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 */
39
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/ktr.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/sysctl.h>
56 #include <sys/taskqueue.h>
57 #include <sys/resourcevar.h>
58
59 #include <machine/atomic.h>
60
61 #include <security/mac/mac_framework.h>
62
63 #include <vm/uma.h>
64
65 #include <fs/unionfs/union.h>
66
67 #define NUNIONFSNODECACHE 16
68 #define UNIONFSHASHMASK (NUNIONFSNODECACHE - 1)
69
70 static MALLOC_DEFINE(M_UNIONFSHASH, "UNIONFS hash", "UNIONFS hash table");
71 MALLOC_DEFINE(M_UNIONFSNODE, "UNIONFS node", "UNIONFS vnode private part");
72 MALLOC_DEFINE(M_UNIONFSPATH, "UNIONFS path", "UNIONFS path private part");
73
74 static struct task unionfs_deferred_rele_task;
75 static struct mtx unionfs_deferred_rele_lock;
76 static STAILQ_HEAD(, unionfs_node) unionfs_deferred_rele_list =
77 STAILQ_HEAD_INITIALIZER(unionfs_deferred_rele_list);
78 static TASKQUEUE_DEFINE_THREAD(unionfs_rele);
79
80 unsigned int unionfs_ndeferred = 0;
81 SYSCTL_UINT(_vfs, OID_AUTO, unionfs_ndeferred, CTLFLAG_RD,
82 &unionfs_ndeferred, 0, "unionfs deferred vnode release");
83
84 static void unionfs_deferred_rele(void *, int);
85
86 /*
87 * Initialize
88 */
89 int
unionfs_init(struct vfsconf * vfsp)90 unionfs_init(struct vfsconf *vfsp)
91 {
92 UNIONFSDEBUG("unionfs_init\n"); /* printed during system boot */
93 TASK_INIT(&unionfs_deferred_rele_task, 0, unionfs_deferred_rele, NULL);
94 mtx_init(&unionfs_deferred_rele_lock, "uniondefr", NULL, MTX_DEF);
95 return (0);
96 }
97
98 /*
99 * Uninitialize
100 */
101 int
unionfs_uninit(struct vfsconf * vfsp)102 unionfs_uninit(struct vfsconf *vfsp)
103 {
104 taskqueue_quiesce(taskqueue_unionfs_rele);
105 taskqueue_free(taskqueue_unionfs_rele);
106 mtx_destroy(&unionfs_deferred_rele_lock);
107 return (0);
108 }
109
110 static void
unionfs_deferred_rele(void * arg __unused,int pending __unused)111 unionfs_deferred_rele(void *arg __unused, int pending __unused)
112 {
113 STAILQ_HEAD(, unionfs_node) local_rele_list;
114 struct unionfs_node *unp, *tunp;
115 unsigned int ndeferred;
116
117 ndeferred = 0;
118 STAILQ_INIT(&local_rele_list);
119 mtx_lock(&unionfs_deferred_rele_lock);
120 STAILQ_CONCAT(&local_rele_list, &unionfs_deferred_rele_list);
121 mtx_unlock(&unionfs_deferred_rele_lock);
122 STAILQ_FOREACH_SAFE(unp, &local_rele_list, un_rele, tunp) {
123 ++ndeferred;
124 MPASS(unp->un_dvp != NULL);
125 vrele(unp->un_dvp);
126 free(unp, M_UNIONFSNODE);
127 }
128
129 /* We expect this function to be single-threaded, thus no atomic */
130 unionfs_ndeferred += ndeferred;
131 }
132
133 static struct unionfs_node_hashhead *
unionfs_get_hashhead(struct vnode * dvp,struct vnode * lookup)134 unionfs_get_hashhead(struct vnode *dvp, struct vnode *lookup)
135 {
136 struct unionfs_node *unp;
137
138 unp = VTOUNIONFS(dvp);
139
140 return (&(unp->un_hashtbl[vfs_hash_index(lookup) & UNIONFSHASHMASK]));
141 }
142
143 /*
144 * Attempt to lookup a cached unionfs vnode by upper/lower vp
145 * from dvp, with dvp's interlock held.
146 */
147 static struct vnode *
unionfs_get_cached_vnode_locked(struct vnode * lookup,struct vnode * dvp)148 unionfs_get_cached_vnode_locked(struct vnode *lookup, struct vnode *dvp)
149 {
150 struct unionfs_node *unp;
151 struct unionfs_node_hashhead *hd;
152 struct vnode *vp;
153
154 hd = unionfs_get_hashhead(dvp, lookup);
155
156 LIST_FOREACH(unp, hd, un_hash) {
157 if (unp->un_uppervp == lookup ||
158 unp->un_lowervp == lookup) {
159 vp = UNIONFSTOV(unp);
160 VI_LOCK_FLAGS(vp, MTX_DUPOK);
161 vp->v_iflag &= ~VI_OWEINACT;
162 if (VN_IS_DOOMED(vp) ||
163 ((vp->v_iflag & VI_DOINGINACT) != 0)) {
164 VI_UNLOCK(vp);
165 vp = NULLVP;
166 } else {
167 vrefl(vp);
168 VI_UNLOCK(vp);
169 }
170 return (vp);
171 }
172 }
173
174 return (NULLVP);
175 }
176
177
178 /*
179 * Get the cached vnode.
180 */
181 static struct vnode *
unionfs_get_cached_vnode(struct vnode * uvp,struct vnode * lvp,struct vnode * dvp)182 unionfs_get_cached_vnode(struct vnode *uvp, struct vnode *lvp,
183 struct vnode *dvp)
184 {
185 struct vnode *vp;
186
187 vp = NULLVP;
188 VI_LOCK(dvp);
189 if (uvp != NULLVP)
190 vp = unionfs_get_cached_vnode_locked(uvp, dvp);
191 else if (lvp != NULLVP)
192 vp = unionfs_get_cached_vnode_locked(lvp, dvp);
193 VI_UNLOCK(dvp);
194
195 return (vp);
196 }
197
198 /*
199 * Add the new vnode into cache.
200 */
201 static struct vnode *
unionfs_ins_cached_vnode(struct unionfs_node * uncp,struct vnode * dvp)202 unionfs_ins_cached_vnode(struct unionfs_node *uncp,
203 struct vnode *dvp)
204 {
205 struct unionfs_node_hashhead *hd;
206 struct vnode *vp;
207
208 ASSERT_VOP_ELOCKED(uncp->un_uppervp, __func__);
209 ASSERT_VOP_ELOCKED(uncp->un_lowervp, __func__);
210 KASSERT(uncp->un_uppervp == NULLVP || uncp->un_uppervp->v_type == VDIR,
211 ("%s: v_type != VDIR", __func__));
212 KASSERT(uncp->un_lowervp == NULLVP || uncp->un_lowervp->v_type == VDIR,
213 ("%s: v_type != VDIR", __func__));
214
215 vp = NULLVP;
216 VI_LOCK(dvp);
217 if (uncp->un_uppervp != NULL)
218 vp = unionfs_get_cached_vnode_locked(uncp->un_uppervp, dvp);
219 else if (uncp->un_lowervp != NULL)
220 vp = unionfs_get_cached_vnode_locked(uncp->un_lowervp, dvp);
221 if (vp == NULLVP) {
222 hd = unionfs_get_hashhead(dvp, (uncp->un_uppervp != NULLVP ?
