1 /*-
2 * SPDX-License-Identifier: BSD-3-Clause
3 *
4 * Copyright (c) 1994 Jan-Simon Pendry
5 * Copyright (c) 1994
6 * The Regents of the University of California. All rights reserved.
7 * Copyright (c) 2005, 2006, 2012 Masanori Ozawa <ozawa@ongs.co.jp>, ONGS Inc.
8 * Copyright (c) 2006, 2012 Daichi Goto <daichi@freebsd.org>
9 *
10 * This code is derived from software contributed to Berkeley by
11 * Jan-Simon Pendry.
12 *
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
15 * are met:
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
21 * 3. Neither the name of the University nor the names of its contributors
22 * may be used to endorse or promote products derived from this software
23 * without specific prior written permission.
24 *
25 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
26 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
27 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
28 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
29 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
30 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
31 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
32 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
33 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
34 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
35 * SUCH DAMAGE.
36 *
37 * @(#)union_subr.c 8.20 (Berkeley) 5/20/95
38 * $FreeBSD$
39 */
40
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
44 #include <sys/lock.h>
45 #include <sys/mutex.h>
46 #include <sys/malloc.h>
47 #include <sys/mount.h>
48 #include <sys/namei.h>
49 #include <sys/proc.h>
50 #include <sys/vnode.h>
51 #include <sys/dirent.h>
52 #include <sys/fcntl.h>
53 #include <sys/filedesc.h>
54 #include <sys/stat.h>
55 #include <sys/resourcevar.h>
56
57 #include <security/mac/mac_framework.h>
58
59 #include <vm/uma.h>
60
61 #include <fs/unionfs/union.h>
62
63 #define NUNIONFSNODECACHE 16
64
65 static MALLOC_DEFINE(M_UNIONFSHASH, "UNIONFS hash", "UNIONFS hash table");
66 MALLOC_DEFINE(M_UNIONFSNODE, "UNIONFS node", "UNIONFS vnode private part");
67 MALLOC_DEFINE(M_UNIONFSPATH, "UNIONFS path", "UNIONFS path private part");
68
69 /*
70 * Initialize
71 */
72 int
73 unionfs_init(struct vfsconf *vfsp)
74 {
75 UNIONFSDEBUG("unionfs_init\n"); /* printed during system boot */
76 return (0);
77 }
78
79 /*
80 * Uninitialize
81 */
82 int
83 unionfs_uninit(struct vfsconf *vfsp)
84 {
85 return (0);
86 }
87
88 static struct unionfs_node_hashhead *
89 unionfs_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 */
108 static struct vnode *
109 unionfs_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 */
145 static struct vnode *
146 unionfs_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 */
185 static void
186 unionfs_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 */
208 int
209 unionfs_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 unp = malloc(sizeof(struct unionfs_node),
253 M_UNIONFSNODE, M_WAITOK | M_ZERO);
254
255 error = getnewvnode("unionfs", mp, &unionfs_vnodeops, &vp);
256 if (error != 0) {
257 free(unp, M_UNIONFSNODE);
258 return (error);
259 }
260 error = insmntque(vp, mp); /* XXX: Too early for mpsafe fs */
261 if (error != 0) {
262 free(unp, M_UNIONFSNODE);
263 return (error);
264 }
265 if (dvp != NULLVP)
266 vref(dvp);
267 if (uppervp != NULLVP)
268 vref(uppervp);
269 if (lowervp != NULLVP)
270 vref(lowervp);
271
272 if (vt == VDIR)
273 unp->un_hashtbl = hashinit(NUNIONFSNODECACHE, M_UNIONFSHASH,
274 &(unp->un_hashmask));
275
276 unp->un_vnode = vp;
277 unp->un_uppervp = uppervp;
278 unp->un_lowervp = lowervp;
279 unp->un_dvp = dvp;
280 if (uppervp != NULLVP)
281 vp->v_vnlock = uppervp->v_vnlock;
282 else
283 vp->v_vnlock = lowervp->v_vnlock;
284
285 if (path != NULL) {
286 unp->un_path = (char *)
287 malloc(cnp->cn_namelen +1, M_UNIONFSPATH, M_WAITOK|M_ZERO);
288 bcopy(cnp->cn_nameptr, unp->un_path, cnp->cn_namelen);
289 unp->un_path[cnp->cn_namelen] = '\0';
290 }
291 vp->v_type = vt;
292 vp->v_data = unp;
293
294 if ((uppervp != NULLVP && ump->um_uppervp == uppervp) &&
295 (lowervp != NULLVP && ump->um_lowervp == lowervp))
296 vp->v_vflag |= VV_ROOT;
297
298 if (path != NULL && dvp != NULLVP && vt == VDIR)
299 *vpp = unionfs_ins_cached_vnode(unp, dvp, path);
300 if ((*vpp) != NULLVP) {
301 if (dvp != NULLVP)
302 vrele(dvp);
303 if (uppervp != NULLVP)
304 vrele(uppervp);
305 if (lowervp != NULLVP)
306 vrele(lowervp);
307
308 unp->un_uppervp = NULLVP;
309 unp->un_lowervp = NULLVP;
310 unp->un_dvp = NULLVP;
311 vrele(vp);
312 vp = *vpp;
313 vref(vp);
314 } else
315 *vpp = vp;
316
317 unionfs_nodeget_out:
318 if (lkflags & LK_TYPE_MASK)
319 vn_lock(vp, lkflags | LK_RETRY);
320
321 return (0);
322 }
323
324 /*
325 * Clean up the unionfs node.
