1/*-
2 * Copyright (c) 1994 Jan-Simon Pendry
3 * Copyright (c) 1994
4 * The Regents of the University of California. All rights reserved.
5 * Copyright (c) 2005, 2006 Masanori Ozawa <ozawa@ongs.co.jp>, ONGS Inc.
6 * Copyright (c) 2006 Daichi Goto <daichi@freebsd.org>
7 *
8 * This code is derived from software contributed to Berkeley by
9 * Jan-Simon Pendry.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 4. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * SUCH DAMAGE.
34 *
35 * @(#)union_subr.c 8.20 (Berkeley) 5/20/95
36 * $FreeBSD: head/sys/fs/unionfs/union_subr.c 190888 2009-04-10 10:52:19Z rwatson $
37 */
38
39#include <sys/param.h>
40#include <sys/systm.h>
41#include <sys/kernel.h>
42#include <sys/lock.h>
43#include <sys/mutex.h>
44#include <sys/malloc.h>
45#include <sys/mount.h>
46#include <sys/namei.h>
47#include <sys/proc.h>
48#include <sys/vnode.h>
49#include <sys/dirent.h>
50#include <sys/fcntl.h>
51#include <sys/filedesc.h>
52#include <sys/stat.h>
53#include <sys/resourcevar.h>
54
55#ifdef MAC
56#include <sys/mac.h>
57#endif
58
59#include <vm/uma.h>
60
61#include <fs/unionfs/union.h>
62
63#define NUNIONFSNODECACHE 16
64
65static MALLOC_DEFINE(M_UNIONFSHASH, "UNIONFS hash", "UNIONFS hash table");
66MALLOC_DEFINE(M_UNIONFSNODE, "UNIONFS node", "UNIONFS vnode private part");
67MALLOC_DEFINE(M_UNIONFSPATH, "UNIONFS path", "UNIONFS path private part");
68
69/*
70 * Initialize
71 */
72int
73unionfs_init(struct vfsconf *vfsp)
74{
75 UNIONFSDEBUG("unionfs_init\n"); /* printed during system boot */
76 return (0);
77}
78
79/*
80 * Uninitialize
81 */
82int
83unionfs_uninit(struct vfsconf *vfsp)
84{
85 return (0);
86}
87
88static struct unionfs_node_hashhead *
89unionfs_get_hashhead(struct vnode *dvp, char *path)
90{
91 int count;
92 char hash;
93 struct unionfs_node *unp;
94
95 hash = 0;
96 unp = VTOUNIONFS(dvp);
97 if (path != NULL) {
98 for (count = 0; path[count]; count++)
99 hash += path[count];
100 }
101
102 return (&(unp->un_hashtbl[hash & (unp->un_hashmask)]));
103}
104
105/*
106 * Get the cached vnode.
107 */
108static struct vnode *
109unionfs_get_cached_vnode(struct vnode *uvp, struct vnode *lvp,
110 struct vnode *dvp, char *path)
111{
112 struct unionfs_node_hashhead *hd;
113 struct unionfs_node *unp;
114 struct vnode *vp;
115
116 KASSERT((uvp == NULLVP || uvp->v_type == VDIR),
117 ("unionfs_get_cached_vnode: v_type != VDIR"));
118 KASSERT((lvp == NULLVP || lvp->v_type == VDIR),
119 ("unionfs_get_cached_vnode: v_type != VDIR"));
120
121 VI_LOCK(dvp);
122 hd = unionfs_get_hashhead(dvp, path);
123 LIST_FOREACH(unp, hd, un_hash) {
124 if (!strcmp(unp->un_path, path)) {
125 vp = UNIONFSTOV(unp);
126 VI_LOCK_FLAGS(vp, MTX_DUPOK);
127 VI_UNLOCK(dvp);
128 vp->v_iflag &= ~VI_OWEINACT;
129 if ((vp->v_iflag & (VI_DOOMED | VI_DOINGINACT)) != 0) {
130 VI_UNLOCK(vp);
131 vp = NULLVP;
132 } else
133 VI_UNLOCK(vp);
134 return (vp);
135 }
136 }
137 VI_UNLOCK(dvp);
138
139 return (NULLVP);
140}
141
142/*
143 * Add the new vnode into cache.
144 */
145static struct vnode *
146unionfs_ins_cached_vnode(struct unionfs_node *uncp,
147 struct vnode *dvp, char *path)
148{
149 struct unionfs_node_hashhead *hd;
150 struct unionfs_node *unp;
151 struct vnode *vp;
152
153 KASSERT((uncp->un_uppervp==NULLVP || uncp->un_uppervp->v_type==VDIR),
154 ("unionfs_ins_cached_vnode: v_type != VDIR"));
155 KASSERT((uncp->un_lowervp==NULLVP || uncp->un_lowervp->v_type==VDIR),
156 ("unionfs_ins_cached_vnode: v_type != VDIR"));
157
158 VI_LOCK(dvp);
159 hd = unionfs_get_hashhead(dvp, path);
160 LIST_FOREACH(unp, hd, un_hash) {
161 if (!strcmp(unp->un_path, path)) {
162 vp = UNIONFSTOV(unp);
163 VI_LOCK_FLAGS(vp, MTX_DUPOK);
164 vp->v_iflag &= ~VI_OWEINACT;
165 if ((vp->v_iflag & (VI_DOOMED | VI_DOINGINACT)) != 0) {
166 LIST_INSERT_HEAD(hd, uncp, un_hash);
167 VI_UNLOCK(vp);
168 vp = NULLVP;
169 } else
170 VI_UNLOCK(vp);
171 VI_UNLOCK(dvp);
172 return (vp);
173 }
174 }
175
176 LIST_INSERT_HEAD(hd, uncp, un_hash);
177 VI_UNLOCK(dvp);
178
179 return (NULLVP);
180}
181
182/*
183 * Remove the vnode.
184 */
185static void
186unionfs_rem_cached_vnode(struct unionfs_node *unp, struct vnode *dvp)
187{
188 KASSERT((unp != NULL), ("unionfs_rem_cached_vnode: null node"));
189 KASSERT((dvp != NULLVP),
190 ("unionfs_rem_cached_vnode: null parent vnode"));
191 KASSERT((unp->un_hash.le_prev != NULL),
192 ("unionfs_rem_cached_vnode: null hash"));
193
194 VI_LOCK(dvp);
195 LIST_REMOVE(unp, un_hash);
196 unp->un_hash.le_next = NULL;
197 unp->un_hash.le_prev = NULL;
198 VI_UNLOCK(dvp);
199}
200
201/*
202 * Make a new or get existing unionfs node.
203 *
204 * uppervp and lowervp should be unlocked. Because if new unionfs vnode is
205 * locked, uppervp or lowervp is locked too. In order to prevent dead lock,
206 * you should not lock plurality simultaneously.
207 */
208int
209unionfs_nodeget(struct mount *mp, struct vnode *uppervp,
210 struct vnode *lowervp, struct vnode *dvp,
211 struct vnode **vpp, struct componentname *cnp,
212 struct thread *td)
213{
214 struct unionfs_mount *ump;
215 struct unionfs_node *unp;
216 struct vnode *vp;
217 int error;
218 int lkflags;
219 enum vtype vt;
220 char *path;
221
222 ump = MOUNTTOUNIONFSMOUNT(mp);
223 lkflags = (cnp ? cnp->cn_lkflags : 0);
224 path = (cnp ? cnp->cn_nameptr : NULL);
225 *vpp = NULLVP;
226
227 if (uppervp == NULLVP && lowervp == NULLVP)
228 panic("unionfs_nodeget: upper and lower is null");
229
230 vt = (uppervp != NULLVP ? uppervp->v_type : lowervp->v_type);
231
232 /* If it has no ISLASTCN flag, path check is skipped. */
233 if (cnp && !(cnp->cn_flags & ISLASTCN))
234 path = NULL;
235
236 /* check the cache */
237 if (path != NULL && dvp != NULLVP && vt == VDIR) {
238 vp = unionfs_get_cached_vnode(uppervp, lowervp, dvp, path);
239 if (vp != NULLVP) {
240 vref(vp);
241 *vpp = vp;
242 goto unionfs_nodeget_out;
243 }
244 }
245
246 if ((uppervp == NULLVP || ump->um_uppervp != uppervp) ||
247 (lowervp == NULLVP || ump->um_lowervp != lowervp)) {
248 /* dvp will be NULLVP only in case of root vnode. */
249 if (dvp == NULLVP)
250 return (EINVAL);
251 }
252
253 /*
254 * Do the MALLOC before the getnewvnode since doing so afterward
255 * might cause a bogus v_data pointer to get dereferenced elsewhere
256 * if MALLOC should block.
257 */
258 unp = malloc(sizeof(struct unionfs_node),
259 M_UNIONFSNODE, M_WAITOK | M_ZERO);
260
261 error = getnewvnode("unionfs", mp, &unionfs_vnodeops, &vp);
262 if (error != 0) {
263 free(unp, M_UNIONFSNODE);
264 return (error);
265 }
266 error = insmntque(vp, mp); /* XXX: Too early for mpsafe fs */
267 if (error != 0) {
268 free(unp, M_UNIONFSNODE);
269 return (error);
270 }
271 if (dvp != NULLVP)
272 vref(dvp);
273 if (uppervp != NULLVP)
274 vref(uppervp);
275 if (lowervp != NULLVP)
276 vref(lowervp);
277
278 if (vt == VDIR)
279 unp->un_hashtbl = hashinit(NUNIONFSNODECACHE, M_UNIONFSHASH,
280 &(unp->un_hashmask));
281
282 unp->un_vnode = vp;
283 unp->un_uppervp = uppervp;
284 unp->un_lowervp = lowervp;
285 unp->un_dvp = dvp;
286 if (uppervp != NULLVP)
287 vp->v_vnlock = uppervp->v_vnlock;
288 else
289 vp->v_vnlock = lowervp->v_vnlock;
290
291 if (path != NULL) {
292 unp->un_path = (char *)
293 malloc(cnp->cn_namelen +1, M_UNIONFSPATH, M_WAITOK|M_ZERO);
294 bcopy(cnp->cn_nameptr, unp->un_path, cnp->cn_namelen);
295 unp->un_path[cnp->cn_namelen] = '\0';
296 }
297 vp->v_type = vt;
298 vp->v_data = unp;
299
300 if ((uppervp != NULLVP && ump->um_uppervp == uppervp) &&
301 (lowervp != NULLVP && ump->um_lowervp == lowervp))
302 vp->v_vflag |= VV_ROOT;
303
304 if (path != NULL && dvp != NULLVP && vt == VDIR)
305 *vpp = unionfs_ins_cached_vnode(unp, dvp, path);
306 if ((*vpp) != NULLVP) {
307 if (dvp != NULLVP)
308 vrele(dvp);
309 if (uppervp != NULLVP)
310 vrele(uppervp);
311 if (lowervp != NULLVP)
312 vrele(lowervp);
313
314 unp->un_uppervp = NULLVP;
315 unp->un_lowervp = NULLVP;
316 unp->un_dvp = NULLVP;
317 vrele(vp);
318 vp = *vpp;
319 vref(vp);
320 } else
321 *vpp = vp;
322
323unionfs_nodeget_out:
324 if (lkflags & LK_TYPE_MASK)
325 vn_lock(vp, lkflags | LK_RETRY);
326
327 return (0);
328}
329
330/*
331 * Clean up the unionfs node.
