2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1994 Jan-Simon Pendry
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>
10 * This code is derived from software contributed to Berkeley by
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
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.
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
37 * @(#)union_subr.c 8.20 (Berkeley) 5/20/95
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/kernel.h>
46 #include <sys/mutex.h>
47 #include <sys/malloc.h>
48 #include <sys/mount.h>
49 #include <sys/namei.h>
51 #include <sys/vnode.h>
52 #include <sys/dirent.h>
53 #include <sys/fcntl.h>
54 #include <sys/filedesc.h>
56 #include <sys/sysctl.h>
57 #include <sys/taskqueue.h>
58 #include <sys/resourcevar.h>
60 #include <machine/atomic.h>
62 #include <security/mac/mac_framework.h>
66 #include <fs/unionfs/union.h>
68 #define NUNIONFSNODECACHE 16
69 #define UNIONFSHASHMASK (NUNIONFSNODECACHE - 1)
71 static MALLOC_DEFINE(M_UNIONFSHASH, "UNIONFS hash", "UNIONFS hash table");
72 MALLOC_DEFINE(M_UNIONFSNODE, "UNIONFS node", "UNIONFS vnode private part");
73 MALLOC_DEFINE(M_UNIONFSPATH, "UNIONFS path", "UNIONFS path private part");
75 static struct task unionfs_deferred_rele_task;
76 static struct mtx unionfs_deferred_rele_lock;
77 static STAILQ_HEAD(, unionfs_node) unionfs_deferred_rele_list =
78 STAILQ_HEAD_INITIALIZER(unionfs_deferred_rele_list);
79 static TASKQUEUE_DEFINE_THREAD(unionfs_rele);
81 unsigned int unionfs_ndeferred = 0;
82 SYSCTL_UINT(_vfs, OID_AUTO, unionfs_ndeferred, CTLFLAG_RD,
83 &unionfs_ndeferred, 0, "unionfs deferred vnode release");
85 static void unionfs_deferred_rele(void *, int);
91 unionfs_init(struct vfsconf *vfsp)
93 UNIONFSDEBUG("unionfs_init\n"); /* printed during system boot */
94 TASK_INIT(&unionfs_deferred_rele_task, 0, unionfs_deferred_rele, NULL);
95 mtx_init(&unionfs_deferred_rele_lock, "uniondefr", NULL, MTX_DEF);
103 unionfs_uninit(struct vfsconf *vfsp)
105 taskqueue_quiesce(taskqueue_unionfs_rele);
106 taskqueue_free(taskqueue_unionfs_rele);
107 mtx_destroy(&unionfs_deferred_rele_lock);
112 unionfs_deferred_rele(void *arg __unused, int pending __unused)
114 STAILQ_HEAD(, unionfs_node) local_rele_list;
115 struct unionfs_node *unp, *tunp;
116 unsigned int ndeferred;
119 STAILQ_INIT(&local_rele_list);
120 mtx_lock(&unionfs_deferred_rele_lock);
121 STAILQ_CONCAT(&local_rele_list, &unionfs_deferred_rele_list);
122 mtx_unlock(&unionfs_deferred_rele_lock);
123 STAILQ_FOREACH_SAFE(unp, &local_rele_list, un_rele, tunp) {
125 MPASS(unp->un_dvp != NULL);
127 free(unp, M_UNIONFSNODE);
130 /* We expect this function to be single-threaded, thus no atomic */
131 unionfs_ndeferred += ndeferred;
134 static struct unionfs_node_hashhead *
135 unionfs_get_hashhead(struct vnode *dvp, struct vnode *lookup)
137 struct unionfs_node *unp;
139 unp = VTOUNIONFS(dvp);
141 return (&(unp->un_hashtbl[vfs_hash_index(lookup) & UNIONFSHASHMASK]));
145 * Attempt to lookup a cached unionfs vnode by upper/lower vp
146 * from dvp, with dvp's interlock held.
148 static struct vnode *
149 unionfs_get_cached_vnode_locked(struct vnode *lookup, struct vnode *dvp)
151 struct unionfs_node *unp;
152 struct unionfs_node_hashhead *hd;
155 hd = unionfs_get_hashhead(dvp, lookup);
157 LIST_FOREACH(unp, hd, un_hash) {
158 if (unp->un_uppervp == lookup ||
159 unp->un_lowervp == lookup) {
160 vp = UNIONFSTOV(unp);
161 VI_LOCK_FLAGS(vp, MTX_DUPOK);
162 vp->v_iflag &= ~VI_OWEINACT;
163 if (VN_IS_DOOMED(vp) ||
164 ((vp->v_iflag & VI_DOINGINACT) != 0)) {
180 * Get the cached vnode.