326 */
327 void
328 unionfs_noderem(struct vnode *vp, struct thread *td)
329 {
330 int count;
331 struct unionfs_node *unp, *unp_t1, *unp_t2;
332 struct unionfs_node_hashhead *hd;
333 struct unionfs_node_status *unsp, *unsp_tmp;
334 struct vnode *lvp;
335 struct vnode *uvp;
336 struct vnode *dvp;
337
338 /*
339 * Use the interlock to protect the clearing of v_data to
340 * prevent faults in unionfs_lock().
341 */
342 VI_LOCK(vp);
343 unp = VTOUNIONFS(vp);
344 lvp = unp->un_lowervp;
345 uvp = unp->un_uppervp;
346 dvp = unp->un_dvp;
347 unp->un_lowervp = unp->un_uppervp = NULLVP;
348 vp->v_vnlock = &(vp->v_lock);
349 vp->v_data = NULL;
350 vp->v_object = NULL;
351 if (vp->v_writecount > 0) {
352 if (uvp != NULL)
353 VOP_ADD_WRITECOUNT(uvp, -vp->v_writecount);
354 else if (lvp != NULL)
355 VOP_ADD_WRITECOUNT(lvp, -vp->v_writecount);
356 } else if (vp->v_writecount < 0)
357 vp->v_writecount = 0;
358 VI_UNLOCK(vp);
359
360 if (lvp != NULLVP)
361 VOP_UNLOCK(lvp, LK_RELEASE);
362 if (uvp != NULLVP)
363 VOP_UNLOCK(uvp, LK_RELEASE);
364
365 if (dvp != NULLVP && unp->un_hash.le_prev != NULL)
366 unionfs_rem_cached_vnode(unp, dvp);
367
368 if (lockmgr(vp->v_vnlock, LK_EXCLUSIVE, VI_MTX(vp)) != 0)
369 panic("the lock for deletion is unacquirable.");
370
371 if (lvp != NULLVP)
372 vrele(lvp);
373 if (uvp != NULLVP)
374 vrele(uvp);
375 if (dvp != NULLVP) {
376 vrele(dvp);
377 unp->un_dvp = NULLVP;
378 }
379 if (unp->un_path != NULL) {
380 free(unp->un_path, M_UNIONFSPATH);
381 unp->un_path = NULL;
382 }
383
384 if (unp->un_hashtbl != NULL) {
385 for (count = 0; count <= unp->un_hashmask; count++) {
386 hd = unp->un_hashtbl + count;
387 LIST_FOREACH_SAFE(unp_t1, hd, un_hash, unp_t2) {
388 LIST_REMOVE(unp_t1, un_hash);
389 unp_t1->un_hash.le_next = NULL;
390 unp_t1->un_hash.le_prev = NULL;
391 }
392 }
393 hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, unp->un_hashmask);
394 }
395
396 LIST_FOREACH_SAFE(unsp, &(unp->un_unshead), uns_list, unsp_tmp) {
397 LIST_REMOVE(unsp, uns_list);
398 free(unsp, M_TEMP);
399 }
400 free(unp, M_UNIONFSNODE);
401 }
402
403 /*
404 * Get the unionfs node status.
405 * You need exclusive lock this vnode.
406 */
407 void
408 unionfs_get_node_status(struct unionfs_node *unp, struct thread *td,
409 struct unionfs_node_status **unspp)
410 {
411 struct unionfs_node_status *unsp;
412 pid_t pid = td->td_proc->p_pid;
413
414 KASSERT(NULL != unspp, ("null pointer"));
415 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), "unionfs_get_node_status");
416
417 LIST_FOREACH(unsp, &(unp->un_unshead), uns_list) {
418 if (unsp->uns_pid == pid) {
419 *unspp = unsp;
420 return;
421 }
422 }
423
424 /* create a new unionfs node status */
425 unsp = malloc(sizeof(struct unionfs_node_status),
426 M_TEMP, M_WAITOK | M_ZERO);
427
428 unsp->uns_pid = pid;
429 LIST_INSERT_HEAD(&(unp->un_unshead), unsp, uns_list);
430
431 *unspp = unsp;
432 }
433
434 /*
435 * Remove the unionfs node status, if you can.