332 */
333void
334unionfs_noderem(struct vnode *vp, struct thread *td)
335{
336 int vfslocked;
337 int count;
338 struct unionfs_node *unp, *unp_t1, *unp_t2;
339 struct unionfs_node_hashhead *hd;
340 struct unionfs_node_status *unsp, *unsp_tmp;
341 struct vnode *lvp;
342 struct vnode *uvp;
343 struct vnode *dvp;
344
345 /*
346 * Use the interlock to protect the clearing of v_data to
347 * prevent faults in unionfs_lock().
348 */
349 VI_LOCK(vp);
350 unp = VTOUNIONFS(vp);
351 lvp = unp->un_lowervp;
352 uvp = unp->un_uppervp;
353 dvp = unp->un_dvp;
354 unp->un_lowervp = unp->un_uppervp = NULLVP;
355
356 vp->v_vnlock = &(vp->v_lock);
357 vp->v_data = NULL;
358 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_INTERLOCK, VI_MTX(vp));
359 if (lvp != NULLVP)
360 VOP_UNLOCK(lvp, 0);
361 if (uvp != NULLVP)
362 VOP_UNLOCK(uvp, 0);
363 vp->v_object = NULL;
364
365 if (dvp != NULLVP && unp->un_hash.le_prev != NULL)
366 unionfs_rem_cached_vnode(unp, dvp);
367
368 if (lvp != NULLVP) {
369 vfslocked = VFS_LOCK_GIANT(lvp->v_mount);
370 vrele(lvp);
371 VFS_UNLOCK_GIANT(vfslocked);
372 }
373 if (uvp != NULLVP) {
374 vfslocked = VFS_LOCK_GIANT(uvp->v_mount);
375 vrele(uvp);
376 VFS_UNLOCK_GIANT(vfslocked);
377 }
378 if (dvp != NULLVP) {
379 vfslocked = VFS_LOCK_GIANT(dvp->v_mount);
380 vrele(dvp);
381 VFS_UNLOCK_GIANT(vfslocked);
382 unp->un_dvp = NULLVP;
383 }
384 if (unp->un_path != NULL) {
385 free(unp->un_path, M_UNIONFSPATH);
386 unp->un_path = NULL;
387 }
388
389 if (unp->un_hashtbl != NULL) {
390 for (count = 0; count <= unp->un_hashmask; count++) {
391 hd = unp->un_hashtbl + count;
392 LIST_FOREACH_SAFE(unp_t1, hd, un_hash, unp_t2) {
393 LIST_REMOVE(unp_t1, un_hash);
394 unp_t1->un_hash.le_next = NULL;
395 unp_t1->un_hash.le_prev = NULL;
396 }
397 }
398 hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, unp->un_hashmask);
399 }
400
401 LIST_FOREACH_SAFE(unsp, &(unp->un_unshead), uns_list, unsp_tmp) {
402 LIST_REMOVE(unsp, uns_list);
403 free(unsp, M_TEMP);
404 }
405 free(unp, M_UNIONFSNODE);
406}
407
408/*
409 * Get the unionfs node status.
410 * You need exclusive lock this vnode.
411 */
412void
413unionfs_get_node_status(struct unionfs_node *unp, struct thread *td,
414 struct unionfs_node_status **unspp)
415{
416 struct unionfs_node_status *unsp;
417 pid_t pid = td->td_proc->p_pid;
418
419 KASSERT(NULL != unspp, ("null pointer"));
420 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), "unionfs_get_node_status");
421
422 LIST_FOREACH(unsp, &(unp->un_unshead), uns_list) {
423 if (unsp->uns_pid == pid) {
424 *unspp = unsp;
425 return;
426 }
427 }
428
429 /* create a new unionfs node status */
430 unsp = malloc(sizeof(struct unionfs_node_status),
431 M_TEMP, M_WAITOK | M_ZERO);
432
433 unsp->uns_pid = pid;
434 LIST_INSERT_HEAD(&(unp->un_unshead), unsp, uns_list);
435
436 *unspp = unsp;
437}
438
439/*
440 * Remove the unionfs node status, if you can.
441 * You need exclusive lock this vnode.
442 */
443void
444unionfs_tryrem_node_status(struct unionfs_node *unp,
445 struct unionfs_node_status *unsp)
446{
447 KASSERT(NULL != unsp, ("null pointer"));
448 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), "unionfs_get_node_status");
449
450 if (0 < unsp->uns_lower_opencnt || 0 < unsp->uns_upper_opencnt)
451 return;
452
453 LIST_REMOVE(unsp, uns_list);
454 free(unsp, M_TEMP);
455}
456
457/*
458 * Create upper node attr.
459 */
460void
461unionfs_create_uppervattr_core(struct unionfs_mount *ump,
462 struct vattr *lva,
463 struct vattr *uva,
464 struct thread *td)
465{
466 VATTR_NULL(uva);
467 uva->va_type = lva->va_type;
468 uva->va_atime = lva->va_atime;
469 uva->va_mtime = lva->va_mtime;
470 uva->va_ctime = lva->va_ctime;
471
472 switch (ump->um_copymode) {
473 case UNIONFS_TRANSPARENT:
474 uva->va_mode = lva->va_mode;
475 uva->va_uid = lva->va_uid;
476 uva->va_gid = lva->va_gid;
477 break;
478 case UNIONFS_MASQUERADE:
479 if (ump->um_uid == lva->va_uid) {
480 uva->va_mode = lva->va_mode & 077077;
481 uva->va_mode |= (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile) & 0700;
482 uva->va_uid = lva->va_uid;
483 uva->va_gid = lva->va_gid;
484 } else {
485 uva->va_mode = (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile);
486 uva->va_uid = ump->um_uid;
487 uva->va_gid = ump->um_gid;
488 }
489 break;
490 default: /* UNIONFS_TRADITIONAL */
491 FILEDESC_SLOCK(td->td_proc->p_fd);
492 uva->va_mode = 0777 & ~td->td_proc->p_fd->fd_cmask;
493 FILEDESC_SUNLOCK(td->td_proc->p_fd);
494 uva->va_uid = ump->um_uid;
495 uva->va_gid = ump->um_gid;
496 break;
497 }
498}
499
500/*
501 * Create upper node attr.
502 */
503int
504unionfs_create_uppervattr(struct unionfs_mount *ump,
505 struct vnode *lvp,
506 struct vattr *uva,
507 struct ucred *cred,
508 struct thread *td)
509{
510 int error;
511 struct vattr lva;
512
513 if ((error = VOP_GETATTR(lvp, &lva, cred)))
514 return (error);
515
516 unionfs_create_uppervattr_core(ump, &lva, uva, td);
517
518 return (error);
519}
520
521/*
522 * relookup
523 *
524 * dvp should be locked on entry and will be locked on return.
525 *
526 * If an error is returned, *vpp will be invalid, otherwise it will hold a
527 * locked, referenced vnode. If *vpp == dvp then remember that only one
528 * LK_EXCLUSIVE lock is held.
529 */
530int
531unionfs_relookup(struct vnode *dvp, struct vnode **vpp,
532 struct componentname *cnp, struct componentname *cn,
533 struct thread *td, char *path, int pathlen, u_long nameiop)
534{
535 int error;
536
537 cn->cn_namelen = pathlen;
538 cn->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
539 bcopy(path, cn->cn_pnbuf, pathlen);
540 cn->cn_pnbuf[pathlen] = '\0';
541
542 cn->cn_nameiop = nameiop;
543 cn->cn_flags = (LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME | ISLASTCN);
544 cn->cn_lkflags = LK_EXCLUSIVE;
545 cn->cn_thread = td;
546 cn->cn_cred = cnp->cn_cred;
547
548 cn->cn_nameptr = cn->cn_pnbuf;
549 cn->cn_consume = cnp->cn_consume;
550
551 if (nameiop == DELETE)
552 cn->cn_flags |= (cnp->cn_flags & (DOWHITEOUT | SAVESTART));
553 else if (RENAME == nameiop)
554 cn->cn_flags |= (cnp->cn_flags & SAVESTART);
555
556 vref(dvp);
557 VOP_UNLOCK(dvp, 0);
558
559 if ((error = relookup(dvp, vpp, cn))) {
560 uma_zfree(namei_zone, cn->cn_pnbuf);
561 cn->cn_flags &= ~HASBUF;
562 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
563 } else
564 vrele(dvp);
565
566 return (error);
567}
568
569/*
570 * relookup for CREATE namei operation.
571 *
572 * dvp is unionfs vnode. dvp should be locked.
573 *
574 * If it called 'unionfs_copyfile' function by unionfs_link etc,
575 * VOP_LOOKUP information is broken.
576 * So it need relookup in order to create link etc.
577 */
578int
579unionfs_relookup_for_create(struct vnode *dvp, struct componentname *cnp,
580 struct thread *td)
581{
582 int error;
583 struct vnode *udvp;
584 struct vnode *vp;
585 struct componentname cn;
586
587 udvp = UNIONFSVPTOUPPERVP(dvp);
588 vp = NULLVP;
589
590 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
591 strlen(cnp->cn_nameptr), CREATE);
592 if (error)
593 return (error);
594
595 if (vp != NULLVP) {
596 if (udvp == vp)
597 vrele(vp);
598 else
599 vput(vp);
600
601 error = EEXIST;
602 }
603
604 if (cn.cn_flags & HASBUF) {
605 uma_zfree(namei_zone, cn.cn_pnbuf);
606 cn.cn_flags &= ~HASBUF;
607 }
608
609 if (!error) {
610 cn.cn_flags |= (cnp->cn_flags & HASBUF);
611 cnp->cn_flags = cn.cn_flags;
612 }
613
614 return (error);
615}
616
617/*
618 * relookup for DELETE namei operation.
619 *
620 * dvp is unionfs vnode. dvp should be locked.