182 static struct vnode *
183 unionfs_get_cached_vnode(struct vnode *uvp, struct vnode *lvp,
191 vp = unionfs_get_cached_vnode_locked(uvp, dvp);
192 else if (lvp != NULLVP)
193 vp = unionfs_get_cached_vnode_locked(lvp, dvp);
200 * Add the new vnode into cache.
202 static struct vnode *
203 unionfs_ins_cached_vnode(struct unionfs_node *uncp,
206 struct unionfs_node_hashhead *hd;
209 ASSERT_VOP_ELOCKED(uncp->un_uppervp, __func__);
210 ASSERT_VOP_ELOCKED(uncp->un_lowervp, __func__);
211 KASSERT(uncp->un_uppervp == NULLVP || uncp->un_uppervp->v_type == VDIR,
212 ("%s: v_type != VDIR", __func__));
213 KASSERT(uncp->un_lowervp == NULLVP || uncp->un_lowervp->v_type == VDIR,
214 ("%s: v_type != VDIR", __func__));
218 if (uncp->un_uppervp != NULL)
219 vp = unionfs_get_cached_vnode_locked(uncp->un_uppervp, dvp);
220 else if (uncp->un_lowervp != NULL)
221 vp = unionfs_get_cached_vnode_locked(uncp->un_lowervp, dvp);
223 hd = unionfs_get_hashhead(dvp, (uncp->un_uppervp != NULLVP ?
224 uncp->un_uppervp : uncp->un_lowervp));
225 LIST_INSERT_HEAD(hd, uncp, un_hash);
236 unionfs_rem_cached_vnode(struct unionfs_node *unp, struct vnode *dvp)
238 KASSERT(unp != NULL, ("%s: null node", __func__));
239 KASSERT(dvp != NULLVP,
240 ("%s: null parent vnode", __func__));
243 if (unp->un_hash.le_prev != NULL) {
244 LIST_REMOVE(unp, un_hash);
245 unp->un_hash.le_next = NULL;
246 unp->un_hash.le_prev = NULL;
252 * Common cleanup handling for unionfs_nodeget
253 * Upper, lower, and parent directory vnodes are expected to be referenced by
254 * the caller. Upper and lower vnodes, if non-NULL, are also expected to be
255 * exclusively locked by the caller.
256 * This function will return with the caller's locks and references undone.
259 unionfs_nodeget_cleanup(struct vnode *vp, struct unionfs_node *unp)
263 * Lock and reset the default vnode lock; vgone() expects a locked
264 * vnode, and we're going to reset the vnode ops.
266 lockmgr(&vp->v_lock, LK_EXCLUSIVE, NULL);
269 * Clear out private data and reset the vnode ops to avoid use of
270 * unionfs vnode ops on a partially constructed vnode.
274 vp->v_vnlock = &vp->v_lock;
275 vp->v_op = &dead_vnodeops;
280 if (unp->un_dvp != NULLVP)
282 if (unp->un_uppervp != NULLVP)
283 vput(unp->un_uppervp);
284 if (unp->un_lowervp != NULLVP)
285 vput(unp->un_lowervp);
286 if (unp->un_hashtbl != NULL)
287 hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, UNIONFSHASHMASK);
288 free(unp->un_path, M_UNIONFSPATH);
289 free(unp, M_UNIONFSNODE);
293 * Make a new or get existing unionfs node.
295 * uppervp and lowervp should be unlocked. Because if new unionfs vnode is
296 * locked, uppervp or lowervp is locked too. In order to prevent dead lock,
297 * you should not lock plurality simultaneously.
