp *l = ap->a_l;
645 int error;
646
647 /*
648 * If there is an existing upper vp then simply open that.
649 */
650 tvp = un->un_uppervp;
651 if (tvp == NULLVP) {
652 /*
653 * If the lower vnode is being opened for writing, then
654 * copy the file contents to the upper vnode and open that,
655 * otherwise can simply open the lower vnode.
656 */
657 tvp = un->un_lowervp;
658 if ((ap->a_mode & FWRITE) && (tvp->v_type == VREG)) {
659 error = union_copyup(un, (mode&O_TRUNC) == 0, cred, l);
660 if (error == 0)
661 error = VOP_OPEN(un->un_uppervp, mode, cred, l);
662 return (error);
663 }
664
665 /*
666 * Just open the lower vnode, but check for nodev mount flag
667 */
668 if ((tvp->v_type == VBLK || tvp->v_type == VCHR) &&
669 (ap->a_vp->v_mount->mnt_flag & MNT_NODEV))
670 return ENXIO;
671 un->un_openl++;
672 vn_lock(tvp, LK_EXCLUSIVE | LK_RETRY);
673 error = VOP_OPEN(tvp, mode, cred, l);
674 VOP_UNLOCK(tvp, 0);
675
676 return (error);
677 }
678 /*
679 * Just open the upper vnode, checking for nodev mount flag first
680 */
681 if ((tvp->v_type == VBLK || tvp->v_type == VCHR) &&
682 (ap->a_vp->v_mount->mnt_flag & MNT_NODEV))
683 return ENXIO;
684
685 FIXUP(un);
686
687 error = VOP_OPEN(tvp, mode, cred, l);
688
689 return (error);
690 }
691
692 int
693 union_close(v)
694 void *v;
695 {
696 struct vop_close_args /* {
697 struct vnode *a_vp;
698 int a_fflag;
699 kauth_cred_t a_cred;
700 struct lwp *a_l;
701 } */ *ap = v;
702 struct union_node *un = VTOUNION(ap->a_vp);
703 struct vnode *vp;
704
705 vp = un->un_uppervp;
706 if (vp == NULLVP) {
707 #ifdef UNION_DIAGNOSTIC
708 if (un->un_openl <= 0)
709 panic("union: un_openl cnt");
710 #endif
711 --un->un_openl;
712 vp = un->un_lowervp;
713 }
714
715 #ifdef DIAGNOSTIC
716 if (vp == NULLVP)
717 panic("union_close empty union vnode");
718 #endif
719
720 ap->a_vp = vp;
721 return (VCALL(vp, VOFFSET(vop_close), ap));
722 }
723
724 /*
725 * Check access permission on the union vnode.
726 * The access check being enforced is to check
727 * against both the underlying vnode, and any
728 * copied vnode. This ensures that no additional
729 * file permissions are given away simply because
730 * the user caused an implicit file copy.
731 */
732 int
733 union_access(v)
734 void *v;
735 {
736 struct vop_access_args /* {
737 struct vnodeop_desc *a_desc;
738 struct vnode *a_vp;
739 int a_mode;
740 kauth_cred_t a_cred;
741 struct lwp *a_l;
742 } */ *ap = v;
743 struct vnode *vp = ap->a_vp;
744 struct union_node *un = VTOUNION(vp);
745 int error = EACCES;
746 struct union_mount *um = MOUNTTOUNIONMOUNT(vp->v_mount);
747
748 /*
749 * Disallow write attempts on read-only file systems;
750 * unless the file is a socket, fifo, or a block or
751 * character device resident on the file system.
752 */
753 if (ap->a_mode & VWRITE) {
754 switch (vp->v_type) {
755 case VDIR:
756 case VLNK:
757 case VREG:
758 if (vp->v_mount->mnt_flag & MNT_RDONLY)
759 return (EROFS);
760 break;
761 case VBAD:
762 case VBLK:
763 case VCHR:
764 case VSOCK:
765 case VFIFO:
766 case VNON:
767 default:
768 break;
769 }
770 }
771
772
773 if ((vp = un->un_uppervp) != NULLVP) {
774 FIXUP(un);
775 ap->a_vp = vp;
776 return (VCALL(vp, VOFFSET(vop_access), ap));
777 }
778
779 if ((vp = un->un_lowervp) != NULLVP) {
780 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
781 ap->a_vp = vp;
782 error = VCALL(vp, VOFFSET(vop_access), ap);
783 if (error == 0) {
784 if (um->um_op == UNMNT_BELOW) {
785 ap->a_cred = um->um_cred;
786 error = VCALL(vp, VOFFSET(vop_access), ap);
787 }
788 }
789 VOP_UNLOCK(vp, 0);
790 if (error)
791 return (error);
792 }
793
794 return (error);
795 }
796
797 /*
798 * We handle getattr only to change the fsid and
799 * track object sizes
800 */
801 int
802 union_getattr(v)
803 void *v;
804 {
805 struct vop_getattr_args /* {
806 struct vnode *a_vp;
807 struct vattr *a_vap;
808 kauth_cred_t a_cred;
809 struct lwp *a_l;
810 } */ *ap = v;
811 int error;
812 struct union_node *un = VTOUNION(ap->a_vp);
813 struct vnode *vp = un->un_uppervp;
814 struct vattr *vap;
815 struct vattr va;
816
817
818 /*
819 * Some programs walk the filesystem hierarchy by counting
820 * links to directories to avoid stat'ing all the time.
821 * This means the link count on directories needs to be "correct".
822 * The only way to do that is to call getattr on both layers
823 * and fix up the link count. The link count will not necessarily
824 * be accurate but will be large enough to defeat the tree walkers.
825 *
826 * To make life more interesting, some filesystems don't keep
827 * track of link counts in the expected way, and return a
828 * link count of `1' for those directories; if either of the
829 * component directories returns a link count of `1', we return a 1.