621 */
622int
623unionfs_relookup_for_delete(struct vnode *dvp, struct componentname *cnp,
624 struct thread *td)
625{
626 int error;
627 struct vnode *udvp;
628 struct vnode *vp;
629 struct componentname cn;
630
631 udvp = UNIONFSVPTOUPPERVP(dvp);
632 vp = NULLVP;
633
634 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
635 strlen(cnp->cn_nameptr), DELETE);
636 if (error)
637 return (error);
638
639 if (vp == NULLVP)
640 error = ENOENT;
641 else {
642 if (udvp == vp)
643 vrele(vp);
644 else
645 vput(vp);
646 }
647
648 if (cn.cn_flags & HASBUF) {
649 uma_zfree(namei_zone, cn.cn_pnbuf);
650 cn.cn_flags &= ~HASBUF;
651 }
652
653 if (!error) {
654 cn.cn_flags |= (cnp->cn_flags & HASBUF);
655 cnp->cn_flags = cn.cn_flags;
656 }
657
658 return (error);
659}
660
661/*
662 * relookup for RENAME namei operation.
663 *
664 * dvp is unionfs vnode. dvp should be locked.
665 */
666int
667unionfs_relookup_for_rename(struct vnode *dvp, struct componentname *cnp,
668 struct thread *td)
669{
670 int error;
671 struct vnode *udvp;
672 struct vnode *vp;
673 struct componentname cn;
674
675 udvp = UNIONFSVPTOUPPERVP(dvp);
676 vp = NULLVP;
677
678 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
679 strlen(cnp->cn_nameptr), RENAME);
680 if (error)
681 return (error);
682
683 if (vp != NULLVP) {
684 if (udvp == vp)
685 vrele(vp);
686 else
687 vput(vp);
688 }
689
690 if (cn.cn_flags & HASBUF) {
691 uma_zfree(namei_zone, cn.cn_pnbuf);
692 cn.cn_flags &= ~HASBUF;
693 }
694
695 if (!error) {
696 cn.cn_flags |= (cnp->cn_flags & HASBUF);
697 cnp->cn_flags = cn.cn_flags;
698 }
699
700 return (error);
701
702}
703
704/*
705 * Update the unionfs_node.
706 *
707 * uvp is new locked upper vnode. unionfs vnode's lock will be exchanged to the
708 * uvp's lock and lower's lock will be unlocked.
709 */
710static void
711unionfs_node_update(struct unionfs_node *unp, struct vnode *uvp,
712 struct thread *td)
713{
714 unsigned count, lockrec;
715 struct vnode *vp;
716 struct vnode *lvp;
717 struct vnode *dvp;
718
719 vp = UNIONFSTOV(unp);
720 lvp = unp->un_lowervp;
721 ASSERT_VOP_ELOCKED(lvp, "unionfs_node_update");
722 dvp = unp->un_dvp;
723
724 /*
725 * lock update
726 */
727 VI_LOCK(vp);
728 unp->un_uppervp = uvp;
729 vp->v_vnlock = uvp->v_vnlock;
730 VI_UNLOCK(vp);
731 lockrec = lvp->v_vnlock->lk_recurse;
732 for (count = 0; count < lockrec; count++)
733 vn_lock(uvp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY);
734
735 /*
736 * cache update
737 */
738 if (unp->un_path != NULL && dvp != NULLVP && vp->v_type == VDIR) {
739 static struct unionfs_node_hashhead *hd;
740
741 VI_LOCK(dvp);
742 hd = unionfs_get_hashhead(dvp, unp->un_path);
743 LIST_REMOVE(unp, un_hash);
744 LIST_INSERT_HEAD(hd, unp, un_hash);
745 VI_UNLOCK(dvp);
746 }
747}
748
749/*
750 * Create a new shadow dir.
751 *
752 * udvp should be locked on entry and will be locked on return.
753 *
754 * If no error returned, unp will be updated.
755 */
756int
757unionfs_mkshadowdir(struct unionfs_mount *ump, struct vnode *udvp,
758 struct unionfs_node *unp, struct componentname *cnp,
759 struct thread *td)
760{
761 int error;
762 struct vnode *lvp;
763 struct vnode *uvp;
764 struct vattr va;
765 struct vattr lva;
766 struct componentname cn;
767 struct mount *mp;
768 struct ucred *cred;
769 struct ucred *credbk;
770 struct uidinfo *rootinfo;
771
772 if (unp->un_uppervp != NULLVP)
773 return (EEXIST);
774
775 lvp = unp->un_lowervp;
776 uvp = NULLVP;
777 credbk = cnp->cn_cred;
778
779 /* Authority change to root */
780 rootinfo = uifind((uid_t)0);
781 cred = crdup(cnp->cn_cred);
782 chgproccnt(cred->cr_ruidinfo, 1, 0);
783 change_euid(cred, rootinfo);
784 change_ruid(cred, rootinfo);
785 change_svuid(cred, (uid_t)0);
786 uifree(rootinfo);
787 cnp->cn_cred = cred;
788
789 memset(&cn, 0, sizeof(cn));
790
791 if ((error = VOP_GETATTR(lvp, &lva, cnp->cn_cred)))
792 goto unionfs_mkshadowdir_abort;
793
794 if ((error = unionfs_relookup(udvp, &uvp, cnp, &cn, td, cnp->cn_nameptr, cnp->cn_namelen, CREATE)))
795 goto unionfs_mkshadowdir_abort;
796 if (uvp != NULLVP) {
797 if (udvp == uvp)
798 vrele(uvp);
799 else
800 vput(uvp);
801
802 error = EEXIST;
803 goto unionfs_mkshadowdir_free_out;
804 }
805
806 if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH)))
807 goto unionfs_mkshadowdir_free_out;
808 unionfs_create_uppervattr_core(ump, &lva, &va, td);
809
810 error = VOP_MKDIR(udvp, &uvp, &cn, &va);
811
812 if (!error) {
813 unionfs_node_update(unp, uvp, td);
814
815 /*
816 * XXX The bug which cannot set uid/gid was corrected.
817 * Ignore errors.
818 */
819 va.va_type = VNON;
820 VOP_SETATTR(uvp, &va, cn.cn_cred);
821 }
822 vn_finished_write(mp);
823
824unionfs_mkshadowdir_free_out:
825 if (cn.cn_flags & HASBUF) {
826 uma_zfree(namei_zone, cn.cn_pnbuf);
827 cn.cn_flags &= ~HASBUF;
828 }
829
830unionfs_mkshadowdir_abort:
831 cnp->cn_cred = credbk;
832 chgproccnt(cred->cr_ruidinfo, -1, 0);
833 crfree(cred);
834
835 return (error);
836}
837
838/*
839 * Create a new whiteout.
840 *
841 * dvp should be locked on entry and will be locked on return.
842 */
843int
844unionfs_mkwhiteout(struct vnode *dvp, struct componentname *cnp,
845 struct thread *td, char *path)
846{
847 int error;
848 struct vnode *wvp;
849 struct componentname cn;
850 struct mount *mp;
851
852 if (path == NULL)
853 path = cnp->cn_nameptr;
854
855 wvp = NULLVP;
856 if ((error = unionfs_relookup(dvp, &wvp, cnp, &cn, td, path, strlen(path), CREATE)))
857 return (error);
858 if (wvp != NULLVP) {
859 if (cn.cn_flags & HASBUF) {
860 uma_zfree(namei_zone, cn.cn_pnbuf);
861 cn.cn_flags &= ~HASBUF;
862 }
863 if (dvp == wvp)
864 vrele(wvp);
865 else
866 vput(wvp);
867
868 return (EEXIST);
869 }
870
871 if ((error = vn_start_write(dvp, &mp, V_WAIT | PCATCH)))
872 goto unionfs_mkwhiteout_free_out;
873 error = VOP_WHITEOUT(dvp, &cn, CREATE);
874
875 vn_finished_write(mp);
876
877unionfs_mkwhiteout_free_out:
878 if (cn.cn_flags & HASBUF) {
879 uma_zfree(namei_zone, cn.cn_pnbuf);
880 cn.cn_flags &= ~HASBUF;
881 }
882
883 return (error);
884}
885
886/*
887 * Create a new vnode for create a new shadow file.
888 *
889 * If an error is returned, *vpp will be invalid, otherwise it will hold a
890 * locked, referenced and opened vnode.
891 *
892 * unp is never updated.
893 */
894static int
895unionfs_vn_create_on_upper(struct vnode **vpp, struct vnode *udvp,
896 struct unionfs_node *unp, struct vattr *uvap,
897 struct thread *td)
898{
899 struct unionfs_mount *ump;
900 struct vnode *vp;
901 struct vnode *lvp;
902 struct ucred *cred;
903 struct vattr lva;
904 int fmode;
905 int error;
906 struct componentname cn;
907
908 ump = MOUNTTOUNIONFSMOUNT(UNIONFSTOV(unp)->v_mount);
909 vp = NULLVP;
910 lvp = unp->un_lowervp;
911 cred = td->td_ucred;
912 fmode = FFLAGS(O_WRONLY | O_CREAT | O_TRUNC | O_EXCL);
913 error = 0;
914
915 if ((error = VOP_GETATTR(lvp, &lva, cred)) != 0)
916 return (error);
917 unionfs_create_uppervattr_core(ump, &lva, uvap, td);
918
919 if (unp->un_path == NULL)
920 panic("unionfs: un_path is null");
921
922 cn.cn_namelen = strlen(unp->un_path);
923 cn.cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
924 bcopy(unp->un_path, cn.cn_pnbuf, cn.cn_namelen + 1);
925 cn.cn_nameiop = CREATE;
926 cn.cn_flags = (LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME | ISLASTCN);
927 cn.cn_lkflags = LK_EXCLUSIVE;
928 cn.cn_thread = td;
929 cn.cn_cred = cred;
930 cn.cn_nameptr = cn.cn_pnbuf;
931 cn.cn_consume = 0;
932
933 vref(udvp);
934 if ((error = relookup(udvp, &vp, &cn)) != 0)
935 goto unionfs_vn_create_on_upper_free_out2;
936 vrele(udvp);
937
938 if (vp != NULLVP) {
939 if (vp == udvp)
940 vrele(vp);
941 else
942 vput(vp);
943 error = EEXIST;
944 goto unionfs_vn_create_on_upper_free_out1;
945 }
946
947 if ((error = VOP_CREATE(udvp, &vp, &cn, uvap)) != 0)
948 goto unionfs_vn_create_on_upper_free_out1;
949
950 if ((error = VOP_OPEN(vp, fmode, cred, td, NULL)) != 0) {
951 vput(vp);
952 goto unionfs_vn_create_on_upper_free_out1;
953 }
954 vp->v_writecount++;
955 *vpp = vp;
956
957unionfs_vn_create_on_upper_free_out1:
958 VOP_UNLOCK(udvp, 0);
959
960unionfs_vn_create_on_upper_free_out2:
961 if (cn.cn_flags & HASBUF) {
962 uma_zfree(namei_zone, cn.cn_pnbuf);
963 cn.cn_flags &= ~HASBUF;
964 }
965
966 return (error);
967}
968
969/*
970 * Copy from lvp to uvp.
971 *
972 * lvp and uvp should be locked and opened on entry and will be locked and
973 * opened on return.