300 unionfs_nodeget(struct mount *mp, struct vnode *uppervp,
301 struct vnode *lowervp, struct vnode *dvp, struct vnode **vpp,
302 struct componentname *cnp)
305 struct unionfs_mount *ump;
306 struct unionfs_node *unp;
314 ump = MOUNTTOUNIONFSMOUNT(mp);
315 lkflags = (cnp ? cnp->cn_lkflags : 0);
316 path = (cnp ? cnp->cn_nameptr : NULL);
319 if (uppervp == NULLVP && lowervp == NULLVP)
320 panic("%s: upper and lower is null", __func__);
322 vt = (uppervp != NULLVP ? uppervp->v_type : lowervp->v_type);
324 /* If it has no ISLASTCN flag, path check is skipped. */
325 if (cnp && !(cnp->cn_flags & ISLASTCN))
328 /* check the cache */
329 if (dvp != NULLVP && vt == VDIR) {
330 vp = unionfs_get_cached_vnode(uppervp, lowervp, dvp);
333 goto unionfs_nodeget_out;
337 unp = malloc(sizeof(struct unionfs_node),
338 M_UNIONFSNODE, M_WAITOK | M_ZERO);
340 error = getnewvnode("unionfs", mp, &unionfs_vnodeops, &vp);
342 free(unp, M_UNIONFSNODE);
347 if (uppervp != NULLVP)
349 if (lowervp != NULLVP)
353 unp->un_hashtbl = hashinit(NUNIONFSNODECACHE, M_UNIONFSHASH,
355 KASSERT(hashmask == UNIONFSHASHMASK,
356 ("unexpected unionfs hash mask 0x%lx", hashmask));
360 unp->un_uppervp = uppervp;
361 unp->un_lowervp = lowervp;
363 if (uppervp != NULLVP)
364 vp->v_vnlock = uppervp->v_vnlock;
366 vp->v_vnlock = lowervp->v_vnlock;
369 unp->un_path = malloc(cnp->cn_namelen + 1,
370 M_UNIONFSPATH, M_WAITOK | M_ZERO);
371 bcopy(cnp->cn_nameptr, unp->un_path, cnp->cn_namelen);
372 unp->un_path[cnp->cn_namelen] = '\0';
373 unp->un_pathlen = cnp->cn_namelen;
379 * TODO: This is an imperfect check, as there's no guarantee that
380 * the underlying filesystems will always return vnode pointers
381 * for the root inodes that match our cached values. To reduce
382 * the likelihood of failure, for example in the case where either
383 * vnode has been forcibly doomed, we check both pointers and set
384 * VV_ROOT if either matches.
386 if (ump->um_uppervp == uppervp || ump->um_lowervp == lowervp)
387 vp->v_vflag |= VV_ROOT;
388 KASSERT(dvp != NULL || (vp->v_vflag & VV_ROOT) != 0,
389 ("%s: NULL dvp for non-root vp %p", __func__, vp));
391 vn_lock_pair(lowervp, false, uppervp, false);
392 error = insmntque1(vp, mp);
394 unionfs_nodeget_cleanup(vp, unp);
397 if (lowervp != NULL && VN_IS_DOOMED(lowervp)) {
399 unp->un_lowervp = NULL;
401 if (uppervp != NULL && VN_IS_DOOMED(uppervp)) {
403 unp->un_uppervp = NULL;
405 if (unp->un_lowervp == NULL && unp->un_uppervp == NULL) {
406 unionfs_nodeget_cleanup(vp, unp);
410 vn_set_state(vp, VSTATE_CONSTRUCTED);
412 if (dvp != NULLVP && vt == VDIR)
413 *vpp = unionfs_ins_cached_vnode(unp, dvp);
414 if (*vpp != NULLVP) {
415 unionfs_nodeget_cleanup(vp, unp);
426 if (lkflags & LK_TYPE_MASK)
427 vn_lock(vp, lkflags | LK_RETRY);
433 * Clean up the unionfs node.
436 unionfs_noderem(struct vnode *vp)
438 struct unionfs_node *unp, *unp_t1, *unp_t2;
439 struct unionfs_node_hashhead *hd;
440 struct unionfs_node_status *unsp, *unsp_tmp;
448 * The root vnode lock may be recursed during unmount, because
449 * it may share the same lock as the unionfs mount's covered vnode,
450 * which is locked across VFS_UNMOUNT(). This lock will then be
451 * recursively taken during the vflush() issued by unionfs_unmount().
452 * But we still only need to lock the unionfs lock once, because only
453 * one of those lock operations was taken against a unionfs vnode and
454 * will be undone against a unionfs vnode.
456 KASSERT(vp->v_vnlock->lk_recurse == 0 || (vp->v_vflag & VV_ROOT) != 0,
457 ("%s: vnode %p locked recursively", __func__, vp));
458 if (lockmgr(&vp->v_lock, LK_EXCLUSIVE | LK_NOWAIT, NULL) != 0)
459 panic("%s: failed to acquire lock for vnode lock", __func__);
462 * Use the interlock to protect the clearing of v_data to
463 * prevent faults in unionfs_lock().