830 */
831
832 vap = ap->a_vap;
833
834 vp = un->un_uppervp;
835 if (vp != NULLVP) {
836 /*
837 * It's not clear whether VOP_GETATTR is to be
838 * called with the vnode locked or not. stat() calls
839 * it with (vp) locked, and fstat calls it with
840 * (vp) unlocked.
841 * In the mean time, compensate here by checking
842 * the union_node's lock flag.
843 */
844 if (un->un_flags & UN_LOCKED)
845 FIXUP(un);
846
847 error = VOP_GETATTR(vp, vap, ap->a_cred, ap->a_l);
848 if (error)
849 return (error);
850 union_newsize(ap->a_vp, vap->va_size, VNOVAL);
851 }
852
853 if (vp == NULLVP) {
854 vp = un->un_lowervp;
855 } else if (vp->v_type == VDIR) {
856 vp = un->un_lowervp;
857 if (vp != NULLVP)
858 vap = &va;
859 } else {
860 vp = NULLVP;
861 }
862
863 if (vp != NULLVP) {
864 error = VOP_GETATTR(vp, vap, ap->a_cred, ap->a_l);
865 if (error)
866 return (error);
867 union_newsize(ap->a_vp, VNOVAL, vap->va_size);
868 }
869
870 if ((vap != ap->a_vap) && (vap->va_type == VDIR)) {
871 /*
872 * Link count manipulation:
873 * - If both return "2", return 2 (no subdirs)
874 * - If one or the other return "1", return "1" (ENOCLUE)
875 */
876 if ((ap->a_vap->va_nlink == 2) &&
877 (vap->va_nlink == 2))
878 ;
879 else if (ap->a_vap->va_nlink != 1) {
880 if (vap->va_nlink == 1)
881 ap->a_vap->va_nlink = 1;
882 else
883 ap->a_vap->va_nlink += vap->va_nlink;
884 }
885 }
886 ap->a_vap->va_fsid = ap->a_vp->v_mount->mnt_stat.f_fsidx.__fsid_val[0];
887 return (0);
888 }
889
890 int
891 union_setattr(v)
892 void *v;
893 {
894 struct vop_setattr_args /* {
895 struct vnode *a_vp;
896 struct vattr *a_vap;
897 kauth_cred_t a_cred;
898 struct lwp *a_l;
899 } */ *ap = v;
900 struct vattr *vap = ap->a_vap;
901 struct vnode *vp = ap->a_vp;
902 struct union_node *un = VTOUNION(vp);
903 int error;
904
905 if ((vap->va_flags != VNOVAL || vap->va_uid != (uid_t)VNOVAL ||
906 vap->va_gid != (gid_t)VNOVAL || vap->va_atime.tv_sec != VNOVAL ||
907 vap->va_mtime.tv_sec != VNOVAL || vap->va_mode != (mode_t)VNOVAL) &&
908 (vp->v_mount->mnt_flag & MNT_RDONLY))
909 return (EROFS);
910 if (vap->va_size != VNOVAL) {
911 switch (vp->v_type) {
912 case VDIR:
913 return (EISDIR);
914 case VCHR:
915 case VBLK:
916 case VSOCK:
917 case VFIFO:
918 break;
919 case VREG:
920 case VLNK:
921 default:
922 /*
923 * Disallow write attempts if the filesystem is
924 * mounted read-only.
925 */
926 if (vp->v_mount->mnt_flag & MNT_RDONLY)
927 return (EROFS);
928 }
929 }
930
931 /*
932 * Handle case of truncating lower object to zero size,
933 * by creating a zero length upper object. This is to
934 * handle the case of open with O_TRUNC and O_CREAT.
935 */
936 if ((un->un_uppervp == NULLVP) &&
937 /* assert(un->un_lowervp != NULLVP) */
938 (un->un_lowervp->v_type == VREG)) {
939 error = union_copyup(un, (vap->va_size != 0),
940 ap->a_cred, ap->a_l);
941 if (error)
942 return (error);
943 }
944
945 /*
946 * Try to set attributes in upper layer,
947 * otherwise return read-only filesystem error.
948 */
949 if (un->un_uppervp != NULLVP) {
950 FIXUP(un);
951 error = VOP_SETATTR(un->un_uppervp, vap,
952 ap->a_cred, ap->a_l);
953 if ((error == 0) && (vap->va_size != VNOVAL))
954 union_newsize(ap->a_vp, vap->va_size, VNOVAL);
955 } else {
956 error = EROFS;
957 }
958
959 return (error);
960 }
961
962 int
963 union_read(v)
964 void *v;
965 {
966 struct vop_read_args /* {
967 struct vnode *a_vp;
968 struct uio *a_uio;
969 int a_ioflag;
970 kauth_cred_t a_cred;
971 } */ *ap = v;
972 int error;
973 struct vnode *vp = OTHERVP(ap->a_vp);
974 int dolock = (vp == LOWERVP(ap->a_vp));
975
976 if (dolock)
977 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
978 else
979 FIXUP(VTOUNION(ap->a_vp));
980 error = VOP_READ(vp, ap->a_uio, ap->a_ioflag, ap->a_cred);
981 if (dolock)
982 VOP_UNLOCK(vp, 0);
983
984 /*
985 * XXX
986 * perhaps the size of the underlying object has changed under
987 * our feet. take advantage of the offset information present
988 * in the uio structure.
989 */
990 if (error == 0) {
991 struct union_node *un = VTOUNION(ap->a_vp);
992 off_t cur = ap->a_uio->uio_offset;
993
994 if (vp == un->un_uppervp) {
995 if (cur > un->un_uppersz)
996 union_newsize(ap->a_vp, cur, VNOVAL);
997 } else {
998 if (cur > un->un_lowersz)
999 union_newsize(ap->a_vp, VNOVAL, cur);
1000 }
1001 }
1002
1003 return (error);
1004 }
1005
1006 int
1007 union_write(v)
1008 void *v;
1009 {
1010 struct vop_read_args /* {
1011 struct vnode *a_vp;
1012 struct uio *a_uio;
1013 int a_ioflag;
1014 kauth_cred_t a_cred;
1015 } */ *ap = v;
1016 int error;
1017 struct vnode *vp;
1018 struct union_node *un = VTOUNION(ap->a_vp);
1019
1020 vp = UPPERVP(ap->a_vp);
1021 if (vp == NULLVP)
1022 panic("union: missing upper layer in write");
1023
1024 FIXUP(un);
1025 error = VOP_WRITE(vp, ap->a_uio, ap->a_ioflag, ap->a_cred);
1026
1027 /*
1028 * the size of the underlying object may be changed by the
1029 * write.