436 * You need exclusive lock this vnode.
437 */
438 void
439 unionfs_tryrem_node_status(struct unionfs_node *unp,
440 struct unionfs_node_status *unsp)
441 {
442 KASSERT(NULL != unsp, ("null pointer"));
443 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), "unionfs_get_node_status");
444
445 if (0 < unsp->uns_lower_opencnt || 0 < unsp->uns_upper_opencnt)
446 return;
447
448 LIST_REMOVE(unsp, uns_list);
449 free(unsp, M_TEMP);
450 }
451
452 /*
453 * Create upper node attr.
454 */
455 void
456 unionfs_create_uppervattr_core(struct unionfs_mount *ump,
457 struct vattr *lva,
458 struct vattr *uva,
459 struct thread *td)
460 {
461 VATTR_NULL(uva);
462 uva->va_type = lva->va_type;
463 uva->va_atime = lva->va_atime;
464 uva->va_mtime = lva->va_mtime;
465 uva->va_ctime = lva->va_ctime;
466
467 switch (ump->um_copymode) {
468 case UNIONFS_TRANSPARENT:
469 uva->va_mode = lva->va_mode;
470 uva->va_uid = lva->va_uid;
471 uva->va_gid = lva->va_gid;
472 break;
473 case UNIONFS_MASQUERADE:
474 if (ump->um_uid == lva->va_uid) {
475 uva->va_mode = lva->va_mode & 077077;
476 uva->va_mode |= (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile) & 0700;
477 uva->va_uid = lva->va_uid;
478 uva->va_gid = lva->va_gid;
479 } else {
480 uva->va_mode = (lva->va_type == VDIR ? ump->um_udir : ump->um_ufile);
481 uva->va_uid = ump->um_uid;
482 uva->va_gid = ump->um_gid;
483 }
484 break;
485 default: /* UNIONFS_TRADITIONAL */
486 uva->va_mode = 0777 & ~td->td_proc->p_fd->fd_cmask;
487 uva->va_uid = ump->um_uid;
488 uva->va_gid = ump->um_gid;
489 break;
490 }
491 }
492
493 /*
494 * Create upper node attr.
495 */
496 int
497 unionfs_create_uppervattr(struct unionfs_mount *ump,
498 struct vnode *lvp,
499 struct vattr *uva,
500 struct ucred *cred,
501 struct thread *td)
502 {
503 int error;
504 struct vattr lva;
505
506 if ((error = VOP_GETATTR(lvp, &lva, cred)))
507 return (error);
508
509 unionfs_create_uppervattr_core(ump, &lva, uva, td);
510
511 return (error);
512 }
513
514 /*
515 * relookup
516 *
517 * dvp should be locked on entry and will be locked on return.
518 *
519 * If an error is returned, *vpp will be invalid, otherwise it will hold a
520 * locked, referenced vnode. If *vpp == dvp then remember that only one
521 * LK_EXCLUSIVE lock is held.
522 */
523 int
524 unionfs_relookup(struct vnode *dvp, struct vnode **vpp,
525 struct componentname *cnp, struct componentname *cn,
526 struct thread *td, char *path, int pathlen, u_long nameiop)
527 {
528 int error;
529
530 cn->cn_namelen = pathlen;
531 cn->cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
532 bcopy(path, cn->cn_pnbuf, pathlen);
533 cn->cn_pnbuf[pathlen] = '\0';
534
535 cn->cn_nameiop = nameiop;
536 cn->cn_flags = (LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME | ISLASTCN);
537 cn->cn_lkflags = LK_EXCLUSIVE;
538 cn->cn_thread = td;
539 cn->cn_cred = cnp->cn_cred;
540
541 cn->cn_nameptr = cn->cn_pnbuf;
542
543 if (nameiop == DELETE)
544 cn->cn_flags |= (cnp->cn_flags & (DOWHITEOUT | SAVESTART));
545 else if (RENAME == nameiop)
546 cn->cn_flags |= (cnp->cn_flags & SAVESTART);
547 else if (nameiop == CREATE)
548 cn->cn_flags |= NOCACHE;
549
550 vref(dvp);
551 VOP_UNLOCK(dvp, LK_RELEASE);
552
553 if ((error = relookup(dvp, vpp, cn))) {
554 uma_zfree(namei_zone, cn->cn_pnbuf);
555 cn->cn_flags &= ~HASBUF;
556 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
557 } else
558 vrele(dvp);
559
560 return (error);
561 }
562
563 /*
564 * relookup for CREATE namei operation.
565 *
566 * dvp is unionfs vnode. dvp should be locked.
567 *
568 * If it called 'unionfs_copyfile' function by unionfs_link etc,
569 * VOP_LOOKUP information is broken.
570 * So it need relookup in order to create link etc.