466 unp = VTOUNIONFS(vp);
467 lvp = unp->un_lowervp;
468 uvp = unp->un_uppervp;
470 unp->un_lowervp = unp->un_uppervp = NULLVP;
471 vp->v_vnlock = &(vp->v_lock);
474 if (unp->un_hashtbl != NULL) {
476 * Clear out any cached child vnodes. This should only
477 * be necessary during forced unmount, when the vnode may
478 * be reclaimed with a non-zero use count. Otherwise the
479 * reference held by each child should prevent reclamation.
481 for (count = 0; count <= UNIONFSHASHMASK; count++) {
482 hd = unp->un_hashtbl + count;
483 LIST_FOREACH_SAFE(unp_t1, hd, un_hash, unp_t2) {
484 LIST_REMOVE(unp_t1, un_hash);
485 unp_t1->un_hash.le_next = NULL;
486 unp_t1->un_hash.le_prev = NULL;
492 writerefs = atomic_load_int(&vp->v_writecount);
493 VNASSERT(writerefs >= 0, vp,
494 ("%s: write count %d, unexpected text ref", __func__, writerefs));
496 * If we were opened for write, we leased the write reference
497 * to the lower vnode. If this is a reclamation due to the
498 * forced unmount, undo the reference now.
501 VNASSERT(uvp != NULL, vp,
502 ("%s: write reference without upper vnode", __func__));
503 VOP_ADD_WRITECOUNT(uvp, -writerefs);
511 unionfs_rem_cached_vnode(unp, dvp);
517 if (unp->un_path != NULL) {
518 free(unp->un_path, M_UNIONFSPATH);
523 if (unp->un_hashtbl != NULL) {
524 hashdestroy(unp->un_hashtbl, M_UNIONFSHASH, UNIONFSHASHMASK);
527 LIST_FOREACH_SAFE(unsp, &(unp->un_unshead), uns_list, unsp_tmp) {
528 LIST_REMOVE(unsp, uns_list);
532 mtx_lock(&unionfs_deferred_rele_lock);
533 STAILQ_INSERT_TAIL(&unionfs_deferred_rele_list, unp, un_rele);
534 mtx_unlock(&unionfs_deferred_rele_lock);
535 taskqueue_enqueue(taskqueue_unionfs_rele,
536 &unionfs_deferred_rele_task);
538 free(unp, M_UNIONFSNODE);
542 * Get the unionfs node status object for the vnode corresponding to unp,
543 * for the process that owns td. Allocate a new status object if one
544 * does not already exist.
547 unionfs_get_node_status(struct unionfs_node *unp, struct thread *td,
548 struct unionfs_node_status **unspp)
550 struct unionfs_node_status *unsp;
553 pid = td->td_proc->p_pid;
555 KASSERT(NULL != unspp, ("%s: NULL status", __func__));
556 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), __func__);
558 LIST_FOREACH(unsp, &(unp->un_unshead), uns_list) {
559 if (unsp->uns_pid == pid) {
565 /* create a new unionfs node status */
566 unsp = malloc(sizeof(struct unionfs_node_status),
567 M_TEMP, M_WAITOK | M_ZERO);
570 LIST_INSERT_HEAD(&(unp->un_unshead), unsp, uns_list);
576 * Remove the unionfs node status, if you can.
577 * You need exclusive lock this vnode.
580 unionfs_tryrem_node_status(struct unionfs_node *unp,
581 struct unionfs_node_status *unsp)
583 KASSERT(NULL != unsp, ("%s: NULL status", __func__));
584 ASSERT_VOP_ELOCKED(UNIONFSTOV(unp), __func__);
586 if (0 < unsp->uns_lower_opencnt || 0 < unsp->uns_upper_opencnt)
589 LIST_REMOVE(unsp, uns_list);
594 * Create upper node attr.