223 uncp->un_uppervp : uncp->un_lowervp));
224 LIST_INSERT_HEAD(hd, uncp, un_hash);
225 }
226 VI_UNLOCK(dvp);
227
228 return (vp);
229 }
230
231 /*
232 * Remove the vnode.
233 */
234 static void
unionfs_rem_cached_vnode(struct unionfs_node * unp,struct vnode * dvp)235 unionfs_rem_cached_vnode(struct unionfs_node *unp, struct vnode *dvp)
236 {
237 KASSERT(unp != NULL, ("%s: null node", __func__));
238 KASSERT(dvp != NULLVP,
239 ("%s: null parent vnode", __func__));
240
241 VI_LOCK(dvp);
242 if (unp->un_hash.le_prev != NULL) {
243 LIST_REMOVE(unp, un_hash);
244 unp->un_hash.le_next = NULL;
245 unp->un_hash.le_prev = NULL;
246 }
247 VI_UNLOCK(dvp);
248 }
249
250 /*
251 * Common cleanup handling for unionfs_nodeget
252 * Upper, lower, and parent directory vnodes are expected to be referenced by
253 * the caller. Upper and lower vnodes, if non-NULL, are also expected to be
254 * exclusively locked by the caller.
255 * This function will return with the caller's locks and references undone.
256 */
257 static void
unionfs_nodeget_cleanup(struct vnode * vp,struct unionfs_node * unp)258 unionfs_nodeget_cleanup(struct vnode *vp, struct unionfs_node *unp)
259 {
260
261 /*
262 * Lock and reset the default vnode lock; vgone() expects a locked
263 * vnode, and we're going to reset the vnode ops.
264 */
265 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
266
267 /*
268 * Clear out private data and reset the vnode ops to avoid use of
269 * unionfs vnode ops on a partially constructed vnode.
270 */
271 VI_LOCK(vp);
272 vp->v_data = NULL;
273 vp->v_vnlock = &vp->v_lock;
274 vp->v_op = &dead_vnodeops;
275 VI_UNLOCK(vp);
276 vgone(vp);
277 vput(vp);
278
279 if (unp->un_dvp != NULLVP)
280 vrele(unp->un_dvp);
281 if (unp->un_uppervp != NULLVP)
282 vput(unp->un_uppervp);
283 if (unp->un_lowervp != NULLVP)
284 vput(unp->un_lowervp);
285 if (unp->un_hashtbl != NULL)
286 hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, UNIONFSHASHMASK);
287 free(unp->un_path, M_UNIONFSPATH);
288 free(unp, M_UNIONFSNODE);
289 }
290
291 /*
292 * Make a new or get existing unionfs node.
293 *
294 * uppervp and lowervp should be unlocked. Because if new unionfs vnode is
295 * locked, uppervp or lowervp is locked too. In order to prevent dead lock,
296 * you should not lock plurality simultaneously.
297 */
298 int
unionfs_nodeget(struct mount * mp,struct vnode * uppervp,struct vnode * lowervp,struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp)299 unionfs_nodeget(struct mount *mp, struct vnode *uppervp,
300 struct vnode *lowervp, struct vnode *dvp, struct vnode **vpp,
301 struct componentname *cnp)
302 {
303 char *path;
304 struct unionfs_mount *ump;
305 struct unionfs_node *unp;
306 struct vnode *vp;
307 u_long hashmask;
308 int error;
309 int lkflags;
310 __enum_uint8(vtype) vt;
311
312 error = 0;
313 ump = MOUNTTOUNIONFSMOUNT(mp);
314 lkflags = (cnp ? cnp->cn_lkflags : 0);
315 path = (cnp ? cnp->cn_nameptr : NULL);
316 *vpp = NULLVP;
317
318 if (uppervp == NULLVP && lowervp == NULLVP)
319 panic("%s: upper and lower is null", __func__);
320
321 vt = (uppervp != NULLVP ? uppervp->v_type : lowervp->v_type);
322
323 /* If it has no ISLASTCN flag, path check is skipped. */
324 if (cnp && !(cnp->cn_flags & ISLASTCN))
325 path = NULL;
326
327 /* check the cache */
328 if (dvp != NULLVP && vt == VDIR) {
329 vp = unionfs_get_cached_vnode(uppervp, lowervp, dvp);
330 if (vp != NULLVP) {
331 *vpp = vp;
332 goto unionfs_nodeget_out;
333 }
334 }
335
336 unp = malloc(sizeof(struct unionfs_node),
337 M_UNIONFSNODE, M_WAITOK | M_ZERO);
338
339 error = getnewvnode("unionfs", mp, &unionfs_vnodeops, &vp);
340 if (error != 0) {
341 free(unp, M_UNIONFSNODE);
342 return (error);
343 }
344 if (dvp != NULLVP)
345 vref(dvp);
346 if (uppervp != NULLVP)
347 vref(uppervp);
348 if (lowervp != NULLVP)
349 vref(lowervp);
350
351 if (vt == VDIR) {
352 unp->un_hashtbl = hashinit(NUNIONFSNODECACHE, M_UNIONFSHASH,
353 &hashmask);
354 KASSERT(hashmask == UNIONFSHASHMASK,
355 ("unexpected unionfs hash mask 0x%lx", hashmask));
356 }
357
358 unp->un_vnode = vp;
359 unp->un_uppervp = uppervp;
360 unp->un_lowervp = lowervp;
361 unp->un_dvp = dvp;
362 if (uppervp != NULLVP)
363 vp->v_vnlock = uppervp->v_vnlock;
364 else
365 vp->v_vnlock = lowervp->v_vnlock;
366
367 if (path != NULL) {
368 unp->un_path = malloc(cnp->cn_namelen + 1,
369 M_UNIONFSPATH, M_WAITOK | M_ZERO);
370 bcopy(cnp->cn_nameptr, unp->un_path, cnp->cn_namelen);
371 unp->un_path[cnp->cn_namelen] = '\0';
372 unp->un_pathlen = cnp->cn_namelen;
373 }
374 vp->v_type = vt;
375 vp->v_data = unp;
376
377 /*
378 * TODO: This is an imperfect check, as there's no guarantee that
379 * the underlying filesystems will always return vnode pointers
380 * for the root inodes that match our cached values. To reduce
381 * the likelihood of failure, for example in the case where either
382 * vnode has been forcibly doomed, we check both pointers and set
383 * VV_ROOT if either matches.