1030 */
1031 if (error == 0) {
1032 off_t cur = ap->a_uio->uio_offset;
1033
1034 if (cur > un->un_uppersz)
1035 union_newsize(ap->a_vp, cur, VNOVAL);
1036 }
1037
1038 return (error);
1039 }
1040
1041 int
1042 union_lease(v)
1043 void *v;
1044 {
1045 struct vop_lease_args /* {
1046 struct vnode *a_vp;
1047 struct lwp *a_l;
1048 kauth_cred_t a_cred;
1049 int a_flag;
1050 } */ *ap = v;
1051 struct vnode *ovp = OTHERVP(ap->a_vp);
1052
1053 ap->a_vp = ovp;
1054 return (VCALL(ovp, VOFFSET(vop_lease), ap));
1055 }
1056
1057 int
1058 union_ioctl(v)
1059 void *v;
1060 {
1061 struct vop_ioctl_args /* {
1062 struct vnode *a_vp;
1063 int a_command;
1064 void *a_data;
1065 int a_fflag;
1066 kauth_cred_t a_cred;
1067 struct lwp *a_l;
1068 } */ *ap = v;
1069 struct vnode *ovp = OTHERVP(ap->a_vp);
1070
1071 ap->a_vp = ovp;
1072 return (VCALL(ovp, VOFFSET(vop_ioctl), ap));
1073 }
1074
1075 int
1076 union_poll(v)
1077 void *v;
1078 {
1079 struct vop_poll_args /* {
1080 struct vnode *a_vp;
1081 int a_events;
1082 struct lwp *a_l;
1083 } */ *ap = v;
1084 struct vnode *ovp = OTHERVP(ap->a_vp);
1085
1086 ap->a_vp = ovp;
1087 return (VCALL(ovp, VOFFSET(vop_poll), ap));
1088 }
1089
1090 int
1091 union_revoke(v)
1092 void *v;
1093 {
1094 struct vop_revoke_args /* {
1095 struct vnode *a_vp;
1096 int a_flags;
1097 struct proc *a_p;
1098 } */ *ap = v;
1099 struct vnode *vp = ap->a_vp;
1100
1101 if (UPPERVP(vp))
1102 VOP_REVOKE(UPPERVP(vp), ap->a_flags);
1103 if (LOWERVP(vp))
1104 VOP_REVOKE(LOWERVP(vp), ap->a_flags);
1105 vgone(vp);
1106 return (0);
1107 }
1108
1109 int
1110 union_mmap(v)
1111 void *v;
1112 {
1113 struct vop_mmap_args /* {
1114 struct vnode *a_vp;
1115 vm_prot_t a_prot;
1116 kauth_cred_t a_cred;
1117 struct lwp *a_l;
1118 } */ *ap = v;
1119 struct vnode *ovp = OTHERVP(ap->a_vp);
1120
1121 ap->a_vp = ovp;
1122 return (VCALL(ovp, VOFFSET(vop_mmap), ap));
1123 }
1124
1125 int
1126 union_fsync(v)
1127 void *v;
1128 {
1129 struct vop_fsync_args /* {
1130 struct vnode *a_vp;
1131 kauth_cred_t a_cred;
1132 int a_flags;
1133 off_t offhi;
1134 off_t offlo;
1135 struct lwp *a_l;
1136 } */ *ap = v;
1137 int error = 0;
1138 struct lwp *l;
1139 struct vnode *targetvp;
1140
1141 /*
1142 * If vinvalbuf is calling us, it's a "shallow fsync" -- don't
1143 * bother syncing the underlying vnodes, since (a) they'll be
1144 * fsync'ed when reclaimed and (b) we could deadlock if
1145 * they're locked; otherwise, pass it through to the
1146 * underlying layer.