974 */
975static int
976unionfs_copyfile_core(struct vnode *lvp, struct vnode *uvp,
977 struct ucred *cred, struct thread *td)
978{
979 int error;
980 off_t offset;
981 int count;
982 int bufoffset;
983 char *buf;
984 struct uio uio;
985 struct iovec iov;
986
987 error = 0;
988 memset(&uio, 0, sizeof(uio));
989
990 uio.uio_td = td;
991 uio.uio_segflg = UIO_SYSSPACE;
992 uio.uio_offset = 0;
993
994 buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK);
995
996 while (error == 0) {
997 offset = uio.uio_offset;
998
999 uio.uio_iov = &iov;
1000 uio.uio_iovcnt = 1;
1001 iov.iov_base = buf;
1002 iov.iov_len = MAXBSIZE;
1003 uio.uio_resid = iov.iov_len;
1004 uio.uio_rw = UIO_READ;
1005
1006 if ((error = VOP_READ(lvp, &uio, 0, cred)) != 0)
1007 break;
1008 if ((count = MAXBSIZE - uio.uio_resid) == 0)
1009 break;
1010
1011 bufoffset = 0;
1012 while (bufoffset < count) {
1013 uio.uio_iov = &iov;
1014 uio.uio_iovcnt = 1;
1015 iov.iov_base = buf + bufoffset;
1016 iov.iov_len = count - bufoffset;
1017 uio.uio_offset = offset + bufoffset;
1018 uio.uio_resid = iov.iov_len;
1019 uio.uio_rw = UIO_WRITE;
1020
1021 if ((error = VOP_WRITE(uvp, &uio, 0, cred)) != 0)
1022 break;
1023
1024 bufoffset += (count - bufoffset) - uio.uio_resid;
1025 }
1026
1027 uio.uio_offset = offset + bufoffset;
1028 }
1029
1030 free(buf, M_TEMP);
1031
1032 return (error);
1033}
1034
1035/*
1036 * Copy file from lower to upper.
1037 *
1038 * If you need copy of the contents, set 1 to docopy. Otherwise, set 0 to
1039 * docopy.
1040 *
1041 * If no error returned, unp will be updated.
1042 */
1043int
1044unionfs_copyfile(struct unionfs_node *unp, int docopy, struct ucred *cred,
1045 struct thread *td)
1046{
1047 int error;
1048 struct mount *mp;
1049 struct vnode *udvp;
1050 struct vnode *lvp;
1051 struct vnode *uvp;
1052 struct vattr uva;
1053
1054 lvp = unp->un_lowervp;
1055 uvp = NULLVP;
1056
1057 if ((UNIONFSTOV(unp)->v_mount->mnt_flag & MNT_RDONLY))
1058 return (EROFS);
1059 if (unp->un_dvp == NULLVP)
1060 return (EINVAL);
1061 if (unp->un_uppervp != NULLVP)
1062 return (EEXIST);
1063 udvp = VTOUNIONFS(unp->un_dvp)->un_uppervp;
1064 if (udvp == NULLVP)
1065 return (EROFS);
1066 if ((udvp->v_mount->mnt_flag & MNT_RDONLY))
1067 return (EROFS);
1068
1069 error = VOP_ACCESS(lvp, VREAD, cred, td);
1070 if (error != 0)
1071 return (error);
1072
1073 if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH)) != 0)
1074 return (error);
1075 error = unionfs_vn_create_on_upper(&uvp, udvp, unp, &uva, td);
1076 if (error != 0) {
1077 vn_finished_write(mp);
1078 return (error);
1079 }
1080
1081 if (docopy != 0) {
1082 error = VOP_OPEN(lvp, FREAD, cred, td, NULL);
1083 if (error == 0) {
1084 error = unionfs_copyfile_core(lvp, uvp, cred, td);
1085 VOP_CLOSE(lvp, FREAD, cred, td);
1086 }
1087 }
1088 VOP_CLOSE(uvp, FWRITE, cred, td);
1089 uvp->v_writecount--;
1090
1091 vn_finished_write(mp);
1092
1093 if (error == 0) {
1094 /* Reset the attributes. Ignore errors. */
1095 uva.va_type = VNON;
1096 VOP_SETATTR(uvp, &uva, cred);
1097 }
1098
1099 unionfs_node_update(unp, uvp, td);
1100
1101 return (error);
1102}
1103
1104/*
1105 * It checks whether vp can rmdir. (check empty)
1106 *
1107 * vp is unionfs vnode.
1108 * vp should be locked.
1109 */
1110int
1111unionfs_check_rmdir(struct vnode *vp, struct ucred *cred, struct thread *td)
1112{
1113 int error;
1114 int eofflag;
1115 int lookuperr;
1116 struct vnode *uvp;
1117 struct vnode *lvp;
1118 struct vnode *tvp;
1119 struct vattr va;
1120 struct componentname cn;
1121 /*
1122 * The size of buf needs to be larger than DIRBLKSIZ.
1123 */
1124 char buf[256 * 6];
1125 struct dirent *dp;
1126 struct dirent *edp;
1127 struct uio uio;
1128 struct iovec iov;
1129
1130 ASSERT_VOP_ELOCKED(vp, "unionfs_check_rmdir");
1131
1132 eofflag = 0;
1133 uvp = UNIONFSVPTOUPPERVP(vp);
1134 lvp = UNIONFSVPTOLOWERVP(vp);
1135
1136 /* check opaque */
1137 if ((error = VOP_GETATTR(uvp, &va, cred)) != 0)
1138 return (error);
1139 if (va.va_flags & OPAQUE)
1140 return (0);
1141
1142 /* open vnode */
1143#ifdef MAC
1144 if ((error = mac_vnode_check_open(cred, vp, VEXEC|VREAD)) != 0)
1145 return (error);
1146#endif
1147 if ((error = VOP_ACCESS(vp, VEXEC|VREAD, cred, td)) != 0)
1148 return (error);
1149 if ((error = VOP_OPEN(vp, FREAD, cred, td, NULL)) != 0)
1150 return (error);
1151
1152 uio.uio_rw = UIO_READ;
1153 uio.uio_segflg = UIO_SYSSPACE;
1154 uio.uio_td = td;
1155 uio.uio_offset = 0;
1156
1157#ifdef MAC
1158 error = mac_vnode_check_readdir(td->td_ucred, lvp);
1159#endif
1160 while (!error && !eofflag) {
1161 iov.iov_base = buf;
1162 iov.iov_len = sizeof(buf);
1163 uio.uio_iov = &iov;
1164 uio.uio_iovcnt = 1;
1165 uio.uio_resid = iov.iov_len;
1166
1167 error = VOP_READDIR(lvp, &uio, cred, &eofflag, NULL, NULL);
1168 if (error != 0)
1169 break;
1170 if (eofflag == 0 && uio.uio_resid == sizeof(buf)) {
1171#ifdef DIAGNOSTIC
1172 panic("bad readdir response from lower FS.");
1173#endif
1174 break;
1175 }
1176
1177 edp = (struct dirent*)&buf[sizeof(buf) - uio.uio_resid];
1178 for (dp = (struct dirent*)buf; !error && dp < edp;
1179 dp = (struct dirent*)((caddr_t)dp + dp->d_reclen)) {
1180 if (dp->d_type == DT_WHT ||
1181 (dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1182 (dp->d_namlen == 2 && !bcmp(dp->d_name, "..", 2)))
1183 continue;
1184
1185 cn.cn_namelen = dp->d_namlen;
1186 cn.cn_pnbuf = NULL;
1187 cn.cn_nameptr = dp->d_name;
1188 cn.cn_nameiop = LOOKUP;
1189 cn.cn_flags = (LOCKPARENT | LOCKLEAF | SAVENAME | RDONLY | ISLASTCN);
1190 cn.cn_lkflags = LK_EXCLUSIVE;
1191 cn.cn_thread = td;
1192 cn.cn_cred = cred;
1193 cn.cn_consume = 0;
1194
1195 /*
1196 * check entry in lower.
1197 * Sometimes, readdir function returns
1198 * wrong entry.
1199 */
1200 lookuperr = VOP_LOOKUP(lvp, &tvp, &cn);
1201
1202 if (!lookuperr)
1203 vput(tvp);
1204 else
1205 continue; /* skip entry */
1206
1207 /*
1208 * check entry
1209 * If it has no exist/whiteout entry in upper,
1210 * directory is not empty.
1211 */
1212 cn.cn_flags = (LOCKPARENT | LOCKLEAF | SAVENAME | RDONLY | ISLASTCN);
1213 lookuperr = VOP_LOOKUP(uvp, &tvp, &cn);
1214
1215 if (!lookuperr)
1216 vput(tvp);
1217
1218 /* ignore exist or whiteout entry */
1219 if (!lookuperr ||
1220 (lookuperr == ENOENT && (cn.cn_flags & ISWHITEOUT)))
1221 continue;
1222
1223 error = ENOTEMPTY;
1224 }
1225 }
1226
1227 /* close vnode */
1228 VOP_CLOSE(vp, FREAD, cred, td);
1229
1230 return (error);
1231}
1232
1233#ifdef DIAGNOSTIC
1234
1235struct vnode *
1236unionfs_checkuppervp(struct vnode *vp, char *fil, int lno)
1237{
1238 struct unionfs_node *unp;
1239
1240 unp = VTOUNIONFS(vp);
1241
1242#ifdef notyet
1243 if (vp->v_op != unionfs_vnodeop_p) {
1244 printf("unionfs_checkuppervp: on non-unionfs-node.\n");
1245#ifdef KDB
1246 kdb_enter(KDB_WHY_UNIONFS,
1247 "unionfs_checkuppervp: on non-unionfs-node.\n");
1248#endif
1249 panic("unionfs_checkuppervp");
1250 };
1251#endif
1252 return (unp->un_uppervp);
1253}
1254
1255struct vnode *
1256unionfs_checklowervp(struct vnode *vp, char *fil, int lno)
1257{
1258 struct unionfs_node *unp;
1259
1260 unp = VTOUNIONFS(vp);
1261
1262#ifdef notyet
1263 if (vp->v_op != unionfs_vnodeop_p) {
1264 printf("unionfs_checklowervp: on non-unionfs-node.\n");
1265#ifdef KDB
1266 kdb_enter(KDB_WHY_UNIONFS,
1267 "unionfs_checklowervp: on non-unionfs-node.\n");
1268#endif
1269 panic("unionfs_checklowervp");
1270 };
1271#endif
1272 return (unp->un_lowervp);
1273}
1274#endif
2 * Copyright (c) 1994 Jan-Simon Pendry
3 * Copyright (c) 1994
4 * The Regents of the University of California. All rights reserved.
5 * Copyright (c) 2005, 2006 Masanori Ozawa <ozawa@ongs.co.jp>, ONGS Inc.