571 */
572 int
573 unionfs_relookup_for_create(struct vnode *dvp, struct componentname *cnp,
574 struct thread *td)
575 {
576 int error;
577 struct vnode *udvp;
578 struct vnode *vp;
579 struct componentname cn;
580
581 udvp = UNIONFSVPTOUPPERVP(dvp);
582 vp = NULLVP;
583
584 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
585 strlen(cnp->cn_nameptr), CREATE);
586 if (error)
587 return (error);
588
589 if (vp != NULLVP) {
590 if (udvp == vp)
591 vrele(vp);
592 else
593 vput(vp);
594
595 error = EEXIST;
596 }
597
598 if (cn.cn_flags & HASBUF) {
599 uma_zfree(namei_zone, cn.cn_pnbuf);
600 cn.cn_flags &= ~HASBUF;
601 }
602
603 if (!error) {
604 cn.cn_flags |= (cnp->cn_flags & HASBUF);
605 cnp->cn_flags = cn.cn_flags;
606 }
607
608 return (error);
609 }
610
611 /*
612 * relookup for DELETE namei operation.
613 *
614 * dvp is unionfs vnode. dvp should be locked.
615 */
616 int
617 unionfs_relookup_for_delete(struct vnode *dvp, struct componentname *cnp,
618 struct thread *td)
619 {
620 int error;
621 struct vnode *udvp;
622 struct vnode *vp;
623 struct componentname cn;
624
625 udvp = UNIONFSVPTOUPPERVP(dvp);
626 vp = NULLVP;
627
628 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
629 strlen(cnp->cn_nameptr), DELETE);
630 if (error)
631 return (error);
632
633 if (vp == NULLVP)
634 error = ENOENT;
635 else {
636 if (udvp == vp)
637 vrele(vp);
638 else
639 vput(vp);
640 }
641
642 if (cn.cn_flags & HASBUF) {
643 uma_zfree(namei_zone, cn.cn_pnbuf);
644 cn.cn_flags &= ~HASBUF;
645 }
646
647 if (!error) {
648 cn.cn_flags |= (cnp->cn_flags & HASBUF);
649 cnp->cn_flags = cn.cn_flags;
650 }
651
652 return (error);
653 }
654
655 /*
656 * relookup for RENAME namei operation.
657 *
658 * dvp is unionfs vnode. dvp should be locked.
659 */
660 int
661 unionfs_relookup_for_rename(struct vnode *dvp, struct componentname *cnp,
662 struct thread *td)
663 {
664 int error;
665 struct vnode *udvp;
666 struct vnode *vp;
667 struct componentname cn;
668
669 udvp = UNIONFSVPTOUPPERVP(dvp);
670 vp = NULLVP;
671
672 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
673 strlen(cnp->cn_nameptr), RENAME);
674 if (error)
675 return (error);
676
677 if (vp != NULLVP) {
678 if (udvp == vp)
679 vrele(vp);
680 else
681 vput(vp);
682 }
683
684 if (cn.cn_flags & HASBUF) {
685 uma_zfree(namei_zone, cn.cn_pnbuf);
686 cn.cn_flags &= ~HASBUF;
687 }
688
689 if (!error) {
690 cn.cn_flags |= (cnp->cn_flags & HASBUF);
691 cnp->cn_flags = cn.cn_flags;
692 }
693
694 return (error);
695
696 }
697
698 /*
699 * Update the unionfs_node.
700 *
701 * uvp is new locked upper vnode. unionfs vnode's lock will be exchanged to the
702 * uvp's lock and lower's lock will be unlocked.
703 */
704 static void
705 unionfs_node_update(struct unionfs_node *unp, struct vnode *uvp,
706 struct thread *td)
707 {
708 unsigned count, lockrec;
709 struct vnode *vp;
710 struct vnode *lvp;
711 struct vnode *dvp;
712
713 vp = UNIONFSTOV(unp);
714 lvp = unp->un_lowervp;
715 ASSERT_VOP_ELOCKED(lvp, "unionfs_node_update");
716 dvp = unp->un_dvp;
717
718 /*
719 * lock update
720 */
721 VI_LOCK(vp);
722 unp->un_uppervp = uvp;
723 vp->v_vnlock = uvp->v_vnlock;
724 VI_UNLOCK(vp);
725 lockrec = lvp->v_vnlock->lk_recurse;
726 for (count = 0; count < lockrec; count++)
727 vn_lock(uvp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY);
728
729 /*
730 * cache update
731 */
732 if (unp->un_path != NULL && dvp != NULLVP && vp->v_type == VDIR) {
733 static struct unionfs_node_hashhead *hd;
734
735 VI_LOCK(dvp);
736 hd = unionfs_get_hashhead(dvp, unp->un_path);
737 LIST_REMOVE(unp, un_hash);
738 LIST_INSERT_HEAD(hd, unp, un_hash);
739 VI_UNLOCK(dvp);
740 }
741 }
742
743 /*
744 * Create a new shadow dir.