1147 */
1148 if (ap->a_flags & FSYNC_RECLAIM)
1149 return 0;
1150
1151 targetvp = OTHERVP(ap->a_vp);
1152 l = ap->a_l;
1153
1154 if (targetvp != NULLVP) {
1155 int dolock = (targetvp == LOWERVP(ap->a_vp));
1156
1157 if (dolock)
1158 vn_lock(targetvp, LK_EXCLUSIVE | LK_RETRY);
1159 else
1160 FIXUP(VTOUNION(ap->a_vp));
1161 error = VOP_FSYNC(targetvp, ap->a_cred, ap->a_flags,
1162 ap->a_offlo, ap->a_offhi, l);
1163 if (dolock)
1164 VOP_UNLOCK(targetvp, 0);
1165 }
1166
1167 return (error);
1168 }
1169
1170 int
1171 union_seek(v)
1172 void *v;
1173 {
1174 struct vop_seek_args /* {
1175 struct vnode *a_vp;
1176 off_t a_oldoff;
1177 off_t a_newoff;
1178 kauth_cred_t a_cred;
1179 } */ *ap = v;
1180 struct vnode *ovp = OTHERVP(ap->a_vp);
1181
1182 ap->a_vp = ovp;
1183 return (VCALL(ovp, VOFFSET(vop_seek), ap));
1184 }
1185
1186 int
1187 union_remove(v)
1188 void *v;
1189 {
1190 struct vop_remove_args /* {
1191 struct vnode *a_dvp;
1192 struct vnode *a_vp;
1193 struct componentname *a_cnp;
1194 } */ *ap = v;
1195 int error;
1196 struct union_node *dun = VTOUNION(ap->a_dvp);
1197 struct union_node *un = VTOUNION(ap->a_vp);
1198 struct componentname *cnp = ap->a_cnp;
1199
1200 if (dun->un_uppervp == NULLVP)
1201 panic("union remove: null upper vnode");
1202
1203 if (un->un_uppervp != NULLVP) {
1204 struct vnode *dvp = dun->un_uppervp;
1205 struct vnode *vp = un->un_uppervp;
1206
1207 FIXUP(dun);
1208 VREF(dvp);
1209 dun->un_flags |= UN_KLOCK;
1210 vput(ap->a_dvp);
1211 FIXUP(un);
1212 VREF(vp);
1213 un->un_flags |= UN_KLOCK;
1214 vput(ap->a_vp);
1215
1216 if (union_dowhiteout(un, cnp->cn_cred, cnp->cn_lwp))
1217 cnp->cn_flags |= DOWHITEOUT;
1218 error = VOP_REMOVE(dvp, vp, cnp);
1219 if (!error)
1220 union_removed_upper(un);
1221 } else {
1222 FIXUP(dun);
1223 error = union_mkwhiteout(
1224 MOUNTTOUNIONMOUNT(UNIONTOV(dun)->v_mount),
1225 dun->un_uppervp, ap->a_cnp, un->un_path);
1226 vput(ap->a_dvp);
1227 vput(ap->a_vp);
1228 }
1229
1230 return (error);
1231 }
1232
1233 int
1234 union_link(v)
1235 void *v;
1236 {
1237 struct vop_link_args /* {
1238 struct vnode *a_dvp;
1239 struct vnode *a_vp;
1240 struct componentname *a_cnp;
1241 } */ *ap = v;
1242 int error = 0;
1243 struct componentname *cnp = ap->a_cnp;
1244 struct lwp *l = cnp->cn_lwp;
1245 struct union_node *dun;
1246 struct vnode *vp;
1247 struct vnode *dvp;
1248
1249 dun = VTOUNION(ap->a_dvp);
1250
1251 #ifdef DIAGNOSTIC
1252 if (!(ap->a_cnp->cn_flags & LOCKPARENT)) {
1253 printf("union_link called without LOCKPARENT set!\n");
1254 error = EIO; /* need some error code for "caller is a bozo" */
1255 } else
1256 #endif
1257
1258
1259 if (ap->a_dvp->v_op != ap->a_vp->v_op) {
1260 vp = ap->a_vp;
1261 } else {
1262 struct union_node *un = VTOUNION(ap->a_vp);
1263 if (un->un_uppervp == NULLVP) {
1264 /*
1265 * Needs to be copied before we can link it.
1266 */
1267 vn_lock(ap->a_vp, LK_EXCLUSIVE | LK_RETRY);
1268 if (dun->un_uppervp == un->un_dirvp) {
1269 dun->un_flags &= ~UN_ULOCK;
1270 VOP_UNLOCK(dun->un_uppervp, 0);
1271 }
1272 error = union_copyup(un, 1, cnp->cn_cred, l);
1273 if (dun->un_uppervp == un->un_dirvp) {
1274 /*
1275 * During copyup, we dropped the lock on the
1276 * dir and invalidated any saved namei lookup
1277 * state for the directory we'll be entering
1278 * the link in. We need to re-run the lookup
1279 * in that directory to reset any state needed
1280 * for VOP_LINK.
1281 * Call relookup on the union-layer to reset
1282 * the state.
1283 */
1284 vp = NULLVP;
1285 if (dun->un_uppervp == NULLVP)
1286 panic("union: null upperdvp?");
1287 error = relookup(ap->a_dvp, &vp, ap->a_cnp);
1288 if (error) {
1289 VOP_UNLOCK(ap->a_vp, 0);
1290 return EROFS; /* ? */
1291 }
1292 if (vp != NULLVP) {
1293 /*
1294 * The name we want to create has
1295 * mysteriously appeared (a race?)
1296 */
1297 error = EEXIST;
1298 VOP_UNLOCK(ap->a_vp, 0);
1299 vput(ap->a_dvp);
1300 vput(vp);
1301 return (error);
1302 }
1303 }
1304 VOP_UNLOCK(ap->a_vp, 0);
1305 }
1306 vp = un->un_uppervp;
1307 }
1308
1309 dvp = dun->un_uppervp;
1310 if (dvp == NULLVP)
1311 error = EROFS;
1312
1313 if (error) {
1314 vput(ap->a_dvp);
1315 return (error);
1316 }
1317
1318 FIXUP(dun);
1319 VREF(dvp);
1320 dun->un_flags |= UN_KLOCK;
1321 vput(ap->a_dvp);
1322
1323 return (VOP_LINK(dvp, vp, cnp));
1324 }
1325
1326 int
1327 union_rename(v)
1328 void *v;
1329 {
1330 struct vop_rename_args /* {
1331 struct vnode *a_fdvp;
1332 struct vnode *a_fvp;
1333 struct componentname *a_fcnp;
1334 struct vnode *a_tdvp;
1335 struct vnode *a_tvp;
1336 struct componentname *a_tcnp;
1337 } */ *ap = v;
1338 int error;
1339
1340 struct vnode *fdvp = ap->a_fdvp;
1341 struct vnode *fvp = ap->a_fvp;
1342 struct vnode *tdvp = ap->a_tdvp;
1343 struct vnode *tvp = ap->a_tvp;
1344
1345 if (fdvp->v_op == union_vnodeop_p) { /* always true */
1346 struct union_node *un = VTOUNION(fdvp);
1347 if (un->un_uppervp == NULLVP) {
1348 /*
1349 * this should never happen in normal
1350 * operation but might if there was
1351 * a problem creating the top-level shadow
1352 * directory.