597 unionfs_create_uppervattr_core(struct unionfs_mount *ump, struct vattr *lva,
598 struct vattr *uva, struct thread *td)
601 uva->va_type = lva->va_type;
602 uva->va_atime = lva->va_atime;
603 uva->va_mtime = lva->va_mtime;
604 uva->va_ctime = lva->va_ctime;
606 switch (ump->um_copymode) {
607 case UNIONFS_TRANSPARENT:
608 uva->va_mode = lva->va_mode;
609 uva->va_uid = lva->va_uid;
610 uva->va_gid = lva->va_gid;
612 case UNIONFS_MASQUERADE:
613 if (ump->um_uid == lva->va_uid) {
614 uva->va_mode = lva->va_mode & 077077;
615 uva->va_mode |= (lva->va_type == VDIR ?
616 ump->um_udir : ump->um_ufile) & 0700;
617 uva->va_uid = lva->va_uid;
618 uva->va_gid = lva->va_gid;
620 uva->va_mode = (lva->va_type == VDIR ?
621 ump->um_udir : ump->um_ufile);
622 uva->va_uid = ump->um_uid;
623 uva->va_gid = ump->um_gid;
626 default: /* UNIONFS_TRADITIONAL */
627 uva->va_mode = 0777 & ~td->td_proc->p_pd->pd_cmask;
628 uva->va_uid = ump->um_uid;
629 uva->va_gid = ump->um_gid;
635 * Create upper node attr.
638 unionfs_create_uppervattr(struct unionfs_mount *ump, struct vnode *lvp,
639 struct vattr *uva, struct ucred *cred, struct thread *td)
644 if ((error = VOP_GETATTR(lvp, &lva, cred)))
647 unionfs_create_uppervattr_core(ump, &lva, uva, td);
655 * dvp should be locked on entry and will be locked on return.
657 * If an error is returned, *vpp will be invalid, otherwise it will hold a
658 * locked, referenced vnode. If *vpp == dvp then remember that only one
659 * LK_EXCLUSIVE lock is held.
662 unionfs_relookup(struct vnode *dvp, struct vnode **vpp,
663 struct componentname *cnp, struct componentname *cn, struct thread *td,
664 char *path, int pathlen, u_long nameiop)
669 cn->cn_namelen = pathlen;
671 cn->cn_nameiop = nameiop;
672 cn->cn_flags = (LOCKPARENT | LOCKLEAF | ISLASTCN);
673 cn->cn_lkflags = LK_EXCLUSIVE;
674 cn->cn_cred = cnp->cn_cred;
675 cn->cn_nameptr = cn->cn_pnbuf;
678 if (nameiop == DELETE) {
679 cn->cn_flags |= (cnp->cn_flags & DOWHITEOUT);
680 } else if (nameiop == RENAME) {
682 } else if (nameiop == CREATE) {
683 cn->cn_flags |= NOCACHE;
689 if ((error = vfs_relookup(dvp, vpp, cn, refstart))) {
690 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
694 KASSERT(cn->cn_pnbuf == path, ("%s: cn_pnbuf changed", __func__));
700 * relookup for CREATE namei operation.
702 * dvp is unionfs vnode. dvp should be locked.
704 * If it called 'unionfs_copyfile' function by unionfs_link etc,
705 * VOP_LOOKUP information is broken.
706 * So it need relookup in order to create link etc.
709 unionfs_relookup_for_create(struct vnode *dvp, struct componentname *cnp,
714 struct componentname cn;
717 udvp = UNIONFSVPTOUPPERVP(dvp);
720 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
721 cnp->cn_namelen, CREATE);
738 * relookup for DELETE namei operation.
740 * dvp is unionfs vnode. dvp should be locked.
743 unionfs_relookup_for_delete(struct vnode *dvp, struct componentname *cnp,
748 struct componentname cn;
751 udvp = UNIONFSVPTOUPPERVP(dvp);
754 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
755 cnp->cn_namelen, DELETE);
772 * relookup for RENAME namei operation.
774 * dvp is unionfs vnode. dvp should be locked.
777 unionfs_relookup_for_rename(struct vnode *dvp, struct componentname *cnp,
782 struct componentname cn;
785 udvp = UNIONFSVPTOUPPERVP(dvp);
788 error = unionfs_relookup(udvp, &vp, cnp, &cn, td, cnp->cn_nameptr,
789 cnp->cn_namelen, RENAME);
804 * Update the unionfs_node.
806 * uvp is new locked upper vnode. unionfs vnode's lock will be exchanged to the
807 * uvp's lock and lower's lock will be unlocked.