384 */
385 if (ump->um_uppervp == uppervp || ump->um_lowervp == lowervp)
386 vp->v_vflag |= VV_ROOT;
387 KASSERT(dvp != NULL || (vp->v_vflag & VV_ROOT) != 0,
388 ("%s: NULL dvp for non-root vp %p", __func__, vp));
389
390 vn_lock_pair(lowervp, false, LK_EXCLUSIVE, uppervp, false,
391 LK_EXCLUSIVE);
392 error = insmntque1(vp, mp);
393 if (error != 0) {
394 unionfs_nodeget_cleanup(vp, unp);
395 return (error);
396 }
397 if (lowervp != NULL && VN_IS_DOOMED(lowervp)) {
398 vput(lowervp);
399 unp->un_lowervp = lowervp = NULL;
400 }
401 if (uppervp != NULL && VN_IS_DOOMED(uppervp)) {
402 vput(uppervp);
403 unp->un_uppervp = uppervp = NULL;
404 if (lowervp != NULLVP)
405 vp->v_vnlock = lowervp->v_vnlock;
406 }
407 if (lowervp == NULL && uppervp == NULL) {
408 unionfs_nodeget_cleanup(vp, unp);
409 return (ENOENT);
410 }
411
412 vn_set_state(vp, VSTATE_CONSTRUCTED);
413
414 if (dvp != NULLVP && vt == VDIR)
415 *vpp = unionfs_ins_cached_vnode(unp, dvp);
416 if (*vpp != NULLVP) {
417 unionfs_nodeget_cleanup(vp, unp);
418 vp = *vpp;
419 } else {
420 if (uppervp != NULL)
421 VOP_UNLOCK(uppervp);
422 if (lowervp != NULL)
423 VOP_UNLOCK(lowervp);
424 *vpp = vp;
425 }
426
427 unionfs_nodeget_out:
428 if (lkflags & LK_TYPE_MASK)
429 vn_lock(vp, lkflags | LK_RETRY);
430
431 return (0);
432 }
433
434 /*
435 * Clean up the unionfs node.
436 */
437 void
unionfs_noderem(struct vnode * vp)438 unionfs_noderem(struct vnode *vp)
439 {
440 struct unionfs_node *unp, *unp_t1, *unp_t2;
441 struct unionfs_node_hashhead *hd;
442 struct unionfs_node_status *unsp, *unsp_tmp;
443 struct vnode *lvp;
444 struct vnode *uvp;
445 struct vnode *dvp;
446 int count;
447 int writerefs;
448
449 /*
450 * The root vnode lock may be recursed during unmount, because
451 * it may share the same lock as the unionfs mount's covered vnode,
452 * which is locked across VFS_UNMOUNT(). This lock will then be
453 * recursively taken during the vflush() issued by unionfs_unmount().
454 * But we still only need to lock the unionfs lock once, because only
455 * one of those lock operations was taken against a unionfs vnode and
456 * will be undone against a unionfs vnode.
457 */
458 KASSERT(vp->v_vnlock->lk_recurse == 0 || (vp->v_vflag & VV_ROOT) != 0,
459 ("%s: vnode %p locked recursively", __func__, vp));
460 if (lockmgr(&vp->v_lock, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0)
461 panic("%s: failed to acquire lock for vnode lock", __func__);
462
463 /*
464 * Use the interlock to protect the clearing of v_data to
465 * prevent faults in unionfs_lock().
466 */
467 VI_LOCK(vp);
468 unp = VTOUNIONFS(vp);
469 lvp = unp->un_lowervp;
470 uvp = unp->un_uppervp;
471 dvp = unp->un_dvp;
472 unp->un_lowervp = unp->un_uppervp = NULLVP;
473 vp->v_vnlock = &(vp->v_lock);
474 vp->v_data = NULL;
475 vp->v_object = NULL;
476 if (unp->un_hashtbl != NULL) {
477 /*
478 * Clear out any cached child vnodes. This should only
479 * be necessary during forced unmount, when the vnode may
480 * be reclaimed with a non-zero use count. Otherwise the
481 * reference held by each child should prevent reclamation.
482 */
483 for (count = 0; count <= UNIONFSHASHMASK; count++) {
484 hd = unp->un_hashtbl + count;
485 LIST_FOREACH_SAFE(unp_t1, hd, un_hash, unp_t2) {
486 LIST_REMOVE(unp_t1, un_hash);
487 unp_t1->un_hash.le_next = NULL;
488 unp_t1->un_hash.le_prev = NULL;
489 }
490 }
491 }
492 VI_UNLOCK(vp);
493
494 writerefs = atomic_load_int(&vp->v_writecount);
495 VNASSERT(writerefs >= 0, vp,
496 ("%s: write count %d, unexpected text ref", __func__, writerefs));
497 /*
498 * If we were opened for write, we leased the write reference
499 * to the lower vnode. If this is a reclamation due to the
500 * forced unmount, undo the reference now.
501 */
502 if (writerefs > 0) {
503 VNASSERT(uvp != NULL, vp,
504 ("%s: write reference without upper vnode", __func__));
505 VOP_ADD_WRITECOUNT(uvp, -writerefs);
506 }
507 if (lvp != NULLVP)
508 VOP_UNLOCK(lvp);
509 if (uvp != NULLVP)
510 VOP_UNLOCK(uvp);
511
512 if (dvp != NULLVP)
513 unionfs_rem_cached_vnode(unp, dvp);
514
515 if (lvp != NULLVP)
516 vrele(lvp);
517 if (uvp != NULLVP)
518 vrele(uvp);
519 if (unp->un_path != NULL) {
520 free(unp->un_path, M_UNIONFSPATH);
521 unp->un_path = NULL;
522 unp->un_pathlen = 0;
523 }
524
525 if (unp->un_hashtbl != NULL) {
526 hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, UNIONFSHASHMASK);
527 }
528
529 LIST_FOREACH_SAFE(unsp, &(unp->un_unshead), uns_list, unsp_tmp) {
530 LIST_REMOVE(unsp, uns_list);
531 free(unsp, M_TEMP);
532 }
533 if (dvp != NULLVP) {
534 mtx_lock(&unionfs_deferred_rele_lock);
535 STAILQ_INSERT_TAIL(&unionfs_deferred_rele_list, unp, un_rele);
536 mtx_unlock(&unionfs_deferred_rele_lock);
537 taskqueue_enqueue(taskqueue_unionfs_rele,
538 &unionfs_deferred_rele_task);
539 } else
540 free(unp, M_UNIONFSNODE);
541 }
542
543 /*
544 * Get the unionfs node status object for the vnode corresponding to unp,
545 * for the process that owns td. Allocate a new status object if one
546 * does not already exist.
547 */
548 void
unionfs_get_node_status(struct unionfs_node * unp,struct thread * td,struct unionfs_node_status ** unspp)549 unionfs_get_node_status(struct unionfs_node *unp, struct thread *td,
550 struct unionfs_node_status **unspp)
551 {
552 struct unionfs_node_status *unsp;
553 pid_t pid;
554
555 pid = td->td_proc->p_pid;
556
557 KASSERT(NULL != unspp, ("%s: NULL status", __func__));
558 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), __func__);
559
560 LIST_FOREACH(unsp, &(unp->un_unshead), uns_list) {
561 if (unsp->uns_pid == pid) {
562 *unspp = unsp;
563 return;
564 }
565 }
566
567 /* create a new unionfs node status */
568 unsp = malloc(sizeof(struct unionfs_node_status),
569 M_TEMP, M_WAITOK | M_ZERO);
570
571 unsp->uns_pid = pid;
572 LIST_INSERT_HEAD(&(unp->un_unshead), unsp, uns_list);
573
574 *unspp = unsp;
575 }
576
577 /*
578 * Remove the unionfs node status, if you can.
579 * You need exclusive lock this vnode.