1353 */
1354 error = EXDEV;
1355 goto bad;
1356 }
1357
1358 fdvp = un->un_uppervp;
1359 VREF(fdvp);
1360 }
1361
1362 if (fvp->v_op == union_vnodeop_p) { /* always true */
1363 struct union_node *un = VTOUNION(fvp);
1364 if (un->un_uppervp == NULLVP) {
1365 /* XXX: should do a copyup */
1366 error = EXDEV;
1367 goto bad;
1368 }
1369
1370 if (un->un_lowervp != NULLVP)
1371 ap->a_fcnp->cn_flags |= DOWHITEOUT;
1372
1373 fvp = un->un_uppervp;
1374 VREF(fvp);
1375 }
1376
1377 if (tdvp->v_op == union_vnodeop_p) {
1378 struct union_node *un = VTOUNION(tdvp);
1379 if (un->un_uppervp == NULLVP) {
1380 /*
1381 * this should never happen in normal
1382 * operation but might if there was
1383 * a problem creating the top-level shadow
1384 * directory.
1385 */
1386 error = EXDEV;
1387 goto bad;
1388 }
1389
1390 tdvp = un->un_uppervp;
1391 VREF(tdvp);
1392 un->un_flags |= UN_KLOCK;
1393 vput(ap->a_tdvp);
1394 }
1395
1396 if (tvp != NULLVP && tvp->v_op == union_vnodeop_p) {
1397 struct union_node *un = VTOUNION(tvp);
1398
1399 tvp = un->un_uppervp;
1400 if (tvp != NULLVP) {
1401 VREF(tvp);
1402 un->un_flags |= UN_KLOCK;
1403 }
1404 vput(ap->a_tvp);
1405 }
1406
1407 error = VOP_RENAME(fdvp, fvp, ap->a_fcnp, tdvp, tvp, ap->a_tcnp);
1408 goto out;
1409
1410 bad:
1411 vput(tdvp);
1412 if (tvp != NULLVP)
1413 vput(tvp);
1414 vrele(fdvp);
1415 vrele(fvp);
1416
1417 out:
1418 if (fdvp != ap->a_fdvp) {
1419 vrele(ap->a_fdvp);
1420 }
1421 if (fvp != ap->a_fvp) {
1422 vrele(ap->a_fvp);
1423 }
1424 return (error);
1425 }
1426
1427 int
1428 union_mkdir(v)
1429 void *v;
1430 {
1431 struct vop_mkdir_args /* {
1432 struct vnode *a_dvp;
1433 struct vnode **a_vpp;
1434 struct componentname *a_cnp;
1435 struct vattr *a_vap;
1436 } */ *ap = v;
1437 struct union_node *un = VTOUNION(ap->a_dvp);
1438 struct vnode *dvp = un->un_uppervp;
1439 struct componentname *cnp = ap->a_cnp;
1440
1441 if (dvp != NULLVP) {
1442 int error;
1443 struct vnode *vp;
1444
1445 FIXUP(un);
1446 VREF(dvp);
1447 un->un_flags |= UN_KLOCK;
1448 VOP_UNLOCK(ap->a_dvp, 0);
1449 error = VOP_MKDIR(dvp, &vp, cnp, ap->a_vap);
1450 if (error) {
1451 vrele(ap->a_dvp);
1452 return (error);
1453 }
1454
1455 error = union_allocvp(ap->a_vpp, ap->a_dvp->v_mount, ap->a_dvp,
1456 NULLVP, cnp, vp, NULLVP, 1);
1457 if (error)
1458 vput(vp);
1459 vrele(ap->a_dvp);
1460 return (error);
1461 }
1462
1463 vput(ap->a_dvp);
1464 return (EROFS);
1465 }
1466
1467 int
1468 union_rmdir(v)
1469 void *v;
1470 {
1471 struct vop_rmdir_args /* {
1472 struct vnode *a_dvp;
1473 struct vnode *a_vp;
1474 struct componentname *a_cnp;
1475 } */ *ap = v;
1476 int error;
1477 struct union_node *dun = VTOUNION(ap->a_dvp);
1478 struct union_node *un = VTOUNION(ap->a_vp);
1479 struct componentname *cnp = ap->a_cnp;
1480
1481 if (dun->un_uppervp == NULLVP)
1482 panic("union rmdir: null upper vnode");
1483
1484 if (un->un_uppervp != NULLVP) {
1485 struct vnode *dvp = dun->un_uppervp;
1486 struct vnode *vp = un->un_uppervp;
1487
1488 FIXUP(dun);
1489 VREF(dvp);
1490 dun->un_flags |= UN_KLOCK;
1491 vput(ap->a_dvp);
1492 FIXUP(un);
1493 VREF(vp);
1494 un->un_flags |= UN_KLOCK;
1495 vput(ap->a_vp);
1496
1497 if (union_dowhiteout(un, cnp->cn_cred, cnp->cn_lwp))
1498 cnp->cn_flags |= DOWHITEOUT;
1499 error = VOP_RMDIR(dvp, vp, ap->a_cnp);
1500 if (!error)
1501 union_removed_upper(un);
1502 } else {
1503 FIXUP(dun);
1504 error = union_mkwhiteout(
1505 MOUNTTOUNIONMOUNT(UNIONTOV(dun)->v_mount),
1506 dun->un_uppervp, ap->a_cnp, un->un_path);
1507 vput(ap->a_dvp);
1508 vput(ap->a_vp);
1509 }
1510
1511 return (error);
1512 }
1513
1514 int
1515 union_symlink(v)
1516 void *v;
1517 {
1518 struct vop_symlink_args /* {
1519 struct vnode *a_dvp;
1520 struct vnode **a_vpp;
1521 struct componentname *a_cnp;
1522 struct vattr *a_vap;
1523 char *a_target;
1524 } */ *ap = v;
1525 struct union_node *un = VTOUNION(ap->a_dvp);
1526 struct vnode *dvp = un->un_uppervp;
1527 struct componentname *cnp = ap->a_cnp;
1528
1529 if (dvp != NULLVP) {
1530 int error;
1531
1532 FIXUP(un);
1533 VREF(dvp);
1534 un->un_flags |= UN_KLOCK;
1535 vput(ap->a_dvp);
1536 error = VOP_SYMLINK(dvp, ap->a_vpp, cnp, ap->a_vap,
1537 ap->a_target);
1538 return (error);
1539 }
1540
1541 vput(ap->a_dvp);
1542 return (EROFS);
1543 }
1544
1545 /*
1546 * union_readdir works in concert with getdirentries and
1547 * readdir(3) to provide a list of entries in the unioned
1548 * directories. getdirentries is responsible for walking
1549 * down the union stack. readdir(3) is responsible for
1550 * eliminating duplicate names from the returned data stream.