745 *
746 * udvp should be locked on entry and will be locked on return.
747 *
748 * If no error returned, unp will be updated.
749 */
750 int
751 unionfs_mkshadowdir(struct unionfs_mount *ump, struct vnode *udvp,
752 struct unionfs_node *unp, struct componentname *cnp,
753 struct thread *td)
754 {
755 int error;
756 struct vnode *lvp;
757 struct vnode *uvp;
758 struct vattr va;
759 struct vattr lva;
760 struct componentname cn;
761 struct mount *mp;
762 struct ucred *cred;
763 struct ucred *credbk;
764 struct uidinfo *rootinfo;
765
766 if (unp->un_uppervp != NULLVP)
767 return (EEXIST);
768
769 lvp = unp->un_lowervp;
770 uvp = NULLVP;
771 credbk = cnp->cn_cred;
772
773 /* Authority change to root */
774 rootinfo = uifind((uid_t)0);
775 cred = crdup(cnp->cn_cred);
776 /*
777 * The calls to chgproccnt() are needed to compensate for change_ruid()
778 * calling chgproccnt().
779 */
780 chgproccnt(cred->cr_ruidinfo, 1, 0);
781 change_euid(cred, rootinfo);
782 change_ruid(cred, rootinfo);
783 change_svuid(cred, (uid_t)0);
784 uifree(rootinfo);
785 cnp->cn_cred = cred;
786
787 memset(&cn, 0, sizeof(cn));
788
789 if ((error = VOP_GETATTR(lvp, &lva, cnp->cn_cred)))
790 goto unionfs_mkshadowdir_abort;
791
792 if ((error = unionfs_relookup(udvp, &uvp, cnp, &cn, td, cnp->cn_nameptr, cnp->cn_namelen, CREATE)))
793 goto unionfs_mkshadowdir_abort;
794 if (uvp != NULLVP) {
795 if (udvp == uvp)
796 vrele(uvp);
797 else
798 vput(uvp);
799
800 error = EEXIST;
801 goto unionfs_mkshadowdir_free_out;
802 }
803
804 if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH)))
805 goto unionfs_mkshadowdir_free_out;
806 unionfs_create_uppervattr_core(ump, &lva, &va, td);
807
808 error = VOP_MKDIR(udvp, &uvp, &cn, &va);
809
810 if (!error) {
811 unionfs_node_update(unp, uvp, td);
812
813 /*
814 * XXX The bug which cannot set uid/gid was corrected.
815 * Ignore errors.
816 */
817 va.va_type = VNON;
818 VOP_SETATTR(uvp, &va, cn.cn_cred);
819 }
820 vn_finished_write(mp);
821
822 unionfs_mkshadowdir_free_out:
823 if (cn.cn_flags & HASBUF) {
824 uma_zfree(namei_zone, cn.cn_pnbuf);
825 cn.cn_flags &= ~HASBUF;
826 }
827
828 unionfs_mkshadowdir_abort:
829 cnp->cn_cred = credbk;
830 chgproccnt(cred->cr_ruidinfo, -1, 0);
831 crfree(cred);
832
833 return (error);
834 }
835
836 /*
837 * Create a new whiteout.
838 *
839 * dvp should be locked on entry and will be locked on return.
840 */
841 int
842 unionfs_mkwhiteout(struct vnode *dvp, struct componentname *cnp,
843 struct thread *td, char *path)
844 {
845 int error;
846 struct vnode *wvp;
847 struct componentname cn;
848 struct mount *mp;
849
850 if (path == NULL)
851 path = cnp->cn_nameptr;
852
853 wvp = NULLVP;
854 if ((error = unionfs_relookup(dvp, &wvp, cnp, &cn, td, path, strlen(path), CREATE)))
855 return (error);
856 if (wvp != NULLVP) {
857 if (cn.cn_flags & HASBUF) {
858 uma_zfree(namei_zone, cn.cn_pnbuf);
859 cn.cn_flags &= ~HASBUF;
860 }
861 if (dvp == wvp)
862 vrele(wvp);
863 else
864 vput(wvp);
865
866 return (EEXIST);
867 }
868
869 if ((error = vn_start_write(dvp, &mp, V_WAIT | PCATCH)))
870 goto unionfs_mkwhiteout_free_out;
871 error = VOP_WHITEOUT(dvp, &cn, CREATE);
872
873 vn_finished_write(mp);
874
875 unionfs_mkwhiteout_free_out:
876 if (cn.cn_flags & HASBUF) {
877 uma_zfree(namei_zone, cn.cn_pnbuf);
878 cn.cn_flags &= ~HASBUF;
879 }
880
881 return (error);
882 }
883
884 /*
885 * Create a new vnode for create a new shadow file.