580 */
581 void
unionfs_tryrem_node_status(struct unionfs_node * unp,struct unionfs_node_status * unsp)582 unionfs_tryrem_node_status(struct unionfs_node *unp,
583 struct unionfs_node_status *unsp)
584 {
585 KASSERT(NULL != unsp, ("%s: NULL status", __func__));
586 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), __func__);
587
588 if (0 < unsp->uns_lower_opencnt || 0 < unsp->uns_upper_opencnt)
589 return;
590
591 LIST_REMOVE(unsp, uns_list);
592 free(unsp, M_TEMP);
593 }
594
595 /*
596 * Create upper node attr.
597 */
598 void
unionfs_create_uppervattr_core(struct unionfs_mount * ump,struct vattr * lva,struct vattr * uva,struct thread * td)599 unionfs_create_uppervattr_core(struct unionfs_mount *ump, struct vattr *lva,
600 struct vattr *uva, struct thread *td)
601 {
602 VATTR_NULL(uva);
603 uva->va_type = lva->va_type;
604 uva->va_atime = lva->va_atime;
605 uva->va_mtime = lva->va_mtime;
606 uva->va_ctime = lva->va_ctime;
607
608 switch (ump->um_copymode) {
609 case UNIONFS_TRANSPARENT:
610 uva->va_mode = lva->va_mode;
611 uva->va_uid = lva->va_uid;
612 uva->va_gid = lva->va_gid;
613 break;
614 case UNIONFS_MASQUERADE:
615 if (ump->um_uid == lva->va_uid) {
616 uva->va_mode = lva->va_mode & 077077;
617 uva->va_mode |= (lva->va_type == VDIR ?
618 ump->um_udir : ump->um_ufile) & 0700;
619 uva->va_uid = lva->va_uid;
620 uva->va_gid = lva->va_gid;
621 } else {
622 uva->va_mode = (lva->va_type == VDIR ?
623 ump->um_udir : ump->um_ufile);
624 uva->va_uid = ump->um_uid;
625 uva->va_gid = ump->um_gid;
626 }
627 break;
628 default: /* UNIONFS_TRADITIONAL */
629 uva->va_mode = 0777 & ~td->td_proc->p_pd->pd_cmask;
630 uva->va_uid = ump->um_uid;
631 uva->va_gid = ump->um_gid;
632 break;
633 }
634 }
635
636 /*
637 * Create upper node attr.
638 */
639 int
unionfs_create_uppervattr(struct unionfs_mount * ump,struct vnode * lvp,struct vattr * uva,struct ucred * cred,struct thread * td)640 unionfs_create_uppervattr(struct unionfs_mount *ump, struct vnode *lvp,
641 struct vattr *uva, struct ucred *cred, struct thread *td)
642 {
643 struct vattr lva;
644 int error;
645
646 if ((error = VOP_GETATTR(lvp, &lva, cred)))
647 return (error);
648
649 unionfs_create_uppervattr_core(ump, &lva, uva, td);
650
651 return (error);
652 }
653
654 /*
655 * relookup
656 *
657 * dvp should be locked on entry and will be locked on return.
658 *
659 * If an error is returned, *vpp will be invalid, otherwise it will hold a
660 * locked, referenced vnode. If *vpp == dvp then remember that only one
661 * LK_EXCLUSIVE lock is held.
662 */
663 int
unionfs_relookup(struct vnode * dvp,struct vnode ** vpp,struct componentname * cnp,struct componentname * cn,struct thread * td,char * path,int pathlen,u_long nameiop)664 unionfs_relookup(struct vnode *dvp, struct vnode **vpp,
665 struct componentname *cnp, struct componentname *cn, struct thread *td,
666 char *path, int pathlen, u_long nameiop)
667 {
668 int error;
669 bool refstart;
670
671 cn->cn_namelen = pathlen;
672 cn->cn_pnbuf = path;
673 cn->cn_nameiop = nameiop;
674 cn->cn_flags = (LOCKPARENT | LOCKLEAF | ISLASTCN);
675 cn->cn_lkflags = LK_EXCLUSIVE;
676 cn->cn_cred = cnp->cn_cred;
677 cn->cn_nameptr = cn->cn_pnbuf;
678
679 refstart = false;
680 if (nameiop == DELETE) {
681 cn->cn_flags |= (cnp->cn_flags & DOWHITEOUT);
682 } else if (nameiop == RENAME) {
683 refstart = true;
684 } else if (nameiop == CREATE) {
685 cn->cn_flags |= NOCACHE;
686 }
687
688 vref(dvp);
689 VOP_UNLOCK(dvp);
690
691 if ((error = vfs_relookup(dvp, vpp, cn, refstart))) {
692 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
693 } else
694 vrele(dvp);
695
696 KASSERT(cn->cn_pnbuf == path, ("%s: cn_pnbuf changed", __func__));
697
698 return (error);
699 }
700
701 /*
702 * relookup for CREATE namei operation.
703 *
704 * dvp is unionfs vnode. dvp should be locked.
705 *
706 * If it called 'unionfs_copyfile' function by unionfs_link etc,
707 * VOP_LOOKUP information is broken.
708 * So it need relookup in order to create link etc.
709 */
710 int
unionfs_relookup_for_create(struct vnode * dvp,struct componentname * cnp,struct thread * td)711 unionfs_relookup_for_create(struct vnode *dvp, struct componentname *cnp,
712 struct thread *td)
713 {
714 struct vnode *udvp;
715 struct vnode *vp;
716 struct componentname cn;
717 int error;
718
719 udvp = UNIONFSVPTOUPPERVP(dvp);
720 vp = NULLVP;
721
722 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
723 cnp->cn_namelen, CREATE);
724 if (error)
725 return (error);
726
727 if (vp != NULLVP) {
728 if (udvp == vp)
729 vrele(vp);
730 else
731 vput(vp);
732
733 error = EEXIST;
734 }
735
736 return (error);
737 }
738
739 /*
740 * relookup for DELETE namei operation.
741 *
742 * dvp is unionfs vnode. dvp should be locked.
743 */
744 int
unionfs_relookup_for_delete(struct vnode * dvp,struct componentname * cnp,struct thread * td)745 unionfs_relookup_for_delete(struct vnode *dvp, struct componentname *cnp,
746 struct thread *td)
747 {
748 struct vnode *udvp;
749 struct vnode *vp;
750 struct componentname cn;
751 int error;
752
753 udvp = UNIONFSVPTOUPPERVP(dvp);
754 vp = NULLVP;
755
756 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
757 cnp->cn_namelen, DELETE);
758 if (error)
759 return (error);
760
761 if (vp == NULLVP)
762 error = ENOENT;
763 else {
764 if (udvp == vp)
765 vrele(vp);
766 else
767 vput(vp);
768 }
769
770 return (error);
771 }
772
773 /*
774 * relookup for RENAME namei operation.
775 *
776 * dvp is unionfs vnode. dvp should be locked.