1551 */
1552 int
1553 union_readdir(v)
1554 void *v;
1555 {
1556 struct vop_readdir_args /* {
1557 struct vnodeop_desc *a_desc;
1558 struct vnode *a_vp;
1559 struct uio *a_uio;
1560 kauth_cred_t a_cred;
1561 int *a_eofflag;
1562 u_long *a_cookies;
1563 int a_ncookies;
1564 } */ *ap = v;
1565 struct union_node *un = VTOUNION(ap->a_vp);
1566 struct vnode *uvp = un->un_uppervp;
1567
1568 if (uvp == NULLVP)
1569 return (0);
1570
1571 FIXUP(un);
1572 ap->a_vp = uvp;
1573 return (VCALL(uvp, VOFFSET(vop_readdir), ap));
1574 }
1575
1576 int
1577 union_readlink(v)
1578 void *v;
1579 {
1580 struct vop_readlink_args /* {
1581 struct vnode *a_vp;
1582 struct uio *a_uio;
1583 kauth_cred_t a_cred;
1584 } */ *ap = v;
1585 int error;
1586 struct vnode *vp = OTHERVP(ap->a_vp);
1587 int dolock = (vp == LOWERVP(ap->a_vp));
1588
1589 if (dolock)
1590 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1591 else
1592 FIXUP(VTOUNION(ap->a_vp));
1593 ap->a_vp = vp;
1594 error = VCALL(vp, VOFFSET(vop_readlink), ap);
1595 if (dolock)
1596 VOP_UNLOCK(vp, 0);
1597
1598 return (error);
1599 }
1600
1601 int
1602 union_abortop(v)
1603 void *v;
1604 {
1605 struct vop_abortop_args /* {
1606 struct vnode *a_dvp;
1607 struct componentname *a_cnp;
1608 } */ *ap = v;
1609 int error;
1610 struct vnode *vp = OTHERVP(ap->a_dvp);
1611 struct union_node *un = VTOUNION(ap->a_dvp);
1612 int islocked = un->un_flags & UN_LOCKED;
1613 int dolock = (vp == LOWERVP(ap->a_dvp));
1614
1615 if (islocked) {
1616 if (dolock)
1617 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1618 else
1619 FIXUP(VTOUNION(ap->a_dvp));
1620 }
1621 ap->a_dvp = vp;
1622 error = VCALL(vp, VOFFSET(vop_abortop), ap);
1623 if (islocked && dolock)
1624 VOP_UNLOCK(vp, 0);
1625
1626 return (error);
1627 }
1628
1629 int
1630 union_inactive(v)
1631 void *v;
1632 {
1633 struct vop_inactive_args /* {
1634 const struct vnodeop_desc *a_desc;
1635 struct vnode *a_vp;
1636 struct lwp *a_l;
1637 } */ *ap = v;
1638 struct vnode *vp = ap->a_vp;
1639 struct union_node *un = VTOUNION(vp);
1640 struct vnode **vpp;
1641
1642 /*
1643 * Do nothing (and _don't_ bypass).
1644 * Wait to vrele lowervp until reclaim,
1645 * so that until then our union_node is in the
1646 * cache and reusable.
1647 *
1648 * NEEDSWORK: Someday, consider inactive'ing
1649 * the lowervp and then trying to reactivate it
1650 * with capabilities (v_id)
1651 * like they do in the name lookup cache code.
1652 * That's too much work for now.
1653 */
1654
1655 if (un->un_dircache != 0) {
1656 for (vpp = un->un_dircache; *vpp != NULLVP; vpp++)
1657 vrele(*vpp);
1658 free(un->un_dircache, M_TEMP);
1659 un->un_dircache = 0;
1660 }
1661
1662 VOP_UNLOCK(vp, 0);
1663
1664 if ((un->un_flags & UN_CACHED) == 0)
1665 vgone(vp);
1666
1667 return (0);
1668 }
1669
1670 int
1671 union_reclaim(v)
1672 void *v;
1673 {
1674 struct vop_reclaim_args /* {
1675 struct vnode *a_vp;
1676 } */ *ap = v;
1677
1678 union_freevp(ap->a_vp);
1679
1680 return (0);
1681 }
1682
1683 int
1684 union_lock(v)
1685 void *v;
1686 {
1687 struct vop_lock_args /* {
1688 struct vnode *a_vp;
1689 int a_flags;
1690 } */ *ap = v;
1691 struct vnode *vp = ap->a_vp;
1692 int flags = ap->a_flags;
1693 struct union_node *un;
1694 int error;
1695 #ifdef DIAGNOSTIC
1696 int drain = 0;
1697 #endif
1698
1699 /* XXX unionfs can't handle shared locks yet */
1700 if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1701 flags = LK_EXCLUSIVE | (flags & ~LK_TYPE_MASK);
1702 }
1703
1704 genfs_nolock(ap);
1705 /*
1706 * Need to do real lockmgr-style locking here.
1707 * in the mean time, draining won't work quite right,
1708 * which could lead to a few race conditions.