886 *
887 * If an error is returned, *vpp will be invalid, otherwise it will hold a
888 * locked, referenced and opened vnode.
889 *
890 * unp is never updated.
891 */
892 static int
893 unionfs_vn_create_on_upper(struct vnode **vpp, struct vnode *udvp,
894 struct unionfs_node *unp, struct vattr *uvap,
895 struct thread *td)
896 {
897 struct unionfs_mount *ump;
898 struct vnode *vp;
899 struct vnode *lvp;
900 struct ucred *cred;
901 struct vattr lva;
902 int fmode;
903 int error;
904 struct componentname cn;
905
906 ump = MOUNTTOUNIONFSMOUNT(UNIONFSTOV(unp)->v_mount);
907 vp = NULLVP;
908 lvp = unp->un_lowervp;
909 cred = td->td_ucred;
910 fmode = FFLAGS(O_WRONLY | O_CREAT | O_TRUNC | O_EXCL);
911 error = 0;
912
913 if ((error = VOP_GETATTR(lvp, &lva, cred)) != 0)
914 return (error);
915 unionfs_create_uppervattr_core(ump, &lva, uvap, td);
916
917 if (unp->un_path == NULL)
918 panic("unionfs: un_path is null");
919
920 cn.cn_namelen = strlen(unp->un_path);
921 cn.cn_pnbuf = uma_zalloc(namei_zone, M_WAITOK);
922 bcopy(unp->un_path, cn.cn_pnbuf, cn.cn_namelen + 1);
923 cn.cn_nameiop = CREATE;
924 cn.cn_flags = (LOCKPARENT | LOCKLEAF | HASBUF | SAVENAME | ISLASTCN);
925 cn.cn_lkflags = LK_EXCLUSIVE;
926 cn.cn_thread = td;
927 cn.cn_cred = cred;
928 cn.cn_nameptr = cn.cn_pnbuf;
929
930 vref(udvp);
931 if ((error = relookup(udvp, &vp, &cn)) != 0)
932 goto unionfs_vn_create_on_upper_free_out2;
933 vrele(udvp);
934
935 if (vp != NULLVP) {
936 if (vp == udvp)
937 vrele(vp);
938 else
939 vput(vp);
940 error = EEXIST;
941 goto unionfs_vn_create_on_upper_free_out1;
942 }
943
944 if ((error = VOP_CREATE(udvp, &vp, &cn, uvap)) != 0)
945 goto unionfs_vn_create_on_upper_free_out1;
946
947 if ((error = VOP_OPEN(vp, fmode, cred, td, NULL)) != 0) {
948 vput(vp);
949 goto unionfs_vn_create_on_upper_free_out1;
950 }
951 error = VOP_ADD_WRITECOUNT(vp, 1);
952 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d", __func__, vp,
953 vp->v_writecount);
954 if (error == 0) {
955 *vpp = vp;
956 } else {
957 VOP_CLOSE(vp, fmode, cred, td);
958 }
959
960 unionfs_vn_create_on_upper_free_out1:
961 VOP_UNLOCK(udvp, LK_RELEASE);
962
963 unionfs_vn_create_on_upper_free_out2:
964 if (cn.cn_flags & HASBUF) {
965 uma_zfree(namei_zone, cn.cn_pnbuf);
966 cn.cn_flags &= ~HASBUF;
967 }
968
969 return (error);
970 }
971
972 /*
973 * Copy from lvp to uvp.
974 *
975 * lvp and uvp should be locked and opened on entry and will be locked and
976 * opened on return.
977 */
978 static int
979 unionfs_copyfile_core(struct vnode *lvp, struct vnode *uvp,
980 struct ucred *cred, struct thread *td)
981 {
982 int error;
983 off_t offset;
984 int count;
985 int bufoffset;
986 char *buf;
987 struct uio uio;
988 struct iovec iov;
989
990 error = 0;
991 memset(&uio, 0, sizeof(uio));
992
993 uio.uio_td = td;
994 uio.uio_segflg = UIO_SYSSPACE;
995 uio.uio_offset = 0;
996
997 buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK);
998
999 while (error == 0) {
1000 offset = uio.uio_offset;
1001
1002 uio.uio_iov = &iov;
1003 uio.uio_iovcnt = 1;
1004 iov.iov_base = buf;
1005 iov.iov_len = MAXBSIZE;
1006 uio.uio_resid = iov.iov_len;
1007 uio.uio_rw = UIO_READ;
1008
1009 if ((error = VOP_READ(lvp, &uio, 0, cred)) != 0)
1010 break;
1011 if ((count = MAXBSIZE - uio.uio_resid) == 0)
1012 break;
1013
1014 bufoffset = 0;
1015 while (bufoffset < count) {
1016 uio.uio_iov = &iov;
1017 uio.uio_iovcnt = 1;
1018 iov.iov_base = buf + bufoffset;
1019 iov.iov_len = count - bufoffset;
1020 uio.uio_offset = offset + bufoffset;
1021 uio.uio_resid = iov.iov_len;
1022 uio.uio_rw = UIO_WRITE;
1023
1024 if ((error = VOP_WRITE(uvp, &uio, 0, cred)) != 0)
1025 break;
1026
1027 bufoffset += (count - bufoffset) - uio.uio_resid;
1028 }
1029
1030 uio.uio_offset = offset + bufoffset;
1031 }
1032
1033 free(buf, M_TEMP);
1034
1035 return (error);
1036 }
1037
1038 /*
1039 * Copy file from lower to upper.