777 */
778 int
unionfs_relookup_for_rename(struct vnode * dvp,struct componentname * cnp,struct thread * td)779 unionfs_relookup_for_rename(struct vnode *dvp, struct componentname *cnp,
780 struct thread *td)
781 {
782 struct vnode *udvp;
783 struct vnode *vp;
784 struct componentname cn;
785 int error;
786
787 udvp = UNIONFSVPTOUPPERVP(dvp);
788 vp = NULLVP;
789
790 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
791 cnp->cn_namelen, RENAME);
792 if (error)
793 return (error);
794
795 if (vp != NULLVP) {
796 if (udvp == vp)
797 vrele(vp);
798 else
799 vput(vp);
800 }
801
802 return (error);
803 }
804
805 /*
806 * Update the unionfs_node.
807 *
808 * uvp is new locked upper vnode. unionfs vnode's lock will be exchanged to the
809 * uvp's lock and lower's lock will be unlocked.
810 */
811 static void
unionfs_node_update(struct unionfs_node * unp,struct vnode * uvp,struct thread * td)812 unionfs_node_update(struct unionfs_node *unp, struct vnode *uvp,
813 struct thread *td)
814 {
815 struct unionfs_node_hashhead *hd;
816 struct vnode *vp;
817 struct vnode *lvp;
818 struct vnode *dvp;
819 unsigned count, lockrec;
820
821 vp = UNIONFSTOV(unp);
822 lvp = unp->un_lowervp;
823 ASSERT_VOP_ELOCKED(lvp, __func__);
824 ASSERT_VOP_ELOCKED(uvp, __func__);
825 dvp = unp->un_dvp;
826
827 VNASSERT(vp->v_writecount == 0, vp,
828 ("%s: non-zero writecount", __func__));
829 /*
830 * Update the upper vnode's lock state to match the lower vnode,
831 * and then switch the unionfs vnode's lock to the upper vnode.
832 */
833 lockrec = lvp->v_vnlock->lk_recurse;
834 for (count = 0; count < lockrec; count++)
835 vn_lock(uvp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY);
836 VI_LOCK(vp);
837 unp->un_uppervp = uvp;
838 vp->v_vnlock = uvp->v_vnlock;
839 VI_UNLOCK(vp);
840
841 /*
842 * Re-cache the unionfs vnode against the upper vnode
843 */
844 if (dvp != NULLVP && vp->v_type == VDIR) {
845 VI_LOCK(dvp);
846 if (unp->un_hash.le_prev != NULL) {
847 LIST_REMOVE(unp, un_hash);
848 hd = unionfs_get_hashhead(dvp, uvp);
849 LIST_INSERT_HEAD(hd, unp, un_hash);
850 }
851 VI_UNLOCK(unp->un_dvp);
852 }
853 }
854
855 /*
856 * Create a new shadow dir.
857 *
858 * udvp should be locked on entry and will be locked on return.
859 *
860 * If no error returned, unp will be updated.
861 */
862 int
unionfs_mkshadowdir(struct unionfs_mount * ump,struct vnode * udvp,struct unionfs_node * unp,struct componentname * cnp,struct thread * td)863 unionfs_mkshadowdir(struct unionfs_mount *ump, struct vnode *udvp,
864 struct unionfs_node *unp, struct componentname *cnp, struct thread *td)
865 {
866 struct vnode *lvp;
867 struct vnode *uvp;
868 struct vattr va;
869 struct vattr lva;
870 struct nameidata nd;
871 struct mount *mp;
872 struct ucred *cred;
873 struct ucred *credbk;
874 struct uidinfo *rootinfo;
875 int error;
876
877 if (unp->un_uppervp != NULLVP)
878 return (EEXIST);
879
880 lvp = unp->un_lowervp;
881 uvp = NULLVP;
882 credbk = cnp->cn_cred;
883
884 /* Authority change to root */
885 rootinfo = uifind((uid_t)0);
886 cred = crdup(cnp->cn_cred);
887 change_euid(cred, rootinfo);
888 change_ruid(cred, rootinfo);
889 change_svuid(cred, (uid_t)0);
890 uifree(rootinfo);
891 cnp->cn_cred = cred;
892
893 memset(&nd.ni_cnd, 0, sizeof(struct componentname));
894 NDPREINIT(&nd);
895
896 if ((error = VOP_GETATTR(lvp, &lva, cnp->cn_cred)))
897 goto unionfs_mkshadowdir_abort;
898
899 if ((error = unionfs_relookup(udvp, &uvp, cnp, &nd.ni_cnd, td,
900 cnp->cn_nameptr, cnp->cn_namelen, CREATE)))
901 goto unionfs_mkshadowdir_abort;
902 if (uvp != NULLVP) {
903 if (udvp == uvp)
904 vrele(uvp);
905 else
906 vput(uvp);
907
908 error = EEXIST;
909 goto unionfs_mkshadowdir_abort;
910 }
911
912 if ((error = vn_start_write(udvp, &mp, V_WAIT | V_PCATCH)))
913 goto unionfs_mkshadowdir_abort;
914 unionfs_create_uppervattr_core(ump, &lva, &va, td);
915
916 /*
917 * Temporarily NUL-terminate the current pathname component.
918 * This function may be called during lookup operations in which
919 * the current pathname component is not the leaf, meaning that
920 * the NUL terminator is some distance beyond the end of the current
921 * component. This *should* be fine, as cn_namelen will still
922 * correctly indicate the length of only the current component,
923 * but ZFS in particular does not respect cn_namelen in its VOP_MKDIR
924 * implementation
925 * Note that this assumes nd.ni_cnd.cn_pnbuf was allocated by
926 * something like a local namei() operation and the temporary
927 * NUL-termination will not have an effect on other threads.
928 */
929 char *pathend = &nd.ni_cnd.cn_nameptr[nd.ni_cnd.cn_namelen];
930 char pathterm = *pathend;
931 *pathend = '\0';
932 error = VOP_MKDIR(udvp, &uvp, &nd.ni_cnd, &va);
933 *pathend = pathterm;
934
935 if (!error) {
936 /*
937 * XXX The bug which cannot set uid/gid was corrected.
938 * Ignore errors.
939 */
940 va.va_type = VNON;
941 VOP_SETATTR(uvp, &va, nd.ni_cnd.cn_cred);
942
943 /*
944 * VOP_SETATTR() may transiently drop uvp's lock, so it's
945 * important to call it before unionfs_node_update() transfers
946 * the unionfs vnode's lock from lvp to uvp; otherwise the
947 * unionfs vnode itself would be transiently unlocked and
948 * potentially doomed.
949 */
950 unionfs_node_update(unp, uvp, td);
951 }
952 vn_finished_write(mp);
953
954 unionfs_mkshadowdir_abort:
955 cnp->cn_cred = credbk;
956 crfree(cred);
957
958 return (error);
959 }
960
961 static inline void
unionfs_forward_vop_ref(struct vnode * basevp,int * lkflags)962 unionfs_forward_vop_ref(struct vnode *basevp, int *lkflags)
963 {
964 ASSERT_VOP_LOCKED(basevp, __func__);
965 *lkflags = VOP_ISLOCKED(basevp);
966 vref(basevp);
967 }
968
969 /*
970 * Prepare unionfs to issue a forwarded VOP to either the upper or lower
971 * FS. This should be used for any VOP which may drop the vnode lock;
972 * it is not required otherwise.