6 * Copyright (c) 2006 Daichi Goto <daichi@freebsd.org>
7 *
8 * This code is derived from software contributed to Berkeley by
9 * Jan-Simon Pendry.
10 *
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
13 * are met:
14 * 1. Redistributions of source code must retain the above copyright
15 * notice, this list of conditions and the following disclaimer.
16 * 2. Redistributions in binary form must reproduce the above copyright
17 * notice, this list of conditions and the following disclaimer in the
18 * documentation and/or other materials provided with the distribution.
19 * 4. Neither the name of the University nor the names of its contributors
20 * may be used to endorse or promote products derived from this software
21 * without specific prior written permission.
22 *
23 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
24 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
25 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
26 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
27 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
28 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
29 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
30 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
31 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
32 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 * SUCH DAMAGE.
34 *
35 * @(#)union_subr.c 8.20 (Berkeley) 5/20/95
36 * $FreeBSD: head/sys/fs/unionfs/union_subr.c 190888 2009-04-10 10:52:19Z rwatson $
37 */
38
39#include <sys/param.h>
40#include <sys/systm.h>
41#include <sys/kernel.h>
42#include <sys/lock.h>
43#include <sys/mutex.h>
44#include <sys/malloc.h>
45#include <sys/mount.h>
46#include <sys/namei.h>
47#include <sys/proc.h>
48#include <sys/vnode.h>
49#include <sys/dirent.h>
50#include <sys/fcntl.h>
51#include <sys/filedesc.h>
52#include <sys/stat.h>
53#include <sys/resourcevar.h>
54
55#ifdef MAC
56#include <sys/mac.h>
57#endif
58
59#include <vm/uma.h>
60
61#include <fs/unionfs/union.h>
62
63#define NUNIONFSNODECACHE 16
64
65static MALLOC_DEFINE(M_UNIONFSHASH, "UNIONFS hash", "UNIONFS hash table");
66MALLOC_DEFINE(M_UNIONFSNODE, "UNIONFS node", "UNIONFS vnode private part");
67MALLOC_DEFINE(M_UNIONFSPATH, "UNIONFS path", "UNIONFS path private part");
68
69/*
70 * Initialize
71 */
72int
73unionfs_init(struct vfsconf *vfsp)
74{
75 UNIONFSDEBUG("unionfs_init\n"); /* printed during system boot */
76 return (0);
77}
78
79/*
80 * Uninitialize
81 */
82int
83unionfs_uninit(struct vfsconf *vfsp)
84{
85 return (0);
86}
87
88static struct unionfs_node_hashhead *
89unionfs_get_hashhead(struct vnode *dvp, char *path)
90{
91 int count;
92 char hash;
93 struct unionfs_node *unp;
94
95 hash = 0;
96 unp = VTOUNIONFS(dvp);
97 if (path != NULL) {
98 for (count = 0; path[count]; count++)
99 hash += path[count];
100 }
101
102 return (&(unp->un_hashtbl[hash & (unp->un_hashmask)]));
103}
104
105/*
106 * Get the cached vnode.
107 */
108static struct vnode *
109unionfs_get_cached_vnode(struct vnode *uvp, struct vnode *lvp,
110 struct vnode *dvp, char *path)
111{
112 struct unionfs_node_hashhead *hd;
113 struct unionfs_node *unp;
114 struct vnode *vp;
115
116 KASSERT((uvp == NULLVP || uvp->v_type == VDIR),
117 ("unionfs_get_cached_vnode: v_type != VDIR"));
118 KASSERT((lvp == NULLVP || lvp->v_type == VDIR),
119 ("unionfs_get_cached_vnode: v_type != VDIR"));
120
121 VI_LOCK(dvp);
122 hd = unionfs_get_hashhead(dvp, path);
123 LIST_FOREACH(unp, hd, un_hash) {
124 if (!strcmp(unp->un_path, path)) {
125 vp = UNIONFSTOV(unp);
126 VI_LOCK_FLAGS(vp, MTX_DUPOK);
127 VI_UNLOCK(dvp);
128 vp->v_iflag &= ~VI_OWEINACT;
129 if ((vp->v_iflag & (VI_DOOMED | VI_DOINGINACT)) != 0) {
130 VI_UNLOCK(vp);
131 vp = NULLVP;
132 } else
133 VI_UNLOCK(vp);
134 return (vp);
135 }
136 }
137 VI_UNLOCK(dvp);
138
139 return (NULLVP);
140}
141
142/*
143 * Add the new vnode into cache.
144 */
145static struct vnode *
146unionfs_ins_cached_vnode(struct unionfs_node *uncp,
147 struct vnode *dvp, char *path)
148{
149 struct unionfs_node_hashhead *hd;
150 struct unionfs_node *unp;
151 struct vnode *vp;
152
153 KASSERT((uncp->un_uppervp==NULLVP || uncp->un_uppervp->v_type==VDIR),
154 ("unionfs_ins_cached_vnode: v_type != VDIR"));
155 KASSERT((uncp->un_lowervp==NULLVP || uncp->un_lowervp->v_type==VDIR),
156 ("unionfs_ins_cached_vnode: v_type != VDIR"));
157
158 VI_LOCK(dvp);
159 hd = unionfs_get_hashhead(dvp, path);
160 LIST_FOREACH(unp, hd, un_hash) {
161 if (!strcmp(unp->un_path, path)) {
162 vp = UNIONFSTOV(unp);
163 VI_LOCK_FLAGS(vp, MTX_DUPOK);
164 vp->v_iflag &= ~VI_OWEINACT;
165 if ((vp->v_iflag & (VI_DOOMED | VI_DOINGINACT)) != 0) {
166 LIST_INSERT_HEAD(hd, uncp, un_hash);
167 VI_UNLOCK(vp);
168 vp = NULLVP;
169 } else
170 VI_UNLOCK(vp);
171 VI_UNLOCK(dvp);
172 return (vp);
173 }
174 }
175
176 LIST_INSERT_HEAD(hd, uncp, un_hash);
177 VI_UNLOCK(dvp);
178
179 return (NULLVP);
180}
181
182/*
183 * Remove the vnode.
184 */
185static void
186unionfs_rem_cached_vnode(struct unionfs_node *unp, struct vnode *dvp)
187{
188 KASSERT((unp != NULL), ("unionfs_rem_cached_vnode: null node"));
189 KASSERT((dvp != NULLVP),
190 ("unionfs_rem_cached_vnode: null parent vnode"));
191 KASSERT((unp->un_hash.le_prev != NULL),
192 ("unionfs_rem_cached_vnode: null hash"));
193
194 VI_LOCK(dvp);
195 LIST_REMOVE(unp, un_hash);
196 unp->un_hash.le_next = NULL;
197 unp->un_hash.le_prev = NULL;
198 VI_UNLOCK(dvp);
199}
200
201/*
202 * Make a new or get existing unionfs node.
203 *
204 * uppervp and lowervp should be unlocked. Because if new unionfs vnode is
205 * locked, uppervp or lowervp is locked too. In order to prevent dead lock,
206 * you should not lock plurality simultaneously.
207 */
208int
209unionfs_nodeget(struct mount *mp, struct vnode *uppervp,
210 struct vnode *lowervp, struct vnode *dvp,
211 struct vnode **vpp, struct componentname *cnp,
212 struct thread *td)
213{
214 struct unionfs_mount *ump;
215 struct unionfs_node *unp;
216 struct vnode *vp;
217 int error;
218 int lkflags;
219 enum vtype vt;
220 char *path;
221
222 ump = MOUNTTOUNIONFSMOUNT(mp);
223 lkflags = (cnp ? cnp->cn_lkflags : 0);
224 path = (cnp ? cnp->cn_nameptr : NULL);
225 *vpp = NULLVP;
226
227 if (uppervp == NULLVP && lowervp == NULLVP)
228 panic("unionfs_nodeget: upper and lower is null");
229
230 vt = (uppervp != NULLVP ? uppervp->v_type : lowervp->v_type);
231
232 /* If it has no ISLASTCN flag, path check is skipped. */
233 if (cnp && !(cnp->cn_flags & ISLASTCN))
234 path = NULL;
235
236 /* check the cache */
237 if (path != NULL && dvp != NULLVP && vt == VDIR) {
238 vp = unionfs_get_cached_vnode(uppervp, lowervp, dvp, path);
239 if (vp != NULLVP) {
240 vref(vp);
241 *vpp = vp;
242 goto unionfs_nodeget_out;
243 }
244 }
245
246 if ((uppervp == NULLVP || ump->um_uppervp != uppervp) ||
247 (lowervp == NULLVP || ump->um_lowervp != lowervp)) {
248 /* dvp will be NULLVP only in case of root vnode. */
249 if (dvp == NULLVP)
250 return (EINVAL);
251 }
252
253 /*
254 * Do the MALLOC before the getnewvnode since doing so afterward
255 * might cause a bogus v_data pointer to get dereferenced elsewhere
256 * if MALLOC should block.
257 */
258 unp = malloc(sizeof(struct unionfs_node),
259 M_UNIONFSNODE, M_WAITOK | M_ZERO);
260
261 error = getnewvnode("unionfs", mp, &unionfs_vnodeops, &vp);
262 if (error != 0) {
263 free(unp, M_UNIONFSNODE);
264 return (error);
265 }
266 error = insmntque(vp, mp); /* XXX: Too early for mpsafe fs */
267 if (error != 0) {
268 free(unp, M_UNIONFSNODE);
269 return (error);
270 }
271 if (dvp != NULLVP)
272 vref(dvp);
273 if (uppervp != NULLVP)
274 vref(uppervp);
275 if (lowervp != NULLVP)
276 vref(lowervp);
277
278 if (vt == VDIR)
279 unp->un_hashtbl = hashinit(NUNIONFSNODECACHE, M_UNIONFSHASH,
280 &(unp->un_hashmask));
281
282 unp->un_vnode = vp;
283 unp->un_uppervp = uppervp;
284 unp->un_lowervp = lowervp;
285 unp->un_dvp = dvp;
286 if (uppervp != NULLVP)
287 vp->v_vnlock = uppervp->v_vnlock;
288 else
289 vp->v_vnlock = lowervp->v_vnlock;
290
291 if (path != NULL) {
292 unp->un_path = (char *)
293 malloc(cnp->cn_namelen +1, M_UNIONFSPATH, M_WAITOK|M_ZERO);
294 bcopy(cnp->cn_nameptr, unp->un_path, cnp->cn_namelen);
295 unp->un_path[cnp->cn_namelen] = '\0';
296 }
297 vp->v_type = vt;
298 vp->v_data = unp;
299
300 if ((uppervp != NULLVP && ump->um_uppervp == uppervp) &&
301 (lowervp != NULLVP && ump->um_lowervp == lowervp))
302 vp->v_vflag |= VV_ROOT;
303
304 if (path != NULL && dvp != NULLVP && vt == VDIR)
305 *vpp = unionfs_ins_cached_vnode(unp, dvp, path);
306 if ((*vpp) != NULLVP) {
307 if (dvp != NULLVP)
308 vrele(dvp);
309 if (uppervp != NULLVP)
310 vrele(uppervp);
311 if (lowervp != NULLVP)
312 vrele(lowervp);
313
314 unp->un_uppervp = NULLVP;
315 unp->un_lowervp = NULLVP;
316 unp->un_dvp = NULLVP;
317 vrele(vp);
318 vp = *vpp;
319 vref(vp);
320 } else
321 *vpp = vp;
322
323unionfs_nodeget_out:
324 if (lkflags & LK_TYPE_MASK)
325 vn_lock(vp, lkflags | LK_RETRY);
326
327 return (0);
328}
329
330/*
331 * Clean up the unionfs node.