1040 *
1041 * If you need copy of the contents, set 1 to docopy. Otherwise, set 0 to
1042 * docopy.
1043 *
1044 * If no error returned, unp will be updated.
1045 */
1046 int
1047 unionfs_copyfile(struct unionfs_node *unp, int docopy, struct ucred *cred,
1048 struct thread *td)
1049 {
1050 int error;
1051 struct mount *mp;
1052 struct vnode *udvp;
1053 struct vnode *lvp;
1054 struct vnode *uvp;
1055 struct vattr uva;
1056
1057 lvp = unp->un_lowervp;
1058 uvp = NULLVP;
1059
1060 if ((UNIONFSTOV(unp)->v_mount->mnt_flag & MNT_RDONLY))
1061 return (EROFS);
1062 if (unp->un_dvp == NULLVP)
1063 return (EINVAL);
1064 if (unp->un_uppervp != NULLVP)
1065 return (EEXIST);
1066 udvp = VTOUNIONFS(unp->un_dvp)->un_uppervp;
1067 if (udvp == NULLVP)
1068 return (EROFS);
1069 if ((udvp->v_mount->mnt_flag & MNT_RDONLY))
1070 return (EROFS);
1071
1072 error = VOP_ACCESS(lvp, VREAD, cred, td);
1073 if (error != 0)
1074 return (error);
1075
1076 if ((error = vn_start_write(udvp, &mp, V_WAIT | PCATCH)) != 0)
1077 return (error);
1078 error = unionfs_vn_create_on_upper(&uvp, udvp, unp, &uva, td);
1079 if (error != 0) {
1080 vn_finished_write(mp);
1081 return (error);
1082 }
1083
1084 if (docopy != 0) {
1085 error = VOP_OPEN(lvp, FREAD, cred, td, NULL);
1086 if (error == 0) {
1087 error = unionfs_copyfile_core(lvp, uvp, cred, td);
1088 VOP_CLOSE(lvp, FREAD, cred, td);
1089 }
1090 }
1091 VOP_CLOSE(uvp, FWRITE, cred, td);
1092 VOP_ADD_WRITECOUNT_CHECKED(uvp, -1);
1093 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d", __func__, uvp,
1094 uvp->v_writecount);
1095
1096 vn_finished_write(mp);
1097
1098 if (error == 0) {
1099 /* Reset the attributes. Ignore errors. */
1100 uva.va_type = VNON;
1101 VOP_SETATTR(uvp, &uva, cred);
1102 }
1103
1104 unionfs_node_update(unp, uvp, td);
1105
1106 return (error);
1107 }
1108
1109 /*
1110 * It checks whether vp can rmdir. (check empty)
1111 *
1112 * vp is unionfs vnode.
1113 * vp should be locked.
1114 */
1115 int
1116 unionfs_check_rmdir(struct vnode *vp, struct ucred *cred, struct thread *td)
1117 {
1118 int error;
1119 int eofflag;
1120 int lookuperr;
1121 struct vnode *uvp;
1122 struct vnode *lvp;
1123 struct vnode *tvp;
1124 struct vattr va;
1125 struct componentname cn;
1126 /*
1127 * The size of buf needs to be larger than DIRBLKSIZ.