973 * The unionfs vnode shares its lock with the base-layer vnode(s); if the
974 * base FS must transiently drop its vnode lock, the unionfs vnode may
975 * effectively become unlocked. During that window, a concurrent forced
976 * unmount may doom the unionfs vnode, which leads to two significant
977 * issues:
978 * 1) Completion of, and return from, the unionfs VOP with the unionfs
979 * vnode completely unlocked. When the unionfs vnode becomes doomed
980 * it stops sharing its lock with the base vnode, so even if the
981 * forwarded VOP reacquires the base vnode lock the unionfs vnode
982 * lock will no longer be held. This can lead to violation of the
983 * caller's sychronization requirements as well as various failed
984 * locking assertions when DEBUG_VFS_LOCKS is enabled.
985 * 2) Loss of reference on the base vnode. The caller is expected to
986 * hold a v_usecount reference on the unionfs vnode, while the
987 * unionfs vnode holds a reference on the base-layer vnode(s). But
988 * these references are released when the unionfs vnode becomes
989 * doomed, violating the base layer's expectation that its caller
990 * must hold a reference to prevent vnode recycling.
991 *
992 * basevp1 and basevp2 represent two base-layer vnodes which are
993 * expected to be locked when this function is called. basevp2
994 * may be NULL, but if not NULL basevp1 and basevp2 should represent
995 * a parent directory and a filed linked to it, respectively.
996 * lkflags1 and lkflags2 are output parameters that will store the
997 * current lock status of basevp1 and basevp2, respectively. They
998 * are intended to be passed as the lkflags1 and lkflags2 parameters
999 * in the subsequent call to unionfs_forward_vop_finish_pair().
1000 * lkflags2 may be NULL iff basevp2 is NULL.
1001 */
1002 void
unionfs_forward_vop_start_pair(struct vnode * basevp1,int * lkflags1,struct vnode * basevp2,int * lkflags2)1003 unionfs_forward_vop_start_pair(struct vnode *basevp1, int *lkflags1,
1004 struct vnode *basevp2, int *lkflags2)
1005 {
1006 /*
1007 * Take an additional reference on the base-layer vnodes to
1008 * avoid loss of reference if the unionfs vnodes are doomed.
1009 */
1010 unionfs_forward_vop_ref(basevp1, lkflags1);
1011 if (basevp2 != NULL)
1012 unionfs_forward_vop_ref(basevp2, lkflags2);
1013 }
1014
1015 static inline bool
unionfs_forward_vop_rele(struct vnode * unionvp,struct vnode * basevp,int lkflags)1016 unionfs_forward_vop_rele(struct vnode *unionvp, struct vnode *basevp,
1017 int lkflags)
1018 {
1019 bool unionvp_doomed;
1020
1021 if (__predict_false(VTOUNIONFS(unionvp) == NULL)) {
1022 if ((lkflags & LK_EXCLUSIVE) != 0)
1023 ASSERT_VOP_ELOCKED(basevp, __func__);
1024 else
1025 ASSERT_VOP_LOCKED(basevp, __func__);
1026 unionvp_doomed = true;
1027 } else {
1028 vrele(basevp);
1029 unionvp_doomed = false;
1030 }
1031
1032 return (unionvp_doomed);
1033 }
1034
1035
1036 /*
1037 * Indicate completion of a forwarded VOP previously prepared by
1038 * unionfs_forward_vop_start_pair().
1039 * basevp1 and basevp2 must be the same values passed to the prior
1040 * call to unionfs_forward_vop_start_pair(). unionvp1 and unionvp2
1041 * must be the unionfs vnodes that were initially above basevp1 and
1042 * basevp2, respectively.
1043 * basevp1 and basevp2 (if not NULL) must be locked when this function
1044 * is called, while unionvp1 and/or unionvp2 may be unlocked if either
1045 * unionfs vnode has become doomed.
1046 * lkflags1 and lkflag2 represent the locking flags that should be
1047 * used to re-lock unionvp1 and unionvp2, respectively, if either
1048 * vnode has become doomed.
1049 *
1050 * Returns true if any unionfs vnode was found to be doomed, false
1051 * otherwise.
1052 */
1053 bool
unionfs_forward_vop_finish_pair(struct vnode * unionvp1,struct vnode * basevp1,int lkflags1,struct vnode * unionvp2,struct vnode * basevp2,int lkflags2)1054 unionfs_forward_vop_finish_pair(
1055 struct vnode *unionvp1, struct vnode *basevp1, int lkflags1,
1056 struct vnode *unionvp2, struct vnode *basevp2, int lkflags2)
1057 {
1058 bool vp1_doomed, vp2_doomed;
1059
1060 /*
1061 * If either vnode is found to have been doomed, set
1062 * a flag indicating that it needs to be re-locked.
1063 * Otherwise, simply drop the base-vnode reference that
1064 * was taken in unionfs_forward_vop_start().
1065 */
1066 vp1_doomed = unionfs_forward_vop_rele(unionvp1, basevp1, lkflags1);
1067
1068 if (unionvp2 != NULL)
1069 vp2_doomed = unionfs_forward_vop_rele(unionvp2, basevp2, lkflags2);
1070 else
1071 vp2_doomed = false;
1072
1073 /*
1074 * If any of the unionfs vnodes need to be re-locked, that
1075 * means the unionfs vnode's lock is now de-coupled from the
1076 * corresponding base vnode. We therefore need to drop the
1077 * base vnode lock (since nothing else will after this point),
1078 * and also release the reference taken in
1079 * unionfs_forward_vop_start_pair().
1080 */
1081 if (__predict_false(vp1_doomed && vp2_doomed))
1082 VOP_VPUT_PAIR(basevp1, &basevp2, true);
1083 else if (__predict_false(vp1_doomed)) {
1084 /*
1085 * If basevp1 needs to be unlocked, then we may not
1086 * be able to safely unlock it with basevp2 still locked,
1087 * for the same reason that an ordinary VFS call would
1088 * need to use VOP_VPUT_PAIR() here. We might be able
1089 * to use VOP_VPUT_PAIR(..., false) here, but then we
1090 * would need to deal with the possibility of basevp2
1091 * changing out from under us, which could result in
1092 * either the unionfs vnode becoming doomed or its
1093 * upper/lower vp no longer matching basevp2. Either
1094 * scenario would require at least re-locking the unionfs
1095 * vnode anyway.
1096 */
1097 if (unionvp2 != NULL) {
1098 VOP_UNLOCK(unionvp2);
1099 vp2_doomed = true;
1100 }
1101 vput(basevp1);
1102 } else if (__predict_false(vp2_doomed))
1103 vput(basevp2);
1104
1105 if (__predict_false(vp1_doomed || vp2_doomed))
1106 vn_lock_pair(unionvp1, !vp1_doomed, lkflags1,
1107 unionvp2, !vp2_doomed, lkflags2);
1108
1109 return (vp1_doomed || vp2_doomed);
1110 }
1111
1112 /*
1113 * Create a new whiteout.
1114 *
1115 * udvp and dvp should be locked on entry and will be locked on return.