332 */
333void
334unionfs_noderem(struct vnode *vp, struct thread *td)
335{
336 int vfslocked;
337 int count;
338 struct unionfs_node *unp, *unp_t1, *unp_t2;
339 struct unionfs_node_hashhead *hd;
340 struct unionfs_node_status *unsp, *unsp_tmp;
341 struct vnode *lvp;
342 struct vnode *uvp;
343 struct vnode *dvp;
344
345 /*
346 * Use the interlock to protect the clearing of v_data to
347 * prevent faults in unionfs_lock().
348 */
349 VI_LOCK(vp);
350 unp = VTOUNIONFS(vp);
351 lvp = unp->un_lowervp;
352 uvp = unp->un_uppervp;
353 dvp = unp->un_dvp;
354 unp->un_lowervp = unp->un_uppervp = NULLVP;
355
356 vp->v_vnlock = &(vp->v_lock);
357 vp->v_data = NULL;
358 lockmgr(vp->v_vnlock, LK_EXCLUSIVE | LK_INTERLOCK, VI_MTX(vp));
359 if (lvp != NULLVP)
360 VOP_UNLOCK(lvp, 0);
361 if (uvp != NULLVP)
362 VOP_UNLOCK(uvp, 0);
363 vp->v_object = NULL;
364
365 if (dvp != NULLVP && unp->un_hash.le_prev != NULL)
366 unionfs_rem_cached_vnode(unp, dvp);
367
368 if (lvp != NULLVP) {
369 vfslocked = VFS_LOCK_GIANT(lvp->v_mount);
370 vrele(lvp);
371 VFS_UNLOCK_GIANT(vfslocked);
372 }
373 if (uvp != NULLVP) {
374 vfslocked = VFS_LOCK_GIANT(uvp->v_mount);
375 vrele(uvp);
376 VFS_UNLOCK_GIANT(vfslocked);
377 }
378 if (dvp != NULLVP) {
379 vfslocked = VFS_LOCK_GIANT(dvp->v_mount);
380 vrele(dvp);
381 VFS_UNLOCK_GIANT(vfslocked);
382 unp->un_dvp = NULLVP;
383 }
384 if (unp->un_path != NULL) {
385 free(unp->un_path, M_UNIONFSPATH);
386 unp->un_path = NULL;
387 }
388
389 if (unp->un_hashtbl != NULL) {
390 for (count = 0; count <= unp->un_hashmask; count++) {
391 hd = unp->un_hashtbl + count;
392 LIST_FOREACH_SAFE(unp_t1, hd, un_hash, unp_t2) {
393 LIST_REMOVE(unp_t1, un_hash);
394 unp_t1->un_hash.le_next = NULL;
395 unp_t1->un_hash.le_prev = NULL;
396 }
397 }
398 hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, unp->un_hashmask);
399 }
400
401 LIST_FOREACH_SAFE(unsp, &(unp->un_unshead), uns_list, unsp_tmp) {
402 LIST_REMOVE(unsp, uns_list);
403 free(unsp, M_TEMP);
404 }
405 free(unp, M_UNIONFSNODE);
406}
407
408/*
409 * Get the unionfs node status.
410 * You need exclusive lock this vnode.
411 */
412void
413unionfs_get_node_status(struct unionfs_node *unp, struct thread *td,
414 struct unionfs_node_status **unspp)
415{
416 struct unionfs_node_status *unsp;
417 pid_t pid = td->td_proc->p_pid;
418
419 KASSERT(NULL != unspp, ("null pointer"));
420 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), "unionfs_get_node_status");
421
422 LIST_FOREACH(unsp, &(unp->un_unshead), uns_list) {
423 if (unsp->uns_pid == pid) {
424 *unspp = unsp;
425 return;
426 }
427 }
428
429 /* create a new unionfs node status */
430 unsp = malloc(sizeof(struct unionfs_node_status),
431 M_TEMP, M_WAITOK | M_ZERO);
432
433 unsp->uns_pid = pid;
434 LIST_INSERT_HEAD(&(unp->un_unshead), unsp, uns_list);
435
436 *unspp = unsp;
437}
438
439/*
440 * Remove the unionfs node status, if you can.
441 * You need exclusive lock this vnode.
442 */
443void
444unionfs_tryrem_node_status(struct unionfs_node *unp,
445 struct unionfs_node_status *unsp)
446{
447 KASSERT(NULL != unsp, ("null pointer"));
448 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), "unionfs_get_node_status");
449
450 if (0 < unsp->uns_lower_opencnt || 0 < unsp->uns_upper_opencnt)
451 return;
452
453 LIST_REMOVE(unsp, uns_list);
454 free(unsp, M_TEMP);
455}
456
457/*
458 * Create upper node attr.
459 */
460void
461unionfs_create_uppervattr_core(struct unionfs_mount *ump,
462 struct vattr *lva,
463 struct vattr *uva,
464 struct thread *td)
465{
466 VATTR_NULL(uva);
467 uva->va_type = lva->va_type;
468 uva->va_atime = lva->va_atime;
469 uva->va_mtime = lva->va_mtime;
470 uva->va_ctime = lva->va_ctime;
471
472 switch (ump->um_copymode) {
473 case UNIONFS_TRANSPARENT:
474 uva->va_mode = lva->va_mode;
475 uva->va_uid = lva->va_uid;
476 uva->va_gid = lva->va_gid;
477 break;
478 case UNIONFS_MASQUERADE:
479 if (ump->um_uid == lva->va_uid) {
480 uva->va_mode = lva->va_mode & 077077;
481 uva->va_mode |= (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile) & 0700;
482 uva->va_uid = lva->va_uid;
483 uva->va_gid = lva->va_gid;
484 } else {
485 uva->va_mode = (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile);
486 uva->va_uid = ump->um_uid;
487 uva->va_gid = ump->um_gid;
488 }
489 break;
490 default: /* UNIONFS_TRADITIONAL */
491 FILEDESC_SLOCK(td->td_proc->p_fd);
492 uva->va_mode = 0777 & ~td->td_proc->p_fd->fd_cmask;
493 FILEDESC_SUNLOCK(td->td_proc->p_fd);
494 uva->va_uid = ump->um_uid;
495 uva->va_gid = ump->um_gid;
496 break;
497 }
498}
499
500/*
501 * Create upper node attr.
502 */
503int
504unionfs_create_uppervattr(struct unionfs_mount *ump,
505 struct vnode *lvp,
506 struct vattr *uva,
507 struct ucred *cred,
508 struct thread *td)
509{
510 int error;
511 struct vattr lva;
512
513 if ((error = VOP_GETATTR(lvp, &lva, cred)))
514 return (error);
515
516 unionfs_create_uppervattr_core(ump, &lva, uva, td);
517
518 return (error);
519}
520
521/*
522 * relookup
523 *
524 * dvp should be locked on entry and will be locked on return.
525 *
526 * If an error is returned, *vpp will be invalid, otherwise it will hold a
527 * locked, referenced vnode. If *vpp == dvp then remember that only one
528 * LK_EXCLUSIVE lock is held.
529 */
530int
531unionfs_relookup(struct vnode *dvp, struct vnode **vpp,
532 struct componentname *cnp, struct componentname *cn,
533 struct thread *td, char *path, int pathlen, u_long nameiop)
534{
535 int error;
536
537 cn->cn_namelen = pathlen;
538 cn->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
539 bcopy(path, cn->cn_pnbuf, pathlen);
540 cn->cn_pnbuf[pathlen] = '\0';
541
542 cn->cn_nameiop = nameiop;
543 cn->cn_flags = (LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME | ISLASTCN);
544 cn->cn_lkflags = LK_EXCLUSIVE;
545 cn->cn_thread = td;
546 cn->cn_cred = cnp->cn_cred;
547
548 cn->cn_nameptr = cn->cn_pnbuf;
549 cn->cn_consume = cnp->cn_consume;
550
551 if (nameiop == DELETE)
552 cn->cn_flags |= (cnp->cn_flags & (DOWHITEOUT | SAVESTART));
553 else if (RENAME == nameiop)
554 cn->cn_flags |= (cnp->cn_flags & SAVESTART);
555
556 vref(dvp);
557 VOP_UNLOCK(dvp, 0);
558
559 if ((error = relookup(dvp, vpp, cn))) {
560 uma_zfree(namei_zone, cn->cn_pnbuf);
561 cn->cn_flags &= ~HASBUF;
562 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
563 } else
564 vrele(dvp);
565
566 return (error);
567}
568
569/*
570 * relookup for CREATE namei operation.
571 *
572 * dvp is unionfs vnode. dvp should be locked.
573 *
574 * If it called 'unionfs_copyfile' function by unionfs_link etc,
575 * VOP_LOOKUP information is broken.
576 * So it need relookup in order to create link etc.
577 */
578int
579unionfs_relookup_for_create(struct vnode *dvp, struct componentname *cnp,
580 struct thread *td)
581{
582 int error;
583 struct vnode *udvp;
584 struct vnode *vp;
585 struct componentname cn;
586
587 udvp = UNIONFSVPTOUPPERVP(dvp);
588 vp = NULLVP;
589
590 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
591 strlen(cnp->cn_nameptr), CREATE);
592 if (error)
593 return (error);
594
595 if (vp != NULLVP) {
596 if (udvp == vp)
597 vrele(vp);
598 else
599 vput(vp);
600
601 error = EEXIST;
602 }
603
604 if (cn.cn_flags & HASBUF) {
605 uma_zfree(namei_zone, cn.cn_pnbuf);
606 cn.cn_flags &= ~HASBUF;
607 }
608
609 if (!error) {
610 cn.cn_flags |= (cnp->cn_flags & HASBUF);
611 cnp->cn_flags = cn.cn_flags;
612 }
613
614 return (error);
615}
616
617/*
618 * relookup for DELETE namei operation.
619 *
620 * dvp is unionfs vnode. dvp should be locked.