1128 */
1129 char buf[256 * 6];
1130 struct dirent *dp;
1131 struct dirent *edp;
1132 struct uio uio;
1133 struct iovec iov;
1134
1135 ASSERT_VOP_ELOCKED(vp, "unionfs_check_rmdir");
1136
1137 eofflag = 0;
1138 uvp = UNIONFSVPTOUPPERVP(vp);
1139 lvp = UNIONFSVPTOLOWERVP(vp);
1140
1141 /* check opaque */
1142 if ((error = VOP_GETATTR(uvp, &va, cred)) != 0)
1143 return (error);
1144 if (va.va_flags & OPAQUE)
1145 return (0);
1146
1147 /* open vnode */
1148 #ifdef MAC
1149 if ((error = mac_vnode_check_open(cred, vp, VEXEC|VREAD)) != 0)
1150 return (error);
1151 #endif
1152 if ((error = VOP_ACCESS(vp, VEXEC|VREAD, cred, td)) != 0)
1153 return (error);
1154 if ((error = VOP_OPEN(vp, FREAD, cred, td, NULL)) != 0)
1155 return (error);
1156
1157 uio.uio_rw = UIO_READ;
1158 uio.uio_segflg = UIO_SYSSPACE;
1159 uio.uio_td = td;
1160 uio.uio_offset = 0;
1161
1162 #ifdef MAC
1163 error = mac_vnode_check_readdir(td->td_ucred, lvp);
1164 #endif
1165 while (!error && !eofflag) {
1166 iov.iov_base = buf;
1167 iov.iov_len = sizeof(buf);
1168 uio.uio_iov = &iov;
1169 uio.uio_iovcnt = 1;
1170 uio.uio_resid = iov.iov_len;
1171
1172 error = VOP_READDIR(lvp, &uio, cred, &eofflag, NULL, NULL);
1173 if (error != 0)
1174 break;
1175 if (eofflag == 0 && uio.uio_resid == sizeof(buf)) {
1176 #ifdef DIAGNOSTIC
1177 panic("bad readdir response from lower FS.");
1178 #endif
1179 break;
1180 }
1181
1182 edp = (struct dirent*)&buf[sizeof(buf) - uio.uio_resid];
1183 for (dp = (struct dirent*)buf; !error && dp < edp;
1184 dp = (struct dirent*)((caddr_t)dp + dp->d_reclen)) {
1185 if (dp->d_type == DT_WHT || dp->d_fileno == 0 ||
1186 (dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1187 (dp->d_namlen == 2 && !bcmp(dp->d_name, "..", 2)))
1188 continue;
1189
1190 cn.cn_namelen = dp->d_namlen;
1191 cn.cn_pnbuf = NULL;
1192 cn.cn_nameptr = dp->d_name;
1193 cn.cn_nameiop = LOOKUP;
1194 cn.cn_flags = (LOCKPARENT | LOCKLEAF | SAVENAME | RDONLY | ISLASTCN);
1195 cn.cn_lkflags = LK_EXCLUSIVE;
1196 cn.cn_thread = td;
1197 cn.cn_cred = cred;
1198
1199 /*
1200 * check entry in lower.
1201 * Sometimes, readdir function returns
1202 * wrong entry.
1203 */
1204 lookuperr = VOP_LOOKUP(lvp, &tvp, &cn);
1205
1206 if (!lookuperr)
1207 vput(tvp);
1208 else
1209 continue; /* skip entry */
1210
1211 /*
1212 * check entry
1213 * If it has no exist/whiteout entry in upper,
1214 * directory is not empty.
1215 */
1216 cn.cn_flags = (LOCKPARENT | LOCKLEAF | SAVENAME | RDONLY | ISLASTCN);
1217 lookuperr = VOP_LOOKUP(uvp, &tvp, &cn);
1218
1219 if (!lookuperr)
1220 vput(tvp);
1221
1222 /* ignore exist or whiteout entry */
1223 if (!lookuperr ||
1224 (lookuperr == ENOENT && (cn.cn_flags & ISWHITEOUT)))
1225 continue;
1226
1227 error = ENOTEMPTY;
1228 }
1229 }
1230
1231 /* close vnode */
1232 VOP_CLOSE(vp, FREAD, cred, td);
1233
1234 return (error);
1235 }
1236
1237 #ifdef DIAGNOSTIC
1238
1239 struct vnode *
1240 unionfs_checkuppervp(struct vnode *vp, char *fil, int lno)
1241 {
1242 struct unionfs_node *unp;
1243
1244 unp = VTOUNIONFS(vp);
1245
1246 #ifdef notyet
1247 if (vp->v_op != unionfs_vnodeop_p) {
1248 printf("unionfs_checkuppervp: on non-unionfs-node.\n");
1249 #ifdef KDB
1250 kdb_enter(KDB_WHY_UNIONFS,
1251 "unionfs_checkuppervp: on non-unionfs-node.\n");
1252 #endif
1253 panic("unionfs_checkuppervp");
1254 }
1255 #endif
1256 return (unp->un_uppervp);
1257 }
1258
1259 struct vnode *
1260 unionfs_checklowervp(struct vnode *vp, char *fil, int lno)
1261 {
1262 struct unionfs_node *unp;
1263
1264 unp = VTOUNIONFS(vp);
1265
1266 #ifdef notyet
1267 if (vp->v_op != unionfs_vnodeop_p) {
1268 printf("unionfs_checklowervp: on non-unionfs-node.\n");
1269 #ifdef KDB
1270 kdb_enter(KDB_WHY_UNIONFS,
1271 "unionfs_checklowervp: on non-unionfs-node.\n");
1272 #endif
1273 panic("unionfs_checklowervp");
1274 }
1275 #endif
1276 return (unp->un_lowervp);
1277 }
1278 #endif
Cache object: cbfd5e0bde6fea98bd92be306caa3f56
|