1116 */
1117 int
unionfs_mkwhiteout(struct vnode * dvp,struct vnode * udvp,struct componentname * cnp,struct thread * td,char * path,int pathlen)1118 unionfs_mkwhiteout(struct vnode *dvp, struct vnode *udvp,
1119 struct componentname *cnp, struct thread *td, char *path, int pathlen)
1120 {
1121 struct vnode *wvp;
1122 struct nameidata nd;
1123 struct mount *mp;
1124 int error;
1125 int lkflags;
1126
1127 wvp = NULLVP;
1128 NDPREINIT(&nd);
1129 if ((error = unionfs_relookup(udvp, &wvp, cnp, &nd.ni_cnd, td, path,
1130 pathlen, CREATE))) {
1131 return (error);
1132 }
1133 if (wvp != NULLVP) {
1134 if (udvp == wvp)
1135 vrele(wvp);
1136 else
1137 vput(wvp);
1138
1139 return (EEXIST);
1140 }
1141
1142 if ((error = vn_start_write(udvp, &mp, V_WAIT | V_PCATCH)))
1143 goto unionfs_mkwhiteout_free_out;
1144 unionfs_forward_vop_start(udvp, &lkflags);
1145 error = VOP_WHITEOUT(udvp, &nd.ni_cnd, CREATE);
1146 unionfs_forward_vop_finish(dvp, udvp, lkflags);
1147
1148 vn_finished_write(mp);
1149
1150 unionfs_mkwhiteout_free_out:
1151 return (error);
1152 }
1153
1154 /*
1155 * Create a new vnode for create a new shadow file.
1156 *
1157 * If an error is returned, *vpp will be invalid, otherwise it will hold a
1158 * locked, referenced and opened vnode.
1159 *
1160 * unp is never updated.
1161 */
1162 static int
unionfs_vn_create_on_upper(struct vnode ** vpp,struct vnode * udvp,struct unionfs_node * unp,struct vattr * uvap,struct thread * td)1163 unionfs_vn_create_on_upper(struct vnode **vpp, struct vnode *udvp,
1164 struct unionfs_node *unp, struct vattr *uvap, struct thread *td)
1165 {
1166 struct unionfs_mount *ump;
1167 struct vnode *vp;
1168 struct vnode *lvp;
1169 struct ucred *cred;
1170 struct vattr lva;
1171 struct nameidata nd;
1172 int fmode;
1173 int error;
1174
1175 ump = MOUNTTOUNIONFSMOUNT(UNIONFSTOV(unp)->v_mount);
1176 vp = NULLVP;
1177 lvp = unp->un_lowervp;
1178 cred = td->td_ucred;
1179 fmode = FFLAGS(O_WRONLY | O_CREAT | O_TRUNC | O_EXCL);
1180 error = 0;
1181
1182 if ((error = VOP_GETATTR(lvp, &lva, cred)) != 0)
1183 return (error);
1184 unionfs_create_uppervattr_core(ump, &lva, uvap, td);
1185
1186 if (unp->un_path == NULL)
1187 panic("%s: NULL un_path", __func__);
1188
1189 nd.ni_cnd.cn_namelen = unp->un_pathlen;
1190 nd.ni_cnd.cn_pnbuf = unp->un_path;
1191 nd.ni_cnd.cn_nameiop = CREATE;
1192 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | ISLASTCN;
1193 nd.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
1194 nd.ni_cnd.cn_cred = cred;
1195 nd.ni_cnd.cn_nameptr = nd.ni_cnd.cn_pnbuf;
1196 NDPREINIT(&nd);
1197
1198 vref(udvp);
1199 if ((error = vfs_relookup(udvp, &vp, &nd.ni_cnd, false)) != 0)
1200 goto unionfs_vn_create_on_upper_free_out2;
1201 vrele(udvp);
1202
1203 if (vp != NULLVP) {
1204 if (vp == udvp)
1205 vrele(vp);
1206 else
1207 vput(vp);
1208 error = EEXIST;
1209 goto unionfs_vn_create_on_upper_free_out1;
1210 }
1211
1212 if ((error = VOP_CREATE(udvp, &vp, &nd.ni_cnd, uvap)) != 0)
1213 goto unionfs_vn_create_on_upper_free_out1;
1214
1215 if ((error = VOP_OPEN(vp, fmode, cred, td, NULL)) != 0) {
1216 vput(vp);
1217 goto unionfs_vn_create_on_upper_free_out1;
1218 }
1219 error = VOP_ADD_WRITECOUNT(vp, 1);
1220 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
1221 __func__, vp, vp->v_writecount);
1222 if (error == 0) {
1223 *vpp = vp;
1224 } else {
1225 VOP_CLOSE(vp, fmode, cred, td);
1226 }
1227
1228 unionfs_vn_create_on_upper_free_out1:
1229 VOP_UNLOCK(udvp);
1230
1231 unionfs_vn_create_on_upper_free_out2:
1232 KASSERT(nd.ni_cnd.cn_pnbuf == unp->un_path,
1233 ("%s: cn_pnbuf changed", __func__));
1234
1235 return (error);
1236 }
1237
1238 /*
1239 * Copy from lvp to uvp.
1240 *
1241 * lvp and uvp should be locked and opened on entry and will be locked and
1242 * opened on return.
1243 */
1244 static int
unionfs_copyfile_core(struct vnode * lvp,struct vnode * uvp,struct ucred * cred,struct thread * td)1245 unionfs_copyfile_core(struct vnode *lvp, struct vnode *uvp,
1246 struct ucred *cred, struct thread *td)
1247 {
1248 char *buf;
1249 struct uio uio;
1250 struct iovec iov;
1251 off_t offset;
1252 int count;
1253 int error;
1254 int bufoffset;
1255
1256 error = 0;
1257 memset(&uio, 0, sizeof(uio));
1258
1259 uio.uio_td = td;
1260 uio.uio_segflg = UIO_SYSSPACE;
1261 uio.uio_offset = 0;
1262
1263 buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK);
1264
1265 while (error == 0) {
1266 offset = uio.uio_offset;
1267
1268 uio.uio_iov = &iov;
1269 uio.uio_iovcnt = 1;
1270 iov.iov_base = buf;
1271 iov.iov_len = MAXBSIZE;
1272 uio.uio_resid = iov.iov_len;
1273 uio.uio_rw = UIO_READ;
1274
1275 if ((error = VOP_READ(lvp, &uio, 0, cred)) != 0)
1276 break;
1277 if ((count = MAXBSIZE - uio.uio_resid) == 0)
1278 break;
1279
1280 bufoffset = 0;
1281 while (bufoffset < count) {
1282 uio.uio_iov = &iov;
1283 uio.uio_iovcnt = 1;
1284 iov.iov_base = buf + bufoffset;
1285 iov.iov_len = count - bufoffset;
1286 uio.uio_offset = offset + bufoffset;
1287 uio.uio_resid = iov.iov_len;
1288 uio.uio_rw = UIO_WRITE;
1289
1290 if ((error = VOP_WRITE(uvp, &uio, 0, cred)) != 0)
1291 break;
1292
1293 bufoffset += (count - bufoffset) - uio.uio_resid;
1294 }
1295
1296 uio.uio_offset = offset + bufoffset;
1297 }
1298
1299 free(buf, M_TEMP);
1300
1301 return (error);
1302 }
1303
1304 /*
1305 * Copy file from lower to upper.
1306 *
1307 * If you need copy of the contents, set 1 to docopy. Otherwise, set 0 to
1308 * docopy.
1309 *
1310 * If no error returned, unp will be updated.