621 */
622int
623unionfs_relookup_for_delete(struct vnode *dvp, struct componentname *cnp,
624 struct thread *td)
625{
626 int error;
627 struct vnode *udvp;
628 struct vnode *vp;
629 struct componentname cn;
630
631 udvp = UNIONFSVPTOUPPERVP(dvp);
632 vp = NULLVP;
633
634 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
635 strlen(cnp->cn_nameptr), DELETE);
636 if (error)
637 return (error);
638
639 if (vp == NULLVP)
640 error = ENOENT;
641 else {
642 if (udvp == vp)
643 vrele(vp);
644 else
645 vput(vp);
646 }
647
648 if (cn.cn_flags & HASBUF) {
649 uma_zfree(namei_zone, cn.cn_pnbuf);
650 cn.cn_flags &= ~HASBUF;
651 }
652
653 if (!error) {
654 cn.cn_flags |= (cnp->cn_flags & HASBUF);
655 cnp->cn_flags = cn.cn_flags;
656 }
657
658 return (error);
659}
660
661/*
662 * relookup for RENAME namei operation.
663 *
664 * dvp is unionfs vnode. dvp should be locked.
665 */
666int
667unionfs_relookup_for_rename(struct vnode *dvp, struct componentname *cnp,
668 struct thread *td)
669{
670 int error;
671 struct vnode *udvp;
672 struct vnode *vp;
673 struct componentname cn;
674
675 udvp = UNIONFSVPTOUPPERVP(dvp);
676 vp = NULLVP;
677
678 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
679 strlen(cnp->cn_nameptr), RENAME);
680 if (error)
681 return (error);
682
683 if (vp != NULLVP) {
684 if (udvp == vp)
685 vrele(vp);
686 else
687 vput(vp);
688 }
689
690 if (cn.cn_flags & HASBUF) {
691 uma_zfree(namei_zone, cn.cn_pnbuf);
692 cn.cn_flags &= ~HASBUF;
693 }
694
695 if (!error) {
696 cn.cn_flags |= (cnp->cn_flags & HASBUF);
697 cnp->cn_flags = cn.cn_flags;
698 }
699
700 return (error);
701
702}
703
704/*
705 * Update the unionfs_node.
706 *
707 * uvp is new locked upper vnode. unionfs vnode's lock will be exchanged to the
708 * uvp's lock and lower's lock will be unlocked.
709 */
710static void
711unionfs_node_update(struct unionfs_node *unp, struct vnode *uvp,
712 struct thread *td)
713{
714 unsigned count, lockrec;
715 struct vnode *vp;
716 struct vnode *lvp;
717 struct vnode *dvp;
718
719 vp = UNIONFSTOV(unp);
720 lvp = unp->un_lowervp;
721 ASSERT_VOP_ELOCKED(lvp, "unionfs_node_update");
722 dvp = unp->un_dvp;
723
724 /*
725 * lock update
726 */
727 VI_LOCK(vp);
728 unp->un_uppervp = uvp;
729 vp->v_vnlock = uvp->v_vnlock;
730 VI_UNLOCK(vp);
731 lockrec = lvp->v_vnlock->lk_recurse;
732 for (count = 0; count < lockrec; count++)
733 vn_lock(uvp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY);
734
735 /*
736 * cache update
737 */
738 if (unp->un_path != NULL && dvp != NULLVP && vp->v_type == VDIR) {
739 static struct unionfs_node_hashhead *hd;
740
741 VI_LOCK(dvp);
742 hd = unionfs_get_hashhead(dvp, unp->un_path);
743 LIST_REMOVE(unp, un_hash);
744 LIST_INSERT_HEAD(hd, unp, un_hash);
745 VI_UNLOCK(dvp);
746 }
747}
748
749/*
750 * Create a new shadow dir.
751 *
752 * udvp should be locked on entry and will be locked on return.
753 *
754 * If no error returned, unp will be updated.
755 */
756int
757unionfs_mkshadowdir(struct unionfs_mount *ump, struct vnode *udvp,
758 struct unionfs_node *unp, struct componentname *cnp,
759 struct thread *td)
760{
761 int error;
762 struct vnode *lvp;
763 struct vnode *uvp;
764 struct vattr va;
765 struct vattr lva;
766 struct componentname cn;
767 struct mount *mp;
768 struct ucred *cred;
769 struct ucred *credbk;
770 struct uidinfo *rootinfo;
771
772 if (unp->un_uppervp != NULLVP)
773 return (EEXIST);
774
775 lvp = unp->un_lowervp;
776 uvp = NULLVP;
777 credbk = cnp->cn_cred;
778
779 /* Authority change to root */
780 rootinfo = uifind((uid_t)0);
781 cred = crdup(cnp->cn_cred);
782 chgproccnt(cred->cr_ruidinfo, 1, 0);
783 change_euid(cred, rootinfo);
784 change_ruid(cred, rootinfo);
785 change_svuid(cred, (uid_t)0);
786 uifree(rootinfo);
787 cnp->cn_cred = cred;
788
789 memset(&cn, 0, sizeof(cn));
790
791 if ((error = VOP_GETATTR(lvp, &lva, cnp->cn_cred)))
792 goto unionfs_mkshadowdir_abort;
793
794 if ((error = unionfs_relookup(udvp, &uvp, cnp, &cn, td, cnp->cn_nameptr, cnp->cn_namelen, CREATE)))
795 goto unionfs_mkshadowdir_abort;
796 if (uvp != NULLVP) {
797 if (udvp == uvp)
798 vrele(uvp);
799 else
800 vput(uvp);
801
802 error = EEXIST;
803 goto unionfs_mkshadowdir_free_out;
804 }
805
806 if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH)))
807 goto unionfs_mkshadowdir_free_out;
808 unionfs_create_uppervattr_core(ump, &lva, &va, td);
809
810 error = VOP_MKDIR(udvp, &uvp, &cn, &va);
811
812 if (!error) {
813 unionfs_node_update(unp, uvp, td);
814
815 /*
816 * XXX The bug which cannot set uid/gid was corrected.
817 * Ignore errors.
818 */
819 va.va_type = VNON;
820 VOP_SETATTR(uvp, &va, cn.cn_cred);
821 }
822 vn_finished_write(mp);
823
824unionfs_mkshadowdir_free_out:
825 if (cn.cn_flags & HASBUF) {
826 uma_zfree(namei_zone, cn.cn_pnbuf);
827 cn.cn_flags &= ~HASBUF;
828 }
829
830unionfs_mkshadowdir_abort:
831 cnp->cn_cred = credbk;
832 chgproccnt(cred->cr_ruidinfo, -1, 0);
833 crfree(cred);
834
835 return (error);
836}
837
838/*
839 * Create a new whiteout.
840 *
841 * dvp should be locked on entry and will be locked on return.
842 */
843int
844unionfs_mkwhiteout(struct vnode *dvp, struct componentname *cnp,
845 struct thread *td, char *path)
846{
847 int error;
848 struct vnode *wvp;
849 struct componentname cn;
850 struct mount *mp;
851
852 if (path == NULL)
853 path = cnp->cn_nameptr;
854
855 wvp = NULLVP;
856 if ((error = unionfs_relookup(dvp, &wvp, cnp, &cn, td, path, strlen(path), CREATE)))
857 return (error);
858 if (wvp != NULLVP) {
859 if (cn.cn_flags & HASBUF) {
860 uma_zfree(namei_zone, cn.cn_pnbuf);
861 cn.cn_flags &= ~HASBUF;
862 }
863 if (dvp == wvp)
864 vrele(wvp);
865 else
866 vput(wvp);
867
868 return (EEXIST);
869 }
870
871 if ((error = vn_start_write(dvp, &mp, V_WAIT | PCATCH)))
872 goto unionfs_mkwhiteout_free_out;
873 error = VOP_WHITEOUT(dvp, &cn, CREATE);
874
875 vn_finished_write(mp);
876
877unionfs_mkwhiteout_free_out:
878 if (cn.cn_flags & HASBUF) {
879 uma_zfree(namei_zone, cn.cn_pnbuf);
880 cn.cn_flags &= ~HASBUF;
881 }
882
883 return (error);
884}
885
886/*
887 * Create a new vnode for create a new shadow file.
888 *
889 * If an error is returned, *vpp will be invalid, otherwise it will hold a
890 * locked, referenced and opened vnode.
891 *
892 * unp is never updated.
893 */
894static int
895unionfs_vn_create_on_upper(struct vnode **vpp, struct vnode *udvp,
896 struct unionfs_node *unp, struct vattr *uvap,
897 struct thread *td)
898{
899 struct unionfs_mount *ump;
900 struct vnode *vp;
901 struct vnode *lvp;
902 struct ucred *cred;
903 struct vattr lva;
904 int fmode;
905 int error;
906 struct componentname cn;
907
908 ump = MOUNTTOUNIONFSMOUNT(UNIONFSTOV(unp)->v_mount);
909 vp = NULLVP;
910 lvp = unp->un_lowervp;
911 cred = td->td_ucred;
912 fmode = FFLAGS(O_WRONLY | O_CREAT | O_TRUNC | O_EXCL);
913 error = 0;
914
915 if ((error = VOP_GETATTR(lvp, &lva, cred)) != 0)
916 return (error);
917 unionfs_create_uppervattr_core(ump, &lva, uvap, td);
918
919 if (unp->un_path == NULL)
920 panic("unionfs: un_path is null");
921
922 cn.cn_namelen = strlen(unp->un_path);
923 cn.cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
924 bcopy(unp->un_path, cn.cn_pnbuf, cn.cn_namelen + 1);
925 cn.cn_nameiop = CREATE;
926 cn.cn_flags = (LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME | ISLASTCN);
927 cn.cn_lkflags = LK_EXCLUSIVE;
928 cn.cn_thread = td;
929 cn.cn_cred = cred;
930 cn.cn_nameptr = cn.cn_pnbuf;
931 cn.cn_consume = 0;
932
933 vref(udvp);
934 if ((error = relookup(udvp, &vp, &cn)) != 0)
935 goto unionfs_vn_create_on_upper_free_out2;
936 vrele(udvp);
937
938 if (vp != NULLVP) {
939 if (vp == udvp)
940 vrele(vp);
941 else
942 vput(vp);
943 error = EEXIST;
944 goto unionfs_vn_create_on_upper_free_out1;
945 }
946
947 if ((error = VOP_CREATE(udvp, &vp, &cn, uvap)) != 0)
948 goto unionfs_vn_create_on_upper_free_out1;
949
950 if ((error = VOP_OPEN(vp, fmode, cred, td, NULL)) != 0) {
951 vput(vp);
952 goto unionfs_vn_create_on_upper_free_out1;
953 }
954 vp->v_writecount++;
955 *vpp = vp;
956
957unionfs_vn_create_on_upper_free_out1:
958 VOP_UNLOCK(udvp, 0);
959
960unionfs_vn_create_on_upper_free_out2:
961 if (cn.cn_flags & HASBUF) {
962 uma_zfree(namei_zone, cn.cn_pnbuf);
963 cn.cn_flags &= ~HASBUF;
964 }
965
966 return (error);
967}
968
969/*
970 * Copy from lvp to uvp.