810 unionfs_node_update(struct unionfs_node *unp, struct vnode *uvp,
813 struct unionfs_node_hashhead *hd;
817 unsigned count, lockrec;
819 vp = UNIONFSTOV(unp);
820 lvp = unp->un_lowervp;
821 ASSERT_VOP_ELOCKED(lvp, __func__);
822 ASSERT_VOP_ELOCKED(uvp, __func__);
825 VNASSERT(vp->v_writecount == 0, vp,
826 ("%s: non-zero writecount", __func__));
828 * Update the upper vnode's lock state to match the lower vnode,
829 * and then switch the unionfs vnode's lock to the upper vnode.
831 lockrec = lvp->v_vnlock->lk_recurse;
832 for (count = 0; count < lockrec; count++)
833 vn_lock(uvp, LK_EXCLUSIVE | LK_CANRECURSE | LK_RETRY);
835 unp->un_uppervp = uvp;
836 vp->v_vnlock = uvp->v_vnlock;
840 * Re-cache the unionfs vnode against the upper vnode
842 if (dvp != NULLVP && vp->v_type == VDIR) {
844 if (unp->un_hash.le_prev != NULL) {
845 LIST_REMOVE(unp, un_hash);
846 hd = unionfs_get_hashhead(dvp, uvp);
847 LIST_INSERT_HEAD(hd, unp, un_hash);
849 VI_UNLOCK(unp->un_dvp);
854 * Create a new shadow dir.
856 * udvp should be locked on entry and will be locked on return.
858 * If no error returned, unp will be updated.
861 unionfs_mkshadowdir(struct unionfs_mount *ump, struct vnode *udvp,
862 struct unionfs_node *unp, struct componentname *cnp, struct thread *td)
871 struct ucred *credbk;
872 struct uidinfo *rootinfo;
875 if (unp->un_uppervp != NULLVP)
878 lvp = unp->un_lowervp;
880 credbk = cnp->cn_cred;
882 /* Authority change to root */
883 rootinfo = uifind((uid_t)0);
884 cred = crdup(cnp->cn_cred);
886 * The calls to chgproccnt() are needed to compensate for change_ruid()
887 * calling chgproccnt().
889 chgproccnt(cred->cr_ruidinfo, 1, 0);
890 change_euid(cred, rootinfo);
891 change_ruid(cred, rootinfo);
892 change_svuid(cred, (uid_t)0);
896 memset(&nd.ni_cnd, 0, sizeof(struct componentname));
899 if ((error = VOP_GETATTR(lvp, &lva, cnp->cn_cred)))
900 goto unionfs_mkshadowdir_abort;
902 if ((error = unionfs_relookup(udvp, &uvp, cnp, &nd.ni_cnd, td,
903 cnp->cn_nameptr, cnp->cn_namelen, CREATE)))
904 goto unionfs_mkshadowdir_abort;
912 goto unionfs_mkshadowdir_abort;
915 if ((error = vn_start_write(udvp, &mp, V_WAIT | V_PCATCH)))
916 goto unionfs_mkshadowdir_abort;
917 unionfs_create_uppervattr_core(ump, &lva, &va, td);
919 error = VOP_MKDIR(udvp, &uvp, &nd.ni_cnd, &va);
922 unionfs_node_update(unp, uvp, td);
925 * XXX The bug which cannot set uid/gid was corrected.
929 VOP_SETATTR(uvp, &va, nd.ni_cnd.cn_cred);
931 vn_finished_write(mp);
933 unionfs_mkshadowdir_abort:
934 cnp->cn_cred = credbk;
935 chgproccnt(cred->cr_ruidinfo, -1, 0);
942 * Create a new whiteout.
944 * dvp should be locked on entry and will be locked on return.
947 unionfs_mkwhiteout(struct vnode *dvp, struct componentname *cnp,
948 struct thread *td, char *path, int pathlen)
957 if ((error = unionfs_relookup(dvp, &wvp, cnp, &nd.ni_cnd, td, path,
970 if ((error = vn_start_write(dvp, &mp, V_WAIT | V_PCATCH)))
971 goto unionfs_mkwhiteout_free_out;
972 error = VOP_WHITEOUT(dvp, &nd.ni_cnd, CREATE);
974 vn_finished_write(mp);
976 unionfs_mkwhiteout_free_out:
981 * Create a new vnode for create a new shadow file.
983 * If an error is returned, *vpp will be invalid, otherwise it will hold a
984 * locked, referenced and opened vnode.
986 * unp is never updated.