1311 */
1312 int
unionfs_copyfile(struct unionfs_node * unp,int docopy,struct ucred * cred,struct thread * td)1313 unionfs_copyfile(struct unionfs_node *unp, int docopy, struct ucred *cred,
1314 struct thread *td)
1315 {
1316 struct mount *mp;
1317 struct vnode *udvp;
1318 struct vnode *lvp;
1319 struct vnode *uvp;
1320 struct vattr uva;
1321 int error;
1322
1323 lvp = unp->un_lowervp;
1324 uvp = NULLVP;
1325
1326 if ((UNIONFSTOV(unp)->v_mount->mnt_flag & MNT_RDONLY))
1327 return (EROFS);
1328 if (unp->un_dvp == NULLVP)
1329 return (EINVAL);
1330 if (unp->un_uppervp != NULLVP)
1331 return (EEXIST);
1332 udvp = VTOUNIONFS(unp->un_dvp)->un_uppervp;
1333 if (udvp == NULLVP)
1334 return (EROFS);
1335 if ((udvp->v_mount->mnt_flag & MNT_RDONLY))
1336 return (EROFS);
1337
1338 error = VOP_ACCESS(lvp, VREAD, cred, td);
1339 if (error != 0)
1340 return (error);
1341
1342 if ((error = vn_start_write(udvp, &mp, V_WAIT | V_PCATCH)) != 0)
1343 return (error);
1344 error = unionfs_vn_create_on_upper(&uvp, udvp, unp, &uva, td);
1345 if (error != 0) {
1346 vn_finished_write(mp);
1347 return (error);
1348 }
1349
1350 if (docopy != 0) {
1351 error = VOP_OPEN(lvp, FREAD, cred, td, NULL);
1352 if (error == 0) {
1353 error = unionfs_copyfile_core(lvp, uvp, cred, td);
1354 VOP_CLOSE(lvp, FREAD, cred, td);
1355 }
1356 }
1357 VOP_CLOSE(uvp, FWRITE, cred, td);
1358 VOP_ADD_WRITECOUNT_CHECKED(uvp, -1);
1359 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
1360 __func__, uvp, uvp->v_writecount);
1361
1362 vn_finished_write(mp);
1363
1364 if (error == 0) {
1365 /* Reset the attributes. Ignore errors. */
1366 uva.va_type = VNON;
1367 VOP_SETATTR(uvp, &uva, cred);
1368 }
1369
1370 unionfs_node_update(unp, uvp, td);
1371
1372 return (error);
1373 }
1374
1375 /*
1376 * It checks whether vp can rmdir. (check empty)
1377 *
1378 * vp is unionfs vnode.
1379 * vp should be locked.
1380 */
1381 int
unionfs_check_rmdir(struct vnode * vp,struct ucred * cred,struct thread * td)1382 unionfs_check_rmdir(struct vnode *vp, struct ucred *cred, struct thread *td)
1383 {
1384 struct vnode *uvp;
1385 struct vnode *lvp;
1386 struct vnode *tvp;
1387 struct dirent *dp;
1388 struct dirent *edp;
1389 struct componentname cn;
1390 struct iovec iov;
1391 struct uio uio;
1392 struct vattr va;
1393 int error;
1394 int eofflag;
1395 int lookuperr;
1396
1397 /*
1398 * The size of buf needs to be larger than DIRBLKSIZ.
1399 */
1400 char buf[256 * 6];
1401
1402 ASSERT_VOP_ELOCKED(vp, __func__);
1403
1404 eofflag = 0;
1405 uvp = UNIONFSVPTOUPPERVP(vp);
1406 lvp = UNIONFSVPTOLOWERVP(vp);
1407
1408 /* check opaque */
1409 if ((error = VOP_GETATTR(uvp, &va, cred)) != 0)
1410 return (error);
1411 if (va.va_flags & OPAQUE)
1412 return (0);
1413
1414 /* open vnode */
1415 #ifdef MAC
1416 if ((error = mac_vnode_check_open(cred, vp, VEXEC|VREAD)) != 0)
1417 return (error);
1418 #endif
1419 if ((error = VOP_ACCESS(vp, VEXEC|VREAD, cred, td)) != 0)
1420 return (error);
1421 if ((error = VOP_OPEN(vp, FREAD, cred, td, NULL)) != 0)
1422 return (error);
1423
1424 uio.uio_rw = UIO_READ;
1425 uio.uio_segflg = UIO_SYSSPACE;
1426 uio.uio_td = td;
1427 uio.uio_offset = 0;
1428
1429 #ifdef MAC
1430 error = mac_vnode_check_readdir(td->td_ucred, lvp);
1431 #endif
1432 while (!error && !eofflag) {
1433 iov.iov_base = buf;
1434 iov.iov_len = sizeof(buf);
1435 uio.uio_iov = &iov;
1436 uio.uio_iovcnt = 1;
1437 uio.uio_resid = iov.iov_len;
1438
1439 error = VOP_READDIR(lvp, &uio, cred, &eofflag, NULL, NULL);
1440 if (error != 0)
1441 break;
1442 KASSERT(eofflag != 0 || uio.uio_resid < sizeof(buf),
1443 ("%s: empty read from lower FS", __func__));
1444
1445 edp = (struct dirent*)&buf[sizeof(buf) - uio.uio_resid];
1446 for (dp = (struct dirent*)buf; !error && dp < edp;
1447 dp = (struct dirent*)((caddr_t)dp + dp->d_reclen)) {
1448 if (dp->d_type == DT_WHT || dp->d_fileno == 0 ||
1449 (dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1450 (dp->d_namlen == 2 && !bcmp(dp->d_name, "..", 2)))
1451 continue;
1452
1453 cn.cn_namelen = dp->d_namlen;
1454 cn.cn_pnbuf = NULL;
1455 cn.cn_nameptr = dp->d_name;
1456 cn.cn_nameiop = LOOKUP;
1457 cn.cn_flags = LOCKPARENT | LOCKLEAF | RDONLY | ISLASTCN;
1458 cn.cn_lkflags = LK_EXCLUSIVE;
1459 cn.cn_cred = cred;
1460
1461 /*
1462 * check entry in lower.
1463 * Sometimes, readdir function returns
1464 * wrong entry.
1465 */
1466 lookuperr = VOP_LOOKUP(lvp, &tvp, &cn);
1467
1468 if (!lookuperr)
1469 vput(tvp);
1470 else
1471 continue; /* skip entry */
1472
1473 /*
1474 * check entry
1475 * If it has no exist/whiteout entry in upper,
1476 * directory is not empty.
1477 */
1478 cn.cn_flags = LOCKPARENT | LOCKLEAF | RDONLY | ISLASTCN;
1479 lookuperr = VOP_LOOKUP(uvp, &tvp, &cn);
1480
1481 if (!lookuperr)
1482 vput(tvp);
1483
1484 /* ignore exist or whiteout entry */
1485 if (!lookuperr ||
1486 (lookuperr == ENOENT && (cn.cn_flags & ISWHITEOUT)))
1487 continue;
1488
1489 error = ENOTEMPTY;
1490 }
1491 }
1492
1493 /* close vnode */
1494 VOP_CLOSE(vp, FREAD, cred, td);
1495
1496 return (error);
1497 }
1498
1499