1709 * the following test was here, but is not quite right, we
1710 * still need to take the lock:
1711 if ((flags & LK_TYPE_MASK) == LK_DRAIN)
1712 return (0);
1713 */
1714 flags &= ~LK_INTERLOCK;
1715
1716 un = VTOUNION(vp);
1717 #ifdef DIAGNOSTIC
1718 if (un->un_flags & (UN_DRAINING|UN_DRAINED)) {
1719 if (un->un_flags & UN_DRAINED)
1720 panic("union: %p: warning: locking decommissioned lock", vp);
1721 if ((flags & LK_TYPE_MASK) != LK_RELEASE)
1722 panic("union: %p: non-release on draining lock: %d",
1723 vp, flags & LK_TYPE_MASK);
1724 un->un_flags &= ~UN_DRAINING;
1725 if ((flags & LK_REENABLE) == 0)
1726 un->un_flags |= UN_DRAINED;
1727 }
1728 #endif
1729
1730 /*
1731 * Don't pass DRAIN through to sub-vnode lock; keep track of
1732 * DRAIN state at this level, and just get an exclusive lock
1733 * on the underlying vnode.
1734 */
1735 if ((flags & LK_TYPE_MASK) == LK_DRAIN) {
1736 #ifdef DIAGNOSTIC
1737 drain = 1;
1738 #endif
1739 flags = LK_EXCLUSIVE | (flags & ~LK_TYPE_MASK);
1740 }
1741 start:
1742 un = VTOUNION(vp);
1743
1744 if (un->un_uppervp != NULLVP) {
1745 if (((un->un_flags & UN_ULOCK) == 0) &&
1746 (vp->v_usecount != 0)) {
1747 /*
1748 * We MUST always use the order of: take upper
1749 * vp lock, manipulate union node flags, drop
1750 * upper vp lock. This code must not be an
1751 */
1752 error = vn_lock(un->un_uppervp, flags);
1753 if (error)
1754 return (error);
1755 un->un_flags |= UN_ULOCK;
1756 }
1757 #ifdef DIAGNOSTIC
1758 if (un->un_flags & UN_KLOCK) {
1759 vprint("union: dangling klock", vp);
1760 panic("union: dangling upper lock (%p)", vp);
1761 }
1762 #endif
1763 }
1764
1765 /* XXX ignores LK_NOWAIT */
1766 if (un->un_flags & UN_LOCKED) {
1767 #ifdef DIAGNOSTIC
1768 if (curproc && un->un_pid == curproc->p_pid &&
1769 un->un_pid > -1 && curproc->p_pid > -1)
1770 panic("union: locking against myself");
1771 #endif
1772 un->un_flags |= UN_WANTED;
1773 tsleep(&un->un_flags, PINOD, "unionlk2", 0);
1774 goto start;
1775 }
1776
1777 #ifdef DIAGNOSTIC
1778 if (curproc)
1779 un->un_pid = curproc->p_pid;
1780 else
1781 un->un_pid = -1;
1782 if (drain)
1783 un->un_flags |= UN_DRAINING;
1784 #endif
1785
1786 un->un_flags |= UN_LOCKED;
1787 return (0);
1788 }
1789
1790 /*
1791 * When operations want to vput() a union node yet retain a lock on
1792 * the upper vnode (say, to do some further operations like link(),
1793 * mkdir(), ...), they set UN_KLOCK on the union node, then call
1794 * vput() which calls VOP_UNLOCK() and comes here. union_unlock()
1795 * unlocks the union node (leaving the upper vnode alone), clears the
1796 * KLOCK flag, and then returns to vput(). The caller then does whatever
1797 * is left to do with the upper vnode, and ensures that it gets unlocked.
1798 *
1799 * If UN_KLOCK isn't set, then the upper vnode is unlocked here.
1800 */
1801 int
1802 union_unlock(v)
1803 void *v;
1804 {
1805 struct vop_unlock_args /* {
1806 struct vnode *a_vp;
1807 int a_flags;
1808 } */ *ap = v;
1809 struct union_node *un = VTOUNION(ap->a_vp);
1810
1811 #ifdef DIAGNOSTIC
1812 if ((un->un_flags & UN_LOCKED) == 0)
1813 panic("union: unlock unlocked node");
1814 if (curproc && un->un_pid != curproc->p_pid &&
1815 curproc->p_pid > -1 && un->un_pid > -1)
1816 panic("union: unlocking other process's union node");
1817 if (un->un_flags & UN_DRAINED)
1818 panic("union: %p: warning: unlocking decommissioned lock", ap->a_vp);
1819 #endif
1820
1821 un->un_flags &= ~UN_LOCKED;
1822
1823 if ((un->un_flags & (UN_ULOCK|UN_KLOCK)) == UN_ULOCK)
1824 VOP_UNLOCK(un->un_uppervp, 0);
1825
1826 un->un_flags &= ~(UN_ULOCK|UN_KLOCK);
1827
1828 if (un->un_flags & UN_WANTED) {
1829 un->un_flags &= ~UN_WANTED;
1830 wakeup( &un->un_flags);
1831 }
1832
1833 #ifdef DIAGNOSTIC
1834 un->un_pid = 0;
1835 if (un->un_flags & UN_DRAINING) {
1836 un->un_flags |= UN_DRAINED;
1837 un->un_flags &= ~UN_DRAINING;
1838 }
1839 #endif
1840 genfs_nounlock(ap);
1841
1842 return (0);
1843 }
1844
1845 int
1846 union_bmap(v)
1847 void *v;
1848 {
1849 struct vop_bmap_args /* {
1850 struct vnode *a_vp;
1851 daddr_t a_bn;
1852 struct vnode **a_vpp;
1853 daddr_t *a_bnp;
1854 int *a_runp;
1855 } */ *ap = v;
1856 int error;
1857 struct vnode *vp = OTHERVP(ap->a_vp);
1858 int dolock = (vp == LOWERVP(ap->a_vp));
1859
1860 if (dolock)
1861 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1862 else
1863 FIXUP(VTOUNION(ap->a_vp));
1864 ap->a_vp = vp;
1865 error = VCALL(vp, VOFFSET(vop_bmap), ap);