989 unionfs_vn_create_on_upper(struct vnode **vpp, struct vnode *udvp,
990 struct unionfs_node *unp, struct vattr *uvap, struct thread *td)
992 struct unionfs_mount *ump;
1001 ump = MOUNTTOUNIONFSMOUNT(UNIONFSTOV(unp)->v_mount);
1003 lvp = unp->un_lowervp;
1004 cred = td->td_ucred;
1005 fmode = FFLAGS(O_WRONLY | O_CREAT | O_TRUNC | O_EXCL);
1008 if ((error = VOP_GETATTR(lvp, &lva, cred)) != 0)
1010 unionfs_create_uppervattr_core(ump, &lva, uvap, td);
1012 if (unp->un_path == NULL)
1013 panic("%s: NULL un_path", __func__);
1015 nd.ni_cnd.cn_namelen = unp->un_pathlen;
1016 nd.ni_cnd.cn_pnbuf = unp->un_path;
1017 nd.ni_cnd.cn_nameiop = CREATE;
1018 nd.ni_cnd.cn_flags = LOCKPARENT | LOCKLEAF | ISLASTCN;
1019 nd.ni_cnd.cn_lkflags = LK_EXCLUSIVE;
1020 nd.ni_cnd.cn_cred = cred;
1021 nd.ni_cnd.cn_nameptr = nd.ni_cnd.cn_pnbuf;
1025 if ((error = vfs_relookup(udvp, &vp, &nd.ni_cnd, false)) != 0)
1026 goto unionfs_vn_create_on_upper_free_out2;
1035 goto unionfs_vn_create_on_upper_free_out1;
1038 if ((error = VOP_CREATE(udvp, &vp, &nd.ni_cnd, uvap)) != 0)
1039 goto unionfs_vn_create_on_upper_free_out1;
1041 if ((error = VOP_OPEN(vp, fmode, cred, td, NULL)) != 0) {
1043 goto unionfs_vn_create_on_upper_free_out1;
1045 error = VOP_ADD_WRITECOUNT(vp, 1);
1046 CTR3(KTR_VFS, "%s: vp %p v_writecount increased to %d",
1047 __func__, vp, vp->v_writecount);
1051 VOP_CLOSE(vp, fmode, cred, td);
1054 unionfs_vn_create_on_upper_free_out1:
1057 unionfs_vn_create_on_upper_free_out2:
1058 KASSERT(nd.ni_cnd.cn_pnbuf == unp->un_path,
1059 ("%s: cn_pnbuf changed", __func__));
1065 * Copy from lvp to uvp.
1067 * lvp and uvp should be locked and opened on entry and will be locked and
1071 unionfs_copyfile_core(struct vnode *lvp, struct vnode *uvp,
1072 struct ucred *cred, struct thread *td)
1083 memset(&uio, 0, sizeof(uio));
1086 uio.uio_segflg = UIO_SYSSPACE;
1089 buf = malloc(MAXBSIZE, M_TEMP, M_WAITOK);
1091 while (error == 0) {
1092 offset = uio.uio_offset;
1097 iov.iov_len = MAXBSIZE;
1098 uio.uio_resid = iov.iov_len;
1099 uio.uio_rw = UIO_READ;
1101 if ((error = VOP_READ(lvp, &uio, 0, cred)) != 0)
1103 if ((count = MAXBSIZE - uio.uio_resid) == 0)
1107 while (bufoffset < count) {
1110 iov.iov_base = buf + bufoffset;
1111 iov.iov_len = count - bufoffset;
1112 uio.uio_offset = offset + bufoffset;
1113 uio.uio_resid = iov.iov_len;
1114 uio.uio_rw = UIO_WRITE;
1116 if ((error = VOP_WRITE(uvp, &uio, 0, cred)) != 0)
1119 bufoffset += (count - bufoffset) - uio.uio_resid;
1122 uio.uio_offset = offset + bufoffset;
1131 * Copy file from lower to upper.
1133 * If you need copy of the contents, set 1 to docopy. Otherwise, set 0 to
1136 * If no error returned, unp will be updated.