971 *
972 * lvp and uvp should be locked and opened on entry and will be locked and
973 * opened on return.
974 */
975static int
976unionfs_copyfile_core(struct vnode *lvp, struct vnode *uvp,
977 struct ucred *cred, struct thread *td)
978{
979 int error;
980 off_t offset;
981 int count;
982 int bufoffset;
983 char *buf;
984 struct uio uio;
985 struct iovec iov;
986
987 error = 0;
988 memset(&uio, 0, sizeof(uio));
989
990 uio.uio_td = td;
991 uio.uio_segflg = UIO_SYSSPACE;
992 uio.uio_offset = 0;
993
994 buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK);
995
996 while (error == 0) {
997 offset = uio.uio_offset;
998
999 uio.uio_iov = &iov;
1000 uio.uio_iovcnt = 1;
1001 iov.iov_base = buf;
1002 iov.iov_len = MAXBSIZE;
1003 uio.uio_resid = iov.iov_len;
1004 uio.uio_rw = UIO_READ;
1005
1006 if ((error = VOP_READ(lvp, &uio, 0, cred)) != 0)
1007 break;
1008 if ((count = MAXBSIZE - uio.uio_resid) == 0)
1009 break;
1010
1011 bufoffset = 0;
1012 while (bufoffset < count) {
1013 uio.uio_iov = &iov;
1014 uio.uio_iovcnt = 1;
1015 iov.iov_base = buf + bufoffset;
1016 iov.iov_len = count - bufoffset;
1017 uio.uio_offset = offset + bufoffset;
1018 uio.uio_resid = iov.iov_len;
1019 uio.uio_rw = UIO_WRITE;
1020
1021 if ((error = VOP_WRITE(uvp, &uio, 0, cred)) != 0)
1022 break;
1023
1024 bufoffset += (count - bufoffset) - uio.uio_resid;
1025 }
1026
1027 uio.uio_offset = offset + bufoffset;
1028 }
1029
1030 free(buf, M_TEMP);
1031
1032 return (error);
1033}
1034
1035/*
1036 * Copy file from lower to upper.
1037 *
1038 * If you need copy of the contents, set 1 to docopy. Otherwise, set 0 to
1039 * docopy.
1040 *
1041 * If no error returned, unp will be updated.
1042 */
1043int
1044unionfs_copyfile(struct unionfs_node *unp, int docopy, struct ucred *cred,
1045 struct thread *td)
1046{
1047 int error;
1048 struct mount *mp;
1049 struct vnode *udvp;
1050 struct vnode *lvp;
1051 struct vnode *uvp;
1052 struct vattr uva;
1053
1054 lvp = unp->un_lowervp;
1055 uvp = NULLVP;
1056
1057 if ((UNIONFSTOV(unp)->v_mount->mnt_flag & MNT_RDONLY))
1058 return (EROFS);
1059 if (unp->un_dvp == NULLVP)
1060 return (EINVAL);
1061 if (unp->un_uppervp != NULLVP)
1062 return (EEXIST);
1063 udvp = VTOUNIONFS(unp->un_dvp)->un_uppervp;
1064 if (udvp == NULLVP)
1065 return (EROFS);
1066 if ((udvp->v_mount->mnt_flag & MNT_RDONLY))
1067 return (EROFS);
1068
1069 error = VOP_ACCESS(lvp, VREAD, cred, td);
1070 if (error != 0)
1071 return (error);
1072
1073 if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH)) != 0)
1074 return (error);
1075 error = unionfs_vn_create_on_upper(&uvp, udvp, unp, &uva, td);
1076 if (error != 0) {
1077 vn_finished_write(mp);
1078 return (error);
1079 }
1080
1081 if (docopy != 0) {
1082 error = VOP_OPEN(lvp, FREAD, cred, td, NULL);
1083 if (error == 0) {
1084 error = unionfs_copyfile_core(lvp, uvp, cred, td);
1085 VOP_CLOSE(lvp, FREAD, cred, td);
1086 }
1087 }
1088 VOP_CLOSE(uvp, FWRITE, cred, td);
1089 uvp->v_writecount--;
1090
1091 vn_finished_write(mp);
1092
1093 if (error == 0) {
1094 /* Reset the attributes. Ignore errors. */
1095 uva.va_type = VNON;
1096 VOP_SETATTR(uvp, &uva, cred);
1097 }
1098
1099 unionfs_node_update(unp, uvp, td);
1100
1101 return (error);
1102}
1103
1104/*
1105 * It checks whether vp can rmdir. (check empty)
1106 *
1107 * vp is unionfs vnode.
1108 * vp should be locked.
1109 */
1110int
1111unionfs_check_rmdir(struct vnode *vp, struct ucred *cred, struct thread *td)
1112{
1113 int error;
1114 int eofflag;
1115 int lookuperr;
1116 struct vnode *uvp;
1117 struct vnode *lvp;
1118 struct vnode *tvp;
1119 struct vattr va;
1120 struct componentname cn;
1121 /*
1122 * The size of buf needs to be larger than DIRBLKSIZ.
1123 */
1124 char buf[256 * 6];
1125 struct dirent *dp;
1126 struct dirent *edp;
1127 struct uio uio;
1128 struct iovec iov;
1129
1130 ASSERT_VOP_ELOCKED(vp, "unionfs_check_rmdir");
1131
1132 eofflag = 0;
1133 uvp = UNIONFSVPTOUPPERVP(vp);
1134 lvp = UNIONFSVPTOLOWERVP(vp);
1135
1136 /* check opaque */
1137 if ((error = VOP_GETATTR(uvp, &va, cred)) != 0)
1138 return (error);
1139 if (va.va_flags & OPAQUE)
1140 return (0);
1141
1142 /* open vnode */
1143#ifdef MAC
1144 if ((error = mac_vnode_check_open(cred, vp, VEXEC|VREAD)) != 0)
1145 return (error);
1146#endif
1147 if ((error = VOP_ACCESS(vp, VEXEC|VREAD, cred, td)) != 0)
1148 return (error);
1149 if ((error = VOP_OPEN(vp, FREAD, cred, td, NULL)) != 0)
1150 return (error);
1151
1152 uio.uio_rw = UIO_READ;
1153 uio.uio_segflg = UIO_SYSSPACE;
1154 uio.uio_td = td;
1155 uio.uio_offset = 0;
1156
1157#ifdef MAC
1158 error = mac_vnode_check_readdir(td->td_ucred, lvp);
1159#endif
1160 while (!error && !eofflag) {
1161 iov.iov_base = buf;
1162 iov.iov_len = sizeof(buf);
1163 uio.uio_iov = &iov;
1164 uio.uio_iovcnt = 1;
1165 uio.uio_resid = iov.iov_len;
1166
1167 error = VOP_READDIR(lvp, &uio, cred, &eofflag, NULL, NULL);
1168 if (error != 0)
1169 break;
1170 if (eofflag == 0 && uio.uio_resid == sizeof(buf)) {
1171#ifdef DIAGNOSTIC
1172 panic("bad readdir response from lower FS.");
1173#endif
1174 break;
1175 }
1176
1177 edp = (struct dirent*)&buf[sizeof(buf) - uio.uio_resid];
1178 for (dp = (struct dirent*)buf; !error && dp < edp;
1179 dp = (struct dirent*)((caddr_t)dp + dp->d_reclen)) {
1180 if (dp->d_type == DT_WHT ||
1181 (dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1182 (dp->d_namlen == 2 && !bcmp(dp->d_name, "..", 2)))
1183 continue;
1184
1185 cn.cn_namelen = dp->d_namlen;
1186 cn.cn_pnbuf = NULL;
1187 cn.cn_nameptr = dp->d_name;
1188 cn.cn_nameiop = LOOKUP;
1189 cn.cn_flags = (LOCKPARENT | LOCKLEAF | SAVENAME | RDONLY | ISLASTCN);
1190 cn.cn_lkflags = LK_EXCLUSIVE;
1191 cn.cn_thread = td;
1192 cn.cn_cred = cred;
1193 cn.cn_consume = 0;
1194
1195 /*
1196 * check entry in lower.
1197 * Sometimes, readdir function returns
1198 * wrong entry.
1199 */
1200 lookuperr = VOP_LOOKUP(lvp, &tvp, &cn);
1201
1202 if (!lookuperr)
1203 vput(tvp);
1204 else
1205 continue; /* skip entry */
1206
1207 /*
1208 * check entry
1209 * If it has no exist/whiteout entry in upper,
1210 * directory is not empty.
1211 */
1212 cn.cn_flags = (LOCKPARENT | LOCKLEAF | SAVENAME | RDONLY | ISLASTCN);
1213 lookuperr = VOP_LOOKUP(uvp, &tvp, &cn);
1214
1215 if (!lookuperr)
1216 vput(tvp);
1217
1218 /* ignore exist or whiteout entry */
1219 if (!lookuperr ||
1220 (lookuperr == ENOENT && (cn.cn_flags & ISWHITEOUT)))
1221 continue;
1222
1223 error = ENOTEMPTY;
1224 }
1225 }
1226
1227 /* close vnode */
1228 VOP_CLOSE(vp, FREAD, cred, td);
1229
1230 return (error);
1231}
1232
1233#ifdef DIAGNOSTIC
1234
1235struct vnode *
1236unionfs_checkuppervp(struct vnode *vp, char *fil, int lno)
1237{
1238 struct unionfs_node *unp;
1239
1240 unp = VTOUNIONFS(vp);
1241
1242#ifdef notyet
1243 if (vp->v_op != unionfs_vnodeop_p) {
1244 printf("unionfs_checkuppervp: on non-unionfs-node.\n");
1245#ifdef KDB
1246 kdb_enter(KDB_WHY_UNIONFS,
1247 "unionfs_checkuppervp: on non-unionfs-node.\n");
1248#endif
1249 panic("unionfs_checkuppervp");
1250 };
1251#endif
1252 return (unp->un_uppervp);
1253}
1254
1255struct vnode *
1256unionfs_checklowervp(struct vnode *vp, char *fil, int lno)
1257{
1258 struct unionfs_node *unp;
1259
1260 unp = VTOUNIONFS(vp);
1261
1262#ifdef notyet
1263 if (vp->v_op != unionfs_vnodeop_p) {
1264 printf("unionfs_checklowervp: on non-unionfs-node.\n");
1265#ifdef KDB
1266 kdb_enter(KDB_WHY_UNIONFS,
1267 "unionfs_checklowervp: on non-unionfs-node.\n");
1268#endif
1269 panic("unionfs_checklowervp");
1270 };
1271#endif
1272 return (unp->un_lowervp);
1273}
1274#endif