1866 if (dolock)
1867 VOP_UNLOCK(vp, 0);
1868
1869 return (error);
1870 }
1871
1872 int
1873 union_print(v)
1874 void *v;
1875 {
1876 struct vop_print_args /* {
1877 struct vnode *a_vp;
1878 } */ *ap = v;
1879 struct vnode *vp = ap->a_vp;
1880
1881 printf("\ttag VT_UNION, vp=%p, uppervp=%p, lowervp=%p\n",
1882 vp, UPPERVP(vp), LOWERVP(vp));
1883 if (UPPERVP(vp) != NULLVP)
1884 vprint("union: upper", UPPERVP(vp));
1885 if (LOWERVP(vp) != NULLVP)
1886 vprint("union: lower", LOWERVP(vp));
1887 if (VTOUNION(vp)->un_dircache) {
1888 struct vnode **vpp;
1889 for (vpp = VTOUNION(vp)->un_dircache; *vpp != NULLVP; vpp++)
1890 vprint("dircache:", *vpp);
1891 }
1892
1893 return (0);
1894 }
1895
1896 int
1897 union_islocked(v)
1898 void *v;
1899 {
1900 struct vop_islocked_args /* {
1901 struct vnode *a_vp;
1902 } */ *ap = v;
1903
1904 return ((VTOUNION(ap->a_vp)->un_flags & UN_LOCKED) ? 1 : 0);
1905 }
1906
1907 int
1908 union_pathconf(v)
1909 void *v;
1910 {
1911 struct vop_pathconf_args /* {
1912 struct vnode *a_vp;
1913 int a_name;
1914 int *a_retval;
1915 } */ *ap = v;
1916 int error;
1917 struct vnode *vp = OTHERVP(ap->a_vp);
1918 int dolock = (vp == LOWERVP(ap->a_vp));
1919
1920 if (dolock)
1921 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
1922 else
1923 FIXUP(VTOUNION(ap->a_vp));
1924 ap->a_vp = vp;
1925 error = VCALL(vp, VOFFSET(vop_pathconf), ap);
1926 if (dolock)
1927 VOP_UNLOCK(vp, 0);
1928
1929 return (error);
1930 }
1931
1932 int
1933 union_advlock(v)
1934 void *v;
1935 {
1936 struct vop_advlock_args /* {
1937 struct vnode *a_vp;
1938 void *a_id;
1939 int a_op;
1940 struct flock *a_fl;
1941 int a_flags;
1942 } */ *ap = v;
1943 struct vnode *ovp = OTHERVP(ap->a_vp);
1944
1945 ap->a_vp = ovp;
1946 return (VCALL(ovp, VOFFSET(vop_advlock), ap));
1947 }
1948
1949
1950 /*
1951 * XXX - vop_strategy must be hand coded because it has no
1952 * vnode in its arguments.
1953 * This goes away with a merged VM/buffer cache.
1954 */
1955 int
1956 union_strategy(v)
1957 void *v;
1958 {
1959 struct vop_strategy_args /* {
1960 struct vnode *a_vp;
1961 struct buf *a_bp;
1962 } */ *ap = v;
1963 struct vnode *ovp = OTHERVP(ap->a_vp);
1964 struct buf *bp = ap->a_bp;
1965
1966 #ifdef DIAGNOSTIC
1967 if (ovp == NULLVP)
1968 panic("union_strategy: nil vp");
1969 if (((bp->b_flags & B_READ) == 0) &&
1970 (ovp == LOWERVP(bp->b_vp)))
1971 panic("union_strategy: writing to lowervp");
1972 #endif
1973
1974 return (VOP_STRATEGY(ovp, bp));
1975 }
1976
1977 int
1978 union_getpages(v)
1979 void *v;
1980 {
1981 struct vop_getpages_args /* {
1982 struct vnode *a_vp;
1983 voff_t a_offset;
1984 struct vm_page **a_m;
1985 int *a_count;
1986 int a_centeridx;
1987 vm_prot_t a_access_type;
1988 int a_advice;
1989 int a_flags;
1990 } */ *ap = v;
1991 struct vnode *vp = ap->a_vp;
1992 int error;
1993
1994 /*
1995 * just pass the request on to the underlying layer.
1996 */
1997
1998 if (ap->a_flags & PGO_LOCKED) {
1999 return EBUSY;
2000 }
2001 ap->a_vp = OTHERVP(vp);
2002 simple_unlock(&vp->v_interlock);
2003 simple_lock(&ap->a_vp->v_interlock);
2004 error = VCALL(ap->a_vp, VOFFSET(vop_getpages), ap);
2005 return error;
2006 }
2007
2008 int
2009 union_putpages(v)
2010 void *v;
2011 {
2012 struct vop_putpages_args /* {
2013 struct vnode *a_vp;
2014 voff_t a_offlo;
2015 voff_t a_offhi;
2016 int a_flags;
2017 } */ *ap = v;
2018 struct vnode *vp = ap->a_vp;
2019 int error;
2020
2021 /*
2022 * just pass the request on to the underlying layer.
2023 */
2024
2025 ap->a_vp = OTHERVP(vp);
2026 simple_unlock(&vp->v_interlock);
2027 if (ap->a_flags & PGO_RECLAIM) {
2028 return 0;
2029 }
2030 simple_lock(&ap->a_vp->v_interlock);
2031 error = VCALL(ap->a_vp, VOFFSET(vop_putpages), ap);
2032 return error;
2033 }
2034
2035 int
2036 union_kqfilter(void *v)
2037 {
2038 struct vop_kqfilter_args /* {
2039 struct vnode *a_vp;
2040 struct knote *a_kn;
2041 } */ *ap = v;
2042 int error;
2043
2044 /*
2045 * We watch either the upper layer file (if it already exists),
2046 * or the lower layer one. If there is lower layer file only
2047 * at this moment, we will keep watching that lower layer file
2048 * even if upper layer file would be created later on.
2049 */
2050 if (UPPERVP(ap->a_vp))
2051 error = VOP_KQFILTER(UPPERVP(ap->a_vp), ap->a_kn);
2052 else if (LOWERVP(ap->a_vp))
2053 error = VOP_KQFILTER(LOWERVP(ap->a_vp), ap->a_kn);
2054 else {
2055 /* panic? */
2056 error = EOPNOTSUPP;
2057 }
2058
2059 return (error);
2060 }
2061