1139 unionfs_copyfile(struct unionfs_node *unp, int docopy, struct ucred *cred,
1149 lvp = unp->un_lowervp;
1152 if ((UNIONFSTOV(unp)->v_mount->mnt_flag & MNT_RDONLY))
1154 if (unp->un_dvp == NULLVP)
1156 if (unp->un_uppervp != NULLVP)
1158 udvp = VTOUNIONFS(unp->un_dvp)->un_uppervp;
1161 if ((udvp->v_mount->mnt_flag & MNT_RDONLY))
1164 error = VOP_ACCESS(lvp, VREAD, cred, td);
1168 if ((error = vn_start_write(udvp, &mp, V_WAIT | V_PCATCH)) != 0)
1170 error = unionfs_vn_create_on_upper(&uvp, udvp, unp, &uva, td);
1172 vn_finished_write(mp);
1177 error = VOP_OPEN(lvp, FREAD, cred, td, NULL);
1179 error = unionfs_copyfile_core(lvp, uvp, cred, td);
1180 VOP_CLOSE(lvp, FREAD, cred, td);
1183 VOP_CLOSE(uvp, FWRITE, cred, td);
1184 VOP_ADD_WRITECOUNT_CHECKED(uvp, -1);
1185 CTR3(KTR_VFS, "%s: vp %p v_writecount decreased to %d",
1186 __func__, uvp, uvp->v_writecount);
1188 vn_finished_write(mp);
1191 /* Reset the attributes. Ignore errors. */
1193 VOP_SETATTR(uvp, &uva, cred);
1196 unionfs_node_update(unp, uvp, td);
1202 * It checks whether vp can rmdir. (check empty)
1204 * vp is unionfs vnode.
1205 * vp should be locked.
1208 unionfs_check_rmdir(struct vnode *vp, struct ucred *cred, struct thread *td)
1215 struct componentname cn;
1224 * The size of buf needs to be larger than DIRBLKSIZ.
1228 ASSERT_VOP_ELOCKED(vp, __func__);
1231 uvp = UNIONFSVPTOUPPERVP(vp);
1232 lvp = UNIONFSVPTOLOWERVP(vp);
1235 if ((error = VOP_GETATTR(uvp, &va, cred)) != 0)
1237 if (va.va_flags & OPAQUE)
1242 if ((error = mac_vnode_check_open(cred, vp, VEXEC|VREAD)) != 0)
1245 if ((error = VOP_ACCESS(vp, VEXEC|VREAD, cred, td)) != 0)
1247 if ((error = VOP_OPEN(vp, FREAD, cred, td, NULL)) != 0)
1250 uio.uio_rw = UIO_READ;
1251 uio.uio_segflg = UIO_SYSSPACE;
1256 error = mac_vnode_check_readdir(td->td_ucred, lvp);
1258 while (!error && !eofflag) {
1260 iov.iov_len = sizeof(buf);
1263 uio.uio_resid = iov.iov_len;
1265 error = VOP_READDIR(lvp, &uio, cred, &eofflag, NULL, NULL);
1268 KASSERT(eofflag != 0 || uio.uio_resid < sizeof(buf),
1269 ("%s: empty read from lower FS", __func__));
1271 edp = (struct dirent*)&buf[sizeof(buf) - uio.uio_resid];
1272 for (dp = (struct dirent*)buf; !error && dp < edp;
1273 dp = (struct dirent*)((caddr_t)dp + dp->d_reclen)) {
1274 if (dp->d_type == DT_WHT || dp->d_fileno == 0 ||
1275 (dp->d_namlen == 1 && dp->d_name[0] == '.') ||
1276 (dp->d_namlen == 2 && !bcmp(dp->d_name, "..", 2)))
1279 cn.cn_namelen = dp->d_namlen;
1281 cn.cn_nameptr = dp->d_name;
1282 cn.cn_nameiop = LOOKUP;
1283 cn.cn_flags = LOCKPARENT | LOCKLEAF | RDONLY | ISLASTCN;
1284 cn.cn_lkflags = LK_EXCLUSIVE;
1288 * check entry in lower.
1289 * Sometimes, readdir function returns
1292 lookuperr = VOP_LOOKUP(lvp, &tvp, &cn);
1297 continue; /* skip entry */
1301 * If it has no exist/whiteout entry in upper,
1302 * directory is not empty.
1304 cn.cn_flags = LOCKPARENT | LOCKLEAF | RDONLY | ISLASTCN;
1305 lookuperr = VOP_LOOKUP(uvp, &tvp, &cn);
1310 /* ignore exist or whiteout entry */
1312 (lookuperr == ENOENT && (cn.cn_flags & ISWHITEOUT)))
1320 VOP_CLOSE(vp, FREAD, cred, td);